Pervasive midline glioma accurate drug screening method, prognosis model and application
By integrating transcriptome data and machine learning methods, accurate drugs for diffuse midline glioma of H3K27M mutant type were screened, solving the problem of low screening efficiency in the prior art, and achieving effective killing and prognosis prediction of H3K27M mutations.
Patent Information
- Application Number
- CN202510486088.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-17
- Publication Date
- 2025-09-02
AI Technical Summary
It is difficult for the prior art to accurately screen out drugs that are effective for diffuse midline glioma (H3K27M mutant type), and traditional drug screening strategies ignore the global apparent effects of H3K27M mutation, resulting in low screening efficiency and inability to achieve killing of pan-H3K27M mutations.
By integrating transcriptome data, differential genes were analyzed using Limma package, combining WGCNA algorithm to identify gene modules, machine learning methods such as LASSO, Random Forest and XGBoost were used to screen key characteristic genes, and CTRP and GDSC databases were used to predict drug sensitivity, perform high-throughput drug screening, and verify drug effects in vitro and in vivo.
The precise drugs with broad-spectrum killing ability for H3K27M mutant diffuse midline glioma were successfully screened, providing a new path to drug discovery, significantly improving therapeutic potential, and establishing an effective prognostic prediction model.
Smart Images
Figure CN120581067A_ABST
Abstract
Description
Field of the Invention
[0001] The present application belongs to the field of tumor treatment. Specifically, the present application provides a precise drug screening method, prognosis model and application for diffuse midline glioma. Background Art
[0002] Diffuse midline glioma (DMG) is one of the most common brain tumors in children. Characterized by high mortality and disability rates, difficult treatment, and high recurrence rates, DMG has become the leading cause of death from brain tumors in children. H3K27M mutation is a hallmark of DMG, accounting for up to 80% of pediatric brainstem gliomas. This mutation results in loss of methylation at lysine 27 of histone H3, leading to a significant decrease in global H3K27 trimethylation (H3K27me3) through epigenetic reprogramming, thereby activating oncogene expression and inhibiting differentiation-related pathways. This epigenetic dysregulation not only drives malignant tumor progression but also enhances tumor cell resistance to radiotherapy and chemotherapy through metabolic reprogramming, making traditional drugs difficult to inhibit tumor progression. Furthermore, DMG is also anatomically restricted. DMG is often located in deep midline structures such as the brainstem and thalamus. Its infiltrative growth pattern makes surgical resection extremely risky, necessitating the development of new precision medicine treatments.
[0003] In recent years, drug screening for DMG has not achieved significant breakthroughs, primarily due to outdated screening methods and a lack of research models. Most current drug screening studies are still based on small sample sizes and commercially available drug libraries, or are developed through multidisciplinary discussions. This creates redundancy between targets and drugs, is subjective, and often overlooks the global epigenetic effects of H3K27M mutations, resulting in low screening efficiency. In contrast, in studies of other central nervous system tumors (such as supratentorial glioblastoma and lymphoma), multi-omics sequencing is often used to classify tumor subtypes and screen for potential targeted drugs based on exonic mutations or signaling pathway regulation. While this strategy has achieved some success in relatively common tumor types, it has been very limited in rare intracranial tumors such as brainstem gliomas, particularly diffuse midline gliomas. This is primarily due to the fact that drug libraries designed based on tumor mutation subtypes or signaling pathways are limited to a few DMG subtypes. Even if sensitive drugs are identified, they often fail to achieve pan-H3K27M targeting, resulting in uncertain translational prospects. The fundamental reason is that, apart from the H3K27M mutation, DMG lacks other mutation sites or signaling pathways that have been clearly closely related to the pathogenesis, making it difficult for existing drug screening strategies to accurately match the molecular characteristics of DMG.
[0004] H3K27M is an exonic mutation of the structural protein histone. Currently, no drugs are available to treat this target. Furthermore, this structural protein is closely associated with important functions such as DNA replication, cell cycle regulation, and chromatin stability, making the design of drugs targeting this mutated structural protein impractical. Therefore, it is more reasonable and realistic to shift the focus from exonic mutations to the global changes in the transcriptome caused by mutations, making this the more reasonable and realistic research focus. Furthermore, transcriptome data is the most accessible and cost-effective high-throughput sequencing data. As the most common mRNA differential expression in omics data, it has the potential to bridge clinical and pathological diagnoses. Summary of the Invention
[0005] The inventors have established the world's largest transcriptome cohort of brainstem glioma (covering DMG) patients, and simultaneously completed transcriptome determination of patient-derived brainstem glioma cell lines (covering DMG). By integrating and analyzing transcriptome data to cross-identify H3K27M robust differentially expressed genes and incorporating machine learning to screen key characteristic genes, it is ultimately possible to achieve transcriptome-guided DMG characteristic differentially expressed gene picking, pathway enrichment, key characteristic gene screening, and drug sensitivity prediction. Furthermore, high-throughput precision drug screening based on the above four steps was completed to obtain sensitive drugs for the precise targeted treatment of DMG. At the same time, a prognostic model based on key characteristic genes was established to provide a potential means of diagnosis, treatment, and monitoring for future clinical transformation.
