Diagnostic role of Wnt pathway gene promoter methylation in non small cell lung cancer

Wnt signal pathway genes are known to be involved with cancer development. Here we tested the hypothesis whether DNA methylation of genes part of the Wnt signaling pathway could help the diagnosis of non-small cell lung cancer (NSCLC). The methylation levels of SFRP1, SFRP2, WIF1 and PRKCB in 111 NSCLC patients were evaluated by quantitative methylation-specific PCR (qMSP). Promoter methylation levels of four candidate genes were significantly higher in tumor tissues compared with the adjacent tissues. SFRP1, SFRP2 and PRKCB genes were all shown to be good predictors of NSCLC risk (SFRP1: AUC = 0.711; SFRP2: AUC = 0.631; PRKCB: AUC = 0.650). The combined analysis showed that the methylation status of the four genes had a sensitivity of 70.3% and a specificity of 73.9% in the prediction of NSCLC risk for study cohort. A higher diagnostic value with an AUC of 0.945 (95% CI: 0.923–0.967, sensitivity: 90.6%, specificity: 93.0%) was found in TCGA cohort. In addition, SFRP1 and SFRP2 hypermethylation events were specific to male patients. Further TCGA data mining analysis suggested that SFRP1_cg15839448, SFRP2_cg05774801, and WIF1_cg21383810 were inversely associated with the host gene expression. Moreover, GEO database analysis showed that 5′-Aza-deoxycytidine was able to upregulate gene expression in several lung cancer cell lines. Subsequent dual-luciferase reporter assay showed a crucial regulatory function of PRKCB promoter. In summary, our study showed that a panel of Wnt signal pathway genes (SFRP1, SFRP2, WIF1 and PRKCB) had the potential as methylation biomarkers in the diagnosis of NSCLC.


INTRODUCTION
Lung cancer is the most common cancer and the leading cause of cancer death in China [1]. As a heterogeneous disease, lung cancer can be classified into small cell lung cancer and non-small cell lung cancer (NSCLC). NSCLC includes lung squamous carcinoma (LUSC), lung adenocarcinoma (LUAD) and large cell carcinoma [2]. Although the diagnosis and treatment of NSCLC have been improved in the recent years, its 5-year overall survival rate (OS) is still poor (~15%) [3].
DNA methylation is one of the most well-known epigenetic modifications. The cytosine of CG nucleotide in the DNA sequence is selectively added a methyl group [4]. Aberrant DNA methylation of CpG islands may lead to the transcriptional silencing of tumor suppressor genes Research Paper [5]. Since DNA methylation is frequently altered in lung cancer [6], increasing number of highly sensitive assays have been developed to assess gene promoter methylation in biological fluids in order to non-invasively identify early cancer in the high risk population.
Wingless-type (Wnt) signal pathway components were reported to be markedly associated with the initiation and development of NSCLC [7]. Two functional classes of extracellular Wnt antagonists have been identified [8]. The secreted frizzled related protein (SFRP) family (SFRP1-SFRP5) and Wnt inhibitory factor-1 (WIF1) proteins inhibit Wnt signaling by directly binding to Wnt molecules [8]. SFRP1 promoter methylation has been reported to contribute to the tumorigenesis and progression of many human cancers [9][10][11][12][13], including lung cancer [14]. SFRP2 and WIF1 promoter methylation was previously suggested to play important roles in Wnt signaling in lung cancer [15,16]. Protein kinase C beta (PRKCB) belongs to PKC family members, which is involved in Wnt pathway [17]. PRKCB was reported to be hypermethylated in lung adenocarcinoma stage I while hypomethylated in stage II [18], suggesting a potential of PRKCB hypermethylation assessment for NSCLC early detection.
In the current study, we evaluated the association between the promoter methylation of four Wnt signaling pathway genes (SFRP1, SFRP2, WIF1 and PRKCB) and NSCLC risk. The goal of our study was to investigate whether they could be used as a panel of methylation biomarkers in the diagnosis of NSCLC risk.

