RAPTOR Silencing Inhibits Head and Neck Squamous Cell Carcinoma Progression via Regulation of S100A8/A9 and Urokinase Plasminogen Activator

Introduction

Head and neck squamous cell carcinoma (HNSCC) is a highly aggressive malignant neoplasm originating from the mucosal epithelium of the upper aerodigestive tract, with primary anatomical involvement of the oral mucosa, vermilion border, paranasal sinuses (including maxillary and ethmoid sinuses), glottic region, and nasopharynx [1]. Global cancer epidemiological data show that HNSCC ranks as the sixth most prevalent malignancy worldwide, with an annual incidence exceeding 650 000 cases and an associated mortality of 90 000 deaths per year [2]. Established etiological factors include chronic tobacco exposure, persistent infection with high-risk human papillomavirus, and chronic ethanol consumption [3]. Despite advances in treatment, the 5-year overall survival (OS) for HNSCC remains poor at 48.3% and drops to 32.7% in patients with local recurrence or distant metastasis (particularly to lung and bone). In advanced-stage (III–IV) disease, multimodal therapeutic interventions (including radical surgical resection with adjuvant chemoradiotherapy) frequently induce substantial functional impairments, such as dysphonia, dysphagia, and ageusia. These treatment-related toxicities significantly compromise health-related quality of life [4,5].

A major clinical challenge in HNSCC management is diagnostic delay (median time from symptom onset to diagnosis: 4.2 months), primarily attributable to the nonspecific clinical presentation of early-stage lesions (eg, persistent pharyngeal paresthesia or mild pain only). Current evidence suggests that identifying novel molecular diagnostic biomarkers may represent a pivotal strategy to improve early detection rates and subsequently enhance patient survival outcomes. Emerging proteogenomic approaches that integrate mass spectrometry–based proteomic profiling with next-generation sequencing offer promise for identifying clinically actionable biomarkers for early diagnosis and personalized therapeutic interventions [6–8]. Proteogenomics enables systematic validation of gene expression at the protein level through peptide-spectrum matching, thereby providing a powerful framework for biomarker discovery and tumor biology research [9]. Previous studies have demonstrated the utility of proteogenomics in tumor prognosis, such as the identification of C-C motif chemokine ligands as key regulators of angiogenesis and progression in Epstein-Barr virus–associated nasopharyngeal carcinoma [10]. In HNSCC, aberrant overexpression of proteins including CD44, S100A7, and S100P has been reported as potential biomarkers for early malignant transformation [11–13].

In addition to biomarker discovery, aberrant activation of the PI3K/AKT/mTOR signaling pathway has been widely implicated in HNSCC tumorigenesis and progression. Pharmacological inhibition of mTORC1 using rapamycin and its analogs shows antitumor effects in preclinical models. However, most existing studies have focused on upstream regulators of global mTOR activity, whereas the specific role of RAPTOR, the essential scaffolding component of mTORC1, remains poorly defined in HNSCC. Moreover, protein-based prognostic models integrating large-scale proteomic and transcriptomic data are still limited, and the downstream mechanisms linking RAPTOR to tumor invasion and inflammation-related pathways have not been fully elucidated. Previous TCGA/TCPA-based analyses proposed alternative HNSCC protein signatures, but these studies relied on earlier data releases with shorter follow-up and limited clinical annotation. Our study fills this gap by leveraging updated datasets and systematically evaluating prognostic proteins, aiming to refine survival prediction and improve clinical relevance. Therefore, this study integrates protein expression data from The Cancer Proteome Atlas (TCPA) with transcriptomic profiles from The Cancer Genome Atlas (TCGA), combined with experimental validation, to systematically identify clinically relevant protein biomarkers. We further investigate the functional and mechanistic role of RAPTOR in HNSCC progression, aiming to provide novel prognostic tools and potential therapeutic targets for precision medicine.

Material and Methods

DATA COLLECTION:

Proteomic data were obtained from the TCPA portal (http://tcpaportal.org/), which systematically documents protein expression profiles from 8167 tumor samples generated using Reverse Phase Protein Array (RPPA) technology. The RPPA platform provides a quantitative and highly robust evaluation of protein expression and post-translational modifications. These proteomic datasets show high concordance with genomic data from the TCGA (https://portal.gdc.cancer.gov/), thereby offering a reliable resource for multi-omics integration and functional model selection.

For this study, standardized Level 4 RPPA proteomic data were retrieved from TCPA, while matched transcriptomic (RNA-sequencing) data and corresponding clinical information were obtained from the TCGA HNSCC project via the Genomic Data Commons (GDC) Data Portal. Rigorous data acquisition and integration procedures were applied to ensure consistency across platforms.

A total of 353 patients with primary HNSCC were included in the final analysis. Inclusion criteria were: (1) availability of matched RPPA proteomic data from TCPA and RNA-sequencing transcriptomic data from TCGA, and (2) complete OS information. Patients were excluded if survival time or survival status was missing, or if samples were flagged for poor data quality in the original TCPA or TCGA datasets. The proportions of missing data for key clinical variables were 4.8% for tumor grade, 8.2% for tumor stage, and 5.7% for T stage; these missing values were treated as separate “unknown” categories in subsequent analyses.

