Introduction

Glioblastoma multiforme (GBM) is the most common and aggressive malignant brain tumor in adults, with an estimated 5-year survival rate of 4%1. The current standard of care is an intensive multimodal treatment including neurosurgical resection, radiotherapy2 and concomitant and adjuvant chemotherapy with temozolomide (TMZ)3 and other drugs. Several targeted therapies such as immunotherapy, are under clinical investigation4. However, there is a need for novel approaches to address high disease recurrence and low patient survival.

Artificial intelligence (AI) and computational tools are increasingly used to identify promising drugs and new disease targets more efficiently5,6,7,8,9,10,11.These approaches are valuable for both discovering novel drug candidates and identifying new uses for existing ones12,13,14.

This study exemplifies the power of computational methods in uncovering new therapeutic indications for existing therapeutics, such as endoxifen, by leveraging large-scale data integration and advanced bioinformatics tools15. (Z)-endoxifen (endoxifen) is one of the most potent and bioactive metabolites of tamoxifen, a well-established oral selective estrogen receptor modulator (SERM) efficacious in reducing recurrence risk of ER + breast cancer16. Endoxifen consists of two isomeric forms, (E) and (Z), with (Z)-endoxifen being the active antiestrogenic form17. To date, endoxifen has been used in clinical trials for a limited set of indications, including ER + invasive breast cancer, ductal carcinoma in situ, mammographic breast density and in a few patients with gynecologic cancers and desmoid tumors18,19,20, with most recently promising antitumor effects demonstrated in ER + /HER- breast cancer patients21. Also, endoxifen manufactured by Intas Pharmaceuticals under the brand name Zonalta is approved for the acute treatment of manic episodes of bipolar I disorder in India22,23. Endoxifen has significantly fewer adverse side effects than tamoxifen24, which may offer potential advantages for its clinical application19. Moreover, endoxifen is hypothesized to have complex mechanisms of action (MOA) that are concentration-dependent, distinguishing it from other antiestrogens and broadening its clinical applications25. At lower concentrations, endoxifen has been shown to inhibit ERα via antagonistic properties, and at higher concentrations it has been shown to allosterically inhibit protein kinase C beta (PKCβ1) activity26 and degrade PKCβ1, ultimately leading to apoptosis in breast cancer cells27. Endoxifen stabilizes ERβ protein and induces ERα/ERβ heterodimerization in a concentration-dependent manner, while ERβ in turn enhances the sensitivity of breast cancer cells to the anti-estrogenic effects of endoxifen28. Moreover, endoxifen has been shown to function as an aromatase inhibitor29, thereby introducing additional complexity and breadth to its MOA.

We hypothesized that endoxifen’s multi-faceted mechanism of action, including modulation of PKCβ1 and ER signaling pathways, could be therapeutically beneficial in certain non-breast cancers. To test this, we applied an AI-based platform (PandaOmics) to analyze over 900 cancer types, identify high-potential indications, and selected GBM for experimental validation based on its high predicted efficacy score, unmet clinical need, and prior evidence supporting the activity of tamoxifen in GBM models. Laboratory experiments were then performed to test endoxifen’s effects on GBM cells. The results highlight the versatility of endoxifen’s MOA and its ability to modulate key signaling pathways involved in GBM progression, paving the way for further preclinical and clinical investigations to validate its efficacy in GBM treatment.

Materials and methods

Data collection and integration using PandaOmics AI platform

Gene expression data were obtained from Gene Expression Omnibus (GEO)30, ArrayExpress31, and PRIDE32. These data were collected in PandaOmics 4.0 (https://pandaomics.com/), an AI-driven target discovery platform33 (http://www.pharma.ai/pandaomics), designed to solve various biomedical problems, including discovering therapeutic targets and biomarkers, compound identification, and selection of indication hypotheses. The platform allows combining comparisons with similar experimental designs into meta-analyses by processing each dataset separately and performing differential expression analysis for each analysed comparison (e.g., treated versus vehicle control, disease versus healthy control) to extract well-founded insights across diverse datasets. Dataset selection and sample group preparation were performed manually by experienced biology analysts. All available endoxifen omics datasets were collected and are described in detail in Supplementary Table 1. GBM (EFO_0000519) meta-analysis comprised 30 multi-omic datasets (RNA-seq, microarray, proteomics, and methylation) with 2,689 total samples which are described in Supplementary Table 3. All omics datasets were pre-processed according to the PandaOmics pipeline, which automatically defines data type and normalizes the data for further analysis33.

Data analysis

Differential expression analysis and combined log-fold changes (LFC)

Differential expression analysis was performed in PandaOmics using the limma R package. Each dataset was processed according to standard protocols. Microarrays were preprocessed according to platform specific protocols, RNA-seq counts were upper-quantile normalized, and VSN (Variance Stabilization Normalization) normalization was applied for proteomics datasets. Gene-wise p-values were corrected by the Benjamini–Hochberg procedure. Genes with q-value < 0.05 in either proteomic or transcriptomic datasets were retained. Combined log-fold changes (LFC) between transcriptomics and proteomics datasets were calculated using the ‘Expression’ feature in PandaOmics. This feature reconciles differences between platforms by first normalizing LFC values within each comparison, from either proteomic or transcriptomic dataset, by their standard deviation. These normalized LFCs are then averaged across comparisons used in meta-analysis to create a combined measure. For p-value integration, p-values are separated into left- and right-tailed values according to the direction of change, and then aggregated across comparisons using Stouffer’s method. The final combined p-value is selected based on the direction of the combined LFC, with false discovery rate (FDR) correction applied. Finally, min–max normalization is applied to the combined LFC values. This approach ensures that integration reflects both the magnitude and statistical significance of changes, while minimizing bias from platform-specific expression ranges.

Genes were considered differentially expressed (DEGs) if they had an adjusted p-value (FDR) < 0.05 from the respective meta-analyses, upregulated if LFC > 0 and downregulated if LFC < 0. Genes exhibiting dose-dependent expression changes were identified as those showing consistent increases or decreases in LFC values with increasing doses, accompanied by adjusted p-values (FDR) < 0.05.

Gene set enrichment analysis (GSEA)

GSEA was performed with the GSEApy package34 using the enrichr() function according to standard protocols. DEGs were used as a gene list for pathway enrichment analysis. The MSigDB database was selected as the gene set from GSEApy internal library for signaling pathway enrichment analysis.

