Introduction
In advanced stages of head and neck squamous cell carcinoma (HNSCC), an aggressive malignancy, mortality, and morbidity are high [
1]. With approximately 800,000 new cases and 430,000 new deaths worldwide each year, HNSCC is one of the most common causes of cancer-related deaths [
2,
3]. Surgery, radiotherapy, chemotherapy, and combination therapy are still the standard treatment methods for HNSCC [
4]. Unfortunately, due to the frequent local and distant metastasis of HNSCC and the resistance to chemotherapeutics, there is currently no entirely satisfactory treatment for advanced HNSCC, which leads to a high mortality rate in patients [
5,
6]. Although in recent years, the clinical application of immune checkpoint inhibitors (ICIs), such as nivolumab and pembrolizumab, has completely changed the treatment outcome of metastatic or recurrent HNSCC. However, its objective response rate is still 20%. The risk of immune-related adverse events (irAEs) that can contribute to severe or fatal toxicities for patients hinder the wide application of ICI therapy [
7‐
9]. In the combination therapy of ICIs, the incidence of side effects is higher than that of ICIs monotherapy, and the side effects also occur faster [
10,
11]. In addition, targeting CD44, a marker for cancer stem cell-like cells (CSCs), has recently been regarded as a promising therapeutic target for HNSCC treatment; however, further clinical applications are still being explored [
12,
13]. As a type of RNA without protein-coding ability, long-chain non-coding RNA (lncRNA) not only participates in gene regulation processes, such as regulating mRNA splicing, chromatin, histone remodeling, and transcription regulation [
14], but also participates in biological regulation processes such as tumor occurrence, development, and metastasis [
15,
16]. Increasing evidence shows that lncRNAs may play a vital role in the proliferation, migration, and treatment of HNSCC [
17‐
20]. In recent years, lncRNAs have been identified to facilitate the resistance to cisplatin, paclitaxel, 5FU, and other chemotherapeutic drugs in various ways [
21‐
25]. Overexpression of lncRNA-UCA1 can protect the expression of PDL1 from miRNAs, thereby upregulating the expression of PDL1 and ultimately promoting the immune escape of GC cells [
26]. Given higher tissue specificity and easier detection than mRNA, lncRNA is more suitable as a biomarker for tumor diagnosis and prognosis [
27,
28]. Consequently, a growing number of studies note that ir-lncRNAs signals can predict the prognosis and treatment sensitivity of various cancers, such as melanoma, lung adenocarcinoma, and endometrial cancer [
29‐
31]. Unfortunately, most of these predictive signatures are combinations of single lncRNAs. In contrast, the dual biomarker combination is superior to a single marker in terms of the accuracy of the cancer diagnosis model [
32]. To achieve higher accuracy, it is indispensable to develop a model based on the combination of double lncRNAs for HNSCC prognosis. Due to the technical differences between different platforms, it is difficult for the detected gene expression levels to be of the same standard [
33]. Recently, novel gene pair signatures have been developed to circumvent this problem subtly. By comparing the expression of two genes in each patient, the researchers assigned a value of 1 (expression of gene A >expression of gene B) or 0 (expression of gene A <expression of gene B) to this gene pair [
33]. It is evident that such a combination of lncRNAs meets the need for a dual biomarker combination with higher accuracy.
LncRNAs have been demonstrated to regulate cancer progression through immune regulation, mainly by changing the immune microenvironment. Xiong et al. found that with the upregulation of lncRNA-POU3F3 expression in cancer-related cells, circulating regulatory T cells (Tregs) increased in gastric cancer patients [
34]. In vitro experiments further confirmed that lncRNA-POU3F3 promoted Treg differentiation by activating TGFβ signaling, thereby promoting the proliferation of tumor cells [
34]. In hepatocellular carcinoma (HCC), Jiang et al. also found that the highly expressed lncRNA-EGFR stimulated Treg production and continuous activation, resulting in the suppression of cytotoxic T cells [
35]. In addition, RP11-284N8.3.1 and AC104699.1.1 are related to T cell activation and differentiation and are associated with the increasing survival rate of ovarian cancer [
36].
