Concise Review: Prostate Cancer Stem Cells: Current Understanding

Abstract

Prostate cancer (PCa) is heterogeneous harboring phenotypically diverse cancer cell types. PCa cell heterogeneity is caused by genomic instability that leads to the clonal competition and evolution of the cancer genome, and by epigenetic mechanisms that result in subclonal cellular differentiation. The process of tumor cell differentiation is initiated from a population of prostate cancer stem cells (PCSCs) that possess many phenotypic and functional properties of normal stem cells. Since the initial reports on PCSCs in 2005, there has been much effort to elucidate their biological properties including unique metabolic characteristics. In this review, we discuss the current methods for PCSC enrichment and analysis, the hallmarks of PCSC metabolism and the role of PCSCs in tumor progression.

1. Prostate cancer hierarchy, heterogeneity and plasticity

1.1. An introduction to cancer stem cells

Cellular heterogeneity represents an omnipresent feature of most human tumors. Cancer cell heterogeneity can result, in principle, from both clonal competition and evolution (i.e., selection of the ‘fittest’ clone) driven by genetic instability inherently high in tumor cells as well as intra-clonal (subclonal) cellular maturation (differentiation) program driven by epigenetic mechanisms (1). The latter process of cell diversification is initiated from a subset of unique cancer cells that possess many phenotypic and functional properties of normal stem cells (SCs), hence cancer stem cells (CSCs). Phenotypically, CSCs, like normal SCs, may express certain unique or common markers that can be used for prospective enrichment and purification. Functionally, CSCs may share with normal SCs the two most important properties: self-renewal and differentiation. Using these two functional properties, and, in the most strict sense, a single ‘pluripotent’ CSC should be able to regenerate a tumor that histo-structurally recapitulates the parent (patient) tumor that consists of all different cancer cell types. No such a CSC has ever been reported in any tumor. In reality, CSCs are operationally defined as cell subpopulations that are enriched in tumor-regenerating and (serial) tumor-propagating activities, which measure self-renewal, and CSC-derived tumors should harbor diverse cell types present in the parent tumor (i.e., CSC differentiation) (2). CSCs should be distinguished from cancer cell-of-origin as the latter refers to the initial normal cell or the cell type that became tumorigenically transformed whereas CSCs refer to the cell population that drives clonal tumor evolution (2).

Like most human solid tumors, prostate cancer (PCa) is heterogeneous harboring many cancer cell types. PCa cell heterogeneity tends to become accentuated during disease progression and, in particular, upon treatment. This is best evidenced by PCa cells that express AR (androgen receptor), the master pro-differentiation transcription factor in normal prostatic epithelial cells, and PSA (prostate specific antigen), a downstream target of AR and the best lineage differentiation marker for differentiated human prostate luminal cells. Although most untreated primary tumors contain AR⁺ and PSA⁺ cells as the majority, AR^−/lo and PSA^−/lo PCa cells become gradually enriched in high-grade untreated tumors, and PSA^−/lo PCa cells frequently become the predominant cell population in castration-resistant PCa (CRPC) (3). Ever since the initial reports by three independent groups (4–6), in 2005, of PCa stem cells (PCSCs), there has been now substantial evidence for the presence and functions of PCSCs, which we shall further elaborate below.

In this review, we discuss the current methods for PCSC enrichment and analysis, the hallmarks of PCSC biology and the role of PCSCs in therapy resistance, tumor progression and metastases. Relevant literature published in peer-reviewed journals up to March of 2018, was identified via systematic search of PubMed. The scientific quality of the included literature was independently assessed by all authors. Conference abstracts and other unpublished materials were not included. This review is based on the pre-clinical basic research, clinical studies and recent reviews that provide indispensable information for current understanding of the biology of PCSCs.

1.2. Current methods for PCSC enrichment and analysis

The first proof-of-principle studies for prospective identification of PCSCs exploited different phenotypes to isolate self-renewing and tumorigenic cells from patient-derived prostate tumors and cell lines (4–6). These phenotypes included expression of surface markers CD44⁺/α2β1^hi/CD133⁺, expression of transporter protein ATP-binding cassette sub-family G member 2 (ABCG2) that is involved in cell detoxification and the “side population” phenotype, which is defined by Hoechst dye exclusion by tumor cells (4–6). Since then, a few other markers have been described for human PCSC populations (Table1), including CD44/CD133 (7–9), aldehyde dehydrogenase (ALDH) (10), CD166 (11), CD44⁺/CD24⁻ (12). PCa cells expressing these markers have capacities to self-renew and to generate the heterogeneous tumor cell subpopulations. Using antibodies to the membrane CSC markers and fluorescence activated cell sorting (FACS) or magnetic cell sorting (MACS), marker-positive and marker-negative cells can be isolated from cancer cell lines, xenograft tumors and patient-derived specimens, and CSC properties of these cell populations can be analyzed and compared (Table 1).

Table 1.

Methods for detection and isolation of PCSC populations.

Marker-based methods
Technique	Method	Markers	Ref	Advantages	Limitations
Staining cell surface proteins / fluorescence detection of enzymatic activity	Fluorescence - based cell sorting, IF microscopy, IHC	CD44	(8)	Analysis of the experimental (xenograft animal models) and clinical (patient-derived tissues, TMA) samples	These markers are not specific for PCSCs and express on other cells including normal tissues; PCSCs can be heterogeneous for the marker expression, and marker positive cells can be functionally heterogeneous (e.g. by their self-renewal properties)
		CD133/ CD44	(7–9)
		ALDH/CD44/α2β1	(28)
		CD166	(11)
		Trop2	(213)
Gene reporter systems	Fluorescence - based cell sorting, fluorescence microscopy, in vitro and in vivo cell tracking	PSA	(28)	In vitro and in vivo tracking of CSC populations, in vivo CSC imaging, analysis of CSC differentiation
		26S proteasome activity	(14)
		NANOG	(13, 18)
		CD44	(168)

Marker-free methods
Technique	Method	Ref	Advantages	Limitations
Sphere forming assay	Cell cultivation under non-adherent conditions accompanied by FACS analysis to confirm CSC enrichment in spheres	(214, 215)	Elucidation of molecular and physiological properties of CSCs; CSC characteristics associated with treatment resistance	This methods cannot be used to obtain a highly enriched CSC populations; Sphere forming properties in vitro are not indicative for tumorigenicity in vivo
Therapy-induced CSC enrichment	Cancer cell cultivation in the presence of chemotherapeutic drugs or repetitive cell exposure to ionizing radiation	(21, 216)
Side population	Fluorescence sorting based on the Hoechst 33342 or Rhodamine 123 efflux by CSCs	(6, 23)	Various samples can be analyzed e.g. cell lines, xenograft tumors and tumor specimens	Hoechst 33342 and Rhodamine 123 are cytotoxic; absence of the standard dye concentration; low specificity
CSC isolation based on the cell sizes	Microfluidic techniques, FACS	(27, 217, 218)	Isolation of cell populations with different morphology, size and metabolic activity	Low specificity

In addition to the isolation of PCSC populations based on marker expression, other strategies exploiting PCSC-specific gene reporters are being developed for real-time analysis of PCSC behavior at a single-cell level in vitro and in animal models. These systems are based on the stable expression of a reporter gene (e.g. fluorescent protein) driven by the promoters of PCSC- or lineage-specific genes such as NANOG, PSA, SOX2, and OCT4 (13–18) (Table 1). For example, Qin et al. (3) took advantage of the fact that PSA represents one of the most lineage-specific differentiation markers in human prostate epithelia and addressed the important question of whether the phenotypically differentiated (i.e., PSA^+/hi) and undifferentiated (PSA^−/lo) PCa cell populations might be functionally distinct. By building and employing a series of lentiviral-based lineage-tracing vectors using the PSA promoter/enhancer (i.e., PSAP), they provided concrete evidence that the PSA^−/lo PCa cell population harbors authentic self-renewing cancer cells that in vitro can undergo asymmetric cell division under time-lapse videomicroscopy and in vivo can serially propagate xenograft tumors (3). Of clinical relevance, the PSA^−/lo PCa cells are intrinsically resistant to antiandrogens, chemotherapeutic drugs, pro-oxidants, and radiation, and can readily regenerate and serially propagate castration-resistant xenograft tumors (3). These results evidenced that the PSA^−/lo human PCa cell population harbors true PCSCs.

