Multiomics Reveals NDRG1-Driven Oral Squamous Cell Carcinoma Progression: Competitive Endogenous RNAs Regulation and Proangiogenic Microenvironment Remodelling Authors: Lei Feng, Yi Ou, Linqing Tian, Wenjie Jiang, Min Zhang, Xun Sheng Published: 2026 ABSTRACT Oral squamous cell carcinoma (OSCC) is characterized by high aggressiveness. This study aims to elucidate the role of NDRG1 in the evolutionary heterogeneity and spatial microenvironmental remodelling of OSCC. By integrating bulk transcriptomics, single-cell RNA sequencing (scRNA-seq), and spatial transcriptomics (ST), complemented by in vivo and in vitro functional assays, we systematically explored the regulatory logic of the MSTRG.47889/miR-1299/NDRG1 axis. The MSTRG.47889/miR-1299/NDRG1 ceRNA regulatory axis was identified and validated, demonstrating its significant role in promoting OSCC proliferation and invasion while impairing cellular adhesion. Single-cell analysis revealed a significant expansion of the NDRG1-high subpopulation in tumour tissues, which drives cellular evolution along a pseudotime trajectory toward a partial epithelial-mesenchymal transition (p-EMT) and a high glycolytic state. Spatial transcriptomics analysis revealed that NDRG1 is highly expressed within ‘hypoxia-metabolic’ niches. Our integrative analysis suggests that these regions may coordinate with endothelial cells, highlighting a potential role for the ANGPTL4-CDH5 signalling axis in promoting a proangiogenic microenvironment. These findings provide preliminary insights into how NDRG1 serves as a pivotal regulator driving p-EMT and coordinating niche remodelling. NDRG1 serves as a pivotal regulator driving p-EMT and proangiogenic niche remodelling, representing a potential novel target for the diagnosis and treatment of OSCC. Introduction Oral squamous cell carcinoma (OSCC) is the most common malignancy of the oral and maxillofacial region, accounting for over 90% of all oral cancers.1 Globally, OSCC accounts for 2% of new cancer cases and 1.9% of cancer-related deaths.2 Despite advances in diagnosis and treatment, OSCC still faces challenges such as high invasiveness, high rates of lymph node metastasis, and low 5-year survival rates; recurrence, metastasis, and lymphatic invasion are critical factors influencing the prognosis of OSCC patients.3,4 Therefore, a thorough exploration of its molecular pathogenesis is of paramount importance. Epithelial-mesenchymal transition (EMT) confers highly aggressive invasive capabilities to tumour cells.5,6 Recent studies have identified SOX11 and GLIS2 as novel biomarkers that drive the progression of oral leucoplakia to OSCC by regulating EMT and angiogenesis. As a member of the NDRG family.7 As a member of the NDRG family,8 N-myc downstream-regulated gene 1 (NDRG1) exhibits a tissue-specific ‘double-edged sword’ effect across various cancers. While regarded as a metastasis suppressor in gastric cancer,9 its high expression is closely associated with poor prognosis in cervical adenocarcinoma, breast cancer, and liver cancer.10,11 However, significant cellular heterogeneity exists within OSCC, and traditional in vitro experiments are insufficient to elucidate the actual role of NDRG1 within complex tissue spaces and dynamic evolution.12,13 Noncoding RNAs play a regulatory role in the progression of OSCC.14 Long noncoding RNAs (lncRNAs) can function as competitive endogenous RNAs (ceRNAs) to regulate target gene expression.15,16 Similarly, lncRNAs such as AFAP1-AS1 have been shown to promote OSCC malignancy through ubiquitin-mediated proteolysis and ceRNA-related pathways.17 MiR-1299 has been demonstrated to possess tumour-suppressive potential in various malignancies.18 However, in OSCC, how the MSTRG.47889/miR-1299/NDRG1 axis drives the p-EMT program19 and induces communication between endothelial cells and proangiogenic factors (such as ANGPTL4/VEGF) – thereby constructing a ‘hypoxia-induced proinvasive niche’ that facilitates metastasis – remains a critical scientific question to be elucidated.20 This study aims to systematically characterize the role of NDRG1 in OSCC by integrating bulk transcriptomics, single-cell RNA sequencing (scRNA-seq), and spatial transcriptomics (ST). We will identify the MSTRG.47889/miR-1299/NDRG1 axis and specifically reveal how this axis drives tumour progression through the remodelling of the spatial niche, providing a novel molecular basis for the precision treatment of OSCC. Materials and methods Specimen collection and tissue processing Twenty-nine pairs of OSCC and matched adjacent normal tissues (2 cm from the margin) were obtained from patients at the Affiliated Stomatological Hospital of Kunming Medical University and confirmed by pathological examination. The study protocol was approved by the Ethics Committee of the Affiliated Stomatological Hospital of Kunming Medical University (KYKQ2025MEC0034), and informed consent was obtained from all participants. Cell culture and model selection OSCC cell lines (CAL27, SCC-15, TSCCA) and human oral keratinocytes (HOK) were purchased from ZQXZBIO and cultured in DMEM supplemented with 10% FBS or specialized keratinocyte media, respectively. CAL27 was selected for subsequent functional