Regulation of Neural Specification from Human Embryonic Stem Cells by BMP and FGF

Abstract

Inhibition of bone morphogenetic proteins (BMPs) signaling is required for vertebrate neural induction and fibroblast growth factors (FGFs) may affect neural induction through phosphorylation at the linker region of the Smad1 thus regulating BMP signaling. Here we show that human embryonic stem cells (hESCs) efficiently converted to neuroepithelial cells in the absence of BMP antagonists, or even when exposed to high concentrations of exogenous BMP4. Molecular and functional analyses revealed multiple levels of endogenous BMP signaling inhibition that may account for the efficient neural differentiation. Blocking FGF signaling inhibited neural induction, but did not alter the phosphorylation of the linker region of Smad1, suggesting that FGF enhances human neural specification independently of BMP signaling.

Introduction

One tenet of developmental neurobiology is that during the conversion of embryonic ectoderm to the neural ectoderm, inhibition of bone morphogenetic protein (BMP) signaling is required [1-3]. BMPs, members of the transforming growth factor (TGF)-β family of secreted morphogens, activate a set of serine/threonine kinase receptors. BMP receptors (BMPRs) in turn phosphorylate the BMPR-bound transcription factor(s) Smad1/5/8 (Smad1). Once phosphorylated, Smad1 is released from the BMPR complex and binds to the co-smad, Smad4. This complex is then translocated to the nucleus where it affects global gene expression [4].

The requirement for BMP inhibition during neural induction (NI) was first recognized in developing Xenopus embryos, where a set of secreted proteins, Noggin [5], Chordin [6] and Follistatin [7], were shown to have neuralizing activity, acting by binding BMPs and preventing them from engaging their cognate receptors [8-10]. Consistent with the idea that BMPs were inhibitory to NI, exogenous application of BMPs were shown to inhibit neural tissue formation [11, 12]. The requirement of BMP inhibition during mammalian NI was later demonstrated using mouse embryonic stem cells (mESCs). Like in Xenopus embryos, exposure of differentiating mESCs to BMP4 drastically reduced the percentage of neural progenitors formed [13-15].

Several lines of evidence from animal models and mESCs suggested that FGF signaling also played a role in NI. In developing Xenopus embryos FGF2 was shown to work in synergy with noggin to specify neural tissue [16]. The expression of a dominant negative FGF receptor inhibited neural tissue formation in Xenopus [17]. In epiblast explants from chick embryos pharmacological inhibition of FGF signaling blocked neural induction [18, 19]. NI was also blocked in mESCs using pharmacological reagents and the over expression of dominant negative FGFRs [14, 15]. These observations suggested that NI might be more complicated than simply inhibiting BMP signaling.

The opposing effects that these two signaling pathways exert on Xenopus NI were recently found to converge on Smad1. BMPR-phosphorylated Smad1, which inhibits NI, can be regulated by FGF signaling through MAPK-mediated phosphorylation of the linker domain of Smad1 [20-22]. In the present study, we addressed the questions of whether inhibition of BMP signaling is required for induction of the neuroectoderm from human ES cells and if FGF facilitates NI through Smad1 phosphorylation. Using a chemically defined system [23-25], we found that in the absence of any known neural inducing morphogens, hESCs were converted to a nearly uniform population of neural epithelial cells, which are characterized by their rosette morphology and their expression of Pax6. Neural specification of hESCs was surprisingly resistant to inhibition by BMP4, because of an intrinsic program of BMP signaling inhibition, which was active at multiple levels of the BMP signaling cascade. As in other vertebrates, FGF signaling was required for the efficient conversion of hESCs to NE, but this was independent of its role in inhibiting Smad1 through linker phosphorylation.

Material and Methods

Buffers

FACS buffer is PBS/2% donkey serum/0.01% NaN₃.

Cytoplasmic lysis buffer is 0.5% TritonX-100, 50 mM Tris pH 7.4, 150 mM NaCl, 10% glycerol, 10 mM Na pyrophosphate, 10 mM Na vanadate, 10 mM EDTA, and protease inhibitors (Sigma, MO). Nuclei lysis buffer is 0.5% SDS, 0.5% TritonX-100, 50 mM Tris pH 7.4, 150 mM NaCl, 10% glycerol, 10 mM Na pyrophosphate, 10 mM Na vanadate, 10 mM EDTA, and protease inhibitors.

