The data used in this study were primarily sourced from the TCGA (The Cancer Genome Atlas) ¹ and HNSC-CPTAC databases. ² The TCGA-HNSC dataset comprises 548 tumor samples and 44 adjacent normal tissue samples. Detailed clinical baseline characteristics of the patient cohort are presented in Supplementary Table S1. RNA sequencing data, clinical information, and relevant survival data for HNSC patients were retrieved from TCGA using the TCGAbiolinks package in R. For the analysis of protein expression levels, data from the HNSC-CPTAC dataset were utilized.

Differential expression analysis of redox metabolism-related genes between normal and tumor samples in HNSC was performed using the limma package. A set of predefined redox metabolism-related genes was extracted from the GOBP_CELL_REDOX_HOMEOSTASIS gene set (obtained from the Molecular Signatures Database). ³ From this gene set, differential expression was assessed with an adjusted p-value threshold of 0.05, yielding redox metabolism-related DEGs. The results were visualized using a heatmap generated with the pheatmap package. Genes exhibiting significant differential expression were further analyzed for their prognostic potential using univariate Cox regression analysis.

Univariate and multivariate Cox regression analyses were performed to evaluate the prognostic significance of the identified redox metabolism-related DEGs. Cox regression models were constructed using the survival package. The univariate analysis first identified individual prognostic markers with a significance threshold of p < 0.05, followed by multivariate Cox regression to assess the independent prognostic value of each gene. Forest plots were generated to illustrate hazard ratios (HRs) and their corresponding 95% confidence intervals (CIs). The prognostic significance of ERP44 expression was analyzed using Kaplan-Meier survival analysis. The survminer package was used to assess the relationship between ERP44 expression and various survival outcomes, including overall survival (OS), disease-specific survival (DSS), progression-free interval (PFI), and disease-free interval (DFI). Log-rank tests were applied to compare survival curves between high and low ERP44 expression groups. Additionally, Wilcoxon rank-sum tests were conducted to compare ERP44 expression levels between tumor and normal tissues in the TCGA-HNSC dataset.

Gene Set Variation Analysis (GSVA) was performed to evaluate the association between ERP44 expression and various oncogenic pathways in the TCGA-HNSC dataset. The GSVA package was used to calculate GSVA scores for 14 predefined hallmark pathways (MSigDB) related to angiogenesis, epithelial-mesenchymal transition (EMT), proliferation, hypoxia, and other tumor-related processes. Spearman’s correlation coefficients (|R| >0.10, FDR-adjusted p < 0.05) were calculated to assess the relationship between ERP44 expression and GSVA scores. Only pathways meeting both correlation strength (|R| >0.10) and statistical significance thresholds were reported in the main text.

The relative proportion of immune cell types in the TCGA-HNSC dataset was estimated using the ssGSEA algorithm. The correlation between ERP44 expression and immune cell infiltration levels was assessed using the spearman’s correlation analysis (|R| > 0.10, FDR-adjusted p < 0.05). Differential immune cell infiltration between high and low ERP44 expression groups (stratified by median expression) was evaluated, and statistical significance was determined using Wilcoxon rank-sum tests (p < 0.05).

Single-cell RNA-sequencing data from the GSE103322 dataset (18 primary HNSC patients, 5,902 cells total) were analyzed using the Seurat package to investigate the expression of ERP44 at the single-cell level. Uniform Manifold Approximation and Projection (UMAP) plots were generated to visualize the clustering of various cell types, including malignant cells, fibroblasts, T cells, and macrophages. The distribution of ERP44 expression levels across these cell types was visualized using box plots and bar plots to determine the proportion of cells expressing ERP44 in each identified cell type.

The correlation between ERP44 expression and chemotherapy sensitivity was analyzed using data from the GDSC (Genomics of Drug Sensitivity in Cancer) database. ⁴ The sensitivity to chemotherapeutic agents, including Paclitaxel, Vinblastine, Gemcitabine, Sorafenib, and Rapamycin, was evaluated by calculating the IC50 values and performing Spearman’s correlation analysis (|R| >0.10, FDR-adjusted p < 0.05) between ERP44 expression and drug sensitivity. A p-value threshold of 0.05 was considered significant for the correlation analysis.