Indication prioritization using weighted endoxifen signature

The PandaOmics indication prioritization and expansion function using over 20 AI, omics, and text-based models was utilized to identify indications where endoxifen may exert beneficial clinical effects33. This approach utilizes a repository of pre-calculated disease meta-analyses that encompass over 8000 diseases, with special emphasis on more than 500 manually curated meta-analyses involving human-patient-derived omics data from disease-relevant tissues. Typically, this method uses a single gene as input and ranks diseases based on their potential association with the gene. However, in this study, instead of a single target gene, the weighted endoxifen gene expression signature comprising endoxifen-dependent genes was applied as input. This gene expression signature comprises a list of genes that are differentially expressed following endoxifen treatment, as determined by a combined meta-analysis, with increased weight assigned to genes demonstrating dose-dependent expression changes or associations with endoxifen’s MOA, as supported by PPI networks, signaling pathway databases, and a proprietary NLP-driven knowledge graph. Scores for each gene were calculated, and the results were aggregated considering gene weights. The focus was on cancer indications, narrowing the list of diseases to approximately 900 cancer terms.

To validate the scoring approach, two validation metrics were used: Log-transformed fold change of enrichment (ELFC) and hypergeometric p-value (HGPV). ELFC showed how much the top of the list was enriched by positive controls, and HGPV stood for the statistical significance of the effect and showed how likely the same level of enrichment could be achieved from the random list of indications. Both metrics have been described in detail6,35. Threshold values of ELFC > 1 and HGPV > 1.3 were used to indicate significant positive results. The resultant list of prioritized indications for endoxifen demonstrated statistically significant enrichment with positive controls, as evidenced by elevated HGPV (7.68) and ELFC (5.36) metrics. Positive controls were defined as diseases where endoxifen had been previously tested in clinical trials.

Transcription factor (TF) enrichment analysis

TF enrichment analysis was performed using the GSEApy package with the enrichr() function according to standard protocols. The primary library “ARCHS4_Coexpression” from the ChEA3 database36 was selected as the gene set for TF enrichment analysis. The total number of tested TFs was 1628 (1331 passed FDR adjusted p-value < 0.05 for GBM meta-analysis and 980 for endoxifen meta-analysis). The FDR adjusted p-values were used for ranking TFs separately for GBM and endoxifen meta-analyses to identify the most highly perturbed TFs. A scatter plot was used to show significantly enriched TFs. Only TFs that were significantly altered in both the GBM meta-analysis and the endoxifen meta-analysis (FDR adjusted p-value < 0.05) and had ≥ 75% of dysregulated downstream genes were included. Individual TF enrichment results were visualized on a bar plot using the matplotlib.pyplot Python package. Each bar on the plot corresponds to a significantly perturbed gene regulated by the corresponding TF.

Single-cell RNA-seq gene expression analysis

GSE8446537 and GSE15941638 were downloaded from the GEO database and preprocessed with a standard pipeline using Scanpy package. Initially, single cell data was filtered to include only cells with at least 200 genes detected. Genes present in less than three cells were discarded from the analysis. The quality control pipeline included removal of cells with mitochondrial fraction more than 20% and doublet removal, ensuring accurate cell population identification and improving overall data reliability. Data were initially normalized with a scale factor of 10,000 and then log2 transformed. Variable genes were identified using the sc.pp.highly_variable_genes function() with n_top_genes = 4000 and flavor = “seurat_v3”. The PCA was performed using the sc.pp.pca function with the previously defined variable genes and n_comps = 50. A batch effect between patients samples was corrected with sce.pp.harmony_integrate() function. The neighborhood graph was computed on the first 50 principal components derived after batch correction and the results were visualised using Uniform Manifold Approximation and Projection for Dimension Reduction (UMAP). Louvain clustering methods were used for cell types annotation. Cell cluster annotations were based on marker genes as described in corresponding papers (37 for GSE84465, 38 for GSE159416). Differential expression between case and control for each cell type was calculated using tl.rank_genes_groups function (method = ‘wilcoxon’). Gene expression changes were compared between tumor core and peripheral tissues in GSE84465 and visualized on boxplots and heatmaps. Gene expression changes between Mesenchymal and Proneural/Classical GBM across all cell types were compared in GSE159416 and visualized on boxplots and heatmaps.

Patient survival analysis

Survival analysis was conducted on the The Cancer Genome Atlas (TCGA-GBM) dataset39 using the KaplanMeierFitter function from Lifelines python package. Median function was applied for normalized gene expression data and median value for each gene of interest was used as a threshold for patients’ stratification. Patients with the expression value of the gene of interest ≥ or < than median value were considered as patients with “high” or “low” expression of a particular gene, respectively. A log-rank test and Cox proportional hazards regression were performed to calculate the statistical significance for the survival analyses.

In vitro validation experiments

Cell lines, chemicals and reagents

The CRT435 cell line is a proprietary patient-derived GBM cell line (Certis Oncology Solutions; San Diego, CA, USA) and was used due its relevance to human GBM biology. The cell line was isolated from a 61 year old female with recurrence after treatment with Temodar and radiotherapy. The cell line underwent Short Tandem Repeat (STR) testing prior to culture at Certis Oncology Solutions to confirm its identity. Cells were cultured in appropriate media according to supplier recommendations, with the media and reagents prepared following ATCC guidelines unless specified otherwise. To identify the optimal seeding density CRT435 cells were seeded on one 96-well cell plate with the density range from 75 to 10,000 cells per well, and transduced with the IncuCyte Nuclight Lentivirus Reagent (Sartorius, Göttingen, Germany). Based on the resultant growth curves, densities that ensured consistent and exponential growth rates without early confluence or delayed growth due to low density were identified, leading to the selection of 2500 cells per well for the efficacy study.

TMZ (temozolomide) was purchased from Medchem Express (USA) and DMSO (dimethyl sulfoxide) was purchased from Sigma-Aldrich (USA). Endoxifen was synthesized by OLON S.p.A. (Rodano, Italy). For proliferation study, cells were transduced with the IncuCyte Nuclight Lentivirus Reagent (Sartorius, Göttingen, Germany). For cytotoxicity assays, cells were stained with IncuCyte Annexin V Dyes for Apoptosis (Sartorius, Göttingen, Germany).

Treatments

For efficacy study, on Day 0, cells were seeded at an appropriate density in 96-well plates, with a volume of 100 µL per well. During Day 1, following confirmation that the transduced cells were fluorescing, the desired treatments were administered. These included 9 concentrations of endoxifen (0.08–20 µM), one (250 µM) concentration of TMZ, and 9 × 2 combinations of endoxifen (0.08–20 µM) with TMZ (37.5 µM and 150 µM). Each treatment was applied to three independent biological replicates, including vehicle control (equivalent to the dose of highest test agent dose) and a positive control for cell killing (10% DMSO).

