Development and validation of an immune-related gene signature for the prognostic and immune landscape prediction in head and neck squamous cell carcinoma by integrated analysis of machine learning and Mendelian randomization

Introduction

Head and neck squamous cell carcinoma (HNSCC) is acknowledged as the sixth most prevalent malignancy worldwide, presenting in a variety of anatomical regions, encompassing the oral cavity, oropharynx, hypopharynx, and larynx (1-5). A wide range of etiological factors have been implicated in the development of HNSCC, such as intake of alcohol, utilization of tobacco, and infection by human papillomavirus (6-8). Due to its occurrence in anatomically concealed sites, early diagnosis of HNSCC remains difficult, with most cases being detected only at more advanced stages (4). Prognosis for advanced-stage HNSCC continues to be poor, regardless of the treatment modality used, including surgical intervention, radiation therapy, or chemotherapy (9). Recently, immunotherapy, specifically immune checkpoint inhibitors (ICIs), has emerged as a promising treatment strategy for HNSCC (10).

The initiation and advancement of tumors are closely associated with the biological characteristics of tumor cells, the immune system, and the immune microenvironment (11). The tumor microenvironment (TME) is composed of a dynamic, bidirectional, and complex network that includes diverse stromal cell types, immune cells, and extracellular components. TME is essential in tumor advancement by influencing local drug resistance, cancer metastasis, and immune evasion (12-15). Immune cells are particularly susceptible to the effects of TME, with numerous immune cells and immune-related genes (IRGs) being pivotal in their regulation (16). For instance, regulatory T cells (Tregs) are known for their secretion of immunosuppressive cytokines, encompassing transforming growth factor-beta and interleukin-10, as well as their expression of cytotoxic T lymphocyte-associated protein 4, which markedly correlates with tumor progression in HNSCC (17).

Immunotherapy has demonstrated significant therapeutic efficacy across various cancers, encompassing lung, head and neck, bladder, renal cancer, and melanoma (18,19). Current therapeutic strategies encompass ICIs, tumor vaccines, dendritic cell immunotherapy, antibody-drug conjugates, and adoptive T-cell transfer therapy (19). Nivolumab and pembrolizumab, both monoclonal anti-PD-1 antibodies, were among the first ICIs approved for treating recurrent HNSCC (20). Nevertheless, immunotherapy response rates remain under 20% in non-selected HNSCC cases (21,22). As a result, discovering new biological indicators and establishing advanced predictive frameworks remains essential for anticipating patient outcomes and directing individualized therapeutic strategies (23).

In recent years, machine learning algorithms have been widely utilized for the construction of predictive models (24-26). Recent advancements in machine learning have facilitated novel approaches for the integration of large-scale omics data, thereby enhancing the identification of robust biomarkers with substantial predictive capabilities. Nonetheless, the exclusive application of machine learning may remain constrained by confounding variables and biases intrinsic to observational data (27,28). To address these challenges, we employed a combination of 10 distinct machine learning algorithms alongside Mendelian randomization (MR) analysis. By integrating these methodologies, our study sought to transcend the limitations inherent in traditional observational studies, thereby facilitating a more precise and reliable identification of biomarkers for HNSCC. This approach not only enhances the robustness of our findings but also holds the potential to uncover novel therapeutic targets by differentiating between mere associations and causal relationships (27).

This research conducted a comprehensive examination to analyze IRG expression patterns and their prognostic significance in HNSCC patients. Subsequently, machine learning techniques were utilized to develop an IRGs-based signature for calculating a risk score to forecast overall survival (OS) and assess the biological characteristics of patients classified into low- and high-risk categories. The tumor immune microenvironment, immune cell infiltration, and reaction to immunotherapy were also investigated to uncover prospective approaches for specific HNSCC treatments. Furthermore, MR was applied to explore whether these IRGs correspond to genetic variants contributing to the increased risk of developing HNSCC. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2024-2665/rc).

Methods

Data collection

The methodology of the study is similar to our previous study (29). Transcriptome profiles, quantified as fragments per kilobase million measurements from The Cancer Genome Atlas-Head and Neck Squamous Cell Carcinoma (TCGA-HNSC) initiative, encompassing 522 HNSCC and 44 normal tissue specimens, were procured from the TCGA database (https://portal.gdc.cancer.gov/). The GSE65858 dataset, sourced from the Gene Expression Omnibus (GEO) database, included transcriptome data from 270 HNSCC patients. Clinicopathological features from both cohorts were downloaded and subsequently subjected to analysis. After excluding cases with missing survival data or survival times under 30 days, this study included 511 HNSCC cases from TCGA as the training cohort and 267 from the GEO as the test cohort. Clinicopathological features of the HNSCC patients in this study are shown in Table 1. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

IRG expression in HNSCC

A sum of 2,483 IRGs was identified from the IMMPORT database (https://immport.org/). Differences in IRG expression between HNSCC and normal samples were conducted utilizing the “limma” R package. Prognostic IRGs were ascertained through univariate Cox regression analysis (UCRA) executed with the “survival” R package, employing a threshold of P<0.001.