We aspire to obtain differentially expressed immune-related lncRNAs (DE-irlncRNAs) by combining the HNSCC RNA sequencing data obtained from the TCGA database and the immune-related genes obtained from the ImmPort database. The effective paired DE-irlncRNAs were then screened to find the prognostic-related DE-irlncRNAs pairs (PR-DE-irlncRNAs pairs). Following this, a prognostic model was designed based on 20 pairs of PR-DE-irlncRNAs, and the sample risk score was calculated. We performed Risk plots, Receiver Operating Characteristic (ROC) curves, Kaplan–Meier (K-M) curves, Cox regression analysis, and clinical features subgroup analysis to verify the accuracy of the model’s predictive ability. In parallel, we investigated their possible mechanisms of action in HNSCC by gene set enrichment analysis (GSEA), tumor immune microenvironment (TIME) analysis, and mutation analysis. Finally, to determine whether prognostic models can be applied to treatment guidance and prognosis evaluation for patients with HNSCC, immunophenotypic score (IPS) analysis, drug sensitivity analysis, ICIs/m6A/multidrug resistance-related genes analysis and nomogram were used.
Discussion
Currently, most of these predictive signatures were combinations of single lncRNAs [
50,
51]. However, in contrast, the dual biomarker combination has been shown to outperform a single marker in terms of the accuracy of cancer diagnostic models [
32]. Inspired by the strategy of matching immune-related genes, we paired the DE-irlncRNAs into lncRNA pairs that was not affected by the expression level. We constructed a prognostic model based on 20 PR-DE-irlncRNAs pairs. First, we selected the PR-DE-irlncRNAs pairs that were effectively matched based on the data obtained from the TCGA database and subsequently developed an effectively prognostic model for HNSCC patients. The predictive value of the model was then verified through a variety of methods. Moreover, our model was closely related to ICIs/m6A/multidrug resistance-related genes’ expression, with excellent clinical applicability. Finally, our analysis revealed that immune cells, immune function, and TP53 mutations may be implicated in the progress of HNSCC.
As a current research hotspot, irlncRNAs have been used in the signatures of various cancers. AC007038.1|AC084018.1, ZNF687−AS1|AL354733.3, TMPO−AS1|ATP1B3−AS1, AC132192.2|AC004148.1, LINC01063|AC116914.2, RUSC1−AS1|AC004687.1, LINC00944|AC004687.1, CHKB−DT|AL365330.1, MIR924HG|MIR9−3HG, and AL365330.1|MIR9−3HG in model played negative significant roles, while AL390719.2|AL133243.2, AC096992.2|HOXC−AS1, C5orf66−AS1|KDM4A−AS1, C5orf66−AS1|AP000251.1, AC106820.3|KDM4A−AS1, PTOV1−AS2|SNHG25, PTOV1−AS2|AL132712.1, PTOV1−AS2|LINC00205, AC008735.2|SNHG25, and AL360181.2|AC098487.1 played positive roles in HNSCC patients’ survival. The PR-DE-irlncRNAs used in this study to establish a prognostic model have also been determined to possess a brilliant predictive value for the prognosis of patients in other cancers.
Xu et al. identified AL390719.2 as one of the key prognostic lncRNAs for both 10- and 5-year survival rates in colorectal cancer [
52]. C5orf66-AS1 prevents oral squamous cell carcinoma by inhibiting cell growth and metastasis [
53] and has been verified as a biomarker for various cancers [
54,
55]. Studies have demonstrated that AL354733.3 exhibits a positive correlation with autophagy genes and can serve as an independent prognostic indicator for OSCC patients [
56]. TMPO‐AS1 regulates the proliferation and migration of triple‐negative breast cancer cells by modulating transforming growth factor‐β and E2F signaling pathways [
57]. Moreover, TMPO-AS1 has the potential to enhance LCN2 transcriptional activity by binding to transcription factor E2F6, thus stimulating ovarian cancer progression [
58]. Feng et al. revealed that AC116914.2 is significantly related to the expression of PD-L1 in primary head and neck squamous cell carcinoma [
59]. Cheng et al. stated that
LINC01063 is a risk-related autophagy-related lncRNA with a poor prognosis in colorectal cancer [
60], which was confirmed again in the study of Zhou et al. [
61]. Ye et al. reported that
AC004687.1 is significantly related to recurrence-free survival of hepatocellular carcinoma patients [
62]. It was shown that RUSC1-AS1 correlated with the prognosis of various cancers [
63‐
65]. For instance, it activates NOTCH signaling via the hsa-miR-7-5p/NOTCH3 axis, promoting the proliferation and reducing the apoptosis of HCC cells [
66]. Moreover, RUSC1-AS1 promotes the aggressiveness of cervical cancer in vitro and in vivo by upregulating miR-744-Bcl-2 axis output [
67]. De Santiago et al. showed that
LINC00944 is in response to ADAR1 up- and downregulation in breast cancer cells, and the low expression of
LINC00944 is correlated to poor prognosis in breast cancer patients [
68]. MIR9-3HG was identified as a key lncRNA with diagnostic and prognostic value for HNSCC [
69] and liver hepatocellular carcinoma [
70]. LncRNA HOXC-AS1 promotes nasopharyngeal carcinoma progression by sponging miR-4651 to upregulate FOXO6 [
71]. Deng et al. identified signature lncRNAs that could serve as predictors of the OS rate of hepatocellular carcinoma [
72]. PTOV1-AS2 was used to construct a tp53-associated nomogram to predict the OS in patients with pancreatic cancer [
73]. High expression levels of the
LINC00205 correlate with a better OS in pancreatic cancer [
74]. The study of Yang et al. revealed that OS was significantly shortened in the SNHG25 high expression group and significantly upregulated in clear cell renal cell carcinoma (ccRCC) tissues [
75]. Another study identified AL360181.2 and AC008735.2 as potential prognostic markers to construct a model for predicting the prognosis of ccRCC patients [
76].