Proliferation and cytotoxicity assays

The IncuCyte S3 Live-Cell Analysis System (Sartorius, Göttingen, Germany) with the IncuCyte Cell-by-Cell Analysis Software Module (PN 9600 0031) was used to measure cytotoxicity and proliferation. Stable cell lines were produced using the lentiviral transduction with the IncuCyte Nuclight Lentivirus Reagent (Sartorius, Göttingen, Germany). The transduced cells were seeded into 96-well plates at optimal densities and treated as described above. Scans were taken every 6 h starting at Day 0 and over an 8-day period, with the objective magnification 10×, 5 images per well. Proliferation was measured by counting the nuclear green objects per image, while cytotoxicity was assessed via red fluorescence intensity and the area of red fluorescence (RCU × µm2/image) after staining the cells with IncuCyte Annexin V Dyes for Apoptosis (Sartorius, Göttingen, Germany). Control wells included cells treated with the vehicle (DMSO) equivalent to the highest concentration of drug treatments and a positive control using 10% DMSO to induce cell death.

In vivo studies

The efficacy of endoxifen, with and without TMZ, was further investigated in an in vivo GBM model. Female athymic nude mice, aged 6–12 weeks, were housed in sterile, individually ventilated cages (IVCs) with a maximum of five animals per cage. The animals were maintained under specific pathogen-free (SPF) conditions, with controlled temperature (20–24 °C) and humidity (40–60%). A 12-h light/dark cycle was provided, and the mice had ad libitum access to autoclaved food and water. All procedures were conducted in accordance with institutional guidelines and approved by the relevant ethics committee.

The efficacy study included eight groups with six-eight mice each, treated for 28 days. Mice were implanted subcutaneously into the right rear flank with an inoculum fragment of 13.5 mm3 tumor, and once tumors reached an average size of 60–200 mm3, with no single tumor exceeding 300 mm3, mice were randomized into the following treatment groups: endoxifen high dose (75 mg/kg), endoxifen medium dose (50 mg/kg), endoxifen low dose (25 mg/kg), TMZ (25 mg/kg), TMZ (25 mg/kg) + endoxifen high dose (75 mg/kg), TMZ (25 mg/kg) + endoxifen medium dose (50 mg/kg), TMZ (25 mg/kg) + endoxifen low dose (25 mg/kg), or a vehicle control group (0.1% Tween 80 (v/v) and 0.5% methyl cellulose (w/v) in deionized ultra-filtered (DIUF) water). Endoxifen was given orally once a day for 28 days, 28 doses total. TMZ was administered intratumorally once a day for 5 days, followed by 2 days off, 4 cycles, 20 doses total.

Body weight was measured at least once weekly until tumor growth was observed, then twice weekly. The average body weight of mice for the duration of the study was roughly 26 g. Clinical observations were conducted weekly, then twice weekly post-tumor growth. Tumor volumes were measured twice weekly using digital calipers. Tumor volume was calculated using the formula: Volume = 0.5 × Length × Width2, where the length (L) and width (W) were measured in millimeters (mm). The resulting volume was expressed in cubic millimeters (mm3). The primary endpoint was tumor volume reduction.

During the study, some animals reached humane endpoints, and were euthanised. All living mice were terminated at the end of the study on Day 28 of treatment. Cardiac perfusion for terminal blood was used as the method for euthanasia. The animals were under anesthesia (ketamine (100 mg/kg) and xylazine (10 mg/kg), intraperitoneally) for cardiac perfusion and then, cervical dislocation was performed to fully terminate animals, which is in line with the guidelines by the American Veterinary Medical Association. All animal studies were performed by the contract research organization, Certis Oncology. Certis is an AAALAC accredited facility that adheres to the most recent AVMA and ARRIVE guidelines (https://doi.org/10.1371/journal.pbio.3000410) for study reporting. An internal ethics committee at Certis reviewed and approved the study to ensure it meets the AVMA guidelines. Certis adheres to AVMA Guide for all standards, criteria, and procedures to ensure the highest level of animal welfare is reflected in all Animal Care and Use Protocols as well as in all procedures involving research animals.

Statistical analysis

Data analysis was carried out using GraphPad Prism 5 software (GraphPad Software Inc., San Diego, CA, USA). Each experiment was performed at least in triplicate.

Statistical analysis for treatment effects on cell proliferation and apoptosis was performed using a two-way ANOVA (Proliferation/Apoptosis ~ Treatment * Time) followed by Tukey’s HSD post-hoc test for multiple comparisons.

In vivo effect analysis of body weight was performed with a mixed effects model and Sidak’s multiple comparison test. Mean tumor volume was assessed using a mixed effects model with Dunnett’s multiple comparison test with individual variances computed for each comparison. To enhance the sample size, statistical analysis of tumor volumes in the treatment groups of mice was conducted on Day 16 following the initiation of treatment.

For in vitro and in vivo studies data are presented as mean ± standard error (SE).

Paper draft preparation

The initial draft of this paper was created using Draft Outline Research Assistant (DORA) (https://dora.insilico.com/), Insilico Medicine’s LLM-based paper drafting assistant. DORA streamlines publication creation with the help of over 30 AI agents powered by Large Language Models (LLMs) and integrated databases. Using Retrieval-Augmented Generation (RAG), these agents ensure comprehensive data collection, reduce hallucinations, and provide relevant PubMed links for transparency. The authors reviewed and manually refined the draft, supplementing it with their own text and references to finalize the article.

Results

Publicly available data search and integration

Generation of the endoxifen gene expression signature was described elsewhere40. Briefly, relevant datasets were found in public repositories and included four microarrays from MCF-7 cells treated with endoxifen under different conditions (Supplementary Table 1). Endoxifen doses ranged from 20 to 1000 nM, with or without co-treatments, including: no co-treatment; 1 nM or 10 nM estradiol; or a combination of 10 nM estradiol, 300 nM tamoxifen, 7 nM 4-hydroxytamoxifen, and 700 nM N-desmethyl-tamoxifen. The most common time point was 24 h, with one study describing 3 d of treatment and one study analyzing xenograft mouse models treated with endoxifen for 4 weeks. Identified datasets were integrated into PandaOmics, a cloud-based software platform, which utilizes bioinformatics methodologies and AI to generate corresponding gene expression signatures with the highest levels of completeness and data quality, enabling robust detection of changes33 (Fig. 1).

