Prognostic value of aberrantly expressed methylation gene profiles in lung squamous cell carcinoma: A study based on The Cancer Genome Atlas

1 INTRODUCTION

Lung cancer ranks first among all cancers in terms of incidence, and it is also the most important cause of cancer deaths all over the world (Torre et al., 2015). According to histological differences, lung cancer can be divided into non-small-cell lung cancer (NSCLC) and small-cell lung cancer (SCLC), of which NSCLC accounts for approximately 85%, and 30% of NSCLC cases can be classified as lung squamous cell carcinoma (LUSC; Piperdi, Merla, & Perez-Soler, 2014). The incidence of LUSC is high, and the prognosis is poor (5-year survival rate < 15%); LUSC causes approximately 400,000 deaths per year worldwide. Recent studies have shown that LUSC is closely related to smoking and is more common in men than in women (Kenfield, Wei, Stampfer, Rosner, & Colditz, 2008). Although the occurrence of LUSC can be prevented to a certain extent by regulating smoking (Meza, Meernik, Jeon, & Cote, 2015), the mechanism of occurrence and development of LUSC is still not fully characterized. In recent years, there has been great progress in molecular targeted therapies for LUSC (Drilon, Rekhtman, Ladanyi, & Paik, 2012; Lee et al., 2013; Sun et al., 2016). However, due to targeted drug resistance and the toxicity of treatment, some patients do not benefit from it. Therefore, to decrease mortality and improve the management of LUSC, detection and risk stratification of LUSC is urgently needed to identify new early diagnostic biomarkers and therapeutic targets. In the current study, we aimed to find effective potential molecular biomarkers for predicting survival in LUSC.

NSCLC poses a great threat to human health. Epigenetic abnormalities are one of the most important mechanisms in the development of lung cancer (Yang & Schwartz, 2011). The abnormal methylation of DNA is the most common and important modification in epigenetics and is an important means of regulating the function of the genome. Selective methylation and demethylation of genes to form specific tissue types and regulate the expression of genes during development and differentiation are considered to be major factors leading to the development of multiple tumors (Kulis & Esteller, 2010). The study of the relationship between methylation and lung cancer has become an area that has drawn more attention. Studies have suggested that the altered DNA methylation patterns in tumor tissues may silence tumor-suppressor genes and activate oncogenes through hyper/hypomethylation. For example, the hypermethylation of the promoter of the suppressor gene SOX17 was found to silence the expression of this gene in 60.2% of primary lung cancer samples, thereby eliminating the inhibition of cell proliferation in lung cancer (Yin et al., 2012). Identification of gene methylation abnormalities can explain the instability and redundancy of the lung cancer genome and provide the basis for targeted therapy or risk prediction.

The Cancer Genome Atlas (TCGA; Tomczak, Czerwińska, & Wiznerowicz, 2015) has created a genomic data analysis process that can effectively collect, select, and analyze human tissue for genome changes. The TCGA database provides analysis of high-throughput data for various genomic changes, including DNA methylation. Bioinformatics analysis along with patient diagnostics, treatment information, and tumor pathology information can be combined to advance oncology research and lay the foundation for improved cancer prevention, early detection, and treatment. In this study, LUSC-related abnormally methylated and expressed genes were identified based on TCGA data, and the expression of abnormally methylated genes and related differential genes in LUSC patients was clearly defined. We conducted a series of bioinformatics and survival analyses, screened for biomarkers associated with methylation, and constructed a prognostic risk model to help us to predict the prognosis of LUSC patients. These results could provide new insight into the molecular mechanisms based on methylation in LUSC.

2 MATERIALS AND METHODS

3 RESULTS

3.1 TCGA data analysis and construction of the PPI network

A total of 370 samples of methylated LUSC and 42 control samples were included in our analysis. Based on the relevant LIMMA software package (http://bioconductor.riken.jp/packages/3.0/bioc/html/limma.html), the abnormal methylation expression data from TCGA were extracted and analyzed. To narrow down the data range, we used |log FC| > 1, p < 0.05 as the screening conditions to obtain 211 abnormally methylated genes (Figure 1). To further investigate the potential links between these abnormally methylated genes, the STRING online database was used to mine and describe the PPI network, and a confidence score >0.4 was used as the cut-off criterion. Then, we used the Cytoscape software to visualize the generated PPI network; a total of 103 abnormally methylated genes were filtered into the complex PPI network, which contained 103 nodes and 225 edges (Figure 2), 118 of the 211 genes did not fall into the PPI network. Those genes were used for further functional and survival analysis to explore functional abnormalities caused by the aberrant methylation of genes in normal and diseased patients.

