Introduction

Glioblastoma multiforme (GBM) is the most common malignant tumor of the central nervous system (CNS) in adults and remains highly fatal1. The unfavorable prognosis of GBM is intrinsically related to its highly aggressive and infiltrative traits, as comprehensively documented2,3. This infiltration is facilitated by various mechanisms, including the secretion of matrix metalloproteinases (MMPs), which degrade the extracellular matrix and enable tumor cells to migrate through brain tissue4,5. The tumor microenvironment (TME) is a pivotal determinant of tumor heterogeneity and therapeutic response, and its complexity necessitates advanced analytical approaches6. Recent studies have highlighted the critical role of the TME in cancer progression, as demonstrated by dynamic network biomarker analyses that identify stage-specific molecular transitions and microenvironmental alterations in thyroid carcinoma7. Utilizing algorithm-driven Multiomics analysis, such as genetic algorithm-driven multi-omics analysis, has demonstrated the critical role of TME characteristics in predicting immune checkpoint blockade efficacy8. These advances highlight the central role of the TME in precision oncology, offering biomarkers for risk stratification and therapeutic optimization. The invasive and migratory characteristics of GBM are closely linked to the TME. Inflammatory cells, particularly tumor-associated macrophages (TAMs), can promote tumor growth and invasion by secreting cytokines and growth factors that enhance the migratory capabilities of GBM cells9,10. Moreover, the altered composition of the extracellular matrix within the TME can significantly influence cell adhesion and migration11. These interactions create a dynamic and supportive environment for tumor progression, making it increasingly difficult to achieve effective therapeutic outcomes. Understanding these complex mechanisms is crucial for developing innovative strategies to combat GBM and to improve patient survival rates.

NUF2 is an important cell cycle protein that serves as a component of the NDC80 kinetochore complex and participates in binding of the kinetochore to microtubules during mitosis12. Emerging evidence has substantiated the role of NUF2 in a spectrum of human cancer models, including breast cancer13, ovarian cancer14, gastric cancer15 and hepatocellular carcinoma16. NUF2 plays a crucial role in regulating the cell cycle in tumor cells, particularly in facilitating the transition from the G1 phase to S phase. NUF2 overexpression has been shown to accelerate cell cycle progression, leading to increased proliferation of cancer cells17. Furthermore, NUF2 has been implicated in tumor cell migration and invasion. By modulating the reorganization of the cytoskeleton and the expression of cell adhesion molecules, NUF2 enhances the invasive potential of cancer cells. The role of NUF2 in the TME has also gained attention. Studies have indicated that NUF2 can influence the infiltration and activity of immune cells, thereby affecting the immune evasion mechanisms in tumors18,19. Elevated NUF2 expression may alter the function of TAMs and other immune cells, thereby promoting tumor growth and metastasis18. Moreover, NUF2 influences various signaling pathways that are critical for tumor biology. One of the most notable pathways is the Wnt/β-catenin signaling pathway. The activation of NUF2 enhances the nuclear translocation of β-catenin, increases its transcriptional activity, and promotes the expression of genes associated with cell proliferation and migration13. Additionally, NUF2 may regulate the PI3K/Akt and MAPK/ERK pathways, which are involved in cell survival and proliferation20,21,22. However, the specific role of NUF2 in GBM remains unclear. A recent study indicated that FOXM1-driven NUF2 enhances the impact of temozolomide (TMZ) on human GBM cells through the regulation of proliferation, apoptosis, and autophagy23; however, the downstream regulatory mechanisms remain to be elucidated. Accordingly, predicting NUF2 status in GBM may serve as a diagnostic tool, particularly in patients who are unable to undergo surgery. However, methods to determine the NUF2 expression status of GBM based on radiological features using magnetic resonance imaging (MRI) or computed tomography (CT) modalities have not been explored.

Radiogenomics is an emerging interdisciplinary field that integrates imaging data with genomic information to uncover correlations between the radiographic features and molecular characteristics of GBM24. By linking non-invasive imaging techniques, such as MRI or CT, with genetic and epigenetic profiles, radiogenomics provides valuable insights into GBM biology, enabling personalized medicine approaches25,26. Moreover, radiogenomics offers a non-invasive method to predict molecular markers, such as IDH mutation status, 1p/19q co-deletion, and MGMT promoter methylation, which are critical for GBM classification and treatment strategies selection24. Specific MRI features have been associated with distinct genetic alterations, such as contrast-enhancement patterns and tumor heterogeneity27. This allows clinicians to obtain molecular profiles without the need for invasive biopsies, particularly in cases in which surgical access is challenging. By utilizing radiogenomics, we can gain a deeper understanding of the genomic characteristics of GBMs, such as the spatial heterogeneity of tumors, treatment responses, molecular classifications, and immune infiltration within the tumor microenvironment26,28,29, Radiogenomic models have been developed to stratify patients based on their molecular subtypes, aid treatment decisions and outcome prediction30,31,32.

In this study, we aimed to investigate the novel function of NUF2 in GBM progression and elucidate the underlying molecular mechanisms. Specifically, we examined how NUF2 contributes to GBM cell proliferation, invasion, and migration, focusing on the β-catenin/MMP9 signaling axis. We sought to develop and validate a novel MRI-based radiogenomic model to non-invasively predict NUF2 expression in GBM patients, thereby providing a clinically applicable tool for patient stratification and personalized treatment management.

