Identification of biomarkers in laryngeal cancer by weighted gene co-expression network analysis

Abstract

Objective

Laryngeal cancer (LC) is a globally prevalent and highly lethal tumor. Despite extensive efforts, the underlying mechanisms of LC remain inadequately understood. This study aims to conduct an innovative bioinformatic analysis to identify hub genes that could potentially serve as biomarkers or therapeutic targets in LC.

Methods

We acquired a dataset consisting of 117 LC patient samples, 16 746 LC gene RNA sequencing data points, and 9 clinical features from the Cancer Genome Atlas (TCGA) database in the United States. We employed weighted gene co-expression network analysis (WGCNA) to construct multiple co-expression gene modules. Subsequently, we assessed the correlations between these co-expression modules and clinical features to validate their associations. We also explored the interplay between modules to identify pivotal genes within disease pathways. Finally, we used the Kaplan-Meier plotter to validate the correlation between enriched genes and LC prognosis.

Results

WGCNA analysis led to the creation of a total of 16 co-expression gene modules related to LC. Four of these modules (designated as the yellow, magenta, black, and brown modules) exhibited significant correlations with 3 clinical features: The age of initial pathological diagnosis, cancer status, and pathological N stage. Specifically, the yellow and magenta gene modules displayed negative correlations with the age of pathological diagnosis (r=-0.23, P<0.05; r=-0.33, P<0.05), while the black and brown gene modules demonstrated negative associations with cancer status (r=-0.39, P<0.05; r=-0.50, P<0.05). The brown gene module displayed a positive correlation with pathological N stage. Gene Ontology (GO) enrichment analysis identified 77 items, and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis identified 30 related signaling pathways, including the calcium signaling pathway, cytokine-cytokine receptor interaction, neuro active ligand-receptor interaction, and regulation of lipolysis in adipocytes, etc. Consequently, central genes within these modules that were significantly linked to the overall survival rate of LC patients were identified. Central genes included CHRNB4, FOXL2, KCNG1, LOC440173, ADAMTS15, BMP2, FAP, and KIAA1644.

Conclusion

This study, utilizing WGCNA and subsequent validation, pinpointed 8 genes with potential as gene biomarkers for LC. These findings offer valuable references for the clinical diagnosis, prognosis, and treatment of LC.

Laryngeal cancer (LC), primarily of squamous cell carcinoma type, remains one of the most prevalent head and neck carcinomas. LC can originate in different parts of the larynx, including the glottis, supraglottic, and subglottic regions, with overall survival rates being influenced by the tumor's location ^[1]. LC may spread by direct extension into surrounding tissues, metastasize to regional cervical lymph nodes, or disseminate to distant sites through the bloodstream. Symptoms of LC include hoarseness or other voice changes, a neck lump, sore throat, persistent cough, stridor, and difficulty swallowing^[2]. However, early diagnosis can be challenging due to these non-specific symptoms. Some LC patients are diagnosed at an advanced clinical stage, resulting in a decreased quality of life and a poorer prognosis. While significant advancements have been made in LC treatment, including surgical intervention, radiotherapy, and novel therapies, the development of a prognostic multigene classifier holds promise for early LC diagnosis, assessing the risk of recurrence, or guiding treatment decisions in the future^[3].

