Xenopus msx1 mediates epidermal induction and neural inhibition by BMP4

The development of the vertebrate embryo, which includes both the induction and patterning of germ layers, requires signaling among cells. In the amphibian embryo, where the molecular nature of cell-cell signaling has been the subject of intense scrutiny, various growth factors have been identified that participate in these embryonic decisions (Kessler and Melton, 1994; Slack, 1994). The patterning of the embryonic ectoderm has been suggested to be under the control of a secreted peptide growth factor called Bone Morphogenetic Protein 4 (BMP4), (Weinstein and Hemmati-Brivanlou, 1997). BMP4 is a member of a large family of signaling factors collectively known as Transforming Growth Factor-β (TGF-β) ligands. During early gastrulation in the frog Xenopus laevis, the cells of the ectoderm decide between two possible fates: the cells in the ventral side follow a molecular pathway that ultimately leads to the differentiation of epidermis while, in the dorsal side, the cells adopt a neural fate and ultimately give rise to the entire nervous system (Hemmati-Brivanlou and Melton, 1997). BMP4 has been shown to be a key regulator in this cell fate decision and can directly induce epidermal fate and inhibit the formation of neural tissue, thus imposing a ventral fate to the ectoderm (Wilson and Hemmati-Brivanlou, 1995). Elimination of BMP4 signaling by secreted antagonists derived from the organizer, such as noggin and chordin, unveils the neural fate on the dorsal side (Piccolo et al., 1996; Zimmerman et al., 1996). Thus epidermal fate is an induced fate and neural fate is interpreted as a ‘default’ or ‘ground’ state (Hemmati-Brivanlou and Melton, 1997).

Because of this type of key regulatory function during embryogenesis, and because of many other physiological functions, such as bone and cartilage induction during later stages of development, the molecular pathway downstream of the BMPs has been the subject of intense scrutiny. It is now clear that two type of receptors (type I and type II) bind dimers of BMP ligands (Yamashita et al., 1996). Binding of ligand to either type of receptor triggers the recruitment of the other type and, upon dimerization, the type II receptor phosphorylates serine and threonine residues in the intracellular domain of the type I receptor (Wrana et al., 1994). This initiates a signal transduction cascade that is mediated by a family of proteins collectively known as Smads (Massagué, 1996). In the case of BMPs, Smad1 is phosphorylated by the activated type I BMP receptors and the consequence of this activation is that Smad1 translocates into the nucleus (Hoodless et al., 1996; Liu et al., 1996; Kretzschmar et al., 1997). The function of Smad1 also requires the activity of another family member, Smad4 (Lagna et al., 1996). Once in the nucleus, Smads regulate the transcription of a subset of target genes and thus an extracellular signal is converted to a cellular response (Chen et al., 1996; Liu et al., 1996). In Xenopus embryos, Smad1 and Smad5, a close relative of Smad1, can mimic many BMP functions and activate BMP-inducible genes (Graff et al., 1996; Thomsen, 1996; Suzuki et al., 1997a).

While a large number of genes have been shown to be induced or repressed by BMP signaling in different cell types and tissues, little is known about immediate early response genes for the BMP signaling pathway. The identification of early response genes would allow the clarification of molecular pathways involved in vertebrate cell fate determination. Toward this aim, we made a survey of genes that are positively regulated in response to BMPs and ask if any would qualify as an immediate early response gene. Msx genes, vertebrate homologues of Drosophila msh (muscle segment homeobox) gene, are good candidates for BMP response genes (Davidson, 1995). Three different msx genes have been identified in the mouse (Shimeld et al., 1996), and at least two of them, msx1 and msx2, have been isolated from human (Jabs et al., 1993) and Xenopus (Su et al., 1991). Their expression pattern and sequence similarities strongly suggest that msx1 and msx2 would have similar functions during vertebrate development (Catron et al., 1996). Expression of msx1 and msx2 overlaps that of BMP genes during development (Davidson, 1995). BMP4 protein can induce or upregulate msx gene expression in dental mesenchyme, hindbrain, neural tube, limb bud, paraxial ectomesoderm and facial primordia (Vainio et al., 1993; Graham et al., 1994; Liem et al., 1995; Wang and Sassoon, 1995; Shimeld et al., 1996; Watanabe and Le Douarin, 1996; Barlow and Francis-West, 1997). In this manuscript, we provide evidence that the homeobox gene Xenopus msx1 (previously known as Xhox-7.1; Su et al., 1991) is an immediate early response to BMP4. The timing of expression and embryonic distribution parallels that described for BMP4. Moreover, we demonstrate that msx1 can mimic the activities of BMPs. Overexpression of msx1 in early Xenopus embryos leads to ventralization. In the context of dissociated ectodermal explants, msx1, like BMP4, can induce epidermis and inhibit neural differentiation. Finally, msx1 can also rescue neuralization imposed by a dominant negative BMP receptor (tBR) in intact ectodermal explants. These results suggest that Xenopus msx1 acts downstream of BMP4 and mediates BMP signaling in epidermal induction and inhibition of neural differentiation.

