1 Introduction
Head and neck squamous cell carcinoma (HNSCC) is the sixth most prevalent malignancy worldwide, with around 600,000 new cases of HNSCC diagnosed around the world annually [
1‐
3]. HNSCC is a highly aggressive tumor arising from the oral cavity, oropharynx, hypopharynx, and larynx [
4,
5]. Alcoholism, smoking, and infection with human papillomavirus are recognized as common risk factors [
5]. Early diagnosis of HNSCC is challenging due to its concealed physiological location, resulting in the majority of patients being diagnosed at advanced stages [
5]. Patients with advanced HNSCC have an unfavorable prognosis, irrespective of receiving interventions such as surgery, radiotherapy, and chemotherapy [
6]. Immunotherapy, including immune checkpoint inhibitors (ICIs), is a promising strategy to treat HNSCC [
7]. However, the overall response rate of immunotherapy in unselected HNSCC patients falls below 20% [
8,
9]. Thus, it is crucial to investigate new biomarkers and establish innovative prognostic signatures for predicting the outcomes of individuals with HNSCC and enabling personalized precision therapy.
Disulfidptosis, a recently identified cell death form, is induced by disulfide stress, as reported in recent research [
10]. Liu et al. noted a novel type of cell death termed disulfidoptosis, which exhibited substantial differences from apoptosis and ferroptosis. They reported that elevated expression of SLC7A11 in combination with glucose starvation is the underlying cause of this biological process [
10]. In cells with high SLC7A11 expression under glucose starvation, high cystine uptake combined with the decreased NADPH supply results in the depletion of NADPH, aberrant disulfide bonding in actin cytoskeleton proteins, collapse of the actin filament network, eventually leading to disulfidptosis [
10,
11]. In addition, Rac activated the WAVE regulatory complex to promote lamellipodia formation, which facilitated disulfidptosis [
12]. It was found that SLC7A11, SLC3A2, RPN1, and NCKAP1 were the top four disulfidptosis-promoting genes [
11]. Recent studies reported the association between disulfidptosis-related genes (DRGs) and prognosis in thyroid carcinoma, hepatocellular carcinoma, and bladder cancer [
13‐
15]. Nevertheless, despite the available literature, the relationship between disulfidptosis and HNSCC remains obscure.
Long noncoding RNAs (lncRNAs), a form of essential regulators involved in diverse biological processes, assume a significant role in tumorigenesis, such as proliferation, apoptosis, and metastasis [
16]. According to previous studies, lncRNAs have been considered functionally active in the progression and development of HNSCC [
17,
18] and hold promise as novel biomarkers and therapeutic strategies for HNSCC [
19,
20]. Nonetheless, the association between lncRNAs and disulfidptosis in HNSCC is still largely unexplored. Hence, in this study, disulfidptosis-related lncRNAs (DRlncRNAs) were retrieved at The Cancer Genome Atlas (TCGA,
https://portal.gdc.cancer.gov/) [
21] database using Pearson correlation analysis. Afterward, utilizing a combined approach of Cox proportional hazard regression and Least Absolute Shrinkage and Selection Operator (LASSO) regression, a DRlncRNA signature for HNSCC was developed based on 10 prognostic lncRNAs. Moreover, the signature was examined further concerning its prognostic value. Further investigations were conducted to assess its relevance in various aspects, such as immunotherapy response, infiltration of immune cells, and tumor microenvironment (TME).
2 Methods
2.1 HNSCC datasets and clinical data retrieval
The gene expression datasets of HNSCC, along with comprehensive clinical information, were acquired from the TCGA database (last accessed: July 15, 2023). The current research involved a total of 44 adjacent normal control cases and 522 HNSCC cases. The transcriptome data were retrieved in the form of fragments per kilobase million. The corresponding clinicopathologic information was also extracted, encompassing age, sex, clinical stage, grade, as well as T-, N-, and M-categories, along with overall survival (OS) time and status. Detailed clinicopathologic information of HNSCC patients used in this study was shown in Supplementary Table 1.