A pan-cancer analysis of ERP44 expression was performed using RNA-seq data from 33 cancer types in TCGA. Box plots were generated to compare ERP44 expression (log2[TPM+1] values) between tumor and normal tissues, with statistical significance assessed using Wilcoxon signed-rank tests (FDR-adjusted p < 0.01). For survival analysis, patients were dichotomized into high/low ERP44 expression groups based on median expression in each cancer type. A forest plot was created to display univariate Cox regression results (HR with 95% confidence intervals) for overall survival across all 33 cancers, with statistical significance threshold set at p < 0.05 (two-sided).

The human oral squamous cell carcinoma line SCC15 and the normal human immortalized oral epithelial cell (HIOEC) line were obtained from Otwo Biotech (Shenzhen, China). Both cell lines were maintained in RPMI-1640 medium (Life Technologies, Shanghai) under standard culture conditions: 37°C, 5% CO₂, and high humidity. Total RNA extraction was performed using Thermo Fisher’s RNA isolation reagent, followed by cDNA synthesis with the PrimeScript RT Kit (Thermo Fisher). Gene expression analysis was carried out via quantitative RT-PCR (qRT-PCR) on an Applied Biosystems ABI 7900HT platform, with relative expression levels calculated using the 2^−ΔΔCT method.

All statistical analyses were conducted using R (version 4.1.0). P-values less than 0.05 were considered statistically significant. Data visualization was performed using the ggplot2, pheatmap, and forestplot packages.

To unveil the differential expression of genes associated with redox metabolism between tumor and normal samples, we employed the limma package for comprehensive analysis. The resultant redox metabolism-related differentially expressed genes (DEGs) were visualized using a heatmap (Figure 2A), demonstrating distinctive expression profiles across the groups. The heatmap clearly indicates a set of genes exhibiting upregulated (in red) and downregulated (in blue) expression levels in the 548 tumor samples compared to the 44 adjacent normal samples. To assess the prognostic value of the identified DEGs, we performed univariate Cox regression analysis. The analysis results were illustrated in a forest plot, showcasing the hazard ratios (HRs) with their corresponding 95% confidence intervals (CIs) and p-values (Figure 2B). Several DEGs, such as ERP44, PRDX6, TXNRD1, and SELENOT, showed significant associations with patient survival (p < 0.05), warranting further investigation into their potential as prognostic biomarkers in HNSC. For further biological interpretation, enrichment analysis of the identified redox metabolism-related DEGs was conducted. The analysis revealed significant enrichment in pathways and processes crucial for cellular redox balance and oxidative stress management. Specifically, the DEGs were significantly associated with pathways including Parkinson’s disease, selenocompound metabolism, protein-disulfide reductase activity, oxidoreductase activity acting on a sulfur group of donors, glutathione metabolism, mitochondrial matrix, cellular detoxification, cellular oxidant detoxification, antioxidant activity, and cell redox homeostasis (Figure 2C). Overall, these analyses delineate a comprehensive profile of redox metabolism-related alterations in HNSC, underscoring their potential roles in disease progression and prognosis.

To determine the prognostic value of redox metabolism-related DEGs, we performed Cox regression analyses (Supplementary Figure S1). Univariate Cox regression analysis of the TCGA data indicated that ERP44 (p = 0.001), among other redox-related DEGs, exhibited the strongest association with poor prognosis in HNSC patients and was thus prioritized for further investigation. Additionally, age (p = 0.019), gender (p = 0.048), and metastasis status (M stage) (p = 0.002) were also found to be significant predictors of patient outcomes. Subsequent multivariate Cox regression analysis confirmed the independent prognostic value of ERP44 (p = 0.003). Notably, ERP44 retained significance after adjusting for clinical covariates, underscoring its robustness as a candidate biomarker. Age (p = 0.028) and M stage (p = 0.010) remained significant in the multivariate model, while other clinical factors such as grade, stage, T stage, and N stage did not show significant prognostic value in the multivariate analysis.