Methylation levels of SFRP1, SFRP2, PRKCB and WIF1 genes in NSCLC
As illustrated in Figure 1A and 1D, the percentage of methylated reference (PMR) levels of SFRP1, SFRP2, PRKCB and WIF1 in 111 NSCLC tumor tissues and 111 paired adjacent tissues were quantified by SYBR greenbased quantitative methylation-specific PCR (qMSP). The median (interquartile range) PMR levels of SFRP1, SFRP2, PRKCB and WIF1 in tumor tissues were 5.01% (2.32-15.82%), 15 Our results showed that methylation levels of four genes in tumor tissues were significantly higher than that in the adjacent tissues (all P < 0.05; Mann-Whitney test). Subgroup analysis by histological type showed that significantly higher methylation events of two (SFRP1 and SFRP2) and three genes (SFRP1, SFRP2 and PRKCB) were observed in LUSC and LUAD, respectively.
To assess DNA hypermethylation frequencies in our cohort and The Cancer Genome Atlas (TCGA) cohort, cut-off values for each gene were obtained from the quantile representing the upper 95% of methylation levels in nontumor samples [19,20]. Therefore, the percentages of hypermethylation for SFRP1, SFRP2, WIF1 and PRKCB were 25.2% (28/

Diagnostic values of SFRP1, SFRP2, PRKCB and WIF1 methylation in NSCLC
As shown in Figure 2A, SFRP1, SFRP2 and PRKCB methylation levels could discriminate NSCLC tissues from adjacent non-tumor samples. Using a PMR cut-off value of 3.65, SFRP1 yielded a significant area under the curve (AUC) of 0.711 (95% CI: 0.642-0.780) with a sensitivity of 62.2% and a specificity of 77.5%. Using a PMR cutoff value of 18.76, SFRP2 yielded a significant AUC of 0.631 (95% CI: 0.557-0.704) with a sensitivity of 47.7% and a specificity of 80.2%. PRKCB yielded a significant AUC of 0.650 (95% CI: 0.578-0.721) with a sensitivity of 59.5% and a specificity of 68.5% from a PMR cut-off value of 12.53.
Although WIF1 methylation lacked in good predictive capability, a combined analysis of the four markers improved test accuracy with a sensitivity of 70.3% and a specificity of 73.9% with AUC of 0.747 (95% CI: 0.683-0.811). As shown in Figure 2B and 2C, combined receiver operating characteristic (ROC) curve indicated that a good diagnostic value of four genes was found in both LUAD (an AUC of 0.798 with a sensitivity of 78.3% and a specificity of 72.5%) and LUSC (an AUC of 0.728 with a sensitivity of 69.0% and a specificity of 71.4%).
Subsequently, we have used the TCGA cohort to confirm our findings. As shown in Figure 2D, a higher diagnostic value of the epigenetic panel among NSCLC patients has been found and the combination AUC of four genes was 0.932 (95% CI: 0.914-0.950). The sensitivity and the specificity were 83.3% and 96.0%, respectively. Likewise, the combined ROC curve in LUAD indicated a good diagnostic value with a significant AUC of 0.961 (95% CI: 0.943-0.979, sensitivity: 88.4%, specificity: 96.7%, Figure 2E). The combined ROC curve of LUSC www.impactjournals.com/oncotarget also showed a similar value with an AUC of 0.945 (95% CI: 0.923-0.967, sensitivity: 90.6%, specificity: 93.0%, Figure 2F).

Correlation between DNA methylation and gene expression
We also utilized the available methylation data and mRNA data from TCGA datasets ( Figure 3). In 818 NSCLC patients, methylation levels of three genes were shown to be inversely correlated with gene expression (SFRP1_cg15839448: Spearman r = −0.367, P < 0.001, Figure 3A; SFRP2_cg05774801: Spearman r = −0.095, P = 0.007, Figure 3C; WIF1_cg21383810: Spearman r = −0.263, P < 0.001, Figure 3D). It was intriguing that positive correlations were found between two CpG sites of PRKCB (cg24250393 and cg08406370) and gene expression (Spearman r = 0.152, P < 0.001, Figure 3E; Spearman r = 0.116, P = 0.001, Figure 3F, respectively), suggesting that there might be a complex epigenetic regulation in PRKCB gene expression. No significant correlation was observed between SFRP2_cg05874561 methylation and its host gene expression (P > 0.05, Figure  3B).