2.2 Expression data of DRlncRNAs in HNSCC
An assessment of prior research resulted in acquiring a total of 10 DRGs, including OXSM, GYS1, LRPPRC, NDUFS1, NUBPL, RPN1, NCKAP1, NDUFA11, SLC3A2, and SLC7A11 [
10,
22]. Through the Pearson correlation analysis, DRlncRNAs were selected employing the "limma" R packages. This analysis was executed after preparing 10 DRGs and all lncRNA expression data from the TCGA-HNSCC dataset. The lncRNAs were established as being linked to 10 DRGs: |Correlation coefficient|> 0.4 and
p < 0.001. The screened lncRNAs were then classified as DRlncRNAs.
2.3 Construction of the DRlncRNA-based prognostic model
Cases with corresponding prognostic information in the entire dataset were randomly categorized into training and test cohorts at a 1:1 ratio for further assessment. Screening of the survival-related DRlncRNAs in the training cohort was conducted through univariate Cox regression at a value of
p < 0.05. Afterward, LASSO regression with tenfold cross-validation and a
p-value of 0.05 was applied to further refine the selection of DRlncRNAs for constructing the DRlncRNA-based prognostic model. To prevent overfitting, this process was repeated 1000 times. The final selection of DRlncRNAs was conducted on the basis of multivariate Cox regression analysis. For all HNSCC patients, the formula mentioned below was utilized for quantifying the risk score:
$$\mathrm{Risk\, score} = {\sum }_{\mathrm{i }= 1}^{{\text{n}}}{\text{coefficient}} \times \mathrm{DRlncRNA\, expression}$$
The median risk score in the training set was employed for categorizing the training and test cohorts into high-risk group (HRG) and low-risk group (LRG). The R packages employed in this process encompassed "survival", "caret", "glmnet", "survminer", and "timeROC".
2.4 Validation of the risk model
To examine potential correlations between the calculated risk score and survival, we generated plots illustrating the risk score distribution of individuals with HNSCC. The Kaplan–Meier (KM) curves of OS and progression-free survival (PFS) were utilized to illustrate the predictive performance of the risk signature. The log-rank test was employed to compare the difference in survival between HRG and LRG. Moreover, heatmaps were utilized for the visualization of the expression of DRlncRNAs in each cohort. These analyses utilized R packages including "survival", "survminer", and "pheatmap". The next step involved utilizing principal component analysis (PCA) to show the distinguish of all genes, DRGs, DRlncRNAs, and risk model lncRNAs. The analysis was conducted utilizing the R packages "limma" and "scatterplot3d". Following this, the independent predictive value of the risk scores for OS was examined through Cox regression analyses (both univariate and multivariate). To comprehensively investigate the effectiveness of the DRlncRNA model, the receiver operating characteristic (ROC) curves of 1–, 3–, and 5–years were generated. Further assessment of the predictive performance of the risk score was executed through a comparison with clinical information (age, sex, grade, and stage) using ROC curves and C-index. These clinical features were also used for subgroup survival analyses. Additionally, the DRlncRNA signature was assessed to determine its potential for superior predictive performance. This was accomplished by comparing time-dependent ROC curves and C-index values with four previously established signatures [
23‐
26]. The R packages "survival", "survminer", "timeROC", "dplyr", "rms", and "pec" were employed for this assessment.
2.5 Tumor mutation burden (TMB) analysis
In the waterfall plot, the top 15 genes with the highest mutation frequencies were shown for both HRG and LRG. An analysis was then conducted to determine the mutation counts of all the HNSCC samples. The KM plotter and log-rank test were performed to test the OS of subgroups based on risk score and TMB utilizing the R packages "survival" and "survminer".
2.6 Establishment and validation of the prognostic nomogram
A nomogram was developed for survival prediction considering such factors as age, sex, histologic grade, clinical stage, T–, M–, and N–categories and risk score. Time-dependent ROC curves and calibration curves were employed to assess the precision of the nomogram. The R packages used for these analyses included "survival", "regplot", "rms", "survcomp", "survminer", and "timeROC".
