Early developmental specification of the thyroid gland depends on han-expressing surrounding tissue and on FGF signals

The endoderm gives rise to pharynx and intestine, and also to thyroid,lung, pancreas and liver. Development of these organs is initiated in different areas of the primitive gut, where, after induction of different transcriptional programmes, the organ primordia bud off. The signalling mechanisms that control the early development of these endoderm-derived organs have only recently begun to be investigated. Work in mice has shown that mesenchymal cells from the lateral plate mesoderm(Kumar et al., 2003);endothelial cells (Lammert et al.,2001); and the notochord (Kim et al., 1997) are involved in induction of the pancreas from the endoderm. For both liver and lung primordia, cardiac tissue adjacent to the endodermal layer has been identified as a source of inducing signals(Gualdi et al., 1996; Jung et al., 1999; Serls et al., 2005). Understanding the processes specifying different domains of the primitive gut can provide insight into congenital defects in humans and into regenerative responses to tissue damage.

Whereas increasing details of the inductive steps in the lung, liver and pancreas are emerging, our knowledge of such early developmental processes in thyroid development is scarce. As an anterior derivative of the primitive gut,the thyroid primordium buds from the ventral midline of the primitive pharynx(reviewed in De Felice and Di Lauro,2004). During a relocation process, the primordium loses contact with the pharynx, adopting a species-specific position in the hypopharyngeal mesenchyme. In many vertebrates, including mice and man, the primordium bifurcates in the neck area, leading to its final bilobed shape, whereas, in zebrafish and other teleosts, thyroid tissue forms an elongated strand of tissue along the ventral aorta (Wendl et al., 2002). Development of the thyroid is comparable between fish and mammals on the molecular level. The thyroid-specific transcriptional programme, including the transcription factors Nkx2.1 (also known as Nk2.1a and Titf1a/TITF1), Pax8 and Hhex, is conserved with respect to expression patterns and function between zebrafish and mouse(Elsalini et al., 2003; Wendl et al., 2002). In this study, we use zebrafish as a model to investigate the initiation of thyroid development.

In mouse development, induction of lung and liver by cardiac mesoderm was anticipated because of the close association of cardiac mesoderm with lung and liver primordia (Gualdi et al.,1996; Jung et al.,1999; Serls et al.,2005). Similarly, early mouse thyroid markers start to be expressed in the primitive pharynx adjacent to the aortic sac(Fagman et al., 2005)-the cardiac region that gives rise to the embryonic outflow tract and cervical arteries. This spatial correlation appears to be conserved in zebrafish, in which initial thyroidal nk2.1a expression starts, at 24 hours post-fertilisation (hpf), adjacent to the outflow tract of the heart(Rohr and Concha, 2000). Earlier in development, during the zebrafish somitogenesis stages, the anterior lateral plate mesoderm (aLPM), from which the heart later develops,as well as the endoderm, converge in parallel processes to the midline(Keegan et al., 2004; Warga and Nusslein-Volhard,1999). Parallel development and close association of both tissues is reflected in a functional relationship. The endoderm is necessary for normal cardiac morphogenesis and, in its absence, the converging halves of the aLPM fail to fuse (Alexander and Stainier,1999).

Fibroblast growth factors (FGFs) constitute a large family of signalling molecules that have been shown to act in multiple ways on endoderm-derived organ development. FGF1 and FGF2 are crucial for induction of lung and liver in mammals (Jung et al., 1999; Serls et al., 2005), and tissue explant assays suggest cardiac tissue to be the source of the signals. Moreover, in tissue-explant assays, these FGFs act in a concentration-dependent manner, with high concentrations required for lung,and lower concentrations for liver, induction(Serls et al., 2005). However,an exclusive role is not supported by the phenotype of FGF1/FGF2 double-knock-out mice, which are viable(Miller et al., 2000b).

