SERPINE1-centric inflammatory signature associates with treatment resistance and survival in laryngeal squamous cell carcinoma

Abstract

Background

Laryngeal squamous cell carcinoma (LSCC) prognosis remains poor despite treatment advances. More accurate prognostic assessment models can help guide individualized treatment and improve prognosis. Chronic inflammation contributes to tumorigenesis, yet inflammatory response-related genes (IRGs) in LSCC prognosis are underexplored. This study aimed to construct an IRG prognostic signature for LSCC and further dissect core IRG-mediated mechanisms of immune escape and chemoresistance.

Methods

Transcriptional profiles and clinical data from LSCC patients were retrieved from The Cancer Genome Atlas (TCGA). IRGs were sourced from Gene Set Enrichment Analysis (GSEA) hallmark gene set. We identified differentially expressed IRGs linked to survival outcomes in LSCC. Key IRGs were subsequently selected using least absolute shrinkage and selection operator (LASSO) Cox regression analysis to establish an inflammatory risk score model. This model underwent internal validation within the TCGA cohort and external validation using independent Gene Expression Omnibus (GEO) datasets. We further assessed the model’s association with the tumor immune microenvironment and the impact of IRGs on chemotherapy response. Finally, the functional roles of interested signature IRG were experimentally validated in LSCC cell lines.

Results

Four significant IRGs (AQP9, ITGA5, LCK, SERPINE1) were identified to build the risk score model. The model stratified LSCC patients into distinct prognostic groups: TCGA cohort: 5-year area under the curve (AUC) =0.836, P<0.001; GSE25727 cohort: 5-year AUC =0.706, P=0.02; GSE27020 cohort: 5-year AUC =0.798, P<0.01. Multivariate analysis confirmed the risk score as an independent prognostic factor (P<0.05). High-risk patients showed reduced immune cell infiltration (CD8+ T cells, dendritic cells) and suppressed immune pathways. Multi-algorithm immune analysis further revealed defective antigen presentation and reduced anti-tumor immune infiltration in high-risk LSCC, promoting tumor immune escape. GSEA/Gene Ontology (GO) enrichment combined with drug sensitivity prediction further revealed that high-risk tumors activate invasive signaling and acquire broad chemoresistance alongside impaired anti-tumor immunity. SERPINE1 might be associated with chemotherapy resistance and exhibited the highest alteration frequency (predominantly amplification) and overexpression in LSCC tissues. Its knockdown significantly suppressed proliferation, migration, invasion and chemoresistance in LSCC cells. Immunohistochemistry (IHC) confirmed tumor SERPINE1 overexpression (P=0.002 vs. normal tissues), correlating with poor survival (P<0.001).

Conclusions

The 4-IRG risk signature is a reliable prognostic indicator reflecting immune dysfunction in LSCC. SERPINE1 is validated as a therapeutic target and biomarker, enriching our understanding of gene regulation dynamics in LSCC.

Highlight box.

Key findings

• A 4-gene inflammatory risk signature (AQP9, ITGA5, LCK, SERPINE1) was built and validated across The Cancer Genome Atlas and two Gene Expression Omnibus cohorts as an independent prognostic marker for laryngeal squamous cell carcinoma (LSCC).

• Multi-omics analyses revealed high-risk LSCC features impaired antigen presentation, suppressed anti-tumor immunity, activated invasive pathways and broad chemoresistance, driving immune escape.

• SERPINE1, the core oncogenic gene in this signature, promotes LSCC progression and drug resistance, which was confirmed by in vitro assays and tissue immunohistochemistry.

What is known and what is new?

• Chronic inflammation fuels LSCC development, but inflammatory gene-based prognostic models remain insufficiently studied.

• We developed a dedicated inflammatory signature for LSCC and systematically uncovered its immune-escape and chemoresistance mechanisms via multi-algorithm bioinformatics.

What is the implication, and what should change now?

• This signature enables risk stratification to guide individualized chemo-immunotherapy; SERPINE1 is a promising prognostic biomarker and therapeutic target for LSCC.

• Prospective clinical validation and further functional research on AQP9/ITGA5/LCK are required to translate these findings.

Introduction

Laryngeal cancer represents the third most common head and neck malignancy, with laryngeal squamous cell carcinoma (LSCC) accounting for >90% of cases (1). LSCC comprises 25–30% of all head and neck squamous cell carcinomas (HNSCCs) (2). Globally, laryngeal cancer constituted 0.9% of new cancer diagnoses in 2022, ranking 20th among malignancies (3). Over the past four decades, the 5-year survival rate declined from 66% to 61%, with mortality increasing concurrently. Early diagnosis, timely treatment, disease surveillance, and accurate prognostication are therefore critical for improving LSCC outcomes.

Despite therapeutic advances, LSCC mortality primarily results from local recurrence and metastasis (4). Developing refined prognostic models to guide individualized treatment thus remains essential for improving outcomes. Accumulating evidence reveals significant histological and molecular parallels between tumorigenesis and inflammation (5). Both processes involve overlapping inflammatory cells, mediators, and molecular pathways during initiation and progression. The tumor inflammatory microenvironment - comprising macrophages, dendritic cells, interleukins, interferons, and chemokines - mirrors complex immune responses observed in infection and wound healing (6). In LSCC specifically, elevated derived neutrophil-to-lymphocyte ratio (dNLR) correlates with greater invasiveness and advanced tumor, node, metastasis (TNM) stage (7), highlighting the inflammatory-immune microenvironment’s critical role in disease progression. Chronic inflammation consequently represents a recognized tumorigenesis driver (8), promoting carcinogenesis through direct oncogene mutagenesis or cytokine-mediated responses (9). While inflammatory gene markers show prognostic utility in cancers like hepatocellular carcinoma and lung adenocarcinoma (10,11), their prognostic value in LSCC remains insufficiently explored.

This study aimed to characterize inflammatory response-related genes (IRGs) predictive of survival in LSCC. Utilizing The Cancer Genome Atlas (TCGA) dataset, we constructed an inflammatory risk score model based on identified IRGs to establish its prognostic utility in LSCC. Model performance was rigorously validated through time-dependent receiver operating characteristic (TD-ROC) and Kaplan-Meier analyses. We further examined associations between constituent IRGs in our model and tumor immune infiltration characteristics, as well as their correlations with drug sensitivity profiles. Finally, serpin family E member 1 (SERPINE1) was prioritized from signature genes for functional validation, assessing its expression and impact on LSCC cell biology. We present this article in accordance with the MDAR and TRIPOD reporting checklists (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1190/rc).

