When a PCa patient is diagnosed, his or her concern is not where the tumor came from (i.e., from which cell type). Of most concern is how bad the tumor is, what’s the prognosis after surgery and/or radiation and other therapies, and whether the tumor has already spread outside the prostate. For advanced PCa patients, androgen-deprivation therapy (ADT) is the mainstay; however, most treated patients become refractory to ADT and show signs of recurrence within 6-20 months. For those patients with recurrent PCa, there is NO effective and specific therapeutics. If all PCa cells homogeneously express AR, why would some PCa cells escape ADT? After impressive reduction in primary tumor burden, what cells give rise to recurrent, castration-resistant PCa? If we can put our fingers on such cells, can we develop novel therapeutics to specifically target them? These are the fundamental questions our lab has been focusing on, mostly from the perspective of CSCs generating cellular heterogeneity, conferring therapy resistance, and mediating castration-resistant recurrence.
We started our CSC work early 2002, just after the CSC concept was revived in 2001. We primarily utilized PCa as the model and first adopted the marker-independent side population (SP) strategy (Cancer Res., 65, 6207-6219, 2005). We find that the SP, purified from glioma and breast cancer (MCF7) cells, is enriched in tumor-initiating cells. Importantly, limiting-dilution assays (LDAs) of tumor transplantations in NOD/SCID mice have determined that the tumor-initiating frequencies (TIFs) of SP and non-SP cells in LAPC9 xenograft prostate tumors are 1/513 and 1/273,072, respectively, indicating a 500 fold enrichment of tumor-initiating cells in the SP. Importantly, the LAPC9 tumors derived from SP cells can be serially transplanted and contain SP cells as the minority with most cells being non-SP, suggesting that the SP cells can self-renew in vivo regenerating both SP and non-SP cells. These results provided the first piece of in vivo data that human PCa contains stem-like cells. Compared to the SP, cancer cells that express ABCG2 (i.e., ABCG2+ cells), the multi-drug resistant protein that mediates the SP phenotype in HSCs, either do not display increased tumor-initiating capacity or exhibit only ~10-fold enrichment in tumorigenicity over the ABCG2- cells, suggesting that ABCG2 is only partially involved in mediating the SP phenotype in (prostate) cancer cells.
A major technical drawback in the SP strategy is the lack of appropriate control for non-SP cells, which chronically accumulate the potentially toxic Hoechst 33342 dye. Another deficiency in the SP strategy is that among nearly a dozen PCa cells analyzed we could only reliably detect the SP in LAPC9 xenograft tumors. Consequently, in 2003, we started using, as a marker, cell surface adhesion molecule CD44, which is highly expressed in multiple normal stem/progenitor cells and in the basal-cell layer of human prostatic glands (Figure 2). We utilized FACS to purify CD44-high (i.e., CD44+) and CD44-low (i.e., CD44-) PCa cells from both cultured cells (Du145, PC3, NHP/LT, etc) and xenograft tumors (Du145, LAPC4, and LAPC9) and then compared their biological properties in vitro and tumorigenic potential in vivo. Tumor transplantation LDAs have revealed a 6-30 fold enrichment in tumor-initiating capacity in CD44+ vs. CD44- PCa cells. Importantly, most metastatic activity resides in CD44+ cells. Other evidence also supports that the CD44+ PCa cell population HARBORS stem-like cells. First, at the single-cell level, CD44+ PCa cells possess substantial clonal/clonogenic capacities. Second, CD44+ PCa cells are largely quiescent but possess great proliferative potential. BrdU pulse-chase experiments reveal that many CD44+ PCa cells are long-term or intermediate label-retaining cells (LRCs). Third, CD44+ PCa cells express low levels of AR but can clonally regenerate ARhiPSA+ ‘differentiated’ cells. Fourth, some CD44+ PCa cells are able to undergo asymmetric cell division (ACD) generating two daughter cells with only one retaining CD44 expression. ACD is a DEFINING characteristic of stem cells (Oncogene 25, 1696-1708, 2006).
Along with our efforts in dissecting the tumorigenic heterogeneity of PCa cells, we proposed functional (and practical) criteria that can be used to define potential prostate CSCs. First, the presumptive CSC must be prospectively purified. In other words, simple marker expression on IHC analysis does not indicate CSCs. Second, in vivo tumorigenicity experiments by LDAs must be performed to show that such cell populations are enriched in tumor-initiating cells. When feasible, serial tumor transplantation should be carried out to determine whether the tumors derived from the putative CSCs can be transplanted for multiple generations. Histologically, the reconstituted as well as serially transplanted tumors should resemble the original patient tumor. Third, importantly, the presumptive CSC population, or a subpopulation within, has to be studied to show that they possess certain intrinsic biological properties normally associated with SC discussed above. Only when these conditions are fulfilled can one confidently claim that the candidate population of tumor cells under investigation is enriched in potential CSCs (Mol. Carcinogenesis, 46, 1-14, 2007). Throughout our studies, we have adhered to these stringent criteria in claiming CSC enrichment. For CSC definitions, please also see Qin Tang (2008) and Tang (2009). For methodologies in assays prostate CSCs that we have been using, refer to Li et al., MMB, 2009.
