Recently, several human cancers including leukemia and breast and brain tumors were found to contain stem-like cancer cells called cancer stem cells (CSC). Most of these CSCs were identified using markers that identify putative normal stem cells. In some cases, stem-like cancer cells were identified using the flow cytometry-based side population technique. In this study, we first show that ∼30% of cultured human cancer cells and xenograft tumors examined (∼30 in total) possess a detectable side population. Purified side population cells from two cell lines (U373 glioma and MCF7 breast cancer) and a xenograft prostate tumor (LAPC-9) are more tumorigenic than the corresponding non–side population cells. These side population cells also possess some intrinsic stem cell properties as they generate non–side population cells in vivo, can be further transplanted, and preferentially express some “stemness” genes, including Notch-1 and β-catenin. Because the side population phenotype is mainly mediated by ABCG2, an ATP-binding cassette half-transporter associated with multidrug resistance, we subsequently studied ABCG2+ and ABCG2− cancer cells with respect to their tumorigenicity in vivo. Although side population cells show increased ABCG2 mRNA expression relative to the non–side population cells and all cancer cells and xenograft tumors examined express ABCG2 in a small fraction (0.5-3%) of the cells, highly purified ABCG2+ cancer cells, surprisingly, have very similar tumorigenicity to the ABCG2− cancer cells. Mechanistic studies indicate that ABCG2 expression is associated with proliferation and ABCG2+ cancer cells can generate ABCG2− cells. However, ABCG2− cancer cells can also generate ABCG2+ cells. Furthermore, the ABCG2− cancer cells form more and larger clones in the long-term clonal analyses and the ABCG2− population preferentially expresses several “stemness” genes. Taken together, our results suggest that (a) the side population is enriched with tumorigenic stem-like cancer cells, (b) ABCG2 expression identifies mainly fast-cycling tumor progenitors, and (c) the ABCG2− population contains primitive stem-like cancer cells.
- side population
- stem cells
- progenitor cells
- cancer stem cells
- stemness genes
Stem cells, which have now been found in multiple adult tissues and organs, have several fundamental properties. First, they are generally very rare. For example, the long-term hematopoietic stem cells (LT-HSC) in mouse bone marrow constitute ∼0.02% and the short-term HCSs (ST-HSC) ∼0.1% of the total cells ( 1). Second, stem cells in their normal microenvironment (i.e., niche) rarely divide, although they possess tremendous proliferative potential ( 2). Third, stem cells can self-renew; that is, they can regenerate themselves when they divide to give rise to progenitor cells ( 2). Fourth, stem cells possess multipotential, oligopotential, or unipotential differentiation ability ( 3). Many adult stem cells also seem to have the ability to trans-differentiate into other cell types, although this phenomenon is still being hotly debated ( 3, 4). Finally, stem cells may express unique markers or properties that can allow their enrichment and identification.
Indeed, many stem cells are identified by their expression of unique markers. For example, mouse LT-HSCs, ST-HSCs, and multipotent progenitors can be identified and separated by their marker phenotypes, c-kithiSca-1hiThy1.1loLin−/loFlk−, c-kithiSca-1hiThy1.1loLin−/loFlk+, and c-kithiSca-1hiThy1.1−Lin−/loFlk+, respectively ( 1). Unfortunately, most organ-restricted stem cells or progenitors lack unique and specific markers. One way to identify them relies on the slow-cycling properties of stem cells. When pulsed by bromodeoxyuridine (BrdU) for a period of time followed by “chasing” for longer time intervals, slow-dividing stem cells will be identified as label-retaining cells (LRC; refs. 2, 5). The LRCs purified from the stem cell niche (i.e., the bulge in the hair follicle) in transgenic animals possess many of the expected stem cell properties ( 2, 5). Another way to identify putative adult stem cells was pioneered by Goodell et al. ( 6), who observed that when bone marrow–derived cells are incubated with Hoechst dye 33342 and then analyzed by dual-wavelength flow cytometry, a small population of cells does not accumulate an appreciable amount of dye and is thus identified as a Hoechstlo side population. Remarkably, the side population is highly enriched in HSCs ( 6). Since its initial application in bone marrow HSCs, the side population technique has been adapted to identify putative stem cells and progenitors in multiple tissues/organs including umbilical cord blood ( 7), skeletal muscle ( 8– 10), mammary glands ( 11, 12), lung ( 13– 15), liver ( 16), epidermis ( 17, 18), forebrain ( 19), testis ( 20, 21), heart ( 22), kidney ( 23), limbal epithelium ( 24), and prostate ( 25).
The side population–enriched stem cells are rare (∼0.01-5%; refs. 18, 23, 26) and heterogeneous, varying with tissue types, stages of development, and methods of preparation ( 10, 27, 28). For example, the bone marrow side population cells contain not only HSCs but also mesenchymal stem cells ( 29) and do not capture all HSCs ( 30) but only a subset of long-term repopulating HSCs ( 31). The skeletal muscle side population cells are composed of mostly bone marrow–derived cells ( 8, 9) with only a minor population of resident muscle stem/progenitor cells (i.e., satellite cells). The lung side population cells are also heterogeneous comprising both CD45+ (i.e., bone marrow derived) and CD45− cells ( 13– 15). Similarly, the testis side population cells may be enriched in spermatogonial, germinal, as well as mesenchymal (i.e., Leydig) stem cells ( 20, 21). Although the side population cells in most cases seem enriched in primitive stem cells, there are also reports that suggest that the side population cells do not identify stem cells ( 17, 18, 32).
