Runx3 Protects Gastric Epithelial Cells Against Epithelial-Mesenchymal Transition-Induced Cellular Plasticity and Tumorigenicity§


  • Author contributions: D.C.V.: conception and design, collection of data, data analysis/interpretation, and manuscript writing; H.J.W., J.K.W.K., T.A.P.N., and Y.T.H.: collection of data and data analysis/interpretation; Y.S.C. and J.P.T.: concept and design, collection, and analysis/interpretation of data (imaging and microscopy); S.L.C.: conception and design and data interpretation (sphere culture); K.I. and H.F.: provision of study materials (GIF cell lines); Y.I.: conception and design, financial support, and final approval of manuscript. D.C.V. and H.J.W. 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 August 16, 2012.


The transcription factor RUNX3 functions as a tumor suppressor in the gastrointestinal epithelium, where its loss is an early event in carcinogenesis. While RUNX3 acts concurrently as a mediator of TGF-β signaling and an antagonist of Wnt, the cellular changes that follow its loss and their contribution to tumorigenicity are not fully understood. Here, we report that the loss of Runx3 in gastric epithelial cells results in spontaneous epithelial-mesenchymal transition (EMT). This produces a tumorigenic stem cell-like subpopulation, which remarkably expresses the gastric stem cell marker Lgr5. This phenomenon is due to the compounding effects of the dysregulation of the TGF-β and Wnt pathways. Specifically, Runx3−/−p53−/− gastric epithelial cells were unexpectedly sensitized for TGF-β-induced EMT, during which the resultant induction of Lgr5 was enhanced by an aberrantly activated Wnt pathway. These data demonstrate a protective role for RUNX3 in safeguarding gastric epithelial cells against aberrant growth factor signaling and the resultant cellular plasticity and stemness. STEM Cells2012;30:2088–2099


Cell proliferation and differentiation in the gastrointestinal epithelium are highly regulated processes governed by intrinsic factors, such as cell cycling, and extrinsic signaling cues from the tissue niche. Prominent among these extracellular signals are the Wnt and TGF-β/BMP signaling pathways, whose disruption causes a breakdown in this delicate balance and results in the development of cancer [1, 2]. The aberrant activation of the Wnt pathway leads to uncontrolled growth and is the best-known cause of human colorectal cancer [1]. Conversely, a blockade of the TGF-β/BMP pathway due to somatic or hereditary mutations results in unchecked neoplastic growth [3, 4]. However, during the later stages of carcinogenesis, TGF-β signaling plays a paradoxical role in the promotion of the invasion and metastasis of cancer through its induction of epithelial-mesenchymal transition (EMT) [5]. During normal development, EMT consists of a reprogramming of epithelial cells that leads to the loss of epithelial characteristics, notably polarity and cell adhesion, and the acquisition of migratory properties [5]. However, during cancer progression, the reprogramming properties of EMT lead to increased invasion, migration, and the greater adaptability of cancer cells [6].

In the gastric epithelium, the Runt-domain transcription factor RUNX3 functions as a key mediator of the TGF-β pathway, where it modulates cellular proliferation and apoptosis through transcriptional regulation of genes such as the growth regulator p21WAF1 and the proapoptotic gene Bim [7–10]. In Runx3-deficient mice, the loss of these functions results in cellular hyperproliferation in the gastric epithelium due to dysregulated cell growth and the impairment of TGF-β-mediated apoptosis [7]. This is accompanied by gross antralization and intestinalization, disrupted differentiation of chief cells, and the development of metaplasia, which are indicative of a preneoplastic state [11].

Consistent with its tumor suppressive properties, the loss of RUNX3 function is observed in up to 80% of human primary gastric cancers due to epigenetic silencing by promoter hypermethylation and protein mislocalization [7, 12]. In support of these in vivo observations, immortalized cell lines (termed GIF lines) established from E16.5 Runx3−/−p53−/− fetal gastric epithelia are tumorigenic when transplanted into nude mice but not their Runx3+/+p53−/− counterparts [7, 13]. Furthermore, Runx3-null GIF lines exhibit dysregulated differentiation, frequent intestinal transdifferentiation, and disrupted epithelial polarity, due in part to the loss of the tight junction component Claudin-1 [13–15].

In addition to mediating TGF-β signaling, RUNX3 also functions as a negative regulator of the Wnt pathway through a direct and mutually antagonizing interaction with the T-cell factor (TCF)/β-catenin complex [11, 16]. In mice, this relationship is reflected in the increased incidence and malignancy of intestinal tumors in Runx3+/−Apcmin/+ compound mutant mice, compared with mice bearing single mutations [16]. In humans, the RUNX3R122C mutation, which was identified in a human gastric cancer patient, is unable to attenuate the Wnt signal [11]. However, although the molecular basis of the interactions between RUNX3, TGF-β/BMP, and the Wnt pathway has been established, the overall cellular changes associated with the loss of RUNX3 that render gastrointestinal epithelial cells preneoplastic remain obscure.

In this study, we report that gastric epithelial cells undergo spontaneous EMT in the absence of Runx3 and p53, which induces the formation of a tumorigenic subpopulation in immortalized Runx3−/−p53−/− gastric epithelial (GIF) cells. Remarkably, the mesenchymal-like subpopulation also possesses stem-like properties, as reflected in the enrichment of the gastrointestinal stem cell marker Lgr5 and greater sphere-initiating and colony-forming abilities. Contrary to their resistance to TGF-β-mediated apoptosis, Runx3-null GIF lines are unexpectedly sensitized to TGF-β-induced EMT, compared with their Runx3-normal equivalents, indicating a rerouting of the TGF-β signal. In addition, a greatly increased Wnt-responsiveness was observed in Runx3-null GIF cells, which acts synergistically with TGF-β to induce Lgr5 expression. Together, these observations indicate that Runx3 acts as a tumor suppressor by safeguarding gastric epithelial differentiation and phenotypes; in its absence, gastric epithelial cells are prone to spontaneous EMT and aberrant TGF-β and Wnt signaling, giving rise to a tumorigenic stem cell-like subpopulation.


