Defining a Population of Stem-like Human Prostate Cancer Cells That Can Generate and Propagate Castration-Resistant Prostate Cancer Free

Human tumors are heterogeneous, containing many phenotypically and functionally distinct cancer cells, and tumor cell heterogeneity can arise as a result of genetic diversity and/or epigenetic maturation of stem cell–like cancer cells or cancer stem cells (CSC; refs. 1, 2). CSCs in many human solid tumors have been implicated in tumor initiation, progression, metastasis, and therapy resistance (3, 4). Like other human cancers, prostate cancer is a heterogeneous malignancy containing phenotypically differentiated cancer cells and immature cancer cells with stem cell properties, i.e., prostate CSCs (PCSC; refs. 4–19).

Prostate cancer is one of the most common malignancies affecting American males, with an estimated 220,800 new cases and 27,540 cancer-associated deaths in 2015 (20). Localized prostate cancer at an early stage can be treated by radical prostatectomy with a good prognosis. Advanced prostate cancer patients are mostly treated by androgen-deprivation therapy (ADT), which fails eventually leading to the development of incurable and lethal castration-resistant prostate cancer (CRPC). Although many molecular mechanisms have been proposed to explain CRPC, the cellular origin for CRPC remains largely unknown (4).

In the field, the cell-of-origin of primary prostate cancer, i.e., whether human prostate cancer is derived from basal, luminal, or intermediate (progenitor) cells, has been an area of intensive studies. In mouse models, several groups have shown that both basal and luminal murine prostatic epithelial cells can function as the targets of tumorigenic transformation (21). In the human prostate, the basally localized stem cell–enriched population, i.e., CD49f^hiTrop2⁺, can be transformed by AKT, ERG, and androgen receptor (AR) and can function as the cell-of-origin for human prostate cancer (22). Whether this cell population can serve as the cell-of-origin for CRPC is presently unknown. Recent evidence indicates that some PCSC populations in hormone-naïve prostate cancer xenograft models and patient tumors express low levels of AR and could potentially function as the cell-of-origin of CRPC (6, 13, 16, 19), and vice versa, several cell populations, such as the CD166⁺ (23) and N-Cadherin⁺ (24) cells, become enriched in CRPC models. Using a lentiviral reporter tracking system, we have recently demonstrated that long-term self-renewing and tumor-propagating PCSCs are enriched in the undifferentiated prostate cancer cell population that lacks prostate-specific antigen (PSA) expression (i.e., PSA^−/lo; ref. 16). Importantly, the PSA^−/lo cell population is intrinsically resistant to androgen deprivation and enzalutamide in vitro (16, 19) and possesses high tumor-regenerating activity in castrated NOD/SCID mice (16), suggesting that it may function as the cellular origin of CRPC. Nevertheless, the PSA^−/lo cell population is still heterogeneous and the subset of cells within this population that actually mediates castration resistance remains ill defined (16, 19).

Interestingly, cDNA microarray analysis comparing gene expression in PSA⁺ and PSA^−/lo LAPC9 prostate cancer cells identified the ALDH^hiCD44⁺α2β1⁺ subpopulation, within the PSA^−/lo population, that manifested very high tumor-regenerating activity in castrated mice such that as few as 10 ALDH^hiCD44⁺α2β1⁺ cells could regenerate an androgen-independent tumor (16). The main goals of the current study are to (i) better define the ALDH^hiCD44⁺α2β1⁺ cells, which we term the triple marker-positive or TM⁺ prostate cancer cell population, with respect to its long-term tumorigenic potential in castrated hosts; (ii) test the hypothesis that the TM⁺ cell population can function as the cell-of-origin for CRPC; (iii) explore the potential molecular mechanisms underlying the CRPC mediated by the TM⁺ cell population. The results presented here have important translational and clinical significance as they shed light on CRPC evolution at the cellular level.

PC3, Du145, PPC-1, 22Rv1, and LNCaP cells were obtained from ATCC and cultured in RPMI-1640 plus 7% heat-inactivated fetal bovine serum (FBS), 10,000 μg/mL streptomycin, and 10,000 U/mL penicillin (PS; HyClone). These cell lines were regularly authenticated by our institutional CCSG Cell Line Characterization Core using short tandem repeat (STR) analysis and checked to be free of mycoplasma contamination using the Agilent MycoSensor QPCR Assay Kit (cat. #302107) every 8 weeks. All cell lines used were passaged fewer than 6 months after receipt. Antibodies used in this study are summarized in Supplementary Table S1. Immunodeficient mice (NOD/SCID and NOD/SCID-IL2Rγ^−/−, i.e., NSG) were purchased from the Jackson Laboratory, and breeding colonies were maintained in our animal facility core under standard conditions. All animal-relevant studies in the current project have been approved by the MD Anderson Cancer Center IACUC (Institutional Animal Care and Use Committee; ACUF#00000923-RN00).