The side population phenotype is mediated by the ABC family of transporter proteins. One of the major mediators seems to be ABCG2 or BCRP ( 33), which was initially identified in drug-selected MCF7 breast cancer cells and later found to efflux multiple chemotherapeutic drugs and xenobiotics ( 34). The strongest evidence linking ABCG2 and the side population phenotype comes from the nearly complete loss of the bone marrow side population phenotype in abcg−/− mice ( 35). Other supporting evidence is that side population cells preferentially express ABCG2 ( 13, 16, 18, 20, 22, 24, 36, 37) and that ABCG2 expression is detected in known stem/progenitor cells such as HSCs ( 33), nestin-positive islet-derived progenitors ( 36), hepatic oval cells ( 16), limbal basal cells ( 24), and neural stem cell ( 38). On the other hand, it should be noted that only a fraction of the side population cells expresses ABCG2 (e.g., ref. 24) and that both side population and known stem/progenitor cells also express other ABC transporters such as MDR-1 (i.e., ABCB1 or P-glycoprotein), MRP-1 (ABCC1), and ABCA2 ( 36, 37, 39), suggesting that these latter molecules may also be involved in mediating the side population phenotype. In support, enforced expression of MDR-1 in murine bone marrow cells leads to the expansion of the side population ( 40).
Recently, several human cancers including leukemia ( 1, 41) and breast ( 42) and brain ( 37, 43– 46) tumors were found to contain stem-like cells (SLC) called cancer stem cell (CSCs; ref. 47). Most of these tumorigenic SLCs were identified using markers that identify putative normal stem cells. Interestingly, SLCs have also been identified in immortalized cell lines ( 39), long-term cultured cancer cells ( 37, 48), or patient tumor samples ( 37) using the side population technique. These observations suggest that even long-term cultured (cancer) cells may retain SLCs and that the side population approach represents a valid marker-independent method to identify such cells. In a recent study, 3 we have also provided independent evidence for the existence of SLCs in multiple human tumor cell cultures as well as xenograft tumors. In the current study, we first seek to confirm the utility of the side population technique to identify tumorigenic SLCs in cultured human cancer cells and xenograft tumors. Then we focus on the question whether the higher tumorigenicity associated with the side population might be related to the expression of ABCG2, a major mediator of the side population phenotype in bone marrow cells. Our results surprisingly show that in contrast to the tumorigenic differences between the side population and non–side population cells, the ABCG2+ and ABCG2− cancer cells show very similar tumorigenicities in vivo.
Materials and Methods
Cells, reagents, and animals. Various cancer cell lines including those of prostate (PPC-1, Du145, LNCaP, and PC3 cells), breast (T47D, MCF-7, MDA-MB231, MDA-MB435, and MDA-MB468), colon (COLO 320, DLD-1, RKO, and HCT116), bladder (RT4, UC14, UC1, and UC3), glioma (D54, U87, U251, and U373), melanoma (WM266-4 and WM562-4), cervix (HeLa), and ovary (SKOV-3) were obtained from American Type Culture Collection (Manassas, MA) or collaborators and cultured in the recommended medium containing 5% to 10% of heat-inactivated fetal bovine serum (FBS). Xenograft human prostate tumors LAPC-4 and LAPC-9 were kindly provided by C. Sawyers (Department of Medicine, University of California, Los Angeles, CA) ( 49) and maintained in nonobese diabetic/severe combined immunodeficiency (NOD/SCID) mice. Du145 xenograft tumors were established in our lab using early-passage cells and maintained in NOD/SCID mice. All animals were obtained from The Jackson Laboratory (Bar Harbor, ME) and maintained in standard conditions according to the Institutional guidelines. The monoclonal anti-ABCG2, FITC-, and PE-conjugated anti-ABCG2 monoclonal antibodies, isotype control antibody, and secondary antibodies were all obtained from Chemicon (Temecula, CA). Monoclonal antibodies against Ber-EP4 and CD31 were obtained from DakoCytomation, Inc. (Carpinteria, CA) and BD PharMingen (San Diego, CA), respectively. All chemicals were obtained from Sigma (St. Louis, MO) unless specified otherwise.
Side population analysis. The protocol was based on Goodell et al. ( 6) with slight modifications. Briefly, cells (1 × 106 cells/mL) were incubated in prewarmed DMEM/5% FBS containing freshly added Hoechst 33342 (5 μg/mL final concentration) for 90 minutes at 37°C with intermittent mixing. In some experiments, cells were incubated with the Hoechst dye in the presence of verapamil (50 μmol/L) or reserpine (100 μmol/L). At the end of incubation, cells were spun down in the cold and resuspended in ice-cold PBS. 7-AAD (2 μg/mL final concentration) was added for 5 minutes before fluorescence-activated cell sorting (FACS) analysis, which allows for the discrimination of dead versus live cells. As positive controls, we used HL60 promyelocytic leukemia cells selected by chronic exposure to doxorubicin. Samples were analyzed on a Coulter Epics flow cytometer. The Hoechst dye was excited with the UV laser at 351 to 364 nm and its fluorescence measured with a 515-nm side population filter (Hoechst blue) and a 608 EFLP optical filter (Hoechst red). A 540 DSP filter was used to separate the emission wavelengths.