assays and intervention experiments due to its significant differential expression Yof NDRG1 and MSTRG.47889 compared to HOK cells (Supplementary Figure 1). Plasmid construction and lentiviral infection NDRG1 and MSTRG.47889 knockdown (shRNA) and overexpression (ORF) vectors were constructed by Abiowell and packaged into lentiviruses to infect CAL27 cells. Stable cell lines were established through gradient selection with puromycin (Supplementary Figure 2). Bulk transcriptome sequencing analysis Total RNA was extracted for library construction and sequenced on the Illumina Novaseq 6000 platform using a PE150 strategy. Data were aligned using HISAT2, assembled via StringTie, and differentially expressed genes (DEGs) were identified using edgeR based on FPKM values. Single-cell RNA-seq data processing OSCC scRNA-seq datasets (GSE181919) were analysed using the Seurat R package for normalization, dimensionality reduction (PCA/UMAP), and subclustering of malignant epithelial cells. Functional scores for proliferation(MKI67,PCNA), glycolysis(GAPDH,LDHA), and mesenchymal signatures were calculated using the AddModuleScore function. Spatial transcriptomics and pseudotime analysis 10X Visium spatial transcriptomics data (GSM6339631) were normalized via SCTransform, and NDRG1 distribution was visualized using SpatialFeaturePlot. Evolutionary trajectories of malignant subpopulations were inferred using Monocle3, with normal epithelial cells defined as the root to calculate pseudotime and visualize dynamic gene expression. Intercellular communication analysis (CellChat) Intercellular communication between NDRG1-high/low subpopulations and vascular endothelial cells was analysed using the CellChat R package. Communication probabilities of ligand-receptor pairs were calculated via the computeCommunProb function based on the CellChatDB database. Signalling outputs of the VEGF and ANGPTL pathways were compared between NDRG1-high and NDRG1-low subpopulations using netVisual_bubble to evaluate communication intensity with endothelial cells. Cellular phenotype assays Cell proliferation, migration, invasion, and wound healing capacity were assessed using CCK-8, Transwell, and scratch assays, respectively. For adhesion assays, 2000 cells per well were seeded in 96-well plates; nonadherent cells were removed after 1 hour, and adherence was quantified via CCK-8. Molecular validation (WB, RT-qPCR, FISH) For Western blotting, proteins were extracted, separated by SDS-PAGE, and transferred onto PVDF membranes for incubation with primary antibodies and ECL-based visualization. Relative expression of NDRG1, MSTRG.47889, and miR-1299 was quantified using the 2^-ΔΔCt method, with primer sequences provided in Supplementary Table 1. For FISH, samples were fixed in 4% PFA, digested with Proteinase K, and hybridized overnight at 42 °C with the probe: 5’-DIG-AATGAGAAGAAAGGGAAGAGACTGGAAAGAAC-DIG-3’. Signal amplification was performed using AntiDIG HRP and Cy3 TSA, followed by DAPI counterstaining and fluorescence microscopy imaging. For dual-luciferase assays, wild-type or mutant 3′UTR sequences were cloned into pMIR-REPORT and cotransfected with miRNA mimics into HEK293 cells for relative luciferase activity measurement. Dual luciferase assay Primers containing specific restriction enzyme sites were designed to amplify the 3′UTR sequence of the miRNA target gene and its mutated version via PCR. Detailed procedures for mutation and amplification are provided in Supplementary Figure 3. The components of the enzyme digestion and ligation systems are listed in Supplementary Tables 2 and 3. Recombinant plasmids were transformed into E. coli, and positive clones were selected using ampicillin resistance and validated by sequencing. The primary components and structure of the recombinant Luc-3′UTR plasmid are shown in Supplementary Figure 4. HEK293 cells were transfected with recombinant plasmids using Lipofectamine 2000, and luciferase activity was measured after 48 hours. The interaction between the miRNA and the target 3′UTR sequence was assessed by comparing relative luciferase activities. The specific binding sites of miR-1299 with NDRG1 and lncRNA MSTRG.47889 are illustrated in Supplementary Figure 5. LncRNA and miRNA prediction and selection Transcripts were assembled with StringTie, and differentially expressed lncRNAs were identified using CPC and CNCI to evaluate coding potential. Potential miRNAs interacting with NDRG1 and MSTRG.47889 were predicted using Miranda software with a binding score threshold of 160. Immunohistochemistry For IHC, deparaffinized sections underwent antigen retrieval and overnight primary antibody incubation at 4 °C, followed by DAB detection and haematoxylin counterstaining. In vivo validation Twenty-four male nude mice (Balb/c-nu, 6 weeks old, weighing 15-20 g) were obtained from the Experimental Animal Department of Kunming Medical University (Kunming, China). The mice were maintained in a specific-pathogen-free (SPF) facility. For tumour implantation, mice were anaesthetized via intraperitoneal injection of 1% pentobarbital sodium (50 mg/kg) to minimize pain and distress. Then, 4 × 106 stably transfected CAL27 cells were subcutaneously injected into