Skip to main content
NCBI home page
Cancer Cell International logoLink to Cancer Cell International
. 2025 Mar 18;25:102. doi: 10.1186/s12935-025-03739-5

A combined gene signature model for predicting radiotherapy response and relapse-free survival in laryngeal squamous cell carcinoma

Shiqi Gong 1,✉,#, Liyun Yang 2,#, Meng Xu 3,#, Mingliang Xiang 1, Juntian Lang 1, Hao Zhang 1,✉, Yamin Shan 1,✉
PMCID: PMC11916850  PMID: 40102978

Abstract

Background

Radioresistance is a major challenge in radiotherapy for laryngeal squamous cell carcinoma (LSCC), and there is currently no effective method to predict radiosensitivity in LSCC patients. This study aimed to establish a prediction model for radiotherapy response based on gene expression.

Methods

The datasets of LSCC were obtained from the ENT department of Shanghai Ruijin Hospital and The Cancer Genome Atlas (TCGA). Lasso regression and Cox regression were used to establish the prediction model based on gene expression. Weighted gene coexpression network analysis (WGCNA) was used to analyze the correlation between gene expression and clinical characteristics. RT-qPCR was used to detect gene expression in tumor tissue to verify the accuracy of the prediction model.

Results

Using a cohort of LSCC cases receiving radiotherapy collected in the TCGA database, the 3 protein-coding genes (PCGs) signature model was identified for the first time as the predictor of relapse-free survival and radiosensitivity in LSCC patients. And we explored the potential clinical value of 3 PCGs and screened out 2 long non-coding RNAs (lncRNAs) potential associated with 3 PCGs. More importantly, the LSCC cases collected by our department were used to preliminarily verify the predictive power of the 3 PCGs signature model for the radiosensitivity of LSCC, and the significant correlation between the expression levels of the 3 PCGs and the 2 lncRNAs.

Conclusion

We successfully establish a radiosensitivity prediction model based on the 3 PCGs Riskscore, which provides a theoretical basis for the decision-making of LSCC treatment options. Meantime, we preliminarily screen the potential associated lncRNAs of the 3 PCGs for further basic and clinical research.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12935-025-03739-5.

Keywords: Laryngeal squamous cell carcinoma, Radiation sensitivity prediction model, Lasso regression, Cox regression, WGCNA

Introduction

Laryngeal squamous cell carcinoma (LSCC) is one of the most common head and neck carcinoma originating from the laryngeal mucosal epithelium [1]. The onset of LSCC is occult, and approximately 60% of patients are at the advanced stages (clinical stages III and IV) when they are initially diagnosed [2]. Though the rapid development of comprehensive treatment approaches for LSCC, unfortunately, its 5-year survival rate has declined from 66 to 63% over the past 40 years [3]. The patient's risk of recurrence is highest about 2 years after surgery, due to some adverse factors that significantly reduce the efficacy of long-term treatments such as radiotherapy and so on [4]. The prognosis of patients with recurrent or metastatic LSCC is poor [5]. However, there is no effective method to predict radiotherapy sensitivity in patients with LSCC.

Radiotherapy is a significant component of the comprehensive treatment of malignant tumors, which is widely used in head and neck cancer [6], breast cancer [7], lung cancer [8], esophageal cancer [9], rectal cancer [10], etc. According to statistics, 65–75% of cancer patients have received radiotherapy during the treatment process. With the rapid development of three-dimensional conformal, intensity-modulated, and image-guided radiotherapy, the efficacy of radiotherapy continues to improve [11]. However, radioresistance remains a major challenge in the treatment [12]. Due to the molecular heterogeneity of malignant tumors, they have different responses to radiotherapy [13], such as p53 and p16 overexpression enhancing tumor cells' sensitivity to radiotherapy [14, 15].

Radiotherapy is one of the important measures of LSCC comprehensive treatment. It is widely used in the radical treatment of early-stage (T1 and T2) LSCC [16], intensive treatment for the location and proximity to critical structures encompassing the resected disease site and “at-risk” areas after surgery [17], salvage treatment for recurrence after first-line surgery [18], and palliative treatment for advanced LSCC [19], etc. Radiotherapy and surgical operation are different treatment options for patients with early-stage LSCC and recurrence after first-line surgery [20, 21]. Whether the patient is sensitive to radiotherapy is one of the key factors affecting the decision of treatment option. Therefore, there is an urgent need to establish the predictive model for radiotherapy sensitivity of LSCC to provide patients with more effective and personalized comprehensive treatments.

In this study, we identified for the first time the 3 protein-coding genes (PCGs) signature model as the predictor of relapse-free survival (RFS) and radiosensitivity in LSCC patients by Cox regression and Lasso regression models using a cohort of LSCC cases receiving radiotherapy collected in The Cancer Genome Atlas (TCGA) database. We explored the potential clinical value and screened the potential associated long non-coding RNAs (lncRNAs) of the 3 PCGs using weighted gene co-expression network analysis [22] (WGCNA), etc. More importantly, we used the 20 LSCC cases collected by our department to preliminarily verify the predictive power of the 3 PCGs signature model for the radiosensitivity of LSCC, and the significant correlation between the expression levels of the 3 PCGs and the 2 lncRNAs. We successfully establish a radiosensitivity prediction model based on the 3 PCGs Riskscore, which provides a theoretical basis for the decision-making of LSCC treatment options. In the meantime, we preliminarily screen the potential associated lncRNAs of the 3 PCGs for further basic and clinical research.

