CXCL10 as a migration-associated biomarker in oral squamous cell carcinoma

Abstract

Background

Oral squamous cell carcinoma (OSCC) is marked by aggressive local invasion and early metastasis, yet molecular biomarkers linked to tumor cell motility remain insufficiently defined. CXCL10, an interferon-inducible chemokine, exhibits context-dependent roles across cancers, but its functional relevance to OSCC migration has not been fully elucidated.

Methods

Three GEO transcriptomic datasets (GSE13601, GSE160042, GSE227919) were analyzed to identify differentially expressed genes (DEGs). Functional enrichment, STRING–Cytoscape protein–protein interaction analysis, and CytoHubba/MCODE were used to prioritize hub genes. A random-effects meta-analysis quantified the pooled expression differences of key candidates. CXCL10 expression was validated using UALCAN, CPTAC, and an independent OSCC tissue microarray. The functional role of CXCL10 was examined via siRNA-mediated knockdown and recombinant CXCL10 treatment in HSC3 and SCC4 cell lines using wound-healing and Transwell migration assays.

Results

A total of 263 overlapping DEGs were identified across the three cohorts, enriched in pathways related to extracellular matrix organization, focal adhesion, immune regulation, and antiviral responses. Network analyses identified five hub genes, among which CXCL10 demonstrated the highest pooled overexpression in OSCC (log₂FC = 4.27; 95% CI 2.02–6.52). UALCAN and CPTAC analyses confirmed elevated CXCL10 transcript and protein levels in tumor tissues, which were further validated by tissue-array immunohistochemistry. Functionally, CXCL10 knockdown significantly reduced OSCC cell migration, whereas recombinant CXCL10 dose-dependently enhanced motility.

Conclusions

CXCL10 is markedly overexpressed in OSCC and contributes to tumor cell motility in vitro. These findings support CXCL10 as a migration-associated candidate biomarker with potential clinical relevance, warranting further validation in OSCC cohorts.

Introduction

Oral squamous cell carcinoma (OSCC) is the predominant malignancy of the oral cavity, comprising over 90% of oral cancers and representing a major subset of head and neck squamous cell carcinoma (HNSC). Despite therapeutic advances, the 5-year overall survival rate has remained at approximately 50–60%, largely due to locoregional invasion, recurrence, and metastasis [1, 2]. Clinical parameters alone insufficiently predict outcome, and although immune checkpoint inhibitors such as pembrolizumab and nivolumab provide benefit in recurrent/metastatic disease, only a minority of patients achieve durable responses [2, 3]. Thus, the identification of robust biomarkers that reflect tumor progression, metastatic potential, or immune responsiveness remains an unmet need in OSCC management.

Metastasis is the principal cause of cancer mortality and requires coordinated processes involving epithelial–mesenchymal transition (EMT), extracellular matrix (ECM) remodeling, directional migration, intravasation, and colonization at distant sites. Increasing evidence highlights the tumor microenvironment (TME) and chemokine signaling as major regulators of these steps. Chemokine axes such as CXCL12–CXCR4, CXCL8–CXCR1/2, and CCL2–CCR2 promote invasion, EMT, angiogenesis, and immunosuppression across cancers. In tongue cancer, CXCL12–CXCR4 enhances migration via PI3K/AKT signaling, while CXCL8 from tumor-associated macrophages accelerates metastasis and stemness in breast cancer [4–8]. The CCL2–CCR2 axis also drives macrophage recruitment, immunosuppression, and metastatic niche formation in multiple cancers [9, 10]. These findings underscore the importance of chemokine-mediated communication in promoting invasion and metastatic niche formation—mechanisms highly relevant to OSCC, where early nodal metastasis strongly predicts poor prognosis.

Among interferon-inducible CXC chemokines, CXCL10 (IP-10) has garnered attention for its dual roles in tumor immunity and tumor progression. CXCL10 signals mainly through CXCR3, which has isoforms with divergent biological effects. In several malignancies, CXCL10 has been linked to enhanced proliferation, migration, and invasion. For example, in colorectal cancer, CXCL10 promotes invasiveness via CXCR3A-dependent MMP9 induction [11], and in osteosarcoma, CXCL10-CXCR3 signaling supports lung metastasis through AKT/PAK1 activation [12]. CXCL10 may also shape metastatic microenvironments, including fibroblast-derived CXCL9/10 in breast cancer and astrocyte-derived CXCL10 in brain metastasis [13, 14], and it can promote bone metastasis through effects on osteoclast precursors [15]. Conversely, CXCL10 can enhance anti-tumor immunity by recruiting CXCR3 + cytotoxic T cells and NK cells, and higher CXCL10 expression has been associated with immune infiltration and favorable outcomes in multiple cancers, including HNSC [16–19]. Induction of CXCL10 by radiation, STING activation, HDAC3 inhibition, or AID/APOBEC3H upregulation can enhance CD8 + T-cell trafficking and suppress tumor growth [20–23]. In HNSC, CXCL10-positive TAM or CAF subsets contribute to distinct immune niches that may either support or restrict T-cell activity [24, 25]. Together, these findings suggest that CXCL10 exerts context-dependent effects shaped by tumor-intrinsic signaling, CXCR3 isoform usage, stromal composition, and immune status. However, its mechanistic role in OSCC remains incompletely defined, particularly with respect to tumor cell motility and metastatic potential. Although bioinformatic studies suggest CXCL10 upregulation in OSCC and possible links to immune escape [26, 27], experimental validation and functional characterization in OSCC are still limited.