Basic procedures were previously described (16, 19). For LDA, FACS purified prostate cancer cells were mixed with Matrigel and injected subcutaneously (s.c.) at increasing numbers in NOD/SCID mice. For serial tumor transplantation assays, tumor cells of a specific phenotype were sorted from the first-generation (1°) tumors of the same phenotype for 2° tumor transplantations. Sequential tumor transplantations were carried out using similar strategies. For tumor studies in castrated mice, male NOD/SCID or NOD/SCID-IL2Rγ^−/− (NSG) mice (8–10 weeks) were surgically castrated 1 to 2 weeks prior to tumor cell injections. At the endpoint, tumors were harvested and various parameters were recorded, including tumor incidence, weight, latency, and images. Tumor-initiating frequency (TIF) was calculated using the Limdil function of the Statmod package (http://bioinf.wehi.edu.au/software/elda/).

Basic procedures have been previously described (16, 19). Briefly, we produced lentivirus in 293T packaging cells (Clontech), which was then titered using GFP positivity in HT1080 cells. Prostate cancer cells were infected at a multiplicity of infection (MOI) of 10 to 20 for 48 to 72 hours at 37°C. All GIPZ-shRNA-encoding lentiviral vectors used in the current study were purchased from GE Dharmacon (Supplementary Materials and Methods). The knockdown effect on the target molecules was assessed by qRT-PCR.

We used an unpaired two-tailed Student t test to compare significance in cell numbers, percentages of CD44⁺ and/or α2⁺ cells, cloning and sphere-forming efficiencies, tumor weights, knockdown efficiency, mRNA levels of multiple genes and other related parameters. We used a χ² test to compare tumor incidence. All results were presented as mean ± SD with a P < 0.05 considered statistically significant.

See also Supplementary Materials and Methods.

In our earlier cDNA microarray analysis, we compared gene expression profiles between PSA^−/lo versus PSA⁺ LAPC9 prostate cancer cells and found that PSA^−/lo prostate cancer cells overexpressed several dozens of stem cell–related genes, including CD44, integrin α2, and ALDH1A1 (16). ALDH^hiCD44⁺α2β1⁺or TM⁺ LAPC9 cells regenerated much larger tumors when implanted in castrated mice than the corresponding ALDH^loCD44⁻α2β1⁻or TM⁻ cells (16), suggesting that TM⁺ prostate cancer cells may play an important role in CRPC development.

To directly test this suggestion, we established serially passaged androgen-independent (AI, i.e., castration-resistant) xenograft models, including LAPC9, LAPC4, LNCaP, and HPCa101 (25) from their respective androgen-dependent (AD) parental tumors (Fig. 1A). As illustrated in Fig. 1B, both LAPC9 and LAPC4 AI tumors showed a prominent upregulation of N-Cadherin, a molecule known to be involved in CRPC (24). In contrast, E-Cadherin levels did not significantly change in AI tumors in comparison to AD tumors (Fig. 1B). Interestingly, the LAPC4 AI tumors showed increased AR protein, whereas the LAPC9 AI tumors gradually lost AR, similar to earlier reports by others (24, 26). However, both AI tumor models showed decreased amounts of PSA (Fig. 1B), consistent with our earlier observations that castration resistance is associated with decreasing tumor cell PSA levels and increasing PSA^−/lo PCSCs (16, 19). Together, these results indicate that we have successfully established experimental CRPC models.

Quantitative RT-PCR (qRT-PCR) analysis revealed that the LAPC9 AI tumors expressed significantly higher levels of integrin α2, ALDH1A1, and ALDH7A1 (11) mRNAs than AD tumors (Fig. 1C). A trend of increased CD44 mRNA in LAPC9 AI tumors was also observed (Fig. 1C). Importantly, the CD44 and integrin a2 mRNA levels in TM⁺ LAPC9 cells purified from serially passaged AI tumors were significantly higher than those in the corresponding TM⁻ cells (Fig. 1D). FACS analysis, using the gating strategies we developed, showed that the percentage of TM⁺ cells dramatically increased in serially passaged LAPC9 AI tumors (Supplementary Fig. S1).