Methods

Data source

RNA-seq data from LSCC cases (STAR-counts workflow; format: tab-separated values) were obtained from TCGA (https://portal.gdc.cancer.gov/). After excluding cases lacking complete clinical data (age, gender, grade, stage, overall survival), 111 LSCC and 12 normal samples were retained. A gene set of 200 IRGs corresponding to “HALLMARK_INFLAMMATORY_RESPONSE” was acquired from MSigDB (http://www.gsea-msigdb.org/gsea/msigdb/). The GENCODE annotation file (gencode.v36.annotation.gtf.gz) enabled gene identifier conversion. Using the limma R package, differentially expressed IRGs were identified with thresholds of |log₂ fold change (FC)| >1.0 and false discovery rate (FDR) <0.05. Prognostic IRGs were determined via univariate Cox regression and log-rank tests (survival R package). Genes with inter-gene correlation >0.20 were analyzed, and overlapping differentially expressed/prognostic IRGs (P<0.05) were visualized using the venn R package.

Construction and validation of the inflammatory risk score

Least absolute shrinkage and selection operator (LASSO) Cox regression identified optimal weight coefficients for prognostic IRGs. Following 1,000 iterations of cross-validation to maximize the likelihood penalty estimate, the optimal λ value was selected using the minimum criterion, establishing the inflammatory risk score. To account for potential heterogeneity in baseline characteristics and clinical management across validation cohorts, we determined cohort-specific optimal cut-offs for risk stratification using TD-ROC analysis. This exploratory approach maximized the model’s discriminatory power in each independent population. Validation proceeded through TD-ROC analysis and Kaplan-Meier comparisons of high/low-risk groups. Internal validation used the TCGA-LSCC cohort (n=111), while external validation employed two Gene Expression Omnibus (GEO) datasets (GSE25727, n=56; GSE27020, n=109).

Independent prognostic analysis

Univariate and multivariate Cox regression analyses determined whether age, gender, grade, clinical stage, and risk score were independent prognostic indicators. LSCC grading criteria: G1: well differentiated, G2: moderately differentiated, G3: poorly differentiated, G4: undifferentiated. Clinical staging followed American Joint Committee on Cancer (AJCC) 8th edition TNM. Variables with P<0.05 in univariate analysis entered multivariate modeling. Factors retaining P<0.05 in multivariate analysis were deemed independent prognostic indicators. The entire cohort had sufficient data for independent analyses of the inflammatory risk score in univariate and multivariate Cox regression analyses.

Immune subtypes, infiltration, and pathway analyses

Immune subtyping: TCGA LSCC samples were classified into six immune subtypes per Thorsson et al. (12): C1 (wound healing), C2 (IFN-γ dominant), C3 (inflammatory), C4 (lymphocyte depleted), C5 (immunologically quiet), and C6 (TGF-β dominant). C5 was excluded due to absence in LSCC. Risk score variations across subtypes were assessed using Kruskal-Wallis tests (P<0.05) with visualization via ggplot2.

Immune cell infiltration: single-sample Gene Set Enrichment Analysis (ssGSEA) implemented through the GSVA R package quantified immune cell enrichment levels across risk groups within the tumor microenvironment.

Immune pathways: enrichment scores for immune pathways in high/low-risk groups were calculated using published gene sets (13).

In addition to ssGSEA and quanTIseq, three extra immune deconvolution algorithms (CIBERSORT, MCP-counter, EPIC) were used to cross-validate immune cell proportions. The ESTIMATE algorithm was applied to calculate stromal score, immune score, composite ESTIMATE score and tumor purity of each sample. We adopted published gene signatures to calculate mRNA-based stemness index (mRNAsi) and antigen presentation machinery (APM) scores via ssGSEA. Wilcoxon rank-sum tests were used for subgroup comparisons, and Spearman correlation analysis was used to assess associations with riskScore; all P values were corrected via the Benjamini-Hochberg FDR method to reduce false positives.

Differential expression analysis between high- and low-risk subgroups was performed via limma-trend. GSEA was conducted against MSigDB Hallmark and Kyoto Encyclopedia of Genes and Genomes (KEGG) gene sets to compare pathway enrichment trends, with normalized enrichment score (NES) and FDR calculated. Gene Ontology (GO) enrichment was implemented on risk-group differential expressed genes (DEGs) for cross-verification. The oncoPredict R package trained on the GDSC database was used to predict half-maximal inhibitory concentration (IC₅₀) values of common LSCC chemotherapeutics and EGFR inhibitors for each sample. Wilcoxon rank-sum test and Benjamini-Hochberg FDR correction were applied for all group comparisons.

Immune correlation and network analyses

TCGA LSCC RNA-seq data (STAR-counts, level 3) with clinical metadata were analyzed. Spearman’s correlation (P<0.05) assessed associations between non-normally distributed variables, visualized via pheatmap. Immuno-correlation analysis was conducted according to the following steps: (I) STAR-counts were converted to transcripts per million (TPM) and log₂(TPM+1)-normalized. (II) Clinical data-filtered samples were retained. (III) Immune scores were estimated using immunedeconv (quanTIseq algorithm). (IV) Networks were constructed and visualized via ggClusterNet (significance threshold: P<0.05).

Immune checkpoint correlation analysis

We analyzed eight immune checkpoint genes: CD274, CTLA4, HAVCR2, LAG3, PDCD1, PDCD1LG2, TIGIT, and SIGLEC15. Expression values for these transcripts and IRGs were extracted from RNA-seq datasets. IRG expression groups (high/low) were stratified by median expression and visualized using ggplot2 and pheatmap.

Chemosensitivity assessment

Chemotherapeutic responses were predicted for TCGA LSCC samples using the GDSC database (https://www.cancerrxgene.org/)—the largest public pharmacogenomics resource. The pRRophetic R package estimated IC₅₀ via ridge regression with: default parameters; combat batch correction; all tissue types included; duplicate genes mean-aggregated.