Integrative machine learning algorithms developed an IRG signature

An IRG-based predictive model was executed utilizing 10 integrative machine learning algorithms for HNSCC analysis. The model exhibiting superior C-index values across training and test cohorts was deemed optimal. Upon identifying the optimal predictive IRG model, the IRGs and their respective coefficients were extracted. By applying gene expression data in conjunction with the coefficients, a risk score for every HNSCC patient could be calculated. Individuals were subsequently split into two cohorts—low-risk and high-risk—utilizing the computed risk score’s median threshold. R packages such as “seqinr”, “plyr”, “survival”, “randomForestSRC”, “glmnet”, “plsRcox”, “superpc”, “gbm”, “mixOmics”, “survcomp”, “CoxBoost”, “survivalsvm”, “BART”, “snowfall”, “ComplexHeatmap”, and “RColorBrewer” were employed during these analytical procedures.

Verification of the developed risk model

The OS differences between low- and high-risk cohorts were compared using Kaplan-Meier survival curves and the log-rank test. The investigation proceeded with univariate and multivariate Cox regression examinations to ascertain whether risk scores and various clinicopathological factors (age, sex, grade, and stage) functioned independently as OS prognostic indicators. The study generated receiver operating characteristic (ROC) curves at 1-, 3-, and 5-year intervals to gauge the prediction capabilities of the IRG signature. Predictive precision increases as the area under the ROC curve (AUC) approaches unity. The study also analyzed risk scores along with patient demographics and clinical factors. These included age, sex, grade, and stage. ROC curves and C-index measurements were used for evaluation. These clinical parameters were subsequently integrated into stratified survival examinations. The analytical process utilized R software packages, specifically “survival”, “survminer”, “timeROC”, “dplyr”, “rms”, and “pec”.

Tumor mutation burden (TMB) analysis

A waterfall plot illustrated the 15 most commonly altered genes in the two risk cohorts. The aggregate mutation frequencies for every HNSCC specimen underwent examination. Subsequently, survival analysis was performed to assess and contrast OS measurements among individuals with varying risk scores and TMB frequencies, using “survival” and “survminer” packages.

Development and confirmation of the nomogram

A nomogram was constructed by integrating factors like grade, sex, clinical stage, age, T- and N-stages, and risk score, providing a visual tool to predict patient survival outcomes (30). The efficacy of the nomogram was evaluated using ROC curves and calibration plots. The R packages “survival”, “regplot”, “rms”, “survcomp”, “survminer”, and “timeROC” were employed for these analyses.

Immune landscape analysis and immunotherapy prediction

Gene set enrichment analysis (GSEA) pathway evaluation was implemented using R packages, including “limma”, “org.Hs.eg.db”, “clusterProfiler”, and “enrichplot”. The investigation computed tumor purity, ImmuneScore, StromalScore, and ESTIMATEScore for HNSCC subjects through the “ESTIMATE” package. Subsequently, immune functionality between groups underwent comparison via single-sample GSEA (ssGSEA). The CIBERSORT technique was applied to quantify the distribution of 22 immune cell subtypes within every specimen, seeking to understand the link between risk scores and immune cell presence. Additionally, the study contrasted immune checkpoint-associated and human leukocyte antigen (HLA)-associated gene expression between risk groups. Moreover, the Tumor Immune Dysfunction and Exclusion (TIDE) score and Immunophenoscore (IPS) were implemented to assess immunotherapeutic responsiveness in individuals with HNSCC. Reduced TIDE scores and elevated IPS values indicate enhanced immunotherapy response potential.

MR study of model gene expression and the onset of oral cavity carcinoma

Expression quantitative trait loci (eQTLs) represent genetic regions that modulate gene expression. By utilizing eQTLs as instrumental variables, alterations in phenotype stemming from gene expression variations can be investigated. eQTL information of the model gene, in conjunction with genetic information for oral cavity carcinoma, was obtained from the IEU Open genome-wide association studies (GWAS) database of the MRC Integrative Epidemiology Unit. Upon evaluating factors including sample size, sequencing coverage, and ethnicity, the dataset labeled “ieu-b-4961” was chosen as the outcome data for this study. A two-sample MR methodology, executed via R software, explored the causative relationship between eQTLs of genes and oral cavity carcinoma progression. Multiple approaches, encompassing the inverse variance weighted (IVW) technique, weighted median (WM), simple median (SM), weighted median estimator (WME), and MR-Egger regression, were implemented to confirm the findings. The Cochran Q test assessed heterogeneity across results. When P<0.05, it suggested heterogeneity among single nucleotide polymorphisms (SNPs), and causative inference employed the random effects model in IVW. To evaluate potential pleiotropy, researchers conducted an MR-Egger intercept test. The regression intercept’s P value measured horizontal pleiotropy extent, with P<0.05 denoting the lack of horizontal pleiotropy. Lastly, a leave-one-out sensitivity assessment determined whether individual SNPs influenced the combined IVW estimate. Notable shifts in MR outcomes following the exclusion of specific instrumental variables indicated result sensitivity to those variables. Results are shown as odds ratios (ORs) accompanied by 95% confidence intervals (CIs), and P<0.05 indicates statistical significance.

Statistical analysis

Statistical analyses were executed employing R 4.2.2 software, generating data visualizations. Variable comparisons utilized the Wilcoxon signed-rank test or Chi-squared test. The links among variables underwent Pearson correlation analysis. Statistical significance was set at P value <0.05. All P values were two-sided.

