Cells with Characteristics of Cancer Stem/Progenitor Cells Express the CD133 Antigen in Human Endometrial Tumors Free

In Western countries, endometrial cancer is the most common gynecologic pelvic malignancy, accounting for 6% of all cancers in women (1). Endometrial cancer is usually diagnosed at an early stage because most patients become symptomatic and suffer from abnormal vaginal bleeding. Approximately 80% of all patients are diagnosed as having stage I disease and 13% as having stage II disease (2). Although the 5-year overall survival is as high as 88% for women with surgically staged stage I disease, subgroups of patients with early stages have significantly decreased 5-year overall survival rates (2). Based on retrospective data, various prognostic factors have been identified and validated, such as tumor grade, histologic type, and high-grade lesions, and deep invasion of the uterus. The risk of endometrial cancer recurrence has been well characterized and ranges from 7.7% to 63.3%, depending on the presence or absence of specific prognostic factors (3). For these reasons, novel targeted therapies are currently being developed with the aim to achieve greater specificity for a selected population of cancer cells (1).

Cancer stem cells (CSC) can be defined as a population of undifferentiated tumorigenic cells responsible for tumor initiation, maintenance, and spreading (4). In accordance with the paradigm already established for hematopoietic stem cells (5), CSC display unlimited proliferation potential, ability to self-renew, and capacity to generate a progeny of differentiated cells that constitute the major tumor population. To date, CSC have been isolated from leukemias (6) and solid tumors and have been expanded in vitro as tumor spheres (4, 7). In the recently proposed CSC model, only a tiny proportion of cells in the tumor is expected to proliferate and self-renew extensively, thus sustaining in vivo tumor growth, whereas the bulk of tumor cells proceed to differentiate into heterogeneous tumor cells that become the phenotypic signature of the tumor (4, 7, 8). It is, therefore, conceivable that only the eradication of CSC or, alternatively, the induction of CSC differentiation into cells lacking self-renewal potential may lead to an effective cancer cure.

CD133/prominin-1 is a 120-kDa glycoprotein and a founding member of the prominin family of pentaspan membrane proteins (9). CD133 is expressed by normal primitive hematopoietic stem cells from adult blood, bone marrow, and umbilical cord blood and on endothelial, neural, and epithelial cells (10–12). By virtue of its highly restricted expression on plasma membrane protrusions of epithelial cells, as well as its association with membrane cholesterol, a role for members of the CD133 family in plasma membrane topology and maintenance of an appropriate lipid composition has been proposed (reviewed in ref. 9). Currently available monoclonal antibodies to CD133 mostly recognize the AC133 antigen, a glycosylation-dependent epitope of CD133 (13).

The expression of CD133 has been recently associated with CSC isolated from prostate (8), lung (14), brain (15), and ovarian cancers (16) and from childhood acute lymphoblastic leukemia (17). CD133⁺ cells with tumorigenic potential have also been identified in human colon cancer (18–20). In an in vivo xenograft model consisting of implantation of human colon CSC into irradiated NOD/SCID mice, an estimate of 20 colon cancer–initiating cells per 57,000 unfractionated tumor cells has been calculated, implying that not all the cells within a tumor may be able to initiate and sustain neoplastic growth (19). In most studies published so far, CD133 has been detected by its glycosylated epitope, AC133, consistent with the knowledge that the glycosylation state of a cell may change upon malignant transformation.

CD133 has also been identified in glandular and luminal epithelial cells from normal endometrial tissues using the AC141 monoclonal antibody that recognizes an epitope different from that targeted by the AC133 clone (21). In that study, CD133⁺ cells isolated from human endometrium were endowed with a low colony-forming unit capacity (21). It is presently unknown whether CD133⁺ cells with tumorigenic potential can be prospectively isolated from primary endometrial cancer. The present study was designed and conducted with the aim to determine whether CD133 may be expressed by human endometrial cancer, whether CD133⁺ cells from primary tumors manifest any of the currently recognized stem/progenitor cell properties, including the ability to self-renew and initiate tumor development in vivo, and whether CD133 expression may correlate with clinically relevant features, such as histologic grade, International Federation of Gynecology and Obstetrics (FIGO) clinical stage, and presence of metastatic lymph node lesions.

Characteristics of patients. Tumor tissue specimens were obtained at time of primary surgery from 113 patients with endometrial carcinoma. Written informed consent to tumor tissue collection and use for isolation of putative tumor stem cells was obtained by each patient according to the research protocol approved by the local ethical committee. As detailed in Table 1, ages of patients were ≤65 and >65 years in 67 and 46 cases, respectively. Tumors were categorized following the FIGO criteria (22). In 86 cases, tissue samples were obtained from FIGO stage I to stage II endometrial cancers, whereas, in 27 cases, biopsies were obtained from FIGO stage III to stage IV tumors. Tumor grade was G₁-G₂ and G₃ in 58 and 55 patients, respectively. Most tumors showed endometrioid histology (n = 92) and negative nodal status (n = 79). CA-125 serum levels were available in 79 of 113 patients and were >35 IU/mL in 27 of them (Table 1).