Materials and methods

Dataset construction

117 LSCC cases were collected from TCGA (https://cancergenome.nih.gov/), which included 79 patients who received radiotherapy. 33 cases with incomplete transcriptome sequencing data or clinical data were excluded. Finally, 46 LSCC cases were selected for our study, whose clinical characteristics were shown in Supplementary Table 1A. After the standardization process, each sample contains 1583 lncRNAs and 16882 PCGs expression data. The patients who had the complete response to radiotherapy as well as no recurrence and metastasis were defined as the complete response (CR) group (n = 26). And the patients who had the partial response or progressive disease to radiotherapy and/or recurrence and/or metastasis were defined as the non-complete response (non-CR) group (n = 20).

Meanwhile, we collected the tumor tissues, clinical and follow-up characteristics of 20 LSCC patients who received radiotherapy in the ENT department of Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medicine. According to the same grouping criteria, 20 LSCC patients were divided into the CR group (n = 12) and the non-CR group (n = 8), whose clinical characteristics were shown in Supplementary Table 1B.

Differential expression analysis

The R software (version 3.6.2) edgeR package was used to identify differentially expressed lncRNAs and PCGs between the CR group and the non-CR group. | log twofold change (FC) |> 1 and false discovery rate (FDR) < 0.05 were set as the threshold.

Establishing the predictive model

The R software survival package was used for univariate and multivariate Cox regression analysis, which was used to construct the model and analyze whether the model was independent of clinical characteristics. And the glmnet package was used for Lasso regression analysis to shrink and select variables to finally obtain genes with independence for establishing model.

Kaplan–Meier survival analysis and model evaluation

The Kaplan–Meier survival analysis (log-rank test) was used to determine the difference in RFS between the two groups. The sensitivity and specificity of the predictive model were assessed by calculating the area under curve (AUC) of receiver operating characteristic (ROC) curve using the R software survivalROC package. The principal component analysis (PCA) was used to assess the discriminative power of the predictive model for the observed variables.

Weighted gene co-expression network analysis

The expression profile of 1583 lncRNAs was used to construct a gene co-expression network using the WGCNA package in R software. An adjacency matrix was constructed using the WGCNA function adjacency by calculating the Pearson correlation between all pairs of lncRNAs in all selected samples. In this study, the power of β = 5 was used as a soft threshold parameter to ensure a scale-free network. To further identify functional modules in the co-expression network with 1583 lncRNAs, the adjacency matrix was used to calculate the topological overlap measurement (TOM) representing the overlap in the shared neighbors. The module eigengenes (MEs) were a representation of the gene expression profile in the module. The correlation and P value between the module and clinical characteristics were evaluated by calculating the MEs.

Reverse transcription-quantitative polymerase chain reaction (RT-qPCR)

The TRIzol reagent was used to isolate total RNA from tumor tissue. The HiScript III RT SuperMix for qPCR Kit (Vazyme Biotech, Nanjing, China) was used to synthesize cDNA. The ChamQ Universal SYBR qPCR Master Mix (Vazyme Biotech, Nanjing, China) and ABI7500 system (Applied Biosystems, Foster City, CA, United States) were used to analyze cDNA. The amplification program was as follows: initial denaturation step at 95 ℃ for 30 s, followed by 40 cycles at 95 ℃ for 5 s, and 60 ℃ for 30 s. The standard curve was drawn to analyze the normalized expression of target PCGs and lncRNAs. The primer sequences were shown in Supplementary Table 2.

Statistical analysis

The Chi-square test was performed with SPSS (version 25.0). The Pearson correlation analysis was performed using R software to evaluate the correlation between variables. The results of RT-qPCR were analyzed by Graphpad Prism (version 8.3.1), and differences between groups were assessed using independent sample t-test. P < 0.05 was considered statistically significant.

Results

Differentially expressed lncRNA and PCG between CR and non-CR groups

According to | log 2 FC |> 1 and FDR < 0.05, compared with the CR group, we found that 12 lncRNAs and 99 PCGs were upregulated, 25 lncRNAs and 166 PCGs were downregulated in the non-CR group, which were shown in Supplementary Table 3. p53 and p16 showed a trend of downregulation in the non-CR group (Supplementary Table 1C).

Establishing and evaluating a predictive model based on 3 PCGs

We first used univariate Cox regression analysis to assess the association between the expression levels of 302 differentially expressed lncRNAs as well as PCGs and patient RFS. According to the criteria of P < 0.1, we found that 4 lncRNAs and 20 PCGs were significantly associated with patient RFS. After Lasso regression analysis, among 4 lncRNAs and 20 PCGs, we deleted those with high correlation or subordination candidates, and 4 PCGs with high independence remained, i.e., TENM2, KLHL4, BASP1, and ZNF665 (Fig. 1A, B). Then stepwise multivariate Cox regression analysis was performed to establish the predictive model, which was defined as a linear combination of the expression levels of 3 PCGs (Table 1) whose relative coefficient weights were as follows: Riskscore = (6.99E-05 × expression value of TENM2) + (0.0015 × expression value of KLHL4)—(0.0109 × expression value of ZNF665).

Fig. 1.

Fig. 1

Open in a new tab

A, B Deleting genes with high correlation or subordination candidates by Lasso regression analysis. C The distribution of Riskscore. D The survival duration and status of patients. E A heatmap of the expression levels of 3 PCGs

Table 1.

