Biomarker for neuroendocrine prostate cancer and application thereof

Abstract

The invention provides a neuroendocrine prostate cancer biomarker and application thereof, and relates to the technical field of prostate cancer diagnostic markers. The biomarkers of the neuroendocrine prostate cancer are ASCL1 and WDFY4; the biomarker can be applied to preparation of a prostatic cancer risk assessment kit and a prostatic cancer risk assessment chip, and through single-cell RNA sequencing analysis, copy number variation assessment by InferCNV, pseudo-time analysis and a BayesPrism method, NEPC prognosis characteristics are further researched, a program for distinguishing is determined, clinical prognosis of a patient is assessed, and the like. The invention overcomes the defects in the prior art, reveals the heterogeneity of prostate cancer cells and the key transcription program thereof, successfully verifies the novel biomarker of NEPC, and provides a solid foundation for early prediction and clinical prognosis management of NEPC.

Description

Technical Field

[0001] The present invention relates to the technical field of prostate cancer diagnostic markers, and in particular to a biomarker for neuroendocrine prostate cancer and an application thereof. Background Art

[0002] Neuroendocrine prostate cancer (NEPC) is a lethal malignancy with limited treatment options. The identification of effective NEPC biomarkers is important for its early detection and intervention, but the current research on such biomarkers is still significantly insufficient.

[0003] Castration-resistant prostate cancer (CRPC) poses a major challenge in the management of advanced prostate cancer (PCa), characterized by continued tumor progression despite androgen deprivation therapy (ADT). Approximately 10-20% of prostate cancer patients progress to CRPC, and the median survival of CRPC patients is 9 to 30 months. This stage of prostate cancer is often accompanied by limited treatment options and poor clinical prognosis3. The NCCN guidelines on CRPC recommend the use of abiraterone, enzalutamide, apalutamide, darolutamide, or docetaxel. Although the existing treatment options have gradually increased and the survival of patients has gradually prolonged, overall, CRPC remains a highly lethal malignant disease. Despite substantial progress in elucidating the molecular mechanisms of prostate cancer, the factors driving CRPC progression and its associated drug resistance are still not fully understood. A comprehensive understanding of the molecular basis and cellular heterogeneity of CRPC is essential for identifying new therapeutic targets and improving patient outcomes.

[0004] The prevalence of neuroendocrine prostate cancer (NEPC) is expected to rise as patients undergo multiple treatments. Neuroendocrine tumor cells can be histologically distinguished from other cell types in the complex prostate cancer microenvironment, but their scarcity, especially in the early stages of the disease, often leads to misdiagnosis of early NEPC. Currently, the diagnosis of NEPC mainly relies on immunohistochemical analysis of several biomarkers, including negative androgen receptor (AR), high MKI67 expression, and positive expression of neuroendocrine markers such as chromogranin A (CHGA), synaptophysin (SYP), enolase 2 (ENO2), and neural cell adhesion molecule 1 (NCAM1). However, the heterogeneity of expression of these proteins in neuroendocrine tumor cells significantly reduces their diagnostic sensitivity.

[0005] The advent of next-generation sequencing technologies has enabled in-depth exploration of the molecular features associated with NEPC. Several important factors contributing to NEPC have been identified, notably FOXA1 and SPOP mutations found in primary prostate cancer, lineage flexibility resulting from RB1 loss and TP53 damage, and activation of Polycomb Repressive Complex 2 (PRC2), including EZH2, in advanced prostate cancer cases. In addition, these studies have identified more than a dozen gene sets associated with NEPC, which together contain thousands of differentially expressed genes (DEGs). However, these genomic assessments have shown significant variability, which may be related to the following factors: a) the studies were mainly conducted on cohorts with a small number of NEPC cases; b) the gene expression profile of castration-resistant prostate cancer adenocarcinoma (CRPC-adenocarcinoma) is highly similar to NEPC; and c) the gene sets mainly rely on transcriptome information obtained from a large number of tumors rather than being derived exclusively from neuroendocrine tumor cells. Therefore, there is an urgent need to establish sensitive and specific NEPC biomarkers to meet the needs of basic research and clinical applications. Summary of the invention

[0006] In view of the deficiencies in the prior art, the present invention provides a biomarker for neuroendocrine prostate cancer and its application, which is of great significance for the early detection and intervention of neuroendocrine prostate cancer.