Materials and methods

Cell lines, reagents, datasets selection and patients

The GBM cell lines U87 and U251 were purchased from the Chinese Academy of Sciences (Shanghai, China). Dulbecco’s modified Eagle’s medium (DMEM), penicillin/streptomycin, and fetal bovine serum (FBS) were purchased from Gibco.Lentiviral-mediated short hairpin RNA(shRNA) plasmids were obtained from the Jikai Gene (Shanghai, China). Puromycin, LiCl (GSK-3β inhibitor) and temozolomide (TMZ ) were purchased from Sigma-Aldrich (St. Louis, MO, USA).

HTSeq-count data for GBM were obtained from the UCSC Xena platform (https://xena.ucsc.edu) and CGGA (https://www.cgga.org.cn). Data from the LM22 basis matrix from CIBERSORT (http://cibersort.stanford.edu) and the Genomics of Drug Sensitivity in Cancer (GDSC) database (www.cancerRxgene.org) were downloaded for immune infiltration assessment and prediction of small-molecule drug sensitivity. Single-cell RNA-seq data were downloaded from the GSE173278 database. In addition, MRI images of 61 patients with GBM were downloaded from The Cancer Imaging Archive (TCIA) dataset (https://www.cancerimagingarchive.net/).

The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of the First Affiliated Hospital of Shenzhen University. For the clinical cohort, GBM patients were selected based on the following criteria: (1) histopathologically confirmed primary GBM according to the 2021 WHO classification; (2) availability of preoperative MRI scans of adequate quality; and (3) complete clinical follow-up data. Patients with a history of other malignancies or incomplete clinical data were excluded. The imaging dataset from TCIA included T1-weighted contrast-enhanced (T1CE) MRI sequences acquired on 1.5T or 3T scanners with the following parameters: TR/TE = 8-15ms/3-6ms; flip angle = 10–15°; slice thickness = 1–1.5 mm; matrix size = 256 × 256; field of view = 240–260 mm. All images underwent quality assessment by two neuroradiologists with > 5 years of experience to ensure diagnostic quality before inclusion. Patient demographics included 61 GBM patients (33 males and 28 females) with a mean age of 51.23 years (range: 24–76 years). The MGMT promoter methylation status was available for 53 patients (86.9%), of whom 39 (73.6%) showed methylation. The 1p/19q codeletion status was determined in 60 patients (98.4%), of whom 10 (60.0%) exhibited codeletion.

Cell culture and lentiviral infection

U87 and U251 cells were cultured in DMEM supplemented with 10% fetal bovine serum, 100 µg/mL streptomycin, and 100 U/mL penicillin in a 37 °C incubator with 5% carbon dioxide. Cells were passaged when reaching 80–90% confluence using 0.25% trypsin-EDTA. Lentivirus-mediated RNA interference (RNAi) was performed according to the manufacturer’s instructions. Briefly, cells were seeded at 4 × 10^3 cells/well in 96-well plates for 24 h before transfection. Lentiviral particles containing shRNA constructs (multiplicity of infection = 10) were added to the cells in the presence of 4 µL HiTransG. After 16 h of incubation, the medium was replaced with a fresh complete medium. Stably transfected cells were selected with 2.5 µg/mL puromycin for 48 h beginning 72 h post-transfection. The shRNA sequences targeting NUF2 were as follows: shU87#: CCAAGCGTTTGAATGAACTAA; shU251#: CTACATGAGAGCCTTACAAAT. After 16 h of infection with lentivirus, the cells were passaged and selected with 2.5 µg/mL puromycin for 48 h to eliminate uninfected cells before harvesting for further analysis. Knockdown efficiency was determined by western blotting.

Cell viability

Cell viability was measured using a CCK-8 assay kit following the manufacturer’s protocol (Dojindo Molecular Technologies, Inc.). NUF2 knockdown U251(shNUF2-U251), U87 (shNUF2-U87) and null-transformed cells (shNC-U251 and shNC-U87) were inoculated into 96-well plates at a density of approximately 2000/well. Cells were cultured overnight, and when the cell density reached 50%-60% of the Petri dish surface area, TMZ (250 µM ) or LicL (10 mmol /L) was added for 72 h. The medium was then replaced with 10 µL of CCK-8 solution and 100 µL of DMEM. After further incubation for 1 h at 37 °C, the absorbance of each well was measured at 450 nm wavelength using a microplate reader (Multiskan Go 1510; Thermo Fisher Scientific, Inc.). The cell viability was calculated as follows: cell viability (%)=(experimental blank)/(control blank) × 100%.

Cell cycle analysis

For the cell cycle assay, a cell cycle detection kit (KeyGen BioTECH, Nanjing, China) was used to analyze the cell cycle after GBM cells were cultured for 24 h according to the manufacturer’s protocol. Briefly, shNC-U251, shNUF2-U251, shNC-U87 and shNUF2-U87cells were starved in serum-free medium for 24 h for synchronization, then changed to complete culture medium for another 24 h, after which the cells were harvested and fixed in ice-cold 70% ethanol at 4 °C overnight. Cells were centrifuged and resuspended in DAPI Staining Solution (Beyotime Biotechnology, China) for 30 min at room temperature in the dark. Finally, the cells were analyzed by FACS Calibur flow cytometry (Beckton Dickinson, San Jose, CA, USA), and the percentage of cells in the G0/G1, S, or G2/M phases was quantified.