To address this knowledge gap, we integrated multi-cohort transcriptomic comparisons, network-based gene prioritization, and statistical meta-analysis to identify consistently dysregulated and migration-related genes in OSCC. CXCL10 emerged as the top overexpressed hub gene across datasets. We validated CXCL10 expression using public databases and an independent OSCC tissue microarray, and performed functional assays with siRNA and recombinant CXCL10 to characterize its impact on OSCC cell motility. This integrative experimental and bioinformatic framework provides new evidence supporting CXCL10 as a migration-associated gene and potential biomarker contributing to OSCC progression.

Materials and methods

Data acquisition and dataset selection

Three publicly available transcriptomic datasets (GSE13601, GSE160042, and GSE227919) were retrieved from the Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo/) to identify DEGs between OSCC and normal tongue tissues. Datasets were selected based on sample size, the availability of tumor and non-tumor controls, and the use of validated microarray or RNA-seq platforms. GSE13601 (GPL8300; Affymetrix Human Genome U95 v2 Array): Obtained on Oct 19, 2025, includes 31 tongue carcinoma samples and 26 normal tongue samples. GSE160042 (GPL18180; Agilent Human LncRNA v4 platform): Obtained on Oct 21, 2025, includes 10 tongue tumor tissues and 10 matched non-tumor controls. GSE227919 (GPL18573; Illumina NextSeq 500 RNA-seq): Obtained on Oct 17, 2025, includes 8 oral cancer samples and 18 healthy oral mucosa samples. Because GSE227919 is an RNA-seq dataset, it provides high-resolution quantitative expression data and supports cross-platform validation. Combining microarray- and sequencing-based datasets reduced platform-specific bias and improved the robustness of downstream differential expression, enrichment, and network analyses (Fig. 1).

Fig. 1.

Workflow of the study design. A schematic overview of the integrative analysis pipeline used to identify and characterize migration-associated genes in OSCC

Identification of DEGs

Differential gene expression analysis between OSCC and normal oral tissues was conducted using the GEO2R analytical platform (https://www.ncbi.nlm.nih.gov/geo/geo2r/; accessed Oct 30, 2025). GEO2R utilizes the GEOquery and limma R packages to perform standardized statistical comparisons across GEO series datasets [28]. For each dataset, raw expression values were normalized prior to analysis to ensure comparability between tumor and control groups. Genes exhibiting significant differential expression were identified using the following thresholds: Benjamini–Hochberg adjusted p-value < 0.01, and absolute log₂ fold change (|log₂FC|) ≥ 1.0. These criteria were applied uniformly across all datasets to minimize platform-related variability and to retain genes with both statistical and biological significance. To determine consistently dysregulated genes, DEG lists from all three datasets were intersected using the InteractiVenn tool (https://www.interactivenn.net/; accessed Nov 11, 2025) [29]. This integrative approach enabled the identification of overlapping DEGs, thereby strengthening the reliability of downstream enrichment and network analyses. The resulting consensus DEG set reflects robust transcriptional alterations reproducibly associated with OSCC across independent cohorts.

Hallmark enrichment, Gene Ontology (GO) annotation, and KEGG pathway analysis

To investigate the biological relevance of the overlapping DEGs from GSE13601, GSE160042, and GSE227919, hallmark enrichment analysis was performed using the Cancer Hallmarks Analytics Tool (https://cancerhallmarks.com; accessed Nov 11, 2025). This platform uses a hypergeometric test to evaluate enrichment across ten cancer hallmark categories [30], and statistical significance was assessed using the adjusted p-values reported by the tool. Results were summarized as a Hallmark Enrichment Plot with five concentric rings. The outer ring displays the ten hallmark categories; segment size reflects the number of mapped DEGs and color indicates hallmark type. Inner rings represent adjusted p-value ranges (1.0 to 0.0001), and non-significant categories (adjusted p-value > 0.05) are shown in gray. Functional annotation was performed using ShinyGO v0.82 (http://bioinformatics.sdstate.edu/go/; accessed Nov 12, 2025). GO enrichment was analyzed across Biological Process (BP), Molecular Function (MF), and Cellular Component (CC), using p < 0.01 and FDR < 0.05. KEGG pathway enrichment was performed in ShinyGO using the same thresholds [31–33].