In this study, we show that the zebrafish mutant hands off(han, hand2) has severe defects in early thyroid development. The han locus encodes the bHLH transcription factor Hand2(Yelon et al., 2000). Research on this mutant so far has concentrated on defects correlating with known sites of han expression, including the cardiac mesoderm, the fin buds and the pharyngeal arches (Angelo et al.,2000; Miller et al.,2003; Yelon et al.,2000). Two han alleles have been isolated, han^s6, which has a deletion spanning a maximum of 100 kb,including the han locus, and han^c99, which has an insertion in the han locus. han^s6 is a null mutation, and homozygotes exhibit a stronger phenotype than han^c99 mutants (Yelon et al., 2000). A role of han in thyroid development has not been described before and represents a novel aspect in thyroid research. In grafting experiments, we show that the han gene is required in the surrounding tissue for thyroid development. Further studies suggest that it is han-expressing anterior plate mesoderm or cardiac mesoderm that is crucial for thyroid specification. We further show that FGF signalling is required for thyroid development in zebrafish, and that FGF-coated beads are able to restore thyroid development in han^s6 mutants. Thus, our study provides a first step towards understanding the role of surrounding tissue during thyroid specification.

Zebrafish care, in situ hybridisation, immunohistochemistry and sections were carried out as described previously(Elsalini and Rohr, 2003; Rohr and Concha, 2000; Wendl et al., 2002). We used the han^s6, han^c99(Yelon et al., 2000) and ace^ti282a (fgf8^ti282a)(Brand et al., 1996) alleles. The identity of homozygotes was possible based on morphology. In addition, the identity of homozygous han mutant embryos was always confirmed by MF20 immunostaining visualising myocardial morphology.

As the lineage tracer for grafting experiments, we injected biotin-dextran(10,000 M_r, 5 mg/ml; Molecular Probes) into zebrafish embryos and detected biotin-labelled donor-derived cells after in situ hybridisation using the ABC kit (Vector Laboratories). In grafted embryos, the peroxidase reaction against biotin-dextran was carried out in DAB medium containing 67 mM NiCl₂, resulting in black donor-derived cells. This first reaction was followed by MF20 immunostaining using normal DAB medium, resulting in brown staining of the myocardium.

For fate mapping of thyroid precursor cells, photoactivation of caged fluorescein was essentially carried out as described previously(Keegan et al., 2004). Morpholino oligonucleotides targeted against endothelin 1 (edn1-MO)(Miller and Kimmel, 2001), lockjaw (tfap2a; 3.1-MO)(Knight et al., 2003) and foxi1 (foxi1-MO) (Mackereth et al., 2005), as well as an unspecific control morpholino (Gene Tools), were used as described previously. Implantation of beads was performed as described (Reifers et al.,2000) in low-melting-agarose-embedded embryos. Beads (45 μm Microspheres, Polysciences) were soaked overnight in 250 μg/ml human recombinant FGF1 (Sigma, St Louis, USA) or 100 μg/ml human recombinant FGF2(Roche, Indianapolis, USA), in both FGFs together, or in 250 μg/ml mouse recombinant FGF8b (R&D Systems, Minneapolis, USA) or 250 μg/ml BSA, all dissolved in PBS.

In zebrafish larvae, the thyroid gland can be visualised using an antibody detecting thyroid hormone (T4) at the apical membrane of follicles from approximately 60 hours post fertilisation (hpf). We found that the hands off mutant han^s6 lacks the differentiated thyroid gland completely at the larval stages (Fig. 1A,B). han^s6 has been identified based on defective heart, pharynx and fin development, and so we wondered to what extent endoderm is affected. At 24 hpf, when the thyroid starts to develop,the endoderm appeared to be normal, based on marker expression, indicating that endoderm specification was normal in han^s6 mutants(Fig. 1C,D). Thus, han^s6 is a good model in which to investigate the mechanisms of thyroid development.