Results

Expression of IRGs in HNSCC and establishment of the IRG signature

After obtaining data from the TCGA-HNSC dataset, differential expression analysis was performed, and the expression profiles of the top 25 enhanced and the top 25 suppressed genes in HNSCC and normal tissues were represented through a heatmap (Figure 1A). Subsequently, 18 prognostic IRGs linked to OS were ascertained utilizing UCRA (Figure 1B). These 18 IRGs were then subjected to an integrative process incorporating 10 machine learning algorithms, which facilitated the development of the IRG signature. Figure 2A presents the C-index values for each distinct prediction model across both cohorts. The model using the random survival forest (RSF) + CoxBoost method emerged as superior, yielding the highest mean C-index of 0.621. A list of 10 IRGs was selected to establish the IRG signature, including DKK1, CCR7, STC2, OLR1, PTX3, PLAU, ZAP70, SPINK5, PSMD2, and VEGFC.

Figure 1 Expression of IRGs in TCGA-HNSC. (A) Heatmap illustrates the differentially expressed IRGs between HNSCC and normal samples. (B) The prognostic significance of IRGs is contingent upon univariate Cox proportional hazards regression (P<0.001). HNSCC, head and neck squamous cell carcinoma; IRGs, immune-related genes; TCGA-HNSC, The Cancer Genome Atlas-Head and Neck Squamous Cell Carcinoma.

Figure 2 Prognostic IRG signature developed by integrative machine learning study. (A) The C-index for every prognostic model was established utilizing 10 machine learning methods and different combinations in the training and test cohorts. Kaplan-Meier survival curves for OS in the training cohort (B) and test cohort (C). Univariate (D) and multivariate (E) analyses indicated that the risk score served as an independent prognostic indicator in HNSCC patients. (F) The ROC curve of the IRG signature for 1-, 3-, and 5-year OS. (G) ROC curve comparing risk score with various clinicopathological features. (H) C-index of risk score and various clinicopathological features. AUC, area under the curve; GEO, Gene Expression Omnibus; HNSCC, head and neck squamous cell carcinoma; IRG, immune-related gene; OS, overall survival; ROC, receiver operating characteristic; RSF, random survival forest; SuperPC, supervised principal components; SVM, support vector machine; TCGA, The Cancer Genome Atlas.

Prognosis value of individuals with HNSCC based on IRGs signature score

HNSCC patients were categorized into low-risk and high-risk cohorts utilizing the median risk score. The Kaplan-Meier survival curves showed that high-risk patients had worse OS than low-risk patients in both the training (Figure 2B) and test cohorts (Figure 2C). Univariate (Figure 2D) and multivariate (Figure 2E) studies revealed that the risk score, age, and stage were identified as independent prognostic determinants for individuals with HNSCC. The AUC values for 1-, 3-, and 5-year OS ROC curves were 0.694, 0.731, and 0.656, respectively (Figure 2F). In comparing the 3-year AUC value (0.731) with those for other features, the IRG signature exhibited enhanced predictive efficacy in comparison to other clinical variables (Figure 2G). This result was confirmed by C-index (Figure 2H). Furthermore, survival outcomes were compared between two cohorts across diverse clinical parameters, with individuals in the low-risk cohort showing more favorable prognoses concerning age, sex, grade, and stage (Figure 3A-3D).

Figure 3 Subgroup survival analysis. Kaplan-Meier survival curves for age (A), sex (B), histologic grade (C), and clinical stage (D) in the TCGA-HNSC dataset. TCGA-HNSC, The Cancer Genome Atlas-Head and Neck Squamous Cell Carcinoma.

Correlations between TMB and the risk model

Somatic mutation data were utilized to compute the mutation frequencies of all genes. The alteration rates of the top 15 genes exhibiting the highest mutation rates were depicted using a waterfall plot for both low- (Figure 4A) and high-risk cohorts (Figure 4B). A shorter OS was observed in patients with elevated TMB when compared to those with reduced TMB (Figure 4C). The findings of the stratified subgroup survival evaluation suggested that patients with increased TMB exhibited reduced OS versus those with decreased TMB, regardless of the risk cohort (Figure 4D).

Figure 4 TMB analysis. The waterfall plot depicts the top 15 mutated genes in the low-risk cohort (A) and high-risk cohort (B). (C) Survival assessment between high and low TMB cohorts. (D) Survival assessment for patients classified by both TMB and risk score. H, high; L, low; TMB, tumor mutation burden.

Development of the predictive nomogram

A nomogram was established to predict OS at 1, 3, and 5 years for HNSCC individuals (Figure 5A). The predictive capacity of the nomogram was assessed through ROC analysis and calibration curves. In Figure 5B, the AUC values for the nomogram in predicting OS at 1, 3, and 5 years were 0.714, 0.757, and 0.731, respectively. Moreover, the calibration curves for 1-, 3-, and 5-year OS substantially aligned with the theoretical line (Figure 5C), demonstrating the nomogram’s strong performance.