Cancer is indeed a complex disease, involving a multitude of risk factors, intricate biological processes, and far-reaching consequences throughout the body. Using a single molecule as a biomarker for precision medicine in LC patients has its limitations ^[4]. One powerful method to address this complexity is the construction of co-expression modules, which allows for the development of gene co-expression networks on a large scale. Weighted gene co-expression network analysis (WGCNA) is a comprehensive collection of R functions that has been extensively utilized for large-scale data analysis. It helps explore the relationship between gene-based networks and clinical phenotypes using RNA sequencing or microarray data from various samples^[5]. One powerful method to address this complexity is the construction of co-expression modules, which allows for the development of gene co-expression networks on a large scale. WGCNA is a comprehensive collection of R functions that has been extensively utilized for large-scale data analysis. It helps explore the relationship between gene-based networks and clinical phenotypes using RNA sequencing or microarray data from various samples. WGCNA has been successfully applied to transform gene expression data into co-expression modules and identify candidate biomarkers for several types of cancer, often associated with significant clinical features. This approach offers a more holistic understanding of the molecular underpinnings of cancer and has the potential to uncover novel insights for precision medicine in LC and other malignancies. Through single-cell transcriptome analysis and WGCNA, a study^[7] published in Cell journal has revealed the molecular characteristics of CD133⁺/ glial fibrillary acidic protein (GFAP)^- ependymal (E) cells within the adult mouse forebrain neurogenic zone. Surprisingly, it has been found that prominent hub genes within the unique gene network of ependymal CD133⁺/GFAP^- quiescent cells were enriched in immune-responsive genes and genes encoding receptors for angiogenic factors. These comprehensive findings shed light on subsequent neural lineage differentiation and migration^[7]. However, it’s worth noting that WGCNA is not commonly employed for studying LC to identify prognostic biomarkers^[8]. Therefore, this study aims to describe the correlation patterns among genes using a systematic biological approach based on WGCNA and to identify novel biomarkers associated with LC prognosis.

1. Materials and methods

1.1. The Cancer Genome Atlas data of LC patients

The Cancer Genome Atlas (TCGA) research network marked the end of the TCGA program by publishing the Pan-Cancer Atlas: A collection of cross-cancer analyses delving into overarching themes on cancer, including cell-of-origin patterns, oncogenic processes and signaling pathways. The data remain available to the public for further mining through the Genomic Data Commons (https://cancergenome.nih.gov/). Level 3 RNA-seq V2 and clinical data were obtained from 117 LC patients in the TCGA database. The gene expression missing value (expression=0) that more than 20% was excluded. Nine prognostic factors of LC data were extracted, including age at initial pathologic diagnosis (patients were aged 38 to 83), alcohol history (yes/no), sex, neoplasm histologic grade (G1, G2, G3, G4, GX), pathologic N stage (N0, N1, N2, N3, NX), pathologic T stage (T1, T2, T3, T4, TX), pathologic stage (stage I-IV), presence of perineural invasion (yes/no), and cancer status (with tumor/tumor-free).

1.2. Weighted correlation network analysis of genes in LC patients

WGCNA was used to investigate the relationship between gene expression pattern and clinical phenotypes. In this co-expression network, weighted gene co-expression modules were constructed using gene expression data through WGCNA package of R software (https://www.r-project.org/)^[9]. The analysis was performed as follows. Co-expression network was constructed as undirected and weighted gene network. The nodes of such a network correspond to gene expression profiles, and edges between genes are determined by the pairwise correlations between gene expressions. By raising the absolute value of the correlation to a power β (soft thresholding), the weighted gene co-expression network construction emphasizes high correlations at the expense of low correlations. Optionally, the user can also specify a signed co-expression network where the adjacency is defined as aij=|[1+cor(xi, xj)]/2| ^β (Xi and Xj were vectors of expression values for genes i and j. aij encoded the network connection strength between genes i and j). In this study, when β value (>=1) was at least 3, the corresponding scale free R² value was 0.91 to obtain a good scale-free topology model. The modules are clusters of highly interconnected genes. In an unsigned co-expression network, modules correspond to clusters of genes with high absolute correlations. Intramodular connectivity measures how connected a given gene are with respect to the genes of a particular module. The intramodular connectivity may be interpreted as a measure of module membership (MM). Then, cluster dendrogram was transformed into topology matrix to form network heatmap plot. The adjacency matrix algorithm was used to generate the topological overlap matrix (TOM). The average linkage hierarchical clustering was conducted according to TOM-based dissimilarity measure, which classified similar expression profiles of gene-modules into the same gene-module cluster with the DynamicTreeCut algorithm. Heatmap tool package was plotted to analyze the strength of network interactions. The relationships between modules and 9 LC prognostic factors (age at initial pathologic diagnosis, cancer status, and pathologic N stage) were analyzed by calculating the Pearson correlation coefficient and visualized by heat map with a statistical significance level at P<0.05. In addition, the gene significance (GS) was defined as mediated P-value of each gene (GS=lgP) in the linear regression between gene expression and the clinical traits.