Indirect immunofluorescence, flow cytometry analysis of ABCG2 expression, and fluorescence-activated cell sorting of ABCG2+ cells. For fluorescence microscopy ( 50), 3 to 10 × 103 cells were plated on glass coverslips. Next day, cells were fixed in 4% paraformaldehyde (10 minutes at room temperature) and blocked in 10% goat whole serum. Coverslips were incubated sequentially with monoclonal antibody anti-ABCG2, biotinylated goat anti-mouse IgG, and FITC-conjugated streptavidin with washings in between. Generally 1,200 to 1,500 cells were counted for each cell type to quantify the ABCG2+ cells. For flow cytometry, cells were gently dissociated with Accutase (Innovative Cell Technologies, Inc., San Diego, CA) and washed (twice) in serum-free medium. Cells were stained live in the staining solution containing bovine serum albumin and insulin and FITC- or PE-conjugated monoclonal anti-ABCG2 (15 minutes at 4°C). Samples were analyzed on a Coulter flow cytometer. A minimum of 500,000 viable cells was measured per sample, and cell debris and cell aggregates were electronically gated out. For FACS, 2 to 4 × 107 cells were similarly stained for ABCG2 and used to sort out ABCG2+ and ABCG2− cells. For the positive population, only the top 10% most brightly stained cells were selected. For the negative population, only the bottom 5% to 10% most dimly stained cells were selected. The purity of ABCG2+ cells, as determined by both post-sorting flow analyses as well as restaining followed by fluorescence microscopy analyses, was ≥98% and the purity of the ABCG2− cells was ≥99%.
Xenograft tumor experiments and in vivo tumorigenicity. Xenograft prostate (Du145, LAPC-4, and LAPC-9) and glioma (U373) tumors were aseptically dissected out from animals and minced into ∼1 mm3 pieces in DMEM supplemented with 10% FBS. After rinsing in the same medium (twice) followed by PBS to wash out serum, tumor tissues were incubated with 1× Accumax (1,200-2,000 units/mL proteolytic activity containing collagenase and DNase; Innovative Cell Technologies) at 10 mL/1 g tissue in DPBS for 20 minutes at 37°C under rotating conditions. At the end, residual tissues were allowed to precipitate to the bottom of tubes and dissociated cells collected from the supernatant. When necessary, the residual tumor pieces were subjected to one or more rounds of Accumax digestion and dissociated cells pooled. A single cell suspension was obtained by filtering the supernatant through a 40-μm cell strainer (BD Falcon, Bedford, MA). Upon viability count using erythrosin B (American Type Culture Collection), cell suspension was gently loaded onto a layer of Histopaque-1077 gradient (Sigma-Aldrich, St. Louis, MO) and centrifuged at 400 × g for 30 minutes at room temperature. RBC, dead cells, and debris were removed from the bottom of the tube and live nucleated cells collected at the interface. Then the cell mixture was depleted of lineage-positive host cells using the MACS Lineage Cell Depletion Kit (Miltenyi Biotec, Auburn, CA). Briefly, cells were first incubated (10 minutes at 4°C) in the staining solution [PBS (pH 7.2), 0.5% FBS, 0.5 μg/mL insulin] containing biotinylated antibodies against a panel of lineage antigens (CD5, CD45R, CD11b, anti-Ly-6G, 7-4, and Ter-119). Cells were then incubated with the anti-biotin Microbeads (15 minutes at 4°C) and the Lin− cells were eluted using the MS columns. The eluted prostate cancer cells were all human epithelial cells as confirmed by their expression of Ber-EP4, a surface marker unique to human epithelial cells, indicating that we had obtained highly purified human tumor cells using this protocol.
For tumor experiments, various numbers of cells, either unsorted or sorted populations (i.e., side population, non–side population, ABCG2+, and ABCG2−) were injected in 40 μL of medium/Matrigel (1:1) s.c. into either male or female NOD/SCID mice (4-8 weeks old). MCF7 cells were injected into female mice. Tumor development was monitored starting from the second week. The primary tumor sizes were measured with a caliper on a weekly basis and approximate tumor weights determined using the formula 0.5ab2, where b is the smaller of the two perpendicular diameters. Tumorigenicity was measured mainly by tumor incidence (i.e., the number of tumors/number of injections) and latency (i.e., time from injection to detection of palpable tumors). All animals were terminated at 6 to 9 months after tumor cell injection. Tumors harvested were fixed in formalin and paraffin sections were made for H&E staining or immunohistochemistry for CD31.
Relationship between ABCG2 expression and cell proliferation. Two sets of experiments were done. In one, cells undergoing mitotic division were determined in ABCG2+ and ABCG2− populations of cells. In the other, purified ABCG2+ and ABCG2− cancer cells, cultured for various time intervals, were pulsed with BrdU (2.5 μmol/L × 3 hours) and processed for BrdUrd staining ( 50).