[0006] In one aspect, the present application provides a method for precise drug screening of diffuse midline glioma, wherein the diffuse midline glioma is an H3K27M mutant, and the method comprises:
[0007] (1) Analyze the differentially expressed genes between H3K27M mutant and wild-type diffuse midline gliomas, identify gene modules associated with H3K27M mutations, analyze the intersection of differentially expressed genes and gene modules, obtain robust differentially expressed genes, and analyze their enriched pathways;
[0008] (2) Drug screening targeting H3K27M key differential pathways: Drug screening is performed among drugs that can intervene in the enriched pathways obtained in step (1);
[0009] (3) Drug screening targeting H3K27M robust differentially expressed genes: Analyze the genes associated with prognosis among the robust differentially expressed genes obtained in step (1), and screen for drugs that can intervene in these prognosis-related robust differentially expressed genes;
[0010] (4) Drug screening targeting H3K27M key characteristic genes: Analyze genes associated with poor prognosis and screen drugs that affect the expression of these poor prognosis-related genes.
[0011] Furthermore, in step (1), the Limma package was used to analyze the differentially expressed genes between H3K27M mutant and wild-type diffuse midline gliomas; the WGCNA algorithm was used to identify gene modules related to H3K27M mutations; and the intersection of the differentially expressed genes and the gene modules was obtained through the Venn diagram.
[0012] Furthermore, the enriched pathways obtained in step (1) are microtubule pathways and actin pathways.
[0013] Furthermore, in step (2), drug screening is performed among microtubule inhibitors and actin inhibitors.
[0014] Furthermore, in step (3), COX regression analysis is used to identify genes related to prognosis among the robust differentially expressed genes obtained in step (1).
[0015] Furthermore, the prognosis-related robust differentially expressed genes in step (3) were TOP2A, RRM2, and TTK.
[0016] Furthermore, in step (4), the Lasso regression algorithm, random forest algorithm and Xgboost algorithm are used to analyze genes related to poor prognosis, and the intersection of the three algorithms is taken as the genes related to poor prognosis through the Venn diagram.
[0017] Furthermore, genes associated with poor prognosis were DEPDC1, NUF2, and ZFHX4. Drugs that could reduce the expression of DEPDC1 or NUF2 and increase the expression of ZFHX4 were predicted using the GDSC and CTRP databases, and screening was performed among drugs that could reduce the expression of DEPDC1 or NUF2 and increase the expression of ZFHX4.
[0018] Furthermore, the method further includes (5) applying the drugs screened in steps (2)-(4) to cells, sequencing the transcriptome of the cell line before and after administration, and performing pathway enrichment analysis on differentially expressed genes; and verifying the effects of the drugs screened in steps (2)-(4) in animal models and diffuse midline glioma organoids.
[0019] On the other hand, the present application provides the use of Verubulin, Filanesib, BAY1217389 or STF-31 in the preparation of a drug for treating H3K27M mutant diffuse midline glioma.
[0020] Furthermore, the drug is in the form of intravenous injection, in situ administration or intrathecal administration.
[0021] The verubulin is a compound with CAS No. 827031-83-4, Filanesib is a compound with CAS No. 885060-09-3, BAY1217389 is a compound with CAS No. 1554458-53-5, and STF-31 is a compound with CAS No. 724741-75-7. The above compounds / drugs in this application include pharmaceutically acceptable salt forms thereof.
[0022] On the other hand, the present application provides a prognosis prediction model for diffuse midline glioma with H3K27M mutation, in which the expression levels of DEPDC1, NUF2 and ZFHX4 are used as variables for COX regression analysis.
[0023] On the other hand, the present application provides the use of a reagent for detecting the expression levels of DEPDC1, NUF2 and ZFHX4 in preparing a prognosis judgment kit for diffuse midline glioma with H3K27M mutation.
[0024] Furthermore, the reagent for detecting the expression levels of DEPDC1, NUF2 and ZFHX4 is a qRT-PCR reagent or a microarray chip.
[0025] Furthermore, when the kit is used, COX regression analysis is performed using the obtained expression levels of DEPDC1, NUF2 and ZFHX4 as variables.