ESTABLISHING PROGNOSIS-RELATED GENE SIGNATURES:

Univariable Cox proportional hazards regression was first performed to identify prognosis-related genes and develop gene signature models. Variables with potential prognostic relevance were then subjected to least absolute shrinkage and selection operator (LASSO)-penalized Cox regression to select the most informative features and construct the final gene signature model. The gene signature risk score was calculated using the following formula: Risk score=∑ (β_i × Exp_i). where β_i represents the regression coefficient of the ith gene, and Exp_i denotes the expression level of the ith gene (i=1, 2..., n). The proportional hazards assumption was assessed for each candidate biomarker using Schoenfeld residuals. Biomarkers showing significant violations of this assumption (P<0.05) were further modeled using extended Cox proportional hazards models with time-varying coefficients. Specifically, for HER3_pY1289, MERIT40_pS29, and Hexokinase-I, an interaction term between biomarker expression and log-transformed time was introduced, such that the regression coefficient was parameterized as β(t) = β1 + β2 × ln (t)β(t) = β1 + β2 × ln(t), allowing the corresponding hazard ratios to vary continuously over follow-up. All remaining biomarkers satisfied the proportional hazards assumption and were modeled using time-fixed coefficients. For internal validation, the entire cohort was randomly divided into a training set and a testing set at a 1: 1 ratio. Patients were stratified into low- and high-risk groups according to the median risk score derived from the training cohort. Dynamic risk score trajectories were generated for visualization purposes only to illustrate temporal changes in predicted risk for representative patients. These plots were not used for risk stratification, model training, or performance evaluation. Model performance was evaluated using Kaplan-Meier survival analysis with the log-rank test to compare survival differences between risk groups. Time-dependent receiver operating characteristic (ROC) curves were generated to assess predictive accuracy, and the area under the curve (AUC) was calculated at 1, 3, and 5 years. Calibration plots were further used to evaluate the agreement between predicted and observed survival outcomes. A 2-sided P value <0.05 was considered statistically significant.

ESTABLISHING A PREDICTIVE NOMOGRAM:

We aimed to develop a prognostic nomogram for HNSCC by integrating a protein-derived signature with established clinical factors. The modeling process followed a sequential, data-driven feature-selection pipeline designed to ensure robustness and minimize overfitting.

1) CANDIDATE PREDICTOR SCREENING AND MISSING DATA HANDLING:

Candidate predictor screening was first performed using univariable Cox proportional hazards regression to evaluate the associations between all protein expression features, as well as standard clinicopathological variables, including age, sex, and tumor-node-metastasis (TNM) stage, and overall survival. Variables with a P-value < 0.05 were retained as candidate predictors for subsequent multivariable modeling. Sex was treated as a categorical variable with female as the reference category and male as the comparison category, such that hazard ratios represent the relative risk for males compared with females (male vs female).

Given the presence of missing values in several clinical variables, multiple imputation was applied to preserve sample size and minimize potential bias. Missing values were imputed using the Multiple Imputation by Chained Equations (MICE) algorithm implemented in the mice R package (version 3.16.0). Five imputed datasets (m=5) were generated with 10 iterations to ensure convergence. Continuous variables (eg, age) were imputed using predictive mean matching (PMM), whereas categorical variables (eg, sex, grade, and stage) were imputed using polytomous logistic regression. All subsequent analyses were conducted separately within each imputed dataset, and final parameter estimates were pooled according to Rubin’s rules. Because the M stage variable exhibited a high proportion of missing values (67.1%), it was excluded from the multivariable Cox regression analysis and from subsequent nomogram construction to avoid potential bias and distortion of effect estimates.

2) HIGH-DIMENSIONAL FEATURE SELECTION FOR PROTEOMIC DATA:

Due to the high dimensionality of the proteomic dataset, protein candidates identified in the univariable screening step were further refined using LASSO–penalized Cox regression. The optimal penalty parameter was determined via 10-fold cross-validation (lambda.min). This procedure yielded a minimal and non-redundant set of prognostic protein biomarkers, which were subsequently combined into a composite protein-based risk score using the linear predictor derived from the LASSO model.

3) INTEGRATED PROGNOSTIC MODEL CONSTRUCTION AND NOMOGRAM DEVELOPMENT:

The protein-based risk score was then integrated with the selected clinical variables in a multivariable Cox proportional hazards model. To derive the most parsimonious and informative prognostic model, a bidirectional stepwise selection procedure based on the Akaike Information Criterion (AIC) was applied. Only the following clinical variables were retained for the final prognostic model: age, sex, grade, overall clinical stage (Stage I–II vs Stage III–IV), T stage (T1–T2 vs T3–T4), and N stage (N0 vs N1–N3), together with the protein-based risk score. In this model, gender was consistently coded as female (reference) and male (comparison), and the corresponding hazard ratio reflects the relative risk of death for males vs females. The regression coefficients from the final multivariable model were used to construct the prognostic nomogram, which provides individualized estimates of 1-, 3-, and 5-year OS probabilities by visualizing the relative contribution of each independent predictor.

4) MODEL PERFORMANCE ASSESSMENT:

The discriminative performance of the nomogram was evaluated using time-dependent ROC curve analysis, with the AUC calculated along with corresponding 95% confidence intervals. Calibration was assessed by plotting observed versus nomogram-predicted survival probabilities at 1, 3, and 5 years.

PRIMARY TISSUE SPECIMENS AND IMMUNOHISTOCHEMISTRY:

For immunohistochemistry validation of RAPTOR expression, paired tumor tissues and matched adjacent non-tumor epithelial tissues were collected from 30 independent patients with oral squamous cell carcinoma (OSCC), a major and well-recognized subtype of HNSCC, who underwent primary surgical resection at the First Affiliated Hospital of Nanchang University. Each pair was obtained from a different patient (30 tumor tissues and 30 matched adjacent tissues in total). Adjacent tissues were sampled at least 2 cm from the tumor margin to avoid microscopic infiltration. The patient cohort included individuals aged 27–90 years (median, 59.37), consisting of 17 males and 13 females. Primary tumor sites were distributed among buccal (n=8), tongue (n=11), floor of mouth (n=3), gingival (n=5), lip (n=2), and palatal (n=1). Disease stages ranged from I to IV according to the AJCC 8^th edition, and all specimens were pathologically confirmed as squamous cell carcinoma. None of the patients had received preoperative chemotherapy or radiotherapy. All samples were formalin-fixed and paraffin-embedded (FFPE) and collected with informed consent. The study was conducted in accordance with the Declaration of Helsinki and approved by the institutional ethics committee (Approval No. (2024) CDYFYYLK (12–312)).

