Authors contributions: H.Y.O and T. Yamazaki: conception and design, data analysis and interpretation, and writing manuscript, and final approval of manuscript; S.K., M.S., O.N., A.K., K.M., and T.O.: collection and assembly of data and data analysis and interpretation; Y.J.C., M.Y., E.H., Y.W., H.M., M.A., C.K., and T.W.: collection and assembly of data; T. Yoshikubo, N.T., and M.K.: data analysis and interpretation; S.F. and K.Y.: provision of study material (new antibodies); A.J.L.: other (support of manuscript). S. K. and H.Y.O. contributed equally to this article.
Disclosure of potential conflicts of interest is found at the end of this article.
First published online in STEM CELLSEXPRESS November 7, 2012.
The cancer stem cell (CSC) concept has been proposed as an attractive theory to explain cancer development, and CSCs themselves have been considered as targets for the development of diagnostics and therapeutics. However, many unanswered questions concerning the existence of slow cycling/quiescent, drug-resistant CSCs remain. Here we report the establishment of colon cancer CSC lines, interconversion of the CSCs between a proliferating and a drug-resistant state, and reconstitution of tumor hierarchy from the CSCs. Stable cell lines having CSC properties were established from human colon cancer after serial passages in NOD/Shi-scid, IL-2Rγnull (NOG) mice and subsequent adherent cell culture of these tumors. By generating specific antibodies against LGR5, we demonstrated that these cells expressed LGR5 and underwent self-renewal using symmetrical divisions. Upon exposure to irinotecan, the LGR5+ cells transitioned into an LGR5− drug-resistant state. The LGR5− cells converted to an LGR5+ state in the absence of the drug. DNA microarray analysis and immunohistochemistry demonstrated that HLA-DMA was specifically expressed in drug-resistant LGR5− cells, and epiregulin was expressed in both LGR5+ and drug-resistant LGR5− cells. Both cells sustained tumor initiating activity in NOG mice, giving rise to a tumor tissue hierarchy. In addition, anti-epiregulin antibody was found to be efficacious in a metastatic model. Both LGR5+ and LGR5− cells were detected in the tumor tissues of colon cancer patients. The results provide new biological insights into drug resistance of CSCs and new therapeutic options for cancer treatment. STEM CELLS 2012;30:2631–2644
Tumors arise from normal tissues by the progression of multiple mutations resulting in malignant cells. The origin of the cells harboring these mutations, whether stem cells (SCs), progenitor cells, or mature differentiated cells, remains unclear. The heterogeneity of tumor cell types and the prevalence of drug resistance have led to the hypothesis for the existence of cancer stem cells (CSCs), although this theory is still an ongoing debate [1–7].
Evidence for the existence of colon CSCs has been the most convincing, with LGR5-positive (LGR5+) cells of particular interest in CSC studies [8–12]. Lgr5, a Wnt target gene, was first identified as a marker for normal SCs in the intestine . It was also reported that Lgr5-positive (Lgr5+) cells formed adenomas upon deletion of Apc and that Lgr5 is expressed in colon cancer cell lines . The cells with high Wnt activity, thereby rendering them LGR5+, are functionally designated colon CSCs . Clearly, LGR5 is an important molecule to identify colon CSCs.
In the normal intestine, Tian et al.  described that Lgr5-negative (Lgr5−) SCs serve as a reserve population of Lgr5+ cells that are themselves therefore dispensable for normal small intestine cell reproduction. It was also reported that slow cycling SCs positive for Hopx are present at the position 4 (the SC crypt), and that there is an interconversion between Hopx+ slow cycling SCs and Lgr5+ proliferating SCs located at the crypt base . Similarly in CSCs, several reports suggest the existence of distinct states of CSCs [17–21]. However, it remains unknown how proliferating CSCs acquire a drug-resistant phenotype and whether interconversion between proliferating and slow cycling/quiescent CSCs occurs. Difficulties in investigating CSCs are due to the heterogeneity of cell types and the rare presence of CSCs in cancer tissues. Many attempts have been made to enrich and isolate CSCs by spheroid cultures in vitro, cell sorting with CSC markers, and direct xenotransplantation of cancer cells to immunodeficient mice [22–30]. Although spheroid cultures enrich CSCs, they result in heterogeneous populations of cells and are not efficient enough to isolate and maintain pure CSC populations .
Here we report the establishment of human colon cancer cell lines that express LGR5 and possess CSC properties. The cell lines were created using serial passages of colon cancer cells in xenotransplantion in NOD/Shi-scid, IL-2Rγnull (NOG) mice followed by adherent culture of cells. For this purpose, we generated antibodies that are specific to LGR5. The obtained LGR5+ cells transitioned to LGR5-negative (LGR5−) cells upon exposure to an anticancer drug, and such LGR5− cells reverted to LGR5+ cells after re-seeding and culturing without an anticancer drug. By gene expression profiling of the cell lines, we demonstrated that HLA-DMA, which belongs to the HLA class II alpha chain paralogs, is expressed in drug-resistant LGR5− cells, and epiregulin (EREG), a member of the epidermal growth factor family, which can function as a ligand of epidermal growth factor receptor and most members of the ErbB family of tyrosine-kinase receptors, is expressed in both proliferating LGR5+ and drug-resistant LGR5− cells. Using antibodies against LGR5, HLA-DMA, and EREG, we show the existence of LGR5+ and LGR5− cells in xenotransplanted tumor tissues and in human colon cancer tissues from patients. Furthermore, the anti-EREG antibody exhibited antitumor activity against tumors derived from the LGR5+ cells in a metastatic model. This is the first demonstration of the establishment of stable cell lines having CSC properties and the ability to transition between the two distinct states, a proliferating and a drug-resistant state. Thus, LGR5+ colon CSCs interconvert with drug-resistant LGR5− cells and are capable of tumor reconstitution. This suggests the physiological importance of CSCs in tumor recurrence after drug treatment. Further, using the anti-EREG antibody, we provide an option for CSC targeting therapy.
