Molecular Cancer BioMed Central

Background Selective cyclooxygenase (COX)-2 inhibitors elicit anti-proliferative responses in various tumours, however the underlying anti-tumour mechanisms are unclear. Mutational inactivation of the tumour suppressor p53 gene is frequent in malignant gliomas. The role of p53 mutation in the anti-tumour responses of the selective COX-2 inhibitor celecoxib in human glioblastoma cells is unknown. In this study, we used human glioblastoma cells with various p53 status; U87MG (with high and low p53 functional levels), LN229 (functional p53) and U373MG (mutant p53) cells. Inhibition of p53 was achieved in U87MG cells transfected with E6 oncoprotein (U87MG-E6) and treated with pifithrin-α, a reversible inhibitor of p53 (U87MG-PFT). We investigated whether the anti-glioblastoma responses of celecoxib were p53-dependent, and whether celecoxib induced DNA damage leading to p53-dependent G1 cell cycle arrest, followed by autophagy or apoptosis. Results Our findings demonstrated that celecoxib concentration-dependently reduced glioblastoma cell viability, following 24 and 72 hours of treatment. Inhibition of functional p53 in glioblastoma cells significantly reduced the anti-proliferative effect of celecoxib. In U87MG cells, celecoxib (8 and 30 μM) significantly induced DNA damage and inhibited DNA synthesis, corresponding with p53 activation. Celecoxib induced G1-phase cell cycle arrest, accompanied with p21 activation in U87MG cells. Cell cycle progression of U87MG-E6 and U87MG-PFT cells was not affected by celecoxib. In parallel, celecoxib induced G1 cell cycle arrest in LN229 cells, but not in U373MG cells. Autophagy was induced by celecoxib in U87MG and LN229 cells, as shown by the significantly greater population of acridine orange-stained cells and increased levels of LC3-II protein (in comparison with non-treated controls). Celecoxib did not induce significant autophagy in U87MG-PFT, U87MG-E6 and U373MG cells, which lack functional p53. Regardless of p53 status, celecoxib caused no significant difference in apoptosis level of U87MG, U87MG-PFT, U87MG-E6 and U373MG cells. Conclusion Our findings reveal that p53 increases human glioblastoma sensitivity to celecoxib. Celecoxib inhibits glioblastoma cell viability by induction of DNA damage, leading to p53-dependent G1 cell cycle arrest and p53-dependent autophagy, but not apoptosis.


