|Home | About | Journals | Submit | Contact Us | Français|
Prostate cancer (PCa) is heterogeneous and contains both differentiated and undifferentiated tumor cells, but the relative functional contribution of these two cell populations remains unclear. Here we report distinct molecular, cellular, and tumor-propagating properties of PCa cells that express high (PSA+) and low (PSA−/lo) levels of the differentiation marker PSA. PSA−/lo PCa cells are quiescent and refractory to stresses including androgen deprivation, exhibit high clonogenic potential, and possess long-term tumor-propagating capacity. They preferentially express stem cell genes and can undergo asymmetric cell division generating PSA+ cells. Importantly, PSA−/lo PCa cells can initiate robust tumor development and resist androgen ablation in castrated hosts, and harbor highly tumorigenic castration-resistant PCa cells that can be prospectively enriched using ALDH+CD44+α2β1+ phenotype. In contrast, PSA+ PCa cells possess more limited tumor-propagating capacity, undergo symmetric division and are sensitive to castration. Together, our study suggests PSA−/lo cells may represent a critical source of castration-resistant PCa cells.
PCa is heterogeneous manifesting variegated cellular morphologies and histopathological presentations. PCa also exhibits great intra-tumor histological and immunophenotypic heterogeneities, with low-grade tumors often harboring poorly differentiated areas and high-grade tumors containing relatively differentiated foci. The cellular basis for the histological and cellular heterogeneity of PCa remains unclear.
Androgen and androgen receptor (AR) signaling has been implicated in PCa. Androgen-deprivation therapy (ADT) blocks androgen production or AR signaling and is the mainstay treatment for advanced and recurrent PCa but such interventions only achieve short-term efficacy due to the emergence of castration-resistant disease (i.e., CRPC). Although many mechanisms, mostly centered on AR, have been proposed for CRPC development (Shen and Abate-Shen 2010; Wang Q et al., 2009), the cell-of-origin and molecular identity of CRPC cells remain undefined.
Numerous studies have demonstrated that PSA (prostate-specific antigen) protein expression in PCa positively correlates with its overall degree of differentiation (e.g., Abrahamsson et al., 1988; Feiner and Gonzales, 1986; Gallee et al., 1990). At the cellular level, PCa contains differentiated cancer cells expressing high levels of PSA (i.e., PSA+) as well as PCa cells that express little or no PSA (i.e., PSA−/lo). The PSA−/lo cells appear to be rare in early-stage tumors but become more abundant in high-grade and locally advanced tumors and some cases of PCa may completely lack PSA expression. PCa patients with their tumors containing >50% PSA+ PCa cells tend to have longer survival (Roudier et al., 2003; Shah et al., 2004). These clinical observations raise a fundamental question: could PSA−/lo PCa cells be intrinsically distinct from PSA+ cells and thus play differential roles in tumor maintenance and progression to CRPC? Herein, we address this clinically relevant question using a PSA promoter-driven lentiviral reporter system to separate bulk PCa cells into PSA−/lo and PSA+ subpopulations.
We first performed a semi-quantitative PSA immunohistochemical (IHC) analysis in cohorts of untreated Gleason 7 (GS7, n = 10), Gleason 9 or 10 (GS9/10, n = 10), and treatment-failed (n = 23) PCa (Figure S1; Table S1). Most tumor glands in GS7 tumors stained strongly for PSA but there existed poorly differentiated areas of PSA−/lo cells (Figure S1A). In contrast, in GS9/10 tumors, the main histological pattern was undifferentiated tumor mass in which most tumor cells were PSA−/lo with PSA+ foci only occasionally present (Figure S1B). In 23 recurrent PCa cases (mainly CRPC), some tumors resembled untreated GS9/10 tumors but most tumors completely lacked PSA+ PCa cells (Figure S1C–F). Quantification revealed significantly increased numbers of PSA−/lo PCa cells in untreated GS9/10 and treatment-failed PCa compared to untreated GS7 tumors (Figure 1A).
Consistent with the IHC results, analysis of multiple microarray data sets in Oncomine revealed that tumor PSA mRNA levels were significantly decreased in high-grade primary tumors and in recurrent and metastatic PCa (Figure S2; data not shown). Importantly, reduced tumor PSA mRNA levels correlated with lymph node positivity, tumor recurrence, metastasis, and shortened patient survival (Figure S2; data not shown; also see Figure 7A). Together, the PSA IHC and mRNA analysis indicates that advanced and recurrent PCa have lower PSA mRNA and more undifferentiated PSA−/lo cells.
To separate PSA−/lo from PSA+ PCa cells, we employed the PSAP-GFP lentivector, in which the PSA promoter (PSAP) drives eGFP expression (Yu et al., 2001) (Figure S3A). The PSAP was originally isolated from a PCa patient with high serum PSA and was highly specific and sensitive for PSA-positive prostate (cancer) cells. We also generated two modified PSAP-GFP vectors (Figure S3A).
Using the PSAP-GFP vector, we infected LNCaP cells at an MOI of 25 (Figure 1B), at which virtually all cells were infected as evidenced by PCR detection of the GFP sequence in genomic DNA of randomly picked clones (Figure 1C). We then used fluorescence-activated cell sorting (FACS) to purify out the top 10% GFP-bright (GFP+) and bottom 2–6% GFP-negative/GFP-dim (i.e., GFP−/lo) LNCaP cells. The purity of GFP−/lo and GFP+ cells was 98–100% and ≥97%, respectively (e.g., Figure S3B). LNCaP cells routinely cultured in RPMI-7% FBS contained 2.7 ± 1.8% (0.3 – 6.0%; n = 15) GFP−/lo cells. When LNCaP cells were infected with PSAP-GFP-Psv40-neo (Figure S3A) followed by G418 selection for several weeks, we observed 2.7 ± 1.7 % (n = 7) GFP−/lo cells.