The methodology exploiting marker-based CSC isolation and enrichment markedly contributed to the progress in CSC research; nevertheless it is necessary to bear in mind that the majority of the above-described markers were also identified in the malignant tumors of other origins, such as breast, colorectal, ovarian and lung carcinomas and glioblastoma (9, 19, 20). It is also known that these markers are not specifically expressed in malignant tissues, but also found on the normal embryonic and adult stem cells. Therefore, further investigation of additional reliable and more CSC-specific markers is needed to improve the marker-based CSC harvesting techniques.

Since some CSC populations are described to be treatment-resistant, there is a strategy to enrich CSCs by the repetitive use of chemotherapeutics or radiation therapy (21). PCa cell cultivation in the presence of cytostatic drugs or upon radiation exposure led to the development of acquired treatment resistance and enrichment of carcinoma cells positive for CSC markers and having high tumorigenic capacities (21). Some of the treatment-resistant CSC populations are characterized by upregulation of ABC family members including ABCG2 transporter, which provides the rationale for using the side population technique for CSC isolation (22). For this, tumor cells from cell cultures or dissociated xenograft or primary human tumors are stained with Hoechst 33342 or Rhodamine 123 dyes. Further cell analysis by flow cytometry enables detection of the cells with increased dye efflux (side population) (6, 23). It is suggested that CSCs with a high expression of ABC transporters actively extrude the dye out of the cells. Unfortunately, this method has a number of limitations. First, Hoechst 33342 and Rhodamine 123 are toxic to cells. Next, an efficiency of CSC isolation based on side population might be hampered by low specificity and inconsistency of the existing staining protocols. Although carcinoma cells with acquired therapy resistance frequently demonstrate upregulation of CSC biomarkers, treatment-resistant cells do not necessarily represent a homogeneous cell population, but rather a mixture of phenotypically different cell subsets with different properties (24). A number of studies indicate that therapy might enrich CSCs by selecting for preexisting CSC populations, and/or by inducing tumor cell dedifferentiation (3, 21, 25).

CSCs can also be identified and isolated using the differences in sizes between non-CSCs and CSCs. It was described that CSCs are markedly smaller than more differentiated cells (26, 27). Recent study of Li et al. showed that a population of small PC3 PCa cells (<10 m) has a tendency of being more tumorigenic than the corresponding larger (≥20 or 30 m) cells (27). Further investigation is warranted to clarify the mechanisms controlling cell size in homeostasis and cancer and whether cell size can be effectively used for CSC identification and isolation.

Growing established cell lines or patient-derived tumor cells under serum-free and sphere-forming conditions was one of the first methods to enrich CSCs in vitro. Most of the cells cultured under these harsh conditions undergo anoikis and die. The survived cell populations are enriched in sphere-forming cells, which can sustain their proliferation independently of the cell adhesion mechanisms. This assay also allows investigation of the cancer cell populations with self-renewal properties if the sphere formation is analyzed in multiple passages, because only self-renewing cells are capable of maintaining their spherogenicity in multiple generations (3, 28). It should be noted that sphere-forming cancer cells, though frequently showing an enrichment of CSC markers, may not necessarily display enhanced tumorigenic abilities in xenograft models (29). For example, Kuch et al. (2012) showed that the capacities of carcinoma cells to form spheres cannot be used as a reliable surrogate assay to detect the tumor-initiating capacities in animals (29).

In that respect, in vivo limiting dilution assay should be performed as a standard technique to analyze the frequency of tumor-initiating cells in cancer cell populations defined by putative CSC markers. For this analysis, the same numbers of marker-positive and marker-negative cancer cells isolated either from cancer cell lines or dissociated xenograft or human tumor specimens are injected into immunodeficient mice by orthotopic or ectopic implantation. The cells are injected at different dilutions: from a few or even a single cell to thousands or millions cells, and tumor growth is monitored by caliper measurements or/and tumor bioluminescence for a few months (3, 6, 30). Analysis of tumor incidence in different doses of injected cells enables an estimate of CSC frequencies in the injected cell populations (2). To validate if putative CSC populations indeed possess the long-term self-renewal capacities in vivo, the CSCs of interest are further purified out from regenerated tumors, and then serially transplanted in mice. Cell populations that can maintain long-term self-renewal properties (CSCs) should sustain long-term tumor regeneration during serial transplantations. In contrast, cell populations with a limited self-renewal will lose their tumorigenicity upon serial xenografting (2, 31). For lineage tracing of CSC cells in vivo, a genetically modified mouse model can be established where CSC populations are labeled by using CSC specific gene reporters and monitored during tumor development in living mice (32).

New approaches that facilitate genome editing such as transcription activator-like effector nucleases (TALENs), zinc finger nucleases (ZFN), and the clustered regularly interspaced palindromic repeats/Cas9 associated (CRISPR/Cas9) technologies are expected to offer new possibilities to investigate endogenous gene functions at the genome-wide scale and to establish a direct link between the genome and cellular properties including multipotency and self-renewal.

2. Hallmarks of PCSC metabolism

2. 1. Metabolism of normal prostate and PCa epithelial cells

In contrast to other epithelial tissues, normal prostate epithelial cells have distinct glucose metabolism profiles attributed to the main function of the prostate gland: to produce and secrete fluid that protects and nourishes sperm. The luminal cell compartment of the prostate gland produces and secretes a large amount of citrate into the prostatic fluid. The high level of citrate is important for the maintenance of sperm viability by supplying energy source and by calcium chelation (33). In most normal tissues, citrate, which is produced in the mitochondria or delivered into the cells via membrane transporters, is utilized as a major substrate for energy production in the tricarboxylic acid cycle (TCA) (34). High production of citrate in the prostate gland is a consequence of the low activity of mitochondrial aconitase (m-ACNT) and subsequent inhibition of citrate oxidation (35). In normal luminal cells, suppression of m-ACNT enzymatic activity results from a high mitochondrial concentration of zinc that is maintained by a high expression level of zinc transporter proteins (35, 36). As a consequence of dampened citrate oxidation, normal prostate epithelial cells have a low level of TCA and glycolytic metabolism. Due to a relatively quiescent state, normal prostate SCs have low levels of TCA, oxidative phosphorylation (OXPHOS), and glycolysis (37) (Figure 1).

Figure 1. Metabolism of normal and tumor prostate.