the axilla (Supplementary Figure 6). Tumour diameters were measured every 2 days using callipers, and volumes (V) were calculated as: V = a × b2 × 0.5 (where ‘a’ and ‘b’ represent length and width, respectively). On day 21, mice were euthanized by CO2 inhalation (30% volume displacement rate per minute) followed by cervical dislocation to ensure the animals were unconscious throughout the procedure and death was confirmed. Tumours were harvested for weighing and subsequent IHC and Western blot analysis. The experimental protocol was approved by the Experimental Animal Ethics Committee of Kunming Medical University (KMMUX2025020572). Statistical analysis Statistical analyses were performed using SPSS 22.0 and GraphPad Prism 8.0, employing t-tests for two-group comparisons and one-way ANOVA for multiple groups. All experiments were independently repeated at least three times, and P < .05 was considered statistically significant. Results Transcriptomic screening and validation of NDRG1 as a key pathogenic factor in OSCC To identify key drivers of OSCC initiation and progression, we first performed bulk transcriptomic sequencing analysis on OSCC tissues and matched adjacent normal tissues (n = 3). The clustering heat map revealed differentially expressed genes (DEGs) between tumour and adjacent tissues (Figure 1A), which were significantly enriched in cell adhesion-related pathways (Figure 1B, C). Among the nine core DEGs associated with cell adhesion, eight genes excluding NCAM1 showed significant differences, with NDRG1 exhibiting the most prominent up-regulation (Figure 1D), a pattern highly consistent with findings from The Cancer Genome Atlas (TCGA) database (Figure 1E). Subsequently, we evaluated NDRG1 expression levels in an independent validation cohort of 29 pairs of OSCC clinical samples (clinical data provided in Supplementary Table 4).Fig. 1Expression profiles of NDRG1 and identification of the MSTRG.47889/miR-1299/NDRG1 axis in OSCC. (A) Hierarchical clustering heatmap of DEGs in OSCC and normal tissues. (B, C) GO and KEGG enrichment analysis of DEGs. (D) mRNA levels of cell adhesion-related genes. (E) NDRG1 mRNA expression in TCGA-OSCC cohort. (F, G) Western blot, HE, and IHC analysis of NDRG1 in clinical specimens. (H, I) Differentially expressed lncRNAs and lncRNA-target gene interaction network. (J, K) RT-qPCR and correlation analysis of NDRG1 and MSTRG.47889 in clinical samples. (L) FISH assay of MSTRG.47889 localization in CAL27 cells. (M, N) Candidate miRNA prediction and RT-qPCR analysis in clinical samples. (O) Dual-luciferase reporter assay for miR-1299 binding to NDRG1 and MSTRG.47889. (P) Western blot of NDRG1 protein expression following modulation of the MSTRG.47889/miR-1299 axis. Statistics: Data are presented as mean±SD (n = 3 independent experiments). Statistical significance was determined by two-tailed Student’s t-test or one-way ANOVA. *P < .05, P < .01, ***P < .001.Fig 1 dummy alt text Expression profiles of NDRG1 and identification of the MSTRG.47889/miR-1299/NDRG1 axis in OSCC. (A) Hierarchical clustering heatmap of DEGs in OSCC and normal tissues. (B, C) GO and KEGG enrichment analysis of DEGs. (D) mRNA levels of cell adhesion-related genes. (E) NDRG1 mRNA expression in TCGA-OSCC cohort. (F, G) Western blot, HE, and IHC analysis of NDRG1 in clinical specimens. (H, I) Differentially expressed lncRNAs and lncRNA-target gene interaction network. (J, K) RT-qPCR and correlation analysis of NDRG1 and MSTRG.47889 in clinical samples. (L) FISH assay of MSTRG.47889 localization in CAL27 cells. (M, N) Candidate miRNA prediction and RT-qPCR analysis in clinical samples. (O) Dual-luciferase reporter assay for miR-1299 binding to NDRG1 and MSTRG.47889. (P) Western blot of NDRG1 protein expression following modulation of the MSTRG.47889/miR-1299 axis. Statistics: Data are presented as mean±SD (n = 3 independent experiments). Statistical significance was determined by two-tailed Student’s t-test or one-way ANOVA. *P < .05, P < .01, ***P < .001. Consistent results from Western blot and immunohistochemistry assays demonstrated that NDRG1 protein levels were significantly elevated in OSCC tissues compared to adjacent tissues (Figure 1F), characterized by strong positive plaque-like distribution within tumour nests (Figure 1G). The MSTRG.47889/miR-1299 axis mediates the high expression of NDRG1 in OSCC To elucidate the upstream regulatory mechanism underlying NDRG1 overexpression, we predicted and screened for potential noncoding RNA regulators (Figure 1H-I). A novel long noncoding RNA, designated as MSTRG.47889, was identified on chromosome 8 (chr8: 133320796-133323239). The expression of lncRNA MSTRG.47889 was significantly upregulated in OSCC tissues and exhibited a significant positive correlation with NDRG1 levels (Figure 1J-K). Fluorescence in situ hybridization (FISH) assays confirmed that MSTRG.47889 is predominantly localized within the cytoplasm of OSCC cells (Figure 1L), suggesting a physical prerequisite for its function as a competitive endogenous RNA (ceRNA). Further computational analysis using miRanda software predicted miR-1299 as a common target for both MSTRG.47889 and NDRG1 (Figure 1M). MiR-1299 was significantly