In zebrafish, thyroid markers, such as nk2.1a, hhex and pax2.1, start being expressed in presumptive thyroid precursor cells in the endoderm at around 24 hpf (Elsalini et al., 2003; Rohr and Concha,2000). In 24-28 hpf han^s6 mutants, expression of pax2.1 and hhex was always absent(Fig. 1E-H). nk2.1aexpression was absent at this stage in most mutants, but, in about 10% of mutants, could be detected in a few endodermal cells(Fig. 1I-K), indicating that the thyroid phenotype is variable or not fully penetrant. Later, at 55 hpf,expression of nk2.1a, pax2.1 and hhex was not detectable in any domain that would indicate a thyroid in han^s6 (data not shown). In some clutches, occasional faint nk2.1a expression in few cells of the hypopharyngeal area indicated that some thyroid cells might be specified in han^s6 mutants (barely detectable; data not shown).

The differentiation marker slc5a5, encoding the sodium iodide symporter (NIS), is expressed exclusively in thyroid follicle cells from about 40 hpf (Alt et al., 2006). slc5a5 expression was usually not detectable in han^s6 mutants at 60 hpf(Fig. 1L,M), although we found a strongly reduced expression domain in 23 out of 258 homozygous specimen (9%; Fig. 1N). Taken together, the absence of the thyroid gland in han^s6 mutants can be explained by the lack of the early primordium in most specimens. In a small proportion of homozygous embryos, a reduced number of thyroid precursor cells were present and started the differentiation programme, but eventually failed to form a mature gland. Because it is hardly conceivable that slc5a5is expressed in the complete absence of thyroid-specific developmental genes,we assume that, in some mutants, remaining low levels of developmental genes are sufficient to initiate slc5a5 expression in some cells, but not sufficient for all aspects of terminal differentiation.

To find out whether increased cell death might be responsible for the absence of the thyroid primordium in han^s6 mutants, we carried out TUNEL assays. However, we did not observe visibly increased cell death in the pharyngeal endoderm or in the area where the thyroid would develop [tested at the 16-somite stage (ss), 20 hpf and 24 hpf; data not shown]. It should be noted that the thyroid primordium is very small, so it is possible to miss its few precursor cells undergoing cell death.

In han^s6 mutants, the deletion might affect the expression of a neighbouring locus, and so we tested the second available,hypomorphic han^c99 allele for thyroid defects. Here, from the beginning of detectable marker gene expression, the thyroid primordium was reduced in size, albeit not absent (Fig. 1O-Q). The reduced size of the primordium persisted during development and resulted in a smaller differentiated gland at the larval stages (Fig. 1Q). The fact that thyroid development in han^c99 mutants followed a similar,albeit less-severe, phenotypic trend to that of han^s6mutants confirms that it is the han locus in the han^s6 deletion that is involved in thyroid development. Furthermore, this observation suggests a dosage-sensitive requirement for Hand2 during thyroid development.

We reinvestigated han expression in the area in which thyroid markers start to be expressed. Here, han was expressed in the heart tube and the roots of the first pair of branching arteries, and, in addition,in the neural crest-derived mesenchyme of the pharyngeal arches(Fig. 2A,B). Furthermore,strong han expression was detectable in a set of bilateral cells at the border between the first and second arch, on the same anteroposterior(a-p) level as thyroid marker expression, but more lateral. These cells were directly adjacent to tyrosine-hydroxylase-positive cells called arch-associated neurons (AANs; Fig. 2A-D), which are presumably the precursor cells of the carotid bodies (Holzschuh et al.,2001). It is likely that these two bilateral and distinct groups of han-expressing cells form part of the carotid bodies.

In addition to previously described han expression in cardiovascular and pharyngeal structures, we found weak expression in the pharyngeal endoderm, including in tissue that expresses early thyroid markers(Fig. 2E,F). We were not able to detect han expression in the thyroid after evagination from the endoderm (data not shown). Taken together, multiple tissues surrounding the site where first thyroid marker expression is initiated, including mesoderm and endoderm, express han. han expression in the posterior lateral plate mesoderm and in the fin buds is far away from the foregut endoderm and can be ignored with respect to thyroid development. In most han^s6 mutant embryos, the thyroid primordium was missing from the beginning, raising the possibility that its induction or the competence of the endoderm to respond to inductive signals is impaired. We therefore addressed the question of where thyroid progenitors reside in the zebrafish embryo before the onset of earliest thyroid markers, and how does this position relate to han expression?