Normal prostate epithelial cells have a low level of TCA and glycolytic metabolism as a result of the impaired citrate oxidation. In contrast, prostate adenocarcinoma cells increase citrate oxidation and OXPHOS. This oxidative phenotype is controlled by AR signaling and can be affected by hypoxia. Mutations in mtDNA and tumor suppressor genes PTEN and TP53, which play a role in the metabolic shift from OXPHOS to aerobic glycolysis, can promote tumor progression. Metabolically active aggressive primary tumors and castration-resistant metastatic disease exhibit the Warburg effect and have a high level of nutrient consumption. ROS released by PCa induces oxidative stress in neighboring CAFs. In turn, CAFs secrete a high level of lactate which is utilized by PCa cells for ATP production via OXPHOS. The glycolytic features of advanced PCa can be used for clinical imaging by using positron emission tomography (PET) for glucose uptake imaging with FDG. Beside glucose, highly proliferative cancer cells require additional supplies for their biosynthesis that cannot be met by glucose consumption such as glutamine. Metabolic features of PCa can be potentially used for anti-cancer therapy to increase efficiency of conventional treatment such as chemo- or radiotherapy. TCA - tricarboxylic acid cycle; AR - androgen receptor; OXPHOS - oxidative phosphorylation; mtDNA - mitochondrial DNA; FDG - Fluoro-2-deoxyglucose; RWE - reverse Warburg effect; CAFs - cancer associated fibroblasts; ROS - reactive oxygen species; ADT - androgen deprivation therapy; HADs - hypoxia activated drugs; CAFs - cancer associated fibroblasts; WE - Warburg effect; HIF - hypoxia inducible factor signaling.

In contrast to the normal luminal cells, prostate adenocarcinoma cells contain a low level of mitochondrial zinc that abrogates the inhibition of m-ACNT activity and enables utilization of citrate in the metabolic pathways (34–36). Reprogramming of cellular energy metabolism plays an important role in tumor initiation, progression, and therapy resistance. In contrast to normal epithelial cells, PCa cells increase citrate oxidation and OXPHOS. This oxidative phenotype is controlled by AR signaling and can be affected by hypoxia (26, 38). Oncogenic mutations of mitochondrial DNA (mtDNA) and tumor suppressor genes PTEN and TP53 play a role in the metabolic shift from OXPHOS to aerobic glycolysis known as the Warburg effect (39, 40). To meet high energetic demands, fast growing tumor cells that exhibit the Warburg effect need to reprogram their metabolic pathways for a high level of nutrient consumption, and this is especially true in case of metabolically active aggressive primary tumors and castration-resistant metastatic disease. Reactive oxygen species (ROS) released by PCa into microenvironment induces oxidative stress in neighboring cancer-associated fibroblasts (CAFs). In turn, CAFs undergo aerobic glycolysis and secrete a high level of energy-rich fuels such as lactate. Consequently lactate is utilized by PCa cells for ATP production via OXPHOS (the phenomena referred to as the reverse Warburg effect) (41). The glycolytic features of advanced prostate tumors can be used in clinical setting for detection of metastatic disease and assessment of therapeutic response using positron emission tomography (PET) for glucose uptake imaging with 2-[18F]Fluoro-2-deoxyglucose (FDG) (42). In addition to increased glucose uptake, highly proliferative cancer cells require additional supplies for their biosynthesis that cannot be met by glucose consumption (Figure 1). A significant proportion of the biosynthetic needs can be covered by the metabolism of glutamine, which is the most abundant amino acid in human plasma. Similarly to the bulk of tumor cells, PCSCs can exhibit glycolytic or oxidative metabolic phenotype depending on the acquired mutations and activation of specific signaling pathways, which we shall further discuss below.

2.2. MYC - dependent metabolic reprogramming and maintenance of PCSCs

Glutamine is an important donor of nitrogen and carbon for the growth-promoting pathways. Although most normal tissues can synthesize glutamine, it becomes conditionally essential for the fast growing tissues. Tumor cells with enhanced expression of MYC oncogene, such as PCa, are particularly dependent on glutamine metabolism to sustain their viability (43–45). In these cells, deprivation of glutamine results in depression of TCA cycle, decrease in ATP level, and inhibition of cell cycle (43, 46). MYC is broadly acting transcription regulator that activates the embryonic stem cell program in human cancer and induces tumor-initiating cells (47). Recent findings indicate that targeting of MYC inhibits PCSC maintenance and tumorigenicity (48, 49). MYC contributes to metabolic adaptations of tumor cells by regulating glucose transporter GLUT1 and the enzymes involved in the glucose metabolism such as hexokinase 2 (HK2), phosphofructokinase (PFK1), enolase 1 (ENO1), and lactate dehydrogenase A (LDHA) (43).

Together with AR and well-established tumor driver the mammalian target of rapamycin, mTOR, MYC regulates expression of glutamine transporters genes SLC1A4 and SLC1A5, which are frequently overexpressed in PCa and thereby drives increased glutamine uptake (50). Through the suppression of miR-23a/b, MYC also regulates the expression levels of GLS1 glutaminase, which converts glutamine to glutamate at the first step of glutamine utilization (43). Analysis of human prostate tissues has revealed upregulation of MYC mRNA in the majority of primary PCa lesions (51). This overexpression of MYC can be attributed to MYC oncogene amplification that occurs in up to 29% of hormone-refractory recurrent prostate carcinomas (52). In addition, MYC expression can be regulated at the mRNA level as a result of APC hypermethylation and Wnt/β-catenin pathway activation, deletion of FOXP3 gene as well as TMPRSS2-ERG rearrangement that activates ERG-dependent transcription program (53–56). MYC overexpression alone is sufficient to bypass senescence and immortalize prostate cells derived from benign prostate tissue specimens (57). Transgenic expression of human MYC in the mouse prostate (i.e., Hi-Myc mice) efficiently induces the precursor lesion prostatic intraepithelial neoplasia (PIN) that rapidly progress to invasive adenocarcinoma (58). MYC-driven tumors show a MYC gene dosage-dependent phenotype, and share molecular signatures with human PCa such as loss of Nkx3.1 tumor suppressor and upregulation of Pim-1 protein kinase that increases energy metabolism and protein synthesis (58, 59). The two-step transformation of prostate basal or luminal cells from benign prostate tissues, either by MYC overexpression and AKT activation or by overexpression of MYC and knockout of the tumor suppressor phosphatase and tensin homolog (PTEN), can induce development of heterogeneous tumors from both cellular compartments (60–63). Significantly, high expression of MYC and loss of PTEN contribute to androgen-independent growth and are associated with highly invasive and metastatic prostate tumors (62–65). Interestingly, MYC is a context-dependent regulator of the expression of glutamine transporters SLC1A4 and SLC1A5, and does not regulate these transporters in PTEN wild-type tumors (50).

2.3. Androgen signaling and PCa metabolism

Dependence on androgen signaling is a hallmark of initial stages of prostate tumor development. Because of this, androgen deprivation therapy (ADT) is one of the main treatments for PCa besides prostatectomy and radiotherapy. While most of the patients initially respond to ADT, about 90% of them eventually develop androgen-refractory tumors with a fatal prognosis (66). ‘Androgen-independent’ tumor progression depends on a variety of mechanisms including AR gene amplification and gene mutation, ligand-independent activation of AR signaling pathway and alterations in the expression of AR co-regulators leading to an increase in AR-dependent transcription activity (67).

Androgen independent PCa growth and resistance to anti-androgen therapy can be conferred by COOH-terminal truncated AR variants (AR-V) (68–70). These AR variants lack ligand-binding domain but retain transcriptional activity. The most frequently occurring AR splice variants associated with poor prognosis include AR-V1, AR-V7, and AR-V567es (68, 70, 71). Studies by Antonarakis at al. demonstrated association between expression of AR splice variant-7, AR-V7 in circulating tumor cells (CTCs) and resistance of PCa patients to the new generation of antiandrogen drugs such as enzalutamide and abiraterone (72). Since then, a growing body of evidence suggested the potential application of AR-V7 detection in PCa patients’ blood as a potential biomarker for prediction of anti-androgen treatment failure (73). Recent experimental evidence suggested that AR-V7 might contribute to PCa progression through triggering epithelial-mesenchymal transition (EMT) (discussed in 3.1, below), induction of PCSC characteristics and metabolic reprogramming through increasing dependence on glutaminolysis (74, 75).

