Retinoids signal directly to zebrafish endoderm to specify insulin-expressing β-cells

The vertebrate endoderm gives rise to the epithelial lining of the gut and to organs such as the thymus, liver and pancreas. This remarkable diversity of adult endodermal tissues derives from a homogeneous and multipotent precursor cell population. The later morphology of the gut and associated organs is produced by the precisely orchestrated divergence of subsets of endodermal progenitors towards different developmental fates in response to local inductive signals from adjacent mesoderm(Horb and Slack, 2001; Kim et al., 1997; Kumar et al., 2003; Rossi et al., 2001; Wells and Melton, 2000).

Several signaling molecules have been implicated in induction of specific endodermal cell types. Application of fibroblast growth factor 4 (FGF4)protein in the chick induces posterior endodermal markers in anterior(foregut) endoderm (Wells and Melton,2000). Activin-βB and FGF2, which function to suppress endodermal Sonic hedgehog (Shh) signaling, are permissive for pancreas formation, similar to the role of the embryonic notochord(Hebrok et al., 1998). Kumar and colleagues (Kumar et al.,2003) used chick explant cultures to show that pancreatic markers are induced in anterior endoderm by lateral plate mesoderm from the pre-pancreatic region. They also showed that activin, bone morphogenetic proteins (BMPs) or retinoic acid (RA) can all induce pancreatic markers in anterior endoderm co-cultured with anterior mesoderm(Kumar et al., 2003). Of these molecules, only RA has also been found necessary for pancreas development. Zebrafish or mice defective in RA synthesis or RA receptor (RAR) function lack pancreatic cell types (Martin et al.,2005; Molotkov et al.,2005; Stafford and Prince,2002). In addition, exogenous application of RA to zebrafish embryos induces ectopic pancreatic cells throughout the anterior endoderm. Work in amphibian and avian systems has yielded similar results, indicating that the requirement for RA in endocrine pancreas specification is probably a general feature of vertebrate development(Chen et al., 2004; Stafford et al., 2004). Together the gain- and loss-of-function studies suggest that RA plays an instructive role in specifying the position of the pancreas along the anteroposterior (AP) axis of the gut.

The mechanism of RA signal transduction is well known(Bastien and Rochette-Egly,2004). The rate-limiting step in RA synthesis is conversion of retinaldehyde to RA, catalyzed by retinaldehyde dehydrogenase (RALDH). RA signals are transduced by two types of nuclear hormone receptors: retinoic acid receptors (RARs) and retinoid receptors (RXRs). RARs and RXRs bind as heterodimers to retinoic acid response elements (RAREs) and either activate or repress transcription based on the presence or absence of ligand,respectively. The expression of RALDH enzymes (RALDH1-3) defines the tissues that produce RA; the RA essential for early vertebrate embryogenesis depends on RALDH2 function (Begemann et al.,2001; Grandel et al.,2002; Niederreither et al.,1999). In all vertebrates that have been examined, Raldh2is expressed during gastrulation in the mesendoderm, and becomes localized to lateral plate and paraxial mesoderm during segmentation(Begemann et al., 2001; Berggren et al., 1999; Niederreither et al., 1997; Swindell and Eichele, 1999; Wang et al., 1996; Zhao et al., 1996). Similarly,the expression of RARs and RXRs defines the tissues capable of responding to RA, and these exhibit broad, dynamic expression patterns during development(Begemann et al., 2001; Joore et al., 1994; Mollard et al., 2000; Sharpe, 1992). Consistent with these broad expression domains, loss-of-function studies have revealed that RA signaling is important for the development of derivatives of all germ layers(reviewed by Ross et al.,2000). The expression of RARs in a given tissue does not necessarily imply active RA signaling, as RARs can act as transcriptional repressors when not bound by RA (Koide et al., 2001).

Although RA clearly plays a crucial role in pancreas development, the precise nature of the inductive interaction is unknown. RA synthesized in the mesoderm could act directly on pancreas precursor cells in the endoderm. However, as RA also patterns other germ layers, its influence on the endoderm may be indirect. Distinguishing between these direct and indirect signaling models is crucial to understanding the molecular mechanism of pancreas specification.

