Correction: OPCML Is a Broad Tumor Suppressor for Multiple Carcinomas and Lymphomas with Frequently Epigenetic Inactivation

Background: Identification of tumor suppressor genes (TSGs) silenced by CpG methylation uncovers the molecular mechanism of tumorigenesis and potential tumor biomarkers. Loss of heterozygosity at 11q25 is common in multiple tumors including nasopharyngeal carcinoma (NPC). OPCML, located at 11q25, is one of the downregulated genes we identified through digital expression subtraction.


Introduction
Epigenetic silencing of tumor suppressor genes (TSGs) is frequently involved in tumor development and progression [1]. Aberrant methylation of promoter CpG islands (CGI) represents a major mechanism of this epigenetic inactivation, which leads to the binding of transcription repressors, compressed chromatin, and transcription silencing [2]. Identification of candidate TSGs silenced by promoter methylation thus uncovers the epigenetic mechanism of carcinogenesis and also identifies new epigenetic tumor markers for early cancer detection [3].
We previously used a new strategy to search for candidate TSGs genome-wide in NPC, through combining Differential Gene Expression Displayer (DGED) analysis with reported loss of heterozygosity (LOH) data of NPC (Liu & Tao, manuscript in preparation). This strategy revealed a number of putative TSGs that were down-regulated in NPC and also located at LOH loci. One of the in silico identified genes is OPCML (opioid binding protein/cell adhesion molecule-like gene), also known as OBCAM (opioid binding cell adhesion molecule), belonging to the IgLON (OPCML, LSAMP, NEGR1 and HNT) family of glycosylphosphatidylinositol (GPI)-anchored cell adhesion molecules that are highly expressed in the nervous system [15][16][17][18] and involved in cell adhesion and cell-cell recognition [19]. Located at 11q25, OPCML was the first IgLON member linked to tumorigenesis. It was initially identified as a TSG for epithelial ovarian cancer, being frequently inactivated by hemizygous deletion and promoter methylation [20]. More recent studies also demonstrated that OPCML is highly methylated in lung adenocarcinoma [21] and down-regulated in gastric and brain carcinomas [22,23], however no study has been reported for NPC and other tumors yet.
We thus systematically studied its alteration in a series of common tumors. As alternative splicing is a feature of OPCML [23] and other IgLONs (e.g. LSAMP) [24], we first studied its alternative splicing. We then examined its epigenetic inactivation in NPC and multiple other tumors which have not been studied for this gene, including esophageal, lung, gastric, hepatocellular, colorectal, breast, cervical and prostate carcinomas, as well as non-Hodgkin and Hodgkin lymphomas. We further found that OPCML is a stress-and p53-responsive gene; however, its stress response is epigenetically disrupted when the promoter becomes methylated. Ectopic expression of OPCML in tumor cell lines with endogenous silencing led to strong inhibition of cell colony formation, demonstrating that OPCML acts as a broad tumor suppressor.