[0026] The bioinformatics analysis described in the present invention used the software Rstudio (v4.2) and R (v4.2.3).
[0027] The present invention uses the Limma package for differential gene analysis. Limma (Linear Models for Microarray Data) is an R toolkit designed for gene expression data analysis. It was originally used for microarray data and is now widely used in RNA-seq analysis. It uses linear models to assess gene expression differences and combines empirical Bayesian methods to improve statistical stability. LIMMA has functions such as data preprocessing, normalization, batch effect adjustment, and multiple testing correction, and is an important tool in high-throughput genomics research. Before analysis, transcriptome sequencing data were converted to counts or TPMs; two groups of brainstem glioma (including DMG) transcriptome cohorts (GSE50774; EGAS00001004341) and one group of brainstem glioma cell line (including DMG) transcriptome data were included. The Limma package was used to identify differentially expressed genes between H3K27M mutations and H3 wild-type (|Log2FC>1|, p.adjust<0.05) for the three sets of transcriptome data, draw volcano plots, and perform differential gene pathway enrichment analysis.
[0028] The pathway enrichment analysis of the present invention involves the KEGG database, the GO database, and the Msigdb_HALLMARK dataset, and uses the clusterProfile package to complete the pathway enrichment analysis (p < 0.05, q < 0.05). Pathway enrichment analysis is a computational method used to reveal the functional trends of genes or protein sets in biological pathways, which can help researchers identify which pathways are significantly activated or inhibited in specific biological processes. clusterProfiler is specifically used for gene function annotation and pathway enrichment analysis. It supports multiple databases and can be used for methods such as hypergeometric test and gene set enrichment analysis (GSEA). The results are beautified and visualized using the enrichplot and ggplot2 packages.
[0029] The present invention uses the WGCNA algorithm to identify key gene expression modules representing the H3K27M mutation signature. WGCNA (Weighted Gene Co-expression Network Analysis) is an R package for constructing and analyzing gene co-expression networks. It constructs a weighted network by calculating gene expression similarity and clusters genes using a topological overlap matrix (TOM) to identify gene modules with high co-expression. These modules are typically associated with specific biological features or disease states. The WGCNA algorithm was used to identify the gene modules most relevant to H3K27M in the EGAS00001004341 cohort.
[0030] In the present invention, the three groups of differentially expressed genes were intersected with the green module of WGCNA, and the Venn diagram was used to obtain the robust differentially expressed genes in diffuse midline glioma. The robust differentially expressed genes were subjected to the above pathway enrichment analysis to identify the H3K27M-specific pathway.
[0031] The present invention uses univariate Cox regression analysis associated with prognosis to screen for robust differentially expressed genes (H3K27M-Robust Genes) associated with the prognosis of diffuse midline glioma (p < 0.05). Univariate Cox regression analysis is a statistical method used to evaluate the impact of a single variable on survival time. Based on the Cox Proportional Hazards Model, it calculates the hazard ratio (HR, Hazard Ratio) and its significance for each variable to determine whether the variable is associated with survival outcomes. This method can retain differentially expressed genes associated with the prognosis of patients with H3K27M mutations.
[0032] The present invention also uses a variety of machine learning methods to identify and screen robust differentially expressed genes after screening by univariate COX regression analysis, retaining key characteristic genes. LASSO (Least Absolute Shrinkage and Selection Operator) is a regression method used for feature selection and model compression. It shrinks some regression coefficients to zero by adding an L1 norm penalty term to the regression model, thereby screening out key variables and avoiding model overfitting. LASSO is particularly commonly used in high-dimensional data analysis, such as feature screening in gene expression data, to help identify important biomarkers associated with disease or survival outcomes. In survival analysis, LASSO can be combined with COX regression (LASSO-Cox) to screen variables with the greatest prognostic value and construct a risk prediction model. Random Forest is an ensemble learning method consisting of multiple decision trees. Training data is randomly sampled using the bootstrap method, and each tree is constructed in combination with random feature selection. The prediction results are ultimately obtained by voting or averaging. Compared to a single decision tree, Random Forest can effectively reduce overfitting and has stronger noise resistance and generalization capabilities. It is widely used in tasks such as classification, regression, and survival analysis. In bioinformatics, it can be used for gene feature screening, disease classification, and prognosis model construction. XGBoost (eXtreme GradientBoosting) is an efficient gradient boosting algorithm that uses a weighted decision tree to gradually optimize the model and improve prediction accuracy. It enhances computational efficiency and prevents overfitting through mechanisms such as greedy algorithms, second-order gradient optimization, and regularization terms. XGBoost has parallel computing capabilities and is suitable for large-scale data processing. It performs well in tasks such as classification, regression, and survival analysis. In bioinformatics, XGBoost is commonly used for gene feature screening, disease risk prediction, and clinical prognosis modeling. The present invention uses the above three methods (LASSO, random forest, and XGBoost) to screen robust differentially expressed genes related to prognosis, and takes the intersection to obtain three key feature genes: DEPDC1 / NUF2 / ZFHX4 (H3K27M-KeyGenes).