Here, we take advantage of the ability to transplant cells into specific germ layers in zebrafish to pinpoint the source of RA and to demonstrate that it acts directly on endodermal cells to induce pancreatic fates. To establish the source, we generated chimeric embryos with a mixture of normal and RALDH2-deficient cells, an approach similar to previous studies of neural patterning (Begemann et al.,2001). The presence of wild-type donor cells in anterior mesoderm(somites 1-7) of a host embryo otherwise unable to synthesize RA rescues the formation of insulin-expressing cells, demonstrating that the crucial source of RA is the paraxial mesoderm. Conversely, disruption of RAR function in the endoderm with antisense morpholinos (MOs) and dominant-negative receptors blocks β-cell development, whereas loss of RAR function in other germ layers has no effect on embryonic insulin expression. In addition, a constitutively active RAR can drive ectopic expression of insulin in anterior endoderm. Our studies are the first to demonstrate that β-cell development requires RA produced in the paraxial mesoderm and that pancreatic precursors receive the RA signal directly.

A full-length rarα2a cDNA was isolated using PCR primers upstream of the START codon and overlapping the STOP codon (Accession Number NM_131406): RARα2a-START, TTGAATTCGCCAGCCGTTGTCTTGAGGAAT;RARα2a-STOP, CATCCTCGAGTGATGGTTGTCACGGTGACTG. A DN-RARα2a containing a premature stop codon to remove the C-terminal activation domain was modeled on Xenopus DN-RARs(Blumberg et al., 1997; Sharpe and Goldstone, 1997). To generate dominant-negative (DN) RARα2a, the above START primer was used with a primer that introduced a STOP codon at amino acid residue 403:DN-RARα2a-403-STOP, CATCCTCGAG CTATGAGCCTGGGATCTCATTT. PCR products were digested with EcoRI and XhoI and subcloned into pCS2+(Turner and Weintraub,1994).

Synthetic capped mRNAs encoding zebrafish SOX32, GFP, RARα2a,DN-RARα2a and Xenopus VP16-RARα1 were synthesized using the Ambion Megascript kit according to manufacturer's instructions. mRNA was suspended in 0.2 M KCl and ∼100 pl pressure-injected into one-cell stage embryos. Antisense morpholinos were designed by Gene Tools to target the raldh2 and rar genes: raldh2-MO, 5′GCAGTTCTTCACTGGAGGTCAT; RARα2b-MO, 5′ CCACAACGTCCACGCTCTCGTACAT;RARα2a-MO, 5′ GGTTCACATCCACACTCTCATACAT; RARγ-MO, 5′CCAGAGCCTCCATACAGTCGAACAT.

Approximately 100 pl of morpholino was injected at the yolk/blastoderm interface at the one- to two-cell stage, at concentrations ranging from 2 mg/ml to 4 mg/ml in injection buffer (0.25% Phenol Red, 120 mM KCl, 20 mM HEPES-NaOH pH7.5). Reagents were injected at the following final concentrations unless otherwise stated: SOX32 mRNA, 20 ng/μl; GFP mRNA, 15 ng/μl; RARα2a mRNA, 100 ng/μl; DN-RARα2a mRNA, 125 ng/μl;VP16-RARα1, 1 μg/μl; raldh2-MO, 4 mg/ml; RARα2b-MO, 2 mg/ml;RARα2a-MO, 2 mg/ml; RARγ-MO, 2 mg/ml; sox32-MO(Sakaguchi et al., 2001), 5 mg/ml; 40 kDa lysinated fluorescein dextran (Molecular Probes), 1 mg/ml.