Reagents

BMP4, Noggin and antibodies against Smad1 and Smad4 were obtained from R&D systems (Minneapolis, MN). Oct4 mAb and Abnoggin were purchased from Santa Cruz Biotechnology (Santa Cruz, CA) and Pax6 mAb were from Developmental Hybridoma Bank (Iowa City, IA). Antibodies against phospho-Smad1, Smad6, MAPK Erk1/2 and p-MAPK Erk1/2 from Cell Signaling Technology (Danvers, MA), actin from Sigma (Saint Louis, MO), histone 2B and α-tubulin from abCam antibodies (Cambridge, MA) were used. The p-Smad1^MAPK antibody was a generous gift from Dr. E. DeRobetis (UCLA).

hES cell differentiation

The maintenance and neural differentiation of hESC lines H9 (p16-35), H1 (p20-35), and H7 (p22-35) were preformed as previously described [24, 25]. Briefly, neural differentiation was initiated by dissociating hESCs with 1mg/ml Dispase (Invitrogen, CA) and allowing clusters of cells to grow as floating aggregates for 4 days in the hESC media (HESCM) consisting of DMEM/F12, 20% knockout replacement serum, 1 × non-essential amino acids, 2 mM glutamine,100 μM β-mercaptoethanol (all from Invitrogen, CA). ESC aggregates were then switched to serum-free minimal media (SFM media) consisting of DMEM/F12, N2 supplement, 1 × non-essential amino acids, 2mM glutamine and 2 μg/ml heparin (all from Invitrogen, CA). Cells remained floating in SFM media for 2 days before attaching to laminin (Invitrogen, CA) coated tissue culture plates. Cells were grown as adherent colonies, which differentiated into radial columnar cells over the following 4 days.

BMP4 and noggin treatments

At day 4 of differentiation, BMP4, Noggin, or both were added to the cultures for 4 days (until day 8 of differentiation). Cells were harvested at day 10 of differentiation for immunocytochemical and FACS analyses.

Immunocytochemistry

hES cells and neural progenitors were stained using previously described methods [24, 25].

Flow cytometry

Cells were harvested with trypsin and filtered through 70μM cell strainer. The resulting single cell suspension was fixed in 0.1% paraformaldehyde and permeablized in 90% methanol before being resuspended in FACS buffer. 10⁶ cells were incubated, overnight at 4°C, with primary Antibodies or normal mouse IgG as control, followed by washing and staining with secondary Antibodies for 1 hour.

Mircoarray sample preparation

RNA was isolated with Trizol™ (Invitrogen, Carlsbad, CA) and RNeasy Plus™ (Qiagen, CA) from the cells. For each time point of differentiation, RNA was collected from three independent differentiation cultures, and pooled. This was repeated 3 times generating three samples at each time point for a total of 9 samples. 2 μg of RNA from each sample was analyzed by Northern analysis to assess the quality of the RNA (data not shown). 5 μg of each pooled sample (9 samples total) was used for generating cRNA probes. cRNA probe synthesis and array hybridizations were carried out at the NIH Neuroscience Microarray Consortium (http://arrayconsortium.tgen.org/np2/home.do).

Mircoarray data analysis

Preliminary analysis was performed using Affymetrix Microarray Suite 5.0 (MAS 5.0) and Data Mining Tool softwares. The data were deposited at the NIH Neuroscience Microarray Consortium (http://arrayconsortium.tgen.org/np2/home.do). Further analysis was performed using GeneSpring GX ™ (Agilent Technologies, CA). Expression signal values were compared across three time points of differentiation. Day 0 (ESCs) was used as a base line and fold change was calculated for the day 6 and day 10 time points.

qRT-PCR

All quantitative reverse transcriptase PCR (qRT-PCR) was performed at the University of Wisconsin-Madison Genome Center (http://www.biotech.wisc.edu/GEC/) using the following primer sets (Supplemental Table. cDNA was generated using Superscript II Reverse Transcriptase (Invitrogen, CA) using manufactures protocol.

