Network-based integrative transcriptomic analysis reveals immune-regulatory and oncogenic signatures of EOMES and SPRYD3 in head and neck squamous cell carcinoma

Abstract

Background

Head and neck squamous cell carcinoma (HNSCC) exhibits marked heterogeneity driven by both tumor-intrinsic programs and immune context. However, the relationship between epithelial proliferation and immune activity remains incompletely defined.

Methods

We combined bulk transcriptomic analyses of HNSCC cohorts (GSE83519 and TCGA-HNSC) with single-cell RNA sequencing data derived from OSCC (GSE172577) to identify gene programs associated with tumor progression. Co-expression network analysis was used to identify tumor-related modules, and candidate genes were evaluated by survival analysis and experimental validation. Single-cell analysis of an OSCC cohort and CopyKAT-based epithelial cell identification were performed to define cell-type-specific expression patterns.

Results

EOMES and SPRYD3 were identified as candidate prognostic markers associated with distinct biological programs. Single-cell analysis showed that SPRYD3 expression was enriched in aneuploid epithelial cells, whereas EOMES expression was restricted to cytotoxic lymphocyte populations. Tumor epithelial cells with high SPRYD3 expression exhibited activation of cell cycle and DNA replication pathways, while immune-related signaling was reduced. In contrast, EOMES expression was associated with interferon response and T-cell effector programs. Experimental validation supported increased SPRYD3 expression, whereas EOMES exhibited a tendency toward lower expression in OSCC tissues.

Conclusions

SPRYD3 and EOMES represent distinct transcriptional programs associated with epithelial proliferation and immune activity in HNSCC. These findings provide a framework for understanding tumor–immune interactions and highlight the potential prognostic and biological relevance of EOMES and SPRYD3 in HNSCC.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12885-026-16379-3.

Introduction

Head and neck squamous cell carcinoma (HNSCC) encompasses a group of aggressive malignancies arising from the mucosal epithelium of the oral cavity, pharynx, and larynx. Despite advances in surgery, radiotherapy, chemotherapy, and immunotherapy, the overall survival rate for patients with advanced disease remains limited [1–3]. A substantial proportion of patients are diagnosed at a locally advanced stage with regional lymph node involvement, which contributes to unfavorable clinical outcomes [4–8].

HNSCC development is influenced by multiple etiological factors. Tobacco use, excessive alcohol consumption, and betel-nut chewing are well-established risk factors, and their combined exposure can synergistically promote tumorigenesis in certain populations [9, 10]. In addition, infection with high-risk human papillomavirus (HPV), particularly HPV16 and HPV18, defines a distinct subset of HNSCC characterized by unique molecular features and immune interactions [11–13]. Emerging evidence further suggests that the oral microbiome and host genetic susceptibility may also contribute to disease initiation and progression [14, 15].

At the molecular level, HNSCC is characterized by complex alterations involving both oncogenes and tumor suppressors [16]. Recurrent genetic events include amplification of EGFR and PIK3CA, mutations in TP53 and CDKN2A, and dysregulation of pathways related to cell cycle control, apoptosis, and DNA repair [17, 18]. Large-scale studies, including The Cancer Genome Atlas (TCGA), have highlighted the marked heterogeneity of HNSCC at the genomic and transcriptomic levels. However, translating these findings into clinically useful biomarkers remains challenging [19]. Many studies have focused on individual genes or bulk transcriptional signatures [20–28], which may not fully capture coordinated biological processes or the contribution of distinct cellular compartments.

Weighted gene co-expression network analysis (WGCNA) provides a systems-level approach to identify modules of co-regulated genes associated with disease phenotypes [29]. Compared with conventional differential expression analysis, WGCNA captures gene–gene correlation structures and facilitates the identification of hub genes within biologically relevant networks [30, 31]. At the same time, single-cell RNA sequencing enables resolution of gene expression patterns at the cellular level, allowing the assignment of transcriptional programs to specific cell populations. Integrating these approaches offers an opportunity to bridge network-level analysis with cellular context [32].