Background
Epithelial ovarian cancer is the most lethal gynecological cancer among women worldwide. Metastasis occurs in most patients before diagnosis, and the majority of patients have high subsequent recurrent rate after completing surgery and chemotherapy [1]. Survival is at about 3 years for patients in advanced stages [2]. The poor prognosis of ovarian cancer is due to the difficulty in early diagnosis and the detrimental processes of invasion and metastasis. A better understanding of the molecular mechanisms of cancer development and progression will help to improve the diagnosis and treatment of the disease.
Insulin-like growth factor binding proteins (IGFBPs) are circulating transport proteins for IGF, with IGFBP-3 being the predominant IGFBP in circulation [3]. IGFBP-3 can regulate cell growth and death, either dependent or independent of its interaction with IGF [3,4]. Recently, we have further identified IGFBP-3 as an invasion suppressor gene using an established ovarian cancer cell line OVTW59-P0 and its sublines P1 to P4, which were obtained from a gradual invasion model with sequential increase in invasiveness [5]. Many tumor suppressor genes involved in cancer formation and progression are frequently epigenetically silenced through aberrant hypermethylation at their promoter regions [6]. In ovarian cancer, genes with promoter hypermethylation are frequently found to be related to cancer progression [7,8]. Aberrant promoter hypermethylation of IGFBP-3 and gene silencing are observed in many cancers, such as lung, hepatocellular, gastric, colorectal, breast, and ovarian cancers [9][10][11][12][13]. However, the association of IGFBP-3 promoter hypermethylation with poor clinical outcome was identified only at early stages in lung and ovarian cancers [9,10].
p53 is a known transcription factor for IGFBP-3 expression [9]. Induction of IGFBP-3 by p53 has been shown to cause cell apoptosis in an IGF-independent manner [14]. Eleven p53 binding sites have been identified within IGFBP-3 gene based on the homology to the p53 binding consensus sequence, and confirmation by electrophoretic mobility shift analyses [15]. Promoter hypermethylation at these p53 binding sites caused gene silencing and resistance to p53 [16]. Therefore, it has been suggested that p53 could mediate cross-talk to the IGF axis through IGFBP-3 regulation [14].
Pathologically, epithelial ovarian cancer is classified into four major histological subtypes: serous, mucinous, endometrioid and clear cell carcinoma. Each subtype is associated with distinct molecular alterations [17]. In our previous study, we found 51.4% of the ovarian endometrioid carcinoma (OEC) subtype showing lower IGFBP-3 expression, which is associated significantly with poor patient outcome [5]. From literature review, p53 overexpression is more frequently reported in the serous subtype of ovarian cancer [18][19][20]. The association of altered p53 expression in tumor tissue with patient survival in ovarian cancer is still under debate [19][20][21], but a significant correlation is often reported in the serous subtype [18,19].
Despite the close molecular regulation between p53 and IGFBP-3 being known for more than a decade [9], little information concerning p53 regulation of IGFBP-3 or the effect of epigenetic inactivation of IGFBP-3 exist with defined clinical endpoints. In the current study, we analyzed the association between p53 and IGFBP-3 expression in ovarian cancer progression. We explored the clinical significance of aberrant promoter hypermethylation at the p53 binding sites of IGFBP-3 in OEC. Functional analysis of p53 regulation on IGFBP-3 expression was further assessed using the OVTW59-P0 and P4 cell lines. Our study indicates that methylation at the p53 transcription factor binding sites in the IGFBP-3 promoter can silence IGFBP-3 expression and hence lead to cancer progression in OEC.

IGFBP-3 expression was correlated with IGFBP-3 promoter methylation, but not p53 expression
A total of 60 OEC patients with an average age at surgery of 51.5 ± 12.4 years old were recruited. Among these patients, 41.7% were in the advanced disease stages (stages 3 and 4) and 61.8% at higher histological grades (grades 2 and 3) ( Table 1). By immunohistochemical staining, IGFBP-3 was located in the tumor cytoplasm and p53 was immunoreactive at the nucleus ( Figure 1).
The amount of methylation at IGFBP-3 promoter in these patient samples was measured by quantitative real time methylation specific PCR (qMSP). Twenty samples were unable to be bisulfite converted probably due to failed in DNA extraction from very old paraffin blocks. In total, qMSP analysis was performed on 40 patient samples. There were no differences in IGFBP-3 and p53 expressions or in clinicopathological features between the samples which were unable to be analyzed by qMSP and those which were (Additional file 1, Table S1). Among these 40 samples, the calculated promoter methylation of IGFBP-3 ranged from 0.01 to 59.8%. By using receiver operating characteristic (ROC) curve and area under the curve (AUC) analysis methods, the optimal cutoff value was set at a threshold of > 3% as hypermethylation. At this threshold, the sensitivity of IGFBP-3 methylation with low IGFBP-3 expression was 0.86 and the specificity was 0.73 (Additional file 1, Figure S1). Table 1 shows the correlation among IGFBP-3 and p53 expression and qMSP results with the clinical parameters. High IGFBP-3 expression was found in tumors of lower grades and lower stages, and p53 tended to be overexpressed in tumors with higher grades. No correlation was found between p53 overexpression and IGFBP-3 expression or IGFBP-3 promoter methylation. Cases that showed higher IGFBP-3 methylation (at threshold > 3%) were significantly correlated with lower IGFBP-3 expression and higher histological grades.

Immunohistochemical staining
The calculated IGFBP-3 methylation from qMSP analysis was further subdivided into four quartiles, and Table 2 shows the relation between IGFBP-3 expression and clinicopathological features for each quartile. Again, we observed a close association of high IGFBP-3 expression with ≤ 3% methylation and low IGFBP-3 expression with > 3% methylation. However, 2 cases showed low IGFBP-3 expression and ≤ 3% methylation and 8 cases showed high IGFBP-3 expression and > 3% methylation. These associations were opposite to the parallel relationship of methylation and IGFBP-3 suppression and were thus grouped into the unparallel category (Table 3).