Multi-platform validation of IRGs

Risk score IRGs were validated across biological scales: (I) tissue-level: TCGA RNA-seq data (https://portal.gdc.com). (II) Cellular-level: Cancer Cell Line Encyclopedia (CCLE; https://portals.broadinstitute.org/ccle). (III) Mutational frequency: cBioPortal (http://www.cbioportal.org/) using “Head and Neck Carcinoma, Other (CPTAC GDC 2025)” cohort (n=49 LSCC). Immunohistochemical (IHC) validation of protein expression in normal squamous epithelium and HNSCC tissues was performed via Human Protein Atlas.

Cell lines and culture

Human LSCC cell lines TU212 and TU138 were obtained from the Shanghai Institute of Biochemistry and Cell Biology (Shanghai, China). Cells were cultured in RPMI-1640 medium (Gibco, Shanghai Branch, China) supplemented with 10% fetal bovine serum, 100 µg/mL penicillin, and 100 µg/mL streptomycin at 37 ℃ in 5% CO₂ humidified atmosphere.

siRNA transfection

SERPINE1 was silenced using two siRNAs (GenePharma, Suzhou, China):

• ❖ siRNA#1: 5'-UCUCUGCCCUCACCAACAUUCTT-3'.

• ❖ siRNA#2: 5'-AGACCAACAAGUUCAACUAUATT-3'.

Transient transfection was performed with Lipofectamine 3000 (Invitrogen, Carlsbad, CA, USA) according to manufacturer’s protocol, with analysis 48 h post-transfection.

RNA extraction and qRT-PCR

Total RNA was isolated using RaPure Total RNA Micro Kit (Magen, Guangzhou, China). cDNA was synthesized with RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher, Waltham, MA, USA). qPCR was performed using ChamQ SYBR Master Mix (Vazyme, Shanghai, China, Q311-02) with glyceraldehyde-3-phosphate dehydrogenase (GAPDH) normalization. Primers:

• ❖ SERPINE1: F 5'-ACCGCAACGTGGTTTTCTCA-3',

• ❖ R 5'-TTGAATCCCATAGCTGCTTGAAT-3'.

• ❖ GAPDH: F 5'-ATCACCATCTTCCAGGAGCGA-3',

• ❖ R 5'-CCTTCTCCATGGTGGTGAAGAC-3'.

• ❖ Relative expression was calculated by 2^−ΔΔCt method.

Cell proliferation assay

Proliferation was assessed using Cell Counting Kit-8 (CCK-8; Dojindo, Kumamoto, Japan). Cells seeded in 96-well plates (2,000 cells/well) were transfected with SERPINE1 siRNA or negative control. At 24, 48, 72, and 96 h post-transfection, cells were incubated with CCK-8 reagent for 2 h. Viability was determined by measuring absorbance at 450 nm.

Colony formation assay

SERPINE1-knockdown and control cells (500 cells/well) were cultured in 6-well plates for 14 days. Colonies were fixed with 4% paraformaldehyde (10 min), stained with crystal violet (15 min), and imaged. Clones containing >50 cells were counted using ImageJ (NIH).

Migration and invasion assays

Wound healing: confluent monolayers (90%) in 6-well plates were transfected for 48 h, scratched with 100-µL pipette tips, washed with phosphate-buffered saline (PBS), and imaged at 0/24/48 h. Wound closure was calculated as: (initial width − final width)/initial width × 100%.

Transwell invasion: performed using BD BioCoat Matrigel Invasion Chambers per manufacturer’s protocol (BD Biosciences, Franklin Lakes, NJ, USA).

Immunohistochemistry (IHC)

The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. All IHC experiments were approved by the Institutional Research Ethics Committee of Sun Yat-sen University Cancer Center (No. B2022-628-01) and separate patient written consent for archived de-identified biobank specimens was waived. Twenty paraffin-embedded LSCC tissue sections from the Tumor Biobank of Sun Yat-sen University Cancer Center (surgically resected 2010-2025, no chemotherapy) were immunostained with anti-SERPINE1 antibody (1:100, sc-5297, Santa Cruz, Santa Cruz, CA, USA) and secondary antibody (1:200, GB23303, Servicebio, Wuhan, China). Two blinded pathologists independently quantified staining intensity and percentage in tumor cells.

Statistical analysis

Continuous variables were dichotomized. Risk score-clinicopathologic correlations were assessed by chi-square test. Patients were stratified by median risk score using R packages ’survival’/’survminer’. Predictive accuracy was evaluated by TD-ROC curves with area under the curve (AUC) calculated (’survivalROC’). Kaplan-Meier analysis with log-rank tests determined survival differences. Analyses used SPSS 24.0 and R 3.6.2 (significance threshold: P<0.05). All experiments included three biological replicates.

Results

Development of an inflammatory risk score derived from prognosis-related IRGs

The overall methodological framework is summarized in Figure 1A. Transcriptomic analysis of TCGA-LSCC data revealed differential expression of 85/200 IRGs from the ‘HALLMARK_INFLAMMATORY_RESPONSE’ gene set (GSEA) in LSCC tissues (Table S1), with 75 upregulated and 10 downregulated. Additionally, 12 IRGs exhibited significant associations with LSCC prognosis (Figure S1, Table S2). Intersection of these gene sets identified six prognosis-related IRGs: aquaporin 9 (AQP9), bone marrow stromal cell antigen 2 (BST2), integrin subunit alpha 5 (ITGA5), lymphocyte-specific protein tyrosine kinase (LCK), poliovirus receptor (PVR), and SERPINE1 (Figure 1B). A heatmap visualizing their expression patterns in LSCC versus normal tissues is presented in Figure 1C. These six IRGs functioned as independent prognostic indicators and demonstrated significant inter-gene correlations. AQP9, ITGA5, PVR, and SERPINE1—all associated with increased mortality risk—showed positive mutual correlations. Conversely, BST2 and LCK (potential protective factors) were positively correlated with each other but negatively correlated with the high-risk IRG group, except for BST2-PVR (Figure 1D,1E).

Figure 1.