Figure 5 Construction and validation of the nomogram. (A) Nomogram for determining the 1-, 3-, or 5-year OS rate in persons with HNSCC. (B) ROC curves of the nomogram. (C) Calibration curves of the nomogram. AUC, area under the curve; CI, confidence interval; HNSCC, head and neck squamous cell carcinoma; N, lymph node; OS, overall survival; ROC, receiver operating characteristic; T, tumor.

Immune landscape analysis

Furthermore, GSEA identified pathways, such as allograft rejection and asthma, that exhibited enrichment within the low-risk cohort (Figure 6A), whereas pathways related to extracellular matrix receptor interaction and focal adhesion were predominantly enriched in the high-risk cohort (Figure 6B). The heatmap illustrated fluctuations in tumor purity, ESTIMATEScore, and ssGSEA scores related to immune functions between two cohorts (Figure 7A). ESTIMATE analysis indicated that the low-risk cohort demonstrated higher ImmuneScore and ESTIMATEScore versus the high-risk cohort (Figure 7B). In addition, ssGSEA analysis of immune cell populations and functions demonstrated that the low-risk cohort harbored a more pronounced presence of immune-related functions compared to the high-risk cohort (Figure 7C). Then, an investigation was conducted to examine the connection between risk scores and tumor-immune cell infiltration, with the results visually summarized in Figure 8A as a heatmap. The low-risk cohort generally displayed a greater infiltration of naïve B cells, plasma cells, CD8 T cells, activated memory CD4 T cells, follicular helper T cells, Tregs, monocytes, resting dendritic cells, resting mast cells, and neutrophils, while showing diminished infiltration of resting memory CD4 T cells, resting natural killer (NK) cells, M0 macrophages, M2 macrophages, and activated mast cells (Figure 8B). Correlation analysis demonstrated that naïve B cells, resting dendritic cells, resting mast cells, neutrophils, plasma cells, activated memory CD4 T cells, CD8 T cells, follicular helper T cells, and Tregs were negatively linked to risk scores (Figure 8C). Conversely, activated dendritic cells, M0 macrophages, M2 macrophages, activated mast cells, resting NK cells, and resting memory CD4 T cells were positively linked to risk scores (Figure 8C).

Figure 6 Enrichment analysis. GSEA between low-risk (A) and high-risk (B) cohorts. ECM, extracellular matrix; GSEA, gene set enrichment analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Figure 7 Effect of IRG signature on TME in HNSCC. (A) The heatmap depicts the link between risk score and tumor immune microenvironment. (B) ImmuneScore, StromalScore, and ESTIMATEScore in both risk cohorts. (C) The differential infiltration levels of immune cells and immunological-related functions across two cohorts. *, P<0.05; **, P<0.01; ***, P<0.001. aDCs, activated dendritic cells; APC, antigen-presenting cell; CCR, C-C chemokine receptor; DCs, dendritic cells; HLA, human leukocyte antigen; HNSCC, head and neck squamous cell carcinoma; iDCs, immature dendritic cells; IFN, interferon; IRG, immune-related gene; MHC, major histocompatibility; NK, natural killer; pDCs, plasmacytoid dendritic cells; TIL, tumor infiltrating lymphocyte; Tfh, follicular helper T cell; TME, tumor microenvironment; Treg, regulatory T cell.

Figure 8 Impact of IRG signature on immune cell infiltration. (A) The precise extent of immune cell infiltration for every HNSCC sample. (B) The analysis examines the infiltration levels of immune cells across the two cohorts. (C) Relationship between immune cell infiltration, risk model IRG expression, and risk score. *, P<0.05; **, P<0.01; ***, P<0.001. HNSCC, head and neck squamous cell carcinoma; IRG, immune-related gene; NK, natural killer.

Correlation between risk models and immunotherapy in the HNSCC

The examination of immune checkpoints demonstrated that most checkpoints exhibited enhanced activation in the low-risk cohort (Figure 9A). Following this, the HLA-related gene level was compared between two cohorts. Of the 24 examined HLA-related genes analyzed, 15 demonstrated increased expression levels in the low-risk cohort versus the high-risk cohort (Figure 9B). The investigation implemented the TIDE algorithm to generate TIDE scores for individual patients, serving as a metric to evaluate immunotherapy responsiveness. Figure 9C illustrates elevated TIDE scores in the high-risk cohort, indicating diminished immunotherapy effectiveness. Additionally, the low-risk cohort demonstrated notably superior IPS scores, whether considering monotherapy with anti-PD-1 (Figure 9D), anti-CTLA4 (Figure 9E), or their combined administration (Figure 9F).

Figure 9 Forecasting immunotherapy response of the model. (A) Expression of immune checkpoint-related genes. (B) Comparative analysis of HLA-related gene expression. (C) Comparative analysis of TIDE scores. (D) Significant increase of IPS-PD-1 in the low-risk cohort. (E) Significant elevation of IPS-CTLA4 in the low-risk cohort. (F) Notable increase of IPS-PD-1 + CTLA4 in the low-risk cohort. *, P<0.05; **, P<0.01; ***, P<0.001. HLA, human leukocyte antigen; IPS, Immunophenoscore; TIDE, Tumor Immune Dysfunction and Exclusion.