Fig. 1
Fig. 1The alternative text for this image may have been generated using AI.Fig. 1The alternative text for this image may have been generated using AI.
Full size image

A diagram illustrating the pipeline of data integration, hypothesis generation, and deep dive analysis.

As was shown in the study by Remmel et al., gene set enrichment analysis (GSEA) of endoxifen gene expression signature revealed substantial differences in MCF7 cells treated with endoxifen compared to control, including estrogen signaling modulation, inhibition of cell cycle and cancer development-related pathways including E2F targets, G2-M Checkpoint, Myc targets, mitotic spindle, mTORC1 signaling, activation of hypoxia, reactive oxygen species (ROS) and PI3K/AKT/mTORC1 hallmarks, and bi-directional modulation of glycolysis, oxidative phosphorylation, and apoptosis40. Further analysis showed the existence of estrogen-dependent and independent effects of endoxifen emphasizing the wide range of its MOAs in cancer cells that may be different from that of its prodrug tamoxifen40.

Indication prioritization and nomination of indication hypotheses

Given the known anticancer effects of endoxifen, its molecular MOAs, and the significant unmet clinical need for new treatments, we focused on identifying human cancer types where endoxifen could provide the greatest patient benefit. To address this question, PandaOmics-driven Indication Prioritization approach was applied, which integrates 23 models derived from omics, text, and financial data to rank approximately 900 cancer terms33. A list of endoxifen-responsive genes, incorporating dose-dependent effects, gene locations within the protein–protein interaction (PPI) network, their participation in key signaling pathways, and corroborative evidence from a proprietary NLP-driven knowledge graph, which represents the weighted gene expression signature of endoxifen, was utilized as input. Individual gene scores were calculated and aggregated, considering gene weights (Fig. 1). The top of the resultant list of cancer indications was statistically significantly enriched with diseases where endoxifen has been tested in clinical trials, namely, breast carcinoma (rank 11), ovarian carcinoma (rank 16), endometrial cancer (rank 22) and cervical cancer (rank 31)20, validating the scoring approach (see Materials and Methods). New indications, that were scored even higher than well-established breast cancer indication, included lung cancer (rank 2) (both non-small cell lung cancer (rank 1) and small-cell lung cancer subtypes (ranks 10)), colorectal cancer (rank 3), melanoma (rank 4), GBM (rank 6), hepatocellular and pancreatic carcinomas (ranks 7 and 9, accordingly), as well as acute leukemia (rank 8) (Supplementary Table 2).

Among these, GBM stood out due to several compelling factors. First, previous preclinical and clinical studies using tamoxifen, the prodrug of endoxifen, have shown encouraging results in GBM41,42,43,44,45. Second, endoxifen is already approved for the acute treatment of manic episodes in bipolar I disorder22,23, providing evidence of its efficacy in central nervous system (CNS) conditions. Taking into account the particularly high unmet clinical need in GBM, the lack of effective treatments, the promising efficacy of the prodrug in this indication, and endoxifen’s demonstrated success in another CNS disorder, the authors ultimately selected GBM as the most promising candidate for further investigation.

Computational analysis of endoxifen for GBM

To investigate the potential impacts of endoxifen, we conducted an analysis of endoxifen-induced transcriptomic alterations and their correlation with gene expression changes observed in GBM. Manually curated multi-omics GBM PandaOmics meta-analysis which consisted of 30 studies including 2085 GBM and 604 control samples (Supplementary Table 3) was used to generate the GBM gene signature (Fig. 1). GBM and endoxifen gene signature intersection analysis detected 1463 genes in the overlap. Among these, 560 genes were upregulated in GBM and downregulated by endoxifen, and 264 genes were downregulated in GBM and upregulated by endoxifen in MCF7 cells (Fig. 2A, Supplementary Table 4).

Fig. 2
Fig. 2The alternative text for this image may have been generated using AI.
Full size image

(A) Analysis of genes intersecting between bulk GBM and endoxifen data. The red arrow pointing up and the blue arrow pointing down describe a group of genes that were upregulated in GBM and downregulated by endoxifen in MCF7 cells. The red arrow pointing down and the blue arrow pointing up describe a group of genes that were downregulated in GBM and upregulated by endoxifen in MCF7 cells. (B) GSEA analysis for 560 genes upregulated in GBM and downregulated after endoxifen treatment in MCF7 cells. Ranking is based on adjusted p-value. Coloring is based on Combined score =  − log(p-value) * Odds Ratio. Dot size is based on the number of genes related to the pathway. All significant hallmarks are presented.

Next, we focused on genes upregulated in GBM and downregulated after endoxifen treatment in MCF7 cells and applied GSEA analysis on this subset of genes (Fig. 2B). Among the top pathways hyperactivated in GBM that could be potentially reversed by endoxifen were cell-cycle related hallmarks (G2-M checkpoint, E2F, mitotic spindle, Myc targets) and mTORC1 pathway. Importantly, estrogen and androgen response hallmarks were also highly enriched, underlying the role of estrogen and androgen signaling in GBM. Among other pathways hyperactivated in GBM and inhibited by endoxifen, apoptosis, immune response-related hallmarks (TNF-alpha signaling via NF-κB, interferon gamma and alpha response, IL-2/STAT5 signaling), hypoxia, glycolysis and EMT were identified.

To further explore key regulators relevant to GBM and endoxifen gene expression signatures, transcription factor (TF) enrichment analysis was performed. TFs enriched with downstream genes that were upregulated in GBM and downregulated after endoxifen treatment in MCF7 cells were identified. Among the highly perturbed TFs were E2F7, E2F8, PRMT3, ZNF146, and CENPA (Fig. 3A). Ninety-five percent of genes regulated by E2F8 were significantly upregulated in GBM; conversely, 94% of E2F8-regulated genes were downregulated after endoxifen treatment (Fig. 3B). Similar results were obtained for E2F7 where 95% of E2F7-regulated genes were upregulated in GBM and 83% of E2F7-regulated genes were downregulated after endoxifen treatment. This was in agreement with pathway enrichment analyses (Fig. 2B), underscoring the potential of endoxifen to modulate E2F-regulated signaling in GBM.