Despite the increasing availability of transcriptomic resources for head and neck squamous cell carcinoma (HNSCC), the cellular context underlying candidate prognostic genes remains incompletely understood. Single-cell transcriptomic analysis provides an opportunity to resolve cell-type-specific expression patterns and distinguish tumor-intrinsic programs from immune-associated processes. Oral squamous cell carcinoma (OSCC), the most common subtype of HNSCC, is supported by well-characterized public single-cell datasets and accessible clinical specimens in our center. Therefore, OSCC datasets and tissue samples were used for cellular-resolution and experimental validation analyses in the present study.

By integrating bulk transcriptomic analyses of HNSCC with single-cell and tissue-level validation in OSCC, we sought to identify candidate genes with prognostic relevance and defined cellular distribution. Through a combined framework incorporating co-expression network analysis and single-cell transcriptomic profiling, we identified SPRYD3 and EOMES as representative markers of distinct biological programs. SPRYD3 was primarily associated with malignant epithelial cells and proliferative signaling, whereas EOMES was enriched in cytotoxic immune populations. These findings provide insight into the coordination between tumor growth and immune activity in HNSCC and may facilitate future biomarker development and mechanistic investigations.

Materials and methods

Data acquisition and preprocessing

Gene expression profiles were obtained from the GEO database (GSE83519), which includes 22 paired HNSCC tumor and adjacent normal mucosal samples. Raw CEL files were processed in R using the affy package for background correction and quantile normalization. Probe annotations were mapped to gene symbols based on platform information, and when multiple probes corresponded to the same gene, the probe with the highest median expression was retained. Outlier samples were assessed by hierarchical clustering and principal component analysis (PCA) to ensure data quality.

Differential expression analysis

Differential expression between tumor and matched normal samples was assessed using the limma package [33] with paired design incorporated into the linear model. Statistical significance was defined as adjusted p < 0.05 with |log2 fold change| ≥ 1. Identified genes were used for subsequent network and functional analyses.

Functional enrichment analysis

Biological interpretation of differentially expressed genes was performed using Gene Ontology (GO) and KEGG pathway analysis implemented in clusterProfiler [34]. Only terms with false discovery rate (FDR) < 0.05 were considered significant. The top enriched terms were visualized with dot plots, in which dot size represented gene counts and color scale represented statistical significance.

Construction of co-expression network

To explore coordinated transcriptional patterns, a weighted gene co-expression network was constructed using the weighted gene co-expression network analysis (WGCNA) framework. Genes with high expression variability were selected to reduce noise. A soft-thresholding power was chosen based on scale-free topology criteria, and adjacency relationships were transformed into a topological overlap matrix. Modules of co-expressed genes were identified using hierarchical clustering and dynamic tree cutting, with a minimum module size of 30. Modules showing limited gene overlap with those derived from normal tissues were considered tumor-associated.

Hub gene prioritization

Within tumor-associated modules, genes were ranked according to their connectivity and correlation with tumor phenotype. Highly connected genes were further evaluated using protein interaction networks through STRING portal and visualized in Cytoscape [35] and selected as candidate hub genes for downstream analysis.

Survival analysis

The prognostic significance of candidate genes was evaluated in the TCGA-HNSC cohort. Patients were dichotomized into high- and low-expression groups according to the median gene expression level. Kaplan–Meier survival curves were generated, and differences between groups were assessed using the log-rank test. P < 0.05 was considered statistically significant.

