A Hypoxia Associated STC2 High Endothelial Subpopulation is Linked to Lipid Metabolic Reprogramming, Pathological Angiogenesis, and Immune Remodeling in Head and Neck Squamous Cell Carcinoma

Abstract

Purpose

Hypoxia drives malignant advancement and immune escape in HNSCC, but its cell‑type‑specific regulatory circuitry within the TME remains insufficiently elucidated. This study systematically characterizes hypoxia‑associated molecular programs and identifies core genes with relevance to prognosis and therapeutic response.

Methods

Bulk, single‑cell, and spatial transcriptomic datasets from independent patient cohorts were integrated. Hypoxia‑associated gene signatures and two molecular subtypes were constructed through unsupervised clustering. A consensus of eight machine‑learning algorithms identified STC2 as the key hypoxia mediator. Bioinformatics analyses encompassing enrichment, CellChat, scMetabolism, and scTenifoldKnk were conducted. STC2 expression was confirmed by RT‑qPCR, immunohistochemistry, and multiplex immunofluorescence. Functional experiments using STC2‑silenced HUVECs included scratch wound‑healing, CCK‑8, and immunofluorescence staining to assess endothelial migration, viability, and cytoskeletal organization. Candidate drugs were explored through pharmacogenomic databases, molecular docking, and molecular dynamics simulations.

Results

The hypoxia‑derived signature predicted reduced survival, enhanced angiogenesis, and an immunosuppressive TME. STC2 was predominantly enriched in a distinct endothelial subpopulation marked by active hypoxia‑driven fatty acid metabolism and ANGPT2‑expressing angiogenic signaling. In silico knockout of STC2 weakened hypoxia‑induced angiogenic and immune‑related pathways. CellChat analysis revealed altered GALECTIN‑ and ANGPTL‑mediated communication between this endothelial subset and immune populations. High STC2 levels were associated with immunotherapy resistance and with increased sensitivity to four candidate compounds. PLX4032 and Midostaurin demonstrated stable binding interactions with STC2 in computational analyses.

Conclusion

STC2 is preferentially expressed in a hypoxia‑associated endothelial subpopulation characterized by lipid metabolic activity, angiogenic features, and immune‑regulatory crosstalk. Functional assays support a role for STC2 in endothelial viability, migration, and angiogenic function, while its involvement in lipid metabolism and immune modulation requires further investigation. These findings highlight the potential value of STC2 for patient stratification and as a target for future therapeutic exploration in HNSCC.

Graphical Abstract

Introduction

Head and neck squamous cell carcinoma (HNSCC) stands as the sixth most prevalent malignancy globally and imposes a formidable public health burden,^1,2 with annual incidence now exceeding 800,000 cases worldwide.³ Among patients with locally advanced disease, postoperative recurrence surpasses 50% and five‑year survival remains dismal, fueling an escalating disease trajectory.⁴ Although multimodal regimens encompassing surgery, radiotherapy, chemotherapy, and immune checkpoint inhibitors (ICIs) have enhanced locoregional control in selected cases, therapeutic resistance, recurrence, and distant metastasis persistently undermine long‑term prognosis.^5,6 Consequently, deciphering the molecular circuitry governing HNSCC initiation and progression, coupled with the identification of novel biomarkers and actionable targets, is imperative for refining clinical management paradigms.

A quintessential hallmark of solid tumors, hypoxia operates as a principal driver of malignant progression across diverse cancers.^7,8 Rapid tumor cell proliferation in concert with aberrant vasculature engenders insufficient oxygen delivery, which not only intensifies tissue hypoxia but also stimulates tumor‑associated endothelial cell (TEC) expansion and angiogenesis.^9,10 The resultant neovasculature, however, is structurally chaotic and functionally compromised, further deepening regional hypoxia and perpetuating a vicious cycle of hypoxia, angiogenesis, and exacerbated hypoxia.¹¹ Clinical evidence substantiates that the synergistic interplay between hypoxia and pathological angiogenesis markedly aggravates prognosis in HNSCC.¹² Moreover, the intricate crosstalk between hypoxia‑related genes (HRGs) and angiogenic signaling cascades has severely constrained the efficacy of monotherapies targeting either process in isolation.¹³ Hence, systematic deconvolution of the HRG regulatory network and its dialogue with the tumor microenvironment (TME) is essential to pinpoint combinatorial targets that concurrently neutralize hypoxic responses and aberrant angiogenesis, thereby enabling innovative therapeutic strategies.

Advances in single‑cell transcriptomics have afforded unprecedented granularity in defining the cellular landscape of HNSCC.^14,15 Comprehensive single‑cell profiling studies have resolved transcriptionally heterogeneous malignant, stromal, endothelial, and immune populations and have tracked their dynamic reprogramming across tumor evolution.¹⁶ The subsequent integration of spatial transcriptomic technologies has enabled these diverse cellular states and their underlying molecular signatures to be localized within their authentic anatomical context, providing direct insight into the spatially organized architecture of the HNSCC microenvironment.^17,18 In parallel, integrative analyses that merge bulk and single‑cell transcriptomic datasets have systematically linked hypoxia‑associated molecular patterns to clinical prognosis, immune contexture, and intercellular communication networks.^19,20 Emerging multi‑omics frameworks that embed machine learning, single‑cell resolution, and spatial transcriptomic mapping have further validated the power of these convergent approaches to extract clinically relevant, cell‑type‑specific molecular features.² Collectively, these methodological innovations provide a rigorous foundation for interrogating hypoxia‑driven transcriptional programs at cellular and spatial resolution, a departure from the interpretive constraints imposed by conventional bulk tumor profiling.

STC2 is a secreted glycoprotein recognized as a classical hypoxia‑responsive target under direct transcriptional control of HIF1.²¹ Its contribution to HNSCC progression has been substantiated through multiple lines of evidence: STC2 was shown to enhance proliferation, migration, invasion, and metastasis via the PI3K‑AKT‑Snail signaling axis,²² to function as a downstream effector of the HOTAIR-miR‑206 regulatory network,²³ and to correlate with aggressive disease features and unfavorable survival in hypoxia‑based molecular subtyping analyses.^24,25 Although these findings firmly establish a hypoxia–STC2–HNSCC aggressiveness axis, existing studies have largely been restricted to bulk tumor profiling or malignant epithelial cell biology. Consequently, whether STC2 is enriched in a specific non‑malignant stromal compartment, most notably the tumor endothelium, and whether an STC2‑high endothelial state is coupled to coordinated alterations in hypoxia‑driven transcriptional activity, lipid metabolism, angiogenic signaling, and immune‑oriented communication remain unresolved. To address these knowledge gaps, this investigation integrated multiple independent bulk transcriptomic cohorts, three single‑cell RNA‑seq datasets, and spatial transcriptomic data to systematically chart the hypoxia‑associated molecular landscape of HNSCC and nominate core regulators. Consensus machine‑learning screening across eight algorithms converged on STC2 for subsequent characterization. Single‑cell and spatial transcriptomic analyses were then applied to determine its cell‑type‑specific distribution and to define an STC2‑high endothelial subpopulation marked by heightened hypoxia‑related transcriptional activity, fatty acid biosynthetic programs, angiogenic features, and altered communication circuits with immune cells. Functional experiments in STC2‑silenced HUVECs further assessed its role in sustaining endothelial viability, migration, cytoskeletal organization, and angiogenic phenotypes. By repositioning the investigative lens from bulk‑tumor and malignant‑cell‑centric STC2 biology toward its endothelial cellular context, this study delivers a cell‑type‑resolved framework for understanding the interplay between hypoxia‑associated STC2 expression and vascular microenvironmental remodeling in HNSCC.

Materials and Methods

Data Acquisition and Processing

Bulk transcriptomic data were collected from seven independent HNSCC cohorts obtained from The Cancer Genome Atlas, ArrayExpress, and the Gene Expression Omnibus. The TCGA HNSC cohort contained 502 HNSCC samples and 44 adjacent normal samples. The E MTAB 8588 cohort obtained from ArrayExpress comprised 117 HNSCC samples and 98 adjacent normal samples. Five additional cohorts were obtained from GEO, including GSE41613 with 97 HNSCC samples, GSE42743 with 74 HNSCC and 29 adjacent normal samples, GSE65858 with 270 HNSCC samples, GSE75538 with 14 HNSCC and 14 adjacent normal samples, and GSE41116 with 43 HNSCC samples. Gene expression matrices from all cohorts were converted to a log2 scale before integration. Gene identifiers were harmonized to official gene symbols according to the corresponding platform annotation files. Genes shared across the included datasets were retained for subsequent integrated analyses. Batch effects across cohorts were corrected using the ComBat algorithm implemented in the sva R package. Principal component analysis was performed before and after batch correction to evaluate the effectiveness of cohort integration. The resulting integrated expression matrix contained 1117 HNSCC samples and 185 adjacent normal samples. Samples lacking overall survival information were excluded only from survival related analyses.