1.3. Pathway and gene ontology enrichment analysis of co-expression modules

The Kyoto Encyclopedia of Genes and Genomes (KEGG) is a reference knowledge base involving systems information, genomic information, and chemical information. KEEG pathway (https://david.ncifcrf.gov/home.jsp) enrichment analyses within co-expression modules with a statistical significance level at P<0.05. Gene ontology (GO) information can be useful for further prioritizing intramodular co-expression module genes (https://cytoscape.org/). GO analysis on biological process (BP), molecular function (MF), and cellular component (CC) was performed with Cytoscape ClueGO (two-sided hypergeometric test, adjusted P<0.05 corrected with Benjamini-Hochberg).

1.4. Identifying hub molecules with molecular complex detection

The RNA-RNA interactions were analyzed with Cytoscape software (version 3.2.1; National Resource for Network Biology) to obtain the network. The criteria of hub-molecule searching were set as the molecular complex detection (MCODE) score>6, and statistical significance of P<0.05. The identified hub genes in co-expression modules were plotted expression correlation network with Rstudio, and their Pearson correlation coefficients were calculated to determine the significant correlation pairs. Furthermore, overall survival analysis of hub genes in LC patients was performed with Kaplan-meier plotter by R package.

1.5. Statistical analysis

All data were analyzed by R software 3.4.1 (https://www.r-project.org/). In all cases, P<0.05 was considered as statistical significance. For the pair of molecules, Pearson correlation coefficient was calculated. Benjamini-Hochberg for multiple testing, and false discovery rate were calculated to correct the P-value. Survival curves were created using the Kaplan-Meier method, and any differences in the survival curves were compared by the log-rank test.

2. Results

2.1. Construction of co-expression modules of LC

Before beginning WGCNA analysis, the raw data should be pre-treatment. The expression values of all genes (n=16 746) in 117 LC samples were used to WGCNA analysis. In this study, the dataset with the top genes (n=5 000) were selected for constructing co-expression modules with WGCNA package after going on WGCNA package filtration. The clinical characteristics of these eligible LC patients were summarized. Sample clustering was used to detect outliers according to RNA expression data. The red line was the cut-off value of data filtering in the step of data preprocessing, and 116 LC samples included (Figure 1A). Sample dendrogram and trait heatmap were plotted based on RNA expression data and clinical information (Figure 1B). Firstly, the power value that based on LC gene expression data was plotted. When the power value was equal to 3, the scale R ² was up to 0.91, and the intramodular average connectivity degree was higher (Figure 2A). Therefore, the power value (β=3) was be chosen to construct the distinct gene co-expression modules in LC patients. Cluster dendrogram of all selected genes was clustered based on a TOM-based dissimilarity matrix. These co-expression modules were represented in different colors (Figure 2B). Their correlation was analyzed between different co-expression modules. Heatmap plot of topological overlap model was plotted based on this gene network. The network heatmap plot of all genes and module assignment were shown (Figure 2C). In the heatmap, blue color represented low adjacency (negative correlation), while red represented high adjacency (positive correlation). Squares of red color along the diagonal were the meta-modules (Figure 2D). Therefore, the power value used to construct co-expression module and the results show that identifying 16 distinct gene co-expression modules (black, tan, purple, magenta, yellow, salmon, blue, brown, cyan, turquoise, midnight blue, red, pink, green, greenyellow, and grey modules) in LC. The number of genes in the 16 co-expression modules was shown in Table 1.

Figure 1. Sample cluster analysis based on RNA data from TCGA database.

A: Sample clustering was performed to identify outliers using RNA data. The red line represents the data filtering cut-off in the data preprocessing step. B: Sample dendrogram and trait heatmap were generated based on gene expression data and clinical information.

Figure 2. Construction of co-expression modules of laryngeal cancer.

