Specification of posterior hypothalamic neurons requires coordinated activities of Fezf2, Otp, Sim1a and Foxb1.2

The hypothalamus is an evolutionarily ancient integrative center that plays a pivotal role in the survival and propagation of vertebrate species. It orchestrates complex adaptive behaviors that regulate numerous functions including stress responses, food intake, thermoregulation, fluid homeostasis and reproductive behavior. Reflecting the diversity of its function, the organization of the hypothalamus is highly complex with various cell types and fiber pathways tightly packed into a small space. Based on Nissl-stained tissues, the hypothalamus can be broadly divided into four major areas in the anteroposterior direction (preoptic, anterior, tuberal and mammillary) (Saper, 2004; Simerly, 2004; Swanson, 1987). More than a dozen nuclei with distinct functions and different cell types are interspersed within the different hypothalamic regions. Although a large number of studies have assigned distinct physiological functions to them, the mechanisms by which the different hypothalamic areas and neurons develop are poorly understood.

In particular, very little is known about the development of the neurons in the posteriormost part of the hypothalamus, which is referred to as the mammillary area (MA). One of the most prominent structures of the MA is the mammillary body (MB; also called the mammillary nucleus), which forms connections with the hippocampus, tegmentum and anterior thalamus and plays an important role in memory both in humans and in rodents (Vann, 2010; Vann and Aggleton, 2004). The MB is a large neuronal group comprising two nuclei, termed the lateral and medial mammillary nuclei (Swanson, 1987; Vann, 2010), each with distinct connections and different electrophysiological properties (Radyushkin et al., 2005; Vann, 2005; Vann, 2010; Vann and Aggleton, 2004). In addition to the MB, other nuclei found within the MA are the tuberomammillary nucleus, the supramammillary and the premammillary area, which have diverse functions such as defense behavior, anxiety, arousal, sleep, memory and feeding (Aguiar and Guimarães, 2011; Canteras et al., 2008; Haas and Panula, 2003; Pan and McNaughton, 2004).

Several transcription factors that control the specification of hypothalamic neurons have been identified. Genetic analyses in mouse revealed factors that control the development of neuroendocrine cells in the rostral hypothalamus including SIM1, SIM2, BRN2 (POU3F2), ARNT2 and OTP (Goshu et al., 2004; Michaud et al., 2000; Michaud et al., 1998; Nakai et al., 1995; Schonemann et al., 1995; Wang and Lufkin, 2000). The function of hypothalamic regulators seems to have been highly conserved during evolution and zebrafish orthologs have similar roles to their mouse counterparts (Machluf et al., 2011). For example, zebrafish Otp and Sim1/Arnt2 are required for the development and function of neuroendocrine cells as well as for the development of diencephalic dopaminergic (DA) neurons (Amir-Zilberstein et al., 2012; Blechman et al., 2007; Del Giacco et al., 2006; Eaton and Glasgow, 2006; Eaton and Glasgow, 2007; Eaton et al., 2008; Löhr et al., 2009; Ryu et al., 2007). Zebrafish otpb is regulated by the zinc-finger protein Fezf2 (Blechman et al., 2007), which was originally identified as a factor upregulated by overexpression of Dkk1 in zebrafish (Hashimoto et al., 2000). Fezf2 is expressed early on in the developing forebrain and controls regionalization of the diencephalon in both zebrafish and mouse (Hirata et al., 2006; Jeong et al., 2007; Jeong et al., 2006; Scholpp and Lumsden, 2010; Shimizu and Hibi, 2009). Later in development, Fezf2 is important for the development of certain forebrain neurons in mouse (Chen et al., 2005a; Chen et al., 2005b; Hirata et al., 2004; Komuta et al., 2007; Molyneaux et al., 2005) and for the development of DA, Oxytocin-like (Oxtl), GABAergic and serotonergic neurons in zebrafish (Blechman et al., 2007; Guo et al., 1999; Jeong et al., 2006; Levkowitz et al., 2003; Yang et al., 2012).