Protein–Protein Interaction (PPI) network construction and hub gene identification

To examine functional relationships among the identified DEGs, a PPI network was constructed using the STRING database (http://string-db.org/; accessed Nov 13, 2025) [34, 35]. Interactions with a minimum required interaction score of medium confidence (combined score > 0.4) were retained. Nodes represent proteins encoded by DEGs, and edges represent STRING-supported interactions derived from multiple evidence sources. The resulting PPI network consisted of 263 nodes and 1,309 edges, with an average node degree of 9.95 and an average local clustering coefficient of 0.449. Protein–protein interaction enrichment analysis revealed a statistically significant enrichment (p < 1.0 × 10⁻¹⁶), indicating that the observed interactions were unlikely to occur by chance. STRING-derived interaction data were imported into Cytoscape v3.10.2 for visualization and topological assessment [36, 37]. MCODE-based clustering analysis was performed on the entire STRING-derived PPI network using interaction data exported in TSV format, with the following parameters: degree cutoff = 2, node score cutoff = 0.2, k-core = 2, maximum depth = 100, and the haircut option enabled. Among the extracted modules, the highest-scoring MCODE cluster (MCODE score = 27.357), consisting of 29 nodes and 383 edges, was selected as the densest functional module for downstream hub-gene identification. Within the selected MCODE cluster, node connectivity was evaluated by degree centrality to assess local connectivity. To complement this with a global network-flow perspective, CytoHubba analysis was applied using the Bottleneck algorithm (betweenness-based), which identifies nodes occupying critical control points on shortest paths throughout the network. Genes ranking among the top candidates in both degree and Bottleneck analyses were defined as core hub genes. Using this integrated prioritization strategy, interferon induced transmembrane protein 3 (IFITM3), 2’-5’-oligoadenylate synthetase 2 (OAS2), CXCL10, TNF superfamily member 10 (TNFSF10), and XIAP-associated factor 1 (XAF1) were identified as the top five hub genes within the dominant MCODE cluster.

Meta-analysis of hub gene expression across GEO datasets

To quantitatively compare the expression of hub genes across independent OSCC transcriptomic datasets, we performed a random-effects meta-analysis using log₂FC values derived from GSE13601, GSE160042, and GSE227919. For each dataset, tumor–normal log₂FC values and corresponding standard errors were calculated from GEO2R-generated differential expression results. Meta-analyses were conducted using the DerSimonian–Laird random-effects model, which accounts for between-study heterogeneity. Forest plots were generated to visualize pooled effect sizes and 95% confidence intervals (CIs).

Clinical expression validation

CXCL10 mRNA expression in HNSC was examined using TCGA data queried through the UALCAN portal (http://ualcan.path.uab.edu/; accessed Nov 14, 2025), which provides interactive access to publicly available cancer OMICS datasets. CXCL10 protein expression was assessed using CPTAC HNSC proteomic data as implemented in UALCAN [38, 39]. In UALCAN, CPTAC protein abundance is visualized as jitter plots, where Z-values represent standard deviations from the median across samples within the given cancer type. According to the UALCAN/CPTAC implementation, log₂ spectral count ratio values are normalized within each sample profile and then normalized across samples prior to visualization. For each comparison (tumor vs. normal and clinicopathologic subgroup analyses), the number of samples with available protein expression data was taken from the UALCAN output. All statistical comparisons and data processing followed the methods implemented by the UALCAN platform.

Reagents and antibodies

The anti-CXCL10 primary antibody (GTX642109) was purchased from GeneTex (Irvine, CA, USA). Recombinant human CXCL10 (IP-10) (PeproTech, #250 − 16; Thermo Fisher Scientific, Waltham, MA, USA) was used for gain-of-function experiments. Culture inserts for wound healing assays (Cat. No. 80209) were obtained from ibidi GmbH. Unless otherwise specified, all additional reagents were sourced from Sigma-Aldrich.

Cell culture

Human OSCC cell lines SCC4 (ATCC, Manassas, VA, USA) and HSC3 (Sigma-Aldrich) were maintained in DMEM supplemented with 10% fetal bovine serum, penicillin (100 U/mL), and streptomycin (100 µg/mL). Cells were cultured at 37 °C in a humidified incubator containing 5% CO₂.

