Cancer cell heterogeneity can result from the differentiation program driven by a subset of unique cancer cells called cancer stem cells (CSCs), which have self-renewal potential and can differentiate into various types of cells in a symmetrical or asymmetrical cell divided manner [30,31,32]. In 2005, with the successful isolation of CD44+/α2β1^hi/CD133+ enriched cells from a human prostate tumor, Collins et al. found that this small population of PCa cells possessed self-renewal ability and were enabled to differentiate into non-clonal tumor cells [33]. Other studies also demonstrated that CD44+ and PSA^-/lo PCa cells shared similar properties with CD44+/α2β1^hi/CD133+ enriched cells [34,35,36]. With decades of research, scientists have gradually recognized the existence and significance of these fractional cells in PCa using various biomarkers (Table 1) [37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61], which have stem cell characteristics and are then referred as PCSCs. Now, it is clear that PCSCs are involved in the initiation, progression and therapeutic resistance of PCa.

A number of literature studies have highlighted cancer stem cells as the origin of tumor initiation, including PCa. In the prostate gland, a small population of prostate stem cells (PSCs) undergo an oncogenic transformation to PCSCs when the external environment is unfavorable (Figure 1). Then, the omnipotent PCSCs give rise to various sub-types of prostate cancer cells, either AR-dependent or -independent, forming PCa heterogeneity. It seems that luminal-like PCa originates from progenitor cells in luminal epithelial cells. However, α2β1^hi/CD133+ prostate stem cells were previously identified to be located in the basal layer [62]. Moreover, Witte et al. impressively found that it was the basal epithelial cells, but not luminal epithelial cells, growing to the prostate gland structure when they were subcutaneously implanted into nonobese diabetic scid gamma (NSG) mice [63]. Another study revealed that only basal epithelial cells with oncogenic induction led to prostatic intraepithelial neoplasia (PIN) and PCa in NSG mice [64]. All these findings together suggest that PCSCs in basal epithelial cells are the origin of PCa.

However, the origin of PCa in mice is different from that in humans. Studies by lineage tracing system elucidated that PSCs can be found in both the luminal and basal layers [65,66]. Further investigation illuminated that phosphatase and tensin homolog (PTEN) loss in either the basal layer or luminal layer led to PCa development. In the view of a CSCs-based hypothesis, PSCs in either the luminal layer or basal layer will transform to PCSCs when suffering from genetic alteration such as PTEN loss [67]. Although there is a little discrepancy between mouse PCa and human PCa regarding their origin (Figure 1), the concept of PCSCs as PCa origin has been widely acknowledged by many scientists: PSCs in the prostate epithelial cells undergo genomic instability to give rise to PCSCs when the environment is unfavorable, finally leading to PCa development.

PCa with 2–3 years ADT treatment will eventually develop to castration-resistant disease, a lethal stage owing to its metastatic potential. By bypassing this systemic androgen suppression, AR signaling hijacks several means in order to efficiently respond to the castrated androgen level. AR amplification, AR mutation and selective up-regulation of AR co-regulators are the more frequently observed biological processes in CRPC as compared with primary tumors [12,68,69,70].

Growing evidence also illustrates that ADT treatment expands the population of PCSCs [29,71,72], which have no response to androgenic sources. Continuous AR signaling inhibition of CRPC is fulfilled by the treatment of FDA approved anti-androgen enzalutamide [23,26], which competes androgen to bind AR and suppresses AR nuclear imports [73]. However, it only benefits CRPC by extending to a five month survival [74]. One popular mechanism responsible for the acquired enzalutamide resistance is the selective expression of truncated AR variants. For example, ARv7, which lacks an androgen responsive domain, is a crucially causal factor controlling enzalutamide resistance [75,76,77]. Another important mechanism responsible for anti-androgen resistance is lineage plasticity [78], which is possibly driven by PCSCs. Accumulative evidence suggests that AR-targeted therapies including ADT and anti-androgen treatment considerably enrich the population of PCSCs by facilitating the transition of AR-dependent PCa cells to AR-independent cells with stem-cell-like states or multi-lineage stages [79,80]. PCSCs then re-differentiate into AR-therapy-resistant sub-clones such as neuroendocrine PCa cells or other lineage sub-clones inherently resistant to enzalutamide treatment. In this regard, PCSCs play critical roles in PCa progression and anti-androgen resistance.