In vitro and in vivo self-renewal and clonal analyses. Purified ABCG2+ and ABCG2− cancer cells were plated at clonal density (i.e., 100-400 cells per well; depending on cell type) in the flat-bottomed 6-well culture dishes. Cells were cultured for different time periods. The percentage of cells that initiated a clone was presented as cloning efficiency. The clone sizes (i.e., the number of cells/clone) were determined for some time points. Triplicate samples were run for each cell type and experiments repeated when feasible. For in vivo self-renewal experiments, tumor cells purified from the tumors derived from unsorted, ABCG2+, or ABCG2− cells were stained for ABCG2 and used to quantify the ABCG2-expressing cells by flow cytometry.
Preparation of mouse bone marrow and newborn mouse keratinocytes. To prepare bone marrow, we sacrificed C57BL/6 mice (6-8 weeks old) by cervical dislocation and removed the skin covering the femurs. The bones were removed by cutting below the knee and cutting at the hip. The muscle was then cleaned from both femurs. The bone marrow cells were flushed out of the femur using a 27-gauge needle in 50 mL of PBS, washed once, and used in the side population analysis. For keratinocytes, newborn pups were cleaned and anesthetized on ice for at least 30 minutes. The tail and limbs were removed and discarded. The skin was removed and floated on 2 mL trypsin (0.25%) solution with dermal side down overnight at 4°C. Epidermis was removed from dermis, placed in Waymouth's Medium (Life Technologies, Gaithersburg, MD) containing 10% FBS and 1% penicillin-streptomycin and minced. The minced epidermis was gently stirred for 20 minutes in a sterile glass beaker. The resulting solution was filtered through a 30-μm mesh and plated at 3 × 106 cells per 35-mm dish for 2.5 hours. The medium was changed to Keratinocyte Growth Medium - 2 (KGM-2; Cambrex BioScience Walkersville, Inc., Walkersville, MD) with 0.5 mmol/L calcium. Cells were either used freshly or cultured for a short period of time (to expand) and used in the side population analysis.
Reverse transcription-PCR analysis. Total RNA was isolated using the Absolutely RNA Nanoprep Kit (Stratagene, La Jolla, CA) and used in semiquantitative reverse transcription-PCR (RT-PCR) analysis ( 50). The PCR primers included ABCG2 (sense, 5′-CTGAGATCCTGAGCCTTTGG-3′; antisense, 5′-TGCCCATCACAACATCATCT-3′); Notch-1 (sense, 5′-ATCGGGCACCTGAACGTGGCG-3; antisense, 5′-CACGTCTGCCTGGCTCGG CTC-3′); β-catenin (sense, 5′-ACTGGCAGCAACAGTCTTACC-3′; antisense, 5′-TTTGAAGGCAGTCTGTC GTAAT-3′); SMO (sense, 5′-ATCTCCACAGGAGAGACTGGTTCGG-3′; antisense, 5′-AAAGTGGG GCCTTGGGAACATG-3′); Oct-4 (sense, 5′-GTGGAGGAAGCTGCAAACAATGAAA-3′; antisense, 5′-GACCGAGGAGTTACAGTGCAGTGAAG-3′); and glyceraldehyde-3-phosphate dehydrogenase (sense, 5′-ACCACAGTCCATGC CATCAC-3′; antisense, 5′-TCCACCACCCTGTTGCTGTA-3′).
Some cultured human cancer cells and xenograft tumors have a side population. Several articles have recently reported the presence of stem cell–enriched side population in long-term cultured mouse C2C12 myogenic cells ( 39), rat C6 glioma cells ( 48), or human brain tumor (i.e., glioma and medulloblastoma; ref. 37) cells. We first sought to determine whether this phenomenon is generally applicable to other human tumor cells in culture, in particular, the human epithelial cancer cells. To that end, we first established the side population protocol on our flow cytometer using as experimental controls the HL60-Dox cells (courtesy of Dr. M. Andreef, Department of Blood and Marrow Transplantation, UT MD Anderson Cancer Center, Houston, TX); i.e., HL60 leukemia cells chronically exposed to a low concentration of doxorubicin. The HL60-Dox cells overexpressed the multidrug resistance proteins that allowed them to efflux various drugs and xenobiotics including the Hoechst 33342 dye. 4 As shown in Fig. 1 , unselected HL60 cells did not show a side population (A) whereas >90% of the HL60-Dox cells were in the side population (B), which presented as a distinct “tail” on the histogram and was completely inhibited by either verapamil (C) or reserpine (D), chemicals previously shown to block the side population phenotype (e.g., refs. 6, 33). To determine the sensitivity of our system, we titrated the HL60-Dox cells into HL60 cultures and then did side population analysis. The results ( Fig. 1E-H) revealed that our system could reliably detect a side population of ∼0.01%. Indeed, using these experimental conditions, we reliably detected ∼0.01% and 0.5% side population in mouse bone marrow cells ( Fig. 1I) and newborn mouse keratinocytes ( Fig. 1L), respectively, and the side population phenotypes in these cells could also be blocked by verapamil or reserpine ( Fig. 1J and K; data not shown).