Figure 3 illustrates the prognostic significance and expression analysis of ERP44 in HNSC. Kaplan-Meier survival analyses based on TCGA-HNSC data revealed that high ERP44 expression is significantly associated with reduced overall survival (OS) (Figure 3A, p = 0.006), disease-specific survival (DSS) (Figure 3C, p = 0.001), and progression-free interval (PFI) (Figure 3D, p = 0.002). Although a similar trend was observed for disease-free interval (DFI), the association was not statistically significant (Figure 3B, p = 0.058). Gene expression analysis demonstrated that ERP44 expression levels were significantly higher in tumor tissues compared to normal tissues within the TCGA-HNSC dataset (Figure 3E, p < 0.001). Protein expression analysis using the HNSC-CPTAC dataset further corroborated these findings, showing significantly elevated protein levels of ERP44 in tumor tissues compared to normal tissues, as determined by a Wilcoxon rank sum test (Figure 3F, p < 0.001). In addition, our cell experiments further confirmed that the ERP44 gene was significantly upregulated in SCC15 cells (Supplementary Figure S2). As shown in Figure 4, our data reveal that patients classified in the low-risk group generally exhibit lower risk scores and a higher survival probability compared to those in the high-risk group. The survival status is indicated with blue dots for alive patients and red dots for deceased patients. It is apparent that a higher number of deceased patients (red dots) are found in the high-risk group, suggesting a negative correlation between higher risk scores and patient survival. Additionally, the heatmap below the plot illustrates ERP44 expression levels, with color gradients from blue (lower expression) to red (higher expression). Patients in the high-risk group predominantly show higher ERP44 expression levels, whereas those in the low-risk group exhibit lower ERP44 expression. This inverse relationship between ERP44 expression and survival underscores the potential prognostic value of ERP44 in HNSC. In addition, Supplementary Figure S3 presents the expression levels of ERP44 across various clinical subgroups within the TCGA-HNSC dataset. These findings indicate that ERP44 expression varies significantly with tumor grade and treatment outcome. These results collectively underscore the potential role of elevated ERP44 expression as a biomarker for poor prognosis in HNSC patients.

Figure 5 illustrates the GSVA of ERP44 in the TCGA-HNSC dataset, assessing the relationship between ERP44 expression and various oncogenic pathways. Significant negative correlations were observed between ERP44 expression and key oncogenic processes such as angiogenesis (R = −0.16, p = 0.00015), differentiation (R = −0.26, p = 2.7e-09), DNA damage (R = −0.14, p = 0.0011), DNA repair (R = −0.12, p = 0.0078), epithelial-mesenchymal transition (EMT) (R = −0.14, p = 0.0019), proliferation (R = −0.21, p = 1.1e-06), and stemness (R = −0.33, p = 6.9e-15). Conversely, a significant positive correlation was found between ERP44 expression and hypoxia (i.e., elevated ERP44 levels corresponded to higher hypoxic conditions) (R = 0.17, p = 9.7e-05). These results suggest that high ERP44 expression is associated with decreased activity in key oncogenic pathways such as angiogenesis, cell cycle, and differentiation, while showing a positive association with hypoxia in HNSC. This highlights the multifaceted role of ERP44 in the modulation of tumor-related processes.

As shown in Figure 6A, the box plots illustrate significant differences in immune cell infiltration levels between the high and low ERP44 expression groups. Notably, higher ERP44 expression is associated with increased infiltration of several immune cell types, including Tgd and Th2 cells (p < 0.05). Conversely, lower ERP44 expression shows a significant association with increased infiltration of B cells, DC, mast cells, pDC, T cells, Th17 cells, and regulatory T cells (TReg) (p < 0.05). As shown in Figure 6B, the correlation analysis reveals that ERP44 expression is positively correlated with Th2 cells (R = 0.204, p < 0.001) and Th1 cells (R = 0.136, p < 0.01). Negative correlations were observed between ERP44 expression and B cells (R = −0.280, p < 0.001), pDC (R = −0.248, p < 0.001), mast cells (R = −0.217, p < 0.001), and Th17 cells (R = −0.176, p < 0.001). These results suggest that ERP44 expression in HNSC is associated with distinct patterns of immune cell infiltration, implicating its potential role in the tumor immune microenvironment.

