Identification of hub genes and key modules in laryngeal squamous cell carcinoma

Abstract

Background

Laryngeal squamous cell carcinoma (LSCC) is the prominent cancer in head and neck, which greatly affects life quality of patients. The pathogenesis of LSCC is not clear. Presently, the LSCC treatments include chemotherapy, surgery and radiotherapy; however, these methods have poor efficacy in patients with recurrent and persistent cancer. Therefore, the study identified the hub genes accompanied with LSCC, which may be a potential therapeutic target in the future.

Methods

We extracted whole transcriptome high-throughput sequencing (HTS) LSCC data from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases and calculate differentially expressed genes (DEGs) between LSCC and normal samples using statistical software RStudio. Through weighted gene co-expression network analysis (WGCNA), enrichment examination of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways and Gene Ontology (GO) functions, and examination of protein-protein interaction (PPI) network, we obtained network hub genes and validated the hub genes prognostic value and expression levels of protein.

Results

Through analysis of differential gene expression, from the GEO and TCGA databases 2,139 and 2,774 DEGs were obtained, respectively, 13 and 15 modules were screened from TCGA-LSCC and GSE127165 datasets by WGCNA, respectively. The most significant positive and negative correlation modules in the WGCNA and DEG lists were overlapped, and overall 36 co-expressed overlapping genes were retrieved. Through enrichment analysis of GO and KEGG, it was found that the gene functions were highly concentrated in cell junction assembly, basement membrane, extracellular matrix (ECM) structural constituent etc., and the pathways were mainly concentrated in ECM receptor interaction, focal adhesion, small cell lung cancer, and toxoplasmosis. Through analysis of PPI network analysis, 10 network hub genes (SNAI2, ITGA6, LAMB3, LAMC2, CAV1, COL7A1, GJA1, EHF, OAT, and GPT) were obtained. Finally, survival analysis and protein expression validation of these genes confirmed that low OAT expression and high CAV1 expression remarkably influenced the survival of patient’s prognosis with LSCC.

Conclusions

We recognized the hub genes and key modules nearly associated to LSCC and these genes were validated by survival analysis and the database of Human Protein Atlas (HPA), which is of high importance for unveiling the pathogenesis of LSCC and probing for new precise biological marker and potential therapeutic targets.

Highlight box.

Key findings

• We recognized the hub genes and key modules nearly associated to laryngeal squamous cell carcinoma (LSCC) and these genes were validated by survival analysis and the database of Human Protein Atlas.

What is known and what is new?

• LSCC is a common cancer of the head and neck, which greatly affects patients’ quality of life. The pathogenesis of LSCC remains unclear.

• We identified hub genes and key modules that are closely associated with LSCC incidence and growth, and verified these genes by survival analysis and protein expression.

What is the implication, and what should change now?

• The discovery of hub genes and key modules related to LSCC is of great significance for revealing the pathogenesis of LSCC and exploring new precise biomarkers and potential therapeutic targets.

Introduction

Laryngeal squamous cell carcinoma (LSCC) is the prominent cancer in head and neck, which greatly affects life quality of patients. According to the latest European Head and Neck Society (EHNS)-European Society for Medical Oncology (ESMO)-European Society for Radiotherapy and Oncology (ESTRO) clinical practice guidelines, the incidence is 4.6/100,000 and the year relative survival rate is 61% (1). The pathogenesis of LSCC is not clear and is presently believed in the involvement of the synergistic effect of numerous carcinogenic factors like alcohol consumption, smoking, human papillomavirus (HPV) infection and environmental factors (2). Presently, the LSCC treatments include chemotherapy, surgery and radiotherapy; however, these methods have poor efficacy in patients with recurrent and persistent cancer (3). Therefore, there is an urgent need to recognize the hub genes accompanied with LSCC and identify novel therapeutic targets.

With modern creation evolution of genomics and high-throughput sequencing (HTS) technology, numerous gene datasets store a high quantity disease gene expression information and clinical information (4,5), which lays a foundation for evaluating the mechanism of disease at the molecular level and biological gene functions through bioinformatic analysis. Analysis of weighted gene co-expression network analysis (WGCNA) is a systematic bioinformatic analysis which is used to recognize modules of co-expressed genes which are greatly associated with clinical practice. WGCNA promotes screening methods of complex-based differential genes that can recognize potential biological markers and therapeutic targets (6).

In the present study, WGCNA and examination of differential expression of genes were conducted on whole transcriptome of LSCC sequencing data from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) datasets, and co-expressed differential genes and key modules were obtained. Through Gene Ontology (GO) and enrichment analyses of Kyoto Encyclopedia of Genes and Genomes (KEGG), examination of protein-protein interaction (PPI), survival analysis, protein expression validation, we further explored the molecular mechanism of the incidence and growth of LSCC.

