Deciphering Disulfidptosis‐Linked lncRNA Patterns as Potential HNSCC Biomarkers

ABSTRACT

Background

Our investigation sought to uncover the intrinsic features of Head and Neck Squamous Cell Carcinoma (HNSCC), particularly the role of long non‐coding RNAs implicated in disulfidptosis (DRLs).

Materials and Methods

We carried out lncRNA‐mRNA RNA‐Seq studies on HNSCC cells and harnessed the data from The Cancer Genome Atlas (TCGA), which includes 522 HNSCC tumors and 44 normal specimens. Bioinformatics evaluations aided in recognizing DRLs and estimating their prognostic value. Furthermore, we built a predictive model related to the chosen DRLs to scrutinize its linkage with the patients' prognosis. We also dug into tumor mutation loads and responses to chemotherapy.

Results

Our study identified three key DRLs (LINC02434, AC245041.2, and LINC02762) with considerable correlation to HNSCC prognosis. The risk model, utilizing these DRLs, successfully categorized patients into high‐risk and low‐risk clusters, uncovering differential survival trajectories. Moreover, the same risk model conveyed unique prognostic potential in HNSCC. Surveying the tumor microenvironment unfolded disparities between the groups, hinting toward potential implications for tactics in immunotherapy. We recognized distinct chemotherapeutic drugs with fluctuating responses across the risk clusters and molecular categories.

Conclusion

This investigation not only sheds light on prospective therapeutic pathways but also enhances our grasp of the molecular intricacies of HNSCC.

1. Introduction

Head and neck squamous cell carcinomas (HNSCCs) manifest themselves from the mucosal epithelium of the oral cavity, pharynx, and larynx (Johnson et al. 2020; Sung et al. 2021). Occupying the grim seventh position among global malignancies, these variants of cancer exhibit a discouraging survival trajectory, annually afflicting over 660,000 subjects and resulting in nearly 325,000 casualties (Johnson et al. 2020; Sung et al. 2021). HNSCC springs from a complex web of causative factors, encompassing exposure to carcinogenic substances such as tobacco, areca nut, betel quid, ethanol, and infections involving high‐risk species of human papillomavirus (HPV) and Epstein–Barr virus (EBV) (Isayeva et al. 2012; Johnson et al. 2020; Wong et al. 2014). Despite ongoing research, the genetic complexity of HNSCC, the difficulty in early identification, and the lack of dependable markers contribute to a 5‐year survival rate that remains under 50% (Elbers et al. 2020; Vos et al. 2021). This situation highlights the pressing requirement for an effective prognostic tool that can improve patient survival rates and inform more precise therapeutic approaches.

Cancer cell metabolism, driven by bioenergetic and biosynthetic demands, is integral to tumor initiation and progression (Hsieh et al. 2019). The strategic targeting of metabolic pathways presents an enticing arena for therapeutic exploitation (Martínez‐Reyes and Chandel 2021; Stine et al. 2022). Among the various regulated cell death (RCD) pathways identified, disulfidptosis—a novel form involving disulfide bond reactions—has shown potential in cancer therapy (Carneiro and El‐Deiry 2020; Wabnitz et al. 2010). The discovery of disulfidptosis and its role in the collapse of the cytoskeleton underpins a new frontier in understanding cell death mechanisms in cancer (Tang et al. 2019; Zheng, Liu, et al. 2023). Preliminary evidence suggests that a balanced state of disulfidptosis could serve as a potential therapeutic bullseye, likely contributing toward treatment efficacy enhancement in HNSCC (Liu et al. 2023; Meng et al. 2023). Nonetheless, its affiliation with HNSCC prognosis and therapeutic outcomes warrants in‐depth rationalization.

Long non‐coding RNAs (lncRNAs), an assorted class of gene transcripts that exceed 200 nucleotides in length, have been found to engulf a realm of cell differentiation and tumorigenesis (Mercer et al. 2009). Recent studies have highlighted the potential of disulfidptosis‐related lncRNAs (DRLs) as prognostic indicators in an array of malignancies, suggesting their potential in modulating cytoskeletal dynamics and immune responses (Li et al. 2023; Shen et al. 2023; Xue et al. 2023). Though still nascent, the exploration of DRLs in HNSCC promises remarkable potential in unraveling the complex symbiosis of cellular death channels and immune evasion techniques.

In our study, we harnessed the power of RNA Sequencing (RNA‐Seq) to pinpoint differentially expressed (DE) DRLs and mRNAs in HNSCC cell varieties. Aided by a distinctive prognostic model based on these DRLs, our objective was to deliver an encompassing analysis involving survival trajectories, immune scenarios, and responses to chemotherapy. Our findings not only enhance the prognostic prediction accuracy for HNSCC but also pave the way for the discovery of new clinical targets. This signature is poised to revolutionize the predictive capabilities for patient prognosis and the development of precision medicine strategies for HNSCC management.