IHC and IF staining in formalin-fixed paraffin-embedded (FFPE) samples confirmed that CD44⁺ cells were highly enriched in LAPC9 AI tumors compared with AD tumor (Fig. 1E and F; and Supplementary Fig. S2A). Recent studies have linked aldehyde dehydrogenase (ALDH) activity in PCSCs to prostate cancer development (10, 11, 19, 27, 28) that might be conferred by several isoforms, including ALDH1A1 (10, 27, 28) and ALDH7A1 (11). However, whether these ALDH isoforms are expressed in CRPC samples is unclear. We found that the cells with high ALDH activity were increased in LAPC9 AI tumors (Supplementary Fig. S1, e), and the abundance of ALDH1A1⁺ and ALDH7A1⁺ prostate cancer cells was also higher in AI versus AD tumors (Supplementary Fig. S2B and S2C). In addition, integrin α2⁺ cells were increased in both LAPC9 (Fig. 1G and H) and LAPC4 (Supplementary Fig. S2D) AI tumor models.

We used FACS analysis to further investigate TM⁺ cells in several different models. In the LAPC9 model, ∼1.7% TM⁺ cells were present in AD tumors, but this percentage gradually increased with increasing cycles of castration to nearly 40% at 6° generation (Fig. 1I). The TM⁺ cells were rare in AD LNCaP and HPCa101 tumors, but their abundance still increased in the corresponding AI tumors (Fig. 1J). Finally, several commonly used AI prostate cancer cell lines all contained significant percentages of TM⁺ cells (Fig. 1K). Altogether, these results suggest that the TM⁺ prostate cancer cells are enriched in experimental CRPC models and are abundantly present in AI prostate cancer cell lines.

The ‘gold standard’ to measure CSC activity is to determine if a candidate CSC population has the capacity to generate serially xenotransplantable tumors in immunodeficient mice at increasing cell doses (an assay termed limiting-dilution tumor regeneration assay or LDA), and if the regenerated tumors phenocopy, at least partially, parental tumor histopathologically (4, 29). Therefore, LDA and serial transplantation assays are combined in our study to characterize the CSC properties of the TM⁺ and related prostate cancer cell populations.

Because TM⁺ cells preexist in AD tumors (Fig. 1I–J; Supplementary Fig. S1A), we first determined their tumor-initiating and tumor-propagating activities in androgen-proficient hosts, using strategies similar to those we previously used to characterize the PSA^−/lo cells (16). Thus, TM⁺ cells and several other cell populations, including ALDH^hiCD44⁻α2β1⁻, ALDH^loCD44⁺α2β1⁺, and TM⁻, were purified from LAPC9 AD tumors and s.c. injected in hormonally intact NOD/SCID male mice at increasing doses (from 10 to 10,000 cells; Table 1). Surprisingly, at 1°, although TM⁺ cells exhibited a trend of higher tumorigenicity than corresponding TM⁻ cells (TIF 1/1,312 vs. 1/2,972; P = 0.0647), two intermediate cell populations, ALDH^hiCD44⁻α2β1⁻ (TIF 1/315) and ALDH^loCD44⁺α2β1⁺ (TIF 1/421), were significantly more tumorigenic than TM⁺ cells (Table 1). However, as tumor cells of a specific phenotype were sorted from the 1° tumors of the same phenotype (e.g., purifying TM⁺ cells from 1° TM⁺ cell-derived tumors) for the 2° transplantations in intact male mice, TM⁺ cells maintained the highest tumorigenicity compared with ALDH^hiCD44⁻α2β1⁻, ALDH^loCD44⁺α2β1⁺, and TM⁻ cells (Table 1). Similarly, TM⁺ cells were greatly enriched in tumor-regenerating capacity compared with all other cell populations at the 3° transplantations in intact male mice (Table 1). Notably, during serial passages, tumors regenerated from TM⁺ cells at each generation contained TM⁺ cells as the minority with the majority of cells being non-TM⁺ cells (Supplementary Fig. S3, left), suggesting that the TM⁺ cells possess self-renewal potential in vivo. Interestingly, protein levels of two well-characterized differentiation markers, i.e., AR and PSA, were significantly decreased in TM⁺ cell-derived AD tumors than TM⁻ cell-derived counterparts even after two consecutive passages in intact male mice (Supplementary Fig. S4), supporting the general notion that CSCs are less differentiated (4). Taken together, these observations demonstrate that TM⁺ LAPC9 cells are endowed with CSC (i.e., long-term tumor-propagating) properties in hormone-proficient conditions.