Using the above protocol, we surveyed ∼30 cultured human tumor cell lines of the prostate, breast, colon, glioma, bladder, ovary, cervix, glioma, and melanoma and we reliably detected a side population (0.04-0.2%) in ∼30% of the cell lines ( Table 1 ; data not shown). We also analyzed side population in tumor cells freshly purified from three xenograft human prostate tumors (i.e., Du145, LAPC-4, and LAPC-9), and we only detected a distinct side population (0.07%) in the LAPC-9 tumor ( Table 1; data not shown). These results suggest that only some cultured human cancer cells and tumors cells freshly purified from xenograft tumors contain a detectable side population and that most cultured human cancer cells may have a side population too small (i.e., <0.01%) to be reliably detected.
Side population cells are more tumorigenic than the non–side population cells. The side population cells isolated from the rat C6 glioma have been shown to be more tumorigenic than the non–side population cells ( 48). To determine whether the side population cells we identified in human cancer cells might also be more tumorigenic, we did several small-scale tumor experiments. We first purified the side population cells from the U373 glioma cells, which had ∼0.1% side population ( Table 1; Fig. 2A, a-b ). When 1,000 U373 side population cells were injected into the NOD/SCID mice, a prominent tumor arose within about 1 month ( Table 2 ; Fig. 2A, c). The side population tumor histologically resembled clinical samples in that it had palisades-like structures ( Fig. 2A, d) and was highly vascularized ( Fig. 2A, e). In contrast, no tumor was observed in 7 months when 50,000 non–side population U373 cells were injected ( Table 2).
We similarly isolated side population cells from MCF7 breast cancer cells, which had ∼0.2% side population ( Table 1; Fig. 2B, a). When injected into the NOD/SCID mice, we observed a cell number–dependent tumor development ( Table 2). For example, we observed a 17% (one of six) tumor incidence with 1,000 side population cells injected and 50% (three of six) tumor incidence with 10,000 side population cells injected ( Table 2). Tumor latency was also reduced with increased numbers of side population cells injected ( Table 2). In contrast to the side population MCF7 cells, the non–side population MCF7 at 1,000 cells did not give rise to any tumors (zero of six). With 10,000 cells injected, we observed one tumor with six injections ( Table 2). Even with 250,000 cells injected, we only observed one of two tumor incidence ( Table 2). In addition, the tumors from the non–side population MCF7 cells arose with longer latencies ( Table 2).
Finally, we purified from the LAPC-9 prostate xenograft tumors the side population cells, which constituted ∼0.07% of the total tumor ( Table 1; Fig. 2C, a-b). When injected into the NOD/SCID mice, as few as 100 cells generated a tumor (25% incidence; Table 2). With 1,000 side population cells injected, we observed a tumor incidence of 75% (three of four; Table 2). With 1,500 side population cells injected, we observed two of two tumor formation with shorter latencies ( Table 2). Tumors derived from the side population cells histologically resembled the unsorted LAPC-9 tumors (data not shown). In contrast, no tumor generation was observed with 1,500 (zero of six) or 15,000 (zero of four) non–side population LAPC-9 cells injected. Even 150,000 non–side population LAPC-9 cells injected did not give rise to tumors within 9 months, although tumor did arise with 300,000 non–side population cells ( Table 2). These results suggest that the side population LAPC-9 tumor cells are probably >100 times more tumorigenic than the non–side population LAPC-9 cells.
Side population cells have some stem cell properties. To determine whether the higher tumorigenicity in the side population cells might be associated with some of the intrinsic stem cell properties, we first used the purified side population and non–side population U373 cells in a clonogenic assay, which partially measures the self-renewal capacity of the cells. As shown in Fig. 2A, f, whereas ∼10% of the side population U373 cells could sustain a clonal growth, <0.01% of the non–side population U373 cells were clonogenic. Then we asked whether the side population cell-generated tumors contain non–side population cells and can be further passaged in vivo. To that end, we took the U373 side population tumor obtained from the preceding experiments (above) and purified tumor cells out and did side population analysis. The results revealed a side population of ∼0.1% (data not shown), which was similar to the percentage of side population cells in the first-generation tumor ( Table 1; Fig. 2A, a). These observations suggest that the side population cells can give rise to non–side population cells in vivo. When injected into the NOD/SCID mice, 100 side population cells generated tumors with ∼60% (7 of 11) efficiency and 1,000 side population U373 cells also generated a tumor ( Table 2). By contrast, no tumors were observed with 1,000 non–side population cells injected (zero of six). Even 200,000 non–side population U373 cells injected did not generate a tumor ( Table 2), suggesting that the U373 side population cells possess self-renewal capacities in vitro and in vivo and they are probably >200 more tumorigenic than the non–side population cells.
To determine whether the side population cells have some other intrinsic properties of stem cells such as preferential expression of “stemness” genes, which are important for stem cell self-renewal, proliferative capacity, or fate determination ( 2, 47, 51), 3 we did semiquantitative RT-PCR analysis of Notch-1, β-catenin (a crucial molecule in the Wnt signaling pathway), and Smoothened (SMO; the activating receptor for the Hedgehog signaling pathway). As shown in Fig. 2B, b, the side population MCF7 cells expressed higher levels of Notch-1 and β-catenin mRNAs than the non–side population MCF7 cells although both populations expressed similar levels of SMO mRNA.
Taken together, these results suggest that the side population cancer cells have some intrinsic properties of stem cells, similar to observations in various normal stem cell populations (see Introduction).