[0007] To achieve the above objectives, the technical solution of the present invention is implemented through the following technical solutions:

[0008] A biomarker for neuroendocrine prostate cancer, wherein the biomarkers are ASCL1 and WDFY4.

[0009] The biomarkers can be used to prepare a prostate cancer risk assessment kit and a prostate cancer risk assessment chip.

[0010] The biomarkers can be used to establish an early detection and prognostic analysis model for neuroendocrine prostate cancer.

[0011] A method for establishing a neuroendocrine prostate cancer diagnostic model comprises the following steps:

[0012] S1. Obtain single-cell transcriptome data, pre-process the single-cell transcriptome data and analyze them using single-cell RNA sequencing to discover new markers of neuroendocrine prostate cancer, ASCL1 and WDFY4;

[0013] S2. We evaluated copy number variation using InferCNV and characterized transcriptional programs using consensus non-negative matrix factorization, ultimately identifying a program that exhibited a characteristic expression pattern of NEPC, activated DNA replication and cell cycle-related pathways, and expressed classic NEPC markers.

[0014] S3. Using pseudo-time analysis to analyze the differentiation trajectory during prostate cancer progression;

[0015] S4. Combined with the BayesPrism method, the single-cell results were integrated with the Cancer Genome Atlas-PRAD sequencing data to further study the prognostic features;

[0016] S5. Enrichment analysis, BayesPrism deconvolution analysis, and immunohistochemistry were performed to verify the high expression of ASCL1 and WDFY4 in NEPC.

[0017] Preferably, the method of processing the single-cell transcriptome data in step S1 includes the following steps:

[0018] S1-1. The raw data were preprocessed using the Seurat package, including excluding cells with less than 200 or more than 2500 detected transcripts, and cells with a mitochondrial gene percentage of more than 10%;

[0019] S1-2. To address the impact of the cell cycle on single-cell transcriptome data, the CellCycleScoring function in Seurat was used to perform cell cycle scoring;

[0020] S1-3, use the NormalizeData function of the LogNormalize method to perform logarithmic transformation and standardization;

[0021] S1-4, use the FindVariableFeatures function to identify highly variable genes through the variance stabilization transformation method, and retain the top 2000 genes with the highest variability;

[0022] S1-5. Use the ScaleData function to remove batch effects between samples;

[0023] S1-6, use the RunPCA function in Seurat to perform principal component analysis, and retain the first 50 principal components;

[0024] S1-7. Use the FindNeighbors function to identify the K nearest neighbor relationship between cells, and use the Louvain method through the FindClusters function to cluster cells.

[0025] Preferably, the specific method of step S2 comprises the following steps:

[0026] S2-1, after preliminary classification of cell types using the Seurat package, InferCNV was applied to calculate CNVs of all autosomes of each cell type;

[0027] S2-2, for 10×Genomics single-cell data, 0.1 was used as the cutoff value;

[0028] S2-3, using consensus non-negative matrix factorization through the cNMF module of the omicverse package, cNMF decomposes the gene expression matrix X into the gene weight matrix W and the gene program activation matrix H, satisfying X≈W×H;

[0029] S2-4. Through multiple random initializations and runs, consensus heat maps were generated and co-correlation coefficients were calculated. The number of factors k = 5 was selected to maximize the stability of matrix decomposition and the biological explanatory power. The metagenes finally generated showed the typical transcriptional status of malignant epithelial cells.

[0030] Preferably, a higher proportion of patients showing the characteristic expression pattern of NEPC in step S2 is significantly associated with a worse clinical prognosis, a higher Gleason score and a later T stage.

[0031] Preferably, the specific method in step S3 includes the following steps:

[0032] S3-1, preprocess the gene expression matrix to filter low-quality cells and genes, and then perform standardization and variance stabilization;

[0033] S3-2. Select highly variable genes to ensure robust trajectory inference. Then, use Monocle2 to sort cells along the pseudo-timeline according to their transcriptional profiles, and use the DDRTree method for dimensionality reduction and trajectory construction.

[0034] Preferably, the expression of ASCL1 and WDFY4 increases significantly as the NEPC cell fate progresses.