Zebrafish embryos were injected at the one-cell stage with 25 pg of a RA-responsive reporter (tk-[βRARE]₂-luc, a gift from B. Blumberg) either with or without 500 pg DN-RARα2a RNA. Embryos were then treated with various concentrations of RA diluted in embryo medium and harvested at 24 hpf. Six embryos were used for each experimental condition,lysed in 10 μl/embryo lysis buffer (50 mM Tris pH 7.5, 1 mM DTT, 2 mM EDTA)and centrifuged at 16,000 g to remove cell debris. Lysate (10μl) was assayed using the Enhanced Luciferase Assay kit (Pharmingen);Monolight 2010 luminometer measurements were performed in duplicate.

In situ probes were used as previously described(Prince et al., 1998; Stafford and Prince, 2002). GFP protein was detected by immunohistochemistry using rabbit anti-GFP polyclonal antibody (Molecular Probes) at a dilution of 1:2000, processed with the Vectastain Universal ABC kit and visualized with TSA substrate (NEN)according to manufacturer's instructions. Confocal microscopy was performed on a Zeiss LSM510.

Wild-type embryos used in transplants were from the ^*AB strain or a locally obtained pet store variety. Transplantation was performed as previously described (Ho and Kane,1990). For transplants to test where RA signals are produced, the host embryos were injected with ∼100 pl of 4 mg/ml raldh2-MO, and donor embryos were labeled with 40 kDa fixable fluorescein dextran at the one-cell stage. Groups of 10-15 uncommitted cells from 4 hpf donors were transplanted along the blastoderm margin of equivalent stage host embryos. The embryos were then raised to 24 hpf and insulin expression detected to assay for the presence of endocrine β-cells. For transplants to test whether RARs function in mesoderm or ectoderm to specify pancreas, the hosts were injected with RKD reagents (DN-RARα2a at 125 ng/μl plus RARα2b-MO at 2 mg/ml) at the one-cell stage.

For transplants using SOX32 to target cells to the endoderm, donor embryos were injected at the one-cell stage with SOX32 mRNA. The embryos were co-injected with fixable fluorescein dextran, or with GFP mRNA, to allow visualization of donor cells, and with RKD reagents where appropriate. Approximately 25-30 cells from 4 hpf donors were distributed along the blastoderm margin of stage-matched hosts. Donor embryos die by 9 hpf as a consequence of being composed exclusively of endoderm.

Zebrafish embryos mutant in the raldh2/neckless gene(Begemann et al., 2001), or treated with a pan-RAR inhibitor (BMS493), lack all pancreatic cell types(Stafford and Prince, 2002). Timed treatments with BMS493 indicate that RA signals required to specify the pancreas occur at the end of gastrulation (9-12 hours post fertilization, hpf)(Stafford and Prince, 2002). As RA production during these stages depends on RALDH2 activity, we examined the distribution of raldh2 mRNA in endoderm and mesoderm of the region that forms the pancreas in greater detail. raldh2 is expressed in mesendodermal progenitor cells at 6 hpf(Begemann et al., 2001; Grandel et al., 2002). By 10 hpf, expression appears more restricted to anterior trunk mesoderm but may still include endodermal cells (Fig. 1A). To investigate this, we made use of SOX32/Casanova, a transcription factor expressed specifically in endoderm that has the capacity to confer endodermal fates on transplanted cells(Kikuchi et al., 2001; Sakaguchi et al., 2001). This allowed us to label the endoderm by transplanting fluorescein dextran-labeled cells from donor embryos ubiquitously expressing SOX32 into uninjected hosts,and to test if endodermal cells express raldh2. Using confocal analysis, we found co-localization of fluorescein and raldh2 mRNA in grafted endodermal cells in addition to raldh2 expression in adjacent host endoderm and mesoderm (Fig. 1B-D).