Tissue sections were deparaffinized, rehydrated, and incubated with 3% hydrogen peroxide. Antigen retrieval was performed in 10 mM sodium citrate buffer (pH 6.0) for 10 min at sub-boiling temperature, followed by blocking with 5% BSA and incubation with primary antibodies. immunohistochemistry was visualized using a DAB kit (ZSGB, China) and counterstained with hematoxylin. Slides were dehydrated, mounted, and scanned with an Olympus NanoZoomer scanner. The RAPTOR index was quantified using ImageJ 1.53(Bethesda, MD) as the proportion of RAPTOR–positive cells relative to total cells in randomly selected fields.

CELL CULTURE AND LENTIVIRAL TRANSDUCTION:

HNSCC cell lines (Cal27, HSC3, HN30) were obtained from Procell (Wuhan, China) and cultured in DMEM (Gibco) with 10% FBS and 1% penicillin-streptomycin at 37 °C, 5% CO₂. Cells at 60–70% confluence were transduced with shNC or shRPTOR lentivirus (ObiO Technology, Shanghai, China) at an MOI of 10 in 8 μg/mL polybrene, followed by medium replacement and puromycin selection (2 μg/mL) for 5–7 days. Transduction efficiency was confirmed by RT-qPCR and western blot. All cell lines were routinely screened for mycoplasma contamination and confirmed to be free of contamination. In addition, the identity of each cell line was verified by short tandem repeat (STR) profiling. Cells were used for experiments before passage 10 to ensure experimental consistency.

CELL PROLIFERATION AND COLONY FORMATION:

Cell proliferation was assessed using CCK-8, seeding 6,000 cells per well in 96-well plates and measuring absorbance at 450 nm at 24, 48, 72, and 96 h. Rapamycin (0–100 nM; MedChemExpress, USA) was applied for 48 h where indicated. For colony formation, cells were plated in 6-well plates for 10–14 days, fixed with 4% paraformaldehyde, stained with 0.05% crystal violet, and colonies containing ≥50 cells were counted.

MIGRATION AND INVASION ASSAYS:

Wound healing assays were performed on cells grown to >90% confluence in 6-well plates; wounds were created with a pipette tip, and images were captured at 0 and 24 h. Cell invasion was assessed using Matrigel-coated Transwell inserts (8 μm; Corning) with 2×10⁴ cells in serum-free medium in the upper chamber and 700 μL DMEM with 20% FBS in the lower chamber. After 48 h, invaded cells were fixed, stained, and counted in four random fields (100×).

RNA ISOLATION AND REVERSE TRANSCRIPTION-QUANTITATIVE POLYMERASE CHAIN REACTION (RT-QPCR):

Briefly, total RNA was isolated from HNSCC cells using TRIzol (Invitrogen, Carlsbad, CA, USA) and then reverse-transcribed to synthesize cDNA. RT-qPCR was conducted using ChamQ Universal SYBR qPCR Master Mix (Vazyme, China). Results are expressed as ΔΔCt values normalized to GAPDH as relative transcript levels compared with controls. The primer sequences used in this study were as follows:

and

WESTERN BLOTTING:

Cells were lysed in RIPA buffer containing protease and phosphatase inhibitors (Solarbio, China), and protein concentrations were measured using a BCA assay. Equal amounts of protein were separated by SDS-PAGE and transferred onto PVDF membranes (Millipore). Membranes were blocked with 5% skim milk, incubated with the following primary antibodies: anti-RAPTOR (Proteintech, Cat#20984-1-AP; 1: 2000 for western blot and 1: 200 for immunohistochemistry), anti-S100A8 (Proteintech, Cat#15792-1-AP; 1: 1000), anti-S100A9 (Proteintech, Cat#26992-1-AP; 1: 1000), anti-uPA (Cell Signaling Technology, Cat#15800; 1: 1000), anti-uPA/Urokinase (Proteintech, Cat#17968-1-AP; 1: 1000), anti-PI3K p85α (Immunoway, Cat#YM8045; 1: 1000), phospho-PI3K p85α (Tyr607; Immunoway, Cat#YP0765; 1: 1000), anti-AKT (Proteintech, Cat#10176-2-AP; 1: 2000), phospho-AKT (Ser473; Proteintech, Cat#66444-1-Ig; 1: 2000), anti-mTOR (Immunoway, Cat#YM8208; 1: 2000), phospho-mTOR (Ser2448; Immunoway, Cat#YM8326; 1: 2000), anti-LC3A/B (Cell Signaling Technology, Cat#12741; 1: 1000), anti-Ki67 (Immunoway, Cat#YM8189; 1: 200 for immunohistochemistry), and anti-β-tubulin (Servicebio, Cat#GB11017-100; 1: 1000). Membranes were then incubated with horseradish peroxidase-conjugated secondary antibodies, and protein bands were visualized using an enhanced chemiluminescence (ECL) system (UElandy, China).

RNA SEQUENCING:

RNA sequencing was performed by Majorbio Bio-Pharm Biotechnology Co., Ltd. (Shanghai, China). Total RNA was extracted from the indicated cell samples using TRIzol (Invitrogen), and libraries were prepared and sequenced by OEbiotech (Shanghai, China). Differentially expressed genes (DEGs) were identified using limma (|fold change| >2, adjusted P<0.05). Gene ontology (GO) enrichment and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses were performed using Goatools and Python SciPy, respectively.