A: Analysis of network topology using various soft-threshold powers, plotting the scale-free fit index (Y-axis) against the mean connectivity (degree, Y-axis). This step aimed to confirm the presence of a scale-free topology. The adjacency matrix was defined using soft-thresholds with β=3. B: Clustering dendrograms for genes was generated, with dissimilarity calculated based on topological overlap, accompanied with assigned module colors. As a result, 16 co-expression modules were constructed. C: Heatmap depicts the topological overlap matrix (TOM) among genes, which is based on co-expression modules. D: Using a heatmap plot was employed to visualize the gene network.

Table 1.

Number of genes in the 16 co-expression modules

Module colors	Gene frequency
Black	890
Tan	1 047
Purple	535
Magenta	453
Yellow	320
Salmon	292
Blue	283
Brown	211
Cyan	186
Turquoise	180
Midnightblue	145
Red	129
Pink	127
Green	94
Greenyellow	76
Grey	32

2.2. Gene co-expression modules correspond to clinic traits

The association analysis was performed between common expression eigengene pattern in co-expression module and the particular clinic traits dataset from TCGA database. Heatmap of the correlation between module eigengenes and clinical traits of LC showed correlation coefficient and P value (Figure 3). The results from the eigengene dendrogram and heatmap analysis demonstrated significant associations between 4 co-expression modules and various clinical factors (Figure 4). The yellow module exhibited a significant correlation with age at initial pathologic diagnosis (r=0.25, P=6.0E-03). Similarly, the magenta module showed a significant correlation with age at initial pathologic diagnosis (r=0.31, P=7.0E-04) and pathologic N stage (r=0.28, P=2.0E-03). The black module was associated with cancer status (r=0.30, P=1.0E-03). Lastly, the tan module displayed correlations with age at initial pathologic diagnosis (r=0.21, P=3.0E-02), pathologic N stage (r=0.26, P=5.0E-03), and cancer status (r=0.33, P=3.0E-04). Besides, we also plotted a scatterplot of GS vs. MM in the four modules respectively. The results revealed that MM in co-expression yellow module (r=-0.23, P=3.3E-05) and co-expression magenta module (r=-0.33, P=6.1E-06) was correlated with age at initial pathologic diagnosis. MM in co-expression black module (r=-0.39, P=4.5E-0.9) and co-expression tan module (r=-0.50, P=2.2E-0.9) was correlated with cancer status. MM in co-expression tan module (r=-0.31, P=3.9E-04) was correlated with pathologic N stage.

Figure 3. Analysis of module-trait relationships of LC based on TCGA data Each row corresponds to a module eigengene, and column to a trait. a: Age at initial pathologic diagnosis; b: Alcohol history; c: Sex; d: Neoplasm histologic grade; e: Pathologic N stage; f: Pathologic T stage; g: Pathologic stage; h: Presence of perineural invasion; i: Cancer status. LC: Laryngeal cancer; TCGA: The Cancer Genome Atlas.

Figure 4. Identifying correlated eigengenes and meta-modules.

A: In the dendrogram, the yellow and magenta modules exhibit strong correlations with age at initial pathologic diagnosis. This relationship is further illustrated in the scatterplot, which depicts gene significance (GS) for age at initial pathologic diagnosis against module membership (MM) in the yellow and magenta modules. B: Similarly, the dendrogram highlights the strong connection between the black and tan modules and cancer status. The scatterplot demonstrates the correlation between GS for cancer status and MM in the black and tan modules. C: Dendrogram reveals a close relationship between the black and tan modules and pathologic N stage. This association is evident in the scatterplot, which showcases GS for pathologic N stage against MM in the tan module.