Colony formation assay

To determine the effect of NUF2 on GBM development in vitro, we performed colony formation assays, as previously described methods in the literature33. Briefly, cells were seeded in 6-well plates at a density of 1,000 cells/well and cultured for 13 days until the colonies were large enough to be observed. Colonies were washed with PBS, fixed with 40 g/L paraformaldehyde, and stained with 5% crystal violet for 30 min at room temperature. After drying, the plates were scanned and the number of colonies was calculated using the Image J software (version 1.53q). The experiments were performed in triplicates.

Cell scratch, invasion and migration assays

Cell scratch, invasion and migration assays were performed as previously described5. Briefly, single-cell suspensions were inoculated into 6-well plates at a density of 5 × 105 cells/mL. When the cells were adherent to the wall, the bottom of the well plate was scratched with a 200 µL pipette tip, washed with PBS, changed to serum-free medium, and observed under an optical microscope for 0 h to determine the state of cell scratching. Licl (10 mM) was then added to each well, and incubation was continued for 24 h. Subsequently, photos were taken under a light microscope, and the wound healing ratio was calculated as (area 0 h-area 24 h)/area 0 h.

Cell migration and invasion capabilities were measured using Transwell chambers (No. 3422, BD Biosciences, USA) and Corning® Biocoat™ Matrigel® Invasion Chambers (No. 354480; Corning, USA) according to the manufacturer’s protocol. All chambers with 8-µm poreswere inserted into 24-well plates, and for the migration or invasion experiment, 3 × 104 transfected cells were seeded in the upper chamber medium lacking FBS, whereas the lower chamber was filled with culture medium containing 10% FBS. After incubation for 48 h, the chambers were fixed with 4% paraformaldehyde and stained with crystal violet (Beyotime Biotechnology, China). The cells that migrated or invaded the pores were imaged and counted.

Western blotting

Total protein was extracted from GBM tissue samples and cells using RIPA lysis buffer. Brain tissues derived from patients with epilepsy were used as the normal control, and the protein concentration was determined using a BCA kit (Beyotime Institute of Biotechnology). Following denaturation, proteins were separated by SDS-PAGE and transferred to a polyvinylidene difluoride (PVDF) membrane. The membrane was blocked with 5% skimmed milk at room temperature for 1 h. The following antibodies were then incubated on the PVDF membrane overnight at 4 °C: NUF2 (Cat#49882,dilution of 1:2,000; Signalway Antibody), PI3K (Cat#4249,dilution of 1:1,000; Cell Signaling Technology), ERK (Cat#9102, dilution of 1:1,000; Cell Signaling Technology), Cyclin B (Cat#181593, dilution of 1:2,000;Abcam), β-catenin (Cat#9562,dilution of 1:1,000; Cell Signaling Technology), MMP9 (Cat#13667, dilution of 1:1,000; Cell Signaling Technology), and β-actin (Cat#sc47778, dilution of 1:2,000; Santa Cruz Biotechnology). Subsequently, the PVDF membrane was incubated with IgG-HRP antibodies (anti-mouse, Cat#7076; anti-rabbit, Cat#7074; dilution of 1:4,000; Cell Signaling Technology) for 1 h. Finally, the protein bands were visualized using Western Bright™ reagent (Cat# K12043-d10; Advansta), and the grayscale values of the bands were analyzed using the Image-Pro Plus image analysis system.

Gene set enrichment analysis (GSEA) and immune infiltration analysis

GSEA is an analytical method used to assess the trend of distribution of genes of a predefined gene set in a table of genes sorted by phenotypic relevance, and thus to determine their contribution to the phenotype. The MSigDB gene set (https://www.gsea-msigdb.org/gsea/msigdb) was selected for enrichment analysis between the NUF2 high- and low-risk groups using GSEA software, which compares the differences in the enrichment of individual signaling pathways between the high- and low-risk groups. Pathway analysis was performed using KEGG (Kyoto Encyclopedia of Genes and Genomes) database (www.kegg.jp/kegg/kegg1.html)34,35,36, For previous uses, the Kanehisa laboratory have happily provided permission (Permission number: 252029).

For immune infiltration analysis, the abundance of 22 common immune infiltrating cells in the samples was estimated using the CIBERSORT algorithm. Spearman’s correlation analysis was performed on the 22 infiltrating immune cells and correlation Laplace plots were created using the “Corrplot” package.

Drug sensitivity analysis and cell subpopulation analysis

The GDSC is the largest freely accessible pharmacogenomics database for predicting the therapeutic response of small-molecule anticancer drugs to tumors and exploring differences in drug treatment effects between NUF2 risk groups of GBM patients.Drug sensitivity scores were assessed in GDSC2 using the oncoPredict software package.

The GSE173278 single-cell dataset was quality controlled and downscaled using the “Seurat” R software package. Cells were clustered according to the top 2,000 genes with the highest expression differences between cells and then downscaled to show the 2D spatial location of the cells using UMAP plots. The genes with the largest expression differences between each cell cluster and other clusters were obtained by analyzing the gene expression differences between cell clusters. Each cell cluster was identified by comparing the above genes with the marker genes recorded in the Cellmarker and Panglao DB databases. The clustered cells were then annotated and cell clustering maps were drawn using “ single R”.

