The origin of NEPC remains a subject of controversy. Some studies suggest that normal NE cells in the prostate may be selectively preserved and undergo malignant proliferation during ADT (9,10). However, the lineage plasticity hypothesis, which is supported by more substantial evidence, posits that prostate adenocarcinoma cells originating from luminal or basal cells may lose their adenocarcinoma characteristics and acquire a NE phenotype during prolonged ADT (9). This hypothesis suggests a similarity between the origins of NE cells and stem cells. Derivatives of therapy-resistant NE tumors consistently exhibit molecular characteristics of the primary adenocarcinoma, including specific somatic mutations (9). This supports the hypothesis that aggressive NE tumors evolve through lineage plasticity rather than originating as a second primary cancer. An open question remains as to whether lineage conversion involves a dedifferentiation phase, in which adenocarcinoma cells initially lose AR-specific gene expression and acquire stem-like properties before differentiating into NE cells, or whether transdifferentiation occurs directly, bypassing this intermediate stem-like stage (11). Based on a comprehensive analysis, the present study proposed that the development of NEPC, as illustrated in Fig. 1, may involve multiple potential origins.

The incidence of NEPC appears to be increasing, potentially linked to the use of more effective AR-targeting therapies. A rapid autopsy study found that 13.3% of patients had metastases with an AR-negative NE phenotype and an additional 23.3% had AR-negative metastases without NE features (6). These rates exceed those reported in autopsies conducted prior to the approval of potent AR signaling inhibitors such as enzalutamide and abiraterone acetate (5.4% AR-negative, 6.3% NE) (6). This suggests that with the increasing use of ADT, the progression of normal prostate adenocarcinomas to NEPC is likely to become more common.

NEPC can be somewhat imprecise. While a number of prostate adenocarcinomas contain occasional benign NE cells within the epithelium, only 5–10% of tumors exhibit distinct, large, multifocal clusters of NE cells, which are classified as adenocarcinomas NED (12). These tumors typically present as small, infiltrating glands with prominent nucleoli, hyperchromasia, absence of intraluminal secretions or pale blue mucin and isolated tumor cells, often associated with eosinophilic granules (13,14). Immunohistochemical staining for at least one NE marker, such as synaptophysin, chromogranin A, neuron-specific enolase (NSE), or CD56, is essential for confirming the diagnosis of NE differentiation (14).

The association between NED and poorer oncologic outcomes in prostate cancer remains controversial. While some studies suggest a link, NED in typical adenocarcinoma has shown only a weak correlation with prognosis, insufficient to be considered clinically significant (11,13). Most research indicates that NED does not affect outcomes, with each study examining NED in prostate cancer through needle biopsies and transurethral resections of the prostate (15–18). Different pathological subtypes yield varying outcomes: NEPC tumors with well-differentiated Paneth cell-like and carcinoid differentiation tend to have an improved prognosis, while poorly differentiated NE tumors, particularly small cell carcinoma, exhibit greater aggressiveness (11).

One study compared 36 cases of pure small cell carcinoma with 51 cases (58.6%) of mixed small cell carcinoma and adenocarcinoma. The findings revealed that pure small cell carcinoma had significantly shorter overall survival compared with mixed histology (8.9 months vs. 26.1 months following NEPC diagnosis; P<0.001) (19). The transition from adenocarcinoma to NEPC is not abrupt but involves an intermediate state. In its final stage, prostate cancer with NE differentiation presents the following four clinical features (6,14,19): i) Highly aggressive with poor prognosis: A systematic review found that the median survival following an NEPC diagnosis is only 7 months (20). ii) Predominantly osteolytic/visceral metastasis: Tumors tend to be large with low PSA levels, with a median PSA level of less than 4 ng/ml. iii) Non-responsive to androgen deprivation therapy. iv) Short-term response to platinum-based chemotherapy. The factors driving well-prognosed adenocarcinoma and adenocarcinoma with NED to transform into the poorly prognosed NEPC during this lineage switch remain an open question.

The role of NED in prostate cancer is closely associated with the development of castration resistance, yet the high malignancy of NEPC is more critically attributed to the molecular events that occur during NED. The extensive activation of EMT and stemness during NED probably contributes to the aggressive nature of NEPC, which is characterized by high metastatic potential, resistance to radiotherapy and chemotherapy and poor prognosis. The activation of EMT and stemness may be an indispensable component of NEPC.