Clinical correlation of IGFBP-3 expression, p53 overexpression and IGFBP-3 promoter methylation with survival rates
Cox-regression analysis revealed low IGFBP-3 expression to be associated with lower progression-free survival (PFS) and lower overall survival (OS) rates. p53 overexpression  Advanced stage was defined as stage 3 or 4.
Higher histological grade was defined as grade 2 or 3. was not associated with patient survivals while high IGFBP-3 promoter methylation was marginally associated with lower OS. When combining these variables for survival analysis, low IGFBP-3 expression, high IGFBP-3 promoter methylation and normal p53 expression were correlated most significantly with lower PFS and OS. However, since IGFBP-3 expression and IGFBP-3 promoter methylation were correlated with disease stages and tumor grades (Table 1), the associations of these variables with survival rates was not independent of age, stage and grade (Table 3). Figure 2 shows Kaplan Meier survival estimates of patients classified by IGFBP-3 expression, methylation levels of IGFBP-3 promoter and p53 expression. For comparison, Additional file 1, Figure S2 shows the survival estimates classified by IGFBP-3 expression and IGFBP-3 promoter methylation levels in all patients with tumors that showed either normal or overexpressed p53. The association of patient PFS and OS with IGFBP-3 expression and IGFBP-3 promoter methylation were less significant comparing to the results when only cases with normal p53 were included ( Figure 2d).

Correlation of IGFBP-3 promoter methylation and IGFBP-3 expression in ovarian cancer cell lines
To have a better understanding of the molecular regulation of IGFBP-3, we included two established OEC sublines OVTW59-P0 and P4 for further study. As reported previously, P4 was more invasive and metastatic than P0; and P4 expressed very low IGFBP-3 compared to P0 [5].
Using MSP, we detected both 129 bp methylated and 158 unmethylated bands in P0 indicating that P0 was partially methylated at the IGFBP-3 promoter sites. P4 showed only a 129 bp band indicating a complete methylation of the IGFBP-3 promoter (Figure 3a). 5-aza-deoxycytidine (5-aza-dC) treatment restored IGFBP-3 expression in P4. The amount of 5-aza-dC used for treatment directly correlated with IGFBP-3 mRNA and protein expressions (Figure 3b).

Transcriptional regulation of p53 on IGFBP-3 promoter
Both P0 and P4 expressed similar levels of wild type (wt) p53 by direct sequencing after amplifying exons 5, 6, 7, and 8 of p53 gene, and Western blot analysis (data not shown). To test the transcriptional regulation of IGFBP-3 by p53, the -253 ~+61 portion of the promoter and first exon of IGFBP-3 were transfected into P0 and P4. After treatment with a known p53 inhibitor, pifithrin-α, the luciferase activity representing IGFBP-3 expression was completely abolished in these two sublines (Figure 3c). This result indicates that promoter sites between -253 +61 are essential for transcriptional regulation by p53 on IGFBP-3 expression

Identification of methylation hotspots at the p53 consensusbinding sequence in IGFBP-3 promoter
Bisulfite-PCR and sequencing (BSP) was performed in P0 and P4, focusing on the -330 to +140 sites of IGFBP-3 that included the reported 11 p53 consensus binding sites [22]. We identified four CpG sites that were frequently methylated in these two sublines, distributed at the -210, P4 showed a higher frequency of methylation at these loci compared to P0 (60.7% versus 25.0%) (Figure 4b).
Since p53 binding at these sequences has been reported to be diminished by methylation [16], we established a series of site-directed mutation constructs to test the importance of these four loci on the transcriptional regulation by p53 of IGFBP-3 (Figure 4c). In addition to using P0 and P4, a human embryonic 293T kidney cell line with documented wt p53 was also included in analysis. The transcription activity of IGFBP-3 was significantly reduced after mutagenesis at these loci and the amount of luciferase activity corresponding to IGFBP-3 expression decreased as the number of mutated loci increased ( Figure  4d). These findings indicate that these four loci are critical spots for transcriptional regulation by p53 of IGFBP-3. In addition, methylation at these sites could quantitatively silence the process of IGFBP-3 induction by p53.

