Overcoming the leptomeningeal seeding of medulloblastoma by targeting HSP70

Highlights

• •Established a serial xenograft-derived medulloblastoma leptomeningeal seeding cell model.
• •Identified HSP70 as a key molecular vulnerability in seeding cells.
• •VER155008 selectively inhibited seeding cell viability in vitro and in vivo.
• •Combination of VER155008 and ifosfamide showed synergistic antitumor effects.
• •Proposed a novel strategy to target leptomeningeal dissemination in medulloblastoma.

Abstract

Leptomeningeal seeding (LMS) via cerebrospinal fluid is a common and often fatal progression in medulloblastoma (MB), significantly worsening prognosis. However, its molecular drivers are poorly understood, and effective treatments remain limited. This study aimed to establish a physiologically relevant MB LMS model and identify novel therapeutic targets through detailed characterization of seeding cell biology.

Using three rounds of serial orthotopic xenograft transplantation and in vivo selection, we isolated distinct seeding (S3) and non-seeding (N3) MB cell populations. Functional and transcriptomic analyses revealed unique phenotypes and differentially expressed genes (DEGs). Based on DEGs, we screened inhibitors and assessed the therapeutic efficacy of the HSP70 inhibitor VER155008, alone and with chemotherapeutics (ifosfamide or cisplatin), both in vitro and in a preclinical LMS mouse model established by cerebellar implantation of S3 cells.

S3 cells showed slower proliferation, altered migration behavior, and increased adhesion to collagen IV versus N3 cells. Transcriptomic profiling identified HSP70 as the most upregulated gene in S3 cells, with strong enrichment in metabolic pathways. Among six candidate compounds, VER155008 most effectively suppressed S3 cell viability. In vitro, VER155008 combined with 4-hydroperoxycyclophosphamide (active ifosfamide metabolite) produced a synergistic antitumor effect. This synergy was confirmed in vivo, where VER155008 with ifosfamide significantly reduced spinal LMS.

These findings highlight HSP70 as a promising therapeutic target for MB LMS. The observed synergy between VER155008 and ifosfamide supports a selective combination strategy, offering a novel therapeutic avenue to improve outcomes in patients with leptomeningeal dissemination of MB.

Introduction

In 19b5, Harvey Cushing and Percival Bailey first described a distinct malignant cerebellar tumor in children, later termed medulloblastoma (MB). They notably identified its unique propensity to disseminate via the subarachnoid space along the brain and spinal cord surfaces—a process now known as leptomeningeal seeding (LMS) [1]. Approximately 30% of MB patients already exhibit LMS at diagnosis, a factor strongly associated with poor prognosis and markedly reduced survival [[2], [3], [4]]. To mitigate this devastating complication, patients often require craniospinal irradiation at higher doses and intensified chemotherapy regimens, which unfortunately increase the risk of severe long-term neurocognitive and systemic complications [5].

LMS is one of the most aggressive relapse patterns in MB, characterized by poor prognosis and limited therapeutic options. Tumor cells disseminated within the cerebrospinal fluid are exposed to substantial metabolic and environmental stress, necessitating reliance on adaptive stress-response pathways for survival. Recent studies suggest that stress-related chaperone systems play an important role in supporting tumor cell persistence under these conditions. Although cisplatin and ifosfamide remain standard components of medulloblastoma therapy, their efficacy against leptomeningeal disease is limited, underscoring the need to identify novel therapeutic vulnerabilities.

LMS represents a key metastatic mechanism underlying the majority of MB-related deaths. While molecular subgrouping of MB has significantly advanced our understanding of its inherent heterogeneity, the precise molecular drivers of LMS remain elusive. Among MB subgroups, Group 3 MB, characterized by MYC amplification, exhibits the highest LMS risk and worst prognosis [6]. Interestingly, despite this clinical correlation, transgenic models overexpressing c-MYC have failed to fully recapitulate extensive LMS, and critically, LMS can also occur in other MB subgroups lacking MYC alterations [7,8]. This underscores the need for more representative models and a deeper understanding of universal LMS mechanisms.

A major obstacle in LMS research has been the limited availability of human LMS tissue, as current clinical protocols do not routinely involve tumor sampling from CSF-disseminated sites. Consequently, orthotopic xenograft models have been developed to more faithfully mimic human LMS in vivo [9,10]. However, complementary in vitro models are also critically needed to elucidate both cell-autonomous and microenvironment-dependent mechanisms governing LMS, offering a controllable system for mechanistic research.

Unlike conventional orthotopic xenograft models, which primarily reflect primary tumor growth and generate leptomeningeal dissemination inconsistently, the serial orthotopic transplantation strategy used in this study applies physiologically relevant selection pressure within the CSF environment to enrich for LMS-competent tumor cells.

Tumor evolution is profoundly driven by genomic instability and selective pressures that collectively foster aggressive phenotypes [[10], [11], [12]]. Such pressures include nutrient scarcity, immune surveillance, intercellular competition, and therapeutic interventions [13,14]. Notably, the CSF constitutes a metabolically constrained niche, exerting substantial selective pressure on disseminated tumor cells and profoundly shaping their survival and adaptation strategies [15,16].

Based on this concept of selective pressures driving metastatic adaptation, we established clinically relevant MB seeding (LMS-enriched) cells via serial orthotopic transplantation to directly compare with non-seeding cells. Through comprehensive molecular and phenotypic analyses, we identified HSP70 as a critical vulnerability in seeding cells and demonstrated that its inhibition by VER155008, alone or combined with ifosfamide, effectively suppresses leptomeningeal dissemination in preclinical models.