[0035] The present invention provides a biomarker for neuroendocrine prostate cancer and its application, which has the following advantages over the prior art:

[0036] The present invention reveals the heterogeneity of prostate cancer cells and their key transcriptional programs by combining single-cell and large-scale RNA sequencing analysis, and successfully verifies the new biomarkers ASCL1 and WDFY4 of NEPC, providing a solid foundation for the early prediction and clinical prognosis management of NEPC. Five different expression programs of PCa malignant epithelial cells were identified, among which program 3 showed the characteristic expression pattern of NEPC, activated DNA replication and cell cycle-related pathways, and expressed classic NEPC markers. Patients with a higher proportion of program 3 were significantly associated with poorer clinical prognosis, higher Gleason scores, and later T stages. Pseudo-time analysis revealed key trajectory-dependent genes involved in NEPC transformation. The expression of ASCL1 and WDFY4 increased significantly with the progression of NEPC cell fate, and was further verified by IHC analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 Schematic diagram of the characteristics of cell populations in prostate cancer samples in an embodiment of the present invention, wherein A is a unified manifold approximation and projection (UMAP) visualization of 11 prostate cancer (PCA) samples; B is a UMAP visualization that distinguishes three major cell types; C is a UMAP visualization that shows seven different clusters; D is a UMAP visualization that shows the expression pattern of marker genes; E is a dot plot showing the expression of marker genes in three cell types; F is a violin plot showing the average expression values ​​of differentially expressed genes in three cell types; G is a heat map showing the correlation between epithelial, stromal and immune cell populations; H is the proportion of each cell type in 11 prostate cancer samples; I is the Ro / e score showing the tissue preference distribution of three cell types in 11 samples.

[0038] Figure 2Schematic diagram of comprehensive analysis of malignant programs in an embodiment of the present invention; A is the inferred large-scale copy number variations (CNVs), which distinguish cancer cells (blue) from non-cancerous cells (red); the x-axis shows the chromosome region, and the y-axis represents tumor and normal cells; B is a scatter plot of CNV correlation and CNV score, which classifies cells into malignant, non-malignant and other groups; C is a consensus heat map showing the Euclidean distance between transcriptional programs, highlighting different cell clusters based on gene expression profiles; D is the stability and error rate evaluation of the cNMF model under different numbers of components, and the optimal number of components is selected as 6 based on high stability and low error rate; E is a UMAP visualization of six malignant transcriptional programs: F is the proportion of five transcriptional modules in eight prostate cancer samples.

[0039] Figure 3 Schematic diagram of functional enrichment and gene module characteristics in the embodiment of the present invention; A is the expression pattern of representative genes in five transcriptional programs and their related biological functions and pathways based on GO, KEGG and WikiPathways databases; B is the GO enrichment analysis of programs 1-5, showing the biological processes related to each program; C is an enrichment map showing the association of transcriptional programs in castration-resistant prostate cancer (CRPC); D is the expression of neuroendocrine prostate cancer (NEPC) marker genes in five programs.

[0040] Figure 4 Schematic diagram of pseudo-time analysis of differentiation trajectories of five malignant cell types in an embodiment of the present invention; wherein AC is a pseudo-time trajectory analysis using Monocle 2, exploring the trajectory of prostate cancer (PCa) cells based on highly variable genes; each point in the tree structure represents a cell, which is color-coded according to sample A, program B or gene expression level C; D is the top 20 trajectory-dependent genes identified in castration-resistant prostate cancer (CRPC) and primary PCa; E is the top 20 trajectory-dependent genes in five programs; F is the significant differential expression of marker genes between primary PCa and CRPC (****P≤0.0001); G is a cluster analysis showing the differential expression of the top 20 marker genes in five modules; H is an immunohistochemistry (IHC) analysis of neuroendocrine prostate cancer (NEPC) and PCa tissues, highlighting the differential expression of NSE, CHGA, ASCL1 and WDFY4 in NEPC and primary PCa;

[0041] Figure 5 Schematic diagram of the association between program 3 and poor prognosis in an embodiment of the present invention; wherein A and B are the distribution of Gleason score A and T stage B in the TCGA-PRAD cohort; C is the prognostic analysis of five modules in the TCGA-PRAD cohort, highlighting the impact of program 3 and program 5 on patient prognosis; D and E are the proportions of program 3 in different Gleason scores D and T stages E. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention is clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] Example:

[0044] 1. Materials and methods

[0045] 1. Data acquisition

[0046] The single-cell transcriptome data are derived from the PRJNA699369 database, which is related to a published study (Cheng, Q. et al. Pre-existing Castration-resistant Prostate Cancer-like Cells in Primary Prostate Cancer Promote Resistance to Hormonal Therapy).

