S100B induces angiogenesis via the clathrin/FOXO1/β-catenin signaling pathway and contributes to Bevacizumab resistance in epithelial ovarian cancer

Graphical abstract

Highlights

• •A signature was developed to predict the prognosis of BEV treatment in EOC patients.
• •S100B is a major molecule that regulates the sensitivity of EOC to BEV.
• •Exogenous S100B enters endothelial cells through clathrin-mediated endocytosis.
• •S100B promotes tube formation of endothelial cells through the FOXO1/β-catenin pathway.
• •S100B inhibitor Pentamidine collaborates with BEV to inhibit tumor angiogenesis.

Abstract

Introduction

Bevacizumab (BEV), the most common antiangiogenic agent for treating ovarian cancer, prolongs progression-free survival (PFS) but does not significantly improve overall survival (OS). Improving the limited clinical benefit of BEV remains a major challenge in ovarian cancer treatment. Although several studies have explored the mechanisms underlying tumor resistance to BEV, the clinical translation of these findings to overcome BEV resistance has been limited.

Objectives

To identify the key molecules and mechanisms that modulate ovarian cancer sensitivity to BEV.

Methods

RNA sequencing was conducted on BEV-sensitive and BEV-resistant mouse ovarian cancer tissue to identify differentially expressed genes (DEGs). A prognostic assessment was performed and a risk signature was constructed using these DEGs and the BEV-related sequencing datasets. S100B was identified and assessed in angiogenesis using tube formation, 3D fibrin bead sprouting, wound healing, and migration assays. Downstream targets and signaling pathways of S100B in HUVECs were identified by proteomics and validated by western blot. The effect of S100B inhibitors on BEV efficacy was evaluated using in vivo experiments.

Results

A BEV-related prognostic signature comprising 11 genes was constructed. Of these, S100B expression significantly increased in BEV-resistant mouse ovarian cancer tissue and significantly correlated with poor PFS and OS of ovarian cancer patients treated with a BEV combination chemotherapy. HUVECs co-cultured with S100B-overexpressing ovarian cancer cells promoted tube formation, sprouting, and migration. Exogenous S100B entered HUVECs via clathrin-mediated endocytosis, downregulated FOXO1 expression, and promoted β-catenin nuclear translocation and transcriptional activity, ultimately enhancing tube formation. The S100B inhibitor pentamidine significantly increased BEV responsiveness and prolonged survival in ovarian tumor-bearing mice.

Conclusion

S100B is a key molecule regulating ovarian cancer sensitivity to BEV. Paracrine S100B secreted by ovarian cancer cells acts on HUVECs, promoting angiogenesis through the FOXO1/β-catenin pathway. Pentamidine combined with BEV holds potential for overcoming BEV resistance in clinical use.

Introduction

Ovarian cancer is one of three major gynecologic malignancies in women, with epithelial ovarian cancer (EOC) being the most common pathological subtype, accounting for approximately 90 % of cases [1]. Due to the absence of obvious early symptoms, around 75 % of patients are diagnosed at an advanced stage [1], leading to a poor prognosis and significant threats to the health and quality of life of the patient. In recent years, the management of EOC has evolved into a comprehensive model combining more precise and extensive cytoreductive surgery, platinum-based chemotherapy, and targeted maintenance therapy [2]. Despite advances in treatments and improvements in short-term outcomes, long-term survival remains unsatisfactory due to the high rates of tumor recurrence and metastasis. Consequently, the 5-year survival rate is only about 46 % [3].

Bevacizumab (BEV), an anti-angiogenic drug targeting vascular endothelial growth factor A (VEGFA), inhibits tumor growth by reducing the formation of new blood vessels within tumors [4]. BEV is approved by the U.S. Food and Drug Administration (FDA) for treating both newly diagnosed and recurrent epithelial ovarian cancer. Clinical studies have demonstrated that BEV significantly improves progression-free survival (PFS); however, it has not shown a substantial benefit in overall survival (OS) [5]. The development of adaptive resistance to BEV is thought to be the primary limitation of its long-term efficacy. Mechanisms underlying this resistance include increased alternative angiogenic factor expression, metabolic reprogramming, myeloid-derived suppressor cell recruitment, and tumor stroma remodeling [6,7]. However, efforts to leverage these insights into clinical strategies to overcome resistance have yielded only limited success. One clinical study found that nidanib (a tyrosine kinase inhibitor targeting VEGF, FGF, and PDGF receptors) had minimal activity in bevacizumab-resistant EOC patients, with a partial response rate of only 7.4 %; further, 37 % of patients exhibited stable disease and 56 % of patients showed progressive disease (PD) [8]. Therefore, it is of great clinical significance to screen the molecules associated with BEV resistance and explore for potential targeting molecules that can improve BEV sensitivity.

The flowchart of this study is presented in Fig. 1. We used RNA sequencing of tissue from ovarian cancer animal models combined with large-scale public database analyses to construct a BEV-related prognostic signature. We identified S100B as a potential key molecule regulating BEV sensitivity in ovarian cancer. S100B is a member of the EF-hand calcium-binding protein S100 family, which regulates various intracellular functions by interacting with target proteins and exhibits cytokine-like activity as a secreted protein [9]. S100B is predominantly found in the central nervous system, where it is synthesized and secreted by astrocytes and serves as a critical marker of neural distress in clinical practice [10]. Moreover, elevated expression and secretion of S100B have been observed in gliomas [11] and melanomas [12], with serum S100B levels often used as an important diagnostic and therapeutic marker. Existing studies have demonstrated that S100B binds to TP53 in tumor cells, promoting TP53 degradation and inhibiting its transcriptional function, thereby facilitating tumorigenesis and progression [9]. However, research on S100B in other tumors remains limited. In ovarian cancer, only one study has suggested that S100B may participate in regulating cancer stemness maintenance and chemoresistance by suppressing TP53 expression and phosphorylation [13]. In a previous study, serum S100B levels were examined in nine patients with stage III melanoma before and after BEV treatment. Decreased S100B levels were associated with post-treatment tumor necrosis, while increased levels were observed in patients without tumor necrosis following BEV treatment [14]. These findings suggested that S100B may be a valuable biomarker indicating BEV efficacy in melanoma patients. In terms of angiogenesis, current studies have shown that S100B promotes proliferation and increased permeability of vascular endothelial cells [15,16] and smooth muscle cells [17,18]. However, there are few studies on its mechanism of action, and some studies suggest that it may potentially promote angiogenesis by regulating the NF-kappaB pathway to enhance the expression of VEGFs [[19], [20], [21], [22]]. However, the role of S100B in regulating BEV efficacy and its impact on angiogenesis remain unclear. In this study, we determined that S100B promoted tube formation, sprouting, and migration of endothelial cells through the FOXO1/β-catenin pathway and was unaffected by BEV. The combination of the S100B inhibitor pentamidine with BEV inhibited tumor growth, reduced blood vessel formation, and significantly prolonged survival in an ovarian cancer mouse model.

Fig. 1.

Flowchart of the study. First, BEV-sensitive and BEV-resistant ovarian cancer mouse models were constructed, and RNA sequencing was performed on tumor tissue. Then, a BEV-related prognostic signature was established using machine learning, and S100B was identified as the most important molecule regulating BEV sensitivity in ovarian cancer. Its function and mechanism were analyzed in vitro. Finally, BEV efficacy when combined with an S100B inhibitor was verified in vivo. BEV: Bevacizumab; DEGs: Differentially expressed genes.

Materials and methods

Animal experiment

Animal experiments were performed using 5–6-week-old female BALB/c nude mice, sourced from Vital River (Beijing, China), and housed in a specific pathogen-free environment. The mice underwent a 1-week acclimation period before the experiment. A2780-luciferase cells (1 × 10⁷ cells) were injected into the peritoneal cavity of each mouse, and tumor progression was monitored weekly using an animal imager (Bruker MI) by D-Luciferin salt (Yeasen, 40901ES03). After 2 weeks, all mice had developed tumors, and they were randomly assigned to experimental groups based on tumor imaging results.

To construct the BEV adaptive resistance model, mice were randomly assigned to a control group (n = 5) and a BEV treatment group (n = 10) 2 weeks after basal tumors had formed. Mice in the control group received normal saline (80 μL, twice per week) via intraperitoneal injection, while mice in the BEV treatment group received BEV (Avastin, Roche). A BEV stock solution (25 mg/mL) was diluted with normal saline to a concentration of 2.5 mg/mL and injected intraperitoneally twice a week at a dosage of 10 mg/kg. After 3 weeks of treatment, when mice in the control group could no longer tolerate the tumors and their body weight was reduced by more than 20 %, the tumor tissue was harvested. In contrast, tumor growth in BEV-treated mice was significantly inhibited during weeks 2–4 of treatment compared to that in the control group, and the harvested tumor tissue was designated as the sensitive group (n = 5). The remaining mice continued BEV treatment. After 7 weeks, tumor growth significantly accelerated compared to that of week 4 and the body weights were reduced by more than 20 %. The tumor tissue was harvested and classified as the BEV-resistant group (n = 5).

For the combination drug treatment model, two weeks after the formation of basal tumors, mice were randomly assigned to four groups, with five mice in each group: control, pentamidine treatment, BEV treatment, and BEV combined with pentamidine treatment. The control group was administered intraperitoneal injections of normal saline. Mice in the BEV treatment group received BEV (10 mg/kg) intraperitoneally twice a week. The pentamidine treatment group was administered pentamidine (Selleck, China) at a dosage of 10 mg/kg intraperitoneally three times a week [23]. Mice in the BEV combined with pentamidine treatment group were given intraperitoneal injections of both drugs. Mice were euthanized when their body weight decreased by more than 20 %, and tumor tissue was harvested.

At the end of the experiment, the mice were euthanized, and the abdominal cavity was dissected to collect the ovarian tumor tissue. The following procedures were performed: 1) The tumor tissue was trimmed to remove necrotic and non-tumor tissues, divided into three sections, and stored in three pre-packaged RNAlater vials (Servicebio, G3019) for subsequent RNA sequencing. 2) Additional tumor tissue was trimmed to remove necrotic and non-tumor tissues, frozen in liquid nitrogen for 15 min, placed in cryovials, and preserved at − 80 °C for subsequent protein extraction and western blot. 3) Representative tumor tissue was selected and fixed in 4 % paraformaldehyde (Servicebio, G1101) for paraffin embedding and sectioning.

RNA sequencing analysis

Fresh mouse tissue stored in RNAlater (Servicebio, G3019-25) was used for RNA extraction. RNA concentrations were measured using a NanoDrop spectrophotometer (Thermo Fisher Scientific). The minimum detectable RNA concentration was set to 500 ng. Sequencing was performed on an Illumina NovaSeq 6000 platform in PE150 mode. The raw sequencing data underwent quality control and filtering and were normalized using the log2(FPKM + 1) method. Differential expression analysis was conducted using the “DESeq2” R package, identifying differentially expressed genes with a threshold of |log2(fold change)| ≥ 2 and P-value ≤ 0.05.