Single-cell RNA sequencing data acquisition, preprocessing, and quality control

Given that OSCC represents the most common subtype of HNSCC and is supported by well-characterized public single-cell datasets, GSE172577 [36] from GEO database was selected for cellular-resolution analyses. Raw gene-barcode matrices in 10x Genomics format (barcodes.tsv, features.tsv, matrix.mtx) were obtained for six primary tumor samples (SYSMH1–SYSMH6). All samples were processed in R (version 4.3.2) using the Seurat package (version 5.0.3) [37]. For each sample, a Seurat object was created with genes detected in at least three cells and cells expressing at least 200 genes. Objects were then merged into an integrated dataset for downstream analysis. Low-quality cells were removed by applying the following criteria: (i) fewer than 200 or more than 6,000 detected genes, and (ii) > 15% of transcripts derived from mitochondrial genes. After filtering, remaining cells were log-normalized, and highly variable genes were identified using the “vst” method [38]. PCA was performed and the top 20 principal components were used for clustering (FindNeighbors, FindClusters, resolution = 0.5) and Uniform Manifold Approximation and Projection (UMAP) visualization.

Dimensionality reduction, clustering, and visualization

Data were scaled and subjected to PCA. The top 20 principal components were used for construction of a k-nearest neighbor graph and clustering by the Louvain algorithm at a resolution of 0.5. Non-linear dimensional reduction was performed using UMAP for visualization. Cluster-specific marker genes were identified using the Wilcoxon rank-sum test implemented in FindAllMarkers, with thresholds of log2 fold change > 0.25, adjusted p-value < 0.05, and expression in ≥ 25% of cells [39].

Cell type annotation

Cell types were annotated based on canonical marker genes using module scoring (AddModuleScore) and known lineage signatures, including epithelial/tumor (EPCAM, KRT8, KRT18), CD8⁺T (CD3D, CD8A, GZMB, NKG7), CD4⁺T/Treg (IL7R, FOXP3, IL2RA), B (MS4A1, CD79A), NK (KLRD1, PRF1), myeloid (LYZ, FCGR3A), fibroblast (COL1A1, DCN), and endothelial (VWF, PECAM1) signatures. Each cluster was assigned to the cell type with the highest module score margin, and cell-type-specific markers were confirmed through differential expression analysis (FindAllMarkers). Annotation was manually curated and cross-validated with automated reference-based classification using SingleR [40] with the BlueprintEncodeData reference.

Analysis of EOMES and SPRYD3 expression

Expression levels of EOMES and SPRYD3 were visualized across clusters and annotated cell types using FeaturePlot, VlnPlot, and DotPlot functions in Seurat. Differential expression of these genes between tumor and normal epithelial subsets was further assessed using Wilcoxon tests.

Inference of malignant versus normal epithelial populations

To distinguish malignant from non-malignant epithelial cells, we applied CopyKAT (version 1.0.8), which infers large-scale chromosomal copy number variations (CNVs) from scRNA-seq data [41]. Raw counts of epithelial-enriched cells (top 30% by EPCAM/KRT module score) were used as input. CopyKAT classified cells into aneuploid (tumor epithelial), diploid (normal epithelial), or uncertain groups. The malignancy label was integrated into the Seurat object and visualized on the UMAP. Proportions of tumor versus normal epithelial cells were quantified for each patient. Differential gene expression between malignant and normal epithelial subsets was performed using Seurat’s FindMarkers with the same thresholds as above.

To elucidate the biological functions and signaling pathways associated with transcriptional differences between tumor and normal epithelial cells, GO and KEGG enrichment analyses were conducted using the clusterProfiler package (v4.12.0) in R. DEGs were derived from Seurat-based comparisons between CopyKAT-defined tumor epithelial and normal epithelial subsets, using the criteria |log₂FC| > 0.25 and adjusted p < 0.05.

Clinical sample collection

OSCC tissues and adjacent normal mucosa were obtained from five patients undergoing surgical resection at Second Xiangya Hospital, Central South University. All patients provided written informed consent, and the study was approved by the Institutional Review Board (IRB Protocol ID: 20250557). Clinical characteristics of the patients, including age, sex, tumor depth, lymph-node status, and stage, are summarized in Table S1.