For single‑cell dissection, three scRNA‑seq datasets (GSE182227, GSE185965, and GSE193766) encompassing 25 HNSCC and 5 normal specimens were curated. Quality control and preprocessing executed via Seurat excluded cells meeting any of the following thresholds: fewer than 200 or more than 5000 detected genes, mitochondrial gene proportion exceeding 15%, or hemoglobin gene proportion exceeding 3%. Inter‑sample batch variation was mitigated with Harmony, after which unsupervised clustering was projected through uniform manifold approximation and projection (UMAP). Cell identities were subsequently ascribed based on canonical lineage markers referenced in prior literature. After annotation of the major cellular populations, endothelial cells were extracted for secondary analysis. Harmony was reapplied to the endothelial cell subset using the first 20 dimensions, followed by dimensionality reduction and clustering at a resolution of 0.9. No additional normalization step was performed after endothelial cell extraction because the normalized expression matrix generated during the initial preprocessing of the complete single cell dataset was retained for subsequent analysis. Endothelial cells were then ranked according to STC2 expression. Cells in the upper quartile were defined as the STC2+ subset, whereas cells in the lower quartile were defined as the STC2-. Cells within the intermediate 50% were excluded from direct comparisons. This quartile based strategy was used to compare cells at the two ends of the continuous STC2 expression distribution while avoiding reliance on an extreme expression cutoff.

Spatial transcriptomic (ST) profiles were acquired from Benjamin et al²⁶ and the hypoxia gene set was retrieved from the Molecular Signatures Database (MSigDB).

Data regarding immunotherapy response and CTRP/PRISM drug sensitivity were retrieved from the BEST database (https://rookieutopia.hiplot.com.cn/app_direct/BEST/).

Identification of Differentially Expressed Hypoxia-Related Genes

Differential expression across the merged HNSCC cohort was resolved through the limma package, employing thresholds of |log2 fold change (FC)| > 0.5 and false discovery rate (FDR) < 0.05. The resulting differentially expressed genes (DEGs) were subsequently intersected with the predefined hypoxia-related gene set to identify differentially expressed hypoxia-related genes (DE-HRGs).

Functional Enrichment Analysis

Functional enrichment was orchestrated via the clusterProfiler package, with Gene Ontology (GO) interrogation spanning molecular function (MF), cellular component (CC), and biological process (BP) categories and KEGG pathway analysis conducted in parallel to delineate dysregulated signaling cascades. Complementary whole-transcriptome Gene Set Enrichment Analysis (GSEA) further resolved significantly enriched pathways, thereby extending the gene-level enrichment framework.

Calculation of the Hypoxia Score

Differentially expressed hypoxia-related genes were first identified by intersecting the differentially expressed genes between HNSCC and normal tissues with the predefined hypoxia-related gene set obtained from the Molecular Signatures Database. Univariate Cox regression analysis was subsequently performed to identify genes associated with overall survival, and the 14 prognosis-associated differentially expressed hypoxia-related genes were collectively defined as the hypoxia signature.

The enrichment activity of this 14-gene signature was quantified for each bulk transcriptomic sample using single-sample Gene Set Enrichment Analysis (ssGSEA). The resulting enrichment score was defined as the Hypoxia Score, with a higher score indicating greater relative enrichment of the prognosis-associated hypoxia signature. No additional transformation or normalization was applied to the ssGSEA score after calculation.

Evaluation of Immune Cell Infiltration and Tumor Microenvironment-Related Scores

Immune infiltration levels and gene set activity scores were quantified via single-sample gene set enrichment analysis (ssGSEA), the relative proportions of 22 tumor‑infiltrating immune cell subtypes were deconvoluted with CIBERSORT, and tumor microenvironment (TME)-associated indices were derived employing the ESTIMATE algorithm.

Unsupervised Clustering Analysis

Prognosis-associated DE‑HRGs were discriminated through univariate Cox regression at a p < 0.05 threshold, after which unsupervised consensus clustering via ConsensusClusterPlus partitioned the HNSCC cohort into two molecular subtypes defined by the expression milieu of these prognostic DE‑HRGs.

Machine Learning Analysis

Eight machine learning algorithms were employed to prioritize candidate hypoxia-associated genes, including Adaptive Boosting, Linear Discriminant Analysis, Logistic Regression, Multilayer Perceptron, K Nearest Neighbor, Quadratic Discriminant Analysis, Support Vector Machine, and Random Forest. The integrated dataset was randomly divided into a training set and an independent test set at a ratio of 8:2 with a fixed random seed of 42. Sample-level data partitioning was performed to guarantee that each individual sample was exclusively assigned to either the training cohort or the test cohort. All model construction and optimization procedures were strictly implemented based only on the training dataset. Five-fold cross-validation was conducted on the training set for model training and internal validation. Hyperparameter tuning was performed to optimize model performance, and no data or information from the independent test set was involved throughout the entire pipeline, including model training, feature selection, cross-validation, and hyperparameter optimization. Following model construction, the generalized performance of the final models was evaluated using the reserved independent test set. Machine learning models were constructed to identify key hypoxia-related genes associated with tumor hypoxia status. Feature importance analysis was performed to quantify the predictive contribution of each input gene. Candidate genes were ranked according to their feature importance scores derived from each algorithm, and the top 10 most important genes were retained for each model. Consensus hypoxia-associated candidate genes were finally determined by intersecting the significant gene sets obtained from all eight machine learning models. All machine learning analyses were implemented using Python 3.9.7, along with the packages scikit-learn 1.0.2, TensorFlow 2.8.0, and pandas 1.3.5. A unified random seed of 42 was applied to all random sampling procedures to ensure stable and reproducible analytical results.

Spatial Transcriptomic Data Analysis

The Seurat R package was harnessed for ST data processing. Spatial cellular constituents were subsequently resolved through reference‑guided deconvolution, employing FindTransferAnchors and TransferData under default parameterization to attain single‑cell resolution cell‑type assignment within the ST landscape.

Immunohistochemistry (IHC)

Immunohistochemical staining of frozen tissue sections was undertaken as follows. Following air‑drying for 5 to 10 min, sections were fixed in 4% paraformaldehyde for 10 to 30 min and then rinsed three times with phosphate‑buffered saline (PBS; pH 7.4). Endogenous peroxidase activity was quenched with 3% hydrogen peroxide for 10 min at ambient temperature. To block nonspecific binding, the sections were incubated with 3% bovine serum albumin (BSA) for 30 min before overnight exposure at 4°C to the designated primary antibodies diluted in PBS within a humidified chamber. After a washing step, horseradish peroxidase (HRP)‑conjugated species‑specific secondary antibodies were applied for 50 min at room temperature. Immunoreactivity was visualized with freshly prepared diaminobenzidine (DAB). Counterstaining was performed using Harris hematoxylin, followed by differentiation in 1% acid alcohol, bluing in ammonia water, dehydration through graded ethanol, clearing in xylene, and mounting with neutral resin. IHC analysis was performed on three independent pairs of HNSCC tumor and adjacent normal tissues, with n = 3 independent patients.

HUVEC Cell Culture

The human umbilical vein endothelial cell line (HUVEC) was cultured in Ham’s F-12K supplemented with 10% fetal bovine serum (ZETA, USA), 1% penicillin-streptomycin solution (Gibco, USA), 0.1 mg/mL heparin, and 0.03–0.05 mg/mL endothelial cell growth supplement (ECGS) at 37 °C with 5% CO₂.

For STC2 knockdown, HUVECs were transfected with the STC2 Human shRNA Plasmid Kit (Cat# TL309053, Origene, USA), which contained four distinct 29-mer shRNA lentiviral GFP constructs targeting human STC2 (Gene ID: 8614) and a scrambled non-targeting shRNA negative control (TR30021) provided by the manufacturer. Transfection was conducted following the manufacturer’s standard protocols. 72 hours post-transfection, total cellular RNA was extracted, and knockdown efficiency of STC2 was validated via Real-Time Quantitative Polymerase Chain Reaction (RT-qPCR).