Given that TM⁺ cells are prominently enriched in AI tumors, we ponder a critical question: Can TM⁺ cells in AD tumors initiate tumor regeneration in androgen-ablated hosts? To answer this question, we purified TM⁺ and TM⁻ cells from maintenance of LAPC9 AD tumors and performed LDA by implanting different doses of cells into castrated immunodeficient NOD/SCID-IL2Rγ^−/− (NSG) mice. Remarkably, TM⁺ cells demonstrated ∼43-fold higher tumor-initiating capacity in castrated hosts compared with TM⁻ cells (TIF 1/49 vs. 1/2,142; Fig. 2A). As few as 100 TM⁺ cells initiated 7/8 tumors, whereas 100 TM⁻ cells did not regenerate a single tumor in castrated hosts (Fig. 2A). Differences in both tumor incidence and weight between the two groups were statistically highly significant (Fig. 2A). This experiment provides direct evidence that the TM⁺ LAPC9 cells can function as the cell-of-origin for CRPC.

To determine whether the TM⁺ cells can in the long term propagate CRPC in androgen-deficient hosts, we purified TM⁺ and TM⁻ cell populations from LAPC9 AI tumors and carried out LDA and serial transplantation assays in fully castrated hosts, i.e., castrated male NOD/SCID mice also treated with bicalutamide (ref. 16; Fig. 2B and C; Table 1). In the 1° generation, the TM⁺ cells demonstrated much higher tumorigenic potential than TM⁻ cells in castrated mice (TIF 1/2, 590 vs. 1/21,299, respectively; Fig. 2B; Table 1). In the 2° generation tumor experiments, the TM⁺ cells, in a cell dose–dependent manner, regenerated robust AI tumors in castrated hosts such that as few as 100 cells generated 5/8 tumors, whereas the TM⁻ cells did not generate tumors at 100 and 1,000 cell doses (Fig. 2C; Table 1). Overall, the TM⁺ cells manifested significantly higher 2° tumor-regenerating capacity than the TM⁻ LAPC9 cells (TIF 1/283 vs. 1/13,802, respectively; P = 2.23e−09; Table 1). These results suggest that the TM⁺ LAPC9 cells possess long-term CRPC-propagating capability. In full agreement with this conclusion, both primary and secondary LAPC9 AI tumors derived from the TM⁺ cells contained a similar small population (i.e., 10%–20%) of TM⁺ cells as the parental AI tumors (Supplementary Fig. S3, right), suggesting that the TM⁺ cells in AI tumors self-renewed in vivo.

To further determine the intrinsic castration resistance in TM⁺ prostate cancer cells, we sorted TM⁺ and non-TM⁺ (i.e., TM-depleted) cells from LAPC9 AI tumors, and carried out similar LDA in castrated mice. We observed that the TM⁺ cells displayed much higher tumor-regenerating activity than the isogenic TM-depleted cells (Fig. 2D; Table 1). Likewise, TM⁺ cells were able to self-renew for subsequent generations (Table 1). It was apparent that the TM⁺ cells outcompeted other subpopulations (e.g., ALDH^hiCD44⁺α2β1⁻, ALDH^loCD44⁺α2β1⁺) with respect to tumor-initiating and tumor-propagating abilities in castrated mice (Table 1). In support of the above tumor experiments, the TM⁺ LAPC9 cells purified out from the AI tumors demonstrated much higher sphere formation (Fig. 2E) and clonogenic (Fig. 2F) activities than the TM⁻ cells in “castrated” culture conditions.

Lastly, we applied similar strategies to study additional AI models. TM⁺ PC3 cells were significantly more clonogenic than TM⁻ PC3 cells in castrated culture conditions (Fig. 2G and H). Importantly, the TM⁺ PC3 cells manifested much higher tumorigenic potential than TM⁻ PC3 cells generating more and larger tumors (Fig. 2I). Purified TM⁺ LAPC4 AI cells also exhibited increased clonogenicity as compared with TM⁻ LAPC4 AI cells in “castrated” culture conditions (data not shown). Combined, these results demonstrate that (i) the TM⁺ cells in LAPC9 AD tumors can function as the cell-of-origin for LAPC9 AI tumors, (ii) the TM⁺ cells in LAPC9 AI and PC3 models possess great CRPC-regenerating activity, (iii) the LAPC9 AI TM⁺ cells possess long-term CRPC-propagating activities, and (iv) the AI TM⁺ cells from multiple models exhibit intrinsic castration resistance in vitro.