[0033] The present invention uses the CTRP and GDSC databases to predict potentially effective therapeutic drugs for H3K27M mutations based on three key characteristic genes. The CTRP (Cancer Therapeutics Response Portal) and GDSC (Genomics of Drug Sensitivity in Cancer) databases are two important cancer drug sensitivity datasets used to study drug responses and sensitivities in cancer treatment and help develop precision medicine treatment strategies. They provide response data of various cancer type cell lines to different chemotherapy drugs and targeted drugs, helping researchers understand the mechanism of drug response. The database reveals the differences in sensitivity of tumor cells to drugs through correlation analysis between gene expression data and drug sensitivity, which helps provide guidance for personalized treatment.
[0034] The present invention further screens drugs targeting the aforementioned H3K27M-specific pathways, robust genes, and key genes in a multi-mutation panel of patient-derived brainstem glioma cell lines (both DMG and non-DMG) constructed in vitro, as well as a normal brainstem stem cell line (PPC). All inhibitors of these targets / pathways were extracted from the CellCycle / DDR drug compound library (MCE). Echo sonar automated drug addition was performed in vitro, and cell viability was measured using a Cell-titer assay. Relative cell viability (%) was calculated as: fluorescence intensity of the drug-treated group / fluorescence intensity of the control group × 100%, and effective drugs were screened.
[0035] The present invention also established an orthotopic brainstem tumor model in mice and tested the efficacy and safety of the screened drugs in vivo. Based on the drug's physicochemical properties, an in vivo administration route that matches the clinical setting was selected. For example, Verubulin can cross the blood-brain barrier, allowing systemic administration; BAY1217389 and Filanesib cannot cross the blood-brain barrier, allowing for in situ administration; and STF-31, while unable to cross the blood-brain barrier, has a good safety profile and can be considered for intrathecal administration.
[0036] After identifying the precise drug, the present invention conducted transcriptome sequencing analysis of the cell lines after drug administration. The transcriptome comparison of differentially expressed genes and enriched pathways before and after drug administration demonstrated the effectiveness of the precise screening strategy.
[0037] Based on the brainstem glioma organoid model cultivation method described in patent CN118126952A, this paper constructs patient-derived diffuse midline glioma organoids. Using these organoids, a clinical screening simulation for the aforementioned precision medicine was performed. Simulated dosing was performed once a week, and morphological changes in the organoids were observed after two doses to monitor the tumor damage reflected by the drug.
[0038] The present invention provides a predictive model for the prognosis of diffuse midline glioma. Based on the aforementioned key characteristic genes DEPDC1 / NUF2 / ZFHX4 (H3K27M-Key Genes), a multivariate COX regression analysis model was performed, and a Norman plot was drawn. At the same time, a calibration curve, a clinical decision curve, and an ROC curve were drawn, and the area under the curve (AUC) was calculated to characterize the robustness and effectiveness of the model. Multivariate Cox Regression Analysis is a survival analysis method based on the Cox proportional hazard model, which is used to simultaneously evaluate the effects of multiple variables on survival time. Unlike univariate COX regression, multivariate analysis can adjust confounding factors and more accurately identify independent prognostic factors.
[0039] Beneficial effects:
[0040] Innovative Drug Screening Strategy: This study successfully screened for precision medicine targeting the H3K27M mutation characteristic of diffuse midline glioma (DMG) through deep bioinformatics analysis. By incorporating multiple advanced machine learning methods (such as LASSO, Random Forest, and XGBoost), the team identified H3K27M-specific pathways, robust differentially expressed genes, and key genes, enabling precision drug screening. This breakthrough in traditional drug screening provides a new path for drug discovery.
[0041] The selected drugs have significant therapeutic advantages: In vitro and in vivo experiments have demonstrated broad-spectrum anti-H3K27M activity, effectively combating diffuse midline gliomas and demonstrating significant therapeutic potential. The drugs' extremely low IC50 values provide a wider therapeutic window for clinical application.
[0042] Innovative screening model that matches clinical practice: The in vivo validation experiment of this invention fully considers the physicochemical properties of the drug and its potential clinical application pathways, and selects a dosage method that adapts to the physicochemical properties and efficacy of the drug; at the same time, the new screening model based on organoids provides a new paradigm for future clinical transformation.