The most lethal malignancies, including lung, gastric, liver, breast and prostate cancers, as well as hematological cancers such as leukemia, contain cells with varying stem cell-like properties (29). As early as 1877, Cohnheim, a student of Virchow, observed this population of cells with embryonic characteristics (30). Today, these cells are referred to as cancer stem cells (CSCs) or tumor-initiating cells (TICs) and are recognized as the driving force behind tumor establishment and progression (31). Although CSCs comprise <1% of the primary tumor (32,33), their effect is profound. Stemness is characterized by the capacity for self-renewal, multilineage differentiation and extended cellular lifespan functions vital for normal physiology and tissue repair (34). As with normal stem cells, cancer stem cells have the ability to divide, grow, undergo clonal expansion and differentiate (35). This ability closely mirrors the uncontrolled proliferation observed in cancer cells, making CSCs genetically akin to cancer cells and more susceptible to mutations that drive oncogenesis. However, unlike normal stem cell niches, which possess tumor-suppressive functions and various signaling pathways to regulate growth, CSC niches provide a supportive microenvironment for cancer cells. This niche includes cancer-associated macrophages and cancer-associated fibroblasts, which promote cancer cell growth and differentiation through the secretion of various molecules, including growth factors, cytokines and vesicles such as exosomes (36,37). CSCs can adapt to stress conditions such as oxidative damage, hypoxia and exposure to exogenous substances by altering their gene expression programs, leading to phenotypes associated with migration, invasion and drug resistance (38,39).

Current evidence suggests that the primary reservoirs of normal stem cells in the prostate are luminal and basal cells (24). Under normal conditions, stem cells typically exhibit a basal cell phenotype (40). However, following ADT, stem-like cells with a luminal phenotype are more likely to serve as the origin of tumor re-initiation in CRPC (41). The cellular origin of NEPC is also luminal epithelial cells (9) and the stemness of these luminal epithelial cells may facilitate NED.

Tumors acquire stemness under the pressure of endocrine therapy, chemotherapy and radiotherapy through a series of related molecular mechanisms. During chemotherapy or radiotherapy, certain tumor cells may enter a state of senescence known as therapy-induced senescence (TIS), with the senescence-associated secretory phenotype (SASP) promoting the emergence of stem cell-like subpopulations. Over time, the antitumor effects of TIS may wane, allowing the cancer to regain stemness, which can contribute to tumor recurrence (39). Research has shown that the tumorigenic potential of cancer cells is linked to the increased expression of CD44 or other CSC-like characteristics, driven by SASP-related cytokines such as IL-8 and IL-6 (42). The acquisition of cancer stemness is driven not only by the evasion of senescence but also by the plasticity of non-CSCs and CSCs. This plasticity is influenced by endocrine therapy, where the loss of AR signaling can induce EMT and NED, processes that further promote stemness. The effects of SASP and plasticity are difficult to distinguish, with key interleukins involved in SASP being IL-1, IL-6 and IL-8, which can influence the plasticity phenotype of CSCs (43). Additionally, the development of CSCs remains a topic of ongoing debate, with various models proposed. Genetic and epigenetic alterations, along with changes in the ‘stem cell niche’, can result in unchecked stem cell proliferation, endowing these stem-like cells with oncogenic and drug-resistant potential (39). CSCs may arise from aberrant differentiation of stem cells, mutations and dedifferentiation of normal cells, or dedifferentiation of tumor cells.

CSCs exhibit resistance to androgen deprivation, pro-oxidants, radiation and chemotherapy (44–46). In other studies, CSCs have demonstrated significant resistance to various treatments, including tyrosine kinase inhibitors, taxanes and topoisomerase inhibitors (47,48). The resistance of CSCs to chemotherapeutic agents arises from a combination of intrinsic and extrinsic factors (49). Intrinsic factors contributing to CSC therapy resistance include efficient slow cell cycle progression, DNA repair mechanisms, tumor cell vasculogenic mimicry, active anti-apoptotic pathways, poor immunogenicity and the maintenance of stemness characteristics. Extrinsic factors involve the high cell density of CSCs, lineage plasticity, the tumor microenvironment (TME) and epigenetic influences (31,50–52). These features are also evident in NEPC. Although NEPC initially shows sensitivity to platinum-based chemotherapy, resistance and disease progression often develop rapidly and the response to other types of drugs is generally poor. Additionally, similar to stem cells, NEPC exhibits alterations in DNA repair mechanisms (53), tumor angiogenesis (54,55), anti-apoptotic pathways (56) and autophagy (57,58).