The percentage of GFP−/lo LNCaP cells was very close to that of PSA−/lo cells in LNCaP cultures (2.2 ± 1.5%; n = 4). Real-time (qPCR; Figure 1D) and semi-quantitative (Figure S3C) RT-PCR revealed lower PSA mRNA levels in GFP−/lo LNCaP cells compared to the corresponding GFP+ cells. Also, most purified GFP+ LNCaP cells stained strongly positive for PSA protein whereas GFP−/lo cells were weak or negative for PSA (Figure 1E). GFP−/lo LNCaP cells also expressed lower levels of AR mRNA (Figure 1D; Figure S3C) and protein (Figure 1F–G) compared to GFP+ cells. These results indicate that the PSAP-GFP lentiviral system faithfully reports endogenous PSA expression. Hence, in many forgoing experiments we refer to GFP+ and GFP−/lo cells as PSA+ and PSA−/lo cells, respectively.
AR staining revealed ~82% and 18% GFP+ LNCaP cells showing strong and intermediate nuclear AR, respectively, and no GFP+ LNCaP cells were negative for AR (Figure 1F). In contrast, 46% of the PSA−/lo LNCaP cells were completely negative for AR whereas 41% and 13% PSA−/lo LNCaP cells had weak and strong AR, respectively (Figure 1F). These results suggest that the majority of PSA+ PCa cells are high in AR whereas PSA−/lo cells express a gradient of AR, from completely negative to strong nuclear staining.
When PSAP-GFP infected LNCaP cells were cultured in androgen-deprived conditions, i.e., using charcoal dextran-stripped serum (CDSS) or with bicalutamide (an antiandrogen), PSA+ cells dramatically decreased with a concomitant expansion of PSA−/lo cells (Figure S3D). Purified PSA−/lo LNCaP cells also displayed higher survival and holoclone (Li et al., 2008) forming capacity in the absence of androgen (Figure S3E). These results suggest that PSA−/lo PCa cells are resistant to androgen deprivation.
Whole-genome transcriptome profiling in purified PSA−/lo and PSA+ LNCaP cells revealed distinct gene expression patterns in the two isogenic subpopulations (Figure 1H). A total of 726 probes representing 561 unique genes were significantly overexpressed whereas 557 probes representing 403 genes were under-expressed (fold change [F.C] ≥1.4, P <0.05) in PSA−/lo LNCaP cells (Figure S3F; Figure S3G shows qPCR of several genes). A combination of Gene Ontology (GO) analysis and literature-based curation put many of these differentially expressed genes into distinct functional categories (Figure 1H; Table S2). Strikingly, as many as 10% of the genes overexpressed in PSA−/lo LNCaP cells were involved in anti-stress responses, which included detoxification (metallothioneins, GSTT2, etc), hypoxia-responsive (HIF1α, THBS1, PLAU, APLN), p53 signaling (e.g, ZBTB7A, PSME3), and DNA-damage sensing/repair (e.g., MSH6, XPA, REV1) genes (Figure 1H–I; Table S2). The PSA−/lo LNCaP cells also overexpressed Bcl-2 and under-expressed many proapoptotic genes (Table S2).
Differential expression of anti-stress and proapoptotic genes suggests that the PSA−/lo cells would be more resistant to not only androgen deprivation but also other stresses. Indeed, when LNCaP cells infected with PSAP-GFP were treated with CDSS plus bicalutamide, etoposide, paclitaxel (taxol), or H2O2, PSA−/lo cells expanded with concomitant decreases in PSA+ cells (Figure 1J). FACS analysis indicated that these treatments preferentially induced apoptosis in PSA+ LNCaP cells (not shown).
The PSA−/lo LNCaP cells under-expressed dozens of cell cycle and mitosis-related genes (Figure 1H; Figure S3H; Table S2), suggesting that PSA−/lo PCa cells may be more quiescent than PSA+ cells. Several lines of evidence supported this suggestion. First, cell-cycle analysis revealed a smaller percentage of PSA−/lo LNCaP cells in S and G2/M phases (Figure 1K). Second, the PSA−/lo and PSA+ LNCaP populations had 4.2% and 12%, respectively, of Ki-67+ cells (P <0.0001). Third, BrdU label-retaining experiments demonstrated that many more PSA−/lo LNCaP cells retained the BrdU label upon an 11-d chase (Figure 1L).
The observations that PSA−/lo LNCaP cells are quiescent and resist stress stimulations suggest that the population may be enriched in stem cells (SCs) (Laffin and Tang, 2010). In support, the PSA−/lo LNCaP cells, in androgen/serum-free medium, possessed higher capacity to establish holoclones (Figure S3E) and anchorage-independent prostaspheres (Figure 2A). The PSA−/lo cell-derived spheres were much larger (Figure 2A, insets) and generated significantly more secondary spheres than the PSA+ cell-originated spheres (Figure 2B). PSA−/lo LNCaP cells also preferentially expressed many SC and developmental genes such as ASCL1, CTED2, GATA6, IGF-1R, KLF5, LRIG1, NKX3.1, and TBX15 (Figure 1H; Figure 2C; Table S2). We employed tetracycline inducible pTRIPZ lentiviral shRNAmir system to knock down three representative SC molecules, i.e., ASCL1 (Jiang et al., 2009), NKX3.1 (Wang X. et al., 2009), and IGF-1R (Chan et al., 1998) (Figure S3I) in PSA−/lo LNCaP cells. Knocking down each of these molecules reduced sphere formation of PSA−/lo LNCaP cells (Figure 2D) without affecting the inherently low sphere-forming activity in PSA+ LNCaP cells (not shown). Furthermore, ASCL1 knockdown significantly inhibited (P<0.05) whereas IGF-1R or NKX3.1 knockdown partially reduced the expansion of PSA−/lo cells caused by androgen-deprivation and etoposide (Figure 2E). These results suggest that at least some of the ‘stemness’ genes over-expressed in the PSA−/lo LNCaP cells are functionally important.