Cell migration and wound healing assay

For Transwell migration assays, 3 × 10⁴ cells were treated with recombinant CXCL10 (0–50 ng/mL) and seeded into 8-µm pore-size inserts in serum-free medium. After 24 h, non-migrated cells on the upper membrane surface were removed, while migrated cells on the lower surface were fixed with 4% formaldehyde, stained with 0.05% crystal violet, and imaged microscopically. Migratory activity was quantified using ImageJ software (v1.52a, NIH). For wound-healing assays, SCC4 and HSC3 cells (3 × 10⁴) were seeded into ibidi two-well culture inserts and allowed to adhere for 24 h. Inserts were then removed to generate a uniform wound region. Cells were subsequently treated with recombinant CXCL10 (0–50 ng/mL) for 24 h. Images captured at 0 and 24 h were analyzed using ImageJ to determine wound closure rates.

Transient transfection

Transient knockdown of CXCL10 was achieved using Lipofectamine 3000 (Thermo Fisher Scientific) in accordance with the manufacturer’s protocol. Cells were transfected with either control siRNA or CXCL10-targeting siRNA (Sigma-Aldrich). The sequences for control siRNA were 5’-GAUCAUACGUGCGAUCAGA-3’ and 5’-UCUGAUCGCACGUAUGAUC-3’. CXCL10 siRNA sequences were 5’-CCAAUGAUGGUCACCAAAU-3’ and 5’-AUUUGGUGACCAUCAUUGG-3’. Transfection efficiency was verified using fluorescently labeled FAM-siRNA in HSC3 cells. Clear intracellular fluorescence was observed after transfection under fluorescence microscopy, whereas control siRNA showed no detectable signal, confirming successful siRNA delivery.

Immunohistochemical (IHC) staining

Paraffin-embedded tissue microarray sections (Catalog T273; TissueArray, Derwood, MD, USA) were processed for IHC staining. Sections were deparaffinized in xylene, rehydrated through graded ethanol, and subjected to heat-induced antigen retrieval using sodium citrate buffer. Endogenous peroxidase activity was quenched, and non-specific binding was blocked using the Novolink Polymer Detection System (Leica Biosystems, Nussloch, Baden-Württemberg, GER). Slides were incubated with anti-CXCL10 primary antibody (1:200 dilution) overnight at 4 °C, followed by polymer detection and DAB visualization. Hematoxylin was used for nuclear counterstaining, and slides were mounted and examined using a Nikon ECLIPSE Ti microscope. CXCL10 staining was semi-quantitatively assessed by three independent, blinded evaluators using the H-score method, based on staining intensity (0–3) and the percentage of positive cells. The H-score was calculated as: H-score = Σ (Pi × Ii) × 100, where Pi represents the percentage of cells at each staining intensity and Ii denotes the corresponding intensity category.

Statistical analysis

Statistical analyses were performed using GraphPad Prism (version 8.0.2) unless otherwise stated. All in vitro experiments were conducted with at least four independent biological replicates (n ≥ 4). Quantitative data are presented as mean ± SD. For cell-based functional assays, data distribution and assumptions for parametric testing were evaluated in GraphPad Prism prior to analysis. For comparisons involving two factors, statistical significance was assessed using two-way ANOVA, followed by Sidak’s multiple comparisons test or Tukey’s multiple comparisons test, as appropriate. For TCGA/UALCAN and CPTAC/UALCAN analyses, statistical comparisons and data processing followed the methods implemented by the respective platforms. Unless otherwise specified, a p value < 0.05 was considered statistically significant.

Results

Identification of differentially expressed genes in OSCC

To characterize transcriptional alterations associated with OSCC, differential expression analysis was performed across three GEO datasets—GSE13601, GSE160042, and GSE227919—each containing tumor and normal tongue tissues. Genes with |log₂FC| ≥ 1 and adjusted p < 0.01 were considered significantly dysregulated. Volcano plots from each dataset clearly distinguished upregulated (red) and downregulated (blue) genes relative to non-significant transcripts (gray) (Fig. 2A–C). The datasets consisted of 31 tumor and 26 normal samples in GSE13601, 10 matched tumor–normal pairs in GSE160042, and 8 tumor and 18 normal samples in GSE227919 (Fig. 2D). Differential expression analysis identified 1,852 DEGs in GSE13601, 3,820 in GSE160042, and 2,440 in GSE227919. To increase robustness and reduce dataset-specific bias, we intersected DEGs across all three cohorts, yielding a set of 263 consistently dysregulated genes (Fig. 2E). This integrative DEG signature represents reproducible transcriptomic changes in OSCC and served as the foundation for subsequent enrichment, network, and functional analyses.

Fig. 2.