This study showed a significant correlation between the patient’s risk score and prognosis, with different survival probabilities observed between different risk subgroups. Furthermore, the clinical stratification analysis showed that risk score still maintained the ability to distinguish the prognosis of patients with high- and low-risk across different subgroups. These all highlighted the accuracy and optimality of the predictive ability of the prognostic model.
We analyzed 230 differential genes between high-risk and low-risk populations. We discovered that the enriched molecular functions were related to immunity, and immune-related functions were also observed in the enriched cellular components and molecular functions. This indicated a close relationship of our model with immunity. Therefore, we further explored the relationship between tumor immunity and our risk model from the perspective of the immune microenvironment, including immune infiltrating cells and immune function. Immune infiltrating cells in tumors played an essential role in the occurrence and development of tumors, ultimately impacting patient prognosis [
77]. Therefore, understanding tumor immune infiltrating cells could explore the prognosis of tumor patients and the new direction of HNSCC treatment in the future. We observed that B cell, T cell CD8+, myeloid dendritic cell, B cell memory, B cell plasma, T cell gamma delta, Tfh cell, Treg cell, mast cell activated, and NK cell activated negatively correlated with the risk score, while mast cell resting and eosinophil were positively correlated. Even further, the significant differences in cell content between high- and low-risk groups supported the credibility of the results. B cells played an important role in anti-tumor immunity. The presence of NK cells and NK T cells in most solid tumors often meant a good prognosis [
78].
In addition, the scores of CCR, APC co-stimulation [
79], type II IFN response [
80], checkpoint, cytolytic activity, HLA [
81], inflammation-promoting, T cell co-stimulation, and T cell co-inhibition were observed to be negatively correlated with risk score (
p < 0.05). Among them, higher scores of checkpoint, cytolytic activity, HLA, inflammation-promoting, T cell co-stimulation, and type II IFN response were observed in the low-risk group (
p < 0.05). CCR 5 could recruit MDSC to tumors closely related to tumor immunity [
82]. Type II IFN-γ has the potential to induce tumor cell apoptosis and regulate cancer immune activity [
80]. Previous studies have demonstrated that primary colorectal cancer and corresponding metastases usually exhibit downregulation or loss of HLA-I expression [
81]. Tumor may drive the immune escape by changes in HLA expression (or by other means) [
83], which might be developed into auxiliary tumor markers in the future. These conclusions implied that a multitude of immune infiltrating cells in the low-risk group may participate in the anti-HNSCC response through a series of immune functions, ultimately leading to improved patient prognoses.
Among the mutations in HNSCC, TP53 mutation was found to be the most common mutation in both high-risk and low-risk groups. It was well-established that the number of p53-regulated lncRNA increased rapidly, indicating their widespread involvement downstream of p53 activation [
84]. Transcription factor p53 was a most prominent human tumor suppressor that played an essential role in cellular responses to DNA damage stimuli [
85]. As anticipated, patients in the TP53 mutation group had higher risk scores and worse prognoses. Moreover, we also found that the contents of T cell CD8, T cell CD4 memory activation, Tfh cell, Treg cell, macrophage M1, mast cell rest, and mast cell activation were significantly lower in the TP53 mutation group. Conversely, there was a significant increase in the contents of T cell CD4 memory rest, macrophage M0, and dendritic cell rest within this same group. In head and neck cancers, the presence of TP53 mutations was associated with lower estimates of various immune infiltrating cells, such as T, B, and NK cells [
86]. Increased levels of M0 macrophages were associated with poor clinical outcomes in lung adenocarcinoma [
87]. Furthermore, relevant studies have shown that M0 macrophages promoted malignant progression and were affected by tumor development [
87].