Immunofluorescence staining of tissue sections

According to the existing steps of immunofluorescence staining, we used the indirect method to process and stain the paraffin sections [24]. The paraffin sections were dewaxed, rehydrated, and underwent to antigen retrieval in EDTA antigen retrieval buffer (Servicebio, G1207-1L) in a microwave oven. The sections were then slightly dried, and the tissue margins were outlined using a tissue pen. Sections were blocked for 30 min with BSA (Beyotime, ST023-50 g), and then the blocking solution was gently removed. Two primary antibodies, diluted in PBS at specified ratios, were applied to the sections, which were incubated overnight at 4 °C. The following day, secondary antibody solutions were applied to the tissue sections and incubated at room temperature for 50 min. The sections were counterstained with a DAPI solution (Beyotime, C1002) and kept in the dark at room temperature for 10 min. The sections were slightly dried, and a self-fluorescence quencher (Beyotime, P0126) was applied for 5 min, followed by a 10-min wash with water. Once dried, the sections were mounted with anti-fade mounting medium. Finally, the prepared sections were imaged using a fluorescence scanner. The primary antibodies utilized for immunofluorescence were as follows: HIF-1α (Proteintech, 20960–1-AP, 1:250) and CD31 (Servicebio, GB11063-2, 1:500).

Metascape database

Metascape (https://metascape.org/gp/index.html) integrates more than 40 independent functional databases, including GO, KEGG, and UniProt, providing a wide range of capabilities, such as functional enrichment, interactive group analysis, and gene annotation [25]. We used this database for enrichment analysis of our RNA sequencing and proteomics data, with minimum overlap = 1, P-value cutoff = 0.05, and minimum enrichment = 1.5 as thresholds.

Download and analysis of gene expression Omnibus database

The sequencing datasets GSE140082 [26], GSE72951 [27], GSE118828 [28], GSE154600, GSE184880 [29], and GSE189955 [30] were downloaded from the Gene Expression Omnibus database (GEO, https://www.ncbi.nlm.nih.gov/geoprofiles). Data processing and visualization were performed using R software (version 3.6.1).

Development and validation of BEV-related prognostic signatures (BRPSs)

For the GSE140082 and GSE72951 datasets, the “intersect” function was employed to combine RNA-seq expression data with clinical information. Before constructing the prognostic signature, univariate Cox regression analysis was conducted with the “coxph” function from the “Survival” R package, which identified 16 variables significantly correlated with OS in the GSE140082 dataset. The GSE140082 samples were randomly allocated into a training set and a validation set in a 7:3 ratio with the “sample” function. In the training set, various classical machine learning algorithms were integrated to construct the signature, including random forest (RSF), least absolute shrinkage and selection operator (LASSO), partial least squares regression for Cox, gradient boosting machine, supervised principal components, CoxBoost, elastic net (Enet), survival support vector machine, and stepwise Cox (StepCox). Four algorithms—RSF, LASSO, CoxBoost, and StepCox (including backward and both methods)—were used for preliminary variable screening and dimensionality reduction and subsequently combined with other machine learning methods to establish the final prognostic signature. The concordance index (C-index) of each prognostic signature was calculated using the GSE140082 training set, GSE140082 internal validation set, and GSE72951 external validation set, and the mean values were computed, sorted, and visualized.

Based on the selected prognostic signature, risk scores were calculated for each sample in the GSE140082 training set, GSE140082 internal validation set, and GSE72951 external validation set. Kaplan–Meier survival analysis was performed to identify high- and low-risk patients using the “Survival” R package. The “timeROC” R package was employed to perform time-dependent receiver operating characteristic (ROC) analysis.

Construction of a BRPS-related nomogram

Univariate and multivariate Cox regression analyses were conducted to determine OS-related variables in the complete GSE140082 dataset. Based on these variables, a nomogram was developed with the “rms” R package. A decision curve analysis curve was drawn using the “stdca” function from “rmda” R package. A ROC analysis related to the nomogram was conducted by the “timeROC” R package.

Gene set enrichment analysis (GSEA)

In the complete GSE140082 dataset, patients were categorized into high-risk and low-risk groups based on the risk scores determined by the BRPS. GSEA was performed by GSEA software (v. 4.0.3) [31], with “H.all.v2024.1.Hs.symbols.gmt” as a reference gene set. Enrichment analysis was conducted on the two groups divided by risk, and a nominal P-value < 0.05 was used as the threshold to recognize meaningfully enriched hallmark pathways.

Processing and annotation of single-cell sequencing data

The “Seurat” R package was used to process and visualize four single-cell RNA sequencing datasets: GSE118828, GSE154600, GSE184880, and GSE189955. Cells expressing fewer than 50 genes or containing more than 5 % mitochondrial genes were removed. After normalizing the data using the “NormalizeData” function, the “FindVariableFeatures” function was applied to identify genes with significant variation in the different cells. Initial dimensionality reduction was performed using PCA, followed by nonlinear structure exploration using t-SNE. Cell markers for ovarian cancer tissue were collected from the CellMarker database (http://biocc.hrbmu.edu.cn/CellMarker/) (Supplementary Table S1), and the samples were manually annotated.

Immunohistochemical staining (IHC)

Based on the classic steps of microwave antigen repair and IHC staining of paraffin sections [32], we formulated the following protocol:Paraffin sections underwent dewaxing and rehydration, and antigen retrieval was performed using microwave heating in 10 mM sodium citrate solution (Servicebio, G1202). The sections were blocked with serum, treated with primary antibodies, and incubated overnight at 4 °C. The following day, secondary antibodies were applied and the sections were incubated at room temperature, followed by hematoxylin (Beyotime, ST2067) counterstaining, dehydration, and sealing with neutral resin. The sections were then observed and photographed by an inverted microscope. The primary antibodies used in IHC were as follows: S100B (Proteintech, 66616–1-Ig, 1:250) and CD31 (Servicebio, GB11063-2, 1:500). A semi-quantitative method was used to evaluate S100B staining. Staining intensity was graded as 0 (no staining), 1 (light yellow), 2 (brown-yellow), and 3 (dark brown). The percentage of positive cells in five randomly selected high-power fields was graded as 0 (<5%), 1 (5 %–25 %), 2 (26 %–50 %), 3 (51 %–75 %), and 4 (>75 %). The final score was obtained by multiplying the staining score by the percentage score. The microvascular density (MVD) was assessed by counting the number of vessel cross-sections with positive CD31 staining in five randomly selected high-power fields. The average value was calculated to represent the final MVD. Each tissue section was independently reviewed by two researchers to minimize scoring errors.

Cell culture and treatment

Ovarian cancer cell lines OVCAR3 was cultured in RPMI 1640 (Cytiva, USA) medium containing 10 % fetal bovine serum (FBS) (Thermo Fisher Scientific, Waltham, MA, USA). HEY and A2780 were cultured in DMEM (Cytiva, USA) with 10 % FBS. Human umbilical vein endothelial cells (HUVECs) were cultured in Ham's F-12 K (Kaighn's) medium (Genom, China) with 10 % FBS. All cells were maintained at 37 °C with 5 % CO₂.

HUVECs and ovarian cancer cells were transfected with a lentivirus (GeneChem, China) to establish stable S100B or FOXO1 overexpressing cell lines, followed by selection with 4 μg/mL puromycin for 1 week. During phenotype experiments with HUVECs, recombinant S100B protein (MCE, HY-P70659) was added at concentrations of 10, 100, and 1000 mg/mL. Protein extraction and phenotype experiments were performed after HUVEC pretreatment with endocytosis inhibitors. HUVECs were treated with 1 μM or 10 μM Pitstop-2 (MCE, HY-115604), a clathrin-mediated endocytosis inhibitor, for 6h; 20 μM or 40 μM nystatin (MCE, HY-17409), a caveolae/caveolin1-mediated endocytosis inhibitor, for 1 h; or 5 μM or 10 μM of FPS-ZM1 (MCE, HY-19370), a RAGE inhibitor.

Data-independent acquisition (DIA) proteomics

HUVECs (2 × 10⁶ cells) were cultured in a 10-cm dish and treated with either a control or 100 ng/mL exogenous recombinant S100B. After 48 h, the cell medium was removed. Cells were washed, scraped, and transferred to 1.5-mL tubes. These steps were repeated three times to ensure sufficient replicates. After sample collection, proteins were extracted and subjected to enzymatic digestion. Prior to analysis, iRT, a Biognosys quality control reagent, was mixed to each sample to calibrate the peptide retention time during chromatography. Pulsar software generated a database from data-dependent acquisition (DDA), and data-independent acquisition (DIA) data were analyzed against it for protein identification. Proteins detected in at least one sample were included in the qualitative analysis, and quantitative data for all samples were exported. The spectral peak intensity of the samples was normalized using the local normalization method in the Pulsar software. Differentially expressed proteins were screened with a threshold of |log2(fold change)| ≥ 1.0.

Protein extraction and western blot

Proteins were extracted from tissue and cells using RIPA buffer with 20 % protease inhibitor cocktail (0.4 mM aprotinin, 25 mM bestatin, 7.5 mM E64, and 10 mM leupeptin in H₂O). Protein concentration was quantified by a BCA assay kit (Epizyme, China). Proteins were separated by 10 % (Epizyme, PG112) or 15 % (Epizyme, PG114) sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) in a tris–glycine buffer system (Servicebio, G2027) and transferred to polyvinylidene fluoride membranes using a tris–glycine transfer buffer (Servicebio, G2145). The membranes were blocked for 10 min using a fast blocking buffer (Yoche, YWB0500) at room temperature and incubated overnight at 4 °C with primary antibodies. Membranes were treated with HRP-conjugated anti-mouse (Epizyme, LF101) or anti-rabbit (Epizyme, LF102) secondary antibodies for 1 h at room temperature. Finally, protein bands were detected with an enhanced chemiluminescence kit (Epizyme, China) and captured using an Amersham Imager 600 system (GE Healthcare).

The primary antibodies employed for western blotting were as follows: rabbit monoclonal S100B antibody (Abcam, ab52642, 1:2500), rabbit monoclonal FOXO1 antibody (Abcam, ab179450, 1:5000), rabbit monoclonal RAGE antibody (Abcam, ab216329, 1:1000), rabbit monoclonal β-catenin antibody (Abcam, ab32572, 1:7500), rabbit monoclonal MMP7 antibody (Abcam, ab207299, 1:1000), mouse monoclonal lamin B1 antibody (Proteintech, CL750-66095, 1:5000), and mouse monoclonal β-actin antibody (Proteintech, 20536–1-AP, 1:20,000).

Quantitative analysis of band intensity was performed using ImageJ software. The relative expression level of target proteins was calculated as the ratio of target protein band intensity to internal reference protein band intensity. Following three independent replicates, statistical analysis was conducted using GraphPad Prism software (version 8.0.2).

RNA extraction and RT-qPCR

An RNA purification kit (EZBioscience, B0004D) was used for RNA purification. The RNA concentration was determined using a microvolume UV–Vis spectrophotometer (Thermo Fisher Scientific, Waltham, MA, USA). cDNA was generated from 1000 ng of RNA using a reverse transcription kit (TAKARA, RR047, Shiga, Japan). A SYBR Green real-time PCR kit (Takara, RR820A, Shiga, Japan) was used to configure the 10-μL amplification system. The amplification procedure was performed using a 7500 fast real-time PCR system (Thermo Fisher Scientific). Primers are listed in Supplementary Table S2.