Multiplex Immunofluorescence Staining

Multiplex immunofluorescence (mIF) employing tyramide signal amplification was carried out on frozen tissue sections. Following air‑drying for 5 to 10 min, sections were fixed in 4% paraformaldehyde for 10 to 30 min, rinsed with PBS, quenched with 3% hydrogen peroxide for 10 min, and blocked with 3% BSA for 30 min. Overnight incubation at 4°C with anti‑STC2 antibody (DF12324, Affinity Biosciences; 1:500 in PBS) was subsequently performed, after which an HRP‑conjugated goat anti‑rabbit secondary antibody was applied for 50 min at room temperature and coverage with 570‑TSA reagent proceeded for 10 min in the dark. Upon PBS washing, antibody elution was achieved at 37°C for 10 min and sections were reblocked with 3% BSA. A second labeling cycle utilized anti‑HIF‑1α antibody (66730‑1‑Ig, Proteintech) or anti‑CD31 antibody (YM4916, Immunoway), each diluted 1:500 in PBS and incubated overnight at 4°C, followed by HRP‑conjugated goat anti‑mouse secondary antibody and 520‑TSA reagent under identical conditions. Nuclear counterstaining was accomplished with DAPI for 8 min, the slides were coverslipped with antifade mounting medium, and images were acquired using a slide scanner. mIF analysis was performed using three independent pairs of HNSCC tumor and adjacent normal tissues, with n = 3 representing independent biological specimens.

HUVECs transfected with STC2-targeting lentiviral shRNA or scrambled negative control shRNA from the STC2 Human shRNA Plasmid Kit (Cat# TL309053, Origene) were seeded onto sterile glass coverslips. At 72 h post-transfection, cells were fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton X-100, and blocked with 3% BSA at room temperature. Cells were incubated with anti-CD31 primary antibody (YM4916, Immunoway) overnight at 4 °C. After thorough PBS washing, cells were incubated with goat anti-rabbit IgG secondary antibody conjugated to Alexa Fluor 488 (Cat# A11034, Thermo Fisher, green fluorescence for CD31). Subsequently, F-actin cytoskeleton was stained with red fluorescent phalloidin (Cat# C2207, Beyotime). Cell nuclei were counterstained with DAPI. Fluorescence images were acquired using a laser scanning confocal microscope to compare the expression and localization of CD31 and cytoskeleton between the two groups.

RT-qPCR Analysis

Total RNA was isolated from paired tumor and adjacent normal tissues using TRIzol reagent (Thermo Fisher Scientific, Inc.). First‑strand complementary DNA (cDNA) synthesis subsequently utilized the SuperScript II First‑Strand cDNA Synthesis Kit (TaKaRa, Japan) with 1 μg of total RNA serving as template. Real‑time quantitative polymerase chain reaction (RT‑qPCR) was employed to assay the mRNA abundance of target genes. All expression values were normalized to the endogenous control GAPDH, and relative quantification relied on the 2^−ΔΔCt algorithm. Table S1 catalogues the complete primer sequences applied in this study. RT qPCR analyses were performed using three independent pairs of HNSCC tumor and adjacent normal tissues from three patients, with n = 3 representing independent biological specimens.

Wound Healing Assay

HUVECs were uniformly seeded into 6-well plates. Upon reaching appropriate confluence, they were transfected with STC2-targeting shRNA lentiviral plasmids and scrambled non-targeting control shRNA plasmids from the STC2 Human shRNA Plasmid Kit (Cat# TL309053, Origene). After 72 hours, a standardized scratch was made in the cell monolayer using a sterile 100 μL pipette tip. Wound closure was then monitored and imaged with an inverted microscope at 0 h and 24 h post-scratching for comparison.

CCK-8 Cell Viability Assay

HUVECs were evenly seeded into 96-well plates. When reaching suitable confluence, cells were transfected with STC2-targeting lentiviral shRNA plasmids or scrambled negative control shRNA plasmids supplied by the STC2 Human shRNA Plasmid Kit (Cat# TL309053, Origene). 72 hours after transfection, 10 μL CCK-8 reagent was added to each well, followed by incubation at 37 °C for 1 h. The absorbance value at 450 nm was detected using a microplate reader to quantify the relative cell viability of the two groups.

CellChat

We employed the CellChat R package to explore the crosstalk pattern between cells. Following the prescribed procedure, we generated the CellChat entity using a standardized count matrix. We performed permutation tests to evaluate the statistical significance and used default on method parameters.

ScMetabolism

To quantify the activity of metabolic pathways across distinct endothelial subpopulations, the R package scMetabolism was applied to calculate metabolism scores at single‑cell resolution. Built upon gene set enrichment theory, this tool employs the AUCell (Area Under the Curve) algorithm to compute enrichment scores for predefined metabolic gene sets in individual cells. The relative activity of each metabolic pathway is quantified by measuring the area under the cumulative distribution curve of pathway genes within the cellular transcriptional profile.

scTenifoldKnk

The R package scTenifoldKnk was used to perform in silico knockout of STC2 in tumor endothelial cells. Its core algorithm compares the gene regulatory landscapes between wild‑type and computationally knocked‑out cell populations, thereby identifying differentially regulated target genes modulated by the gene of interest.

Patients and Samples

Patients undergoing surgical resection at Yongkang First People’s Hospital of Wenzhou Medical University generously provided tissue specimens, comprising HNSCC tumors and paired adjacent normal tissues. All participants furnished written informed consent prior to sample collection. The investigation was conducted in strict compliance with pertinent local, national, and international guidelines and regulations. Ethical approval for the study protocol was granted by the Human Research Ethics Committee of Jinhua Yongkang First People’s Hospital of Wenzhou Medical University (approval No. 2026-LW-011; approved on March 27, 2026).

Statistical Analysis

All statistical analyses relied on R (v4.2.1) and GraphPad Prism 9.0 (GraphPad Software, San Diego, CA, USA). For sample level single cell compositional analyses, the proportion of each cell population was calculated separately within each biological specimen. Individual specimens rather than individual cells were treated as the independent observational units for comparisons of cell type proportions. Two-group comparisons adopted Student’s t-test for normally distributed continuous variables, whereas the Mann–Whitney U-test handled non-normally distributed data. Spearman’s rank correlation coefficient evaluated associations between variables. Kaplan‑Meier methodology generated survival curves, which the log‑rank test subsequently contrasted. A two-sided p value below 0.05 was deemed indicative of statistical significance. Because the clinical validation component was exploratory and was based on the availability of eligible paired specimens during the predefined collection period, no formal prospective sample size calculation was performed. Exact biological sample sizes are now reported for each assay, together with effect estimates and confidence intervals where applicable.

Results

Identification of a Hypoxia-Related Signature Associated with Malignant Phenotype in HNSCC

A multi‑cohort integration framework employing ComBat batch correction across seven transcriptomic datasets was established to systematically characterize the hypoxia‑associated molecular landscape of HNSCC (Figure 1A). Differential expression analysis revealed extensive transcriptional reprogramming between tumor and normal tissues (Figure 1B). GO enrichment demonstrated predominant associations with aldehyde dehydrogenase activity, oxygen binding, membrane rafts, and cell adhesion complexes, together with biological processes centered on hypoxia responsiveness and oxygen‑level adaptation (Figure 1C). KEGG analysis further highlighted metabolic pathways including tyrosine metabolism, cytochrome P450‑mediated xenobiotic metabolism, and glycolysis/gluconeogenesis (Figure 1D), while GSEA confirmed elevated hypoxia‑signature activity and diminished oxidative metabolism in tumors (Figure 1E). Collectively, these findings portray HNSCC as a malignancy defined by extensive oxygen‑responsive transcriptional remodeling coupled with pronounced metabolic reprogramming.

Figure 1.