Identification of Differentially Expressed Genes (DEGs) in OSCC across three GEO datasets. A–C Volcano plots of DEGs in GSE13601, GSE160042, and GSE227919, using |log₂FC| ≥ 1 and adjusted p < 0.01. Upregulated genes are shown in red, downregulated genes in blue, and non-significant genes in gray. D Sample distributions in each dataset, showing tumor and normal tissue counts. E Venn diagram illustrating the intersection of DEGs across the three datasets, yielding 263 consistently dysregulated genes shared among all cohorts

Functional enrichment analysis of OSCC-associated DEGs

To elucidate the biological roles of the 263 consistently dysregulated genes identified across OSCC datasets, we performed cancer hallmark enrichment analysis. The DEGs were significantly enriched in processes related to tissue invasion and metastasis, evading immune destruction, resisting cell death, sustained angiogenesis, tumor-promoting inflammation, and proliferative signaling (Fig. 3A). Enrichment in hallmarks such as evading growth suppressors and replicative immortality further suggests broad involvement in core oncogenic programs. GO analysis provided more detailed functional insights. In the MF domain, DEGs were enriched for ECM structural constituents, integrin binding, serine-type peptidase activity, and cell adhesion molecule binding (Fig. 3B). In the CC domain, DEGs were enriched in basement membrane, collagen-containing ECM, extracellular vesicles, and extracellular space, consistent with prominent extracellular remodeling (Fig. 3C). In the BP domain, enrichment was observed in antiviral and immune responses, cell adhesion, proliferation, responses to external stimuli, and cell-surface receptor signaling (Fig. 3D). KEGG pathway analysis identified ECM-receptor interaction, focal adhesion, PI3K-Akt signaling, and several viral/immune-related pathways (e.g., influenza A, measles, and HPV infection) as significantly enriched (Fig. 3E). Overall, these results highlight ECM organization, adhesion, and migration-related signaling as dominant themes in OSCC. Interferon- and immune-related programs were also represented, consistent with the biology of CXCL10, but the primary enrichment pattern supported our focus on motility-associated genes.

Fig. 3.

Functional enrichment analysis of the 263 overlapping DEGs. A Cancer hallmarks analytics tool visualization showing enrichment of DEGs in hallmark processes. Adjusted p < 0.05 was considered statistically significant. B–D ShinyGO-based GO enrichment of MF, CC, and BP categories. E KEGG pathway. The size of each bubble corresponds to the number of genes enriched in each term, while color represents the –log₁₀(FDR); FDR < 0.05 indicates significant

Network topology identifies five central hub genes in OSCC

To further delineate key regulatory genes among the 263 overlapping DEGs, a PPI network was constructed using STRING and visualized in Cytoscape. Network clustering with the MCODE algorithm revealed several functional modules, among which the largest cluster contained 29 nodes and 383 edges, representing the densest interaction subnetwork (Fig. 4A-C). This core cluster was subjected to topological analysis to identify genes with central regulatory influence. First, Degree centrality was used to rank the 29 nodes within the cluster (Fig. 4D). Subsequently, the CytoHubba plugin was applied using the Bottleneck algorithm, which prioritizes nodes located on critical shortest-path routes and thus essential for network information flow. Genes that ranked highly by both Degree and Bottleneck metrics were designated as hub genes. Through this integrative network-based approach, five top-scoring hub genes—IFITM3, OAS2, CXCL10, TNFSF10, and XAF1—were identified (Fig. 4E). These genes represent the most influential nodes within the OSCC-associated interaction network and were selected for downstream expression validation and functional assessment.

Fig. 4.

STRING–Cytoscape PPI network, module detection, and hub gene identification. A A PPI network was constructed from 263 overlapping DEGs using the STRING database (Homo sapiens), retaining interactions with a minimum required interaction score of medium confidence (combined score > 0.4). Nodes represent proteins encoded by DEGs, and edges represent STRING-supported interactions derived from multiple evidence sources. B MCODE-based clustering analysis was performed on the entire STRING-derived PPI network using interaction data exported in TSV format. The resulting clusters were visualized using a circular layout to illustrate global network topology and module organization. C The highest-scoring MCODE cluster (MCODE score = 27.357), consisting of 29 nodes and 383 edges, is highlighted as the densest functional module and was selected for downstream hub-gene analysis. D Degree centrality ranking of the 29 nodes within the selected MCODE cluster, reflecting local node connectivity. E CytoHubba analysis was applied to the selected cluster using the Bottleneck algorithm (betweenness-based). This panel displays exclusively the five top-ranked hub genes—IFITM3, OAS2, CXCL10, TNFSF10, and XAF1—identified from the selected MCODE cluster

Meta-analysis identifies CXCL10 as the most strongly overexpressed hub gene in OSCC

To quantitatively compare expression changes of the five hub genes across independent datasets, a random-effects meta-analysis was performed using GSE13601, GSE160042, and GSE227919. The pooled results demonstrated that CXCL10 exhibited the highest and most significant overexpression in OSCC relative to normal tissues (log₂FC = 4.27; 95% CI: 2.02–6.52), followed by XAF1, TNFSF10, OAS2, and IFITM3 (Fig. 5). Among all hub genes, CXCL10 showed the most pronounced and consistently elevated expression across cohorts, supporting its prioritization for subsequent experimental validation and functional characterization.