Earlier fate-mapping studies have shown that both endoderm and cardiac mesoderm converge medially to the embryonic axis during the somitogenesis stages (Keegan et al., 2004; Warga and Nusslein-Volhard,1999). Comparison of han expression with the endodermal marker foxa3 (fkd2) shows that, at the 8 ss, bilateral stripes of endodermal cells are distributed with aLPM cells in a partially overlapping fashion, with some endodermal cells being closer to the midline(Fig. 3A). We wanted to know where prospective thyroid precursor cells are located in relation to the han expression in the aLPM.

For this fate-mapping approach, we injected caged fluorescein into embryos at the one-cell stage and photoactivated the fluorescein dye at around the 8 ss. This stage was chosen because it is when the endoderm has not yet reached a position ventral to the neural tube during its convergence movements and is therefore accessible for photoactivation. As landmarks along the a-p axis, we used the posterior end of the eye, the midbrain-hindbrain boundary (MHB) and the anterior tip of the notochord to subdivide the aLPM into four zones(Fig. 3B). Nomarski optics allowed for visualisation of the aLPM edge. During photoactivation, we targeted cells along the medial border of the aLPM in zone 1 to zone 4(Fig. 3C-E) and probed their contribution to the thyroid at 55 hpf.

In total, 224 embryos were uncaged in 20 sets of experiments. In 100 embryos (45%), cells contributed to the pharynx epithelium, showing that the medial border of the aLPM is also the area where the anterior endodermal cells reside, as predicted from han/foxa3 double staining. Photoactivated cells contributed to the thyroid primordium in eight embryos(Fig. 3F). This low number reflects the small size of the thyroid primordium in comparison to the pharyngeal endoderm. In all of these eight embryos, photoactivated cells were derived from areas within zone 1 or zone 2(Fig. 3G). By contrast, cells derived from zone 3 or zone 4 never contributed to the thyroid. Taken together, endodermal thyroid precursors are, at the 8 ss, at the medial border of the aLPM, on the a-p level of the MHB. The thyroid primordium is also associated with the MHB at 24 hpf, when the first thyroid markers start to be expressed (Wendl et al.,2002).

To get a rough estimation of han expression in relation to the MHB during subsequent somitogenesis stages, we compared the expression of han with that of pax2.1, a MHB marker. hanexpression in the aLPM had its anterior border on the level at the MHB throughout the somitogenesis stages (Fig. 3H-M). Thus, han expression in the cardiac mesoderm is always ventral to the MHB. Even if thyroid precursors are only roughly associated with the MHB along the a-p level, it is likely that the han-expressing cardiac mesoderm is continuously close to thyroid precursors throughout the somitogenesis stages.

To find out whether han is cell-autonomously required in the endoderm for initial thyroid development or is required non-cell-autonomously in adjacent structures, we created genetic mosaics by the transplantation of wild-type donor cells into han^s6 mutant hosts. We first tested whether wild-type grafted cells are capable of expressing hanin the han^s6 environment. Embryos were fixed at the 12 ss,when han is broadly expressed in the anterior lateral plate mesoderm. In han^s6 mutant embryos, han expression was completely missing because of the deletion, but single wild-type cells ending up in the region of the anterior lateral plate mesoderm expressed han(Fig. 4A-C).