MATERIALS AND METHODS
Preparation of Monoclonal Antibodies Against LGR5
Anti-LGR5 monoclonal antibodies, 2L36 and 2U2E-2, were obtained by DNA immunization and protein immunization, respectively. For DNA immunization, plasmid DNA containing LGR5 was transferred once a week six times to the abdominal skin of 6-week-old female MRL/lpr mice (MRL/MpJ-Tnfrsf6<lpr>/Crlj) (Charles River Japan, Yokohama, Japan, http://www.crj.co.jp) using a Helicos Gene Gun (Bio-Rad, Hercules, CA, http://www.bio-rad. com) at a pressure of 200–300 psi. At the final immunization, 1 × 106 cells of CHO DG44 (Life Technologies, Rockville, MD, http://www.lifetech.com) expressing LGR5 were intravenously injected. The splenocytes were resected 3 days after the final immunization and fused with P3-X63-Ag8U1 mouse myeloma cells (ATCC, Manassas, VA, http://www.atcc.org). 2L36 was obtained by screening the culture supernatants of hybridoma by flow cytometry .
The N-terminal region of LGR5 (amino acid 1–555) was expressed as a fusion protein with the Fc region of mouse IgG2a in CHO DG44 cell. The LGR5-Fc protein secreted in the culture medium was purified with HiTrap Protein A FF column (GE Healthcare, Little Chalfont, United Kingdom, http://www.gehealthcare.com), and then 6-week-old female BALB/c mice (Charles River Japan) were immunized subcutaneously with 50 μg of the LGR5-Fc protein emulsified in Freund's Complete Adjuvant (Becton Dickinson, Franklin Lakes, NJ, http://www.bd.com). Immunization was repeated once a week for 2 weeks with the same amount of the LGR5-Fc protein in Freund's Incomplete Adjuvant (Becton Dickinson). Three days before cell fusion, mice were injected intravenously with 25 μg of the LGR5-Fc protein. Hybridomas were generated as described above, and the antibody 2U2E-2 was selected by ELISA with the LGR5-Fc protein.
Establishment of Human Colon Cancer Xenografts Using NOG Mice
Colon cancer specimens were obtained from consenting patients, as approved by the ethical committee at PharmaLogicals Research and Parkway Laboratory Services in Singapore. Pieces of tumors were minced by scissors and implanted into the flank of NOG mice (Central Institute for Experimental Animals, Kawasaki, Japan, http://www.ciea.or.jp). The human colon cancer xenografts were maintained by passages in NOG mice. All studies and procedures involving animal subjects were approved by the Animal Care and Use Committee at PharmaLogicals Research and the Institutional Animal Care and Use Committee at Chugai Pharmaceutical Co., Ltd. The animals used in this experiment were treated in accordance with the Animal Research Guideline of PharmaLogicals Research and the Guidelines for the Care and Use of Laboratory Animals at Chugai Pharmaceutical Co., Ltd.
Establishment of Colon Cancer Cell Lines with CSC Properties
Single cell suspension of cancer cells from the xenografts was prepared by mincing the tissues with scissors, incubated in Dulbecco's phosphate buffered saline (DPBS) containing collagenase/dispase (Roche, Basel, Switzerland, http://www.roche-applied-science.com) and DNase I (Roche) at 37°C for 3 hours followed by filtrating 40 μm cell strainer (BD Biosciences, San Diego, CA, http://www.bdbiosciences.com) and suspending in lysing buffer (BD Biosciences). The cells were cultured in a SC medium [Dulbecco's modified Eagle's medium/F12 medium (Life Technologies) supplemented with N-2 supplement (Life Technologies), 20 ng/ml human epidermal growth factor (Life Technologies), 10 ng/ml human basic fibroblast growth factor (Sigma-Aldrich, St. Louis, MO, http://www.sigmaaldrich.com), 4 μg/ml heparin (Sigma-Aldrich), 4 mg/ml bovine serum albumin (BSA) (Life Technologies), 20 μg/ml human insulin, zinc solution (Life Technologies), and 2.9 mg/ml glucose (Sigma-Aldrich)] at 37°C under 5% CO2 . Culture flasks treated with polystyrene (BD Biosciences) and ultra-low-attachment cell culture flasks (Corning Life Sciences, Acton, MA, http://www.corning.com/lifesciences) were used for adherent cultures and the spheroid cultures, respectively. Drug-resistant LGR5− cells were obtained by treating the adherent LGR5+ cells with 10 μg/ml irinotecan (Hospira, Lake Forest, IL, http://www.hospira.com/) for 3 days.
Sorting of the LGR5+ and LGR5− Cells
The primary cells from xenografts were incubated with the anti-LGR5 antibody (2L36, 2 μg/ml) and then R-phycoerythrin (PE)-labeled anti-mouse IgG2a (Life Technologies, 1/200 dilution). Mouse cells were discriminated from the human colon cancer cells by staining with anti-mouse major histocompatibility complex (MHC) class I antibody (Abcam, Cambridge, U.K., http://www.abcam.com, 0.1 μg/ml) and allophycocyanin (APC)-labeled anti-rat IgG (BioLegend, San Diego, CA, http://www.biolegend.com, 1/100 dilution). Anti-CD133 antibody (Miltenyi Biotec, Bergisch Gladbach, Germany, http://www. miltenyibiotec.com, 5 μg/ml) and Alexa 488-labeled anti-mouse IgG1 (Life Technologies, 1/100 dilution) were used to detect CD133. Dead cells were removed by 7- aminoactinomycin D (7-AAD) viability dye (Beckman Coulter, Brea, CA, http://www.beckmancoulter.com). Flow cytometry analysis and cell sorting were performed using a MoFlo XDP (Beckman Coulter) cell sorter.