2.3. Functional enrichment analysis of genes in interested co-expression modules

KEGG pathway analysis revealed 30 statistically significant signaling pathways to involve genes identified in those interested co-expression modules. Interestingly, it was demonstrated that LC exhibited an increased dependence on multiple signaling pathways, including calcium signaling pathway, cytokine-cytokine receptor interaction, neuroactive ligand-receptor interaction, regulation of lipolysis in adipocytes, adrenergic signaling in cardiomyocytes, signaling pathways regulating pluripotency of stem cells, and gastric acid secretion according to the co-expression black module. GO enrichment analysis of genes in interested co-expression modules revealed CC, MF, and BP. The genes in yellow co-expression module were mainly distributed in negative regulation of lipid metabolic process, monocarboxylic acid catabolic process, retinoid metabolic process, gland morphogenesis, sensory organ morphogenesis, regulation of cation channel activity, diencephaton development, eye morphogenesis, negative of neuron differentiation, mechanosensitived protassium channel activity, and positive regulation of cation transmembrane transport (Figure 5A). The genes in magenta co-expression module were mainly distributed in cation channel complex, presynaptic process involved in chemical synaptic transmission, sensory perception of sound, microtubule associated complex, anchored component of membrane, neuron migration, negative regulation of lipid metabolic process, regulation of canonical WNT signaling pathway (Figure 5B). The genes in black co-expression module were mainly distributed in chondrocyte differentiation, chondrocyte differentiation, amino acid transmembrane transporter activity, nerve development, presynaptic membrane, positive regulation of synapse assembly, positive regulation of extrinsic apoptotic signaling pathway, regulation of extrinsic apoptotic signaling pathway in absence of ligand, bile acid metabolic process, and regulation of biomineral tissues development (Figure 5C). The genes in tan co-expression module were mainly distributed in endocrine process, adult locomotory behavior, sensory perception of pain, Z disc, embryonic limb morphogenesis, regulation of smooth muscle contraction, and anterior/posterior pattern specification (Figure 5D).

Figure 5. Gene ontology analysis of co-expression modules.

A: Co-expression yellow modules; B: Co-expression magenta modules; C: Co-expression black modules; D: Co-expression tan modules.

2.4. Hub genes and survival-associated genes

The intramodular connectivity of each gene was calculated by summing the connection strengths with other module genes and dividing this number by the maximum intramodular connectivity. Genes with high intramodular connectivity (MCODE score >6 and P<0.05) are considered as intramodular hub genes. A total of 18 hub genes were identified from 212 genes in the black co-expression module, while 19 hub-mRNAs were identified from 179 genes in the magenta module. Additionally, 11 hub-mRNAs were identified from 126 genes, and 22 hub-mRNAs were identified from 319 genes in the tan co-expression module. The Kaplan-Meier Kaplan-Meier plot analysis demonstrated a significant association between high expression levels of four out of 18 hub genes in the black co-expression module and poor outcomes in LC (P<0.05). These genes include a disintegrin and metalloproteinase with thrombospondin motifs 15 (ADAMTS15) (P=0.017), bone morphogenetic protein 2 (BMP2) (P=0.023), fibroblast activation protein (FAP) (P=0.035), and protein shisa-like-1 (KIAA1644) (P=0.043), as depicted in Figure 6A. Additionally, the Kaplan-Meier plot analysis revealed that high expression levels of 4 out of 12 hub genes in the tan co-expression module were associated with worse overall survival in LC (P<0.05). These genes inluded neuronal acetylcholine receptor subunit beta-4 (CHRNB4) (P=0.014), forkhead box protein L2 (FOXL2)(P=0.008), potassium voltage-gated channel subfamily G member 1 (KCNG1) (P=0.017), and long noncoding RNA 440173 (LOC440173) (P=0.041), as illustrated in Figure 6B. Moreover, R-Studio software was used to calculate co-expressions relationship of these identified hub genes from the interest co-expression modules (Figure 6C), and output their correlation coefficients, and P-values.

Figure 6. Analysis of overall survival-related hub genes and their co-expressions relationship in laryngeal cancer.

A: Overall survival-related hub genes were identified in the co-expression black modules, namely ADAMTS15, BMP2, FAP, and KIAA1644. B: Overall survival-related hub genes were identified in the co-expression tan modules, including CHRNB4, FOXL2, KCNG1, LOC440173. C: Co-expressions relationship among these identified hub genes are presented, shedding light on their potential interactions and roles in laryngeal cancer.