Embryos injected with an antisense morpholino oligonucleotide (MO) directed against the raldh2 transcript phenocopy neckless(nls) mutants (Begemann et al.,2001) and fail to form a pancreas(Fig. 1E,F). We used these embryos as hosts in cell-transplantation experiments to establish which tissue(s) require RALDH2 function in pancreas development. Based on the zebrafish embryonic fate map (Kimmel et al., 1990), we transplanted wild-type cells into the mesoderm or ectoderm of RALDH2-deficient host embryos (schematized in Fig. 1G), and assayed for insulin-expressing endocrine β-cells. In the 36 cases in which donor tissue was located in anterior paraxial mesoderm (defined as somites 1-7), 70% developed insulin-expressing cells(Fig. 1H). Transplants that contributed to more posterior mesoderm (n=11), or to the notochord,floorplate or neural tube, did not induce insulin expression(Fig. 1I-J). Similarly,transplants from SOX32-expressing donor embryos that contributed to endoderm only did not induce insulin expression (n=12; data not shown). We conclude that RA derived from anterior, pre-somitic mesoderm, the site of high levels of raldh2 expression, is sufficient for pancreas induction.

RA signals are transduced by RARs; three zebrafish genes encoding RARs have previously been described (rara2a, rara2b and rarg)(Begemann et al., 2001; Joore et al., 1994; Stafford and Prince, 2002). At 10 hpf, rara2b is expressed at higher levels than rara2a in the presumptive pancreatic region (Fig. 2A,B). Using the endoderm-labeling method described above and confocal analysis, we found that rara2b is expressed in both endoderm and mesoderm (Fig. 2D-F), while rara2a expression is excluded from endoderm (data not shown). These findings are consistent with our previous analysis of sections, which also showed that rarg is localized to endoderm(Stafford and Prince,2002).

To attempt to address the specific roles of individual RARs in pancreatic development, we designed morpholinos targeted against the 5′ coding regions. Morpholinos targeted against rara2a and rara2b did not cause a significant reduction in the number of insulin-expressing cells when injected alone or together (Fig. 2C,H; data not shown). Injection of morpholino targeted against rarg caused a minor reduction in number of insulin-expressing cells (Fig. 2H), and this effect was exacerbated by co-injection of rarg with rara2b morpholino(Fig. 2H), although even in co-injected specimens the pancreas retained more than 50% of its normal size. By contrast, injection of a dominant-negative (DN) RARα2a lacking the C-terminal activation domain (Blumberg et al., 1997; Sharpe and Goldstone, 1997) caused severe defects. Zebrafish embryos injected with synthetic mRNA encoding DN-RARα2a exhibited hindbrain and ear defects similar to those typically caused by reductions in RA signaling, as well as a dose-dependent reduction in size of the endocrine pancreas (as assayed by insulin expression; Fig. 2G). Furthermore, we found that co-injection of rara2b-morpholino with DN-RARα2a mRNA produced a significantly more severe pancreas phenotype than injection of the DN-RAR alone (Fig. 2H,I).

To confirm the specificity of the DN-RAR, we made use of several assays. We microinjected zebrafish embryos with a reporter construct in which luciferase expression is under RARE control and, as expected, found that this reporter was unable to respond to RA treatments in the presence of DN-RAR(Fig. 3A). To test further the specificity of the DN-RAR in vivo, we attempted to rescue pancreas development by co-injection of full-length RARα2a mRNA. However, this mRNA had no effect in the presence of the DN-RAR, and when injected alone led to a reduction in pancreas size (data not shown). Excess RAR protein has been shown to bind RA independently of RXR dimerization and DNA binding, which may reduce the amount of RA available to interact with RARs bound to RAREs(Blumberg et al., 1997). As an alternative test of specificity, we therefore examined the ability of our DN-RAR to block activation of Hox gene transcription by exogenous RA(Bel-Vialar et al., 2002; Conlon and Rossant, 1992). Treatment of 9 hpf embryos with 10^-6M RA for 1 hour induced ectopic hoxa4a expression in the anterior of the embryo(Fig. 3B,C), suggesting that hoxa4a is likely to be a direct RA target. Our DN-RAR blocked early expression of hoxa4a (Fig. 3D), and in addition we found that the ability of RA to induce ectopic hoxa4a expression in DN-RAR-injected embryos was severely abrogated (Fig. 3E), indicating that this reagent acts in a dominant-negative capacity. Similarly, we found that the capacity of RA to induce ectopic insulin-expressing cells was lost in DN-RAR-injected embryos (data not shown).