Tumor segmentation and image preprocessing

The Cancer Imaging Archive (TCIA) MRI data were saved in the DICOM format. Leveraging 3D-Slicer 5.6.1 (https://download.slicer.org/), an experienced radiologist and two neurosurgeons independently performed manual slice-by-slice delineation of the tumor region of interest (ROI). The software automatically reconstructed a 3D volume of interest (VOI) from the annotations. To ensure objectivity, the contouring team was blinded to all clinicopathological data, except for the anatomical location of the tumor. Prior to feature extraction, all MRI images underwent standardized preprocessing: First, T1CE images in DICOM format were converted to the NIFTI format using the dcm2nii software package. Next, coordinate repositioning function of FSL was used to align the T1CE images of all patients to the Left-Posterior-Superior (LPS) coordinate system, standardizing their spatial orientation. Subsequently, the FSL registration tool (FLIRT) was employed to register the T1CE images to the standard T1-weighted template within the MNI152 atlas, unifying all brain tissues into a standard space with a resolution of 182 × 182 × 218 and a voxel size of 1 × 1 × 1 mm³. Finally, the skull was removed using the FSL brain extraction tool (BET), and the ANTs tool was applied to perform N4 bias field correction on the images, addressing intensity inhomogeneities. This standardized preprocessing workflow alleviates discrepancies in the imaging data acquired from different institutions and devices.

Radiomic feature extraction and radiomic feature selection

A total of 1037 features were extracted from the T1CE sequences in the tumor region. These included 14 shape features that described the 3D geometric characteristics of the VOI. 198 intensity features representing the first-order distribution of the intensities. A total of 825 texture features were extracted to capture patterns in high-order intensity distributions using five methods; the gray-level co-occurrence matrix (GLCM), gray-level run-length matrix (GLRLM), gray-level size zone matrix (GLSZM), gray-level dependence matrix (GLDM), and neighborhood gray-tone difference matrix (NGTDM). These features encompass the original features, those features derived from log transformations with bases of 3.0 and 5.0, and features obtained through wavelet transformations. An open-source Python package, PyRadiomics (version 2.2.0; https://pyradiomics.readthedocs.io/), was used to calculate radiomic features.

Before feature selection, patients from TCIA were randomly divided into training and test sets at a ratio of 7:3. Initially, the features were standardized using z-score normalization to achieve a mean of zero and a unit variance for the training set. The distribution of features in the training set was then used to normalize both the test sets and validation sets. For feature pairs with correlation coefficients exceeding 0.90, the feature with superior univariate predictive power (indicated by a smaller Mann–Whitney U test p-value) was retained. The number of features was reduced from 1037 to 589. Subsequently, the Least Absolute Shrinkage and Selection Operator (LASSO) regression model was applied to the training subset for further feature selection and signature construction. LASSO shrinks all regression coefficients toward zero based on the model parameter λ, effectively setting the coefficients of many irrelevant features to zero. To determine the optimal λ, ten-fold cross-validation with a minimum criteria approach was employed. In our analysis, the final λ value that minimized the cross-validation error was 0.1542, resulting in the optimal LASSO model, which retained six features with non-zero coefficients.

Reproducibility of radiomic measurements

Intraclass coefficients (ICCs) were calculated for the inter-observer agreement. The average ICCs for the three reviewers in measuring the radiomic features was 0.846 (range, 0.788–0.889) for T1CE.

Development of the prediction model

We developed a random forest classifier using the R package ‘randomForest’ (v4.7-1.1) to distinguish between NUF2 high- and low-expression GBM subtypes, with the model constructed using 500 decision trees (ntree = 500) via the Classification and Regression Tree (CART) methodology, where node splitting was based on the Gini impurity. The number of features considered at each split (mtry) was optimized via cross-validation, defaulting to the square root of the input features (mtry = 3 for the final six features), and the terminal node size was set to five to prevent overfitting. The feature importance was evaluated using the mean decrease in the Gini index. For model optimization, considering the moderate sample size (n = 61), we employed a repeated stratified 10-fold cross-validation strategy on the training set: 10-fold stratification ensured a balanced representation of NUF2 expression classes in each fold, while to 3–5 repetitions struck a practical balance between robust performance estimation and computational efficiency. The diagnostic performance of the model was assessed using metrics such as the AUC, accuracy, precision, recall, and F1-score. Additionally, clinical usefulness and net benefit were assessed using clinical decision curve analysis (DCA) and calibration curves to assess model consistency.

Statistical analysis

Results are presented as mean ± standard deviation (SD). Statistical analyses were performed using GraphPad Prism software (version 8.0). Student’s t-test was used to compare data between two groups, and one-way analysis of variance (ANOVA) was used to compare three or more groups. The Wilcoxon rank-sum test was used to analyze differences between non-normally distributed variables, and correlations between NUF2 and ECM, NUF2, and IDH were analyzed using Spearman’s correlation. Survival curves were estimated using the Kaplan-Meier method and compared using the log-rank test. For multiple comparisons, Bonferroni or Tukey’s post hoc test was applied, as appropriate. Statistical significance was defined as P < 0.05 (*), P < 0.01 (**), and P < 0.001 (***).