RNA extraction and quantitative real-time PCR (RT-qPCR)

Total RNA was isolated from frozen OSCC and adjacent tissues using TRIzol reagent (Invitrogen). RNA concentration and purity were assessed with a NanoDrop spectrophotometer, and integrity was verified by agarose gel electrophoresis. One microgram of RNA was reverse-transcribed into cDNA using a high-capacity cDNA reverse transcription kit. Quantitative PCR was performed with SYBR Green Master Mix (Takara Bio) on an ABI 7500 Real-Time PCR System (Applied Biosystems, USA). GAPDH was used as the internal control. Relative gene expression was calculated using the 2^−ΔΔCt method. Primer sequences were designed using Primer Premier 5 and synthesized commercially. Sequences are listed below:

EOMES_Fwd: GCTCAAATTCCACCGCCACC;

EOMES_Rev: CGCCAGCACCACCTCTACGAA;

SPRYD3_Fwd: ACCCGCAGCCACTACTTCGAG;

SPRYD3_Rev: GCATGATAAGCCACAGACCCTC;

GAPDH_Fwd: ACCCTGAAGTACCCCATCGAG;

GAPDH_Rev: AGCACAGCCTGGATAGCAAC.

Protein extraction and western blotting

OSCC and matched normal tissues were homogenized in RIPA lysis buffer (Sigma-Aldrich) supplemented with protease inhibitors. Protein concentration was determined by bicinchoninic acid (BCA) assay. Equal amounts of protein (20–40 µg) were resolved by SDS–PAGE and transferred to polyvinylidene difluoride (PVDF) membranes. Membranes were blocked with 5% nonfat milk and incubated overnight at 4 °C with primary antibodies: anti-SPRYD3 (Thermo Fisher), anti-EOMES (Abzoom), and anti-GAPDH (Abcam). After washing, membranes were incubated with horseradish peroxidase–conjugated secondary antibodies for 1 h at room temperature. Protein bands were visualized using enhanced chemiluminescence (ECL) and quantified with ImageJ. Uncropped blots are presented in Figure S1.

Statistical analysis

All statistical analyses were performed using R (v4.5) and GraphPad Prism (v9.0). Continuous variables are expressed as mean ± standard deviation (SD). Paired t-tests or Wilcoxon signed-rank tests were applied for tumor versus matched normal comparisons, as appropriate. For multiple testing in enrichment analyses, FDR control was used. A two-tailed P < 0.05 was considered statistically significant.

Results

Differential expression analysis and functional annotation

We first compared gene expression profiles between HNSCC tissues and matched normal mucosa to identify tumor-associated transcriptional changes. A total of 893 genes met the predefined significance criteria, including 518 upregulated and 375 downregulated genes (Fig. 1A, Table S2). Projection of these genes by dimensionality reduction clearly separated tumor from normal samples, indicating consistent transcriptomic differences across the cohort (Fig. 1B). Functional annotation showed that genes with higher expression in tumors were mainly related to epithelial differentiation and metabolic processes(Fig. 1C), whereas genes with lower expression were enriched in cell adhesion, tissue organization, and immune-related functions(Fig. 1D). Pathway analysis further indicated that tumor-upregulated genes were associated with signaling pathways linked to cellular adaptation and proliferation(Fig. 1E), while downregulated genes were enriched in pathways involved in inflammatory and immune responses(Fig. 1F). These results indicate a coordinated shift in tumor tissues toward increased proliferative activity and reduced immune-associated signaling.

Fig. 1.