Interestingly, PSA−/lo LNCaP cells, compared to PSA+ cells, overexpressed some (e.g., EED, HDAC4, PHF8) whereas under-expressed other (e.g., DNMT3B, PHF19) chromatin modifiers/epigenetic regulators (Figure S3J; Table S2). The functional significance of these changes in regulating the epigenetic landscape of PSA−/lo PCa cells is currently explored by genome-wide ChIP-Seq analysis.
LNCaP cultures in RPMI-7% FBS contained ~1.4% of GFP−/lo cells with the bulk being GFP+ (Figure S4A). When purified cells were cultured continuously for ~3 weeks, GFP+ LNCaP cells remained all GFP+ (Figure S4B) whereas GFP− cultures became heterogeneous containing 1.8% GFP− cells and ~75% GFP-bright cells (Figure S4C). The 20 GFP+ LNCaP cultures derived from 10 GFP− cells continued to remain all GFP+ after an additional 17-day culture (Figure S4D) whereas the 20 GFP− cultures continued to regenerate both GFP− and GFP+ cells (not shown). Clonal development assays (Patrawala et al., 2005, 2006) revealed that cells in the clones derived from single GFP+ LNCaP cells remained 100% GFP+ at 2 (Figure S4E) and 4 (not shown) weeks. In contrast, single GFP−/lo LNCaP cells developed into 3 distinct types of clones: type I with all cells being GFP+, type II containing both GFP+ and GFP−/lo cells, and type III containing all GFP−/lo cells (Figure S4F–H). Quantitative analysis demonstrated that by 2 weeks, 70–80% of all clones derived from single GFP−/lo LNCaP cells were type I and ~20% were type II whereas the rest were type III (Figure S4, I and J). Type I clones were likely derived from the cells that at the sorting, had already committed to differentiation. Type III clones might all be PSA−/lo cells that underwent symmetric self-renewal based on PCR exclusion of non-infection (Figure 1C). Regardless, the emergence of type II clones indicated that ~20% PSA−/lo LNCaP cells were able to undergo ACD regenerating PSA−/lo and giving rise to PSA+ cells.
Since ACD is the cardinal feature of SCs (Knoblich, 2008), we used time-lapse videomicroscopy to further study the clonal development of PSA+ vs. PSA−/lo LNCaP cells. In agreement with our ‘static’ clonal analysis (above), live imaging of single GFP+ cells showed that the PSA+ LNCaP cells only underwent symmetric division generating clones that contained all PSA+ cells (Figure 2F; Supplemental movie 1). By contrast, single GFP− cells generated type I (Figure 2G; Supplemental movie 2), II (Figure 2H; Supplemental movies 3), and III (Figure 2I; Supplemental movie 4) clones. Approximately 15% of the GFP− LNCaP cells underwent ACD during the first cell division with one daughter cell becoming GFP+ (Figure 2J). Analysis of the end-point clones derived from single GFP− cells showed that 21% and 11% clones were of type II and III, respectively (Figure 2K).
To further explore asymmetric PCa cell division, we examined Numb partition during or right after mitosis. Numb is a Notch antagonist preferentially segregated into the differentiated daughter cells during asymmetric divisions of neuronal, hematopoietic, and muscle SCs (Knoblich, 2008; Wu et al., 2007). We observed that in 242 GFP− LNCaP cells that had just undergone mitosis, 15% of the cells preferentially segregated Numb to the daughter cell that also expressed more PSA (Figure 2L, Figure 3A). In such cells, Numb showed typical cortical concentration (Figure 3A), consistent with its well-established roles in cell polarity and ACD. Using ‘mitotic shake-off’ strategy, we observed similar asymmetric co-segregation of PSA and Numb in one daughter cell in some GFP− LNCaP cells (Figure 3B; a–d) whereas in LNCaP cells that underwent symmetric division, Numb was also equally distributed in both cells (Figure 3B, e–h). Finally, we coinfected LNCaP cells with PSAP-GFP and a Numb-DsRed fusion retroviral reporter. The DsRed+/GFP− LNCaP underwent ACD at 6 h when Numb was partitioned in only one daughter cell and from 24 h, the Numb+ daughter cell also started to express GFP (i.e., PSA; Figure 3C). These observations indicate that a subset of PSA−/lo LNCaP cells can undergo authentic ACD associated with Numb co-segregation into the differentiated PSA+ daughter cells.
We used PSAP-GFP or the modified lentivectors to establish LAPC9 (and LAPC4) ‘reporter’ tumors (Figure S5A). The LAPC4 and LAPC9 xenograft models contain both differentiated and undifferentiated PCa cells and as such, are very useful in elucidating the cellular heterogeneity of PCa (Patrawala et al., 2005, 2006, 2007). Immunostaining using LAPC4 and LAPC9 cells purified from the reporter tumors revealed that most GFP+ cells stained strongly for PSA whereas GFP−/lo tumor cells were generally negative or weak for PSA (Figure S5B–C). Western blotting (Figure 3D) and qPCR (not shown) also revealed lower protein and mRNA levels of PSA and AR in GFP−/lo LAPC9 cells.
