A molecular subtyping method for meningiomas based on single-cell tumor cell status and its application

Abstract

This application relates to the biomedical field and discloses a molecular subtyping method for meningiomas based on single-cell tumor cell states and its application. The method includes collecting tumor tissue samples; acquiring sample data and preprocessing it; using an unsupervised clustering algorithm to perform dimensionality reduction analysis on the single-cell data to identify different tumor cell states in the meningioma; using a deconvolution algorithm to map the tumor cell states to spatial transcriptome data, calculating the compositional proportions of different tumor cell states, and analyzing the spatial distribution patterns of different cell states in the tissue; and subtyping meningioma patients based on the compositional characteristics and functional pathway activities of tumor cell states in each sample. This invention, by comprehensively integrating multi-omics data, has significant technical advantages in tumor cell state identification, subtyping framework construction, prognostic prediction, risk stratification, and treatment decision support, providing an innovative technical solution for the precision treatment and clinical application of meningiomas.

Description

Technical Field

[0001] This application relates to the field of biomedicine, and more specifically, it relates to a molecular subtyping method for meningiomas based on the state of single-cell tumor cells and its application. Background Technology

[0002] Meningiomas are the most common primary intracranial tumors, generally classified into benign and malignant types. Although histological classification of meningiomas is widely used clinically, this method still has certain limitations in clinical practice. Existing grading systems, especially the World Health Organization (WHO) meningioma grading system, primarily rely on the morphological characteristics of the tumor, such as mitotic figures, brain tissue invasion, and specific morphological subtypes. However, while these classification systems provide preliminary diagnostic criteria, their role in prognostic prediction and individualized treatment decisions remains insufficient. Particularly, the biological behavior of tumors varies considerably among patients with meningiomas of the same grade, indicating that relying solely on histological characteristics is insufficient for adequate risk stratification and prognostic prediction.

[0003] Currently, the classification methods for meningiomas mainly focus on classification based on tumor cell morphology and the detection of some biomarkers. With the development of molecular biology techniques, researchers have attempted to assist in classification by detecting the expression of specific genes or proteins. However, these methods are often limited by the quality and quantity of samples and lack technologies that can comprehensively reflect the tumor microenvironment and cellular heterogeneity.

[0004] In recent years, single-cell transcriptomics technology has provided a new perspective for tumor research. By analyzing gene expression profiles at the single-cell level, researchers can gain a deeper understanding of tumor cell heterogeneity, identify different tumor cell subtypes, and reveal the interactions between the tumor microenvironment and tumor cells. This technology offers a new breakthrough for precision medicine. However, most current research still focuses on tumor cell populations, lacking comprehensive typing methods that combine single-cell data with Bulk RNA-seq data, and these methods have not yet been fully validated in clinical applications.

[0005] Although numerous studies have attempted to utilize single-cell technology for meningioma subtyping and prognostic prediction, these methods have yet to develop a mature subtyping framework, and most have failed to effectively translate single-cell analysis results into clinically applicable prognostic assessment tools. Therefore, how to construct an operational molecular subtyping method for meningiomas by integrating single-cell transcriptome data, bulk RNA-seq data from clinical samples, and other omics data, and apply it to clinical prognostic assessment and risk stratification, remains a pressing issue. Summary of the Invention

[0006] This invention aims to address the shortcomings of existing technologies that cannot effectively combine single-cell transcriptome data with bulk RNA-seq data, and provides a molecular subtyping method for meningiomas based on tumor cell status. This method not only enables in-depth analysis of tumor cell heterogeneity in meningiomas, but also provides reliable prognostic assessment and risk stratification criteria for clinical practice.

[0007] To achieve the above-mentioned objectives, this application provides a molecular subtyping method for meningiomas based on the state of single-cell tumor cells and its application, employing the following technical solution:

[0008] In a first aspect, this application provides a molecular subtyping method for meningiomas based on the state of single-cell tumor cells, comprising the following steps: Step 1: Collect tumor tissue samples from meningioma patients and collect corresponding clinicopathological data. After processing, the samples are used for single-cell transcriptome sequencing, spatial transcriptome analysis and bulk RNA-seq analysis, respectively. Step 2: Obtain single-cell transcriptome sequencing data and bulk RNA-seq data from the samples, and perform preprocessing. Step 3: Use an unsupervised clustering algorithm to perform dimensionality reduction analysis on the single-cell data to identify different tumor cell states in meningiomas, including immune, stromal, metabolic, and proliferative tumor cell states. Step 4: Use a deconvolution algorithm to map the tumor cell states identified in Step 3 to spatial transcriptome data, calculate the composition ratio of different tumor cell states, and analyze the spatial distribution pattern of different cell states in the tissue. Step 5: Based on the tumor cell state composition characteristics and functional pathway activity in each sample, an unsupervised clustering algorithm is used to classify meningioma patients and identify three transcriptomic subtypes: classical, metabolic, and dedifferentiated.

[0009] Furthermore, in step 2, the preprocessing of single-cell transcriptome sequencing data includes standardization and noise reduction; the preprocessing of Bulk RNA-seq data includes standardization, and the transcriptional features of tumor cells are mapped to the spatial transcriptome and Bulk RNA-seq samples using a deconvolution algorithm.