SUBCUTANEOUS XENOGENEIC MODEL:

Five-week-old female nude mice (Hangzhou Ziyuan, Nanjing, China) were randomly assigned to experimental groups (n=6 per group). All animal procedures were performed in accordance with protocols approved by the Animal Research Committee of the First Affiliated Hospital of Nanchang University (Approval No. CDYFY-IACUC-202505GR062).

Cal27 cells (3×10⁶; NC or shRPTOR) were suspended in 50 μL PBS/Matrigel (3: 1) and injected subcutaneously into the dorsal region. For rapamycin treatment, 3×10⁶ Cal27 cells were implanted subcutaneously, and once tumors reached ~100 mm³, mice received intraperitoneal injections of vehicle or rapamycin (5 mg/kg/day in 5% DMSO/corn oil) for 2 weeks. Tumor growth was measured twice weekly, and volume was calculated as V=(length×width²)/2. Tumor burden was assessed by serial tumor volume measurements; tumor weight was not recorded at the experimental endpoint. Body weight and activity were monitored for systemic toxicity. At the endpoint, tumors were harvested for histopathology and immunohistochemistry. The Ki67 index was quantified using ImageJ 1.53 (Bethesda, MD) as the percentage of Ki67–positive nuclei among total nuclei in randomly selected fields.

STATISTICAL ANALYSES:

Data were analyzed with GraphPad Prism 9 (San Diego, CA) and are presented as mean±SEM. Comparisons between the 2 groups were performed using unpaired t-test, while multiple-group comparisons were analyzed by one-way ANOVA followed by Tukey’s post hoc test. P values <0.05 were considered statistically significant.

Results

PROTEINS WITH PROGNOSTIC RELEVANCE IN HNSCC:

Univariable Cox regression identified 18 prognostic factors significantly associated with OS in HNSCC patients, including 9 that displayed a protective effect (HR <1) and 9 that were associated with increased risk (HR>1). Forest plots showed hazard ratios (HRs) and 95% confidence intervals for each protein (Figure 1A). Specifically, the univariable analysis demonstrated HR for CD45 (HR=0.603), RAPTOR (HR=4.725), SETD2 (HR=0.637), MERIT40_pS29 (HR=0.134), HER3_pY1289 (HR=4.814), Hexokinase-I (HR=2.079), and BETACATENIN (HR=0.809). A volcano plot visualized their differential expression patterns (Figure 1B). All 18 candidates were subsequently subjected to LASSO regression analysis for feature selection, which refined the protein signature (Figure 1C, 1D).

A multivariable Cox regression model was constructed, incorporating 7 key proteins: CD45, RAPTOR, SETD2, MERIT40_pS29, HER3_pY1289, Hexokinase-I, and BETACATENIN. In the multivariable analysis, the adjusted HRs were CD45 (HR=0.74), RAPTOR (HR=1.74), SETD2 (HR=0.83), MERIT40_pS29 (HR=0.22), HER3_pY1289 (HR=1.90), Hexokinase-I (HR=1.28), and BETACATENIN (HR=0.83). The final prognostic risk score for each patient was calculated using the following formula derived from the model coefficients: Risk Score = (−0.29) × CD45 + 0.55 × RAPTOR + (−0.17) × SETD2 + (−1.5) × MERIT40_pS29 + 0.64 × HER3_pY1289 + 0.29 × Hexokinase-I + (−0.17) × BETACATENIN, where each variable represents its normalized protein expression level. In this model, positive coefficients indicate an association with increased risk, whereas negative coefficients correspond to reduced risk. The proportional hazards assumption was evaluated using Schoenfeld residuals. To appropriately account for non-proportional hazards, we constructed a time-dependent Cox model. Biomarkers that violated the proportional hazards assumption were modeled using time-varying coefficients, while those satisfying the assumption were retained as time-fixed covariates. Biomarkers including HER3_pY1289, MERIT40_pS29, and Hexokinase-I violated this assumption (P<0.05), indicating that their hazard ratios varied significantly over time. In contrast, biomarkers such as BETACATENIN, SETD2, RAPTOR, and CD45 satisfied the assumption, suggesting stable, time-fixed effects. This classification was formally confirmed by chi-square tests for time-dependence, which yielded consistent results (Figure 1E–1G). Individual dynamic risk score trajectories were plotted for representative patients stratified into low- and high-risk groups. The trajectories demonstrated that the separation in predicted risk between the 2 groups was maintained over the observed follow-up period (Figure 1H).

Baseline clinical characteristics of the total cohort, along with the training and testing sets, are summarized. Data are presented as number (percentage). The cohort included 353 patients, 177 in the training set and 176 in the testing set. Categorical variables included Age (>65 vs ≤65 years), Sex, Tumor Grade (G1–G2, G3–G4, and unknown), Tumor stage (I–II, III–IV, and unknown), T stage (T1–T2, T3–T4, and unknown), N stage (N0, N1–N3, and unknown), and M stage (M0, M1, and unknown). Unknown categories were included for missing data: Grade unknown in 17 (4.8%) cases, Stage unknown in 29 (8.2%) cases, T stage unknown in 20 (5.7%) cases, N stage unknown in 48 (13.6%) cases, and M stage unknown in 237 (67.1%) cases. Distribution between training and testing sets for these unknown categories was similar: Grade unknown (training 7, 4.0%; testing 10, 5.7%), Stage unknown (training 16, 9.0%; testing 13, 7.4%), T unknown (training 11, 6.2%; testing 9, 5.1%), N unknown (training 24, 13.6%; testing 24, 13.6%), and M unknown (training 127, 71.8%; testing 110, 62.5%). P-values were calculated using Chi-square or Fisher’s exact test for categorical comparisons between the training and testing sets (Figure 1I). No significant differences were observed between the training and testing sets across all categorical variables (P>0.05), confirming balanced data partitioning. Furthermore, a heatmap visualizing key clinical features across risk-stratified patients illustrated the distribution of sex, tumor grade, and tumor stage, highlighting the clinical relevance of the risk groups (Figure 1J).