Discussion
We demonstrated previously that IGFBP-3 is an invasion suppressor in OEC [5]. In this study, we evaluated the molecular mechanism of IGFBP-3 regulation in cancer cells and found that methylation at the CpG sites of the IGFBP-3 promoter was significantly correlated with IGFBP-3 silencing in OEC tumors. Clinically, patients with lower IGFBP-3 expression and higher IGFBP-3 promoter methylation have significantly lower PFS and OS. This association was strongest in cases with normal p53 expression. In an in vitro study using OEC cell line OVTW59, we identified a quantitative correlation between IGFBP-3 suppression and methylation in the first 200 base pairs surrounding the transcription start site of IGFBP-3. Our data suggest that IGFBP-3 silencing through aberrant hypermethylation at its promoter is a major regulatory pathway for OEC tumorigenesis.
Methylation specific PCR (MSP) has been widely used to study promoter methylation in clinical specimens. By MSP, aberrant methylation at the IGFBP-3 promoter was found in 44% of epithelial ovarian cancer and in 28.6% (12/42) of OEC subtype [7,12]. However, IGFBP-3 silencing has been reported unrelated to IGFBP-3 promoter methylation [7,12]. A significant association between IGFBP-3 promoter methylation and disease progression has been reported, but only in early-stage ovarian cancer [12]. It was concluded that methylation at IGFBP-3 promoter is a common tumorigenesis process in the early steps of ovarian cancer progression [12]. By setting the calculated percentage of IGFBP-3 methylation at 3% as a cutoff point for qMSP analysis, we found methylation at the IGFBP-3 promoter significantly correlated with IGFBP-3 silencing in OEC. Furthermore, we found patients with lower IGFBP-3 expression and higher IGFBP-3 promoter methylation to be significantly associated with poor outcomes. A possible reason is that we have used a more sen-sitive method, qMSP, to study these epigenetic events. In addition, we have focused our study on the OEC subtype of ovarian cancer. It has been observed that promoter methylation of specific genes in cancer occur frequently in a tumor-type and cell-type specific manner [23]. Our results demonstrate that the clinical significance of aberrant IGFBP-3 promoter methylation is more commonly observed in the OEC subtype of ovarian cancer.
Ten cases in our study showed unparallel results between IGFBP-3 expression and IGFBP-3 promoter methylation, i.e. either high IGFBP-3 and IGFBP-3 promoter methylation > 3% or low IGFBP-3 and IGFBP-3 promoter methylation ≤ 3%. Though threshold of > 3% as hypermethylation was selected based on ROC curve analysis that showed 0.86 sensitivity and 0.73 specificity of low IGFBP-3 expression, we found five (50%) of these unparallel cases were at the third quartile of methylation, i.e. high IGFBP-3 expression and promoter methylation levels between 3% and 6.36% (Table 2). This could be due to inadequate sensitivity and specificity of qMSP assay. Alternatively, the unparallel result could be explained by multiple mechanisms of IGFBP-3 regulation. IGFBP-3 could also interact with several other growth-inhibitory agents to mediate wide varieties of growth suppression signal in the absence of IGF [24,25]. In addition, other CpG methylation sites in the IGFBP-3 promoter that were not studied might also contribute to IGFBP-3 silencing We further explored the association between p53 and IGFBP-3 promoter methylation and the mechanism of p53 regulation on IGFBP-3. Thirty-five percent of these cases showed p53 overexpression that was correlated with higher tumor grade at marginal significance. However, overexpression of p53 was not correlated with IGFBP-3 expression or IGFBP-3 promoter methylation and it did not correlate with OEC patient survival. This suggested that p53 is not important in the progression of OEC, and that IGFBP-3 silencing through promoter methylation is a mechanism different from p53 overexpression. Since OEC is similar in histological pattern to the endometrioid (EC) subtype of endometrial cancer, we reviewed the molecular characteristics of different subtypes of endometrial cancer. In endometrial cancer, p53 alteration was specifically reported to be present in the non-endometrioid (non-EC) serous subtype, which has been classified as a type 2 endometrial cancer. The EC subtype of endometrial cancer is classified as type 1 endometrial cancer [26]. Accordingly, Kurman et al. proposed an epithelial ovarian