Materials and methods

Cell culture

MB UW426-effLuc cell line, genetically modified to overexpress the enhanced firefly luciferase gene (effLuc) via retroviral infection [17], was maintained in Dulbecco’s Modified Eagle’s Medium (DMEM; Invitrogen, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (FBS; Invitrogen) and 1% penicillin-streptomycin (Invitrogen). Cells were incubated at 37°C in a humidified atmosphere of 5% CO₂ and 95% air. Mycoplasma contamination was routinely excluded using PlasmoTest (InvivoGen, San Diego, CA, USA).

Intracerebellar orthotopic MB xenograft models

To model leptomeningeal seeding under physiologically relevant conditions, we used an intracerebellar orthotopic xenograft approach that enables tumor cells to interact with the native cerebellar microenvironment and access the cerebrospinal fluid (CSF) space. To enrich for LMS-competent cells in a reproducible manner, we performed three rounds of serial in vivo selection by re-isolating spinal LMS cells and re-implanting them orthotopically.

All animal experiments were carried out at the animal facility of Seoul National University Hospital (SNUH), approved by the Institutional Animal Care and Use Committee (IACUC) at SNUH (No. 18-0272-C2A1), in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals (NIH publication no. 80-23). Female BALB/c-nude mice (6–7 weeks) were kept under specific pathogen-free conditions. The mice were anesthetized with an intraperitoneal injection of 20 mg/kg zoletil and 10 mg/kg xylazine. UW426-effLuc cells (1.2 × 10⁵) were suspended in 3 μL of PBS before injection. The entire injection procedure was performed as described previously [18], using stereotactic coordinates for the right cerebellar hemisphere: 1 mm lateral to the midline, 1 mm posterior to the coronal suture, and 3 mm depth [19]. Mice were monitored daily for neurological symptoms, including gait instability, hindlimb weakness or paralysis, as well as changes in general activity indicative of neurological impairment. At the experimental endpoint, brains and spinal cords were harvested for subsequent histopathological examination.

To establish the LMS-enriched mouse model and isolate seeding cell populations (Fig. 1A), mice were monitored for spinal LMS development using bioluminescence imaging (BLI). Upon detection of distal spinal cord LMS by BLI, cerebellar tumors (non-seeding cells, N1) and spinal LMS cells (seeding cells, S1) were separately isolated and cultured [[20], [21], [22], [23], [24]]. Once cell numbers reached 1 × 10⁷ cells, magnetic-activated cell sorting (MACS; Miltenyi Biotec, Bergisch Gladbach, Germany) was performed using monoclonal anti-CD90.1 conjugated to magnetic microbeads, as previously described [17]. The sorted UW426-effLuc cells (S1 and N1) were then re-injected orthotopically into naive mice (n = 3 per group). This serial transplantation and re-isolation process was repeated three times to generate the third-passage seeding (S3) and non-seeding (N3) cell populations (Fig. 1A).

Fig. 1.

Schematic representation of the establishment of a medulloblastoma leptomeningeal seeding (LMS) cell model. (A) The serial selection process for seeding cells was repeated three times. (B) A representative longitudinal section from a mouse brain to the spinal cord showing tumorigenesis and LMS status by H&E staining in a seeding model. Tumor tissues marked with a red box were excluded to obtain tumor tissues for isolation of seeding cells and non-seeding cells. (C) Cells isolated from each tumor tissue were selected through three rounds of MACS sorting, and their luciferase activity was verified. The graph shows the luciferase activity of the third (final) selected cells. 1.25 × magnification. (D) The third selected seeding cells (S3) demonstrated more rapid LMS progression compared to non-seeding cells (N3) in vivo. Data represent the average results from at least three independent experiments. Scale bars: 100 μm.

Histological analysis

At the experimental endpoint, mice were transcardially perfused with saline containing 2.5 U/mL heparin, followed by fixation with 4% paraformaldehyde under deep anesthesia. Brains and spinal cords were then decalcified using 10% EDTA (pH 7.2), dehydrated, embedded in OCT compound (Tissue-Tek®; Sakura, Tokyo, Japan), and sagittally sectioned into 10 µm-thick sections using a cryostat, as previously reported [17,25]. Frozen sections were stained with hematoxylin and eosin (H&E) and stored at -80°C for subsequent immunofluorescence analysis.

In vitro luciferase activity assay

Luciferase activity was quantified using a luciferase assay kit (Promega, Madison, WI, USA) as previously described [6].

In vivo bioluminescence imaging (BLI)

Non-invasive in vivo tumor growth monitoring was performed using BLI. Mice were intraperitoneally injected with D-Luciferin (150 mg/kg, Caliper Life Sciences, Hopkinton, MA, USA) according to the manufacturer's protocol, prior to BLI measurement. Images were acquired using an IVIS-100 imaging system (Xenogen Corp., Alameda, CA), with mice under anesthesia (2% isoflurane in 100% O₂ via nose cone). Luminescence signals from defined regions of interest (ROIs) were integrated for 1–3 minutes and quantified using Living Image software to assess the viability of implanted tumor cells [17].

Microarray analysis

Total RNA was isolated using the RNeasy Mini Kit (Qiagen, Hilden, Germany), quantified with a NanoDrop 2000, and validated using an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). cDNA was synthesized using the GeneChip WT (Whole Transcript) Amplification Kit (Affymetrix) according to the manufacturer’s instructions. Data were generated on the Affymetrix Human 2.0ST array platform. For microarray-based transcriptome screening, S3 and N3 cell populations were each established from pooled tumor tissues to obtain sufficient material and analyzed once per group.

For mRNA analysis, data were normalized using a robust multi-average (RMA) method implemented in Affymetrix® Power Tools (APT). Differentially expressed gene (DEG) analysis was performed using fold-change and independent t-tests, with a cut-off of log₂|Fold Change| > 1.5. The false discovery rate (FDR) was controlled by adjusting the P-value using the Benjamini-Hochberg algorithm. Hierarchical cluster analysis for DEG sets was conducted using complete linkage and Euclidean distance. Gene enrichment and functional annotation analyses for significant probe lists utilized Gene Ontology (http://geneontology.org) and KEGG (http://kegg.jp). All data analysis and visualization of differentially expressed genes were conducted using R version 3.3.2 (www.r-project.org).