[0047] A small population of cells that exhibit CRPC characteristics before hormone therapy was identified in primary prostate cancer. These cells are inherently castration-resistant rather than developing as an adaptive response to hormone therapy and are associated with biochemical recurrence and distant metastasis. Data integration between samples was performed using the R package Seurat (version 4.3.0), which was used to construct Seurat objects and annotate the grouping information of each sample. Clinical characteristics and corresponding gene expression data were obtained from The Cancer Genome Atlas (TCGA, https: / / portal.gdc.cancer.gov / ), and TCGA-PRAD data were retrieved using the R package TCGAbiolinks (version 2.22.0). Gene expression levels for all three groups were expressed as transcripts per million (TPM). Messenger RNAs with TPM values ​​less than 1 and expressed in more than 90% of the samples were considered background noise and excluded from further analysis. The primary endpoint was progression-free survival (PFS).

[0048] 2. Single-cell transcriptome data processing

[0049] The raw data were preprocessed using the Seurat package, including excluding cells with less than 200 or more than 2500 detected transcripts and cells with more than 10% mitochondrial gene percentage. To address the impact of the cell cycle on single-cell transcriptome data, the CellCycleScoring function in Seurat was used for cell cycle scoring. Log transformation and normalization were performed using the NormalizeData function of the LogNormalize method. The FindVariableFeatures function was used to identify highly variable genes using the variance stabilizing transformation (vst) method, and the top 2000 genes with the highest variability were retained. The ScaleData function was then used to remove batch effects between samples. After data preprocessing, dimensionality reduction was performed to facilitate further analysis of high-dimensional single-cell transcriptome data. Principal Component Analysis (PCA) was performed using the RunPCA function in Seurat, and the top 50 principal components were retained. Next, the FindNeighbors function was used to identify the K-nearestneighbor (KNN) relationship between cells, and the Louvain method was used to cluster cells using the FindClusters function. Cell clusters were manually annotated based on established marker genes (Cheng, Q. et al. Pre-existing Castration-resistant Prostate Cancer-like Cells in Primary Prostate Cancer PromoteResistance to Hormonal Therapy; Song, H. et al. Single-cell analysis of humanprimary prostate cancer reveals the heterogeneity of tumor-associatedepithelial cell states. Nat Commun 13, 141 (2022); Messex, JK & Liou, GY Impact ofImmune Cells in the Tumor Microenvironment ofProstate CancerMetastasis. Life (Basel) 13 (2023)).

[0050] 3. Identification of tumor cells through InferCNV analysis

[0051] The formation of polyploid or aneuploid cells is a pathological hallmark of malignancy, characterized by copy number variations (CNVs). CNVs in single-cell RNA-seq data were analyzed using InferCNV (https: / / github.com / broadinstitute / inferCNV). InferCNV visualizes CNVs in cells based on single-cell RNA-Seq expression data. Initial CNV estimates were obtained by analyzing genes and their chromosomal locations and averaging their relative expression values. After preliminary classification of cell types using the Seurat package, InferCNV was applied to calculate CNVs for all autosomes of each cell type. For 10×Genomics single-cell data, 0.1 was used as the cutoff value.

[0052] 4. Revealing the diverse gene expression patterns of malignant cells in prostate cancer

[0053] To reveal the transcriptional program in malignant epithelial cells, consensus non-negative matrix factorization (cNMF) was used through the cNMF module of the omicverse package (version 1.6.4). This unsupervised method decomposes gene expression data into meta-genes, showing different transcriptional states. Before analysis, the gene expression data were preprocessed as follows:

[0054] (1) Standardization to eliminate size differences between samples;

[0055] (2) Correction of batch effects to improve cross-sample consistency;

[0056] (3) Screening of highly variable genes, focusing on the top 2,000 genes with significant variability in malignant cells, thereby reducing noise interference.

[0057] When running cNMF, a multiple iteration method was used to determine the optimal number of factors (k). Specifically, cNMF decomposes the gene expression matrix X into a gene weight matrix W and a gene program activation matrix H, satisfying X≈W×H. After multiple random initializations and runs, consensus heat maps were generated and co-correlation coefficients were calculated, and the number of factors k=5 was selected to maximize the stability of matrix decomposition and biological explanatory power. The resulting metagenes showed the typical transcriptional state of malignant epithelial cells. The definition of these metagenes laid the foundation for further functional enrichment analysis, which can deeply explore their potential functional significance in tumor biology.