Like MYC, AR regulates expression of many genes involved in glutamine and glucose metabolism. Androgen signaling stimulates glucose consumption as well as production and export of lactate by upregulating the expression of GLUT1/GLUT3, phosphofructokinase 1 (PFK 1), monocarboxylate transporter 4 (MCT4) and calmodulin-dependent kinase kinase 2 (CAMKK2) genes (76, 77). CAMKK2 is highly upregulated in PCa, and CAMKK2-dependent activation of AMP-regulated kinase (AMPK) results in an increased PFK1 activity and promotes glucose uptake and lactate production (78). Metformin, which is the first-line drug for treating diabetes, enhances AMPK activity as a result of inhibition of the mitochondrial complex I activity (79). Metformin treatment was clinically associated with a significant decrease in early biochemical relapse and cancer-specific mortality in PCa patients (80), and also led to an increase in tumor oxygenation and response to radiotherapy in mouse xenograft models (80). Metformin combined with doxorubicin completely inhibits growth of PC3 PCa cells in xenograft models (81). Metformin treatment of docetaxel-resistant PC3 (PC3-DR) cells that manifest an increased oxidative metabolism, inhibits their proliferation and invasiveness without affecting parental PC3 cells (82). In breast and prostate cancer cells, metformin inhibits entry of glucose carbon into the TCA cycle and promotes the contribution of glutamine carbon to the TCA cycle making the cells more dependent on glutamine anaplerosis (83). However, the molecular effects of metformin on PCSCs still need to be clarified. Studies of the signaling pathways deregulated by metformin-induced metabolic stress in breast CSCs showed that metformin selectively inhibits activation of nuclear factor kappa B (NF-κB) and signal transducer and activator of transcription 3 (STAT3) - dependent transcription and expression of a number of inflammatory genes in CSCs as opposed to non-CSCs isolated from the same tumor (84). Interestingly, metformin suppresses AR-dependent expression of MYC in Hi-Myc tumors and inhibits the development of PIN and PCa lesions (85). Other studies have shown that metformin inhibits androgen dependent and androgen independent PCa by downregulation of AR mRNA levels and by inhibition of autophagy triggered by AR blockage (86, 87).

An increase in MYC copy number in tumor cells was observed as a result of ADT in PCa patients (88). Of interest, overexpression of MYC antagonizes the AR-dependent transcription program and confers the growth advantage in tumor cells in the absence of androgens. Therefore, increase in MYC expression may promote androgen-independent prostate tumor progression (65, 89).

2.4. Metabolic reprogramming of prostate tumor epigenetics

Interestingly, the MYC expression level inversely correlates with the global level of H3K27me3 repressive histone mark in human PCa and MYC-driven mouse prostate tumors (90). Decreased level of H3K27me3 correlates with a higher Gleason score and pathological stage. The basal cell compartment of normal prostate epithelium has a lower level of H3K27me3 as compared to the luminal cells (90). In support of this observation, MYC expression and MYC-dependent transcriptional program were shown to be upregulated in basal cells (91). Recent work has revealed that normal stem cells such as embryonic stem cells (ESCs) have an increased consumption of glutamine, which is utilized to maintain a high level of α-ketoglutarate (AKG). AKG is a co-factor for the Jumonji C domain-containing histone lysine demethylases as well as ten-eleven translocation (or TET) hydroxylases that regulate DNA demethylation (92, 93). In ESCs, AKG promotes histone demethylation including the H3K27me3 mark and therefore suppresses differentiation (21, 94). Targeting glutamine uptake via inhibition of ASCT2 (SLC1A5), the major glutamine transporter in cancer cells, suppresses the growth of PCa xenografts (45). Prostate tumor-initiating cell populations also exhibit a high dependence on glutaminolysis. Recent studies employed two clonal populations derived from PC3 PCa cells with different orthotopic and metastatic growth in xenograft models to characterize metabolic features associated with PCSCs (95, 96). These studies showed that PC3 CSC-like cells exhibit a high level of anaerobic glycolysis. These cells have a high level of lactate production and increased consumption of glutamine to compensate the acidification derived from the Warburg effect through the release of ammonia (96). In support of this data, other studies showed that advanced PCa have an increased glycolytic phenotype and a high activity of lactate dehydrogenase (LDH) correlates with poor clinical prognosis (26, 97). Knocking out genes that control OXPHOS, such as pyruvate dehydrogenase E1 component subunit alpha (PDHA1) or mitochondrial pyruvate carrier 1 (MPC1), results in an enrichment of PCa cells with CSC phenotype and increases therapy resistance and migration ability (98, 99).

3. The interaction between PCSCs and tumor microenvironment

3.1. Epithelial-mesenchymal transition and PCSCs

Epithelial-mesenchymal transition (EMT) is a process of switching from the epithelial cell characteristics to more migratory mesenchymal phenotype that is associated with loss of the epithelial markers (e.g. E-cadherin) and gain of mesenchymal signatures (e.g., vimentin, fibronectin, N-cadherin). The EMT process is associated with an increase in the expression levels of the transcriptional repressors of E-cadherin including SNAI1/2, SLUG, Twist, ZEB1/2 proteins and downregulation of microRNAs miR-200 and miR-34 that target ZEB1 and SNAI1 (100). Although EMT was initially identified as integral processes for embryogenesis and would healing, it was recently recognized as one of the drivers of tumor progression and metastases (100). Increasing evidences suggest that EMT plays a role in the regulation of PCSCs, metastatic ability and therapy resistance of PCa cells, reviewed in (101). A number of different factors might contribute to EMT in PCa including autocrine and paracrine signaling molecules released by tumor cells and stroma e.g. transforming growth factor beta 1 (TGF-β1), fibroblast growth factor (FGF), interleukin 6 (IL-6), hypoxia-inducible factor (HIF) and Wnt ligands as reviewed previously (102).

EMT can also be triggered by anticancer therapy including ADT, chemotherapy, and radiotherapy. Androgen withdrawal causes EMT and switch to the pro-metastatic phenotype by induction of the hypoxic stress, activation of ZEB1 mRNA expression and reduction of miR-200b (103–105). The diabetic drug metformin, which also exhibits anti-inflammatory activity, reverses TGF-β1/STAT3 pathway activated by anti-androgen treatment with enzalutamide and therefore inhibits EMT (106). Similar to androgen deprivation therapy, tissue injury induced by castration, chemotherapy and radiotherapy also inhibits or destroys tumor vasculature and promotes hypoxia and hypoxia-associated paracrine and autocrine responses in tumor cells and stroma that might trigger EMT primarily through activation of HIF - dependent gene expression (101, 107–109). HIF1 signaling is associated with the development of PCa metastases and activates transcription of genes involved in glycolytic flux and Warburg effect (110, 111). Stromal cells in the tumor microenvironment are also known to promote EMT, induce tumor cell reprogramming, and regulate the maintenance of PCSCs in their specific niches.