Fig. 5.

Meta-analysis of hub gene expression across GEO datasets. Forest plots showing pooled log₂FC values with 95% confidence intervals (CI) for the five hub genes across GSE13601, GSE160042, and GSE227919 using a random-effects model. A XAF1. B TNFSF10. C CXCL10. D OAS2. E IFITM3. I² indicates between-study heterogeneity, and w denotes the study weight

CXCL10 is significantly upregulated in OSCC tissues at both the transcript and protein levels

To evaluate the clinical relevance of CXCL10, we first examined its expression using TCGA-HNSC RNA-seq data through the UALCAN portal. CXCL10 transcript levels were markedly elevated in tumor tissues compared with normal tissues (Fig. 6A). Stage stratification showed consistently higher CXCL10 expression across all tumor stages (Fig. 6B). Protein-level validation using the CPTAC HNSC proteomic dataset revealed a similar pattern, with increased CXCL10 abundance in tumor samples relative to normal controls (Fig. 6C). In stage-stratified analyses, CXCL10 protein levels were higher in stage 1–2 tumors, whereas differences in stage 3–4 were not significant, likely reflecting limited sample size and variability in the available CPTAC data for these groups (Fig. 6D). To corroborate these findings in an independent clinical cohort, we performed immunohistochemistry on an OSCC tissue microarray. CXCL10 showed stronger cytoplasmic staining in tumor epithelium than in normal tongue tissue, consistent with its transcriptional and proteomic upregulation (Fig. 6E).

Fig. 6.

CXCL10 expression is elevated in OSCC at transcript, protein, and tissue levels. A TCGA-HNSC RNA-seq data (UALCAN) showing the CXCL10 transcript abundance in tumor versus normal tissues. B CXCL10 expression across clinical stages in HNSC. C CPTAC-HNSC proteomic analysis accessed through UALCAN showing CXCL10 protein expression in tumor tissues (n = 108) relative to normal controls (n = 71). D Stage-stratified CPTAC protein abundance showing CXCL10 expression in normal tissues (n = 71) and stage 1 (n = 7), stage 2 (n = 25), stage 3 (n = 30), and stage 4 (n = 46) HNSC tumors. In UALCAN, CPTAC protein expression is visualized as jitter plots, where values are reported as Z-scores (standard deviations from the median) derived from normalized log₂ spectral count ratios. E Immunohistochemical staining of OSCC tissue microarrays showing CXCL10 staining level in the tumor and normal tongue tissue. CXCL10 staining intensity was quantified using the H-score. Representative images are shown at 40×, 200×, and 400× magnification; scale bars correspond to 200 μm (40×), 50 μm (200×), and 50 μm (400×). * p < 0.05 indicates significant

CXCL10 functionally contributes to migratory activity in OSCC cells

To investigate the functional contribution of CXCL10 to OSCC cell motility, loss- and gain-of-function experiments were performed in HSC3 and SCC4 cells. Transient silencing of CXCL10 using siRNA markedly impaired migratory behavior, as evidenced by a significant reduction in wound closure and decreased numbers of cells traversing Transwell membranes (Fig. 7A–B). In contrast, treatment with recombinant human CXCL10 led to a dose-dependent increase in wound healing and Transwell migration (Fig. 7C–D). These results demonstrate that CXCL10 positively regulates OSCC cell motility in vitro.

Fig. 7.

CXCL10 enhances migratory activity of OSCC cells in vitro. A, B Wound-healing and Transwell migration assays in HSC3 and SCC4 cells following transfection with control siRNA or CXCL10-targeting siRNA (n = 4). C, D Wound-healing and Transwell migration assays performed after treatment with recombinant CXCL10 (0–50 ng/mL) for 24 h (n = 4). * p < 0.05 indicates significant

Discussion

In this study, we integrated multi-cohort OSCC transcriptomics with network topology analysis and in vitro functional assays to nominate CXCL10 as a migration-associated candidate. Across independent datasets, CXCL10 consistently emerged as a top-ranked, highly connected gene within a dense interaction module, motivating downstream validation.

Using public cohorts and an independent OSCC tissue microarray, we observed concordant CXCL10 upregulation at both transcript and protein levels. These observations are consistent with prior HNSCC and oral cancer studies linking CXCL10 to immune-inflamed phenotypes and, in some cohorts, favorable clinical outcomes [26, 40–44], supporting its clinical relevance while highlighting potential context dependence. In tumor-cell monoculture assays, CXCL10 knockdown reduced, whereas recombinant CXCL10 increased, OSCC cell migration, indicating a tumor cell–intrinsic pro-migratory effect. Similar CXCL10/CXCR3-associated motility has been reported in other malignancies through mechanisms such as MMP induction and AKT/PAK1 signaling [11–15, 45, 46].