Identification of differentially expressed prognosis-related IRGs in LSCC. (A) Workflow of the screening of prognostic IRGs and construction of the proportional hazard regression model. (B) Venn plot showing 6 differentially expressed and prognosis-related IRGs genes. (C) Heat map of the 6 screened IRGs. (D) Forest plot of the 6 screened IRGs. (E) Correlation analysis between the 6 screened IRGs. Gene pairs lacking statistically significant correlations were not connected by lines. (F) Variable selection via LASSO Cox regression with 1,000 bootstrap iterations. X-axis (top): number of variables. X-axis (bottom): log(λ). Y-axis: partial likelihood deviance. (G) LASSO coefficient curves of candidate IRGs. Y-axis represents the regression coefficient weight of each gene. CI, confidence interval; DEG, differentially expressed gene; HNSCC, head and neck squamous cell carcinoma; IRG, inflammatory response-related gene; LASSO, least absolute shrinkage and selection operator; LSCC, laryngeal squamous cell carcinoma; ROC, receiver operating characteristic; TCGA, The Cancer Genome Atlas.

Using maximum likelihood estimation and LASSO Cox regression applied to the six prognosis-related IRGs, we developed a prognostic model optimized through 1,000 bootstrap iterations (Figure 1F). The regularization parameter lambda (λ) was selected via the one-standard-error criterion to determine optimal feature weights (Figure 1G). This process yielded a 4-IRG signature, with the inflammatory risk score calculated as:

Inflammatory risk score = (0.3319 × expression of AQP9) + (0.2090 × expression of ITGA5) − (0.5965 × expression of LCK) + (0.1184 × expression of SERPINE1).

We performed Kaplan-Meier survival analyses of the 4 IRGs using TCGA-LSCC data. AQP9, ITGA5 and SERPINE1 were verified as high-risk genes associated with poor survival, while LCK exerted a protective effect, clearly demonstrating their respective prognostic impacts on the risk model (Figure S2).

Validation of the prognostic inflammatory risk score

The prognostic performance of the inflammatory risk score was first evaluated internally in the TCGA-LSCC cohort. TD-ROC analysis demonstrated AUC values of 0.662, 0.784, and 0.836 for 1-, 3-, and 5-year overall survival, respectively (Figure 2A). Using the maximum Youden index to determine the optimal cutoff, patients were stratified into risk groups. Kaplan-Meier analysis revealed significantly worse survival in high-risk patients (log-rank P<0.001; Figure 2B), with dot plots confirming superior outcomes in the low-risk group (Figure 2C).

Figure 2.

Validation of the inflammatory risk score model. (A) Time-dependent ROC analysis of the risk model in TCGA-LSCC. (B) Survival difference between risk groups (TCGA-LSCC). (C) Outcome comparison dot plot (TCGA-LSCC). (D) Time-dependent ROC analysis (GSE25727 cohort). (E) Survival difference (GSE25727). (F) Outcome dot plot (GSE25727). (G) Time-dependent ROC analysis (GSE27020 cohort). (H) Survival difference (GSE27020). (I) Outcome dot plot (GSE27020). AUC, area under the curve; LSCC, laryngeal squamous cell carcinoma; ROC, receiver operating characteristic; TCGA, The Cancer Genome Atlas.

External validation was performed using two independent GEO cohorts. In GSE25727 (n=56), the model achieved AUCs of 0.682 (1-year), 0.694 (3-year), and 0.706 (5-year) (Figure 2D). Consistently, high-risk patients exhibited poorer survival (log-rank P=0.02; Figure 2E), with dot plots verifying risk stratification (Figure 2F). Similar results were observed in GSE27020 (n=109), with AUC values of 0.635 (1-year), 0.724 (3-year), and 0.798 (5-year) (Figure 2G). High-risk patients again demonstrated inferior overall survival (log-rank P<0.01; Figure 2H) and clear outcome differentiation (Figure 2I).

Independent prognostic value of the risk score

Cox proportional hazards regression incorporating both the risk score and clinical characteristics (age, gender, grade, clinical stage) identified independent prognostic factors for LSCC. Univariate analysis demonstrated significant associations between overall survival and the risk score (P<0.05), tumor grade (P<0.05), and clinical stage (P<0.05; Figure 3A). These factors remained significant independent predictors in multivariate analysis (P<0.05; Figure 3B). Notably, risk scores showed no significant association with age (P=0.18) or gender (P=0.66), but were positively correlated with both tumor grade and clinical stage (P<0.05; Figure 3C-3F), indicating elevated risk scores in advanced disease.

Figure 3.

Univariate and multivariate Cox proportional hazards regression analyses, and patient risk scores in different clinical subgroups. (A) Forest plot of the univariable analysis. (B) Forest plot of the multivariable analysis. (C-F) Patient risk scores in different clinical subgroups, including age (>65 vs. ≤65 years), gender (female vs. male), grade (grade 1–2 vs. grade 3–4), and clinical stage (stage I–II vs. stage III–IV). Data are shown as the mean ± SD. CI, confidence interval; SD, standard deviation.

Elevated inflammatory risk score is associated with diminished immune infiltration in LSCC

Given the critical interplay between inflammation and tumor immunity, we employed ssGSEA to quantify immune function enrichment and assess correlations between risk scores and immune status in TCGA LSCC cases. Risk scores varied significantly across LSCC immune subtypes (P<0.05), indicating interactions between signature IRGs and immune biology (C1: wound healing; C2: IFN-γ dominant; C3: inflammatory; C4: lymphocyte depleted; C6: TGF-β dominant). Notably, higher risk scores were significantly associated with C1 and C4 subtypes (Figure 4A).

Figure 4.

Immune-based analysis of the inflammatory risk score. (A) The level of risk score among different immune subtypes in LSCC. Relative enrichment scores of 16 types of immune cells (B) and 13 immune-related pathways (C) in the high and low risk groups according to the median score were shown. (D) Immune score correlation network diagram. The lower left heat map indicates the correlation between different immune scores. The upper right line indicates the correlation between inflammatory risk score and immune scores. Data are shown as the mean ± SD. **, P<0.01; ***, P<0.001; ns, not significant. LSCC, laryngeal squamous cell carcinoma; SD, standard deviation.