[0043] A prognostic prediction model for diffuse midline glioma has been established: This study successfully constructed a prognostic prediction model based on multivariate COX regression analysis of H3K27M signature genes (DEPDC1, NUF2, and ZFHX4). This model effectively assesses a patient's prognostic risk and, combined with tools such as Norman plots, calibration curves, and receiver operating characteristic (ROC) curves, validates its accuracy and robustness, providing a reliable basis for clinical decision-making. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 This is a schematic diagram of the process of the precise drug screening method according to the present invention;
[0045] Figure 2A Differential gene analysis process for extracting H3K27M features of multiple cohorts in this invention: cluster dendrogram;
[0046] Figure 2B The differential gene analysis process for extracting H3K27M features of multiple cohorts in this invention: the association diagram between modules and phenotypic features;
[0047] Figure 2C The differential gene analysis process for extracting H3K27M features in multiple cohorts for this invention: Correlation analysis between expression levels and modules; WGCNA analysis process identifies that the green module is closely related to H3K27M mutations;
[0048] Figure 2D The differential gene analysis process for extracting H3K27M features of multiple cohorts for this invention was as follows: the differential genes of different transcriptome cohorts were intersected with the green module genes identified by WGCNA, and 34 genes were simultaneously present in the above gene set (CL-DEGs: differential genes of cell line transcriptomes);
[0049] Figure 3A The results of KEGG pathway enrichment analysis of 34 genes;
[0050] Figure 3B The results of GO pathway enrichment analysis of 34 genes;
[0051] Figure 3C The results of Msigdb_HALLMARK pathway enrichment analysis of 34 genes;
[0052] Figure 4A Results of high-throughput drug screening for tubulin inhibitors;
[0053] Figure 4B The IC50 curves of tubulin inhibitor cell lines, high-throughput drug screening results of actin inhibitors, and IC50 curves of actin inhibitor cells are shown;
[0054] Figure 5Univariate COX regression analysis showed that 28 genes were associated with poor prognosis due to H3K27M mutations, among which TOP2A, TTK, and RRM2 were target genes that could be directly intervened by drugs;
[0055] Figure 6A Results of high-throughput drug screening for TOP2A inhibitors and IC50 curves of TOP2A inhibitors in cell lines;
[0056] Figure 6B Results of high-throughput drug screening for TTK inhibitors and RRM inhibitors, as well as IC50 curves for TTK inhibitors and RRM inhibitors;
[0057] Figure 7A The present invention uses a variety of machine learning methods to analyze key characteristic genes: Lasso regression model establishment process;
[0058] Figure 7B For the present invention, key feature genes were analyzed by various machine learning methods: random forest model;
[0059] Figure 7C The present invention uses a variety of machine learning methods to analyze key feature genes: XGBoost model;
[0060] Figure 7D For this invention, key feature genes were analyzed by various machine learning methods: a total of 3 genes were identified as robust candidate genes by taking the intersection of the three models;
[0061] Figure 8 Western blotting and real-time fluorescence quantitative PCR were performed to verify the key characteristic genes of the present invention; Part A: Western blotting experiments confirmed that the expression of the three candidate genes was significantly different between the H3K27M mutation and wild-type groups; Part B: Real-time fluorescence quantitative PCR confirmed that the expression of the three candidate genes was significantly different between the H3K27M mutation and wild-type groups;
[0062] Figure 9A For the present invention, sensitive drug prediction and drug screening are performed on key characteristic genes: CTRP database predicts drugs;
[0063] Figure 9B For the present invention, sensitive drug prediction and drug screening are performed on key characteristic genes: GDSC database predicts drugs;
[0064] Figure 9C For the present invention, sensitive drug prediction and drug screening are performed on key characteristic genes: predicted high-throughput drug screening results of cell lines with potential drug sensitivity;
[0065] Figure 9DFor the present invention, sensitive drug prediction and drug screening are performed on key characteristic genes: cell line IC50 curves of candidate drugs;
[0066] Figure 10A Transcriptome pathway enrichment analysis after administration of the sensitive drugs finally screened for the present invention: GO enrichment analysis of the transcriptome after administration of BAY1217389;
[0067] Figure 10B Transcriptome pathway enrichment analysis after administration of the sensitive drugs finally screened for the present invention: Transcriptome GO enrichment analysis after administration of Verubulin;
[0068] Figure 10C Transcriptome pathway enrichment analysis after administration of the sensitive drugs finally screened for the present invention: KEGG enrichment analysis after administration of Filanesib;
[0069] Figure 10D Transcriptome pathway enrichment analysis after administration of the sensitive drugs finally screened for the present invention: GO enrichment analysis after administration of STF-31;
[0070] Figure 11 To conduct in vivo testing and verification of the final drug screening of the present invention;
[0071] Figure 12 The present invention simulates actual clinical screening scenarios based on organoids;
[0072] Figure 13A Prognostic prediction model and validation curve established for the present invention: Norman diagram;
[0073] Figure 13B The prognostic prediction model and validation curve established for the present invention: prognostic KM curve based on Norman diagram risk score;
[0074] Figure 13C The prognostic prediction model and validation curve established for the present invention: a calibration curve of the prognostic model;
[0075] Figure 13D The prognostic prediction model and validation curve established for the present invention: clinical decision curve;
[0076] Figure 13E The prognostic prediction model and validation curve established for the present invention: ROC curve of 6-month survival;
[0077] Figure 13F The prognostic prediction model and validation curve established for the present invention: ROC curve of 1-year survival. DETAILED DESCRIPTION
[0078] The overall implementation flow chart is as follows Figure 1 shown.