Tissue collection, isolation, and culture of CD133⁺ cancer cells. At time of tissue collection, the tumor specimen was cut in two halves; one half was used to confirm the pathologic diagnosis, and the second half was used for isolation, purification, and culture of CD133⁺ cells. Within 30 min from surgery, tumors were mechanically and enzymatically dissociated with trypsin-EDTA (Life Technologies-Invitrogen) for 15 min and then with collagenase I (1 mg/mL) for 3 h at 37°C.

After washing with PBS, cell suspensions were incubated with ammonium chloride (StemCell Technologies) for 10 min at 4°C to lyse contaminating erythrocytes. The negative selection of nonhematopoietic CD45⁻ cells was done with CD45 MicroBeads (Miltenyi Biotec) and was followed by positive selection of CD133⁺ cells (CD133 cell isolation kit, Miltenyi Biotec) on an AutoMACS platform. Antibodies to CD133/2 (AC141 clone, Miltenyi Biotec) were used to assess the purity of selected CD133/1⁺ populations.

To culture endometrial cancer spheres, cells recovered after the enzymatic dissociation of primary tumors were resuspended in serum-free DMEM (Life Technologies-Invitrogen) supplemented with 50 μg/mL insulin, 100 μg/mL apotransferrin, 10 μg/mL putrescine, 0.03 μmol/L sodium selenite, 2 μmol/L progesterone, 0.4% bovine serum albumin, 20 ng/mL epidermal growth factor, 10 ng/mL fibroblast growth factor-2 (Sigma Chemicals), as previously detailed (14). The medium was replaced or supplemented with fresh growth factors weekly, until cells started to grow and form floating aggregates. Cultures were expanded weekly by mechanical dissociation of the spheres, followed by replating of small single cells and residual small aggregates in complete fresh medium. Isolated CD133⁺ cells were maintained in RPMI 1640 culture medium (Miltenyi Biotec) supplemented with 10% fetal bovine serum. Aliquots of cultured CD133⁺ cells were harvested weekly and used to assess viable cell number and expression of CD133 by flow cytometry.

Umbilical cord blood CD133⁺ hematopoietic stem cells were obtained from consented mothers after full-term delivery and were used as controls in selected experiments, as detailed in the figure legends.

Cell renewal potential at the single-cell level. For single-cell colony assays, purified CD133⁺ cells or CD133-depleted tumor cells were plated at a final density of one per well in 96-well microwell plates in 200 μL of MyeloCult H5100 (StemCell Technologies) supplemented with 10^-6 mol/L hydrocortisone. Cell cultures were fed with fresh medium every 3 d for two consecutive weeks. Wells were scored as positive when at least 30 viable cells could be identified on day 14. All other wells were scored as negative and usually contained no detectable cells.

Cell proliferation assays were done by plating CD133⁺ and CD133⁻ cells in T25 flasks with RPMI 1640 supplemented with 10% fetal bovine serum at a density of 90,000 cells/mL. Cell counts were obtained on days 6 and 12 using a Neubauer counting chamber. Cell viability was determined by trypan blue exclusion.

Flow cytometry and immunofluorescence analysis. The quantification of CD133/1 expression was done as follows. Cells were incubated for 10 min at 4°C with FcR blocking reagent (Miltenyi Biotech), followed by labeling with monoclonal antibodies to a glycosylation-dependent epitope of CD133 (CD133/1 or AC133, Miltenyi Biotec) and to the pan-hematopoietic marker CD45 (Becton Dickinson). After 30 min at 4°C, cells were extensively washed with ice-cold PBS and resuspended in 20 μg/mL 7-amino-actinomycin-D for 10 min at room temperature, shielded from light, before flow cytometry analysis. Fluorochrome-conjugated, isotype-matched monoclonal antibodies from the same manufacturers were used to establish background fluorescence. To rule out any contamination of CD133⁺ cells with endothelial cells, aliquots of purified CD133⁺ cells were labeled with monoclonal antibodies against vascular endothelial growth factor (VEGF) receptor II (KDR; clone 89106, R&D Systems), CD105 (clone SN6, Serotec), CD34 (clone 8G12, Becton Dickinson), and VE-cadherin (Bender MedSystems). In any of the described experiments, fluorochrome-conjugated, isotype-matched monoclonal antibodies from the same manufacturers were used to establish background fluorescence.

To detect intracellular cytokeratin-7 (CK-7), CD133⁺ cells were sequentially treated with commercially available fixation and permeabilization media (Fix & Perm reagent, Caltag Laboratories), followed by incubation for 1 h at 4°C with monoclonal antibodies reacting against human CK-7 (DakoCytomation). FITC-conjugated goat antimouse antibodies from the same manufacturer were used as secondary reagents. The difference between test and control histogram distributions was calculated with the Kolmogorov-Smirnov statistics (D_KS). A D_KS value of >0.15 is conventionally considered to express significant differences in histogram distributions.

For DNA content analysis, cells were fixed for 1 h at 4°C with 70% ice-cold ethanol. After washings with PBS, cells were resuspended in appropriate volumes of DNA staining solution, containing 50 μg/mL propidium iodide, 6.25 μg/mL Rnase A, and 12.5 μL/mL Nonidet-P40.

Cells were run through a fluorescence-activated cell sorting (FACS) Canto flow cytometer (Becton Dickinson) with standard equipment. A minimum of 30,000 events was collected and acquired in list mode using the FACS Diva software (Becton Dickinson). The analysis of cell cycle distribution was done with the ModFit software (Verity Software House), as previously detailed (9).