Methods

Collation of data from TCGA and GEO database

Data of gene expression in LSCC were copied from TCGA database (7) (https://portal.gdc.cancer.gov) and GEO databases (8) (https://www.ncbi.nlm.nih.gov/gds). All expression matrices and correlated clinical data of LSCC were copied through the R package TCGAbioLinks (9) in TCGA database. Overall, 123 samples were collected, including 111 cases of LSCC and 12 cases of normal tissue (see Table 1). In the GEO database, the large-sample HTS database GSE127165 was obtained using the R package GEOquery (10) and included 57 LSCC tissue samples and 57 adjacent normal mucosal tissue samples (see Table 2) (11). The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013).

Table 1. Basic characteristics of 111 LSCC samples in dataset TCGA.

Parameters	Number of cases (%)
Age (years)
≤60	47 (42.3)
>60	64 (57.7)
Gender
Female	20 (18.0)
Male	91 (82.0)
T staging
T1	7 (6.3)
T2	12 (10.8)
T3	25 (22.5)
T4	54 (48.6)
Unknown	13 (11.7)
Metastasis of cervical lymph node
N0	39 (35.1)
N1	12 (10.8)
N2	39 (35.1)
N3	2 (1.8)
Unknown	19 (17.1)
Distant metastasis
M0	40 (36.0)
M1	1 (0.9)
Unknown	70 (63.1)
Clinical stage
I	2 (1.8)
II	9 (8.1)
III	14 (12.6)
IV	71 (64.0)
Unknown	15 (13.5)

LSCC, laryngeal squamous cell carcinoma; TCGA, The Cancer Genome Atlas.

Table 2. Basic characteristics of 57 LSCC samples in dataset GSE127165.

Parameters	Number of cases (%)
Age (years)
≤60	31 (54.4)
>60	26 (45.6)
Gender
Female	4 (7.0)
Male	53 (93.0)
Primary site
Glottic	30 (52.6)
Supraglottic	21 (36.8)
Subglottic	2 (3.5)
Transglottic	4 (7.0)
Differentiation
High	8 (14.0)
Medium	37 (64.9)
Low	12 (21.1)
T staging
T1	21 (36.8)
T2	13 (22.8)
T3	17 (29.8)
T4	6 (10.5)
Metastasis of cervical lymph node
N0	43 (75.4)
N+	14 (24.6)
Distant metastasis
M0	57 (100.0)
M1	0 (0.0)
Clinical stage
I	20 (35.1)
II	10 (17.5)
III	15 (26.3)
IV	12 (21.1)

LSCC, laryngeal squamous cell carcinoma.

Screening of differentially expressed genes (DEGs)

RStudio statistical software (12) (version 1.3.959; https://rstudio.com/) and Bioconductor packages (13) (http://www.bioconductor.org/) were employed in the bioinformatic analysis of LSCC and normal tissue samples. First, raw data from the TCGA database and GEO dataset GSE127165 were gathered and brought up in accordance to the RStudio input file format software. Heatmaps and volcano plots of DEGs were drawn by the R packages limma, edgeR, ggplot2, and pheatmap (14-17). The adjusted (adj.) P values <0.05 and |logfold change (logFC)| >1 were determined statistically significant.

WGCNA

WGCNA is a systematic bioinformatic method. It is a network-based gene screening method that explores the complex relationships between highly correlated gene modules and disease phenotypes in different samples (18,19). R package WGCNA (6) in RStudio were used to carry out WGCNA for the LSCC gene expression data in TCGA and GEO databases. First, the sorted raw data were input, sample information and clinical data were read, and the LSCC tissue samples and normal samples were gathered. To construct the biological hub network, we fixed the optimal power value using function PickSoftThreshold to calculate the gene co-expression fitting index and average connectivity (20). The corresponding relationship was then transfigured to a topological overlap matrix (TOM), that was employed for sorted gathering of genes and modules of rigidity was recognized, and identical modules were gathered and merged (21). Then, the listing of genes of every single module, tree diagram of gene-module, and module-trait relationship diagram were output (22).

Overlapping of DEG analysis and WGCNA

We used R package VennDiagram (23) and intersected DEGs and WGCNA results to get the associated genes of LSCC. First, the modules with the most significant positive and negative associations with LSCC samples in module-trait relationship diagrams were obtained, intersected the module genes with positive and negative correlations with DEGs, and then drew two Venn diagrams. Finally, the hub intersection genes in the two Venn diagrams (24) were combined to obtain the final genes related to LSCC.