By contrast, very little is known about the mechanism that controls the development of distinct neurons in the posterior hypothalamus. Several studies have elucidated important roles for Shh, BMP, Wnt, FGF and Nodal signaling in proper patterning of the posterior hypothalamus and the development of the hypothalamic progenitors (Alvarez-Bolado et al., 2012; Kapsimali et al., 2004; Lee et al., 2006; Manning et al., 2006; Mathieu et al., 2002; Ohyama et al., 2008; Pearson et al., 2011; Tsai et al., 2011; Wang et al., 2012). However, downstream factors that contribute to posterior hypothalamic neuronal specification are not well understood. In mouse mutants for the winged helix-loop-helix protein FOXB1 the MB does not develop properly, suggesting a crucial role for this transcription factor (Alvarez-Bolado et al., 2000; Labosky et al., 1997; Wehr et al., 1997). More recent studies revealed additional transcription factors, including SIM1, PAX6 and PITX2, that are involved in late aspects of MB differentiation, such as the development of the mammillothalamic tract (Marion et al., 2005; Skidmore et al., 2012; Szabó et al., 2011). However, only these few regulators of MA neuron development are currently known. Interestingly, many of the known hypothalamic regulators are broadly expressed in the hypothalamus, suggesting that they play roles in the formation of multiple cell types. For example, although to date Otp function has been mainly analyzed in the neuroendocrine hypothalamus, it is expressed in other hypothalamic regions including the retrochiasmatic region, ventral tuberal region and the MA in mouse (Shimogori et al., 2010; Simeone et al., 1994) and in the posterior hypothalamus in zebrafish (Blechman et al., 2007; Ryu et al., 2007). Similarly, Fezf2 is found in both the anterior and posterior hypothalamus in zebrafish (Levkowitz et al., 2003), yet its role in the posterior hypothalamus is unknown.

Here we demonstrate crucial roles of Fezf2, Otp, Sim1a and Foxb1.2 in the development of distinct MA neuron types. First, we show that Fezf2 is required for early development of the posterior hypothalamus. At a stage when neuronal specification takes place, Fezf2, Otp, Foxb1.2 and Sim1a form distinct subdomains within the MA. We show that neuron types producing two hypothalamic hormones, Vasoactive intestinal peptide (Vip) and Urotensin 1 (Uts1), develop in different subdomains. Using loss-of-function and temporally controlled gain-of-function studies, we demonstrate that VIP neuron specification requires Otp and Sim1a, whereas Uts1-positive neurons require Fezf2, Sim1a and Foxb1.2. Thus, our study provides the first molecular map of the posterior hypothalamus in zebrafish and identifies, for the first time, transcription factor requirements for the specification of distinct MA neurons.

Zebrafish maintenance and breeding were carried out under standard conditions at 28.5°C (Westerfield, 2000). To avoid pigmentation, embryos were incubated in 0.2 mM 1-phenyl-2-thiourea. Embryos were fixed in 4% paraformaldehyde in phosphate-buffered saline (PBS) at the desired time points. The otpa^m866 mutant allele and genotyping protocol have been described (Ryu et al., 2007). For the transgenic lines overexpressing full-length cDNAs of otpa, sim1a, fezf2 or foxb1.2, the viral 2A peptide (Tang et al., 2009) was placed between the cDNA and EGFP or tdTomato and placed after the hsp70 heat-shock promoter (Halloran et al., 2000) in a multi-cistronic gene expression cassette. The full-length cDNAs of otpa, sim1a, fezf2 and foxb1.2 were PCR amplified and cloned into the pCRII vector (Invitrogen) using the primers listed in supplementary material Table S1. A single G0 founder was used to create a stable F1 family. The following stable transgenic lines were established: Tg(hsp70:fezf2-2A-EGFP)hd6, Tg(hsp70:otpa-2A-EGFP)hd7, Tg(hsp70:sim1a-2A-lyntdTomato)hd8 and Tg(hsp70:foxb1.2-2A-EGFP)hd9. Heat-shock treatments were performed at 34 hours post-fertilization (hpf) or 2.5 days post-fertilization (dpf) by adding 42°C egg water to the embryos and incubating them overnight at 37°C.