Gene name	GenBank #	Forward primer	Reverse primer
BMP2	NM_00120 0.2	CTTCTAGCGTTGCTGCTTCCC	GCATCTGTTCTCGGAAAACCTG
BMP4	NM_00120 2.3	TTCACTGCAACCGTTCAGAGG	CCGCATGTAGTCCGGAATGAC
BMPR1 A	NM_00432 9.2	ATTTGGGAGATGGCTCGTCG	TGTGAGTCTGGAGGCTGGATTG
BMPR2	NM_00120 4.5	GGCCAAGCATGTTTGATTCC	TAGTGGCCGTTCCCTCCTG
Chordin	NM_00374 1.2	CCATCCTGACTCTAGAAGGCCC	CCATTTCCTAGCAGCGTGAGG
Gremlin	NM_01337 2.5	CTGAAGCGAGACTGGTGCAAA	CACTCATGCACACGAACTACGC
Noggin	NM_00545 0.2	TGCGGAGGAAGTTACAGATGTG	TGCATTACAGGAACCAGAAAGC
Smad1	NM_00590 0.2	TTCTTGGGTGGAAACAGGGC	GGCATGTGAGGCTCATTTTGTC
Smad6	NM_00558 5.3	GGAGATCCTCCTCAACAACCC	TGGTATTAGCTTTTTGTCTGGGC
ZEB2 (SIP1)	NM_01479 5.2	CCTTTATGAATGGTGGGCTTG	TTGAGAGCATGGATCCTTCATG

Immunoblotting

Cells were lysed using RIPA buffer. For fractionation study, cells were first lysed in cytoplasmic lysis buffer. Nuclei were isolated by centrifugation, washed once with cytoplasmic lysis buffer and then lysed in nuclei lysis buffer. Lysates were resolved by SDS-PAGE and western blotting was carried out using horseradish peroxidase-conjugated IgG as a secondary antibody and ECL system for detection.

Immunoprecipitation

Cells were harvested and washed one time with ice cold PBS, and snap frozen in a dry ice ethanol bath, then lysed in the cytoplasmic lysis buffer. 1 mg of protein was used for each IP. Immunoprecipitations were performed by incubating cell lysates with appropriate antibodies as indicated for overnight at 4 °C, followed by incubation with protein G-sepharose for 1 hour. Beads were washed 3 times with lysis buffer. Precipitates were eluted from beads by SDS sample buffer.

Results

We have previously established the transcription factor Pax6 as an early marker of differentiating neural epithelia (NE) in a hESC culture [24, 25]. Immunocytochemical analysis in the present study confirmed that hESCs were positive for the pluripotent marker Oct4 and negative for Pax6 (Figure 1A). Conversely, 10 days of differentiation was sufficient to produce a nearly uniform population of NE cells, which were positive for Pax6 and negative for Oct4 (Figure 1A). Quantification of this transition using fluorescence activated cell sorting (FACS) indicated that the percentage of cells that were Oct4 positive remained high at day 4 of differentiation (85+/- 5%; Figure 1B, green histogram plot). However, by day 6, Oct4 expression was significantly diminished and was largely absent by day 7. In contrast, Pax6 was expressed by a small percentage of cells at day 6 of differentiation (17+/-2%; Figure 1B, red histogram plot). At day 7, about half of the cells were positive for Pax6 and by day 10, greater than 90% (93+/- 3%) of cells were positive for Pax6 (Figure 1B). This high efficiency of conversion of hESCs to hNE, on the same time course, from day 4 to day 10, was seen with all 3 lines that were tested (Data not shown). Thus, there is a critical period, from day 4 to day 10, for the transition of ESCs to hNE.

Figure 1.

hESCs convert to neural epithelia in the absence of exogenous morphogens. (A) Immunofluorescence demonstrates that hESCs are uniformly positive for Oct4 and negative for Pax6, while hESC-derived NE are uniformly positive for Pax6 and negative for Oct4 at day 10. (HO: Hoechst, Scale bar: 50μM) (B) Flow cytomerty reveals the temporal induction of Pax6 expression (red histograms) with the concomitant repression of Oct4 expression (green histograms) in differentiating hESCs. (The percentage of positive cells are indicated in the flow histogram.)