The relationship between CSCs and tumor metastasis is complex, with evidence suggesting that the role of stemness in metastasis is not to promote the initiation of metastasis, but rather to facilitate post-migration colonization (38). Current research indicates that EMT promotes the emergence of stemness (59), while stemness may, in turn, facilitate MET, aiding in cellular colonization; a coherent process. Mesenchymal cells lack the ability to colonize; however, CSCs, which constitute a small subset of malignant cells within most tumors, are ultimately responsible for relapse and metastasis (38). Epithelial cell adhesion molecule (EpCAM) is one of the earliest identified CSC markers. The expression of EpCAM is associated with a CSC-like phenotype that promotes bone metastasis in mice and is often associated with the expression of other CSC markers, such as CD44 and CD166 (60). Additionally, as an epithelial cell marker, EpCAM is downregulated during EMT and upregulated during MET, with larger metastatic tumors (≥30 cells) and primary tumors showing high levels of EpCAM expression (61). Therefore, the expression of EpCAM in CSCs reflects their potential for adhesion and post-migration colonization, which is associated with MET. In a mouse model of basal-like breast cancer, continuous MET activation induces stem cell properties (62). In pancreatic and prostate cancers, both MET and hepatocyte growth factor are predominantly expressed in stem-like tumor cells (38). Studies have shown that hematopoietic stem cells undergo a phenotypic shift toward a more mesenchymal state when multiple stemness genes such as Sox2, Myc and Klf4 are knocked out (38,46). By using magnetic-activated cell sorting, CSC populations have been isolated from mesenchymal cancer cells (MCCs). Compared with MCCs, CSCs demonstrate slower proliferation, greater resistance to apoptosis, drug resistance and heightened clonogenic potential, although their invasiveness is reduced (38).

CSCs and disseminated tumor cells (DTCs) share several biological characteristics, such as dormancy (63). Research indicates that DTCs expressing a CSC phenotype constitute only a small fraction of the total DTC population (63). However, these CSC-like DTCs are more effective at displacing hematopoietic stem cells (HSCs) from their interaction with osteoblasts a mechanism HSCs use to localize to the bone niche (64). By displacing HSCs, CSCs bind to components of the HSC niche, enriching the dormant population of prostate cancer cells and enhancing their ability to resist various chemotherapeutic attacks (65,66). Dormant-enriched human PCa cell populations bind to osteoblasts within the bone marrow HSC niche, inducing the expression of TANK-binding kinase 1 (67). This, in turn, inhibits mTOR signaling, promoting dormancy, maintaining the CSC phenotype (CD44+/CD133+) and contributing to drug resistance (61). Increasing evidence suggests that CSCs, though a small population within most tumors, are largely responsible for relapse and metastasis (38,68). Therefore, the activation of stemness in NEPC may promote the survival and colonization of tumor cells after metastasis, leading to the development of visceral and osteolytic metastases (14).

Induced by ADT and inversely correlated with AR signaling, features such as EMT, CSCs and NED are commonly observed (24,71,72). These features emerge under the stress of ADT treatment, with tumor cells undergoing transdifferentiation to evade drug-related pressures (73). For example, LuCaP 35 ×enografts derived from human PCa show concurrent EMT and increased expression of stem cell markers such as WNT 5a and WNT 5b upon ADT treatment (73). Androgen deprivation can also induce the upregulation of EMT-associated molecules, including vimentin (VIM), Zinc finger E-box binding homeobox 1 (ZEB1), N-cadherin (CDH2), Snail Family Zinc Finger 2 (SLUG) and Twist Family BHLH Transcription Factor 1 (TWIST1) (73). Notably, a bidirectional negative feedback loop exists between AR and ZEB1. After ADT treatment, stem-like cells reach their peak levels (74). PSA-/low cell populations exhibit gene expression profiles similar to those of stem cells, characterized by enhanced self-renewal capacity and resistance to ADT and chemotherapeutic agents (75). CSCs isolated from AR-negative DU145 cells show high expression of stem cell markers, including CD24, integrin α2β1, CD44 and the reprogramming factor SRY-box 2 (SOX2). These cells display tumor-initiating potential and self-renewal capabilities (76). These changes may represent adaptive mechanisms under the pressure of ADT treatment, driving the tumor cells toward similar pathways.