In situ hybridization probes for otpa and otpb (Ryu et al., 2007), sim1a (Löhr et al., 2009), foxb1.2 (Thisse et al., 2001), shh (Krauss et al., 1993), emx2 (Morita et al., 1995), nkx2.1a (Rohr and Concha, 2000), wnt8b (Kelly et al., 1995), fgf8 (Reifers et al., 1998), avpl (avp - Zebrafish Information Network) (Eaton et al., 2008), oxtl (oxt - Zebrafish Information Network) (Eaton and Glasgow, 2006), fgf3 (Maroon et al., 2002), arx (Miura et al., 1997), dlx2a (Akimenko et al., 1994), mcm5 (Ryu et al., 2005), pit1 (pou1f1 - Zebrafish Information Network) (Nica et al., 2004), prl, pomca (Herzog et al., 2003) and crabp1a (Liu et al., 2005) have been described. Probes for fezf2, lef1, lhx6, irx5a, gsx2, vip, uts1 and nr5a1a were PCR amplified using cDNA of 2-5 dpf embryos with the primers listed in supplementary material Table S2 and cloned into the pCRII vector. Whole-mount in situ hybridization and multicolor fluorescence in situ hybridization were performed as described (Hauptmann and Gerster, 1994; Lauter et al., 2011). Fluorescent immunohistochemistry was performed adapting an existent protocol (Pearson et al., 2011) using primary antibodies against Otpa (made by S.R.) and acetylated Tubulin (Sigma-Aldrich).

MOs (Gene Tools) are listed in supplementary material Table S3. The standard control, otpb, sim1a, fezf2, fgf8 and wnt8b MOs were described previously (Fürthauer et al., 2001; Jeong et al., 2006; Löhr et al., 2009; Riley et al., 2004; Ryu et al., 2007). The efficiency of the foxb1.2 MO was tested with the transient expression of a construct containing the foxb1.2 MO binding site in frame with GFP. In MO-injected embryos, no GFP expression was detected. Both DNA and MOs were diluted in water containing 0.05% Phenol Red and injected into one-cell stage embryos. PhotoMorph caging strand (Gene Tools) was designed to bind and block the fezf2 MO. Before injection, caging strand and MO were mixed in a molar ratio of 1.7:1 and hybridized at 65°C for 30 minutes. After hybridization the mixture was cooled, stored overnight at 4°C, and was injected at the one-cell stage. The caged MO solution and the injected embryos were kept in the dark until uncaging. Uncaging of the MO was performed at 24 or 48 hpf by transferring the embryos to glass dishes and exposing them to UV light at 312 nm using a UV table (Intas, Göttingen, Germany) for 30 minutes. The efficiency of caging and uncaging was confirmed by PCR analysis of cDNA derived from injected embryos using the primers listed in supplementary material Table S4. IWR-1 treatment was performed with 200 μM and 100 μM in DMSO for 12 hours prior to fixation and 50 μM in DMSO 24 hours prior to fixation at 2 dpf.

In order to identify potential regulators of posterior hypothalamic neuronal development, we began our study with Otp, as this transcription factor controls multiple aspects of hypothalamic neuronal development and is expressed broadly in the hypothalamus. To find factors that might mediate Otp function, we performed an in situ hybridization-based screen in zebrafish embryos lacking Otp activity. More than 80 transcription factors expressed in the hypothalamus were selected from the zebrafish expression database at the Zebrafish Information Network (www.zfin.org) and tested in 2-dpf otpa^m866 mutant embryos injected with 4 ng otpb morpholino (MO) (Ryu et al., 2007).

Two transcription factors showed changes in expression in the diencephalon of embryos lacking Otp activity. Expression of the zinc-finger transcription factor fezf2 was reduced in the preoptic area (PO) and expanded in the posterior part of the hypothalamus (Fig. 1A-D). The expression domain of the winged helix-loop-helix transcription factor foxb1.2, which encodes the zebrafish ortholog of FOXB1 (Odenthal and Nüsslein-Volhard, 1998), was also expanded in this region (Fig. 1E,F).