Hypoxia related signature is associated with malignant phenotypes and adverse prognosis in HNSCC. (A) Principal component analysis plots showing sample clustering before and after ComBat based batch correction across seven transcriptomic cohorts. The black arrow indicates the batch correction procedure. (B) Volcano plot showing differentially expressed genes between HNSCC and normal tissues in the integrated cohort. Red dots indicate upregulated genes, blue dots indicate downregulated genes, and gray dots indicate genes without significant differential expression. (C) Gene Ontology enrichment analysis of the differentially expressed genes. Colors indicate biological process, cellular component, and molecular function categories, while dot size represents the number of genes enriched in each term. (D) Kyoto Encyclopedia of Genes and Genomes pathway enrichment analysis of the differentially expressed genes. Bar length represents enrichment significance expressed as negative log transformed adjusted P values. (E) Gene set enrichment analysis showing representative hypoxia and metabolism associated gene sets enriched in HNSCC. Ridge colors represent P values, and normalized enrichment scores indicate the direction and magnitude of gene set enrichment. (F) Venn diagram showing the intersection between 1269 differentially expressed genes and 200 hypoxia related genes, yielding 30 hypoxia related differentially expressed genes. Blue and pink circles represent differentially expressed genes and hypoxia related genes, respectively. (G) Forest plot of univariate Cox regression analysis identifying 14 prognosis associated hypoxia related differentially expressed genes. Squares represent hazard ratios and horizontal lines indicate 95% confidence intervals. Red and blue squares indicate genes with hazard ratios above and below 1, respectively, while the vertical dashed line indicates a hazard ratio of 1. (H) Boxplots comparing the expression levels of the 14 prognostic hypoxia related genes between normal and tumor tissues. Blue and red boxes represent normal and tumor tissues, respectively. (I) Comparison of Hypoxia Scores between normal and HNSCC tissues. Blue and red boxes represent normal and tumor tissues, respectively. (J) Kaplan Meier survival curves for overall survival according to the Hypoxia Score. Red and blue curves represent the high and low Hypoxia Score groups, respectively, and shaded regions indicate 95% confidence intervals. (K) Mantel test based correlation analysis showing associations between the Hypoxia Score and malignant phenotype related signatures. Edge color represents the Mantel test P value, edge width represents the Mantel correlation coefficient, and the heatmap color scale represents Pearson correlation coefficients among the indicated signatures. (L and M) Lollipop plots showing correlations between the Hypoxia Score and immune cell infiltration estimated using CIBERSORT and single sample gene set enrichment analysis, respectively. The direction and magnitude of the correlation coefficients are indicated along the horizontal axis, while dot color represents the corresponding P value. (N) Univariate Cox regression analysis evaluating the prognostic associations of the Hypoxia Score and clinicopathological variables with overall survival. Squares represent hazard ratios, horizontal lines indicate 95% confidence intervals, and the vertical dashed line indicates a hazard ratio of 1. (O) Multivariate Cox regression analysis evaluating the independent prognostic value of the Hypoxia Score after adjustment for clinicopathological variables. Squares represent adjusted hazard ratios, horizontal lines indicate 95% confidence intervals, and the vertical dashed line indicates a hazard ratio of 1. (P) Violin plots showing the expression patterns of canonical marker genes used for major cell type annotation in the single cell transcriptomic dataset. (Q) UMAP visualization of the single cell transcriptomic landscape showing five major cell populations, including endothelial cells, epithelial cells, fibroblasts, immune cells, and other cells. Colors indicate the annotated cell populations. (R) Stacked bar plot showing the relative proportions of major cell populations in normal and tumor tissues. Colors represent the indicated cell types. (S) Comparison of Hypoxia Scores between normal and tumor tissues across endothelial cells, epithelial cells, immune cells, and fibroblasts. Blue and Orange boxes represent cells derived from normal and tumor tissues, respectively. (T and V) Spatial distribution of major cell types in HNSCC tissue. Colors indicate the annotated cell populations. (U and W) Spatial distribution of the Hypoxia Score within HNSCC tissue. The color scale represents the relative Hypoxia Score. Colors in the UMAP and spatial maps indicate the corresponding clusters. ***P < 0.001.

Intersecting 1269 differentially expressed genes with 200 hypoxia‑related genes curated from GSEA identified 30 differentially expressed hypoxia‑associated candidates (Figure 1F). Univariate Cox regression narrowed these to 14 genes with significant prognostic relevance (Figure 1G), of which 12 were adverse risk factors prominently upregulated in tumors (Figure 1H). A hypoxia score derived from these 14 genes was significantly elevated in tumor tissue (Figure 1I), and patients with higher scores experienced markedly poorer overall survival (Figure 1J). The score correlated positively with malignant programs including epithelial‑mesenchymal transition, Pan‑F‑TBRS, and angiogenesis (Figure 1K), and its elevation was accompanied by enrichment of immunosuppressive populations such as M0 macrophages and regulatory T cells, with reciprocal depletion of CD8⁺ T cells and natural killer cells (Figure 1L and M).

The hypoxia score emerged as a powerful independent prognostic indicator in both univariate (HR 20.508, 95% CI 4.294–97.941, P < 0.001) (Figure 1N) and multivariate analyses adjusted for clinical T, M, N, and overall stage (HR 17.694, 95% CI 3.597–87.039, P < 0.001) (Figure 1O), demonstrating prognostic value beyond conventional staging.

Single‑cell transcriptomic profiling of 106,699 cells (83,149 tumor; 23,530 normal) following stringent quality control and Harmony‑based batch correction (Figure S1A–H) resolved five principal cellular compartments: endothelial, epithelial, fibroblast, immune, and other cells (Figure 1P and Q). Endothelial and fibroblast populations were markedly expanded in tumors (Figure 1R). Sample‑level distributions of major populations and STC2⁺/STC2⁻ endothelial subsets are provided in Figure S6. AUCell analysis localized elevated hypoxia activity predominantly to endothelial cells and fibroblasts (Figure 1S).

Spatial transcriptomic analysis provided orthogonal validation of the heterogeneous organization of hypoxia within the HNSCC microenvironment. Quality‑controlled spatial transcriptomes underwent unsupervised clustering and spatial mapping (Figure S1I–Q), with lineage mapping revealing pronounced spatial heterogeneity (Figure 1T and V). Hypoxia‑score projection demonstrated preferential enrichment of high‑score regions within endothelial‑dominant areas (Figure 1U and W).

Construction of Hypoxia Related Clusters in HNSCC Based on Prognostic DE-HRGs

Harnessing the expression profiles of the prognostic DE‑HRGs, unsupervised consensus clustering was employed to resolve the heterogeneity of hypoxia patterns across the integrated HNSCC cohort. A two‑cluster configuration (k = 2) emerged as the optimal partition, stratifying patients into two molecularly distinct subtypes (Figures 2A and S2). Principal component analysis (PCA) corroborated a clear demarcation between these clusters (Figure 2B). Kaplan–Meier survival estimation subsequently demonstrated divergent clinical trajectories, with patients in Cluster 2 experiencing markedly abbreviated overall survival relative to those in Cluster 1 (Figure 2C). Consistent with a more pronounced hypoxia phenotype, Cluster 2 was characterized by pervasive upregulation of canonical hypoxia markers (STC2, SERPINE1, TGFBI, and STC1) and concurrent downregulation of SELENBP1 and RORA (Figure 2D).

Figure 2.

Identification of two hypoxia-related clusters in HNSCC. (A) Consensus clustering heatmap showing the classification of HNSCC samples into C1 and C2. Colors indicate consensus values from 0 to 1. (B) PCA showing the separation between C1 and C2. Blue and red indicate C1 and C2, respectively, and dashed ellipses indicate the distribution of each cluster. (C) Kaplan Meier analysis of overall survival in C1 and C2. Blue and red curves indicate C1 and C2, respectively, and dashed lines indicate median survival. (D) Heatmap showing the expression of prognostic DE HRGs in C1 and C2. Blue and red indicate relatively low and high expression, respectively. (E and F) ESTIMATE and stromal scores in C1 and C2. (G and H) CIBERSORT and ssGSEA analyses of immune cell infiltration in C1 and C2. (I) Expression of immune checkpoint genes in C1 and C2. (J) Comparison of malignant biological signatures between C1 and C2. (K) GO and KEGG enrichment analyses of differentially expressed genes between C1 and C2. Bar length represents enrichment significance, and colors indicate ontology categories. (L) GSEA of representative pathways differentially enriched between C1 and C2. Curve color corresponds to the adjusted P value shown by the color scale, and NES denotes normalized enrichment score. Asterisks indicate statistical significance where shown, with *P < 0.05, **P < 0.01, and ***P < 0.001. ns indicates no statistical significance.