[0010] Furthermore, step 2 also includes a step of using a data integration algorithm to jointly analyze single-cell transcriptome, spatial transcriptome, and bulk RNA-seq data to eliminate batch effects and preserve biological differences.

[0011] Furthermore, the deconvolution algorithm mentioned in step 4 is one or more of the BayesPrism algorithm, CIBERSORTx algorithm, or RCTD algorithm.

[0012] Furthermore, the clustering algorithm described in step 5 is one or more of consensus clustering, K-means, or hierarchical clustering.

[0013] Secondly, this application provides a molecular subtyping system for meningiomas, comprising: The data input module is used to receive gene expression data from the sample to be tested; The computational analysis module is used to execute the aforementioned molecular subtyping method for meningiomas; The results output module is used to output the molecular typing results of the calculation and analysis module and the corresponding prognostic risk assessment information.

[0014] Thirdly, this application provides the application of the described molecular subtyping method for meningiomas in the preparation of diagnostic products for prognostic assessment and / or risk stratification of meningiomas.

[0015] Fourthly, this application provides the application of the aforementioned molecular subtyping method for meningiomas in screening therapeutic drugs.

[0016] Fifthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned molecular subtyping method for meningiomas.

[0017] In a sixth aspect, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned meningioma molecular typing method.

[0018] The molecular typing method for meningiomas based on single-cell transcriptome analysis, spatial transcriptome data integration, and Bulk RNA-seq deconvolution provided by this invention has significant advantages over existing technologies in the following aspects: 1. Precise identification and classification of tumor cell status: Existing methods for classifying meningiomas largely rely on histological features or single molecular markers, which cannot comprehensively reflect the cellular heterogeneity and microenvironment characteristics of meningiomas. This invention, however, through single-cell transcriptome analysis combined with differential expression analysis and functional enrichment analysis, successfully identified four main functional states of tumor cells (immunogenic, stromal, metabolic, and proliferative), providing a new perspective for accurate meningioma classification. These cellular state characteristics not only reflect the biological behavior of tumor cells but also reveal the interaction between the tumor microenvironment and tumor cells, enhancing the biological basis and clinical significance of the classification.

[0019] 2. Cross-scale data integration and the construction of fractal models: This invention overcomes the limitations of traditional typing methods on a single data platform by integrating single-cell transcriptome data, spatial transcriptome data, and bulk RNA-seq data. After data integration, more comprehensive tumor biological characteristics can be obtained at both the tumor cell and spatial tissue levels. Using this multi-omics data, this invention constructs a typing framework based on tumor cell state, capable of accurately identifying classic, metabolic, and dedifferentiated meningioma subtypes. This typing framework effectively reflects the functional state and genomic characteristics of different tumor cell states, thus providing a more precise basis for personalized treatment and clinical risk assessment.

[0020] 3. Improved prognostic prediction and risk stratification capabilities The typing method of this invention, through progression-free survival (PFS) analysis, validated that the typing system based on tumor cell status is superior to the traditional WHO classification. Compared with the WHO classification, the typing system of this invention can function as an independent prognostic factor in a multivariate Cox regression model, significantly improving the accuracy of prognostic prediction. Time-dependent ROC curve and decision curve analyses further demonstrate that this typing has a relatively stable advantage in terms of clinical net benefit and predictive efficacy, thus providing more scientific risk stratification and treatment decision support for meningioma patients.

[0021] 4. Scalability and reliability of cross-queue verification: This invention, through cross-cohort validation, ensures the stability and reproducibility of the classification framework across different data sources and patient populations. Even in cohorts with relatively limited sample sizes, the classification structure remains stable, and the prognostic trends are consistent. This advantage indicates that the classification system of this invention has high generalization ability and can be applied to meningioma patient populations from different sources, demonstrating broad clinical application prospects.

[0022] 5. Spatial organizational characteristics analysis of tumor microenvironment and cell state: This invention combines spatial transcriptomics technology to reveal the spatial distribution characteristics of different tumor cell states in tissues. Through spatial deconvolution analysis, this invention can quantitatively estimate the proportion of different cell states in tumor samples and further analyze the spatial relationships between tumor cells and non-tumor cells such as immune cells and stromal cells. This innovative analytical method, which combines cell state with spatial tissue information, can provide a new theoretical basis for the study of the microenvironment of meningiomas and the design of targeted therapy strategies.

[0023] 6. Supporting personalized treatment: The typing method of this invention can effectively predict the biological characteristics and clinical outcomes of different meningioma subtypes, especially in terms of the distribution characteristics of immune, stromal, metabolic, and proliferative tumor cells, providing a more refined risk assessment. This provides strong support for the development of individualized treatment plans for meningiomas, particularly through the interaction between tumor cell status and the microenvironment, which can provide more precise strategies for targeted therapy, immunotherapy, and other treatments.

[0024] In summary, compared with existing technologies, this invention provides a precise and reliable molecular subtyping method for meningiomas by comprehensively integrating multi-omics data. It has significant technical advantages in tumor cell status identification, subtyping framework construction, prognostic prediction, risk stratification, and treatment decision support, and solves many defects in existing technologies. It provides an innovative technical solution for the precision treatment and clinical application of meningiomas. Attached Figure Description

[0025] Figure 1 Study Design and Analysis Flowchart: This method integrates transcriptomic data from multiple cohorts, including single-cell RNA sequencing (n=24, with 6 cases in the training cohort and 18 cases in the independent validation cohort from GEO), spatial transcriptomics (n=5), and bulk RNA sequencing data (n=724). The bulk RNA sequencing data includes one training cohort (n=401) and five independent validation cohorts (sample sizes of 180, 47, 43, 30, and 90, respectively).