In addition, we found that the expression of PDL1 (CD274) decreased significantly in the TP53 mutation group, which may lead to an increase in the tumor and cancer stem cell phenotype in cholangiocarcinoma. At the same time, it was found that low CD274 had high tumor initiation potential [
88]. At present, the PD-L1 signal contributes to human cancer immune escape, thereby blocking PD-L1 has been applied to clinical cancer treatment [
89,
90]. The significant efficacy of PD-L1 blockers in cancer immunotherapy was expected to control cancer by regulating the expression of PD-L1 [
91,
92], which has been shown to have a potential predictive effect in melanoma, non-small-cell lung cancer, renal cell carcinoma, prostate cancer, or colorectal cancer [
93,
94]. These results suggested that in HNSCC, TP53 mutation may promote the progress of HNSCC by suppressing these immune cells and inhibiting anti-tumor immunity, ultimately leading to a poor prognosis. Furthermore, in HNSCC, patients with TP53 mutations may benefit less from PD-1 treatment. Nowadays, immunotherapy is an emerging strategy for anti-tumor therapy. Therefore, our study investigated the relationship between immune-related genes and the prognosis model. We found that the expression of CD44 and CD276 had a significantly positive correlation with a risk score. As a CSC marker of HNSCC, CD44 participates in the DNA damage response of G2/M phase arrest. Overexpression of CD44 provided relative protection for HNSCC cells against cell death response [
95]. CD276, a member of the B7 family, was considered a factor that regulated antigen-specific T cell immune response through costimulatory and co-inhibitory receptors. The expression of CD276 was negatively correlated with the number of tumor-infiltrating CD8 + T cells, and the upregulated expression was related to the poor prognosis in esophageal cancer [
96,
97]. In our study, patients in the high-risk group with more CD44 and CD276 expression exhibited a worse prognosis, which was consistent with the conclusions of these related studies.
More and more evidence showed that m6A RNA methylation played a crucial role in tumorigenesis. m6A modification of some genes may result in changes in mRNA behavior and expression, thus accelerating tumor development, whereas the lack of m6A modification of other genes might still lead to tumor progression [
98]. We studied the relationship between m6A-related genes and prognostic models, from which the expression level of HNRNPC was positively correlated with the risk score. Overexpression of HNRNPC was found in the central regulators of colon rectum cancer cells and cancer progression-related genes [
99]. The expression of HNRNPC might be related to poor prognosis, similar to our findings, providing valuable insights into the study m6A-related genes in HNSCC.
Our study observed that IPS-CTLA4 and IPS-PD1 + CTLA4 were negatively correlated with a risk score. Some experiments showed that in HNSCC, the scores of IPS with CTLA4 blocker, IPS with CTLA4, and PD1/PDL1/PdL2 blocker in the low-risk group were significantly higher than those in the high-risk group, which was consistent with our experimental results [
100]. This meant that our prognostic model had a certain predictive value for the efficacy of patients receiving corresponding immunotherapy. Following calculation, we found that the IC50 of methotrexate was positively correlated with the risk score, while the IC50 of doxorubicin and docetaxel was negatively correlated with the risk score. From this, our prognostic model possesses a certain guiding significance for the use of chemotherapeutic drugs. Nomogram could provide personalized prognostic assessment for both surgeons and patients, serving as a reference for treatment planning [
101]. The Nomogram drawn according to the relevant data from the training samples had an excellent ability to predict survival, which provided a new insight into the prognosis of HNSCC.
It is worth acknowledging that our research had certain shortcomings and limitations. Firstly, due to the lack of data sets containing complete lncRNA and mRNA transcription data in other shared databases, we only relied on a distinct data set from TCGA to build and validate our model, which may lead to randomness in the results. The lack of an external validation set posed a challenge to the reliability of model performance. To compensate for this limitation, three data sets obtained by randomly splitting the TCGA data set were used to thoroughly verify the model’s performance. In addition, it is noteworthy that the limited number of normal samples also poses challenges to the accuracy of differential analysis results. Furthermore, due to the large number of lncRNAs in the model and limited experimental conditions, it was challenging for us to perform qRT-PCR to verify the differential expression of these lncRNAs. However, we obtained many valuable conclusions through multi-perspective analysis, regarding both clinical application and the underlying mechanism of HNSCC progress. Nevertheless, these conclusions need to be further verified in subsequent experimental studies.