ADT suppresses AR signaling, leading to a transformation of epithelial cells into a NE phenotype, which is widely accepted as a key process in prostate cancer progression. The focus here is on the downstream molecular mechanisms through which ADT-induced AR suppression promotes NED. ADT enhances NED by increasing the expression of Forkhead transcription factor (Fox)a2 and knockout of Foxa2 reverses the conversion of the glandular to the NE lineage (77). Notably, ten days after enzalutamide treatment, significant increases in chromatin accessibility were observed compared with 3 days and untreated CRPC, with enrichment in stem cell and neuronal pathways. Knockout experiments further confirmed that ASCL1 is essential for establishing neuronal and stem cell-like lineages (78). ASCL1 and FOXA2 are both considered lineage-determining factors in NED, as they remodel chromatin into a permissive state, facilitating subsequent transcription factor binding and indirectly highlighting the critical role of ADT in NED (79). ADT alleviates AR-mediated suppression of H19, a long non-coding (lnc)RNA involved in cellular plasticity, driving either the NE phenotype (H19 overexpression) or the luminal phenotype (H19 knockdown) (80). Furthermore, ADT reduces EHF transcription by disrupting AR binding to androgen response elements, thus promoting the expression and enzymatic activity of Enhancer of Zeste Homolog 2 (EZH2), which catalyzes histone H3 trimethylation at lysine 27 to repress downstream genes and facilitate NED (81). Enzalutamide promotes lncRNA-p21 transcription through AR inhibition, altering EZH2 function from a histone methyltransferase to a non-histone methyltransferase, leading to STAT3 methylation and promoting NED (82). ADT also activates cAMP response element-binding protein (CREB) and the CREB-EZH2-TSP1 pathway is a newly identified mechanism driving therapy-induced NED in prostate cancer (55). Overexpression of DPYSL5 in prostate cancer cells induces a neuron-like phenotype, enhances invasion and proliferation, upregulates stemness and NE-related markers and negatively correlates with AR and PSA (83). Mechanistically, DPYSL5 promotes prostate cancer cell plasticity via EZH2-mediated polycomb repressive complex 2 (PRC2) activation (83). Depletion of DPYSL5 reduces proliferation, induces G₁ phase cell cycle arrest, reverses the NE phenotype and upregulates luminal genes. Notably, EZH2 emerges as a key downstream molecule in these ADT-induced NED pathways. Enzalutamide also induces NED via the MAOA/mTOR/HIF-1α signaling axis (84). Additionally, AR inhibition depletes REST, leading to NED (85). BRN2 expression, directly suppressed by AR and negatively correlated with AR activity, drives the NE phenotype by binding to SOX2 upon release from suppression (86). AR binds directly to the cis-enhancer region of SOX2 and AR inhibition enhances SOX2 expression (87). Following AR pathway inhibition (ARPI), the expression of PEG10 is derepressed, regulated by both AR and the E2F-RB1 transcription factor pathway during the early and late stages of NEPC development (88). Targeting these downstream pathways associated with AR inhibition may provide opportunities to reverse EMT, stemness and NED in prostate cancer.

Various EMT-related molecules, including ZEB1, PDGF-D, Snail and E-cadherin, play pivotal roles in promoting stemness in cancer cells (24). The relationship between stemness and EMT is complex. For example, silencing E-cadherin in spheroids derived from the PC3 human prostate cancer cell line stimulates the EMT process (89), whereas its expression induces stemness gene expression and spheroid formation in DU145 prostate cancer cells (90,91). Elevated levels of the EMT regulator ZEB1 promote stem cell-like traits in prostate cancer cells, leading to increased SOX2 expression (92). Additionally, CD44-low cells (non-CSCs) can transition to a CD44-high phenotype (CSCs), forming mammospheres in response to TGF-β-driven ZEB1 expression (93). Overexpression of platelet-derived growth factor D in PC3 cells results in morphological changes characteristic of EMT, enhanced clonogenicity and increased spheroid formation, along with higher expression of stemness-related transcription factors, including Sox2, Oct4, Nanog, Lin28B and components of the PRC2 (94). Loss of E-cadherin is crucial in conferring oncogenic properties, particularly in enhancing both stemness and metastatic potential (95). Activation of TGF-β signaling or transcription factors such as Snail can convert non-cancer stem cells (CD133- or ALDH-low) into cancer stem cells/progenitor cells (CD133+ or ALDH-high) (96).