Cell proliferation assay

Cells (4 × 10³ cells/well) were seeded in a 96-well CellCarrier-96 Ultra plate (PerkinElmer, Germany) and incubated at 37°C with 5% CO₂. Cell proliferation was monitored and quantified consecutively for 64 hours using a high-content screening (HCS) system (Operetta CLS, PerkinElmer). Cytotoxicity was simultaneously assessed using the HCS LIVE/DEAD Green kit (Thermo Fisher Scientific, Eugene, OR, USA), following the manufacturer’s instructions.

Cell viability assay

Cell viability was assessed using the EZ-cytox kit (DAEIL Lab, Seoul, Korea), following the manufacturer’s protocol. Cells (2 × 10³ cells/well) were seeded in a 96-well plate and cultured for 72 hours in a 5% CO₂ environment at 37°C. Ten microliters of EZ-cytox kit reagent were added to each well and incubated for 2 hours under standard culture conditions. Plates were gently shaken for 1 minute at room temperature before absorbance measurement. Optical density (OD) values at 540 nm were determined using a micro-ELISA reader (Molecular Devices, Sunnyvale, CA, USA). Data were expressed as a percentage of viability relative to non-seeding cells.

Cell wound healing/scratch assay

Cell wound healing/scratch assays were performed using wound-making pins by HCS, as previously described [26]. Cells (1 × 10⁴ cells/well) were seeded into a 96-well CellCarrier-96 Ultra plate (PerkinElmer) and grown to 90% confluence. A scratch was then made at the bottom of each well using a transfer device with a 0.9 mm diameter pin. Cells were rinsed 2-3 times with serum-free media and labeled with calcein AM (cell permeant dye) for 30 minutes. After two more rinses, cell culture media was added. Wound healing was observed and quantified using an HCS device at 12-hour intervals for up to 48 hours.

Cell migration assay

Cell migration assays were conducted using a Cytoselect 24-well cell migration kit (Cell Biolabs, Inc. San Diego, CA, USA) according to the manufacturer’s instructions. Cells (5 × 10⁴/ well) in 200 µL of serum-free medium were seeded into the upper chamber of a transwell apparatus. The lower chamber was filled with complete cell culture media. After 24 hours of incubation, non-migrated cells on the inside perimeter of the insert were washed and removed. Migratory cells were fixed, stained, and photographed using an inverted microscope. Each insert was then transferred to an empty well, and an extraction solution was added per well. After 10 minutes of incubation on a shaker, each sample’s absorbance was measured at OD560 nm in a microplate reader.

Cell adhesion assay

Cell adhesion assays were performed using Cytoselect 48-well plates (Cell Biolabs) following the manufacturer’s instructions. Briefly, cells were seeded onto the 48-well plate and incubated for 24 hours. Media was gently removed, and cells were washed 3-5 times before staining. After washing 3-5 more times and air-drying, plates were photographed. Subsequently, plates were incubated with extraction solution for 10 minutes on a shaker, and absorbance was measured at OD595 nm. Bovine serum albumin (BSA) served as a negative control.

siRNA-mediated knockdown of HSPA1A and HSPA1B

Small interfering RNAs (siRNAs) targeting HSPA1A and HSPA1B were purchased from Qiagen (Hilden, Germany). Two independent siRNA sequences per target gene were tested in naive, N3, and S3 cells (siRNA-HSPA1A-1/2 and siRNA-HSPA1B-1/2). Detailed information on siRNA product names, catalog numbers, and target sequences is provided in Supplementary Table S1.

Cells were transfected with siRNAs at final concentrations of 50–100 nM using Lipofectamine™ RNAiMAX (Invitrogen, Thermo Fisher Scientific) according to the manufacturer’s instructions. A non-targeting control siRNA was used as a negative control in all experiments.

Knockdown efficiency was assessed at 48 hours post-transfection, and cell viability was evaluated at 72 hours post-transfection. For viability assays, 4,000 cells per well were seeded in 96-well plates, and cell viability was measured using standard colorimetric assays. All experiments were performed with at least three independent biological replicates.

Drug treatment

VER155008, Apoptozole, HA15, TRC051384, JG98, Elesclomol, cisplatin (Selleckchem, Houston, TX. USA), and 4-hydroxyifosfamide (4HC, Cayman Chemical, Ann Arbor, MI, USA) were prepared in dimethyl sulfoxide (DMSO). For in vitro experiments, 4HC the active metabolite of ifosfamide was used instead of the parent drug, since it does not require metabolic activation and is thus suitable for direct application in cell-based assays. Drugs were administered at varying concentrations and incubated for 72 hours. The 50% inhibitory concentration (IC₅₀) values were calculated using GraphPad Software [25].

Drug combination assay

The combined effect of VER155008 with conventional anti-cancer drugs (4HC and cisplatin) was determined by calculating the combination index (CI) using CompuSyn software (Paramus, NJ, www.combosyn.com.), based on the Chou-Talalay method. Combined drug doses were prepared based on the half maximal inhibitory concentration (IC₅₀) of each individual drug, specifically at 5 constant ratios: 0.25 × IC₅₀, 0.5 × IC₅₀, 1.0 × IC₅₀, 2.0 × IC₅₀, and 4.0 × IC₅₀. Synergy, additivity, and antagonism were defined as CI < 1, CI = 1, and CI > 1, respectively. A fraction affected (Fa) value < 0.5 was considered irrelevant [25].