MR analysis of model gene expression and the onset of oral cavity cancer

The MR analysis revealed that only the eQTL of CCR7, among the genes included in the model, exhibited a causative link to the development of oral cavity cancer. The regression patterns derived from multiple methods are illustrated in the scatter plot (Figure 10A). The forest plot illustrates the effect sizes of diverse SNPs within the eQTL of CCR7 (Figure 10B). The leave-one-out sensitivity analysis confirmed that no individual study substantially impacted the results (Figure 10C). The intercept examination through MR-Egger for CCR7 eQTL revealed no indication of horizontal pleiotropy (P=0.71), suggesting that the selected instrumental variables had no substantial effect on the outcome through other routes. The Cochran Q test for heterogeneity demonstrated no notable heterogeneous patterns in the CCR7 eQTL (P=0.88 for the IVW approach).

Figure 10 The causality of CCR7 expression on HNSCC risk. (A) Scatter plot. The gradient of each line signifies the projected impact per MR methodology. (B) Forest diagram. Horizontal lines represent 95% confidence intervals. (C) Leave-one-out analysis. Black dots signify the application of the IVW approach to assess the causal influence, excluding a particular variation from the study. The red point is the IVW estimate that includes all SNPs. HNSCC, head and neck squamous cell carcinoma; IVW, inverse variance weighted; MR, Mendelian randomization; SNP, single nucleotide polymorphism.

Discussion

Given the unfavorable prognosis and limited treatment efficacy observed in patients with HNSCC, establishing an innovative and dependable prognostic model becomes essential to forecast clinical outcomes and direct personalized therapeutic approaches (1). An expanding body of evidence highlights the critical role of immune cell infiltration and IRG expression in carcinogenesis and tumor progression (31,32). Examining the correlation between IRGs and cancer prognosis offers the potential for novel insights into predicting outcomes in HNSCC patients (33). Accordingly, research was undertaken to explore the connection between IRGs and HNSCC, resulting in the development of an innovative evaluation method engineered to anticipate patient survival and treatment outcomes.

In this investigation, IRGs were procured from the IMMPORT database. Subsequently, 18 survival-related IRGs associated with HNSCC were selected through UCRA. An integrated pipeline was established to develop an IRG signature, employing 10 distinct machine learning algorithms. Among 101 evaluated prognostic models, the RSF + CoxBoost-based model was the most efficient, yielding the highest mean C-index of 0.621. The Kaplan-Meier survival assessments and multivariate Cox regression substantiated the risk model’s capacity to reveal the IRG signature’s potential as an independent prognostic indicator for HNSCC across both training and test cohorts. Furthermore, using multivariate ROC curves and C-index evaluation suggested that the signature displayed superior predictive accuracy compared to traditional prognostic factors such as age, sex, grade, and stage. Subgroup analysis based on these clinical parameters further demonstrated that the predictive signature was consistently effective for every subgroup, underscoring its extensive applicability. In aggregate, these observations suggested that the signature identified in our investigation displayed a high level of robustness in forecasting the prognosis of HNSCC patients.

Previous investigations have suggested that TMB could serve as a crucial factor in promoting immune cell infiltration and affecting the therapeutic outcomes of immunotherapy (34,35). Elevated TMB levels have been linked to less favorable prognoses among individuals with HNSCC (36,37), a trend also observed in this investigation. A subsequent analysis of the mutation landscape of HNSCC revealed that TP53 mutations are the most prevalent. TP53 is recognized as a vital tumor suppressor, essential for maintaining genomic stability and regulating both the cell cycle and apoptosis (38,39). Existing literature supports the association between TP53 mutations and a worse prognosis in HNSCC (40-42). Notably, a markedly higher proportion of TP53 mutations was observed in the high-risk cohort than the low-risk cohort. Furthermore, the result suggested that the high-risk cohort experienced worse OS regardless of TMB status, reinforcing the predictive accuracy of this prediction signature. Additionally, a nomogram was developed to estimate the risk and survival likelihood of patients with HNSCC. The calibration and ROC curves demonstrated the nomogram’s strong predictive performance. These findings collectively suggest that the prognostic model exhibits a robust and reliable predictive capacity.

Tumor biology is shaped by the TME, particularly the immune microenvironment (43,44). ImmuneScore and StromalScore are linked to the levels of immune and matrix components present in the TME (45). A further examination was conducted to assess whether significant differences existed between the two cohorts concerning StromalScore, ImmuneScore, and ESTIMATEScore. Analysis revealed that the low-risk cohort displayed increased ImmuneScore and StromalScore versus the high-risk cohort, suggesting an enhanced degree of immune cell infiltration. This observation was further corroborated by the ssGSEA, which revealed relatively high immune infiltration in the low-risk cohort. These observations indicate that low-risk samples may exhibit heightened immune activation. To better understand immune cell infiltration, a correlation analysis was performed. Cui et al. demonstrated that HNSCC patients with abundant immune cell infiltration, encompassing B cells, CD8⁺ T cells, and neutrophils, tend to have a better prognosis (46). In line with these findings, the current study identified a negative correlation between risk scores and the infiltration of B cells, CD8⁺ T cells, and neutrophils. Numerous studies have established that macrophage infiltration is linked to unfavorable outcomes in individuals with HNSCC, suggesting that the extent of macrophage presence may serve as a valuable prognostic marker (47,48). Similarly, this investigation found that the presence of M0 and M2 macrophages exhibited a positive link to higher risk scores. Furthermore, NK cell activity has been reported to correlate with a favorable prognosis in HNSCC patients (49). In this investigation, resting NK cells were detected to be positively linked to higher risk scores. Based on these observations, it is plausible to infer that IRGs may significantly influence the immune microenvironment in HNSCC. Additionally, individuals in the low-risk cohort may display a more robust antitumor immune response, potentially contributing to a better prognosis.