downregulated in OSCC (Figure 1N), displaying an expression profile inverse to that of NDRG1. Dual-luciferase reporter assays validated that miR-1299 directly binds to the predicted target sites on both MSTRG.47889 and NDRG1 (Figure 1O). Subsequently, Western blot (WB) assays in CAL27 cells – the cell line exhibiting the most pronounced NDRG1 expression variance – revealed that NDRG1 protein levels decreased upon miR-1299 overexpression and increased following its inhibition (Figure 1P); conversely, NDRG1 protein levels significantly rose with MSTRG.47889 overexpression and fell with its knockdown. Taken together, these results demonstrate the successful identification and validation of the MSTRG.47889/miR-1299/NDRG1 ceRNA regulatory axis. The MSTRG.47889/miR-1299/NDRG1 axis promotes OSCC proliferation and invasion while impairing cell adhesion Following the establishment of the regulatory axis, we investigated its impact on the biological behaviours of OSCC. Functional assays demonstrated that overexpression of either MSTRG.47889 or NDRG1 significantly enhanced the proliferation, migration, and invasive capacities of CAL27 cells (Figure 2A-D). Given that previous enrichment analyses suggested a close association between NDRG1 and cell adhesion, adhesion assays confirmed that activation of this axis significantly reduced homotypic cellular adhesion (Figure 2E), potentially providing a physical basis for cancer cell detachment and infiltration. Further rescue experiments revealed that overexpression of miR-1299 mimics effectively reversed the enhanced proliferation and invasion induced by NDRG1 while restoring cellular adhesion capacity (Figure 2F-J). Concurrently, the phenotypic inhibition resulting from MSTRG.47889 knockdown was offset by NDRG1 overexpression (Figure 2F), confirming the functional dependency of this axis. Finally, subcutaneous xenograft experiments in nude mice corroborated that NDRG1 overexpression significantly accelerated tumour growth in vivo (Figure 3K-L).Fig. 2Impact of the MSTRG.47889/miR-1299/NDRG1 axis on OSCC malignant phenotypes and in vivo tumour growth. (A, B) Transwell migration and invasion assays in CAL27 cells following modulation of NDRG1 or MSTRG.47889 expression. (C, D) CCK-8 proliferation and wound healing assays (24 h and 48 h) across indicated experimental groups. (E) Homotypic cell adhesion assays evaluating cellular attachment across different groups. (F) Western blot analysis of EMT-related markers and NDRG1 protein levels in rescue experiments. (G-J) Functional rescue assays, including Transwell, wound healing, cell adhesion, and CCK-8 analysis, evaluating the interaction between miR-1299 and NDRG1. (K) Representative macroscopic morphology of excised subcutaneous xenograft tumours at the experimental endpoint (day 21). (L) Statistical analysis of mouse body weight, tumour growth curves, and final tumour weights. For in vitro experiments, data are expressed as mean±SD from three independent biological replicates (n = 3). For in vivo experiments, n = 6 mice per group. Statistical significance was assessed using two-tailed Student’s t-test or one-way ANOVA. *P < .05, P < .01, ***P < .001.Fig 2 dummy alt textFig. 3Single-cell transcriptomic analysis of NDRG1 heterogeneity and evolutionary trajectory. (A) UMAP visualization of major cell populations in OSCC tissues. (B) UMAP and violin plots illustrating NDRG1 expression across identified cell types. (C) Subclustering of malignant cells and comparison of NDRG1-high cluster proportions (Clusters 0, 2, and 6) in primary (CA) versus metastatic (LN) tissues. (D) Comparison of glycolysis and metabolism signature scores between NDRG1-high and NDRG1-low malignant cells. (E) Heatmap showing genes coexpressed with NDRG1. (F, G) Monocle 3 pseudotime trajectory and correlation analysis of NDRG1 with invasion-related markers (S100A8, S100A9, CDH1, and LAMC2). (H) Western blot and quantitative analysis of EMT markers in CAL27 cells following NDRG1 knockdown or overexpression. For Western blot analysis, data are presented as mean ± SD from three independent biological replicates (n = 3). P values were calculated using two-tailed Student’s t-test or one-way ANOVA as indicated. Significance levels: *P < .05, P < .01, ***P < .001.Fig 3 dummy alt text Impact of the MSTRG.47889/miR-1299/NDRG1 axis on OSCC malignant phenotypes and in vivo tumour growth. (A, B) Transwell migration and invasion assays in CAL27 cells following modulation of NDRG1 or MSTRG.47889 expression. (C, D) CCK-8 proliferation and wound healing assays (24 h and 48 h) across indicated experimental groups. (E) Homotypic cell adhesion assays evaluating cellular attachment across different groups. (F) Western blot analysis of EMT-related markers and NDRG1 protein levels in rescue experiments. (G-J) Functional rescue assays, including Transwell, wound healing, cell adhesion, and CCK-8 analysis, evaluating the interaction between miR-1299 and NDRG1. (K) Representative macroscopic morphology of excised subcutaneous xenograft tumours at the experimental endpoint (day 21). (L) Statistical analysis of mouse body weight, tumour growth curves, and final tumour weights. For in vitro experiments, data are expressed as mean±SD