As shown in Figure 2 , the carcinogenesis of CRPC could be divided into two primary aspects, i.e., the AR-related mechanisms, and the non-AR-based mechanisms. Among them, AR-related mechanisms have been widely concerned by researchers and clinical practitioners, including AR overexpression, mutations, and splice variants, abnormal AR transcription and modifications, AR-related alternative pathway activation, and abnormal androgen synthesis (5).

As a famous star in CRPC development, targeting AR signaling axis has already been the first-line approach for CRPC therapy. As illustrated in Figure 3 , most of the studies focus on the upstream regulation of AR pathway during CRPC evolution. For example, USP16, KDM4B, and RNF8 could regulate AR signaling by mediating the expression of c-myc (6–8). Interestingly, Larsson et al. found that FcγRIIIa receptors could interact with AR receptors and affect the progression of CRPC in xenograft mouse models (9). COP1 promoted GATA2 degradation to inhibit AR expression and activation (10), however, Shen et al. found that MAPK4 activated GATA2 to regulate AR transcription in mice (11).

CRPC exhibits substantial heterogeneity in terms of its sensitivity to ADT, tissue histopathological types, and genetic profiles. In patients with metastatic CRPC (mCRPC), the occurrence rate of SPOP mutations is relatively low, however, patients carrying SPOP mutations have a relatively better prognosis and are more sensitive to treatment with novel anti-androgen drugs (12, 13). In addition, Taplin et al. found that the detection rate of ARv7 in mCRPC patients was less than 10% in a randomized trial (14), but ARv7 positive patients had poor response to treatment with novel anti-androgen drugs (15). Such variances directly impacted the responsiveness (or resistance) of patients with the same histopathological type to medications. In clinical practice, corresponding theoretical support for a uniform treatment approach to CRPC is still limited, making it challenging to achieve the desired therapeutic outcomes. Deeper exploration of the heterogeneity of CRPC among patients, identifying relevant molecular targets, understanding how these targets vary among different patient subgroups or racial populations, and how this affects treatment outcomes are significant for the personalized management of CRPC patients. With the rapid advance of sequencing techniques, the next-generation sequencing (NGS) is increasingly being widely utilized in clinical diagnosis and treatment. Therefore, it has become essential to analyze the mechanisms and pathological characteristics of CRPC, to categorize CRPC patients accordingly, and to develop personalized drug dosing plans to achieve optimal treatment outcomes.

The understanding of non-AR-based mechanisms of CRPC opens new avenues for the development of novel therapies against resistance. As shown in Figure 2 , the non-AR-based mechanisms are scattered but could be summarized as the following aspects according to recent literature reports.

Neuroendocrine cell-related mechanisms: Neuroendocrine CRPC could be induced by treatments such as ADT, radiotherapy, and chemotherapy, where neuroendocrine differentiation of PCa cells is the main driving force of disease development. Neuroendocrine CRPC exhibits resistance to hormone therapy with rapid progresses but does not reveal an elevation in PSA levels (16). Previous studies indicated that neuroendocrine cells were negative for PSA, and were more abundant in CRPC tumors (17). Moreover, neuroendocrine cells expressed IL-8, and CXCR2, and IL-8/CXCR2 had a significant role in benign and malignant neuroendocrine cells by interacting with p53 signaling (18). Li et al. demonstrated that CXCR2 expression could alter the phenotype of PCa cells, and the inhibition of CXCR2 expression in neuroendocrine PCa cells had the significance to re-sensitized enzalutamide-resistant PCa to enzalutamide (19).

Prostate stem cell-related mechanisms: It mainly includes the transformation of normal stem cells into malignant cells and the activation of tumor stem cells from differentiated tumor cells in response to external stimuli. Here a small subset of cells expressing CD44+/α2β1/CD133+ and lacking of AR expression are identified as prostate cancer stem cells (PCSC), and they hold the ability of proliferating even in androgen-depleted environments or under ADT (20). The research progress in targeted therapy for PCSC includes approaches targeting the prostate CSC microenvironment, targeted nanoparticles, and CAR-T cells targeting the CSC marker epithelial cell adhesion molecule (EpCAM), and some of them have already been entered into clinical trials (21, 22).

The molecular heterogeneity and variability of cellular populations within tumor microenvironment (TME): Chen et al. performed single-cell sequencing and discovered the activated endothelial cells, KLK3-high T-cell clusters, and KLK3-positive T cells in TME for CRPC progression to elucidate the significant variability presented in PCa and offered insights for pinpointing therapeutic targets and developing robust tumor biomarkers (23). In recent years, cancer immunotherapy has garnered increasing attention in cancer therapeutics. Some small-molecule tyrosine kinase inhibitors, whether used as single agents or in combination with other immunotherapies, may potentially improve clinical outcomes (24).