Results

Elevated expression of NUF2 correlated with unfavorable prognosis in GBM patients

NUF2 protein expression was significantly upregulated in GBM tissues compared to normal brain tissues, as determined by western blot analysis (P < 0.01; Fig. 1A, B). Similarly, both U87 and U251 GBM cell lines exhibited markedly higher NUF2 protein levels than normal human astrocytes (HA).  Notably, NUF2 expression showed a strong positive correlation with PI3K, ERK, and Cyclin B protein levels (Fig. 1A, B), which are established markers of poor GBM prognosis. Unsurprisingly, the Kaplan-Meier survival curve indicated that primary or recurrent GBM patients with high NUF2 levels had significantly shorter overall survival (OS) than those with low NUF2 expression (Fig. 1C, D). In summary, these results suggest that NUF2 expression levels are elevated in GBM tissues and cells and are associated with poor clinical outcomes.

Fig. 1
figure 1

NUF2 is upregulated in GBM and correlated with poor clinical prognosis. (A) Repersentative protein expression of NUF2 and PI3K in GBM cell lines (top pannel) and expression of NUF2, ERK and Cyclin B in tissues of GBM  (bottom pannel) assayed by western blot. (B) Quantitative analysis of protein expression in (A). *p＜0.05, **p＜0.01, ***p＜0.001.   (C,D) Kaplan-Meier analysis of overall survival (OS) in primary (C) or recurrent (D) GBM patients based on the expression of NUF2.

Full size image

NUF2 enrichment analysis, immune infiltration analysis, and cell subtype distribution

Considering the important role of NUF2 in GBM, this study explored the potential biological functions of NUF2 by using GSEA. GO analysis indicated that biological processes significantly associated with NUF2 expression included monoamine transport, voltage-gated potassium channel activity, and neurotransmitter receptor activity. KEGG analysis revealed related pathways such as DNA replication, ribosomes, and spliceosomes (Fig. 2A, B). Futhermore, we detected NUF2 in U87 and U251 GBM cell lines, and its expression level was significantly downregulated after lentivirus-mediated knockdown of NUF2 (Fig. 2C, D).

Fig. 2
figure 2

The function of NUF2 was analyzed, and its knockout status was confirmed. (A,B) GSEA enrichment analysis of NUF2. (C,D) Western blot analysis of the stable knockdown of NUF2 in U87 and U251 Cells. *p < 0.05.

Full size image

Single-cell analysis revealed a broad distribution of NUF2 in the GBM tissues. The results showed significantly elevated NUF2 expression in the astrocyte subpopulations (Fig. 3A). Immune infiltration analysis showed that several immune cell types, including gamma-delta T cells, macrophages, and CD4 + naive T cells, were positively correlated with NUF2 expression, whereas monocyte and regulatory T cell infiltration were negatively correlated (Fig. 3B). Additionally, NUF2 expression was positively correlated with that of extracellular matrix components, such as ELN, CSPG4, LAMC1, and MMPs (Fig. 3C). Notably, NUF2 expression was also positively correlated with IDH mutations as well as the expression of GSK3β, β-catenin, and MMP9 (Fig. 3E, F). Finally, drug sensitivity analysis preliminary predicted that NUF2 expression was positively correlated with sensitivity to Venetoclax, MK-8776, Lapatinib, and Pyridostatin, but negatively correlated with sensitivity to Entospletinib, Rapamycin, and SCH772984 (Fig. 3D). However, these findings remain speculative without experimental validation. Further studies are needed to confirm the association between NUF2 expression and the therapeutic efficacy of Venetoclax, MK-8776, and Lapatinib in GBM treatment. Such validation would strengthen the translational relevance of NUF2 as a therapeutic target and guide potential repurposing of these drugs for GBM patients based on the expression of NUF2. These results suggest that NUF2 plays a significant role in GBM development.

Fig. 3
figure 3

Comprehensive bioinformatics analysis for NUF2. (A) Visualization of NUF2 expression in various cell subpopulations. Correlation Analysis of NUF2 with immune infiltrating Cells (B), ECM (C), and small molecule drugs (D). (E) Correlation between NUF2 expression and IDH1/IDH2. (F) NUF2 Expression was positively correlated with β-catenin, MMP9, and GSK3β.

Full size image

NUF2 knockdown inhibits proliferation, colony formation, and cellcycle progression in GBM cells

We used shRNA technology to achieve stable depletion of NUF2 expression in the U251 and U87 GBM cell lines, and infection efficiency was assessed based on brightness and GFP expression observed under a fluorescence microscope. Furthermore, western blot analysis confirmed the knockdown effect of NUF2, showing that NUF2 protein expression was significantly reduced after shRNA-NUF2 transfection (shNUF2) compared with that in the negative control group (shNC) (Fig. 2C, D). CCK-8 assays revealed that NUF2 knockdown significantly reduced the proliferation of U251 and U87 cells compared with that of the negative control (shNC) (Fig. 4B). In addition, flow cytometry was used to assess the proportion of cells in different phases of the cell cycle. As shown in Fig. 4A, cell cycle progression in the shNUF2 groups was inhibited compared to the shNC group, as evidenced by a significant increase in the proportion of U87 and U251 cells in the G0/G1 phase. The ability of NUF2 expression to influence colony formation was assessed through clonogenic assays in the U251 and U87 cell lines. These results indicated that NUF2 knockdown affected the colony-forming capacity of both U251 and U87 cells. As shown, the number of colonies in the shNUF2 group was significantly lower than that in the shNC group (Fig. 4C). These data indicate that silencing NUF2 expression significantly slowed the growth rate of GBM cells, suggesting that NUF2 plays an important role in regulating the malignant characteristics of GBM.