Fig. 3
Fig. 3The alternative text for this image may have been generated using AI.
Full size image

TF enrichment analysis for downstream genes that were upregulated in GBM and downregulated after endoxifen treatment in MCF7 cells. (A) Scatter plot for TFs significantly enriched with downstream genes (FDR adjusted p-value < 0.05) in both GBM and endoxifen meta-analyses. The first 10 TFs with the highest proportion of downstream genes upregulated in GBM and downregulated after endoxifen treatment in breast cancer cells are labeled. (B) Graph for E2F8 downstream genes with their combined normalized log fold changes (LFC) in GBM project and in endoxifen project. Significantly (FDR corrected p-value < 0.05) upregulated and downregulated genes are colored green and red, respectively.

To confirm bulk expression data results and explore endoxifen’s potential MOA in GBM with greater resolution, single-cell expression analysis was applied. Single-cell RNA-seq (scRNA-seq) dataset GSE84465 was analyzed, which included a cohort of 4 primary GBM patients (IDH1-negative, grade IV GBMs) with 3589 cells obtained from the tumor core and surrounding peripheral tissue37. Gene expression changes were compared between tumor core and peripheral tissues in specific cell types, and neoplastic cells were used for further analysis. There were 113 genes in the overlap between bulk GBM, scRNA-seq GBM and endoxifen gene signatures. A subset of 73 genes that are upregulated in bulk GBM and scRNA-seq GBM and downregulated after endoxifen treatment (Fig. 4A) (Supplementary Table 5) was used for GSEA analysis. Enrichment with cell-cycle related hallmarks (G2-M checkpoint, E2F targets, Myc targets), in addition to apoptosis and immune response hallmarks (inflammatory response, interferon gamma response, IL-2/STAT5 signaling and coagulation) were determined (Fig. 4B).

Fig. 4
Fig. 4The alternative text for this image may have been generated using AI.
Full size image

(A) Analysis of genes intersecting between bulk GBM, scRNA-seq GBM and endoxifen data. The red arrow pointing up and the blue arrow pointing down describe a group of genes that were upregulated in bulk and scRNA-seq GBM and downregulated by endoxifen in MCF7 cells. (B) GSEA analysis for 73 genes upregulated in bulk and scRNA-seq GBM and downregulated after endoxifen treatment in MCF7 cells. Ranking is based on adjusted p-value. Coloring is based on Combined score =  − log(p-value) * Odds Ratio. Dot size is based on the number of genes related to the pathway. All significant hallmarks are presented.

To elucidate the impact of this subset of genes on the disproportionate survival of GBM patients, we conducted survival analysis using the TCGA-GBM dataset39. From the 73 genes upregulated in both bulk and scRNA-seq GBM data, and downregulated by endoxifen, three genes had high expression associated with poor survival in GBM patients, namely, HSPB1 (Cox p = 0.0045, HR = 1.64, 95% CI 1.17–2.30, log-rank p = 0.0040), RPA3 (Cox p = 0.0193, HR = 1.51, 95% CI 1.07–2.12, log-rank p = 0.0184), and NFKBIZ (Cox p = 0.0318, HR = 1.46, 95% CI 1.03–2.06, log-rank p = 0.0328) (Fig. 5A). In the bulk GBM data, all three genes were strongly upregulated in numerous comparisons (Fig. 5B). On the single-cell level HSPB1, RPA3 and NFKBIZ were enriched in neoplastic cells in the tumor core compared to surrounding peripheral tissue (Fig. 5C,D). Endoxifen significantly downregulated the expression of these genes in MCF7 cells (Fig. 5E).

Fig. 5
Fig. 5The alternative text for this image may have been generated using AI.Fig. 5The alternative text for this image may have been generated using AI.Fig. 5The alternative text for this image may have been generated using AI.
Full size image

Results of the analysis of the genes associated with a patient’s survival and GBM subtypes. (A) Kaplan–Meier survival analysis based on the TCGA-GBM dataset. Heatmaps based on PandaOmics expression analysis in GBM (B) and (E) endoxifen meta-analyses. (C) UMAPs of HSPB1, PRA3 and NFKBIZ expression in GBM samples from GSE84465 scRNA-seq. (D) Boxplots for HSPB1, PRA3 and NFKBIZ genes from GSE84465 scRNA-seq GBM study. (F) UMAPs and boxplot of HSPB1 expression in GBM samples with Mesenchymal (MES) and Proneural/Classical (PN&CL) subtypes based on GSE159416 scRNA-seq GBM study. *p-value < 0.05, **p-value < 0.01, ***p-value < 0.0001.

An additional computational analysis was conducted to evaluate the correlation between gene expression and various GBM subtypes. For this analysis the scRNA-seq dataset GSE159416 was used in which 18 GBM patient samples were profiled to uncover tumor subtypes38. Out of 73 genes that were significantly upregulated in bulk/scRNA-seq GBM data and downregulated after endoxifen treatment, 24 genes were also more highly expressed in the more aggressive Mesenchymal GBM subtype compared to Proneural/Classical subtypes (Supplementary Table 5). Among these genes, HSPB1 was identified (Fig. 5F), which, along with its association with patient’s survival, indicates its important role in GBM progression.

GBM experimental confirmation in vitro and in vivo

In vitro assays demonstrated that endoxifen, alone or in combination with TMZ, markedly reduced cell proliferation and induced apoptosis in the CRT435 GBM cell line. However, in vivo evaluation of the same treatments in CRT435 xenograft-bearing mice did not reproduce these effects, with no significant tumor regression observed under the conditions tested.

In vitro testing of endoxifen alone or in combination with TMZ in the CRT435 GBM cell line

In vitro analysis was performed to investigate the effects of endoxifen in GBM cell proliferation and stimulating cell death. The effects of endoxifen alone or in combination with the standard of care drug TMZ were evaluated in a proprietary patient-derived GBM cell line CRT435 (Certis Oncology Solutions; San Diego, CA, USA). A 9-point dose curve for endoxifen (0.08–20 µM), a 4-point dose curve for TMZ (37.5–300 µM), and combination treatments were investigated. Cells were treated after seeding with optimal densities and incubating for 24 h as described in Materials and Methods. Cell proliferation and death were monitored continuously through imaging every 6 h for up to 8 days.

A strong cytotoxic effect of 10% dimethyl sulfoxide (DMSO) (positive control) was observed (Fig. 6A) (Δ = 267.2, p < 0.05, where Δ, delta, represents the mean differences between treatment and vehicle). While TMZ treatment decreased cell proliferation, it exhibited milder potency in a concentration of 250 µM compared to DMSO (Δ = 142.6, p < 0.05). Conversely, treatment with 20 μM endoxifen, either alone (Δ = 269.7, p < 0.05) or in combination with 150 µM TMZ (Δ = 272.8, p < 0.05), resulted in a more pronounced inhibitory effect on cell proliferation. Importantly, treatment with endoxifen alone and in combination with 150 µM TMZ decreased cell proliferation significantly higher compared to 250 µM TMZ alone (Δ = 127.1 and Δ = 130.2, respectively, p < 0.05, here delta represents the mean differences between endoxifen or endoxifen/TMZ combination treatment and TMZ).