In serum-containing medium, PSA−/lo LAPC9 cells initiated spheres that gradually enlarged and expanded and could be passaged for at least 4 generations whereas PSA+ cell-initiated spheres aborted by 20 generation despite that they formed slightly more 1° spheres (Figure 3E; Figure S5D), suggesting that PSA−/lo LAPC9 cells possess high sphere-propagating capacity. When PSA+ and PSA−/lo LAPC9 cells were cultured in medium containing CDSS, PSA−/lo cells formed much more (Figure 3F) and larger (Figure S5E) spheres than PSA+ cells. Interestingly, purified PSA−/lo LAPC4 cells founded more and larger spheres in both serum- (Figure S5F, a–c) and bicalutamide-containing (Figure S5F, d–f) media. Similar to PSA−/lo LNCaP cells, the PSA−/lo LAPC9 cells in the tumors were quiescent as assessed by in vivo BrdU LRC (Figure 3G) and PKH26 dye-retaining (Pece et al., 2010) (Figure S6A) assays. Finally, we infected LAPC9 cells with PSAP-GFP/Pcmv-DsRed (Figure S3A), plated the purified PSA−/lo (i.e., DsRed+/GFP−) cells on fibroblast feeder, and tracked their developmental fates. Although most PSA−/lo LAPC9 cells underwent symmetric cell division (Figure 3H, top), ~5% cells underwent ACD generating PSA+ LAPC9 cells (i.e., DsRed+/GFP+, yellow; Figure 3H, bottom).
Microarray profiling revealed that ~200 genes were over-expressed whereas ~300 genes were under-expressed (F.C ≥1.4, P<0.05) in PSA−/lo LAPC9 cells, which fall into distinct functional categories (Figure 3I, Table S3, Table S4). Most prominently, ~27% of genes (>50) overexpressed in PSA−/lo LAPC9 cells were associated with SCs and development, which included SPP1 (osteopontin or OPN), FGFs, ALDH1A1, integrin α2, c-KIT, Bcl-2, IGF-1, CD44, and Nanog (Figure 3I, top; Table S3, Table S4). Overexpression of some of these molecules was confirmed by Western blotting (Figure 4A) and/or qPCR (Figure S6B). Many of the upregulated genes including Bcl-2, IGF-1, IGFBP3, REG4, and Nanog have been implicated in resistance to androgen deprivation (Jeter et al., 2011). Intriguingly, the PSA−/lo LAPC9 cells overexpressed about 20 neural/glial-related genes (Table S4), suggesting that PSA−/lo cells might be related to or have the ability to generate neuroendocrine-like cells. Finally, many genes preferentially expressed in PSA−/lo LAPC9 cells were shared with those expressed in ESCs or with the genes having either bivalent or H3K27me3 chromatin marks (Figure S6C). The major class of genes upregulated in PSA+ LAPC9 cells (26%) was involved in intermediated metabolism and, interestingly, NumbL, the mammalian homolog of Numb, was overexpressed in PSA+ cells (Figure 3I, bottom; Table S3).
Next, we performed limiting-dilution (LDAs) and serial tumor transplantation assays by monitoring tumor latency, incidence, growth rate, and/or endpoint weight. We first implanted 10,000 each of PSA−/lo (i.e., GFP−/lo) and PSA+ (GFP+) LAPC9 cells subcutaneously in hormonally intact male NOD/SCID mice. Surprisingly, PSA+ LAPC9 cells readily regenerated primary (1°) tumors that were about twice as large as those derived from PSA−/lo cells (Figure 4A; Figure S7A). When we infected LAPC9 cells with PSAP-GFP/Pcmv-DsRed and purified out PSA+ (GFP+DsRed+) and PSA−/lo (GFP−DsRed+) cells for LDAs, the former demonstrated higher tumor-regenerating capacity (Table 1) and developed larger tumors (not shown). Similarly, when PSA+ and PSA−/lo LAPC9 cells were implanted orthotopically in the dorsal prostate (DP), PSA+ cells initiated more (Table 1) and larger (not shown) tumors. The PSA+ LNCaP cells implanted in testosterone-supplemented male NOD/SCID mice also initiated larger tumors (Table 1). These findings suggest that ‘differentiated’ PSA+ PCa cells are, unexpectedly, tumorigenic in androgen-proficient hosts.
Nevertheless, when PSA+ and PSA−/lo LAPC9 cell-derived tumors were serially passaged in intact male mice, PSA−/lo cells maintained relatively constant tumorigenicity whereas PSA+ cells displayed decreasing tumorigenicity (Figure 4A–B; Figure S7A). By 2° generation, tumor weights between the two groups became almost equal and starting from the 3° generation, PSA+ cells generated tumors 2–3 times smaller than PSA−/lo cell-derived tumors (Figure 4B; Figure S7A). Tumor growth rates also showed contrasting patterns – although the 1° PSA+ LAPC9 tumors grew faster than PSA−/lo tumors, starting from the 3° generation, the PSA−/lo tumors grew much faster (not shown). Importantly, although initially there was no significant difference in tumor incidence between the PSA+ and PSA−/lo groups, by the 5° generation tumor incidence was lower for PSA+ cells and, by the 6° generation, tumor incidence was significantly lower (P = 0.006) for PSA+ cells (Figure 4B; Figure S7A). Comparing tumor incidence across PSA+ generations revealed that the 6° tumor incidence was much lower than that in the earlier (i.e., 1° – 4°) generations (P = 0.007; proportion trend test). These observations indicate that PSA−/lo LAPC9 cells are endowed with long-term tumor-propagating capacity in androgen-proficient male hosts.