CXCL10 signals predominantly through CXCR3, and downstream signaling has been reported to engage JAK/STAT, MAPK/ERK, and PI3K/AKT pathways, converging on cytoskeletal remodeling and adhesion dynamics that regulate directional migration [19, 47, 48]. In tumor cells, CXCL10–CXCR3 signaling has been linked to activation of AKT and PAK1 and induction of matrix-remodeling programs, collectively promoting motility and invasion in a context-dependent manner [11, 12, 45]. Consistent with these reports, our enrichment analyses revealed that OSCC-overlapping DEGs were significantly enriched in ECM–receptor interaction, focal adhesion, and PI3K–Akt pathways. In several cancers, CXCL10 co-occurs with ECM-remodeling factors such as MMP9, MMP12, or adhesion molecules [11, 15, 49], suggesting that it may influence motility by modifying adhesion or matrix degradation. However, we did not directly assess CXCR3 isoform usage or downstream pathway activation in OSCC cells. Future studies using CXCR3-specific inhibition and pathway perturbation will be necessary to establish a causal signaling framework for CXCL10-driven migration in OSCC.

A recent preprint by Handel et al. investigated CXCL10 primarily from an immunotherapy perspective in HNSC, demonstrating that intratumoral CXCL10 delivery enhances anti-tumor immunity, promotes CD8⁺ T- and NK-cell recruitment and activation, reduces T-cell exhaustion, inhibits angiogenesis and lymphangiogenesis, and synergizes with anti–PD-1 therapy in murine models [50]. While highly relevant, the scope of that study differs from ours in important ways. Specifically, our work focuses on OSCC and integrates multi-cohort human transcriptomic analyses, network-based hub prioritization, and clinical-scale validation to identify CXCL10 as a consistently overexpressed hub gene across independent patient datasets. Importantly, our functional assays were performed in tumor-cell monocultures, supporting a tumor cell–intrinsic pro-migratory response to CXCL10 in vitro. This finding complements immune-centric studies by suggesting that CXCL10 may exert context- and compartment-dependent effects beyond immune recruitment. In many settings, CXCL10 promotes antitumor immunity by recruiting CXCR3⁺ CD8⁺ T cells, Th1 cells, and NK cells and amplifying interferon-driven immune activation [16, 18, 20, 22, 23, 51–54]. In HNSC, higher CXCL10 expression is frequently associated with immune “hot” phenotypes and improved responses to immune checkpoint blockade [40, 41, 55, 56]. However, clinical associations may vary by compartment and disease context; for example, tumor CXCL10 has been linked to better outcomes in colorectal cancer, whereas circulating CXCL10 may correlate with poorer prognosis [48, 57]. Accordingly, the CXCL10/CXCR3 axis has been described as a “double-edged sword,” with net effects shaped by CXCR3 isoforms, tissue context, and immune composition [19, 49, 50]. Because we did not assess immune infiltration or outcomes within the same OSCC cohort, we cannot determine the net clinical impact of CXCL10 in OSCC. Nevertheless, our in vitro data support the possibility of stage- or niche-dependent roles, consistent with dichotomous functions reported in other tumor settings [22, 49, 58, 59].

This study has several limitations. First, DEG and hub-gene identification relied on public datasets, which may introduce batch effects and clinical heterogeneity, although multi-cohort overlap and meta-analysis likely reduced these biases. Second, functional validation was limited to two OSCC cell lines and focused on migration; invasion, EMT markers, and in vivo metastasis were not assessed. Third, we did not assess CXCR3 isoforms or downstream signaling mediators, which precludes OSCC-specific mechanistic inference. In addition, we did not investigate upstream regulatory mechanisms of CXCL10 expression in OSCC, such as interferon-related signaling or epigenetic regulation. Fourth, although our study describes CXCL10 as a “migration-associated biomarker”, its prognostic value requires OSCC-specific survival validation. Future work should therefore: (1) correlate CXCL10 with CXCR3 isoforms, immune infiltration, and survival in OSCC cohorts; (2) test whether CXCL10-driven migration depends on MMPs, PI3K-Akt, or ERK signaling and define contributions of CXCR3A versus CXCR3B; and (3) evaluate whether CXCL10 predicts responses to radiotherapy, STING agonists, HDAC inhibitors, or immune checkpoint blockade [16, 20, 22, 23, 51–53, 55].