To determine whether foxb1.2 and fezf2 are expressed in the posteriormost part of the hypothalamus, which could represent the mammillary area (MA), and not in the neighboring tuberal hypothalamus, we compared their expression domains with those of markers for the tuberal hypothalamus, including pomca (Song et al., 2003), dlx2a (Bulfone et al., 1993a; Bulfone et al., 1993b) and nr5a1a (supplementary material Fig. S1). In particular, nuclear receptor 5a 1 (NR5A1; also called steroidogenic factor 1 or SF-1) (Lala et al., 1992) is a specific marker of the tuberal region in the early developing mouse and later a marker of the ventromedial hypothalamus (VMH) (Allen Institute for Brain Science, 2012, http://developingmouse.brain-map.org) (Ikeda et al., 1995; Ikeda et al., 1994). Whereas pomca, dlx2a and nr5a1a are expressed adjacent to the adenohypophysis, further supporting their tuberal location in zebrafish, foxb1.2 and fezf2 were expressed in a distinct location that was more dorsal and posterior. Since FOXB1 is expressed in the MA in mouse (Labosky et al., 1997; Wehr et al., 1997) and both foxb1.2 and fezf2 are expressed in the posteriormost part of the zebrafish hypothalamus and not in the tuberal hypothalamus, we refer to this region as the putative MA.

Thus, our data show expression of fezf2 and foxb1.2 in the putative MA in zebrafish larvae and suggest that Otp negatively regulates fezf2 and foxb1.2 expression in this region.

To gain insight into the roles of Otp, Fezf2 and Foxb1.2 in posterior hypothalamus development, we first examined their early roles by performing co-expression analysis with hypothalamic markers and loss-of-function experiments at 1 dpf. We used nkx2.1a as a marker of the entire hypothalamus (Rohr et al., 2001) and shh and emx2 for the anterior-dorsal and the posterior-ventral hypothalamus, respectively (Mathieu et al., 2002). At this stage, the differentiation between tuberal and mammillary hypothalamus might be difficult because the location of adenohypophysis/neurohypophysis is anterior to the hypothalamus and tuberal markers are not yet well expressed within the hypothalamus (supplementary material Figs S2, S3) (Herzog et al., 2003; Liu et al., 2005; Nica et al., 2004). We observed that both foxb1.2 and fezf2 were expressed within the posterior part of the nkx2.1a-positive hypothalamic domain where fezf2 labels more ventral and foxb1.2 more dorsal domains (Fig. 2A-A″′). The expression domains of shh and fezf2 did not overlap, whereas those of shh and foxb1.2 partially overlapped (Fig. 2B-B″′). emx2 expression was entirely contained within the hypothalamic fezf2 domain (Fig. 2C-C″′). By contrast, the hypothalamic otpb expression overlapped with nkx2.1a, fezf2 and foxb1.2, but showed little overlap with emx2 (Fig. 2D-E″′).

Because of its broad expression in the posterior hypothalamus, we next tested whether fezf2 affects emx2 expression. Embryos that were injected with fezf2 MO (Jeong et al., 2006) showed a strong reduction in emx2 expression in the posterior-ventral hypothalamus and loss of otpa and otpb (Blechman et al., 2007) expression in the posterior hypothalamus (Fig. 2F,G; supplementary material Fig. S4A-D). Hypothalamic foxb1.2 expression was not reduced (Fig. 2H,I). In addition, the previously reported reduction of prethalamus was apparent in fezf2 morphants (Jeong et al., 2007) (Fig. 2F-I; supplementary material Fig. S4E-H).