Effects of candidate gene promoter regions on dual-luciferase activity
Subsequent dual-luciferase reporter assays showed a significantly higher activity of PRKCB promoter specific region (−580 bp to −180 bp) when compared with pGL3-Basic vector (fold change = 2.47, P = 0.002, Figure 5), suggesting that PRKCB promoter fragment was able to upregulate gene expression. However, no significant promoter activity could be found for the recombinant plasmids of SFRP1 (+464 bp to +863 bp), SFRP2 (+261 bp to +660 bp) and WIF1 (−511 bp to 111 bp).

DISCUSSION
Previously, most methylation studies has focused on single gene [21]. Here, we hypothesized that a panel of several genes might improve the diagnostic power for NSCLC. Our findings suggested that the DNA methylation levels of four Wnt pathway genes (SFRP1, SFPR2, PRKCB and WIF1) could jointly predict the risk of NSCLC in both our cohort and TCGA cohort.
Dysregulation of Wnt signaling pathway is often implicated in cancer initiation [22]. Aberrant epigenetic regulation of Wnt pathway genes has been identified to be a common signature in adenocarcinoma [23]. Wnt antagonist or inhibitors are known as tumor suppressors, implying that hypermethylation-induced silencing of these tumor suppressor genes may participate in the pathogenesis or progression of human malignancies [24][25][26]. Therefore, we tested whether four candidate genes from canonical and non-canonical Wnt signaling pathway could accurately detect the risk of NSCLC. REporting recommendations for tumour MARKer prognostic studies (REMARK) guidelines have recently been applied in many journals to prevent relatively usefulness biomarkers and prognostic algorithms from being introduced [27]. In the current study, the Wntgene epigenetic panel should be treated as diagnostic Bold value indicates statistical significance. PMR data are represented as median (quartile). P value is performed by Mann-Whitney test. * Partial information has not been recorded. LUSC: lung squamous carcinoma; LUAD: lung adenocarcinoma; PMR: the percentage of methylated reference biomarkers rather than prognostic biomarkers [27].
The translational value of the epigenetic panel we have found may allow the early detection of lung cancer and represent an increased probability of cure for patients. At present, the diagnosis of NSCLC is mainly based on clinical symptoms, imaging detection and histopathological examination [28]. However, most of the patients' clinical symptoms appear relatively late, which challenges the outcome of NSCLC patients. Aberrant DNA methylation has been suggested as the early event during lung carcinogenesis [29]. Therefore, it shows a higher efficiency in the early detection. Epigenetic biomarkers may be a better tool for early diagnosis due to their preponderance of non-invasion, high sensitivity and high specificity [30]. Downregulation of Wnt inhibitors (e.g., by hypermethylation) is common in NSCLC tumor cell lines and resected samples [31]. Since epigenetic changes are dynamic and reversible, several inhibitors of enzymes controlling epigenetic modifications are promising targets for the development of more effective therapeutic strategies against cancer [32]. The two main DNA methyltransferase inhibitors that have been largely tested in the clinic are 5-azacitidine and decitabine [33].
There are several distinct types of cancer biomarkers based on different areas: proteomics, metabolomics, genetics and epigenetics [34]. Currently, conventional plasma proteins in clinical detection of lung cancer usually lack enough sensitivity in terms of their usages as biomarkers. For example, cytokeratin 19 fragment (CYFRA21-1) had a relatively low AUC of 0.624 (sensitivity: 0.576; specificity: 0.797) to discriminate NSCLC patients and control subjects [35], which suggested a lower diagnostic ability compared with our study. Moreover, previous study had confirmed that plasma CYFRA21-1 appeared more sensitive for NSCLC diagnosis than other tumor biomarkers such as carcinoembryonic antigen (CEA) and neuron-specific enolase (NSE) and progastrin-releasing peptide (ProGRP) [36]. DNA methylation emerged as a promising biomarker for early detection, which was independent of the genomic composition of the primary tumor [37]. However, intratumoral heterogeneity has profound clinical implications, which challenges the current methods of tumor diagnosis and targeted therapy [38]. Previous studies have reported the intratumor DNA methylation heterogeneity in human cancers [39][40][41]. Liquid biopsy was shown as an ideal means of sampling intratumor genetic and epigenetic heterogeneity for diagnostics [42]. Since DNA methylation as a biomarker was more effective in serum than that in tissues [43], future study is needed to check whether the panel of Wnt-gene epigenetic biomarkers has a higher diagnostic value in liquid biopsy. In order to assess the diagnostic capacity of the expression of four candidate genes, we have used the gene expression profiles (IlluminaHiSeq_RNA-SeqV2) of 817 NSCLC tumor samples and 29 non-tumor samples from TCGA database. As shown in Supplementary Figure 1, the results revealed that the AUC values of low expression of SFRP1, SFRP2, WIF1 and PRKCB were 0.705, 0.319, 0.959 and 0.857, respectively. SFRP2 expression did not draw a consistent conclusion like other genes, which implying a low efficiency of early detection. Although mRNA expression could be one of the distinct types of cancer biomarkers, it is advisable to establish a methylation-based diagnostic system for cancers since DNA methylation frequently regulates gene expression and occurs in the early stage of tumorigenesis [44]. Previous studies has reported the diagnostic values of SFRP1 methylation in cutaneous squamous cell carcinoma [45] and esophageal squamous cell carcinoma recurrence [46]. SFRP2 hypermethylation has also been shown in several cancers including acute myeloblastic leukemia [47], mesothelioma [48], bladder cancer [49], liver cancer [50], as well as lung cancer [51]. SFRP2 hypermethylation was previously detected in women and nonsmoking NSCLC patients [52], indicating a gender-specific effect of DNA methylation on gene expression. Interestingly, our results revealed a more frequent SFRP1 and SFRP2 methylation in male NSCLC patients. This divergence may be due to the different patient populations and multiple environmental factors exposure. Since NSCLC is well-known to be composed of heterogeneous groups, sexual hormones have been considered to participate in some signaling pathway in NSCLC [53]. In human ovarian cancer, the ovarian hormones-induced Wnt pathway activation could increase the growth of ovarian cancer precursor lesions [54]. Meanwhile, cell proliferation could be accelerated by androgen and inhibited by estrogen [55]. Estrogen can influence neoplastic diseases by changing the levels of gene expression and DNA methylation [56,57]. Therefore, we speculated that the interaction of sexual hormones and Wnt signal pathway might be involved in the elevated DNA methylation in NSCLC patients.