Expression of ABCG2 in a small subset of cancer cells: association of ABCG2 with cell proliferation. Three considerations made us to turn our attention to ABCG2. First, ABCG2 is one of the primary mediators of the side population phenotype in mouse bone marrow and some other cells ( 35). Second, the majority of the cancer cells we have studied do not possess a detectable side population under our experimental conditions therefore an alternate “marker” is needed to identify the rare tumorigenic cells. Third, one of the potential concerns in the side population analysis is that the chronic accumulation of the low levels of the Hoechst dye in the non–side population cells might be toxic to these cells, although our post-sort viability analysis as well as culture of multiple non–side population cells did not reveal such cytotoxicities (data not shown). Regardless, if ABCG2 could be used as a surrogate marker for the side population cells, it would completely avoid the potential cytotoxicity problem. With these considerations, we first examined the relationship between side population cells and ABCG2 expression. As expected, the side population MCF7 cells expressed higher levels of ABCG2 mRNA than the non–side population cells ( Fig. 2B, b). We then examined the ABCG2 protein expression, using both immunofluorescence staining and flow cytometry, in the same spectrum of human tumor cell lines and xenograft tumor-derived cells. In every case, we detected ABCG2 expression in a small percentage (0.1-4%) of the cells ( Table 1; Fig. 3 ). A similar ABCG2 expression pattern (i.e., 0.1-2%) was also observed in several bladder cell lines ( Fig. 3E, d; data not shown) as well as in LAPC-4 and LAPC-9 xenograft tumors (data not shown).
Interestingly, we observed that many of the ABCG2+ cancer cells were in the process of cell division ( Fig. 4 ). Overall, we observed that ∼30% of the ABCG2+ cancer cells were mitotic, whereas only ∼1% of the ABCG2− cells were mitotic. In clonal analyses, we found that the majority of divided cells that had completed or were about to complete cytokinesis equally distributed ABCG2 to both daughter cells ( Fig. 4G-I and J-L). However, in ∼1% of the dividing cells undergoing cytokinesis ABCG2 seemed to segregate asymmetrically to mainly one daughter cell (e.g., Fig. 4M-O). These observations suggest that ABCG2 might preferentially mark proliferating cells and that some ABCG2-expressing cancer cells might be undergoing asymmetrical cell division, a cardinal feature of stem cells. To determine whether ABCG2 expression might be associated with cell proliferation in general, we prospectively purified ABCG2+ and ABCG2− cancer cells to near homogeneity and used them in a BrdU-labeling experiment. As shown in Fig. 5 , acutely purified ABCG2+ Du145 prostate (A) and MDA-MB435 breast (B) cancer cells that had been BrdU-pulsed and cultured for only 3 hours showed significantly more proliferation than the corresponding ABCG2− cells.
ABCG2+ and ABCG2− tumor cells are similarly tumorigenic. Next, we purified ABCG2+ and ABCG2− cells and did tumor experiments to determine whether the ABCG2+ cancer cells might be more tumorigenic. Much to our surprise, when the two populations of U373 cells were injected into the NOD/SCID mice, we did not observe any major differences with respect to their tumorigenicities ( Table 3 ). In fact, the ABCG2− cells tended to generate tumors slightly faster than the ABCG2+ or unsorted cells ( Table 3).
To determine whether the lack of correlation between ABCG2 expression and tumorigenicity may be a cell type–restricted phenomenon, we carried out similar tumor experiments using purified ABCG2+ and ABCG2− cells from prostate (Du145), breast (MDA-MB435), and colon (HCT116) cancer cell cultures. In every case, we failed to observe any significant difference in tumor incidence or latency periods between the two populations ( Table 3; data not shown). We also compared the ABCG2+ and ABCG2− cells purified from Du145 xenograft tumors and again did not observe any major differences in their tumorigenicities ( Table 3).
Evidence that ABCG2 expression marks proliferating tumor progenitors whereas ABCG2− population contains primitive cancer stem cells. If the ABCG2+ cells proliferate faster than the ABCG2− cancer cells, why are they not more tumorigenic? One possibility is that ABCG2+ cells are mostly fast proliferating tumor progenitors (i.e., transit amplifying cells) rather than primitive, slow-cycling CSCs. To test this possibility, we first determined whether the ABCG2+ tumor cells could generate ABCG2− cells and, more importantly, whether ABCG2− cells could generate ABCG2+ cells. As shown in Table 4 , tumors derived from the ABCG2+ cancer cells all contained only a fraction of ABCG2+ cells ( Table 4), suggesting that ABCG2+ cells generated ABCG2− tumor cells in vivo. Except for one tumor derived from the 1,000 ABCG2+ Du145 cells, we detected only small percentages of ABCG2+ cells in all other tumors derived from the ABCG2+ tumor cells ( Table 4). On the other hand, tumors derived from the ABCG2− cells also contained a small fraction of ABCG2+ cells ( Table 4), suggesting that some ABCG2− cells also have the ability to generate ABCG2+ tumor cells.