Results
Identification of novel splicing variants of OPCML OPCML contains 7 exons and is transcribed from telomere to centromere (Fig. 1A). Among the four IgLON family members, OPCML shares the highest homology to HNT that lies approximately 80 kb centromeric to OPCML in the opposite orientation. Two alternative splice transcripts of OPCML, variant 1 (v1) (NM_002545) and variant 2 (v2) (NM_001012393), were previously identified in human, which differ only in their 59 exons ( Fig. 1B) but encode an identical mature protein [23].
We determined the transcription start sites of OPCML using 59-Rapid Amplification of cDNA Ends (59-RACE) in human brain and testis RNA. We obtained four PCR products of different sizes (Fig. 1C). Sequence analysis of the major PCR product (EU562296) indicated it as the v2 variant. The 59 end of v2 was found to be shorter (,110-bp downstream) than the published data, but its transcription start site matched exactly the DBTSS prediction (DataBase of Transcriptional Start Sites, http://dbtss. hgc.jp/). Three minor splice forms were also identified, designated v4 (EU562298), v5 (EU562299), and v6 (EU562300) (Fig. 1B and 1C left panel). Our 59-RACE gel electrophoresis failed to reveal a PCR band for the major transcript variant v1 in brain and testis tissues, probably due to the presence of a too large exon 1 (,1-kb) for v1, which would result in low amplification efficiency. Thus, we performed further RT-PCR using the 59-RACE product of brain and testis as template and primers specific to v1 (v1F/R4 and v1F0/R4; Fig. 1C right panel), to check whether v1 was expressed in normal tissues. This analysis did confirm the expression of v1 (EU562295) and identified another alternatively spliced variant v3 (EU562297) which is widely expressed in adult tissues (Fig. 1D). Further analysis using primers specific to the common exons (exon 2 and 3) of OPCML variants in cell lines without both v1 and v2 transcripts revealed the presence of even more unidentified, alternative promoter usage (Fig. 1F).
Broad expression of OPCML-v1 and v2 major variants in normal tissues Previously, OPCML was shown to be strongly expressed in brain and normal ovarian epithelia [20]. We further assessed its expression in 33 normal human adult and fetal tissues by semiquantitative PCR with specific primers targeting the v1, v2, or common exons (exon 2 and 3) (Fig. 1B), respectively. OPCML-v1 was widely expressed in all normal adult and fetal tissues except for placenta and peripheral blood mononuclear cells (PBMC), though at varying levels (highly expressed in brain, kidney, spleen, stomach, trachea, testis, cervix, ovary and prostate, and weakly in lung, breast, and bone marrow) (Fig. 1E). Compared to v1, OPCML-v2 displayed a more tissue-specific expression pattern in adult tissues, with expression absent or barely detectable in kidney, spleen, pancreas, breast, testis, lung, colon, liver, testis and bone marrow. In contrast to its expression in adult tissues, OPCML-v2 was expressed at moderate to high levels in all fetal tissues except for placenta. These results suggest that both v1 and v2 are likely to have important functions in embryonic development.

Silencing of OPCML by CpG methylation in tumor cell lines
We identified OPCML as a down-regulated gene through in silico subtraction. We further validated its expression in a large collection of carcinoma and lymphoma cell lines by semiquantitative RT-PCR. It was found that OPCML-v1 expression was dramatically reduced or completely silenced in multiple carcinoma cell lines of nasopharynx, esophagus, breast, cervix, stomach, lung, colon, liver and prostate, as well as in virtually all lymphoma cell lines examined ( Fig. 2C and Figure S1), but readily detected in glioma cell lines (Fig. 2D). In contrast, its expression was readily detected in most non-tumor cell lines, including normal mammary (HMEC, HMEpC) and prostate (PrEC-6) epithelial cell line, and immortalized but non-transformed epithelial cell lines (nasopharyngeal, NP69; esophageal, NE1, NE3 and Het-1A; prostate, MLCSV40). Thus, the downregulation of OPCML-v1 appeared to be tumor-specific. Notably, expression of OPCML-v2 remained undetectable in virtually all cell lines evaluated, including normal cell lines. Given its limited tissue expression pattern and the fact that v2 promoter is not a CpG island with very few CpG sites, the mechanism of OPCML-v2 silencing is not pursued further in the current study.
As methylation of promoter CGI is a well-recognized epigenetic mechanism of TSGs silencing [2], we thus examined the potential promoter regions of the 2 major variants (v1 and v2). The OPCML-v1 (NM_002545) and v2 (NM_001012393) sequence upstream of their exon 1 was retrieved from the NCBI database and analyzed using promoterinspector (http://www.genomatix. de) and CpG Island Searcher (http://ccnt.hsc.usc.edu/cpgis-lands2). This analysis predicted a promoter for OPCML-v1, located within a typical CGI spanning the published transcription start site of v1, which was also confirmed by our 59-RACE analysis ( Fig. 2A). We did not find any obvious CGI or predicted promoter in the region upstream of the exon 1 of v2, indicating that either the v2 promoter is not a typical one or the published 59-end sequence of v2 is not complete yet. Thus, we focused on the role of promoter CGI methylation in the silencing of OPCML-v1. We first validated that our methylation-specific PCR (MSP) was specific, which did not give any non-specific signal for the unbisulfate DNA  Figure S1 and Table 1), while no methylation was detected in the eight normal epithelial cell lines, demonstrating that v1 promoter methylation is well correlated with its expression status (Fig. 2C).
To further confirm the MSP results and examine the methylation status of the v1 CGI in more detail, we performed high-resolution bisulfite genome sequencing (BGS) analysis of 90 CpG sites within the island, spanning almost the entire predicted promoter. The BGS results were consistent with the MSP analysis, with all the promoter alleles extensively methylated in silenced cell lines and only scattered methylated CpG sites detected in non-tumor cell lines (Fig. 3). These results thus revealed a strong correlation between OPCML-v1 promoter methylation and its transcriptional silencing in tumor cell lines.