For analysis of thyroid development, embryos were fixed at 55 hpf and processed for nk2.1a in situ hybridisation. Homozygotes were identified by MF20 immunohistochemistry visualising heart muscle, which is strongly reduced in han^s6 mutants. Unequivocal identification of mutants was possible because grafted wild-type cells did not restore han^s6 heart morphology. Out of 87 homozygous han^s6 hosts (from 327 hosts in total) that received grafted wild-type cells, 75 did not show any sign of a thyroid at 55 hpf. However, 12 embryos (13.8%) showed a strong nk2.1a expression domain in the pharyngeal epithelium or in the pharyngeal mesenchyme(Table 1, Fig. 4D-G). The position of these nk2.1a domains resembled normal nk2.1a expression in the thyroid primordium. In the remaining homozygotes from the same clutches,which were fixed as controls, thyroidal nk2.1a expression was consistently not detectable at 55 hpf. As mentioned earlier, we occasionally observed faint nk2.1a expression in homozygotes of other clutches,but expression was weaker and restricted to a smaller domain. We therefore conclude that wild-type cells can restore nk2.1a expression in han^s6 mutants, or can increase weak levels of expression that are otherwise below detection, or can prolong initially present expression to unusually late time points.

Biotin detection revealed that grafted cells were always close to the restored nk2.1a expression domain in han^s6 mutant embryos. In none of the 12 embryos could wild-type cells be found in the domain of pharyngeal nk2.1a expression itself(Table 1). Thus, for restoration of nk2.1a expression, it is sufficient to bring wild-type cells into the surrounding tissue of the place where the thyroid would develop in han^s6 mutants. This means that han is required in cells other than thyroid cells for nk2.1a expression in the thyroid primordium.

han encodes a transcription factor, and so its cell non-autonomous action in thyroid development depends on its cell autonomous activity in surrounding tissue. In embryos with restored nk2.1a expression,grafted cells were rarely found to have contributed to the pharyngeal epithelium (Table 1). Thus,lack of han expression in the surrounding endoderm cannot be responsible for the loss of thyroidal nk2.1a expression in the mutant. Between the early steps of thyroid development and the time point of fixation at 55 hpf, morphogenesis of both heart and pharyngeal arches is highly abnormal in han^s6 mutants(Miller et al., 2003; Trinh et al., 2005; Yelon et al., 2000). Therefore, grafted cells cannot be unequivocally classified as belonging to either of these structures. To determine whether cardiac or branchial arch tissue is required for thyroid development, we eliminated hanexpression in the pharyngeal arches and putative carotid bodies by morpholino knock-down of upstream genes.

As was previously demonstrated, han expression in the branchial arches depended on the expression of endothelin 1 (edn1) and tfap2a (Knight et al.,2003; Miller and Kimmel,2001; Miller et al.,2000a; Piotrowski et al.,2003). Correspondingly, double morpholino knock-down of these two genes eliminated han expression in the pharyngeal arches completely(Fig. 5A-D). By contrast, han expression in the putative carotid bodies, heart and endoderm was unaffected in these double morphants. Thus, edn1/tfap2adouble morphants can serve as a model for thyroid development in the absence of pharyngeal arch han expression.

nk2.1a and slc5a5 expression is essentially normal in edn1 and tfap2a single morphants as well as in edn1/tfap2a double morphants(Fig. 5E, and data not shown),indicating that thyroid development does not depend on han expression in the pharyngeal arches.

To analyse whether han expression in the arch-associated cells(adjacent to the tyrosine hydroxylase-positive AANs) is required for thyroid development, we analysed foxi1 morphants. It has been shown that the foxi1 (no soul) gene is required for specification of the AANs (Guo et al., 1999). We found that, in foxi1 morphants, not only the AANs, but also the adjacent han-expressing cells located between the first and second pharyngeal arches are missing (Fig. 5F). Nevertheless, foxi1 morphants had a normal thyroid(Fig. 5G), indicating that han expression in arch-associated cells is not required for thyroid development. These data suggest that the remaining other site of detectable han expression, the cardiac mesoderm, is required for thyroid development. To date, it is not possible to ablate this tissue specifically,and so its role in thyroid development awaits further experimental confirmation.