2.7 Analysis of the association between the risk model and clinical characteristics
To validate the clinical relevance of the developed signature, the correlation of the risk score with clinicopathological variables was assessed. These variables encompassed clinical stages, sex, age, grade, T–, and N–categories. The Wilcoxon signed-rank test was applied to detect variance in risk scores for diverse groups of clinicopathological features. The R packages "ComplexHeatmap", "limma", and "ggpubr" were used to perform the necessary statistical analysis.
2.8 Functional enrichment analysis
To identify differentially expressed genes (DEGs) between two groups, the "limma" R tool with a significance threshold of FDR q < 0.05 and |log2FC|≥ 1 was utilized. Next, the identified genes were utilized for gene ontology (GO) annotation and KEGG pathway enrichment analysis.
2.9 Immune landscape and immunotherapy prediction analyses
To gain further insights into the immune landscape of the process, various analyses were conducted utilizing the "ESTIMATE" package, such as StromalScore, ImmuneScore, Tumor purity, and ESTIMATEScore for individuals with HNSCC. Subsequently, the immune functions of the two groups were compared using single-sample gene set enrichment analysis (ssGSEA). The CIBERSORT method was utilized to compute the percentages of 22 types of immune cells in every HNSCC specimen to determine the potential relationship between the risk score and the infiltration of immune cells. Further, the Tumor Immune Dysfunction and Exclusion (TIDE,
http://tide.dfci.harvard.edu/) scores were adopted to predict the response of HNSCC cases to immunotherapy. Conventionally, a lower TIDE score reflects, a more favorable response to immunotherapy. Moreover, to further predict the immunotherapy response of patients with HNSCC, the expression of human leukocyte antigen (HLA)-related and ICIs-related genes in HRG and LRG was comparatively assessed. The R packages utilized for this investigation were "ggpubr", "plyr", "reshape2", "ggplot2", and "ggExtra".
2.10 Statistical analyses
The analyses were carried out using R version 4.2.2, which also serves as a data visualization tool. The Wilcoxon signed-rank test or Chi-square test was employed for comparing the variables. Pearson correlation analysis was utilized for correlation evaluations. P value < 0.05 was deemed as statistically significant.
4 Discussion
Due to the heterogeneous characteristics and complex carcinogenic mechanisms of HNSCC, the most commonly used TNM staging system cannot accurately interpret the prognosis of patients [
27,
28]. Thus, it is crucial to investigate novel biomarkers for HNSCC. Increasing evidence has shown that the stability and tissue-specificity of lncRNAs make them ideal prognostic indicators for tumors [
29,
30]. Moreover, the discovery of disulfidptosis has ushered in a new era in antitumor treatment, unveiling a previously unexplored domain of research [
10]. This discovery has intriguied interest and investigations into innovative therapeutic approaches for tackling tumors. Herein, a novel DRlncRNA signature was developed for prognostic and immune microenvironment prediction of HNSCC.
In this study, we utilized the TCGA-HNSCC dataset to identify 10 prognosis-linked DRlncRNAs to constitute a novel prognostic disulfidptosis-related lncRNAs signature using LASSO and Cox regression analyses in the training cohort. The risk score derived from this signature negatively correlated with the OS of individuals with HNSCC in the training and the test cohorts. Additionally, it was noted to function as an independent risk indicator on the basis of the KM analysis and multivariate Cox regression. Moreover, PCA, utilized to validate the performance of the model, indicated the discriminative capacity of the lncRNAs in the DRlncRNA signature across patients from different risk groups. Furthermore, the prognostic value of the signature was comparatively assessed with diverse conventional prognostic factors (sex, age, grade, and stage) using ROC curves and C-index value. These comparisons highlighted the superior predictive capacity of the risk model. The clinical subgroup comparisons of age, sex, grade, and stage showed the effectiveness of the prognostic signature in all subgroups, further highlighting the universal applicability of the current risk model. Furthermore, a comparison of this signature with other published HNSCC-related prognosis signatures was executed through time-dependent ROC curves and C-index value. The resulting data implied that our signature was superior to the others in terms of prognosis prediction.