In various animal models, as well as during mESC differentiation, inhibition of BMP signaling is required for NI [26]. We tested whether this requirement held true for the formation of hESC-derived NE. Exogenous BMP4 (0, 10, 50, and 100 ng/ml) was added at day 4 of our differentiation protocol, when most cells (85%) were still positive for Oct4 and negative for Pax6 (Figure 1B). Immunofluorescence for Pax6 at day 10 of differentiation showed that nearly all untreated cells were Pax6 positive (Figure 2A_i-iv). In cells treated with a high concentration of BMP4 (100 ng/ml) flat non-columnar cells were observed at the periphery of NE colonies (Figure 2A, v). These flat cells were negative for Pax6 (Figure 2A, vi-viii), as well as Sox2, another marker of hESC derived neuroectoderm (data not shown).

Figure 2.

BMP4 modestly represses neural induction in differentiating hESCs. (A) Light microscopy (i, v, ix, xiii) reveals a small expansion of flat cells surrounding NE clusters in BMP4 treated cultures (v). Noggin alone had no effect (ix) on cell morphology, but was able to block BMP4 induced changes (xiii). Confocal images (ii-iv, vi-viii, x-xii, xiv-xvi) show that cells of altered morphology, following 100 ng/ml treatments with BMP4, were Pax6 negative (v-viii; below the region outlined with dashed line) as compared to the untreated group (ii-iv). In cells treated with BMP4, noggin blocked this effect (xvi). (B) Pooled data from FACS analysis showed that 100 ng/ml of BMP4 slightly, but significantly reduced the percentage of Pax6 positive cells (82% +/- 3%; n=3, p<0.05).

Previous studies have demonstrated that 10ng/ml of BMP4 was effective at inhibiting NI from mESCs [27]. Differentiating hESCs treated with 10 ng/ml of BMP4, showed no significant reduction in the percentage of cells that were Pax6 positive (Figure 2B, 91 +/- 3%, n=3). Cells treated with 50ng/ml BMP4 showed a small reduction in the percentage of cells that were Pax6 positive (87+/- 5%, n=3). A larger and significant reduction was seen in cells treated with 100 ng/ml BMP4, where only 82% (+/- 3%, n=3, p<0.05) of cells were Pax6 positive (Figure 2B). Thus, high concentrations of BMP4 were sufficient to cause a small population of differentiating cells to adopt a non-neural fate. However, the vast majority of cells still adopted a neural fate.

Various hESC neural differentiation protocols have suggested that exogenous BMP antagonism is necessary for efficient NI from hESCs [28, 29]. While we have demonstrated that hESC-derived NE differentiation does not require noggin treatment, we examined whether Noggin could reverse the small BMP4-mediated inhibition of NI. Recombinant human Noggin (100ng/ml), added to differentiating hESCs at day 4 to 10 did not significantly alter the percentage of Pax6 positive cells (Figure 2Aix-xii, and 2B 91% +/-3%, n=3). Simultaneous treatment with 100 ng/ml BMP4 and 100 ng/ml of noggin restored the morphology of NE clusters and percentage of Pax6+ cell population (Figure 2Axiii, Figure 2B, 89% +/- 4%, n=3), as compared to BMP treatment alone. These data demonstrate that during NI of hESCs, noggin was able to antagonize the inhibitory effect of exogenous BMP4.

The marginal effect of a high concentration of BMP4 on differentiating hESCs prompted us to ask if the differentiating hESCs express the necessary cellular machinery to respond to BMP4. Affymetrix Human Genome U133 plus 2.0 microarrays were used to analyze global gene expression levels of key components of the BMP4 signaling pathway at days 0, 6, 10 of hESC neural differentiation (Figure 3A-B). To find genes that were induced or repressed during NI, hESC expression profiles were used as a reference and then compared to the day-6 ESC-derived neural precursor sample expression profiles, and to the day-10 neuroepithelial sample expression profiles. We then confirmed the changes in the expression level of many of these genes using qPCR and/or Western blotting analyses (Supplemental Fig. 2).