[0058] 5. Trajectory inference analysis

[0059] To study cell differentiation trajectories, the Monocle2 package (version 2.22.0) was used for pseudotime analysis. Monocle2 reconstructs lineage trajectories based on single-cell gene expression data. The gene expression matrix was first preprocessed to filter low-quality cells and genes, followed by normalization and variance stabilization. Highly variable genes were selected to ensure robust trajectory inference. Then, Monocle2 was used to sort cells along the pseudotime axis according to their transcriptional profiles, and the DDRTree method was used for dimensionality reduction and trajectory construction. This method helps to identify branch points, representing potential cellular decision events during differentiation.

[0060] 6. Enrichment analysis

[0061] To further reveal the different biological functions of these gene patterns, enrichment analysis was performed and annotations were performed based on the GO, KEGG, and WikiPathways databases. In addition, gene set enrichment analysis (GSEA) was performed. All enrichment analyses performed on single-cell RNA data were completed using the SCP pipeline (https: / / github.com / zhanghao-njmu / SCP).

[0062] 7. BayesPrism deconvolution analysis

[0063] In order to deconvolute large-scale RNA sequencing data into different cell types, BayesPrism algorithm analysis was performed. As a model-driven framework, BayesPrism aims to distinguish large-scale gene expression data based on probabilistic models by using reference single-cell RNA sequencing datasets. This method, combined with Bayesian models, can effectively reduce the noise in batch expression data while dealing with the basic uncertainty associated with single-cell reference datasets.

[0064] 8. Immunohistochemical analysis

[0065] Prostate tissues from patients who underwent radical prostatectomy in the Department of Urology, the First Affiliated Hospital of Anhui Medical University were selected for this analysis. Immunohistochemical staining was performed to evaluate NSE (NSE polyclonal antibody, Cat#10149-1-AP, RRID:AB_2099180, Proteintech, USA), CHGA (staining granulin A polyclonal antibody, Cat#23342-1-AP, RRID:AB_2879259, Proteintech, USA), ASCL1 (ASCL1 polyclonal antibody, Cat#23751-1-AP, RRID:AB_2935459, Proteintech, USA) and WDFY4 (WDFY4 polyclonal antibody, Cat#17558-1-AP, RRID:AB_2288447, Proteintech, USA).

[0066] For detailed immunohistochemistry steps, please refer to Yin, Y. et al. N-Myc promotes therapeutic resistance development of neuroendocrine prostate cancer by differentially regulating miR-421 / ATM pathway. Mol Cancer 18, 11 (2019) and Ma, W. et al. Integrated analysis of histone modification features in clear cell renal cancer for risk stratification and therapeutic prediction. Transl Oncol 47, 102042 (2024).

[0067] Tumor samples were collected and stored in 4% formaldehyde solution for 24 hours. Subsequently, these samples were embedded in paraffin and cut into sections of approximately 5 μm thickness. Tumor sections were dewaxed and hydrated, followed by inhibition of endogenous peroxidase activity and antigen retrieval. After that, tumor sections were treated with 5% BSA solution for 30 minutes to reduce nonspecific binding, followed by overnight incubation with primary antibodies. Next, a one-hour secondary antibody incubation was performed, and finally the tumor sections were visualized using a DAB kit.

[0068] 9. Statistical analysis

[0069] Statistical analysis, data processing, and visualization were performed using R software (version 4.2.2) and Python (version 3.9). Differences between groups were assessed using the Kruskal-Wallis test and the Wilcoxon test, and clinical characteristics were compared using the chi-square test. Two-tailed p values ​​were calculated, and significance was set at p < 0.05.

[0070] 2. Results

[0071] 1. Cell population characteristics of prostate cancer samples

[0072] After eliminating the batch effect ( Figure 1 A), seven clusters were identified ( Figure 1 B), which were subsequently annotated into three major cell populations. The expression of classical epithelial, mesenchymal, and immune marker genes was examined.

[0073] Key marker genes, including EPCAM, KRT19, and CLDN4, are enriched in epithelial cells; PECAM1 and VWF are enriched in mesenchymal cells; and CD3, CD8A, and CD4 are prominent in immune cell populations. Dot plots showing marker gene expression in these cell types confirm distinct expression patterns ( Figure 1 DE).