Similarly, the 1° PSA+ LAPC4 tumors were slightly larger than those derived from PSA−/lo cells but later-generation PSA+ LAPC4 cells regenerated significantly smaller tumors than the corresponding PSA−/lo or early-generation PSA+ cells (Figure S7B). Slightly different from LAPC9, PSA−/lo LAPC4 cells consistently demonstrated higher tumor incidence than PSA+ cells across generations (Figure S7B; Table 1).
Consistent with the PSA−/lo LNCaP and LAPC9 cells being able to undergo ACD generating both PSA−/lo and PSA+ cells whereas PSA+ cells undergoing only symmetric divisions, most tumor cells in PSA+ LNCaP cell-derived tumors in male mice were GFP+/PSA+ whereas tumors derived from PSA−/lo LNCaP cells contained both GFP+/PSA+ and GFP−/PSA− cells (Figure S7C). Likewise, most tumor cells in PSA+ LAPC9 cell-derived tumors serially passaged in male mice were GFP+/PSA+ whereas tumors derived from PSA−/lo cells contained both GFP+/PSA+ and GFP−/PSA− cells (not shown). FACS analysis demonstrated that tumors derived from GFP+ LAPC9 cells contained mostly GFP+ cells whereas tumors derived from PSA−/lo LAPC9 cells contained ~20% GFP−/lo cells with the majority of cells being GFP+ (Figure S7D), indicating that the GFP−/lo PCa cells can undergo self-renewal and recreate the cellular heterogeneity in vivo.
We then implanted purified PSA+ and PSA−/lo LAPC9 cells in castrated male NOD/SCID mice also treated with bicalutamide (50 mg/kg body weight; 3 times/week). In such ‘fully castrated’ mice, PSA−/lo LAPC9 cells developed much larger tumors that grew significantly faster than corresponding PSA+ cells (Figure 4C–D). In female NOD/SCID mice, often used as surrogate androgen-deficient hosts (Klein et al., 1997), PSA−/lo LAPC9 cells similarly initiated larger tumors than PSA+ cells (Figure 4E). Purified PSA−/lo LNCaP cells also regenerated larger and/or more tumors in fully castrated male or female NOD/SCID mice (Table 1). These results suggest that the PSA−/lo PCa cells are more tumorigenic than PSA+ cells in androgen-deficient hosts.
Intriguingly, the PSA−/lo LAPC9 cells did not display significantly higher tumor-initiating frequency whether we utilized PSAP-GFP or PSAP-GFP/Pcmv-DsRed lentivectors to purify PSA+ and PSA−/lo cells (Table 1). We reasoned that the PSA−/lo cell population was still heterogeneous with tumorigenic cells able to initiate CRPC likely representing a minority. cDNA microarray analysis revealed the overexpression of ALDH1A1, integrin α2, and CD44 in PSA−/lo LAPC9 cells (Table S3). ALDH1A1 is the major mediator of Aldefluor phenotype and Aldefluor-hi (i.e., ALDH+) population is enriched in cancer SCs (CSCs) (van den Hoogen et al., 2010) whereas CD44+ PCa cells contain tumor-initiating cells (Patrawala et al., 2006) that can be further enriched by CD44+α2β+ phenotype (Patrawala et al., 2007). Consequently, we purified ALDH+CD44+α2β1+ and ALDH−CD44−α2β1− LAPC9 cells (Figure S7E) from the xenograft tumors maintained in castrated male NOD/SCID mice in which ~90% tumor cells were PSA−/lo and performed serial LDAs in fully castrated mice. Remarkably, ALDH+CD44+α2β1+ cells, in a cell dose-dependent manner, initiated tumor regeneration with as few as 10 cells (Figure 4F; Table 1). In contrast, ALDH−CD44−α2β1− cells only regenerated 1 tumor (out of 22 injections) at the highest cell number (Figure 4F), which likely resulted from cell impurity. Similar differences in tumorigenicity were observed between the two populations in the 2° transplantations (Table 1). The abundance of ALDH+CD44+α2β1+ cells was higher in castrate tumors than tumors in intact male mice and was maintained during serial transplantations (Figure 4G; data not shown), indicating the self-renewal of these cells in vivo. Combined, these results suggest that the ALDH+CD44+α2β1+ phenotype in PSA−/lo population further enriches CRPC cells.
To determine what molecules might be involved in determining the tumorigenicity of PSA−/lo PCa cells, we again resorted to our microarray data, which identified increased expression of Nanog, CD44, and OPN, among many others. Overexpression of Nanog, CD44, and OPN was confirmed by qPCR in independently purified PSA−/lo LAPC9 and other PCa cells (Figure S6B; data not shown). We therefore infected PSA−/lo LAPC9 cells with lentivectors encoding shRNA for Nanog (Jeter et al., 2009), OPN, or CD44 (Liu et al., 2011). Knockdown of OPN, CD44, or Nanog (Figure 4H–I) inhibited tumor regeneration of PSA−/lo LAPC9 cells in fully castrated hosts, consistent with our recent findings that CD44 knockdown inhibits PCa metastasis (Liu et al., 2011) and that Nanog overexpression promotes CSC properties and PCa cell resistance to androgen deprivation (Jeter et al., 2011).