In summary, this study integrates multi-cohort transcriptomics, network topology, clinical validation, and functional assays to identify CXCL10 as a key migration-associated gene in OSCC. We show that CXCL10 is markedly upregulated in OSCC tissues and enhances tumor cell motility in vitro, supporting its potential relevance as a migration-associated biomarker in OSCC. These findings establish a foundation for future investigations into the dual immunologic and motility-related functions of CXCL10 and its potential utility as a diagnostic, migration-associated, or therapeutic biomarker in OSCC.

Conclusion

This study identifies CXCL10 as a consistently overexpressed, migration- associated gene in OSCC through integrated transcriptomic, network, and functional analyses. CXCL10 enhances OSCC cell motility in vitro, supporting its potential role in tumor cell migration. While these findings suggest possible clinical relevance, further in vivo and cohort-based clinical studies are required to determine its prognostic or therapeutic significance in OSCC.

Abbreviations

AID

Activation-induced cytidine deaminase

Akt

Protein kinase B

APOBEC3H

Apolipoprotein B mRNA-editing enzyme catalytic polypeptide-like 3 H

BP

Biological process

CAF

Cancer-associated fibroblast

CC

Cellular component

CCR2

C-C chemokine receptor type 2

CD

Cluster of differentiation

CI

Confidence interval

CPTAC

Clinical Proteomic Tumor Analysis Consortium

CXCL10

C-X-C motif chemokine ligand 10

CXCR3

C-X-C chemokine receptor 3

DAB

3,3’-Diaminobenzidine

DEG(s)

Differentially expressed gene(s)

DMEM

Dulbecco’s Modified Eagle Medium

ECM

Extracellular matrix

EMT

Epithelial–mesenchymal transition

ERK

Extracellular signal–regulated kinase

FDR

False discovery rate

GEO

Gene Expression Omnibus

GO

Gene Ontology

HNSC

Head and neck squamous cell carcinoma

IFN

Interferon

IFITM3

Interferon-induced transmembrane protein 3

IHC

Immunohistochemistry

IL

Interleukin

KEGG

Kyoto Encyclopedia of Genes and Genomes

MAPK

Mitogen-activated protein kinase

MCODE

Molecular Complex Detection algorithm

MF

Molecular Function

MMP(s)

Matrix metalloproteinase(s)

NK

Natural killer

OAS2

2’-5’-Oligoadenylate synthetase 2

OSCC

Oral squamous cell carcinoma

PAK1

p21-activated kinase 1

PPI

Protein–protein interaction

RNA-seq

RNA sequencing

SD

Standard deviation

siRNA

Small interfering RNA

STAT

Signal transducer and activator of transcription

STING

Stimulator of interferon genes

TAM(s)

Tumor-associated macrophage(s)

TCGA

The Cancer Genome Atlas

TME

Tumor microenvironment

TNFSF10

Tumor necrosis factor superfamily member 10 (TRAIL)

UALCAN

University of Alabama at Birmingham Cancer Data Analysis Portal

XAF1

XIAP-associated factor 1

Authors’ contributions

Ying-Sui Sun: Conceptualization, Funding acquisition, Investigation, Methodology, Project administration, Writing - Original Draft; Chi-Jen Chang: Resources, Software, Validation; Tsung-Ming Chang: Formal analysis, Investigation, Methodology, Project administration; Peng Chen: Data Curation, Investigation, Software, Validation; Ju-Fang Liu: Conceptualization, Funding acquisition, Methodology, Supervision, Writing - Review & Editing.

Funding

This work was funded by grants from the Ministry of Science and Technology of Taiwan (NSTC113-2628-B-038-008-MY3) (NSTC112-2221-E-038-006-MY3), and (NSTC112-2811-E-038-004-MY3). Taipei Medical University (DP3-114-62322-04; 114-TMU-NTUST-114-08).

Data availability

All data generated or analyzed in this study were obtained from publicly accessible databases and online bioinformatics platforms without access restrictions. Bulk transcriptomic datasets for OSCC were retrieved from the Gene Expression Omnibus (GEO) repository (GSE13601, GSE160042, and GSE227919; https://www.ncbi.nlm.nih.gov/geo/). All raw and processed files for these series are freely available through GEO. Clinical transcriptomic and proteomic data for HNSC were accessed via the UALCAN portal from the TCGA-HNSC and CPTAC datasets. All external databases and tools used in this work (e.g., GEO, UALCAN, STRING, ShinyGO, Cancer Hallmarks Analytics Tool) are openly available online.

Declarations

Ethics approval and consent to participate

Not applicable. This study used only publicly available, de-identified datasets and commercially obtained anonymized tissue microarrays, and therefore did not require institutional review board approval or individual informed consent.

Consent for publication

Not applicable. This manuscript does not contain any individual person’s data.

Competing interests

The authors declare no competing interests.