Embryos injected with both DN-RARα2a and rara2b-MO,hereafter referred to as RAR knock-down (RKD), were used in subsequent experiments as a source of RAR-deficient cells. The inability of rara2b-MO to block pancreas development in the absence of DN-RAR may indicate that absence of receptor leads to derepression of downstream targets,as has been observed in Xenopus(Koide et al., 2001). In addition, our DN-RAR reagent may block function of RARγ or of other RARs that have not yet been isolated from the zebrafish. In summary, although our results do not directly address which RAR subtype(s) are required for pancreatic development, the RKD reagents allow us to generate embryos that are unable to transduce RA signals while nevertheless expressing the RA synthetic enzyme raldh2.

Although RA is required for pancreas development between 9 and 12 hpf(Stafford and Prince, 2002),the onset of expression of the earliest known pancreatic marker, pdx1, is not until ∼14 hpf. Thus, there is sufficient time for RA to induce an intermediate signal rather than acting directly on the endoderm itself. Potential sources of this intermediate factor(s) are axial mesoderm or neuroectoderm, or paraxial mesoderm itself. To test the hypothesis that RA acts indirectly on the endoderm, we transplanted wild-type cells into RKD-injected hosts. Labeled donor cells were placed into presumptive paraxial mesoderm in somites 1-7, but in contrast to RALDH2-deficient embryos, these never rescued insulin expression(Table 1). Likewise,transplants of axial mesoderm or overlying neural ectoderm never caused a significant increase in the number of insulin-expressing cells. Thus,it is unlikely that a primary response to RA occurs in the mesoderm or ectoderm. As our transplantation techniques did not target lateral plate mesoderm, we did not test the possibility that this tissue relays RA signals. However, our results with endoderm (below) suggest that if such a relay exists, it acts in parallel with RA that signals to endoderm.

To target reagents exclusively to the zebrafish endoderm, we grafted SOX32-expressing donor cells into wild-type or SOX32-deficient hosts. SOX32 functions downstream of the Nodal receptor Taram-A, and is necessary and sufficient for endoderm formation (Fig. 4A) (reviewed by Ober et al.,2003). casanova (cas/sox32) mutants or embryos injected with a SOX32-targeted MO completely lack endoderm and display no expression of insulin, nkx2.3(Lee et al., 1996), or the posterior endodermal marker foxa3(Odenthal and Nüsslein-Volhard,1998) (data not shown and Fig. 4B,C). Cells transplanted from donor embryos injected with sox32 mRNA into SOX32-MO hosts restored development of large regions of the endoderm (Fig. 4D-H). Similar targeting of cells to the endoderm can be achieved by expressing Taram-A in donors (David and Rosa,2001); however, in our hands larval-stage endoderm morphology appeared more normal in transplants using SOX32 (data not shown).

To determine how well the transplantation method itself restores normal endodermal patterning, we examined insulin expression. Endodermal morphology at 24 hpf appeared identical to untransplanted controls in these mosaics, forming a strip along the ventral midline that widened towards the anterior (Fig. 4E,F). In mosaic embryos, however, there were on average only 17.5 insulin-expressing cells (n=37; Fig. 4E; Table 1), compared with 26.3(n=44; Table 1) in controls. These fell into two classes: 19% with fewer than seven insulin-expressing cells, whereas the other 81% averaged 23 insulin-expressing cells per embryo. The failure to form pancreas in some cases may reflect the dorsoventral position along the blastoderm margin where donor cells were grafted and their subsequent migration during gastrulation. Transplanted embryos expressed the posterior endoderm marker foxa3 (n=9; Fig. 4F) and the pharyngeal endoderm marker nkx2.3(n=5; Fig. 4G,H) in the appropriate AP locations. Thus, targeted transplantation using SOX32 restores pancreas development and does not disrupt regionalization of the endoderm.