2. Materials and Methods

2.1. RNA‐Seq Procedure and Data Distillation

With the RNAeasy Animal RNA Isolation Kit at hand (Beyotime, R0026), we extracted total RNA. Abiding by the protocols provided by the manufacturer, we established sequencing libraries using the RNA Library Prep Kit and incorporated distinct index codes for individual samples. We partitioned regulatory ncRNA and mRNA from the complete RNA via rRNA depletion probes, which was soon followed by fragmentation triggered by divalent cations at heightened temperatures within the First Strand Synthesis Reaction Buffer (5X). Employing M‐MuLV Reverse Transcriptase (RNaseH) along with a random hexamer primer, the inaugural strand cDNA was created. The second strand of cDNA was cultivated, and subsequent to which exonuclease/polymerase activities were introduced to establish blunt ends. The 3′ ends were then subjected to adenylation, and NEBNext adaptors were ligated for hybridization. The library fragments were sieved for dimension (370–420 bp) utilizing the AMPure XP system and then treated with USER Enzyme preceding PCR magnification with Phusion High‐Fidelity DNA polymerase supplemented with Index Primers. After the purification of PCR products, an examination of the library quality was conducted on the Agilent 5400 system, followed by QPCR quantification (1.5 nM). Sequencing was completed on Illumina platforms via a PE150 strategy. The raw data were deciphered from fluorescence image files into a FASTQ format, accounting for sequence and quality specifications. Ensuing quality checks were done using Fastp, which furnished rudimentary statistics on raw read quality and proceeded to eliminate reads exhibiting adapter pollution, abundant uncertain bases, or a large fraction of low‐quality bases.

2.2. Bioinformatics Analysis Through GO and KEGG

The DESeq2 R package (version 4.2.2) was utilized for carrying out differential expression scrutiny. This was succeeded by GO and KEGG pathway analysis executed via the R package clusterProfiler. The accumulated results underwent visual rendering through ggplot2.

2.3. HNSCC Cell Lines and Tissue Samples Collection for qRT‐PCR

The human HNSCC cell lines FaDu and SCC25, along with the nasopharyngeal epithelial cell line NP69 (Procell, Wuhan, China), were cultured in specialized media at 37°C in a 5% CO₂ atmosphere. The human tongue cancer specimens and adjacent normal tissues were obtained from Anhui Provincial Hospital, consisting of small residual tissues collected during surgical procedures. Informed consent was obtained from all patients prior to the acquisition of these samples. The qRT‐PCR protocol adhered to the guidelines specified in the TAKARA instruction manual, with relative transcript levels calculated using the 2^−ΔΔCT method.

2.4. Data Procurement

We obtained RNA sequencing information, clinical details, and gene mutation records related to HNSCC from TCGA at our access on 17 May 2023. The dataset encompassed an entirety of 522 HNSCC tumorous and 44 normal tissue samples. The application of the Perl programming language (https://www.perl.org) aided in the extraction and organization of RNA‐seq information, clinical details, and gene mutation data. To augment the credibility of our discovery, 519 patients (who had transcript data available) were separated into a training cohort (n = 260) and a verification cohort (n = 259) with a balanced 1:1 ratio. This division was executed randomly to curtail potential bias arising during the signature validation evaluation (Table 1). Our main objective was a thorough appraisal and confirmation of the introduced signature, ensuring its consistency and universality across the HNSCC patient demography.

TABLE 1.

Clinical characteristics of the patients with HNSC in the training and testing cohort.

Covariates	Type	Total	Test	Train	p
Age	≤ 65	341 (65.7%)	166 (64.09%)	175 (67.31%)	0.4971
Age	> 65	178 (34.3%)	93 (35.91%)	85 (32.69%)
Gender	Female	136 (26.2%)	75 (28.96%)	61 (23.46%)	0.1856
Gender	Male	383 (73.8%)	184 (71.04%)	199 (76.54%)
Grade	G1	62 (11.95%)	27 (10.42%)	35 (13.46%)	0.3642
Grade	G2	303 (58.38%)	147 (56.76%)	156 (60%)
Grade	G3	125 (24.08%)	67 (25.87%)	58 (22.31%)
Grade	G4	7 (1.35%)	5 (1.93%)	2 (0.77%)
Grade	Unknown	22 (4.24%)	13 (5.02%)	9 (3.46%)
Stage	Stage I	27 (5.2%)	16 (6.18%)	11 (4.23%)	0.6405
Stage	Stage II	70 (13.49%)	38 (14.67%)	32 (12.31%)
Stage	Stage III	81 (15.61%)	40 (15.44%)	41 (15.77%)
Stage	Stage IV	266 (51.25%)	129 (49.81%)	137 (52.69%)
Stage	Unknown	75 (14.45%)	36 (13.9%)	39 (15%)
T	T0	1 (0.19%)	0 (0%)	1 (0.38%)	0.2065
T	T1	48 (9.25%)	25 (9.65%)	23 (8.85%)
T	T2	135 (26.01%)	66 (25.48%)	69 (26.54%)
T	T3	99 (19.08%)	59 (22.78%)	40 (15.38%)
T	T4	174 (33.53%)	80 (30.89%)	94 (36.15%)
T	Unknown	62 (11.95%)	29 (11.2%)	33 (12.69%)
M	M0	185 (35.65%)	97 (37.45%)	88 (33.85%)	1
M	M1	1 (0.19%)	1 (0.39%)	0 (0%)
M	Unknown	333 (64.16%)	161 (62.16%)	172 (66.15%)
N	N0	175 (33.72%)	80 (30.89%)	95 (36.54%)	0.5981
N	N1	67 (12.91%)	35 (13.51%)	32 (12.31%)
N	N2	169 (32.56%)	89 (34.36%)	80 (30.77%)
N	N3	8 (1.54%)	4 (1.54%)	4 (1.54%)
N	Unknown	100 (19.27%)	51 (19.69%)	49 (18.85%)