Reverse transcription-PCR. Total RNA was extracted using RNeasy Mini kit (QIAGEN) according to the manufacturer's protocol from tissues homogenized twice for 4 min at 30 Hz using Mixer Mill MM 300 (Retsch) or from cells homogenized by vortexing for 30 s. The RNA concentration of each sample was measured by spectrophotometry (Beckman spectrophotometer DU640). The RNA samples were frozen at −80°C until reverse transcription-PCR (RT-PCR) analysis. One microgram of total RNA was reverse-transcribed with 25 units of Moloney murine leukemia virus reverse transcriptase (PE Applied Biosystem) at 42°C for 30 min in the presence of random hexamers. Two microliters of cDNA products were amplified with 1 unit of AmpliTaq Gold (PE Applied Biosystem) in the presence of primers specific for CD133 (23). For the amplification of glyceraldehyde-3-phosphate dehydrogenase, the following primers were used: 5′-TGACATCAAGAAGGTGGTGA-3′ and 5′TCCACCACCCTGTTGCTGTA-3′ (both synthesized by m-Medical). Reactions were conducted in the PTC-0200 DNA Engine (MJ RESEARCH), as described by Yu and colleagues (23). The PCR products were analyzed on 3% agarose gel and stained with ethidium bromide (2% agarose type II + 1% Nu-Sieve agarose, Sigma Aldrich).

Immunohistochemistry. Immunostaining was done on 3 μm formalin-fixed, paraffin-embedded cancer tissue sections, mounted on poly-l-lysine–coated slides or SuperFrost Plus microscope slides, and dried at 37°C overnight. The slides were then deparaffinized in xylene and rehydrated. Endogenous peroxidase was blocked with 3% H₂O₂ for 5 min. To reduce nonspecific binding, the sections were incubated with 20% normal goat serum for 30 min at room temperature. Cells expressing CD133 were identified after overnight incubation at 4°C with mouse anti-CD133 antibodies (AC133 clone, 1:50 final dilution). CD133 was visualized with the Dako REAL EnVision Detection System, Peroxidase/DAB+ (DakoCytomation). Sections were counterstained with hematoxilyn, dehydrated and cleared in xylene, and then mounted with Eukit.

Detection of estrogen receptors. Cytospins were prepared using 100 μL of cell suspension containing ∼50 × 10³ cells. All cytospins were immediately fixed with 4% paraformaldehyde for 10 min at room temperature. To permeabilize the cells, the slides were incubated for 15 min at room temperature with 0.5% Triton X-100 (Sigma Chemicals). Cytospins were also treated with 20% normal goat serum for 30 min at room temperature to reduce nonspecific binding. The detection of estrogen receptors (ER) was done with mouse antihuman ER monoclonal antibodies (1:40 dilution, 6F11 clone, UCS) in 20% normal goat serum. The expression of ER was visualized with the Dako REAL EnVision Detection System, Peroxidase/DAB+ (DakoCytomation). Sections were counterstained with hematoxylin, dehydrated, cleared in xylene, and mounted with Eukit. Positive controls for ER were prepared with MCF7 breast cancer cells (American Type Culture Collection, LGC Promochem).

Confocal microscopy. Serial formalin-fixed, paraffin-embedded endometrial tumor samples were deparaffinized with xylene, followed by absolute ethanol, 95% ethanol, and distilled water. After blocking of the unspecific sites, the sections were incubated for 1 h at room temperature with mouse monoclonal anti–Tn-MUC1 antibody (5E5, kindly provided by H. Clausen, University of Copenhagen; 1:5 dilution) and then with Texas Red–conjugated goat anti-mouse IgG (H + L, Jackson ImmunoResearch Laboratories; 1:200 dilution). CD133 was visualized with an anti-CD133 antibody (AC133 clone; 1:30 dilution) after 18 h of incubation at 4°C followed by FITC-conjugated goat anti-mouse IgG (H + L, Jackson ImmunoResearch Laboratories). The FITC fluorescence was amplified using FASER kit-FITC (Miltenyi Biotech) according to the manufacturer's instructions. After washes, the coverslips were mounted with VECTASHIELD Mounting Medium (Vector Laboratories, Inc.). Imaging was done by two-photon absorption fluorescence. A conventional confocal laser-scanning microscope (TE2000E, Nikon) and C1 Nikon Plus scanning head was converted for two-photon absorption use. Two-photon absorption fluorescence was excited by a Ti/sapphire ultrafast laser source (Mai Tai Laser 750-850, Spectra Physics) set at a wavelength of 750 nm and a power output of 850 mW, which corresponds to ∼6 mW of average power in the focal plane. The fluorescence signal, collected by the Plan Apochramat 40×/1.40/0.21/Oil objective (Nikon) and selected by the HQ535-50 filter (Chroma, Inc.) was fed to a multimode fiber directed to a photomultiplier (R928, Hamamatsu) in the C1 plus controller. Colocalization was done using SVI software (Scientific Volume Imaging).