[0074] In addition, differentially expressed genes among the three clusters were identified ( Figure 1 F). Average expression values ​​of the three cell types show higher expression levels of KRT19 and EPCAM in epithelial cells, higher expression of PECAM1 and VWF in mesenchymal cells, and higher expression of CD3D, CD8A, and CD4 in immune cells, supporting the robust annotation.

[0075] There was a moderate positive correlation between epithelial and mesenchymal cells, while immune cells showed a lower correlation with either epithelial or mesenchymal cells ( Figure 1 G). Quantified the proportion of each cell type in prostate cancer samples ( Figure 1 H), and the tissue preference of each cluster in 11 samples was evaluated using the Ro / e score ( Figure 1 I).

[0076] 2. Identification of five distinct expression patterns of malignant epithelial cells in prostate cancer

[0077] CNVs in genomic regions were assessed in malignant and non-malignant cells using InferCNV. Heatmap ( Figure 2 A) shows the CNV pattern, with red indicating amplification and blue indicating deletion. The top panel corresponds to the reference (non-malignant) cells, and the bottom panel represents the observed malignant cells, revealing different chromosomal alterations in multiple genomic regions. Scatter plot of CNV correlation and CNV score ( Figure 2B) Used to classify cells into malignant, non-malignant, and other groups.

[0078] To identify transcriptional programs in malignant epithelial cells, cNMF was used. Consensus heatmap ( Figure 2 C) Visualization of the Euclidean distance between transcriptional programs, revealing different cell clusters based on gene expression profiles. Stability and error rate evaluation of cNMF model with different numbers of components ( Figure 2 D) shows the optimal number of components chosen based on high stability and low error rate. Six malignant programs were initially identified ( Figure 2 E). Due to the limited number of cells in program 6, this program was excluded, and five key transcriptional programs were finally identified. The proportions of these programs varied significantly among the samples. Program 2 represented a cell population specific to primary prostate cancer, while program 3 represented CRPC. Programs 1, 4, and 5 were present in both populations and may represent potential primary CRPC cells ( Figure 2 F).

[0079] 3. Functional enrichment and gene module characteristics

[0080] To examine the expression patterns of representative genes in five transcriptional programs and their associated biological functions and pathways ( Figure 3 A). The results showed that programs 1 and 2 were enriched in pathways related to metal ion stress response, inorganic compound detoxification, mineral absorption, and copper homeostasis. Program 3 was highly associated with DNA template replication, chromosome segregation, and cell cycle pathways. Program 4 showed significant enrichment in epidermal growth factor receptor (EGFR) signaling pathway, ERBB pathway, endocytosis, and cell-matrix adhesion. Program 5 was activated in pathways related to temperature response, protein folding, and unfolded protein response. GSEA enrichment analysis further validated these findings.

[0081] In GO enrichment analysis ( Figure 3 B), programs 1 and 2 were enriched in oxidative stress response, transmembrane metal ion transport, systemic homeostasis regulation, and cellular adaptation to arsenic-containing compounds. Program 3 had significant associations in DNA repair and chromatin organization, highlighting its role in maintaining genome integrity under conditions of cellular stress. Program 4 was enriched in integrin-mediated signaling and actin cytoskeleton regulation, highlighting mechanisms that enhance cell adhesion, migration, and structural dynamics. Program 5 highlighted the importance of heat shock protein interactions and proteasome regulation, which are critical for alleviating proteotoxic stress and maintaining protein homeostasis. Overall, these pathways elucidate the complex biological processes of tumor adaptability and progression, further emphasizing the multifaceted roles of these programs in cancer resistance and evolution.

[0082] Programs 3, 4, and 5 play a key role in the development of CRPC ( Figure 3C). Program 3 drives DNA replication and cell cycle progression, contributing to the rapid proliferation and genomic instability that characterize CRPC. Program 4 supports tumor survival and migration through EGFR signaling and cell adhesion pathways, allowing tumors to adapt to a low androgen environment. Program 5 enhances tumor resistance to therapeutic stress by activating protein folding and stress response mechanisms, allowing CRPC cells to tolerate therapy. Together, these programs promote the progression of CRPC, driving tumor growth, survival, and adaptation to therapeutic stress.