Immunotherapy constitutes a critical component in the therapeutic approaches utilized by oncologists for the management of HNSCC (50). However, its effectiveness is often restricted to a limited subset of patients unless precise selection criteria are applied, underscoring the necessity to identify specific subtypes of HNSCC that would benefit most from immunotherapeutic interventions. Prior investigations have suggested that immune evasion may be associated with downregulating genes related to HLA (51,52). An analysis of immune checkpoint genes and HLA-related gene levels was conducted, revealing that the majority of them exhibited higher levels in low-risk patients. This finding points to a potentially enhanced responsiveness to immunotherapy in this cohort. Additionally, the TIDE score, which correlates positively with immune escape, was detected to be markedly diminished in the low-risk cohort (53). This result reflects the positive link between the risk score and tumor-associated immune escape. Further, IPS analysis suggested that the low-risk cohort displayed increased IPS scores, which suggests an augmented level of immunoreactivity. Collectively, the data from this investigation suggest that low-risk patients display increased sensitivity to immunotherapy in contrast to those with higher risk. The findings further imply that the IRG profile demonstrates considerable potential as a predictive indicator for guiding immunotherapy strategies in HNSCC. In addition, the MR analysis indicated that CCR7 was strongly associated with the incidence of oral cavity cancer, highlighting the possible risk.

Several limitations inherent in this investigation must be acknowledged. Firstly, this study used existing secondary datasets without external validation and did not account for confounders such as smoking or human papilloma virus (HPV) status, possibly leading to biased results. Furthermore, the correlation between the risk signature and immunotherapy requires validation through large-scale clinical trials. Moreover, the outcome information used in the MR analysis was confined to oral cavity cancer, with unclear generalizability to other HNSCC subtypes, necessitating further investigation with larger sample sizes to validate the results of these bioinformatics analyses. Finally, the biological mechanisms underlying the prognostic value of the IRG signature remain unexplored, and the functional impact of CCR7 in HNSCC carcinogenesis needs experimental confirmation.

Conclusions

A novel IRG risk model was established in this investigation to predict survival outcomes and delineate the immune landscape in HNSCC patients. This reliable and potent signature demonstrates considerable potential to guide and augment the development of novel therapeutic strategies for HNSCC patients.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2024-2665/rc