Next, we did a series of clonal analyses to determine the relationship between ABCG2+ and ABCG2− cancer cells. Although freshly purified ABCG2+ Du145 ( Fig. 5A; 3 hours) or MDA-MB435 ( Fig. 5B; 3 hours) cells proliferated faster than their corresponding ABCG2− counterparts, culture for as short as 1 day eliminated or reduced this proliferative difference. Continued cultures of these cells revealed increasing proliferating (i.e., BrdU+) cells in the ABCG2− populations ( Fig. 5A-B). These results suggest the possibility that, with time in culture, the ABCG2+ tumor progenitors gradually lose their proliferative capacity whereas primitive CSCs in the ABCG2− population give rise to highly proliferative ABCG2+ tumor progenitors. In support, immunostaining revealed the emergence of ABCG2+ cells from the starting ABCG2− cancer cells within 1 week (data not shown), consistent with the ability of some ABCG2− tumor cells to generate ABCG2+ cells in vivo ( Table 4).
Consistent with the BrdU incorporation assays, clonal analyses revealed that at earlier time points the ABCG2+ tumor cells had higher cloning efficiency; that is, more cells had the ability to establish a clone ( Fig. 5C-D). However, at later time points, the ABCG2− tumor cells picked up and formed similar or higher ( Fig. 5C-D) percentages of clones. Subsequently, we carried out differential clonal analyses in which purified ABCG2+ and ABCG2− tumor cells were plated at clonal density and clonal sizes were quantified at a shorter and a longer time point. As shown in Fig. 5E, 7 days after plating, more ABCG2+ Du145 cells formed larger clones. However, at 14 days post plating, there were significantly more large clones derived from the ABCG2− Du145 cells ( Fig. 5G). The average clonal sizes (cells/clone) of the ABCG2+ versus ABCG2− Du145 cells were 101 and 60 cells at 7 days versus 4,716 and 5,658 cells at 14 days, respectively (P < 0.01 in both cases, ANOVA). Similarly, more ABCG2+ MDA-MB435 cells formed larger clones than the ABCG2− MDA-MB435 cells at 7 days after plating ( Fig. 5F). However, by 11 days after plating, there were significantly more large clones (i.e., >500 cells per clone) derived from the ABCG2− MDA-MB435 cells ( Fig. 5H). The average clonal sizes of the ABCG2+ versus ABCG2− MDA-MB435 cells were 30 and 16 cells at 7 days (P < 0.01) versus 346 and 356 cells at 11 days, respectively. Together, the proliferation ( Fig. 5A-B) and clonal ( Fig. 5C-H) analyses provide strong evidence that the ABCG2+ tumor cells likely represent fast-proliferating tumor progenitor cells, whereas the ABCG2− population contains slow-cycling, primitive CSC cells that, with time, could establish robust clonal growth and also generate fast-cycling progenitor cells.
Finally, we did RT-PCR analysis to assess the mRNA expression of several stemness genes ( Fig. 6A-B ). Consistent with the concept that the ABCG2− population contains primitive, stem-like cancer cells, we found that purified ABCG2− tumor cells expressed higher mRNA levels of Notch-1, β-catenin, and SMO ( Fig. 6A-B). Even Oct-4, a transcription factor essential for embryonic stem cell self-renewal ( 51) and recently shown to be expressed in some adult stem cells ( 52), also showed preferential expression in the ABCG2− cells in three of four cell types ( Fig. 6A). Interestingly, the Notch-1 mRNA was detected nearly exclusively in the ABCG2− tumor cells ( Fig. 6A). Similarly, the β-catenin mRNA was observed only in the ABCG2− Du145 cells ( Fig. 6B). These RT-PCR results provide strong support for the existence of primitive CSC in the ABCG2− tumor cell population.
A long-time puzzle to tumor biologists is the observations that even with long-term cultured cancer cells, in general sufficient numbers of cells have to be injected to initiate an orthotopic tumor in recipient animals (reviewed in refs. 1, 47), suggesting that even in long-term tumor cell cultures, not all cells are equal and only a small population of cells is tumorigenic. Indeed, when multiple human cancer cells, which have been in culture under different conditions for years or even decades are assessed for their clonal growth and clonogenic potentials, we find that only a small percentage of cells possesses such potentials. 3 Furthermore, when tumor cell–derived spheres or xenograft tumors in situ are pulsed with BrdU followed by extended chase, only a very minor population of the cells manifests as the long-term LRCs, 3 which are known to preferentially identify stem cells ( 2, 5). In further support, long-term cultured rat C6 glioma cells ( 48) and some human brain tumor (i.e., glioma and medulloblastoma; ref. 37) cells are found to contain a side population, which is known to be enriched in stem cells (see Introduction). Importantly, the C6 side population cells are more tumorigenic than and can also generate the non–side population cells ( 48), providing the first direct evidence for a population of more tumorigenic cells in long-term tumor cell cultures. It is these observations ( 37, 48) 3 that have prompted us to first determine whether the side population technique can be generally applied to other cultured human tumor cells, in particular, the human epithelial cancer cells.