3.2 Functional enrichment analysis of abnormally methylated genes

To further understand the functional role of abnormally methylated genes in LUSC, we used the BinGo plug-in for functional enrichment analysis of these genes. The results showed that the abnormally methylated genes were particularly rich in molecular function (MF), biological processes (BP), and cell components classification. In the BP group, these genes were mainly enriched in regulation of DNA-dependent transcription, the RNA metabolic process, gene expression, etc. (Figure 3a). The MF is mainly enriched in multiple binding, such as chromatin DNA binding, DNA regulatory region binding, transcription regulator activity, and transcription factor binding (Figure 3b). Also, cellular component (CC) terms are mainly involved in the cytoplasm and nucleoplasm (Figure 3c).

3.3 Prognostic assessment of abnormally methylated genes and clinical features

We conducted a univariate Cox regression between abnormally methylated genes in LUSC patients, and the results showed that a total of 39 genes with p < 0.05 were greatly associated with OS (Table 1). And then, multivariate Cox regression was applied to construct a prognostic risk model. We found that four genes (ventral anterior homeobox 1 [VAX1], cholesterol 25-hydroxylase [CH25H], adenylate cyclase activating polypeptide 1 [AdCyAP1], and iroquois homeobox 1 [Irx1]) were independent prognostic indicators for LUSC (Figure 4), and the risk model constructed by them could be used for prognosis assessment. The risk score was input as follows: (−1.973 ×  expression level of ADCYAP1) + (−1.124 × expression level of IRX1) + (1.384  ×  expression level of VAX1) + (−1.589  × expression level of CH25H). Also, based on the cut-off of the median risk score, a total of 405 patients were classified into a high-risk group (n = 202) and a low-risk group (n = 203) according to the median risk score. The results of survival analysis showed that patients in the high-risk group have a significantly worse OS than that of patients in the low-risk group (p = 3e−05, Figure 5a). Then, we used time-dependent ROC curves to evaluate the prognostic ability of risk model. The area under the curve (AUC) for the four-biomarkers prognostic model was 0.644 at 5 years of OS (Figure 5b).

Table 1. The revealed aberrantly methylated genes that were significantly associated with OS after univariate Cox regression analysis

Gene	HR	Z	p
GHSR	0.226142	−2.83672	0.004558
ADCYAP1	0.206839	−2.75709	0.005832
TRIM58	0.271923	−2.70983	0.006732
ZNF454	0.237009	−2.70699	0.00679
CH25H	0.262343	−2.66953	0.007596
COX11P1	0.3957	−2.64223	0.008236
ZNF730	0.206126	−2.57848	0.009924
PCDH8	0.230702	−2.55218	0.010705
IRX1	0.212528	−2.50359	0.012294
ZNF568	0.182196	−2.45557	0.014066
NKX2–3	0.311258	−2.43929	0.014716
EVX2	0.342645	−2.37066	0.017756
NKX2–6	0.338588	−2.3206	0.020308
SOX14	0.207648	−2.32048	0.020315
ZNF729	0.282189	−2.31958	0.020364
EMX2	0.250729	−2.29638	0.021654
CBLN1	0.268631	−2.25892	0.023888
ZNF790-AS1	0.27809	−2.22204	0.026281
C5orf38	0.289961	−2.19936	0.027852
FOXB2	0.336379	−2.17058	0.029963
ZNF69	0.219363	−2.13281	0.03294
SOX11	0.41728	−2.12322	0.033735
POU3F3	0.324792	−2.11451	0.034472
DBX1	0.240704	−2.10174	0.035576
ZNF492	0.273512	−2.09055	0.036568
ALX1	0.40646	−2.07893	0.037624
SOX1	0.33162	−2.06098	0.039305
ZNF728	0.320848	−2.0513	0.040237
ZNF257	0.308881	−2.05121	0.040246
ABCB10P4	0.413226	−2.0358	0.04177
ACTA1	0.244979	−2.02704	0.042658
NKX6–2	0.2789	−2.016	0.0438
PDX1	0.268326	−2.00947	0.044488
VAX1	0.373035	−2.00838	0.044603
GRAMD1A	0.168909	−2.00591	0.044866
RAX	0.227745	−1.99582	0.045954
TBX20	0.371405	−1.99235	0.046332
SOX17	0.432915	−1.97682	0.048062
ZNF808	0.187345	−1.96339	0.049601

• Note. HR: hazard ratio; OS: overall survival.

3.4 Correlation analysis of differential methylation sites and gene expression

Using p < 0.05 as a meaningful screening criterion for combined survival, the methylation and gene expression levels of the prognosis-related gene IRX1 were significantly correlated with prognosis (Figure 6). Also, the prognosis genes methylation sites based on relevant data mining in TCGA are searched, and correlations between site and gene expression were analyzed, using |Cor| > 0.5 as a screening condition (Table 2). The gene expression of IRX1 and VAX1 was found to be related to the methylation level of multiple sites, and the gene IRX1 showed a negative correlation, while VAX1 showed a positive correlation (Figure 7).