Multivariate Cox regression to establish a prognostic model based on 3 PCGs Riskscore

PCG Multivariate Cox regression analysis Differential expression analysis
Coefficient HR P value logFC P value FDR
TENM2 6.99E-05 1.001 0.017 1.359 1.22E-04 0.016
KLHL4 0.002 1.002 0.088 2.500 3.36E-05 0.006
ZNF665 − 0.011 0.989 0.096 − 1.611 0.001 0.037
Open in a new tab

The differential expression analysis of 3 PCGs (non-CR group compared with CR group)

HR hazard ratio, FC fold change, FDR false discovery rate

For each of the 46 LSCC cases in our study, we calculated the Riskscore based on 3 PCGs expression and classify patients into low-risk (n = 23) and high-risk (n = 23) groups according to the median Riskscore of 0.9168 as the cut-off value (Fig. 1C–E). The Kaplan–Meier curve showed that the RFS of the low-risk group was significantly higher than that of the high-risk group. (Fig. 2A, P = 0.029). The predicting ability of the 3 PCGs signature model was evaluated by calculating the AUC of the ROC curve. The ROC curve of predicting 12-month RFS obtained the AUC of 0.808 (Fig. 2B), and the ROC curve of predicting 24-month RFS obtained the AUC of 0.772 (Fig. 2C), showing good sensitivity and specificity of the 3 PCGs signature model in predicting the RFS of LSCC patients. We evaluated the prognostic value of the clinicopathological characteristics and 3 PCGs signature model by univariate and multivariate Cox regression analysis. We found that high Riskscore (HR = 2.831) was the independent risk factor for RFS of LSCC patients (Table 2).

Fig. 2.

Fig. 2

Open in a new tab

A The RFS of low-risk patients is significantly higher than high-risk patients. B, C The ROC curve of the model predicting 12-month and 24-month RFS of patients. D The radiotherapy response of patients with increasing Riskscore. E The principal component analysis. F The ROC curve of the model predicting radiotherapy response of patients

Table 2.

The prognostic value of clinicopathologic characteristics and 3 PCGs Riskscore in patients with LSCC

Variables Univariate analysis Multivariate analysis
HR (95% CI) P value HR (95% CI) P value
Age at initial diagnosis (> = 60/ < 60) 0.967 (0.424, 2.208) 0.937
Gender (Female/Male) 2.282 (0.663, 7.851) 0.191 2.401 (0.623, 9.252) 0.203
Histologic grade (G3/G1 + G2) 0.890 (0.350, 2.266) 0.807
Clinical stage (IV/II + III) 0.605 (0.192, 1.903) 0.390
T stage (T4/T1 + T2 + T3) 1.667 (0.682, 4.073) 0.262
N stage (N1 + N2/N0) 0.815 (0.300, 2.214) 0.688
M stage (Mx/M0) 1.165 (0.224, 6.064) 0.856
Smoking history (Yes/No) 0.545 (0.237, 1.253) 0.153 0.483 (0.209, 1.121) 0.090
3 PCGs Riskscore (High/Low) 2.605 (1.067, 6.363) 0.036 2.831 (1.150, 6.968) 0.024
Open in a new tab

The P value indicating statistical significance is marked with bold type

HR hazard ratio, CI confidence interval

Predicting radiotherapy response using the 3 PCGs signature model

In the high-risk group, 15 cases responded to radiotherapy with non-CR and 8 cases with CR. In the low-risk group, 5 cases responded to radiotherapy with non-CR and 18 cases with CR (Fig. 2D). The chi-square test showed that the risk group of patients was significantly associated with radiotherapy response (P = 0.003). The PCA showed that the Riskscore could effectively separate patients into CR group and non-CR group (Fig. 2E). The ROC curve of predicting patient response to radiotherapy obtained the AUC of 0.840 (Fig. 2F). These results showed that the 3 PCGs signature model had excellent predictive power for radiotherapy response in patients with LSCC.

Identifying the clinical significance of 3 PCGs

To explore the clinical significance of 3 PCGs respectively, the chi-square test was used to analyze the association between the expression levels of the 3 PCGs and the clinical characteristics of patients (Supplementary Table 4). We found that TENM2 (P = 0.017), KLHL4 (P = 0.003) and ZNF665 (P = 0.017) were all associated with radiotherapy response of patients. In addition, KLHL4 (P = 0.018) was associated with patient age at initial diagnosis, and ZNF665 (P = 0.008) was associated with patient smoking history. These results showed that 3 PCGs might jointly play important roles in the clinical characteristics, including radiotherapy response, of patients.

Weighted gene co-expression network construction and module identification

To further explore the lncRNAs potential associated with the 3 PCGs, we performed WGCNA analysis. The average linkage method and Pearson correlation method were used to cluster 46 LSCC cases receiving radiotherapy, 2 outlier cases were excluded, and finally 44 cases were recruited for subsequent analysis (Fig. 3A, B). The network topology analysis was used to assess various soft-thresholding powers for obtaining relatively balanced scale independence and average connectivity of WGCNA. In this study, we selected the power of β = 5 (scale-free R2 = 0.85) as the soft-thresholding parameter to ensure a scale-free network (Fig. 3C–E). Through dynamic tree cut and merged dynamics, 1583 lncRNAs generated 14 different modules in a hierarchical clustering tree (Fig. 4A). The dissimilarity measure among 14 lncRNA modules was all greater than 50% (Fig. 4B). Then, we analyzed the correlation between 14 lncRNA modules and the age at initial diagnosis, smoking history as well as radiotherapy response of patient, and found that the brown module including 129 lncRNAs was associated with the radiotherapy response of patient (Fig. 4C). The list of lncRNAs for clinically significant modules was shown in Supplementary Table 5.

Fig. 3.