Deregulations in pathways including PI3K-Akt-mTOR, Wnt, Hippo, Hedgehog, and Notch etc: The PI3K-Akt-mTOR signaling pathway played a crucial role in regulating cell survival, proliferation, differentiation, and angiogenesis. It is recognized as one of the important pathway implicated in driving the progression of CRPC (25). In a randomized study conducted on mCRPC patients who had undergone prior docetaxel chemotherapy, the combination of the Akt inhibitor Ipatasertib with abiraterone was compared to abiraterone alone. It was observed that patients with PTEN loss derived a radiographic progression-free survival (rPFS) benefit from varying doses of Ipatasertib in conjunction with abiraterone (26). Robinson et al. identified abnormalities within the Wnt pathway in 18% of mCRPC patients. These abnormalities encompassed periodic alterations in adenomatous polyposis coli, β-catenin, and R-spondins within the pathway, implying a potential pivotal role of this pathway in CRPC progression (27). Currently, small molecule drugs and biological agents directed at the Wnt pathway remain in early stages of research, thus further exploration of the potential anti-tumor mechanisms induced by Wnt pathway inhibition needs to be conducted.

Mutations in the genetic architecture: In addition to epigenetic changes, molecular mutations in specific genes were found to be associated with the prognosis of patients with CRPC, which could guide clinical treatment for patients. In CRPC, inherited or systemic mutations, particularly alterations in the BRCA1 and BRCA2 genes, were linked to an unfavorable prognosis (28). In patients with metastatic hormone-sensitive PCa and mCRPC, TP53 mutations (32%) and PTEN mutations or copy number variations (20%), along with RB1 copy number variations (6%), were commonly observed (29). These genetic alterations were significantly correlated with increased tumor burden and a less favorable clinical prognosis (30, 31).

Biomarkers are important predictors indicating the state change for CRPC management. Previous studies identified plenty of potential biomarkers that may help the early detection, prognosis, and treatment response prediction of CRPC patients. As listed in Table 3 , HSD3B1 is a biomarker that enhanced dihydrotestosterone synthesis from extra-gonadal precursors and has been shown to predict castration resistance in PCa in two retrospective studies (116, 117). PHF8 could promote CRPC progression through the HIF/PHF8/AR axis (118). The expression of Gal-1 in CRPC cells was significantly higher than that in hormone-sensitive PCa cells (119). MiR-32, FOXC2, and miRNA-221/222 have been found to be potential biomarkers for the progression and malignant invasion of CRPC (120–123). LIF was associated with CRPC neuroendocrine and could be used as a serum biomarker for the diagnosis of advanced PCa (124).

Bone metastasis has a negative effect on patient quality of life and contributes to fatal outcomes. Hence, timely intervention holds immense significance. As shown in Table 3 , studies have shown that STAT3 and MUC1 were closely related to bone metastasis in CRPC patients (125, 126). The expressions of LY6D, RGS2, CPT1B, miR-1290, and miR-375 were related to the prognosis of CRPC patients and could be used as potential indicators for predicting prognosis (127–129). In addition, MED15 and stanniocalcin 2 were found to be overexpressed in CRPC and aggressive PCa, while they were expressed at lower levels in benign prostate tissue (130, 131). Furthermore, SLPI was a potential biomarker in the cell proliferation of CRPC under androgen deprivation conditions and its levels were observed to be increased significantly in mCRPC (132). SOX7 and SOX9 belong to the same SOX gene family, however, during castration resistance, SOX9 was found to be significantly increased, while SOX7 was observed to decrease significantly (133).

The identification of biomarkers holds both theoretical and clinical significance for CRPC risk prediction and personalized therapy. For example, the integration of biomarkers including HSD3B1, PHF8, Gal-1, and the SOX gene family facilitated the construction of computational models for CRPC early diagnosis (116, 119). The utilization of factors including LIF, NR6A1, miR-32, FOXC2, and miRNA-221/222 could improve the stratification of patients for applying personalized clinical therapeutics (122–124). Moreover, STAT3, and MUC1 indicated the possibility of bone metastasis, which would be helpful of monitoring the prognosis of CRPC patients into metastatic status (125, 126).