Analysis of 16 immune cell subsets revealed reduced infiltration in high-risk patients compared to low-risk counterparts, including diminished CD8⁺ T, Th1, Th2, and Tfh lymphocytes. Antigen-presenting activity mediated by aDCs and iDCs was also attenuated in the high-risk group (Figure 4B). Furthermore, key immune pathways demonstrated compromised activity with elevated risk scores: CCR, pro-inflammatory signaling, T cell co-stimulation/co-inhibition, and type II IFN response were suppressed in high-risk patients (Figure 4C). These findings suggest LSCCs with high inflammatory risk scores possess immunologically inert microenvironments characterized by impaired immune infiltration and signaling, potentially enabling tumor immune evasion.

We additionally quantified risk score correlations with specific immune cell populations. Consistent with preceding results, the risk score inversely correlated with anti-tumorigenic immune cells including B cells (r=−0.38, P<0.001), NK cells (r=−0.22, P=0.02), CD8⁺ T cells (r=−0.41, P<0.001), and myeloid dendritic cells (r=−0.33, P<0.001). While modest positive associations were observed with M1 macrophages (r=0.18, P=0.04), neutrophils (r=0.21, P=0.02), and CD4⁺ T cells (r=0.17, P=0.05), alongside reduced Treg activity (r=−0.16, P=0.06), the pronounced negative correlations with B and CD8⁺ T cells indicate substantially compromised anti-tumor immunity (Figure 4D).

Multi-algorithm immune landscape, stemness and antigen presentation analyses stratified by the risk score

Multi-algorithm immune deconvolution, immune checkpoint, stemness and antigen presentation analyses were performed in TCGA-LSCC dataset to further dissect how the inflammatory risk score reshape tumor immunity. Cross-verification across 4 independent deconvolution tools revealed consistent immune disparities between subgroups. Cytotoxic immune populations including CD8+ T cells, B cells and dendritic cells were significantly reduced in high-risk LSCC, whereas unpolarized M0 macrophages accumulated prominently (Figure 5A). ESTIMATE analysis further supported this immune-cold phenotype: high-risk LSCC had markedly lower immune scores (Wilcoxon P<0.001), with no significant difference in stromal scores and slightly elevated tumor purity (Figure 5B). The CIBERSORT stacked bar chart intuitively displayed the gradual loss of effector immune cells as risk score increased (Figure 5C).

Figure 5.

Multi-algorithm immune deconvolution, tumor stemness and antigen presentation analyses stratified by the risk score in TCGA-LSCC cohort. (A) Cross-algorithm bubble plot (CIBERSORT/MCP-counter/EPIC/quanTIseq): blue = low-risk enriched immune cells, orange = high-risk enriched cells; bubble size = −log₁₀(FDR). (B) ESTIMATE stromal, immune, composite scores and tumor purity between high/low risk groups, Wilcoxon P labeled. (C) CIBERSORT stacked bar chart showing immune cell fractions sorted by risk score. (D) Spearman correlation bubble plot of risk score and eight immune checkpoints, bubble size = −log₁₀(FDR). (E) Boxplots and scatter correlations of ssGSEA mRNAsi (stemness) and APM scores versus risk score. Wilcoxon test for group comparison, Spearman correlation with Benjamini-Hochberg FDR correction. APM, antigen presentation machinery; FDR, false discovery rate; LSCC, laryngeal squamous cell carcinoma; SD, standard deviation; ssGSEA, single-sample Gene Set Enrichment Analysis; TCGA, The Cancer Genome Atlas.

We next analyzed correlations between risk core and 8 canonical immune checkpoint genes. Risk score exhibited consistent negative correlations with all 8 markers, with significant inverse associations observed for LAG3 (ρ=−0.59), PDCD1 (ρ=−0.57), TIGIT (ρ=−0.50) and CTLA4 (ρ=−0.39) (Figure 5D). Since immune checkpoints are mainly expressed on infiltrated T cells, this trend indicated high-risk tumors suffered from T-cell immune ignorance instead of typical T-cell exhaustion.

We finally evaluated tumor stemness and antigen presentation capacity using ssGSEA scores (Figure 5E). The APM score was significantly decreased in the high-risk subgroup (P=3×10⁻⁴, ρ=−0.38), indicating defective antigen delivery. By contrast, mRNAsi stemness scores showed no statistical difference or linear correlation between risk groups (P=0.55, ρ=0.05).

In summary, these analyses demonstrate that elevated risk score suppresses anti-tumor immune cell infiltration and impairs antigen presentation, jointly constructing an immune-suppressive microenvironment that facilitates tumor immune escape in LSCC.

GSEA and GO enrichment combined with chemosensitivity prediction dissect tumor-related core pathways

We conducted GSEA and GO enrichment on DEGs in risk-group TCGA-LSCC patients, paired with transcriptome-based drug response prediction. GSEA results were grouped into inflammation, proliferation, invasion and chemoresistance modules. Immune and proliferative gene sets (antigen presentation, TCR/BCR signaling, E2F targets) were enriched in low-risk patients, while angiogenesis and extracellular matrix (ECM) remodeling pathways driving metastasis were significantly upregulated in high-risk cases, as validated by representative enrichment curves (Figure 6A,6B). GO enrichment further supported these trends: genes overexpressed in high-risk tumors were enriched in cell adhesion and developmental programs linked to invasion, whereas downregulated genes centered on adaptive immune processes (Figure 6C).

Figure 6.

GSEA, GO enrichment and predicted chemosensitivity analyses stratified by the risk score in TCGA-LSCC cohort. (A) Bar plot of significantly enriched GSEA signatures grouped into immune/inflammation, proliferation and invasion modules. (B) Running enrichment score curves of three representative pathways (angiogenesis, antigen processing, T cell receptor signaling). (C) Bubble plot of GO biological process enrichment for DEGs upregulated/downregulated in high-risk tumors. (D) Box plots of predicted IC₅₀ values for standard chemotherapy and EGFR inhibitors, ns, non-significant. All statistical comparisons adopted Wilcoxon rank-sum test with Benjamini-Hochberg FDR correction. *, P<0.05; **, P<0.01; ***, P<0.001; ns, not significant. BH, Benjamini-Hochberg; DEG, differentially expressed gene; FDR, false discovery rate; GO, Gene Ontology; GSEA, Gene Set Enrichment Analysis; IC₅₀, half-maximal inhibitory concentration; LSCC, laryngeal squamous cell carcinoma; NES, normalized enrichment score; TCGA, The Cancer Genome Atlas.