To test the hypothesis that RAR function is required in endoderm for pancreas development, we targeted RKD-expressing cells to the endoderm of SOX32-deficient hosts. Donors were co-injected with SOX32 mRNA and RKD(schematized in Fig. 5A). These transplants also formed endoderm with approximately normal morphology that expressed markers such as nkx2.3 (data not shown). However, insulin expression in these mosaic embryos was either absent or severely reduced (Fig. 5B). Of 26 transplanted embryos, 24 had fewer than seven insulin-expressing cells (Table 1). This resembled the frequency of β-cell defects observed when RKD reagents were delivered to the entire embryo (94%; Table 1). Thus, disrupting RAR signal transduction in the endoderm alone blocks pancreas as effectively as doing so throughout the entire embryo,suggesting that RAR function in the endoderm is necessary for normal pancreas development.

To determine if RAR function in the endoderm is sufficient for pancreas formation, we transplanted endoderm cells into RKD hosts (schematized in Fig. 5C). Fifty-four percent had seven or more insulin-expressing cells (n=26; Fig. 5D; Table 1), indicating that the presence of endoderm capable of transducing RA signals is sufficient to allow pancreas specification. We conclude that RAR function need only be present in the endoderm for pancreas specification to occur, strengthening our hypothesis that the signal is direct.

Although we have previously shown that exogenous RA treatment is sufficient to induce anterior endoderm cells to take on endocrine pancreas fates(Stafford and Prince, 2002),such global RA application probably alters positional information in all three germ layers. Thus, this experiment does not discriminate between RA acting as an instructive signal, which confers positional information directly to endoderm, or as a permissive signal, which allows previously specified endoderm cells to complete their differentiation program. To distinguish between these possibilities, we activated RA signal transduction only in the endoderm germ layer by using a constitutively active RAR (XenopusVP16-RARα1) (Blumberg et al.,1997). As expected, this activated receptor expands expression of a RA reporter transgene (RARE-YFP) when microinjected into zebrafish embryos(data not shown) (Perz-Edwards et al.,2001). When we transplanted endoderm cells expressing the constitutively active receptor into otherwise normal embryos(Fig. 5E), we found insulin-expressing cells differentiated within anterior endoderm(Fig. 5F-H). All embryos formed a normal-sized pancreas in the wild-type position(Table 1). However, in 21 out of 43 cases, ectopic donor-derived insulin-expressing cells were found anterior to the first somite (cell locations summarized in Fig. 5G). These studies confirm that RA signal transduction in the endoderm is able to promote an endocrine pancreatic fate even when adjacent mesoderm is of anterior character.

Our study has shown that RA derived from the anterior paraxial mesoderm signals directly to adjacent endoderm to induce pancreatic progenitors. Importantly, our results suggest that RA can act as an instructive signal; RA may therefore play a role not only in specification of the pancreas, but also in its localization along the AP axis. The finding that RA induces β-cell fates directly may have important implications for islet stem cell biology and diabetes.

We have used cell transplantations to show that the RA signal required for zebrafish β-cell development is produced in the anterior paraxial mesoderm. This finding is consistent with previous zebrafish transplantation experiments, which revealed a similar reliance of neural ectoderm regionalization on paraxial mesoderm-derived RA signals(Begemann et al., 2001). It should be noted that although most of the zebrafish β-cells derive from the dorsal pancreatic bud, a few β-cells differentiate at later stages from the ventral bud (Field et al.,2003); the stages at which our experiments were analyzed do not allow us to make conclusions about the role of RA in development of this later differentiating population of β-cells. The requirement for mesodermal RALDH2 function in pancreas development is conserved between fish and mouse(Martin et al., 2005; Molotkov et al., 2005): mice mutant for Raldh2 do not form a dorsal pancreatic bud. The expression pattern of mouse Raldh2 suggests that the splanchnic component of the lateral plate mesoderm is the crucial source of RA for pancreas specification in this species, as this mesoderm is juxtaposed with the dorsal pre-pancreatic endoderm at the onset of expression of the pancreas progenitor marker Pdx1 (Martin et al.,2005; Molotkov et al.,2005). Similarly, explant studies in chick have suggested a crucial role for the lateral plate mesoderm in conferring pancreatic identity(Kumar et al., 2003). By contrast, our transplants reveal that RA produced in paraxial mesoderm is sufficient to induce the zebrafish pancreas. As our transplants did not target lateral plate mesoderm, RA from this source might also be capable of inducing pancreas in zebrafish. Alternatively, this apparent discrepancy between species may reflect differences in the relative locations of germ layer derivatives during gastrulation of different vertebrates.