Fezf2 may regulate emx2 directly or affect signaling pathways that control posterior hypothalamus regionalization. In particular, Fezf2 overexpression downregulates wnt8b in the prospective zona limitans intrathalamica (ZLI) and in the mid-hindbrain boundary (MHB) and fgf8 expression in the MHB at the mid-somitogenesis stage (Jeong et al., 2007). We asked whether downregulation of emx2 in fezf2 morphants might be mediated by wnt8b or fgf8. wnt8b morphants showed reduction in hypothalamic emx2 expression, whereas in fgf8 morphants emx2 was unchanged (supplementary material Fig. S5A-C,K). However, neither nkx2.1a nor shh expression was altered in these morphants (supplementary material Fig. S5D-I). Interestingly, in 1-dpf fezf2 morphants, we observed clear expansion of wnt8b and fgf8 expression in the posterior hypothalamus (Fig. 2J-M; supplementary material Fig. S6). Therefore, although Fezf2 regulates wnt8b expression in the posterior hypothalamus, its effect on emx2 is likely to be independent of wnt8b. In fact, wnt8b expression at 1 dpf mostly does not overlap with that of emx2 (supplementary material Fig. S7). In contrast to fezf2 morphants, embryos that lack Otp or Foxb1.2 activity do not show apparent changes in shh, emx2, wnt8b or fgf8 expression (supplementary material Fig. S8, Fig. S9A-G; data not shown). Hypothalamic fgf3 expression was also unchanged in embryos that lack Otp, Fezf2 or Foxb1.2 activity (supplementary material Fig. S9H-L).

Our results thus show that, at 1 dpf, Fezf2 is required for maintaining emx2 expression and limiting the extent of hypothalamic wnt8b and fgf8 expression in the posterior hypothalamus.

In order to investigate neuronal development in the posterior hypothalamus, it is necessary to first define its anatomical features. Since the anatomy of the posterior hypothalamus has not been characterized in detail in larval zebrafish, we first identified putative MA nuclei, taking advantage of the recently reported mouse hypothalamus atlas (Shimogori et al., 2010). Markers that define different MA nuclei in rodents are Lef1 (premammillary nucleus), Lhx6 (tuberomammillary terminal), Irx5 (supramammillary nucleus) and Foxb1 (MB or mammillary nucleus). Strikingly, their zebrafish homologs lef1, lhx6, irx5a and foxb1.2 are also expressed in distinct domains in the posterior hypothalamus (Fig. 3).

To gain insight into the roles of Otp, Fezf2 and Foxb1.2 in the neuronal development of the MA in zebrafish, we first determined in which MA domains they are found at 2 dpf, when many neuron types in the posterior hypothalamus are specified. This we achieved first by comparing foxb1.2 and otpb expression relative to that of lef1, lhx6 and irx5a (Fig. 3A-M″). A schematic map of their relative domains was drawn taking into consideration the location of landmarks in combination with DIC images (Fig. 3G,N; supplementary material Figs S10, S11). Next, we compared the relative expression of foxb1.2, otpb and fezf2 (Fig. 4A-C″). Similar analysis was performed for sim1a and wnt8b given their reported roles in posterior hypothalamus neuronal development (Lee et al., 2006; Marion et al., 2005) (Fig. 4D-J″). fezf2 expression overlapped in the posterior extent of the foxb1.2 domain but mostly formed a separate domain (Fig. 4A-A″). The otpb and otpa domains in the anterior part of the posterior hypothalamus were lateral to the fezf2 and foxb1.2 domains (Fig. 4B-C″; supplementary material Fig. S12). The sim1a domain overlapped with the fezf2 and foxb1.2 domains, but extended more laterally, overlapping with otpb (Fig. 4D-F″). wnt8b was expressed along the midline and in a band posterior to the foxb1.2 and sim1a domains, partially overlapping with fezf2 and otpb expression (Fig. 4G-J″). Thus, strong foxb1.2 expression demarcates a distinct domain within the posterior hypothalamus with contributions by sim1a and fezf2 expression.

This spatial relationship of the transcription factors was further confirmed in the lateral view (supplementary material Fig. S13). Schematic maps of the expression domains in the dorsal and lateral view are shown (Fig. 4K,L). To corroborate the identity of the foxb1.2-expressing domain as the MB, we sought to visualize potential mammillothalamic tracts. Axon bundles in this region were visible (supplementary material Fig. S14), but the precise target area definition requires more refined tracing techniques. Nevertheless, given the conservation of foxb1.2 expression we will refer to this region as the putative MB.