[0026] Figure 2 Single-cell transcriptomics reveals the cellular composition and tumor cell status characteristics of meningiomas at different grades. A: Basic clinical and pathological features of 6 meningioma samples. B: UMAP after integrated analysis shows the clustering results of all cells, identifying 10 major cell populations. C–D: Expression and annotation results of classic marker genes for each cell population. E: Proportional distribution of major cell types in meningiomas of different WHO grades. F: UMAP plot after tumor cell re-clustering. G: Changes in the proportion of each tumor cell subtype in different WHO grades. H: Distribution characteristics of tumor cell subtypes in two-dimensional space.

[0027] Figure 3Spatial transcriptomics reveals the spatial organizational characteristics of tumor cell subtypes. A: Thirteen cell types based on single-cell annotation. B: Spatial dimensionality reduction and clustering results of 5 spatial transcriptome samples. C: Multimodal cross-analysis showing the spatial enrichment patterns of different tumor cell subtypes. D: Proportional distribution of each cell type obtained from spatial deconvolution analysis. E: Comparison of spatial proportions of tumor cell subtypes in samples from different WHO grading levels. F–G: Spatial correlation and network analysis results between tumor cell subtypes and immune cells and stromal cells. H: Spatial expression distribution of marker genes for different tumor cell subtypes. I: Spatial correlation between proliferation-related genes and immune and stromal marker genes. J: Spatial niches and their cellular composition characteristics identified based on nonnegative matrix factorization. K: Proportional distribution of different spatial niches in samples from different WHO grading levels.

[0028] Figure 4 Meningioma transcriptomic subtyping based on cell composition inference. A: Cumulative distribution function curves of consensus clustering under different cluster numbers. B: Delta area plot of the area under the cumulative distribution function curves corresponding to different cluster numbers. C: Consensus matrix when k=3, indicating that the three subtypes have good clustering stability. D: Cell composition distribution of 401 meningioma samples in the three subtypes based on BayesPrism deconvolution results. E: Comparison of the proportion of each cell type among the three subtypes. F–H: Volcano plot of differentially expressed genes in categories 1, 2, and 3. I–K: Gene set enrichment analysis results for categories 1, 2, and 3, showing the enrichment characteristics of immune inflammation-related, metabolic-related, and cell cycle and DNA repair-related functional programs, respectively. Figure 5 : Optimization value of single-cell molecular subtyping for prognostic stratification of meningiomas. A: Sankey diagram showing the correspondence between WHO classification, molecular subtyping, and patient prognostic status. B: Kaplan-Meier curves for progression-free survival of patients with different molecular subtypes. C: Kaplan-Meier curves for progression-free survival of patients with different WHO classifications. D: Univariate Cox regression analysis assessing the relationship between various clinicopathological variables and molecular subtyping and recurrence risk. E: Multivariate Cox regression analysis assessing the independent prognostic value of molecular subtyping after adjusting for other clinical variables. F: Calibration curves of the predictive model at 12, 36, and 48 months. G: Decision curve analysis showing the net benefit of the model under different threshold probabilities. H: Comparison of time-dependent AUC of molecular subtyping and other clinical variables. I–K: Comparison of ROC curves of molecular subtyping, WHO classification, and other clinical variables at 1, 3, and 4 years.

[0029] Figure 6Prognostic validation of meningioma transcriptomic subtyping in independent external cohorts. A–E: Kaplan–Meier curve comparison of progression-free survival based on transcriptomic subtyping in the UW(A), WCH(B), BWH(C), HKU(D), and GEO(E) cohorts. F–J: Kaplan–Meier curve comparison of progression-free survival based on WHO classification in the UW(F), WCH(G), BWH(H), HKU(I), and GEO(J) cohorts. K–O: ROC curves of transcriptomic subtyping versus WHO classification predicting progression-free survival at 1, 3, and 5 years in the UW(K), WCH(L), BWH(M), HKU(N), and GEO(O) cohorts. Detailed Implementation

[0030] The technical solutions and effects of this application will be further described in detail below with reference to the embodiments and accompanying drawings. It should be understood that the specific embodiments described herein are merely for explaining the invention and are not intended to limit the invention.

[0031] This application discloses a molecular subtyping method for meningiomas based on the integration of single-cell transcriptome analysis, spatial transcriptome analysis, and BulkRNA-seq data. This method constructs a clinically valuable meningioma subtyping framework by meticulously analyzing the tumor cell state, cellular composition of the tumor microenvironment, and spatial distribution characteristics within the meningioma, enabling accurate subtyping, prognostic assessment, and risk stratification.