In vitro sensitivity of primary tumors to chemotherapeutic agents. Dissociated CD133⁺ cells and their CD133⁻ counterpart were cultured for 48 h in the presence of escalating concentrations of cisplatin (0.1-15 μmol/L), doxorubicin (10-100 μmol/L), and paclitaxel (1-20 nmol/mL), as detailed in the figure legends. Drug concentrations to be applied in vitro were selected based on previous reports (14, 24–26). To estimate the percentage of apoptotic cells (27), samples were labeled for 20 min at 4°C with FITC-conjugated Annexin V diluted in calcium-containing binding buffer. Negative controls for Annexin V staining were established with calcium-free PBS to impede Annexin V binding to membrane phospholipids. Samples were further stained with 20 μg/mL 7-amino-actinomycin-D for 20 min before flow cytometry analysis. The percentage of Annexin V⁺ 7-amino-actinomycin-D⁻ apoptotic cells was calculated with the FACS Diva software package (Becton Dickinson).

Gene expression profiling. RNAs (at least 1 μg total RNA per sample) were extracted from highly purified tumor-derived CD133⁺ cells and their CD133⁻ counterpart and were hybridized on topic-defined PIQOR oncology and stem cell microarrays (PIQOR Microarray and Bioinformatics Services, Miltenyi Biotec) to identify differentially expressed genes. The PIQOR oncology microarrays comprise 1,503 genes mainly involved in apoptosis, cell cycle and DNA repair, cytokine signaling, metabolism, and genes encoding for transcription factors, hormone growth factors, extracellular matrix, and structural proteins. The PIQOR stem cell microarrays consist of 942 genes relevant for stem cell biology and differentiation. The full gene lists for both oncology and stem cell microarrays are provided as Supplementary Tables S1 and S2, respectively. A linear T7-based amplification step was done before labeling and hybridization. Discriminatory gene lists were generated by a ranking approach. A high-confidence list of differentially regulated genes required at least two measurements with a 2-fold or higher expression difference or, in the absence of replicates, one measurement with a 4-fold expression difference. Additionally, the median expression change, as calculated from all the available replicate measurements, had to be at least 2-fold. The high-confidence gene lists are shown as Supplementary Tables S3 to S6. A functional grouping was subsequently done with the proprietary TreeRanker software (Miltenyi Biotec) and was refined by an annotation enrichment analysis aimed at identifying pathway affiliations or other functional annotations that are significantly enriched in the regulated gene sets. The information on molecular function and involvement in biological signaling pathways has been based on Gene Ontology⁸

Generation of s.c. tumors into NOD-SCID mice. For mice xenografts, tumor cells obtained from tumor tissue dissociation were diluted in growth factor–containing stem cell medium and mixed with an equal volume of Matrigel before s.c. injection in the flank of 4-wk-old to 6-wk-old female NOD/SCID mice (28). Details about tumor characteristics, tumor cell dose, and mice xenografting are shown in Supplementary Table S7. When tumors reached a minimum of 10 mm in diameter, mice were sacrificed and tumor tissue was collected. Part of the tumor was enzimatically dissociated, and single cells were either cultured in stem cell medium to obtain tumor spheres or analyzed by flow cytometry for CD133/1 expression. The remaining tumor tissue was fixed in buffered formalin and subsequently analyzed by immunohistochemistry. H&E staining followed by immunohistochemical analysis were done to evaluate tumor histology and compare mouse xenografts with patient tumors. Animal experiments adhered to the requirements of the Commission Directive 86/609/EEC, the Italian legislation (Decreto Legislativo 116, 27 January 1992), and the UK Coordinating Committee on Cancer Research guidelines for the welfare of animals with experimentally induced neoplasia (29). The studies were approved by the Animal Care and Use Committee (Catholic University and Istituto Superiore di Sanità).

Statistical methods. The approximation of data distribution to normality was preliminarily tested with statistics for kurtosis and symmetry. Results were presented as mean and SD, and all comparisons were done with the Student's t test for paired or unpaired determinations or with ANOVA, as appropriate. Correlations between parameters of interest were explored using the Spearman rank analysis and correlation coefficient (R²). The criterion for statistical significance was defined as P ≤ 0.05.

CD133/1 expression pattern in primary endometrial tumors. In a first set of experiments, we quantified CD133 expression by flow cytometry in surgically resected primary endometrial tumors. Patients' characteristics are illustrated in detail in Table 1. Primary tumor samples (n = 113) exhibited a variable degree of immunoreactivity with CD133/1 antibodies, in agreement with previously published reports on the phenotype of putative CSC in solid tumors (15). The CD133/1 antigen was expressed on a median of 18.1% (range, 1.3-62.6%) of freshly dissociated tumor cells. Flow cytometry profiles from representative endometrial tumors are depicted in Fig. 1A. After immunomagnetic selection, CD133⁺ cells were consistently at >97% pure and stained negatively for CD45, a pan-hematopoietic marker (Fig. 1B). Endothelial cell antigens, such as KDR (Fig. 1B), VE-cadherin, and CD105 (data not shown), were found on negligible percentages of CD133⁺ cells purified from primary tumor samples, suggesting lack of measurable contamination with endothelial progenitors and/or mature endothelial cells. This concept was further reinforced by the absence of immunoreactivity with anti-CD34 monoclonal antibodies (data not shown), recognizing hematopoietic stem cells and mature endothelial cells. Finally, immunoselected CD133⁺ cells expressed intracytoplasmic CK-7 (Fig. 1C), as already described in endometrioid adenocarcinomas (30). In these experiments, umbilical cord blood CD133⁺ cells were used as negative control for CK-7 expression (Fig. 1C).