Fig. 3

Open in a new tab

A Clustering dendrogram of 46 LSCC cases and excluding 2 outlier cases. B Clustering dendrogram of 44 LSCC cases corresponding to clinical characteristics. C Analysis of the scale-free fit index for various soft-thresholding powers. D Analysis of the mean connectivity for various soft-thresholding powers. E Checking the scale-free topology when β = 5

Fig. 4.

Fig. 4

Open in a new tab

A The dendrogram of lncRNAs are clustered based on the dissimilarity measure (1-TOM). B The dissimilarity measure among 14 lncRNA modules is all greater than 50%. C The heatmap of the correlation between module eigengenes (MEs) and clinical characteristics of patient

Screening the potential associated lncRNAs of 3 PCGs

We calculated the Pearson correlation coefficients between the 3 PCGs and the 129 lncRNAs of brown module to determine the co-expression relationship, respectively. According to the criteria of P < 0.001, 7 lncRNAs were screened out as potential associated lncRNAs for TENM2, 6 lncRNAs were screened out as potential associated lncRNAs for KLHL4, and 4 lncRNAs were screened out as potential associated lncRNAs for ZNF665. Totally 14 different lncRNAs were involved (Table 3). To further screen for key lncRNAs, the chi-square test was used to analyze the correlation between the 14 lncRNAs and radiotherapy response of patient. We found that ELF3-AS1 (P = 0.017), PARD3-AS1 (P = 0.003), GABPB1-AS1 (P = 0.017) and ZNF790-AS1 (P = 0.017) were associated with radiotherapy response of patient, among which ELF3-AS1 and PARD3-AS1 had co-expression relationship with TENM2 and KLHL4, GABPB1-AS1 and ZNF790-AS1 had co-expression relationship with ZNF665 (Table 3). Combined with the differential expression analysis, we found that PARD3-AS1 (P = 0.011, FDR = 0.229) and ZNF790-AS1 (P = 0.016, FDR = 0.294) had a down-regulation trend in the non-CR group, and GABPB1-AS1 (P = 0.001, FDR = 0.049) was significantly downregulated in the non-CR group, compared with CR group (Table 3). The above results indicated that PARD3-AS1 was the potential associated lncRNA of TENM2 and KLHL4, GABPB1-AS1 and ZNF790-AS1 were the potential associated lncRNAs of ZNF665.

Table 3.

The potential associated lncRNAs of 3 PCGs, including the Pearson correlation analysis between 3 PCGs and 129 lncRNAs, the chi-square test between the 14 lncRNAs and radiotherapy response of patients, and the differential expression analysis of 14 lncRNAs (non-CR group compared with CR group)

PCG Pearson correlation analysis Chi-square test Differential expression analysis
LncRNA Cor P value P value logFC P value FDR
TENM2 ELF3-AS1 − 0.543 9.80E-05 0.017 − 0.520 0.143 0.684
TENM2 LINC01964 − 0.522 1.98E-04 0.234 − 1.258 0.107 0.632
TENM2 C9orf147 − 0.514 2.62E-04 0.234 − 0.440 0.171 0.720
TENM2 DUBR 0.511 2.89E-04 0.552 0.101 0.734 0.957
TENM2 MSC-AS1 0.505 3.43E-04 0.234 0.715 0.065 0.534
TENM2 LINC02308 − 0.501 3.88E-04 0.234 − 2.720 0.001 0.066
TENM2 PARD3-AS1 − 0.479 7.59E-04 0.003 − 1.106 0.011 0.229
KLHL4 MSC-AS1 0.527 1.70E-04 0.234 0.715 0.065 0.534
KLHL4 PARD3-AS1 − 0.521 2.06E-04 0.003 − 1.106 0.011 0.229
KLHL4 LINC01133 − 0.517 2.35E-04 0.074 − 1.152 0.029 0.388
KLHL4 ELF3-AS1 − 0.507 3.21E-04 0.017 − 0.520 0.143 0.684
KLHL4 LINC01341 − 0.477 8.04E-04 0.074 − 1.658 0.004 0.131
KLHL4 ZKSCAN2-DT − 0.471 9.52E-04 0.552 − 0.344 0.276 0.809
ZNF665 ZNF582-AS1 0.717 2.14E-08 0.552 − 0.596 0.133 0.670
ZNF665 ZNF790-AS1 0.667 4.17E-07 0.017 − 1.221 0.016 0.294
ZNF665 ZNF571-AS1 0.662 5.48E-07 0.234 − 1.342 0.001 0.061
ZNF665 GABPB1-AS1 0.507 3.23E-04 0.017 − 0.878 0.001 0.049
Open in a new tab

The P value indicating statistical significance is marked with bold type

PCG protein-coding gene, Cor correlation coefficient, FC fold change, FDR false discovery rate

Survival analysis of 3 PCGs and their potential associated lncRNAs

The Kaplan–Meier survival curve (log-rank test) was used to evaluate the effect of 3 PCGs and their potential associated lncRNAs on RFS of patients. We found that patients with high expression of TENM2 (Fig. 5A, P = 0.014) and KLHL4 (Fig. 5B, P = 0.043) had shorter RFS. Compared with low expression of ZNF665, the RFS of patients with high expression of ZNF665 tended to increase (Fig. 5C, P = 0.058). Meanwhile, regarding these potential associated lncRNAs, the patients with low expression of PARD3-AS1 (Fig. 5D, P = 0.026) and GABPB1-AS1 (Fig. 5E, P = 0.033) had shorter RFS. The effect of ZNF790-AS1 expression on RFS of patients had not been found (Fig. 5F, P = 0.232). These results suggested that PARD3-AS1 was a key associated lncRNA for TENM2 and KLHL4, and GABPB1-AS1 was a key associated lncRNA for ZNF665.