Cell co-culture

Direct co-culture: Ovarian cancer cells were co-cultured with HUVECs by first seeding HUVECs in a six-well plate. When the cell density reached approximately 30 %, a polycarbonate membrane insert (JERBIOFIL, China) with a pore size of 0.4 μm was placed onto the six-well plate. Ovarian cancer cells were then seeded on the insert. After 48 h of co-culture, the insert was removed, and HUVECs in the six-well plate were collected for subsequent experiments.

Acquisition of conditioned medium (CM): OVCAR3 or HEY ovarian cancer cells (3 × 10⁶) were cultured in a 10-cm dish with 8 mL of complete medium containing 10 % FBS and incubated at 37 °C in a cell incubator. After 24 h, the culture medium was removed, and the cells were washed three times with PBS. Subsequently, 8 mL of serum-free medium was added into the cell culture dish and incubated for 48 h. The collected supernatant was centrifuged at 1500 rpm for 15 min to remove debris and stored at − 80 °C.

HUVEC fibrin bead sprouting assay

Experiments and statistical analyses were conducted based on the existing protocol [33]. HUVECs were coated on pre-treated Cytodex 3 beads (GE, 17048501) and cultured overnight in EGM-2 medium (Lonza, CC3162). The following day, the HUVEC-coated beads were resuspended in a fibrinogen solution (Biosharp, BS943) containing aprotinin (MERK, A1153). Thrombin (MERK, T4648) was added to a 24-well plate, and the bead suspension was added to the same wells. After gentle mixing, the plate was left to stand at room temperature for 5 min and then incubated at 37 °C for 20 min to coagulate. Once the fibrin matrix had formed, an ovarian cancer cell suspension was added as droplets onto the fibrin-bead mixture. After 24–48 h, HUVEC sprouting was observed and captured under a microscope. The Image was quantitatively analyzed using the Image J software to quantify the areas of the total area and the beads area. The sprout area was finally obtained by subtracting the two: sprout area = total segmented area − beads area.

Tube formation assay

Endothelial cell tube formation assay is one of the most widely used and reliable methods for studying angiogenesis in vitro. We conducted experiments and statistical analyses based on existing protocols [34]. A 96-well plate was coated with 50 μL of Matrigel (Corning, 356234) and solidified at 37 °C for 15 min. HUVECs (1.5 × 10⁴ cells) were resuspended in serum-free medium with the following treatments: 1) 100 μL of CM from control/oe-S100B/Si-S100B ovarian cancer cells; 2) 100 μL of CM from control/oe-S100B/Si-S100B ovarian cancer cells + BEV; 3) 100 μL of serum-free F-12 K medium + S100B: and 4) 100 μL of serum-free F-12 K medium + S100B + BEV. The suspensions were added onto the Matrigel-coated wells and incubated at 37 °C. After 3 h, tube formation was observed and captured under a microscope. The images were quantitatively analyzed using the Angiogenesis Analyzer plugin in the ImageJ software. This plugin can automatically quantitatively analyze the number of nodes in tubular images for subsequent data statistics.

Cell counting kit-8 proliferation assay

HUVECs (1,000 cells per well) were seeded uniformly onto 96-well plates. After 24 h, the medium was removed, and wells were rinsed twice with PBS. Then, 100 μL of a 10 % CCK-8 reagent solution (NCM Biotech, C6005) was added to each well. Plates were cultured at 37 °C for 120 min, and the optical density (OD) at 450 nm was measured using a microplate reader. In subsequent experiments, the medium in the remaining wells was replaced with the following treatments: 1) 50 μL of CM from control/oe-S100B ovarian cancer cells + 50 μL of F-12 K medium containing 10 % FBS; 2) 50 μL of CM from control/oe-S100B ovarian cancer cells + 50 μL of F-12 K medium with 10 % FBS + BEV; 3) 100 μL of F-12 K medium with 10 % FBS + S100B; and 4) 100 μL of F-12 K medium with 10 % FBS + S100B + BEV. After 24, 48, and 72 h of incubation, the steps to obtain the OD values at 450 nm using the CCK-8 reagent were repeated.

Transwell migration assay

The experiment was conducted based on the existing protocol [35]. To shorten the experimental time and enhance the migration of endothelial cells, we changed the concentration of 10 % FBS in the protocol to 20 %. Briefly, 500 μL medium with 20 % FBS was added to each well of a 24-well plate. An 8-μm pore size cell culture insert (Corning, 3422) was placed into each well, and 1 × 10⁴ HUVECs were resuspended in serum-free medium with the following treatments: 1) 200 μL of CM from control or oe-S100B ovarian cancer cells; 2) 200 μL of CM from control or oe-S100B ovarian cancer cells + BEV; 3) 200 μL of serum-free F-12 K medium + S100B; and 4) 200 μL of serum-free F-12 K medium + S100B + BEV. The cell suspensions were dripped into the inserts and incubated at 37 °C in a cell incubator. After 24 h, the 24-well plate was removed, and the medium was aspirated. The inserts were fixed with 4 % paraformaldehyde (Servicebio, G1101) at room temperature for 20 min. After removing paraformaldehyde (Servicebio, G1101) and rinsing twice with PBS, cells were stained with 1 % crystal violet at room temperature for 15 min. Following three PBS washes, the non-migrated cells remaining on the upper side of the membrane were gently removed with a cotton swab. Migrated cells on the lower side of the membrane were observed and captured under a microscope. ImageJ software was employed to quantitatively analyze the images.

Wound healing assay

The wound healing test using pipette tips was conducted according to the existing protocol [35]. HUVECs (2 × 10⁴ cells) were seeded onto a 24-well plate. After 24 h, when the cells reached 90 %–100 % confluency, a wound was created using a 10-μL pipette tip. The wells were washed twice with PBS to remove detached cells, and the following serum-free media treatments were added: 1) 500 μL of CM from control/oe-S100B/Si-S100B ovarian cancer cells; 2) 500 μL of CM from control/oe-S100B/Si-S100B ovarian cancer cells + BEV; 3) 500 μL of serum-free F-12 K medium + S100B; and 4) 500 μL of serum-free F-12 K medium + S100B + BEV. The plates were incubated at 37 °C in a cell incubator. Images of the wound area were observed and captured under a microscope at 0 and 24 h. ImageJ software was used to quantitatively analyze wound closure.

Enzyme-linked immunosorbent assay (ELISA)

The experiment was conducted according to the protocol of Indirect ELISA [36]. OVCAR3 and HEY ovarian cancer cells (4 × 10⁶) were cultured in 10-cm dishes with 8 mL of complete medium with 10 % FBS. After 48 h, the cell culture supernatant was collected and centrifuged at 1500 rpm for 15 min to remove cell debris. The treated supernatant was stored at − 80 °C and used within 3 months. S100B and VEGFA levels in the supernatant were measured using S100B (Elabscience, E-EL-H1297) and VEGFA (Elabscience, E-EL-H0111) ELISA kits. For each assay, an enzyme-labeled plate was pre-coated with the corresponding antibody and incubated at 37 °C for 90 min. Then, 100 μL of cell culture supernatant was added to each well. A biotin-labeled antibody working solution was then added and the supernatant was incubated at 37 °C for 60 min. An HRP enzyme conjugate working solution was added and incubated at 37 °C for 10 min. After five washes, a substrate solution (TMB) was added and incubated at 37 °C for 15 min in the dark. Finally, the reaction was terminated by adding a stop solution, and the OD 450 was measured using a microplate reader. The concentrations of S100B and VEGFA were calculated based on a standard curve.

Cellular immunofluorescence

Harvest HUVECs during the logarithmic growth phase by trypsinization and resuspend in complete medium supplemented with 10 % FBS. Seed the cell suspension into 24-well plates pre-loaded with sterile coverslips (JingAn, J24001). Incubate at 37 °C under 5 % CO₂ for 24 h to allow cell adhesion and monolayer formation. Treat cells with exogenous recombinant S100B protein for 48 h. Aspirate the culture medium and wash cells twice with PBS. Fix with 4 % paraformaldehyde (Servicebio, G1101) for 20 min at room temperature. Rinse twice with PBS, air-dry briefly, and store plates at − 20 °C for up to one month prior to staining.Gently remove the cell slides from storage and permeabilize with 0.5 % Triton X-100 (Servicebio, GC204003) in PBS for 5–10 min at room temperature. Block nonspecific binding sites by incubating with 1 % BSA (Servicebio, GC305010) in PBS for 1 h at room temperature. Incubate with the following primary antibodies: Anti-FOXO1 (Invitrogen, MA5-17078, 1:150), Anti-β-catenin (Abcam, Cat# ab32572, 1:50). Perform incubation overnight at 4 °C in a humidified chamber. After washing 3 × with PBS, incubate with a mixture of HRP-conjugated secondary antibodies for 1 h at room temperature: HRP-goat anti-rabbit IgG (Servicebio, GB28301), HRP-goat anti-mouse IgG (Servicebio, GB25301). Wash slides 3 × with PBS, then stain nuclei with DAPI for 5 min. Carefully dry slides, apply anti-fade mounting medium (Servicebio, G1401), and cover with a coverslip. Image using a fluorescence microscope.

RNA interference

Cells in logarithmic growth phase were trypsinized and resuspended in complete medium supplemented with 10 % FBS. The cell suspension was seeded into 6-well plates and allowed to grow until reaching 50–60 % confluence prior to transfection. For transfection complex preparation, 5 μL of Lipofectamine 3000 (Invitrogen, L3000001) was diluted in 125 μL Opti-MEM (Gibco, 51985091) and incubated for 5 min at room temperature. In parallel, 5 μL of siRNA was diluted in 125 μL Opti-MEM. The diluted siRNA was then combined with the Lipofectamine 3000 mixture and incubated for 20 min at room temperature to allow complex formation. Before transfection, the culture medium was aspirated and cells were washed three times with PBS, followed by addition of 1.75 mL fresh Opti-MEM per well. The transfection complexes were added dropwise to the wells, and cells were maintained in culture without medium replacement. After 48 h of incubation, both cells and culture supernatants were collected for subsequent analysis.

Statistic analysis

Data were analyzed and visualized with GraphPad Prism (v.8.0.2) or R (v.3.6.0). T test and one-way ANOVA were used to compare the difference between two groups and multiple groups, respectively. Survival analysis was carried out using the Kaplan–Meier method, and compared using a log-rank test. A P-value ≤ 0.05 indicated statistical significance.