We also stained a randomly selected group of primary tumors (n = 21) with anti-CD133/2 antibodies, recognizing an epitope of the CD133 antigen different from the glycosylation-dependent epitope 1 recognized by the AC133 clone. It has been previously shown that antibodies to CD133/1 and CD133/2 (clone AC141) might recognize tumor cells in a discordant pattern, when used in patients with myelodysplastic syndromes and acute myeloid leukemias (31). As shown in Fig. 1D, a statistically significant positive correlation between the percentage of CD133/1-expressing and CD133/2-expressing cells was found, suggesting that discordant staining by AC133 and AC141 monoclonal antibodies could not serve as a marker for the malignant clone in endometrial carcinomas (32).

We next investigated the expression of CD133 in paraffin-embedded tumor sections. As shown in Fig. 2A, tumor sections were diffusely stained by anti-CD133/1 antibodies. The expression pattern of CK-7 in sections from the same primary tumors is shown for comparison. Finally, mRNA signals for CD133 were detected in dissociated tumor samples, as well as in normal CD133⁺ hematopoietic stem cells used as controls, at variance with normal endometrial tissue (Fig. 2B). The same paraffin-embedded tumors were used to investigate the simultaneous expression of CD133 and the MUC1 tumor-associated antigen. The 5E5 monoclonal antibody was chosen to selectively detect the Tn-MUC1 glycoform (GalNAcα1-O-Ser/Thr of MUC1) that has been shown to be expressed by epithelial tumor cells (33). Figure 2C shows the results of a representative confocal microscopy experiment aimed at visualizing the expression of the antigens of interest. CD133⁺ cells coexpressed the MUC1 Tn glycoform, as shown in Fig. 2C, where colocalization is outlined in white. Colocalization of the two antigens was found to be prominent in the membrane of the tumor cells.

Functional characterization of CD133-expressing cells in primary endometrial tumors. We and others have previously shown that CD133-expressing hematopoietic stem cells from peripheral blood homogeneously reside in the G₀-G₁ phase of the cell cycle (32, 34–36). At variance with normal CD133⁺ hematopoietic stem cells from umbilical cord blood (>99% in the G₀-G₁ phase of the cell cycle), CD133⁺ cells isolated from primary tumor samples exhibited a readily detectable proliferative activity (6.8 ± 3% of cells in the S + G₂M phase of the cell cycle), which was comparable with that of CD133⁻ tumor cells (8.5 ± 4% of cells in the S + G₂M phase of the cell cycle; number of experiments, 10; P = not significant). Collectively, cell cycle studies suggested that tumor-derived CD133⁺ cells shared some biological characteristics with the CD133⁻ tumor counterpart but differed from hematopoietic CD133⁺ cells in their cycling activity.

To analyze whether tumor-derived CD133⁺ cells manifested other features of stem/progenitor cells, such as self-renewing capacity and differentiation potential, we measured the ability of this cell population to survive in suspension culture. The maintenance of CD133⁺ cells in serum-replenished medium translated into a significant reduction of CD133 expression (Fig. 3A), paralleled by an increased expression of CK-7 both in terms of percentage of positive cells and in terms of mean fluorescence intensity (data not shown), suggesting cell differentiation. By day 14 of culture, <50% of cells stained positively for CD133, as depicted in Fig. 3A. Conversely, the number of viable cells in cultures seeded with CD133⁺ cells significantly increased, indicating the occurrence of vigorous cell proliferation (Fig. 3B). This corresponded to an average 4.6-fold expansion by day 14 compared with day 0, as opposed to an average of 2.2-fold expansion of the CD133⁻ cell counterpart (P = 0.02). Interestingly, CD133⁺ cells had a greater cloning efficiency in single-cell colony assays (14.6 ± 5.2%) compared with CD133⁻ tumor cells (2.1 ± 0.5%; number of experiments = 6; P = 0.001).

We then determined whether CD133⁺ and CD133⁻ cells isolated from primary endometrioid tumors differentially responded to short-term culturing in the presence of estradiol. As shown in Fig. 3C, CD133⁺ cells expanded 2.5-fold on average in response to exogenously added estradiol compared with CD133⁺ cells maintained for 72 hours in the absence of estradiol. In sharp contrast, the number of CD133⁻ cells did not appreciably change in cultures done in the presence or absence of estradiol, suggesting lack of in vitro responsiveness to hormone challenge (Fig. 3C). The expression of ERs was then investigated by immunocytochemistry in CD133⁺ and CD133⁻ cells from primary endometrioid tumors and in the MCF7 breast carcinoma cell line, used as positive control. These studies confirmed the preferential expression of ER on CD133⁺ cells compared with CD133⁻ cells (Fig. 3D).

To test the ability of mechanically dissociated bulk tumors to form sphere-like structures, defined as nonadherent colonies of cells derived from a single CSC (15), we plated tumor cell suspensions from 15 randomly selected endometrial cancer specimens for up to 12 weeks under the culture conditions detailed in Materials and Methods. After 5 weeks of culture, primary tumor spheres were evident in 5 of 15 cases herein analyzed (Fig. 4A). Cells grown as tumor spheres maintained the expression of the CD133 antigen and could be propagated in culture for up to 12 weeks (Fig. 4B and C). It must be emphasized that the ability to form sphere-like structures gradually declined starting from week +7 of culture and such sphere-like elements were no longer detectable from week +8 of culture onward (Fig. 4C). This observation is again in line with the assumption that tumor-derived CD133⁺ cells may undergo differentiation after an initial burst of vigorous cell proliferation.