Functional and pathway enrichment analysis

The R packages org.hs.eg. db, clusterProfiler, enrichplot, and ggplot2 (16,25) were used for conducting GO analysis of genes associated to LSCC, consisting of cell components (CCs), biological processes (BPs), and molecular functions (MFs) (26). These genes were analyzed by the KEGG (27) to probe the biological functions and key pathways nearly correlated with the occurrence and development of LSCC (28). Statistical significance was fix at P<0.05.

PPI network establishment and hub genes probing

The PPI network can provide interaction information for various proteins and help identify key genes and important modules implicated in the growth of LSCC (29). The Search Tool for the Retrieval of Interacting Genes (STRING) database (http://www.string-db.org/) was employed for establishing PPI network of candidate genes (30). Genes with minimum required interaction score ≥0.500 were selected to construct a whole network model that was further anticipated in Cytoscape software (http://www.cytoscape.org). Using plugin Cytohubba, the leading 10 genes with maximal clique centrality (MCC) from the candidate genes as network hub genes were chosen and visualized them (31).

Hub genes prognostic value

To validate the hub genes credibility, the interrelationship among overall survival (OS) and hub genes was tested by survival analysis of Kaplan-Meier. Using R package survival and survminer depending on TCGA database clinical data (32). We employed the online tool Gene Expression Profiling Interactive Analysis 2 (GEPIA2; http://gepia2.cancer-pku.cn/) to identify association between disease-free survival (DFS) and hub genes in LSCC patients (33). In this research work, complete follow-up information of patients was checked for analyze further, split as two divisions according to the hub genes median expression. Cut-off criteria genes of P<0.05 were regarded to be hub genes related to survival prognosis.

Protein expressions validation of prognostic-associated hub genes

In addition, immunohistochemistry (IHC) of the Human Protein Atlas (HPA) (https://www.proteinatlas.org/) database was employed to recognize the protein expression levels of genes associated with survivance among LSCC tissues and normal tissue samples (34). The HPA is a noteworthy dataset that gives researchers with transcriptome and proteomic information on a large number of diseased and normal tissue samples (35).

Statistical analysis

All statistical analyses were performed with RStudio software (2021.09.2 Build 382) and Perl (v5.30.3). Differences were statistically significant at P<0.05.

Results

DEGs identification

From the TCGA dataset, LSCC gene expression and clinical data were copied. In total, 123 samples were collected, including 111 cases of LSCC and 12 cases of normal tissue. The HTS dataset GSE127165 was copied from the GEO dataset and 57 LSCC tissue samples and 57 corresponding normal tissue samples were included. Next to data preprocessing and analysis of differential gene expression, 2,774 DEGs were obtained from TCGA database, including 1,404 upregulated DEGs and 1,370 downregulated DEGs (Figure 1A), and 2,139 DEGs, including 1,095 upregulated DEGs and 1,044 downregulated DEGs, were obtained from the GSE127165 dataset (Figure 1B). The top 50 DEGs with the most significant upregulation and downregulation are displayed by heat maps of genes (Figure 1C,1D). Adj. P values <0.05 and |logFC| ≥1 were used as the cut-off criteria.

Figure 1.

DEGs identification among TCGA and GEO databases of LSCC with the cut-off criteria of |logFC| ≥1.0 and adj. P<0.05. Volcano graphs of DEGs in the TCGA (A) and GEO (B) databases. Heat maps of the top 50 DEGs in the TCGA (C) and GEO (D) databases. Upregulated DEGs (red), downregulated DEGs (green). Adj., adjusted; DEG, differentially expressed gene; TCGA, The Cancer Genome Atlas; GEO, Gene Expression Omnibus; LSCC, laryngeal squamous cell carcinoma; FC, fold change.

WGCNA of whole transcriptome gene expression matrix

To identify identical expression trend of genes and analogy of biological functions, weighted gene co-expression network was constructed based on the RNA-sequencing (RNA-seq) count data of 14,556 genes in TCGA-LSCC and 25,522 genes in the GSE127165 database. We recognized 13 modules from the TCGA-LSCC database (Figure 2A) and 15 modules from the GSE127165 dataset (Figure 2B) and module-trait diagrams were plotted to assess the relationship among every module and clinical phenotypes (tumor and normal). As shown in Figure 2C,2D, the MEturquoise module in TCGA-LSCC and the MEred module in GSE127165 had the greatest negative correlation with LSCC tissue, however the MEcyan module in TCGA-LSCC and the MEblack module in GSE127165 had the greatest positive correlation with LSCC tissue.