In vivo drug treatment and imaging

Three days after orthotopic implantation of S3 cells, for the primary analysis of combination therapy effects in vivo, tumor-bearing mice were randomly assigned to four key experimental groups (n = 5 per group) prior to drug administration. The groups included: vehicle control, VER155008 monotherapy, VER155008 combined with cisplatin, and VER155008 combined with ifosfamide. VER155008 was administered intratumorally as a single injection at a dose of 2 mg/kg in 5 μL. Cisplatin (4 mg/kg) and ifosfamide (100 mg/kg) were administered intraperitoneally (i.p.), with VER155008 administered concurrently with the first dose of either cisplatin or ifosfamide. For in vivo studies, ifosfamide the prodrug form was used, as it undergoes hepatic metabolic activation in vivo to produce its cytotoxic metabolite (4HC). The dosing of VER155008 (25 mg/kg) and 4HC was determined based on previously published preclinical studies demonstrating antitumor efficacy with acceptable tolerability [27,28]. Mice were monitored daily for general condition, neurological symptoms, and changes in body weight throughout the treatment period. Tumor burden and treatment response were assessed using a non-invasive bioluminescence imaging system (IVIS-100, Xenogen Corp.). Luminescence signals from the regions of interest (ROIs) were quantified using Living Image software to evaluate the viability of implanted tumor cells, as previously described [17].

Statistical analysis

Statistical significance was assessed using Student’s t-test or one-way ANOVA. All in vitro experiments were independently repeated at least three times, and data are presented as mean ± SD. For in vivo studies, group sizes are indicated in figure legends. P-values < 0.05 were considered statistically significant.

Results

Establishment and medulloblastoma (MB) leptomeningeal seeding (LMS)cells

We established an orthotopic xenograft model to investigate MB LMS. Mice were monitored for tumor growth and LMS development using in vivo BLI. Upon detecting distal spinal cord LMS by BLI, we separately isolated and cultured spinal LMS cells (designated as seeding cells, S1) and cerebellar tumors (non-seeding cells, N1). These first-passage cells (S1 and N1) were then orthotopically re-injected into naive mice for a second round of selection. This serial transplantation process was repeated three times to generate third-passage seeding (S3) and non-seeding (N3) cell populations (Fig. 1A). Through repeated in vivo selection, the resulting S3 subline showed consistent and reproducible leptomeningeal dissemination, indicating reduced heterogeneity across passages. The designation of seeding (S1, S2, S3) and non-seeding (N1, N2, N3) cells was based on their serial passage and isolation location.

Histological examination of the whole neuraxis in the established model confirmed widespread tumor cell dissemination along the leptomeninges, from the subfrontal area to the conus medullaris (Fig. 1B). There was no significant difference in baseline enhanced luciferase signals between S3 and N3 cells in vitro (Fig. 1C). To compare their LMS-forming capabilities, S3 and N3 cells were orthotopically injected into naive mice, and tumor distribution was evaluated by BLI. BLI demonstrated that at day 9, spinal cord signals were already detectable in S3-injected mice, while no spinal dissemination was observed in N3-injected mice at this time point. By day 16, S3 cells exhibited stronger and more extensive signals extending to the lower spinal cord compared to N3 cells (Fig. 1D), although both eventually resulted in widespread LMS. Of note, no mouse mortality was observed in either S3 or N3 groups up to Day 21.

Biological function analysis for differentially expressed genes (DEGs) in LMS cells

Microarray analysis was performed to identify differentially expressed genes (DEGs) between S3 (seeding) and N3 (non-seeding) cells. Our analysis profiled 44,629 human genes, from which 2,594 mRNAs were identified as DEGs, with 1,219 upregulated and 1,374 downregulated in S3 cells compared to N3 cells (Fig. 2A). The top 10 upregulated mRNAs in S3 cells included HSPA1B, HSPA1A, DNAJA1, HSPA8, DNAJB1, APOC1, DNLZ, GSTM5, REC114, and ZNF415, while the top 10 downregulated mRNAs were KLRC3, ASNS, ARRDC4, ATF3, ULBP1, DDIT4, CTH, TXNIP, SLC7A11, and HORMAD1 (Table 1). Notably, HSPA1A and HSPA1B, two paralogous genes encoding the heat shock protein 70 (HSP70), were among the most highly upregulated mRNAs (top 2-fold change). Their coding regions are intron-less and nearly identical [29].

Fig. 2.

Heatmaps and canonical pathway analysis of seeding cells (S3) vs non-seeding cells (N3). (A) Hierarchical clustering of S3 and N3 populations based on differentially expressed genes (DEGs). Each column and row represent a sample and a transcript, respectively, with gene expression levels depicted according to a color scale. (B) Top 20 canonical pathways predicted to be enriched based on genes showing differential expression between S3 and N3 are shown.

Table 1.

Differentially expressed genes in seeding cells (S3) vs non-seeding cells (N3).

Gene_Symbol	Gene_ID	mRNA Accession	S3/N3.fc	Gene Accession	Gene Description
HSPA1B	3304	NM_005346	7.701289	NM_005346	heat shock 70kDa protein 1B
HSPA1A	3303	NM_005345	7.520299	NM_005345	heat shock 70kDa protein 1A
DNAJA1	3301	NM_001314039	3.128780	NM_001314039	DnaJ (Hsp40) homolog, subfamily A, member 1
HSPA8	3312	NM_006597	3.097083	NM_006597	heat shock 70kDa protein 8
DNAJB1	3337	NM_001300914	2.696915	NM_001300914	DnaJ (Hsp40) homolog, subfamily B, member 1
APOC1	341	NM_001645	2.299404	NM_001645	apolipoprotein C-I
DNLZ	728489	NM_001080849	2.287196	NM_001080849	DNL-type zinc finger
GSTM5	2949	NM_000851	2.252910	NM_000851	glutathione S-transferase mu 5
REC114	283677	NM_001042367	2.236248	NM_001042367	REC114 meiotic recombination protein
ZNF415	55786	NM_001136038	2.137523	NM_001136038	zinc finger protein 415
KLRC3	3823	NM_002261	-2.953263	NM_002261	killer cell lectin-like receptor subfamily C, member 3
ASNS	440	NM_001178075	-3.191711	NM_001178075	asparagine synthetase (glutamine-hydrolyzing)
ARRDC4	91947	NM_183376	-3.267959	NM_183376	arrestin domain containing 4
ATF3	467	NM_001030287	-3.469866	NM_001030287	activating transcription factor 3
ULBP1	80329	NM_025218	-3.580348	NM_025218	UL16 binding protein 1
DDIT4	54541	NM_019058	-3.639853	NM_019058	DNA damage inducible transcript 4
CTH	1491	NM_001190463	-3.954239	NM_001190463	cystathionine gamma-lyase
TXNIP	10628	NM_001313972	-3.995234	NM_001313972	thioredoxin interacting protein
SLC7A11	23657	NM_014331	-4.285549	NM_014331	solute carrier family 7 (anionic amino acid transporter light chain, xc- system), member 11
HORMAD1	84072	NM_001199829	-10.341475	NM_001199829	HORMA domain containing 1