In vitro chemosensitivity of tumor-derived CD133⁺ cells. Endometrial cancer-derived CD133⁺ and CD133⁻ cells were exposed to increasing concentrations of chemotherapeutic drugs currently in use in the clinical setting, such as cisplatin, doxorubicin, and paclitaxel. As shown in Fig. 4D, both cisplatin and paclitaxel induced negligible apoptotic responses in CD133⁺ cells compared with their CD133⁻ counterpart, indicating that CD133⁺ cells may be relatively resistant to chemotherapeutic agents. Conversely, apoptotic cells were readily detected among CD133⁻ tumor cells, and their frequency increased with increasing drug concentrations, suggesting sensitivity to cisplatin and paclitaxel after 48 hours of in vitro exposure. However, CD133⁺ and CD133⁻ endometrial cancer-derived cells were equally sensitive to doxorubicin, as indicated by the comparable percentage of apoptotic cells that emerged after in vitro exposure to the drug.

In vivo tumor growth assays. To investigate whether endometrial tumors may harbor a population of CSC with tumor-initiating activity, we developed an in vivo model in which NOD/SCID mice were injected s.c. with cells from human primary endometrial cancers after their enzymatic digestion. Tumor formation was closely monitored, and animals were sacrificed either as tumors appeared or at 3 to 4 months postxenografting. To assess the long-term tumorigenic potential, newly formed tumors obtained from transplanted mice were excised, enzymatically digested, depleted of nonviable cells, and subsequently injected into secondary recipient mice. The injection of primary nondissociated tumors resulted in the formation of xenografts in 8 of 31 transplanted mice (Supplementary Table S7; Fig. 5A and B). We did not detect any difference in tumor characteristics, such as CD133/1 expression levels, histotype, grade, or FIGO stage, when comparing endometrial cancers that initiated tumor formation in vivo with those that did not. Xenograft tumors were next separated into CD133⁺ and CD133⁻ fractions that were independently injected into secondary NOD/SCID mice (Fig. 5C). The purity of each fraction was assessed by FACS analysis before in vivo cell transfer. Although both CD133⁺ and CD133⁻ tumor-derived cells were capable of initiating tumor growth in vivo as evaluated by tumor size (Fig. 5D) and time of tumor appearance (Supplementary Table S4), the percentage of CD133-expressing cells in secondary transplants was higher compared with primary xenografts, suggesting in vivo enrichment in CD133⁺ tumor-initiating cells (Fig. 5C). Xenografted tumors expressed CK-7, recapitulated the histologic features of primary tumors, and formed sphere-like structures in vitro, as shown in Fig. 5E and F. It should be noted that, at variance with xenograft tumor-derived CD133⁺ cells, freshly dissociated CD133⁺ cells were incapable of initiating tumor development in vivo (Supplementary Table S7; number of transplanted mice, 11), likely due to requirement for accessory cell types contained within the CD133⁻ fraction.

Gene expression profile of tumor-derived CD133⁺ cells. CD133⁺ cells for gene expression profiling were magnetically isolated from seven endometrioid carcinomas and from three umbilical cord blood samples. Cell-derived RNAs were pooled and used for the identification of differentially expressed genes with topic-defined PIQOR microarrays (Miltenyi Biotec). Among the 40 genes preferentially expressed by CD133⁺ cells compared with the CD133⁻ counterpart, the strongest enrichment significance was obtained for the 33 genes listed in Supplementary Table S3. Also, the entire list of genes more highly expressed in CD133⁻ cells compared with CD133⁺ cells is provided as Supplementary Table S4. To apply more stringent criteria, we excluded from the analysis any gene with a <2.0-fold expression difference. The results have been depicted in Fig. 6A.

Metalloproteases, a class of proteolytic enzymes involved in the degradation of extracellular matrix, were among the top discriminatory genes. Specifically, MMP12 (macrophage elastase) and MMP9 (gelatinase B) were preferentially expressed by CD133⁺ cells. The CD44 gene was up-regulated in endometrial-derived CD133⁺ cells, in accordance with the recently reported expression of CD44 on breast and colorectal CSC (37, 38). Similarly, CXCR4, a master regulator of trafficking of both normal and CSC (39, 40), and proangiogenetic interleukin-8 (IL-8) were preferentially found on endometrial-derived CD133⁺ cells. Conversely, tumor-derived CD133⁻ cells preferentially expressed a set of 15 genes, most of which belonged to the Gene Oncology categories “extracellular matrix” and “collagens” (Supplementary Table S4). Other up-regulated genes were contained within the Gene Oncology category “TGF-β–inducible genes” and included the Jun family member JunB.