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2024-2665/prf

Funding: This study was supported by the Ningbo Natural Science Foundation (No. 2024J040), Zhejiang Traditional Chinese Medicine Science and Technology Plan (No. 2023ZL648), and Ningbo Social Public Welfare Research Project (No. 2023S078).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2024-2665/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Bray F, Ferlay J, Soerjomataram I, et al. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2018;68:394-424. [Crossref] [PubMed]
2. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2020. CA Cancer J Clin 2020;70:7-30. [Crossref] [PubMed]
3. Cervenka BP, Rao S, Bewley AF. Head and Neck Cancer and the Elderly Patient. Otolaryngol Clin North Am 2018;51:741-51. [Crossref] [PubMed]
4. Johnson DE, Burtness B, Leemans CR, et al. Head and neck squamous cell carcinoma. Nat Rev Dis Primers 2020;6:92. [Crossref] [PubMed]
5. Leemans CR, Braakhuis BJ, Brakenhoff RH. The molecular biology of head and neck cancer. Nat Rev Cancer 2011;11:9-22. [Crossref] [PubMed]
6. McDermott JD, Bowles DW. Epidemiology of Head and Neck Squamous Cell Carcinomas: Impact on Staging and Prevention Strategies. Curr Treat Options Oncol 2019;20:43. [Crossref] [PubMed]
7. Gillison ML, Chaturvedi AK, Anderson WF, et al. Epidemiology of Human Papillomavirus-Positive Head and Neck Squamous Cell Carcinoma. J Clin Oncol 2015;33:3235-42. [Crossref] [PubMed]
8. Zhou J, Michaud DS, Langevin SM, et al. Smokeless tobacco and risk of head and neck cancer: evidence from a case-control study in New England. Int J Cancer 2013;132:1911-7. [Crossref] [PubMed]
9. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2018. CA Cancer J Clin 2018;68:7-30. [Crossref] [PubMed]
10. Burtness B, Harrington KJ, Greil R, et al. Pembrolizumab alone or with chemotherapy versus cetuximab with chemotherapy for recurrent or metastatic squamous cell carcinoma of the head and neck (KEYNOTE-048): a randomised, open-label, phase 3 study. Lancet 2019;394:1915-28.
11. Arneth B. Tumor Microenvironment. Medicina (Kaunas) 2019;56:15. [Crossref] [PubMed]
12. Shen Z, Feng C, Chen X, et al. Prognostic model of lung adenocarcinoma based on immunoprognosis-related genes and related drug prediction. J Thorac Dis 2024;16:5860-77. [Crossref] [PubMed]
13. Binnewies M, Roberts EW, Kersten K, et al. Understanding the tumor immune microenvironment (TIME) for effective therapy. Nat Med 2018;24:541-50. [Crossref] [PubMed]
14. Lu C, Liu Y, Ali NM, et al. The role of innate immune cells in the tumor microenvironment and research progress in anti-tumor therapy. Front Immunol 2022;13:1039260. [Crossref] [PubMed]
15. Quail DF, Joyce JA. Microenvironmental regulation of tumor progression and metastasis. Nat Med 2013;19:1423-37. [Crossref] [PubMed]
16. Segal BH, Giridharan T, Suzuki S, et al. Neutrophil interactions with T cells, platelets, endothelial cells, and of course tumor cells. Immunol Rev 2023;314:13-35. [Crossref] [PubMed]
17. Rutihinda C, Haroun R, Saidi NE, et al. Inhibition of the CCR6-CCL20 axis prevents regulatory T cell recruitment and sensitizes head and neck squamous cell carcinoma to radiation therapy. Cancer Immunol Immunother 2023;72:1089-102. [Crossref] [PubMed]
18. Stein A, Thuss-Patience P. Immunotherapy in Esophagogastric Adenocarcinoma. Visc Med 2019;35:12-6. [Crossref] [PubMed]
19. Chen S, Yang Y, He S, et al. Review of biomarkers for response to immunotherapy in HNSCC microenvironment. Front Oncol 2023;13:1037884. [Crossref] [PubMed]
20. Marei HE, Hasan A, Pozzoli G, et al. Cancer immunotherapy with immune checkpoint inhibitors (ICIs): potential, mechanisms of resistance, and strategies for reinvigorating T cell responsiveness when resistance is acquired. Cancer Cell Int 2023;23:64. [Crossref] [PubMed]
21. Ferris RL, Blumenschein G Jr, Fayette J, et al. Nivolumab for Recurrent Squamous-Cell Carcinoma of the Head and Neck. N Engl J Med 2016;375:1856-67. [Crossref] [PubMed]
22. Hanna GJ, Lizotte P, Cavanaugh M, et al. Frameshift events predict anti-PD-1/L1 response in head and neck cancer. JCI Insight 2018;3:e98811. [Crossref] [PubMed]
23. Zhang C, Ke J, Li J, et al. SKA1/2/3 link to poor prognosis and immune infiltration of head and neck squamous cell carcinomas. Transl Cancer Res 2024;13:6057-69. [Crossref] [PubMed]
24. Wang Z, Cheng L, Huang J, et al. Integrative machine learning and neural networks for identifying PANoptosis-related lncRNA molecular subtypes and constructing a predictive model for head and neck squamous cell carcinoma. Eur Arch Otorhinolaryngol 2024;281:5481-95. [Crossref] [PubMed]
25. Wang Y, Mou YK, Liu WC, et al. Machine learning developed a macrophage signature for predicting prognosis, immune infiltration and immunotherapy features in head and neck squamous cell carcinoma. Sci Rep 2024;14:19538. [Crossref] [PubMed]
26. Huang C, Zhu F, Zhang H, et al. Identification of S1PR4 as an immune modulator for favorable prognosis in HNSCC through machine learning. iScience 2023;26:107693. [Crossref] [PubMed]
27. Du L, Wang P, Qiu X, et al. Integrating machine learning with mendelian randomization for unveiling causal gene networks in glioblastoma multiforme. Discov Oncol 2025;16:38. [Crossref] [PubMed]
28. Coupland H, Scheidwasser N, Katsiferis A, et al. Exploring the potential and limitations of deep learning and explainable AI for longitudinal life course analysis. BMC Public Health 2025;25:1520. [Crossref] [PubMed]