[0032] The method includes the following steps: I. Data Acquisition and Sample Processing 1. Sample collection and classification Tumor tissue samples were collected from patients with meningiomas, along with corresponding clinicopathological data, including age, sex, tumor location, extent of surgical resection, pathological grade, recurrence status, follow-up information, and adjuvant therapy information. All patients signed informed consent forms preoperatively, and the sample collection and usage process complied with medical ethics review requirements. Included samples were classified into WHO grade 1, 2, and 3 according to the World Health Organization (WHO) classification of central nervous system tumors, further encompassing different histological types such as epithelial, fibrotic, atypical, and anaplastic tumors. Tumor tissues were aliquoted within 30 minutes of surgical excision. Samples for single-cell RNA sequencing were immediately digested in pre-cooled preservation solution; samples for spatial transcriptome analysis were embedded in OCT and then rapidly frozen in liquid nitrogen; and samples for Bulk RNA-seq analysis were excised, frozen in liquid nitrogen, and stored at -80°C for later use.

[0033] 2. Single-cell RNA sequencing (scRNA-seq) 1) Fresh meningioma tissue was rinsed with sterile PBS to remove surface blood, and then minced into pieces of approximately 1 mm. 3 Tissue blocks were placed in a digestion solution for enzymatic digestion. The digestion solution consisted of collagenase IV 1 mg / mL and neutral protease 1 U / mL. Digestion was carried out at 37°C with shaking for 30 min, with gentle pipetting every 10 min to promote tissue dissociation. After digestion, DMEM medium containing 10% fetal bovine serum was added to terminate the reaction, and the mixture was filtered through a 70 μm cell filter to obtain a single-cell suspension. Red blood cells were then removed by treatment with erythrocyte lysis buffer for 5 min, followed by centrifugation at 300 × g for 5 min and resuspending in PBS containing 0.04% BSA buffer. Cell viability was assessed using trypan blue staining; samples with a cell viability below 80% were not included in the subsequent library preparation process.

[0034] 2) The cell concentration was adjusted to 700–1200 cells / μL, and single-cell capture and gel bead-in-emulsion construction were performed using the 10xGenomics ChromiumController platform. Library preparation was performed using the ChromiumNext GEM Single Cell 3′ Reagent Kits v3.1, following the kit instructions to complete steps including reverse transcription, cDNA amplification, fragmentation, end repair, adapter ligation, and sample indexing. The cDNA amplification cycle number was set to 12 cycles. Library quality control was performed using an Agilent 2100 Bioanalyzer to detect fragment distribution, and the concentration was determined using the Qubit dsDNA HSAssay Kit. The target number of cells to be captured per sample was set to 5000–10000.

[0035] 3) After library construction, paired-end sequencing was performed on the Illumina NovaSeq 6000 platform. The sequencing mode was set to Read1 28 bp, i7 index 8 bp, and Read2 91 bp. The raw BCL files were demultiplexed, sequence aligned, UMI deduplication was performed, and expression matrix was generated using Cell Ranger 7.1.0. The reference genome was GRCh38. CellRangerCount was run with default parameters, outputting a raw gene × cell count matrix for each sample, which was used for subsequent single-cell data preprocessing and cell annotation analysis.

[0036] 3. Spatial transcriptome analysis 1) Representative meningioma tissue samples were selected for spatial transcriptome analysis. Immediately after ex vivo removal, tumor tissue was embedded in OCT embedding medium and preserved using liquid nitrogen flash freezing. The samples were cut into 10 μm thick serial sections using a cryostat and mounted on a 10x Genomics Visium Spatial Gene Expression Slide. After fixation at -20°C, sections were stained with hematoxylin and eosin (HE), and high-resolution tissue images were acquired under a microscope for subsequent spatial localization and pathological region correspondence analysis. Samples meeting the Visium library construction requirements in terms of tissue coverage, RNA integrity, and section structural integrity were used in subsequent experiments.

[0037] 2) Spatial transcriptome library preparation was performed using the 10x Genomics Visium Spatial Gene Expression Reagent Kits. The standard procedure was followed, including tissue permeation, in situ reverse transcription, second-strand synthesis, cDNA amplification, and library construction. The tissue permeation time was determined to be 12 min in preliminary experiments. The cDNA amplification cycle number was set to 14 cycles. After library construction, the distribution of library fragments was detected using an Agilent 2100 Bioanalyzer, and the library concentration was determined using the Qubit dsDNA HSAssay Kit. Qualified libraries were subjected to paired-end sequencing on an Illumina NovaSeq 6000 platform, with the sequencing mode set to Read 1 28 bp, i7 index 10 bp, i5 index 10 bp, and Read 2 90 bp.

[0038] 3) Raw spatial transcriptome sequencing data were aligned, spatial barcode identified, and expression matrix generated using Space Ranger 2.0.0, with GRCh38 as the reference genome. Default parameters were used when running `spaceranger count`, and tissue region identification and spatial site localization were performed in conjunction with HE images. The generated spatial site expression matrix was imported into the R environment and then combined with single-cell transcriptome data for subsequent deconvolution analysis and cell type mapping, thereby resolving the spatial distribution characteristics of different tumor cell states and microenvironment cells in tissue sections.

[0039] 4. Bulk RNA-seq Data Processing 1) Sample Source and RNA Extraction: Bulk RNA-seq samples were obtained from surgically removed tissue of meningiomas, including samples from our center and external public or collaborative cohorts. After grinding the tissue samples under liquid nitrogen, total RNA was extracted using TRIzol Reagent. RNA purity was determined using NanoDrop 2000, and RNA integrity was detected using Agilent 2100 Bioanalyzer. The quality criteria for samples included in library construction were: A260 / A280 ratio between 1.8 and 2.1, RNA integrity RIN ≥ 7.0, and total RNA amount not less than 1 μg.