Gene Ontology (GO) enrichment analysis, specifically focusing on the DEGs from the S3 vs N3 comparison, classified these genes into various functional categories. As shown in Fig. 2B, 20 significantly enriched GO terms were identified in biological processes, with metabolic pathways being the most significant term in seeding cells (P < 0.05, FDR < 0.01).

To provide a broader context for the distinct molecular profiles observed in S3 and N3 cells, we further analyzed the gene expression patterns across all serially passaged populations. Initial two-way hierarchical clustering of all samples (NC, 1S, 1B, 2S, 2B, 3S, 3B) based on DEG expression demonstrated distinct molecular profiles corresponding to their passage number and seeding/non-seeding characteristics (Supplementary Fig. S1A and S1B). This clustering indicated that the serial selection process effectively differentiated the molecular signatures of the various cell populations. Furthermore, a comprehensive Gene Ontology (GO) functional category analysis performed across all these serially isolated cell populations (NC, 1S, 1B, 2S, 2B, 3S, 3B) revealed distinct patterns of enriched biological processes and pathways, indicating that the selection process led to varied molecular characteristics at each stage of LMS establishment (Supplementary Fig. S1C). This broader analysis complements the specific S3 vs. N3 comparison by illustrating the dynamic molecular adaptation during LMS establishment.

Characterization of seeding cells (S3) compared to non-seeding cells (N3)

Morphologically, no significant difference was observed between S3 and N3 cells (Fig. 3A). We then utilized a high-content screening (HCS) system for real-time cell function analysis. While N3 cells exhibited a continuous increase in proliferation after 24 hours, the proliferation rates of S3 cells were less than 50% of N3 cells, with S3 cell growth at 64 hours being one-third of N3 cells (Fig. 3B, P = 0.0014). Cell viability showed no significant difference between S3 and N3 (Supplementary Fig.S2).

Fig. 3.

Phenotypic differences between LMS-derived seeding (S3) and non-seeding (N3) cells. (A) Morphological differences were not observed between S3 and N3 under bright field microscope. (B) The graph shows that the proliferative capacity of S3 is slower than that of N3s through High-Content Screening (HCS) analysis. (C) The trans-well migration assay reveals that there was no significant difference in the migration ability between S3 and N3 (quantification graph). (D) Representative images from the trans-well migration assay. (E) Representative images of the wound-healing assay show that N3 filled the open area faster than S3. The white area is saturated with green color due to the overlapping of S3 without filling the wound gap. (F) Quantification graph of the wound-healing assay. (G) Adhesion of S3 and N3 cells to extracellular matrix (ECM) components, including fibronectin, collagen I, collagen IV, laminin I, and fibrinogen, was quantified after 24h of incubation. (H) Adhesion values were normalized to control conditions. All data are presented as mean ± SD from n = 3 independent experiments. Statistical comparisons between two groups were performed using a two-tailed Student’s t-test. *P < 0.05, **P < 0.01, ***P < 0.001.

To assess metastatic characteristics, we compared differences in migration, motility and adhesion. No significant difference was observed in cell migration between S3 and N3 (Fig. 3C and 3D). The in vitro wound-healing assay (cell motility) revealed that N3 cells consistently filled the scratched space over time. In contrast, S3 cells did not effectively fill the scratched space and instead appeared to detach from the monolayer over time (12h–36h: P < 0.001, 48h: P < 0.05; Fig. 3E, 3F, and Supplementary Table S2). Adhesion ability was evaluated on various extracellular matrix (ECM) components (fibronectin, collagen I, collagen IV, laminin I, and fibrinogen). S3 cells exhibited significantly better adhesion on collagen I (P < 0.01) and collagen IV (P < 0.001), while no significant differences were found on other ECMs (Fig. 3G and 3H).

Functional validation of HSP70 dependency by siRNA-mediated knockdown

To functionally validate the transcriptomic upregulation of HSP70 family members in leptomeningeal seeding cells, we performed siRNA-mediated knockdown of HSPA1A and HSPA1B. Consistent with the transcriptomic screening results, baseline mRNA expression levels of HSPA1A and HSPA1B were significantly higher in S3 cells than in N3 cells, as confirmed by quantitative real-time PCR using independent biological replicates (n = 3; Supplementary Fig. S3).

Efficient suppression of target gene expression was confirmed following transfection in naïve, N3, and S3 cell populations (Fig. 4A, 4B). Genetic knockdown of HSPA1A or HSPA1B resulted in a significant reduction in cell viability, with a more pronounced effect observed in S3 cells compared with N3 cells (Fig. 4C, 4D). Notably, while HSPA1A knockdown consistently showed greater viability reduction in S3 cells, HSPA1B knockdown exhibited siRNA sequence–dependent effects, with one sequence showing a marked S3-specific reduction and another showing comparable effects between N3 and S3 cells. Exact quantitative values and statistical analyses for N3 versus S3 comparisons are provided in Supplementary Table S3.