29. Wei Z, Hu Y, Zhang Y, et al. Identification of a telomere-related gene signature for the prognostic and immune landscape prediction in head and neck squamous cell carcinoma by integrated analysis of machine learning and Mendelian randomization. Medicine (Baltimore) 2025;104:e42211. [Crossref] [PubMed]
30. Wang S, Jiang J, Xing M, et al. Deciphering the prognostic potential of a necroptosis-related gene signature in head and neck squamous cell carcinoma: a bioinformatic analysis. Transl Cancer Res 2025;14:340-53. [Crossref] [PubMed]
31. Bogolyubova AV, Belousov PV. Inflammatory Immune Infiltration in Human Tumors: Role in Pathogenesis and Prognostic and Diagnostic Value. Biochemistry (Mosc) 2016;81:1261-73. [Crossref] [PubMed]
32. Li Y, Jiang T, Zhou W, et al. Pan-cancer characterization of immune-related lncRNAs identifies potential oncogenic biomarkers. Nat Commun 2020;11:1000. [Crossref] [PubMed]
33. He J, Fang X, Han M. Discovery and Verification of an Immune-Related Gene Pairs Signature for Predicting Prognosis in Head and Neck Squamous Cell Carcinoma. Front Genet 2021;12:654657. [Crossref] [PubMed]
34. Liu L, Bai X, Wang J, et al. Combination of TMB and CNA Stratifies Prognostic and Predictive Responses to Immunotherapy Across Metastatic Cancer. Clin Cancer Res 2019;25:7413-23. [Crossref] [PubMed]
35. Cristescu R, Mogg R, Ayers M, et al. Pan-tumor genomic biomarkers for PD-1 checkpoint blockade-based immunotherapy. Science 2018;362:eaar3593. [Crossref] [PubMed]
36. Yang B, Shen J, Xu L, et al. Genome-Wide Identification of a Novel Eight-lncRNA Signature to Improve Prognostic Prediction in Head and Neck Squamous Cell Carcinoma. Front Oncol 2019;9:898. [Crossref] [PubMed]
37. Zhang L, Li B, Peng Y, et al. The prognostic value of TMB and the relationship between TMB and immune infiltration in head and neck squamous cell carcinoma: A gene expression-based study. Oral Oncol 2020;110:104943. [Crossref] [PubMed]
38. Chang YK, Tseng HH, Leung CM, et al. Targeted Next-Generation Sequencing-Based Multiple Gene Mutation Profiling of Patients with Rectal Adenocarcinoma Receiving or Not Receiving Neoadjuvant Chemoradiotherapy. Int J Mol Sci 2022;23:10353. [Crossref] [PubMed]
39. Jin H, Zheng L, Wang J, et al. Dendritic cell-related hub genes in head-and-neck squamous cell carcinoma: implications for prognosis and immunotherapy. Transl Cancer Res 2024;13:3620-36. [Crossref] [PubMed]
40. Lyu H, Li M, Jiang Z, et al. Correlate the TP53 Mutation and the HRAS Mutation with Immune Signatures in Head and Neck Squamous Cell Cancer. Comput Struct Biotechnol J 2019;17:1020-30. [Crossref] [PubMed]
41. Yu J, Chen Y, Pan X, et al. Relationships of Ferroptosis and Pyroptosis-Related Genes with Clinical Prognosis and Tumor Immune Microenvironment in Head and Neck Squamous Cell Carcinoma. Oxid Med Cell Longev 2022;2022:3713929. [Crossref] [PubMed]
42. Basyuni S, Nugent G, Ferro A, et al. Value of p53 sequencing in the prognostication of head and neck cancer: a systematic review and meta-analysis. Sci Rep 2022;12:20776. [Crossref] [PubMed]
43. Pitt JM, Marabelle A, Eggermont A, et al. Targeting the tumor microenvironment: removing obstruction to anticancer immune responses and immunotherapy. Ann Oncol 2016;27:1482-92. [Crossref] [PubMed]
44. Tang Y, Hu T, Yin W, et al. Bioinformatics analysis reveals VEGFC's prognostic significance in head and neck squamous cell carcinoma and its association with immune cell infiltration. Transl Cancer Res 2024;13:5953-70. [Crossref] [PubMed]
45. Bi KW, Wei XG, Qin XX, et al. BTK Has Potential to Be a Prognostic Factor for Lung Adenocarcinoma and an Indicator for Tumor Microenvironment Remodeling: A Study Based on TCGA Data Mining. Front Oncol 2020;10:424. [Crossref] [PubMed]
46. Cui J, Zheng L, Zhang Y, et al. Bioinformatics analysis of DNMT1 expression and its role in head and neck squamous cell carcinoma prognosis. Sci Rep 2021;11:2267. [Crossref] [PubMed]
47. Ji WT, Chen HR, Lin CH, et al. Monocyte chemotactic protein 1 (MCP-1) modulates pro-survival signaling to promote progression of head and neck squamous cell carcinoma. PLoS One 2014;9:e88952. [Crossref] [PubMed]
48. He KF, Zhang L, Huang CF, et al. CD163+ tumor-associated macrophages correlated with poor prognosis and cancer stem cells in oral squamous cell carcinoma. Biomed Res Int 2014;2014:838632. [Crossref] [PubMed]
49. Patin EC, Dillon MT, Nenclares P, et al. Harnessing radiotherapy-induced NK-cell activity by combining DNA damage-response inhibition and immune checkpoint blockade. J Immunother Cancer 2022;10:e004306. [Crossref] [PubMed]
50. Galizia D, Minei S, Maldi E, et al. How Risk Factors Affect Head and Neck Squamous Cell Carcinoma (HNSCC) Tumor Immune Microenvironment (TIME): Their Influence on Immune Escape Mechanisms and Immunotherapy Strategy. Biomedicines 2022;10:2498. [Crossref] [PubMed]
51. He M, Wang Y, Zhang G, et al. The prognostic significance of tumor-infiltrating lymphocytes in cervical cancer. J Gynecol Oncol 2021;32:e32. [Crossref] [PubMed]
52. Lin Z, Sui X, Jiao W, et al. Exploring the mechanism and experimental verification of puerarin in the treatment of endometrial carcinoma based on network pharmacology and bioinformatics analysis. BMC Complement Med Ther 2022;22:150. [Crossref] [PubMed]
53. Xiang L, Piao L, Wang D, et al. Overexpression of SMS in the tumor microenvironment is associated with immunosuppression in hepatocellular carcinoma. Front Immunol 2022;13:974241. [Crossref] [PubMed]