[0040] 2) Bulk RNA-seq libraries were constructed using a Poly(A) enrichment strategy. The NEBNext UltraII RNA Library Prep Kit for Illumina was used for mRNA enrichment, fragmentation, first- and second-strand cDNA synthesis, end repair, A-tailing, adapter ligation, and PCR amplification. The PCR amplification cycle count was set to 12 cycles. After library construction, fragment lengths were detected using an Agilent 2100 Bioanalyzer, and library concentrations were determined using a Qubit dsDNA HS Assay Kit. Qualified libraries were subjected to paired-end sequencing on an Illumina NovaSeq 6000 platform in PE150 mode.

[0041] 3) Raw Bulk RNA-seq FastQ data were first quality controlled and adapter removed using FastP 0.23.2 with parameters set to -q 20 -u 30 -n 5 -l 50, retaining clean reads that met sequencing quality standards. Clean reads were aligned to the human reference genome GRCh38 using HISAT2 2.2.1 with default parameters. After alignment, transcripts were assembled and gene expression was quantified using StringTie 2.2.1 to generate a gene expression matrix. After importing the expression data into the R environment, normalized expression values ​​were calculated using edgeR 3.40.2 and converted to log2(CPM + 1) form for subsequent state scoring, cluster typing, prognostic modeling, and external cohort validation analysis.

[0042] II. Data Integration and Preprocessing 1. Data Standardization and Integration 1) Standardization and denoising of single-cell transcriptome data. The raw expression matrix was imported into the R 4.2.0 environment, and the Seurat 4.3.0 package was used for quality control, normalization, and subsequent analysis. First, an object was constructed using CreateSeuratObject(min.cells = 3, min.features = 200) to retain the expression matrix of at least 3 cells with at least 200 genes detected in each cell; then, the proportion of mitochondrial genes in each cell was calculated and statistically analyzed using PercentageFeatureSet(pattern = "^MT-"), and cells with fewer than 200 detected genes, more than 7000 detected genes, less than 500 total UMIs, or a mitochondrial gene proportion greater than 15% were removed. After quality control, NormalizeData(normalization.method = "LogNormalize", scale.factor= 10000) was used to normalize the single-cell data, FindVariableFeatures(selection.method ="vst", nfeatures = 2000) was used to screen for 2000 hypervariable genes, and ScaleData(vars.to.regress= c("nCount_RNA", "percent.mt")) was used to scale the expression matrix and regress the technical bias caused by sequencing depth and mitochondrial ratio.

[0043] 2) Bulk RNA-seq data were standardized, and the transcriptional features of tumor cells were mapped to the spatial transcriptome and Bulk RNA-seq samples using a deconvolution algorithm. Raw Bulk RNA-seq FastQ data were aligned to the human reference genome GRCh38 using HISAT 22.2.1 with default parameters. The alignment results were used to assemble transcripts and quantify expression using StringTie 2.2.1, generating a gene-level expression matrix. The expression matrix was used in the R environment with edgeR 3.40.2 to calculate the CPM value, which was then converted to log2(CPM + 1) form for downstream analysis. Subsequently, deconvolution analysis was performed on the Bulk RNA-seq data using the BayesPrism 2.0 package. The tumor cell states and microenvironment cell types defined in the single-cell transcriptome were used as reference expression profiles, with parameters set to n.cores = 8, chain.length = 1000, and burn.in = 500, to estimate the relative proportions of each tumor cell state and microenvironment cell component in each bulk sample.

[0044] 3) Data integration algorithms were used to jointly analyze single-cell transcriptomics, spatial transcriptomics, and bulk RNA-seq data to eliminate batch effects and preserve biological differences. Single-cell multi-sample integration was performed in Seurat 4.3.0. First, standardization and hypervariable gene screening were performed on each sample. Then, anchors were constructed using FindIntegrationAnchors(dims = 1:30, anchor.features = 2000), followed by data integration using IntegrateData(dims = 1:30). For cross-batch correction, comparative analysis was performed using the Harmony 0.1.1 package. After principal component analysis, RunHarmony(group.by.vars = "sample", dims.use = 1:30) was called for batch correction. For data requiring proximity matching integration, the MNN algorithm in the batchelor 1.14.0 package was used, with fastMNN(k = 20, d = 50) called to complete the correction. After integration, the first 30 principal components are used uniformly for downstream clustering and visualization analysis.

[0045] 2. Tumor cell status identification 1) Unsupervised clustering algorithms were used to perform dimensionality reduction analysis on single-cell data to identify different tumor cell states in meningiomas. Principal component analysis (PCA) was performed on the integrated single-cell objects using RunPCA (npcs = 50), and the top 30 principal components were selected based on ElbowPlot results. Subsequently, a neighborhood graph was constructed using FindNeighbors (dims = 1:30, k.param = 20), and Louvain clustering analysis was performed using FindClusters (resolution = 0.5). The clustering results were visualized in two dimensions using RunUMAP (dims = 1:30, n.neighbors = 30, min.dist = 0.3) and RunTSNE (dims = 1:30, perplexity = 30), respectively. The obtained cell clusters were annotated using known marker gene expression, automatic annotation results, and copy number variation characteristics, and four main tumor cell states—immunogenic, stromal, metabolic, and proliferative—were identified in the tumor cell population.