Fig. 5.

Fig. 5

Open in a new tab

The Kaplan–Meier survival curve of 3 PCGs and 3 lncRNAs

Validation of the predictive ability of the 3 PCGs signature model

We determined the Riskscore based on the measured with RT-qPCR using the 3 PCGs signature model in 20 LSCC patients collected by our department. According to the median Riskscore of 0.3791, the 20 patients were divided into the low-risk group (n = 10) and the high-risk group (n = 10). The Fisher's Exact Test showed differences in radiotherapy response of patients between high-risk and low-risk groups (P = 0.020). There was only 1 patient with non-CR in the low-risk group, and 7 patients with non-CR were included in the high-risk group (Fig. 6A). Therefore, the 3 PCGs signature model could effectively pre-screen patients with non-CR to radiotherapy (sensitivity, 87.5%; specificity, 75.0%; positive predictive value, 70.0%; negative predictive value, 90.0%; and overall accuracy, 80.0%).

Fig. 6.

Fig. 6

Open in a new tab

A The expression of 3 PCGs for each LSCC tumor tissue collected by our department (The expression of TENM2 is 100-fold reduced for display). The first 10 cases are in the low-risk group, and the last 10 cases are in the high-risk group. The black case numbers represent the CR group, and the red case numbers represent the non-CR group. B–F The differential expression of TENM2, KLHL4, ZNF665, PARD3-AS1, and GABPB1-AS1 between the CR group and the non-CR group (**P < 0.01; The expression of TENM2 is 100-fold reduced for display). G–I The scatter plot showing that TENM2 and KLHL4 are negatively correlated with PARD3-AS1, and ZNF665 is positively correlated with GABPB1-AS1

In the non-CR group, TENM2 and KLHL4 were significantly up-regulated, while ZNF665, PARD3-AS1and GABPB1-AS1 were significantly down-regulated (Fig. 6B–F), which was consistent with the results of differential expression analysis of the TCGA dataset. Meantime, we observed that TENM2 and KLHL4 were negatively correlated with PARD3-AS1, and ZNF665 was positively correlated with GABPB1-AS1 (Fig. 6G–I), which suggested that PARD3-AS1 was the potential associated lncRNA with TENM2 and KLHL4, and GABPB1-AS1 was the potential associated lncRNA with ZNF665.

Discussion

Radiotherapy is one of the essential measures of LSCC comprehensive treatment. It is widely used in the radical treatment of early-stage (T1 and T2) LSCC, intensive treatment for the location and proximity to critical structures encompassing the resected disease site and “at-risk” areas after surgery, salvage treatment for recurrence after first-line surgery, and palliative treatment for advanced LSCC, etc. However, radioresistance is still a significant challenge in the treatment. Whether the patient is sensitive to radiotherapy is one of the key factors affecting the decision of treatment options. Therefore, there is an urgent need to establish the predictive model for radiotherapy sensitivity of LSCC to provide patients with more effective and personalized comprehensive treatments.

Radiation kills cancer cells by inducing DNA double-strand breaks. Therefore, radiation resistance is enhanced by efficient repair of damaged DNA. Head and neck cancers overexpress EGFR and have a high frequency of p53 mutations, both of which enhance DNA repair [23]. p53 and p16 overexpression enhance tumor cells' sensitivity to radiotherapy [14, 15]. In 46 LSCC cases from TCGA database, p53 and p16 showed a trend of downregulation in the non-CR group, and the differential expression analysis showed P < 0.05, but FDR > 0.05. This may be due to the limitation of sample size and the complex interaction between tumor moleculars.

In this study, we identified for the first time the 3 PCGs, i.e., TENM2, KLHL4, and ZNF665, signature model as the predictor of RFS and radiosensitivity in LSCC patients by Cox regression and Lasso regression models using a cohort of LSCC cases receiving radiotherapy collected in The Cancer Genome Atlas (TCGA) database. We used risk survival curve, ROC curve, principal component analysis, etc. to verify the predictive power of the 3 PCGs signature model for RFS and radiosensitivity in LSCC patients. More importantly, we preliminarily validated the predictive power of the 3 PCGs signature model for radiosensitivity in LSCC patients in an LSCC cohort collected in our department. The results showed that the 3 PCGs signature model had high predictive power for RFS and radiotherapy response in patients with LSCC, and the prediction results were highly consistent. High-risk LSCC patients predicted by the 3 PCGs signature model exhibited radioresistance while having shorter RFS.

Previous studies have shown that the TENM2 mutant type promotes the effectiveness of immunotherapy for cervical cancer (CC) patients, which may contribute to the relatively better survival of the low-PTGS2 group [24]. This suggests that TENM2 is a risk factor for CC. The study of crizotinib-induced hepatotoxicity in ALK-positive non-small cell lung cancer patients found that TENM2 was significantly enriched in crizotinib-resistant liver-derived cell lines [25]. Tumor hypoxia has long been considered a detrimental factor in response to irradiation [26]. The mitochondria are the main site of cellular respiration. Knockdown of TENM2 in preadipocytes led to increased brown adipocyte marker expression levels upon differentiation resulting in enhanced mitochondrial respiration [27]. This suggests that the high expression of TENM2 may inhibit mitochondrial respiration, causing hypoxia in tumor cells, which in turn leads to radioresistance. Previous studies have found that KLHL4 is a prognostic risk factor for lung adenocarcinoma [28]. In contrast, KLHL4 is found to be a novel p53 target gene, inhibits cell proliferation by activating p21 WAF/CDKN1A [29]. In our study, TENM2 and KLHL4 are risk factors for RFS and radiotherapy response in LSCC. They might inhibit cellular respiration and celluar proliferation, leading to insensitivity to radiotherapy.