Fig. 6
Fig. 6The alternative text for this image may have been generated using AI.
Full size image

Summary of in vitro studies on the efficacy of endoxifen in the CRT435 GBM cell line. (A) Proliferation analysis in CRT435 cells treated with endoxifen, TMZ, or combination of both. (B) Cytotoxicity (cellular apoptosis) analysis in CRT435 cells treated with endoxifen, TMZ, or combination of both. Proliferation over time was measured by nuclear green object count per image following IncuCyte NucLight Green lentiviral transduction, while cytotoxicity (apoptosis) was assessed by red fluorescence integrated intensity (RCU × µm2 per image) using IncuCyte Annexin V staining. Desired treatments were added on Day 1, after the transduced cells were confirmed to fluoresce. Scans were taken every 6 h starting on Day 0 and over an 8-day period. Vehicle control was used in equivalent to the dose of highest test agent dose. 10% DMSO was used as a cell killing positive control. For ease of visualization and perception of data in the graph, only the 5-day curves for cells treated with maximum concentrations of substances and their combinations are shown. The full versions of the proliferation graphs and apoptosis graphs can be found in Supplementary Figs. 1 and 2. Representative images of CRT435 cells from the IncuCyte S3 Live-Cell Analysis System after endoxifen treatment can be found in Supplementary Figs. 3–7.

The apoptosis assay detected strong stimulation of cell death by 10% DMSO (Δ =  − 581,979.3, p < 0.05, where delta represents the mean differences between treatment and vehicle). Treatment with 250 µM TMZ had a very mild effect (Δ =  − 192,842.7, p < 0.05), while 20 µM endoxifen induced a higher level of apoptosis (Δ =  − 283,181.9, p < 0.05) (Fig. 6B). Furthermore, combining 20 µM endoxifen with 150 µM TMZ significantly increased apoptosis, producing a greater inhibitory effect than either agent alone (Δ =  − 536,528.8, p < 0.05) comparable to the cytotoxicity observed with 10% DMSO. Importantly, treatment with 20 µM endoxifen in combination with 150 µM TMZ increased apoptosis significantly higher compared to 250 µM TMZ alone (Δ = 343,686.2, p < 0.05, where delta represents the mean differences between endoxifen/TMZ combination treatment and TMZ).

In vivo testing the efficacy of endoxifen alone or in combination with TMZ in CRT435 xenograft mice

Despite the strong in vitro cytotoxic and anti-proliferative effects of endoxifen observed in CRT435 cells, the subsequent in vivo study did not show significant tumor reduction in any endoxifen treatment group, either as monotherapy or in combination with TMZ. The CRT435 cell line was selected for in vivo validation based on its high proliferation rate and responsiveness in vitro. Athymic nude mice were implanted with the CRT435 PDX subcutaneously into the right rear flank. Once tumors formed, mice were randomized into treatment groups, 6 mice per group. Groups included a control group, three groups treated with varying doses of endoxifen (75 mg/kg, 50 mg/kg, 25 mg/kg), a TMZ group (25 mg/kg), and three combination groups with varying doses of endoxifen and 25 mg/kg TMZ. After randomization, animals were dosed for 28 days.

The analysis of treatment effects on tumor volume identified a significant time and treatment interaction (F (35, 189) = 2.609, p < 0.001). Post hoc analysis of between group comparisons did not identify treatment differences between Day 0 and Day 16. There was no significant difference in tumor regression between the endoxifen + TMZ treated groups and the vehicle control group in the doses tested (Fig. 7 and Supplementary Fig. 8). However, treatment with TMZ alone showed promising anti-tumor activity when compared to vehicle-treated groups, although it did not achieve statistical significance.

Fig. 7
Fig. 7The alternative text for this image may have been generated using AI.
Full size image

Summary of in vivo studies on the efficacy of endoxifen in the CRT435 PDX model. Tumor volume measurements in athymic nude mice inoculated with CRT435 tumor fragment and treated with endoxifen alone or in combination with TMZ. Mean tumor volume (TMV) (mm3) ± standard error (SE) are presented.

In general, all treatments were well tolerated. A main effect of time and treatment on body weight were identified (F (3.392, 127.0) = 58.57, p < 0.001 and F (7, 40) = 2.638, p < 0.05). Post hoc analysis showed that all treatments did not impact body weight over the duration of the study (Supplemental Fig. 9).

Taken together, this study did not demonstrate conclusive efficacy of endoxifen combination treatment along with TMZ in the CRT435 PDX athymic nude mouse model.

Discussion

Despite advancements in multidisciplinary treatment approaches, GBM remains a highly lethal brain tumor, with a median survival of 15 months46, which indicates an urgent need for new therapeutic approaches. In this study we applied the AI-based PandaOmics platform33 to identify endoxifen as a potential new treatment for GBM.

Overall, targeting estrogen signaling pathways with estrogen receptor modulators appears to be a promising approach in GBM treatment47. While several SERMs have been explored in preclinical studies for GBM, endoxifen has not yet been investigated for this indication. Its prodrug, tamoxifen, is the only SERM that has been tested in human studies. Tamoxifen has been examined in various preclinical models of GBM, gliomas, and other CNS tumors, predominantly showing cytotoxic effects41,42,43,44,45,48,49. The cytotoxic effects of tamoxifen have been demonstrated to be both ER-dependent and ER-independent, involving autophagy and the induction of apoptosis through different mechanisms. Furthermore, tamoxifen has been shown to enhance the cytotoxic sensitivity of tumor cells to chemotherapeutic drugs and radiation45,50. Notably, both tamoxifen and, to a lesser extent, endoxifen possess the ability to cross the blood–brain barrier (BBB)51. Considering the promising anticancer activity of endoxifen observed in breast cancer compared to tamoxifen in certain settings52, its complex MOA27,40, its hypothesized role in stabilizing ERβ and thus sensitizing cancer cells to chemotherapy28, endoxifen may represent a promising new therapeutic option for GBM. However, its relevance to GBM should be interpreted cautiously given the markedly different tumor biology and microenvironment, as well as the currently limited data on its ability to penetrate the BBB.