3. Discussion

LC is a common neoplasm affecting the head and neck, with over 150 000 new cases diagnosed annually. Despite advancements in medical care, the mortality rate for LC has remained high over the past few decades ^[10]. Currently, early-stage LCs are primarily managed through radiation therapy, chemotherapy, laser resection, and partial laryngectomy. In contrast, advanced LC cases typically necessitate total laryngectomy followed by comprehensive treatment. In the case of early-stage patients, efforts are made to preserve their swallowing functions as much as possible. Unfortunately, patients in advanced stages often experience the loss of natural voice, dysphagia, or require permanent tracheostomy^[11]. The development of novel biomarkers is urgently required to enhance early diagnosis, monitor recurrence, and improve therapeutic strategies to enhance the long-term quality of life for LC patients^[12].

In this study, we conducted a comprehensive analysis using the WGCNA method to construct a total of 16 co-expression gene modules. These modules were derived from a dataset comprising 16 746 genes obtained from 117 human LC samples. Compared to other analytical methods, WGCNA offers distinct advantages as it primarily focuses on elucidating the associations between co-expression modules of interest and clinically significant traits^[13]. Furthermore, our results exhibited a notably higher degree of reliability in identifying hub molecules, disease-related pathways, and significant biological processes compared to a previous WGNAC study of gastric tumors^[14]. Genes within the same co-expression module have been shown to be functionally associated with each other. Consequently, the hub genes and biologically relevant pathways we identified hold significant potential as biomarkers for detection and treatment strategies. In our study, we identified 4 co-expression gene modules that exhibit significant associations with clinical traits, specifically age at initial pathologic diagnosis, pathologic N stage, and cancer status. The issue of a lower average age at the onset of cancer in patients has been gaining increasing attention. Additionally, lymphatic metastasis, recurrence, and distant metastasis of cancer contribute to high mortality rates^[15]. The development and progression of cancer are accompanied by the abnormal expression of suppressor genes or oncogenes. Consequently, the enriched co-expression modules related to clinical traits may potentially regulate a multitude of cellular functions^[16]. The results of the gene ontology analysis further provided substantial evidence that the identified biomarkers associated with clinical traits in LC, based on the co-expression modules, play pivotal roles in various biological processes, molecular functions, and cellular components. These cellular functions are intimately linked to the malignant phenotype of cancer. For instance, they encompass the regulation of lipid metabolic processes (involving genes like AKR1C3, CYP27B1, GPLD1, LPCAT1, UGT1A10, UGT1A6, UGT1A7, UGT1A9 within the co-expression yellow module) and the regulation of ion transmembrane transport (including genes like GSTM2, HAP1, KCNH2, KCNJ11, NOS1AP, NPSR1, SLC6A4, WNK2 within the co-expression yellow module). Additionally, they encompass the regulation of the canonical WNT signaling pathway (involving genes such as FZD9, GLI1, GPC3, IGFBP6, LGR6, MESP1, SOST, WNT4, ZNF703 within the co-expression magenta module). Therefore, we hypothesize that the 4 identified co-expression modules—namely, the yellow module, magenta module, black module, and tan module—exhibiting significant associations with LC clinical traits (age at initial pathologic diagnosis, pathologic N stage, cancer status) are of paramount importance in the tumorigenesis and progression of LC.