Differential expression analysis and functional annotation in HNSCC. A Volcano plot showing differentially expressed genes between tumor and matched normal tissues from the GSE83519 dataset. The x-axis represents log2 fold change, and the y-axis represents –log10 adjusted P value. Statistical significance was defined as adjusted p < 0.05 with |log2 fold change| ≥ 1. B UMAP projection of samples based on differentially expressed genes. C–D Gene Ontology (GO) enrichment analysis of upregulated (C) and downregulated (D) genes. E–F KEGG pathway enrichment analysis of upregulated (E) and downregulated (F) genes. Bubble plots display gene ratio, number of genes, and adjusted P values

Co-expression network construction and module identification

Given the observed transcriptional differences between tumor and normal tissues, we next performed weighted gene co-expression network analysis (WGCNA) to identify gene modules associated with tumor status and to prioritize candidate hub genes. Based on the predefined filtering criteria, a co-expression network was constructed, and hierarchical clustering of samples is shown in Fig. 2A. For the HNSCC samples, a soft-thresholding power of 8 was selected, resulting in the identification of 11 gene modules (Fig. 2B). The size of these modules ranged from 63 genes (purple module) to 481 genes (turquoise module), as summarized in Table S3. For the normal samples, a soft-thresholding power of 12 was applied, yielding 10 modules, with module sizes ranging from 114 genes (magenta module) to 328 genes (turquoise module) (Fig. 2A). To identify tumor-associated modules, we compared gene composition between tumor and normal networks. Most modules showed varying degrees of overlap between the two groups, with the exception of the gray module. Notably, the magenta and green modules identified in the tumor network showed minimal overlap with modules derived from normal tissues (Fig. 2B), suggesting that these modules may represent tumor-specific transcriptional programs in HNSCC.

Fig. 2.

Gene co-expression network analysis. A Hierarchical clustering dendrogram of genes in tumor and normal samples. B Overlap of gene modules between tumor and normal networks. C Network visualization of hub genes in selected modules. Node size represents connectivity, and edge thickness represents interaction strength

Identification of prognostic hub genes and their single-cell expression patterns in OSCC

Co-expression networks derived from tumor-specific modules were further visualized using Cytoscape to identify key regulatory genes. From the green and magenta modules, the top 40 genes ranked by intramodular connectivity were selected for downstream analysis (Fig. 2C). Among these, the five most highly connected genes in each module were considered representative hub genes (green module: ADCY8, B4GALT7, OR8G1, EOMES, UBXN10; magenta module: CD248, OVOL2, PURA, SPRYD3, RPS19).

We next evaluated the clinical relevance of these candidate genes using the TCGA-HNSCC cohort. Survival analysis showed that only EOMES and SPRYD3 were significantly associated with overall survival (Fig. 3A–I). Increased expression of EOMES was linked to favorable prognosis (Fig. 3C), whereas higher SPRYD3 expression was associated with reduced survival (Fig. 3I). These results suggest that EOMES and SPRYD3 may serve as prognostic markers with distinct roles in HNSCC.

Fig. 3.

Survival analysis of candidate hub genes. Kaplan–Meier curves showing overall survival stratified by gene expression levels (A) ADCY8, (B) CD248, (C) EOMES, (D) UBXN10, (E) OR8G1, (F) OVOL2, (G) PURA, (H) RPS19, (I) SPRYD3. Patients were divided into high- and low-expression groups based on median values. P<0.05 was considered statistatically significant

To further characterize candidate genes at the cellular level, we analyzed a publicly available OSCC single-cell RNA-seq dataset (GSE172577). After quality control, 63,482 cells from six tumors (SYSMH1–SYSMH6) were retained. Unsupervised clustering identified multiple cell populations, including epithelial/tumor cells, CD8⁺ T cells, CD4⁺ T/Treg cells, NK cells, B cells, myeloid cells, fibroblasts, and endothelial cells (Fig. 4A–B). Cell-type annotation based on canonical markers and module scoring reproduced expected cellular identities within the tumor microenvironment. This dataset provided a framework to further assess the expression patterns of EOMES and SPRYD3 across different cell populations.

Fig. 4.