The reported leukemia and glioma stem cell pool is heterogeneous harboring subpopulations of cells with differential tumorigenic potential. We also realized that the CD44+ PCa cell subpopulation, despite highly enriched in tumorigenic cells, does not ‘net’ all tumor-initiating cells. We then utilized a double marker purification strategy and demonstrated that the CD44+alpha2beta1+ PCa (LAPC9 and Du145) cell population possesses even higher tumorigenic potential with 100 cells being able to regenerate tumors (Cancer Res. 67, 6796-6805, 2007). Phenotypic characterizations indicate that all ABCG2+, and most SP (~98%) or alpha2beta1+ (~65%) PCa cells are CD44+, suggesting that the CD44+ PCa cell population is HETEROGENEOUS harboring both primitive CSCs and more mature progenitors.
PC3 cells represent the most aggressive human PCa cell type and the above-mentioned strategies and markers including SP, CD44, alpha2beta1, and ABCG2 fail to delineate tumorigenic differences among subsets of PC3 cells. However, using traditional clonal analysis combined with limiting-dilution serial tumor transplantation assays, we have shown that the PC3 cell holoclones contain self-renewing tumorigenic cells, most of which are CD44+, alpha2beta1hi and b -catenin+ (Cancer Res. 68, 1820-1825, 2008).
These studies in xenograft prostate tumors illustrate several critical points. FIRST, PCa cells are not all equal and certain subsets of PCa cells are clearly more tumorigenic than the others. SECOND, the tumorigenic PCa cell pool is still heterogeneous and likely contains other minor populations of tumorigenic cells expressing other markers such as ALDH, integrin alpha6, Trop2, etc. To a certain degree, the process of CSC enrichment is analogous to biochemical fractionations to enrich for an enzymatic activity – in principle one can combine more markers to obtain a more enriched population of tumorigenic cells. This point explains, potentially, why different research groups may report (prostate) CSCs with quite different phenotypes. THIRD, the pool of prostate CSCs may vary from patient to patient and depending on whether the patients have been treated or not. FOURTH, the model also predicts that primitive prostate CSCs, which express low levels of AR, may not respond well to ADT. Consequently, when ADT has eradicated most ARhi/PSA+ PCa cells, prostate CSCs might regenerate castration-resistant tumors. For more discussions on prostate CSCs, see: Patrawala et al (2008) and Honorio et al. (2009).
A key unanswered question is whether PCa cells in patient tumors are also organized in a similar tumorigenic hierarchy to that in xenograft tumors we have studied. Though we have accumulated evidence suggesting that basal-like PCa cells from patient tumors have certain CSC properties (e.g., enhanced clonogenic potential), this important question has so far dodged an answer mainly due to technical hurdles. Most other CSCs reported including, e.g., those from colon, pancreatic, brain, ovarian, head and neck, and liver cancers, upon marker-based flow sorting, are generally injected in Matrigel directly (i.e., without co-injecting with any other supporting cells) into the orthotopic sites or subcutaneously (s.c) in NOD/SCID mice and then differential tumor development is observed. In sharp contrast, we have observed that when unsorted, or marker-sorted PCa cells derived from primary patient tumors are directly implanted, in 50% Matrigel, in several different sites (s.c, kidney capsule, dorsal prostate) in NOD/SCID mice, there has been no reliable tumor regeneration. In fact, even when we ‘recombined’ primary PCa cells with ‘helper’ cells such as urogenital sinus mesenchyme (UGSM) cells or carcinoma-associated fibroblasts (CAFs), we still fail to achieve reproducible tumor reconstitution (Li et al., MMB, 2009). These results highlight unique ‘microenvironmental’ or niche requirement for prostate cancer (stem) cells, which presumably is very difficult to reconstitute in mice.
Without overcoming this technical hurdle, it will be difficult to test the CSC hypothesis in primary prostate tumors. Our lab has been making conscientious efforts towards this goal and has studied, since Feb. of 2005, 120 primary patient tumors (i.e., HPCa). We have successfully established several early-generation xenograft tumors and have made important progress in tumor reconstitution using single cells derived from primary HPCa samples. While we are developing assays that can allow reliable and reproducible reconstitution of human PCa in mice with similar overall histopathology, we shall continue to utilize various xenograft models to address other important questions, e.g., Might PCSCs also be involved in mediating metastasis?