In Vitro Colony Formation Assay
To test the colony formation ability, cells were seeded on a layer of 100% Matrigel (BD Biosciences) at 10,000 cells per well and cultured in a SC medium supplemented with 10% heat-inactivated fetal bovine serum (FBS) and 5% Matrigel.
Tumor Formation In Vivo
Cells suspended in Hank's balanced salt solution (Life Technologies) with 50% Matrigel were subcutaneously inoculated into the flank of NOG mice. For single cell inoculation, cells were stained with fluorescein isothiocyanate (FITC)-labeled anti-EpCAM antibody (Miltenyi Biotec) and seeded in Terasaki plates (Thermo Fisher Scientific, Waltham, MA, http://www.thermofisher.com). After the presence of single cell in each well was confirmed under a fluorescence microscope, the single cell in 50 μl of 50% Matrigel was inoculated into the flank of mice. Estimated CSC density was calculated by the formula available on the WEHI ELDA website .
Small pieces of surgical specimens of human tissues and of the xenograft tumor tissues were fixed with 4% paraformaldehyde at 4°C for 16–24 hours and embedded in paraffin by the AMeX method [34, 35]. After washing the in vitro cultured cells with phosphate buffered saline (PBS)-EDTA, the cells were fixed with 4% paraformaldehyde at 4°C for 2 hours, suspended in 0.5 ml agarose, and embedded in paraffin with AMeX method. Thin sections were subjected to hematoxylin & eosin staining and to immunohistochemisty.
Thin sections from the above-mentioned paraffin blocks were incubated with anti-LGR5 antibody (2U2E-2, 1 μg/ml), anti-EREG antibody (10 μg/ml), anti-E-cadherin antibody (Abcam, 2.5 μg/mL), anti-HLA-DMA antibody (Sigma-Aldrich, 2.5 μg/ml), or FITC-labeled anti-Ki67 antibody (Abcam, 2.5 μg/mL). After the incubation with the primary antibodies, the sections were incubated with a secondary antibody conjugated with polymer-horseradish peroxidase (HRP) (DAKO, Glostrup, Denmark, http://www.dako.com or Vector Laboratories, Burlingame, CA, http://www.vector labs.com) or biotin, and the proteins were visualized by AlexaFluor 488-labeled tyramide (Life Technologies, 1/100 dilution), AlexaFluor 568-labeled tyramide (Life Technologies, 1/100 dilution), or AlexaFluor 568-labeled streptavidin (Life Technologies, 2 μg/ml). For immunofluorescent cytochemistry, cells were fixed with 4% paraformaldehyde and permeabilized with 0.1% Triton-X 100 (Sigma-Aldrich), and incubated with anti-LGR5 antibody (2L36, 2 μg/ml). After the incubation with the primary antibodies, the cells were incubated with AlexaFluor 488-labeled anti-mouse IgG (Life Technologies, 1/100 dilution). Those specimens and cells were also stained with DAPI (Life Technologies).
Induction of the Transition Between LGR5+ and LGR5− States in Single Cell Culture
LGR5+ cells were sorted with an anti-LGR5 antibody, and single LGR5+ cells were cultured in 96-well microplates. To obtain drug-resistant LGR5− cells, LGR5+ cells were treated with 10 μg/ml irinotecan for 3 days. Single LGR5− cells were cultured in 96-well microplates for 4 days. The medium used for the single cell culture contained 10% conditioned medium of the in vitro cultured LGR5+ cells under an adherent condition. LGR5+ and LGR5− states of the cells were confirmed by immunocytochemical analysis with anti-LGR5 antibody.
Determination of Antitumor Activity of Anti-EREG Antibody In Vivo
2 × 106 of LGR5+ cells were suspended in Hank's balanced sodium solution and intravenously injected into the tail vein of Fox Chase severe combined immunodeficiency (SCID) Beige Mouse (CB17.Cg-PrkdcscidLystbg/Crl, Charles River Japan). For treatment with the anti-EREG antibody, the mice were intravenously administered 10 mg/kg of anti-EREG antibody once a week for five times starting 3 days after tumor inoculation. Mice were scarified 5 days after the final administration under deep anesthesia, and lung tissues were collected. The lung tissues trimmed into 11 pieces were fixed in 4% paraformaldehyde for 24 hours, paraffin embedded by AMeX method [34, 35]. After thin sections were prepared and stained with hematoxylin & eosin, the number of tumors was counted. The sizes of the tumors were determined under a microscope with micrometer.
The Mann–Whitney U test was applied to determine the statistical significance of the differences in the numbers of tumor nodules in a metastatic tumor model. The statistical analysis was carried out with an SAS preclinical package (SAS Institute, Inc., Cary, NC, http://www.sas.com). p values smaller than 0.05 were considered significant.