Since classic drug metabolism signatures showed no statistical significance in GSEA, we adopted oncoPredict to estimate IC₅₀ values. High-risk patients exhibited elevated predicted resistance to platinum, taxane, fluoropyrimidine and vinca alkaloid agents, with no intergroup difference for EGFR inhibitors (Figure 6D).

Collectively, high-risk LSCC is characterized by dampened anti-tumor immunity, activated invasive signaling and broad chemoresistance, which jointly explain its unfavorable clinical outcomes.

Correlation of signature IRGs with immune status

Following the risk score-immune infiltration analysis, we examined relationships between individual signature IRGs and immune responses. SERPINE1 and ITGA5 showed coordinated negative correlations with antitumor immune cells (B cells, DCs, and CD8⁺ T cells; r=−0.31 to −0.42, P<0.001), suggesting their roles in promoting immune evasion. Conversely, LCK demonstrated significant positive associations with B cells (r=0.38, P<0.001) and CD8⁺ T cells (r=0.45, P<0.001), indicating its contribution to antitumor immunity. AQP9 exhibited complex immunomodulatory patterns with positive correlations across macrophage subsets, neutrophils, CD8⁺ T cells, and regulatory T cells (Tregs; r=0.18–0.33, P<0.05; Figure 7A).

Figure 7.

Immune correlation analysis of IRGs in the model. (A) Spearman correlation analysis between IRGs and immune scores. (B-E) Association between different IRG levels and the expression of immune checkpoint molecules. Based on the median value, IRG level was stratified into high and low expression groups. Data are shown as the mean ± SD. *, P<0.05; **, P<0.01; ***, P<0.001. IRG, inflammatory response-related gene; SD, standard deviation.

Regarding immune checkpoints critical for tumor evasion, high expression of AQP9, ITGA5, or SERPINE1 correlated with elevated CD274 (PD-L1), HAVCR2, and PDCD1LG2 levels (P<0.05). Notably, the LCK high-expression group showed increased expression across all eight examined checkpoint molecules compared to low-expression counterparts (P<0.01; Figure 7B-7E).

IRG expression and therapeutic sensitivity

Given the clinical relevance of chemo-immunotherapy combinations in advanced LSCC, we assessed relationships between signature IRGs and chemotherapy sensitivity. Elevated AQP9 expression correlated with tumor resistance to cetuximab (r=0.32, P<0.01; Figure 8A), while high ITGA5 was associated with cisplatin resistance (r=0.28, P=0.02; Figure 8B). Conversely, LCK overexpression enhanced sensitivity to cetuximab (r=−0.41, P<0.001), cisplatin (r=−0.38, P<0.001), and 5-fluorouracil (r=−0.35, P<0.01; Figure 8C). SERPINE1 upregulation significantly increased IC₅₀ values for all tested agents (cetuximab, cisplatin, 5-fluorouracil, docetaxel; P<0.05; Figure 8D), indicating its pivotal role in chemotherapy resistance.

Figure 8.

Spearman correlation analysis between IRG expression and drug IC50 values. (A‑D) Correlations of sensitivity to cetuximab, cisplatin, paclitaxel, and 5‑fluorouracil with AQP9 (A), ITGA5 (B), LCK (C), and SERPINE1 (D). IC₅₀, half-maximal inhibitory concentration; IRG, inflammatory response-related gene.

Characterization of signature IRGs using bioinformatic databases

We analyzed expression patterns and mutational profiles of the four signature IRGs across tumor tissues and cancer cell lines. All IRGs showed significantly elevated expression in LSCC tissues, with SERPINE1 demonstrating the highest relative expression (Figure 9A). Analysis of the CCLE revealed pronounced ITGA5 and SERPINE1 overexpression in both LSCC and HNSCC cell lines, while AQP9 and LCK exhibited comparatively lower expression (Figure 9B,9C). Consistently, all four IRGs displayed higher protein expression in HNSCC tissues versus normal squamous epithelium (Figure 9D).

Figure 9.

Analysis of IRGs included in the model using online databases. (A) The mRNA expression in TCGA LSCC samples. (B,C) The mRNA expression in LSCC cell lines and HNSCC cell lines, determined by RNA sequencing of the four IRGs from CCLE. (D) Immunohistochemistry revealed the protein expression levels of IRGs in both normal tissues and HNSCC. (E) Genetic alterations of the 4 IRGs in LSCC obtained from cBioPortal. Data are shown as the mean ± SD. ***, P<0.001. CCLE, Cancer Cell Line Encyclopedia; HNSCC, head and neck squamous cell carcinoma; IRG, inflammatory response-related gene; LSCC, laryngeal squamous cell carcinoma; SD, standard deviation; TCGA, The Cancer Genome Atlas.

Mutation analysis using cBioPortal (Head and Neck Squamous Cell Carcinoma, Other Subtypes cohort; CPTAC GDC 2025; n=49 LSCC) identified distinct mutational distributions (Figure 9E). SERPINE1 showed the highest alteration frequency (predominantly gene amplification). Given SERPINE1’s marked overexpression in LSCC and established associations with immune evasion and chemoresistance, we prioritized it for functional investigation.

Functional validation and clinical relevance of SERPINE1 in LSCC

siRNA-mediated knockdown confirmed SERPINE1 suppression at the mRNA level (Figure 10A). Functional assays demonstrated that SERPINE1 suppression significantly inhibited proliferation in LSCC cell lines (TU138, TU212; P<0.01; Figure 10B) and reduced clonogenic potential (P<0.001; Figure 10C), indicating its essential role in sustaining LSCC growth.

Figure 10.