[0046] 2) Characteristic genes and dominant transcription programs for each tumor cell state were determined through differential expression analysis. First, malignant tumor cells identified by copy number variation analysis were extracted. Seurat objects were reconstructed and secondary clustering was performed. Then, the FindAllMarkers() function in Seurat was used to identify characteristic genes of each tumor cell subpopulation, with parameters set as test.use = "wilcox", only.pos = TRUE, min.pct = 0.25, and logfc.threshold = 0.25. The differential expression analysis results were corrected using the Benjamini-Hochberg method with multiple tests, and a p-value < 0.05 was used as the significance threshold. The differentially expressed genes were further input into the clusterProfiler 4.6.0 software package for functional enrichment analysis. GO analysis used enrichGO (pAdjustMethod = "BH", pvalueCutoff = 0.05, qvalueCutoff = 0.05), KEGG analysis used enrichKEGG (pvalueCutoff = 0.05, qvalueCutoff = 0.05), and Hallmark pathway analysis used msigdbr 7.5.1 to obtain the gene set, combined with fgsea 1.24.0 for enrichment scoring, with parameters set to minSize = 10, maxSize = 500, and nperm = 1000. Based on the differential gene profile and functional enrichment results, the dominant transcriptional program for each tumor cell state was determined.

[0047] 3. Deconvolution of spatial transcriptomics and bulk data 1) The RCTD algorithm was used to map single-cell data annotation results to spatial transcriptome data to analyze the spatial distribution patterns of different cell states in tissues. Spatial transcriptome data was generated using the 10x Genomics Visium platform. Raw sequencing data was aligned and spatially counted using Space Ranger 2.0.0, with GRCh38 as the reference genome. After importing the obtained spatial expression matrix into the R environment, deconvolution analysis was performed using the RCTD algorithm in the spacexr 2.0.0 package. First, single-cell reference objects and spatial objects were constructed. Then, create.RCTD() and run.RCTD() were called to complete the mapping analysis, with parameters set to max_cores = 8, doublet_mode = "doublet", CELL_MIN_INSTANCE = 25, gene_cutoff = 0.000125, and fc_cutoff = 0.5. The calculation results yielded the composition ratio of different cell types and different tumor cell states at each spatial location, and based on this, the spatial distribution pattern of immune, stromal, metabolic and proliferative tumor cells in meningioma tissue was analyzed.

[0048] 2) Deconvolution was performed on the Bulk RNA-seq data to estimate the proportion of tumor cell states in each sample. Bulk RNA-seq deconvolution was performed using BayesPrism 2.0, with the malignant tumor cell states and microenvironment cell types defined in single-cell transcriptomics as the reference matrix. The parameters were set to n.cores = 8, chain.length = 1000, and burn.in = 500. This yielded the proportions of the four tumor cell states in each bulk sample.

[0049] III. Subtype Model Construction and Clinical Application 1. Construction of the classification system Based on the compositional characteristics of tumor cells and functional pathway activities in each sample, an unsupervised consensus clustering algorithm was used to classify meningioma patients, identifying three main transcriptomic subtypes: classical, metabolic, and dedifferentiated. Specifically, the ConsensusClusterPlus 1.62.0 software package was used, with the tumor cell state score matrix and functional pathway activity matrix of the Bulk RNA-seq samples as input. The clustering algorithm was set to clusterAlg = "hc", the distance metric to distance = "pearson", the number of resampling attempts to reps = 1000, and each resampling included samples with pItem = 0.8 and features with pFeature = 1. The number of candidate clusters was set to maxK = 8. Based on the cumulative distribution function curve, cluster consistency heatmap, and delta area curve, the optimal number of clusters was determined to be 3, classifying patients into the three main transcriptomic subtypes: classical, metabolic, and dedifferentiated. Further comparisons were made in clinical cohorts of differences in WHO classification, pathological type, recurrence rate, and molecular characteristics among different subtypes to verify the stability and biological rationality of the subtype system.

[0050] 2. Clinical validation and prognostic analysis The application value of different subtyping in prognostic stratification was evaluated using Kaplan-Meier survival analysis and Cox regression models. Survival analysis was performed using the `survival 3.5-0` and `survminer 0.4.9` software packages. Survival subjects were constructed using `Surv()`, Kaplan-Meier curves were plotted using `survfit()`, and log-rank tests were used to compare survival differences between different subtypings. Univariate and multivariate Cox regression analyses were performed using the `coxph()` function, incorporating subtyping results, WHO classification, age, sex, resection extent, and radiotherapy status into the model to evaluate the independent predictive value of the subtyping system of this invention for progression-free survival (PFS) and overall survival (OS). The significance criterion was set at P < 0.05. Simultaneously, time-dependent ROC curves were plotted using the `timeROC 0.4` software package to evaluate the accuracy of the subtyping model in predicting 1-year, 3-year, and 5-year PFS or OS, and compared with the traditional WHO classification model, thus demonstrating the superiority of the subtyping method of this invention in prognostic assessment.