Previous studies have demonstrated the high level of CpG island methylator phenotype (CIMP) in rectal cancer is significantly associated with increased risk of extramural vascular invasion (EMVI). And EMVI is significantly associated with adverse survival in rectal cancer [30]. Meanwhile, in colorectal cancer, ZNF665 is highly expressed in patients with low levels of CIMP [31]. This suggests that ZNF665 may play a tumor suppressor role in rectal cancer. In our study, ZNF665 is a protective factor for RFS and radiotherapy response in LSCC.

Furthermore, we identified the clinical significance of 3 PCGs. Using WGCNA, Pearson correlation analysis, and survival analysis, we found that PARD3-AS1 was a key associated lncRNA for TENM2 and KLHL4, and GABPB1-AS1 was a key associated lncRNA for ZNF665.

Previous studies have shown that the chromosome segment 10p11.21–11.22 is lost in head and neck squamous cell carcinoma (HNSCC), and the genes of interest in or near peak region are PARD3 and PARD3-AS1 [32]. This suggests that PARD3-AS1 may have a biological function in HNSCC. In our study, PARD3-AS1 is negatively correlated with TENM2 and KLHL4. LSCC patients with high PARD3-AS1 expression have better RFS.

Previous studies have found that erastin upregulated the lncRNA GABPB1-AS1, which downregulated GABPB1 protein levels by blocking GABPB1 translation, leading to the downregulation of the gene encoding Peroxiredoxin-5 (PRDX5) peroxidase and the eventual suppression of the cellular antioxidant capacity in hepatocellular carcinoma. Such effects critically inhibited the cell viability [33]. Lower GABPB1-AS1 expression was found in clear cell renal cell carcinoma. GABPB1-AS1 expression was inversely associated with tumor size, TNM stage, and Furhman stage. High GABPB1-AS1 expression was associated with a better prognosis. GABPB1-AS1 overexpression significantly inhibited proliferation, migration, and invasion by 786-o and caki-1 cells. GABPB1-AS1 overexpression reduced tumor weights in xenograft experiments [34]. This suggests that GABPB1-AS1 is a tumor suppressor. Conversely, other studies have found that long non-coding RNA GABPB1-AS1 augments malignancy of glioma cells by sequestering microRNA-330 and reinforcing the ZNF367/cell cycle signaling pathway [35]. GABPB1-AS1 Promotes the Development of Osteosarcoma by Targeting SP1 and Activating the Wnt/ β-Catenin Pathway [36]. This suggests that GAPPB1-AS1 is an oncogenic factor. In our study, GAPPB1-AS1 is positively correlated with ZNF665. LSCC patients with high GAPPB1-AS1 expression have better RFS.

We successfully establish a radiosensitivity prediction model based on the 3 PCGs Riskscore, which provides a theoretical basis for the decision-making of LSCC treatment options. Meantime, we preliminarily screen the potential associated lncRNAs of the 3 PCGs for further basic and clinical research. However, further prospective clinical studies are still needed to verify the predictive power of this model, and further basic studies are still needed to explore the potential biological mechanisms.

Supplementary Information

Additional file 1 (22.1KB, docx)
Additional file 2 (15.6KB, docx)
Additional file 3 (27.9KB, xlsx)
Additional file 4 (21.4KB, docx)
Additional file 5 (19.4KB, xlsx)

Acknowledgements

Thanks to Xinjun Meng and other otolaryngologists in Ruijin Hospital for their kindly providing the LSCC specimens and financial support.

Author contributions

Shiqi Gong is responsible for research design, data collection, bioinformatic analysis, and manuscript writing. Liyun Yang is responsible for data organization, and RT-qPCR assay. Meng Xu is responsible for statistical analysis. Mingliang Xiang and Juntian Lang are responsible for providing the LSCC specimens. Hao Zhang and Yamin Shan guide research ideas, design, research methods, and manuscript revision. All authors contributed to the article and approved the submitted version.

Funding

Clinical Science and Technology Innovation Project of Shengkang Hospital Development Center, Grant/Award Number: SHDC12015144. Jiading District Natural Science Research Project, Grant/Award Number: JDKW-2020–0018.

Availability of data and materials

The datasets generated for this study can be found in TCGA, https://portal.gdc.cancer.gov/.

Declarations

Ethics approval and consent to participate

This study was approved by the Institutional Research Ethics Committees of the Ruijin Hospital of Shanghai Jiaotong University School of Medicine. We confirmed that all methods were carried out in accordance with relevant regulations and written informed consent was obtained from patients.

Competing interests

The authors declare no competing interests.

Footnotes

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Shiqi Gong, Liyun Yang and Meng Xu have contributed equally.

Contributor Information

Shiqi Gong, Email: gongshiqi@sjtu.edu.cn.

Hao Zhang, Email: zhanghaoent@163.com.

Yamin Shan, Email: microegg@163.com.