We carried out an ADT experiment (Yoshida et al., 2005) to determine whether PSA−/lo PCa cell-derived tumors resist androgen ablation in vivo. We purified PSA+ and PSA−/lo LAPC9 cells and injected them in intact male mice. When tumors became palpable, mice were castrated and also treated with bicalutamide. PSA−/lo cell-derived tumors grew much better (Figure 5A) and larger (Figure 5B) in androgen-depleted hosts than PSA+ cell-derived tumors. We further attempted to mimic the clinical scenario by correlating % GFP+ (PSA+) cells during castration with biochemical (PSA) failure and tumor recurrence (re-growth). When the group of animals bearing LAPC9 tumors was castrated and concomitantly treated with bicalutamide at week 5, tumor growth plateaued, serum PSA levels dipped, and the % GFP+ cells declined by week 6 (Figure 5C). However, by week 8, despite continued decrease in GFP+ cells (Figure 5C, right), tumor growth resumed (Figure 5C, left, inset) and serum PSA rebounded (Figure 5C, middle, inset), signaling biochemical recurrence (BCR) and tumor recurrence. These observations were remarkably similar to what was observed in PCa patients undergoing ADT (Ryan et al., 2006) and provide evidence that androgen ablation enriches PSA−/lo PCa cells.
Are the preceding findings in PCa models (LNCaP, LAPC9, and LAPC4) applicable to patient tumors? Strikingly, low levels of tumor PSA mRNA correlated with reduced BCR-free and overall patient survival (Figure 7A). We purified HPCa cells from (untreated) prostatectomy specimens, infected them with PSAP-GFP, separated PSA+ and PSA−/lo cells, and performed clonal and sphere assays in serum/androgen-free medium (Jeter et al., 2009; Liu et al., 2011). The results from 3 HPCa samples showed that PSA−/lo cells did not express AR protein (not shown) and possessed significantly higher clonal and sphere-forming capacities than corresponding PSA+ cells (Figure 7B–D; Figure S8A). Importantly, we observed clonal development patterns in HPCa cells similar to those observed in LNCaP cells. For instance, most PSA+ HPCa12 cell-derived clones were GFP+ whereas the PSA−/lo cell-derived holoclones contained GFP−/lo as well as GFP+ cells (Figure 7E). Similar type II clones were observed in PSA−/lo cells plated on collagen (Figure 7F) and some PSA−/lo HPCa cells also underwent ACD (Figure 7G). Microarray analysis in 4 pairs of purified PSA−/lo and PSA+ HPCa cells revealed preferential expression of many SC/developmental genes in PSA−/lo HPCa cells (Table S5).
Using one of the early-generation (4°) HPCa xenografts, i.e., HPCa58 (Liu et al., 2011), we established reporter tumors similarly to LAPC9 and LAPC4. The reporter tumor was green (Figure 7H) and expressed PSA mRNA (Figure 7I). PSA immunostaining revealed a good correlation between GFP and PSA positivity (Figure 7J). When PSA+ and PSA−/lo HPCa58 cells were used in sphere assays, the PSA−/lo cells demonstrated higher sphere-forming capacity in both androgen-supplemented (Figure S8B) and androgen-ablated (Figure S8C) conditions. Serial transplantations in male NOD/SCID mice revealed that PSA+ HPCa58 cells initiated larger tumors than the corresponding PSA−/lo cells in the first generation; however, upon passaging, PSA−/lo HPCa58 cells developed larger tumors than the corresponding PSA+ cells (Figure 7K). Finally, when equal numbers (10,000) of PSA+ and PSA−/lo HPCa58 cells were implanted in castrated male NOD/SCID mice treated with bicalutamide, PSA−/lo cells generated larger and more tumors (Figure S8D). Experiments with another HPCa reporter tumor, i.e., HPCa80, revealed that the PSA−/lo HPCa80 cells generated larger tumors than PSA+ HPCa80 cells (Figure S8E).
PSA is normally expressed and secreted by prostate luminal cells and represents one of the best-characterized organ-specific differentiation markers. Early studies have shown that PSA protein expression in PCa positively correlates with its degree of differentiation and that both untreated PCa and CRPC contain PSA+ and PSA−/lo cancer cells. Our own analysis of ~45 patient tumors confirms the two populations of PCa cells and, importantly, demonstrates that the abundance of PSA−/lo PCa cells is enriched in high-grade and treatment-failed tumors. PSA protein is also reduced or lacking in metastases (Varambally et al., 2005). Strikingly, lower tumor PSA mRNA levels positively correlate with worse clinical outcomes including high tumor grade, LN positivity, metastasis, recurrence, and reduced patient survival. The association of PSA−/lo PCa cells and tumor PSA mRNA/protein with poor clinical features is opposite to the positive correlation between serum PSA and the same clinical parameters. Elevated serum PSA levels in advanced PCa may be due to increased access of PCa cells to bloodstream and/or related to increased tumor mass in which PSA−/lo PCa cells can differentiate into PSA+ cells.
PSA has been thought to be strictly regulated by AR. In clinical samples, however, AR and PSA protein expression is often discordant and heterogeneous with some PCa cells showing little expression of either molecule (Hobisch et al., 1995; Mostaghel et al., 2007; Ruizeveld de Winter et al., 1994; Shah et al., 2004). Discordant AR and PSA expression is also reflected at the mRNA levels in individual primary, hormone-refractory and recurrent tumors as well as in metastases (Figure S2B; unpublished observations). The discordant expression patterns of PSA and AR suggest that PSA expression can be regulated in an AR-independent manner (Hsieh et al., 1993) and that prostate tumors contain AR+/PSA+, AR+/PSA−, AR−/PSA+, and AR−/PSA− PCa cells.