The results of the pathway and cell biological function enrichment analysis revealed statistically significant findings in several crucial cancer-related pathways. These pathways include those associated with cell proliferation, substantial metabolism, energy metabolism, cell differentiation processes, and signaling pathways regulating the pluripotency of stem cells. For instance, the co-expression yellow module was notably enriched in the mitogen-activated protein kinase (MAPK) signaling pathway. This module contained several key genes (such as BDNF, NTF3, NTRK2, MAPK8IP2, FGF13, FGF12, CACNB4, MAPK10, CACNA2D3, MAP2K6, CACNA1B) that are crucial in cell proliferation-related pathways. The MAPK pathway represents a chain of protein kinases within the cell, serving as a conduit for transmitting biomedical signals from cell-membrane receptors to the nuclear DNA of the cell^[17]. After nuclear DNA expression leads to the production of specific proteins, it initiates a cascade of biological processes within the cell, such as cell proliferation. This step is often considered crucial in the development of various cancers^[18]. The key genes we have identified in our study have been previously reported in scientific literature. For instance, MAP2K6 has been shown to increase the ratio of Beclin-1/Bcl-2, which are markers associated with autophagy. Research^[19] results indicate that the overexpression of MAP2K6 in cancer cells significantly activates the MAPK pathway, reducing cell migration capability through an autophagy-dependent mechanism. NTRK3 is identified as a novel aberrantly methylated tumor suppressor gene. Dysregulation in the expression of NTRK3 can activate the MAPK pathway. This regulatory mechanism involving NTRK3 suggests potential novel treatment approaches for various cancers^[20]. Over the past decade, there has been a significant focus on understanding the metabolic reprogramming of cancer cells. It is well-established that the abnormal metabolism of tumor cells can profoundly influence tumor progression^[21]. In our study, we observed that the co-expression yellow module exhibited enrichment in various metabolic pathways, including retinol metabolism, glutathione metabolism, arachidonic acid metabolism, and inositol phosphate metabolism. Within these metabolic pathways, several key genes were identified, including GCNT2, NT5C1B, ADH7, ATP6V1B1, ITPKA, PIPOX, ALDH3A1, AKR1C3, ALDH1A1, UGT1A7, GLS2, UGT1A6, UGT1A9, CYP27B1, DGKB, INPP5J, CYP26B1, PLCH1, UGT8, PCYT1B, PLA2G16, CYP26A1, AK7, UGT1A10, CEL, PLCE1, FOLH1, ATP6V0E2, GCK, AKR1B10, BHMT, CYP4F3, RDH16, and ABO. Notably, study^[22] has shown that CYP4F3 plays a role in activating the metabolic pathway related to environmental chemical carcinogens. The mechanism of cancer susceptibility might be associated with CYP4F3 polymorphisms, which can alter its biological functions^[22]. Furthermore, recent study^[23] have indicated that high levels of AKR1B10 expression are detected in various human tumors. This gene may play a critical role in cancer progression and development through its involvement in metabolic pathways, such as retinoic acid homeostatic regulation, carbonyl detoxification, the activation of tobacco smoke carcinogens, and lipid metabolic control^[23]. Furthermore, the co-expression magenta module was observed to be significantly enriched in the WNT signaling pathway, housing several key genes [FZD9, WNT10A, WNT4, SOST, WNT inhibitory factor-1 (WIF1)] that are crucial in cell differentiation-related processes. Initially, WNT signaling was recognized for its role in carcinogenesis, and later for its involvement in embryonic development. Subsequent research unveiled that mutations in genes within the WNT signaling pathway are responsible for cancer development. Notably, hypermethylation of the WIF1 gene promoter has been identified in various cancer types. Encouragingly, omega-3 polyunsaturated fatty acids (PUFA) have been reported to modulate WIF1 gene expression by regulating its promoter. This suggests that the anti-cancer effects on the WNT signaling pathway could provide hope and opportunities for individuals battling cancer^[24]. In addition, stuy^[25] have also demonstrated that SOST is overexpressed in cancer tumor tissues and cell lines. The SOST gene appears to inhibit the WNT pathway, contributing to the progression of metastatic cancer. Investigating the mechanisms through which SOST is involved in cell differentiation may shed light on its role in regulating pathways such as the WNT signaling pathway^[25].