from three independent biological replicates (n = 3). For in vivo experiments, n = 6 mice per group. Statistical significance was assessed using two-tailed Student’s t-test or one-way ANOVA. *P < .05, P < .01, ***P < .001. Single-cell transcriptomic analysis of NDRG1 heterogeneity and evolutionary trajectory. (A) UMAP visualization of major cell populations in OSCC tissues. (B) UMAP and violin plots illustrating NDRG1 expression across identified cell types. (C) Subclustering of malignant cells and comparison of NDRG1-high cluster proportions (Clusters 0, 2, and 6) in primary (CA) versus metastatic (LN) tissues. (D) Comparison of glycolysis and metabolism signature scores between NDRG1-high and NDRG1-low malignant cells. (E) Heatmap showing genes coexpressed with NDRG1. (F, G) Monocle 3 pseudotime trajectory and correlation analysis of NDRG1 with invasion-related markers (S100A8, S100A9, CDH1, and LAMC2). (H) Western blot and quantitative analysis of EMT markers in CAL27 cells following NDRG1 knockdown or overexpression. For Western blot analysis, data are presented as mean ± SD from three independent biological replicates (n = 3). P values were calculated using two-tailed Student’s t-test or one-way ANOVA as indicated. Significance levels: *P < .05, P < .01, ***P < .001. Single-cell transcriptomics reveals the expansion and evolutionary trajectory of the NDRG1-high subpopulation To characterize the heterogeneity of NDRG1 at the single-cell level, we performed single-cell RNA sequencing (scRNA-seq) on OSCC tissues. Dimensionality reduction and clustering analysis identified 10 major cell populations, including epithelial cells, T cells, and fibroblasts (Figure 3A), with NDRG1 transcriptional signals predominantly concentrated in malignant epithelial cells (Figure 3B); further subclustering revealed that a malignant subpopulation specifically overexpressing NDRG1 (Clusters 0, 2, 6) underwent significant expansion in metastatic OSCC tissues compared to primary lesions (Figure 3C). Functional characterization demonstrated that this subpopulation exhibited a high glycolysis signature score (Figure 3D). Through single-cell coexpression analysis (Figure 3D), we found that NDRG1 was highly coexpressed with genes such as ANGPTL4, CA9, and PGF (Figure 3E). Pseudotime trajectory analysis (Monocle 3) revealed a continuous evolutionary path from normal epithelium to a highly invasive malignant state. Along this timeline, NDRG1 expression increased and was synchronized with the up-regulation of markers such as LAMC2 and S100A8/9. In this study, partial EMT (p-EMT) is characterized as a hybrid state where cells retain certain epithelial identities while simultaneously acquiring invasive mesenchymal features. The synchronization of NDRG1 with markers like LAMC2 – which is often associated with collective rather than single-cell motility – suggests that NDRG1 may facilitate a p-EMT program in OSCC (Figure 3F-G). This trend was further evaluated at the protein level: Western blot results indicated that modulating NDRG1 expression led to the downregulation of E-cadherin and the upregulation of Vimentin and Snail. These findings suggest that NDRG1 acts as a pivotal regulator in the EMT-related progression of OSCC. However, to more clearly distinguish p-EMT from conventional EMT, further experimental validation, such as high-resolution spatial colocalization of hybrid markers, will be pursued in our future studies to confirm these phenotypic observations (Figure 3H). Spatial transcriptomics and in vivo experiments reveal the ‘hypoxia-invasive’ niche and the mechanism of vascular recruitment. Spatial transcriptomics (ST) analysis revealed that NDRG1 exhibited plaque-like high expression in situ, which highly overlapped with hypoxia scores and glycolytic levels (Figure 4A). Statistical analysis confirmed that the degree of microenvironmental malignancy (hypoxia and metabolic scores) in NDRG1-high regions was significantly higher than in low-expression regions (Figure 4B). To provide rigorous quantitative evidence for the observed spatial colocalization, we performed a Monte Carlo permutation-based enrichment analysis (1000 iterations). By comparing the observed overlap with a stochastically generated null distribution, we confirmed that the enrichment of NDRG1 within the malignant metabolic niche was statistically nonrandom (P < .001). This spatial profiling quantifies the preference of NDRG1 for localized areas of high metabolic stress, rather than representing a stochastic distribution across the tumour tissue (Figure 4C). Further analysis revealed that proinflammatory signals were also significantly enhanced in the NDRG1-high subpopulation (Figure 4D), suggesting that NDRG1 may participate in constructing a complex protumour inflammatory niche.Fig. 4Spatial transcriptomic and in vivo characterization of the NDRG1-associated niche and potential vascular interactions. (A) In situ spatial distribution maps illustrating the spatial coupling of NDRG1 expression with glycolysis, hypoxia (HIF1A/CA9), and the invasion front in OSCC tissues. (B) Quantitative comparison of hypoxia and glycolytic metabolism signature scores between NDRG1-high and NDRG1-low spots. (C) Permutation-based