cancer model composed of type 1 and type 2 tumors based on: the histological patterns, molecular features, and the process of tumor progression. Type 1 tumors are slow growing, genetically stable, and are characterized by mutations in KRAS, BRAF, PTEN, and beta-catenin. Type 2 tumors are highly aggressive, rapidly growing, genetically instable, and are characterized by mutation of TP53 [20]. Based on Identification of methylation sites at p53 binding sequence of IGFBP-3 promoter this model, the OEC subtype is classified as a type 1 tumor and serous subtype, with characteristic TP53 mutations, is classified as a type 2 tumor [20]. In our study, we confirmed that p53 alteration is not a characteristic of OEC. Instead, low IGFBP-3 and high IGFBP-3 promoter methylation are major prognostic factors for OEC. By survival analysis, patient survival is lowest in cases with low IGFBP-3, high IGFBP-3 promoter methylation, and normal p53. These suggest OEC is a distinct subtype of ovarian cancer that IGFBP-3 silencing through IGFBP-3 promoter methylation could play an important role in cancer development and progression. We found p53 alteration is not important for the tumorigenesis of OEC, but IGFBP-3 silencing through IGFBP-3 promoter methylation might subsequently interrupt the communication network between p53 and the IGF axis, and hence lead to cancer progression of OEC.
Of the eleven p53 binding sites in IGFBP-3 gene that has been identified [22], Hanafusa et al. observed that at least four sites between -210 to -150 are essential for p53 induced expression of IGFBP-3 in human hepatocarcinoma cell line HepG2. They also found that hypermethylation of these sequences could selectively suppress p53 induced IGFBP-3 expression [16]. Using our OEC cell line, we identified four sites in the IGFBP-3 promoter at the -210, -206, -183, and -179 loci, as methylation hot spots. Functional analyses have been performed to study the influence of methylation at these sites on the p53 regulation of IGFBP-3. Hanafusa et al. showed diminished p53 binding and IGFBP-3 repression associated with hypermethylation of these sequences by luciferase assay after co-transfection of p53 and mutant IGFBP-3 promoter constructs, and by electrophoresis mobility shift assay [16]. In our study, we used pifithrin-α to inhibit p53 and found a subsequent decrease in IGFBP-3 expression in OVTW59-P4 and 293T cell lines, both of which contain wt p53. These functional analyses support the observation that p53 can up-regulate IGFBP-3 expression [27], and methylation at these sites suppress the p53 activation of IGFBP-3. Furthermore, our results show a quantitative association between methylation at the IGFBP-3 promoter and IGFBP-3 expression. The amount of methylation at these hot spots were positively related to IGFBP-3 suppressions in the P0 and P4 sublines. Using site-directed mutagenesis, we demonstrated a linear correlation between IGFBP-3 suppression and the amount of methylation at these loci. Our in vitro data support our clinical qMSP finding that IGFBP-3 expression was correlated to IGFBP-3 promoter methylation in OEC tumors.
Our observation that normal p53 is a required factor along with lower IGFBP-3 and higher IGFBP-3 promoter methylation for a significant survival result also suggests that wt p53 is an important molecular characteristic of OEC subtype. Recently, Kawasaki et al. reported an inverse association between IGFBP-3 promoter methylation and microsatellite instability in patients with methylator phenotype colorectal cancers, particularly under the condition of wt p53 [28]. We postulate that the biological environment of IGFBP-3 promoter methylation is strictly regulated, such that wt p53 and un-methylated IGFBP-3 promoter region are necessary factors to maintain a homeostatic condition.
Progression of tumorigenesis involves multiple steps of genetic alteration. From OEC clinical specimens and cell lines, our data indicates that aberrant methylation at p53 consensus sequence binding region of IGFBP-3 promoter could contribute to low IGFBP-3 expression and subsequently to a poor patient survival outcome. Normal p53 plays an important role in this regulation. In conclusion, our studies show clinical evidence on methylationdependent epigenetic silencing of IGFBP-3 expression regulated by p53. This regulatory pathway represents an important mechanism for loss of IGFBP-3 expression during OEC tumorigenesis and/or progression to metastasis.