Xenopus msx1 cDNA, pUC18-Xhox-7.1 (Su et al., 1991), was digested with EcoRI and the 1.8 kb fragment encoding the entire coding region of msx1 was blunted by Klenow fragment. This fragment was subsequently subcloned into the EcoRV site of pSP64TXB (a gift from M. Tada). Xenopus BMP2, BMP4, and BMP7 genes were subcloned into pSP64TXB or pSP64T (Suzuki et al., 1997b). For a dominant negative BMP receptor (tBR), pSP64T-ΔXTFR11 (Suzuki et al., 1995) was used. The plasmid for the activated Xenopus BMP receptor, CA-XALK2, was described in Suzuki et al. (1997c). The Xenopus En-2 gene (Hemmati-Brivanlou et al., 1991) was subcloned into pSP65. For Xenopus Pax6, CS2-Pax6-FLAG was used (Altmann et al., 1997). Transcription from linearized plasmids was carried out using the mMessage mMachine in vitro SP6 transcription Kit (Ambion).

Xenopus embryos were obtained by in vitro fertilization and microinjection into embryos was carried out as described (Thomsen and Melton, 1993). Embryos were staged according to Nieuwkoop and Faber (1967). Isolation of undissociated animal caps was carried out in 0.5× MMR using hair knives and caps were cultured in 0.5× MMR containing 50 μg/ml gentamycin. Epidermal induction assays using dissociated Xenopus ectoderm were adapted from the method of Wilson and Hemmati-Brivanlou (1995). Briefly, animal caps were excised from late blastula embryos in 1× CMFM (Sargent et al., 1986) and transferred into 1× CMFB (Wilson and Hemmati-Brivanlou, 1995). The dark outer cell layer was manually removed from the dissected caps and the inner layer of cells was dissociated. Dissociated cells were transferred into an agar coated dish (1% in water) containing 1× CMFB containing 0.5 mg/ml BSA with or without the addition of 50 ng/ml of recombinant human BMP4. After 3.5 hours, cells were collected in Eppendorf tubes containing modified 1× MBSH (15 mM Tris-HCl, pH 7.6, 110 mM NaCl, 2 mM KCl, 1 mM MgSO₄, 1 mM MgCl₂, 1 mM CaCl₂, 2 mM NaHCO₃, 0.5 mM sodium phosphate, 50 μg/ml gentamycin) by a brief centrifugation (70 g) and allowed to reaggregate. Recombinant human BMP4 was a gift from Genetics Institute, Inc.

Xenopus albino embryos were fixed at the indicated stages by 1× MEMFA (0.1 M MOPS, 2 mM EGTA, 1 mM MgSO₄, 3.7% formalde-hyde, pH 7.4) for 2 hours and gradually dehydrated. Whole-mount in situ hybridization was performed according to the method of Harland (1991) except for using BM purple (Boehringer Mannheim) instead of NBT/BCIP for the chromogenic reaction. In order to eliminate cross-reaction of the labeled probe among homeobox genes, an msx1 cDNA from which the homeobox sequence had been deleted was used. Xenopus msx1, the pUC18-Xhox-7.1 (Su et al., 1991) was digested by EcoRI and HindIII, and an approximately 0.4 kb fragment encoding the N-terminal sequence without a homeodomain was subcloned into BluescriptKS(−) vector. A digoxigenin-labeled probe was prepared by digesting the plasmid with EcoRI and transcribing with T3 RNA polymerase. In order to follow lineage of the cells that received msx1 RNA, RNA encoding β-galactosidase fused to a nuclear translocation signal (Smith and Harland, 1991) was co-injected with the msx1 RNA. The injected embryos were fixed in 1× MEMFA for 20 minutes and washed twice with PBS. Whole-mount β-galactosidase staining of the embryos was carried out as described in Hemmati-Brivanlou and Melton (1994).