[0079] Example 1 Analysis of differentially expressed genes with H3K27M mutations
[0080] Transcriptome data from two external brainstem glioma cohorts (GSE50774 and EGAS00001004341) (covering both DMG and non-DMG) and one brainstem glioma cell line cohort (covering both DMG and non-DMG, with mutational profiles as shown in Table 1) were included. The Limma package was used to identify differentially expressed genes (|Log2FC>1|, p.adjust<0.05) between H3K27M mutants and H3 wild-types in the three transcriptome data sets. GSE50774 had 303 upregulated and 114 downregulated genes; EGAS00001004341 had 1263 upregulated and 1121 downregulated genes; and the cell lines had 721 upregulated and 69 downregulated genes.
[0081] Table 1
[0082]
[0083] The WGCNA algorithm was used to analyze the EGAS00001004341 cohort and identify the gene module most associated with H3K27M mutations ( Figure 2A-2C Among them, the green module has the highest correlation, with a correlation index of 0.62 (P<0.0001).
[0084] The intersection of the three groups of differentially expressed genes and the WGCNA green module was analyzed by Venn diagram, and 34 robust differentially expressed genes were obtained ( Figure 2D ).
[0085] KEGG, GO and Msigdb_HALLMARK pathway enrichment analysis ( Figure 3A-3C ) showed that the main enriched pathways of the 34 differentially expressed genes were actin and microtubule pathways.
[0086] Example 2 Drug Screening Targeting H3K27M Key Differential Pathway
[0087] The actin and microtubule pathways were enriched as target pathways that could be intervened by drugs. High-throughput drug screening was performed on the enriched actin and microtubule pathways. Echo sonar was used for automatic drug addition and Cell-Titer (Promega, ) Experimental detection of cell viability (Relative cell viability calculation formula: fluorescence intensity of drug-treated group / fluorescence intensity of control group × 100%, drug source: MCE, the same detection method below). The results of microtubule pathway inhibitor screening were visualized using the pheatmap package, showing that microtubule inhibitors are generally effective, among which Verubulin performed best at a concentration of 1μM, and had a significant killing effect on H3K27M mutant brainstem glioma cell lines, with an IC50 value of 600pM, which is the lowest reported so far, and was selected as the final drug. The results of actin inhibitor screening also showed that they were generally effective, with Filanesib having the best effect at a concentration of 1μM, with an IC50 value of 600pM ( Figure 4A-4B ), and the final drugs were selected. This completes the drug screening for the H3K27M-specific pathway.
[0088] Example 3 Drug screening targeting H3K27M robust differentially expressed genes
[0089] Univariate COX regression analysis was performed on 34 robust differentially expressed genes, and 28 genes were significantly associated with prognosis (P < 0.05), of which TOP2A, RRM2, and TTK are known drug targets (H3K27M-RobustGenes) ( Figure 5 ). These three genes are highly expressed in the H3K27M mutation group and are associated with poor prognosis. The corresponding inhibitors were selected for high-throughput screening. Amsacrine, Teniposide, BAY1217389, etc. had the best effect at a concentration of 1μM, and had a significant killing effect on H3K27M mutant brainstem glioma cells. The IC50 determination results showed that: TOP2A inhibitors had a weakened effect on cells with simultaneous TP53 mutations, RRM2 inhibitors had a higher IC50, and TTK inhibitors had the lowest IC50. Among them, BAY1217389 performed best with an IC50 of 800pM, and was finally selected as the final drug ( Figure 6A-6B ). This completes the drug screening targeting H3K27M-Robust Genes.