Result

Establishment and analysis of BEV-sensitive and BEV-resistant ovarian cancer xenograft models

Establishment of BEV-sensitive and BEV-resistant xenograft models

Because BEV is used in combination with chemotherapy drugs, it is difficult to define the reactivity of BEV solely based on tumor size and recurrence. Therefore, current studies on BEV resistance mainly begin with animal models using BEV monotherapy [37,38]. We used tumor-bearing mice to simulate the response of ovarian cancer to BEV monotherapy. By the second week after tumor cell injection, all mice had developed tumors. Fifteen mice were randomly divided into two groups: the control group, treated with normal saline, and the BEV treatment group (Fig. 2 A). The control group consisted of five mice, and the BEV treatment group had 10 mice. During the experiment, two mice in the control group died unexpectedly from hypothermia after anesthesia, leaving 13 mice for subsequent experiments. Tumor growth in the remaining mice was monitored, as shown in Fig. 2 B. In the early phase of BEV treatment (weeks 4–6), tumor growth in the BEV treatment group was significantly inhibited compared to that in the control group. However, from weeks 7 to 9, tumor growth resumed and was no longer inhibited, reaching a point at which the mice could no longer tolerate the tumor. To explore the changes in the ovarian cancer tissue at different times during BEV treatment, tumor tissue was collected from mice euthanized at three specific time points (Fig. 2 C): (1) the control group at week 5 (tumor growth was not inhibited), (2) the BEV group at week 6 (tumor growth was significantly inhibited), and (3) the BEV group at week 9 (tumor growth resumed). Additionally, tumors in which growth was inhibited during BEV treatment were defined as the sensitive group, while tumors that continued to grow despite ongoing BEV treatment were categorized as the resistant group.

Fig. 2.

Establishment and analysis of BEV-sensitive and BEV-resistant ovarian tumor-bearing mice. (A) Flowchart of establishing the BEV-sensitive and BEV-resistant ovarian tumor-bearing mice. (B) The line graphs of tumor fluorescence intensity evaluated by small animal in vivo imaging in mice of the control group and bevacizumab treatment group. (C) Tumor fluorescence images of tumor formation in mice (week 2), and tumor tissue harvested in control mice (week 5), BEV-sensitive mice (week 6), and BEV-resistant mice (week 9). (D) CD31 and HIF-1α immunofluorescence staining patterns of control, BEV-sensitive, and BEV-resistant mouse tumor tissue. (E) Tumor microvascular density and (F) HIF-1α staining intensity of control, BEV-sensitive, and BEV-resistant mouse tumor tissue. (G) DEG thermogram of RNA sequencing in BEV-sensitive and BEV-resistant mouse tumor tissue. (H) A bar chart of Metascape for pathway enrichment of DEG obtained from RNA sequencing of tumor tissues in BEV-sensitive and BEV-resistant mice. The forest map of the (I) OS and (J) PFS was analyzed by univariate COX regression analysis of DEGs (identified from the BEV-sensitive and BEV-resistant mouse tumors) using sequencing data from patients receiving BEV combination treatment based on the GSE140082 dataset. BEV:Bevacizumab；DEGs:Differentially expressed genes; *: P < 0.05; **: P < 0.01; ***: P < 0.001. Scale: All panels are 100 μm.

BEV inhibits tumor growth by suppressing tumor angiogenesis, and previous studies indicated that hypoxia, resulting from an extensive reduction of blood vessels, may be a major cause of BEV resistance. Hypoxia promotes the reactivation of angiogenesis by stimulating the secretion of downstream angiogenic factors. The vascular endothelial cell marker CD31 is commonly used to assess MVD. Therefore, we detected CD31 and the hypoxia marker HIF-1α in ovarian cancer tissue from the three experimental groups using immunofluorescence. The results indicated that expression of CD31 (Fig. 2 D, E) and HIF-1α (Fig. 2 D, F) in ovarian cancer tissue from the BEV-sensitive group was significantly lower than that in the control group. In contrast, the expression of CD31 (Fig. 2 D, E) and HIF-1α (Fig. 2 D, F) in tumor tissue from the BEV-resistant group was significantly higher than that in the BEV-sensitive group.

RNA sequencing and DEG enrichment analysis of tumor tissue

To further explore the mechanisms of BEV resistance in ovarian cancer, RNA sequencing was performed on the control, BEV-sensitive, and BEV-resistant tumor tissue. Using a threshold of |log2(fold change)| ≥ 2 and P ≤ 0.05, 323 DEGs were identified between the control group and the BEV-resistant group, with 272 genes significantly upregulated and 51 genes significantly downregulated in the resistant group (Supplementary Fig. S1A). Similarly, applying the same thresholds, 238 DEGs were identified between the BEV-sensitive and BEV-resistant groups, of which 194 genes were upregulated and 44 genes downregulated in the BEV-resistant group (Fig. 2 G). Enrichment analysis using the Metascape database revealed that these DEGs were predominantly involved in cellular functions such as “tube morphogenesis,” “positive regulation of cell adhesion,” “extracellular matrix,” “calcium ion binding,” and “epithelial-mesenchymal transition” (Supplementary Fig. S1B, Fig. 2 H).

Development and evaluation of the BEV-related prognostic signature for ovarian cancer

Screening DEGs associated with prognosis in ovarian cancer patients treated with BEV

As previously mentioned, 238 DEGs were identified through RNA sequencing in BEV-sensitive and BEV-resistant mouse tumor tissues. GSE140082 dataset was employed to further identify genes associated with prognosis in ovarian cancer patients treated with BEV. This dataset is derived from a phase III clinical trial comparing the efficacy of chemotherapy alone versus chemotherapy combined with BEV in patients with epithelial ovarian cancer [5,26]. A total of 380 patients were included, with 181 receiving chemotherapy alone and 199 receiving chemotherapy combined with BEV. A univariate Cox regression analysis was conducted on the 238 DEGs to assess their association with OS and PFS in patients receiving BEV combination chemotherapy. Sixteen genes were significantly associated with OS (Fig. 2 I), and 22 genes correlated with PFS (Fig. 2 J). Notably, high expression of VIL1, XYLT1, PDLIM3, FOLH1, and S100B significantly correlated with poor OS and PFS in these patients (Fig. 2 I, J).

Development and validation of a BEV-related prognostic signature based on machine learning

To construct the BRPS for ovarian cancer, data of 199 patients who received BEV combination chemotherapy in the GSE140082 dataset were randomly divided into two groups at a 7:3 ratio. This resulted in 139 patients for training and 60 for internal validation. Due to the lack of other datasets containing both prognostic and tissue sequencing data for ovarian cancer patients treated with BEV, we selected GSE72951, which includes data of 44 patients with high-grade glioma treated with BEV combined with chemotherapy, as an external validation set.

To predict long-term outcomes for BEV-treated patients, we used 16 genes that were significantly associated with OS based on a univariate Cox regression analysis in a machine learning framework for signature construction with the training set. The C-index was used to assess the consistency between the risk rankings predicted by the signature and the observed outcomes. A higher C-index indicates more accurate predictions [39]. The C-index for each signature, constructed using different combinations of machine learning methods, was calculated for the training set, internal validation set, and external validation set and sorted by the average C-index across the three datasets (Fig. 3 A).

Fig. 3.

BEV treatment-related prognosis signature of ovarian cancer was constructed using a combination of machine learning methods. (A) Heat maps based on calculating the C-index of the prognostic models in the training and validation sets and ranking the models by mean. Kaplan–Meier prognostic analysis of risk scores performed using the (B) GSE140082 training set, (C) GSE140082 validation set, and (D) GSE72951 validation set. ROC curves were analyzed for 1, 2, and 3 years using the (E) GSE140082 training set, (F) GSE140082 validation set, and (G) GSE72951 validation set. (H) Univariate and (I) multivariate COX regression analyses were performed for risk score, stage, grading, and patient age using the complete GSE140082 dataset. (J) Prognostic nomogram for patients in the GSE140082 global dataset using risk scores and staging. (K) Decision curve analysis of the nomogram, staging, and risk score; and (L) 1-year, 2-year, and 3-year ROC curve analysis of the nomogram using the GSE140082 master dataset. BEV:Bevacizumab; Lasso:Least Absolute Shrinkage and Selection Operator; RSF:Random Survival Forest; ROC:Receiver Operating Characteristic.

The signature constructed by combining LASSO with RSF, which had the highest average C-index (C-index = 0.749), and the signature produced by combining CoxBoost with Enet (α = 0.6), which had relatively consistent C-index values across the three datasets (C-index = 0.772, 0.762, and 0.615, respectively), were selected for further analysis. The signature constructed by combining LASSO and RSF consisted of 11 genes. Supplementary Fig. S2A illustrates the gene selection process using the LASSO method. The λ (regularization coefficient) value that minimized the likelihood deviation was chosen (left panel of Supplementary Fig. S2A). After removing variables with a regression coefficient of 0, the remaining 11 genes were used to develop the signature (right panel of Supplementary Fig. S2A). Supplementary Fig. S2B shows the regression coefficients of the 11 genes selected through the LASSO regression. These 11 variables were subsequently ranked by importance using the RSF method (Supplementary Fig. S2C).

Based on this signature, we generated the risk scores for each sample in the training set and the two validation sets. Kaplan–Meier survival analysis was conducted to compare the prognostic differences between patients with high- and low-risk scores. It was found that patients with high-risk scores exhibited poor OS based on data in both the GSE140082 training set (Fig. 3 B, P < 0.001) and the validation set (Fig. 3 C, P = 0.038), whereas a trend toward a poor prognosis was observed in the GSE72951 independent validation set (Fig. 3 D, P = 0.080). Time-dependent ROC curves were further used to estimate the predictive performance of the prognostic signature for patient survival. For 1-year, 2-year, and 3-year OS, the area under the curves (AUCs) based on the GSE140082 training set were 0.960, 0.971, and 0.968, respectively (Fig. 3 E) and the AUCs based on the GSE140082 validation set were 0.998, 0.722, and 0.580, respectively (Fig. 3 F). Because the OS of high-grade glioma patients was shorter than 3 years, we evaluated OS at 0.5, 1, and 2 years using ROC curves. The results showed that the AUCs based on the GSE72951 validation set were 0.680, 0.612, and 0.599, respectively (Fig. 3 G). These findings indicated that the signature demonstrated strong prognostic value in both the ovarian cancer training and validation datasets of patients treated with BEV. Even for high-grade gliomas, the signature provided meaningful predictive insight into the overall survival of BEV-treated patients. Using the same method, we subsequently performed Kaplan–Meier survival and ROC curve analyses of the signature constructed by using CoxBoost combined with Enet (α = 0.6). However, in both the training set and the two validation sets, there was inferior performance compared to the signature constructed using LASSO combined with RSF (Supplementary Fig. S3).

Based on the BRPS, risk scores were calculated on the data of all the 199 patients in the GSE140082 dataset who received BEV combination chemotherapy. The patients were grouped into high-risk and low-risk groups, and GSEA was then performed to identify enriched pathways between the groups. The results revealed that pathways such as “TGF-β signaling,” “epithelial mesenchymal transition,” “hypoxia,” “angiogenesis,” and “apical junction” were significantly enriched in the high-risk group, while pathways such as “oxidative phosphorylation” and “MYC targets” were involved in the low-risk group (Supplementary Fig. S4).