In order to inhibit protein synthesis, cycloheximide (CHX) treatment of ectoderm was carried out according to the method of Rosa (1989) and Cho et al. (1991) with minor modifications.

For cell dissociation, animal caps were excised from blastulae and left in 0.5× MMR containing 5 μg/ml CHX (Sigma) for 15 minutes. The caps were transferred into 1× CMFB plus 5 μg/ml CHX and dispersed to single cells for 15 minutes as described above. Subsequently, the ectoderm cells were kept dissociated in the medium with or without 50 ng/ml BMP4. At late gastrula stages (stage 11.5), the cells were harvested and subjected to RT-PCR assays. As a control for CHX treatment, ectodermal explants pretreated with 5 μg/ml CHX for 30 minutes were cultured in 0.5× MMR containing activin protein (purified PIF; Sokol et al., 1990). The explants were processed to RT-PCR using chordin primers at late gastrula stage (stage 11.5).

RNA was purified using acid guanidinium thiocyanate-phenol-chloroform method (Chomczynski and Sacchi, 1987). Reverse transcription-PCR (RT-PCR) was carried out as described previously (Wilson and Hemmati-Brivanlou, 1995). Sequence of primers used are described in Wilson and Hemmati-Brivanlou (1995) and Suzuki et al. (1995, 1997a). Other primer sequences are as follows:

msx1 (Su et al., 1991)

upstream: 5′-ACT GGT GTG AAG CCG TCC CT-3′

downstream: 5′-TTC TCT CGG GAC TCT CAG GC-3′

ODC (Bassez et al., 1990)

upstream: 5′-AAT GGA TTT CAG AGA CCA-3′

downstream: 5′-CCA AGG CTA AAG TTG CAG-3′

nrp1 (Richter et al., 1990)

upstream: 5′-GGG TTT CTT GGA ACA AGC-3′

downstream: 5′-ACT GTG CAG GAA CAC AAG-3′

Chordin (Sasai et al., 1994)

upstream: 5′-CAG TCA GAT GGA GCA GGA TC-3′

downstream: 5′-AGT CCC ATT GCC CGA GTT GC-3′.

Because of the previously established link between BMP4 and msx1 in other vertebrates (Davidson, 1995), we first wanted to determine if msx1 expression could be induced in Xenopus ectodermal explants (animal caps). To this aim, we injected synthetic RNA encoding either BMP2, BMP4 or BMP7 as well as the wild type and constitutively active ALK2 receptor, into the animal pole of both blastomeres at the 2-cell stage. Animal caps were explanted at blastula stage and tested by RT-PCR for the expression of msx1. Fig. 1A shows that all three BMPs as well as the constitutively active ALK2 (CA-ALK2) strongly induced msx1 expression, when assayed at gastrula stages. ALK2 has been shown to act as a receptor for BMP2 and BMP7 (Yamashita et al., 1996), and CA-ALK2 mimics BMP-like activity in Xenopus embryo (Suzuki et al., 1997c). Thus, our results indicate that msx1 can be induced by BMPs in Xenopus ectoderm. We also note that there is a low level of msx1 expression in the animal cap, which we have also confirmed by whole-mount in situ hybridization (see below).

We next examined if the induction of msx1 in animal caps represented an immediate early response to BMP4 protein. Animal cap explants from blastula embryos were dissociated and incubated with 50 ng/ml of BMP4 protein in the presence or absence of the translational inhibitor cyclohexamide (CHX). The explants were cultured until sibling controls reached gastrula stages. We preincubated the animal cap cells in a medium containing CHX for 30 minutes before adding BMP4 to ensure complete inhibition of protein synthesis (Rosa, 1989; Cho et al., 1991). Fig. 1B shows that BMP4 strongly induces the expression of msx1 in the presence of CHX (lane 8), demonstrating that its induction does not require de novo protein synthesis and thus that msx1 is an immediate early response to BMP4. In order to confirm the CHX treatment, the expression of chordin, whose induction by activin is inhibited by CHX treatment (Sasai et al., 1994), was examined. Under these conditions, induction of chordin expression by activin is greatly reduced by the CHX treatment (lane 13). We note that expression of msx1 is completely turned off in dissociated ectoderm (lane 5) suggesting that diffusible (non cell autonomous) factors are required for the expression that we see in undissociated ectoderm. Because BMP4 signaling occurs in the intact caps, we suggest that inhibition of msx1 expression in dissociated explants might be due to the elimination of BMP4 activity, or perhaps the activity of other BMPs.