2.5. DRLs Compilation

A sum total of 18 genes linked with disulfidptosis have been cited (Liu et al. 2023; Zheng, Zhou, et al. 2023). We assembled these genes, categorizing them as disulfidptosis‐related genes. The “Limma” package was brought into use for this analysis, keeping a consideration for correlations having |correlation coefficients| above 0.1 and p‐values beneath 0.05. Subsequent identification of modular genes in tumorous and normal tissues was assessed by the weighted gene co‐expression network analysis (WGCNA) technique, setting a cut‐off for p‐values beneath 0.05. Lastly, the Sankey diagram was plotted with the “ggplot2” and “ggalluvial” packages to graphically illustrate the associations between disulfidptosis genes and DRLs.

2.6. Risk Model Design and Verification

The Least Absolute Shrinkage and Selection Operator (LASSO) regression technique was utilized to tackle overfitting in the training cohort, while the Cox multivariate analysis method facilitated the assessment of key DRL coefficients. The construction of the risk model involved the selection of three prognostic DRLs, each with individual coefficients calculated using Cox proportional hazard regression analysis, thereby determining the risk score. For model validation, the packages “survival” and “survminer” were employed, facilitating Kaplan–Meier curve construction and overall survival (OS), progression‐free survival (PFS), and disease‐specific survival (DSS) value comparisons. The model's prognostic accuracy was scrutinized using Cox regression analysis. Model efficacy was further confirmed by the calibration plot, the time‐dependent receiver operating characteristic (ROC) curves, and the C‐index. For exploring the efficacy of the risk model in setting the risk status, principal component analysis (PCA) was engaged. Differential lncRNA expressions across risk groups were investigated via heatmap analysis.

2.7. Tumor Mutation Burden (TMB) and Tumor Immune Dysfunction and Exclusion (TIDE) Examination

Using the “maftools” package, mutated genes were recognized and visualized through waterfall diagrams representing the top 15 most frequently mutated genes. TMB was calculated by tallying mutations per exome size. Visualization of difference and correlation analysis was done using the “ggpubr,” “limma,” and “reshape2” packages. OS differentials were evaluated using the “survival” and “survminer” packages. Data for TIDE scoring were taken from a web source, after which an analysis was carried out between high and low‐risk sectors. Prediction of drug IC50 metrics for HNSCC treatment in varying risk sectors was facilitated by the “oncoPredict” package.

2.8. Tumor Grouping

The “ConsensusClusterPlus” package was used to split tumor samples into three differing clusters. An interplay among clusters and risk groups was mapped using a Sankey diagram, created with the “ggalluvial” package. Survival analysis, potential drug identification, and T‐distributed stochastic neighbor embedding (tSNE) were carried out with prior methods and the “Rtsne” package specifically for tSNE.

2.9. Statistical Analysis

All computational analysis was performed using R (version 4.2.2). Corresponding diagrams were also constructed utilizing R, establishing statistical relevance at p < 0.05.

3. Results

3.1. Unraveling Disulfide‐Linked Cell Death Mechanisms in HNSCC via lncRNA‐mRNA Sequencing

Our lncRNA‐mRNA sequencing analysis unearthed substantial gene expression aberrations in HNSCC cell lines when compared with control cohorts. Specifically, we detected 1151 upregulated and 2529 downregulated mRNAs, alongside 2277 upregulated and 2603 downregulated lncRNAs, each displaying a fold change exceeding 2.0 and having a statistical significance of p < 0.05 (Figure 1A,B). A comprehensive list of these DE RNAs, including their fold changes and p values, is presented in Tables S2 and S3.

FIGURE 1.

Overview of variant expression patterns of lncRNAs and mRNAs in HNSCC cells. (A) A gene expression heatmap representing the variance of lncRNAs in control (C) against HNSCC cell lines (F). (B) A volcano diagram depicting gene expression fold alterations of representative variantly expressed mRNAs in the disulfidptosis process. Significant upregulated and downregulated genes are denoted by red and blue dots, respectively (signified by |log2 fold change| > 2.0 and p value < 0.05). (C) Significantly amplified gene ontology (GO) elements of the divergently expressed mRNAs. (D) Recognition of emblematic variantly expressed mRNAs associated with the microtubule cytoskeleton, the response to oxidative stress, and pathways of disulfidptosis.