Next, we explored the potential clinical relevance of the above findings in xenograft studies using cells purified from 16 untreated HPCA samples, all of which have >75% tumor involvement (Supplementary Table S2; Supplementary Fig. S5A). Tumor areas were confirmed by robust expression of AR, PSA, and racemase and lack of expression of CK5 (Supplementary Fig. S5B). We first purified out HPCA cells as the CD45⁻Trop2⁺ epithelial cells (22, 30), which were then analyzed for the abundance of TM⁺ cells (Fig. 3A and B). We found that all these untreated primary tumors contained TM⁺ cells ranging from ∼0.001% to as high as ∼14.5% (Fig. 3B). When the TM⁺ cells were purified from HPCa201 and cultured in castrated conditions (CDSS + enzalutamide), they exhibited much enhanced clonal, clonogenic, and self-renewal capacities compared with TM⁻ cells (Fig. 3C). Notably, the HPCa201 TM⁺ cells also demonstrated significantly higher secondary colony-forming activity than the corresponding TM⁻ cells (Fig. 3C, b). Purified HPCa222 TM⁺ cells generated much bigger and more clones and colonies than corresponding TM⁻ cells in castrated culture conditions (Fig. 3D). Importantly, the HPCa222 TM⁺ cell–derived colonies were positive for TMPRSS–ERG fusion (Supplementary Fig. S5C-D), indicating the tumor cell origin of the colonies. The HPCa 223 TM⁺ cells similarly demonstrated higher clonal and colony-forming capabilities than the corresponding TM⁻ cells (Supplementary Fig. S6A). Interestingly, some HPCA samples contained almost undetectable TM⁺ cells (Fig. 3B) but contained double-, and/or single-positive cells (Fig. 3A; 19). In the HPCA samples analyzed, we observed that double-positive (i.e., ALDH^hiCD44⁺) HPCA cells were much more clonal and clonogenic than double-negative (i.e., ALDH^loCD44⁻) cells with significantly increased colony sizes and clone numbers (Supplementary Fig. S6B and S6C). However, HPCa210 ALDH^hi cells exhibited similar clonal and clonogenic abilities as ALDH^lo cells (data not shown). Taken together, these results indicate that untreated primary HPCA samples also contain TM⁺ cells that are intrinsically resistant to castration.

Next, we attempted to determine whether the phenotypic markers used to define TM⁺ cells are causally involved in tumor regeneration under AI conditions by performing lentiviral-mediated knockdown of integrin α2, CD44, ALDH1A1, and ALDH7A1 followed by assessing the impact on tumor regeneration in castrated male NOD/SCID mice (Supplementary Fig. S7; see Supplementary Fig. S8A and S8B for knockdown efficiency). Knocking down integrin α2 in LAPC9 AI tumor cells decreased tumor incidence (5/9 or 55.6%) as compared with the control (6/6 or 100%; Supplementary Fig. S7A), whereas knocking down integrin α2 in LAPC4 AI tumor cells resulted in significantly smaller tumors (Supplementary Fig. S7B). CD44 knockdown in LAPC9 AI tumor cells not only inhibited tumor initiation but also reduced tumor burden (Supplementary Fig. S7C). Finally, ALDH1A1 knockdown in LAPC9 AI cells led to reduced spheres (Supplementary Fig. S8C) and a 50% inhibition in tumor weight (Supplementary Fig. S7D), whereas ALDH7A1 knockdown partially affected sphere-forming and tumor-initiating capacities in LAPC9 AI cells (Supplementary Fig. S8C; data not shown). Collectively, these results indicate that the phenotypic markers are causally important in imparting the castration-resistant CSC properties.

To further characterize the TM⁺ cells at the molecular level, we purified out TM⁺ and TM⁻ LAPC9 cells from AD and two different generations of AI tumors and performed qRT-PCR analysis of the genes associated with luminal/basal cell differentiation, castration resistance, and CSCs (Fig. 4A–C; Supplementary Table S3). This analysis revealed interesting changes in the gene expression patterns. In LAPC9 AD tumors, the TM⁺ cells overall expressed higher levels of basal cell genes (CK5, CK14, and CD44) whereas the TM⁻ cells expressed higher levels of luminal genes (AR, PSA, and CK18 with the exception of CK8; Fig. 4A). In LAPC9 AI tumors, the TM⁺ cells still expressed higher levels of basal markers CK5 and CK14 but the TM⁻ cells expressed high PSA, CD26 (31), and CK19 mRNA levels while expressing similar levels of other luminal markers (i.e., AR, CK8, CK18, and Nkx3.1) to TM⁺ cells (Fig. 4B and C).