Not unexpectedly, umbilical cord blood CD133⁺ progenitor cells preferentially expressed a set of 33 genes mostly belonging to the Gene Oncology categories “CD normal” and “immune system process,” including sialomucin (CD164) and other antigens associated with human hematopoietic stem cells, such as CD34, CD33, and FLT3 (Fig. 6B; Supplementary Table S5). Furthermore, transforming growth factor-β1 (TGF-β1) was highly expressed, in accordance with previous findings on the role of autocrine TGF-β in the regulation of stem cell proliferation and differentiation (34). Compared with CD133⁺ hematopoietic progenitors, endometrial cancer-derived CD133⁺ cells expressed a set of 81 genes, including components of the extracellular matrix and basal lamina, matrix metalloproteases, and TGF-β–regulated genes, such as IL-6 and JunB, and proangiogenetic molecules, such as IL-8, CYR61, and VEGFA (Fig. 6B; Supplementary Table S6). Collectively, endometrial cancer-derived CD133⁺ cells displayed a peculiar gene expression profile that was distinct from that pertaining both to CD133⁻ tumor cells and to CD133⁺ normal hematopoietic progenitors.

To validate microarray data, real-time quantitative PCR for a group of 13 differentially expressed genes was done as detailed in Supplementary Materials and Methods. The list of oligonucleotide primers used for quantitative PCR experiments is presented as Supplementary Table S8. In addition, the expression of CXCR4 on CD133/1⁺ cells was confirmed at the protein level by flow cytometry. The results of validation experiments are shown in Supplementary Fig. S1 and were fully concordant with microarray data.

Correlation between CD133 expression and clinical features. We finally attempted to correlate the expression of CD133/1 antigen with other potentially clinically relevant features, including histologic grade, FIGO clinical stage, CA-125 levels, and presence of metastatic lymph nodes. Endometrioid and nonendometrioid tumors did not differ significantly in the percentage of CD133-expressing cells (data not shown). In addition, no statistically significant correlation could be found between histologic grade and percentage expression of CD133/1 either in nonendometrioid (data not shown) and in endometrioid tumors (Fig. 6C), maybe attributable to the low number of grade 1 samples in our patient cohort. Of potential interest, higher percentages of CD133/1-expressing tumor cells were found in patients with early-stage endometriod tumors compared with those having FIGO III to IV stage disease (Fig. 6C). In accordance with this finding, a higher expression of CD133/1 was evident in patients with no detectable lymph node metastases from endometrioid tumors (Fig. 6C). No such correlations were found in patients with nonendometrioid tumor types (data not shown). Finally, no differences in percentage of CD133 expression were detected when patients with endometrioid tumors were dichotomized based on high CA-125 serum levels (>35 IU/mL; median percentage of CD133 expression, 10.0%; range, 1-45.14%) and low CA-125 serum levels (<35 IU/mL; median percentage of CD133 expression, 11.9%; range, 1.1-53.9%; P = not significant).

The concept that tumors may contain cell populations with stem/progenitor properties was initially suggested by clonogenic assays, indicating that a small proportion of the cells isolated from primary tumors manifest a high proliferative capacity. Much of our knowledge of CSC biology has come from experiments in normal and malignant hematopoiesis, which led to the identification of leukemia stem cells possessing a CD34⁺CD38⁻ cell surface phenotype and the ability to reproduce the hematopoietic hierarchy upon xenotransplantation (41, 42). More recently, rare populations of cancer-initiating cells have been identified in other cancers of hematopoietic origin (6, 43), as well as in lung (14), brain (15), breast (37), colon (20), hepatocellular (44), ovarian (45), and prostate cancers (8). Importantly, large numbers of primary tumor cells must be injected into immunocompromised mice to obtain tumor formation, indicating that only a minority of cancer cells are tumorigenic (20). The development of treatments that selectively target CSC is of paramount importance and is constrained by several challenges. Firstly, CSC might largely reside in a quiescent state, might express higher levels of members of the ATP-binding cassette transporter family, and might be less sensitive to chemotherapeutic agents than their normal counterpart. Secondly, residual drug-resistant CSC might initiate cancer relapse, thus affecting the long-term outcome of neoplastic diseases.

Criteria have been proposed to define CSC, including self-renewal, restriction to a minority of the total tumor population, reproducible tumor phenotype, multipotent differentiation into nontumorigenic cells, and expression of distinctive cell surface markers, allowing consistent isolation (46). Although the surface phenotype of CSC remains elusive, it is now clear that tumorigenic breast cancer cells may express a CD44⁺CD24⁻ lineage⁻ phenotype (37) and that CD133 may serve as marker of tumor-initiating cells in a variety of human cancers, thus raising the possibility that CSC may be therapeutically targeted through these markers (9). Also, prostate CSC have been identified and isolated based on the expression of the α₂β₁ integrin (47). CD133 has been recently detected in glandular and luminal epithelia, but not in stromal, cells from normal endometrial tissues using the AC141 monoclonal antibody (21). The present study reports for the first time that CD133/1 is expressed by primary human endometrial tumors and that CD133/1⁺ tumor cells lack expression of other endothelial and hematopoietic stem cell–associated antigens. The extent of CD133/1 expression was highly heterogeneous in our patient cohort, ranging from 1.3% to >60% of the dissociated tumor cells. CD133/1⁺ tumor cells were capable of self-renewal, could be propagated in culture for up to 12 weeks, and formed sphere-like structures under appropriate culture conditions. Furthermore, CD133⁺ tumor cells were found to express functional ER. This observation is backed by previous studies on breast cancer, a hormone-dependent malignancy wherein two different populations of stem cells have been identified: a long-lived, quiescent and steroid receptor-negative stem cell and a short-lived, less quiescent and steroid receptor-positive stem cell (48). It is currently believed that, rather than being terminally differentiated cells, ER⁺ cells may represent slowly dividing progenitor cells capable of generating an ER⁻ transiently amplifying population in the presence of estrogen (49). In our study, estrogen stimulation of endometrial CD133⁺ cells activated mitogenic signals, resulting in cell growth. These data suggest that the tumorigenic component of endometrioid tumors may be sustained by ER⁺ CSC and that hormonal intervention with antiestrogens may constrain the expansion of ER⁺ progenitors and lead to shrinkage of the tumor. We also found that CD133⁺ tumor cells stain positively for the Tn-MUC1 glycoform, a well-characterized tumor-associated antigen (33, 50). The coexpression of CD133 and MUC1 on the surface of endometrial tumor cells is expected to render them susceptible to immune system attack and might help design novel immunotherapies for this malignancy. CSC have been reported to exhibit a chemoresistant phenotype, probably accounting for insufficient eradication of tumor progenitors and tumor regrowth (45). In our study, tumor-derived CD133⁺ cells were significantly less sensitive to cisplatin and paclitaxel in vitro when compared with the CD133⁻ counterpart. The possibility of obtaining a virtually unlimited number of CD133-derived endometrial tumor cells has potential implications for the in vitro evaluation of drug sensitivity, and the availability of cells closely resembling the original tumor will allow the identification of novel drugs for individualized anticancer therapies.