Generation and Characterization of Specific Antibodies Against LGR5
Having an antibody specific to LGR5 is critical to isolate and characterize colon CSCs, but such antibody has not been available yet. Therefore, we first attempted to generate anti-LGR5 antibodies that enable us to isolate and analyze cells having colon CSC properties. Two monoclonal antibodies, 2L36 and 2U2E-2, specific to LGR5 were obtained. The regions of the LGR5 protein that contain epitopes of these antibodies are shown in Figure 1A. Both antibodies were tested for immunohistochemistry and flow cytometry using the CHO cells expressing highly related proteins LGR4, LGR5, or LGR6. When used for immunostaining, both 2L36 and 2U2E-2 recognized CHO cells expressing LGR5 but not those expressing LGR4 or LGR6 (Fig. 1B). In flow cytometry analysis, only 2L36 strongly reacted with CHO cells expressing LGR5 (Fig. 1C). Moreover, the antibody 2U2E-2 reacted specifically with crypt base columnar cells in the normal human intestine (Fig. 1D; Supporting Information Fig. S1A). There was also a good correlation between mRNA expression and cell surface staining of the anti-LGR5 antibody in human colon cancer cell lines, which included the CSC lines established in this study and six commercially available lines (Supporting Information Fig. S1B).
Establishment of Human Colon Cancer Cell Lines with CSC Properties
We established 11 human colon cancer xenografts using NOG mice . Ten out of 11 xenografts were derived from moderately differentiated colon cancer, and one was from poorly differentiated colon cancer. Both the moderately differentiated colon cancer xenografts and the poorly differentiated colon cancer xenograft reconstituted almost the same histological morphologies as the original tumors; the moderately differentiated colon cancer xenografts formed clear epithelial ducts and small budding clusters. In contrast, the poorly differentiated colon cancer xenograft showed no clear epithelial duct structure. We used two moderately differentiated colon cancer xenografts, namely PLR59 and PLR123, for the establishment of colon CSC lines. PLR59 and PLR123 were heterozygous for the mutant K-Ras (G12D), and PLR123 carried the mutant p53 (R249M) in one allele. These xenografts were chosen because they grew faster while retaining the ability to reconstitute tumors with epithelial ducts and small budding clusters even after 10 passages in NOG mice (Fig. 1E). In the epithelial ducts of the tumors, differentiated cancer cells that showed goblet cell-like phenotype were also observed in the xenotransplanted tumor tissues throughout the passages (Fig. 1E, inset).
To confirm the existence of CSCs in the xenotransplanted tumor tissues, we used immunohistochemical staining for the LGR5 protein that marks colon CSCs. LGR5+ cells were detected in the original tumor tissues of PLR59 and PLR123 and in their xenotransplanted tumor tissues throughout the passages (Fig. 1F). The frequency of LGR5+ cells in the original tumor tissues was quite low: it was approximately 0.01% in PLR59 and approximately 0.04% in PLR123. In the xenotransplanted tumor tissues, the frequency of LGR5+ cells increased during the passages (Fig. 1F). Tumor initiating activity (TIA) of the primary cells from the PLR123 xenografts was also increased after the passages. The estimated percentage of CSC in the primary cells, as judged from TIA, was approximately 0.1% after five passages, and after 14 passages it increased to approximately 0.4% (Supporting Information Table S1). Schematic representation of the establishment of the colon cancer cell lines is shown in Figure 2A.
CSC Properties of the Established Colon Cancer Cell Lines
The major properties of CSCs are self-renewal, TIA, and the reconstitution of a tumor tissue hierarchy of differentiated cells. In an attempt to establish cell lines possessing CSC properties, we used spheroid and adherent cultures of the cells derived from PLR59 and PLR123 xenografts in which LGR5+ cells were enriched (over 10 passages). When cells derived from PLR59 and PLR123 were cultured as spheroids, their growth was rather slow, and the spheroids contained only a few LGR5+ cells but more differentiated cells that were positive for CK20, which is a commonly used differentiation marker (Supporting Information Fig. S2). On the contrary, cells from PLR59 and PLR123 cultured under an adherent condition grew fast with a doubling time of approximately 2.5 days and showed epithelial morphology (Fig. 2B).
To examine TIA of the cells, subcutaneous injection of 10 cells from the spheroids formed tumors in one (PLR59-derived cells) or two (PLR123-derived cells) out of six injection sites (Supporting Information Table S2), whereas 10 cells from adherent cultures formed tumors in all six injection sites, and even a single cell injection of an adherent cell reconstituted tumors. Although the spheroid culture led to an increase in TIA as compared to that of the primary cells, the adherent culture was more efficiently enriched in cells possessing TIA. The histological morphology of the tumors from the adherent cells was almost the same as the original tumors (Fig. 2C). In addition, TIA of the adherent cells was maintained even after the cells were cultured for more than a month (Supporting Information Table S3).
We examined cell surface markers of the adherent cells from the PLR59 and PLR123 and found that they were clearly positive for all known colon CSC markers reported earlier: LGR5+, ALDH+, CD133+, CD44+, EpCAM+, CD166+, CD24+, CD26+, and CD29+ (Fig. 2D; Supporting Information Fig. S3). In addition, expression of the cell surface markers was unchanged even after 1 month of cell culture. One of the characteristics of CSCs is symmetrical cell division for self-renewal. The LGR5+ adherent cells divided symmetrically under the adherent culture conditions (Fig. 2E). In the presence of Matrigel and FBS, however, the LGR5+ cells underwent asymmetrical cell divisions, as demonstrated by the segregation of LGR5 protein into one of two daughter cells (Fig. 2F), implicating the generation of two different offspring. Asymmetric cell divisions are one of the hallmarks of SCs.