Chemical reagents
General laboratory reagents were purchased from Sigma-Aldrich (St. Louis, MO, USA) and cell culture reagents from Gibco/BRL (Grand Island, NY, USA), unless otherwise specified. 5-aza-dC was obtained from Sigma-Aldrich and pifithrin-α from Calbiochem (Gibbstown, NJ, USA).

Tumor specimen and cell lines
Sixty cases of OEC specimens were collected from the archive of Department of Pathology, National Taiwan University (NTU) Hospital from May, 1994 to September 2006. These patients received surgeries and chemotherapies according to our previous report [29]. Informed consents were obtained for each patient before receiving operations. OEC cell lines OVTW59-P0 and P4 were established in our laboratory previously [5]. Cell lines A549, H1299 and 293T were obtained from Taiwan National Health Research Institute cell bank. All cell lines were maintained in DMEM solution with 5% FCS.

Immunohistochemical staining
Staining was performed in 4 μm paraffin sections as described previously [29]. In brief, sections were dewaxed, rehydrated, heated by microwave and then blocked with 1% H 2 O 2 and normal horse serum. Antibodies against IGFBP-3 (Sigma-Aldrich) and p53 (clone DO-7; DAKO, Glostrup, Denmark), as the primary antibodies, were incubated, followed by biotinylated secondary antibodies, avidin-biotin-peroxidase complex and peroxidase substrate were then added for microscopic observation. Interpretations of immunostaining were performed independently by two authors (PLT and CTL). Immunostaining of IGFBP-3 was defined as low or high when less or more, respectively, than 25% of tumor cells exhibited strong staining of cytoplasm. Immunostaining of p53 was defined as normal or overexpressed when less or more, respectively, than 25% of tumor cells exhibited strong nuclear staining.

Bisulfite modification
Tumor parts from 10 μm paraffin sections were dissected under a stereomicroscope for DNA isolation. An estimation of greater than 80% purity of tumor cells was obtained in each case. Genomic DNA was extracted using QlAamp DNA kit (Qiagene, Valencia, CA, USA) and subjected to EpiTect Bisulfite kit (Qiagene) according to the manufacture's protocol. In brief, 2 μg of genomic DNA was subjected to a denature/incubation repeat cycle in sodium bisulfite solution.

Quantitative real time methylation specific PCR (qMSP)
Bisulphite converted DNA was subjected to real time qMSP using ABI StepOne real time PCR system (Applied Biosystems, Foster City, CA) as previously described [30] with slight modification. In Brief, each reaction contained 12.5 μL of 2 × SYBR green PCR mix (Toyobo, Japan), 160 nM of each primer and 4 μl of bisulphite modified DNA in a total volume of 25 μl at 95°C for 10 min, 40 cycles of 95°C for 15 sec, 67°C for 30 sec, and 72°C for 30 sec. Primers targeting the IGFBP3 promoter region were shown in Table 4. β-actin (ACTB) was used to normalize for input DNA. A region of ACTB devoid of any CpG dinucleotide was amplified using the following primer sequences: forward, 5' TGGTGATGGAGGAGGTTTAG-TAAGT and reverse, 5' AACCAATAAAACCTACTCCTC-CCTTAA. The amount of methylated IGFBP3 and ACTB were determined by the threshold cycle number (Ct) for each sample against a standard curve generated by SssItreated DNA (Millipore, Billerica, MA)-MSP cloned fragment. The sequence of the fragment was confirmed by sequencing reaction. The performance of the standard curve was shown in additional file 1, Figure S3. The percentage of IGFBP3 methylation was calculated as the IGFBP3:ACTB ratio of a sample divided by the same ratio of SssI-treated sperm DNA (Millipore, Billerica, MA) and multiplying by 100

