The Gene Expression Omnibus (GEO) includes high-throughput Gene Expression data submitted by researchers around the world, including memory chips, next-generation sequencing, and other single-cell sequencing data [13, 14]. Wu et al. downloaded five DNA methylation and gene expression profiling datasets (GSE60185, GSE42568, GSE21653, GSE58812 and GSE52865) from GEO database and identified lots of differentially expressed genes and abnormally methylated genes between breast cancer samples and normal samples. The relationship between DMLs and the prognosis of breast cancer was determined, suggesting that these genes could be used as prognostic and diagnostic markers of breast cancer [15]. DNA methylation represents a new and very promising area of research. Many key tumor suppressor genes, which are silenced by DNA methylation, have been evaluated for their prognostic and predictive significance in many types of cancer [16]. At present, there is only one dataset on malignant transformation of sinonasal papilloma, namely, GSE125399. Illumina Human Methylation 450 BeadChip Methylation analysis was used to study DNA Methylation of CpG islands and promoters in 6 SNIP samples and 5 SNIP-SCC samples. For the first time, differences in DNA methylation between SNIP and SNIP-SCC were compared. Through bioinformatics analysis on the methylation dataset GSE125399, we found that MAML2, UCKL1, GSTT1, and HLA-G methylation significantly upregulated and NRGN methylation significantly downregulated may be the key genes of sinonasal papilloma malignant transformation.

In order to reduce the systematic errors and improve the repeatability of the study, we need to optimize the missing replacement of the downloaded methylation data, reduce the batch error effect, and eliminate the outlier samples. The results showed that the processed samples had better stability, SNIP and SNIP-SCC datasets showed two independent populations, and the gene expression showed differences. The normalized data is conducive to further data analysis.

As one of the biomarkers related to cancer, the identification and application of DNA differential methylation patterns is one of the research hotspots. In 75 colorectal cancer samples, 4062 methylation loci were detected, of which 26 promoter hypermethylation genes and 143 promoter CpG hypomethylation genes were found, suggesting that these DNA methylation biomarkers may be important diagnostic markers and therapeutic targets for colorectal cancer [17]. The methylation status of 485,577 CpG loci in 24 gallbladder carcinoma tissues (tumor, adjacent non-tumor tissue and 8 gallstones each) was studied, of which 24,188 (72%) were hypermethylated and 9255 (28%) were hypomethylated. A comparative analysis of differential proteomic data revealed 7 hypermethylated or down-regulated genes (e.g., FBN1, LPP, SOD3) and 61 hypomethylated or up-regulated genes (e.g., HBE1, SNRPF, TPD52), which contributed to the early diagnosis of gallbladder cancer [18]. Compared with adenomas, Galamb et al. found that the methylation levels of AXIN2, DKK1, VANGL1, and WNT5A gene promoters in colorectal cancer were higher, while the methylation levels of SOX17, PRICKLE1, DAAM2 and MYC were lower, and subsequently confirmed the negative correlation between expression and methylation [19]. Among the differentially methylated loci in the malignant transformation dataset of sinonasal papilloma, we obtained 31 genes containing up-methylated loci and 3 genes containing down-methylated loci. The 31 genes containing upregulated methylation loci included MAML2, GLUD2, GRASP, UCKL1, CLIP3, GSTT1, HLA-G, etc. The three genes containing downregulated loci included two non-coding RNAs (LOC100287834 and LOC391322) and NRGN. Studies of differentially expressed genes with methylation are most often used to uncover candidate genes associated with tumors. By comparing gene methylation levels in cancer tissues and adjacent normal tissues, active genes, methylation loci, and possible suppressor genes leading to tumorigenesis and development can be found out, thereby laying a foundation for the study of specific markers and molecular targets of tumors. We analyzed GO enrichment and KEGG pathway enrichment of DMLs and found that DMLs were mainly enriched in single-stranded DNA binding, chromatin binding, double-stranded DNA binding, DNA-dependent ATPase activity, drug metabolism and human papillomavirus infection pathway. Evaluation of GO functional enrichment and KEGG pathway analysis of DMLs can help us better understand the SNIP-SCC genetic regulatory network and the regulatory mechanisms involved in biological processes. KEGG pathway visualization showed that UCKL1 and GSTT1 were upregulated in the drug metabolism pathway, and HLA-G and MAML2 were significantly upregulated in the human papillomavirus infection pathway. It is speculated that the significant upregulation of UCKL1, GSTT1, HLA-G, and MAML2 methylation and the significant downregulation of NRGN methylation may be the key genes of malignant transformation of sinonasal papilloma.