Improvement of predictive efficacy in patients treated with BEV based on BRPS and clinical characteristics

To elucidate the relationship between multiple prognostic clinical factors and the BRPS, we conducted univariate and multivariate Cox regression analyses based on OS using the GSE140082 dataset, which included FIGO stage, grade, patient age, and risk score. The results indicated that a later FIGO stage, age ≥ 65 years, and high risk score were risk factors for OS in ovarian cancer patients, while higher grade was associated with an improved OS (Fig. 3 H). A further multivariate Cox regression analysis revealed that only FIGO stage and risk score were significantly associated with OS (Fig. 3 I). To enhance the accuracy of survival prediction, we developed a nomogram incorporating both FIGO stage and risk scores (Fig. 3 J). Time-dependent ROC analysis demonstrated that the nomogram had strong predictive power, with AUC values of 0.974, 0.906, and 0.850 for 1-year, 2-year, and 3-year OS, respectively (Fig. 3 L). Additionally, decision curve analysis showed that the nomogram provided a higher net benefit for predicting OS compared to both the risk score and FIGO stage. The net benefit of the risk score was higher than that of FIGO staging and only slightly lower than that of the nomogram (Fig. 3 K). These results suggested that the BRPS outperformed the FIGO stage for predicting the prognosis of ovarian cancer patients treated with BEV, and the combined use of FIGO stage and the BRPS risk score provided a more accurate prediction of patient outcomes.

S100B regulates ovarian cancer sensitivity to BEV in a VEGFA-independent manner

Screening S100B as the main molecule regulating BEV sensitivity in ovarian cancer

To further screen the genes among the 11 BRPS genes that independently predicted the therapeutic effect of BEV, data of 380 patients of the GSE140082 dataset were divided into two groups with high or low gene expression levels, and the differences in the PFS and OS between patients treated with non-BEV chemotherapy and BEV combination chemotherapy were analyzed. The results indicated that, in patients with low expression of PDLIM3, SHANK2, and S100B, BEV combination therapy significantly improved OS and PFS compared to non-BEV chemotherapy alone. In contrast, patients with high expression of these genes did not benefit from BEV combination therapy (Fig. 4A, B; Supplementary Fig. S5B, D). Univariate Cox regression analysis of OS and PFS in patients receiving BEV combination chemotherapy revealed that SHANK2 was a risk factor for OS, but not PFS, while PDLIM3 and S100B were significantly associated with both OS and PFS (Fig. 2 I, J). Between these two, the hazard ratio (HR) for S100B was the highest (OS: HR = 2.77 [1.23–6.25], P = 0.014; PFS: HR = 2.05 [1.16–3.63], P = 0.013).

Fig. 4.

Screening S100B as the main molecule regulating the sensitivity of ovarian cancer to BEV. (A) Kaplan–Meier survival analysis of OS (left) and PFS (right) of patients with high S100B expression treated with chemotherapy with or without BEV (from the GSE140082 dataset). (B) Kaplan–Meier survival analysis of OS (left) and PFS (right) of patients with low S100B expression treated with chemotherapy with or without BEV (from the GSE140082 dataset). (C) Western blot analysis of S100B protein expression in tumor tissue from BEV-sensitive and BEV-resistant mice. (D) Western blot analysis of S100B protein levels in mouse tumor tissue. (E) Immunohistochemical staining of S100B protein in BEV-sensitive and BEV-resistant mice. (F) Statistical histochemical staining of S100B protein in mouse tumor tissue. BEV:Bevacizumab; **: P < 0.01.

Focusing on S100B, we first analyzed its expression using single-cell RNA sequencing data from GSE118828, GSE154600, GSE184880, and GSE189955. It was found that S100B expression was generally low in ovarian cancer and that S100B was predominantly expressed in myeloid cells, with a lower level detected in tumor cells, endothelial cells, and T cells (Supplementary Fig. S6A-D). Next, we used western blotting and IHC to further assess S100B protein expression in ovarian cancer tissue from BEV-sensitive and BEV-resistant mice, which showed a significant increase in S100B expression in tissue from BEV-resistant mice (Fig. 4 C, D). IHC staining revealed that S100B was primarily localized in the cytoplasm of ovarian cancer cells, with a smaller amount also detected in the nucleus (Fig. 4 E, F).

Ovarian cancer cells overexpressing S100B enhance angiogenesis and migration of HUVECs

S100B is a small EF-hand calcium-binding protein that primarily exhibits cytokine-like activity as a secreted protein. To investigate the role of S100B in ovarian cancer angiogenesis, we first evaluated its baseline expression across three ovarian cancer cell lines (A2780, HEY, and OVCAR3). Quantitative analysis revealed significantly higher S100B expression in A2780 cells compared to HEY and OVCAR3 (Fig. 5A). Based on this finding, we selected the low-expressing HEY and OVCAR3 cell lines to establish S100B-overexpressing models (Fig. 5B). We subsequently examined the impact of S100B overexpression on alternative angiogenic factors using PCR analysis. Notably, among the factors assessed (VEGFB, VEGFC, PDGFA, PDGFB, FGF-2, and HGF), only HGF demonstrated a modest but detectable increase in OVCAR3-S100B cells (Fig. 5 C, D). The expression levels of other angiogenic factors remained unaltered in both S100B-overexpressing cell lines, suggesting that S100B-mediated angiogenesis regulation in ovarian cancer operates independently of these canonical pro-angiogenic pathways. ELISA was then performed to measure the levels of S100B and VEGFA in the cell culture supernatant. The results showed that S100B levels significantly increased in the culture supernatants of OVCAR3 and HEY cells overexpressing S100B (Fig. 5 E), while the level of the angiogenic factor VEGFA remained unchanged (Fig. 5 F). We next co-cultured these two cell lines with HUVECs and collected the HUVECs 48 h later for phenotypic assays. The results revealed that, compared to the control group, HUVEC tube formation (Fig. 5G, H) and sprouting (Fig. 5 I, J) were significantly enhanced when co-cultured with S100B-overexpressing ovarian cancer cells, regardless of the presence of BEV. Sprouting angiogenesis is a crucial mechanism of blood vessel formation in which tip cells migrate toward hypoxic areas and stalk cells proliferate to form new vessels. To further investigate how this mechanism is affected by BEV and S100B, we examined the proliferation and migration of HUVECs. A Transwell migration assay (Fig. 5 K, L) and wound healing assay (Fig. 5 M, N) showed significantly enhanced migration of HUVECs co-cultured with ovarian cancer cells overexpressing S100B. However, no significant effect on HUVEC proliferation was observed (Supplementary Fig. S7 A, B). These findings suggest that ovarian cancer cells overexpressing S100B promoted endothelial angiogenesis in a VEGFA-independent manner, and this effect was not influenced by BEV treatment.

Fig. 5.

Co-culture of S100B overexpressing ovarian cancer cells promotes HUEVC angiogenesis and migration. (A) Representative Western blot images and their densitometric quantification showing comparative S100B protein expression profiles among the three ovarian cancer cell lines (A2780, HEY, and OVCAR3) (B) Western blot analysis and statistical analysis of OVCAR3 and HEY cells overexpressing S100B. RT-qPCR analysis of alternative angiogenic factor mRNA expression levels in (C) OVCAR3 and (D) HEY ovarian cancer cell lines following S100B overexpression. (E) The level of S100B in the supernatant of OVCAR3 and HEY cells overexpressing S100B significantly increased, as determined by ELISA. (F) There was no significant change in VEGFA levels in the supernatant of OVCAR3 and HEY cells overexpressing S100B. (G, H) HUVEC tube formation significantly increased after co-culture with ovarian cancer cell lines overexpressing S100B, with or without exogenous BEV. (I, J) Co-culture with ovarian cancer cells overexpressing S100B significantly promoted HUVEC sprouting with or without exogenous BEV. (K, L) Co-culture of ovarian cancer cells overexpressing S100B significantly promoted migration of HUVECs in a Transwell system with or without exogenous BEV. (M, N) Co-culture of ovarian cancer cell lines overexpressing S100B significantly promoted healing of HUVECs with or without exogenous BEV in a scratch assay. BEV: Bevacizumab; NC: negative control; OE: overexpression; *: P < 0.05; **: P < 0.01; ***: P < 0.001; ****: P < 0.0001. Scale: All panels are 50 μm.

Based on the high endogenous expression of S100B in A2780 cells, we established an S100B-knockdown model using siRNA (Supplementary Fig. 8A). Among three siRNA constructs, Si-3 demonstrated the most efficient knockdown and was selected for subsequent experiments. ELISA analysis confirmed significant reduction in secreted S100B protein levels in the conditioned medium from S100B-knockdown A2780 cells (Supplementary Fig. 8B). Notably, in both BEV-treated and untreated conditions, HUVECs co-cultured with S100B-knockdown A2780 cells showed no significant differences in tube formation capacity (Supplementary Fig. 8C, D) or wound healing migration (Supplementary Fig. 8E, F) compared to controls. These results suggest that under VEGFA-blockade conditions, compensatory mechanisms mediated by alternative angiogenic factors may maintain endothelial cell function despite S100B reduction. This observation implies that S100B downregulation alone is insufficient to impair angiogenesis when canonical alternative pathways remain active.

Exogenous S100B promotes tube formation and migration of endothelial cells

To further elucidate the role of S100B as a cytokine in promoting endothelial cell tubularization, we treated HUVECs with varying concentrations of recombinant S100B protein. The results demonstrated a dose-dependent effect of recombinant S100B on HUVECs, irrespective of BEV treatment. Specifically, S100B concentrations of 100 ng/mL or higher significantly enhanced HUVEC tubularization (Fig. 6 A, B), sprouting (Fig. 6 C, D), migration (Fig. 6 E, F), and wound healing (Fig. 6 G, H). Consistent with the co-culture experiments with ovarian cancer cells, exogenous S100B did not significantly affect HUVEC proliferation (Supplementary Fig. 7C–E).

Fig. 6.

Exogenous recombinant S100B protein promoted vascular formation and migration of HUEVCs. Different concentrations of recombinant S100B protein, with or without BEV, promoted (A, B) angiogenesis in a tube formation assay, (C, D) sprouting in a fibrin bead sprouting assay, (E, F) migration in a Transwell assay, and (G, H) healing of endothelial cells in a wound healing assay. (I) Tip cell marker expression in HUVECs significantly increased by rt-qPCR detection after treatment with recombinant S100B protein. BEV: Bevacizumab; *: P < 0.05; **: P < 0.01; ***: P < 0.001; ****: P < 0.0001. Scale: All panels are 50 μm.

During sprouting angiogenesis, endothelial cells differentiate into highly migratory tip cells and proliferative stalk cells. Tip cells, in particular, guide blood vessel growth by sensing angiogenic signals and migrating toward targets using pseudopod extensions [40]. We showed that S100B promoted HUVEC migration, but not proliferation. Next, to investigate the potential involvement of S100B in endothelial differentiation, we conducted a preliminary analysis of tip cell markers. The results revealed significant upregulation of the tip cell markers CD34, VEGFR3, NRP1, NRP2, EFNB2, ROBO4, and CXCR4 in HUVECs treated with recombinant S100B protein (Fig. 6 I).

Summarizing, these results suggested that S100B functions as a cytokine, directly promoting the tube formation and migration of endothelial cells, and it has the potential to induce differentiation of endothelial cells into tip cells.