References

  • 1.Siegel RL, Miller KD, Fuchs HE, Jemal A. Cancer statistics, 2022. CA Cancer J Clin. 2022;72(1):7–33. 10.3322/caac.21708. [DOI] [PubMed] [Google Scholar]
  • 2.Wu Y, Zhang Y, Zheng X, Dai F, Lu Y, Dai L, et al. Circular RNA circCORO1C promotes laryngeal squamous cell carcinoma progression by modulating the let-7c-5p/PBX3 axis. Mol Cancer. 2020;19(1):99. 10.1186/s12943-020-01215-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Steuer CE, El-Deiry M, Parks JR, Higgins KA, Saba NF. An update on larynx cancer. CA Cancer J Clin. 2017;67(1):31–50. 10.3322/caac.21386. [DOI] [PubMed] [Google Scholar]
  • 4.Gao W, Zhang Y, Luo H, Niu M, Zheng X, Hu W, et al. Targeting SKA3 suppresses the proliferation and chemoresistance of laryngeal squamous cell carcinoma via impairing PLK1-AKT axis-mediated glycolysis. Cell Death Dis. 2020;11(10):919. 10.1038/s41419-020-03104-6. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Lau A, Yang WF, Li KY, Su YX. Systemic therapy in recurrent or metastatic head and neck squamous cell carcinoma- a systematic review and meta-analysis. Crit Rev Oncol Hematol. 2020;153:102984. 10.1016/j.critrevonc.2020.102984. [DOI] [PubMed] [Google Scholar]
  • 6.Alterio D, Marvaso G, Ferrari A, Volpe S, Orecchia R, Jereczek-Fossa BA. Modern radiotherapy for head and neck cancer. Semin Oncol. 2019;46(3):233–45. 10.1053/j.seminoncol.2019.07.002. [DOI] [PubMed] [Google Scholar]
  • 7.Haussmann J, Corradini S, Nestle-Kraemling C, Bölke E, Njanang FJD, Tamaskovics B, et al. Recent advances in radiotherapy of breast cancer. Radiat Oncol. 2020;15(1):71. 10.1186/s13014-020-01501-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Vinod SK, Hau E. Radiotherapy treatment for lung cancer: current status and future directions. Respirology. 2020;25(Suppl 2):61–71. 10.1111/resp.13870. [DOI] [PubMed] [Google Scholar]
  • 9.Wang SX, Marshall MB. Chemoradiation therapy as definitive treatment of esophageal cancer. Surg Clin North Am. 2021;101(3):443–51. 10.1016/j.suc.2021.03.006. [DOI] [PubMed] [Google Scholar]
  • 10.Jin F, Luo H, Zhou J, Wu Y, Sun H, Liu H, et al. Dose-time fractionation schedules of preoperative radiotherapy and timing to surgery for rectal cancer. Ther Adv Med Oncol. 2020;12:1758835920907537. 10.1177/1758835920907537. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Gao L, Zheng H, Cai Q, Wei L. Autophagy and tumour radiotherapy. Adv Exp Med Biol. 2020;1207:375–87. 10.1007/978-981-15-4272-5_25. [DOI] [PubMed] [Google Scholar]
  • 12.Huang RX, Zhou PK. DNA damage response signaling pathways and targets for radiotherapy sensitization in cancer. Signal Transduct Target Ther. 2020;5(1):60. 10.1038/s41392-020-0150-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Cramer JD, Burtness B, Le QT, Ferris RL. The changing therapeutic landscape of head and neck cancer. Nat Rev Clin Oncol. 2019;16(11):669–83. 10.1038/s41571-019-0227-z. [DOI] [PubMed] [Google Scholar]
  • 14.Lu C, El-Deiry WS. Targeting p53 for enhanced radio- and chemo-sensitivity. Apoptosis. 2009;14(4):597–606. 10.1007/s10495-009-0330-1. [DOI] [PubMed] [Google Scholar]
  • 15.Matsumura Y, Yamagishi N, Miyakoshi J, Imamura S, Takebe H. Increase in radiation sensitivity of human malignant melanoma cells by expression of wild-type p16 gene. Cancer Lett. 1997;115(1):91–6. 10.1016/s0304-3835(97)04714-9. [DOI] [PubMed] [Google Scholar]
  • 16.Gupta A, Wong KH, Newbold K, Bhide S, Nutting C, Harrington KJ. Early-stage glottic squamous cell carcinoma in the era of image-guided radiotherapy. Front Oncol. 2021;11:753908. 10.3389/fonc.2021.753908. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Porceddu SV, Daniels C, Yom SS, Liu H, Waldron J, Gregoire V, et al. Head and neck cancer international group (HNCIG) consensus guidelines for the delivery of postoperative radiation therapy in complex cutaneous squamous cell carcinoma of the head and neck (cSCCHN). Int J Radiat Oncol Biol Phys. 2020;107(4):641–51. 10.1016/j.ijrobp.2020.03.024. [DOI] [PubMed] [Google Scholar]
  • 18.Akbaba S, Held T, Lang K, Hoerner-Rieber J, Zaoui K, Forster T, et al. Salvage radiotherapy for recurrent hypopharyngeal and laryngeal squamous cell carcinoma (SCC) after first-line treatment with surgery alone: a 10-year single-centre experience. Radiat Oncol. 2019;14(1):34. 10.1186/s13014-019-1238-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Staackmann C, Ribbat-Idel J, Perner S, Idel C, Bruchhage KL, Hakim SG, et al. Palliative local radiotherapy for advanced squamous cell carcinoma of the head-and-neck: prognostic factors of survival. Anticancer Res. 2021;41(6):3205–10. 10.2187/anticanres.15107. [DOI] [PubMed] [Google Scholar]
  • 20.Baird BJ, Sung CK, Beadle BM, Divi V. Treatment of early-stage laryngeal cancer: a comparison of treatment options. Oral Oncol. 2018;87:8–16. 10.1016/j.oraloncology.2018.09.012. [DOI] [PubMed] [Google Scholar]