The PSA+ PCa cells isolated based on our reporter systems mostly show strong nuclear AR whereas PSA−/lo population contains both AR− and AR+ cells. Consequently, PSA+ cells resemble AR+/PSA+ cells whereas PSA−/lo cells contain both AR+/PSA− and AR−/PSA− PCa cells. AR expression is sometimes upregulated in advanced and recurrent tumors, which we surmise could be related to the expansion of AR+/PSA− PCa cells. Future work that permits fractionation of AR+/PSA− and AR−/PSA− PCa cells should allow us to directly address this postulate. It should be noted that AR possesses PCa-suppressive functions (Niu et al., 2008), AR signaling is attenuated in some advanced PCa (Tomlins et al., 2007), AR is significantly reduced and only detectable in ~40% PCa cells in hormone-refractory metastases (Davis et al., 2006), and AR requirement in PCa may be context dependent (Memarzadeh et al, 2011).
PSA−/lo PCa cells possess high clonogenic capacity, survive better in androgen-deficient conditions, and are refractory to not only androgen deprivation but also drugs. PSA−/lo PCa cells are quiescent, which could partly explain their resistance to various stresses. Importantly, a fraction of PSA−/lo PCa (~15–20% PSA−/lo LNCaP and 5% PSA−/lo LAPC9) cells can undergo authentic ACD, a cardinal feature of SCs. In contrast, PSA+ cells undergo mainly symmetric divisions. The distinct division patterns between PSA+ and PSA−/lo cells overall are mirrored in the respective tumors they regenerate – although the PSA+ cell-derived tumors contain mostly PSA+ cells, the PSA−/lo cell-originated tumors contain both PSA−/lo and PSA+ cells.
It is presently unclear how PSA−/lo and PSA+ cells, both of which are maintained under identical conditions, embark on different developmental fates. Nevertheless, the distinct division modes of PSA−/lo and PSA+ cells reinforce their intrinsic biological differences. Significantly, the PSA+, differentiated daughter cell derived from asymmetric division of a PSA− PCa cell also preferentially ‘inherits’ Numb, one of the best studied cell fate determinants known to be asymmetrically segregated into differentiated daughter cells (Knoblich, 2008). It is interesting that asymmetric segregation of Numb precedes that of PSA (Figure 3J), raising the possibility that Notch signaling may regulate PCa cell ACD.
PSA−/lo LNCaP and LAPC9 cells preferentially express dozens of genes associated with development and SC functions. These SC-associated molecules are functionally important as demonstrated for ASCL-1, IGF-1, and NKX3.1 in LNCaP cells and Nanog, CD44, and OPN in LAPC9 cells. The PSA−/lo LNCaP and LAPC9 cells commonly overexpress hundreds of genes (e.g., BCL2, IGF1, SOX15, BMPR1B, TGFBR1, etc), which fall into distinct GO categories including SC, development, stress response, and wound healing (unpublished observations).
The PSA−/lo LNCaP and LAPC9 cells do express ‘unique’ gene categories. Thus, PSA−/lo LNCaP cells prominently under-express genes associated with cell-cycle progression and mitosis. In contrast, the PSA−/lo LAPC9 cells overexpress hundreds of signaling molecules whereas under-express genes associated with intermediate metabolism. The observations that PSA−/lo LNCaP cells under-express cell-cycle and mitosis associated genes and that PSA−/lo LAPC9 cells under-express metabolism-associated genes are consistent with the PSA−/lo PCa cells being more quiescent. Intriguingly, PSA−/lo and PSA+ LAPC9 cells frequently exhibit reciprocal gene expression patterns (Table S4), suggesting that the two populations of PCa cells may cross talk and reciprocally regulate each other in a ‘paracrine’ fashion, as hinted by emerging data in other tumor systems (Tang, 2012).
Tumor transplantation experiments in NOD/SCID mice (~2,000 used) reveal that although the tumor-propagating capacities of PSA−/lo PCa cells are maintained across the generations in hormonally intact male mice, the tumor-regenerating ability of the corresponding PSA+ PCa cells gradually declines, suggesting that PSA−/lo cells possess long-term tumor-propagating capacity. The PSA−/lo cell-regenerated tumors recreate the original tumor heterogeneity containing both PSA−/lo and PSA+ cells. That PSA+ cells serially transplanted in androgen-proficient hosts manifest diminishing tumorigenic potential strongly suggests that these cells intrinsically possess more limited self-renewal ability compared to PSA−/lo PCa cells. The unexpected observations that PSA+ cells, at the first generation, often demonstrate higher tumorigenic potential than the isogenic PSA−/lo cells caution us to be careful when using tumor regeneration as a yardstick of measuring CSC properties. Preferably, serial transplantation assays should be performed – otherwise misleading or even opposing/contradictory conclusions may be reached.