[0051] 3. Multi-center validation and application promotion The subtyping system was replicated and validated in multiple independent external cohorts, including the University of Washington cohort, the Sichuan University cohort, and the University of Hong Kong cohort. Each external cohort employed the same data preprocessing, state scoring, and subtyping procedures as the discovery cohort. First, the Bulk RNA-seq data were standardized. Then, BayesPrism was used to calculate the state scores for four types of tumor cells. Finally, samples were categorized based on the three subtyping characteristics identified in the discovery cohort. The consistency of state composition, functional pathway characteristics, and prognostic outcomes across different centers was then compared, and its robustness was verified using Kaplan-Meier analysis and Cox regression models. The results indicate that the subtyping system is stably reproducible across data from different regions, populations, and sequencing sources, demonstrating good reproducibility, generalizability, and clinical application value.

[0052] To better describe the embodiments of the present invention, several examples are provided below: Example The following examples include transcriptome data from Huashan Hospital affiliated with Fudan University and multiple external cohorts, including single-cell RNA sequencing (n=24, with 6 cases in the training cohort and 18 cases in an independent validation cohort from GEO), spatial transcriptome data (n=5), and bulk RNA sequencing data (n=724). The bulk RNA sequencing data includes one training cohort (n=401) and five independent validation cohorts (sample sizes of 180, 47, 43, 30, and 90, respectively). The overall study design is as follows: Figure 1 As shown.

[0053] Example 1: Construction of a molecular subtyping model for meningiomas This embodiment discloses the steps for constructing a molecular subtyping model of meningioma based on Huashan Hospital Affiliated to Fudan University (training set).

[0054] This embodiment includes samples of meningioma patients admitted to the Department of Neurosurgery, Huashan Hospital Affiliated to Fudan University, including: single-cell transcriptome sequencing samples: 6 cases; spatial transcriptome sequencing samples: 5 cases; bulk RNA-seq samples: 401 cases.

[0055] All patients signed informed consent forms before surgery, and sample collection and use were approved by the Ethics Committee of Huashan Hospital affiliated to Fudan University. All enrolled patients had complete clinicopathological data and follow-up information, including age, gender, tumor location, extent of surgical resection, pathological grade, recurrence status, follow-up information, and adjuvant therapy information.

[0056] The steps of single-cell RNA sequencing, spatial transcriptome analysis, Bulk RNA-seq data processing, data standardization and integration, tumor cell status identification, spatial transcriptome and Bulk data deconvolution, and typing system construction are all performed in accordance with the steps described above.

[0057] The analysis results of this embodiment are as follows: Figure 2-5 As shown. Single-cell transcriptome analysis ( Figure 2 This indicates that malignant meningioma cells exhibit significant functional heterogeneity, which can be further divided into four functional states: immune, stromal, metabolic, and proliferative. Among these, proliferative tumor cells highly express proliferation-related genes such as TOP2A, UBE2C, and PRC1, are significantly enriched in cell cycle, DNA replication, and DNA repair pathways, and show higher proportions of G2 / M phase cells, stemness scores, and chromosome copy number variations, suggesting a stronger potential for malignant proliferation and progression. Pseudo-temporal analysis further reveals that tumor cell states are not isolated from each other, but rather involve a continuous dynamic process of evolution from metabolic to proliferative states and further state transitions. Spatial transcriptome analysis (…) Figure 3 Tumor cells exhibit different spatial distribution patterns within tissues depending on their functional state. Proliferating tumor cells are significantly enriched in high-grade meningiomas and show stronger spatial proximity, co-localization, and interaction with endothelial cells, fibroblasts, and immune cells. Their marker genes MKI67, TOP2A, and UBE2C are spatially co-expressed with immune and matrix-related marker genes, suggesting that proliferating tumor cells are not isolated but embedded in a malignant niche highly coupled with the tumor microenvironment. This niche is closely related to the invasiveness, growth activity, and microenvironmental remodeling of high-grade meningiomas. Consensus cluster analysis ( Figure 4 After projecting single-cell defined cell state features onto 401 training set bulk RNA sequencing samples, three molecular subtypes—classical, metabolic, and dedifferentiated—were stably identified. The classical subtype was predominantly characterized by immune-related and differentiation-related features; the metabolic subtype was enriched with metabolic reprogramming features such as lipid metabolism and bile acid metabolism; and the dedifferentiated subtype exhibited a higher proportion of proliferating tumor cells and richer immune microenvironment components, with significant enrichment of cell cycle, DNA replication, and DNA repair-related pathways, suggesting that this subtype has more pronounced dedifferentiation characteristics, proliferative activity, and genomic instability. Prognostic analysis ( Figure 5The constructed three-molecule subtype model can more effectively distinguish different recurrence risk groups than the traditional WHO classification, with dedifferentiated patients exhibiting the worst progression-free survival outcome. Kaplan-Meier analysis, univariate and multivariate Cox regression analysis all show that this molecular subtyping is an independent predictor of meningioma recurrence and progression. Furthermore, calibration curves, decision curves, and time-dependent ROC analyses further demonstrate that this subtyping model is superior to the traditional WHO classification in terms of prognostic accuracy and net clinical benefit. Therefore, the meningioma molecular subtyping system based on single-cell feature projection established in this invention can be used for molecular stratification, recurrence risk assessment, and prognostic prediction of meningioma patients.

[0058] The above results demonstrate that this embodiment successfully constructed a molecular subtyping model for meningiomas based on tumor cell status, which can significantly improve the accuracy of prognostic assessment.