  • 21.Haapaniemi A, Väisänen J, Atula T, Alho OP, Mäkitie A, Koivunen P. Predictive factors and treatment outcome of laryngeal carcinoma recurrence. Head Neck. 2017;39(3):555–63. 10.1002/hed.24642. [DOI] [PubMed] [Google Scholar]
  • 22.Langfelder P, Horvath S. WGCNA: an R package for weighted correlation network analysis. BMC Bioinform. 2008;9:559. 10.1186/1471-2105-9-559. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Hutchinson MND, Mierzwa M, D’Silva NJ. Radiation resistance in head and neck squamous cell carcinoma: dire need for an appropriate sensitizer. Oncogene. 2020;39(18):3638–49. 10.1038/s41388-020-1250-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Zou C, Xu F, Shen J, Xu S. Identification of a ferroptosis-related prognostic gene PTGS2 based on risk modeling and immune microenvironment of early-stage cervical cancer. J Oncol. 2022;2022:3997562. 10.1155/2022/3997562. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Xin S, Fang W, Li J, Li D, Wang C, Huang Q, et al. Impact of STAT1 polymorphisms on crizotinib-induced hepatotoxicity in ALK-positive non-small cell lung cancer patients. J Cancer Res Clin Oncol. 2021;147(3):725–37. 10.1007/s00432-020-03476-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Gallez B, Neveu MA, Danhier P, Jordan BF. Manipulation of tumor oxygenation and radiosensitivity through modification of cell respiration. A critical review of approaches and imaging biomarkers for therapeutic guidance. Biochim Biophys Acta Bioenerg. 2017;1858(8):700–11. 10.1016/j.bbabio.2017.01.002. [DOI] [PubMed] [Google Scholar]
  • 27.Tews D, Fromme T, Keuper M, Hofmann SM, Debatin KM, Klingenspor M, et al. Teneurin-2 (TENM2) deficiency induces UCP1 expression in differentiating human fat cells. Mol Cell Endocrinol. 2017;443:106–13. 10.1016/j.mce.2017.01.015. [DOI] [PubMed] [Google Scholar]
  • 28.Yao Y, Zhang T, Qi L, Liu R, Liu G, Li J, et al. Identification of four genes as prognosis signatures in lung adenocarcinoma microenvironment. Pharmgenomics Pers Med. 2021;14:15–26. 10.2147/pgpm.S283414. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Choi SH, Cho SY, Song J, Hur MW. KLHL4, a novel p53 target gene, inhibits cell proliferation by activating p21(WAF/CDKN1A). Biochem Biophys Res Commun. 2020;530(3):588–96. 10.1016/j.bbrc.2020.07.100. [DOI] [PubMed] [Google Scholar]
  • 30.Williamson JS, Jones HG, Williams N, Griffiths AP, Jenkins G, Beynon J, et al. Extramural vascular invasion and response to neoadjuvant chemoradiotherapy in rectal cancer: influence of the CpG island methylator phenotype. World J Gastrointest Oncol. 2017;9(5):209–17. 10.4251/wjgo.v9.i5.209. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Zhang TM, Huang T, Wang RF. Cross talk of chromosome instability, CpG island methylator phenotype and mismatch repair in colorectal cancer. Oncol Lett. 2018;16(2):1736–46. 10.3892/ol.2018.8860. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Gollin SM. Cytogenetic alterations and their molecular genetic correlates in head and neck squamous cell carcinoma: a next generation window to the biology of disease. Genes Chromosomes Cancer. 2014;53(12):972–90. 10.1002/gcc.22214. [DOI] [PubMed] [Google Scholar]
  • 33.Qi W, Li Z, Xia L, Dai J, Zhang Q, Wu C, et al. LncRNA GABPB1-AS1 and GABPB1 regulate oxidative stress during erastin-induced ferroptosis in HepG2 hepatocellular carcinoma cells. Sci Rep. 2019;9(1):16185. 10.1038/s41598-019-52837-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Gao S, Zhang F, Sun H, Yang X. LncRNA GA-binding protein transcription factor subunit beta-1 antisense RNA 1 inhibits renal carcinoma growth through an MiR-1246/phosphoenolpyruvate carboxykinase 1 pathway. Onco Targets Ther. 2020;13:6827–36. 10.2147/ott.S257275. [DOI] [PMC free article] [PubMed] [Google Scholar] [Retracted]
  • 35.Li X, Wang H. Long non-coding RNA GABPB1-AS1 augments malignancy of glioma cells by sequestering MicroRNA-330 and reinforcing the ZNF367/cell cycle signaling pathway. Neuropsychiatr Dis Treat. 2021;17:2073–87. 10.2147/ndt.S305182. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Chen J, Bian M, Pan L, Liu C, Yang H. GABPB1-AS1 promotes the development of osteosarcoma by targeting SP1 and activating the Wnt/β-catenin pathway. J Oncol. 2022;2022:8468896. 10.1155/2022/8468896. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Additional file 1 (22.1KB, docx)
Additional file 2 (15.6KB, docx)
Additional file 3 (27.9KB, xlsx)
Additional file 4 (21.4KB, docx)
Additional file 5 (19.4KB, xlsx)

Data Availability Statement

The datasets generated for this study can be found in TCGA, https://portal.gdc.cancer.gov/.

Articles from Cancer Cell International are provided here courtesy of BMC

RESOURCES