[0090] Example 4 Prognostic Association Analysis Using Machine Learning Algorithms
[0091] Lasso regression showed that six genes including DEPDC1, NUF2, ZFHX4, CENPE, and NCAPG could predict poor prognosis. Figure 7A The random forest algorithm identifies genes most associated with poor prognosis, and the decision tree generation and VIP values are shown in Figure 2. Figure 7BAs shown in Figure 2, the significant genes are UBE2C, CENPF, TTK, FAM83D, KIF15, CENPU, ASPM, DEPDC1, NUF2, ZFHX4, CENPE, and NCAPG. The Xgboost algorithm selects genes related to prognosis, and the learning curve is shown in Figure 2. Figure 7C As shown, the significant genes are CBX2, TTK, FAM83D, KIF4A, CENPU, DEPDC1, NUF2, ZFHX4, CKS2, and NEK2. Figure 7D ) found that DEPDC1, NUF2, and ZFHX4 were common genes in the three models. Therefore, these three genes were designated as H3K27M-keygenes and associated with poor prognosis.
[0092] Verification of H3K27M-Key Genes: Western blot experiments were performed on DEPDC1, NUF2, and ZFHX4 to verify their expression levels ( Figure 8 The results showed that there was a significant difference between the H3K27M mutant group and the wild-type group. The mRNA expression of H3K27M was further detected by real-time fluorescence quantitative PCR ( Figure 8 The results of the analysis showed that the expression of the three proteins in different groups were significantly different. DEPDC1 was significantly overexpressed in H3K27M mutant cell lines (P < 0.05), NUF2 was also significantly overexpressed in H3K27M mutant cell lines (P < 0.05), while ZFHX4 was significantly underexpressed in H3K27M mutant cell lines (P < 0.01).
[0093] Example 5 Drug screening targeting H3K27M key characteristic genes
[0094] Using GDSC and CTRP databases ( Figure 9A-9B ) predicts potential drugs that can reverse the expression of DEPDC1, NUF2, and ZFHX4. Drugs that reverse the expression of at least two of these genes are identified as potentially effective drugs. By intersecting the results of the two databases, 19 potentially effective drugs were finally obtained and subjected to in vitro multi-cell line high-throughput drug screening ( Figure 9C At a concentration of 1 μM, STF-31 and Niclosamide showed the best effect and had a significant killing effect on H3K27M mutant brainstem glioma cell lines. IC50 determination was further completed ( Figure 9D STF-31 demonstrated significant anti-tumor activity against H3K27M mutations and was selected as a final candidate. This completes the drug screening process for H3K27M-key genes.
[0095] Example 6: Verification of the Effect of the Drug
[0096] The cell line transcriptomes were sequenced before and after administration of the four final-selected drugs, and pathway enrichment analysis was performed on the differentially expressed genes. The mechanism of action of the four final-selected drugs that are sensitive to H3K27M mutations is closely related to the transcriptome background changes caused by H3K27M histone mutations ( Figures 10A-10D ).
[0097] The route of administration was determined based on a comprehensive evaluation of the drug's physicochemical properties, blood-brain barrier permeability, and toxicity. Finally, the microtubule inhibitor Verubulin was selected for intraperitoneal injection, the TTK inhibitor BAY1217389 and the actin inhibitor Filanesib were administered in situ with sustained release, and the GLUT1 inhibitor STF-31 was injected into the lateral ventricle to simulate intrathecal administration for in vivo evaluation experiments ( Figure 11 A). After in situ brainstem tumor formation in mice, 190326-Luc was administered regularly, and the changes in tumor size were observed by in vivo fluorescence imaging ( Figure 11 B- Figure 11 C). The results showed that the tumor fluorescence values of the four drug-administered groups were significantly lower than those of the control group, and some individuals who received the drugs achieved "tumor clearance", indicating that the drugs were effective. Following up on the survival period, the survival time of the four drug-administered groups was significantly longer than that of the control group (all P < 0.01, Figure 11 D).
[0098] In summary, the DMG-P Library for precision drug screening of diffuse midline glioma includes microtubule inhibitors, actin inhibitors, monopolar spindle inhibitors, and GLUT1 inhibitors, specifically: microtubule inhibitor Verubulin (CAS No. 827031-83-4), actin inhibitor Filanesib (CAS No. 885060-09-3), monopolar spindle inhibitor BAY1217389 (CAS No. 1554458-53-5), and GLUT1 inhibitor STF-31 (CAS No. 724741-75-7).
[0099] Individualized drug screening was performed on diffuse midline glioma organoids derived from a patient using the DMG-P Library. Drugs were administered once a week (drug concentration 1 μM) and organoid morphology was observed two weeks after administration. Drugs in the DMG-P Library showed significant therapeutic effects. Organoids disintegrated two weeks after administration, and the effect was better than previously reported drugs (ONC-201) ( Figure 12 ).