In summary, expression of fezf2, otpa, otpb and sim1a was found in domains including and surrounding the putative MB, with fezf2 and wnt8b marking the posterior, otpa/otpb in the lateral and posterior, and sim1a labeling anterior domains. Thus, the expression analysis of otp, fezf2, sim1a, wnt8b and foxb1.2 revealed the existence of unexpectedly complex and distinct subdomains in the region surrounding the putative MB.

What is the mechanism responsible for generating these distinct transcription factor domains surrounding the putative MB? The largely non-overlapping expression of otp and foxb1.2, or of otp and fezf2, could result from the negative regulation of foxb1.2 and fezf2 by otp (Fig. 1). To explore other regulatory interactions that could contribute to this subregionalization, we analyzed the phenotypes of fezf2, foxb1.2 and sim1a (Löhr et al., 2009) morphants. At 2 dpf, fezf2 morphants showed a reduction in otpa and otpb (Blechman et al., 2007) expression in the region surrounding the putative MB, but a clear expansion of sim1a and foxb1.2 (Fig. 5A-F). In foxb1.2 morphants, sim1a and fezf2 expression was expanded (Fig. 5G-J). By contrast, in sim1a morphants we observed no detectable changes in the expression of otpa, otpb, foxb1.2 and fezf2 (data not shown). Similarly, embryos with reduced Otp activity showed no changes in sim1a expression (data not shown). To test the effect of Wnt8b we treated embryos at 1 dpf with IWR-1 in order to disrupt Wnt signaling without affecting early patterning (Chen et al., 2009; Moro et al., 2012). IWR-1-treated embryos showed no changes in the expression of fezf2, foxb1.2, sim1a and otp (data not shown).

Thus, changing the levels of Otp, Fezf2 or Foxb1.2 leads to changes in the expression of the other transcription factors, whereas changes in Sim1a levels and interference with Wnt signaling do not. Otp, Fezf2 and Foxb1.2 might directly regulate each other’s expression domains or levels. Alternatively, it is possible that the changes that we observe are due to their effects on other transcription factors or signaling pathways. In case of fezf2, this is particularly pertinent because we have shown that fezf2 affects early development of the posterior hypothalamus at 1 dpf. In order to distinguish between early and late effects of fezf2, we designed a PhotoMorph (PM) (Tomasini et al., 2009) specific to fezf2 MO sequences, which led to near complete caging of the fezf2 MO and partial uncaging upon UV exposure at 1 dpf (supplementary material Fig. S15). The expansion of foxb1.2 and sim1a was still apparent in the embryos injected with fezf2 PM after uncaging, whereas the otpa reduction was not (supplementary material Fig. S16). This suggests the existence of stage-specific roles of fezf2 and an Otp-independent interaction between fezf2 and foxb1.2 or fezf2 and sim1a.

Thus, taken together our results show that fezf2, foxb1.2 and otp play an important role in the establishment of the MB domains. Our data are also consistent with the possibility that fezf2 might directly regulate foxb1.2 and sim1a in this process.

To test whether the distinct domains surrounding the putative MB give rise to different neuron types, we next cloned and examined the expression patterns of 30 genes encoding different hypothalamic neuropeptides (our unpublished data). We found that Uts1-positive neurons were generated in the foxb1.2-expressing putative MB (Fig. 6A-A″), whereas Vip-positive neurons were generated in the adjacent foxb1.2-free region (Fig. 6E-E″). Uts1-positive neurons were generated in a domain that is also positive for sim1a and fezf2, whereas the expression domains of uts1 and otpb were mostly exclusive (Fig. 6B-D″). By contrast, vip expression overlapped with those of otpb and sim1a, but there was no significant overlap with fezf2 expression (Fig. 6F-H″). Although fezf2 and foxb1.2 were expressed in largely non-overlapping regions, Uts1-positive cells developed in a small area where they overlapped. These results show that different transcription factor domains in and around the putative MB give rise to distinct neuronal types, where VIP cells are otp⁺, sim1a⁺, fezf2^- and foxb1.2^- and Uts1 cells are otp^-, sim1a⁺, fezf2⁺ and foxb1.2⁺.