spatial enrichment analysis (1000 iterations) demonstrating the nonrandom colocalization of NDRG1 with a composite Malignant Niche Score (Hypoxia + Metabolism). The red dashed line indicates the actual observed overlap compared to the null distribution (P < .001). (D-F) IHC, RT-qPCR, and Western blot analysis of EMT-related markers in harvested xenograft tumour tissues following modulation of NDRG1 expression. (G) Assessment of proangiogenic niche potential in NDRG1-high versus NDRG1-low regions. (H) CellChat bubble plot showing predicted ligand-receptor interactions between malignant epithelial cells and endothelial cells, including the VEGFA, ANGPTL4, and PGF axes. For in vitro and molecular assays, data represent mean ± SD from three independent biological replicates (n = 3). Statistical significance for spot-level spatial analysis was determined by Wilcoxon rank-sum test or Monte Carlo permutation test; molecular comparisons were performed using two-tailed Student’s t-test or one-way ANOVA. Significance levels: *P < .05, P < .01, ***P < .001.Fig 4 dummy alt text Spatial transcriptomic and in vivo characterization of the NDRG1-associated niche and potential vascular interactions. (A) In situ spatial distribution maps illustrating the spatial coupling of NDRG1 expression with glycolysis, hypoxia (HIF1A/CA9), and the invasion front in OSCC tissues. (B) Quantitative comparison of hypoxia and glycolytic metabolism signature scores between NDRG1-high and NDRG1-low spots. (C) Permutation-based spatial enrichment analysis (1000 iterations) demonstrating the nonrandom colocalization of NDRG1 with a composite Malignant Niche Score (Hypoxia + Metabolism). The red dashed line indicates the actual observed overlap compared to the null distribution (P < .001). (D-F) IHC, RT-qPCR, and Western blot analysis of EMT-related markers in harvested xenograft tumour tissues following modulation of NDRG1 expression. (G) Assessment of proangiogenic niche potential in NDRG1-high versus NDRG1-low regions. (H) CellChat bubble plot showing predicted ligand-receptor interactions between malignant epithelial cells and endothelial cells, including the VEGFA, ANGPTL4, and PGF axes. For in vitro and molecular assays, data represent mean ± SD from three independent biological replicates (n = 3). Statistical significance for spot-level spatial analysis was determined by Wilcoxon rank-sum test or Monte Carlo permutation test; molecular comparisons were performed using two-tailed Student’s t-test or one-way ANOVA. Significance levels: *P < .05, P < .01, ***P < .001. To validate the impact of this malignant spatial niche on tumour progression, we conducted functional validation in a xenograft model. Consistent results from IHC, Western blot (WB) of animal tissues, and PCR demonstrated that the up-regulation of NDRG1 significantly induced the expression of Vimentin and N-cadherin while leading to the loss of E-cadherin in tumour tissues in vivo (Figure 4E-G), confirming that cells under such ‘hypoxia-inflammatory’ stress indeed acquire enhanced invasive capabilities through the activation of the EMT program. Furthermore, ST functional module scoring indicated that NDRG1-high regions exhibited greater proangiogenic niche potential (Figure 4H). Integrated with CellChat communication analysis, we observed that the NDRG1-high subpopulation exhibited high interaction probabilities with endothelial cells via the ANGPTL4-CDH5 signalling axis through the predicted secretion of factors such as ANGPTL4, VEGFA, and PGF (Figure 4I). In summary, NDRG1 not only drives the EMT transformation of cancer cells themselves but also actively recruits blood vessels and intensifies invasion by remodelling the ‘hypoxia-inflammation-metabolic’ niche, collectively promoting the malignant progression of OSCC. Discussion NDRG1 is frequently regarded as a ‘double-edged sword’ molecule in the study of malignant tumours.21 Its specific functional phenotype often depends on tissue specificity, subcellular localization, and complex microenvironmental pressures. Although early studies in colon cancer, pancreatic cancer, and glioma identified NDRG1 as a metastasis suppressor in several cancers, including pancreatic cancer and glioblastoma.22,23 Notwithstanding its primary role as a tumour suppressor in certain contexts, research has also demonstrated its oncogenic potential in cervical adenocarcinoma, breast cancer, and liver cancer.10,11,24,25 Through the screening and validation of OSCC clinical samples, the present study establishes that NDRG1 functions as a central oncogenic driver in OSCC. Sequencing results (Figure 1B, C) revealed that NDRG1 is one of the most significantly upregulated genes within the cell adhesion pathway, a trend corroborated by both clinical samples and the TCGA database. Notably, immunohistochemistry (IHC) assays revealed that NDRG1 exhibits a ‘plaque-like’ distribution pattern within tumour tissues (Figure 1G). This positive finding is highly significant; it suggests that the in situ expression of NDRG1 is not homogeneous but rather specifically enriched within distinct spatial regions. This observation echoes the ‘malignant subpopulation expansion’ identified through single-cell