Head and neck cancer is the sixth most common malignancy that leads to considerable mortality, with >450,000 deaths reported worldwide in 2020 (International Agency for Research on Cancer, and World Health Organization, 2021). The incidence rate of head and neck squamous cell carcinoma (HNSCC), which is the most common type of head and neck cancer, is approximately 20 per 100000 people in the regions of South America, China, Europe, the Indian subcontinent, and among African Americans in the United States (Stenson, 2021). The well-known risk factors for HNSCC are chronic exposure to alcohol, smoking, different forms of chewing tobacco (such as betel palm), chronic oral trauma, and HPV infection (Gillison et al., 2015). Currently, surgical resection, radiotherapy, chemotherapy, targeted therapy, and immunotherapy have been developed to comprehensively treat HNSCC patients. However, the 5-year overall survival (OS) of HNSCC patients has remained at 50%, which has not significantly improved in the past decade (Torre et al., 2015; Bray et al., 2018). Moreover, mounting evidence highlights that the percentage of locoregional failure is approximately 40–50%, while distant failure is 20–30% in locoregionally advanced HNSCC patients (Samra et al., 2018; Muzaffar et al., 2021). Therefore, studies exploring novel therapeutic targets and developing novel prognostic models to identify patients with different prognoses are urgently required for HNSCC patients.

Iron is an important transition metal for maintaining the rapid proliferation and growth of cancer cells (Zhou et al., 2018). Moreover, iron participates in several important biological processes, such as oxygen transport, DNA synthesis, and ATP generation. However, excess intracellular iron accumulation can trigger reactive oxygen species (ROS), which cause lipid peroxidation and ferroptosis, a unique form of cell death (Bogdan et al., 2016; Fanzani and Poli, 2017). Ferroptosis is defined as an iron-dependent programmed cell death (PCD) that is dependent on ROS accumulation and lipid peroxidation, and the mechanism and morphology of PCD are distinct from other PCDs, such as autophagy, apoptosis, and necroptosis (Dixon et al., 2012; Stockwell et al., 2017). In recent decades, research on ferroptosis in tumors has rapidly increased, and this PCD has been indicated to be correlated with tumor origin, development, and treatment (Liang et al., 2019; Stockwell and Jiang, 2019). Ferroptosis regulatory genes, such as P53, DPP4, and GPX4, have been shown to be correlated with tumorigenesis and progression (Junttila and Evan, 2009; Liu et al., 2018; Enz et al., 2019). Moreover, ferroptosis has been suggested to regulate the sensitivity of tumor cells to radiotherapy and immunotherapy. Radiotherapy has been suggested to induce ferroptosis, which plays a crucial role in radiotherapy-mediated anticancer effects, and improve the sensitivity of tumor cells to ferroptosis inducers (Lang et al., 2019). Furthermore, interferon-gamma and T cells may sensitize tumor cells to ferroptosis (Wang et al., 2019). After treatment of tumor models with immune checkpoint inhibitors, ROS levels were significantly increased, while tumor size was significantly reduced (Mou et al., 2019). Radiotherapy and immunotherapy have been shown to promote lipid oxidation and ferroptosis in tumor cells owing to a synergistic effect (Lang et al., 2019). Nevertheless, the prognostic value and relationship of distinct ferroptosis-related genes with immunotherapy in HNSCC remain predominantly unknown.

In this study, we first integrated The Cancer Genome Atlas (TCGA), Genotype-Tissue Expression (GTEx), and ArrayExpress databases to construct and validate a novel prognostic risk model based on ferroptosis-related messenger RNA (mRNA). Moreover, internal and external validations were first used to assess the risk model accuracy compared with a previous study (He et al., 2021). To the best of our knowledge, the prognostic risk model and nomogram based on the model showed the best prognostic effect in the present study of risk models based on ferroptosis-related mRNA (He et al., 2021). In addition, the values of the model used to predict prognosis with other clinical parameters and immunotherapy of HNSCC patients were further explored. The correlation between the prognostic risk model and drug sensitivity in cancers was first analyzed.

Excluding communicable diseases, approximately 30% of premature deaths are attributable to cancer in adults aged approximately 30–69 years (WHO, 2020). As a malignant tumor with increasing incidence, the OS of HNSCC patients is unsatisfactory with the currently available treatments, and HNSCC recurrence with a 50% ratio in locally advanced disease further reduces the 5-year survival rate (Muzaffar et al., 2021). Therefore, identifying HNSCC patients with poor prognosis and developing more effective treatments to promote individualized treatment and improve clinical outcomes are urgently required.

Ferroptotic cell death, a relatively novel cell death mechanism, has been shown to improve the curative effect of radiotherapy and immunotherapy in cancer (Lang et al., 2019). However, few studies have investigated the relationship between ferroptosis-related genes and OS in HNSCC patients. In this study, a novel prognostic risk model for HNSCC was first built based on ferroptosis-related genes, and a clinical prognostic model was constructed to assess patient prognosis.