[0083] In addition, reference was made to Wang, Z. et al. Single-cell transcriptional regulation and genetic evolution of neuroendocrine prostate cancer. iScience 25, 104576 (2022). NEPC markers were included and their expression in five programs was compared. Figure 3 As shown in D, program 3 showed higher and more NEPC marker expressions, such as NF1A, CPE, COTL1, and ETS2, indicating that program 3 represents the transcriptome expression program of NEPC.

[0084] 4. Pseudo-temporal analysis reveals differentiation patterns of five malignant cell types

[0085] Trajectory analysis revealed a pattern of differentiation trajectories with distinct transcriptional profiles in prostate cancer, simulating the progression from primary PCa to CRPC and NEPC ( Figure 4 AC). The five cell types identified were divided into two major fates, with program 3 representing a unique differentiation endpoint. This trajectory was enriched in CRPC cells and lacked primary PCa cells, suggesting that these cells may represent a mature CRPC state. Dynamic changes in gene expression along the pseudo-time trajectory highlighted the top 20 trajectory-dependent genes ( Figure 4 D, E), including GXYLT2, EDIL3, MEG3, ASCL1, GRP, and WDFY4. These genes gradually increase as malignant cells transition from the primary state to the NEPC phenotype. These expression trends reflect the molecular changes during the transition from primary disease to castration-resistant disease. Significant differential expression of key genes between primary PCA and CRPC was observed ( Figure 4 F). Genes such as GLYATL2, EDIL3, MEG3, FABP4, ASCL1, GRP, and WDFY4 were significantly upregulated in CRPC compared with primary cancer samples, with statistical significance highlighting their potential role in driving the castration-resistant phenotype.

[0086] In addition, cluster analysis ( Figure 4G) Two major clusters, C1 and C2, were identified, representing distinct transcriptional programs. Cluster C1 was primarily associated with CRPC, showing high expression of genes such as GLYATL2, EDIL3, MEG3, FABP4, and ASCL1. Cluster C2 was associated with primary prostate cancer, expressing genes such as MMP13 and POTEC. This clustering highlights the distinct molecular programs that define primary and castration-resistant prostate cancer, suggesting that genes highly expressed in module 3 may serve as specific markers for CRPC.

[0087] Immunohistochemical analysis was performed to verify the expression of ASCL1 and WDFY4, in combination with the known NEPC markers NSE and CHGA. Tissue samples from two NEPC and two primary PCa cases were analyzed. In NEPC tissues, all four markers (ASCL1, WDFY4, NSE, and CHGA) showed strong expression, whereas their expression was significantly reduced in primary PCa samples ( Figure 4 H). These results suggest that ASCL1 and WDFY4 are specifically upregulated in NEPC compared with primary PCa.

[0088] 5. Type 3 prostate cancer is associated with a poorer prognosis

[0089] Using BayesPrism Deconvolution, we estimated the relative contribution of various cell types in prostate cancer samples by Bayesian deconvolution, providing high-resolution insights into cellular composition and distinguishing tumor cells with distinct transcriptional programs. Figure 5 A and Figure 5 B shows the distribution of Gleason score and T stage in the TCGA-PRAD cohort. Survival analysis showed that a higher proportion of cells in program 3 (p<0.001, HR=2.1, 95%CI: 1.43.16) and module 5 (p<0.001, HR=2.56, 95%CI: 1.73.85) was significantly associated with a worse prognosis ( Figure 5 C) Patients with a higher proportion of procedure 3 had a twofold higher risk of recurrence than those with a lower proportion of procedure 3, whereas patients with a higher proportion of procedure 5 had a 2.56-fold higher risk of recurrence than those with a lower proportion of procedure 5.

[0090] In addition, the proportion of procedure 3 in different clinicopathological stages was evaluated ( Figure 5 D, E). The results showed that the proportion of procedure 3 increased with the increase of Gleason score (all p<0.05), and a similar trend was observed with the increase of T stage (all p<0.05), which confirmed the findings of the previous analysis.

[0091] In summary, the analysis identified distinct cell subpopulations and five key transcriptional programs in prostate cancer, with a particular emphasis on program 3's close association with CRPC. Pseudo-time analysis depicted the dynamics of gene expression from primary prostate cancer to CRPC. High expression levels of program 3 were significantly associated with poor prognosis, and the newly identified biomarkers ASCL1 and WDFY4 were validated in immunohistochemical analysis.