Fig. 4.

Functional validation of HSP70 dependency by siRNA-mediated knockdown. (A, B) Relative mRNA expression levels of HSPA1A (A) and HSPA1B (B) following siRNA-mediated knockdown in naïve, non-seeding (N3), and leptomeningeal seeding (S3) cells. Expression levels were normalized to negative control siRNA and calculated using the 2^−ΔΔCt method. (C, D) Cell viability following siRNA-mediated knockdown of HSPA1A (C) or HSPA1B (D) in naïve, N3, and S3 cells. Viability was measured 72 h after transfection and normalized to negative control siRNA. Data are presented as mean ± SD from n = 3 independent experiments. Statistical comparisons between N3 and S3 cells were performed using a two-tailed Student’s t-test. ***P<0.001.

Selective vulnerability of seeding cells to HSP70 inhibition

To explore potential therapeutic strategies, we screened six commercially available HSP70 inhibitors against both S3 and N3 cells. VER155008, J98, HA15, and TRC051384 demonstrated preferential cytotoxicity toward S3 cells (Fig. 5A-F). TRC051384 was evaluated as a modulator of the HSP70 chaperone network, whereas HA15 was analyzed as a selective inhibitor of the ER chaperone BiP/GRP78 (HSPA5). Among the tested compounds, VER155008 showed a consistent dose-dependent cytotoxic effect and a marked differential response between LMS-derived seeding and non-seeding cells, which supported its selection for further investigation (Fig. 5A). Based on these findings, VER155008 was selected for further investigation. Treatment of S3 cells with VER155008 at its IC₅₀ concentration significantly decreased cell viability (Fig. 5G, Supplementary Table S4). Western blot analysis confirmed suppression of HSP70 protein expression and concomitant downregulation of multidrug resistance proteins MRP1 and MRP2 in S3 cells treated with VER155008 (Fig. 5H, 5I). These results suggest that HSP70 inhibition impairs both survival and resistance pathways in seeding cells. In contrast, treatment of naive, N1, and S1 cells with the HSP70 inhibitor VER155008 resulted in dose-dependent reductions in cell viability without marked differences among these early-stage populations, suggesting that HSP70 dependency is not prominent at initial stages of leptomeningeal progression (Supplementary Figure S4). Collectively, these findings demonstrate that pharmacologic inhibition of HSP70 selectively compromises the viability of seeding cells, supporting HSP70 as a key molecular vulnerability in this subpopulation.

Fig. 5.

Pharmacologic inhibition of HSP70 selectively suppresses viability of seeding cells. (A–F) Dose-response bar graphs showing the effects of six different HSP70 inhibitors on cell viability in non-seeding cells (N3, white bars) and seeding cells (S3, black bars) at the specified concentrations for 72 h. These panels demonstrate that S3 cells exhibit greater sensitivity to HSP70 inhibition compared to N3 cells. (G) Cell viability of S3 cells treated with VER155008 at its IC₅₀ concentration, showing a significant reduction compared to untreated controls. (H) Summary quantification graph of the six HSP70 inhibitors, highlighting relative potency in reducing S3 cell viability. (I) Western blot analysis of HSP70, MRP1, and MRP2 protein expression in S3 cells following VER155008 treatment. β-actin was used as a loading control. Data are shown as mean ± SD from n = 3 independent experiments. Statistical analysis was performed using one-way ANOVA followed by Tukey’s post hoc test. **P < 0.01, ***P < 0.001.

Enhanced antitumor effect of VER155008 and ifosfamide in seeding cells (S3)

To investigate the potential for augmenting conventional chemotherapy through HSP70 inhibition, we first evaluated the combinatorial effects of VER155008 with either 4-hydroperoxycyclophosphamide (4HC) or cisplatin in S3 cells. Single-agent treatments yielded IC₅₀ values of 35.24 ± 5.31 µM for VER155008, 34.50 ± 0.74 µM for 4HC, and 11.80 ± 0.21 µM for cisplatin.

Employing Chou-Talalay analysis, we observed distinct drug interactions depending on the chemotherapeutic agent. The combination of VER155008 with cisplatin demonstrated predominantly antagonistic interactions rather than true synergism. Dose-response curves (Fig. 6A) showed only a modest or minimal increase in cytotoxicity upon combination treatment compared to single agents. Correspondingly, combination index (CI) values (Supplementary Table S5) derived from the CI-Fa plot (Fig. 6B) were generally at or above 1 across most fractional effect (Fa) levels, suggesting moderate to strong antagonism depending on the dose. For example, CI values ranged from 1.23 to 2.06 for 0.25x, 0.5x, and 1x IC₅₀ combinations, indicating antagonism, with a slight antagonism (CI = 1.12) even at 4x IC₅₀. Notably, synergism (CI = 0.76) was observed only at 2x IC₅₀ (Fa = 0.91) for this combination. Isobolographic analysis (Fig. 6C) further supported this, with data points aligning with or above the theoretical additive line. Consistently, quantitative analysis of cell viability at specific concentrations also indicated no significant or consistent synergistic effect with cisplatin (Fig. 6D).

Fig. 6.