Cell Culture

The murine gastric epithelial cell lines GIF-5, GIF-9, GIF-13, and GIF-14 were previously established by Fukamachi et al. from E16.5 Runx3+/+p53−/−and Runx3−/−p53−/− mouse fetal stomachs [13]. These cells were maintained in Dulbecco's modified Eagle's medium (DMEM; Invitrogen, Carlsbad, CA, with 4,500 mg/l glucose supplemented with 10% FBS, 100 units/ml penicillin, and 100 μg/ml streptomycin antibiotics (Thermo Scientific Hyclone, Logan, UT, in standard culturing conditions. For the activation or blockade of the TGF-β pathway, cells were treated with 2.5 ng/ml of recombinant TGF-β1 (R&D Systems, Minneapolis, MN, or SB431542 (TOCRIS Bioscience, Bristol, UK,, respectively, for the indicated periods prior to staining or quantitative RT-PCR (qRT-PCR). To activate the Wnt pathway, cells were treated with control- or Wnt3a-conditioned medium (Cm), prepared as described by Willert et al., for 15 hours prior to analyses [17].

Hoechst 33342 and Surface Antigen Staining

Cells were stained with Hoechst 33342 (Sigma-Aldrich; St. Louis, MO, as described previously [18]. Briefly, GIF cells were washed and resuspended to 1 × 106 cells per milliliter in DMEM supplemented with 2% FBS, 10 mM HEPES, and 10 μg/ml Hoechst 33342 with or without 0.2 mM verapamil (Sigma-Aldrich) and incubated at 37°C for 90 minutes with regular mixing. Cells were resuspended in prechilled 1× Hanks' balanced saline solution (HBSS; Invitrogen) containing 2% FBS and 10 mM HEPES. For costaining experiments, cells were stained with fluorochrome-conjugated antibodies against various surface antigens, including anti-EpCAM-PE-Cy7 (Biolegend, San Diego, CA,; #118215; 1:100) and anti-CD133/Prominin-1-PE (Miltenyi Biotec, Bergisch Glodbach, Germany,; #130-092-334; 1:100) monoclonal antibodies.

Flow Cytometry Analysis and Fluorescence-Activated Cell Sorting

Cells were counterstained with 1 μg/ml of propidium iodide and analyzed or sorted on a FACSVantage or a FACSAria II cell sorter (BD Biosciences, San Jose, CA, Hoechst 33342 staining patterns were analyzed with a 350 nm UV laser using a 450/20 nm bandpass (Hoechst blue) and a 675 nm longpass (Hoechst red) optical filters. Lentivirus-infected cells were sorted based on their expression of enhanced green fluorescent protein (EGFP). Flow cytometry data were analyzed using the FlowJo computer software package (Tree Star, Ashkind, OR,

Gene Expression Profiling by qRT-PCR

RNA extractions were performed using the QIAGEN RNeasy Mini Kit (QIAGEN, Hilden, Germany, and cDNA synthesis was performed with the Omniscript RT kit (QIAGEN), using 200 ng to 1 μg of total RNA. qPCR was performed on an Applied Biosystems 9500 real-time PCR system (Applied Biosystems, Foster City, CA, using gene-specific oligonucleotide primers for SYBR Green-based measurements or Taqman probes (Applied Biosystems). A complete list of the oligonucleotide primer sequences and Taqman probes used are provided in Supporting Information Table S1. Unless otherwise stated, all gene expression data were normalized against Gapdh levels.

Nude Mice Transplantation Assay

Female BALB/c nude mice were purchased from the Biopolis Resource Centre (A*STAR, Singapore) and handled in accordance with BRC Institutional Animal Care and Use Committee guidelines. Clonal GIF-14 lines or fluorescence-activated cell sorting (FACS)-purified cells were resuspended in 200 μl of PBS and injected subcutaneously (s.c.) into 7–8-week-old athymic nude mice. Tumor development was monitored weekly and the animals were culled when the tumor sizes reached 15 mm in diameter.

Soft Agar Colony Assay

Cells were suspended in 0.4% agarose and DMEM supplemented with 10% FBS and seeded over a basal layer of 0.6% agarose. The experiments were set up in six-well plates at densities of 200,000 and 500,000 cells per well in triplicates. Colonies ≥150 μm in size were scored after 3 weeks of culture at 37°C. Photographs of the colonies were taken using phase-contrast microscopy (Eclipse TS100, Nikon, Tokyo, Japan,

Sphere-Forming Assay

Cells were seeded in six-well ultralow attachment plates (Sigma-Aldrich) at a density of 2,500 cells per well and cultured in serum-free DMEM:F12 medium (Invitrogen) containing 20 ng/ml human recombinant epidermal growth factor (EGF; PeproTech, Rehovot, Israel,, 10 ng/ml human recombinant basic fibroblast growth factor (bFGF; PeproTech), B27 (Invitrogen), N2 (Invitrogen), 1 ng/ml hydrocortisone (Sigma-Aldrich), 5 μg/ml insulin (Sigma-Aldrich), and 0.4% BSA Fraction V (Sigma-Aldrich). To prevent cell aggregation, methylcellulose (Sigma-Aldrich) was added to a final concentration of 0.5% unless otherwise stated. At the indicated time, the number of spheres of the indicated sizes was counted and imaged under phase-contrast microscopy (Eclipse TS100, Nikon).

Virus Production and Transduction

The coding region of the p44 isoform of human RUNX3 was cloned into iG2-LeGO [19] upstream of IRES-EGFP using engineered 5′EcoRI and 3′ NotI sites; and into a modified pBOBI vector [20] (a kind gift of Vinay Tergoankar, IMCB, A*STAR) using 5′ XbaI and 3′ Xho1 sites. The RUNXR178Q mutation was generated using QuikChange II Site-Directed Mutagenesis Kit (Agilent Technologies, Santa Clara, CA, The sequence integrity of all constructs was verified by capillary DNA sequencing. Lentiviruses were produced according to the protocol described by Tiscornia et al. with minor modifications [21]. Briefly, iG2- or pBOBI-based transfer vectors were cotransfected with ViraPower packaging vectors (Invitrogen) into 293T cells using FuGENE HD (Roche, Basel, Switzerland, Supernatants containing viral particles were harvested 36 and 48 hours post-transfection and pooled. For transduction with lentiviruses, GIF-14 cells were incubated with virus-containing supernatants in the presence of 5 μg/ml polybrene (Sigma-Aldrich) for 24 hours before replenishment with normal culture medium. Infected cells were treated with Wnt3a-Cm or TGF-β1 prior to enrichment by FACS (iG2 lentiviruses) or direct lysis (pBOBI lentiviruses). Ectopic expression of RUNX3 and RUNX3R178Q in infected GIF-14 cells are confirmed by Western blotting (Supporting Information Fig. S1).