Since msx1 represents an immediate early response to BMP4, we wanted to determine if the spatiotemporal expression of msx1 correlated with that of BMP4 or other BMPs. To this aim, we first performed RT-PCR on RNA extracted from different embryonic stages using msx1-specific primers. Fig. 2 shows that there is a very low level of msx1 expressed maternally (lanes 1-3). The zygotic expression started around late blastula to early gastrula stages (lanes 4 and 5). The level of expression is relatively constant from gastrula to tailbud stages (lanes 6-10). Maternal expression of msx1 has not been reported in previous studies which used RNAase protection assay (Su et al., 1991). This could be attributed to the difference in the detection methods.

We next used whole-mount in situ hybridization to assess the embryonic distribution of the msx1 message. Fig. 3A,B show that, at midgastrula stages (stage 11), msx1 transcripts are localized in the ventral side of the ectoderm and mesoderm and are excluded from the dorsal side. At neural plate stages (stage 14, Fig. 3C), msx1 expression is confined dorsally to the lateral edge of the neural plate (Fig. 3C) and ventrally to the heart primordium (Fig. 3D). Interestingly, msx1 expression is not detected in the anterior boundary of the neural plate and the expression on the lateral edge seems to be graded dorsoventrally. At later stages, after neural tube closure, in addition to the expression in the heart and the dorsal neural tube, a posterior ring of expression delineates the future tailbud at the posterior end of the embryo (Fig. 3E,F). Finally, at tailbud and tadpole stages, msx1 expression is confined to sharp lines of cells in the dorsal fin (Fig. 3G) that ultimately expand posteriorly and ventrally (Fig. 3H), and a subset of cells in the posterior end of the cement gland (Fig. 3G). We have also confirmed the expression of msx1 in the ventral gastrula ectoderm by RT-PCR. Dissection of gastrulae was carried out as schematically presented in Fig. 4A. Fig. 4B shows that msx1 is highly expressed in both ventral ectoderm (VE) and ventral mesoderm (VM), and weakly expressed in both dorsal ectoderm (DE) and dorsal mesoderm (DM). The accuracy of dissection was evaluated using the dorsal mesoderm marker chordin (Sasai et al., 1994), ventral mesoderm marker Xwnt-8 (Christian et al., 1991; Smith and Harland, 1991), pan-mesoderm marker Xbra (Smith et al., 1991). These results indicate that the timing of expression and the embryonic distribution of msx1 RNA are similar to the expression of BMP4 (Dale et al., 1992; Suzuki et al., 1993; Fainsod et al., 1994; Hemmati-Brivanlou and Thomsen, 1995; Schmidt et al., 1995).

To assess the embryonic activity of msx1, we injected synthetic sense RNA encoding msx1 into animal or marginal regions of Xenopus embryos and assessed the resulting phenotype. Fig. 5 shows that ectopic expression of msx1 ventralizes the Xenopus embryo. When 1.0 ng of msx1 RNA was injected into the animal pole of 2-cell embryos, the embryos showed partially ventralized phenotype with no eyes and anterior head structures (55%; n=33). At 2.0 ng of msx1 RNA, the embryos were severely ventralized and both head and tail were shortened (47%; n=43). This ventralization phenotype is similar to that obtained by overexpression of BMPs (Dale, 1992; Jones et al., 1992; Clement et al., 1995; Hemmati-Brivanlou and Thomsen, 1995; Moos et al., 1995; Suzuki, 1997b). In addition to the ventralized phenotype, we also observed gastrulation defects in msx1-injected embryos. This could be interpreted as a toxic effect of the RNA or, because overexpression of BMP4 also leads to inhibition of gastrulation movement (Jones et al., 1992), as part of the BMP-like phenotype. To see whether the ventralization caused by msx1 RNA is due to cell death caused by toxicity, we injected β-galactosidase (β-gal) RNA, as a lineage tracer, alone or with msx1 RNA into animal pole of 2-cell embryos. A comparison between the number of β-gal-positive cells in control embryos and in msx1 RNA-injected embryos revealed that similar numbers of cells are stained (data not shown), suggesting that no significant cell death was contributing to the msx1 ventralized phenotype. The ventralizing effect by msx1 is not consistent with previous studies reporting that overexpression of msx1 causes a duplication of primary axial structures (Chen and Solursh, 1995). In addition, we do not see significant differences in the phenotypes of embryos injected with msx1 RNA into animal pole or marginal regions of embryo.