Previous research has established a correlation between TMB and tumor cell behaviors and immunological response [
31,
32]. It was noted that a higher TMB is indicative of a worse prognosis for individuals with HNSCC [
33,
34]. The survival results of this research were consistent with these reports. TP53, a well-recognized tumor suppressor, influences apoptosis and inhibits proliferation in tumor cells [
35]. Ample evidence has previously indicated that TP53 mutation is associated with a worse prognosis in HNSCC [
36,
37]. In this study, a high mutation rate of TP53 was found in HNSCC cases, especially in the HRG. Notably, it was found that the HRG demonstrated worse OS irrespective of the TMB level, highlighting the accuracy of our risk model concerning predictive values for HNSCC. In addition, nomograms have been applied in prognosis prediction in various human malignancies [
38,
39]. Therefore, a nomogram, based on the clinical features and risk score, was established to accurately predict survival in individuals with HNSCC. The respective AUC values for 1–, 3–, and 5–year OS were 0.787, 0.839, and 0.800 for the nomogram, showing that the nomogram model achieved good predictive accuracy. These collective results show the robustness and efficacy of the model for the prognosis prediction of individuals with HNSCC.
The functional enrichment analysis revealed that the DEGs between HRG and LRG were primarily associated with immune-related biological pathways and processes. It has been well recognized that the TME, particularly the immune microenvironment (IME), is a vital component of tumor biology [
40,
41]. Moreover, it was noted that lncRNAs assume an essential role in regulating the tumor IME [
42,
43]. Thus, the link between risk scores and IME was explored. Specifically, in terms of the infiltration of immune cells, individuals with lower risk depicted a higher immune score in comparison with the HRG, indicating a greater degree of infiltration. Additionally, ssGSEA indicated increased activation of immune functions in the LRG in comparison with the HRG, confirming the stronger antitumor immune activity in the lower-risk individuals. The study further performed the analysis of immune cell infiltration utilizing CIBERSORT and the ssGSEA methods. Prior research has reported that CD8 T cells, essential for adaptive immunity, confer crucial functions in antitumor immune responses [
44‐
46]. The resulting data were indicative of the elevated infiltration of the CD8 T cells in the individuals at low risk. Furthermore, Xu et al. reported that T follicular helper cells, which have antitumor functions, were associated with satisfactory HNSCC survival [
47]. Here, more infiltration of T follicular helper cells was noted in the LRG. Meanwhile, correlation analysis indicated that the risk score had a positive correlation with activated dendritic cells, macrophages M2, macrophages M0, activated mast cells, and resting memory CD4 T cells. It has been reported that macrophages M2, a type of immune suppressive cells, are closely linked to angiogenesis and tissue remodeling. These characteristics contribute to the augmentation of tumor-induced immune suppression, thereby facilitating the progression of tumors [
48]. Mast cells, akin to macrophages M2 cells, are frequently recognized as cells that promote tumor growth [
49‐
51]. Concerning the above results, it is reasonable to conclude that disulfidptosis goes along with the modulating of the composition of the IME of the tumor. Additionally, individuals in the HRG may have an attenuated antitumor immune status and worse immunotherapeutic response, which may further lead to poorer prognosis.
There are several limitations in this research that should be considered. Primarily, the training cohort and test cohort were sourced exclusively from the TCGA database. The inclusion of external validation cohorts for analysis would enhance the credibility of the findings. Moreover, the potential relevance of DRlncRNA mechanisms in immunotherapy against HNSCC was not thoroughly investigated in our study, thus warranting further comprehensive research. Additionally, further evidence is required to substantiate the role of ten DRlncRNAs in HNSCC. Hence, it is imperative to design extensive, multicentered prospective studies and wet experiments to validate our findings in subsequent research endeavors.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.