Recognizing the pivotal nature of lncRNAs in modulating the expression of protein‐coding genes (Mattick et al. 2023), we conducted GO enrichment and KEGG pathway analyses to decode the biological implications of the DE lncRNAs. The GO enrichment analysis revealed significant enrichment terms among DE mRNAs, predominantly associated with cell–cell adhesion, microtubule cytoskeleton organization, oxidative stress response, and dynein complex assembly (Figure 1C). This suggests a coordinated response to changes in cellular architecture and stress. Within the framework of oxidative stress, our analysis highlighted a shift toward the formation of protein disulfide bonds, indicative of a cellular response to disulfide stress, which can culminate in the collapse of the cytoskeleton—a critical event in disulfide stress‐induced cell death. This was corroborated by the identification of DE mRNAs that are pivotal to the microtubule cytoskeleton, oxidative stress response, and three genes (SLC7A11, MYH10, and TLN1) (Figure 1B,D). Cumulatively, these findings suggest a potential involvement of disulfide‐triggered cell death within HNSCC, opening up new possibilities for therapeutic initiatives.

3.2. Core DRLs Identification and Prognostic Risk Model Construction

Initiating our exploration to decode the complex link between disulfidptosis and HNSCC, we curated a list of 18 disulfidptosis‐associated genes from existing literature and cross‐checked them with their expression profiles in HNSCC samples (Tables S4 and S5). A visual synopsis of our study's methodology is depicted in a flowchart format as shown in Figure S1. Employing a robust statistical approach, we performed Pearson correlation analysis on a comprehensive set of lncRNAs, yielding a total of 3573 candidates with significant correlation coefficients. We employed a dynamic tree clipping method for module identification, focusing on modules that contained no fewer than 50 lncRNAs with a similarity index exceeding 0.75. The outcome tree diagram of the lncRNA modules is displayed in Figure 2A. A subsequent correlation analysis between these modules and predefined trait groupings, visualized as a heat map in Figure 2B, allowed us to pinpoint the yellow module as a cluster of interest (Table S6). Notably, this module demonstrated significant correlations with one or more disulfidptosis‐related genes, as delineated in Figure 2C.

FIGURE 2.

Determining the disulfidptosis‐centric lncRNA signature for HNSCC patients. (A) The clustering dendrogram of DRLs in HNSCC per TCGA data. (B) The relationships of module‐to‐traits, premised on TCGA data. DRLs in the yellow module were selected for subsequent examination. (C) The Sankey diagram showing the connection between disulfidptosis‐associated genes and DRLs. (D) A forest plot displaying the prognostic relevance corresponding to DRLs (p < 0.05). (E) lncRNA selection via LASSO Cox regression was conducted, by calculating the smallest criterion. (F) Representation of the generalized cross‐validation curve of paired likelihood deviance. (G) Illustration of the correlation between disulfidptosis‐related genes and DRLs. (H) qRT‐PCR outcomes of the three selected DRLs in cell lines. *, p < 0.05; ***, p < 0.001, ****, p < 0.0001. (I) qRT‐PCR outcomes of the three selected DRLs in samples. *, p < 0.05; **, p < 0.01; ***, p < 0.001.

To substantiate the robustness of these findings, we stratified a cohort of 519 HNSCC patients into distinct training (n = 260) and testing (n = 259) sets, ensuring even clinical characteristic representation across both sets (Table 1). Through univariate Cox regression analysis, we filtered and identified 14 lncRNAs with a significant association with patient prognosis, termed disulfidptosis‐associated prognostic lncRNAs (p < 0.05), as illustrated in Figure 2D. In a bid to refine our predictive model and mitigate overfitting, we applied LASSO Cox regression analysis to the training cohort, carefully tuning the lambda value for optimal model performance (Figure 2E,F). This rigorous selection process culminated in the identification of three pivotal lncRNAs, which were then integrated into a predictive signature risk model. The model's formulation was informed by the regression coefficients derived from our analysis, assigning appropriate weights to each of the lncRNAs within the model. The interplay between the three selected lncRNAs and disulfidptosis‐related genes is graphically represented in Figure 2G. Moreover, the risk scores for all samples, computed based on the expression intensity of these three core DRLs, are meticulously documented in Table S7.

3.3. qRT‐PCR Validation of the Identified Trio of DRLs

We performed qRT‐PCR validation on a new set of samples to confirm the expression patterns of the identified disulfidptosis‐related lncRNAs (DRLs), focusing on three key candidates: LINC02434, AC245041.2, and LINC02762 (Table S1). The results demonstrated significant differential expression of all three DRLs in FaDu and SCC cells compared to NP69 cells (Figure 2H). The expression profiles of LINC02434 and AC245041.2 were consistently downregulated in FaDu and SCC cells, aligning with the trends observed in our initial RNA‐sequencing data and database analysis.

To further verify the expression in tissue samples, we extracted human tongue cancer and adjacent normal tissues, and the results were consistent with those found in the cell lines (Figure 2I). This concordance strengthens the case for these lncRNAs as potential regulators of disulfidptosis in HNSCC. However, the expression pattern of LINC02762 deviated from the expected trend, suggesting a complexity in its regulation or function that may not be fully captured by RNA sequencing alone. This discrepancy highlights the importance of utilizing multiple methodological approaches in biological research and underscores the need for further investigation to fully understand the role of LINC02762 in HNSCC.