We have shown so far that both the expression of msx1 and the embryological activity of msx1 in dorsoventral patterning seem to parallel that of BMP4. Since msx1 is an immediate early response to BMPs, it was important to test if msx1 can mimic BMP4 activity in the context of ectoderm, namely in the induction of epidermis, and establish if msx1 is part of the BMP signaling pathway in this context. Fig. 6A shows that injection of msx1 RNA inhibits neuralization and induces epidermis in dissociated cells. The induction of epidermal keratin by msx1 RNA is dose dependent (lanes 4-8). The induction of epidermal keratin by msx1 is specific: overexpression of other Xenopus homeobox genes, En-2 (Hemmati-Brivanlou et al., 1991) and Pax6 (Altmann et al., 1997), does not induce epidermal keratin expression in dissociated cells (Fig. 6B, lanes 7-10). The slight reduction of NCAM expression by Pax6 could be due to the induction of lens fate (lanes 9 and 10) (Altmann et al., 1997). It has been previously shown that dissociation of ectoderm cells leads to neuralization (Grunz and Tacke, 1989; Godsave and Slack, 1991) and addition of BMP4 protein to these cells inhibits this neuralization and imposes an epidermal fate (Wilson and Hemmati-Brivanlou, 1995). Thus, msx1 can mimic the effects of BMP4 in dissociated cells, consistent with msx1 being downstream of BMP4 signaling.

If msx1 is indeed part of the BMP pathway, it might be able to rescue the effects of a dominant negative BMP receptor (tBR) in the context of ectoderm. Fig. 7 shows that this is indeed the case. Because tBR can interfere with BMP signaling, overexpression of tBR in ectoderm induces neural tissue (Sasai et al., 1995; Suzuki et al., 1995; Xu et al., 1995) (lane 4). When tBR and msx1 are co-expressed in ectoderm (lane 5), however, the expression of neural markers NCAM, nrp1 and the cement gland marker XAG1 are suppressed and epidermal keratin expression is restored. This indicates that ectodermal cells receiving both tBR and msx1 RNAs adopt an epidermal fate and do not neuralize, and that msx1 is sufficient to rescue epidermal fate and blocks neuralization by tBR in ectoderm cells. Thus, msx1 acts downstream of BMP receptor.

The use of whole-mount in situ hybridization revealed the ventrally restricted expression of msx1 at midgastrula stages (Fig. 3). This localized expression was confirmed by the dissection of embryos followed by RT-PCR (Fig. 4). Our findings on the localization of msx1 in the ventral side of the embryo sharply contrast with results published by Su et al. (1991) who reported msx1 transcripts to be localized to the dorsal side of the Xenopus gastrula. We have no explanation for this discrepancy.

Previous studies on msx genes in vertebrate embryos mostly concentrated on their expression and function after gastrulation. For example, msx genes are implicated in the formation or patterning of dental mesenchyme, hindbrain, spinal cord and facial tissue (reviewed by Davidson, 1995). The expression of Xenopus msx1 in ventral ectoderm and mesoderm implies the role of vertebrate msx genes in these tissues. The fate map of the gastrula stage embryo indicates that these regions will develop as epidermis and ventral mesoderm (Keller, 1991). In fact, msx1 directs epidermal fates in dissociated ectoderm cells (Fig. 6) and ventralizes embryos (Fig. 5). Thus, Xenopus msx1 is expressed in the right time and place to act as an epidermal inducer/neural inhibitor.