Amplification targeted sequencing was used to analyze 409 gene mutations in 6 papilloma/cancer tissue samples from 4 SNIP-derived SCC patients. The results showed that CDKN2A, KMT2D, NF1, PDE4DIP, CYP2D6, FLT4, and MYH9 were mutated in several cases [20]. The results of SCCA1, SCCA2 and SCC antigen analysis in serum and tissue samples from 18 SNIP patients and 23 sinus SCC patients showed that the serum SCCA1 concentration in SNIP patients was significantly higher than that in SCC patients, and the serum SCCA2 level in SCC group was significantly higher than that in SNIP group [21]. Compared with SNIP tissue, downregulation of DLEC1 cilia and flagellate-associated protein (DLEC1) in SCC tissue is associated with promoter hypermethylation [22]. We collected 115 cases of sinonasal papilloma and malignant transformation, determined the expression of aberrant methylated genes by immunohistochemical staining, and performed a comprehensive analysis to identify promising DNA methylation biomarkers for the diagnosis of malignant transformation of sinonasal papilloma. The results showed that MAML2 expression decreased in a hierarchical manner and the expressions of NRGN and HLA-G were increased, while the expressions of GSTT1 and UCKL-1 were not significantly changed during the deterioration of SNIP-SCC. Further comparison of DNA methylation status revealed that HLA-G expression was upregulated and hypermethylated, indicating that its expression changes in SNIP-SCC may not be regulated by DNA methylation, or there are other potential regulatory mechanisms that we have not yet discovered, which need to be further studied.

MAML2, a member of the Mastermind-like protein family, is a coactivator of the oncogenic NOTCH signaling pathway [23] and plays a key role in cell proliferation, metastasis and epithelial-mesenchymal transition [24,25,26]. MAML2 is abnormally expressed in various cancers [27], for example, MAML2 enhancer region methylation is significantly increased in breast cancer and squamous cell carcinoma [28,29,30]. MAML2 expression was found to decrease after DNA methylation and histone modification [31]. MAML2 enhancer was hypomethylated and its expression was at a high level in breast cancer [32]. A study based on glioma microarray data identified MAML2 as a novel gene associated with glioma [33]. MAML2 overexpression or MAP3K1 deletion induced meki resistance. Consistent with its function as a transcriptional coregulator of NOTCH, MAML2 overexpression leads to the activation of the NOTCH target HES1. In breast cancer and melanoma, activated NOTCH promotes acquired resistance to MAPK inhibitors [34]. Our immunohistochemical staining results showed that MAML2 expression was gradually lost during the malignant transformation of SNIP-SCC, but its gene was hypermethylated.

UCKL1 is a catalytically active protein with uridine cytidine kinase and phosphoribosyltransferase, which can promote the growth of tumor cells. Serum UCKL-1 mRNA levels were found to be increased 100–1000 times in patients with breast cancer, with the highest UCKL-1 expression in Luminal A and HER2 (ERRB2) subtypes [35]. Studies have found that upregulation of UCKL1 expression may be an indicator of adverse prognosis in hepatocellular carcinomas [36]. Immunohistochemistry was used to evaluate the expression of UCKL-1 in HCC tissues, and the results showed that the expression of UCKL-1 in HCC tissues was significantly elevated [37]. Down-regulating the expression of UCKL-1 in leukemia cells by RNA interference was found to aggravate apoptosis and slow down the cell cycle process, thus reducing the growth rate of leukemia cells with small interference with UCKL-1 RNA processing [38]. In this study, immunohistochemical staining of SNIP and SNIP-SCC cases showed no significant difference in UCKL-1 expression between normal nasal mucosa tissues and grade I, II, III and IV tissues of inverted papilloma.

Glutathione S-transferases (GSTs) belong to the superfamily of phase II metabolic enzymes and play an important anticancer role by catalyzing the binding of glutathione to electrophiles that induce ROS production by heterogenin [39, 40]. The main GST enzymes are GSTM1, GSTT1, and GSTP1, among which the phenotypic loss of GSTT1 activity is due to its homozygous loss. Hypermethylation of CpG islands in the GSTT1 promoter region leads to loss of gene expression and is associated with increased susceptibility of cells to carcinogens [41,42,43]. Studies have found that GSTT1 gene promoter hypermethylation changes are found in a variety of human tumors [44]. The promoter methylation of the GSTT1 gene was found to be higher in patients with breast cancer than in healthy controls, a risk factor for breast cancer [45]. It was found that GSTT1 expression may be associated with an increased risk of colorectal cancer in Asians [46]. In 750 patients with oral squamous cell carcinoma (OSCC), GSTT1 protein expression was significantly upregulated, and was significantly higher in smoking and drinking patients than in controls, suggesting that GSTT1 and history of smoking and drinking are closely related to OSCC susceptibility. However, no significant differences in GSTT1 protein levels were observed between patients with cervical squamous cell carcinoma and healthy subjects. In this study, GSTT1 expression was not found to be associated with malignant transformation of SNIP-SCC.