The Uniform Manifold Approximation and Projection (UMAP) plot shows the clustering of different cell types, including CD4 T cells (CD4Tconv), CD8 T cells (CD8T and CD8Tex), endothelial cells, fibroblasts, malignant cells, mast cells, monocytes/macrophages (Mono/Macro), myocytes, myofibroblasts, and plasma cells (Figure 7A). The UMAP plot for ERP44 expression indicates variation in ERP44 expression levels across the identified cell clusters, with a color gradient representing the expression level (Figure 7B). As shown in Figure 7C, a box plot further details the distribution of ERP44 mRNA expression levels within each cell type. The analysis reveals that ERP44 is differentially expressed among the various cell types, with notably higher expression in malignant cells and fibroblasts compared to immune cells such as CD4 T cells, CD8 T cells, and plasma cells. In addition, the bar plot shows the proportion of cells expressing ERP44 (positive, red) versus not expressing ERP44 (negative, blue) within each identified cell type. Notably, malignant cells exhibited the highest proportion of ERP44 expression, with 58.4% of malignant cells being ERP44-positive compared to 24.5% negative (p < 0.01). Myfibroblasts also showed a significant expression of ERP44, with 7.7% positive compared to 16.7% negative (p < 0.05) (Figure 8). These findings indicate that ERP44 is differentially expressed across various cell types in HNSC, with particularly elevated expression in malignant cells and fibroblasts, suggesting its potential role in tumorigenesis and the tumor microenvironment.

A significant negative correlation was observed between ERP44 expression and Paclitaxel sensitivity (R = −0.33, p = 5.7e-15), suggesting that higher ERP44 expression is associated with increased sensitivity to Paclitaxel (Figure 9A). Vinblastine sensitivity also showed a significant negative correlation with ERP44 expression (R = −0.21, p = 1.9e-06), indicating greater sensitivity to Vinblastine in the high ERP44 expression group (Figure 9B). No significant correlation was found between ERP44 expression and Cisplatin sensitivity (R = −0.04, p = 0.38) (Figure 9C). A slight negative correlation was noted between ERP44 expression and Gemcitabine sensitivity (R = −0.14, p = 0.016), implying that higher ERP44 expression may lead to increased sensitivity to Gemcitabine (Figure 9D). Sorafenib sensitivity showed a strong negative correlation with ERP44 expression (R = −0.32, p = 1.5e-13), indicating higher sensitivity to Sorafenib in tumors expressing higher levels of ERP44 (Figure 9E). Contrarily, Rapamycin sensitivity exhibited a significant positive correlation with ERP44 expression (R = 0.21, p = 1.6e-06), suggesting that higher ERP44 expression correlates with decreased sensitivity to Rapamycin (Figure 9F). These results suggest that ERP44 expression levels are significantly associated with the sensitivities to several chemotherapeutic drugs, including Paclitaxel, Vinblastine, Gemcitabine, Sorafenib, and Rapamycin, highlighting the potential of ERP44 as a predictive biomarker for chemotherapy response in HNSC.

Box plots illustrate the differential expression of ERP44 between tumor tissues and normal tissues in multiple cancer types from the TCGA dataset. ERP44 expression is significantly higher in tumor tissues compared to normal tissues in several cancers, including GBM, HNSC, KICH, KIRP, PRAD, and UCEC (p < 0.05) (Figure 10A). A bar plot from the Human Protein Atlas (HPA) highlights the landscape of ERP44 expression in various cancer cell types, showing the highest expression in myeloma, followed by head and neck cancer, esophageal cancer, and others (Figure 10B). A forest plot displays the hazard ratios (HR) for overall survival (OS) associated with ERP44 expression in different cancers. Significant associations were found in ACC (p < 0.001), HNSC (p = 0.001), KIRP (p = 0.023), LGG (p < 0.001), SKCM (p = 0.003), UCEC (p = 0.017), and UVM (p = 0.01) (Figure 10C). Collectively, these results underscore ERP44’s pan-cancer oncogenic role, where its overexpression correlates with poor survival across diverse malignancies. This establishes ERP44 as a compelling biomarker for prognosis and a target for further mechanistic investigation in tumorigenesis. To delineate ERP44’s impact on tumor immunity, Figure 11 reveals its differential correlations with immune cell populations in a cancer-type-specific manner. For instance, significant positive correlations are noted between ERP44 and activated dendritic cells (aDC) in several cancer types such as BLCA, BRCA, HNSC, and LGG, indicating a potential role in promoting immune response. Conversely, negative correlations are observed between ERP44 expression and T regulatory cells (TReg) in cancers like HNSC, KIRC, and THCA, suggesting a possible inhibitory influence on immune suppression mechanisms. Similar nuanced relationships are observed with macrophages, B cells, neutrophils, and T cells. These findings illuminate ERP44’s multifaceted involvement in sculpting the tumor immune microenvironment. Its context-dependent interactions with immune infiltrates directly support its broader role in tumor progression and patient outcomes, offering mechanistic insights into how ERP44 influences cancer biology and reinforcing its value as a therapeutic or prognostic biomarker.