3.4. Analyzing Survival in Relation to Risk Model and DRLs

We conducted a survival analysis to evaluate the prognostic efficacy of our risk model, stratifying patients into high‐risk and low‐risk categories using median risk scores. This classification was consistently applied across the different patient cohorts, including training, testing, and joined sets. Within the training cohort, the low‐risk category demonstrated significantly improved OS compared to the high‐risk group (p < 0.001, Figure 3A). This considerable discrepancy in OS was corroborated in both the testing (p = 0.003, Figure 3B) and the combined patient sets (p < 0.001, Figure 3C). Likewise, the low‐risk group surpassed the high‐risk group in terms of PFS in the training set (p = 0.003, Figure 3D). This PFS pattern was in agreement with the testing (p = 0.031, Figure 3E) and combined patient sets (p < 0.001, Figure 3F). Disease‐specific survival (DSS) also showed the low‐risk group in the training cohort outperforming the high‐risk group (p = 0.002, Figure 3G). This comparison was consistent in the testing (p = 0.013, Figure 3H) and combined patient sets (p < 0.001, Figure 3I). The findings from these survival analyses underscore the prognostic power of the risk model and offer valuable insights on the links between OS, PFS, and DSS in HNSCC patients.

FIGURE 3.

Life expectancy exploration for training, testing, and all sets. (A) Life expectancy trajectory in terms of OS for the training set (p < 0.001). (B) Life expectancy trajectory in terms of OS for the testing set (p = 0.003). (C) Life expectancy trajectory in terms of OS for all groups combined (p < 0.001). (D) Life expectancy trajectory in terms of PFS within the training set (p = 0.003). (E) Life expectancy trajectory in terms of PFS within the testing set (p = 0.031). (F) Life expectancy trajectory in terms of PFS for all groups combined (p < 0.001). (G) Exploration of DSS expectancy in the training set (p = 0.002). (H) Exploration of DSS expectancy in the testing set (p = 0.013). (I) Exploration of DSS expectancy for all sets together (p < 0.001).

3.5. Risk Trajectory Analysis, Expression Dynamics, and Prognostic Validation of the Risk Model

Our scrutiny of the risk curve solidified a distinct association between escalating risk scores and HNSCC patients' mortality rate within the training set (Figure 4A–D). This important trend was consistently echoed in both testing (Figure 4B,E) and combined patient cohorts (Figure 4C,F). Expression studies highlighted LINC02434, AC245041.2, and LINC02762 possessing higher expression in the high‐risk group versus the low‐risk group within the training, testing, and joined patient cohorts (Figure 4G–I), further reinforcing their potential as prognostic indicators.

FIGURE 4.

Prognosis Information Derived from the Three DRLs‐Based Signature and Cox Regression Analysis of the Risk Pattern. Representation and median value of the risk score in the training set (A), testing set (B), and all datasets amalgamated (C). Showcasing of overall survival status and corresponding duration in the training (D), testing (E), and all datasets amalgamated (F). A heatmap exhibits the differential expression of three DRLs in high‐ and low‐risk groups across the training (G), testing (H), and all datasets (I). (J) Univariate Cox regression assessment of the DRLs signature (risk model) for the training dataset (p < 0.001). (K) Univariate Cox regression analysis of DRLs signature for the testing set (p = 0.003). (L) Univariate Cox regression exploration of the DRLs signature across all datasets (p < 0.001). (M) Multivariate Cox regression examination of the DRLs signature in the training dataset (p < 0.001). (N) Multivariate Cox regression analysis of the DRLs signature for the testing dataset (p = 0.001). (O) Multivariate Cox regression scrutiny of the DRLs signature across all datasets (p < 0.001).

To position the risk score as an individual prognostic determinant, we instituted both univariate and multivariate Cox regression analyses within the training cohort. The univariate analysis yielded a hazard ratio (HR) of 1.784 (p < 0.001, Figure 4J), and the multivariate analysis delivered an HR of 1.696 (p < 0.001, Figure 4M). These findings were strongly validated within the testing set, where univariate analysis achieved an HR of 1.549 (p = 0.003, Figure 4K) and multivariate analysis derived an HR of 1.643 (p = 0.001, Figure 4N). The comprehensive patient set analysis consolidated these results, with univariate analysis demonstrating an HR of 1.670 (p < 0.001, Figure 4L) and multivariate analysis signaling an HR of 1.643 (p < 0.001, Figure 4O). These outcomes indicate that the risk score, extrapolated from our risk model based on the discovered DRLs, stands as an independent and trustworthy prognostic indicator for HNSCC patients. This underscores the potential clinical applicability of the risk score in anticipating patient outcomes and informing treatment choices.

3.6. Evaluating the Diagnostic and Prognostic Importance of the Risk Model

To ascertain the diagnostic efficacy of our risk model, we conducted ROC curve analysis for 1‐, 3‐, and 5‐year survival predictions within the training cohort. The AUC values were 0.636, 0.680, and 0.666 consecutively (Figure 5A). Although these values did not exceed the threshold of 0.7, they indicated better predictive accuracy than traditional clinical parameters such as age, gender, grade, and clinical stage (Figure 5B–D). This trend was consistent in the testing set with AUC values of 0.683, 0.683, and 0.619, and in the all patients set with AUC values of 0.660, 0.680, and 0.627 (Figure 5E–L), suggesting that the risk model provides a more comprehensive diagnostic tool than individual clinical characteristics. Further validation of the model's prognostic value was established through the Concordance (C)‐index, which showed higher scores for the risk model across all patient sets (Figure S2A–C). This metric implies that the risk score is a formidable predictor of long‐term outcomes in HNSCC patients. To streamline clinical usage, we designed a nomogram that incorporates both clinical variables and risk scores to estimate individual patient probabilities of achieving 1‐, 3‐, and 5‐year overall survival (OS) (Figure S2D). Calibration plots verified the accuracy of the nomogram's projections, exhibiting an almost exact alignment between predicted and actual outcomes (Figure S2E).