[0092] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A biomarker for neuroendocrine prostate cancer, characterized in that The biomarkers are ASCL1 and WDFY4.

2. Use of the biomarker as claimed in claim 1 in the preparation of a prostate cancer risk assessment kit and a prostate cancer risk assessment chip.

3. A biomarker as claimed in claim 1 is used to establish an early detection and prognostic analysis model for neuroendocrine prostate cancer.

4. A method for establishing a neuroendocrine prostate cancer diagnostic model, characterized in that: The method comprises the following steps: S1. Obtain single-cell transcriptome data, pre-process the single-cell transcriptome data and analyze them using single-cell RNA sequencing to discover new markers of neuroendocrine prostate cancer, ASCL1 and WDFY4; S2. We evaluated copy number variation using InferCNV and characterized transcriptional programs using consensus non-negative matrix factorization, ultimately identifying a program that exhibited a characteristic expression pattern of NEPC, activated DNA replication and cell cycle-related pathways, and expressed classic NEPC markers. S3. Using pseudo-time analysis to analyze the differentiation trajectory during prostate cancer progression; S4. Combined with the BayesPrism method, the single-cell results were integrated with the Cancer Genome Atlas-PRAD sequencing data to further study the prognostic features; S5. Enrichment analysis, BayesPrism deconvolution analysis, and immunohistochemistry were performed to verify the high expression of ASCL1 and WDFY4 in NEPC.

5. The method for establishing a neuroendocrine prostate cancer diagnostic model according to claim 4, characterized in that: The method of processing the single-cell transcriptome data in step S1 includes the following steps: S1-1. The raw data were preprocessed using the Seurat package, including excluding cells with less than 200 or more than 2500 detected transcripts, and cells with a mitochondrial gene percentage of more than 10%; S1-2. To address the impact of the cell cycle on single-cell transcriptome data, the CellCycleScoring function in Seurat was used to perform cell cycle scoring; S1-3, use the NormalizeData function of the LogNormalize method to perform logarithmic transformation and standardization; S1-4, use the FindVariableFeatures function to identify highly variable genes through the variance stabilization transformation method, and retain the top 2000 genes with the highest variability; S1-5. Use the ScaleData function to remove batch effects between samples; S1-6, use the RunPCA function in Seurat to perform principal component analysis, and retain the first 50 principal components; S1-7. Use the FindNeighbors function to identify the K nearest neighbor relationship between cells, and use the Louvain method through the FindClusters function to cluster cells.

6. The method for establishing a neuroendocrine prostate cancer diagnostic model according to claim 4, characterized in that: The specific method of step S2 comprises the following steps: S2-1, after preliminary classification of cell types using the Seurat package, InferCNV was applied to calculate CNVs of all autosomes of each cell type; S2-2, for 10×Genomics single-cell data, 0.1 was used as the cutoff value; S2-3, using consensus non-negative matrix factorization through the cNMF module of the omicverse package, cNMF decomposes the gene expression matrix X into the gene weight matrix W and the gene program activation matrix H, satisfying X≈W×H; S2-4. Through multiple random initializations and runs, consensus heat maps were generated and co-correlation coefficients were calculated. The number of factors k = 5 was selected to maximize the stability of matrix decomposition and the biological explanatory power. The metagenes finally generated showed the typical transcriptional status of malignant epithelial cells.

7. The method for establishing a neuroendocrine prostate cancer diagnostic model according to claim 4, characterized in that: A higher proportion of patients showing the characteristic expression pattern of NEPC in step S2 is significantly associated with a worse clinical prognosis, a higher Gleason score and a later T stage.

8. The method for establishing a neuroendocrine prostate cancer diagnostic model according to claim 4, characterized in that: The specific method in step S3 includes the following steps: S3-1, preprocess the gene expression matrix to filter low-quality cells and genes, and then perform standardization and variance stabilization; S3-2. Select highly variable genes to ensure robust trajectory inference. Then, use Monocle2 to sort cells along the pseudo-timeline according to their transcriptional profiles, and use the DDRTree method for dimensionality reduction and trajectory construction.

9. The method for establishing a neuroendocrine prostate cancer diagnostic model according to claim 4, characterized in that: The expression of ASCL1 and WDFY4 increased significantly as NEPC cell fate progressed.

Images