sequencing. In in vivo experiments, we observed that NDRG1 overexpression significantly increased tumour volume; however, regarding tumour weight, although an upward trend was noted, it did not reach a level of high statistical significance (Figure 2K). This may imply that NDRG1 primarily facilitates progression by altering the invasive patterns of cancer cells or maintaining cell survival under hypoxic conditions, rather than simply increasing total biomass by accelerating the cell cycle. This conclusion prompted us to further elucidate its role in microenvironmental remodelling from both single-cell and spatial dimensions. Noncoding RNA (ncRNA) regulatory networks represent a pivotal epigenetic mechanism in the malignant progression of OSCC. In this study, we successfully identified and validated the MSTRG.47889/miR-1299/NDRG1 ceRNA regulatory axis. Long noncoding RNAs (lncRNAs) typically function as ‘miRNA sponges’ by competitively sequestering miRNAs, thereby alleviating their post-transcriptional inhibition of target mRNAs.26 Our FISH assays confirmed that MSTRG.47889 is predominantly localized in the cytoplasm, providing a physical prerequisite for its function as a ceRNA (Figure 1L). Subsequent dual-luciferase reporter assays directly pinpointed the binding sites between miR-1299 and both MSTRG.47889 and NDRG1. Existing research indicates that miR-1299 suppresses EMT and invasion in various tumours by targeting key regulatory genes.18 Our study observed that MSTRG.47889 overexpression positively regulates NDRG1, significantly activating key EMT markers and promoting tumour cell migration, invasion, and phenotypic transition. Conversely, MSTRG.47889 knockdown reversed these effects. We observed in rescue experiments that miR-1299 mimics significantly reversed the NDRG1-induced enhancement of proliferation and migration (Figure 3F-J). Compared with previous studies, this research is the first to demonstrate that MSTRG.47889 serves as a key regulator for NDRG1 in OSCC. This finding not only elucidates the mechanism of abnormal NDRG1 accumulation in OSCC but also provides a theoretical foundation for the development of future RNA-based targeted therapies. Single-cell transcriptomic analysis provides a high-resolution perspective for resolving the heterogeneous expression of NDRG1 within OSCC. Figure 3A-B shows that NDRG1 transcriptional signals are not uniformly distributed but are highly and specifically concentrated in malignant epithelial cells. More importantly, subclustering analysis revealed that a specific malignant subpopulation overexpressing NDRG1 underwent significant expansion in tumour tissues compared to normal mucosa (Figure 3C). These cells not only exhibited enhanced proliferative potential but also possessed significantly higher glycolytic metabolism scores (Figure 3D). This phenomenon of ‘elevated mesenchymal scores with retained epithelial identity’ suggests that NDRG1 drives a more aggressive state of partial epithelial-mesenchymal transition (p-EMT) rather than the traditional full EMT process. In landmark studies of head and neck squamous cell carcinoma (HNSCC), p-EMT has been identified as the strongest predictor of lymph node metastasis and perineural invasion.19 The pseudotime trajectory analysis in this study (Figure 3F) further corroborates this perspective: as malignant cells evolve along the trajectory toward an invasive endpoint, NDRG1 expression increases linearly, highly synchronized with the upregulation of invasion-associated markers such as LAMC2, S100A8, and S100A9. LAMC2 is a pivotal molecule involved in ‘collective invasion’.27,28 This provides a potential explanation for why cells with high NDRG1 expression, despite having reduced adhesion, tend to infiltrate outward in the form of cell clusters. Spatial transcriptomics (ST) compensates for the loss of positional information inherent in single-cell sequencing.13 In the present study, we observed that NDRG1 exhibits a distinct plaque-like distribution in situ, and these plaques demonstrate high spatial coupling with hypoxia scores and metabolism scores within the tissue (Figure 4A). This ‘trinity’ of spatial distribution reveals a specific survival strategy within OSCC: the ‘hypoxia-induced proinvasive niche’. Statistical analysis confirmed that the degree of microenvironmental malignancy in NDRG1-high regions is significantly higher than in NDRG1-low regions (Figure 4B). This suggests that NDRG1 may be directly driven by hypoxia-inducible factor 1-alpha (HIF-1α). Within tumour nests where oxygen supply is deficient, NDRG1 provides essential energy compensation for malignant cells by upregulating glycolysis, enabling them to survive under extreme stress and maintain an invasive phenotype. Remodelling of the tumour microenvironment (TME) is a hallmark of malignant progression,29,30 in which angiogenesis not only provides nutrients for tumour growth but also serves as a physical conduit for distant metastasis. In this study, CellChat-based communication analysis revealed that the NDRG1-high malignant subpopulation exhibits a more robust capacity for proangiogenic signalling output compared to the low-expression subpopulation (Figure 4G, H). Specifically, NDRG1-high