Pharmacologic and genetic demethylation restored OPCML-v1 expression
To determine whether methylation directly mediates OPCML silencing, carcinoma and lymphoma cell lines (MB231, Hep3B, HepG2, SNU398, SW480 and L1236) were treated with the DNA methyltransferase inhibitor 5-aza-29-deoxycytidine (Aza), together with or without histone deacetylase inhibitor Trichostatin A (TSA). The treatment resulted in the restoration of OPCML-v1 expression in all tumor cell lines (Fig. 4A). OPCML-v1 could also be induced in the colorectal cancer cell line HCT116 which is completely methylated for this gene, by genetic demethylation through double knock-out of both DNA methyltransferases DNMT1 and DNMT3B (DKO cell line) [25] (Fig. 4B). Concomitantly, the v1 promoter alleles were almost completely demethylated in DKO cells as confirmed by high-resolution BGS analysis (Fig. 4C), suggesting that the maintenance of OPCML methylation is mediated by DNMT1 and DNMT3B together, like other bona-fide TSGs that we and others have examined [12,13,26]. Interestingly, OPCML-v2 could not be activated in any drug treated cell line or DKO cell line (Fig. 4A, 4B), suggesting that the expression of OPCML-v2, being tissue-specific, is controlled by other intrinsic mechanism(s), and that its silencing in multiple carcinoma cell lines is controlled by methylation-independent mechanism or, less likely, that its upregulation level is below the limit of detection.

OPCML downregulation was not due to genetic deletion
The downregulation of OPCML in multiple tumor cell lines might also result from genetic deletion, as it resides in the frequently deleted 11q25 locus. Hemizygous deletion of OPCML was also often detected in epithelial ovarian cancer [20]. We thus performed multiplex differential genomic DNA PCR to detect OPCML deletion for a region spanning the frequently deleted marker D11S4085 in epithelial ovarian cancer. No homozygous deletion was detected in any silenced tumor cell line (Fig. 5). Furthermore, our high-resolution 1-Mb array comparative genomic hybridization (aCGH) analysis of NPC and ESCC cell lines [12,26,27] revealed the hemizygous deletion of OPCML in only 2 out of 15 cell lines (data not shown). Thus, downregulation of OPCML appears not to be due to genetic deletion, but rather predominantly to epigenetic silencing.

Frequent methylation of OPCML in multiple primary tumors
We further investigated the OPCML-v1 promoter methylation in a large collection of primary tumors, some with corresponding normal tissues as controls (Fig. 6A and Table 1). OPCML-v1 methylation was detected in 98% (42/43) of NPC, 66% (21/32) of esophageal, 91% (10/11) of breast, 64% (7/11) of gastric, 94% (17/18) of colorectal, 57% (4/7) of hepatocellular and 88% (7/8) of cervical carcinomas, as well as in 100% (10/10) of Burkitt lymphoma and 89% (8/9) of nasal lymphoma. Methylation was also detected with low frequency in paired surgical marginal tissues from patients with esophageal carcinoma at the rate of 16% (5/ 32), and with breast carcinoma at the rate of 25% (1/4), which might be due to the presence of small number of tumor cells disseminated into the adjacent non-tumorious region or an early tumor in the adjacent normal regions. Basically no methylation was detected in normal epithelial tissues (nasopharynx, esophagus and breast) except for very weak methylation in three nasopharyngeal, two esophageal and one breast epithelial tissues (Fig. 6A).
These results further demonstrated that methylation of OPCML-v1 promoter is frequent in multiple tumors. In contrast, no methylation was detected in all five prostate cancer samples (Fig. 6B).  Promoter methylation disrupted the stress response of OPCML-v1 Examination of the OPCML promoter revealed multiple HSF and p53 binding elements (MatInspector, http://genomatix.de), indicating that it is a stress-and p53-responsive gene (Fig. 7A). We thus inspected the response of OPCML to environmental stress stimuli. We found that the expression of OPCML-v1 was dramatically elevated in cell lines with an unmethylated promoter, after exposure to various stresses, such as heat shock, UV irradiation and H 2 O 2 treatment. On the contrary, this response was significantly decreased or abolished in cell lines with a methylated promoter (Fig. 7B). Interestingly, OPCML-v2 was not activated in any stresstreated cell line, indicating that it is not stress-responsive, probably due to its tissue-specific expression feature.
p53 could induce OPCML expression in the H1299 cell line with a partially methylated promoter, in a dosage-dependent manner (Fig. 7C). Taken together, these results demonstrated that OPCML is a stress-responsive and p53-regulated gene but its stress response is impaired by promoter methylation.