Figure 3.

Summary of gene expression changes in the BMP signaling pathway during hESC neural induction. (A) A schematic model of the BMP signaling pathway [4]. (B) A summary of expression changes of key components in the BMP signaling pathway. Reported changes are ratios of expression values of samples from day 6 or day 10 compared to day 0 (ES cell stage). Yellow boxes indicate no change in expression from ESC stage to day 6 or day 10. Red boxes indicate an increase in expression level.

Analyses of the BMP signaling machinery revealed that the well-characterized BMP receptor pair, BMPR1a and BMPR2, was uniformly expressed at the three time points analyzed. This receptor pair is one of the most abundantly expressed receptors in hESCs [30], and hESCs are very responsive to BMP4 [31]. Three BMPs (BMP2, BMP4, and BMP7), known to negatively affect NI, were expressed throughout the conversion of hESCs to neuroectoderm (Figure 3B). Also, the key intracellular mediator of BMP signaling, Smad1, was uniformly expressed at the three time points analyzed. Finally, Smad4, the co-Smad that interacts with Smad1 [32], and is responsible for its translocation to the nucleus, was also uniformly expressed throughout differentiation (Figure 3B). Therefore, all the cellular machinery required for BMP signaling was expressed in differentiating hESCs. This raised the possibility that these differentiating cells undergo autocrine and/or paracrine BMP signaling.

One clue that addressed the paradox of the efficient NI of hESCs, despite the apparent autocrine/paracrine BMP signaling was found in the expression pattern of secreted BMP antagonists. The BMP antagonist follistatin, gremlin and noggin were all greatly induced (3-4 fold) at day 6 of differentiation as compared to ESCs (Figure 3B). For gremlin and noggin the increased expression level was even higher at day 10 (6 and 11 fold respectively). This observation could in part explain two puzzling observations: First, the ability of NI to proceed in differentiating hESCs, despite endogenous BMP signaling; and second, the ability of NI to proceed when the system was challenged with high levels of exogenous BMP4.

To address whether increased expression of BMP antagonist is responsible for the inhibition of BMP signaling in NI of hESCs, we tested the responsiveness of cells to exogenous BMP4 during neural differentiation, by analyzing the level of Smad1 phosphorylation. At day 4 of differentiation a 30-minute treatment with 100 ng/ml of BMP4 dramatically increased the level of Smad1 phosphorylation. However, in cells at day 6 of differentiation, Smad1 was resistant to phosphorylation by a 100 ng/ml BMP4 treatment (Figure 4A). We speculated that this inhibition of Smad1 phosphorylation was a result of BMP antagonists secreted by the differentiating hES cells. To test this, day-6 cells were either treated with BMP4 (100 ng/ml) in conditioned media, or in unconditioned media (washed). Only in cells treated with BMP4 in unconditioned media was Smad1 efficiently phosphorylated (Figure 4B), suggesting that BMP antagonists secreted by differentiating hESCs inhibit the BMP signaling.

Figure 4.

Phosphorylation and nuclear translocation of Smad1 is inhibited in hESC derived neural precursors. (A) Immunoblotting reveals that 100 ng/ml of BMP4 fails to induce the phosphorylation of Smad1 in day-6 hESC derived neural precursors. (B) Immunoblotting of lysates from day-6 hESC derived neural precursors washed and treated in unconditioned SFM, Smad1 is robustly phosphorylated by 100 ng/ml BMP4. (C) Immunoblotting of cytoplasmic and nuclear fractions show p-Smad1 is retained in the cytoplasm of day-6 ESC derived neural precursors, while Smad6 was found to be localized to the nucleus at day-4 and day-6 of differentiation. (D) Immuno-precipitation for p-Smad1 followed by immunoblotting for the co-Smad4 suggest these proteins still complex in day-6 hESC derived neural precursors.

(-) = no treatment; (+) = 100 ng/ml BMP4 for 30 min.