The primary objective of this study was to identify signaling pathways and specific genes that could potentially determine the effectiveness of endoxifen in GBM and to subsequently validate these predicted effects through pilot in vitro and in vivo experiments. Key findings from the computational analysis revealed a subset of 1,463 genes that are strongly and robustly perturbed in GBM and also undergo changes after endoxifen treatment in MCF7 breast cancer cells. Of particular interest among these were 560 genes that were upregulated in GBM and downregulated by endoxifen in MCF7 cells, as they may elucidate the potential anti-cancer effects of endoxifen in GBM. These genes were enriched in hallmarks associated with cell proliferation, mTORC1 pathway, estrogen and androgen signaling, apoptosis, immune response, metabolic reprogramming, and EMT, all of which are known to play crucial roles in GBM53.

The stimulation of cell cycle progression and the proficient ability of GBM cells to repair treatment-induced DNA damage during the G2 phase of cell cycle contribute, at least partially, to their resistance to irradiation and chemotherapy. Several G2/M checkpoint inhibitors have been investigated in preclinical and clinical studies with the aim of reducing tumor growth by interrupting the replicative cycle of cancer cells54. Endoxifen effectively inhibited cell cycle-related pathways in MCF7 cells, and this effect was further confirmed in the current study in proliferation assays using GBM cells in vitro; however, this did not translate into measurable tumor shrinkage in vivo, highlighting the need for further investigation before its therapeutic potential can be fully assessed. Among other endoxifen-regulated hallmarks Myc targets may play a pivotal role in mitotic stimulation and in maintaining the self-renewal capacity and tumorigenic potential of glioma cells55,56. Additionally, high-level amplification of Myc has been detected in a subset of GBM, driving multiple cancer phenotypes, including altered metabolism to support rapid cell growth and proliferation57,58,59.

Modulation of mTORC1 signaling was also revealed as a potential MOA of endoxifen in GBM in this study. High doses of endoxifen were shown to inhibit AKT signaling and downstream pathways, including mTOR27. At the same time, abnormal activation of the PI3K/Akt/mTOR pathway has been associated with the development of GBM59,60, and is associated with alterations in several signaling proteins that are hallmarks of GBM pathogenesis, such as PTEN loss of function and EGFR amplification/mutation61,62. Therefore, inhibition of the PI3K/Akt/mTOR pathway represents an important therapeutic strategy against GBM63, and may contribute to anticancer effects of endoxifen in this disease.

ER signaling has a complex interplay with GBM behavior; while GBM prevalence is higher in males64, this disparity decreases after menopause65, implicating estrogen in tumor development. Estradiol (E2) has been shown to exert tumor promoting effects on GBM, including stimulation of cell proliferation66, migration, invasion and epithelial-mesenchymal transition (EMT)67 in human GBM cells. These effects are concentration-dependent68 and also dependent on the predominant signaling of the receptor subtype, ERα and ERβ, both of which are widely expressed in astrocytoma tissues69. Conflicting findings exist regarding the impact of ERα, with some studies indicating ERα to promote GBM progression67,70,71, while others show improved survival with high ERα expression72. Conversely, ERβ's antineoplastic role is more consistently supported, with evidence suggesting that its expression decline correlates with tumor grade progression73,74. Given these findings, strong inhibition of ERα signaling52 combined with the stabilizing activity toward ERβ28 underscore the therapeutic potential of endoxifen in the treatment of GBM.

Among the signaling pathways that may contribute to the anticancer effects of endoxifen in GBM, the E2F pathway was identified through both GSEA and transcription factor enrichment analysis. E2F7 and E2F8 are known to promote cell proliferation, cell-cycle progression, and tumorigenicity in GBM, highlighting their potential as therapeutic targets75. Key oncogenic pathways associated with E2F7 and E2F8 include EMT, NF-κB, STAT3, and pathways involved in angiogenesis76. Additionally, E2F7 and E2F8 confer significant radioresistance to GBM tumor cells75.

Among the transcription factors that may play a crucial role in the anticancer action of endoxifen, PRMT3 and CENPA, in addition to E2F7 and E2F8, were the most highly enriched in the downstream genes activated in GBM and downregulated in MCF7 cells treated with endoxifen. PRMT3 belongs to the Protein Arginine Methyltransferases (PRMTs) family, which acts as epigenetic regulators of transcription77. A recently published study demonstrated the role of PRMT3 as a driver of GBM progression78. Knockdown of PRMT3 reduced the proliferation and migration of GBM cell lines and patient-derived GBM stem cells, whereas overexpression of PRMT3 increased the proliferative capacity of cells by promoting cell cycle progression78. CENPA encodes the Centromere Protein A (CENP-A), a histone H3 variant crucial for chromosomal segregation during cell division and implicated in various cancer types79. Recently, CENP-A has been identified as a potential prognostic biomarker in glioma patients, with higher CENP-A expression levels significantly associated with tumor grade, isocitrate dehydrogenase (IDH) status, immune infiltration levels, primary therapy outcome and shorter overall survival80. Additionally, downregulation of CENPA has been shown to inhibit glioma cell migration and proliferation81. These findings suggest that PRMT3 and CENP-A may be involved in endoxifen’s mechanism of action in GBM and warrant further investigation as potential therapeutic targets.

To validate the bulk analysis data and identify genes specifically altered in neoplastic cells in GBM that may be influenced by endoxifen, scRNA-seq analysis was conducted. This analysis identified 113 intersecting genes between bulk GBM, scRNA-seq GBM, and endoxifen gene signatures. Among these, 73 genes were upregulated in both bulk and scRNA-seq GBM datasets and downregulated following endoxifen treatment. Although this subset of genes was small, it significantly enriched signaling pathways such as cell cycle-related hallmarks (G2-M checkpoint, E2F targets, Myc targets), as well as apoptosis and immune response hallmarks (inflammatory response, interferon gamma response, IL-2/STAT5 signaling, and coagulation). These findings parallel those from the larger subset of genes identified in the bulk analysis, underscoring the importance of these signaling pathways in the potential efficacy of endoxifen in GBM. The analyses highlighted three genes among those identified that were additionally associated with poor survival in GBM patients: RPA3, NFKBIZ, and HSPB1. A review of the literature confirmed these findings underscoring the importance of these genes in the development of GBM and provided insights into their potential roles in the MOA of endoxifen.