3.2. PCSC niche

Stemness is not a constitutive characteristic of CSCs, but rather represents a transient state that depends on the genetic, epigenetic and microenvironmental factors (1). Similar to normal SCs, the properties of CSCs such as self-renewal capacity, differentiation and therapy resistance are regulated by their cross-talk with specialized microenvironments called niches. In turn, factors produced by CSCs may also pathologically ‘transform’ normal stromal and endothelial cells hijacking them to promote tumor growth. CSC niches are formed, presumably, by multiple cellular components of the tumor microenvironment and shaped and regulated by the secreted factors such as cytokines and growth factors (112). The cellular components may include endothelial, immune and other stromal cells as well as the extracellular matrix produced by these cells. Stromal cells substantially contribute to tumor initiation and progression, become reactive even at the precancerous stage of prostatic intraepithelial neoplasia (113) and can contribute to PCSC transformation. For example, fibroblast growth factor FGF10 produced by urogenital mesenchymal cells induces cancerous transformation of CD49hi murine basal epithelial cells (114). Similar findings have been described for non-tumorigenic human basal epithelial cells that form tumors in recombination with cancer associated fibroblasts, CAFs (115). CAFs are one of the major components of tumor stroma which not only directly influence tumor cells but also influences other types of cells in form of stimulation (e.g. recruitment of endothelial cells) or inhibition (e.g. exclusion of T-cells) leading to the establishment of a tumor-permissive microenvironment (107). An activated CAF phenotype is associated with elevated microenvironmental oxidative stress that might, in turn, stimulate genetic changes in the epithelial cells and normal stromal fibroblasts and their cancer transformation (116). Based on the heterogeneity of expressed markers, CAFs have been suggested to derive from different cellular origins, e.g. activation of resident fibroblasts, recruitment of bone marrow derived progenitor cells to the tumor site, and trans-differentiation through the endothelial-mesenchymal or epithelial-mesenchymal transition as reviewed recently by Levesque and Nelson (107). In prostate tumors, CAFs induce a metabolic shift toward the mitochondrial oxidative phosphorylation called “reverse Warburg” phenotype. Intercellular contact of PCa cells and CAFs results in a mutual metabolic reprogramming in tumor cells and stroma. This includes a shift of CAFs metabolism toward glycolysis with accelerated expression of glucose transporter GLUT1 and increased lactic acid production and secretion. In turn, CAF-generated lactate stimulates mitochondrial biogenesis and aerobic metabolism in PCa cells (reverse Warburg effect) associated with a decrease in expression of GLUT1 transporter and activation of lactate upload (117–120). Notably, lactate-mediated tumor acidification has been associated with suppression of anticancer immune response mediated by the loss of cytotoxic T-cell and NK cell function, reduction of dendritic cell maturation and increased helper cell activities as reviewed by Choi and co-authors (121).

Accumulating evidences suggest that CAFs induce EMT and stemness in PCa through the production of metalloproteases (MMPs) and activation of the pro-inflammatory pathways in tumor cells involving nuclear factor-κB (NF-κB), cycloxygenase-2 (COX-2) and HIF-1 (119, 122). Inflammation gene signature was found to be highly upregulated in PCSCs, and chronic inflammation contributes to the development of aggressive prostate cancer (123). Inflammation plays a critical role in transformation of prostate epithelial cells and might determine PCa cell of origin. Cytokines and growth factors secreted by immune cells that infiltrate the injured or infected tissues can induce PCSCs. For example, recent study showed that inflammation expands the population of human CD38^low luminal cells that can initiate PCa after oncogenic transformation (124). Experimental studies using murine prostate tumors demonstrated that loss of AR expression results in the activation of inflammatory STAT3 signaling pathway and induction of PCSC populations. This study may suggest the role of STAT3 signaling pathway in the development of hormone-refractory PCa upon hormone deprivation (125). Furthermore, pro-inflammatory Toll-like Receptor 9 (TLR9) and TLR9-mediated activation of NF-κB and STAT3 transcription factors is essential for the maintenance of tumorigenic prostate cells (126). Notably that TLR9⁺ prostate tumor cells promote expansion and activation of myeloid-derived suppressor cells (MDSC) that has immunosuppressive effects on effector CD8⁺ T cells (127). Accumulation of MDSC in PCa correlates with disease progression (128).

Analysis of the CSCs and CAFs in the conditional Pten deletion PCa mouse model demonstrated that signals produced by CAFs positively regulate gland forming and proliferative capacities of CSCs in vivo (129, 130). The Pten^−/− mouse model recapitulates the different consequent stages of prostate tumor development, and CAFs have been shown to regulate CSCs in both androgen dependent and castration resistant (CR) tumors (129, 130). Interestingly that CAFs isolated from more aggressive CR tumors were more effective in supporting the sphere-forming properties of CSCs in vitro and formation of poorly differentiated tumors in vivo. This can be explained by the fact that interaction between tumor and microenvironment is bidirectional, and both cancer cells including CSC populations and tumor stroma mutually influence each other. Indeed, recent study showed that TGF-β1 secreted by androgen insensitive PCa cells serves for the recruitment of mesenchymal stem cells (MSC) to the tumor and their trans-differentiation into CAFs (131). Other study showed that in addition to TGF-β, other factors such as pro-inflammatory cytokine interleukin 6 (IL-6) produced by human PCa cells can activate fibroblasts. These PCa-activated fibroblasts increase PCa aggressiveness by inducing EMT, invasiveness and migration through producing matrix metalloproteinase MMPs, whereas myofibroblasts induce PCa motility by activation of serine proteases μPA/μPARs (119). CAFs synergize with other tumor stroma cells such as cancer associated macrophages (CAMs) in promoting tumor growth (132). Circulating monocytes can be attracted to tumor tissue by monocyte chemotactic protein-1 (MCP-1) produced by PCa cells or by stromal-derived growth factor-1 (SDF-1) produced by CAFs. Under exposure to IL-6 and SDF-1 released by PCa cells or CAFs, respectively, monocytes can consequently differentiate to macrophages which in turn can activate CAFs, induce de novo angiogenesis and cooperate with CAFs to increase PCa cell motility (132).

Anticancer therapy affects tumor stroma biology and its interaction with tumor cells. Recent study showed that castration- induced hypoxia activates HIF-1 and triggers autocrine TGF-β signaling in myofibroblast that promotes their activation and induction of CXCL13, an important regulator of PCa proliferation and invasion (103). Recent study showed that inhibition of CXCL13 signaling axis in prostate cancer cells impairs their tumorigenic properties (133). Tissue injury by castration, chemo- or radiotherapy also results in production of inflammatory cytokines by tumor stroma (e.g., epidermal growth factor (EGF), FGF, lymphotoxin-α (LTα), lymphotoxin-β (LTβ), IL-6, TGF-β and infiltration of immune cells (103, 107). Furthermore, DNA damage caused by cytotoxic chemotherapy in vivo induces κ light polypeptide gene enhancer in B cells 1 (NF-κB) signaling in fibroblasts which upregulates expression of wingless-type MMTV integration site family member 16B (WNT16B), activates the canonical Wnt program in PCa cells through paracrine mechanisms and promotes tumor progression (103). Taken together, these studies suggest that combination of conventional treatment with CAF-targeting therapy could decrease the effects of tumor stroma on tumor recurrence and therapy resistance and improve effectiveness of cancer treatment.

In addition to the cellular and humoral factors, physical conditions such as oxygen concentration, extracellular pH and ion strength also play a role in the CSC maintenance. Similar to CSCs, their microenvironmental niches are also dynamic in nature, and the maintenance of CSCs may be regulated by the distinct niches at the different stages of tumor development starting from tumor initiation to metastasis development (112, 134).

3.3. A -vicious cycle- of PCa bone metastases

The majority of PCa patients with advanced disease develop bone metastases (135). When the tumor spreads to the bone, it cannot be cured anymore (136, 137). This stage of disease is associated with a high risk of morbidity caused by the various complications from bone metastases (138). PCa bone metastases can produce factors that can enhance both osteoblastic and osteoclastic activities and disrupt normal bone homeostasis. Reciprocally, the supportive sites that maintain the survival and proliferation of cancer cells in the bone marrow are defined by special cytokines and growth factors produced by osteoblast, osteoclasts and bone marrow stromal cells or released from the bone matrix upon resorption. This feedback loop between the seeds (prostate tumor cells) and soil (bone microenvironment) leads to a -vicious cycle- of tumor growth as reviewed previously (139).