Functional validation and clinical relevance of SERPINE1 in LSCC. (A) siRNA-mediated knockdown of SERPINE1 in LSCC cells. SERPINE1 depletion inhibited LSCC cell proliferation demonstrated by: CCK-8 assay (B) and colony formation assay (C). Impaired migration and invasion were confirmed via wound healing assay (D, 10× objective) and Transwell assay (E). Migrated cells in (E) were stained with crystal violet. (F) Immunohistochemistry showing elevated SERPINE1 expression in LSCC tissues. (G) High SERPINE1 expression correlates with poor survival (*, P<0.05). (H,I) Dose-response curves and IC50 values of cisplatin and paclitaxel in siNC and siSERPINE1 TU138 cells detected by CCK-8 assay. (J) KEGG pathway enrichment analysis of differentially expressed genes after SERPINE1 knockdown in TU138 cells. Data represent mean ± SD from triplicate experiments with triplicate wells. *, P<0.05; **, P<0.01; ***, P<0.001 vs. NC group. CCK-8, Cell Counting Kit-8; CI, confidence interval; HR, hazard ratio; IC₅₀, half-maximal inhibitory concentration; KEGG, Kyoto Encyclopedia of Genes and Genomes; LSCC, laryngeal squamous cell carcinoma; NC, negative control; OD, optical density; SD, standard deviation.

SERPINE1 silencing also attenuated cell migration and invasion capabilities (P<0.01; Figure 10D,10E), suggesting involvement in metastatic progression. IHC analysis of 20 paired early-stage LSCC specimens confirmed significant tumor upregulation versus adjacent normal tissues (P<0.001; Figure 10F). Patients with high SERPINE1 expression (median IHC cutoff) exhibited worse overall survival (log-rank P<0.001; Figure 10G).

We measured the IC₅₀ of cisplatin and paclitaxel in SERPINE1-silenced TU138 cells, and both drugs exhibited enhanced cytotoxicity after SERPINE1 knockdown (Figure 10H,10I). Transcriptome KEGG enrichment revealed multiple signaling pathways potentially involved in SERPINE1-mediated chemoresistance, including tumor necrosis factor (TNF), interleukin (IL)-17, JAK-STAT, Ras, Hippo, lysosome and ribosome signaling pathways (Figure 10J).

Collectively, SERPINE1 functions as an oncogene in LSCC with prognostic biomarker potential, providing mechanistic insights complementary to our risk prediction model and suggesting novel therapeutic avenues.

Discussion

The inflammation-cancer nexus is well established. Many malignancies either originate from inflammatory milieus or generate inflammation to facilitate progression (14). This paradigm extends to HNSCC, including LSCC. Pro-inflammatory diets elevate serum biomarkers like C-reactive protein and IL-6, while inflammatory transcription factors enhance HNSCC risk (15). Inflammation underpins both tumorigenesis and immune evasion in human papillomavirus (HPV)-associated HNSCC (16). Moreover, constitutive NF-κB activation in HNSCC amplifies inflammatory signaling within tumor microenvironments (17), and disrupted TGF-β and PTEN pathways drive tumor-promoting inflammation (18). Chronic mucosal inflammation from smoking/alcohol exposure constitutes a major LSCC risk factor, yet inflammation-LSCC interactions remain incompletely characterized. Although systemic inflammatory markers show prognostic relevance, molecular exploration of inflammatory mediators in LSCC is nascent (19).

This study integrated inflammation-related genes from GSEA with TCGA-LSCC data to identify prognostic IRGs. Through univariate Cox regression and log-rank testing focusing on differentially expressed IRG intersections, we identified six prognostically significant genes (Figure 1). LASSO regression—selected for its regularization properties that mitigate overfitting in high-dimensional data—refined the signature to four core genes (AQP9, ITGA5, LCK, SERPINE1) with substantial coefficients. We developed an inflammatory risk score model demonstrating good prognostic stratification in our cohort (Figure 2).

Internal validation showed good predictive performance. Using the TD-ROC-optimized cutoff, TCGA-LSCC patients were stratified into high- and low-risk groups, with the latter exhibiting significantly superior survival (log-rank P<0.001; Figure 2). External validation using independent GEO cohorts (GSE25727, GSE27020) confirmed good prognostic discrimination across datasets (Figure 2D-2I). Of note, the model’s predictive performance exhibited evident heterogeneity between different cohorts. The lower 5-year AUC value observed in GSE25727 relative to the TCGA cohort stemmed from inherent objective disparities across public datasets, including disparate transcriptomic detection platforms, unequal sample sizes, unbalanced clinical baseline characteristics, cohort-customized risk thresholds, and distinct molecular pathological compositions among patients. Importantly, all external validation cohorts yielded AUC values above 0.7, demonstrating that this IRG signature retains reliable moderate prognostic capacity for LSCC despite cross-cohort variation. Given such inter-cohort heterogeneity, cohort-specific cutoffs were empirically determined to maximize clinical applicability for each dataset. This exploratory stratification strategy improves the model’s flexibility, though further validation in larger independent cohorts is required to evaluate potential overfitting risks and consolidate its clinical translational value.

Multivariable Cox regression incorporating our inflammatory risk score with clinical parameters identified tumor grade, stage, and the risk score as independent prognostic factors in LSCC (Figure 3). Notably, gender—an established prognostic variable—was not statistically significant in this model. This may plausibly reflect the TCGA-LSCC cohort’s limited sample size and potential selection bias. Thus, our findings may not fully translate to broader LSCC populations where tumor grade, stage, and male gender (epidemiologically predominant) remain validated predictors. Expanded validation cohorts are warranted for generalizability.

Elevated risk scores correlated significantly with advanced clinical stages and higher tumor grades (Figure 3), confirming the score’s reflection of intrinsic tumor aggressiveness. Subsequent analyses linked lower risk scores to immunologically active phenotypes characterized by enhanced CD8⁺ T cells, DCs, NK cells, and Tfh cell infiltration. Conversely, higher scores predominantly associated with lymphocyte-depleted immune subtypes (Figure 4), indicating suppression mediated by tumor-associated IRGs.

Building on the preliminary single-algorithm immune profiling, we supplemented multi-method deconvolution, antigen presentation and stemness analyses to clarify the immune escape mechanism driven by our risk score (Figure 5). Four independent immune algorithms uniformly confirmed the immune-cold feature of high-risk LSCC, which was further validated by the reduced overall immune scores from ESTIMATE quantification. The widespread negative correlation between risk score and immune checkpoints cannot be simply attributed to T-cell exhaustion; instead, it reflects insufficient T-cell recruitment caused by impaired antigen presentation, as confirmed by our APM scoring results. Notably, tumor stemness showed no meaningful correlation with risk stratification, proving the signature’s prognostic effect mainly originates from inflammatory immune dysregulation rather than cancer stem cell activity. These layered results form a complete mechanistic chain: high risk score inhibits cytotoxic immune infiltration and blocks effective antigen delivery, creating an immune-excluded microenvironment that leads to poor survival outcomes in LSCC (20,21).