Recent evidence underscores the crucial role of EMT as a driving force in NED. Key EMT inducers involved in NED, such as Twist, Snail, Slug and E-cadherin repressors such as Zeb1/2, are essential to this process (97). For instance, overexpression of Snail leads to the downregulation of E-cadherin and the upregulation of NE differentiation markers, including ENO2 and CHGA (98). The interplay between the TGF-β signaling network and the AR axis markedly influences EMT-related phenotypic changes. Disruption of epithelial homeostasis via EMT facilitates the transformation of epithelial-derived tumors into invasive forms, enabling them to acquire a NE phenotype and rapidly metastasize (99).

Through extensive literature review, the present study identified that stemness-related factors play a crucial role in the process of NED. Particularly, molecules such as SOX2, MYC, EZH2 and ASCL1.

SOX2, a member of the SRY-related HMG box family, is essential for maintaining embryonic stem cell pluripotency and regulating neuronal differentiation (100–102). Compared with patients with PCa or CRPC-adenocarcinoma, NEPC patients exhibit significant upregulation of SOX2 expression (103). In PCa cells, NED genes are upregulated by SOX2 overexpression and downregulated by its silencing (103). These findings suggest that the loss of PTEN, TP53 and RB1 drives the NE phenotype through SOX2 (104). The molecular mechanism by which SOX2 promotes NED likely involves transactivation of Serine Peptidase Inhibitor Kazal Type 1 (SPINK1) and upregulation of H19 and ASCL1 (105). Additionally, SOX2 regulates global epigenetics through LSD1, leading to reduced ARPC-specific gene expression in NEPC, representing a novel mechanism for SOX2 in NEPC progression (106). Several signals upregulate SOX2 in NEPC. AR chromatin immunoprecipitation (ChIP) studies in CRPC-adenocarcinoma cells (CWR-R1) show that AR directly binds to the enhancer region of SOX2 and ligand activation of AR inhibits SOX2 expression, which is relieved by ADT, eliminating AR ligand-mediated suppression of SOX2 expression (107). RB1 can interact with E2F in fibroblasts and be recruited to the promoter region of SOX2, marking it for repression (108). The core pluripotent stem cell gene LIN28 mediates the derepression of the transcription factor HMGA2, which in turn promotes SOX2 expression by inhibiting let-7 miRNA expression (109). Overexpression of Mucin 1-C-terminal subunit (MUC1-C) induces SOX2 expression through the MUC1-C-MYC-BRN2 axis (110). TBX2 binds to the SOX2 promoter and inhibits the expression of miR-200c-3p, leading to increased expression of its targets, SOX2 and N-MYC (111). Furthermore, SNAI2, TRIM59, TRIB2 and SRRM4 all promote NED by upregulating SOX2 expression (92–95). Overall, SOX2 serves as a critical downstream mechanism in multiple molecular pathways leading to NED. Additionally, the pluripotency factor Oct4 is closely associated with AR-independent NE-like tumors (112).