Colony Forming Activity and Tumorigenicity of the Sorted LGR5+ and the LGR5− Cells
In order to examine the ability of LGR5+ and LGR5− cells to form colonies in vitro and tumors in vivo, we sorted the LGR5+ and LGR5− populations from the primary cells of xenografts generated by the inoculation of the LGR5+ cells. Anti-LGR5 antibody 2L36 was used for the cell sorting. About 93% of the cells in the LGR5+ fraction were LGR5+, and more than 99% of the cells in the LGR5− fraction were LGR5− (Fig. 3A). The sorted LGR5+ cells but not the LGR5− cells efficiently formed colonies on Matrigel in vitro and formed tumors in NOG mice. When 1,000 cells were subcutaneously injected into NOG mice, the sorted LGR5+ cells formed large visible tumors by day 34 after the inoculation, but the LGR5− cells gave rise to only very tiny tumors by day 34 (Fig. 3B). We further examined the relation of LGR5 expression and other CSC markers by double staining the LGR5 with CD133, CD166, or CD44. Nearly all of the LGR5+ cells were positive for CD133 and CD166, but there were large numbers of LGR5− cells that were positive for CD133 or CD166, indicating that LGR5 marks a subpopulation of CD133+ and CD166+ cells (Fig. 3C). Because significant numbers of LGR5+/CD44− cells were present, CD44 does not mark all the LGR5+ cells (Fig. 3C).
We used cell sorting to further characterize the LGR5− cell populations. The cells from the xenografts were stained with the anti-LGR5 and anti-CD133 antibodies, and the LGR5−/CD133−, LGR5−/low/CD133+, and LGR5+/CD133+ cells were separated. More than 90% of the cells in each fraction were LGR5−/CD133−, LGR5−/low/CD133+, and LGR5+/CD133+ (Fig. 3D). The isolated LGR5+/CD133+ and LGR5−/low/CD133+ cells formed colonies on Matrigel, whereas nearly all the LGR5− /CD133− cells died after seeding on culture plates; colony forming efficiency of the sorted LGR5− /CD133−, LGR5−/low/CD133+, and LGR5+/CD133+ cells were about 0.03%, 1.6%, and 4.3%, respectively (Fig. 3E; Supporting Information Fig. S4).
Interconversion Between LGR5+ Proliferating and LGR5− Drug-Resistant States
We next asked whether the LGR5+ cells exhibited a drug-resistant state, which is believed to be a typical characteristic of CSCs . After treatment of the LGR5+ cells with irinotecan for 3 days, the cells stopped proliferation and about half of the cells survived (Supporting Information Fig. S5). One hundred percent of the surviving cells became LGR5−, but they retained other colon CSC markers (Fig. 4A, 4B; Supporting Information Fig. S6, Table S4). The LGR5− cells induced by treating the LGR5+ cells with an anticancer drug were designated as drug-resistant LGR5− cells in this study. LGR5− cells which were pre-existing in xenograft tissues and human tumor tissues are referred to as LGR5− cells. Reverse transcriptase real-time polymerase chain reaction (RT-qPCR) for LGR5, CD133, CD44, CD166, and EPCAM revealed that the LGR5 mRNA was drastically decreased after irinotecan treatment, but the mRNAs of CD133, CD44, CD166, and EPCAM did not decline or even increased after treatment (Supporting Information Fig. S7D). The mRNA level of CK20 did not increase by the irinotecan treatment and remained at a low level (Supporting Information Fig. S2A). Although we cannot rule out the possibility that elimination of epitope by proteolytic cleavage on LGR5 occurred after irinotecan treatment, the cells that were not recognized by any of our anti-LGR5 antibodies were regarded as LGR5 negative. We examined the TIA of these drug-resistant LGR5− cells. Even injection of the 10 LGR5− cells formed tumors in two sites (PLR59-derived cells), and the PLR123-derived cells formed tumors at one site out of six injection sites in the NOG mice (Supporting Information Table S5). Treatment of the LGR5+ cells with 5-fluorouracil or oxaliplatin also gave rise to drug-resistant LGR5− cells which converted to an LGR5+ state after re-seeding and culturing in the absence of the drugs (Supporting Information Fig. S8).
The drug-resistant LGR5− cells did not grow even after irinotecan was removed from the culture medium (Supporting Information Fig. S5C). However, they became positive for LGR5 and resumed proliferation after replating (Fig. 4A, 4B). The transition from the LGR5+ state to an LGR5− state and vice versa was also confirmed by observations with single cells in culture. When single LGR5+ cells were cultured in multiwell plates, the cells transitioned to an LGR5− state within 3 days after irinotecan treatment. When single LGR5− cells that had been treated with irinotecan were then cultured in multiwell plates without irinotecan, 19%–43% of the cells converted to the LGR5+ state within 4 days (Fig. 4C; Supporting Information Table S6). In order to confirm proliferation of the LGR5+ and drug-resistant LGR5− cells, we also used double staining of LGR5 and Ki67 with the in vitro cultures of the LGR5+ and the LGR5− cells. The expression of LGR5 correlated well with Ki67 staining: the LGR5+ cells were positive for Ki67, and the drug-resistant LGR5− cells were negative for Ki67 (Fig. 4D).
Pathway analysis using the results of DNA microarray of both proliferating LGR5+ and drug-resistant LGR5− cells revealed characteristics of two distinct states. As expected from the growth status of the cells, genes involved in cell cycle were downregulated whereas genes in the p53 signaling pathway were upregulated in the drug-resistant LGR5− cells (Supporting Information Fig. S9). Genes whose mRNA expression increased in the LGR5+ cells included those involved in cell growth such as LGR5, FGFBP1, FGFR4, ROR1, NFIA, PIGU, LPAR3, and FZD2 (Supporting Information Fig. S7C).