The bone microenvironment produces or releases various growth factors and cytokines that bind to the receptors on prostate tumor cells and regulate their growth and survival. Among them are Wnt proteins that bind to the Frizzled receptor and its LRP5/6 co-receptor and activate β-catenin - dependent gene expression, C-X-C motif chemokine ligand 12 (CXCL12) that binds to C-X-C chemokine receptor type 4 (CXCR4), as well as growth factors e.g., transforming growth factor beta (TGFβ), insulin like growth factor 1 (IGF-1), bone morphogenetic proteins (BMPs), and fibroblast growth factors (FGFs) that promote metastatic growth (140).

In turn, PCa cells in the bone can produce pro-osteolytic factors such as IL-6, IL-1, parathyroid hormone-related protein (PTHrP), and PSA that stimulate the formation of osteoclasts and promote resorption of bone matrix. Unlike bone metastases from breast cancer and myelomas, PCa metastases almost always form osteoblastic lesions (141). The factors released by PCa cells in the bone include signaling molecules affecting osteoblast proliferation or differentiation such as WNT ligands and TGFβ as well as osteoprotegerin (OPG) that regulate both bone remodeling and tumorigenesis (141, 142). OPG produced by PCa cells and by the bone marrow is an inhibitor of TNF-related apoptosis inducing ligand (TRAIL) and also an important survival factor in hormone independent PCa cells (143). Interestingly, PSA has been shown to inhibit OPG expression and increase mRNA expression of the receptor activator of nuclear factor kappa-B ligand (RANKL) that induces osteoclast differentiation (144). It is worthy mentioning that, in contrast to primary tumors that are almost always PSA positive, up to 40% of prostate metastases are negative for PSA expression (145, 146).

4. CSC evolution during tumor progression

Prostate cancer development is an evolutionary process that reflects dynamic changes of CSC phenotypes and properties over successive CSC generations (Figure 2). Luminal or basal cells in normal prostate can act as a cell of tumor origin after oncogenic transformation, however the link between the tumor initiating cells, or tumor cell of origin and PCSCs that maintain tumor growth is not yet understood (2). Prostate cancer progression is associated with the development of substantial intra-tumor heterogeneity and genomic instability that can be induced by MYC activation, loss of PTEN and mutations in DNA repair genes including BRCA2, ATM and CHEK2 (147–149). This PCSC heterogeneity fosters tumor evolution and development of metastases (Figure 2).

Figure 2. Phenotypes of CSCs, DTCs, MICs and prostate tumor progression.

Prostate cancer development is an evolutionary process that reflects an evolving of CSCs. Luminal or basal cells in normal prostate can act as a cell of tumor origin after oncogenic transformation. The link between the tumor initiating cells, or tumor cell-of-origin and cancer stem cells (CSCs) that maintain tumor growth is not yet understood. Cancer cell-of-origin refers to the initial normal cell or the cell type that became tumorigenically transformed whereas CSCs refer to the cell population that drives clonal tumor evolution. Tumor metastases are driven by the evolved populations of CSCs at their worst. Some tumor cells with malignant potential enter the blood stream (circulating tumor cells, CTCs) and can be disseminated to the distant organs. Single prostate tumor cells can be found in bone marrow and are called disseminated tumor cells (DTCs). Single tumor cells disseminated to lymph nodes are called isolated tumor cells (ITCs, not shown). A small subset of DTCs or ITCs could give rise to metastasis - initiating cells (MICs). Metastatic spread and formation is a long-time process that might take a few years. Once developed, metastases can form secondary metastasis to distant organs. Prostate cancer progression is associated with development of substantial intra-tumor heterogeneity and genomic instability that can be induced by MYC activation, loss of PTEN and mutations in DNA repair genes including BRCA2, ATM and CHEK2. A few PCSC phenotypes have been validated so far in animal models for their tumor initiating properties e.g. PSA^−/lo, ALDH^high, NANOG⁺, Trop2⁺, CD44⁺CD133⁺, ALDH^highCD44⁺α2β1⁺. Although some of the proposed biomarkers (e.g. EpCAM, CD117, c-Met) have been correlated with prostate cancer progression and metastases, there is still a lack of solid experimental evidences that these proteins can be considered as markers of PCSCs.

Acronyms: AR - androgen receptor; CK - cytokeratin; PSA - prostate specific antigen; ALDH - aldehyde dehydrogenase; ABCG2 - ATP binding cassette subfamily G member 2; α2β1 - α2β1 Integrin; ERG1 - early growth response protein 1; PTEN - phosphatase and tensin homolog; EpCAM - epithelial cell adhesion molecule; HER2 - human epidermal growth factor receptor 2; FGF - fibroblast growth factor; MAPK - mitogen-activated protein kinase; PI3K - phosphatidylinositide 3-kinase; NF-κB - nuclear factor ‘kappa-light-chain-enhancer’ of activated B-cells; TGFβ - transforming growth factor β; EGFR - epidermal growth factor receptor; CXCR4 - C-X-C chemokine receptor type 4; EZH2 - enhancer of zeste homolog 2; BRCA2 - breast cancer type 2 susceptibility protein; ATM - ataxia telangiectasia mutated; CHEK2 - checkpoint kinase 2.

Metastatic spread is a multistep process that starts from invasion of cancer cells through the basement membrane and entering the bloodstream (intravasation). At this stage circulating tumor cells (CTCs) can be disseminated to distant organs. Some CTCs that escape anoikis, immune clearance and mechanical stresses can leave the bloodstream and enter the tissues in distant organs in a process called extravasation. At this stage, these disseminated tumor cells (DTCs) can remain dormant at distant site for a few years prior to forming macrometastasis (135, 150). Only a small subset of DTCs can become metastasis-initiating cells (MICs). Selection of DTCs toward a metastatic phenotype is driven by the microenvironmental conditions in end organs as well as by the factors from primary tumors, which define the ability of DTCs to overcome tumor latency and form a metastatic tumor (135, 151–153). Metastasis formation can be considered as an evolutionary process (Figure 2). Plasticity of genetic and epigenetic makeups and selection of the most adapted tumor cells at all stages of the metastatic cascade results in the acquisition of cell-inherent traits that favor tumor dissemination (1, 136). CSCs that possess self-renewal properties and genomic instability are considered to be an engine of tumor evolution, and tumor metastases might as be driven by the constantly evolving populations of CSCs. This can be evidenced by the fact that CTCs and DTCs express many CSC markers, and MICs also share key CSC features including ability to self-renew, establish tumors and activate CSC-specific signaling pathways (136). The phenotypes and key characteristics of CSCs, DTCs and MICs during prostate tumor progression are depicted in Figure 2 and described in Table 2.

Table 2.

The features of prostate cancer stem cells (PCSCs), circulating tumor cells (CTCs), disseminated tumor cells (DTCs) and metastasis initiating cells (MICs).