After phosphorylation of Smad1 by BMPR Smad1 can interact with Smad4 and translocate to the nucleus. Cell fractionation and immunoblotting analyses indicated that p-Smad1 was seen in the nuclear fraction in day-4 ESC aggregates after BMP4 treatment, but very little p-Smad1 was detected in the nucleus in day-6 cell lysates after BMP4 treatment in unconditioned media (Figure 4C). These results indicate that at day 4 the p-Smad1 can be translocated to the nucleus, which corresponds to the mild inhibitory effect of BMP4 on neural differentiation. But by day 6, even when the inhibitory effects of the secreted BMP antagonist can be overcome, p-Smad1 is trapped in the cytoplasm.

We next tested if the failure of nuclear translocation of p-Smad1 is due to lack of interaction with the co-Smad, Smad4. Day-6 ESC-derived neural precursors were washed, and then left untreated or treated with100 ng/ml BMP4. Like p-Smad1, Smad4 was almost exclusively in the cytoplasmic fraction of the lysates (Figure 4C). Co-immunoprecipitation analyses revealed that Smad4 was present in the BMP4 treated samples but not in the untreated samples (Figure 4D), demonstrating that p-Smad1 and Smad4 are able to form a complex in day-6 hESC derived neural precursors. This suggests that the cytosolic retention of p-Smad1 was not due to disruption of its interaction with Smad4.

Phospho-Smad1 can also interact with the inhibitory Smad, Smad6, which prevents Smad1's interaction with Smad4 [33], and facilitates its degradation via the proteasome [34]. We found that the expression level of Smad6 was increased two fold in day-6 ESCs, providing another potential mechanism for hESC-derived neural precursors to inhibit BMP signaling. To understand why increasing levels of Smad6 expression did not inhibit the interaction between p-Smad1 and Smad4 we immuno-blotted for Smad6 from day-6 ESC derived neural precursor samples that had been fractionated into cytoplasmic and nuclear fractions. Surprisingly Smad6 was found exclusively in the nuclear fraction (Figure 4C), which would explain why it did not interfere with the formation of the p-Smad1/Smad4 complex formation (Figure 4D). It has been reported that the nuclear Smad6 disrupt Smad1's activity by recruiting the transcription co-repressor CtBP and inhibitng BMP induced transcription [35, 36]. Our data suggest that increased level of Smad6 in the nucleus during NI of hESCs functions as one of the intrinsic inhibitory mechanisms of BMP signaling. Additionally, Smad Interacting Protein 1 (SIP1, also known as ZEB2), a repressor of Smad1/Smad4 gene activation [37, 38], was increased over 10 fold at day 6 and 20 fold at day 10 of differentiation as compared to levels in hESCs (Fig. 3B). Taken together, these results suggest that multiple levels of BMP signaling inhibition, the high level of BMP antagonists expressed by differentiating hESCs, the nuclear localization of Smad6, the increased expression of SIP1, and the cytoplasmic retention of p-Smad1, are responsible for the high efficient NI in hESCs, even in the presence of exogenous BMP4.

It has been reported that Smad1 translocation to the nucleus can also be inhibited by phosphorylation, by FGF-activated MAPK, in its linker domain. In our hESC culture model of NI treatment with the specific FGF receptor tyrosine kinase inhibitor SU5402 resulted in a dose-dependent inhibition of NI. Treatment with 10 μM of SU5402 reduced the percentage of Pax6 positive cells generated to 40% (+/- 8%; n=3, p < 0.05) (Figure 5A) suggesting that FGF signaling is required for hESC NI.

Figure 5.

Inhibition of FGF signaling inhibits NI induction, but fails to inhibit Smad1 linker phosphorylation in differentiating hESCs. (A) Pooled data from FACS analysis showed that SU5402 reduced the percentage of Pax6 positive cells during hESC neural differentiation (3 μM=65% Pax6 positive, +/- 5%, N=3, p<0.05; 10 μM=40% Pax6 positive, +/- 8%, N=3, p<0.05). (B) Immunoblotting demonstrated that MAPK ERK 1/2 activity is dependent on FGF signaling and the phosphorylation of the linker region of Smad1 is not dependent of FGF signaling in hESCs.