Our GSEA and GO pathway analyses further delineated the full regulatory network controlled by the risk score (Figure 6). Low-risk LSCC retain active anti-tumor immune signaling and physiological proliferative programs, while high-risk LSCC repress antigen presentation and lymphocyte recruitment and simultaneously activate angiogenesis and ECM remodeling pathways to fuel invasion and metastasis. Although classic drug metabolism gene sets showed no prominent enrichment in GSEA, transcriptome-based IC₅₀ prediction consistently confirmed that high-risk patients exhibit broad resistance to mainstream platinum, taxane and fluoropyrimidine chemotherapies. These pathway findings perfectly align with our earlier immune escape results: the four-gene inflammatory signature orchestrates a multi-layered malignant phenotype covering immune suppression, enhanced invasiveness and treatment insensitivity, which collectively explains the unfavorable clinical outcomes of high-risk LSCC patients (22,23).

We hypothesize that poorly differentiated LSCC exhibits immune-excluded phenotypes while well-differentiated tumors permit greater immune infiltration. Notably, independent immune profiling of signature IRGs revealed coordinated associations with immune cell infiltration, immune checkpoint expression and chemoresistance markers (Figures 7,8). Interconnectivity among IRGs forms a regulatory network wherein dysregulation (higher AQP9/ITGA5/SERPINE1; lower LCK) culminates in immunosuppressive microenvironments and potential treatment failure. Thus, our risk score may inform partial immunotherapy/chemotherapy response prediction, with higher scores consistently indicating poorer prognosis.

Among the four prognostic IRGs in our model, AQP9 and ITGA5 demonstrate established survival associations in LSCC. Increased AQP9 expression suppresses malignant behaviors, whereas elevated ITGA5 promotes LSCC progression via EFNB2 signaling (24,25). Paradoxically, our study identifies AQP9 as a prognostic detriment, suggesting context-dependent functions. LCK—essential for T-cell activation and anti-tumor immunity (26-29)—exhibits oncogenic roles in select tumors yet shows reduced LSCC/HNSCC expression (validated in CCLE/cBioPortal; Figure 9), implying tumor-suppressive potential. Conversely, ITGA5 and SERPINE1 display expression patterns supportive of oncogenesis. SERPINE1, the most frequently amplified IRG, remains understudied in LSCC.

As a key plasminogen activator regulator, SERPINE1 serves as a validated prognostic biomarker across diverse cancers (30,31). In breast cancer, SERPINE1 and its substrate uPA inform chemotherapy response prediction (32,33). Mechanistically, it drives oncogenesis by sustaining proliferation, inhibiting apoptosis, promoting angiogenesis, facilitating invasion/metastasis and inducing tumor-promoting inflammation (34,35). SERPINE1 orchestrates inflammatory microenvironments through enhanced IL-6 secretion (36), positioning it as a pivotal regulator of cancer-associated inflammation via immune cell recruitment and tissue homeostasis disruption (37).

In LSCC, SERPINE1 upregulation correlates with poor prognosis (38), and its suppression inhibits proliferation, migration, and invasion. Additional studies link SERPINE1 to EMT and invasive progression (39). Our experimental validation concurs: SERPINE1 knockdown potently attenuated LSCC cell growth/proliferation and suppressed migratory/invasive capabilities. IHC analyses complemented by public database interrogation confirmed consistent SERPINE1 overexpression in LSCC, underscoring its dual utility as a prognostic biomarker and therapeutic target. These congruent findings strengthen our model’s reliability (Figure 10).

Transcriptomic KEGG enrichment uncovered candidate mechanisms for SERPINE1-driven chemoresistance. Enriched TNF, IL-17 and JAK-STAT inflammatory pathways match SERPINE1’s core inflammatory function and are proven to attenuate cisplatin and paclitaxel cytotoxicity. Oncogenic Ras and Hippo signaling were also enriched, which enhance tumor stemness and DNA damage repair to resist chemotherapy. Lysosome pathway enrichment suggests SERPINE1 may trigger intracellular drug sequestration, whereas ribosome and EGFR signals specifically counter paclitaxel-mediated growth arrest. Collectively, cellular and transcriptomic data demonstrate that SERPINE1 induces broad-spectrum resistance to routine LSCC chemotherapies by coordinating inflammatory and oncogenic survival cascades (40,41).

This study has limitations. The external validation cohort for our LSCC model featured a limited sample size. To address potential heterogeneity, we calculated risk stratifications separately across subsets, potentially limiting model generalizability in clinical practice. The analysis of immune and chemotherapeutic sensitivity was correlated based on deconvolution algorithms and GDSC/pRRophetic prediction. However, given the relatively small scale of the present study, real-world validation based on directly observed clinical responses could not be performed temporarily. In addition, functional experiments were only performed on SERPINE1, while the biological roles of AQP9, ITGA5 and LCK need further independent cellular validation in future work. While we briefly delineated fundamental roles of SERPINE1 in LSCC, its precise regulatory mechanism and downstream signaling in LSCC remain to be further elucidated. These areas represent key priorities for future work.

Conclusions

We developed and validated an LSCC prognostic model based on inflammatory gene expression. The risk score served as an independent survival predictor with utility in predicting immunomodulation and chemoresistance. Through in vitro experimentation, we identified SERPINE1 as a functional component of this model, offering mechanistic insights to guide therapeutic decisions in LSCC.

Supplementary

The article’s supplementary files as

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. The study was approved by the Institutional Research Ethics Committee of Sun Yat-sen University Cancer Center (No. B2022-628-01) and separate patient written consent for archived de-identified biobank specimens was waived.

Data Sharing Statement

Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2026-1190/dss

tcr-15-08-584-dss.pdf ^{(89.2KB, pdf)}

DOI: 10.21037/tcr-2026-1190