Hub genes were identified from the co-expression modules we identified. Subsequently, Kaplan-Meier plot analysis revealed significant associations with overall survival in LC for a subset of these hub genes. Specifically, within the black co-expression module, 4 out of 18 hub genes demonstrated a significant association with LC overall survival. These genes include ADAMTS15, BMP2, FAP, and KIAA1644. Similarly, within the tan co-expression module, 4 out of 12 hub genes were significantly associated with LC overall survival. These genes are CHRNB4, FOXL2, KCNG1, and LOC440173. It is noteworthy that some of our findings are consistent with previous reports, underscoring the reliability of the newly identified biomarkers. For instance, the BMP2 gene encodes secretory transforming growth factor-beta superfamily proteins, including TGF-beta ligands. The interaction between these ligands and TGF-beta receptors leads to the recruitment and activation of SMAD family transcription factors, which in turn regulate gene expression^[26]. Head and neck squamous cell carcinomas that exhibit elevated levels of BMP-2 protein are strongly correlated with higher rates of local recurrence. These published findings hold significant potential for utilizing BMP-2 in tissue engineering and reconstructive approaches when dealing with cancer-related defects^[27]. The protein encoded by the FAP gene is a homodimeric integral membrane gelatinase, belonging to the serine protease family. FAP gene expression is selective and primarily observed in stromal fibroblasts of epithelial cancers, granulation tissue, and malignant cells in bone. This protein is believed to play a role in the regulation of epithelial carcinogenesis^[28]. Notably, carcinoma-associated fibroblasts (CAFs), which can influence the malignant characteristics of LC cell lines, exhibit positive staining for FAP in LC. These results suggest that the microenvironment surrounding laryngeal tumors may have a significant impact on disease progression^[29]. Additionally, our study unveiled several novel findings, notably, the identification of ADAMTS15, KIAA1644, CHRNB4, FOXL2, KCNG1, and LOC440173 as previously unreported in the context of LC. This highlights the need for further research to validate these hub genes as potential novel LC biomarkers. In most of the previously reported studies^{[3, 12, 18-19]}, the focus has been on establishing evidence that single factors or gene mutations play a pivotal role in the development and progression of cancer. However, study^[30] have recognized that the prediction models involving multiple genes may offer improved prospects for personalized medicine. In our study, we aimed to construct a reliable biomarker pattern. This implies that by utilizing a set of identified biomarkers to develop a prediction model, we could potentially enhance the current diagnostic and prognostic capabilities for LC.

In conclusion, our study employed WGCNA to uncover 4 co-expression modules (yellow, magenta, black, tan co-expression modules) that exhibit significant correlations with three clinical traits—age at initial pathologic diagnosis, cancer status, and pathologic N stage—in the context of LC. These identified WGCNA modules have unveiled gene signatures or sets that are strongly linked to distinct subtypes of LC. These gene signatures hold great promise as potential biomarkers for early detection, diagnosis, and prognostic assessment of the disease. Furthermore, integrating the WGCNA modules with clinical data has the potential to provide valuable insights into the underlying molecular mechanisms driving the progression of LC. By pinpointing key genes within these modules, researchers can identify potential therapeutic targets that play pivotal roles in the progression of this cancer. The targeting of these specific genes or pathways may pave the way for the development of novel therapeutic strategies and the identification of druggable targets, offering exciting prospects for precision medicine approaches in the treatment of LC.

Admittedly, our study is not without limitations. Initially, our plan was to incorporate validation cohorts into our research. Regrettably, due to resource constraints and limited access to additional samples, we were unable to include these cohorts in the current study. Furthermore, the absence of experimental validation poses a limitation as it restricts our ability to confirm the functional relevance of the identified gene features and co-expression modules. It is evident that further in vitro experiments or in vivo models will be necessary in the future to validate the functional roles and molecular mechanisms associated with the observed relationships. These additional experiments will help affirm the clinical significance of our findings.

Availability of data and materials: The data used to support the findings of this study are available from the corresponding author upon request.

Funding Statement

This work was supported by the Natural Science Foundation of Hunan Province, China (2019JJ40481).

Conflict of Interest

The authors declare that they have no conflicts of interest to disclose.

AUTHORS’CONTRIBUTIONS

ZHANG Fengyu Research design, experimental operation, data collecting and analysis, and paper writing; SHE Li Research design, experimental operation, paper modification; HUANG Donghai Research design, paper supervision and revision. The final version of the manuscript has been approved and read by all authors.

Note

http://xbyxb.csu.edu.cn/xbwk/fileup/PDF/2023081136.pdf