During the course of this study, we developed a novel transplantation technique using activity of the SOX32 transcription factor to direct or exclude reagents from the endoderm germ layer. Our transplantations of RAR-deficient cells into the gut show that the RA signal required for zebrafish pancreas specification is received directly by the endodermal cells. These data rule out an indirect model for RA-mediated pancreas induction in which RA first confers positional information to an intermediate tissue, such as the paraxial mesoderm, which subsequently signals to the endoderm. Although RA had previously been shown to play a role in endoderm regionalization(Chen et al., 2004; Kumar et al., 2003; Martin et al., 2005; Molotkov et al., 2005; Stafford et al., 2004; Stafford and Prince, 2002),this is the first direct demonstration that mesoderm-derived RA signals directly to endoderm cells to induce pancreatic fates. Consistent with our finding, reporters driven by RA response elements (RAREs) are expressed in the mouse pancreatic endoderm (Martin et al.,2005; Molotkov et al.,2005). Martin and colleagues(Martin et al., 2005) have also reported low level RARE reporter activity in adjacent pancreatic mesenchyme, suggesting that RA signals are additionally received by the mesoderm. However, our transplantation results show that, at least in zebrafish, RA receptor function in mesoderm is not required for pancreas specification.

Signals required for pancreas development may function in either a permissive or an instructive fashion: for our purposes we define permissive signals as allowing cells to continue along a pre-specified differentiation program, whereas instructive signals confer positional information. We have shown that a dominant active RA receptor is capable of conferring β-cell fate on anterior endoderm, suggesting that RA acts as an instructive signal. However, it should be noted that in these experiments only a small proportion of the anterior endoderm cells expressing active RA receptor go on to express insulin, and further that the majority of these insulin-positive cells are located posterior to the level of the otic vesicle. These findings may imply that additional signals normally act at foregut level to further bias progenitor cells towards a β-cell fate, and that these signals are lacking from the most anterior part of the embryo. Alternatively, they may merely suggest that the Xenopusdominant-active receptor used does not have full function in the zebrafish. Explant studies in chick (Kumar et al.,2003) have shown that RA is not able to induce pancreatic markers in explants of anterior endoderm cultured in the absence of mesoderm,revealing that a second mesoderm-derived signal in addition to RA is necessary for pancreas specification. Our experiments suggest that mesoderm anterior to the first somite can supply this second signal, implying that the signal is RA independent and permissive in nature.

A major determinant of pancreas position may be the localized source of mesoderm-derived RA. This source is dependent on activity of both the synthetic RALDH2 enzyme and the RA degrading Cyp26 enzymes(Kudoh et al., 2002; Sakai et al., 2001). In future studies, it will be important to determine the precise locations of the pancreatic progenitors during gastrulation stages, in order to deduce whether localized activity of RA is sufficient, in turn, to localize the pancreas. Finally, as an instructive signal, RA is likely to prove a crucial component of protocols to induce stem cells to differentiate into pancreaticβ-cells. Consistent with this, a novel three-step approach to inducingβ-cells from stem cells, using both Activin A and RA, has recently been reported (Shi et al.,2005).

We are grateful to Louis Choi and Tailin Zhang for excellent fish care, and to Bruce Blumberg for tk-[βRARE]₂-luc and XenopusVP16-RARα1 constructs. We thank Ken Cho, Maike Sander, Jon Clarke and members of the Schilling laboratory for helpful comments on the manuscript. This work was supported by grants from JDRF (1-2003-257) and NIH (DK-064973)to V.E.P.; and by NIH (NS-41353, DE-13828), March of Dimes (1-FY01-198) and Pew Scholars Foundation (2615SC) to T.F.S.