Table 2. Correlation between methylation sites and expression of genes VAX1 and IRX1

	Methylation site	Correlation	p
IRX1	cg10530883	−0.502	1.785e−25
	cg09232937	−0.502	1.839e−25
VAX1	cg10143067	0.556	6.501e−32
	cg18459489	0.57	7.532e−34
	cg15668468	0.54	5.872e−30
	cg03851159	0.569	8.79e−34
	cg18709545	0.617	6.186e−41

• Note. IRX1: iroquois homeobox 1; VAX1: ventral anterior homeobox 1.

4 DISCUSSION

Lung cancer, which is mainly associated with smoking, is one of the most common malignancies in humans. Other factors, such as the relationship between viral infection and lung cancer, are currently hot topics. NSCLC is a huge threat to human health, and LUSC accounts for 30% of NSCLC. The study of methylation and its relation to LUSC is a field of growing interest. In recent years, with the clinical application of epidermal growth factor receptor tyrosine kinase inhibitor (EGFR-TKI), the therapeutic effect of lung adenocarcinoma has been significantly improved. At the same time, LUSC is being studied as an independent type of lung cancer to obtain new therapeutic targets. These studies have focused on changes in the molecular biology of LUSC and have made many advances. The rapid development of gene analysis technology provides the possibility for in-depth exploration of the molecular characteristics of LUSC and provides valuable evidence for prognostic evaluation and molecular targeted therapy.

Epigenetics has led to the discovery that not only cytogenetic changes but also epigenetic abnormalities, including DNA methylation, are involved in the pathogenesis of LUSC. Accumulating evidence has shown that DNA methylation, a major molecular mechanism of epigenetic changes, is correlated with human malignant tumors, including lung cancer (Chakravarthi, Nepal, & Varambally, 2016; Huang et al., 2015). Genes with aberrant DNA methylation can be noninvasive biomarkers for the detection and diagnosis of cancer (Chen et al., 2014; Gloss & Samimi, 2014; Luttmer et al., 2016). Therefore, to explore the molecular mechanism of LUSC progression and the epigenetic changes that determine new biomarkers, early diagnosis, treatment, and prognosis of LUSC is crucial. The independence and stability of abnormal DNA methylation analysis make it an effective method for identifying prognostic biomarkers (Dinardo et al., 2017; Marcucci et al., 2014; Ni, Ye, & Huang, 2017). Recent research results showed that the methylation of the P16, RASSF1A, APC, or SHOX2 genes were significantly associated with lung cancer and could serve as potential noninvasive biomarkers in lung cancer (Ni et al., 2017). Other studies have shown that TRIM58/cg26157385 methylation is associated with eight prognostic genes, including A2ML1 and CCNE1, in LUSC (W. Zhang et al., 2018). Also, some research also identified and validated the methylation of the HOXA9 promoter as a prognostic biomarker associated with high risk in Stage I lung adenocarcinoma patients (Robles et al., 2015). Thus, bioinformatics analysis of the functional enrichment and prognostic evaluation of abnormally methylated genes can provide clinicians with new tools that can be used to treat patients and predict prognoses.

In this study, we aimed to investigate abnormally methylated genes between LUSC patients and normal samples and to identify prognostic biomarkers related to methylation-mediated expression. We initially identified 211 abnormally methylated genes and 1,854 differentially expressed genes using the LIMMA software package. To gain insights into the functional roles of these LUSC aberrantly methylated genes, functional enrichment of these genes was performed. These genes were found to be involved in intracellular and nucleoplasm pathways and were mainly enriched related to the regulation of transcription, DNA-dependent processes, RNA metabolic processes and a variety of other functions, such as DNA regulatory region binding and transcription factor binding. These functional enrichment items not only clearly show the changes in gene function that may result from abnormal DNA methylation in different samples but also show the interaction of genes at the functional level.