In the present study, we systematically investigated the correlation between ferroptosis-related genes and HNSCC. A total of 175 differentially expressed ferroptosis-related genes were identified in HNSCC based on TCGA, GTEx, FerrDb, and MSigDB, which provided a reliable foundation for future exploration. The 10 independent prognostic genes were screened to construct the risk score model based on univariate, LASSO, and multivariable logistic regression algorithms, which avoided overfitting and improved the clinical practicability of the model.

In the 10 ferroptosis-related genes, the coefficients of MAP1LC3A, MAP3K5, and SOCS1 were negatively correlated with the prognostic risk score, suggesting that these 3 genes might induce ferroptosis in HNSCC. However, the coefficients of the other 7 genes were positively correlated with the prognostic risk score, suggesting that these 7 genes might inhibit ferroptosis in HNSCC. Moreover, the potential gene functions and pathways of these 10 genes were further explored. Several results have been suggested to be correlated with ferroptosis in the top 10 KEGG and GO enrichment analyses. The top 10 results of GO analysis were primarily enriched in biological processes related to cellular responses to chemical stress, starvation, nutrient levels, and oxidative stress. Previous studies suggested that chemotherapeutic agents, such as sulfasalazine and cisplatin, might induce ferroptosis by acting on system xc- or GSH, which play a key role in ferroptosis development (Mou et al., 2019). Oxidative stress, as an important characteristic, has been identified as a crucial factor in ferroptosis induction (Mou et al., 2019). Moreover, the top 10 gene functions in the GO analysis are related to the cell response to nutrients and starvation, suggesting that deregulating cellular energetics might participate in ferroptosis. Energy stress has been reported to inhibit lipid peroxidation and ferroptosis (Lee et al., 2020). For the KEGG enrichment analysis, the top result was ferroptosis, which further suggested that the 10 genes might participate in ferroptosis and HNSCC development. Autophagy may promote ferroptotic death by ferritinophagy, lipophagy, or clockophagy (Liu et al., 2020). Moreover, cell sensitivity to ferroptosis might be affected by mTOR pathway regulation (Yi et al., 2020). In summary, the most significant gene functions and pathways from GO and KEGG analyses indicated that the mechanism of the 10 genes might be correlated with ferroptosis.

The majority of the 10 signature genes in the prognostic model have been shown to be involved in multiple cancers. However, the role of several genes in HNSCC prognosis remains to be explored. Ovarian tumor domain-containing ubiquitin aldehyde binding protein 1 (OTUB1) is an important deubiquitinating enzyme (DUB) that inhibits the ubiquitination of E2s following cleavage of the ubiquitin chains in target proteins (Wiener et al., 2012). OTUB1 has complex functions in various cancers, promotes tumor migration, and is a tumor suppressor that induces cell apoptosis and cell growth by regulating the DNA damage response (Sun et al., 2012; Herhaus et al., 2013; Kato et al., 2014). Moreover, OTUB1 may influence immune factor production. OTUB1 deletion downregulates the synthesis of the protective chemokine INF-γ, which is secreted by NK cells (Mulas et al., 2021). In addition, OTUB1 participates in metabolic reprogramming, which is an essential mechanism for activated T cell function (Pearce et al., 2013). Acyl-coenzyme A (CoA) synthetase long-chain family member 3 (ACSL3) regulates fatty acid metabolism and is a substrate for both β-oxidation and lipid synthesis after converting free long-chain fatty acids into fatty acyl-CoA esters (Coleman et al., 2002). Lipid metabolic reprogramming is one of the most prominent metabolic changes in cancer cells and has received increasing attention. As an important component of lipid metabolism, fatty acid deregulation might disturb the curative effect of radiotherapy and chemotherapy and affect immunotherapy for cancer patients (Wu et al., 2009; Han et al., 2019; Corn et al., 2020). ACSL enzymes, including ACSL3, have been suggested to induce apoptosis in a subset of TP53-deficient cancer cells (Yamashita et al., 2000; Mashima et al., 2005). Moreover, ACSL3 has been shown to be essential for mutant KRAS lung cancer tumorigenesis (Padanad et al., 2016). Mitogen-activated protein kinase 5 (MAP3K5) has been suggested to be involved in multiple biological activities, including stress-induced apoptosis and cell differentiation (Subramanian et al., 2004; Mizumura et al., 2006). As an important apoptosis signal-regulating kinase, MAP3K5 catalytic activity leads to differentially regulated apoptosis by inducing DNA damage, ROS, and tumor necrosis factor (Nagai et al., 2007). Moreover, MAP3K5 has been shown to play an important role in innate immune responses through the production of proinflammatory cytokine production (Matsuzawa et al., 2005). In general, ferroptosis-related genes are correlated with the human immune system, prompting subsequent analysis of immune function based on the present model.

After the model was constructed, K-M analysis and ROC curves were used to determine the effect of the model to identify HNSCC patients at different risks, and the test method revealed a satisfactory effect in predicting HNSCC patient prognosis.