[0059] Example 2: External Verification This embodiment discloses the steps for constructing a molecular subtyping model of meningioma based on an independent external cohort (external validation set).

[0060] This embodiment includes 180 cases from the University of Washington cohort (UW), 47 cases from the Sichuan University cohort (WCH), 43 cases from the Brigham and Women's Hospital cohort (BWH) affiliated with Harvard Medical School, 30 cases from the University of Hong Kong cohort (HKU), and 90 cases from the GEO public cohort (GSE189672).

[0061] All patients signed informed consent forms before the operation, and the sample collection and use process complied with medical ethics review requirements. All enrolled patients had complete clinical pathology data and follow-up information, including age, gender, tumor location, extent of surgical resection, pathological grade, recurrence status, follow-up information, and adjuvant therapy information.

[0062] The steps of single-cell RNA sequencing, spatial transcriptome analysis, Bulk RNA-seq data processing, data standardization and integration, tumor cell status identification, spatial transcriptome and Bulk data deconvolution, and typing system construction are all performed in accordance with the steps described above.

[0063] The verification results are as follows Figure 6As shown, in all five external cohorts, using the same cell state projection and typing strategy as the training set, classical, metabolic, and dedifferentiated molecular subtypes could be stably identified. The UMAP distribution and Sankey correspondence across different cohorts indicate that this typing system has good transferability and consistency across different data sources, enabling further stratification of patients with the same WHO classification. Survival analysis results show that, after grouping patients according to the molecular typing established in this invention, the overall progression-free survival differences among patients in each external cohort are clearer than those of the WHO classification, with dedifferentiated subtypes generally exhibiting a worse prognostic trend. Simultaneously, time-dependent ROC analysis shows that the predictive efficacy of this typing system at different time points (1 year, 3 years, and 5 years) is generally better than or no less than that of the traditional WHO classification, indicating that the meningioma molecular typing method established in this invention has good cross-cohort stability, generalization ability, and clinical applicability, and can be used for recurrence risk assessment and prognostic stratification of meningioma patients from different centers.

[0064] The results above indicate that applying the subtyping model to patient prognostic analysis can significantly improve the accuracy of prognostic assessment.

[0065] This specific embodiment is merely an explanation of this application and is not intended to limit it. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.

Claims

1. A molecular subtyping method for meningiomas based on the state of single-cell tumor cells, characterized in that, Includes the following steps: Step 1: Collect tumor tissue samples from meningioma patients and collect corresponding clinicopathological data. After processing, the samples are used for single-cell transcriptome sequencing, spatial transcriptome analysis and bulk RNA-seq analysis, respectively. Step 2: Obtain single-cell transcriptome sequencing data and bulk RNA-seq data from the samples, and perform preprocessing. Step 3: Use an unsupervised clustering algorithm to perform dimensionality reduction analysis on the single-cell data to identify different tumor cell states in meningiomas, including immune, stromal, metabolic, and proliferative tumor cell states. Step 4: Use a deconvolution algorithm to map the tumor cell states identified in Step 3 to spatial transcriptome data, calculate the composition ratio of different tumor cell states, and analyze the spatial distribution pattern of different cell states in the tissue. Step 5: Based on the tumor cell state composition characteristics and functional pathway activity in each sample, an unsupervised clustering algorithm is used to classify meningioma patients and identify three transcriptomic subtypes: classical, metabolic, and dedifferentiated.

2. The molecular subtyping method for meningiomas according to claim 1, characterized in that, In step 2, the preprocessing of single-cell transcriptome sequencing data includes standardization and noise reduction; the preprocessing of Bulk RNA-seq data includes standardization, and the transcriptional features of tumor cells are mapped to the spatial transcriptome and Bulk RNA-seq samples using a deconvolution algorithm.

3. The molecular subtyping method for meningiomas according to claim 2, characterized in that, Step 2 also includes using a data integration algorithm to jointly analyze single-cell transcriptome, spatial transcriptome, and bulk RNA-seq data to eliminate batch effects and preserve biological differences.

4. The molecular subtyping method for meningiomas according to claim 1, characterized in that, The deconvolution algorithm mentioned in step 4 is one or more of the BayesPrism algorithm, CIBERSORTx algorithm, or RCTD algorithm.

5. The molecular subtyping method for meningiomas according to claim 1, characterized in that, The clustering algorithm mentioned in step 5 is one or more of consensus clustering, K-means, or hierarchical clustering.

6. A molecular subtyping system for meningiomas, characterized in that, include: The data input module is used to receive gene expression data from the sample to be tested; The computational analysis module is used to execute the molecular typing method for meningiomas as described in any one of claims 1-5; The results output module is used to output the molecular typing results of the calculation and analysis module and the corresponding prognostic risk assessment information.

7. The use of the molecular subtyping method for meningioma according to any one of claims 1-5 in the preparation of diagnostic products for prognostic assessment and / or risk stratification of meningiomas.

8. The application of the molecular subtyping method for meningiomas according to any one of claims 1-5 in screening therapeutic drugs.

9. A computer-readable storage medium storing a computer program, characterized in that, When the program is executed by the processor, it implements the molecular typing method for meningiomas according to any one of claims 1-5.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the molecular subtyping method for meningiomas as described in any one of claims 1-5.

Images