ASSESSMENT OF PROGNOSIS-RELATED PROTEIN MODEL:

A protein-based risk score was calculated for each patient to predict prognosis based on the expression levels of differentially expressed prognostic proteins. Using the R packages “survival” and “survminer,” OS was compared between high- and low-risk groups, revealing significantly better survival in the low-risk group across the overall cohort, training set, and testing set (Figure 2A–2C). Progression-free survival (PFS) analysis also showed significantly improved outcomes in the low-risk group (P<0.001, Figure 2D). Further analysis demonstrated that higher protein risk scores were positively associated with mortality, a trend consistently observed in the overall cohort (Figure 2E, 2H), training set (Figure 2F, 2I), and testing set (Figure 2G, 2J). Expression profiles of prognostic proteins were illustrated by heatmaps for the overall cohort (Figure 2K), training set (Figure 2L), and testing set (Figure 2M).

Correlations between protein risk scores and clinical indicators associated with HNSCC were further evaluated. Patients were stratified into subsets by T stage (T1–T2, T3–T4), N stage (N0, N1–N3), overall stage (I–II, III–IV), grade (G1–G2, G3–G4), sex (male, female), and age (>65, ≤65). Within each subset, patients were classified into high- and low-risk groups, and Kaplan-Meier analysis consistently showed shorter OS in the high-risk groups across all subsets (Figure 3A–3L).

CONSTRUCTION OF NOMOGRAM MODEL:

In univariable Cox regression analysis, Age (HR>1, P<0.05), advanced Stage (Stage IV vs Stage I; HR>1, P<0.05), and Risk Score (HR>1, P<0.05) were significantly associated with worse OS (Figure 4A). In the multivariable Cox model, after adjusting for all covariates, Age (HR>1, P<0.05), Stage IV (vs Stage I; HR>1, P<0.05), and Risk Score (HR>1, P<0.05) retained independent prognostic significance (Figure 4B). Because the nomogram was constructed based on a Cox proportional hazards model, it reflects the average effect of each predictor over the follow-up period. The proportional hazards assumption was assessed using Schoenfeld residuals and showed no significant violation (global test P=0.340), indicating stable covariate effects over time. Importantly, although individual proteins may exhibit time-dependent effects, the composite protein-based risk score demonstrated a stable, time-constant prognostic effect in our cohort. In time-dependent ROC analysis, the model achieved an AUC of 0.717 (95% CI: 0.648–0.786) at 1 year, 0.690 (95% CI: 0.630–0.765) at 3 years, and 0.619 (95% CI: 0.575–0.736) at 5 years, indicating a gradual decline in discrimination with prolonged follow-up. Performance metrics further revealed that the model maintained higher specificity (0.788–0.878), precision (0.400–0.935), and accuracy (0.548–0.742) at earlier time points, while sensitivity remained moderate (0.462–0.580) across all intervals. Collectively, these results indicate robust and consistent prognostic performance, particularly within the first 3 years after diagnosis (Figure 4C–4E). A comprehensive nomogram was constructed to predict 1-, 3-, and 5-year OS by integrating Age, Sex (male vs female), Grade, Stage, T category, N category, and risk–score stratification. The model demonstrated good calibration, with calibration curves showing close agreement between predicted and observed survival probabilities at all 3 time points. Based on the nomogram, the median predicted mortality rates were 26.8% at 1 year, 37.6% at 3 years, and 49.8% at 5 years. This nomogram serves as a practical, individualized tool for survival estimation by aggregating points assigned to each prognostic factor (Figure 4F, 4G). Survival analysis stratified by nomogram-derived risk quartiles revealed significant prognostic discrimination. Kaplan-Meier curves demonstrated clear separation among the 4 risk groups. Patients in the lowest-risk quartile (Q1) exhibited the most favorable survival outcomes, whereas those in the highest-risk quartile (Q4) had the poorest prognosis (Figure 4H). These findings confirm that the nomogram effectively identifies distinct risk subgroups with significantly different long-term survival probabilities.

CLINICAL VALUE OF PROGNOSTIC-RELATED PROTEINS:

Patients were stratified into high- and low-expression groups based on the median expression score of the 7 proteins. Kaplan-Meier survival analysis revealed that all 7 protein signatures were significantly associated with OS (CD45, P<0.001; SETD2, P=0.004; MERIT40_pS29, P<0.001; BETACATENIN, P=0.001; RAPTOR, P<0.001; HER3_pY1289, P<0.001; Hexokinase-I, P=0.010) (Figure 5). High RAPTOR, HER3_pY1289, and Hexokinase-I expression correlated with unfavorable outcomes, whereas elevated CD45, SETD2, MERIT40_pS29, and BETACATENIN expression were significantly associated with improved survival.

PAN-CANCER ANALYSIS OF RPTOR:

RPTOR encodes the core scaffolding protein of the mTORC1 complex, RAPTOR, which serves as a central regulator of anabolic metabolism by recruiting substrates and sensing nutrient and growth factor signals [14]. Through this mechanism, it coordinates protein and lipid synthesis while suppressing autophagy. This gene plays a crucial role in cancer development and progression, as its overexpression markedly promotes tumor cell proliferation and invasion, thereby making it a critical target for anticancer therapy. In addition, RAPTOR is also involved in the regulation of metabolic diseases and developmental processes, highlighting its essential role in maintaining cellular growth and metabolic homeostasis.