Reconstitution of the Epithelial Cell Type Tumor Hierarchy from LGR5+ Cells
The observations that the cells converted from the LGR5+ to the drug-resistant LGR5− state in vitro and that the drug-resistant LGR5− cells formed tumors in vivo prompted us to examine whether the drug-resistant LGR5− cells directly generate a tumor hierarchy of differentiated cell types or first convert to the LGR5+ state in vivo. To detect drug-resistant LGR5− cells, we attempted to identify the genes that are upregulated in the drug-resistant LGR5− cells by comparing the gene expression profiles of the drug-resistant LGR5− cells, the LGR5+ cells, and the primary cells from the xenografts. From the gene expression analyses of DNA microarray, a heat map is shown (Fig. 5A; Supporting Information Fig. S7) for the top 20 genes encoding membrane proteins with the largest change. Genes whose mRNA expression was increased in the drug-resistant LGR5− cells include MHC class II-related genes (HLA-DMA, HLA-DMB), adhesion molecule-related genes (AMIGO2, FLRT3, GJB5, CLDN1), G-protein-coupled receptor protein signaling pathway-related genes (GPR87, GPR110, GPR172B, GNAI1, ABCA1), and immune signaling-related genes (TNFSF15, BLNK, FAS, TMEM173). We further evaluated the genes for which antibodies against their proteins are available (Supporting Information Table S7).
Immunohistochemical staining of the cells cultured in vitro with antibodies confirmed that HLA-DMA was rather specifically expressed in the drug-resistant LGR5− cells (Fig. 5B). HLA-DMA was located in intracellular vesicles, and therefore, it cannot be used for cell sorting. Nevertheless, HLA-DMA can be a useful molecule for identifying the LGR5− cells in xenografts and in clinical specimens. Because HLA-DMA is also expressed in macrophages, we looked for genes that were expressed in both LGR5+ and drug-resistant LGR5− cells and identified EREG (Fig. 5A). Immunohistochemical staining with a monoclonal antibody against EREG confirmed the EREG expression in LGR5+ and LGR5− cells (Fig. 5B; Supporting Information Fig. S10). By combination of these markers, LGR5− cells can be detected as HLA-DMA and EREG double positive cells. After injection of a homogenous population of drug-resistant LGR5− cells into NOG mice, cells weakly expressing LGR5 but still positive for HLA-DMA and EREG appeared within 1 day after the injection, and then the LGR5+/EREG+ cells which were negative for HLA-DMA emerged by day 5 (Fig. 5C). The reconstitution of the epithelial tumor hierarchy of diverse cell types from the drug-resistant LGR5− cells through transition to LGR5+ cells was confirmed (Fig. 5D).
We next examined the possibility of a conversion of LGR5+ cells to a drug-resistant state in vivo. NOG mice bearing tumors derived from the LGR5+ cells were administered intraperitoneally with a maximum tolerated dose (MTD) dose (120 mg/kg) of irinotecan. Tumor growth was nearly completely inhibited (Fig. 5E), and ductal structures were heavily destroyed (Fig. 5F). Under such conditions, the LGR5+ cells were markedly decreased (Fig. 5F; Supporting Information Fig. S11). There was a significant increase in HLA-DMA-positive cells, which are LGR5−, after irinotecan treatment. In contrast, about one-third of the cancer cells in both ducts and budding regions were LGR5+ in the vehicle-treated control mice. Both LGR5+ cells and HLA-DMA+/LGR5− cells were positive for EREG (Fig. 5F). The LGR5+ cells reappeared after termination of irinotecan treatment (Fig. 5F; Supporting Information Fig. S11). We also performed double staining with Ki67 and LGR5 or HLA-DMA that marks LGR5− CSCs. As in the in vitro cultured cells, there was a good correlation of LGR5 expression and Ki67 staining of the cells in the xenografts. The LGR5+ cells were positive for Ki67, and the LGR5−/HLA-DMA+ cells were negative for Ki67 (Fig. 5G). Thus, tumor reconstitution occurred through the LGR5+ cells.
Possible Therapeutic Application of Anti-EREG Antibody
EREG is expressed on the surface of both LGR5+ and drug-resistant LGR5−/HLA-DMA+ cells, but its expression is very low or hardly detectable in differentiated tumor cells and normal tissues (Fig. 5A; Supporting Information Fig. S12A). Therefore, EREG may be a potential therapeutic target. We first examined the growth inhibitory activity and antibody dependent cell cytotoxicity (ADCC) activity of the anti-EREG antibody. The anti-EREG antibody induced ADCC activity against both LGR5+ and drug-resistant LGR5− cells in the presence of human peripheral blood mononuclear cells (PBMC) that contained effector cells, such as natural killer (NK) cells and monocytes, but the antibody did not directly affect the growth of LGR5+ and drug-resistant LGR5− cells in the absence of effector cells in vitro (Supporting Information Fig. S12B, S12C). To test the expression of EREG in vivo, we subcutaneously inoculated the LGR5+ cells into NOG mice in which EREG was highly expressed during early stages of tumor development, and later its expression was rather restricted to budding areas as compared to ducts when tumors formed clear duct structures. EREG-positive cells were also detected after the mice bearing the tumors were administered irinotecan (Fig. 5F). Therefore, we tested the antitumor activity of the anti-EREG antibody after irinotecan treatment. SCID mice were used as a model for efficacy evaluation, because the antibody requires effector cells to elicit ADCC. When the antibody was administered at day 4 and day 11 after the final administration of irinotecan, it only delayed the tumor growth (Supporting Information Fig. S12D).