Human leukocyte antigen G (HLA-G) is a non-classical MHC class I molecule originally identified in trophoblast cells at the maternal–fetal interface and plays a key role in protecting fetal allogeneic tissues from maternal immune rejection [47]. HLA-G is highly expressed in many solid tumor cells and is positively correlated with infiltrating immune cells in the tumor microenvironment [48,49,50,51], which may be a means for tumor cells to avoid immune system regulation by inhibiting natural killer and T cell-mediated lysis [52, 53]. HLA-G is considered as a novel tumor immune checkpoint molecule. Functional enrichment analysis showed that HLA-G was mainly related to T-cell activation, T-cell regulation, and lymphocyte-mediated immunity [54]. The relationship between HLA-G overexpression and aberrant DNA methylation remains elusive. It was found that in human mesenchymal stem cells and placental tissues, the hypomethylation of CpG island occurred not only in the proximal end of the HLA-G promoter, but also in the gene body [55]. The expression of HLA-G was upregulated in breast cancer and malignant melanoma was found, which was partially regulated by DNA methylation [56]. HLA-G was highly expressed in the breast cancer cell line McF-7 and its DNA was hypomethylated [57]. DNA methylation can induce HLA-G overexpression in McF-7 cells, which may be a potential target of new anticancer drugs. Studies have found that high expression of HLA-G is associated with breast cancer metastasis [58]. The expression of HLA-G in oral squamous cell carcinoma was higher than that in normal oral mucosa, which was correlated with tumor stage and lymph node metastasis [59]. However, the expression of HLA-G protein was significantly downregulated in oral squamous cell carcinoma. Transfection of HLA-G overexpression vector could significantly reduce cell viability, migration, and invasion, while inhibition of HLA-G expression could significantly enhance these properties [60]. HLA-G protein was highly expressed in vulvar squamous cell carcinoma, which was correlated with clinical stage, tumor size, and tumor invasion depth, and patients with low HLA-G expression had a good prognosis [61]. The results of this study showed that with the progression and malignant transformation of sinonasal inverted papilloma, the expression of HLA-G increased and was significantly correlated with tumor grade, suggesting that it may be involved in the progression of SNIP-SCC. However, there was no significant difference between the expression of HLA-G in inverted papilloma and that in normal nasal mucosa, suggesting that HLA-G hypermethylation may not be related to the progression of inverted papilloma.

NRGN, a human homolog of the neuron-specific rat RC3/neurogranin gene, exhibits aberrant expression in the brain and plays a role in the development of Parkinson's disease, schizophrenia and Alzheimer's disease [62,63,64]. Little is known about the role of NRGN in tumors, except that NRGN expression is reduced in gliomas and T-cell lymphomas [65]. The low expression of NRGN in GBM plays an antitumor effect in the progression of glioma, which can be used as a new therapeutic target [66, 67]. Studies have shown that NRGN can be produced at high levels outside the brain and has a novel tumor suppressor effect in murine T-cell lymphoma [61]. Studies have revealed that LINC00641 acts as a ceRNA in glioma cells by upregulating NRGN by absorption of miR-4262. Silencing NRGN expression can counteract the inhibition of glioma cell proliferation caused by upregulation of LINC00641 [67]. RBPMS-AS1 promoted NRGN transcription and enhanced radiosensitivity of GBM through miR-301a-3p/CAMTA1 axis [68]. Immunohistochemical staining showed high expression of DNCH2, ARHGEF6, NPM1, and SRI, while low expression of NRGN and TM4SF2 in gliomas [66]. It is not known whether NRGN is regulated by DNA methylation. Our study found a significantly downregulated methylation loci CG26069044 in the NRGN gene in SNIP-SCC, which may be a new finding of NRGN expression dysregulation in cancer. At the same time, immunohistochemical staining showed that the expression of NRGN was increased in SNIP-SCC, suggesting that the expression of NRGN may be involved in the process of malignant transformation.

In general, we found that the expression of MAML2 was downregulated and the gene was hypermethylated, while the expression of NRGN was upregulated and the gene was hypomethylated. There was a significant negative correlation between the expression levels of MAML2 and NRGN and the methylation status, which further confirmed the stability and reliability of our study results. These results suggest that MAML2 and NRGN may serve as methylation biomarkers and potential therapeutic targets for accurate diagnosis and treatment of SNIP-SCC.