The angiogenesis-promoting mechanism of S100B depends on endothelial cell endocytosis

S100B promotes endothelial angiogenesis in a RAGE-independent manner

The advanced glycosylation end product receptor (RAGE) is the most prominent receptor of the S100 protein family, and the S100B/RAGE signaling pathway is the primary mechanism through which S100B exerts its effects. In the developing cerebellum, RAGE functions as a signaling molecule that is essential for axon growth and cell migration [9]. To explore the role of RAGE in endothelial cells, we treated HUVECs with exogenous S100B protein and assessed RAGE. Activated RAGE can induce the transcription of specific RNA molecules by activating downstream signaling pathways, which in turn increases RAGE protein expression on the cell membrane [41]. Consequently, we measured RAGE protein expression levels in HUVECs 48 h after recombinant S100B protein treatment. The results showed that there was no significant change in RAGE protein expression following the exogenous addition of recombinant S100B protein (Fig. 7 A).

Fig. 7.

S100B enters endothelial cells through clathrin-dependent endocytosis to promote angiogenesis and migration. (A) Western blot detection showed that there was no significant change in RAGE receptor expression in endothelial cells after exogenous S100B treatment. (B) Western blot detection showed that protein levels of S100B in endothelial cells significantly increased after co-culture with ovarian cancer cells overexpressing S100B. (C) S100B total protein and nuclear protein in endothelial cells significantly increased after treatment with exogenous S100B, as determined by western blot analysis. (D) Western blot analysis of protein levels of S100B in endothelial cells after pretreatment with the clathrin-mediated endocytosis inhibitor Pitstop-2 or caveolae/caveolin-1-mediated endocytosis inhibitor nystatin. Changes in the (F, G) tubular phenotype, (H, I) sprouting phenotype, (J, K) migration phenotype (Transwell assay), and (L–M) wound healing phenotype (scratch migration assay) of endothelial cells after treatment with the RAGE receptor inhibitor FPS-ZM1, the clathrin-mediated endocytosis inhibitor Pitstop-2, and the caveolae/caveolin-mediated endocytosis inhibitor nystatin. BEV: Bevacizumab; *: P < 0.05; **: P < 0.01; ***: P < 0.001; ****: P < 0.0001. Scale: All panels are 50 μm.

To further investigate whether the effect of S100B on HUVECs is dependent on RAGE activation, we treated the cells with the RAGE-specific inhibitor FPS-ZM1. The results demonstrated that recombinant S100B protein consistently promoted angiogenic and migratory responses in HUVECs, regardless of FS-ZM1 treatment concentration. In functional assays, S100B significantly increased tube formation complexity (measured by node number; Fig. 7E, F) and enhanced sprouting angiogenesis (quantified by sprouting area; Fig. 7G, H). Furthermore, it robustly stimulated cell migration, as evidenced by elevated Transwell migration counts (Fig. 7I, J) and accelerated wound closure in scratch assays (Fig. 7K, L). These findings suggested that the tubulogenic and migratory effects of exogenous S100B protein on HUVECs were independent of the RAGE receptor.

S100B acts through clathrin-mediated endocytosis

As a secreted protein, S100B binds to cell membrane receptors to directly activate downstream signaling pathways, and it also enters cells via endocytosis, where it undergoes sorting in early endosomes to initiate additional signaling events [42,43]. Western blot analysis revealed a significant increase in S100B expression in HUVECs co-cultured with S100B-overexpressing ovarian cancer cells (Fig. 7 B). Furthermore, with exogenous recombinant S100B protein, the total S100B protein levels in HUVECs significantly increased, along with a notable increase in the nuclear accumulation of S100B (Fig. 7 C).

Proteins primarily enter cells through endocytosis, and clathrin-mediated endocytosis is the most common and well-studied endocytosis mechanism in mammals [44]. Of the various clathrin-independent endocytic pathways, caveolae/caveolin-1-dependent endocytosis is the predominant pathway in endothelial cells and is thought to mediate endocytosis across the endothelial barrier [45]. To identify the primary mode of S100B uptake in endothelial cells, HUVECs were pretreated with different concentrations of the clathrin-dependent endocytosis inhibitor Pitstop-2 or the caveolae/caveolin-1-dependent endocytosis inhibitor nystatin, followed by treatment with exogenous recombinant S100B protein. It was observed that Pitstop-2 significantly inhibited the expression of S100B in HUVECs at a concentration of 10 μM, while nystatin did not significantly affect S100B expression (Fig. 7 D). Phenotypic analysis of HUVECs demonstrated that pretreatment with 10 μM Pitstop-2 markedly suppressed exogenous recombinant S100B-induced tubular formation (Fig. 7E, F), sprouting (Fig. 7G, H), Transwell migration (Fig. 7I, J), and wound healing rate (Fig. 7K, L). In contrast, pretreatment with 10 μM or 40 μM nystatin did not significantly alter tube formation (Fig. 7E, F), sprouting (Fig. 7G, H), Transwell migration (Fig. 7I, J), or wound healing (Fig. 7K, L) in HUVECs. These findings suggested that S100B entered endothelial cells via clathrin-dependent endocytosis, thereby promoting vascular formation and endothelial cell migration.

S100B promotes angiogenesis through the FOXO1/β-catenin signaling pathway

S100B promotes nuclear translocation of β-catenin by inhibiting FOXO1 expression in HUVECs

To further elucidate the mechanism of S100B protein on HUVECs, a differential proteomic analysis was performed using DIA proteomic sequencing of HUVECs treated with exogenous S100B protein and the control group. A total of 7,963 proteins were identified. Using a threshold of |log(fold change)| ≥ 1.0, 115 differentially expressed proteins were detected, including 27 upregulated proteins and 88 downregulated proteins (Fig. 8 A). The level of S100B protein significantly increased in HUVECs following treatment with exogenous recombinant S100B (Fig. 8 A). Further enrichment analysis of the differentially expressed proteins suggested that they were associated with “endothelial cell migration” and “FOXO-mediated transcription of cell death genes.” (Fig. 8 B). Additionally, the proteins were enriched in pathways related to “positive regulation of endocytosis,” “endosomal transport,” and “regulation of protein polymerization.” (Fig. 8 B). These enrichment results provided further evidence that exogenous S100B protein may enter endothelial cells via endocytosis to exert its effects. To further investigate the downstream target proteins of S100B that promote angiogenesis, we preliminary evaluated six differentially expressed proteins associated with blood vessel formation based on a review of the literature: MAPK11, CRY1, FOXO1, GAB1, LGALS3BP, and INHBA. It was observed that the mRNA expression of FOXO1 significantly decreased in HUVECs treated with exogenous S100B (Supplementary Fig. 9 A), while the mRNA expression of LGALS3BP significantly increased (Supplementary Fig. 9A). Among the six proteins, only FOXO1 expression was consistent with the proteomic data. Subsequently, we examined the levels of S100B and FOXO1 protein in mouse tumor tissue from the BEV-sensitive and BEV-resistant groups. We found that S100B levels were significantly higher in BEV-resistant tumor tissue compared to those in BEV-sensitive tissue, while the level of FOXO1 in BEV-resistant tumor tissue tended to decrease (Fig. 8 C–E).

Fig. 8.

S100B promotes tube formation and migration of endothelial cells through the FOXO1/β-catenin signaling pathway. (A) DIA proteomics thermogram of differentially expressed proteins in HUVECs with or without exogenous S100B. (B) Function and pathway enrichment analysis by Metascape database of differentially expressed proteins. (C–E) Western blot analysis showed that levels of S100B in ovarian cancer tissue of BEV-resistant mice significantly increased, while FOXO1 levels decreased. (F) Histogram of an RT-qPCR analysis of downstream transcriptional target genes of β-catenin. (G–J) Western blot analysis showed that FOXO1 total protein and nuclear protein levels in endothelial cells significantly decreased after treatment with exogenous S100B, while levels of β-catenin and MMP7 significantly increased. (K–M) Western blot analysis of FOXO1 and β-catenin levels in S100B-treated endothelial cells overexpressing FOXO1 compared with the control group. Endothelial cells transfected with a control virus or FOXO1-overexpressing lentivirus and treated with S100B were evaluated for their (N) tubular phenotype by tube formation assay, (O) sprouting phenotype by fibrin bead sprouting assay, and (P) migration phenotype by Tranwell assay. (Q) Schematic diagram of the mechanism of exogenous S100B on endothelial cell angiogenesis. *: P < 0.05; **: P < 0.01;^: P < 0.05 vs. S100B−/oeFOXO1−; #: P < 0.05 vs. S100B−/oeFOXO1+; &: P < 0.05 vs. S100B+/oeFOXO1 − . Scale: All panels are 50 μm.

FOXO1 is a member of the Forkhead box (FOXO) transcription factor family, which functions as a context-dependent tumor suppressor in various cancers [43]. FOXO1 serves as a critical regulator of vascular endothelial cell biology, regulating multiple functions including cellular metabolism [46,47], proliferation [46] and migration [48], vascular homeostasis maintenance [49], permeability regulation [50], and angiogenesis modulation [51]. Previous studies showed that FOXO1 maintained endothelial cells in a quiescent state by downregulating MYC and its downstream targets, thereby promoting vascular stability and inhibiting angiogenesis [46]. Additionally, cytoplasmic FOXO1 was shown to inhibit the downstream activation of β-catenin by binding to it, preventing β-catenin from entering the nucleus [52,53], or competitively binding to Transcription factor (TCF) and β-catenin in the nucleus [54]. To explore these two potential downstream mechanisms of FOXO1, we conducted preliminary PCR screening. The results revealed that the expression of downstream transcriptional target genes of β-catenin—such as CCND1, MMP2, MMP7, MMP9, WISP1, and Survivin—significantly increased in HUVECs treated with S100B (Fig. 8 F), while MYC expression remained unchanged. Similarly, when we examined the downstream target proteins of MYC, we found that mRNA expression of CDK4, FASN, ENO1, PKM2, LDHB, and LDHA showed no significant changes. However, the expression of the MYC negative regulatory gene MIX1 significantly increased (Supplementary Fig. 9B). These results suggested that the transcriptional activity of β-catenin was activated in HUVECs treated with exogenous S100B, while the expression of MYC and its downstream targets did not change significantly. Notably, although MYC is a downstream target of β-catenin, its expression did not increase significantly, which may be attributed to the elevated expression of its inhibitory factor, MIX1, thus counteracting β-catenin-mediated transcriptional activation of MYC. These findings were consistent with our previous phenotypic studies, which suggested that exogenous S100B significantly promoted HUVEC migration but had no notable effect on proliferation.

Our western blot results indicated that the expression of FOXO1 in both the cytoplasm and nucleus of HUVECs treated with exogenous S100B significantly decreased (Fig. 8 G, H). In contrast, the accumulation of β-catenin in the nucleus significantly increased (Fig. 8 G, I). Furthermore, the protein level of MMP7, a downstream target of β-catenin, also significantly increased (Fig. 8 G, J). To investigate the potential interaction between FOXO1 and β-catenin in endothelial cells, we performed immunofluorescence co-localization analysis in HUVECs. As demonstrated in Supplementary Fig. 10 A, both FOXO1 and β-catenin exhibited predominant cytoplasmic localization with detectable nuclear expression. Quantitative analysis of fluorescence intensities revealed a strong positive correlation between FOXO1 and β-catenin expression patterns (Supplementary Fig. 10B). These suggest significant spatial co-localization of FOXO1 and β-catenin in endothelial cells, supporting the possibility of direct molecular interaction between these two proteins. These findings suggested that S100B may promote endothelial cell migration and angiogenesis by downregulating FOXO1, facilitating the nuclear translocation of β-catenin, and activating its downstream transcriptional targets.