FIGURE 5.

ROC evaluation of the risk arrangement. (A) Time‐dependent ROC curve's AUC for survival forecast based on the risk score within the training set. Comparative AUC values from ROC curves, indicating the prognostic precision of the risk score and clinical factors for 1 year (B), 3 years (C), and 5 years (D) within the training set. (E) Time‐dependent ROC curve's AUC for survival projection centered on the risk score within the testing set. Comparative AUC values from ROC curves, scrutinizing the prognostic precision of the risk score and clinical parameters for a duration of 1 year (F), 3 years (G), and 5 years (H) within the testing set. (I) AUC of time‐dependent ROC curves for the survival prediction leveraging the risk score across all sets. Comparative measures from ROC curves, assessing the prognostic accuracy of the risk score along clinical characteristics for a term of 1 year (J), 3 years (K), and 5 years (L) across all sets.

Moreover, we compared survival probabilities across a host of clinical features, including age, gender, grade, T, N, M, and clinical stage. The analysis consistently revealed that patients deemed high‐risk by our model exhibited poorer OS rates compared to their low‐risk counterparts, regardless of individual clinical variables (Figure S3). These outcomes spotlight the versatility of our risk model across diverse clinical scenarios. In summary, our risk model exhibits both diagnostic and prognostic relevance in HNSCC patients. It surpasses individual clinical attributes in predicting survival outcomes, as endorsed by the AUC values, C‐index scores, and nomogram forecasts.

3.7. Molecular Classification and Survival Consequences in HNSCC Specimens

We utilized a consensus clustering technique based on the expression levels of the three DRLs, presenting three distinct molecular categories within the collection of 519 HNSCC specimens (Figure 6A, Table S8). Consequent survival analysis unveiled that individuals in cluster 2 experienced a significantly elevated survival rate relative to those in clusters 1 and 3 (p < 0.001, Figure 6B). This survival advantage within cluster 2 is potentially attributed to the majority of these patients being allocated into the low‐risk category, which corresponds with a superior prognosis (Figure 6C).

FIGURE 6.

Grouping assessment of 3 DRLs in the total cohort. (A) HNSCC specimens were categorized into three distinct groups with k equalling three. (B) Survival comparison among three clusters: Clusters 1, 2, and 3 (p < 0.001). (C) The Sankey diagram illustrates the relationship between the three tumor clusters and two risk groups. (D) PCA interpretation of the three clusters: Clusters 1, 2, and 3. (E) The t‐SNE analysis segregates and assesses Clusters 1, 2, and 3. (F) Clinical disparity scrutiny of N phase between the three clusters: Clusters 1, 2, and 3 (p = 0.021).

To further probe the molecular variances between the identified tumor clusters, we employed principal component analysis (PCA) and T‐distributed stochastic neighbor embedding (tSNE) visualization methodologies. Both PCA and tSNE representations exhibited a clear distinction among the three clusters, underscoring the solid differentiability of the molecular subtypes (Figure 6D,E). In addition, we executed a differential examination of clinical characteristics across clusters 1, 2, and 3, scrutinizing the dissemination of clinical features. Importantly, we pinpointed a statistically significant discrepancy in the N stage (p = 0.021, Figure 6F), implying that the molecular subdivisions might be associated with varying stages of tumor advancement.

3.8. PCA Analysis and Delving Into TMB and TIDE in Risk Categorization

We harnessed the PCA technique to evaluate the proficiency of our DRLs signature in differentiating between risk states among HNSCC patients. The PCA findings exhibited a distinct separation between high‐risk and low‐risk patient cohorts based on the signature, indicating its significance in risk categorization (Figure S5D). In contrast, analysis of comprehensive genome expression data, solely disulfidptosis‐related genes, or the DRLs independently did not yield efficacious partitioning of risk status (Figure S5A–C), emphasizing the signature's aggregate worth.

Further, seeking to comprehend the influence of somatic mutations on risk groups, we derived mutation data from the TCGA database. Waterfall diagrams spotlighted the top 15 altered genes for both high‐ and low‐risk sections, unveiling unique mutational landscapes (Figure 7A,B). Survival analysis disclosed a substantial survival superiority for individuals in the low‐TMB group compared to the high‐TMB group (p = 0.006, Figure 7C). Remarkably, subgroup analysis implied that patients possessing both low TMB and low risk experienced the most beneficial survival results, further validating the concurrent role of TMB and our risk model in patient prognosis (p < 0.001, Figure 7D). Moreover, the TIDE score analysis signaled that the low‐risk group had a significantly lower TIDE score compared to the high‐risk group (p < 0.05, Figure 7E). This insinuates that low‐risk patients may exhibit a superior response to immunotherapy, possibly attributed to a reduced ability for immune evasion.