Comparison of TME composition disclosed significantly elevated ESTIMATE and stromal scores in Cluster 2 (Figure 2E and F), reflecting a more copious stromal compartment. CIBERSORT deconvolution concurrently uncovered profoundly augmented infiltration of immunosuppressive subsets, notably M0 and M2 macrophages, alongside a marked reduction in antitumor effector cells such as CD8⁺ T cells (Figure 2G). ssGSEA substantiated these findings by confirming a prevailing immunosuppressive infiltrate in Cluster 2, with prominent representation of regulatory T cells and macrophages (Figure 2H). Reflecting this immune topography, the expression landscapes of pivotal functional genes were delineated across the two clusters. As portrayed in Figure S3, genes governing cytotoxic T cell activity (A), chemokine receptors (B), stromal activation (C), antigen presentation (D), and chemokines (E) exhibited divergent expression patterns, further underscoring the more immunosuppressive and pro‑tumorigenic phenotype in Cluster 2. Concordantly, the majority of immune checkpoint molecules displayed extensive upregulation within Cluster 2 (Figure 2I). Cluster 2 additionally accrued significantly elevated malignant phenotype scores, particularly for matrix activation and premetastatic attributes encompassing angiogenesis, EMT, WNT targets, and Pan‑F‑TBRs pathways (Figure 2J).

To further characterize the functional divergence between the two subtypes, enrichment analysis was performed on DEGs differentiating the clusters. GO and KEGG analyses demonstrated that these DEGs were prominently associated with extracellular matrix (ECM)-related processes, including ECM organization and ECM-receptor interaction (Figure 2K). Concordantly, GSEA confirmed the upregulation of TGF-β signaling, focal adhesion, and ECM-receptor interaction within Cluster 2 (Figure 2L). Of marked significance, GSEA concurrently revealed substantial suppression of oxidative metabolic programs in Cluster 2, particularly oxidative phosphorylation and fatty acid metabolism (Figure 2L). Collectively, the hypoxia‑high Cluster 2 constitutes an aggressive HNSCC subtype defined by poor prognosis, heightened immunosuppression, and vigorous stromal remodeling. These observations align closely with the malignant attributes previously linked to the hypoxia score, thereby corroborating the functional significance of the hypoxia‑related signature in HNSCC.

Identification of STC2 as a Critical Hypoxia-Driven Regulator in HNSCC

A systematic evaluation harnessing eight machine learning classifiers (KNN, RF, SVM, AdaBoost, LDA, LR, MLP, and QDA) was undertaken. Intersection of the top 10 genes prioritized by each algorithm converged on three consensus candidates: STC1, STC2, and RORA (Figures 3A and S4). Among the tested classifiers, KNN and RF each assigned the highest feature importance to STC2 (Figure 3A). RT‑qPCR validation subsequently confirmed that STC2 displayed a substantially greater magnitude of upregulation in HNSCC tissues compared with STC1 and RORA (Figure 3B–D). These convergent lines of evidence establish STC2 as the preeminent hypoxia‑driven regulator in HNSCC.

Figure 3.

Identification and validation of STC2 as a key hypoxia-driven regulator in HNSCC. (A) Intersection of candidate genes identified by eight machine learning algorithms, showing STC1, STC2, and RORA as shared genes, with representative feature importance rankings from KNN and random forest models. (B–D) RT qPCR validation of RORA, STC1, and STC2 expression in normal and tumor tissues. (E) Validation of STC2 overexpression in the E MTAB 8588, GSE42743, GSE75538, and TCGA HNSC cohorts. (F) Paired analysis of STC2 mRNA expression between normal and tumor tissues. Lines connect paired samples. (G) Association of STC2 expression with histological grade, clinical stage, and T stage across independent cohorts. (H) Kaplan Meier analysis of overall survival according to STC2 expression. Blue and red curves indicate STC2 low and STC2 high groups, respectively. (I) Representative Human Protein Atlas immunohistochemical images of STC2 expression in normal and tumor tissues. (J) Representative immunohistochemical staining and quantification of STC2 positive area in HNSCC and adjacent normal tissues. *P<0.05, **P<0.01, ***P<0.001. Scale bars, 200 μm.

Experimental and cross cohort validation of STC2 was achieved across multiple independent cohorts. Of particular relevance, STC2 exhibited pronounced upregulation in radiotherapy‑treated patients within TCGA‑HNSC, contrasting with its marked downregulation in HPV‑positive specimens from GSE117973 (Figure S5A and B). Across the E‑MTAB‑8588, GSE42743, GSE75538, and TCGA‑HNSC cohorts, tumor tissues consistently demonstrated elevated STC2 expression relative to adjacent normal counterparts (Figure 3E). Paired‑sample interrogation of the CPTAC proteomic repository further attested to significantly increased STC2 abundance in tumor lesions (Figure 3F). Consistent with these findings, STC2 overexpression was intimately associated with adverse histopathological features; across multiple datasets, heightened STC2 levels correlated with higher histological grade, advanced clinical stage, and elevated T stage (Figure 3G). Kaplan‑Meier analyses subsequently linked high STC2 to unfavorable survival trajectories, encompassing diminished overall survival (OS), disease‑specific survival (DSS), and progress‑free survival (PFS) (Figures 3H and S5C–G). At the proteomic level, both the HPA database and our IHC staining verified enhanced STC2 immunoreactivity in HNSCC tumors, manifesting as larger positive areas and stronger signal intensity (Figure 3I and J). Collectively, these observations cement STC2 as a pivotal hypoxia‑driven regulator intimately coupled with aggressive clinicopathological characteristics and poor prognosis in HNSCC.

Biological Functions and TME Correlations of STC2 in HNSCC

Functional dissection of the STC2‑associated transcriptome by Pearson correlation resolved two opposing biological domains: genes positively correlated with STC2 were enriched for developmental processes, cell adhesion, and ECM organization, whereas negatively correlated genes predominantly mapped to immune‑related functions (Figure 4A). KEGG analysis reinforced this dichotomy, with positive correlations involving cell cycle, focal adhesion, and glycolysis, and negative correlations concentrated in immune pathways (Figure 4B). GSEA further demonstrated that elevated STC2 expression was accompanied by heightened activity of epithelial‑mesenchymal transition, hypoxia, glycolysis, and angiogenesis (Figure 4C). Mantel correlation analysis confirmed that both STC2 and the Hypoxia Score were significantly associated with a comprehensive panel of tumor‑relevant pathways, including glycolysis, IL6/JAK/STAT3, mTORC1, Notch, PI3K/AKT/mTOR, TGFβ, TNFα/NF‑κB, and WNT/β‑catenin signaling (Figure 4D). These convergent functional associations establish STC2 as a molecular nexus linking tumor hypoxia with malignant pathway activation in HNSCC.

Figure 4.

Functional characteristics and tumor microenvironment relevance of STC2 in HNSCC. (A) GO enrichment analysis of genes positively and negatively correlated with STC2. Colors distinguish positively and negatively correlated gene sets. (B) KEGG enrichment analysis of STC2 related genes. Colors distinguish positively and negatively correlated pathways. (C) GSEA showing enrichment of malignant pathways associated with STC2 expression. Ridge colors indicate adjusted P values. (D) Correlation analysis of STC2, the Hypoxia Score, and tumor related pathways. Edge color and width indicate Mantel P values and correlation coefficients, respectively, while tile color represents Pearson correlation coefficients among pathways. (E) UMAP plots showing major cell populations in normal and tumor tissues. Colors indicate annotated cell types, and the colored header bars distinguish normal and tumor samples. (F) UMAP feature plots showing STC2 expression across cells in normal and tumor tissues. Color intensity indicates STC2 expression. (G) Dot plot showing STC2 expression across major cell types. Dot size represents the percentage of STC2 expressing cells, and color intensity indicates expression level. (H and I) Correlation between STC2 expression and endothelial cell infiltration in GSE65858 and TCGA HNSC, respectively. Blue dots represent individual samples, yellow lines indicate fitted regression lines, and gray shaded areas represent 95% confidence intervals. (J) Spatial transcriptomic maps showing the distributions of endothelial cells, PECAM1, and STC2. Color scales indicate relative signal intensity. (K) Representative MIF images and quantitative analysis of CD31 and STC2 in normal and tumor tissues. CD31, STC2, and nuclei are shown in green, red, and blue, respectively. *P < 0.05 and **P < 0.01. Scale bars, 200 μm.

Single‑cell RNA‑seq profiling revealed pronounced cellular heterogeneity of STC2 expression, with endothelial cells exhibiting the highest transcript levels among all major populations (Figure 4E–G). Endothelial enrichment was substantiated by positive correlations between STC2 expression and endothelial abundance in the GSE65858 and TCGA‑HNSC cohorts (Figure 4H and I), by spatial co‑localization of endothelial regions, PECAM1, and STC2 (Figure 4J), and by elevated STC2 and CD31 signals with prominent spatial overlap in clinical HNSCC specimens (Figure 4K). Collectively, these findings position STC2 as a hypoxia‑coupled molecule that is preferentially enriched within the endothelial compartment and functionally integrated into the malignant circuitry of HNSCC.