When transplanted in androgen-deficient hosts, PSA−/lo PCa cells initiate much larger and faster growing tumors than isogenic PSA+ cells. Taken together, the biological, molecular and tumorigenic properties of PSA−/lo cells presented herein, coupled with earlier reports on several prostate CSC populations (e.g., Collins et al., 2005; Huss et al., 2005; Maitland et al., 2011; Patrawala et al., 2006; Rajasekhar et al., 2011), suggest that the PSA−/lo cell population may represent a tumorigenic pool that harbors several subsets of stem-like cancer cells. First, CD133+α2β1hiCD44+ primary PCa cells (Collins et al., 2005), ABCG2+ PCa cells in situ (Huss et al., 2006), and Lin−CD44+ PCa cells in xenografts (Patrawala et al., 2006) all seem to express low levels of AR and lack PSA, suggesting that these PCa cell subsets may overlap with each other and are all harbored in PSA−/lo population. Second, unbiased whole-genome transcriptome analysis reveals preferential expression of CD44, integrin α2, and ALDH1A1 in PSA−/lo LAPC9 cells. Third, prospectively purified ALDH+CD44+α2β1+ subpopulation in PSA−/lo cells greatly enriches for more tumorigenic, castration-resistant PCa cells. Finally, CD44+ PCa cells freshly purified from a dozen untreated primary tumors express much lower levels of PSA mRNAs than the corresponding CD44− PCa cells (Liu et al., unpublished observations). Future work will further elucidate the interrelationship between various subsets of tumorigenic cells and characterize PSA−/lo PCa cells with respect to their relationship with luminal and basal cells.
One of the most significant contributions of the present work is to provide direct experimental evidence that PSA−/lo PCa cells may represent an important source of CRPC cells. First, PSA−/lo cells, in vitro, survive androgen deprivation, resist drug/stress treatments, and robustly found holoclones and self-renewing spheres. Second, when both PSA+ and PSA−/lo cells are implanted in male mice that are subsequently subjected to ADT, the PSA−/lo cell-derived tumors are refractory to castration and continue to develop. Third, androgen deprivation greatly enriches the PSA−/lo cells, which could initiate robust tumor development in castrated hosts. These findings closely resemble the AI progression observed in patients and mirror the observed reduction in PSA-producing cells in patient tumors upon androgen depletion (Ryan et al., 2006). We have provided the first prospective evidence that PSA−/lo PCa cells, which pre-exist in the tumors, are molecularly and functionally distinct from the differentiated counterparts.
We have shown that under normal (i.e., androgen-proficient) conditions, undifferentiated PSA−/lo cells harbor self-renewing CSCs and likely represent one important source of CRPC cells. Future work will address whether under other conditions such as persistent castrations, PSA+ PCa cells may manifest increased plasticity by undergoing de-differentiation, as shown by emerging data in other tumors (Tang, 2012). Altogether, our results suggest that novel therapeutics targeting PSA−/lo cells should be developed and used in conjunction with ADT in order to eradicate all PCa cells and prevent recurrence.
Detailed methods are available online in Supplemental Experimental Procedures (SEP).
GFP+ and GFP− PCa cells were sorted out by FACS from 1° tumors originally derived from GFP+ and GFP− cells, respectively, and implanted s.c to generate 2° tumors in intact male mice. Sequential tumor transplantation was performed using similar strategies by following that GFP+ cells were always purified from tumors originated from purified GFP+ cells whereas GFP− cells were from tumors derived initially from GFP− cells. For tumor experiments in castrated mice, male NOD/SCID mice (6–8 weeks) were surgically castrated 1–2 weeks prior to injection. GFP+/GFP− PCa cells were purified out from reporter tumors and injected s.c into the castrated mice, which also received i.p injections of bicalutamide.
For ADT, GFP+ and GFP− LAPC9 cells were purified out from AD reporter tumors and injected s.c in intact male NOD/SCID mice. When tumors reached ~60 mm3, mice were surgically castrated and treated with bicalutamide. Tumor growth was followed by caliper measurement and volumes of individual tumor were normalized to those on day 0 (day of castration). For ‘recurrence’ experiments, unsorted LAPC9 cells from AD reporter tumors were injected s.c in intact male NOD/SCID mice. Starting from the 4th week, tumor volumes (mm3) were measured using a digital caliper, blood samples (100–200 μl/mouse) were collected from each animal via saphenous vein for serum PSA measurement (ng/ml) and 2–3 tumors were harvested to determine by FACS the % of GFP+ cells in individual tumors on weekly basis. For tumor volumes and serum PSA, the values were presented as fold increases over those from the fourth week. At the fifth week, animals were randomly divided into the control group, in which the animals were mock-castrated, and the castrate group, in which the animals were surgically castrated and also treated with bicalutamide.
Time-lapse fluorescence videomicroscopy was performed using Nikon Biostation Timelapse system (Liu et al., 2011) as described in SEP.
Basic procedures have been described (Bhatia et al., 2008). Total RNA was extracted from pooled purified GFP+ or GFP− LNCaP and LAPC9 cells and microarray experiments were performed in triplicates using the 44 K 60-mer “Human Whole Genome Oligo Microarray Kit from Agilent (Agilent Technologies, Santa Clara, CA) with 500 ng of total RNA. For details, please refer to SEP.
We thank K. Claypool and P. Whitney for FACS, Histology Core for IHC, R. Fagin for HPCa samples, J. Shen and J. Repass for qPCR, L. Shen and S. Tsavachidis for initial microarray analysis, and other members of the Tang lab for helpful discussions. This work was supported in part by grants from NIH (R01-ES015888 and 1R21CA 150009), Department of Defense (W81XWH-11-1-0331), CPRIT (RP120380), Elsa Pardee Foundation, and MDACC UCF, Center for Cancer Epigenetics, and Laura & John Arnold Foundation RNA Center pilot grant (all to D.G.T) and by two Center Grants (CCSG-5 P30 CA166672 and ES007784). We apologize to the colleagues whose work was not cited due to space constraint.
Supplemental Information includes Supplemental Results, eight figures, four tables, six movies, Extended Experimental Procedures, and Supplemental References and can be found with this article online.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.