RPA3 encodes Replication protein A (RPA), which is the major eukaryotic single-stranded DNA-binding protein involved in DNA repair and replication82. Variations in the RPA3 gene were associated with an increased risk of glioma in a Chinese Han population83. The expression of RPA3 in glioma tissue and cells is significantly higher than in normal glial cells and is positively correlated with poor prognosis of patients with gliomas84. Overexpression of RPA3 in glioma cells activates the PI3K/Akt/mTOR pathway, thereby promoting proliferation, migration and invasion84. Downregulation of RPA3 expression by endoxifen may inhibit the PI3K/Akt/mTOR pathway, in addition to its direct modulation of this pathway through PKCβ27.

NFKBIZ (IκBζ) is an atypical member of the ankyrin-repeat family which mediates inflammatory responses to lipopolysaccharides (LPS) and other proinflammatory stimuli by interacting with NF-κB proteins through its ankyrin-repeat domains85. IκBζ is upregulated in a glioma cell line resistant to NF-κB-dependent non-apoptotic cell death and is a biomarker of inflammatory microglia86. In glioma patients, elevated IκBζ expression in tumor specimens correlates with poor prognosis. Following γ-irradiation of glioma cells, increased expression of IκBζ acts as a transcriptional activator for the tumor-promoting cytokines IL-6, IL-8, and CXCL1, which are implicated in glioma-associated inflammatory processes. Conversely, shRNA-mediated knockdown of IκBζ results in reduced expression of these cytokines87. Inhibition of NFKBIZ expression by endoxifen may exert an anti-inflammatory effect in GBM which is supported by the inhibition of TNF-alpha signaling via NF-κB pathway after endoxifen treatment.

The HSPB1-encoded protein Hsp27 (HspB1) functions as a molecular chaperone that facilitates the correct folding of other proteins and plays an important role in the differentiation of a wide variety of cell types88. HSPB1 expression has been shown to increase in glioma cells89 and is a predictive factor of poor prognosis for GBM90. High HSPB1 expression is associated with poor overall survival and significantly correlates with pathologic stages, as well as estrogen and progesterone receptors91. Additionally, treatment of GBM cells with TMZ increases HSPB1 expression, contributing to the chemoresistance92. Conversely, inhibition of HSPB1 reduces drug resistance in glioma cells, making Hsp27-targeted therapies using natural compounds a promising strategy for GBM treatment93,94. The current study’s scRNA-seq analysis also demonstrated an association between the HSPB1 gene and the more aggressive mesenchymal subtype of GBM, further underscoring its role in GBM progression. Notably, previous research has shown that 17β-estradiol treatment significantly increases HSPB1 expression95. In contrast, this study observed that endoxifen decreased HSPB1 expression in MCF7 cells, suggesting that the effect of endoxifen may be mediated through an estrogen-dependent mechanism.

The results from computational analyses yielded substantial insights into the potential MOA of endoxifen in GBM. To validate endoxifen’s efficacy, in vitro proliferation and cytotoxicity assays were conducted using a GBM cell line CRT435. Importantly, the CRT435 cell line is clinically relevant since it is a patient-derived xenograft (PDX) cell culture model derived from a patient with recurrent GBM. Cell proliferation and apoptosis analyses revealed that endoxifen exhibited a strong cytotoxic effect, surpassing that of both the positive control (10% DMSO) and the standard care treatment TMZ at the concentrations tested. Moreover, combining endoxifen with TMZ resulted in a greater induction of cell death than either agent alone, warranting further in vivo experiments. For this an in vivo model utilizing the CRT435 cell line was developed in immunosuppressed mice. Notably, all doses of endoxifen, whether administered as monotherapy or in combination with TMZ, were well-tolerated; they did not affect animal body weight and did not produce adverse clinical signs. However, there were no significant differences observed between treatment groups receiving endoxifen at three different doses compared to the control. Additionally, the combination of endoxifen with TMZ did not produce effect in these experimental settings. The modest results of the in vivo experiment may be partly attributed to the small number of animals in the treatment groups and the use of subcutaneous administration of GBM cells instead of orthotopic administration in the brain, as subcutaneous administration of the PDX cell model fails to account for the CNS tumor microenvironment. Such models lack key features, including the BBB, brain-specific stroma, and immune microenvironment, and therefore do not capture the unique growth patterns or drug penetration challenges of intracranial tumors. It is also possible that the limited efficacy reflects intrinsic biological variability in GBM, including differential expression of ERα/ERβ among tumors, which may influence responsiveness to endoxifen. Furthermore, pharmacokinetic factors (suboptimal drug exposure) could also have contributed to the lack of in vivo activity. These considerations highlight the need for further preclinical studies using orthotopic GBM models and optimized dosing strategies to more accurately evaluate the therapeutic potential of endoxifen.

One limitation of this study is that the global gene expression data following endoxifen exposure were derived from a breast cancer model, in which high ERα signaling is well established to promote tumor growth, which may not fully reflect the variable ERα/ERβ expression and signaling dynamics observed in GBM. In addition, the in vitro assays were performed in a single patient-derived GBM cell line (CRT435), and given the molecular and phenotypic heterogeneity of GBM, further studies across multiple GBM models with diverse receptor profiles will be necessary to determine whether these findings are broadly applicable. Nonetheless, endoxifen demonstrated cytotoxic and anti-proliferative effects in a GBM cell model, supporting its general anticancer molecular MOA. Additional experimental validation is necessary to confirm the identified changes in signaling pathways and specific genes, as well as to uncover new potential effects of endoxifen in the context of GBM.

Another limitation of the current study is its predominant reliance on mRNA changes, which may be challenging to translate to the protein level due to the lack of correlation between mRNA and protein expression, caused by internal adaptive mechanisms such as translational offsetting96. This phenomenon is particularly documented for estrogen-regulated transcripts97 and may significantly affect the analysis of estrogen-dependent mechanisms. Nonetheless, this study underscores the utility of transcriptomic data in analyzing drug mechanisms and highlights the potential of integrating multi-omics data, including proteomics and metabolomics, to achieve a more comprehensive understanding of molecular effects. Future studies will integrate proteomic and metabolomic analyses to complement transcriptomic data and provide a more comprehensive understanding of endoxifen’s effects in GBM.

Conclusions

In summary, this study provides preliminary evidence for the potential of repurposing endoxifen as a novel therapeutic agent for GBM. The comprehensive computational and experimental analyses highlight the versatility of endoxifen’s MOA and its ability to modulate key signaling pathways involved in GBM progression. These findings pave the way for further preclinical and clinical investigations to validate the efficacy of endoxifen in GBM treatment and potentially improve the prognosis for patients with this devastating disease.