Identification of STC2+ Endothelial Subpopulation with Hypoxia and Angiogenesis Activation in HNSCC

Endothelial cells were stratified by STC2 expression into STC2+ (upper 25%) and STC2- (lower 25%) subsets. UMAP visualization and compositional profiling demonstrated a pronounced enrichment of the STC2+ fraction in HNSCC tumor tissues (Figure 5A and B). This subset exhibited significantly elevated hypoxia scores (Figure 5C), and functional enrichment of its defining gene set converged on hypoxia‑responsive GO terms, PI3K‑AKT signaling, and ECM‑receptor interaction pathways (Figure 5D). scMetabolism analysis further revealed heightened fatty acid biosynthesis and unsaturated fatty acid metabolism in STC2+ endothelial cells (Figure 5E). Multiplex immunofluorescence of clinical specimens corroborated these findings, showing markedly increased STC2 and HIF1A signals with substantial spatial colocalization in tumor tissues (Figure 5F).

Figure 5.

Identification of an STC2+ endothelial subpopulation with hypoxia and angiogenesis activation in HNSCC. (A) UMAP visualization of STC2 low and STC2 high endothelial cells in normal and tumor tissues. The lower and upper quartiles of STC2 expression were defined as STC2- and STC2+ endothelial cells, respectively. (B) Relative proportions of STC2- and STC2+ endothelial cells in normal and tumor tissues. (C) Comparison of hypoxia scores between STC2- and STC2+ endothelial cells. (D) GO and KEGG enrichment analyses of highly expressed genes in STC2+ endothelial cells. (E) scMetabolism analysis of metabolic pathway activity in STC2- and STC2+ endothelial cells. Circle size represents the relative enrichment level, and color indicates the pathway activity score. (F) Multiplex immunofluorescence staining of HIF1A, STC2, and DAPI in normal and tumor tissues with fluorescence quantification. Scale bars are indicated in the images. (G and H) CellChat analysis showing the number and strength of interactions between STC2- and STC2+ endothelial cells. Red and blue lines indicate interactions originating from STC2+ and STC2- endothelial cells, respectively. (I) Bubble plot showing ligand receptor interactions between endothelial subpopulations. Circle size indicates statistical significance, and color represents communication probability. (J) ANGPT signaling analysis showing communication patterns, pathway interaction strength, and expression of ANGPT2, TEK, ITGA5, and ITGB1. Red and blue denote STC2+ and STC2- endothelial cells, respectively. (K) Top 20 genes with the largest log fold changes following in silico STC2 knockout in tumor endothelial cells. (L) RT qPCR validation of STC2 knockdown efficiency in HUVECs. (M) CCK8 assay evaluating HUVEC viability following STC2 knockdown. (N) Immunofluorescence staining of CD31, cytoskeleton, and DAPI in control and STC2 knockdown HUVECs. Scale bars are indicated in the images. (O) RT qPCR analysis of VEGF, CD31, vWF, and eNOS expression following STC2 knockdown. (P) Scratch wound assay evaluating HUVEC migration at 0 and 24 h following STC2 knockdown. CD31, STC2, and nuclei are shown in green, red, and blue, respectively. **P < 0.01 and ***P < 0.001. Scale bars, 200 μm.

CellChat analysis uncovered extensive and differential intercellular communication between STC2+ and STC2- endothelial subsets (Figure 5G–I), with the ANGPT signaling axis prominently activated in STC2+ cells, accompanied by abundant expression of ANGPT2 and its receptors TEK, ITGA5, and ITGB1 (Figure 5J). In silico knockout of STC2 via scTenifoldKnk elicited pronounced transcriptional rewiring; the twenty most responsive genes were STC2, XIST, ACKR1, CLU, ESM1, TSPAN7, IL33, MIR4435‑2HG, HLA‑DQA1, LIFR, IL1R1, CA2, UNC5B, HLA‑DQB1, IGFBP3, SELP, KCNE3, ZNF385D, HAPLN1, and ADIRF (Figure 5K). Experimental silencing of STC2 in HUVECs, confirmed by RT-qPCR (Figure 5L), significantly reduced cell viability (Figure 5M), diminished CD31 expression and disrupted cytoskeletal architecture (Figure 5N), downregulated VEGF, CD31, vWF, and eNOS transcripts (Figure 5O), and impaired migratory competence (Figure 5P). Taken together, these data establish STC2+ endothelial cells as a hypoxia-primed, metabolically active population and demonstrate that STC2 is functionally integral to endothelial viability, migration, and angiogenic potential in HNSCC.

Hypoxia-Driven Remodeling of the Intercellular Communication Networks in HNSCC

Heterogeneous correlations between STC2 expression and multiple immune subsets pointed to a significant role for STC2 in the immune microenvironment of HNSCC (Figure 6A and B). Eight major immune populations, comprising B cells, dendritic cells, macrophages, mast cells, monocytes, neutrophils, plasma cells, and T cells, were annotated via canonical lineage markers (Figure 6C) and separated into transcriptionally distinct clusters by UMAP (Figure 6D). A marked redistribution of immune cells was evident in tumors compared with normal tissues (Figure 6E), reflected in a contracted T‑cell compartment and the concomitant expansion of several other immune lineages (Figure 6F).

Figure 6.

Hypoxia-related remodeling of intercellular communication networks in HNSCC. (A and B) Correlation heatmaps showing the associations between STC2 expression and different immune cell subsets. Color intensity represents the direction and magnitude of the correlation, and asterisks indicate statistical significance. (C) Violin plots showing canonical marker genes used for cell type annotation. Colors indicate the annotated cell populations. (D) UMAP visualization of the major cell populations identified in HNSCC. Colors indicate different cell types. (E) UMAP visualization comparing the distribution of cell populations between normal and tumor tissues. Colors indicate different cell types. (F) Relative proportions of cell populations in normal and tumor tissues. (G) CellChat analysis showing the number and strength of intercellular interactions among STC2− endothelial cells, STC2+ endothelial cells, and immune cell populations in normal tissues. Node size represents the relative cell population size, and line width represents interaction number or communication strength. Colors indicate the corresponding cell populations. (H) Bubble plot showing ligand receptor interactions among STC2− endothelial cells, STC2+ endothelial cells, and immune cell populations in normal tissues. Dot size indicates statistical significance, and color represents communication probability. (I and J) GALECTIN and ANGPTL signaling networks in normal tissues, respectively. Heatmaps, violin plots, and chord diagrams illustrate signaling activity, expression patterns of signaling related genes, and intercellular communication patterns. (K) CellChat analysis showing the number and strength of intercellular interactions among STC2− endothelial cells, STC2+ endothelial cells, and immune cell populations in tumor tissues. Node size represents the relative cell population size, and line width represents interaction number or communication strength. Colors indicate the corresponding cell populations. (L) Bubble plot showing ligand receptor interactions among STC2− endothelial cells, STC2+ endothelial cells, and immune cell populations in tumor tissues. Dot size indicates statistical significance, and color represents communication probability. (M and N) GALECTIN and ANGPTL signaling networks in tumor tissues, respectively. Heatmaps, violin plots, and chord diagrams illustrate signaling activity, expression patterns of signaling related genes, and intercellular communication patterns. (O) Spatial CellChat analysis showing the number and strength of intercellular interactions among epithelial cells, T cells, monocytes, fibroblasts, plasma cells, and STC2+ endothelial cells in tumor tissues. Node size represents the relative abundance of each cell population, and line width represents interaction number or communication strength. Colors indicate the corresponding cell populations. (P) Spatial characterization of the GALECTIN signaling network in tumor tissues. Spatial maps, heatmaps, violin plots, and chord diagrams illustrate the spatial distribution of signaling activity, communication probabilities, expression patterns of signaling related genes, and major ligand receptor interactions, including LGALS9 with CD44 and CD45. (Q) Spatial characterization of the ANGPTL signaling network in tumor tissues. Spatial maps, heatmaps, violin plots, and chord diagrams illustrate the spatial distribution of signaling activity, communication probabilities, expression patterns of signaling related genes, and major ligand receptor interactions, including ANGPTL2 with the ITGA5 and ITGB1 receptor complex.

CellChat analysis was employed to decipher the intercellular communication between STC2+ and STC2- endothelial subsets and immune cells. Normal tissues were characterized by sparse endothelial–immune connectivity and modest communication strength (Figure 6G), with ligand–receptor mapping pinpointing the dominant signaling interactions of both endothelial subsets (Figure 6H). The GALECTIN and ANGPTL pathways demonstrated segregated communication profiles across endothelial and immune populations (Figure 6I and J). Under tumor conditions, the number and intensity of interactions involving the two endothelial subsets increased dramatically (Figure 6K), accompanied by an extensive reconfiguration of the ligand–receptor network (Figure 6L). Notably, both the GALECTIN and ANGPTL signaling axes underwent substantial tumor‑driven rewiring (Figure 6M and N).