The MYC protein family plays a significant role in cell reprogramming. As a major transcriptional regulator, MYC promotes a plasticity-permissive, stem-like state. Inhibition of CDC7 suppresses NE transdifferentiation by inducing the degradation of MYC protein, enhancing the response to targeted therapy in in vivo models of NE transformation. This process is linked to both stemness and histological transformation (113). Additionally, CDC7 inhibition markedly prolongs the response to standard cytotoxic agents, such as cisplatin and irinotecan, in small cell prostate cancer models (113). The N-Myc oncogene, encoded by the MYCN gene, is a neurodevelopmental-related transcription factor. N-Myc expression is higher in NEPC compared with non-NE CRPC and it plays an important role in NED (114). In the absence of androgens, upregulation of N-Myc is associated with an enrichment of stem cell characteristics and neural lineage differentiation markers. In castrated 22Rv1 ×enograft mice and GEM models, the N-Myc signature was enriched with genes related to neural lineage pathways, such as neural progenitor cell bivalent genes and genes involved in neural development (such as SOX11, SOX21, NTRK1 and NKX2-1), adult stem cells (such as HOXA2/A3/A9/A10 and WNT5A), ESCs (such as SOX2) and NEPC (such as CHGA), while epithelial lineage-associated genes were downregulated (115). Studies on mouse models with PTEN deficiency and MYCN overexpression have shown that N-Myc interacts with EZH2, AR and various AR co-factors (such as FOXA1 and HOXB13) both in vitro and in vivo (114–116). N-Myc directly binds to the promoters of NSE, SYP and AR, suggesting its critical role in regulating the emergence of NEPC (116). Furthermore, N-Myc interacts with AR by directly binding to AR enhancers, driving NE differentiation through EZH2-mediated transcriptional programs (114). Aurora A stabilizes N-Myc (117) and the Aurora A inhibitor alisertib can disrupt this stabilization, preventing NED (118). Another member of the MYC family, c-MYC, collaborates with Pim1 kinase to drive NE differentiation in aggressive prostate tumors in prostate xenografts (119).

ASCL1, an early lineage determinant of NED, is enriched in regions associated with DNA-binding motifs involved in stem cell and neuronal lineage programming (78,120). Genome-wide mapping of ASCL1 occupancy in NEPC cell lines through chromatin immunoprecipitation sequencing revealed that ASCL1 directly regulates key CSC genes, including SOX2, NANOG and OCT4, as well as NE genes such as CHGA, ENO2 (NSE), NCAM1 and DLL1, all known ASCL1 targets (78). Further supporting this, knockdown of ASCL1 resulted in downregulation of these genes, while deletion of ASCL1 triggered a lineage switch from neuronal to luminal (78,120). Additionally, the ASCL1 cistrome is enriched with targets of PRC2 and deletion of ASCL1 reduces EZH2 activity in NEPC cell lines. Conversely, overexpression of ASCL1 enhances EZH2 activity (78).

EZH2, an epigenetic regulator and catalytic subunit of PRC2, catalyzes the trimethylation of histone H3 at lysine 27 (H3K27me3), promoting transcriptional silencing. Its catalytic activity contributes to lineage-specific cell fate determination by either repressing lineage-specific factors or activating transcription factors that drive stem cell and NE programs (121). Compared with adenocarcinoma and a number of progressing NEPC patients, EZH2 is overexpressed in NEPC samples (121,122). It interacts with miR708 to drive the NE phenotype in PCa (123). In an N-Myc-driven NEPC mouse model, EZH2 catalytic activity facilitates the formation of an N-Myc/AR/EZH2-PRC2 complex, downregulating AR signaling and promoting NEPC (114). Furthermore, AR and EZH2 co-occupy the reprogrammed AR cistrome to transcriptionally modulate stem cell and neuronal gene networks (124). EZH2 also plays a crucial role in promoting NE differentiation during ADT, as outlined in previous sections.

The stem cell marker CD44 is selectively associated with NE tumors, with CD44+ PSA-cells capable of initiating tumors, developing castration resistance and expressing NE markers (75). Recently, the FOXC2 has been identified as a unique marker of the stem cell phenotype in NEPC, where AR+ prostate cancer cells acquire FOXC2 expression, correlating with NE transdifferentiation and resistance to enzalutamide and docetaxel (125). Loss of FOXC2 function is sufficient to trigger a phenotypic switch from NE to luminal epithelial characteristics. O-linked N-acetylglucosamine transferase (OGT), associated with stemness and EMT, is crucial for KIF1A-induced NE phenotype and invasive growth (126). MYBL2, a pluripotency gene, serves as a chromatin-binding partner for key pluripotency factors such as OCT4, NANOG and SOX2. Its expression and activity are enriched in human NEPC and NE-like mouse models (127). Cyclin-dependent kinase 2, a transcriptional target of MYBL2, exhibits potent anti-tumor responses when inhibited in RB1-deficient NEPC (128). Furthermore, L1CAM knockdown in PC3 cells impairs their ability to form tumor spheres and reduces the expression of both cancer stemness and NE markers (129).