Ectopic expression of OPCML-v1 inhibited tumor cell clonogenicity
The frequent silencing of OPCML-v1 in multiple tumor cell lines and primary tumors but not normal epithelial tissues indicates that OPCML-v1 is likely a tumor suppressor. We thus sought to establish whether ectopic expression of OPCML-v1 could inhibit tumor cell clonogenicity. A mammalian expression vector encoding full-length OPCML-v1 was transfected into colorectal  (HCT116), esophageal (KYSE510) and prostate (PC3) carcinoma cell lines which had completely methylated and silenced endogeous OPCML-v1 promoter (Fig. 2C). The colony formation efficiencies of transfected cell line were evaluated by monolayer and soft agar culture. Ectopic expression of OPCML-v1 significantly inhibited the anchorage-dependent growth of three cell lines (down to 30%-60% of vector controls) ( Fig. 8A and 8C). Meanwhile, a significant reduction of colony formation efficiencies was observed in anchorage-independent growth of HCT116 cells (down to 30% of vector control) ( Fig. 8B and 8C). Thus, OPCML-v1 indeed has growth inhibitory activities in tumor cells and can function as a tumor suppressor.

Discussion
We used a novel approach of combining DGED screening for down-regulated genes with reported LOH data of NPC to search for silenced candidate TSGs genome-wide in NPC and identified OPCML. OPCML is frequently silenced by promoter methylation rather than genetic deletion in NPC, as well as multiple other carcinomas and lymphomas. We further showed that OPCML is a stress-responsive and p53-regulated gene, with the response abrogated when the promoter becomes methylated. In addition, ectopic expression of OPCML in carcinoma cells lacking its expression led to dramatic anchorage-dependent and -independent growth inhibition. Thus, our results demonstrate that OPCML is a broad functional tumor suppressor that is epigenetically silenced in multiple tumors.
OPCML belongs to the IgLON family of immunoglobulin (Ig) domain containing glycosylphosphatidylinositol (GPI)-anchored cell adhesion molecules, which includes OPCML, LSAMP, NEGR1 and HNT. The IgLON proteins are highly conserved between species and are typically composed of three Ig domains tethered to the surface of cell membrane by anchoring of their hydrophobic tails to GPI. Limited knowledge about the functions of IgLONs mainly derives from studies in rat and chick brain, the tissues where they are primarily expressed [17,28,29]. In those studies, IgLONs have been suggested to play an important role in cell adhesion and cell-cell recognition, through both homo-and hetero-philic interactions within the family [19]. Recently, it has been proposed that IgLONs function mainly as heterodimers called Diglons [30]. As a cell adhesion molecule, OPCML comprises several protein-protein interaction domains, such as three 'C2' like Ig domains [31] which are more appropriately classified as 'I' set Ig domains [20], commonly found in cellsurface-adhesion and receptor molecules [32]. Through these domains, OPCML may bind directly to growth promoting or inhibitory molecules and modulate their functions in tumor cells. Among the IgLON family, OPCML was the first member reported to possess tumor suppressor functions in epithelial ovarian cancer, being frequently silenced genetically and epigenetically at the early step of ovarian carcinogenesis [20]. This was followed by another report that another IgLON, LSAMP, is also a TSG for renal clear cell carcinoma [33]. Our present study further verifies that OPCML can function as a broad TSG and is frequently inactivated epigenetically in multiple carcinomas and lymphomas, including NPC, esophageal, lung, gastric, hepatocellular, colorectal, breast, cervical and prostate carcinomas. OPCML probably functions as a tumor suppressor through interacting with other IgLONs to form heterodimeric complex [30] involved in signal transduction. Loss of OPCML reduces the intercellular adhesion and heterodimeric complex formation and thus impairs the corresponding signaling pathways, thereby promoting the progress of carcinogenesis.