Currently, studies focusing on RAPTOR in cancer are relatively limited. To further elucidate its potential role in tumorigenesis, we first performed a pan-cancer analysis. Based on TCGA data, RPTOR was found to be differentially expressed across multiple tumor types (Figure 6A). Specifically, RPTOR expression was significantly upregulated in HNSCC, BRCA, COAD, and LIHC (P<0.05), while being downregulated in cancers such as KICH and THCA. Analysis of paired tumor and normal tissues further validated these results, revealing that RAPTOR expression was markedly elevated in cancers such as LUAD, HNSCC, and STAD, whereas it was reduced in KIRP (Figure 6B). These findings suggest that RPTOR exhibits heterogeneous expression patterns across different tumor types and may contribute to tumorigenesis through diverse mechanisms.

We next explored the potential immunological functions of RPTOR. Correlation analysis demonstrated that RPTOR expression was significantly associated with the infiltration of multiple immune cell types (|R|>0.3, P<0.05). RPTOR expression showed positive correlations with CD8+ T cells, macrophages, and dendritic cells, while being negatively correlated with regulatory T cells (Tregs) in certain cancers (Figure 6C). Moreover, RPTOR expression was significantly correlated with tumor microenvironment (TME) indices, including StromalScore and ESTIMATEScore (P<0.05, Figure 6D), suggesting that RPTOR may influence TME remodeling by modulating immune and stromal components.

Analysis of immune-related genes further revealed that RPTOR was significantly associated with a broad spectrum of immune modulators, including pro-inflammatory cytokines, chemokines, and their receptors (P<0.01, Figure 6E). Importantly, RPTOR expression was strongly correlated with canonical immune checkpoint molecules such as PDCD1, CTLA4, and CD274 across multiple cancers (P<0.001), with particularly strong associations observed in HNSCC (Figure 6F). Collectively, these findings indicate that RPTOR plays a potential role in immune signaling regulation and may contribute to tumor immune evasion. Additionally, the correlation between RPTOR and microsatellite instability (MSI) was assessed across TCGA cancer types (Figure 6G). Significant positive correlations were observed in CESC, THCA, PRAD, and MESO (all P<0.01), while negative correlations were found in DLBC, HNSC, and KIRC (all P<0.05). No significant associations were detected in other malignancies. To further validate the bioinformatics analysis results, this study first examined the expression difference of RAPTOR in HNSCC tissues and corresponding adjacent normal tissues. The results showed that RAPTOR was significantly upregulated in tumor tissues compared to adjacent normal tissues (Figure 6H). This finding supports the preliminary bioinformatics conclusions at the clinical sample level.

RAPTOR IS HIGHLY EXPRESSED IN HNSCC, AND SILENCING RAPTOR INHIBITED THE PROLIFERATION AND INVASION OF HNSCC:

To investigate the functional mechanism of RAPTOR in HNSCC, we constructed short hairpin RNA (shRNA) lentiviral vectors targeting RPTOR and selected 3 HNSCC cell lines – Cal27, HSC3, and HN30 – for subsequent experiments. Transfection with 2 silencing viruses (shRPTOR-1 and shRPTOR-2) effectively reduced the mRNA expression level of RPTOR in all 3 cell lines (Figure 7A). Among them, shRPTOR-2 exhibited higher interference efficiency than shRPTOR-1, resulting in a further decrease in RPTOR mRNA expression. Western blot results further confirmed that both shRNAs significantly inhibited RAPTOR expression at the protein level, with shRPTOR-2 showing a more pronounced suppressive effect, consistent with the changes observed at the mRNA level (Figure 7B).

To evaluate the impact of RAPTOR silencing on cell proliferation, we performed CCK-8 assays. RAPTOR silencing significantly inhibited the proliferation of all 3 cell lines, with the shRPTOR-2 treatment group exhibiting a more substantial inhibitory effect, suggesting a positive correlation between the phenotypic outcome and the efficiency of RAPTOR knockdown (Figure 7C). Consistent results were obtained from the colony formation assay: RAPTOR silencing markedly reduced the clonogenic ability of the cells, and shRPTOR-2 produced a stronger inhibitory effect (Figure 7D).Regarding cell migration and invasion ability, results from the wound healing and Transwell assays demonstrated that knockdown of RAPTOR expression significantly suppressed the migration and invasion of all 3 HNSCC cell lines, with the shRPTOR-2 group showing a more potent inhibition (Figure 7E, 7F).

Furthermore, a mouse xenograft model was generated by subcutaneous injection of Cal27 cells transduced with shRPTOR-2. Tumor burden was assessed by serial tumor volume measurements. Compared with the control, the shRPTOR group exhibited a significant reduction in tumor volume (Figure 7G, 7H). Immunohistochemistry analysis demonstrated a significant decrease in Ki67 positivity in shRPTOR tumors, indicating suppressed proliferative activity (Figure 7I).

RAPTOR SILENCING SUPPRESSES HNSCC PROGRESSION THROUGH UPREGULATION OF S100A8/A9 AND DOWNREGULATION OF UPA VIA MTORC1 PATHWAY INHIBITION:

To investigate the downstream molecular alterations associated with RPTOR knockdown, RNA-seq analysis was performed in Cal27 cells transfected with shRPTOR compared with shNC controls. Differential expression analysis identified a distinct transcriptomic landscape, as shown by heatmaps and volcano plots, with multiple genes significantly upregulated or downregulated upon RPTOR silencing (Figure 8A, 8B). Functional enrichment of these DEGs identified significant GO terms across biological processes, cellular components, and molecular functions, suggesting broad involvement in tumor-associated pathways (Figure 8C). Consistently, KEGG analysis demonstrated that DEGs were enriched in signaling cascades relevant to cancer development and progression (Figure 8D). Gene Set Enrichment Analysis (GSEA) revealed that RPTOR knockdown significantly suppressed pathways related to cell proliferation and invasion (Figure 8E, 8F).