Figure 2.

Recognition of modules correlated to clinical phenotype among TCGA and GEO databases of LSCC. Gene-module tree diagrams in the TCGA (A) and GEO (B) datasets. Every single branch indicates a gene, and each color below indicates one co-expressive module. Module-trait relationship diagrams in the TCGA (C) and GEO (D) datasets. Every single row correlate to a color module and column interrelated to a clinical trait (tumor or normal). Every single cell includes consistent correlation and P value. TCGA, The Cancer Genome Atlas; GEO, Gene Expression Omnibus; LSCC, laryngeal squamous cell carcinoma.

Gene recognition among WGCNA modules and the DEGs lists

According to analysis of differential expression and WGCNA, there were 2,774 DEGs in the TCGA-LSCC database and 2,139 DEGs in the GSE127165 database, 3,683 and 1,620 co-expressed genes in the MEturquoise module of TCGA-LSCC and the MEred module of GSE127165, respectively, and 249 and 1,409 co-expressed genes in the MEcyan module of TCGA-LSCC and the MEblack module of GSE127165, respectively. Overall 36 intersecting co-expressed genes were retrieved for further analysis (Figure 3A,3B).

Figure 3.

Venn diagrams. (A) Thirty intersecting genes among the overlapping of DEG list in two datasets [TCGA (TCGA_diff) and GEO databases (GEO_diff)] and two co-expression modules with important positive correlations [MEturquoise module of TCGA-LSCC (TCGA_turquoise) and the MEred module of GSE127165 (GEO_red)]. (B) Six intersecting genes among the overlapping of DEG list in the two datasets (TCGA_diff and GEO_diff) and two co-expression modules with the important negative correlations [MEcyan module of TCGA-LSCC (TCGA_cyan) and the MEblack module of GSE127165 (GEO_black)]. Thirty-six intersecting genes in the list of overlapping of DEG and co-expression modules. GEO, Gene Expression Omnibus; TCGA, The Cancer Genome Atlas; DEG, differentially expressed gene.

Thirty-six genes functional and pathway enrichment examination

GO functional and enrichment examination of KEGG pathway were conducted on the 36 intersecting co-expressed genes by the R package cluster profiler. GO functional enrichment analysis (Figure 4A) exhibited that the BP of the 36 genes were mainly concentrated in hemidesmosome assembly, cell junction assembly, and fat cell differentiation. Regarding CC, these genes were determined to be greatly associated to the basement membrane and collagen-containing extracellular matrix (ECM). In addition, in analysis of MF, ECM structural constituents and coenzyme binding were considered to be associated to these 36 genes. Through enrichment examination of KEGG pathway (Figure 4B), we demonstrated that binding of ECM-receptor interaction, focal adhesion, small-cell lung cancer, toxoplasmosis, and HPV infection are nearly associated to these genes.

Figure 4.

GO (A) and KEGG (B) enrichment examination for the 36 co-expression genes. The color indicates P values, and spots size indicates the number of genes. BP, biological process; CC, cell component; MF, molecular function; ECM, extracellular matrix; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Hub genes acquisition by PPI network analysis

Using STRING database, we constructed a PPI network of the 36 overlapping co-expressed genes (Figure 5A). Using plug-in Cytohubba in the software Cytoscape, hub genes were chosen from the PPI network by the MCC algorithm are shown in Figure 5B. The top 10 hub genes in network are SNAI2, ITGA6, LAMB3, LAMC2, CAV1, COL7A1, GJA1, EHF, OAT, and GPT.

Figure 5.

PPI network visualization and top 10 hub genes. (A) PPI network between the co-expression network genes. The blue node indicates the genes. The edges represent interaction associations among nodes. (B) Hub genes identification from the PPI network by MCC algorithm. The edges indicate the protein-protein associations. The red nodes indicate genes having high MCC scores, while the yellow nodes indicate genes having low MCC scores. PPI, protein-protein interaction; MCC, maximal clique centrality.

Survival analysis and hub genes protein expression validation

Through the above analyses, the leading 10 network hub genes (SNAI2, ITGA6, LAMB3, LAMC2, CAV1, COL7A1, GJA1, EHF, OAT, and GPT) were identified. Also by using the R package and GEPIA2 database survival analysis, OS and DFS examination of the 10 hub genes were carried out to investigate their prognosis in patients with LSCC. OS analysis showed that downregulation of OAT expression was remarkably associated with worse patients’ prognosis with LSCC (P<0.05) (Figure 6A), while DFS analysis recommended that high expressive CAV1 was closely related with worsening prognosis (P<0.05) (Figure 6B). Furthermore, depending on the HPA database, the OAT gene protein expression was remarkably reduced in tumor tissues compared with that in normal tissues (Figure 7A), while the CAV1 expression was remarkably increased (Figure 7B).