To test the efficacy in a metastatic model, we first examined the expression of EREG in metastasized tumors. When the LGR5+ cells were intravenously injected into NOG mice, tumors were developed in several organs including the lung. For the tumors that developed in the lung, the majority of tumor cells were EREG positive (Fig. 6A). Efficacy was then tested using SCID-Beige mice in which macrophages and monocytes can serve as effectors for ADCC. When the antibody was administered once a week for five times starting 3 days after the tumor injection, the number of tumors at different sites was significantly decreased as compared to the control mice (Fig. 6B). In addition, the size of each tumor was also markedly reduced in the antibody treated mice (Fig. 6C, 6D).
The Existence of Both LGR5+ and LGR5− Cells in Human Colon Cancers
We asked whether LGR5+ and LGR5− cells could be detected in tissue sections of clinical colon cancers. Although rare, the LGR5+ cells and the LGR5− cells which were HLA-DMA+/EREG+ were present in primary and metastatic colon cancer tissues from patients (Fig. 7A). Among 12 human colon cancer tissues, both LGR5+ cells and LGR5− cells were detected in eight cases, and either LGR5+ or LGR5− cells were observed in the remaining four cases. The percentages of the LGR5+ and LGR5− cells in those cases ranged between 0.003% and 1.864% for the LGR5+ and 0.001%–0.243% for the LGR5− cells (Supporting Information Table S8). Both LGR5+ and LGR5− cells were detected in ducts and budding areas (Fig. 7A; Supporting Information Table S8). In addition, LGR5+ and LGR5− cells within the ducts were not restricted to particular regions as they were observed randomly throughout the ducts (Fig. 7A). In budding areas, LGR5+ cells were detected as a single cell or in tumor cell clusters consisting of a few tumor cells (Fig. 7A). Also in clinical specimens, the LGR5+ cells were positive for Ki67, and the LGR5−/HLA-DMA+ cells were negative for Ki67 (Fig. 7B).
SC markers such as CD133, CD44, CD166, and ALDH have been used to identify and isolate colon CSCs [25–29]. We observed that colon CSCs reside in subpopulations of the cells positive for these markers; however, none is a definitive marker for colon CSCs. Evidence that Lgr5 marks normal intestinal SCs has accumulated [8, 13]. Despite that evidence, LGR5 remains unexplored in human CSCs, presumably due to a lack of specific antibodies . In this study, we generated monoclonal antibodies that are highly specific to LGR5 and can be applied to immunostaining, flow cytometry, and cell sorting. Using these unique anti-LGR5 antibodies, we were able to define LGR5+ cells as proliferating colon CSCs. To establish pure CSC cell lines, we tested whether spheroid cultures or adherent cultures would be useful to enrich CSCs. Several attempts have been made using spheroid cultures to isolate and enrich CSCs in vitro [9, 14]. The results in this study indicated that spheroid cultures allowed LGR5+ cells to self-renew and differentiate, leading to heterogeneous populations of cells as reported by others [14, 39]. In contrast, adherent cultures kept the LGR5+ cells self-renewing and prevented them from differentiation. Indeed, we did not detect any CK20 expression on the LGR5+ cells during the culture. Thus, the LGR5+ cell lines form a highly homogenous population of cells having CSC properties with strong TIA. Only a few previous reports have described the use of adherent cultures to obtain CSCs such as in glioma and breast cancers [30, 40, 41]. The adherent cultures were rehighlighted to isolate stable cell lines with CSC properties in this study.
Using the established LGR5+ cell lines, definitive evidence for drug-resistant LGR5− CSC subpopulations was obtained after treatment with anti-cancer drugs such as irinotecan. The LGR5+ cells had a number of CSC characteristics such as self-renewal via symmetric and asymmetric division, TIA, and a pathway for producing a tumor hierarchy of different cell types. In addition, these cell lines transition between two distinct states, an LGR5+ proliferating and an LGR5− drug-resistant state. Tumor formation from the drug-resistant LGR5− cells in NOG mice was observed, but the TIA of the drug-resistant LGR5− cells was slightly lower than that of LGR5+ cells. Drug-resistant LGR5− cells first converted to LGR5+ cells in the establishment of tumor hierarchy in vivo. The results of irinotecan treatment in vivo suggest that anticancer drugs induce transition of LGR5+ cells to drug-resistant cells, and such drug-resistant cells revert to LGR5+ cells after drug treatment is terminated. We could detect both LGR5+ and LGR5− cells in ducts and budding sites of the tumors reconstituted from either the LGR5− or LGR5+ cells in mice and also in primary and liver metastasized tumors from patients. These observations may explain why some rare populations of CSCs survive after drug treatments, giving rise to tumor recurrence. If this idea is correct, the CSC cell lines provide a new avenue to test drugs that will kill all of the cancer cells in a tumor.
CSCs self-renew and also give rise to differentiated cancer cells. In fact, the LGR5+ cells exhibited the ability to undergo asymmetric cell divisions, generating two different offspring in vitro and reconstituting tumor hierarchy in vivo. However, it remains unclear whether a transition of differentiated cancer cells to CSCs occurs. Gupta et al.  proposed a stochastic state transition of cancer cells. Using breast cancer cell lines, they demonstrated that differentiated cancer cells possessed plasticity and transitioned to CSCs to maintain phenotypic proportions within tumors, although the frequency was very low (between 0.01% and 0.1%). In this study, the colony forming activity of the sorted cells in the 99.4% pure LGR5−/CD133− population, in which almost all the cells are CSC marker-negative and thereby considered to be differentiated tumor cells, was approximately 0.03%. This number is extremely low but not zero. Therefore, the possibility of a reversion of differentiated cells to CSCs, as proposed by Gupta et al., cannot be ruled out. At the same time, the possibility that colonies were formed by a small number of concomitant LGR5+ cells in this faction also cannot be excluded. Further study which overcomes technological hurdles of cell sorting is necessary to answer this question.