Overexpression of FOXO1 partially reversed the angiogenic effect of S100B

To further confirm that S100B promoted endothelial cell migration and tube formation in a FOXO1-mediated manner, we overexpressed FOXO1 in HUVECs using lentiviral vectors, followed by treatment with S100B. Experiments assessing tube formation (Fig. 8 N), sprouting (Fig. 8 O), and migration (Fig. 8 P) demonstrated that FOXO1 overexpression partially counteracted the promotive effects of S100B. Additionally, the expression of β-catenin in HUVECs overexpressing FOXO1 was not significantly affected by S100B treatment and was markedly lower than that of the control group (Fig. 8 K–M).

Immunofluorescence staining was employed to comprehensively evaluate the effects of exogenous S100B protein on FOXO1 and β-catenin expression dynamics in endothelial cells. The results revealed that treatment with exogenous recombinant S100B protein significantly reduced FOXO1 expression (Supplementary Fig. 10C, D), while increasing both β-catenin expression (Supplementary Fig. 10C, E) and its nuclear accumulation (Supplementary Fig. 10C, F) in HUVECs. However, in FOXO1-overexpressing HUVECs, neither β-catenin expression nor nuclear localization showed significant changes following S100B treatment compared to controls (Supplementary Fig. 10C-F), consistent with our previous findings.

Summarizing, in BEV-resistant ovarian cancer tissue, the expression and secretion of S100B were elevated. Exogenous S100B protein entered endothelial cells through clathrin-mediated endocytosis, where it downregulated FOXO1 expression, thereby promoting the nuclear translocation of β-catenin and activating downstream transcriptional processes. This mechanism enhanced endothelial cell tube formation and migration (Fig. 8 Q).

An S100B inhibitor effectively enhanced the response to BEV in mice with ovarian cancer

Pentamidine is the most widely used and well-established inhibitor of S100B, which functions by competitively binding S100B and other target proteins to inhibit their activity [55]. To investigate whether pentamidine enhanced BEV effectiveness against ovarian cancer, we compared the efficacy of BEV, pentamidine, and their combination in abdominal tumor-bearing mice with ovarian cancer. The experimental workflow is outlined in Fig. 9 A. Two weeks after tumor cell injections, after the tumors had formed, the mice were randomly grouped into four groups. BEV was administered at a dosage of 10 mg/kg via intraperitoneal injection twice a week, pentamidine was administered at a dosage of 10 mg/kg via intraperitoneal injection three times a week, and the combination group received both treatments using the same dosage as the other two groups. The control group was treated with intraperitoneal injections of saline. Tumor growth was monitored weekly using a small animal bioluminescence imager. Mice were euthanized when they lost more than 20 % of their body weight or could no longer tolerate the tumor burden. Fig. 9 B shows the tumor growth curves of the four groups. Fig. 9 C shows tumor luminescence images at different time points: at the beginning of treatment (2 weeks), upon euthanasia of mice in the control and pentamidine-only treatment groups (4–6 weeks), upon euthanasia of mice in the BEV treatment group (8–9 weeks), and upon euthanasia of mice in the BEV with pentamidine treatment group (13–14 weeks). Compared to the control group, pentamidine alone had no significant effect on tumor growth and led to premature euthanasia of mice between the fourth and sixth weeks due to an intolerable tumor burden. However, both the BEV treatment and BEV with pentamidine treatment groups exhibited significant tumor inhibition by the fourth week, although tumors began to regrow after that time. At week 7, significant differences in tumor growth were observed between the BEV treatment group and the BEV with pentamidine treatment group. By weeks 8 and 9, mice in the BEV treatment group were euthanized due to tumor intolerance, while mice in the BEV with pentamidine treatment group continued to survive until weeks 13–14. Kaplan–Meier survival analysis of the four groups revealed that the survival time of mice in the BEV with pentamidine treatment group was markedly longer than that of mice in the BEV treatment group (Fig. 9 D).

Fig. 9.

The S100B inhibitor pentamidine in combination with BEV improves the ovarian cancer response compared to BEV alone. (A) Flowchart of the in vivo experiment. Two weeks after intraperitoneal inoculation of ovarian cancer cells, drug treatment was administered. The mice were randomly divided into a normal saline treatment control group, BEV treatment group, pentamidine treatment group, and BEV with pentamidine treatment group. (B) Line chart of the tumor fluorescence intensity of the four groups of mice. (C) Tumor fluorescence imaging of mice after tumor formation (week 2), after killing the control group and pentamidine treatment group (week 4–6), after killing the BEV treatment group (week 8–9), and after killing the BEV with pentamidine treatment group (week 13–14). (D) Kaplan–Meier survival curves of the four groups of mice. (E) CD31 immunohistochemical staining and (F) MVD statistics of tumor tissue from the four groups of mice. (E) S100B immunohistochemical staining and S100B (G) staining intensity of tumor tissue from the four groups of mice. BEV: Bevacizumab; MVD: Micro-vessel density; *: P < 0.05; **: P < 0.01;^: P < 0.05 vs. NC; #: P < 0.05 vs. pentamidine; &: P < 0.05 vs. BEV.

Immunohistochemical staining of S100B and CD31 in tumor tissue from the four groups revealed that pentamidine had no significant inhibitory effect on S100B expression. Compared to the control and pentamidine treatment groups, long-term BEV treatment significantly increased S100B expression (Fig. 9 E, F). The MVD in the BEV with pentamidine treatment group was significantly lower than that in the other three groups. However, pentamidine alone had no inhibitory effect on tumor angiogenesis (Fig. 9 E, G).

Discussion

Ovarian cancer is one of three major gynecological malignancies in women, characterized by a high degree of malignancy and poor prognosis, with a 5-year survival rate of only 46 %[3]. While the use of BEV has been shown to significantly improve PFS in ovarian cancer patients, it does not offer long-term survival benefits [5,56]. Drug resistance is the primary factor limiting the long-term efficacy of BEV [6]. BEV resistance differs from conventional chemotherapy resistance mechanisms. As an anti-angiogenic agent, BEV uniquely alters tumor perfusion, affecting oxygen and nutrient supply, drug delivery, and molecular transport. These vascular changes induce adaptive responses across multiple components of the tumor microenvironment, collectively creating a complex resistance landscape [[57], [58], [59], [60], [61]]. In ovarian cancer, BEV resistance arises from multiple mechanisms including: (1) overproduction of alternative angiogenic factors, (2) activation of non-sprouting angiogenesis pathways, (3) tumor cell adaptation to microenvironmental stress, and (4) contributions from stromal components and myeloid-derived suppressor cells [6,[62], [63], [64], [65], [66]]. However, the effects achieved by applying these drug resistance mechanisms to clinical or preclinical explorations for overcoming BEV resistance are very limited [8,[67], [68], [69]]. In this study, a BEV-related prognostic signature was developed based on an adaptive BEV resistance model in ovarian cancer mice, which identified S100B as a key molecule regulating BEV sensitivity in ovarian cancer. S100B is a secreted protein that enters endothelial cells through clathrin-mediated endocytosis, where it downregulates FOXO1 expression, promotes β-catenin accumulation, and activates downstream transcriptional processes, thereby enhancing endothelial cell angiogenesis and migration. Pentamidine, a specific inhibitor of S100B, in combination with BEV significantly inhibited tumor growth and prolonged the survival of mice.

Defining clinical resistance to BEV is challenging due to the potential confounding effects of chemotherapy drugs. Therefore, mouse models have become the primary tool for exploring the mechanisms underlying BEV resistance. Previous studies successfully established BEV monotherapy resistance models for lung cancer [37] and ovarian cancer [38]. In this study, we successfully induced adaptive resistance to BEV in ovarian cancer using a peritoneal tumor-bearing mouse model by administering 10 mg/kg BEV via intraperitoneal injection. During the first 4 weeks of BEV treatment, tumor growth was significantly inhibited compared to the control group. However, between 5 and 7 weeks of treatment, the inhibitory effect of BEV weakened, and the tumor growth rate significantly accelerated. In solid tumors, an imbalance between pro-angiogenic and anti-angiogenic factors leads to the development of an immature vascular network [57]. This aberrant vascular proliferation results in distorted, dilated, and poorly perfused blood vessels, creating a hypoxic microenvironment within the tumor. Anti-angiogenic therapies, such as BEV, effectively prune these inefficient blood vessels by eliminating excess endothelial cells, temporarily normalizing the vasculature and reducing hypoxia in the tumor [70]. However, as treatment resistance develops, active angiogenesis is re-initiated [6], eventually promoting tumor recurrence. This process is consistent with the observations of our BEV-resistant ovarian cancer xenograft model. Immunofluorescence staining of mouse tumor tissue revealed that, compared to the control group, BEV treatment significantly inhibited tumor growth and reduced the expression of HIF-1α (a hypoxia marker) and MVD during weeks 3–4. However, by week 7, BEV resistance developed, as evidenced by accelerated tumor growth and a significant increase in HIF-1α and MVD expression. Furthermore, RNA sequencing and enrichment analysis showed that the genes overexpressed in BEV-resistant tumor tissue were primarily involved in pathways related to tube morphogenesis and the epithelial-mesenchymal transition.

Drug resistance is the primary factor limiting the long-term efficacy of BEV [6]. Therefore, we aimed to identify key molecules based on resistance-related genes that impacted OS in ovarian cancer patients receiving BEV treatment, with the goal of constructing a prognostic signature for these patients. Using a combination of machine learning methods, we developed a BEV-related prognostic signature based on DEGs identified between BEV-sensitive and BEV-resistant mouse tumor tissues. A LASSO regression was used to preliminarily screen and simplify the list of DEGs, and RSF was used to rank and model genetic variables. Kaplan–Meier and ROC curve analyses showed that the signature effectively predicted the prognosis of ovarian cancer patients treated with BEV and demonstrated predictive value for glioma patients undergoing BEV therapy. Compared to the FIGO stage, tumor grade, and patient age, the risk score based on our prognostic signature had the highest predictive accuracy for patient OS. Currently, the prognostic signature of ovarian cancer is typically assessed using gene sets representing specific functions, such as collagen remodeling [71] and immune-related gene sets [72]. However, this approach does not accurately predict outcomes of patients undergoing specific treatments. In contrast, our signature, based on BEV resistance-related genes, provides a more precise prediction of long-term survival for BEV-treated patients and therefore helps clinicians develop a more accurate and individualized approach to patient management.

By further analyzing the 11 genes in the prognostic signature, we identified S100B as a potential independent marker for predicting the efficacy of BEV. When patients were classified into two groups, i.e., high and low S100B expression groups, we found that only patients with low S100B expression benefited from BEV combination therapy. As a secreted protein, S100B can be detected in serum and is primarily used as a diagnostic marker for brain injury [73] and melanoma [12]. In ovarian cancer, S100B may promote tumor progression by regulating tumor cell behavior [13]. However, few studies have investigated its relationship with anti-angiogenic therapies, and only one study suggested that S100B plasma levels may affect the response of melanoma to BEV [14]. Our study demonstrated that S100B expression was significantly elevated in BEV-resistant ovarian cancer tissue and correlated with patient responsiveness to BEV therapy. These findings highlighted that S100B modulated BEV sensitivity in ovarian cancer, a finding supported by both animal model data and large-sample clinical data analysis. However, for clinical application, further validation by collecting and analyzing clinical blood samples is necessary.