S100A8/A9 (calprotectin), a calcium-binding heterodimeric complex composed of S100A8 and S100A9 subunits, is a key member of the S100 protein family. In HNSCC, S100A8/A9 expression is notably downregulated, and this reduction is closely associated with malignant progression, largely due to epigenetic silencing mechanisms such as hypermethylation of the CpG islands in the S100A9 promoter. Functionally, loss of intracellular S100A8/A9 leads to dysregulation of the G2/M cell cycle checkpoint and promotes cancer cell proliferation, while its expression negatively regulates multiple oncogenes (e.g., MMP1, MMP2, INHBA, LAMC2), thereby inhibiting tumor cell invasion, migration, and metastasis [12,15–17]. Based on these findings, GSEA analysis of the S100A family was conducted, which indicated upregulated expression of S100A members in the shRPTOR group (Figure 8G). Subsequent RT-qPCR and western blot validation confirmed that RAPTOR knockdown significantly increased the expression levels of S100A8/A9 (Figure 9A–9C).

Urokinase-type plasminogen activator (uPA), encoded by the PLAU gene, is a serine protease that facilitates tumor progression via proteolytic cascades. It converts plasminogen to plasmin, a broad-spectrum protease that degrades key components of the extracellular matrix (ECM) and basement membrane. In HNSCC, uPA binding to its receptor (uPAR) activates downstream signaling pathways such as MAPK and PI3K/Akt, promoting tumor cell proliferation, survival, epithelial–mesenchymal transition (EMT), and conferring treatment resistance [18,19]. Accordingly, we examined PLAU expression using RT-qPCR and western blot. The results demonstrated that RAPTOR knockdown significantly downregulated the expression levels of PLAU (Figure 9A–9C). As RAPTOR is a component of the mTORC1 complex, we also investigated classical mTOR signaling pathways, including PI3K, AKT, and mTOR, as well as the autophagy marker LC3. Western blot results indicated that RAPTOR knockdown suppressed phosphorylation of PI3K, AKT, and mTOR, thereby inhibiting activation of these pathways. Concurrently, increased LC3 expression suggested enhanced autophagy in HNSCC cells upon RAPTOR knockdown (Figure 9D).

RAPAMYCIN RECAPITULATES THE ANTI-TUMOR EFFECTS OF RAPTOR SILENCING BY CONVERGING ON S100A8/A9 UPREGULATION AND UPA SUPPRESSION IN HNSCC:

Due to the lack of highly selective inhibitors targeting RAPTOR directly and considering that RAPTOR is an essential component of the mTORC1 complex, we employed rapamycin – a well-characterized mTOR inhibitor – to further investigate whether pharmacological inhibition of mTORC1 signaling could suppress proliferation and invasion in HNSCC cell lines [20,21]. Using 3 representative HNSCC cell lines (Cal27, HSC3, and HN30), we performed functional validations.

CCK-8 assays showed that rapamycin treatment significantly suppressed cell proliferation of all 3 HNSCC cell lines in a dose-dependent manner, with a concentration of 20 nM already eliciting statistically significant effects, while 50–100 nM produced the most pronounced reduction in viability (Figure 10A). Based on these findings, 50 nM was selected as the working concentration for subsequent colony formation, migration, and invasion assays, as it provided robust inhibition without causing excessive cytotoxicity. Consistently, colony formation assays demonstrated a marked reduction in clonogenic growth (Figure 10B). Wound healing and Transwell assays further indicated that rapamycin impaired both migratory and invasive capacities of HNSCC cells (Figure 10C, 10D).

In vivo, xenograft models demonstrated that rapamycin administration effectively inhibited tumor growth (Figure 10E, 10F). Immunohistochemistry analysis of tumor tissues showed decreased expression of Ki67, indicating reduced proliferative activity (Figure 10H). Body weights remained stable across treatment groups (Figure 10G), and hematoxylin and eosin (H&E) staining of major organs showed no overt pathological changes (Figure 10I), supporting the tolerability and favorable safety profile of rapamycin under these conditions.

To determine whether the mechanistic effects of rapamycin overlap with those of RPTOR knockdown, we examined the expression of key molecular markers. Consistent with the results observed in RPTOR-silenced cells, rapamycin treatment upregulated the expression of S100A8/A9 and downregulated PLAU at both mRNA and protein levels. Additionally, western blot analysis showed that rapamycin suppressed phosphorylation of PI3K, AKT, and mTOR, while promoting the expression of LC3, indicating enhanced autophagy activation (Figure 9E, 9F).

These results collectively demonstrate that rapamycin recapitulates the phenotypic and molecular effects of RAPTOR silencing, inhibiting proliferation and invasion in HNSCC cells through convergent mechanisms involving suppression of PI3K/AKT/mTOR signaling, induction of autophagy, and modulation of S100A8/9 and PLAU expression.

Discussion

Proteomic profiling has emerged as a pivotal tool for dissecting molecular mechanisms underlying tumorigenesis [22]. Unlike genomic or transcriptomic analyses, proteomics directly reflects functional protein activity and cellular signaling, offering a more accurate representation of malignant behavior [23,24]. Accordingly, proteomic approaches provide a complementary framework for identifying functionally relevant prognostic markers in HNSCC.

In this study, we retrieved rigorously quality-controlled Level 4 proteomic data from TCPA database and performed integrated analysis with matched clinical survival data from TCGA. Through rigorous Cox proportional hazards regression analysis employing stringent statistical thresholds (P<0.01 with Benjamini-Hochberg correction), we identified 18 protein biomarkers with significant prognostic value. Notably, elevated expression of RAPTOR, HER3_pY1289 and Hexokinase-I showed significant negative associations with OS, whereas higher levels of CD45, SETD2, MERIT40_pS29, and BETACATENIN correlated with prolonged OS. Consistent with previous studies, HER3 has been implicated in therapy resistance through modulation of EGFR and PI3K/AKT signaling [25,26]. Conversely, CD45 (leukocyte common antigen) is associated with favorable outcomes; elevated CD45 expression correlates with improved locoregional recurrence-free survival and disease-specific survival, while low CD45 levels increase the risk of recurrence twofold [27].