Fig. 4
figure 4

NUF2 knockdown inhibited GBM cells proliferation and colony formation and induced cell cycle arrest. (A) The cell cycle distribution diagram and quantitative bar chart showed that NUF2 knockout led to cell cycle progression arrest in the G0/G1 phase. (B) The CCK8 assay results indicated that the number of viable cells in the shNUF2 group was significantly lower than that in the control group. (C) Representative photos of colonies on the plate and colony counts indicated that NUF2 knockout inhibited colony formation in U87 and U251 cells.*p < 0.05, **p < 0.01, and ***p < 0.001.

Full size image

Silencing NUF2 effectively inhibits the migration and invasion of GBM cells

Scratch assays, migration, and invasion experiments helped us understand the role of NUF2 in the dissemination, invasion, and migration of GBM cells, and the results of the scratch assays showed that compared to the shNC group, NUF2 knockdown in the U251 cell line exhibited reduced scratch healing ability (Fig. 5A). Invasion and transwell assays demonstrated that knockdown of NUF2 significantly reduced the number of migrating and penetrating cells in both U251 and U87 cell lines, indicating a marked decrease in their migration and invasion capabilities (Fig. 5B, C).

Fig. 5
figure 5

NUF2 knockdown inhibited GBM cell migration and invasion. (A) The scratch assay and (B) transwell assay showed the migration abilities of shNC or shNUF2 cells, along with the quantified bar graphs. (C) The invasion assay showed the invasion abilities of shNC or shNUF2 cells, along with the quantified bar graphs. Scale bar = 100 μm.*p < 0.05, ***p < 0.001.

Full size image

NUF2 is involved in GBM progression by regulating the β-catenin/MMP pathway and is associated with GBM sensitivity to TMZ

To elucidate the potential mechanisms of NUF2 in GBM progression, we performed western blot analysis to examine the protein level of β-catenin and MMP9. In U251 and U87 cells, the expression of β-catenin and MMP9 was downregulated in the shNUF2 group compared to that in the negative control group (p < 0.05) (Fig. 6D).

Fig. 6
figure 6

NUF2 regulates GBM cell proliferation and migration through β-catenin and MMP9. (A) Scratch assay showing the migration results of shNC-U251 and shNUF2-U251 cells treated with Licl (0 mM, 10 mM) at 0 h and 24 h, and quantification of the wound healing ratio. (B) After treatment with 10 mM LiCl, the number of U87 and U251 cells transfected with shNC or shNUF2 decreased, but the vitality inhibition caused by NUF2 knockdown was reversed. (C) Cell viability assay of U87 and U251 cells transfected with shNC or shNUF2 and treated with DMSO or 250 µM TMZ showed that NUF2 knockdown increased sensitivity to TMZ treatment. (D) Western blot analysis of NUF2, β-catenin, and MMP9 protein levels in U87 and U251 cells transfected with shNC or shNUF2, along with quantification of the relative protein levels of NUF2, β-catenin, and MMP9, indicated that NUF2 expression was positively correlated with β-catenin and MMP9. *p < 0.05, **p < 0.01, and ***p < 0.001.

Full size image

LiCl, a GSK-3β inhibitor, can activate the β-catenin signaling pathway, thereby promoting tumor cell proliferation, migration, and epithelial-mesenchymal transition (EMT). In the CCK-8 assay, NUF2 lentivirus-transfected U251 and U87 cells showed a reversal of the inhibition of proliferation caused by NUF2 knockdown upon addition of 10 mM LiCl (Fig. 6B). Furthermore, although the scratch-healing ability of NUF2 knockdown U251 cells was diminished after LiCl treatment, it was improved compared with that in the blank treatment group (Fig. 6A). This suggests that NUF2 may regulate GBM cell migration by modulating the β-catenin signaling pathway and MMP9 expression (Fig. 7). Notably, reduced NUF2 expression decreased the sensitivity of U251 and U87 cells to TMZ treatment. Following NUF2 knockdown, shNUF2 cells exhibited substantial cell death in response to TMZ (Fig. 6C), indicating that NUF2 may be involved in the regulation of acquired resistance to TMZ.

Fig. 7
figure 7

Schematic representation of the proposed mechanism by which NUF2 promotes GBM cell proliferation, invasion, and migration through the β-catenin signaling pathway. NUF2 upregulates β-catenin and MMP9 expression, leading to enhanced cell proliferation and invasion.

Full size image

Prediction of NUF2 expression by MRI-based machine learning model

Considering the significant role of NUF2 in GBM, we aimed to develop a non-invasive model for predicting NUF2 expression in GBM patients using MRI radiomics analysis based on TICA data, which included 61 patients (33 males and 28 females; mean age, 51.23 years). A total of 1037 radiomic features were extracted, and after applying variance thresholding and LASSO regression for feature selection (Fig. 8A, B), six features were ultimately identified, including one first-order feature and five textural features (Fig. 8D and Table S1). Supplementary Fig. 1 shows that RF1 and RF6 were significantly overexpressed in GBM cells with low NUF2 expression, whereas RF2-RF5 was significantly overexpressed in GBM cells with high NUF2 expression. In the test set of the prediction model, the area under curve (AUC), accuracy, precision, recall, and F1-score were 0.897, 0.842, 0.857, 0.75, and 0.8, respectively (Fig. 8E and Table S2). The calibration curve and DCA assessment of the model (Figs. 8E-G) indicated that the model predictions were consistent with the actual outcomes and achieved the highest net benefit.