Single-cell analysis and identification of malignant epithelial cells in OSCC. Single-cell RNA-seq data were obtained from an OSCC cohort (GSE172577). A–B UMAP visualization and cell type annotation. C–D Expression of EOMES and SPRYD3 across cell populations. E–F CopyKAT-based inference of copy number variation profiles. G UMAP distribution of malignant and non-malignant epithelial cells. H Expression comparison of EOMES and SPRYD3 between epithelial subpopulations. I–L Functional enrichment analysis of genes upregulated and downregulated in tumor epithelial cells

Cell type–specific expression of EOMES and SPRYD3 in malignant epithelial cells

To examine the functional relevance of the identified hub genes, we next analyzed the expression patterns of EOMES and SPRYD3 across single-cell populations in OSCC. SPRYD3 expression was predominantly detected in epithelial/tumor cells, whereas EOMES expression was mainly enriched in immune compartments, particularly CD8⁺ T cells and NK cells (Fig. 4C–D). Quantitative analysis further showed that SPRYD3 expression was substantially higher in aneuploid epithelial cells compared with diploid epithelial cells, while EOMES expression remained low in epithelial populations but was prominent in cytotoxic lymphocytes. These distinct expression patterns suggest that SPRYD3 and EOMES are associated with tumor-intrinsic and immune-related processes, respectively.

To distinguish malignant from non-malignant epithelial cells, we performed copy number variation inference using CopyKAT at single-cell resolution. This analysis separated epithelial cells into two major populations: aneuploid cells with extensive chromosomal alterations and diploid cells with stable genomic profiles (Fig. 4E–F). Aneuploid cells were defined as tumor epithelial cells, whereas diploid cells were considered non-malignant.

Low-dimensional visualization showed clear separation between these two populations (Fig. 4G), and cell-type composition analysis confirmed that epithelial clusters were the primary contributors to the aneuploid population, while immune and stromal cells remained largely diploid. Across all six OSCC samples, malignant epithelial cells constituted the majority of epithelial populations, with proportions varying between samples (Table S4).

Further visualization demonstrated that SPRYD3 expression was markedly elevated in malignant epithelial cells, whereas EOMES expression was enriched in non-malignant or immune-associated populations (Fig. 4H). These results support the use of CopyKAT-based classification to distinguish malignant and normal epithelial cells and provide a basis for linking gene expression patterns to tumor-specific cellular states.

SPRYD3 and EOMES are associated with proliferative activation and reduced immune signaling in malignant epithelial cells

We next examined gene expression differences between malignant and non-malignant epithelial cells defined by CopyKAT. SPRYD3 expression was significantly increased in tumor epithelial cells, whereas EOMES expression showed a moderate decrease (Table S5). These patterns are consistent with the enrichment of SPRYD3 in proliferative epithelial compartments and the restriction of EOMES to immune-associated cell populations.

To further characterize the biological features of tumor epithelial cells, we performed pathway enrichment analysis based on differentially expressed genes. Genes with higher expression in malignant epithelial cells were primarily associated with cell cycle regulation, DNA replication, and chromosome segregation, indicating enhanced proliferative activity (Fig. 4I). Consistently, pathway analysis showed enrichment of signaling cascades related to cell cycle control and survival, including p53 and PI3K–Akt pathways (Fig. 4J).

In contrast, genes with reduced expression were enriched in processes related to epithelial differentiation, cell–cell adhesion, and immune-associated functions, such as antigen processing and inflammatory signaling (Fig. 4K–L). These transcriptional changes suggest that malignant epithelial cells undergo a shift toward increased proliferation and reduced differentiation, accompanied by decreased immune-related activity.

Taken together, these findings indicate that tumor epithelial cells in OSCC are characterized by coordinated activation of proliferative programs and attenuation of immune-associated pathways, consistent with the distinct expression patterns observed for SPRYD3 and EOMES.

Validation of SPRYD3 and EOMES expression in OSCC tissues

To validate the findings from transcriptomic analyses, the expression levels of SPRYD3 and EOMES were examined in OSCC tissues and matched adjacent samples using RT-qPCR and western blotting. SPRYD3 mRNA expression was significantly higher in tumor tissues, whereas EOMES expression showed a decreasing trend (Fig. 5A–B).