WIF1 is a negative regulator of the Wnt pathway that may have important implications for tumorigenesis [58]. Clinical studies of WIF1 methylation or downregulation were particularly common in lung cancer [59,60]; however, it had a weak diagnostic role in the current study. Methylation changes in carcinogenesis are often heterogeneous, and no single gene has been found to be methylated in every NSCLC specimen [59]. Therefore, further larger number of patients is needed to be investigated.
PRKCB was reported to enhance the expression of cyclin D1 in human breast cancer cells, leading to cell proliferation and cell cycle progression [61]. Its overexpression has been detected in chronic lymphocytic leukemia [62]. These findings suggested that PRKCB was an important target for anticancer therapy. However, in the current study, higher methylation level of PRKCB promoter was observed in tumor tissues than adjacent normal tissues. Our findings were consistent with the pan-cancer study of Li et al. [23]. They identified PRKCB promoter hypermethylation in LUAD, colon cancer and rectal cancer, showing PRKCB as an epigeneticallysilenced gene in Wnt pathway. Besides, we observed a higher promoter activity of PRKCB, suggesting an important regulation of the transcriptional function. It is noteworthy to explore the exact mechanism of PRKCB promoter methylation on gene expression in NSCLC.
With analyzing TCGA data, we have found significant correlations for SFRP1_cg15839448, SFRP2_ cg05774801, WIF1_cg21383810 methylation and their host gene expression, although the correlation coefficients were weak. Notably, results from GEO dataset gave a support that gene expression could be restored in some lung cancer cells after demethylation, to some extent, suggesting a potential effect of DNA methylation on gene silencing. Further dual luciferase assays showed that only PRKCB promoter specific region had a significantly higher activity when compared with pGL3-Basic vector. All the above observations suggested the epigenetic regulation of four candidate genes were complex. There are two familiar mechanisms of DNA methylation on transcriptional repression. First, DNA methylation directly interferes with the combination of transcription factors and cis-element. Second, methyl-CpG-binding protein alters the chromatin structure by recruiting the co-repressor complex [63]. The purpose of the dual luciferase assays we have performed was to show that the promoter regions chosen might be functional. However, it is an intricate network of epigenetic regulation on gene expression. There may be other alternative mechanisms accounting for gene inactivation such as histone modifications (methylation, acetylation, and ubiquitination), chromatin remodeling, and non-coding ribonucleic acids (RNAs). Global profile changes of histone modifications are critical in the initiation and progression of human cancers [64,65]. Cancer cells suffer a global reduction of activation markers H3K4me3 [66] and H4K16ac [67] and a gain in the repressive markers H3K27me3 [68], H3K9me3 [69] and H4K20me3 [67], which is closely related with transcriptional regulation. In esophageal squamous cell carcinoma, DNA methylation and histone acetylation has been found to synergize silencing SFRP1 gene expression [70]. Likewise, an aberrant gain of the repressive mark H3K27me3 could decrease the expression of SFRP1 gene in addition to DNA hypermethylation in prostate cancer cell line [71]. As for renal cell carcinoma, low levels of AcH3, AcH4 and H3K4 and a high level of H3K9, known as repressive histone modifications, were found in the SFRP2 negative cell lines [72]. Therefore, further investigations are required to elucidate the epigenetic control of Wnt signaling pathway in NSCLC.
There were some limitations of our study. First, FFPE samples are commonly used to find methylation and expression, although frequently genetic material in those samples is highly degraded [73]. FFPE samples were likely to yield false positives during formalin fixation process [74]. Future work is needed to validate the diagnostic values of the four candidate genes in fresh frozen biopsies or peripheral blood. Second, next generation sequencing (NGS) technique is superior to yield high-resolution DNA methylation information, such as the pattern information obtainable with bisulfate genomic sequencing or the accurate methylation percentage determination at single CpG site [75]. Although we have obtained a consistent result of the diagnostic value both in NGS-based TCGA data and our study, further validation in other NGS techniques is needed. The qMSP technique is high-throughput and sensitive, and it was often used in the clinical detection [76]. In the present study, due to the limitation of designing multiple suitable primer sets to assess the promoter region, we only evaluated a single CpG site to represent the methylation level of gene promoter region. Further investigations on other CpG sites among larger number of samples are needed to be carried out to clarify the epigenetic regulatory mechanism and its application to diagnostic biomarkers. Third, the sample size of our study was moderate, larger case-control studies from LUAD and LUSC patients need to be performed. Fourth, since the paired specimens have been obtained from the same patients, the data would not bear thorough analysis, i.e., stepwise multivariate analysis.
In summary, our study identified that a methylation panel of Wnt signal pathway genes (SFRP1, SFRP2, WIF1 and PRKCB) might be used as diagnostic biomarkers of NSCLC risk.