Methylation-specific PCR (MSP)
Primers for methylated and unmethylaled region of IGFBP-3 promoter, as previously reported [31], are listed in Table 3. The schematic diagram showing the locations of primers used are shown in Figure 1a. Twenty μL of reaction volume containing one-twentieth of the modified DNA, 4 deoxynucleoside triphosphates, PCR primers (-MSP) and HotStar TaqDNA MasterMix (Qiagen, Valencia, CA) were set for first PCR. The PCR product was then used as DNA template for a second methylated (-M) or unmethylated (-U) IGFBP-3 PCR reaction. Results were resolved by electrophoresis using 2% agarose gels. Cell line A549 treated with SssI was used as a positive methylated control.

Bisulfite-PCR and sequencing (BSP)
The -372 ~+140 bp from the first exon of IGFBP-3 promoter region was amplified from one-twentieth of the modified DNA by PCR using HotStarTaq DNA Master Mix Kit (Qiagene, Valencia, CA). The primer sequences and PCR conditions are shown in Table 4. The pGEM-T Easy Vector (Promega Corp., Madison, WI, USA) was used for TA cloning and high-efficiency competent cells DH5α (Yeastern, Taipei, Taiwan) was used for transformation. Colonies were selected and rechecked by electrophoresis. The recovered plasmids were sequenced at the Sequencing Core Facility of the NTU College of Medicine. Cytosines in CpG dinucleotides that remained unconverted after bisulfite treatment were considered as methylated.

Western blotting
Twenty-fold concentrated conditioned media, collected from cells cultured in FCS-free DMEM media for 24 hours, were subjected to 8% SDS-PAGE (Millipore, Bedford, MA, USA) and then transferred to membranes (Schleicher & Schuell, Germany) using MilliBlot-SDS semi-dry electroblotting system (Millipore). After blocking, the membrane were probed with antibodies against IGFBP-3 (Research Reagents, Texas, USA). Reactions were amplified by biotinylated second antibody, followed by streptavidin-HRP. Results were exposed using ECL system.

Construction of wild and mutant expression vectors
The promoter region of IGFBP-3 (from -253 ~+61) and the site-directed mutagenesis constructs at 4 different mutant sites (C to A at -210, -206, -183 and -179) were generated by PCR amplification with primers and PCR conditions as shown in Table 4. Wild type pGL3-IGFBP-3 and the mutant constructs were generated by subcloning of the PCR fragments to pGL3-Basic plasmid (Promega Corp.). The mutant constructs were named as Mut A (mutation at -210 and -206), Mut B (mutation at -183 and -179) and Mut A+B (mutant at all of the four sites), respectively, as shown in Figure 3c. All constructs were confirmed by DNA sequencing.

Transient transfection and luciferase assays
Cells were transiently transfected using Arrest-In transfection reagents (Open Biosystems, Huntsville, USA). Cells at a density of 1 × 10 5 /well were seeded in 6-well plate. Transfection was performed using 2 μg of plasmid DNA in serum free medium for 6 hours. Cells were then incubated in serum containing medium for another 48 hours. After washing twice with PBS, cells were collected for luciferase assays (Promega Corp.) using a dual luciferase reporter assay system according to the manufacture's instructions. The relative luciferase activity was normalized against renilla activity by co-transfection with 1 μg of pRL-TK (Promega Corp.).

Statistical analysis
The cutoff value of IGFBP-3 methylation as a threshold of hypermethylation was identified using ROC curve and AUC. The expressions of IGFBP-3 and p53, and IGFBP-3 methylation status were compared among patients with different clinical and pathologic features using t test and χ 2 test. Associations among IGFBP-3 expression, p53 overexpression and IGFBP-3 promoter methylation with progression-free survival (PFS) and overall survival (OS) rates were assessed by Cox-regression analysis. Kaplan-Meier analysis with logrank test was used to estimate survival probabilities and to compare survival distributions cate-gorized by IGFBP-3 expression, IGFBP-3 promoter methylation and p53 overexpression. Statistical analysis was carried out using Statistical Analysis System (SAS) version 8.0 (SAS Institute, Cary, NC). Probability values less than 0.05 were regarded as significant.