In normal small intestine, the existence of two types of SCs has been described: slow cycling SCs in the +4 position and proliferating SCs in the crypt base . However, the relationship between the two types of SCs was unclear. More recently, Lgr5−/Bmi1+ SCs were shown to serve as a SC pool: they changed to Lgr5+ SCs when the Lgr5+ SCs were absent . Furthermore, Takeda et al.  demonstrated the interconversion and bidirectional lineage relationship between proliferating Lgr5+ SCs at the crypt base column and slow cycling SCs that expressed an atypical homeobox protein Hopx at +4 position. Tert, telomerase reverse transcriptase, was reported as molecules that mark predominately noncycling, long-lived intestinal SCs that proliferate upon injury . Powell et al.  also demonstrated that expression of Lrig1, a pan ErbB inhibitor, was rather specific to quiescent SCs. However, Wong et al.  indicated the coexpression of Lrig1 in Lgr5+ cells, and intensive analysis with DNA microarray and proteomics revealed that Bmi1, Tert, Hopx and Lrig1 were all robustly expressed in Lgr5+ intestinal SCs . In our RT-qPCR analysis, the HOPX mRNA was not detected in the LGR5+ and drug-resistant LGR5− cells, whereas similar levels of the BMI1 and LRIG1 mRNA were detected in both LGR5+ and drug-resistant LGR5− states (Supporting Information Fig. S13). In addition, expression of TERT was rather specific to the LGR5+ cells; there was a marked decrease of the TERT mRNA after the LGR5+ cells were treated with irinotecan (Supporting Information Fig. S13), which coincided with the results that intestinal SCs contained significant telomerase activity . Except for HOPX mRNA expression, we observed expression of the BMI1, LRIG1, and TERT mRNAs in proliferating colon CSCs. Definitive understanding of the physiological roles and the expression of these genes in normal and cancer SCs await further study.
Because localization of proliferating (Lgr5+) and slow cycling quiescent (Lgr5−) SCs is restricted in normal intestine as described above, localization, proliferation, and transition between the slow cycling and proliferating states of SCs may be controlled by niches which include a gradient of the Wnt ligand . The moderately differentiated colon cancers differed from the normal architecture: it showed duct structures and an epithelial hierarchy, but localization of LGR5+ and LGR5− cells in ducts appeared not to be restricted to particular regions in both xenotransplanted tumors and in colon cancer tissues in patients, and they were observed throughout ducts. Thus, it seems that in tumor tissues, CSCs undergo proliferation and interconversion of their states without the underlying architecture of gradient producing cells at specific locations in the tissue structure.
It is widely believed that the invasive ability of cancer cells is important for metastasis. In addition, tumor budding is suggested to contribute to metastasis in colon cancer . We could identify three markers—LGR5, HLA-DMA, and EREG—that can mark these CSCs in two distinct states. Because both proliferating and drug-resistant colon CSCs were EREG positive, we addressed whether anti-EREG antibody is efficacious against tumor metastasis. The antibody showed only moderate activity against the established xenograft tumors but exhibited a stronger efficacy in a metastatic model tested in this study, suggesting that the anti-EREG antibody is efficacious in the early stage of cancer development when cancers are rich in CSCs. A number of studies suggest that metastatic nodules arise from rare cells in the primary tumor (CSCs) . If this is correct, then therapies targeting CSCs can have profound effects against metastatic tumors, even greater than upon primary tumors. In this way, the CSC cell lines developed here can give rise to novel therapies that could improve the treatment of cancer patients.
During the review process for this article, three papers appeared providing the evidence for the existence of CSCs in solid tumors in mice: Chen J. et al. (Nature 2012: 488: 522-526), Driessens G. et al. (Nature 2012: 488: 527-530), and Schepers AG et al. (Science 2012: 337: 730-735).
We established human colon cancer cell lines that express LGR5 and possess CSC properties. After treatment of the proliferating LGR5+ cells with an anticancer agent, the LGR5+ cells transition to a drug-resistant LGR5− state. In addition, the LGR5− cells converted to an LGR5+ state in the absence of the drug, suggesting a pool of SCs with the ability to interconvert between two distinct states. Using antibodies against LGR5, HLA-DMA, and EREG, we show the existence of LGR5+ and LGR5− cells in xenotransplanted tumor tissues and in human colon cancer tissues from patients. Furthermore, the anti-EREG antibody exhibited antitumor activity against tumors derived from the LGR5+ cells in a metastatic model. This suggests the physiological importance of CSCs in tumor recurrence. Furthermore, using the anti-EREG antibody, we provide an option for CSC targeting therapy.
We thank L.C. Wong, G.N. Yeow, H.S. Ong, Z.X. Wong, and Y. Takai for their technical assistance; Y. Ohnishi, E. Fujii, K. Nakano, Y. Hirata, and K.F-Ouchi for critical discussions; and R. Somerville for proof editing the manuscript. Thanks are also to T. Yamamura and R. Nomura for their continuous support throughout the study. We are also grateful to O. Nagayama, Chairman and CEO of Chugai, for his encouragement. This work is supported in part by a grant from Singapore Economic Development Board.
DISCLOSURE OF POTENTIAL C ONFLICTS OF INTEREST
S.K., H.Y.O., M.S., O.N., A.K., K.M., M.Y., S.F., K.Y., E.H., Y.W., H.M., M.A., C.K., T.W., T.Yo, and T.Ya. are employees of Chugai Pharmaceutical Co., Ltd. Y.J.C. is employee of PharmaLogicals Research Pte. Ltd. The authors indicate no other potential conflict of interest.