We focused on FGFs as putative downstream factors of han in thyroid development, because they have been shown to act downstream of han in tooth development (Abe et al., 2002) and are known to play roles in cardiac development(Reifers et al., 2000). In zebrafish ace mutant embryos, the fgf8 gene is disrupted(Reifers et al., 1998). In this mutant, a reduced size of the thyroid primordium at early stages(Fig. 6A-F) and a reduced number of follicles after differentiation(Fig. 6G,H) indicates that fgf8 is required for normal thyroid development.

FGFs constitute a large family of signalling molecules(Ornitz and Itoh, 2001; Thisse and Thisse, 2005), and the loss of a specific FGF might be compensated for by the overlapping expression of other family members. The FGF-receptor blocker su5402 is an excellent tool to eliminate FGF signalling completely in specific time windows, and has also been used to narrow down the temporal requirement of FGF in zebrafish heart development (Reifers et al., 2000). We treated zebrafish embryos at various stages with 10μM su5402 for a time window of 2 hours and tested at 36 hpf for nk2.1a expression and at 60 hpf for slc5a5 expression. Higher concentrations of su5402 lead to severe malformations, making analysis of thyroid development questionable, so that we confined our analysis to 10μM concentrations only.

In general, su5402 treatment during the somitogenesis stages was sufficient to eliminate thyroidal nk2.1a and slc5a5 expression in around 70% of embryos (Fig. 6I-M). The uniform result for different time windows(Fig. 6M) suggests that washing su5402 out after 2 hours of treatment was probably inefficient, or that FGF signalling is continuously required for thyroid development. Treatment starting at 30 hpf, after the thyroid primordium is induced, affects the thyroid less efficiently, but still eliminates the gland in about 20% of embryos. This could be the result of a reduced influence of FGFs on later thyroid development, but could also be due to a limited potential of the chemical to diffuse into older embryos. su5402 treatment at the somitogenesis stages did not visibly affect endoderm development on the level of foxa2 (axial) expression at 30 hpf (data not shown),indicating that it is not a severe reduction of endoderm that causes the absence of the thyroid. In conclusion, su5402 treatment did not enable precise definition of the time window in which FGF signalling is required for thyroid development. Nevertheless, the drug treatments confirm that FGF signalling is required for thyroid specification and for subsequent differentiation. Furthermore, these data indicate that FGF signalling is not only required during the early steps of thyroid development, as indicated by the reduced early thyroid primordium in ace mutants, but also during later steps,after 30 hpf.

To test a possible role of FGFs downstream of han in thyroid development, we implanted beads soaked with recombinant FGF protein into han^s6 embryos and analysed subsequent thyroid differentiation based on the level of slc5a5 expression. We chose recombinant mouse FGF8 and, in addition, recombinant human FGF1 and FGF2. FGF1 and FGF2 signals from the cardiac mesoderm are suspected to be involved in liver induction in mice (Jung et al.,1999; Serls et al.,2005) and are therefore also good candidates as having a role downstream of han. Using the MHB as a landmark, beads were embedded into the embryo within proximity to the endoderm. Implantation was carried out at the 14-18 ss, because the reduced size of the early thyroid primordium in ace mutants suggests a specific role of FGFs in thyroid development before or around the onset of thyroid marker expression. Numbers of untreated han^s6 mutants with residual slc5a5 expression were not significantly different compared to han^s6 mutants that received BSA-soaked control beads(Fig. 7A,B). The implantation of beads soaked in FGF proteins, however, resulted in significantly increased numbers of han^s6 embryos expressing slc5a5 (BSA control compared to FGF8: X²=3.99, P=0.046; FGF1: X²=4.49, P=0.034; FGF2: X²=5.03, P=0.024; FGF1+FGF2: X²=8.00, P=0.004). Interestingly, all three FGFs were able to restore slc5a5 expression to a similar percentage. Thus,on the protein level, these different FGFs and probably also other members of the family can replace each other functionally. In summary, recombinant FGF protein is able to rescue the thyroid in han^s6 mutants by restoration of slc5a5 expression, showing that FGFs act downstream or in parallel to han in the differentiation of this gland(Fig. 7C).