Time-Lapse Videomicroscopy

Purified P1 and P2 cells were seeded at clonogenic densities on 25 mm glass coverslips (Marienfeld, Lauda Königshofen, Germany, or low evaporation six-well plates (Becton Dickinson, BD Biosciences). Following colony formation, the glass slips were transferred to Sykes Moore (BellCo Glass, Vineland, NJ, chambers with complete CO2-independent medium (Invitrogen), while those grown in six-well plates were replenished with standard growth medium. Skyes Moore chambers or six-well plates were placed onto a Nikon Ti Eclipse microscope (Nikon) and maintained at 37°C (PeCon GmbH, Erbach, Germany, Colonies on glass slips and six-well plates were visualized by differential interference contrast (DIC) microscopy and phase-contrast microscopy, respectively, at ×100 magnification. For comparison of P1 and P2 colonies, DIC images were captured by a CCD camera (HQ2; Photometrics, Tucson, AZ, at 3 minutes intervals over a period of 10 hours. For TGF-β1 treatment, DIC and phase-contrast images of P1 colonies were captured before and post-TGF-β1 treatment at 5 minutes intervals for the indicated time frames. The images were postprocessed using the Metamorph (Molecular Devices, Sunnyvale, CA, and VirtualDub ( software programs for the construction of video images.

Immunostaining of P1 and P2 Colonies

P1 and P2 colonies were generated on coverslips by seeding at clonogenic densities. Once generated, P1 colonies were treated with 2.5 ng/ml TGF-β1 (R&D Systems) or carrier control for 48 hours. Colonies were fixed using 4% paraformaldehyde and permeabilized with 0.5% Triton X-100. After blocking with 1% bovine serum albumin (BSA), the colonies were incubated with anti-β-catenin (Sigma-Aldrich), antidesmoplakin (Millipore, Billerica, MA,, anti-Smad2/3 (BD Transduction Laboratories, San Jose, CA), or anti-phospho-Smad2/3 (Santa Cruz, Santa Cruz, CA, primary antibodies (1:200 dilution). Alexa-488-conjugated anti-mouse (Invitrogen) or Cy3-conjugated anti-rabbit (Invitrogen) was used as secondary antibodies (1:500 dilution) for 45 minutes each at room temperature. F-actin was visualized using phalloidin conjugated to Alexa-350 (Invitrogen; 1:50 dilution). Images were captured using a Nikon Ti Eclipse confocal microscope (Nikon) or a Leica 6000B epifluorescence microscope (Leica, Wetzlar, Germany, and postprocessed using MetaMorph (Molecular Devices).

Protein Isolation and Western Blotting

Cells were lysed in RIPA buffer supplemented with complete protease inhibitor (Roche) and 1 mM phenylmethylsulfonyl fluoride. Whole-cell lysates were resolved in SDS-PAGE and immunoblotted with antibodies at their respective dilutions: anti-RUNX3 (1:1,500; clone R3-5G4, MBL, Woburn, MA,, anti-E-cadherin (1:1,000; clone DECMA-1, Abcam, Cambridge, UK,, anti-Vimentin (1:500, clone V9, Abcam), anti-α-Tubulin (1:2,000; clone DM1A, Sigma-Aldrich), anti-mouse IgG-HRP (1:2,000; Santa Cruz), and anti-rat IgG-HRP (1:3,000; Dako, Glostrup, Denmark,

Statistical Analysis

All data are presented as the mean ± SEM. Student's t test was used in the comparisons of two datasets, and p values <0.05 were considered significant.


Identification of a Tumorigenic Subpopulation in the Runx3−/−p53−/− GIF-14 Gastric Epithelial Cell Line

A prior study reported that Runx3−/−p53−/− GIF cell lines are tumorigenic while Runx3+/+p53−/− lines are not [7]. To investigate this difference, the Runx3-null line GIF-14 was analyzed by flow cytometry following staining with Hoechst 33342, which revealed the presence of two major subpopulations, P1 and P2, with significantly different dye efflux properties. P1 cells were dimly stained due to high efflux efficiency. This staining could be inhibited by verapamil, an ATP-binding cassette protein transporter inhibitor, which induced a pronounced shift in the staining pattern. In contrast, P2 cells were brightly stained and their staining pattern was unaffected by verapamil (Fig. 1A). Subsequently, these two subpopulations were characterized for their surface marker expression by costaining with Hoechst 33342 and specific antibodies. The expression levels of various markers in the P1 and P2 populations are summarized in Supporting Information Table S2. These analyses established that P1 and P2 cells are demarcated by EpCAMhigh/CD133+ and EpCAMlow/CD133 expression patterns, respectively (Fig. 1B). The differential marker expression provides further indication that P1 and P2 represent physiologically distinct subpopulations within GIF-14.

Figure 1.

A GIF-14 subpopulation brightly stained by Hoechst 33342 and marked by EpCAMlow/CD133 is more tumorigenic. (A): Hoechst 33342 staining revealed two subpopulations, P1 and P2, in the GIF-14 line. Dye efflux of P1 cells was specifically blocked in the presence of Verapamil. (B): Costaining with Hoechst 33342 and antibodies against EpCAM and CD133/Prominin-1 revealed distinct surface marker expression by P1 and P2. (C): The tumorigenicities of P1 and P2 cells were determined by allograft transplantation experiments performed in nude mice. (Top) Fluorescence-activated cell sorting-purified P1 or P2 cells were expanded briefly in vitro before transplantation. For each mouse, 2 × 106 of unsorted parental (neck), P1 (left side) and P2 (right side), cells were injected and tumor sizes were scored after 11.5 weeks (n = 9). (Bottom) Clonal P1 and P2 (2 × 106) cells were transplanted and examined after 14 weeks (n = 10). Each dot represents an individual tumor size; black bars denote the average sizes for each class. Student's t test was performed and a p-value ≤.05 was considered significant. (D): The anchorage-independent growth potentials of pooled P1, P2, sorted control, or unsorted parental GIF-14 cells were measured by soft agar colony assays. Colonies ≥150 μm in size were counted after 3 weeks and data are expressed as means ± SEM (n = 3). Representative phase-contrast images of the colonies are shown. Scale bars = 200 μm.

The relative tumorigenicities of these subpopulations were tested in two complementary nude mouse allograft experiments. First, P1 and P2 cells were sorted from GIF-14 cells by FACS and briefly cultured in vitro for the recovery and expansion of the sorted cells. The isolated cells were checked for the maintenance of their Hoechst staining patterns and transplanted subcutaneously into separate abdominal sites on individual nude mice. As controls, unsorted GIF-14 cells were similarly transplanted into the neck region of each mouse. In the second study, clonal lines were established from sorted P1 and P2 cells, and representative lines were tested for their relative tumorigenicity in nude mice. These allograft assays showed that both pooled and clonal P2 cells were significantly more tumorigenic than P1 cells in vivo (Fig. 1C). FACS-enriched P1 and P2 cells were further assessed for their capacity for anchorage-independent growth in vitro (Fig. 1D). Soft agar assays revealed that enriched P2 cells formed colonies far more readily than P1 cells or control cells, despite P1 and P2 cells having very similar proliferation rates (Fig. 1D and Supporting Information Fig. S2). Taken together, these data indicate that the capacity of GIF-14 cells to initiate tumors in nude mice is largely restricted to a distinct, low dye-effluxing subpopulation (P2) that is marked by an EpCAMlow/CD133 profile.

P2 Cells Are Enriched in Stem Cell-Associated Markers and Display Stem/progenitor Cell Properties

To gain insight into the properties of P1 and P2 cells, these cells were fractionated from the parental GIF-14 line by FACS and subjected to gene expression profiling by qRT-PCR (Fig. 2A). The expression levels of the stem cell-associated markers Oct4, Nanog, Lgr5, Sox9, and Cd44 were quantified and expressed as ratios to the levels in total sorted cells. In three separate experiments, P2 cells expressed significantly higher levels of the stem cell markers Oct4 and Nanog (Fig. 2A). Notably, P2 was strongly enriched in Lgr5, which encodes a G protein-coupled receptor that marks a newly described stem cell population in the intestinal crypt and pylorus gastric gland [22, 23]. In addition, P2 also expressed higher levels of Sox9, which is reportedly coexpressed in Lgr5-positive stem cells [22–24]. Lastly, the expression levels of ATP-binding cassette protein transporters such as Abcg2 and Abcb1/mdr, which are associated with the “side population” dye efflux phenotype, are significantly reduced in P2 cells, which might account for their lower dye efflux (Supporting Information Fig. S3).

Figure 2.

P2 cells display increased stem-like properties that are abrogated by RUNX3. (A): Gene expression profiling of P1 and P2 subpopulations. (Top) Parental GIF-14 cells were sorted into P1 and P2 gates based on Hoechst 33342 staining for quantitative RT-PCR (qRT-PCR) analyses. (Bottom) The expression levels of various embryonic or gastrointestinal stem cell markers are presented relative to those of control sorted cells, which included all cells shown on the plot. Data presented are a compilation of three separate experiments and Student's t tests were performed (*, p <.05; **, p <.01). (B): Fluorescence-activated cell sorting (FACS) fractionated P1 and P2 cells were compared to control sorted cells (all cells on FACS plot) and unsorted parental GIF-14 cells for sphere-forming potentials in serum-free cultures. Data are presented as the number of spheres ≥300 μm in size generated per 5,000 cells after 30 days (means ± SEM, n = 5). Phase-contrast images of the spheres are shown. (C): Changes in Lgr5 transcripts in control or sorted P1 and P2 cells following 15 hours treatments with control- or Wnt3a-conditioned media (Cm). Normalized Lgr5 mRNA levels are expressed relative to the basal value of the unsorted control (means ± SEM, n = 4). (D): Comparison of GIF-14 (Runx3−/−) and GIF-13 (Runx3+/+) sphere-initiating potentials. Data are presented as number of spheres ≥250 μm in size generated from 5,000 cells after 25 days in culture (means ± SEM, n = 4). Representative images of the spheres were captured under phase-contrast microscopy. (E): A comparison on the effects of Control-Cm and Wnt3a-Cm on Lgr5 expression in GIF-14 and GIF-13 cells, following 15 hours treatment. Normalized Lgr5 mRNA levels are expressed relative to that of Control-Cm treated GIF-14 cells (means ± SEM, n = 4). (F): Parental GIF-14 cells were infected with pBOBI lentivirus encoding wild-type RUNX3 or DNA-binding defective RUNX3R178Q for 48 hours and treated with control- or Wnt3a-Cm for 15 hours. Expression of Lgr5 expression levels was quantified by qRT-PCR. Normalized values are presented as ratios to control values for each infected sample (means ± SEM, n = 4). (G): Strong abrogation of sphere formation by exogenous RUNX3. Parental GIF-14 cells infected with control or RUNX3-expressing lentiviruses were fractionated by FACS and subjected to a sphere formation assay. Data are presented as number of spheres ≥150 μm in size generated from 8,000 cells after 30 days in culture (means ± SEM, n = 5). (H): Changes in sphere formation in response to Wnt3a-Cm. GIF-14 cells infected with control or RUNX3-encoding lentiviruses were fractionated by FACS and treated with a modified sphere culture medium containing 1.0% methylcellulose and control- or Wnt3a-Cm (1:10 dilution). Data are presented as number of spheres ≥100 μm generated from 8,000 cells after 25 days in culture (means ± SEM, n = 5). Scale bars = 300 μm.

To determine the functional significance of this finding, fractionated P1 and P2 cells were cultured in a serum-free sphere culture medium on an ultralow attachment substratum. Under this culture condition, P2 cells consistently displayed significantly greater sphere-initiating capacities than P1 cells, which is indicative of enhanced stem/progenitor cell-like properties (Fig. 2B). Moreover, some P2-derived spheres could be maintained for an extended period of time in culture (Supporting Information Fig. S4) without obvious signs of differentiation and were able to generate secondary spheres following dissociation (data not shown). In contrast, while P1 cells were capable of initiating spheres, these could not be sustained in long-term culture and were less efficient in generating secondary spheres (Fig. 2B).

Consistent with the role of Lgr5 as a target of the canonical Wnt pathway, treatment with Wnt3a-conditioned medium resulted in a marked induction of Lgr5 transcripts in GIF-14 populations, particularly in P2 cells (Fig. 2C). Increased Lgr5 expression was accompanied by an increase in sphere initiation activity in treated cells (Supporting Information Fig. S5 and Fig. 2H), suggesting a link between Lgr5 expression and the stem/progenitor phenotype in GIF-14 cells.

RUNX3 Strongly Inhibits Sphere-Initiation and Lgr5 Expression

Compared with GIF-14 cells, sphere initiation was significantly lower in Runx3+/+ GIF-13 cells (Fig. 2D). While GIF-13 cells were capable of rapid sphere initiation, these quickly disintegrated into small clusters of mixed cell morphologies and seldom developed like GIF-14-derived spheres (Fig. 2D). This was consistent with the lower basal expression of Lgr5 and markedly reduced Wnt-responsiveness observed in GIF-13 cells (Fig. 2E). To verify the role of Runx3, wild-type RUNX3 or the DNA-binding defective mutant RUNX3R178Q were ectopically introduced into GIF-14 cells by recombinant lentiviruses (pBOBI-RUNX3 or -RUNX3R178Q, respectively). At 48h post-infection, cells were treated with control- or Wnt3a conditioned medium for 15h and subjected to qRT-PCR analyses. These revealed that the induction of Lgr5 by Wnt3a was significantly reduced by ectopic RUNX3 but not RUNX3R178Q expression (Fig. 2F). This observation implies a negative regulatory mechanism that depends on the DNA binding activity of RUNX3. These findings were confirmed by the strong inhibition of sphere initiation observed in GIF-14 cells expressing exogenous RUNX3 in standard sphere culture medium (Fig. 2G) and in the presence of Wnt3a-Cm (Fig. 2H).

Together, these data show the sphere-initiating properties of the Lgr5-expressing P2 subpopulation in the GIF-14 line and indicate that the activation of the Wnt pathway further enhances this effect. In contrast, RUNX3 acts as a negative regulator of this process, as shown in its strong blockade of the Wnt induction of Lgr5 and sphere initiation.

P2 Cells Are Products of Spontaneous EMT Within the GIF-14 Line

In addition to the greater stem/progenitor-like properties of P2 cells, P1 and P2 cells showed distinct cell morphologies. P1 cells appeared as typical epithelial cells with polygonal shapes that formed compact colonies with low cell motility (Fig. 3A and 3B). In contrast, P2 cells displayed properties resembling mesenchymal cells, including dendrite cell morphology and high cell motility, and they formed expansive, poorly organized colonies (Fig. 3A and 3B). Time-lapse DIC video microscopy revealed that P1 colonies expanded in a radial manner while maintaining strong cell-to-cell contact (Fig. 3B; and Supporting Information Video S1). In comparison, P2 colonies expanded rapidly and cells showed a general loss of adherence, with cells at boundaries frequently detaching from the colony and migrating outward (Fig. 3B; and Supporting Information Video S2). The greater migratory and invasive properties of P2 cells could be further demonstrated in Transwell and wound healing assays (Supporting Information Figs. S6, S7).

Figure 3.

P2 cells are the products of spontaneous epithelial-mesenchymal transition (EMT) in GIF-14 cells. (A): Phase-contrast microscopy of fluorescence-activated cell sorting (FACS)-purified P1 and P2 cells revealed marked morphological differences. (B): Differential interference contrast microscopy images obtained from time-lapse imaging of the growth patterns of P1 and P2 colonies (Supporting Information Videos S1, S2). P1 colonies display a characteristic epithelial cobblestone appearance, while P2 colonies are expansive and are composed of highly motile cells with mesenchymal properties. (C): Expression of epithelial, mesenchymal, and EMT-associated markers in P1 and P2 subpopulations purified by FACS (Fig. 2A). Expression levels are presented relative to those of control sorted cells (all populations). Data are compiled from three separate experiments and Student's t tests were performed (**, p <.01). (D): Induction of EMT-associated markers in three separate batches of GIF-14 cells following treatment with TGF-β1 for 24 or 48 hours. Changes in gene expression were determined by quantitative RT-PCR and normalized values are expressed relative to the untreated values (means ± SEM, n = 3). (E): Changes in Hoechst 33342 staining patterns in response to TGF-β1 treatment. GIF-14 cells were cultured in the absence or presence of TGF-β1 for 10 days. At the days indicated, cells were subjected to Hoechst 33342 staining and flow cytometry, which revealed a distinct shift toward a P2 pattern. (F): TGF-β1 induced a concomitant shift toward an EpCAMlow/CD133 expression pattern in treated GIF-14 cells. Scale bars = 50 μm.

Consistent with these characteristics, qRT-PCR analyses of P2 cells revealed reduced expression of Epcam and enrichment in several mesenchymal markers, such as Vimentin (Vim) and Fibronectin (Fn1) (Fig. 3C). Importantly, P2 cells expressed higher levels of Snai1 and Snai2, which are key inducers of EMT during development and cancer metastasis (Fig. 3C). As these two transcription factors are well-known suppressors of E-cadherin (E-cad), the maintenance of its expression is unexpected. While Western blot analysis confirms that P1 and P2 express comparable levels of E-cadherin, its subcellular localization is profoundly perturbed in P2 cells, reflecting a loss of epithelial phenotype (Supporting Information Fig. S8).

As Snai1 and Snai2 are known to act downstream of TGF-β1, P2 cells could be products of TGF-β-induced EMT. To test this, parental GIF-14 cells were treated with TGF-β1 and analyzed for changes in the mesenchymal/EMT markers that are characteristically enriched in P2 cells. As shown in Figure 3D, robust induction of these markers was observed within 24 hours of treatment, indicating that EMT could be readily induced in parental GIF-14 cells, giving rise to a P2-like gene expression signature, which remarkably included an induction of Lgr5.

In concordance with the gene expression data, Hoechst staining of TGF-β1-treated cells showed the expansion of the P2 subpopulation and the reduction of the P1 following 3 days of treatment, which was steadily enhanced during a 10-day treatment regimen (Fig. 3E). Concurrently, the induction of the P2 subpopulation was accompanied by a pronounced shift toward an EpCAMlow/CD133 phenotype (Fig. 3F). Conversely, a TGF-β type I receptor inhibitor (SB431542) strongly promoted a P1-like phenotype marked by the induction of epithelial cell morphology; a suppression of P2 gene signature in induced or resting GIF-14 cells; and a P1-like Hoechst and EpCAMhigh/CD133+ staining profiles (Supporting Information Fig. S9). Together, these data strongly suggest that P2 is the product of spontaneous EMT in GIF-14 cells, driven by a constitutively active TGF-β pathway.

TGF-β1 Induces a Direct Transition of the P1 to the P2 Phenotype in GIF-14 Cells

To demonstrate that the induction of the P2 phenotype was not due to negative selection against P1 cells by TGF-β1, but rather a direct transition of the epithelial to the mesenchymal phenotype, the effects of TGF-β1 on single-cell-derived P1 colonies were studied by time-lapse video imaging using phase-contrast (Fig. 4A; Supporting Information Videos S3A, S3B) and DIC microscopy (Supporting Information Videos S4A, S4B). These experiments clearly demonstrated that TGF-β1 induced a direct transition to a mesenchymal-like phenotype that closely resembled that of P2-derived colonies (Supporting Information Video S2). With time, cells progressively delaminated at the periphery of the colonies and became motile, while cells at the center of the colonies adopted elongated morphologies (Supporting Information Videos S3A, S4A). These changes were in striking contrast to the behavior of untreated controls (Fig. 4A and Supporting Information Videos S3B, S4B).

Figure 4.

TGF-β1 induces a direct transition from the P1 to the P2 phenotype in GIF-14 cells. (A): Time-lapse microscopy shows the induction of mesenchymal-like morphological changes in P1 colonies for more than 2 days of TGF-β1 treatment. Phase-contrast images of representative control and TGF-β1-treated P1 colonies were simultaneously captured 20 hours prior to and 67 hours post-treatment at 5 minutes intervals (Supporting Information Videos S3A, S3B). (B): Immunofluorescent staining of desmoplakin (top panel), F-actin (middle panel), and β-catenin (bottom panel) in P1, P2, and TGF-β1-treated P1 colonies. Scale bars = 50 μm.

To compare the phenotype of TGF-β1-induced P1 colonies to that of P2 colonies, immunofluorescent staining of desmoplakin, F-actin, and β-catenin was performed (Fig. 4B). These imaging experiments revealed that TGF-β1 induced a distinct redistribution of desmoplakin in P1 cells from the plasma membrane to a cytoplasmic/nuclear localization that is indicative of desmosome disassembly, a hallmark of EMT (Fig. 4B). Consistent with the high motilities of P2 and TGF-β1-induced P1 cells, phalloidin staining of F-actin showed prominent actin stress fiber formation in these cells, particularly in those located at the periphery (Fig. 4B). Lastly, TGF-β1 treatment resulted in the loss of plasma membrane-associated β-catenin staining in P1 colonies, indicating the disassembly of adherent junction complexes independent of the expression levels of β-catenin and desmoplakin (Fig. 4B and Supporting Information Fig. S10). Importantly, the staining patterns of these markers induced in P1 cells by TGF-β1 were indistinguishable from the pattern observed in P2-derived colonies (Fig. 4B). Interestingly, the staining of these markers in P2 colonies revealed a gradient of increasing mesenchymal characteristics towards the periphery of the colonies, reflecting a state of plasticity and bipotency in P2 cells (Supporting Information Fig. S11). Indeed, the plasticities of P1 and P2 cells were supported by their interchangeability during prolong treatment of TGF-β1 or SB431542, respectively (Supporting Information Fig. S12). Taken together, the changes in immunofluorescent staining patterns, gene expression, cell morphology and behavior induced in P1 cells by TGF-β1 treatment provide strong evidence that P2 cells are derived from the EMT of P1 cells.

RUNX3 Inhibits Spontaneous and TGF-β-Induced EMT in Multiple GIF Lines

The relationship between the absence of Runx3 and the observed spontaneous EMT was investigated in multiple Runx3-normal or Runx3-null GIF gastric epithelial cell lines. Analyses of the expression of surface markers revealed that the P2-associated EpCAMlow/CD133 profile was significantly represented only in the Runx3-null GIF-5 and GIF-14 lines (Fig. 5A). These similarities suggest a common mechanism for the safeguarding of the gastric epithelial phenotype by Runx3. Treatment of these GIF cell lines with TGF-β1 revealed marked differences in their responses. In Runx3-null GIF-5 and GIF-14 lines, a clear induction of a mesenchymal/EMT expression signature was observed within 24 hours and persisted for 48 hours. In sharp contrast, the induction of these genes was muted in Runx3+/+ GIF-9 and GIF-13 cells (Fig. 5B).

Figure 5.

RUNX3 protects gastric epithelial cells from spontaneous and TGF-β1-mediated epithelial-mesenchymal transition (EMT) and cellular plasticity. (A): EpCAM/CD133 expression profiles in four separate Runx3−/− and Runx3+/+ GIF lines. The mesenchymal-like EpCAMlow/CD133 population was only significantly represented in Runx3−/− lines. (B): Striking differences are evident in the TGF-β1 responsiveness of mesenchymal and EMT-associated markers between Runx3−/− and Runx3+/+ GIF lines. Cells were treated with TGF-β1 for 24 and 48 hours and analyzed by quantitative RT-PCR (qRT-PCR). Normalized values are expressed relative to uninduced control values and presented as means ± SEM (n = 3). (C): GIF-14 (Runx3−/−) and GIF-13 (Runx3+/+) cells were treated with TGF-β1 for the indicated periods and cotreated with Control-Cm or Wnt3a-Cm for 15 hours prior to qRT-PCR analysis of Lgr5 and Snai1 transcripts. Normalized expression levels are presented relative to the expression levels of untreated controls. (D): Exogenous RUNX3, but not RUNX3R178Q, strongly suppresses stem cell markers Lgr5 and Sox9 and attenuates TGF-β1 induction of Snai1 and Hmga2. Cells were infected with the indicated lentiviruses for 6 days prior to TGF-β1 treatment for 24 hours. Each replicate represents a separate infection of parental GIF-14 cells. Cells were fractionated by fluorescence-activated cell sorting and subjected to qRT-PCR. The normalized data are presented relative to the untreated control value (means ± SEM, n = 3). (E): Effects of exogenous RUNX3 on the Hoechst 33342 staining patterns of GIF-14 cells. GIF-14 cells were infected with iG2-Control, -RUNX3, or -RUNX3R178Q lentiviruses for 48 hours prior to treatment with TGF-β1 for 4 days. (F): Changes in EpCAM and CD133/Prominin-1 surface expression following TGF-β1 treatment in infected cells in three separate experiments. The relative levels of EpCAMhigh/CD133+ and EpCAMlow/CD133 populations are presented as a percentage of their combined values.

A potential tripartite relationship between the loss of Runx3, EMT, and the induction of stemness was next investigated by measuring changes in Lgr5 expression as a readout for stemness and plasticity. In this study, Runx3−/− (GIF-14) and Runx3+/+ (GIF-13) cells were pretreated with TGF-β1 for varying durations before a 15 hours induction by Wnt3a- or control-conditioned medium. The results showed striking differences in the induction of Lgr5 and Snai1 transcripts in these two lines (Fig. 5C). In GIF-14 cells, Snai1 was strongly induced by TGF-β1 within 15 hours of treatment but was unresponsive to Wnt3a. In contrast, the expression of Lgr5 was initially suppressed by TGF-β1 treatment but was induced upon prolonged exposure. Remarkably, the Wnt-responsiveness of Lgr5 was greatly augmented in cells treated with TGF-β1 for 48 hours. These data indicate that the TGF-β1 induction of Lgr5 followed a different kinetics to the rapid activation of Snai1. Moreover, its strong synergy with Wnt3a in activating Lgr5 transcription implies an independent but cooperative mechanism (Fig. 5C). Importantly, such induction patterns were entirely absent in Runx3-positive GIF-13 cells. These data reveal a critical function of Runx3 in modulating TGF-β and Wnt signaling in gastric epithelial cells. Through this function, Runx3 protects these cells against aberrant growth factor signaling and dedifferentiation.

Finally, to further investigate the observed negative correlation between the loss of Runx3 and an increased mesenchymal-like phenotype in GIF lines, GIF-14 cells were transduced with iG2-RUNX3 or iG2-RUNX3R178Q for 6 days and treated with TGF-β1 for 24 hours. Infected cells were purified by FACS and subjected to gene expression profiling. As shown in Figure 5D, the expression of exogenous RUNX3 resulted in a strong suppression of basal Lgr5 expression and the complete abrogation of its induction by TGF-β1. These changes were closely mimicked by Sox9, another gastrointestinal stem cell marker closely associated with Lgr5 in vivo. Decreases in Lgr5 and Sox9 expression are correlated to the reduced TGF-β1 induction of Snai1 and Hmga2. Further studies indicate that the suppression of Snai1 transcription by RUNX3 was not mediated by the direct inhibition of Snai1 promoter activity, but rather through an indirect mechanism (data not shown).

Consistent with the changes in gene expression and TGF-β1 responsiveness, ectopic RUNX3 markedly altered the Hoechst 33342 staining and marker expression patterns. In separate lentiviral infection studies, the reintroduction of RUNX3 consistently resulted in the ablation of the P2 population (Fig. 5E). These effects were clearly dependent on the DNA-binding activity of RUNX3, as ectopic RUNX3R178Q resulted in the same staining profile as that of control. RUNX3 also changed the P1 population similar to verapamil, possibly due to its specific suppression of Abca3 and Abcb1/Mdr (Supporting Information Fig. S13). This promotion of an epithelial-like P1 phenotype was further supported by the concurrent induction of the EpCAMhighCD133+population in infected cells (Fig. 5F). Together, these data demonstrate a role for Runx3 in safeguarding gastric epithelial cells against the aberrant induction of EMT and cellular plasticity by TGF-β1.


In previous studies, RUNX3 was established as a tumor suppressor in the gastrointestinal epithelium [7, 11]. In the stomach, the loss of RUNX3 expression or its mislocalization is strongly associated with the onset and progression of human gastric cancer [7, 12]. In mice, genetic targeting of Runx3 leads to a hyperproliferative gastric epithelium with altered differentiation of gastric epithelial cells that is consistent with a precancerous state [7]. As C57BL/6 Runx3−/− mice suffer neonatal death, the GIF cell lines were established from the fetal stomach in a p53-null background to enable detailed studies of the cellular functions of Runx3 [13]. Of note, Runx3−/− GIF cells are susceptible to intestinal transdifferentiation and the loss of epithelial properties, such as the ability to form columnar epithelial structures and to maintain a polarized phenotype, including tight junctions [13, 14]. Despite these clear in vivo and in vitro phenotypes, the molecular and cellular basis of the tumor suppressor function of Runx3 remains to be fully understood. This study shows that, in the absence of Runx3, gastric epithelial cells are prone to spontaneous EMT that gives rise to a tumorigenic and stem-like subpopulation (P2) marked by the gastrointestinal stem cell marker Lgr5. This process was shown to be driven by the dysregulation of the TGF-β and Wnt signaling pathways as a consequence of the loss of Runx3.

RUNX transcription factors are integral components of the TGF-β/BMP pathways in multiple tissue contexts, including hematopoiesis, immunity, osteogenesis, and epithelial growth regulation [25]. In the gastric epithelium, RUNX3 mediates the tumor suppressive effects of TGF-β by cooperating with Smad proteins to transcriptionally regulate p21WAF-1 and Bim [8–10]. Consistent with this, Runx3−/− GIF lines are resistant to TGF-β1-mediated apoptosis and are tumorigenic when transplanted into immunodeficient nude mice [7]. The results of this study show that, paradoxical to their resistance to TGF-β1-induced growth suppression, Runx3−/− GIF lines were sensitized to TGF-β1 to undergo EMT, a phenomenon that was not observed in Runx3+/+ lines. Evidently, in the absence of Runx3, the TGF-β signal is redirected from its growth suppressive to its morphogenetic functions in gastric epithelial cells. The pleiotropic effects of TGF-β pathway in cancer have often been confounding: while it functions as a prominent tumor suppressor pathway early in gastrointestinal carcinogenesis, it is a major promoter of tumor progression and metastasis in later stages [26]. The latter phenomenon has been attributed to the ability of TGF-β to activate EMT, which during the normal developmental process enables the reprogramming and migration of epithelial cells [5]. However, aberrant reactivation of EMT in carcinoma endows carcinoma cells with greater adaptability, migratory capabilities, and chemotherapeutic resistance [27].