To construct a prognostic model, multivariable Cox regression analysis was performed on 7 proteins (CD45, RAPTOR, SETD2, MERIT40_pS29, HER3, Hexokinase-I, and BETACATENIN), and the risk score was calculated as follows: Risk score = (−0.29) × CD45 + 0.55×RAPTOR + (−0.17)×SETD2 + (−1.5)×MERIT40_pS29 + 0.64×HER3 + 0.29×Hexokinase-I + (−0.17)×BETACATENIN. Stratification into high- and low-risk groups revealed significantly poorer prognosis in the high-risk group (P<0.05), consistently observed across clinically relevant subgroups including tumor stage (T1–2/T3–4), nodal status (N0/N1–3), clinical stage (I–II/III–IV), histological grade (G1–G2/G3–G4), sex, and age (≤65/>65 years). Multivariable analysis confirmed that the risk score is an independent prognostic factor (P<0.001). The constructed nomogram demonstrated robust predictive performance, underscoring its potential utility for risk stratification in HNSCC.

Notably, the 7-protein prognostic signature identified in our study differs from that reported in a previous TCGA-based proteomic analysis, which identified ERα, HER3, BRAF, P27, RAPTOR, and E2F using a similar analytical framework [28]. These discrepancies may be attributed to several factors. First, the TCGA and TCPA databases are periodically updated; our analysis leveraged a more recent data release with extended follow-up duration and revised clinical annotations compared with the earlier study. Second, the stochastic nature of resampling-based feature selection algorithms, particularly in high-dimensional proteomic datasets with correlated features, can yield variation in selected predictors across independent analyses.

Despite these differences, a key point of convergence between the 2 studies is the independent identification of RAPTOR and HER3 as prognostically relevant markers. Given that RAPTOR is an obligate component of the mTORC1 complex and a central integrator of metabolic and growth-related signals, its recurrent selection across independent prognostic models highlights its context-specific biological importance in HNSCC rather than a mere analytical overlap. This convergence further reinforces the relevance of mTOR signaling in HNSCC progression and provides a strong rationale for the subsequent functional investigations conducted in the present study.

To experimentally validate these findings, RAPTOR was selected for functional studies. Lentiviral-mediated RAPTOR knockdown in HNSCC cell lines significantly inhibited cell proliferation, as demonstrated by CCK-8 and colony formation assays. Subsequent Transwell and wound healing assays further revealed that RAPTOR silencing impaired migration and invasion. Furthermore, in a murine xenograft model, RAPTOR knockdown effectively reduced tumor growth. To elucidate the alterations following RAPTOR knockout, we conducted RNA sequencing. The results indicated suppression of proliferation- and invasion-related pathways, and downregulation of S100A8/A9 and PLAU, which was confirmed at both mRNA and protein levels by RT-qPCR and western blot. Additionally, RAPTOR silencing reduced phosphorylation of PI3K, AKT, and mTOR, while increasing LC3 protein expression, indicating enhanced autophagy. These findings provide insight into the role of RAPTOR in regulating key signaling pathways and cell behaviors in HNSCC, supporting its importance in tumor progression.

Given the pronounced effects of RPTOR silencing, we further explored its therapeutic potential. Although no direct RAPTOR-targeting drugs are currently available, we employed rapamycin, an mTOR inhibitor, to target the RAPTOR-containing mTOR complex. Consistent with genetic silencing, rapamycin treatment suppressed proliferation and invasion in vitro and tumor growth in vivo, downregulated S100A8/A9 and uPA expression, attenuated phosphorylation of PI3K, AKT, and mTOR, and increased LC3 expression. We acknowledge that rapamycin is not a RAPTOR-specific inhibitor and may exert effects beyond mTORC1, including potential involvement of additional mTOR-related pathways. The phenotypic concordance observed between RAPTOR knockdown and rapamycin treatment therefore suggests overlapping involvement of the mTORC1 signaling pathway, rather than exclusive dependence on RAPTOR alone. Future studies integrating combinatorial genetic and pharmacological approaches will be required to further delineate RAPTOR-specific functions from broader mTOR signaling activities and to refine therapeutic strategies targeting this pathway in HNSCC.

Several limitations warrant consideration. First, the predictive model was constructed and validated using retrospective public datasets (TCGA and TCPA), predominantly representing European and North American populations, which may limit generalizability. Future studies should include prospective, multi-center, and ethnically diverse cohorts to validate its robustness. Second, the precise upstream and downstream mechanisms by which RAPTOR regulates S100A8/A9 and uPA remain to be fully elucidated. Further investigations into these molecular interactions are needed to gain deeper insights into their functional relevance in HNSCC.

Conclusions

In summary, we developed and validated a 7-protein prognostic model for HNSCC (CD45, RAPTOR, SETD2, MERIT40_pS29, HER3_pY1289, Hexokinase-I, and BETACATENIN) that robustly stratifies patient risk across multiple cohorts and clinical subgroups. Functional analyses identified RAPTOR as a key regulator of HNSCC progression, associated with modulation of the S100A8/A9–uPA axis and PI3K/AKT/mTOR signaling. Both genetic silencing and pharmacologic inhibition of RAPTOR recapitulated these effects, supporting its functional involvement in HNSCC progression. Within the limitations of retrospective datasets and experimental models, these findings highlight the relevance of protein-based prognostic signatures and RAPTOR-associated pathways in HNSCC.