Until now, research on the genetics of thyroid development has mainly concentrated on transcription factors expressed in thyroid precursor cells,such as Nkx2.1 (Nk2.1a), Pax proteins and Hhex. In both zebrafish and mice these factors are required for the differentiation of follicle cells(De Felice and Di Lauro, 2004). By contrast, the genetics of thyroid specification are poorly understood. Neither the factors involved in induction, nor in defining the competence of endodermal cells to become thyroid have been identified. Our grafting experiments show that han is required for thyroid development in a cell non-autonomous manner. Because han encodes a transcription factor, it is conceivable that the Hand2 protein is necessary for the development of certain tissues (most probably lateral plate mesoderm or heart)in the vicinity of the endoderm. Thyroid development, in turn, depends on the proper development of these surrounding structures. Such an indirect role could also account for the incomplete penetrance of the han^s6 phenotype at early stages of thyroid development,and for the han^c99 phenotype, in which both heart and thyroid are less severely affected. If the development of adjacent tissue is impaired, local sources of signalling molecules, such as FGFs, are not necessarily completely abolished. Interestingly, final thyroid differentiation leading to hormone (T4) production always failed in han^s6mutants, despite initial nk2.1a expression in some embryos, and despite occasional later expression of the differentiation marker slc5a5. It is possible that reduced nk2.1a levels are not sufficient for normal differentiation. Alternatively, the severely abnormal heart and pharyngeal arches in han^s6 mutants might influence later thyroid development independently of the early steps.

Our edn1/tfap2a and foxi1 morpholino experiments strongly suggest that han-expressing pharyngeal arch mesenchyme and arch-associated cells, respectively, are dispensable for thyroid specification. It remains unresolved which other han-dependent tissue exerts such a function. We discovered so-far unnoticed han expression in the endoderm. However, in han^s6 embryos, in which grafted wild-type cells restored nk2.1a expression, these cells only contributed in some cases to the pharyngeal epithelium, indicating that nk2.1a expression is not dependent on han-expressing neighbouring endoderm. Therefore, the most likely candidate of han-expressing tissue to act on thyroid development is the aLPM or cardiac tissue.

Because it is unknown exactly when han-expressing tissue is required for thyroid specification, at least two different scenarios are possible. Our fate mapping indicates that aLPM is continuously close to thyroid progenitors in the converging endoderm and, in han^s6 mutants, normal expansion of the aLPM fails during the somitogenesis stages. Thus, one possible model is that the aLPM signals to the endoderm, specifying thyroid precursors early during the somitogenesis stages. Alternatively, because cardiac development is severely disrupted in han^s6 mutants, it could also be that cardiac structures such as heart muscle or the outflow tract are required for thyroid specification. In this model, interactions between heart and pharyngeal endoderm would occur later than in the first model, just before the onset of thyroid markers, at around 24 hpf. Because the aLPM is considered to give rise to cardiac structures, both models are similar in that they support a central role of heart development in thyroid specification. Moreover, in both han and ace mutants, defects in cardiac morphogenesis(Reifers et al., 2000; Yelon et al., 2000) are correlated with a similar thyroid phenotype.

Interactions between cardiac and thyroid development are further supported by human syndromes. In DiGeorge (22q11) syndrome, caused by variable deletions on chromosome 22 in humans, congenital heart defects occur. In conjunction, an increased risk of congenital thyroid defects has been described(Bassett et al., 2005). Similarly, in human patients suffering from congenital hypothyroidism, an increased incidence of congenital heart defects has been observed(Olivieri et al., 2002). Furthermore, ectopic thyroid tissue can occasionally be found in human cardiac tissue (Casanova et al.,2000).