Features╲Cells	Non-CSCs (tumor bulk)	CSCs	CTCs	DTCs	MICs
Androgen receptor expression	AR+ (luminal cells); ARlow/− (basal cells) (60, 91, 219)	AR− (primary PC) or AR−/+ (CRPC) (3, 11, 125, 220)	AR− or AR+ (221–225)	AR− or AR+ (226)	AR− or AR+ (227, 228)
Anrogen dependence	Primary PC: Yes CRPC: No (66, 229)	No (3, 28, 55, 220)	Yes / No (221, 230, 231)	Yes / No (228)	Yes / No (228)
PSA expression	Yes (luminal cells) (3, 60)	No (3) or low (232)	PSA+ or PSA− (230, 231, 233)	PSA+ or PSA− (234)	No data
Phenotype	Luminal or basal epithelial cells (229)	Luminal or basal epithelial cells (3, 32, 61, 62, 79, 115, 219, 235, 236)	Epithelial or Mesenchymal (EMT) (163, 222, 225, 233, 237, 238)	Epithelial (234), Mesenchymal (EMT) (163), Hematopoietic (234, 239)	No data
Tumorigenicity	Upon dedifferentiation (14, 232)	Yes (3, 32, 61, 62, 79, 115, 219, 235, 236)	Probably (240)	Probably (241, 242)	Yes (243, 244)
Metabolism	Primary PC: OXPHOS CRPC: glycolysis (Warburg effect) (26, 42)	Glycolysis (96) or/and OXPHOS (80, 245)	No data	No data	No data
Heterogeneity	Yes (246)	Yes (3, 32, 61, 62, 79, 115, 219, 235, 236)	Yes (222, 224, 225, 230, 233, 247)	Yes (226, 228, 234)	Yes (243, 244)
Cellular state	Proliferative (229)	Quiescent (3)	Proliferative or quiescent (248, 249)	Dormant or quiescent (150, 250)	No data
Genetic evolving	Yes (244, 251)	Yes (251)	Yes, reflection of tumor evolution (160, 161, 252)	Yes (159) (135, 151–153)	Yes (243, 244, 253)

In addition, CSC-specific gene signatures correlate with tumor progression and development of PCa metastasis (96, 154, 155). Therefore, enumeration of CTCs in blood and of DTCs in the bone marrow has prognostic and predictive value in PCa (97, 135, 156). Metastatic tumors, CTCs, and DTCs also show varying degrees of concordance with molecular signatures of the primary tumor of origin (157–161). Recent studies, based on the analysis of a large set of RNA expression data from more than 4,000 clinical PCa specimens, demonstrated that the tumor subtype that exhibits a high expression of luminal markers EZH2, AR, MK167 and low expression of luminal markers NKX3–1, PSA, ERG, KLK2 and basal markers, e.g., TP63 and KRT5 is associated with the highest risk of progression to metastatic disease (162). Interestingly, this gene signature was highly expressed in 58% of CTCs from patients with antiandrogen therapy-resistant tumors (162). This study demonstrated that molecular characteristics of CTCs obtained by noninvasive “liquid biopsy” could be used to identify the patients at risk of metastatic disease and treatment failure. These results are consistent with previous studies showing that PSA negative (PSA^−/low) prostate tumor cells are resistant to chemotherapy, androgen deprivation, and stress simulation, and exhibit long-term tumor-propagation capacities (3). Another study showed that CTCs and DTCs cells isolated from mouse xenografts of human prostate tumors exhibit increased potential to metastasize in vivo and resistance to the chemotherapeutic agents mitoxantrone and doxorubicin (163).

5. PCSCs and clinical application

Taking in account CSC properties such self-renewal, tumor-initiating and therapy resistance, CSC-related signatures can be employed in clinical practice for the different purposes e.g.: (1) to predict tumor development; (2) to identify patients with a poor prognosis who can potentially benefit from therapy intensification; (3) to visualize the most critical intratumoral areas containing CSCs that should be taken into account for the local therapy planning (surgery, radiotherapy); (4) to monitor an efficacy of therapeutic approaches administered in cancer patients; (5) to diagnose local or distant tumor recurrences after therapy; and (6) to develop novel specific anti-CSC therapeutics to effectively destroy the most aggressive and treatment-resistant CSCs.

Although a number of strategies for PCSC enrichment and analysis are considered in this review article, there are no specific PCSC markers known up to date which are already implemented in clinical practices. Nevertheless, there is growing evidence that PCSC specific characteristics are predictive of patient outcome (as illustrated in Table 3) that lends credence to PCSCs as potential clinical biomarkers and therapeutic targets.

Table 3.

Selected PCSC - specific characteristics as potential clinical biomarkers.

Marker / parameter	Analysis	Number of analyzed patients	Clinical correlation	Ref
- ALDH activity;	- Fluorescence-activated cell sorting (FACS) based ALDEFLUOR analysis;	n = 39	- ALDH activity and ALDH1A1 gene expression significantly correlates with Gleason score	(192)
- ALDH1A1 gene expression;	- Real-time PCR	n = 39
- ALDH1A1 protein expression	- IHC staining of TMA including BPH, PIN, early stage PCa (T1a/b), radical prostatectomy and castration resistant specimens	n = 290	- ALDH1A1 protein expression levels is higher in advanced stage PCa and significantly correlates with poor survival
CD166	- Affymetrix GeneChip microarray of benign and tumor tissue for patients who underwent radical prostatectomy for prostate cancer	n = 133	CD166/MEMD gene expression levels significantly correlate with biochemical recurrence free survival	(254)
CD117/c-Kit	CD117/EpCAM positive cells were analyzed in PCa patients’ blood by FACS	n = 99	Number of CD117/EpCAM cells is higher in high-grade patients and in patients after biochemical recurrence	(255)
PSAlow	Illumina DASL expression microarray	n = 596	PSA gene expression level correlates with biochemical recurrence free and overall PCa patient survival.	(3, 256)
CD44	IHC staining of tissues from patients with clinically localized PCa who underwent radical prostatectomy	n = 160	CD44v6 expression level correlates with biochemical recurrence free survival	(257)

There is an increasing body of evidence that PCSCs possess higher chemo- and radioresistance compared to the tumor bulk (3, 21, 164). Over the last decades, different strategies have been suggested for eradicating PCSCs that are resistance to conventional therapeutic agents. One of these strategies is targeting the signaling pathways that regulate the maintenance and survival of PCSCs such as WNT/β-catenin and PI3K/AKT (164–166). Inhibition of these signaling routes results in the sensitization of the experimental tumor models to the different types of therapy (164, 167–169). Metabolic features of PCSCs can also be potentially used for anti-cancer therapy to increase efficiency of conventional treatment such as chemo- or radiotherapy (Figure 1). However, further clinical studies are warranted to understand if these drugs can increase efficiency of PCa treatment without ablating normal stem cell functions.

Another experimental approach for PCSC targeting is based on microRNAs (miRNAs) which are non-coding RNAs that regulate gene expression by translational silencing. A number of miRNAs have been implicated in regulation both normal SC and CSCs in the different tumor entities. Recent studies showed that some miRNAs could negatively regulate PCSCs. This, for example, includes miR-34a which directly regulates CD44 expression (170); miR-141 which targets multiple genes involved in tumor development and metastatic dissemination (171); miR128 which downregulates the stem cell factors NANOG, polycomb complex protein BMI-1 and Transforming Growth Factor Beta Receptor 1 (TGFBR1); miR-199a-3p which inhibits mitogenic signaling including epidermal growth factor receptor (EGFR), c-MYC and cyclin D1 (CCND1) (48); let-7 microRNA which acts as a tumor suppressor and PCSC inhibitor by targeting key oncogenes such as MYC, RAS and High Mobility Group A2 gene (HMGA2) (172) and miR-200b which inhibits PCa growth, EMT and metastases (105). Collectively, these data provide a strong rationale for development of microRNA-based therapy against PCSCs, and ongoing studies are exploring the approaches for microRNA delivery in vivo (173).