In HEK293 cells the MAPK dependent linker phosphorylation of Smad1, induced by FGF signaling, allows the ubiquitin ligase Smurf1 to bind Smad1, which causes the cytoplasmic retention of p-Smad1 [22]. To test if this is the mechanism of cytoplasmic retention of Smad1 employed by differentiating hESCs, we treated the day-6 differentiating hESCs with BMP4 in the presences or absence of the SU5402. Immuno-blotting demonstrated that MAPK ERK1/2 activation was dependent on FGF signaling. The expression level and subcellular localization of p-Smad1, Smad4, and Smad6 was not altered in SU5402 treated cells, despite the complete inhibition of MAPK ERK1/2 (Figure 5B). Immuno-blotting using antibodies specific for the linker phosphorylation [39] revealed that despite the complete inhibition of MAK ERK1/2 by SU5402, the linker region of Smad1 was still phosphorylated at levels comparable to the levels in the absence of SU5402 (Figure 5B). These results suggest that the inhibition of hESC NI caused by FGF signaling inhibition was independent of Smad1 linker phosphorylation.

Discussion

Analyses of neuroepithelial specification from hESCs presented in this study confirmed the necessity of BMP signaling inhibition and FGF signaling activation in NI. Our study revealed that differentiating hESCs in our chemically defined colony cultures employ multiple levels of intrinsic program of BMP signaling inhibition, ranging from the high level of BMP antagonists expressed by differentiating hESCs to the cytoplasmic retention of p-Smad1, the nuclear localization of Smad6, and the increased expression of SIP1. The presence of the endogenous BMP inhibition machineries explains the limited effect of BMP4 on hESC NI. Our finding is also in a general agreement with the phenomenon observed in other laboratories that the BMP antagonist Noggin improves hESC neural differentiation when hESCs are differentiated at a high density or via co-culture with stroma cells [40], in which inhibitory molecules including BMPs are likely produced in a larger quantity. BMP inhibition is playing a role not only in our present suspension culture, but also in adherent cultures. Treatment of adherent hESC cultures on matrigel or polylysine under a very similar culture medium to ours significantly promotes neural differentiation [29]. However, the effect of BMPs and their antagonists is dependent on the developmental stages of the target cells. When the undifferentiated hESCs were treated with BMP4 in the ESC growth medium with conditioned media from MEFs and FGF2, the hESCs differentiated to the trophoblasts [31]. In our differentiating hESCs (day 6), the effect of BMP4 is antagonized by the multiple levels of BMP inhibition, which hence did not result in trophoblast differentiation. Therefore, the same signaling pathway has different effects on the differentiation of hESCs depending on the developing stages of the cells. It also holds true to many other growth factors including FGF2 that is required for self renewal of hESCs [41] as well as neural differentiation of hESCs at a later stage [23, 42-44].

Our study also demonstrated the necessity of FGF signaling for efficient neural specification of hESCs. The involvement of FGF in neural induction has been reported in multiple vertebrate species including mouse ESC differentiation [45, 46]. One of the effects of FGF in neural induction is to inhibit BMP signaling by MAPK-mediated phosphorylation of Smad1 at the linker domain [22, 39]. Our results revealed that MAPK Erk1/2 activation is completely dependent on FGF signaling in differentiating hESCs. Yet, FGF signaling inhibition failed to alter the phosphorylation of the linker region of p-Smad1. This contrasts to the phenomenon observed in HEK293 cells in which FGF-activated MAPK leads to phosphorylation of Smad1 at the linker region, thus resulting in the cytoplasmic retention of p-Smad1 [22]. The most reasonable explanation for the discrepancy is that another MAPK, either JNK or p38, is active in FGF inhibitory condition used in this study and responsible for this phosphorylation event in differentiating hESCs. Despite the continued inhibition of p-Smad1 through linker phosphorylation, the efficiency of NI is greatly reduced with SU5402 treatments. This is consistent to the result suggested in recent study, in which FGF/Erk can induce neural specification of mouse ES cells independently of BMP activity [45]. We also observed that treatment of U0126, ERK1/2 inhibitor, decreased NI of hESC at a similar degree as FGFR inhibitor SU5402 (data not shown). Our results suggest that the positive effects of FGF signaling on human neuroepithelial specification may be distinct from BMP signaling inhibition.