cells exhibit robust interaction potential with receptors such as CDH5 and VEGFR2 on the surface of endothelial cells by secreting proangiogenic factors, including ANGPTL4, VEGFA, and PGF. The pathogenic role of ANGPTL4 in the metastasis of oral squamous cell carcinoma (OSCC) has been preliminarily established. Previous research indicates that the expression level of ANGPTL4 in OSCC tissues is significantly and positively correlated with metastasis, and secreted ANGPTL4 has been shown to markedly enhance the in vitro invasive potential of OSCC cells.31 The findings of this study not only support this conclusion but also further pinpoint the cellular source of ANGPTL4 – specifically, a distinct malignant subpopulation driven by NDRG1 – at single-cell resolution. This implies that NDRG1 not only regulates intrinsic cellular invasiveness but also confers tumour cells with the capacity to remodel the surrounding vascular microenvironment by upregulating the secretory factor ANGPTL4. Furthermore, this study provides novel insights into the potential crosstalk by which this regulatory axis may modulate vascular endothelial integrity. Our CellChat analysis suggests that ANGPTL4 secreted by NDRG1-high cells is predicted to act on CDH5 receptors on the surface of endothelial cells. Rather than merely promoting an increase in vascular density, the predicted activation of this axis may facilitate the dissociation of VE-cadherin (CDH5) and claudin-5 protein clusters between endothelial cells via integrin signaling.32 Notably, some studies have previously reported that NDRG1 exerts antiangiogenic effects in pancreatic and colon cancers.22 This polarity shift in function highlights the organ-specificity inherent in the biological roles of NDRG1. This polarity shift in function highlights the organ-specificity inherent in the biological roles of NDRG1. In OSCC – a squamous cell carcinoma characterized by high blood supply dependence and frequent extensive lymph node metastasis – NDRG1 appears to function as a proangiogenic driver. This ANGPTL4-induced disruption of vascular junctions leads to a pronounced ‘leakiness’ in the tumour microvasculature. Integrated with the p-EMT evolutionary trend identified in our pseudotime analysis, we can infer that NDRG1-driven malignant cells gain migratory momentum through p-EMT while simultaneously weakening the barrier function of surrounding microvessels via ANGPTL4 secretion. This synergy between intrinsic and extrinsic factors provides a physical ‘conduit’ for cancer cells at the invasive front to enter the circulatory system and undergo distant lymph node metastasis. The suggested spatial coupling between the MSTRG.47889/miR-1299/NDRG1 axis and the ‘hypoxia-metabolic’ niche offers several hypothetical avenues for the development of translational therapeutic strategies in OSCC. As MSTRG.47889 is predicted to function as a cytoplasmic ceRNA that stabilizes NDRG1 expression, future RNA-targeted interventions – such as antisense oligonucleotides (ASOs) or siRNA-mediated silencing – could potentially be explored to attenuate the associated p-EMT program and invasive phenotypes. Furthermore, our bioinformatic models suggest that NDRG1-high clusters may coordinate with endothelial cells through the ANGPTL4-CDH5 axis. Consequently, investigating the pharmacological disruption of this signalling pathway might represent a strategy to address the formation of a prometastatic environment. Conceptually, integrating these axis-specific inhibitors with conventional chemotherapy or radiotherapy might help sensitize malignant cells within stress-resilient niches, providing a multifaceted framework for future efforts to intercept OSCC progression. While our multiomics integration provides a robust high-resolution map of the OSCC microenvironment, the specific functional contribution of the ANGPTL4-CDH5 axis to vascular recruitment remains to be further validated through direct cell-cell interaction assays. These preliminary findings establish a critical foundation for our upcoming research into targeted disruption of the OSCC metastatic niche. Conclusion This study demonstrates that NDRG1 is a central driver of malignancy in oral squamous cell carcinoma (OSCC) through the following mechanisms: The MSTRG.47889/miR-1299/NDRG1 axis: This newly identified ceRNA network promotes tumour proliferation and invasion while significantly impairing cell-to-cell adhesion. Malignant Evolution: Single-cell analysis confirms that NDRG1-high subpopulations expand during metastasis, driving cells toward a partial EMT (p-EMT) and a high glycolytic state. Niche remodelling: Spatial transcriptomics identifies that NDRG1-rich ‘hypoxia-metabolic’ niches are associated with potential vascular coordination, likely facilitated by the predicted activation of the ANGPTL4-CDH5 signalling axis. In summary, NDRG1 serves as a pivotal regulator that couples epigenetic regulation with spatial microenvironmental remodelling. While further functional validation of the predicted vascular interactions is warranted, targeting this axis offers a promising and novel therapeutic strategy to inhibit the malignant progression of OSCC. Conflict of interest None disclosed.