Fig. 5.

Experimental validation of SPRYD3 and EOMES expression. A–B Relative mRNA expression of SPRYD3 and EOMES in five paired OSCC and adjacent normal tissues measured by RT-qPCR. C–D Protein expression of SPRYD3 and EOMES in five paired OSCC and adjacent normal tissues determined by western blotting. Data were analyzed using paired Student’s t-test. P < 0.05 was considered statistically significant. T, tumor tissue; N, adjacent normal tissue

At the protein level, SPRYD3 abundance was consistently increased in OSCC samples compared with adjacent tissues (Fig. 5C). In contrast, EOMES protein expression was lower in tumor tissues, although this difference did not reach statistical significance (Fig. 5D). These findings support increased SPRYD3 expression in OSCC and are generally consistent with transcriptomic observations suggesting lower EOMES expression, although the difference in EOMES protein abundance did not reach statistical significance in the validation cohort.

Discussion

Head and neck squamous cell carcinoma remains a biologically heterogeneous disease in which tumor-intrinsic programs and immune context jointly shape clinical outcomes [5, 9, 42]. It should be noted that the single-cell analyses and experimental validation were performed in OSCC cohorts. These findings primarily reflect biological features of OSCC, although OSCC constitutes the largest subgroup of HNSCC. Therefore, in the present study, we applied an integrative framework combining bulk transcriptomics, co-expression network analysis, single-cell profiling, and experimental validation to identify candidate prognostic markers associated with OSCC progression. Our results converge on two functionally distinct but complementary axes: a tumor-intrinsic proliferative program represented by SPRYD3 and an immune-associated cytotoxic program represented by EOMES.

Rather than functioning as isolated biomarkers, these genes appear to reflect coordinated biological processes operating across different cellular compartments. At the single-cell level, SPRYD3 expression was predominantly restricted to aneuploid epithelial cells defined by CopyKAT-based CNV inference, whereas EOMES expression was largely confined to cytotoxic lymphocyte populations, including CD8⁺ T cells and NK cells. This spatial segregation highlights a key feature of OSCC biology, in which malignant epithelial expansion is accompanied by heterogeneous and often attenuated immune activation.

Functional analyses further support this dichotomy. Tumor epithelial cells characterized by high SPRYD3 expression exhibited enrichment of cell cycle progression, DNA replication, and PI3K–Akt signaling pathways, consistent with enhanced proliferative and survival capacity. In contrast, genes associated with immune activation, including antigen processing and inflammatory signaling, were relatively suppressed in these cells. These findings suggest that malignant epithelial cells may adopt transcriptional programs that simultaneously promote growth and limit immune recognition, thereby contributing to tumor progression.

EOMES, a T-box transcription factor, has been widely implicated in regulating cytotoxic lymphocyte differentiation and effector function [43–45]. Its role in cancer appears to be context-dependent. In several solid tumors, EOMES expression has been associated with enhanced antitumor immunity and favorable prognosis, whereas in other settings it has been linked to T-cell dysfunction or exhaustion.In hepatocellular and renal cell carcinomas, elevated EOMES expression correlates with enhanced antitumor immunity and favorable prognosis [46], whereas in bladder cancer, acute myeloid leukemia, and non-small cell lung cancer, overexpression has been linked to T-cell exhaustion and poor outcomes [47–49]. Importantly, our single-cell analyses indicate that EOMES expression was enriched in cytotoxic lymphocyte populations rather than malignant epithelial cells and correlated with improved survival, suggesting that it may reflect preserved effector function rather than exhaustion in OSCC. Therefore, the favorable prognostic association observed in bulk transcriptomic analyses is more likely to reflect cytotoxic immune infiltration and antitumor immune activity than a tumor-cell-intrinsic effect. This interpretation is supported by its co-expression with canonical cytotoxic markers and immune activation signatures.