Since expression of msx genes overlaps with that observed for BMPs during development, BMPs have been considered strong candidates for msx-inducing factors (Davidson, 1995). Recombinant BMP4 is capable of inducing msx genes in mouse dental mesenchyme (Vainio et al., 1993) or in chick hindbrain and spinal cord (Graham et al., 1994; Liem et al., 1995). These previous studies, however, did not address whether the induction of msx genes is an immediate response to BMP4 or not. We show here that msx1 is induced by BMP signaling in the early amphibian ectoderm, including BMP ligands and an activated BMP receptor, and that the induction of msx1 expression by BMP4 protein does not require do novo protein synthesis. Moreover, msx1 rescues neuralization by a dominant negative BMP receptor. These observations strongly suggest that msx1 acts downstream of a BMP receptor and is an immediate early response to BMP signaling in Xenopus embryos.

How specific is msx induction by growth factors? In other words, do other members of the TGF-β family, such as activin, Vg1 and nodal, induce msx1 expression? Two observations suggest that activin-like factors are not likely to induce msx1 expression in Xenopus embryos. First, overexpression of activin, Vg1 and nodal genes dorsalize Xenopus embryos (Thomsen et al., 1990; Thomsen and Melton, 1993; Jones et al., 1995), but msx1 does not (Fig. 4). Second, overexpression of a downstream signal transducer for activinA and TGF-β, Smad2 (Baker and Harland, 1996; Eppert et al., 1996; Graff et al., 1996), cannot upregulate expression of msx1 in animal caps (data not shown).

We have detected a basal level of msx1 expression in animal caps at early gastrula stages and found that this expression was downregulated after cell dissociation during gastrulation. This suggests that local cell-cell communication is important to maintain msx1 expression. This is consistent with the idea that local BMP4 signaling within the gastrula ectoderm is required for maintaining epidermal fate (Wilson and Hemmati-Brivanlou, 1995). Wang and Sassoon reported that disruption of cell-cell interactions in limb mesenchyme by cell dissociation results in a dramatic decrease in msx1 expression levels and that BMP4 can restore the expression in the mesenchyme cells in culture (Wang and Sassoon, 1995). Therefore, the maintenance of msx1 expression by local BMP4 signaling might be a common mechanism during vertebrate development.

In this study, we show that msx1 can induce epidermis in dissociated ectoderm cells. This finding is the first demonstration that overexpression of an immediate early response gene for BMP4 is sufficient to mimic the epidermal inducing activity of BMP4. In dissociated ectoderm cells, BMP4 has been thought to induce epidermal fate and inhibit neural fate simultaneously (Wilson and Hemmati-Brivanlou, 1995). Neural inhibition by msx1 suggests that msx1 could be directly involved in the repression of neural-specific genes. As mouse and human msx1 can inhibit muscle differentiation by inhibiting myoD expression in cultured myoblast (Song et al., 1992; Woloshin et al., 1995), the repression of transcription by msx1 may be a general feature of msx1 genes. It will be of interest to find the target genes of msx1 in the neural inhibition and epidermal induction pathways.

It is important to emphasize that, while msx1 can trigger induction of epidermis, it is not clear whether it can mediate all the activities of BMP4, such as mesoderm induction. In fact, mesoderm induction by BMP homodimers has been subject to controversy, since it can only be achieved by injection of high doses of RNA, and not by exposure of explants to physiological concentrations of BMP4 protein. This contrasts with the activity of other related family members such as activin, where pM amount of the ligand triggers a strong mesoderm induction response (Thomsen et al., 1990). In fact, no homeobox genes have yet been shown to be capable of triggering mesodermal differentiation.

Null mutation of the msx1 gene in the mouse has been reported: these mutant mice show abnormalities of craniofacial and tooth development, but epidermis formation is normal (Satokata and Maas, 1994). Although msx gene expression during mouse gastrulation has not been well analyzed, msx1 and msx2 expression overlap at many locations in later development (Davidson, 1995). Therefore, it is possible that, in msx1 knock-out mice, the function of msx1 is compensated by msx2. It will be important to compare the expression pattern of msx1 and msx2 during vertebrate gastrulation, and to test the functional redundancy of the two genes in Xenopus epidermal induction.

A. S. thanks the 1995 Cold Spring Harbor Xenopus development course. We also thank P. Wilson, D. Weinstein and D. Sassoon for critical reading of the manuscript, Genetics Institute Inc. for recombinant BMP4 and F. Ramirez for Xenopus msx1 plasmid. This work was supported by NIH grant (#HD 32105-01) to A. H.-B. and ‘Research for the Future’ program of the Japan Society for the promotion of Science to N. U.; A. S. is a research fellow of the Human Frontier Science Program; A. H.-B is a Searle and Mcknight scholar.