In addition to the targeting of the pro-survival and self-renewal pathways, some novel therapeutic strategies are based on the immunologic targeting of PCa- and PCSC-specific markers such as CD44, prostate stem cell antigen (PSCA) and ALDH. To target CSC marker CD44, recent studies employed phage display library to identify high affinity peptide that target CD44v6 protein isoform and can be potentially used for systemic delivery of diagnostic and therapeutic molecules in PCa patients (174). Several other studies demonstrated potential application of anti-CD44 antibodies to target CSC cells in acute myeloid leukemia (AML) and pancreatic xenograft models as reviewed in (175).

Notwithstanding its name, PSCA is not a marker of stem cells but intermediate prostate epithelial cells that co-express basal and luminal cytokeratins (176). Nevertheless, recent study showed that PSCA is also expressed in stem cells derived from transgenic adenocarcinoma of the mouse prostate (TRAMP) mice (177). PSCA is AR-responsive gene which is highly expressed in PCa tissues and PCa bone metastases, and correlates with a poor prognosis (178, 179). This makes PSCA an attractive molecule for prostate tumor imaging and treatment. Positron emission tomography (PET) imaging with 124I-Anti-PSCA A11 minibody is currently in clinical trial for metastatic prostate cancer (ClinicalTrials.gov Identifier: NCT02092948). There are also early clinical studies in advanced castration-resistant prostate cancer patients for self-adjuvanted mRNA vaccination using prostate-cancer vaccine CV9103 which contains mRNA encoding the antigen proteins PSCA, PSA, prostate-specific membrane antigen (PSMA) and six-transmembrane epithelial antigen of prostate member 1 (STEAP1) (180). Another approach to treat localized and metastatic prostate cancers is a fully human monoclonal antibody against PSCA called AGS-PSCA which is also entered early phase clinical trial (181).

Study of Jachetti et al. on the PCSCs derived from TRAMP mouse model showed that they induce both innate and adaptive immunity. This study showed significant inhibition of tumor growth in mice immunized with PCSC-pulsed dendritic cells and therefore suggested that PCSC-derived antigens can be potentially used for the development of novel immunotherapy for prostate cancer treatment (177). As PCSCs might also express prostate antigens such as PSCA, the PCSA-targeted therapy might substantially augment tumor specific immune response and can be particularly efficient by simultaneous administration with treatment that neutralizes tumor-induced immunosuppression (177).

Studies for other tumor types such as murine squamous cell cancer, murine melanoma and human colorectal cancer demonstrated anti-tumor efficiency of other types of CSC vaccines such as dendritic cells loaded with cell lysates of ALDH positive sorted tumor cells (182, 183). These studies demonstrated that anti-tumor efficiency of the CSC-based vaccine can be significantly improved in combination with an immune checkpoint inhibitor such as anti-PD-L1 antibody which blocks the immunosuppressive effect of the tumor microenvironment (182). An ongoing Phase I/II clinical trial is testing this novel ALDH CSC-derived vaccine for patients with metastatic colorectal cancer (ClinicalTrials.gov Identifier: NCT02176746). Taking into account that ALDH is one of the clinically relevant markers of PCSCs, the results of these studies might be of high importance for the development of new anti-PCa immunotherapy. Further rigorous animal and clinical studies are warranted to demonstrate if these CSC-targeted immunotherapies alone or in combination with inhibitions of tumor immune suppression could increase the efficiency of conventional treatment approaches (175).

6. Common mechanisms underlying metastasis and therapy resistance

A growing body of evidence begins to suggest common mechanisms that underlie the development of therapy resistance and metastatic phenotypes during tumor progression (184–186). In particular, the CXCL12/CXCR4 axis is shown to regulate extravasation and dissemination of PCa cells to the bone. Furthermore, activation of the CXCL12/CXCR4 and WNT/β-catenin signaling pathways that are important regulators of PCSCs promotes tumor radioresistance and contributes to the development of skeletal metastases (21, 187–189). Aldehyde dehydrogenase (ALDH) activity can be used to enrich for PCa cells with increased tumorigenic and metastatic capability in vivo and with high radioresistance (21, 167, 190). ALDH plays an important role in the cellular response to oxidative stress and can potentially serve as one of the pro-survival mechanisms in CTCs enabling them to withstand oxidative stress in the bloodstream (191). Although several ALDH isoforms are highly expressed in PCa, only ALDH1A1 gene expression has been shown to be significantly correlated with ALDH activity and increased at high levels in advanced-stage compared to early-stage PCa and benign prostate hyperplasia (192). WNT ligands produced by tumor cells or by bone cells are acting in autocrine or paracrine fashion by enhancing tumor cell proliferation and inhibiting apoptosis (193). In addition, WNT ligands have a paracrine effect and induce osteoblastic environment promoting bone mineralization in PCa bone metastases, whereas WNT inhibitors promote an osteolytic activity (193, 194). Activation of canonical β-catenin-dependent WNT signaling is associated with malignant transformation, metastasis and radioresistance of PCa (21, 195). In different cancer models, WNT signaling pathway has been shown to regulate expression of CXCR4 and ALDH1A1 genes (21, 196, 197) whereas canonical WNT signaling pathway can be activated by CXCR4 - dependent mechanisms (198, 199). Bone-targeted therapies such as bisphosphonate drugs exhibit inhibitory effects on osteoclast-mediated bone resorption as well as on PCa cells (200, 201). Recent studies showed that bisphosphonate treatment regulates the WNT signaling pathway in PCa, e.g., by inhibition of WNT5A and FZD5 gene expression (202). TGFβ signaling activated in bone metastases regulates the pro-metastatic and pro-survival properties of PCa cells including expression of CXCR4 and increase in ALDH activity (190, 203–205). PCa cells have deregulated expression of integrins (206) that play critical roles in prostate tumor cell invasion and metastatic development (207). The PCa cell population positive for α2β1 integrin expression is enriched for tumor-initiating cells (30, 208). Recent studies showed that integrin signaling might also contribute to the resistance to androgen ablation and radiotherapy in PCa cells (209, 210).

Therefore, activation of the signaling pathways detected at the genomic and transcriptomic levels in DTCs, CTCs from the blood of the patients, and CSCs in primary tumors can be prognostic for metastasis development and may also be involved in the response of primary and metastatic tumors to therapy. Further studies are warranted to investigate the molecular mechanisms regulating metastasis-initiating and therapy-resistant prostate tumor cells and characterize their clinical relevance.

7. Summary and future directions

PCSCs that possess self-renewal properties and genomic instability might serve as a driving force of tumor evolution and metastatic dissemination. A number of recent studies demonstrated that analysis of CSCs in patients’ tumors might serve as prognostic or predictive biomarker (211, 212). In combination with other extrinsic and intrinsic parameters that might affect the properties and complexity of CSC populations, for example heterogeneity indices and hypoxia, these markers could in the future be used for patient stratification and more personalized treatment selection. However, the co-existence of multiple PCSC populations and their evolution during tumor progression and treatment could challenge the clinical application of PCSCs as prognostic biomarkers. This PCSC heterogeneity and plasticity provide a strong rationale to further development of more robust CSC-related prognostic signatures consisting of multiple markers instead of single-marker analysis, for monitoring PCSCs and CTCs during the course of treatment and for zoom in on single tumor cells analysis to identify residual PCSCs that might initiate tumor regrowth. Understanding the genetic relationship between PCSCs, CTCs, DTCs and MICs and better characterization of the MIC phenotypes and properties may be pivotal in the future for prevention of tumor cell dissemination and treatment of metastatic disease. Clinical validation of the targeting agents which are specific for PCSCs and well tolerated by normal SCs might represent a great step closer to the complete prostate cancer cure.