The K-M analysis of our model showed a more optimistic trend; the difference in survival curve in the high- and low-risk groups was more significant compared with that of the previous study (He et al., 2021). Moreover, the prognostic risk model of HNSCC patients was more accurate than the models reported in certain previous studies with a preferable AUC for 1 (0.665 vs. 0.634 and 0.663), 3 (0.743 vs. 0.672 and 0.686), and 5 years (0.755 vs. 0.642 and 0.622) (Yang et al., 2020; Liu B. et al., 2021). Interestingly, the AUC values of our model increased annually, which further strengthened the prognostic effect of the model. Moreover, the internal validation in the test cohort and external validation in the entire cohort and the MTAB-8588 dataset further improved model stringency. Overall, the practicability of the model was validated using multiple methods.

To further explore the clinical value of the model in HNSCC patients, the risk score was integrated with the clinical parameters of HNSCC patients. The risk score based on the model was an independent prognostic parameter that had synergistic effects with other independent prognostic characteristics to further improve the value of the nomogram to calculate individual prognoses, which were validated by the C-index, ROC curve, calibration plot, and DCA. The nomogram designed to predict HNSCC patient prognosis was more accurate than the models reported in a previous study with a superior C-index (0.752 vs. 0.640) and AUC for 1 (0.729 vs. 0.597), 3 (0.828 vs. 0.706), and 5 years (0.853 vs. 0.645) (He et al., 2021). Surprisingly, the AUC values of the nomogram at 1, 3, and 5 years increased annually, and the trend was consistent with the AUC values of the risk score model, which further strengthened the reliability of the risk score model and nomogram. Moreover, internal and external validation revealed optimistic effects for the nomogram.

Recently, tumor immunotherapy has been considered a breakthrough in oncotherapy. However, the response to immunotherapy differs between cancers and within cohorts with the same cancer (Ventola, 2017). Thus, the TMB level was developed to predict the response to immunotherapy in cancer patients, and high TMB has been shown to be positively correlated with the curative effect of immunotherapy (Galuppini et al., 2019). However, it has been reported that a high TMB fails to predict the immune checkpoint blockade response across all cancers (McGrail et al., 2021).

After analyzing the correlation between the risk score and TMB in HNSCC patients, immune cell infiltration in the tumor microenvironment and the differences in immune checkpoints of high- and low-risk HNSCC patients were further investigated to explore the curative effect of immunotherapy based on the risk score model. The relationship between ferroptosis and the immune microenvironment is complicated. Cancer cells with ferroptosis can increase immunogenicity by releasing immune-stimulating signals, which allow immune cells, such as macrophages and dendritic cells (DCs), to locate the cancer cell site (Friedmann Angeli et al., 2019; Tang et al., 2021). Early ferroptotic cancer cells can elicit a vaccination-like effect by promoting phenotypic DC maturation (Efimova et al., 2020). The increased immunogenicity of cancer cells may enhance the efficacy of immune checkpoint inhibitors. Moreover, immune cells, such as CD8 + cells, can trigger lipid peroxidation and cause ferroptosis in tumor cells, which suggests that immune checkpoint inhibitors targeting CD8 + cells might act synergistically with ferroptosis (Wang et al., 2019). In addition to predicting HNSCC patient prognosis, differences in the immune microenvironment based on the prognostic risk model might be beneficial for developing individualized immunotherapy and improving the curative effect of HNSCC patients.

Moreover, the analysis of drug sensitivity based on the model suggested the possibility of targeting the 10 gene risk signatures and reducing the prognostic risk of HNSCC patients. The drugs might regulate the ferroptosis level of cancer cells by targeting the 10 ferroptosis-related genes. ACSL3 was found to contribute the most to drug sensitivity. ARRY-162 (MEK162), as the drug with the greatest correlation with ACSL3, is a highly selective MEK1/2 inhibitor that could offer a new combination therapy with ferroptosis and immunotherapy (Erkes et al., 2020; Gagliardi et al., 2020).

The present study has several limitations. Certain in vitro and in vivo experiments, such as the in-depth molecular mechanisms of the ferroptosis-related genes, needed to construct the prognostic model require further verification in experimental studies. Moreover, the study was based only on research data from public databases, which might have contributed to the selection bias. Thus, a multicenter and large-scale study is required to further validate the clinical utility of our model.

In summary, the present study systematically investigated the correlation between ferroptosis-related genes and HNSCC patient prognosis, and a novel prognostic model of HNSCC patients based on 10 ferroptosis-related genes was established and validated. Furthermore, the values of the model to be used in predicting prognosis with other clinical parameters, immunotherapy, and drug sensitivity of HNSCC patients were positive, indicating that the novel ferroptosis-related gene signatures might be beneficial in developing individualized treatments, thereby improving the OS of HNSCC patients.

The datasets used and analysed for this study were obtained from TCGA 487 (https://portal.gdc.cancer.gov/) and ArrayExpress database 488 (https://www.ebi.ac.uk/arrayexpress/).

DH and WH: conceptualization and project administration. SL and LX: methodology and investigation. SL, LX, LH, and LC: formal analysis. DH: writing – original draft preparation. WH: writing – review and editing and supervision. DH, LH, and MY: visualization. MY and SL: funding acquisition. All authors support publishing, contributed to the article, and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.