The TM⁺ cells in the 6° generation LAPC9 AI tumors consistently expressed higher levels, compared with the TM⁻ cells, of RegIV, SOX9, and UBE2C, which have been implicated in castration resistance and CRPC progression (32–35), as well as several CSC-associated molecules, including TGFBR-1, Bcl-2, and Stat3 (Fig. 4C). Preferential expression of RegIV and SOX9 in TM⁺ cells was further confirmed in a 15° generation LAPC9 AI tumor (Fig. 4D). To determine whether RegIV and SOX9 are causally involved in conferring on TM⁺ LAPC9 cells, certain CSC properties and castration resistance, we knocked down RegIV and SOX9 in LAPC9 AI bulk cells (Supplementary Fig. S8D–S8E), which significantly attenuated their sphere-forming abilities (Fig. 4E). RegIV and SOX9 knockdown in TM⁺ LAPC9 AI cells also reduced their clonogenicity in castrated culture conditions (Fig. 4F) and tumor growth in castrated mice (Fig. 4G). These data suggest that RegIV and SOX9 regulate the CSC properties of the TM⁺ LAPC9 AI cells.

Our laboratory has been studying the mechanisms whereby microRNAs (miRNA) regulate PCSCs and prostate cancer development and metastasis (36–38). To understand how miRNAs might regulate the TM⁺ castration-resistant prostate cancer cells, we used a miRNA SmartChip that contained >1,000 human miRNAs to compare miRNA expression profiles of TM⁺ with TM-depleted LAPC9 AI cells. We found that 59 miRNAs were significantly upregulated, whereas 22 miRNAs were downregulated in the TM⁺ cells (Fig. 5A; Supplementary Table S4). As an example to establish a cause-and-effect relationship between the differentially expressed miRNAs and castration resistance, we focused on miR-499-5P, which was among the most upregulated miRNAs in the TM⁺ cells (Fig. 5A). miR-499-5P is a relatively novel and understudied miRNA. A recent paper reported its potential oncogenic functions in colorectal cancer cells via targeting FOXO4 and PDCD4 (39). We found that miR-499-5P was expressed at higher levels in purified TM⁺ LAPC9 AI cells (Fig. 5B) and in the LAPC9 AI tumors (Fig. 5C). Overexpression of miR-499-5P in LAPC9 AD tumor cells by transfection of miR-499-5P oligos promoted sphere formation compared with cells transfected with negative control oligos (Fig. 5D). In contrast, transfection of the anti-miR-499-5P oligos in the TM⁺ LAPC9 AI cells suppressed sphere formation (Fig. 5E). Importantly, overexpression of miR-499-5P in the TM-depleted LAPC9 AI cells promoted tumor regeneration in castrated mice (Fig. 5F), whereas overexpression of anti-miR-499-5P in the TM⁺ LAPC9 AI cells significantly inhibited tumor growth (Fig. 5G). Together, these results demonstrate that the TM⁺ prostate cancer cells express a unique miRNA profile that causally regulates their tumorigenic and castration-resistant properties.

Two principal mechanisms, i.e., intrinsic and acquired, underlie the resistance of cancer cells to both general and targeted anticancer therapeutics (40). Intrinsic resistance implies that prior to treatments, therapy-insensitive cells preexist in the tumor and become selected and enriched during treatment, whereas acquired resistance is caused by treatment-induced gene mutations and other adaptive responses. Both mechanisms of therapy resistance have been reported in clinical tumors (40–42). Here, we provide a prototypical example for a population of cancer cells that can mediate intrinsic therapy resistance.

As early as 1981, Isaacs and Coffey, working on the rat Dunning R-3327-H prostate adenocarcinoma (H-tumor) model, performed a fluctuation analysis to show that the H-tumor relapse after ADT was due to continuous proliferative growth of preexisting AI prostate cancer cells after castration (43). Exactly what cell population in this model that became selected during castration was uncharacterized (43). In 1999, Craft and colleagues, by performing a similar fluctuation analysis, provided evidence in the LAPC9 model that ADT selected and clonally expanded the preexistent AI cells, resulting in outgrowth of CRPC (44). Nevertheless, which cell population in LAPC9 mediated the CRPC emergence was unclear. Here, we demonstrate that the TM⁺ cell population in the LAPC9 model can function as both a cell-of-origin for CRPC emergence and the tumor-propagating population for the established CRPC.