When assessing the gene expression profile of tumor-derived CD133⁺ cells compared with CD133⁻ cells, endometrial cancer-derived CD133⁺ cells preferentially expressed a set of 30 genes, including metalloproteases, CD44, CXCR4, and IL-8. Relative to umbilical cord blood CD133⁺ cells, endometrial cancer-derived CD133⁺ cells up-regulated a set of 81 genes, including extracellular matrix and basal lamina components, matrix metalloproteases, TGF-β–regulated genes, such as IL-6 and JunB, and proangiogenetic molecules, such as IL-8, CYR61, and VEGFA. These experiments indicated that the gene expression profile of CD133⁺ tumor cells differed from that of both cancer CD133⁻ cells and CD133⁺ hematopoietic progenitors.

The definition of CSC ultimately relies on in vivo xenograft assays, although there is now increasing recognition of the inherent difficulties imposed by the nature of the xenograft model (51). We were able to show tumor growth in 8 of 31 xenografted NOD/SCID mice. This may reflect a residual ability of the murine immune system to reject human tumor cells and/or an inhospitable murine milieu, which may not be fully supportive for human cells. In accordance with previous reports on metastatic colon cancer (52) and glioblastoma multiforme (53, 54), tumor-initiating cells were found both in CD133⁺ and in CD133⁻ fractions from endometrial tumors that were passaged in vivo into NOD/SCID mice. Although CD133⁺ and CD133⁻ fractions grew at apparently similar rates in NOD/SCID mice, as suggested by mean tumor size and time of tumor appearance, secondary transplantation of endometrial cancer cells was associated with an increase in the percentage of CD133-expressing cells, indicating in vivo enrichment of CD133⁺ cells and/or further acquisition of the CD133 antigen during tumor progression. As expected, xenografted tumors recapitulated the histology of the primary tumor and maintained the ability to grow as nonadherent sphere-like structures. In sharp contrast to xenograft tumor-derived CD133⁺ cells, freshly dissociated CD133⁺ cells were incapable of supporting tumor formation in vivo, underscoring the need for accessory cell types contained within the CD133⁻ tumor fraction in cancer initiation and progression (55).

In our study, CD133 was preferentially expressed by early-stage tumors, as defined by FIGO stage and absence of lymph node metastases. This may indicate that selective targeting of CD133-expressing cancer cells has a greater potential to eradicate the neoplastic disease and to limit disease recurrence in this subgroup of patients. Interestingly, the down-regulation of CD133 expression in colon cancer cells has been associated with the acquisition of metastatic potential, pointing to CD133⁻ colon cancer cells as genetically unstable derivatives of CD133⁺ colon cancer cells with a remarkable ability of forming tumors in immunodeficient mice (52, 56). In line with this observation, it is tempting to speculate that the lower expression of CD133/1 that we detected in metastatic endometrial cancers might reflect the emergence of CD133⁻ tumor cells with a clinically more aggressive behavior. A recent study in 20 patients with glioblastoma multiforme has shown that CD133^low glioblastomas are particularly aggressive, as determined by invasiveness, tumor multiplicity, and ventricle involvement on magnetic resonance imaging scans, and have higher disease progression rates after chemotherapy and radiotherapy compared with CD133^high glioblastomas (53). Collectively, we provide evidence that CD133⁺ cells isolated from human endometrial cancers can be extensively propagated in vitro and that they coexpress a tumor-specific glycoform of the MUC1 antigen. This study sheds light into the biology of human endometrial tumors and points to CD133 as a potentially useful antigen for the identification of endometrial cancer cells with a unique gene expression profile and that exhibit stem cell characteristics (57).

No potential conflicts of interest were disclosed.

We thank Prof. Raimondo De Cristofaro (Department of Medicine and Geriatrics, Catholic University Medical School) for critically reading the manuscript and for useful suggestions.