OPCML shares the highest homology to HNT among the four IgLON family members. Notably, the coding region in exon 1 of OPCML-v1 and HNT is identical, and so is the exon 2 except for only several bases. The first Ig domains of these two proteins share 92% identity, while the second and third Ig domains share 70% and 66% identity, respectively. This raises the possibility that OPCML and HNT may originate from the same ancestor by gene conversion during evolution. Thus, primers must be cautiously designed for these two genes to avoid cross-amplification with PCR-based techniques.
Our results also reveal that OPCML transcripts v1 and v2 have different tissue expression patterns. Whereas OPCML-v1 was widely expressed in normal adult tissues, OPCML-v2 showed a more tissue-specific expression profile, being highly expressed in few tissues including brain. Previously, a genome-wide searching for the neuron specific silencer REST/NRSF binding sites (RE1/ NRSE) revealed that there were three NRSE located at intron 1 of OPCML-v2 (http://bioinformatics.leeds.ac.uk/group/online/ RE1db/re1db_home.htm), suggesting that v2 may be a more neuron specific transcript. We also identified other novel isoforms of OPCML (v3, v4, v5, v6), derived from alternative splicing or promoter usages. Using primers specific to the common exons of OPCML transcripts, we found the expression of OPCML in several tumor cell lines (Hep3B, H292, SW480, L1236), where the OPCML-v1 and v2 were totally silenced (Fig. 1F), indicating transcription of OPCML from alternative unknown promoters. Our present study mainly focused on the expression and functional analysis of transcript variant 1, whereas the mechanism of variant 2 silencing was not pursued further. Further studies are needed to characterize these novel splicing variants, their promoter usages and possible biologic functions.
Epigenetic gene silencing is associated with the onset and progression of various cancers [2]. The frequent, predominant epigenetic inactivation of OPCML in multiple malignancies points to the importance of this gene in tumorigenesis. OPCML is a stressand p53-responsive gene, but this response was often epigenetically impaired by promoter methylation. We speculate that epigenetic silencing of OPCML would impair the cellular protective response to environmental stresses in normal cells, thus promoting the development of cancers. As promoter methylation of OPCML was pharmacologically and genetically reversible, pharmacologic demethylation therapy will restore its response to stress and p53. The role of OPCML in DNA damage repair, apoptosis and cell cycle arrest with respect to stress response remains to be further investigated. We also noticed that in some cell lines (like HCC), OPMCL-v1 was silenced without promoter methylation detected by MSP. It could be that for some cell lines, the methylation is not evenly distributed through the CGI (like HepG2 in Fig. 3) and is thus missed by MSP analysis, or additional alternative mechanism such as histone modification is involved.
In summary, we found that the expression of OPCML-v1 (NM_002545), a major transcript of this TSG, is frequently silenced or down-regulated in multiple tumors. This inactivation is due to its promoter methylation, which further impairs its response to environmental stresses. We further demonstrated that OPCML acts as a broad tumor suppressor for multiple tumor types. The high incidence of epigenetic inactivation of OPCML in NPC and esophageal carcinoma, both prevalent in our locality, indicates that OPCML methylation could be an epigenetic biomarker for the molecular diagnosis of these tumors.

Digital expression subtraction
We searched for downregulated genes genome-wide through Differential Gene Expression Displayer (DGED) analysis (cDNA DGED and SAGE DGED) (http://cgap.nci.nih.gov). This analysis identified a number of downregulated genes in tumors. The candidate gene list was further filtered with the reported loss of heterozygosity (LOH) data of NPC. Genes located at published LOH regions in NPC were extracted using UCSC genome database (http://genome.ucsc.edu).