Sample collection
Following the approval by the local Institutional Ethics Committee, a group of 111 primary NSCLC patients were recruited consecutively from Huzhou First People's Hospital, China from August 2010 to October 2013. None of the patients received radiotherapy or chemotherapy before operation. These patients were diagnosed with LUAD or LUSC based on histopathological evaluation. There were 42 LUSC patients and 69 LUAD patients treated with cisplatinum-gemcitabine and cisplatinumpaclitaxel, respectively. Therapeutic effect was evaluated according to the RECIST 1.1 criteria [77]. Clinical features studied included age, gender, smoking history, clinical stage, tumor location and differentiation. The mean age of NSCLC patients (73 males and 38 females) was 63.59 ± 10.19 years. Staging was based on the TNM (tumor, node, metastasis) classification system of the International Union Against Cancer 2009 (UICC). Disease status after surgery for this cohort was updated every two months by telephone follow-up. OS refers to the data from the date of primary surgery to the date of death or the date of last follow-up. The study end point should be the death caused by NSCLC. The study set included 111-paired formalin-fixed, paraffin-embedded (FFPE) lung tumor tissues (4 µm in thickness) and non-tumor tissues (4 µm in thickness) from Department of Pathology archives. All patients gave informed consent for collection and analysis of their tissue specimens for research purposes.