AR plays a pivotal role in PCa, particularly in cases of clinical CRPC. In PCa cell models, AR overexpression has been frequently observed and established as a primary driving factor for PCa progression (134). Over the past few decades, numerous anti-AR drugs have been developed and approved for use across different stages of PCa. In the 1980s and 1990s, the FDA approved the first-generation AR antagonists, including flutamide, nilutamide, and bicalutamide, which had efficacy in the early stages of the disease but ultimately led to the development of resistance and progression to CRPC. With the in-depth research into the AR, second-generation AR antagonists that target the ligand-binding domain (LBD), such as Enzalutamide, Apalutamide, and Darolutamide, have been developed and applied. These agents possess higher AR binding affinity, allowing for more effective suppression of AR expression, and have led to significant improvements in patient survival rates (135–137). However, with the rapid development of resistance, these drugs only provide short-term effects and may potentially give rise to central nervous system toxicity and cardiovascular toxicity (138, 139). It is acknowledged that PCa demonstrates significant inter- and intra-tumor heterogeneity. Targeting a single molecule (e.g., AR) does not benefit all patients, and does not affect all tumor cells equally. Recent studies indicated that many non-AR-based mechanisms were involved in CRPC development, including neuroendocrine cell-related mechanisms, prostate stem cell-related mechanisms, alternations in TME, and deregulations in non-AR genes and pathways. Understanding the role of non-AR-based mechanisms in the development of castration resistance in PCa is also important for identifying new therapeutic targets or strategies against castration resistance. Moreover, as shown in Figure 5 the integration of both AR and non-AR strategies, e.g., inhibiting of AR and neuroendocrine cell-expressed CXCR2 simultaneously, may achieve a better therapeutic effect on CRPC (19).

Tumor cells demonstrate the ability to evade apoptosis, which is an important cause of drug resistance and recurrence in cancer therapy. In recent years, novel regulated cell death pathways such as ferroptosis and pyroptosis have gained increasing attention as representatives in cancer drug discovery and application (140). In PCa studies, Wang et al. triggered ferroptosis in mice using a stable GPX4 inhibitor in a genetically engineered model, and it inhibited the growth and spread of RB-deficient PCa tumors. This finding offered promising prospects for the treatment of RB1-deficient malignant PCa (141). Wu et al. confirmed that inhibiting CDC20 could promote pyroptosis in PCa cells and boost tumor immunity in a mouse model of PCa (142). Wang et al. synthesized a series of aggregation-induced emission materials to mediate the process of ferroptosis and pyroptosis for enhancing PCa immunotherapy (143, 144). As shown in Figure 5 , the identification and application of novel programmed cell death-based approaches would be an emerging direction for CRPC treatment, especially for patients with failure under traditional CRPC therapeutics.

Identifying molecular targets and understanding how these targets vary among different patient subgroups or racial groups and how this affects treatment outcomes are of clinical interest for CRPC personalized management. It is reported that there was a higher similarity at pathway level than that at single gene level in the expression of genes across different PCa datasets, which could partly explain why the single-gene based approaches cannot benefit all patient cohorts and indicate the significance for the development of network medicine-based strategies to fight against therapeutic heterogeneity in cancers (145). In the era of big data and artificial intelligence (AI), computer-aided modeling has now become an emerging approach for translational cancer researches. Compared with traditional experimental methods, computational algorithms simulate the diversity and dynamicity of disease occurrence and progression under a systems biology framework, which would promote the identification and characterization of key signatures for disease early diagnosis and personalized therapy (146, 147). The development of CRPC is a heterogenous process in which genetic, epigenetic, and environmental factors generate large-scale biological networks and contribute to the complexity in PCa phenotype from androgen dependence to castration resistance, thus it is of great significance to integrate multi-omics molecular data with image and clinical information as prior knowledge for multi-step AI model training (148). Here the AI models could be simply divided as two sub-categories based on the methods of feature selection, i.e., traditional models that manually characterize features for training, and deep learning-based techniques automatically extracting features for optimization. The typical applications of AI models for PCa studies are pathological evaluation and classification of multiple PCa status such as benign and malignant lesion identification, PCa grading and molecular subtyping, prognosis and risk stratification, prediction of time to CRPC (149–151), etc. Although there is a promising perspective of AI modeling in PCa and CRPC, the limitations and challenges are still worthy to be concerned. First, the clinical data have the characteristics of small sample size but high dimension and heterogeneity, hence how to reduce the overfitting results and address difficulties in model generalization are the leading issues to be considered. Second, the quality of datasets will directly affect the accuracy of model output. Currently, there is still a lack of in-depth research on clinical data standardization and privacy protection. Construction of PCa-related ontologies would be a possible and feasible way to provide a systematical framework for decoding the large amounts of PCa data and knowledge, and this will contribute to the development of data sharing and integration for model analyses (152). Finally, the interpretability of AI needs to be improved continually, and clinical urologists and pathologists should strengthen their professional behaviors to avoid the biases of missed diagnosis caused by AI models (153).