Sprouting angiogenesis is the main mode of angiogenesis and is largely driven by various growth factors (e.g., VEGFs) that stimulate the proliferation, differentiation, and migration of quiescent endothelial cells. Previous studies showed that S100B promoted VEGF secretion in different cell types [[19], [20], [21], [22]] and increased endothelial cell permeability [16]. However, our study found that the angiogenic effect of S100B on endothelial cells was not affected by a BEV-mediated VEGFA blockade. An ELISA analysis revealed no significant difference in VEGFA levels in the supernatant of ovarian cancer cells overexpressing S100B compared to the control group. These findings suggested that S100B promoted endothelial angiogenesis in a VEGFA-independent manner and therefore acted as an independent cytokine.

RAGE is the primary receptor for S100B [74]. The S100B/RAGE signaling pathway is the most important focus of the research on the mechanism of secretory S100B, which is implicated in the occurrence and development of melanoma and other tumors [75,76]. However, our results showed no significant change in RAGE protein expression in HUVECs after 48 h of S100B treatment. We then treated cells with different concentrations of the RAGE receptor-specific inhibitor FPS-ZM1 and found that regardless of concentration, FPS-ZM1 could not counteract the promotive effect of S100B on vascular endothelial cell formation and migration. This suggested that S100B regulated endothelial cells in a RAGE-receptor-independent manner. A previous study also found a RAGE-independent mode of action of exogenous S100B in mouse endothelial cells [77]. In addition to directly activating downstream signaling pathways via membrane receptor binding, extracellular proteins can also enter cells through receptor-mediated endocytosis [44]. Previous studies showed that exogenous S100B entered rat astrocytes through endocytosis [42]. Our experiment found that S100B expression in endothelial cells was significantly higher after being co-cultured with S100B-overexpressing ovarian cancer cells or after treatment with exogenous S100B. A proteomics analysis also confirmed that S100B expression in endothelial cells was significantly elevated after treatment with exogenous S100B, and the differentially expressed proteins were enriched in functions such as endocytosis (“positive regulation of endocytosis”) and vesicular transport (“endosomal transport”). In general, after entering cells through endocytosis, proteins dissociate from receptors in the endosomes and eventually are degraded by lysosomes. Therefore, protein endocytosis is often considered to downregulate signal transduction [78]. However, emerging evidence suggests that endocytosis also plays a critical role in signal transduction and amplification for many secreted proteins [79]. Notably, secreted proteins, such as the growth factors HGF, EGF, PDGF, and VEGF, undergo endocytosis along with their respective receptors, in which early endosomes regulate signaling pathways by sorting receptor-ligand complexes in both spatial and temporal dimensions [43]. Targeting common endocytic pathways has thus become a key focus in cancer research [80]. In mammals, endocytosis is primarily mediated by clathrin; however, of the clathrin-independent endocytosis mechanisms, caveolae/caveolin-1-mediated endocytosis is currently considered the predominant mode of endocytosis in endothelial cells [44]. Therefore, Pitstop-2, a clathrin-specific inhibitor, and nystatin, a caveolae/caveolin-1 inhibitor, were used to pretreat endothelial cells. We found that in endothelial cells treated with exogenous S100B, Pitstop-2, but not nystatin, significantly inhibited S100B expression, tube formation, and cell migration. Previous studies have demonstrated that clathrin-mediated endocytosis plays a regulatory role in angiogenesis. Specifically, VEGF stimulation promotes VEGFR endocytosis and trafficking through clathrin recruitment in endothelial cells. Research has revealed that sprouting endothelial cells in retinal vasculature exhibit enhanced VEGF uptake capacity, along with accelerated VEGF endocytosis and turnover rates. [81]. The internalized receptor retains its signaling capacity within intracellular compartments, where it continues to promote angiogenesis and endothelial sprouting through activation of the downstream ERK1/2 pathway [82,83]. Modulation of clathrin-mediated endocytosis regulatory proteins has been demonstrated to effectively control angiogenic processes [84,85]. Beyond its canonical functions, VEGF signaling modulates endothelial permeability and migration through endocytosis-mediated redistribution of membrane receptors [[86], [87], [88]]. However, the potential involvement of clathrin-dependent endocytic pathways in BEV resistance remains un-explored. Summarizing, our findings indicated that S100B promoted endothelial cell tube formation and migration via clathrin-mediated endocytosis, independent of its interaction with the classical RAGE receptor. This study provides preliminary evidence for the role of clathrin-mediated endocytosis in S100B-induced endothelial cell function.

To explore the mechanism of S100B on endothelial cells, we performed proteomic sequencing on HUVECs treated with exogenous S100B and on control HUVECs. We identified 115 differentially expressed proteins. “Endothelial cell migration” emerged as the most enriched pathway, further supporting our phenotypic results showing that S100B significantly promoted endothelial cell migration. PCR screening of the mRNA of six differentially expressed proteins associated with angiogenesis revealed that FOXO1 mRNA levels were significantly reduced in HUVECs treated with exogenous S100B, consistent with our proteomic findings. We analyzed the downstream target proteins of MYC and β-catenin based on the two known downstream mechanisms of FOXO1. The results showed that the expression of the downstream target protein of β-catenin significantly increased, while the expression of MYC and its downstream target protein was not significantly changed. Notably, the expression of the MYC inhibitor MIX1 significantly increased, indicating that S100B regulation of endothelial cell proliferation may be jointly affected by multiple pathways that ultimately maintain a stable expression level. FOXO1 is known to inhibit β-catenin nuclear entry and directly suppress its downstream transcription in the nucleus [53,54]. β-catenin, encoded by CTNNB1, is a crucial proto-oncogene that regulates cell proliferation, survival, epithelial-mesenchymal transition, migration, and metastasis by modulating the function of downstream transcription factors in the nucleus. It is dysregulated in nearly all stages of tumorigenesis, not only promoting angiogenesis by regulating the secretion of angiogenic factors and the function of vascular endothelial cells [89], but also contributing to therapy resistance through its regulation of tumor cell growth [90]. Therefore, we assessed protein expression in the cytoplasm and nucleus of HUVECs. After exogenous S100B treatment, we observed a significant decrease in FOXO1 expression in both the cytoplasm and nucleus, while β-catenin nuclear accumulation and the expression of its downstream target, MMP7, significantly increased. Further experiments showed that exogenous S100B did not increase β-catenin expression in endothelial cells overexpressing FOXO1. Additionally, FOXO1 overexpression partially counteracted the effects of S100B on endothelial cell tube formation, sprouting, and migration. Thus, our results suggested that exogenous S100B promotes endothelial cell tube formation and migration by downregulating FOXO1 expression, leading to increased nuclear accumulation of β-catenin and activation of its downstream transcription.

The anti-protozoal drug pentamidine exerts an inhibitory effect on S100B by competitively binding to its hydrophobic binding pocket [55]. In bladder cancer [91], glioma [92], and renal cell carcinoma [93], pentamidine was shown to inhibit tumor growth in vivo. However, its role in gynecological tumors remains unexplored. Our in vivo study results demonstrated that, compared to the control group, pentamidine alone did not inhibit the growth of ovarian tumors. However, when combined with BEV, pentamidine significantly reduced the tumor growth rate, prolonged the therapeutic effect of BEV, and markedly decreased the MVD of tumors. These findings suggested that S100B did not have a major role in endothelial cell tube formation in ovarian cancer when VEGFA was not inhibited. However, when VEGFA/VEGFR2 signaling was blocked by BEV, which significantly inhibited angiogenesis, S100B expression and secretion increased, promoting tube formation. In this context, the S100B inhibitor pentamidine effectively inhibited both tumor growth and angiogenesis. Notably, pentamidine has been widely used in the clinical treatment of pneumocystis pneumonia [94] and has demonstrated a favorable safety profile in patients, providing a strong basis for its potential application in ovarian cancer treatment.

In conclusion, this study integrated mouse models and publicly available clinical sample databases to develop a BEV-related prognostic signature. S100B appears to be a key molecule regulating the BEV response in ovarian cancer, and its specific inhibitor, pentamidine, enhanced the therapeutic efficacy of BEV. S100B promoted endothelial cell tube formation in a VEGFA-independent manner, which may be mediated by FOXO1/β-catenin activation following clathrin-mediated endocytosis of S100B into endothelial cells. Our findings uncovered a novel mechanism of S100B, provided new insights into the regulation of adaptive resistance to BEV in ovarian cancer, and suggested potential approaches for identifying patients who would benefit from BEV treatment, ultimately improving long-term outcomes for ovarian cancer patients receiving BEV therapy. While this study provides novel insights, certain limitations should be acknowledged. First, our clinical correlation analysis relied solely on public sequencing datasets. Future validation using collected clinical specimens (tumor tissues and matched blood samples) would strengthen the findings by enabling direct assessment of S100B histopathological expression and circulating cytokine levels. Second, the mechanistic details of S100B internalization remain to be fully elucidated − particularly how clathrin-mediated endocytosis facilitates S100B delivery to specific downstream targets in endothelial cells. This aspect requires systematic exploration to establish complete pathway coherence. Finally, advanced experimental approaches including matrigel plug assay and endothelial-specific knockout mouse models would provide more physiologically relevant validation of our findings in subsequent investigations.

Conclusions

In combination with transcriptome sequencing of BEV-sensitive and BEV-resistant mouse models and a public database, we constructed a signature that effectively predicts the prognosis of patients with ovarian cancer treated with BEV. Among the 11 signature genes, S100B was selected as the main role molecule in regulating BEV sensitivity of ovarian cancer. In vitro experiments have shown that exogenous S100B protein can promote tube formation, sprouting and migration of endothelial cells. Further mechanism research indicates that S100B enters endothelial cells through clathrin-mediated endocytosis and activates the downstream FOXO1/β-catenin pathway. S100B inhibitor Pentamidine collaborates with BEV to inhibit angiogenesis and tumor growth of ovarian cancer.

Fundings and acknowledgements

This work is supported by grants from National Natural Science Foundation of China (Grant No. 82072877). We thank LetPub (https://www.letpub.com.cn) for its linguistic assistance during the preparation of this manuscript.

Ethics statement

All experiments involving animals were conducted according to the ethical policies and procedures approved by the ethics committee of the Obstetrics and Gynecology Hospital, Fudan University, China (Approval no. 2024-FCKYY-217).

Compliance with ethics requirement

All Institutional and National Guidelines for the care and use of animals (fisheries) were followed. All experiments involving animals were conducted according to the ethical policies and procedures approved by the ethics committee of the Obstetrics and Gynecology Hospital, Fudan University, China (Approval no. 2024-FCKYY-217).

CRediT authorship contribution statement

Haoya Xu: Writing – original draft, Methodology. Wenzhi Li: Methodology. Huiran Yue: Methodology. Yang Bai: Methodology. Jun Li: Methodology. Xin Lu: Conceptualization, Funding acquisition. Jieyu Wang: Project administration, Supervision.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary material

The following are the Supplementary data to this article: