BMP4 Promotes EMT and Mesodermal Commitment in Human Embryonic Stem Cells via SLUG and MSX2



Bone morphogenetic proteins (BMPs) initiate differentiation in human embryonic stem cells (hESCs) but the exact mechanisms have not been fully elucidated. We demonstrate here that SLUG and MSX2, transcription factors involved in epithelial–mesenchymal transitions, essential features of gastrulation in development and tumor progression, are important mediators of BMP4-induced differentiation in hESCs. Phosphorylated Smad1/5/8 colocalized with the SLUG protein at the edges of hESC colonies where differentiation takes place. The upregulation of the BMP target SLUG was direct as shown by the binding of phosphorylated Smad1/5/8 to its promoter, which interrupted the formation of adhesion proteins, resulting in migration. Knockdown of SLUG by short hairpin RNA blocked these changes, confirming an important role for SLUG in BMP-mediated mesodermal differentiation. Furthermore, BMP4-induced MSX2 expression leads to mesoderm formation and then preferential differentiation toward the cardiovascular lineage. Stem Cells 2014;32:636–648


One of the hallmarks of gastrulation is the formation of primitive streak and specification of the three germ layers from which all cells of the body form. Proliferation and migration of epiblast cells takes place as they undergo an epithelial–mesenchymal transition (EMT). The availability of human pluripotent stem cells, both embryonic stem cells (hESCs) and induced pluripotent stem cells (hiPSC) through reprogramming, that resemble epiblast has allowed studies typically carried out in mice or lower vertebrates to be extended to humans. Understanding signaling cascades that control differentiation and reprogramming not only sheds light on the early stages of human development but also leads to improvements in differentiation protocols necessary for future applications in understanding disease and regenerative medicine (reviewed in ref. [1, 2])

Many gene ablation studies in mice have demonstrated that the transforming growth factor β (TGF-β) superfamily has direct effect on embryonic development not only by enhancing self-renewal but also in directing their differentiation into the mesodermal lineage (reviewed in ref. [3]). The TGF-β superfamily also has a major role in both self-renewal and differentiation of hESC and mouse ESC (mESC), although there is some species dependence which is thought may be related to their different developmental states [4, 5]. The two ESC origins result in some disparity when the effect of a single cytokine is compared [6-8].

TGF-β is a multipotent growth factor that transduces signals from the membrane to the nucleus by binding to a heteromeric complex of serine/threonine kinase receptors known as TGF-β types I and II receptors that propagate the phosphorylation signal to receptor-regulated R-Smads. Activated R-Smads complex with Smad4 and accumulate in the nucleus where they regulate transcriptional activity of their target genes (reviewed in ref. [9, 10]). TGF-β signaling through ALK5 and TGF-β family members Activin and Nodal signaling through ALK4 result in the phosphorylation of Smad2 and Smad3, whereas signaling of another family member bone morphogenetic protein (BMP) is through ALK2, ALK3, and ALK6, resulting in phosphorylation of Smad1, Smad5, and Smad8 (reviewed in ref. [11, 12]).

BMPs and their downstream effectors are particularly important in determining mESC's fate not only by facilitating self-renewal, but in the absence of JAK/STAT activation, causing them to become mesodermal instead of differentiating into the ectodermal lineage [13, 14]. BMPs phosphorylate Smad1/5/8 in mESCs and hESCs [6, 7, 15, 16] and upregulate Id proteins (inhibitors of differentiation) [8, 17]. Id proteins inhibit basic helix-loop-helix transcription factors, thereby blocking differentiation of mESC [17]. The effects on hESCs are somewhat less clear and appear to depend of the presence of other cofactors: for example, Xu et al. showed that repression of BMP signaling sustains undifferentiated hESCs, whereas BMP stimulation in conditioned medium containing bFGF promotes trophoblast differentiation [18] Pera et al. reported that BMP2 addition in the presence of serum enhances primitive endoderm differentiation in hESCs [15]. BMP4 can also induce hESC differentiation into mesoderm [16, 19-23]. Similarities between developmental and tumorigenic EMT suggest some common cooperative signaling pathways that include TGF-β superfamily members, the downstream Smads interact with many EMT-associated transcription factors [24] (reviewed in ref. [25]). EMT has been studied extensively at a molecular level in the mouse gastrula but equivalent studies in humans have not been carried out because embryos at the appropriate developmental stage are inaccessible. hESCs now represent a useful surrogate model to study EMT in pluripotent cells.

Here, we examined the early effects of BMP and TGF-β on hESCs and the later consequences for differentiation. We found that the EMT-associated transcription factors SLUG and MSX2 were strongly downregulated in TGF-β-treated hESCs compared with cells treated with the ALK4/5/7 inhibitor SB-431542 and highly upregulated in BMP4-treated hESCs compared with untreated cells. Immunostaining showed that Smad1/5/8 colocalized with SLUG at the very edges of the colonies at the time that differentiation starts to occur. We showed that Smad1/5/8 bound directly to the SLUG promoter using the chromatin immunoprecipitation (ChIP) approach. Microarray analysis showed that mesodermal and, specifically, cardiac genes were enriched in BMP-treated hESCs compared with untreated cells. There were more areas of beating cardiomyocytes in hESCs treated with BMP4, indicating that enhanced expression of mesodermal genes also increased the differentiation of at least one mesodermal derivative. Thus, SLUG and MSX2 appear to be early target genes of BMP in hESC that lead to EMT and subsequent formation of mesoderm with cardiomyogenic potential.

Materials and Methods

Cell Culture and Differentiation of hESC

HUES9, HES2, and HES3 hESCs were obtained from the Melton Lab, Howard's Hughes institute [26]) and WiCell Research Institute (Madison, WI) as trypsin adapted variant from the Keller laboratory and cultured according to each laboratory protocol. The Icelandic Bioethics committee has approved the usage of hESC lines for this research. hESC lines were routinely cultured on a monolayer of Mitomycin C-treated primary mouse embryonic fibroblasts (MEFs CF-1 strain, ATCC, Manassas, VA, in medium containing 80% Dulbecco's modified Eagle's medium (DMEM)/F12 (Invitrogen, Carlsbad, CA, supplemented with 20% knockout serum replacement (Invitrogen), 1% Pen/Strep (Invitrogen), 200 nM l-glutamine (Invitrogen), 1% nonessential amino acids (Invitrogen), 55 mM 2-mercaptoethanol (Invitrogen), and 4 ng/ml basic fibroblast growth factor (bFGF) (PeproTech, London, UK, For feeder-free hESC culture, cells were dissociated with TrypLE (Invitrogen) and seeded in serum-free medium including 25% TeSR1 medium (Stem Cell Technologies, Vancouver, Canada, on Matrigel (Becton Dickinson (BD),; diluted 1:100 in DMEM/F12 medium).

hESC Differentiation Using the Spin EB Method

hESCs grown on MEFs were split (1:2) on growth factor reduced (GFR) Matrigel-coated dishes (1:100) in conditioned medium containing hESC medium supplemented with 10% MEF conditioned medium. Alternatively, mTeSR1 medium was used as indicated. Next day, the cells were dissociated with TrypLE and 3,000 cells were transferred to 96-well V-shaped low attachment plates in BSA polyvinylalchohol essential lipids (BPEL) medium [27] with or without cytokines. The cells were then forced to aggregate and form embryoid bodies (EBs) by centrifugation (at 1,100 rpm for 2 minutes). Seven-day-old EBs were transferred onto gelatin-coated 96-well flat-bottom plates and checked for the formation of beating areas over time. The number of beating areas was counted in two 96-well plates per condition and compared. Videos were taken of beating areas (Supporting Information Video 1).

hESC Treatment with Growth Factors

hESCs were treated with different growth factors and inhibitors at indicated time points or were left untreated then medium and growth factors refreshed daily as appropriate. In addition to bFGF in the medium, growth factors were added as follows: TGF-β (5 ng/ml, PeproTech), Activin A (10 ng/ml, PeproTech), BMP4 (10 ng/ml, PeproTech), the ALK4/5/7 inhibitor (TGF-β inhibitor) SB-431542 (10 μM, Tocris, Bristol, UK,, and the BMP antagonist Noggin (300 ng/ml, PeproTech).

Reverse Transcriptase PCR

Total RNA from hESC was extracted using RNeasy kit (Qiagen), and cDNA was synthesized with Superscript II reverse transcriptase (Invitrogen) according to the manufacturer's protocol. The genes of interest were amplified for 25–35 cycles using primer sets and annealing conditions given in Supporting Information Table S1.

Quantitative Real-Time PCR

Total RNA was isolated with TRIzol reagent (Invitrogen) according to the manufacturer's protocol. cDNA was synthesized using SuperScriptIII First-Strand Synthesis System (Invitrogen). Real-time polymerase chain reaction (PCR) were performed with Maxima SYBR Green/ROX qPCR Master Mix (2X; Fermentas, St.LeonRot, Germany, according to manufacturer's instructions using the 7,500 real-time PCR cycler system (Applied Biosystems). Human ARP gene was used as an internal control to equalize cDNA loading. The genes of interest were amplified for 40 cycles using 60°C for 1 minute as annealing temperature. Primer sets are given in Supporting Information Table S1. Relative expression was calculated using the comparative Ct method. Three separate experiments were performed and quantification of three replicates of a typical experiment is shown.

Western Blotting

Cells were lysed in lysis buffer (20 mM Tris, pH 7.4, 150 mM sodium chloride (NaCl), 10% glycerol, and 1% Triton X-100) including a cocktail of protease inhibitors (Roche), sonicated and boiled. Protein samples were separated on 10% SDS-polyacrylamide gel electrophoresis and transferred to nitrocellulose filter, Hybond-C Extra membrane (GE Healthcare, Pittsburgh, PA,, or alternatively to Immobilon-FL polyvinylidene difluoride (PVDF) membrane (0.45 μm; Merck Millipore). After blocking in tris-buffered saline-tween 20 (TBS-T) and 5% bovine serum albumin (BSA), blots were incubated overnight at 4°C with primary antibodies against phospho-Smad1 (rabbit polyclonal, 1:1000, Cell Signaling,, phospho-Smad2 (rabbit polyclonal, 1:1000, Cell Signaling), ACTIN (mouse monoclonal, 1:20000, Millipore), T-BRACHYURY (rabbit monoclonal, 1:100, Santa Cruz Biotechnology, Santa Cruz, CA,, and hemagglutinin (HA) (1:1000, mouse monoclonal, Roche) in TBS-T + 5% BSA and secondary antibody (horse radish peroxidase-linked rabbit IgG from donkey 1:5000, GE Healthcare). Detection was performed by enhanced chemiluminescence (GE Healthcare). Alternatively, proteins were detected by Odyssey Infrared Imaging system (Li-Cor Biosciences Bad Homburg, Germany, In this case, blots were blocked in 5% BSA and incubated with secondary antibodies, anti-mouse IgG (H + L; DyLight 800 Conjugate), or anti-rabbit IgG (H + L; DyLight 680 Conjugate; Cell Signaling) in 0.01% SDS in TBS-T.


hESCs were cultured in four-well chamber slides (BD Falcon) on MEFs or slides were coated with GFR Matrigel and CELLstart (Invitrogen), respectively. After seeding 1.5 × 104 hESCs/well, cells were treated with and without growth factors and inhibitors in differentiation medium at indicated time points. Cells were fixed in 2% paraformaldehyde (PFA) for 30 minutes, washed, and permeabilized with 0.1% Triton X-100 at room temperature (RT) for 8 minutes. The cells were preincubated with 4% goat serum (Invitrogen) and normal rabbit IgG (1:500; Santa Cruz Biotechnology) at RT for 1 hour before incubation with primary antibody overnight at 4°C against OCT4 (mouse monoclonal, 1:300, Santa Cruz Biotechnology), SOX-2 (rabbit polyclonal, 1:250, Millipore Bedford, MA,, phospho-Smad1 (rabbit polyclonal, 1:100, Cell Signaling), phospho-Smad2 (rabbit polyclonal, 1:100, Cell Signaling), SLUG (A7, mouse monoclonal, 1:250, Santa Cruz Biotechnology), T-BRACHYURY (rabbit polyclonal, 1:100, Santa Cruz Biotechnology), MSX2 (M-70, rabbit polyclonal, 1:100, Santa Cruz Biotechnology and AB10211, rabbit polyclonal, 1:100, Merck Millipore), E-cadherin (BD610182, mouse monoclonal, 1:200, BD Bioscience), and ISLET-1 (mouse monoclonal, 1:250, DSHB, Iowa city, IA, After washing with PBS–0.05% Tween, the cells were incubated with secondary antibody at RT for 1 hour against Alexa Fluor 488 goat-anti-rabbit IgG (1:1000; Invitrogen) and Cy3-conjugated anti-mouse IgG (1:250; Jackson ImmunoResearch, After washing with PBS–0.05% Tween and optional nuclear staining with Topro3 (Invitrogen; 1:1000 in water), the coverslips were embedded in Fluoromount (Sigma-Aldrich). Immunofluorescence was visualized using a confocal Zeiss LSM 5 Pascal laser scanning microscope.

Chromatin Immunoprecipitation

hESCs were dissociated using TrypLE and deprived of MEFs on CeLLStart in 50% hESC medium + 50% MEF CM medium on 10 cm dishes (three plates per condition) and either stimulated (or not) as before for 3 days. Cells were fixed with 1% formaldehyde at RT for 10 minutes with agitation. After centrifugation, cross-linking was stopped with 0.125 M glycine. Cells were washed with PBS and resuspended in cold cell lysis buffer. Cells were centrifuged and nuclei resuspended in radioimmunoprecipitation assay buffer (RIPA) with protease inhibitors. After storage on ice for 10 minutes, samples were sonicated (Diagenode Bioruptor Liege, Belgium) and centrifuged at 4°C for 5 minutes. Samples were resuspended in RIPA buffer and precleared with Dynabeads Protein G (Invitrogen) for 1 hour at 4°C. For immunoprecipitation, 500 μl of chromatin samples were used per reaction including antibody of interest and incubated at 4°C overnight. After magnetic isolation, samples were washed four times in RIPA buffer and once with IP washing buffer and at last washed with TE. Antibody/protein/DNA complexes were eluted and reverse cross-linked by incubation in 0.3 M NaCl at 65°C for 6 hours. Genomic DNA was precipitated and PCR analysis was carried out (Supporting Information Table S1).

RNA Extraction and Microarray Expression Analysis

Total RNA was isolated from hESCs using RNeasy Mini kit (Qiagen). The quantity and quality of RNA was measured on Nanodrop spectrophotometer and on Agilent 2,100 Bioanalyzer RNA chips. Human reference sample (Stratagene, La Jolla, CA) was used as a control between different microarray experiments. The experimental procedures for human expression analysis were performed according to the Roche-NimbleGen expression analysis technical manual. Briefly, total RNA was used to synthesize cDNA, which was labeled with Cy3 and then hybridized to a human 72 K 4-plex array. Data were analyzed in ArrayStar 2.0 software from DNASTAR.

Adenoviral Constructs

hESCs on Matrigel (GFR) are infected with adenovirus-expressing LacZ or caALK2/HA using a multiplicity of infection of 100. After 24 hours, the cells were washed and allowed to recover for 24 hours before using in the indicated assays. To confirm a successful adenoviral infection of a given construct, cells were harvested for Western blotting.

Lentiviral Constructs

LLCIEP control overexpression vector (EV; kind gift from Dr. W. Gallagher) or MSX2, pLKO.1 control vector (P27) or with shMSX2 plasmid (equal amount of clone #20434, #20435, #20436, #20437, #20438) and pLKO.1 control vector or with shSLUG (equal amount of clone #4321, #4322, #4323, #4324, #4325; Mission SIGMA-Aldrich oligos, were transfected together with packing and envelope plasmids (psPAX2 and pMD2.G; Addgene, in human embryonic kidney (HEK)−293T cells using Lipofectamine PLUS (Invitrogen) transfection reagent according to manufacture's instructions. The medium was changed 3 hours after transfection using hESC-medium. The viral supernatant was harvested after 24 and 48 hours, combined, centrifuged (5 minutes; 140g), and supernatant was filtered (0.45 μm). hESCs were dissociated with TrypLE and seeded in mTeSR1 medium on Matrigel-coated 12-well dish with indicated viral particles and 10 μg/ml polybrene (Sigma-Aldrich) for viral infection. After 24 hours, 1 ml of mTesR1 medium was added to each well, and the day after, medium was replaced with 2 ml F12/DMEM–mTesR1 (50:50) medium. Puromycin (Merck) of 0.8 μg/ml was used to select successfully transduced cells. To confirm the knockdown of short hairpin RNA (shRNA)-SLUG and shRNA-MSX2, cells were harvested for quantitative real-time PCR (qRT-PCR) and immunofluorescence.

hESC Migration Assay

Migration was measured using Transwell inserts. A Falcon nucleopore membrane (8 μm pore) was coated with GRF Matrigel (1:100) overnight at 4°C. The lower chamber was filled with hESC medium with or without BMP4. hESCs were dissociated with TrypLE and suspended at a final concentration of 1 × 105cells/ml. The suspension of 150 μl was added to the upper chamber and incubated for 3 days. Cells were washed and either upper or lower surface was wiped to remove cells depending on the assessment of nonmigrating or migrating cells. The membranes were fixed in 3% PFA and methanol followed by staining with crystal violet and counting.

Growth Curve

A total of 2 × 104 cells per well of the infected hESC (MSX2/control overexpression vector EV and shMSX2/control vector P27) were plated onto Matrigel coated 24-well plate and cultured under standard conditions. Three wells of each culture condition were harvested and counted every 12 hours over a period of 4 days. The mean cell number per measurement of each cell line was used to calculate the growth curve.

Statistical Analysis

Experiments were conducted in triplicate, and the data were represented as mean value ± SD. Significance of differences was measured by an unpaired Student's t-test.


BMP4 Induces hESC Differentiation Through the BMP4-Smad1/5/8 Pathway

To investigate the mechanisms through which the TGF-β superfamily affects hESC behavior, we stimulated the HUES9 cells with the two main TGF-β superfamily members, TGF-β1 and BMP4 and their inhibitors (Fig. 1A). To determine which concentration to use for an active BMP4 signaling cascade in hESCs, the phosphorylation levels of endogenous Smad1/5/8 were examined after stimulation of three different concentrations of growth factors at different time points (Fig. 1B, 1C). The BMP cascade is only active for up to 6 hours; hence medium with fresh growth factors was changed daily. HUES9 were cultured for 3 days in serum-free hESC medium including bFGF, with or without different stimuli. Untreated and TGF-β-treated hESC colonies maintained their pluripotent characteristics, growing in discrete colonies on MEFs with large nuclei and little cytoplasm (Fig. 1D, a, d). Moreover, hESCs were immunofluorescently labeled with antibodies against OCT4 and SOX2. Untreated and TGF-β-induced hESCs highly expressed both transcription factors, indicating pluripotency (Fig. 1D, b, c, e, f). The expression of OCT4 and SOX2 decreased when Smad2/3 activation was blocked by the ALK4/5/7-inhibitor SB-431542 (Fig. 1D, h, i). When hESCs were treated with BMP4, colonies became unusually large, ranging from 400 to 600 μm in diameter, instead of the usual 200–300 μm when cells were untreated (Fig. 1D, j). In the additional presence of the natural BMP antagonist Noggin [28], however, this did not occur (Fig. 1D, l, n, o). Interestingly, initial observations did not indicate that the cells were differentiating on BMP4 treatment as they maintained strong OCT4 expression (Fig. 1D, k). However, further examination revealed that the cells had at the same time lost expression of SOX2 (Fig. 1D, l). This indicated that expression of SOX2 was reduced early in the differentiation process, while the cells maintain expression of OCT4 for a longer period of time.

Figure 1.

Bone morphogenetic protein 4 (BMP4) induces human embryonic stem cell (hESC) differentiation via the BMP-Smad1/5/8 pathway. (A): Overview of experimental setup. (B): Western blot analysis on hESC extracts. pSmad1/5/8 was examined in hESCs after stimulation with different concentrations of BMP4 and transforming growth factor β (TGF-β) for 1 hour. Asterisks indicate nonspecific bands. (C): pSmad1/5/8 was analyzed in hESCs after BMP4 treatment (10 ng/ml) in a time-dependent manner. (D): Bright field and immunofluorescent staining of untreated or treated hESCs with TGF-β or SB-431542, BMP4, or Noggin for 3 days. Cells were fixed and stained with the primary antibodies mouse anti-OCT4 and rabbit anti-SOX2. Secondary antibodies used were anti-mouse Cy3 (red) and anti-rabbit AlexaFluor488 (green), respectively. Abbreviations: BMP, bone morphogenetic protein; IF, immunofluorescence; TGF-β, transforming growth factor β; WB, western blotting.

Thus, our results showed that BMP4 is a potent inducer of transition from an OCT4+/SOX2+ undifferentiated hESCs to an OCT4+/SOX2 partially differentiated state.

The EMT-Related Transcription Factors SLUG and MSX2 are Upregulated in BMP4-Induced hESC

We performed whole-genome microarray analysis on total RNA (72 K 4-plex expression arrays from Roche-NimbleGen) to explore differential gene expression regulated by TGF-β superfamily signaling in hESCs. As before, HUES9 cells were untreated or treated with TGF-β, SB-431542 inhibitor, or BMP4 for 3 days in serum-free hESC medium, including bFGF. To confirm that the procedure was successful, we started by comparing TGF-β-treated hESCs with cells treated with SB-431542 (Supporting Information Table S2). The reason for using SB-431542-treated cells as a comparison was because untreated cells already have strong activation of Smad2/3 likely through paracrine mechanisms and, therefore, have similar characteristics and gene profile as TGF-β-induced cells (Supporting Information Fig. S1A). The majority of genes important for self-renewal maintenance, such as NODAL, NANOG, SOX2, OCT4, TDGF1 (CRIPTO-1), LEFTY-A, LEFTY-B, GDF3, and IDO were highly upregulated in TGF-β-induced hESCs. This indicated that the approach was reliable as TGF-β-Smad2 activation is believed to maintain self-renewal in hESC via Smad2-NANOG interaction [29] (Supporting Information Fig. S1B). Interestingly, some EMT-related transcription factors were strongly downregulated in TGF-β-induced hESCs. SLUG and MSX2 were downregulated by nearly ninefold compared with SB-431542-subjected cells. SLUG and MSX2 are common mediators of EMT, induced, for instance, by the TGF-β superfamily, a process important in gastrulation and later development including that of the heart [24].

In contrast to downregulation following TGF-β addition, SLUG and MSX2 genes were upregulated in BMP4-induced hESC (Fig. 2A; Supporting Information Table S3). In another hESC line, HES3, we confirmed BMP4-induced upregulation of SLUG and MSX2 expression at mRNA level by qRT-PCR (Fig. 2B). Interestingly, the expression of other known EMT-related transcription factors such as SNAIL, TWIST 1 and 2, ZEB 1 and 2, and HMGA2 did not show opposite responses to TGF-β and BMP4 like SLUG and MSX2 (Fig. 2A), although TWIST was upregulated upon BMP4 stimulation in HES3 at the RNA level (Supporting Information Fig. S2). The results suggest a role of BMP4 in facilitating differentiation of hESC preferentially via upregulation of the EMT-related transcription factors SLUG and MSX2.

Figure 2.

SLUG and MSX2 are upregulated in bone morphogenetic protein 4 (BMP4)-induced human embryonic stem cells (hESCs) and downregulated in transforming growth factor β (TGF-β)-stimulated cells. (A): Gene expression profile of epithelial–mesenchymal transitions (EMT) transcription factors in TGF-β-stimulated cells versus SB-431542 inhibitor-treated cells and in BMP4-stimulated cells compared with untreated cells for 3 days. (B): Quantitative real-time polymerase chain reaction for EMT factors MSX2 and SLUG in HES3 cells in response to BMP4 at different time points. Each bar represents the mean ± SEM. Human ARP was used as an internal control. All values were normalized to the mRNA level (= 1) of the control cells (ctrl) without BMP4 stimulation. Abbreviations: BMP, bone morphogenetic protein; EMT, epithelial–mesenchymal transitions; hESC, human embryonic stem cell; TGF-β, transforming growth factor β.

BMP4-Induced Smad1/5/8 Phosphorylation Colocalizes with SLUG and Directly Binds to the SLUG Promoter

James et al. and Besser reported that BMP4-induced Smad1/5/8 activation correlates with differentiation [7, 30], but the underlying mechanisms are still debated. To address this, we first investigated the relative distributions of phosphorylated Smad1/5/8 and SLUG in HES3 colonies exposed to BMP4 in serum-free medium with bFGF by immunostaining (Fig. 3A). Smad1/5/8 phosphorylation is stronger in BMP4-induced hESCs than untreated hESCs (Fig. 3A, g, h). It is noteworthy that Smad1/5/8 phosphorylation concentrated at the edges of the hESC colonies, exactly where cells are more prone to differentiate (Fig. 3A, h, i). The expression of SLUG was similarly much stronger in BMP4-induced hESCs than untreated cells and in fact colocalized with phospho-Smad1/5/8 (Fig. 3A, d–f, k–m). The expression of SLUG was abolished when HES2 were treated with the BMP4-antagonist, Noggin (Supporting Information Fig. S3A). When hESCs were infected with adenovirus containing construct of constitutively active (ca) BMP type I receptor ALK2/HA and lacZ, respectively, robust differentiation occurred, that is, morphological changes appeared, compared with lacZ infected cells (Supporting Information Fig. S3A). Smad1/5/8 was highly activated in caALK2/HA expressing cells and colocalized with SLUG expression, recapitulating the observations in BMP4-treated cells (Supporting Information Fig. S3A). Western blot was performed using an HA antibody on cell lysates from infected cells to confirm caALK2 overexpression (Supporting Information Fig. S3B). This proved that BMP4-induced Smad1/5/8 phosphorylation and SLUG colocalized within the outer boundaries of the hESC colonies and BMP4 could induce this via the ALK2 receptor.

Figure 3.

Bone morphogenetic protein 4 (BMP4)-induced differentiation is mediated through activation of the transcription factor Smad1/5/8 and colocalizes with SLUG at the edges of colonies. (A): Immunofluorescent staining of untreated and BMP4-induced human embryonic stem cells (hESCs) showing Smad1/5/8 phosphorylation on the edges of the colony using different magnifications (h, i). Cells were treated with BMP4 for 1 day and then fixed and stained with primary antibody rabbit anti-phosphorylated Smad1/5/8 and mouse anti-SLUG. Secondary antibody used was anti-rabbit AlexaFluor488 (green), anti-mouse Cy3 (red), and Topro3 (blue) as counter staining. (B): BMP4-treated hESCs were harvested after 3 days for ChIP analysis. Samples were immunoprecipitated with anti-phospho Smad1/5/8 antibody or IgG antibody as a negative control. Polymerase chain reaction on samples was performed after DNA purification using primers for the human SLUG promoter. Abbreviations: BMP, bone morphogenetic protein; shSLUG, short hairpin RNA SLUG.

Because phospho-Smad1/5/8 and SLUG colocalized, we hypothesized that direct binding occurs between BMP4-induced Smad1/5/8 and the SLUG promoter, which in turn induces SLUG expression and thereby promoting hESC differentiation. To demonstrate direct binding, we performed ChIP assays on hESCs treated with BMP4 for 3 days and immunoprecipitated the cross-linked DNA–protein complexes with antibodies raised against phosphorylated Smad1/5/8. Primers for SLUG proximal promoter region were used in PCR on the samples. The core region of the SBE (Smad Binding Element) is AGAC and the proximal promoter region is 1.1 kb upstream of the first exon of the human SLUG gene [31]. Figure 3B shows the binding of phospho-Smad1/5/8 to the SLUG promoter in ChIP samples from BMP4-induced HES2 cells. This observation indicates that BMP4-induced differentiation likely occurs via direct binding of phospho-Smad1/5/8 to the SLUG promoter that in turn upregulates SLUG expression resulting in morphological changes in hESCs that reflect EMT-like differentiation.

BMP4-Induced EMT Mediated by SLUG Disrupts E-Cadherin and hESCs Become Migratory

Pluripotent hESCs are known to express an epithelial plasma membrane protein profile. BMP4-mediated SLUG differentiation is most likely disrupting the formation of these epithelial adhesion proteins in hESCs. We demonstrated that E-Cadherin is downregulated or at least becomes disrupted in BMP4-treated HES3 cells compared with untreated cells (Fig. 4A; Supporting Information Fig. S4A). SLUG seemed to play an important role in disrupting the E-cadherin arrangement because cells deprived of SLUG expression using lentiviral constructs expressing shRNA SLUG (shSLUG; Fig. 4B), expressed more E-Cadherin than control cells when stimulated with BMP4 (Fig. 4A). If cell adhesion decreases, cells become migratory. To show that BMP4-induced SLUG expression had biological relevance in hESCs, we studied their migratory behavior in Transwell inserts. We specifically blocked the expression of SLUG with lentiviral shSLUG constructs and examined whether knocking down SLUG would prevent migration of BMP4-induced hESCs. This would show whether SLUG was an essential intermediate in the BMP4-induced EMT in hESCs. We confirmed that SLUG expression was blocked by shSLUG by immunostaining (same lentiviral infection as in Fig. 4B). The cells were seeded on Transwell inserts to study their migratory behavior. BMP4-induced hESCs migration for 3 days was prevented in the shSLUG-treated cells whereas those treated with empty vector (P27) underwent EMT/differentiation in Transwells as controls (Fig. 4C). We conclude that BMP4/Smad1/5/8-induced SLUG expression in hESCs disrupts the formation of adhesion proteins, such as E-cadherin, in hESCs allowing them to become mesenchymal and migrate.

Figure 4.

E-Cadherin is disrupted during bone morphogenetic protein 4 (BMP4)-induced epithelial–mesenchymal transitions (EMT), and human embryonic stem cells (hESCs) undergo migration. (A): Immunofluorescent staining of untreated and treated HES3 with BMP4 for 4 days showing downregulated or disrupted E-Cadherin (E-Cad) expression in BMP4-treated HES3 cells. Cells were fixed and stained with primary antibody mouse anti-E-Cad, secondary antibody anti-mouse Cy3 (red), and Topro3 (blue) as counter staining. (B): Immunofluorescent staining of HES3 cells infected with control vector P27 and short hairpin RNA SLUG (shSLUG) construct. Cells were treated with BMP4 for 4 days, fixed and stained with primary antibody mouse anti-SLUG, secondary antibody anti-mouse Cy3 (red) and Topro3 (blue) as counter staining. (C): hESCs were infected with sh-SLUG lentiviral construct or control vector (P27). Infected cells were seeded in the upper chamber of a Matrigel-coated Transwell insert. In the lower chamber, serum-free medium was included with bFGF and with or without BMP4. Three days later, the cells that had migrated to the bottom of the filter were counted after staining with crystal violet. *, p < .01, #, p < .01 by Student's t-test. Abbreviations: BMP, bone morphogenetic protein; shSLUG, short hairpin RNA SLUG.

BMP4 Mediates Cardiomyocyte Differentiation in hESCs

Our microarray analysis showed that when compared with untreated hESCs, cells that were BMP4-induced for 3 days, a group of 30 genes were expressed approximately threefold higher (Supporting Information Table S3). These genes include T-BRACHYURY, ISL-1, GATA-3, cardiac muscle one alpha, HAND1, and MIXL1, which are all important in mesoderm or cardiac development although GATA-3 and MIXL1 are rather more associated with commitment toward hematopoietic cells [32]. T-BRACHYURY is one of the early mesodermal markers and is highly upregulated at the protein level upon BMP4 stimulation (Fig. 5A, e) compared with untreated hESCs and cells pretreated with Noggin (Fig. 5A, f). To extend these data to biological function, we used the spin EB method to induce cardiomyocyte differentiation. HES2 cells were cultured in BPEL medium with bFGF and were treated or not with growth factors/inhibitors for 3 days. After 12–18 days of differentiation (which included transfer of the EBs to a gelatinized 96-well flat bottomed plates), the number of beating areas formed was determined. Results showed that BMP4 is a potent inducer of cardiomyocyte differentiation in hESCs, as described previously by others [21, 23, 33-35]. The BMP4 effect was inhibited by Noggin indicating that it was BMP specific (Fig. 5B). TGF-β and ActivinA had no effect on the number of beating cardiomyocytes compared with untreated cells (Supporting Information Fig. S4B). Our data suggest that even at the concentrations as low as 10 ng/ml in serum-free culture with bFGF, BMP4 induces preferential differentiation of the hESCs toward the mesoderm and the cells start to express both mesodermal- and cardiac markers.

Figure 5.

Low concentrations of bone morphogenetic protein 4 (BMP4) stimulate differentiation of human embryonic stem cells (hESCs) into beating cardiomyocytes. (A): (a–c) Bright field and (d–f) immunofluorescent staining of hESCs. Cells were treated with or without low-dose of BMP4 or Noggin for 3 days. Cells were then fixed and stained with primary antibody rabbit anti-T-BRACHYURY (BraT). Secondary antibody used was anti-rabbit AlexaFluor488 (green). (B): hESCs were treated with or without BMP4 (10 ng/ml) and/or Noggin (300 ng/ml) in a serum-free medium including bFGF for 3 days and the number of beating areas assessed after 12 days of differentiation using the spin EB method. Beating cardiomyocytes were assessed in 96-well gelatin-coated plates. Abbreviation: BMP, bone morphogenetic protein.

MSX2 Mediates BMP4-Induced Cardiomyocyte Differentiation in hESCs

Our microarray data also showed strong upregulation of MSX2 by BMP4 in hESCs after 3 days (Fig. 2A). Based on the link between BMP4-Smad1/5/8 stimulation and SLUG-mediated EMT that leads to mesoderm then cardiomyocyte differentiation, we investigated whether MSX2 also played a role in this process. First, we studied the kinetics of BMP4-induced MSX2 expression in HES3 cells. MSX2 mRNA was already expressed after one day of BMP4 stimulation and strongly increased for up to 4 days. T-BRACHYURY was accordingly strongly upregulated on day 3 and 4 upon BMP4 stimulation (Fig. 6A) as well as HAND1 and ISLET-1 (Supporting Information Fig. S4C). As MSX2 appeared a direct target of the BMP/Smad1/5/8 pathway, we investigated the effect of MSX2 knockdown (shRNA) using lentiviral constructs on hESCs behavior. We infected hESCs with MSX2 shRNA constructs and stimulated the cells with BMP4 for 4 days on Matrigel after selection. The knockdown was verified quantitatively at the mRNA level using qRT-PCR. The expression of MSX2 was significantly reduced upon BMP4 stimulation at both RNA- (Fig. 6B) and protein- (Fig. 6C) levels. Interestingly, the early mesodermal marker T-BRACHYURY, was undetectable after BMP4 treatment in the shMSX2 hESCs both at the RNA- (Fig. 6B) and protein level (Fig. 6D). Furthermore, our immunostaining analysis indicated low levels of the early cardiac marker ISLET-1 in MSX2 knockdown hESCs and strong upregulation in MSX2 overexpressed hESCs after stimulation with BMP4 for 4 days on Matrigel (Fig. 6E). Morphological differences in hESCs culture were evident depending on the particular construct (MSX2/control overexpression vector EV and shMSX2/control vector P27); shMSX2-infected hESCs formed smaller and denser colonies in comparison with the other constructs (Supporting Information Fig. S5A). Growth curve analysis showed that neither overexpression nor knockdown of MSX2 affected the proliferation rate of hESCs (Supporting Information Fig. S5B). Hence, we suggest that MSX2 is an early target of BMP4 in hESCs that mediates mesodermal differentiation by upregulating T-BRACHYURY that in turn leads to cardiomyocyte differentiation. It likely co-operates with SLUG, although we have no direct evidence to demonstrate this.

Figure 6.

MSX2 mediates bone morphogenetic protein 4 (BMP4)-induced cardiomycocytes. (A): Quantitative real-time polymerase chain reaction (qRT-PCR) analysis of MSX2 and T-BRACHYURY (BraT) after BMP4 treatment in human embryonic stem cells (hESCs) at indicated time points. All values were normalized to the mRNA level (= 1) of the control cells (ctrl) without BMP4 stimulation. Human ARP was an internal control. Each bar represents the mean ± SEM. (B): qRT-PCR analysis after MSX2 knockdown with and without BMP4 treatment for 4 days in hESCs. All values were normalized to the mRNA level (= 1) of the control cells P27 without BMP4 stimulation. Human ARP was an internal control. Each bar represents the mean ± SEM (*, p = .0002; #, p < .0001 by Student's t-test). (C): Immunofluorescent staining of hESCs infected with control vector P27 and shMSX2 construct after treatment with BMP4 for 4 days. Cells were stained with primary antibody rabbit anti-MSX2 (Millipore), secondary antibody anti-rabbit AlexaFluor488 (green), and Topro3 (blue) as counter staining. (D): Western blot analysis of hESCs infected with control vector P27 and shMSX2 construct. T-BRACHYURY (BraT) was examined in hESCs after stimulation without and with BMP4 stimulation for 4 days. (E): Immunofluorescent staining of hESCs infected with control vector P27 and shMSX2 construct as well as control overexpression vector EV and MSX2 construct after treatment with BMP4 for 3–4 days in serum-free medium including bFGF. MSX2 knockdown and MSX2 overexpression hESCs stained with rabbit anti-MSX2 (Santa Cruz Biotechnology) and mouse anti-ISLET1 antibodies. Abbreviations: BMP, bone morphogenetic protein; EV, empty vector.


In this study, we described the effect of BMP4 on temporal gene expression in hESCs and showed that it had an effect on mesoderm commitment and possibly specification through EMT induced by SLUG possibly in collaboration with MSX2 (Fig. 7A).

Figure 7.

Schematic model for the roles of SLUG and MSX2 in early differentiation. Bone morphogenetic protein 4 (BMP4) activates Smad1/5/8 and in turn upregulates SLUG and MSX2. Upon SLUG upregulation, human embryonic stem cells (hESCs) undergo EMT and start migrating. MSX2 acts upstream of T-BRACHYURY and enhances mesodermal development of hESC. Abbreviations: BMP, bone morphogenetic protein; hESC, human embryonic stem cell; TGF-β, transforming growth factor β.

Microarray analysis indicated that expression of the EMT marker SLUG was strongly downregulated in hESCs activated by the TGF-β-Smad2/3 signaling cascade. By contrast, SLUG was highly upregulated in BMP-Smad1/5/8-induced hESCs. MSX2, which has also been linked to EMT, had the same expression profile as SLUG, suggesting a role for both transcription factors in differentiation in early human development. The expression profile of other well-known EMT-mediated transcription factors was also analyzed but none showed changes upon BMP4 stimulation of the same order as SLUG and MSX2. Nonetheless, the known EMT-related transcription factors SNAIL, TWIST, ZEB1, ZEB2, and HMGA2 were downregulated in TGF-β-induced hESCs, despite not being upregulated in BMP4 subjected cells in microarray analysis. It is possible that for these transcription factors the absence of TGF-β is more important than the presence of BMP4 although TWIST was shown to be upregulated upon BMP4 treatment in qRT-PCR.

From our data, we confirm that the onset of hESC differentiation is promoted by BMP4-induced Smad1/5/8 [15, 36] and postulate that this in turn upregulates expression of SLUG and phosphorylated Smad1/5/8 that are coexpressed within the outer boundaries of hESC colonies. This is in agreement with other reports that have shown expression of markers characteristic of EMT in hESCs on Matrigel [37-39]. Recent observation has shown that EMT occurs during hESC differentiation in Matrigel cultures based on decreased expression of E-cadherin followed by upregulation of N-cadherin and Vimentin [38]. Vimentin expression is under debate because it can be found in migrating epithelial cells and in cells undergoing mechanical stress [40]. The EMT mediators secreted by the adjacent myocardium are TGF-β, BMP, β-catenin, MSX2, and SLUG [24]. Loss of connection to the basement membrane and acquisition of migratory traits is a hallmark of EMT (reviewed in ref. [41]), and the proposition that BMP induces SLUG activation leading to EMT is therefore reasonable. Most likely, SLUG interferes with the E-cadherin-mediated formation of adherent junctions in early embryos as we observed in our studies in BMP4-treated hESCs.

The beginning of gastrulation in mammals is marked by the formation of the primitive streak in the region of the epiblast that will later form the posterior part of the embryo. Mapping studies have shown that different regions of the primitive streak produce different signals responsible for the induction of specific lineages. Members of the TGF-β, Nodal and Wnt families as well as a functional BMP gradient are essential for these developmental steps. BMP is also involved in several differentiation processes that require cell migration, such as neural crest formation [42]. Differentiation of hESCs in response to BMP stimulation can vary greatly depending on culture conditions. Different BMP concentrations, the duration of treatment, the absence or presence of feeder cells or serum in the medium and other growth factors, such as bFGF, can all play a role in controlling hESCs commitment into various lineages [15, 16, 18, 19, 21, 23, 43, 44]. Of note, serum is an unknown variable as it likely contains factors that interfere with BMP and TGF-β signaling cascades. Therefore, we analyzed the effects of low concentrations of BMP on hESCs in serum-free cultures with bFGF to provide more defined conditions to determine the effects of selected TGF-β members. MSX2 is a well-known BMP target in development [45, 46]; more precisely, a link has been shown between BMP signaling and MSX2 in regulation of endocardial cushion formation in mice [47, 48]. In addition, it has been shown to include a BMP-responsive upstream enhancer at least in mice [49]. MSX2 is a BMP target in both mESCs and hESCs [17, 50]. We showed here evidence of MSX2 being necessary for early differentiation of hESCs into mesodermal lineages; the cells started to express early mesodermal marker T-BRACHYURY, which is not expressed when MSX2 is knocked down. This is in line with recent reports showing that relatively low concentrations of BMP4 (12.5–25 ng/mL) induce differentiation into beating cardiomyocytes [21, 23]. Several studies have shown that BMP alone or with other growth factors induces cardiomyocyte differentiation [21, 23, 33, 35], which belong to the mesodermal lineage. It has been suggested that the effects of BMP4 on cardiac induction have a time window that is restricted to early stages of differentiation, and therefore identifying the temporal window in which BMP functions in cardiomyocyte differentiation is very important in understanding the mechanisms of commitment [33]. Recent studies have revealed that BMP4 induces hESCs toward subpopulations of mesoderm. These findings have led to the proposal that different subpopulations of mesoderm can be induced by different signaling pathways [21, 23] (reviewed in ref. [51]). It has already been shown that it is important to optimize each hESC line for Activin/Nodal and BMP signaling for specification of cardiac mesoderm [52]. Furthermore, recent data from Yook et al. showed that inhibition of ALK4/5/7 along with bFGF signaling leads to cardiac differentiation by increasing Smad1/5/8 phosphorylation [34]. They further state that BMP4 on its own failed to induce the cardiac differentiation and instead promoted trophoblast formation, as reported earlier [15, 18]. One explanation for various developmental pathways being induced by BMP stimulation could depend on the duration of BMP stimulation as suggested by Zhou et al. [50]. Another explanation is that it could reside in a different phosphorylation states of Smad1/5/8 that would lead to Smads 1, 5, and 8 being arranged in mixed combination of trimeric complexes. This would result in different transcriptional responses that would thereby facilitate different developmental pathways of hESCs. This paradigm has been suggested for TGF-β signaling in endothelial cells (reviewed in ref. [53]). Interestingly, previous observations have shown that miR-124a controls cell migratory activity during developmental transition of hESCs to EBs in vitro [39]. Moreover, miR-124 is expressed in hESCs but is gradually downregulated during EB formation, thereby allowing the expression of SLUG and IQGAP1 that promote cell migratory activity. Of note, besides SLUG and IQGAP1, the authors showed a list of potential target genes, one of which was Smad5. The question arises of whether the attenuation of miR-124 is not the only factor inducing SLUG expression but rather that the BMP cascade via Smad5 could play a role in facilitating SLUG-induced EMT in hESCs, as our results strongly suggest.

Recent observations by Li et al. showed that MET, the reverse of EMT, is required for nuclear reprogramming of mouse fibroblasts into iPSC [54]. Snail was suppressed by SOX2/OCT4, TGF-β components were downregulated by c-MYC while E-cadherin was induced by KLF4. Furthermore, some reports have shown that inhibition of TGF-β/Activin/Nodal signaling can improve both mouse and human fibroblast reprogramming [55, 56]). In contrast to mESC however, hESCs rely on TGF-β/Activin/Nodal signaling to maintain their pluripotent state. Hence, it is difficult to understand why inhibition of TGF-β/Activin/Nodal signaling should induce MET in human fibroblasts. Interestingly, in a recent paper investigators demonstrated that caALK2 inhibits the generation and maintenance of human iPSC when fibroblasts from patients with Fibrodysplasia ossificans progressive were used for reprogramming [57]. These results were in contrast to previous work suggesting a role for an active BMP signaling during the early-stage of MET in mouse iPSC generation [58], although this controversy can be explained by different molecular mechanisms in the mouse and human cells. Our data suggest that human fibroblast reprogramming could rather be induced by inhibiting the BMP/Smad1/5/8 pathway as the EMT transcription factors SLUG and MSX2 are strongly downregulated in TGF-β-treated hESCs. The data may provide improvements in human iPSC generation efficiency and quality in using nonintegrating approaches.


In summary, using a serum-free in vitro culture system for human embryonic stem cells, we demonstrated that along with bFGF, BMP4 promotes EMT via SLUG and MSX2, which further enhances mesodermal commitment of hESCs.


This work was supported by the Postgenomics-Biomedicine fund (grant 72004001) – Icelandic Centre for Research (RANNIS) and The Netherlands Institute of Regenerative Medicine. Adenoviral constructs were kindly provided by Drs. Kohei Miyazono and Peter ten Dijke. We thank Dr. Gordon Keller for trypsin-adapted hESCs and Dr. William Gallagher for the LLCIEP control overexpression vector (EV) and MSX2.

Author Contributions

A.R.: collection and assembly of data, data analysis and interpretation, manuscript writing; L.V., H.E.H., J.F.R., and A.R.O.: collection and assembly of data; D.W.-v.O.: provision of study material; C.M.: conception and design, data analysis and interpretation, manuscript writing; G.V.: conception and design, collection and/or assembly of data, data analysis and interpretation, manuscript writing, final approval of manuscript.

Disclosure of Potential Conflicts of Interest

The authors indicate no potential conflicts of interest.