[0100] Example 7 H3K27M mutation prognostic prediction model for diffuse midline glioma
[0101] Based on the aforementioned key characteristic genes DEPDC1, NUF2 and ZFHX4, a prognostic prediction model for diffuse midline glioma with H3K27M mutation was constructed. The Norman diagram of the results of multivariate COX regression analysis (Table 2) is shown in Figure 2. Figure 13A As shown in Figure 2, the C index of the model is 0.701, which performs well. Based on the expression levels of the above three genes in tumor tissue, the risk score was calculated, and the patient cohort was divided into high-risk group and low-risk group according to the risk score. The survival curve ( Figure 13B ) showed that the two groups were clearly distinguished. Subsequently, the model was subjected to a calibration curve test ( Figure 13C ), the results show that the model calibration effect is good. Further, the clinical decision curve of the model was drawn ( Figure 13D ), showing that this model can significantly increase the benefit of clinical decision-making compared with single-gene prediction.
[0102] Table 2. Multivariate COX regression model
[0103]
[0104]
Claims
1. A precise drug screening method for diffuse midline glioma, wherein the diffuse midline glioma is an H3K27M mutant, characterized in that: The method comprises: (1) Analyze the differentially expressed genes between H3K27M mutant and wild-type diffuse midline gliomas, identify gene modules associated with H3K27M mutations, analyze the intersection of differentially expressed genes and gene modules, obtain robust differentially expressed genes, and analyze their enriched pathways; (2) Drug screening targeting H3K27M key differential pathways: Drug screening is performed among drugs that can intervene in the enriched pathways obtained in step (1); (3) Drug screening targeting H3K27M robust differentially expressed genes: Analyze the genes associated with prognosis among the robust differentially expressed genes obtained in step (1), and screen for drugs that can intervene in these prognosis-related robust differentially expressed genes; (4) Drug screening targeting H3K27M key characteristic genes: Analyze genes associated with poor prognosis and screen drugs that affect the expression of these poor prognosis-related genes.
2. The method according to claim 1, wherein in step (1), the Limma package is used to analyze the differentially expressed genes between H3K27M mutant and wild-type diffuse midline gliomas; the WGCNA algorithm is used to identify gene modules related to H3K27M mutations; the intersection of the differentially expressed genes and the gene modules is obtained by a Venn diagram; and the resulting enriched pathways are the microtubule pathway and the actin pathway.
3. The method according to claim 2, wherein in step (2), drug screening is performed among microtubule inhibitors and actin inhibitors.
4. The method according to any one of claims 1 to 3, wherein in step (3), COX regression analysis is used to identify genes related to prognosis among the robust differentially expressed genes obtained in step (1); the prognosis-related robust differentially expressed genes in step (3) are TOP2A, RRM2, and TTK.
5. The method according to any one of claims 1 to 4, wherein in step (4), the Lasso regression algorithm, the random forest algorithm, and the Xgboost algorithm are used to analyze genes associated with poor prognosis, and the intersection of the three algorithms is taken as the genes associated with poor prognosis through the Venn diagram; the genes associated with poor prognosis are DEPDC1, NUF2, and ZFHX4; the GDSC and CTRP databases are used to predict drugs that can reduce the expression of DEPDC1 or NUF2 and drugs that increase the expression of ZFHX4, and the drugs that can reduce the expression of DEPDC1 or NUF2 and drugs that increase the expression of ZFHX4 are screened.
6. The method according to any one of claims 1 to 5, further comprising (5) applying the drug screened in steps (2) to (4) to cells, sequencing the transcriptome of the cell line before and after administration, and performing pathway enrichment analysis on differentially expressed genes; and verifying the effects of the drug screened in steps (2) to (4) in an animal model and diffuse midline glioma organoids.
7. Use of Verubulin, Filanesib, BAY1217389 or STF-31 in the preparation of drugs for treating H3K27M mutant diffuse midline glioma.
8. The use according to claim 7, wherein the drug is in the form of intravenous system administration, in situ administration or intrathecal administration.
9. Use of reagents for detecting the expression levels of DEPDC1, NUF2 and ZFHX4 in the preparation of a prognostic test kit for diffuse midline glioma with H3K27M mutation.
10. The use according to claim 9, wherein the reagent for detecting the expression levels of DEPDC1, NUF2, and ZFHX4 is a qRT-PCR reagent or a microarray chip; when the kit is used, COX regression analysis is performed using the obtained expression levels of DEPDC1, NUF2, and ZFHX4 as variables.
Citation Information
Patent Citations
Helical seamed structural vessel, method and apparatus of forming same
CA1217389A