Figure 6.

The leading 10 hub genes survival analysis of LSCC patients. (A) OS examination for OAT in LSCC by survival of R package. (B) Analysis of DFS for CAV1 in LSCC from the GEPIA2 database. Stratified the patients into high-level group (red) and low-level group (green) based on the median expression of gene. P<0.05 was a statistically remarkable variation. TPM, transcript per million; HR, hazard ratio; LSCC, laryngeal squamous cell carcinoma; OS, overall survival; DFS, disease-free survival; GEPIA2, Gene Expression Profiling Interactive Analysis 2.

Figure 7.

IHC images of the OAT and CAV1 gene in tumor and normal tissues from the HPA dataset. Protein expression level of (A) OAT (https://www.proteinatlas.org/ENSG00000065154-OAT/pathology/head+and+neck+cancer#img; https://www.proteinatlas.org/ENSG00000065154-OAT/tissue/nasopharynx#img) and (B) CAV1 (https://www.proteinatlas.org/ENSG00000105974-CAV1/pathology/head+and+neck+cancer#img; https://www.proteinatlas.org/ENSG00000105974-CAV1/tissue/nasopharynx#img) in head and neck cancer and normal nasopharynx tissues). HPA, Human Protein Atlas; NOS, not otherwise specified; IHC, immunohistochemistry.

Discussion

Laryngeal carcinoma (LC) is a malignant neoplasm that occurs in supraglottic, glottic, and subglottic regions. Squamous cell cancer (SCC) is the prominent type of cancer. At present, LSCC mainly adopts the comprehensive treatment strategy of surgical adjuvant chemoradiotherapy; however, due to the absence of accurate molecular targets, especially for patients with higher clinical stages, the prognosis is often poor. Therefore, recognizing novel biological indicators and therapeutic targets is crucial for improving prognosis and LSCC patient’s survival. In the present study, we employed bioinformatic analysis methods such as WGCNA to recognize 36 co-expressed DEGs in the TCGA-LSCC and GSE127165 databases. The gene functions were mainly exacerbated in cell junction assembly, basement membrane, and ECM structural constituents, and the pathways were highly concentrated in ECM-receptor interaction, focal adhesion, small cell lung cancer and toxoplasmosis. In addition, the top 10 hub genes (SNAI2, ITGA6, LAMB3, LAMC2, CAV1, COL7A1, GJA1, EHF, OAT, and GPT) were identified using the established PPI network. Low expressed OAT and high expressed CAV1 remarkably influenced the LSCC patients’ survival prognosis. Finally, our conclusions were confirmed by OS, DFS, and IHC analyses of OAT and CAV1.

OAT, a mitochondrial enzyme, is found highly in the liver, intestine, kidneys, and brain. An important role is to regulate the signaling molecules and mediators generation. OAT deficiency leads to cyclotron atrophy, an uncommon but serious genetic disorder, then illustrating the significance of this enzyme (36). Also the differential expression of OAT has been shown to be resulted in hepatocellular carcinoma (37), non-small cell lung cancer (38), and retinoblastoma (39). In this research work, the OAT expression level in LSCC tissues was significantly downregulated and nearly associated with survival prognosis. CAV1 is an oncogenic membrane protein involved in endocytosis, metabolic alterations, cell migration, and signal transduction (40). One study found that the overexpression and secretion of CAV1 is related to the occurrence and progression of prostate cancer (41), and is closely linked with the migration, apoptosis, proliferation and drug impedance of lung cancer cells, and can be labeled as a prognostic factor for lung cancer (42). In addition, experimental data have shown that CAV1 can induce the expression of fucosyl transferase Pofut1 and enhance the metastasis and infiltration of mouse liver cancer cells in vitro and in vivo by activating the Notch and MAPK signaling pathways (43). The results indicate that WGCNA and other comprehensive bioinformatic analyses can reveal novel pathogenic genes and supply rich reference resources for further experimental verification.

Conclusions

In conclusion, through comprehensive bioinformatics analysis, we identified hub genes and key modules that are closely associated with LSCC incidence and growth, and verified these genes by survival analysis and protein expression. Our study is of high importance to uncover the pathogenesis of LSCC and probe for new precise biological markers and potential therapeutic targets. However, studies on different subtypes and stages of LSCC are limited. Many experiments are required to further validate the genes associated with pathogenesis and prognosis of LSCC.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki (as revised in 2013).