Fig. 8
figure 8

Development and validation of a radiogenomics-based NUF2 expression model for GBM patients. (A,B) LASSO coefficient profiles of the radiomics features. The vertical dashed line indicates the optimal lambda value selected by 10-fold cross-validation. (C) Heatmap showing the association between NUF2 expression, clinical features (age, gender, MGMT status, 1p/19q codeletion), and overall survival status in GBM patients. (D) Variable importance plot showing the top radiomics features (RF1-RF6) ranked by MeanDecreaseGini in the random forest model. (E) Receiver operating characteristic (ROC) curve of the radiomics signature for predicting NUF2 expression in GBM patients. The area under the curve (AUC) was 0.897. (F) Calibration curve of the radiomics signature for predicting NUF2 expression in GBM patients. The dashed line represents the ideal calibration, and the solid lines represent the logistic and nonparametric calibration curves. (G) Decision curve analysis (DCA) of the radiomics signature compared to the “All” and “None” strategies. The net benefit was plotted against the threshold probability.

Full size image

The Mann–Whitney U test was used to analyze the relationship between the six radiomic features and MGMT demethylation. The results showed that RF1 was significantly overexpressed in MGMT-methylated GBM, whereas RF2 and RF3 were significantly overexpressed in MGMT-demethylated GBM. However, the expression of RF4-RF6 was not associated with MGMT status (Fig. S1). Additionally, in the clinical correlation heatmap (Fig. 8C), high expression of RF2, RF3, and RF5 was observed in GBM patients with 1p/19q non-codeletion, and these features were predominantly found in older patients. To evaluate the predictive ability of the six radiomic features for GBM prognosis, univariate and multivariate COX analyses were conducted. Table S3 shows that these features were predictive factors for poor prognosis in GBM, with RF2 and RF5 serving as independent predictors. In summary, the predictive model based on high and low NUF2 expression can non-invasively and effectively predict NUF2 expression through MRI, and its radiomic features are independent predictors of poor prognosis.

Discussion

NUF2, a significant component of the Ndc80/Nuf2 complex, is required for the stable attachment of kinetochores to microtubules. Chromosome alignment during mitosis has been shown to play a role in the regulation of apoptosis and proliferation of tumor cells. However, its possible effects on GBM progression and potential mechanisms have not been fully elucidated. The major findings of this study include the following: (i) NUF2 is markedly upregulated in human GBM cells (U87 and U251 cells), which correlates with poor clinical outcomes; (ii) knockdown of NUF2 ameliorates GBM cell proliferation and inhibits the cell cycle; (iii) knockdown of NUF2 significantly attenuates GBM cell invasion and migration; (iv) NUF2 knockdown enhances β-catenin phosphorylation and degradation, further promoting the downregulation of MMP9 expression, and thus inhibiting GBM cell invasion and migration; and (v) a non-invasive radiogenomic model for predicting NUF2 expression based on MRI combined with machine learning has been established, and the model was rigorously evaluated and demonstrated preferable practicality and accuracy. To our knowledge, this is the first comprehensive study to demonstrate that NUF2 promotes GBM cell proliferation, invasion, and migration by activating the β-catenin/MMP9 signaling axis. Furthermore, we developed and validated a robust MRI-based radiogenomic model for the non-invasive prediction of NUF2 expression, which was independently associated with poor clinical outcomes in patients with GBM. Unlike previous study on NUF2 in TMZ resistance37, our study uniquely demonstrates its role in invasion/migration through β-catenin/MMP9 and provides the first non-invasive prediction method.

The evidence suggests that NUF2 depletion inhibits cell growth through cell cycle arrest and apoptosis, as evidenced by the abnormal expression of related cell cycle and apoptosis proteins37. A previous study revealed that NUF2 activated by FOXM1 supports the effects of TMZ on human GBM cells by regulating proliferation, apoptosis, and autophagy. It has also been reported to FOXM1 accelerates the development of malignant peripheral nerve sheath tumors in neurofibromatosis type 1 via a NUF2-dependent pathway. However, the mechanisms by which NUF2 functions in GBM remain unclear. This is the first study on the role of NUF2 in cell invasion and migration and provides preliminary insights into the vital role of NUF2 in GBM. Additionally, we found that the expression of NUF2 in GBM tissues or cells was positively correlated with the expression of prognostic biomarkers associated with poor prognosis in GBM, such as PI3K, ERK, and Cyclin B, which are overexpressed in GBM tissues or cells. This suggests that NUF2 may serve as a candidate molecular biomarker for predicting the prognosis of GBM, which is consistent with previous reports on the potential role of NUF2 in the prognosis of other tumors19,38,39,40. Furthermore, our data indicated that different types of immune cell infiltration were positively correlated with NUF2 expression, including T cells gamma delta, macrophages, and CD4 T cells, but negatively correlated with monocytes and T cell filtrates. Notably, our immune infiltration analysis relied on the CIBERSORT algorithm without IHC validation, which may limit the robustness of the observed correlations. Future studies incorporating IHC or flow cytometry will be warranted to confirm these associations. Reduced NUF2 expression reduced the sensitivity of GBM to TMZ, which was consistent with previously reported results23. Mechanistically, NUF2 may regulate the migration of GBM cells by modulating the β-catenin signaling pathway and altering MMP9 expression. MMP9 degrades collagen and other components of the ECM and promotes migration and invasion of GBM cells41. However, the effect of NUF2 on ECM was not determined in this study. The mechanism by which NUF2 regulates the ECM through MMP9 requires further investigation. Moreover, future studies should explore real-time NUF2 expression tracking during targeted therapy, particularly for β-catenin inhibitors that show preclinical efficacy in NUF2-high GBM.