LAPC9 is a well-known prostate cancer xenograft expressing PSA and wild-type AR, which our laboratory has been using for the past 10 years to study PCSCs (6, 7, 16, 19, 36). Our earlier work has shown that CD44⁺ prostate cancer cells in several xenograft models, including LAPC9, are highly enriched in PCSCs (6). Follow-up work demonstrates that the CD44⁺α2β1⁺ LAPC9 cells are even more tumorigenic than CD44⁺ cells (7). The current project emanated from our recent observations that the PSA^−/lo cell population, which preexists in the AD tumors but becomes enriched in AI tumors, can regenerate larger tumors than PSA⁺ cells in fully castrated hosts (16). This study provided the first evidence for a population of preexisting human prostate cancer cells that preferentially generate AI tumors (16). Nonetheless, both PSA^−/lo and PSA⁺ LAPC9 cells regenerated AI tumors with similar incidences (16), suggesting that only a subset of the PSA^−/lo cells is actually mediating the AI regeneration. Indeed, microarray analysis led to the identification of the TM⁺ subset that was greatly enriched in AI tumor-regenerating activities (16).

By establishing serially passaged AI tumors and by focusing on the LAPC9 system, we have made many important and novel findings. First, the TM⁺ cells in AD tumors possess long-term tumor-propagating activities in androgen-proficient hosts. Second, the TM⁺ cells purified from the LAPC9 AD tumors can preferentially initiate xenograft tumors in fully castrated hosts, suggesting that this population can preferentially function as the cell-of-origin for AI prostate cancer. Third, the TM⁺ cells gradually become enriched in serially transplanted AI tumors in vivo. Fourth, the TM⁺ LAPC9 cells purified from the AI tumors can serially propagate the xenograft tumors in androgen-ablated hosts compared with either TM⁻ or TM-depleted cells, suggesting that the TM⁺ LAPC9 cells can also function as robust CRPC-propagating cells. Fifth, the phenotypic markers used to define TM⁺ cells are functionally important for the TM⁺ cell activities. Finally, the TM⁺ LAPC9 cells seem to possess a unique miRNA expression profile. To our knowledge, the (serial) tumor transplantation studies presented here (Fig. 2; Table 1) represent the most comprehensive effort to define a population of human prostate cancer cells that function as the cell-of-origin as well as the tumor-propagating cells for CRPC.

Is the TM⁺ cell population unique to the LAPC9 model? At least in one other model, i.e., PC3, which is all PSA⁻, the TM⁺ cells exhibit significantly higher tumor-regenerating activity than the corresponding TM⁻ cells in castrated animals. It is interesting that several other AI cell lines examined, including PPC-1, Du145, and 22Rv1, all contain significant percentages of TM⁺ cells, which, we believe, may also possess high AI tumor-regenerating capacities compared with the corresponding TM⁻ cells. In several other xenograft models, including LNCaP and HPCa101, the fraction of TM⁺ cells in the AD tumors is small, which, nevertheless, also increases in the AI tumors, although the significance of the TM⁺ cell fraction in these models remains to be determined. Of significance, however, we have detected the TM⁺ cells in more than a dozen of untreated HPCA samples. The relative abundance of TM⁺ cells in primary HPCA samples vary widely, but freshly purified TM⁺ HPCA cells manifested significantly higher colony-forming abilities than the corresponding TM⁻ cells in castrated culture conditions. Intriguingly, the ALDH^hiCD44⁺ double-positive cells isolated from two HPCA samples also display significant castration-resistant CSC properties. Taken together, our studies in multiple culture/xenograft models and primary HPCA samples suggest that (i) the TM⁺ prostate cancer cell population seems to be widely present, (ii) the TM⁺ cells in some tumor systems, exemplified by LAPC9 and PC3, possess great tumor-initiating and tumor-propagating activities in androgen-deficient hosts, (iii) the TM⁺ HPCA cells in primary tumors also possess great survival advantages and colony-forming capabilities under androgen-deficient culture conditions, and (iv) in some HPCa samples, the ALDH^hiCD44⁺ double-positive cells may behave similarly to the TM⁺ cells.