Spatial CellChat analysis reinforced these findings by uncovering dense communication among epithelial cells, T cells, monocytes, fibroblasts, plasma cells, and STC2+ endothelial cells within tumor regions (Figure 6O). The GALECTIN network was spatially organized, with LGALS9 signaling through CD44 and CD45 distributed across multiple cellular compartments (Figure 6P), while ANGPTL signaling, including ANGPTL2 engagement of the ITGA5/ITGB1 integrin complex, exhibited a comparable spatial pattern (Figure 6Q). Collectively, these results indicate that STC2‑associated endothelial heterogeneity is accompanied by a global reprogramming of endothelial–immune crosstalk, supported by spatially resolved signaling circuits within the HNSCC microenvironment.

Exploratory Assessment of the Therapeutic Relevance of STC2 in HNSCC

In the analyzed external anti PD L1 cohort, elevated STC2 expression was associated with poorer treatment response and inferior post treatment survival outcomes (Figure 7A and B). Analysis of external anti PD 1 cohorts further showed an association between STC2 expression and therapeutic response, supporting its exploratory value for treatment response stratification (Figure 7C and D). Subsequent interrogation centered on the relationship between STC2 expression and drug sensitivity. Within the CTRP repository, PLX‑4032 and Procarbazine emerged as candidate compounds exhibiting preferential sensitivity in STC2‑high samples, while Gemcitabine and Midostaurin demonstrated analogous sensitivity profiles in the PRISM dataset (Figure 7E and F). Molecular docking and molecular dynamics simulations suggested potential structural compatibility between these compounds and STC2 (Figure 7G). Computational modeling via Maestro and PyMOL further delineated their distinctive binding configurations (Figure 7H–K). Dynamic simulation trajectories were further used to characterize the predicted stability of the modeled complexes. Indices of this stabilization included convergence of RMSD values beyond a 100 ns equilibration phase, a consistent radius of gyration reflecting structural compactness, stable solvent‑accessible surface area (SASA) profiles preserving protein topology, low RMSF values indicative of restricted residue fluctuations, and persistent hydrogen‑bonding networks fundamental to complex integrity (Figure 7L–P).

Figure 7.

STC2 may serve as a biomarker of immunotherapy response and a potential therapeutic target in HNSCC. (A) Comparison of STC2 expression between responders and nonresponders in the Wolf anti PD L1 cohort. Blue and red indicate nonresponders and responders, respectively. (B) Kaplan Meier analysis of overall survival according to STC2 expression in the IMvigor210 cohort. Blue and red curves indicate STC2 low and STC2 high groups, respectively. (C and D) ROC curves showing the predictive performance of STC2 for immunotherapy response in the Homet and Gao cohorts. The diagonal gray line represents random classification. (E and F) Drug sensitivity analyses based on CTRP and PRISM datasets. Colors represent correlations between STC2 expression and drug sensitivity according to the color scales. Red text highlights the selected candidate compounds. (G) Binding energies and MM PBSA binding free energies of four candidate compounds with STC2. Colors distinguish the four STC2 drug complexes. (H–K) Two dimensional interaction maps and three dimensional docking conformations of STC2 with Gemcitabine, PLX 4032, Midostaurin, and Procarbazine. Dashed boxes indicate enlarged views of the predicted binding sites. (L–P) Molecular dynamics analyses of the four STC2 drug complexes, including SASA, RMSD, RMSF, radius of gyration, and hydrogen bond number. Line colors distinguish the four complexes as indicated in each panel.

Multi-Dimensional Pan-Cancer Landscape of STC2

Pan‑cancer characterization of STC2 was undertaken to evaluate the broader translational relevance of our observations. Integrative analysis of TCGA and GTEx cohorts unveiled pervasive STC2 overexpression across the preponderance of solid malignancies, encompassing BLCA, LIHC, CESC, and STAD (Figure 8A), a pattern further corroborated by paired tumor‑normal comparisons spanning diverse cancer types (Figure 8B). Aligned with these transcriptomic profiles, IHC data from the HPA database substantiated robust STC2 protein upregulation in multiple tumor entities (Figure 8C). Motivated by this ubiquitous expression signature, the TME association landscape was subsequently interrogated. Across most cancer types, STC2 transcript levels exhibited a pronounced positive correlation with stromal score while manifesting an inverse association with immune score (Figure 8D). Concordantly, correlation analyses disclosed a positive linkage between STC2 expression and the infiltration of macrophages, M0 macrophages, and Th2 cells (Figure 8E and F). Pan‑cancer univariate Cox regression demonstrated that elevated STC2 expression portended inferior OS, DSS, and PFI (Figure 8G). Notably, high STC2 expression was associated with shortened DSS in BLCA, CESC, ESCA, HNSC, KIRP, LIHC, LUAD, MESO, PRAD, and THYM; with worse OS in BLCA, ESCA, HNSC, KIRP, LIHC, LUAD, MESO, SARC, and THYM; and with poorer PFI in ACC, BLCA, CESC, COAD, HNSC, KICH, KIRP, LIHC, LUAD, MESO, PRAD, and THYM.

Figure 8.

Pan-cancer landscape of STC2 expression, immune associations, and prognostic value. (A) Differential STC2 mRNA expression between normal and tumor tissues across cancer types. Blue and red indicate normal and tumor tissues, respectively. (B) Paired comparison of STC2 expression between normal and tumor tissues. Blue and red indicate normal and tumor tissues, respectively, and lines connect paired samples. (C) Representative Human Protein Atlas immunohistochemical images showing STC2 protein expression in normal and tumor tissues across cancer types. Black boxes indicate regions shown at higher magnification, with dashed lines connecting the corresponding regions. Black scale bars indicate the image scale. (D) Correlations of STC2 expression with stromal score, immune score, and ESTIMATE score across cancer types. (E and F) Correlations between STC2 expression and immune cell infiltration across cancer types. Red and blue indicate positive and negative correlations, respectively, and color intensity represents correlation strength. (G) Forest plots showing the prognostic associations of STC2 with disease specific survival, overall survival, and progression free interval. Points represent hazard ratios and horizontal lines represent 95% confidence intervals. Scale bars, 200 μm and 50 μm.

Discussion

Hypoxia constitutes a quintessential hallmark of solid tumors, including HNSCC, wherein it propels aggressive phenotypes, metabolic reprogramming, and immune evasion.²⁷ Nevertheless, the comprehensive landscape of hypoxia‑driven molecular perturbations and their functional ramifications within the HNSCC TME remains to be exhaustively delineated.²⁸ To surmount this knowledge gap, an integrative multi‑omics analytical framework was devised for HNSCC, aimed at deconvoluting the regulatory influence of hypoxia in the TME. Simultaneously, the systematic amalgamation of machine learning algorithms, multi‑center cohort data, and pharmacological repositories permitted the nomination of cardinal hypoxia‑driven effectors and the subsequent formulation of targeted therapeutic strategies.^29–31

A converging body of evidence has cemented hypoxia as a principal driver of aggressive tumor behavior and unfavorable prognosis in HNSCC, although the majority of extant hypoxia‑related signatures are constructed exclusively from bulk RNA profiles.^32,33 To bridge this gap, a hypoxia score was derived from our integrated multi‑omics framework. This score exhibited uniform activation across bulk RNA, scRNA‑seq, and ST datasets and was intimately associated with abbreviated overall survival, thereby reinforcing the prognostic significance of hypoxia documented in prior investigations.³⁴ Consistent with these findings, the hypoxia score displayed robust positive correlations with stromal activation, angiogenesis, and immunosuppression, and downstream unsupervised clustering further authenticated hypoxia‑associated malignant phenotypes, collectively delivering a more granular depiction of the hypoxia‑shaped TME. Mechanistic dissection of hypoxia‑driven malignancy was subsequently achieved through convergent machine learning pipelines, which pinpointed STC2 as the paramount hypoxia‑regulated gene.^35,36 STC2 manifested salient aggressive characteristics in HNSCC and received independent histological confirmation.