In addition, through univariate and multivariate Cox analyses of the abnormally methylated genes, it was found that the prognostic risk model (high risk and low risk) which was constructed by the four abnormally methylated genes (VAX1, CH25H, ADCYAP1, and IRX1) could predict the survival rate, and each of these four genes was an independent prognostic factor for LUSC. The predictive power of the risk model was evaluated by a time-dependent ROC curve, and the result of AUC was more than 0.6. Despite its limitations, it could be concluded that the prognostic model we constructed had a certain accuracy and sensitivity in assessing the prognosis of patients. As a transcription factor, methylated VAX1 is thought to be highly correlated with bladder cancer (BC) recurrence and may serve as a useful biomarker for predicting BC recurrence (Zhao et al., 2012). Also, VAX1 were methylated in more than 80% of the adenocarcinomas, which is including lung adenocarcinoma (Rauch et al., 2012). However, there are few studies on the methylation of VAX1 in LUSC. In this study, hypermethylation of VAX1 showed a high survival rate and could be used as one of the prognostic indicators of LUSC. And the expression of VAX1 gene was positively correlated with the methylation levels of multiple sites. This may be because the hypermethylation of VAX1 enhances the expression of the gene and improves the patient’s quality of life. Of course, this requires further research to confirm our result. CH25H is a gene that codes for a cholesterol 25-hydroxylase involved in cholesterol and lipid metabolism (Bauman et al., 2009). Studies have shown that downregulation of CH25H can reduce plasma cell responses after immune challenge (Hannedouche et al., 2011). Also, low expression of CH25H is closely associated with a poor prognosis in breast cancer (Mittempergher et al., 2013) and may be a potential target for distant metastasis control in breast cancer. We have found that the downregulation of CH25H expression in LUSC patients may shorten the survival time by reducing the patient’s immune response, which requires further verification through experiments. CH25H, which has been less reported in lung cancer, may serve as a new marker for further study of LUSC. The ADCYAP1 gene is involved in various biological processes, including cell growth, proliferation, and differentiation (Wolman, Heppner, & Wolman, 1997). Research showed that ADCYAP1 hypermethylation is closely related to cervical cancer and could be used as an effective and sensitive methylation biomarker for the early diagnosis of cervical cancer (Jung et al., 2011). Also, gene expression of ADCYAP1 is associated with multiple cancers, such as neuroblastoma (Isobe et al., 2004) and breast cancer (García-Fernández et al., 2004). Pituitary adenylate cyclase-activating polypeptide, a protein encoded by gene ADCYAP1, can be used as a potential serum marker for non-small-cell lung cancer (H. Zhang, Chen, & Huang, 2012). In this study, compared with hypomethylation, hypermethylation showed a high survival rate, and methylation abnormalities affected the expression of the ADCYAP1 gene and affected LUSC-related biological processes, such as positive regulation of cell proliferation and cell–cell signalling.

IRX1, a member of the Iroquois homeobox family of transcription factors, was identified as a potential tumor-suppressor gene in head and neck squamous cell carcinoma (Bennett et al., 2008). In this study, IRX1 expression was downregulated and showed hypermethylation. Correlation analysis showed that the expression level was negatively correlated with the methylation level of the site cg09232937 and cg10530883. This may be due to the hypermethylation of the sites leading to IRX1 transcriptional silencing, which results in dysregulation of the expression, which in turn affects the progression of the disease and the prognosis of the patient. Overexpression of Irx genes may cause lung dysplasia by inducing lung dysmorphogenesis with thickened mesenchyme (Doi, Lukosiute, Ruttenstock, Dingemann, & Puri, 2011), in which case it can further lead to lung cancer. Moreover, the hypomethylation of IRX1 is associated with a high risk of lung metastasis (Lu et al., 2015). In our study, we found that the survival rate of patients with hypermethylation of the four genes was higher than that of patients with hypomethylation of those genes, and p < 0.05 was statistically significant. A combined analysis of gene expression and methylation levels revealed that the survival rate of patients with hypermethylation/low gene expression of IRX1 was significantly higher than that of patients with hypomethylation/high gene expression of IRX1. Hypermethylation often leads to transcriptional silencing and chromosome instability, causing gene dysregulation and cell differentiation disorder. Also, the expression of IRX1 was negatively correlated with the hypermethylation level of the site cg09232937 and cg10530883, which suggested that DNA hypermethylation may be an important reason for the downregulation of IRX1 expression in LUSC patients. These results provide a basis for bioinformatics and a related theoretical basis for further experimental validation.

5 CONCLUSION

Identification of differentially expressed genes is widely used in molecular biology studies to obtain markers of cancer diagnosis, treatment, and prognosis. In conclusion, in this study, we performed whole-genome methylation analysis of LUSC patients based on the TCGA database and found some abnormally methylated genes related to the development of LUSC. Univariate and multivariate Cox regression analysis showed that the prognostic risk model constructed by four abnormally methylated genes (VAX1, CH25H, ADCYAP1, and IRX1) proved to be an independent prognostic factor for LUSC. Also, the expression and methylation levels of the key gene IRX1 are significantly correlated with prognosis and are negatively correlated with the methylation of the site cg09232937 and cg10530883. Although experimental verification is required, our study can provide a basis for future diagnosis, treatment, and prognosis of LUSC. Also, our research has important theoretical significance in guiding follow-up studies of LUSC.

CONFLICTS OF INTEREST

The authors declare that there are no conflicts of interest.