What are the TM⁺ cells? qRT-PCR–based phenotyping suggests that the TM⁺ LAPC9 cells in both AD and AI tumors are basal-like, preferentially expressing CK5, CK14, and CD44 mRNAs and less differentiated expressing lower levels of PSA and CD26 mRNAs. This would be consistent with the fact that this subpopulation of the cells was initially uncovered from the PSA^−/lo prostate cancer cell population (16). This is also consistent with our current observations that the AI cell lines, such as PPC-1, PC3, and Du145, all of which are PSA^−/lo, have significantly higher percentages of the TM⁺ cells than the AD LAPC9, LAPC4, and LNCaP tumors, in which the majority of cells are PSA⁺. It should be noted, though, that cell populations that can mediate and propagate CRPC may not necessarily be all basal-like. For instance, in the BM18 xenograft model, cells that mediate CRPC prominently express luminal markers CK18 and NKX3.1 (14).

What confers the TM⁺ prostate cancer cells the ‘hardy’ properties to readily regenerate and propagate the AI tumors? Part of the answer seems to reside with the three defining phenotypic markers, i.e., CD44, integrin α2, and ALDH1A1, whose knockdown all diminished the tumor-regenerating activities of the TM⁺ cells in castrated conditions. This is not surprising considering that the CD44 and integrin α2 function as important adhesion and signaling molecules and that ALDH1A1, and perhaps ALDH7A1, may further extend cell survival by detoxification (e.g., eliminating ROS). Another reason is that the TM⁺ prostate cancer cells preferentially express genes associated with (cancer) stem cells and castration, illustrated by molecules such as Reg IV, SOX9, UBE2C, and TGFBR-1. Knocking down RegIV and SOX9 in the TM⁺ LAPC9 cells partially inhibits tumor regeneration in androgen-deficient hosts, implicating their causal functions in TM⁺ cells. Finally, the TM⁺ cells express a unique profile of miRNAs. Using miR-499-5P as an example, which is overexpressed in the TM⁺ cells compared with the TM-depleted cells, we show that this miRNA is also conferring on the TM⁺ LAPC9 cells castration-resistant properties. Other microRNAs uncovered in this screening may also likely be involved in modulating the biologic properties of the TM⁺ prostate cancer cells. Collectively, the results make it clear that multiple molecules and mechanisms likely come together to endow the TM⁺ prostate cancer cells unique capabilities to function as both the cell of origin and tumor-propagating cells for CRPC.

Our recent work has illustrated that the PSA^−/lo prostate cancer cell population represents a therapeutic target in treating CRPC (16, 19). The current work further pinpoints the TM⁺ subset within the PSA^−/lo population as the driving force of CRPC emergence and propagation. Ongoing work from our laboratory is focusing on devising strategies to target the highly tumorigenic TM⁺ cell population and on elucidating the potential relationship between TM⁺ cells and several other populations of prostate cancer cells, e.g., CD166⁺ (23) and N-Cadherin⁺ (24), that may also be able to propagate castration-resistant tumors.

No potential conflicts of interest were disclosed.

Conception and design: X. Chen, X. Liu, D.G. Tang

Development of methodology: X. Chen, Q. Li, X. Liu, B. Liu, D.G. Tang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): X. Chen, Q. Li, J. Shen

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): X. Chen, Q. Li, C. Liu, B. Liu, K. Lin, Y. Lu, H.-P. Chao, D.G. Tang

Writing, review, and/or revision of the manuscript: X. Chen, Q. Li, R. Liu, K. Rycaj, C. Jeter, K. Lin, Y. Lu, D.G. Tang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): X. Chen, C. Liu, D. Zhang, T. Calhoun-Davis, D.G. Tang

Study supervision: B. Liu, D.G. Tang

Other (coordinating several collaborative groups): D.G. Tang

The authors thank Animal Core for animal care and maintenance, Molecular Biology Core for help in Wafergen SmartChip human miRNA array, Histology Core for IHC studies, Ms. P. Whitney for assistance in FACS sorting and analysis, and other lab members for helpful discussions and suggestions.

This project was supported, in part, by grants from NIH (R01-CA155693), Department of Defense (W81XWH-13-1-0352 and W81XWH-14-1-0575), CPRIT (RP120380), and the MDACC Center for Cancer Epigenetics (D.G.T). X. Chen and C. Liu were supported, in part, by DOD post-doctoral fellowship PC141581 and PC121553, respectively. Both J. Shen and Y. Lu were supported by CPRIT Core Facility Support Award RP120348.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.