STC2 (stanniocalcin 2), a secreted glycoprotein of the highly conserved stanniocalcin hormone family, is robustly upregulated under hypoxic conditions.³⁷ A compelling body of evidence has positioned STC2 as a pro‑tumorigenic mediator across a spectrum of human malignancies.³⁸ Qie et al demonstrated that STC2 orchestrates cancer cell adaptation to nutrient deprivation through attenuation of oxidative stress.³⁹ Hu et al further established that STC2 propels colorectal cancer progression by enhancing cellular resistance to anoikis.⁴⁰ Our pan‑cancer interrogation additionally disclosed that STC2 overexpression is intimately associated with advanced tumor stage and unfavorable prognosis. Functionally, STC2 expression manifested a positive correlation with stromal score and a reciprocal negative correlation with immune score, thereby underscoring a tight linkage with the TME. Elevated STC2 levels were accompanied by heightened infiltration of pro‑tumorigenic immune cells, most notably macrophages and Th2 cells, implicating a contributory role in sculpting an immunosuppressive microenvironment. Collectively, these observations position STC2 as a molecule closely aligned with clinically relevant molecular and immune features in HNSCC, warranting its further evaluation as a candidate biomarker. Within a clinical landscape marked by substantial diagnostic complexity and pathological heterogeneity, particularly among uncommon histological subtypes of head and neck malignancies, accurate disease classification remains contingent upon expert pathological assessment. Molecular indicators such as STC2 may therefore furnish complementary biological insights, provided their application is embedded within established pathological and clinicopathological frameworks rather than interpreted as standalone diagnostic determinants.⁴¹

Harnessing single‑cell transcriptomic data, a high‑resolution functional dissection of STC2 was conducted within the HNSCC microenvironment. Departing from the prevailing narrative that predominantly attributes STC2 activity to malignant epithelial cells, the present analysis uncovered a pronounced endothelial localization, an observation robustly corroborated by independent transcriptomic and spatial transcriptomic evidence.²³ This investigation delineated an STC2‑enriched endothelial subset distinguished by elevated hypoxia scores, heightened fatty acid biosynthetic activity, and enhanced ANGPT signaling inferred through CellChat, with ANGPT2 emerging as a prominent ligand and multiplex immunofluorescence confirming the spatial colocalization of STC2 with hypoxic markers. In silico knockout of STC2 via scTenifoldKnk provoked marked transcriptional rewiring; among the most responsive genes, ESM1, a hypoxia‑inducible endothelial factor governing angiogenesis and fatty acid synthesis,^42–44 and ACKR1, an atypical chemokine receptor directing leukocyte trafficking,⁴⁵ exhibited prominent dysregulation. These convergent computational findings position endothelial STC2 at the nexus of hypoxia adaptation, angiogenic programming, lipid metabolic remodeling, and immune‑endothelial crosstalk in HNSCC.

Single‑cell cartography of the immune compartment revealed extensive TME restructuring, characterized by a contracted T‑cell pool and reciprocal expansion of alternative immune lineages, hallmarks of the immunosuppressive microenvironment in this malignancy.^2,46 CellChat analysis captured heightened interaction density and intensity within tumors, accompanied by pronounced reconfiguration of GALECTIN and ANGPTL signaling networks. Given the elevated hypoxic activity and metabolic reprogramming of STC2+ endothelial cells, together with the recognized status of STC2 as a canonical HIF target, these observations implicate hypoxia‑adapted endothelial cells in shaping the altered immune communication landscape.⁴⁷ GALECTIN signaling governs T‑cell apoptosis, exhaustion, and macrophage polarization, processes centrally involved in tumor immune escape,^48,49 and pathological evidence from gingival squamous cell carcinoma has linked Galectin‑1 expression to tumor immunity and adverse prognosis.⁵⁰ Recent spatially resolved investigations have delineated heterogeneous immune niches, with distinct stromal states associated with differential immunotherapy responses and immune checkpoint microenvironments.⁵¹ These findings provide mechanistic context for the CellChat results and suggest that altered GALECTIN signaling involving STC2+ endothelial cells may contribute to spatially compartmentalized immune suppression within the TME. Concurrent ANGPTL network remodeling may further couple endothelial dynamics to immune regulation through angiogenic signaling and recruitment of immunosuppressive myeloid populations,⁵² with certain ANGPTL members capable of engaging inhibitory receptors such as LILRB2.⁵³ STC2+ endothelial cells therefore appear positioned at a signaling interface linking hypoxia adaptation, metabolic reprogramming, angiogenesis, and immune modulation, although the computationally inferred communication axes await direct experimental validation.^2,54

Exploratory analyses associated STC2 expression with immunotherapy response, though its predictive utility in HNSCC remains to be established.⁵⁵ Elevated STC2 levels correlated with attenuated therapeutic benefit and inferior post‑treatment outcomes in external immunotherapy cohorts, providing preliminary evidence for a relationship with ICI efficacy. Drug sensitivity profiling nominated PLX‑4032, Procarbazine, Gemcitabine, and Midostaurin as candidate compounds exhibiting preferential sensitivity in STC2‑high tumors, with molecular docking and molecular dynamics simulations predicting favorable binding modes and complex stability. PLX‑4032, an FDA‑approved BRAF inhibitor indicated for BRAF‑mutant melanoma,⁵⁶ and Midostaurin, a multikinase inhibitor approved for FLT3‑mutant acute myeloid leukemia,⁵⁷ displayed relatively favorable binding energetics. Critically, none of these candidate compounds has undergone experimental validation in STC2‑high HNSCC models, and their computationally predicted interactions with STC2 do not constitute evidence of direct target engagement or pharmacological efficacy. These compounds should therefore be regarded as hypothesis‑generating leads, pending rigorous biochemical and preclinical substantiation in STC2‑defined HNSCC systems.

Several limitations should be acknowledged. Although STC2 silencing confirmed its functional contribution to endothelial viability, migration, and angiogenic competence, its involvement in lipid metabolic reprogramming and immune modulation remains experimentally unresolved. CellChat analysis infers intercellular communication from ligand–receptor co‑expression and therefore cannot establish direct functional interactions between STC2+ endothelial cells and immune populations. The immunotherapy associations were derived from external retrospective cohorts and require prospective validation in HNSCC‑specific patient populations. Similarly, the candidate compounds identified through drug sensitivity screening, molecular docking, and molecular dynamics simulations demand biochemical and preclinical evaluation before any therapeutic relevance to STC2 can be substantiated.

Conclusion

This integrative multi‑omics study positioned STC2 as a hypoxia‑coupled factor preferentially enriched in a specialized endothelial subset within HNSCC. STC2+ endothelial cells exhibited pronounced hypoxic activity alongside lipid metabolic, angiogenic, and immune‑communication phenotypes. Functional experiments corroborated an association linking STC2 to endothelial viability, migration, and angiogenesis‑related processes. Clinically, elevated STC2 expression was associated with adverse prognostic features, and exploratory interrogation of external immunotherapy cohorts raised the possibility of a relationship with therapeutic responsiveness. Computational screening further prioritized several candidate compounds connected to STC2, although these in silico leads mandate rigorous experimental validation in HNSCC‑specific systems. In sum, the present evidence endorses STC2 as a candidate biomarker reflecting hypoxia‑associated endothelial biology and provides provisional hypotheses for future therapeutic exploration, rather than establishing a clinically validated predictive biomarker or a definitive molecular target.

Funding Statement

This work was supported by the Zhejiang Province Traditional Chinese Medicine Science and Technology Plan Project (2026ZL0078), National Natural Science Foundation of China (82501180), Medical and Health Science Program of Zhejiang Province (2025HY0609, 2025KY942), Zhejiang Provincial Natural Science Foundation of China (LQN25H140004).

Data Sharing Statement

All raw and processed data generated in this study can be obtained from the corresponding author, Mouyuan Sun (sunmouyuan777@zju.edu.cn), upon reasonable written request.

Ethics Approval and Consent to Participate

Ethical clearance for this single-center study was issued by the Human Research Ethics Committee of Jinhua Yongkang First People’s Hospital of Wenzhou Medical University (approval No. 2026-LW-011; date: 2026-03-27). Every procedure related to this research strictly followed the ethical tenets set forth in the Declaration of Helsinki. Written informed consent was acquired from all eligible patients before their inclusion in the study.

Author Contributions

All authors made a significant contribution to the work reported, whether that is in the conception, study design, execution, acquisition of data, analysis and interpretation, or in all these areas; took part in drafting, revising or critically reviewing the article; gave final approval of the version to be published; have agreed on the journal to which the article has been submitted; and agree to be accountable for all aspects of the work.

Disclosure

The authors declare that they have no competing interests.