Genomic DNA extraction and bisulfite modification
The genomic DNA was isolated using the QIAamp DNA FFPE Tissue Kit following the manufacturer's instructions (Qiagen, Hilden, Germany). DNA was quantitated by the NanoDrop 2000 (Thermo Scientific, Wilmington, USA). Sodium bisulfite modification was performed using EZ DNA Methylation Gold ™ Kit (Zymo Research, Orange, CA). According to conversion principle, target sequence is modified converting unmethylated, but not methylated, cytosines to uracils, and uracils are subsequently converted into thymine following PCR reaction.

Methylation assay by qMSP method
For qMSP, primers specific for the CpG island sequences of the target genes (SFRP1, SFRP2, WIF1 and PRKCB) were designed to amplify bisulfite-modified DNA. ACTB was used as an independent reference gene. The primer sequences were shown in Table 3. All experiments were performed on LightCycler 480 system (Roche Diagnostics Ltd, Lewes, UK) utilizing a 384-well plate platform. The amplification reaction was carried out in a final volume of 10 μL containing 5 μL LightCycler 480 SYBR Green I Master (Roche Diagnostics Ltd, Lewes, UK), 0.5 μL DNA template, 0.25 μL of each primer (10 μM) and 4.0 μL H 2 O. The PCR program was as follows: holding at 95°C 10 min for enzyme activation, 45 cycles of denaturation at 95°C for 20 s, annealing at 56°C or 58°C for 20 s, and extension at 72°C for 30 s. After amplification, melting curve analysis was performed for PCR product identification that consisted of one cycle of 95°C for 15 s, 60°C for 60 s, and 95°C for continuous acquisition mode. SssI methyltransferase (Thermo Scientific, Wilmington, USA) and bisulfitetreated leukocyte DNA from healthy person served as a positive methylation control. Water without DNA served Figure 5: Luciferase activity determined by a dual-luciferase assay system. The reporter gene plasmids containing different gene promoter regions were constructed, including pGL3-SFRP1: +464 to +863, pGL3-SFRP2: +261 to +660, pGL3-WIF1: −511 to −111, and pGL3-PRKCB: −580 to −180. HEK-293T cells were transfected with each reporter gene plasmid, together with renilla luciferase reporter plasmid. Empty pGL3-Basic and pGL3-Promoter vectors were used as the negative and positive control, respectively. **P < 0.01; ****P < 0.0001. as a control for contamination was included in each assay. Each reaction was performed in duplicate. PCR products were subjected to the Qsep100 DNA fragment analyzer (Bioptic, Taiwan, China) and visualized in Gel-view format. Subsequently, some qMSP products' sequences were verified by Applied Biosystems ® 3730 DNA Analyzer (Applied Biosystems, Warrington, UK) ( Supplementary  Figures 2-5).

Statistical analysis
PMR at a specific region was calculated by dividing the gene : ACTB ratio of a sample by the gene : ACTB ratio of fully methylated DNA and multiplying by 100. The PMR cut-off point (i.e., the value with the maximum Youden index, defined as sensitivity plus specificity minus 1) was calculated using the ROC curve. Data were presented as median (interquartile range). The Mann-Whitney test was used to study the differences in the nonparameter variables. ROC curves were constructed and the AUCs were calculated to determine the diagnostic role of four genes methylation for NSCLC. Multivariate analysis was performed for OS after excluding the insignificant variables on univariate analysis (P > 0.01). Cox proportional hazard models were fitted with calculating hazard ratios (HR) and the corresponding 95% confidence intervals (95% CI). Nonparametric Spearman's criterion was used to calculate the coefficient of correlation between the levels of mRNA expression and promoter methylation from TCGA datasets. The results of luciferase reporter assay represented the means ± standard deviation of triplicate determinations. Statistical analysis was performed using SPSS version 18.0, with all tests being two-tailed and P < 0.05 was considered to be statistically significant.