59-Rapid Amplification of cDNA Ends (59-RACE)
We determined the OPCML transcription start site using 59-RACE version 2.0 (Invitrogen). Briefly, the first-strand cDNA was synthesized from brain RNA using primer OPCML-DxR, 59-TCCAGGTACTCATCCTCACT. Homopolymeric tails were then added to the 39ends with terminal deoxynucleotidyl transferase. PCR was done using Abridged Anchor Primer and a second gene-specific primer OPCML-R2, 59-CTGCCAATAG-CAAGACACAG. The RACE product was enriched by reamplifying with the Abridged Universal Amplification Primer and OPCML-R, 59-TATGGACCACTTGTCATTCC, cloned and sequenced.

Semi-quantitative RT-PCR analysis
Reverse transcription-PCR (RT-PCR) was performed for 36 or 37 cycles with hot-start, using AmpliTaq Gold DNA Polymerase (Applied Biosystems, Foster City, CA) and GAPDH as a control [36]. RT-PCR primers were designed to span introns to prevent amplification of genomic DNA. Primer sequences are provided in Table S1.

Bisulfite treatment and promoter methylation analysis
Bisulfite modification of DNA, methylation-specific PCR (MSP) and bisulfite genomic sequencing (BGS) were carried out as previously described [26,36]. Both MSP and BGS were performed for 40 cycles using AmpliTaq Gold with hot-start. MSP primers were tested first for not amplifying any unbisulfited DNA. For BGS, the PCR products were cloned into pCR4-TOPO (Invitrogen, Carlsbad, CA), with 5-8 colonies randomly chosen and sequenced. Primer sequences are shown in Table S1.

Stress treatments
Heat shock was done as previously described [7], except for an incubation at 42uC for 1 hour with recovery at 37uC for 2 hours. For UV treatment, medium was removed and the flask was turned upside down to face the light source in a UV cross-linker (Amersham Biosciences, Piscataway, NJ). Cells were irradiated for a dose of 70 J/ m 2 . After irradiation, fresh medium was added, and the cells were recovered at 37uC for 1 hour and then harvested. For H 2 O 2 treatment, cells were exposed to 0.5 mM of H 2 O 2 for 1 hour and then harvested.

Deletion analysis of OPCML by multiplex PCR
Homozygous deletion of OPCML was examined using multiplex genomic DNA PCR as previously described [7]. Primer sequences are shown in Table S1. The final concentration of OPCML and GAPDH primers is 0.4 mM and 0.2 mM, respectively. PCR products were analyzed on 1.8% agarose gels.

Colony formation assays
The full-length OPCML-v1 ORF was subcloned from the pcDNA3.1 Zeo/OPCML plasmid [20] into pcDNA3.1(+) to generate pcDNA3.1+/OPCML-v1. HCT-116, KYSE510 and PC3 cells were seeded at 1610 5 /well in a 12-well plate and allowed to grow for 24h. Cells were then transiently transfected with 0.5 mg of pcDNA3.1+/OPCML-v1 or pcDNA 3.1 vector alone, using Fugene6.0 (Roche, Switzerland). For colony formation assay using monolayer culture, cells were collected and plated in a 6-well plate 48h post-transfection, and selected for 1 to 2 weeks with G418 (0.4mg/ml). Surviving colonies ($50 cells/colony) were counted after staining with Gentian Voilet (ICM Pharma, Singapore). For colony formation assay using soft agar culture, at 48h posttransfection, cells were suspended in RPMI 1640 containing 0.35% agar, 10% fetal bovine serum and 0.4 mg/ml G418 and layered on RPMI containing 0.5% agar, 10% fetal bovine serum and G418 in a 6-well plate. Colonies were photographed at day 20 posttransfection. All the experiments were performed in triplicate wells for three times. Data were presented as relative colony formation ability6SD. Statistical analysis was carried out by Student's t-test, p,0.01 was considered as statistically significant difference.

Supporting Information
Table S1 PCR primers used in this study.