In vitro and in vivo evaluation of combination therapy targeting HSP70 in LMS model. (A, E) Dose-response curves for single and combination treatments of VER155008 with cisplatin (A) or 4HC (E) in S3 cells. The leftward shift of the combination curve (pink line) compared to monotherapy (black line) indicates enhanced efficacy. (B, F) Combination Index (CI) vs. fraction affected (Fa) plots for VER155008 + cisplatin combination (B) and VER155008 + 4HC combination (F), generated by CompuSyn software. CI values below 1 indicate synergy. (C, G) Isobologram analysis for VER155008 + cisplatin combination (C) and VER155008 + 4HC combination (G), showing dose pairs required to achieve 50% (Fa = 0.5, blue), 75% (Fa = 0.75, red), and 90% (Fa = 0.9, green) growth inhibition. (D) Cell viability comparison between single-agent and combination treatments of VER155008 with cisplatin in S3 cells. (H) Cell viability comparison between single-agent and combination treatments of VER155008 with 4HC in S3 cells. The combination showed significantly greater inhibitory effect than monotherapy. (I) Representative bioluminescence imaging (BLI) of LMS-bearing mice treated with vehicle, VER155008 alone, or in combination with cisplatin or ifosfamide (IFO) at days 1, 10, and 17. (J) Longitudinal quantification of BLI signal intensity in the brain region of interest (ROI) over time for each treatment group. (K) Longitudinal quantification of BLI signal intensity in the spinal cord region of interest (ROI) over time for each treatment group. Quantification of photon flux was performed for brain and spinal cord regions of interest (ROI) at the indicated time points. In vivo data are presented as mean ± SD, with n = 5 mice per group. Statistical analysis was conducted using repeated-measures ANOVA followed by post hoc testing. *P < 0.05, **P < 0.01, ***P < 0.001.

Conversely, the combination of VER155008 with 4HC demonstrated synergistic antitumor activity at several key concentrations in vitro. The dose-response curve (Fig. 6E) revealed a significant reduction in cell viability in the combination group compared to either monotherapy. CI values summarized in Supplementary Table S5 and derived from the CI-Fa plot (Fig. 6F) showed synergism at 0.25 × (CI = 0.87), 1 × (CI = 0.77), and strong synergism at 2 × (CI = 0.43) IC₅₀ concentrations. However, antagonistic interactions were observed at 0.5 × (CI = 1.53) and 4x (CI = 2.01) IC₅₀ concentrations, indicating a dose-dependent interaction profile. Isobolographic analysis (Fig. 6G) corroborated these findings, with combination data points for synergistic interactions falling clearly below the additive line. Cell viability assays at defined concentrations further confirmed the stronger synergistic effect of VER155008 and 4HC at effective synergistic concentrations (Fig. 6H).

Based on these in vitro findings, we proceeded to validate the combinatorial strategy in our established LMS mouse model using S3 cells. Tumor burden in both the brain (Fig. 6J) and spinal cord (Fig. 5K), monitored longitudinally via BLI (Fig. 6I), showed distinct therapeutic outcomes, with the most significant observations recorded at Day 1, 9 and 17 (Supplementary Table S6). At Day 17, analysis of brain tumor burden (Fig. 6J) revealed a significant reduction in luminescence signal in the VER155008 monotherapy group compared to the control group (P < 0.001). The combination of VER155008 with ifosfamide also demonstrated a significant reduction in brain tumor burden compared to control (P < 0.05). In contrast, the combination of VER155008 with cisplatin did not show a statistically significant advantage over VER155008 monotherapy in the brain (P > 0.05). For spinal tumor burden (Fig. 6K), VER155008 monotherapy significantly reduced luminescence signal compared to the control group (P < 0.001). The combination of VER155008 with cisplatin also showed a significant reduction compared to control (P < 0.01). Furthermore, the VER155008 and ifosfamide combination resulted in a highly significant decrease in spinal tumor burden compared to control (P < 0.001). Monitoring of body weight indicated no significant changes in body weight across the experimental groups. No treatment-related mortality, significant body weight loss, or overt neurological toxicity was observed in any treatment group during the study period. These data collectively demonstrate that pharmacological inhibition of HSP70 with VER155008 exerts a significant antitumor effect against medulloblastoma seeding cells in vivo. Moreover, our findings suggest that its combination with ifosfamide, but not cisplatin, may represent a promising therapeutic approach in the context of leptomeningeal seeding, providing a rationale for further investigation into selective combination chemotherapy strategies.

Discussion

In this study, we successfully isolated MB LMS cells from distal spinal cord coatings using a serial orthotopic xenograft mouse model. Through repeated transplantation and selective in vivo passage, we generated distinct seeding (S3) and non-seeding (N3) cell populations, allowing for functional and molecular comparisons. Seeding cells exhibited unique characteristics, including slower proliferation, altered adhesion, and specific molecular signatures compared to non-seeding cells. Gene expression profiling revealed significant upregulation of HSPA1A and HSPA1B, encoding the molecular chaperone HSP70, in seeding cells. Importantly, these cells demonstrated increased sensitivity to HSP70 inhibition by VER155008, as observed in both in vitro and in vivo models.

Orthotopic xenograft mouse models of MB LMS have previously been shown to recapitulate key clinical and pathological features observed in human patients [17,30]. Although targeting LMS in such models has been explored [31], dissecting the precise cellular mechanisms underlying LMS requires appropriate cell lines. While patient-derived metastatic MB cell lines would be ideal, they are rarely obtainable in sufficient quantities. The D283 cell line, often used for metastasis studies, was derived from extracranial (peritoneal) metastases rather than LMS and exhibits too slow growth [17,30,32]. Thus, our establishment of LMS-derived seeding cells provides a more relevant model to investigate LMS-specific biology and therapeutic vulnerabilities.

Through three rounds of serial in vivo selection, we isolated LMS-derived S3 cells that exhibit stable dissemination-associated phenotypes, including altered migratory behavior and enhanced adhesion to collagen IV, a key component of the meningeal basement membrane. These properties distinguish S3 cells from parental and non-seeding populations and support the use of this model as a more clinically relevant platform for studying medulloblastoma leptomeningeal dissemination.