Our su5402 experiments indicate that FGF signals are required for thyroid development, and the ace mutant phenotype shows that Fgf8 is involved in this process. fgf8 was not expressed in the thyroid primordium at visible levels (data not shown), suggesting that Fgf8 acts in a non-cell-autonomous manner in thyroid development. This is supported by the observed effects of FGF-soaked beads in han^s6 mutants. fgf8 is expressed in the aLPM(Reifers et al., 2000), tissue that is continuously close to thyroid precursors and therefore a candidate for being the source of FGF signals. A search for FGFs acting together with fgf8 and probably redundantly in thyroid development has not been successful as yet (T.W., D.A. and K.B.R., unpublished observations). For instance, morpholino knock-down of other zebrafish FGFs (Fgf1, Fgf2, Fgf3) in conjunction with Fgf8 did not lead to a more-severe thyroid phenotype.

Genes encoding downstream factors or modifiers of the intracellular signalling cascade of Fgf8, such as spry2, spry4 or sef(also known as il17rd-Zebrafish Information Network)(Furthauer et al., 2002), were not found to be expressed at visible levels in the thyroid or in the endoderm at corresponding stages (T.W., D.A. and K.B.R., unpublished observations),suggesting that Fgf8 is unlikely to signal directly to the pharyngeal endoderm. It is therefore possible that further, unknown factors link FGF signalling and thyroid development.

FGFs have been implicated to play a role in thyroid development previously. In mouse embryos deficient for the FGF receptor 2 IIIb, multiple defects in organogenesis occur, including dysgenesis of the thyroid(Revest et al., 2001). A similar phenotype of FGF10 knock-out mice suggests that FGF10 is a major ligand acting via FGF receptor 2 IIIb(Ohuchi et al., 2000). However, initial thyroid development still occurs in the absence of the receptor 2 IIIb isoform (De Felice and Di Lauro, 2004), suggesting that FGF activity via this isoform is not responsible for early specification of the thyroid. Taken together, it can be anticipated that several FGFs act at different time points in thyroid development.

In han^s6 mutants, FGF proteins are able to restore thyroid differentiation, therefore acting downstream or in parallel of han in thyroid development (Fig. 7C). fgf8 expression in the aLPM appears to be normal in han^s6 mutants (J.J.S. and D.Y., unpublished data),indicating that here fgf8 expression does not depend on Hand2. Thus, fgf8 rather acts in parallel to Hand2 in thyroid development, and it is possible that morphogenetic changes in han^s6 mutants alter the temporal or spatial relation of FGF-expressing tissue to pharyngeal endoderm.

In the bead-implantation experiments, the thyroid was never restored at the wrong level along the a-p axis. This argues against inductive activity of FGFs, which would be likely to cause ectopic primordia. In particular, because slc5a5 expression is seen in a percentage of untreated mutants, we would expect a second thyroid in some FGF-bead-implanted embryos, which was not the case. Alternatively, we favour the possibility that FGFs act permissively in thyroid development, together with other signals. Our su5402 data, as well as the abovementioned mouse data, suggest that FGFs are also,and probably continuously, required for later thyroid differentiation, at which point structures in addition to the aLPM or cardiac tissues might act as a source.

Taken together, han and ace mutants represent two models that shed light on the role of the surrounding tissue in thyroid specification. Our study identifies the aLPM or cardiac structures to be key in this process. It will be interesting to analyse the role of Hand transcription factors as well as FGF signals and their downstream pathway components with respect to congenital thyroid defects in humans, in particular in those cases where they are associated with congenital heart defects.

We thank Didier Stainier and the members of his laboratory for support during the initiation of the project; our colleagues from the zebrafish community for plasmids and morpholino samples; and the members of the Brand,Rohr and Yelon labs for discussion and support. We thank Alexander Picker for advice on bead implantation, Tom Shilling for helpful information and Julia von Gartzen for excellent technical assistance. T.W., D.A. and K.B.R. were supported by DFG: SFB 572.