Our results reveal that ∼30% cultured human cancer cell lines and xenograft tumor-derived cells possess a side population that can be reliably detected under the current experimental conditions. Several notable points are worthy of mention. First, in most literature reports, the side population is identified on either a MoFlo or FACS Vantage flow cytometer as a continuous tail of the non–side population. Therefore, the discrimination of the side population from the non–side population is, to a certain extent, arbitrary and can vary significantly from experiment to experiment. In contrast, using a Coulter Epics flow cytometer and our modified protocol, we have in most cases identified the side population as a distinct “side” population separate from the main non–side population. This could potentially give us relatively pure side population cells. Second, in support of this possibility, our system seems to identify the side population cells in a more stringent manner and the percentages of the side population cells may be more representative of putative CSCs in the cultures or xenograft tumors. Therefore, only ∼30% cultured human cancer cell lines and xenograft tumor-derived cells possess a side population of 0.04% to 0.2%. These percentages are similar to the side population of multiple normal stem cell or progenitor cell populations [i.e., 0.01-5%; refs. 18, 23, 26; e.g., mouse bone marrow (0.01%; Fig. 1), newborn mouse keratinocyte progenitors (0.5%; Fig. 1), and human bone marrow (∼0.03%; ref. 8)]. The majority of the cancer cell lines or xenograft tumors examined seems to possess too small a side population to be reliably detected.
Importantly, the side population cells purified from two cell lines and one xenograft tumor are more tumorigenic than the non–side population counterparts. Furthermore, the side population cells are found to possess several intrinsic properties of stem cells: self-renewal, preferential expression of some stemness genes, and an ability to give rise to non–side population cells. These results thus extend the others studies on rat C6 glioma cells ( 48) and support the concept that the side population is indeed enriched in stem-like tumorigenic cells. It should be noted that with higher numbers of the non–side population epithelial cancer cells (i.e., MCF7 and LAPC-9) injected, we also observed tumor development. These results may suggest that the non–side population also contains a very small percentage of tumorigenic cells, although the results might have stemmed from the contamination of small numbers of the side population cells in the non–side population as epithelial cancer cells are very “sticky”.
The major focus of the current study is to address whether the higher tumorigenicity associated with the side population cells is related to ABCG2 expression. Because the side population phenotype in some cell types is mainly mediated by ABCG2, the side population cells (e.g., in MCF7) express higher levels of ABCG2 mRNA, ABCG2 is expressed only in a small subset of cancer cells, and ABCG2 expression in cancer cells seems associated with cell proliferation, it stands to reason that the ABCG2+ cancer cells might or should be more tumorigenic than the ABCG2− tumor cells. Surprisingly, however, highly purified ABCG2+ cells from several types of tumor cells are not more tumorigenic than the corresponding ABCG2− cells. Further proliferation assays, clonal analyses, self-renewal, and molecular studies suggest a model in which the ABCG2− population contains primitive stem-like cancer cells with higher self-renewal (because of higher levels of stemness genes) and proliferative potentials but are normally slow cycling ( Fig. 6C). These cells then give rise to ABCG2+ tumor progenitor cells that are more actively proliferating but possess reduced self-renewal and long-term proliferative capacities ( Fig. 6). The ABCG2+ tumor progenitor cells eventually give rise to ABCG2−, partially or even fully differentiated tumor cells that constitute the bulk of tumor cell mass ( Fig. 6).
The side population has been shown to be very heterogeneous ( 10, 27, 28). Therefore, the side population detected in cancer cells might contain several subsets of cells, one of which expresses ABCG2 thus explaining the increased ABCG2 expression in the side population. The higher tumorigenicity associated with the side population might be conferred by combined effects of several other subpopulations of cells in addition to the ABCG2+ cells. For example, cells expressing other ABC family members might also contribute to the cancer cell side population phenotype. Indeed, it has been shown that only a fraction of side population cells expresses ABCG2 ( 24) and that both side population and known stem/progenitor cells also express other ABC transporters such as MDR-1 (i.e., ABCB1 or P-glycoprotein), MRP-1 (ABCC1), and ABCA2 ( 36, 37, 39). In addition, enforced expression of MDR-1 in murine bone marrow cells is sufficient to expand the side population ( 40). We are currently using cultured cancer cells as well as xenograft and primary human tumor samples to determine the molecular basis of the higher tumorigenicity associated with the side population (e.g., amplification of oncogenes and/or mutations of specific tumor suppressors), dissect different subpopulations of the side population with respect to their tumorigenic potentials, and elucidate the interrelationship between the side population and several other tumorigenic populations we have identified.
Grant support: NIH grants CA90297, AG023374, and P30 CA16672; NIEHS grant ES07784; American Cancer Society grant RSG MGO-105961; Department of Defense grant DAMD17-03-1-0137; Prostate Cancer Foundation; and M.D. Anderson Cancer Center (PCRP and IRG).
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.
We thank C. Conti, T-J. Liu, M. Andreef, and C. Sawyers for providing cells; C. Carter for assistance in preparing mouse bone marrow cells; the Histology Core for excellent assistance in tissue processing and immunohistochemistry; the Animal Facility Core for help in tumor experiments; E. Richie for her insights; and members of the Tang lab for support and helpful discussion.
Note: J. Zhou is currently at the Dermatology Branch, National Cancer Institute, NIH, Building 10, Room 12N262, 10 Center Drive, MSC 1908, Bethesda, MD 20892-1908.
↵3 C. Jeter et al. Stem-like cancer cells in culture and xenograft tumors: expression and roles of stemness genes, submitted for publication.
↵4 Unpublished observations.
- Received February 21, 2005.
- Revision received April 16, 2005.
- Accepted April 28, 2005.
- ©2005 American Association for Cancer Research.