The principle of selective pressure shaping tumor evolution is well established [13,14]. In our study, seeding cells isolated through serial transplantation represent populations that have adapted to the CSF microenvironment. These cells demonstrated distinct behaviors: they did not effectively close gaps in wound-healing assays and instead appeared to detach from the monolayer over time, and they showed enhanced adhesion to collagen IV among various ECM substrates. The selective increase in adhesion to collagen I and collagen IV, but not to other ECM components, suggests niche-specific adaptation rather than a generalized increase in adhesive capacity. Collagen IV is a major structural component of the meningeal and perivascular basement membranes, which may represent critical interfaces for tumor cell attachment during leptomeningeal dissemination. These observations suggest that MB seeding may place a greater emphasis on survival within the CSF microenvironment, rather than on active invasion or migration mechanisms commonly observed in other metastatic cancers [2,33]. Anatomically, MB's proximity to the fourth ventricle and subarachnoid space facilitates passive CSF-mediated dissemination, consistent with LMS preferentially occurring in the spinal cord and following CSF flow patterns [27].

Transcriptomic analysis revealed strong enrichment in metabolic pathways in seeding cells, suggesting adaptation to the nutrient-poor CSF environment. Recent studies have highlighted that CSF lacks critical micronutrients such as iron, and expression of survival-related molecules (e.g., lipocalin-2, GABA transaminase) may facilitate LMS [15,16]. HSP70 is frequently overexpressed in a variety of cancers, including brain tumors, and plays a key role in maintaining proteostasis under cellular stress conditions such as nutrient deprivation and metabolic challenge [[34], [35], [36]]. In this context, the increased expression of HSP70 observed in LMS-derived cells may reflect a stress-adaptive response associated with tumor persistence and invasiveness, warranting further investigation in models of CNS tumor dissemination.

The preferential sensitivity of LMS-derived S3 cells to both VER155008 and the BiP/GRP78 inhibitor HA15 suggests involvement of a broader HSP chaperone network, encompassing both cytosolic and ER-resident components, in leptomeningeal seeding cells. Among the HSP70 inhibitors evaluated, VER155008 showed a consistent activity pattern in LMS-derived seeding cells across the tested concentration range, which supported its selection for further investigation. Mechanistically, VER155008 is an ATP-competitive inhibitor targeting the nucleotide-binding domain of HSP70, and may therefore interfere with stress-adaptive chaperone functions that could be important for LMS cell persistence under harsh microenvironmental conditions. Together with the differential sensitivity between N3 and S3 cells, these data suggest that HSP70 dependency may become more pronounced during serial in vivo selection for leptomeningeal competence, rather than representing a dominant feature of the parental or early-stage populations. HSP70 and related chaperone-network stress-response programs have been implicated in malignant brain tumors, including processes relevant to tumor persistence and invasiveness [34,36,37]. Therefore, we speculate that stress-response pathways involving HSP70 may also be relevant to pediatric CNS tumors that disseminate through the CSF, warranting validation in additional models

Building upon these findings, we further investigated the potential of combining HSP70 inhibition with conventional chemotherapy. While VER155008 monotherapy effectively delayed LMS progression in our mouse model, it did not completely eradicate tumor growth. Notably, in vitro studies demonstrated a synergistic antitumor effect when VER155008 was combined with 4HC, the active metabolite of ifosfamide. This synergistic effect was also observed in vivo, where the combination of VER155008 and ifosfamide resulted in more pronounced suppression of leptomeningeal spread compared to monotherapies. These results highlight the therapeutic promise of combining HSP70 inhibitors with standard chemotherapy, potentially enhancing efficacy and allowing for reduced doses to mitigate systemic toxicity.

Despite these promising findings, several limitations warrant further investigation. First, we evaluated only one MB cell line with high LMS potential; additional models, including patient-derived cells, should be examined to generalize these results. Second, VER155008 is a preclinical tool compound, and future development of clinically optimized HSP70 inhibitors will be required for translation Third, although concurrent administration was supported by in vitro CI and isobologram analyses (Fig. 5B, 5C, 5F, 5G), alternative schedules (e.g., sequential dosing or different dosing intervals) were not systematically evaluated and should be addressed in future preclinical studies. Fourth, future studies should further optimize the therapeutic strategy by exploring alternative administration routes (e.g., intrathecal delivery) and assessing long-term outcomes, including survival and treatment-related toxicity, in preclinical models.

In conclusion, our study is the first to isolate and functionally characterize MB LMS-derived seeding cells through serial in vivo selection. We demonstrate the potential of targeting HSP70 as a therapeutic strategy against LMS, providing a rationale for further preclinical and clinical evaluation of HSP70 inhibitors, particularly in combination regimens, to improve outcomes in patients with MB leptomeningeal dissemination.

Availability of data and materials

The datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.

Funding

This study was jointly supported by grants from the National Research Foundation (NRF) of Korea (No. RS-2023-NR076773 to JHP, RS-2022-NR069269 to SAC), from the Seoul National University Hospital Research Fund (No. 03-2018-0440 to JHP), and from the Creative-Pioneering Researchers Program through Seoul National University (No. 800-20210571 to JHP), the SNUH Kun-hee Lee Child Cancer & Rare Disease Project, Republic of Korea (26A-003-0100 to JHP).

CRediT authorship contribution statement

Seung Ah Choi: Writing – original draft, Writing – review & editing, Formal analysis, Software, Validation, Visualization, Methodology, Investigation, Data curation, Conceptualization, Funding acquisition. Sokhoeun Heng: Writing – review & editing, Formal analysis, Software, Validation, Methodology, Investigation. Saehim Ha: Methodology, Data curation, Project administration. Seung-Ki Kim: Writing – review & editing. Do Won Hwang: Resources, Data curation, Writing – review & editing. Hyewon Youn: Resources, Writing – review & editing. Ji Hoon Phi: Conceptualization, Supervision, Funding acquisition, Writing – review & editing.

Declaration of competing interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential competing of interest.