Radiogenomics, an emerging research field, combines MRI and genomic information to better classify and understand the heterogeneity and complexity of GBM42,43,44,45. These studies revealed significant heterogeneity in GBM and categorized it into different molecular subtypes, such as pre-neural, neural, classical, and mesenchymal, allowing researchers to better identify GBM biomarkers associated with treatment responses by integrating imaging and genomic data42,46. Several studies have shown that radiomic features have an additional prognostic value in GBMs46,47,48. In addition, recent studies have shown that radiomics can improve survival prediction for the LGG subset of IDHwt [49,50,51,52 ]49,50,51,52. However, owing to the high heterogeneity of GBM and its complex TME, the development of effective targeted diagnostic markers and therapeutic strategies remains a challenge. Our findings demonstrated that quantitative MRI-derived features, particularly textural heterogeneity in T1CE sequences, are significantly correlated with NUF2 overexpression, aligning with its established role in mitotic regulation, DNA damage repair, and chromatin instability53,54. This non-invasive biomarker strategy addresses the critical limitations of tissue sampling in heterogeneous GBMs, where spatial and temporal variations in NUF2 expression may influence therapeutic resistance and increased NUF2 levels may drive GBM proliferation by altering the ECM, leading to more aggressive tumors. Our multivariate COX model incorporating radiomic-NUF2 signatures stratified patients into distinct survival cohorts (Supplementary Table 3), which is complementary to the traditional histopathological grading. Notably, the radiogenomic model combined with NUF2 expression also revealed associations between the molecular pathological patterns of GBM and the level of NUF2, suggesting its potential for molecular targeting therapy. Nevertheless, this study had several limitations. First, our study did not incorporate comprehensive molecular profiling, such as H3K27, TERT, ATRX, and TP53 mutations, which may further refine the prognostic stratification. Second, the retrospective nature and relatively small sample size of the imaging cohort may have introduced selection bias. Future prospective multicenter studies with larger cohorts and integrated multi-omics analyses are required to validate and expand our findings. Therefore, integration of radiogenomics to predict NUF2 expression represents a transformative and promising approach for refining GBM prognosis and therapeutic decisions.

In addition, our radiogenomic model not only predicts NUF2 expression but also reveals important associations with established molecular biomarkers of GBM, particularly MGMT promoter methylation and 1p/19q codeletion. The identified radiomic features, RF1, RF2, and RF3, showed significant correlations with MGMT methylation status, which is a well-established predictor of response to temozolomide chemotherapy. This finding suggests that NUF2 expression patterns may interact with DNA repair mechanisms, potentially influencing the treatment response pathways beyond the β-catenin/MMP9 axis identified in this study. The association between certain radiomic features (RF2, RF3, and RF5) and 1p/19q codeletion status is particularly noteworthy, as this molecular alteration is typically associated with oligodendroglial tumors and a better prognosis. The methylation status of MGMT has been extensively studied and is considered to be a significant predictor of prognosis and treatment response in patients with GBM. Study has shown that MGMT promoter methylation is associated with better survival rates and progression-free survival, particularly in patients treated with alkylating agents such as TMZ55. Additionally, 1p/19q coding deletions are favorable prognostic markers, especially in oligodendrogliomas with IDH mutations, where this deletion is linked to better treatment responses and outcomes56. In clinical practice, testing for MGMT and 1p/19q has been recommended as a routine step in personalized treatment of patients with malignant gliomas. Testing these biomarkers can help identify patients who might benefit from personality treatment, thereby optimizing treatment strategies and improving overall survival57. Furthermore, there was a study has developed molecular classification and prediction models based on DNA methylation-driven genes, which can predict the individual prognosis and immunotherapy response of patients with GBM. These models offer more precise survival predictions and treatment options by integrating multiple molecular markers with clinical features58.

In conclusion, our results provide preliminary evidence for the role of NUF2 in GBM progression, demonstrating that its upregulation in GBM cells can lead to increased cellular growth, invasion and migration. Furthermore, NUF2 promotes GBM invasion and migration by mediating the β-catenin/MMP9 signaling pathway. Further studies are required to elucidate the mechanisms underlying these changes. Additionally, NUF2 expression, as predicted by MRI-based machine learning models, may serve as a negative prognostic indicator in patients with GBM. Future research needs to identify more datasets that meet the same criteria (matching imaging data and gene expression data) to be used for validation and model optimization.