FIGURE 7.

Analysis of TMB and TIDE. (A) A waterfall diagram displays the top 15 mutated genes in the high‐risk category. (B) A waterfall chart showcases the top 15 mutated genes in the low‐risk category. (C) Survival assessment between the groups with high and low TMB values (p = 0.006). (D) Subgroup survival comparison among high‐TMB‐high‐risk, high‐TMB‐low‐risk, low‐TMB‐high‐risk, and low‐TMB‐low‐risk cohorts (p < 0.001). (E) TIDE evaluation for the low‐ and high‐risk groups (*, p < 0.05).

3.9. Inspection of Drug Response Across Risk Segments and Tumor Clusters in HNSCC

As we ventured into the treatment dimensions for HNSCC, we utilized the ‘oncoPredict’ tool to scan for chemotherapy agents that might display varying sensitivity across the two identified risk sections and three molecular tumor clusters in our research. This approach stands pivotal for tailoring medicine, aiming to align treatment to the specific molecular attributes of a patient's tumor. Our structured and detailed investigation, visually depicted in Figure S4A–P, unveiled a suite of drugs that revealed notable sensitivity, as represented by the lowest IC50 values, in the low‐risk cohorts or cluster 2. Importantly, Alpelisib, Axitinib, AZD1208, Doramapimod, GSK269962A, JQ1, Sinularin, and Tozasertib stood out for their efficacy in these contexts (all p < 0.001), suggesting that individuals within these groups are more probable beneficiaries of these targeted therapies. The detection of these drugs is founded on their potential to target the molecular vulnerabilities of tumors in the low‐risk cohorts, or cluster 2, justifying their deployment in treating distinct molecular subtypes of HNSCC. This assessment of sensitivity amplifies our comprehension of drug response in HNSCC and also lays the groundwork for selecting treatments that have a higher likelihood of success in individual patients.

4. Discussion

We have presented a broad analysis of disulfidptosis patterns within HNSCC cell lines in this research using high‐throughput sequencing and bioinformatics breakdowns. The detection of 3428 DE lncRNAs and 5132 mRNAs highlights the heterogeneity of HNSCC while also hinting at a possible function for these lncRNAs in differentiating HNSCC from normal controls. The function enrichment study, involving GO and KEGG pathway annotations, illuminated pathways like cell‐to‐cell adhesion, calcium ion binding, MAPK and PI3K‐AKT signaling, and the IL‐17 signaling pathway, indicating potential significance in the pathogenesis of HNSCC.

Existing theories suggest a linkage between disulfidptosis and disulfide stress happening among intracellular and extracellular protein bodies within the cytoskeleton (Liu et al. 2023). This process incites morphological transformations and impairs routine protein function, resulting in cell death, such as disulfidptosis. Our study's outcomes hint at crucial roles of cytokine‐cytokine receptor interaction and cell adhesion molecules in HNSCC cell death and tumorigenesis (Zheng, Zhou, et al. 2023). Our study's outcomes hint at crucial roles of cytokine‐cytokine receptor interaction and cell adhesion molecules in HNSCC cell death and tumorigenesis (Feng et al. 2023; Liberzon et al. 2015). However, the precise mechanisms connecting disulfide bond creation to cell death remain an area for potential further investigation.

At present, the regulatory processes of disulfidptosis are not thoroughly comprehended, especially concerning lncRNAs (Wang et al. 2023). Through our research, we have pinpointed two particular DRLs, LINC02434 and AC245041.2, using lncRNA sequencing and bioinformatics analysis, which may have important functions in the progression of HNSCC. We confirmed the substantial abnormal expression of chosen DRLs in HNSCC cell lines through qPCR experiments. Further examination revealed all to be linked with an unfavorable prognosis in HNSCC, hinting at a potential role of disulfidptosis in the condition. The upsurge of LINC02434 has been suggested within HNSCC samples, proposing its potential merit as a diagnostic tumor tissue marker for HNSCC patients (Jiang et al. 2021). The lncRNA AC245041.2 and mRNA LAMA3 demonstrated a strong correlation of expression levels in pancreatic adenocarcinoma, reiterating its underlying role in tumor progression (Tian et al. 2021). As an additional discovery, the heightened expression of lncRNA AC245041.2 has been categorized as a novel ferroptosis‐associated lncRNA signature for predicting prognosis in gastric cancer, offering a new hopeful tumor treatment (Wei et al. 2021). The risk scoring and analysis by subgroups of these discovered lncRNAs further bolster their potential as markers for forecasting progression in HNSCC patients, adding to the expanding body of evidence in this domain. However, the exact mechanisms through which the distinguished DRLs influence tumor growth and immune processes in HNSCC remain hypothetical and need stronger validation.