In recent years, the reprogramming properties associated with EMT have been linked to increased cellular plasticity in carcinoma cells, as demonstrated in its ability to induce a “cancer stem cell” phenotype in immortalized human mammary epithelial cells and in breast carcinoma [28–30]. We now report the same phenomenon in gastric epithelial cells, where the activation of EMT in Runx3−/−p53−/− GIF-14 cells was accompanied by the unexpected induction of Lgr5, an exclusive gastric stem cell marker (Fig. 5C) [22]. The induction of Lgr5 during TGF-β-induced EMT is surprising because it is hitherto known only as a Wnt target. Interestingly, the induction kinetics of Lgr5 differed significantly from the transient induction of Snai1, implicating a secondary effect that might involve protracted epigenetic mechanisms, such as chromatin remodeling. Indeed, a greater accessibility to the Lgr5 locus following prolonged TGF-β1 treatment was supported by a sharp increase in the Wnt3a inducibility of Lgr5 (Fig. 5C). These observations are significant for several reasons: first, the reactivation of Lgr5 indicates a profound increase in cellular plasticity and the induction of stem-like cells [22, 24]. Moreover, the synergistic action of TGF-β and Wnt suggests that these remodeled cells are concurrently sensitized to the oncogenic effects of the Wnt pathway due to the absence of Runx3, a known molecular antagonist [16]. Lastly, the dysregulation of these two pathways provides a molecular basis for the tumorigenicity observed for the associated P2 subpopulation. Taken together, these data indicate that a loss of Runx3 in gastric epithelial cells would render them susceptible to aberrant TGF-β and Wnt signaling, spontaneous EMT, and increased cellular plasticity. This in turn induces a subpopulation of mesenchymal- and stem-like cells that displays greater tumorigenicity in vitro and in vivo (as summarized in Fig. 6).

Figure 6.

A model depicting the role of Runx3 in safeguarding gastric epithelial cells against EMT and tumorigenicity. The loss of Runx3 and p53 functions in gastric epithelial cells causes increased cellular plasticity that renders gastric epithelial cells susceptible to EMT due to dysregulated TGF-β signaling (P1). This activity gives rise to a mesenchymal-like subpopulation that displays increased stem-like properties, such as Lgr5 expression (P2). The present data suggest that an aberrantly active Wnt pathway further fuels these EMT-induced stem-like cells to be tumorigenic. Abbreviation: EMT, epithelial-mesenchymal transition.

It bears highlighting that all GIF lines were established against a p53-null background, as its absence could account for part of the observed plasticity. Indeed, the restrictive role of p53 is a common feature of EMT and the phenomenon of somatic cell reprogramming, both of which induce plasticity in differentiated cells in vitro. Inactivation of the p53 network greatly enhances the generation of induced pluoripotent stem cells through the escape of p53-mediated stress responses, such as apoptosis and senescence [31, 32]. Similarly, the evasion of senescence through the inactivation of p53 facilitates the induction of EMT by transcription factors Twist1, Twist2, and Zeb1; as well as growth factors, such as EGF and TGF-β [33–37]. Furthermore, it was recently reported that p53 regulates EMT and stem cell properties through its regulation of EMT-inhibiting miRNAs [38, 39]. Therefore, it is likely that the p53-null background of the GIF cell lines would reduce the cellular threshold for entry into EMT. In this light, it is remarkable that Runx3+/+p53−/− GIF lines remain resistant to EMT in the absence of p53, underscoring the protective role of Runx3 against cellular plasticity.

It is important to note that this novel role of RUNX3 in gastric epithelial cells is consistent with the overall functions of RUNX proteins. All known members of the RUNX family play critical roles in cell fate determination and lineage decisions in different tissues and species [40–45]. In particular, Runx1 is a “master regulator” of definitive hematopoiesis and is indispensable from the emergence of the first hematopoietic stem cells to the proper differentiation of multiple lineages, often acting in conjunction with Runx3 [42, 46–49]. The presence of these RUNX proteins during hematopoiesis guides and safeguards differentiation toward the full range of hematopoietic lineages. In this study, a similar function for Runx3 in the safeguarding of gastric epithelial cells against EMT-induced phenotypic plasticity has been revealed. This anti-EMT property of Runx3 is concordant with its analogous role recently reported in lung epithelial cells [50]. In GIF-14 cells, this was demonstrated by the strong reduction of Lgr5 and Sox9 expression and the ablation of P2 cells following the reintroduction of RUNX3 (Fig. 5D, 5E). Furthermore, exogenous RUNX3 also blocked TGF-β1 induction of Snai1 and Hmga2. While the precise mechanism underlying Runx3 modulation of TGF-β signaling in these cells is currently obscure, its negative regulation of Hmga2 is significant, as it transcriptionally regulates Snai1 and is associated with EMT in multiple cancer types [51, 52]. Importantly, Hmga2 plays an important role in the maintenance of stemness in normal and cancer stem cells [53, 54].

Finally, the protective effects of Runx3 on the gastric epithelial phenotype revealed in the current study fits neatly with the disrupted differentiation and altered epithelial phenotype observed in Runx3−/− GIF cells [13, 15]. Recently, adult BALB/c Runx3-null mice were reported to display gross antralization and intestinalization of the gastric epithelium, which were marked by a loss of mature chief cells and the ectopic expression of Cdx2, respectively [11]. Upon carcinogenic insults, these mice develop striking TFF2- and Muc6-expressing lesions akin to spasmolytic polypeptide expressing metaplasia, a condition strongly associated with human gastric carcinogenesis [11, 55]. Together, these observations indicate that the Runx3-null gastric epithelium exists in a precancerous state, reflecting the tumor suppressor function of Runx3. This study provides a novel molecular and cellular basis for understanding these in vivo observations.


Although the disruption of RUNX3 function is frequently observed in gastric cancer, the cellular consequences are not fully understood. This study reports that in the absence of Runx3, gastric epithelial cells are prone to spontaneous EMT due to the simultaneous dysregulation of TGF-β and Wnt signaling. This gives rise to a tumorigenic stem cell-like subpopulation that expresses the gastric stem cell marker Lgr5. These data demonstrate a protective role for RUNX3 in safeguarding gastric epithelial cells against EMT-induced plasticity and tumorigenicity. Furthermore, these data implicate a significant contribution of EMT-induced epithelial plasticity during the early stages of gastric carcinogenesis.


We express our gratitude to Kristoffer Weber and Carol Stocking for providing the LeGO lentiviral vectors; Vinay Tergaonkar for sharing his technical expertise in lentivirus generation; Lynnette Chen for technical support with FACS and flow cytometry; and Motomi Osato and Vaidehi Krishnan for helpful discussions. Lastly, we wish to thank Juin Hsien Chai for his technical support. This work is supported in part by the National Research Foundation (NRF), Translational and Clinical Research (TCR) Flagship Programme, and the Singapore Stem Cell Consortium (SSCC).


The authors declare no conflict of interest.