SPRYD3, a member of the SPRY domain-containing protein family, mediates protein–protein interactions involved in cell-cycle regulation, chromosomal organization, and inflammatory signaling [50, 51]. For instance, SPRYD4 overexpression in hepatocellular carcinoma was linked to favorable prognosis [52]. In contrast, SPRYD3 remains poorly characterized in cancer biology. Members of the SPRY domain–containing protein family are known to participate in protein–protein interactions and signaling regulation, however, their roles in tumor progression are not well defined. In this study, SPRYD3 was consistently upregulated in malignant epithelial cells and associated with adverse prognosis. Its expression pattern, together with enrichment of proliferation- and metabolism-related pathways, suggests a potential role in sustaining tumor growth. Whether SPRYD3 directly regulates these pathways or acts as part of a broader regulatory complex remains to be determined.

An important strength of this study lies in the integration of network-level and single-cell analyses. WGCNA enabled identification of tumor-associated gene modules, while single-cell profiling provided cellular resolution of gene expression patterns. For example, by combining WGCNA with single-cell CNV analysis through CopyKAT, we identified tumor-specific co-expression modules and directly linked them to cell-level heterogeneity within OSCC. This combination allowed us to connect system-level transcriptional programs with specific cell populations within the tumor microenvironment. Clinical validation analyses were generally consistent with the transcriptomic findings, providing strong support for SPRYD3 upregulation and partial support for reduced EOMES expression in OSCC tissues.

Several limitations should be considered when interpreting the present findings. First, although the discovery analyses were conducted using HNSCC cohorts, the single-cell and experimental validation datasets were restricted to OSCC. Therefore, the applicability of these observations to other HNSCC subsites requires further investigation. Second, the bulk transcriptomic analyses were primarily based on a single GEO cohort together with TCGA-HNSC data, and additional independent cohorts would strengthen the robustness and generalizability of the findings. Third, while single-cell analyses provided insights into the cellular distribution and biological context of EOMES and SPRYD3, the underlying molecular mechanisms by which these genes contribute to tumor progression and immune regulation remain to be fully elucidated. In addition, the experimental validation cohort consisted of only five paired OSCC samples, which may have limited the statistical power to detect differences in EOMES protein expression. Finally, potentially relevant clinical variables, including HPV status, smoking history, and treatment heterogeneity, were not incorporated into stratified analyses and may influence both gene expression patterns and clinical outcomes [53].

Future studies should aim to validate these findings in larger, independent cohorts and incorporate detailed clinical annotations. Functional experiments, including gene perturbation and co-culture systems, will be necessary to clarify the mechanistic roles of SPRYD3 in tumor epithelial cells and EOMES in immune regulation. Moreover, integrating additional omics layers such as epigenomic and proteomic data may further elucidate upstream regulatory mechanisms and downstream effectors.

Conclusions

In summary, our study identifies SPRYD3 and EOMES as markers of distinct biological programs in OSCC, linking epithelial proliferation and immune activity, respectively. These findings improve our understanding of tumor–immune interactions in HNSCC and support further investigation of EOMES and SPRYD3 as prognostic biomarkers in HNSCC.

Authors’ contributions

Y.L. conceived and supervised the study. L.H., Y.Z., and Y.G. performed data analysis. X.X. contributed to data processing and interpretation. L.H., X.X., and Y.L. drafted the manuscript. All authors revised the manuscript critically for important intellectual content and approved the final version.

Funding

Not applicable.

Data availability

All data analyzed in this study are publicly available. Bulk RNA-seq data were obtained from The Cancer Genome Atlas and the Gene Expression Omnibus (GEO; accession: GSE83519). Single-cell RNA-seq data were obtained from GEO (GSE172577). Processed data and analysis scripts are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the Second Xiangya Hospital, Central South University (Approval No. 20250557).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.