The creation of molecular subgroups based on disulfidptosis‐related genes, together with the establishment of a prognostic risk model, lays the groundwork for personalized therapeutic approaches. Potential treatments for varying patient categories, informed by transcriptome profiling, clinical information, and gene mutation data, emphasize the clinical importance of disulfide stress in determining prognosis and suggesting treatment strategies for HNSCC. The critical part played by immune cells within the TME in the formation of different tumors is broadly recognized (Wei et al. 2020). Existing research on HNSCC immunotherapies has found that targeting specific immune checkpoints could restore T cell responses and create beneficial treatment strategies (Chen and Han 2015; Chen et al. 2020; Zou et al. 2016). In our study, we've used a consensus clustering methodology centered on the expression of unique DRLs to identify the molecular subclasses among HNSCC. Remarkably, cluster 3, epitomizing the low‐risk set, showed the greatest expression levels of immune checkpoint genes and presented better clinical outcomes. Moreover, our risk signature demonstrated a strong capacity for predicting OS as an independent prognosis marker in HNSCC, due to its potential ability to overturn immune evasion mechanisms (Wang et al. 2021). Therefore, our observations emphasize the links between DRL types and changes in the immunological tumor microenvironment in HNSCC, highlighting their significance in customizing treatment methodologies. However, current therapeutic measures for HNSCC are still insufficient and frequently lead to drug resistance. In response to this issue, the discovery and development of innovative treatment strategies, along with distinct molecular or cellular indicators, are critical to enhance treatment results and foresee survival for HNSCC patients (Martin et al. 2014). Even though our investigation has provided valuable observations regarding the possible roles of these DRLs, further experimental and clinical studies are needed to clarify their target mRNAs and affirm their functional importance in HNSCC.

The present study has several limitations. First, our analysis was based on publicly available datasets, such as TCGA. While these datasets provide extensive and reliable information, they may not fully capture the heterogeneity of HNSCC across diverse populations. Thus, validation using larger and more diverse patient cohorts is necessary to enhance the generalizability of our findings. Second, although we performed correlation analyses between lncRNAs and clinical features, the molecular pathways involving these lncRNAs, particularly their roles in EMT and disulfidptosis, have not yet been experimentally validated. Functional studies are essential to elucidate their mechanistic contributions to HNSCC progression. Finally, while our findings suggest that the identified lncRNAs show potential as biomarkers or therapeutic targets, additional research is required to evaluate their clinical relevance, safety, and effectiveness in targeted therapies and immunotherapy.

To sum up, our research intertwines the molecular complexities of disulfidptosis with the clinical needs of HNSCC diagnosis and treatment. By illustrating the roles of DRLs and their implications for TME and immunotherapy, we have laid a pathway toward more personalized and efficacious therapeutic strategies in HNSCC management.

Author Contributions

Qi Chen: investigation, conceptualization, writing – original draft, methodology, data curation. Xiao Shi: writing – original draft, investigation, conceptualization, methodology, data curation. Yuanyuan Bao: data curation, software. Yue Chen: writing – review and editing, software, data curation, supervision.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1. Flow chart of our study.

Figure S2. Evaluating and validating for the prognostic value of DRLs signature. C‐index compared the prognostic accuracy of the risk score and clinical factors in training (A), testing (B), and all sets (C). (D) The nomogram showing risk and clinicopathological features for predicting 1‐, 3‐, and 5‐year OS in the total cohort (*, p < 0.05; **, p < 0.01; ***, p < 0.001). (E) Calibration curves corresponding to the nomogram for predicting the 1‐, 3‐, and 5‐year OS.

Figure S3. Survival analysis of low‐ and high‐risk patients with different clinical characteristics. (A) Patients with age ≤ 65 (p < 0.001). (B) Male patients (p < 0.001). (C) Patients with G1–2 (p = 0.012). (D) Patients with G3–4 (p < 0.001). (E) Patients with M0 (p < 0.001). (F) Patients with N0–1 (p = 0.023). (G) Patients with N2–3 (p < 0.001). (H) Patients with Stage III−IV (p < 0.001). (I) Patients with T1–2 (p = 0.016). (J) Patients with T3–4 (p < 0.001).

Figure S4. Sensitivity analysis of chemotherapeutic drugs based on the two risk groups and the three tumor clusters, including Alpelisib, Axitinib, AZD1208, Doramapimod, GSK269962A, JQ1, Sinularin, and Tozasertib (A—P).

Figure S5. PCA analysis for risk stratification in HNSCC utilizing DRLs Signature. (A) PCA analysis of whole genome expression data in HNSCC patients. B. PCA analysis focusing on disulfidptosis‐related genes. C. PCA analysis based on individual DRLs, where risk status differentiation remains inefficient. D. PCA analysis utilizing the combined DRLs signature.

Table S1. Primers used in this study.

Table S2. Differentially expressed lnRNAs in NP69 and FaDu cells.

Table S3. Differentially expressed mRNAs in NP69 and FaDu cells.

Table S4. The list of disulfidptosis‐related genes.

Table S5. The expression matrix of disulfidptosis‐related genes.

Table S6. The expression matrix of disulfidptosis‐related lncRNAs (yellow module).

Table S7. The risk scores for all samples based on three key disulfidptosis‐related lncRNAs.

Table S8. The tumor clusters for all samples based on three key disulfidptosis‐related lncRNAs.

Data Availability Statement

RNA‐sequencing data and clinical data of HNSCC were downloaded from The Cancer Genome Atlas (TCGA) (https://portal.gdc.cancer.gov/). Further inquiries can be directed to the corresponding author.