To test whether transcription factors co-expressed with Vip- or Uts1-positive neurons are indeed necessary for their development, we performed loss-of-function analyses. Reduction of Otp or Sim1a led to fewer Vip-positive cells in the MA, consistent with the fact that Vip-positive neurons express both otpb and sim1a (Fig. 7A-C,M; Fig. 6F-F″,H-H″). Interestingly, although fezf2 is not co-expressed with vip at this stage, fezf2 morphants exhibited fewer Vip-positive cells (Fig. 7A,D,M; Fig. 6G-G″). However, this effect is likely to be due to an early function of fezf2 because in embryos that were injected with fezf2 PM and uncaged at 1 dpf no such reduction is seen (Fig. 7A,E,M). The number of Vip-positive cells did not change in foxb1.2 morphants, consistent with the fact that foxb1.2 is not expressed in these cells (Fig. 7A,F,M; Fig. 6E-E″).

The loss of Sim1a or Fezf2 led to a reduction in Uts1-positive neurons, which normally express both sim1a and fezf2 (Fig. 7G,I,J,N; Fig. 6B-C″). However, the loss of Otp or Foxb1.2 activity resulted in an increase in Uts1-positive cells (Fig. 7G,H,L,N), which is likely to be due to the increase in fezf2 in these embryos. In contrast to Vip-positive cells, a decrease in Uts1-positive cells was still detected in embryos that were injected with fezf2 PM and uncaged at 2 dpf, suggesting an important role of Fezf2 for the specification of these neurons (Fig. 7G,K,N). Injection of 8 ng standard control MO led to no apparent changes in the expression of the investigated transcription factors or in the expression of vip and uts1 (supplementary material Table S5).

To test whether the transcription factors expressed in Vip- or Uts1-positive cells are sufficient to promote their fates, we performed temporally controlled gain-of-function analyses. We generated transgenic zebrafish lines that overexpress otpa, sim1a, foxb1.2 or fezf2 cDNA. The heat-shock experiments were performed at 34 hpf for Vip or at 2.5 dpf for Uts1. Heat-shock treatments induced clear overexpression of the specific transcription factors (supplementary material Fig. S17). Since vip is co-expressed with otp and sim1a (Fig. 6F-F″,H-H″) but not with fezf2 and foxb1.2 (Fig. 6E-E″,G-G″), we hypothesized that overexpression of Otpa, Sim1a and/or a combination of both might lead to supernumerary Vip-positive cells. Indeed, we found more Vip-positive cells in embryos overexpressing Otpa, Sim1a or a combination of both compared with wild-type embryos that were heat shocked (Fig. 7O; supplementary material Fig. S18A-D). Notably, the overexpression of Fezf2 or Foxb1.2 did not lead to an increase in the number of Vip-positive cells, consistent with the fact that Vip-positive cells are fezf2- and foxb1.2-negative at this stage (Fig. 7O; supplementary material Fig. S18A,E,F).

Since uts1 was co-expressed with fezf2, sim1a and foxb1.2 (Fig. 6A-C″), we tested the effect of overexpression of these factors on the number of Uts1-positive cells. Although overexpression of Sim1a alone had no effect, overexpression of Fezf2, Sim1a plus Fezf2, or Foxb1.2 led to an increase in Uts1-positive cells (Fig. 7P; supplementary material Fig. S18G-K). The increase in cell number was small, but nonetheless significant and consistent across multiple experiments.

Thus, taken together, our loss-of-function and overexpression analyses demonstrate that Vip-positive neuron specification requires Otp and Sim1a, whereas Uts1-positive neurons require Fezf2, Sim1a and Foxb1.2.

Despite its functional importance, the mechanism underlying the development of the posteriormost part of the hypothalamus, the MA, is poorly understood. In this study we provided the first molecular map of the MA in zebrafish and showed that in zebrafish embryos the MA consists of domains with strong conservation of marker genes expressed in the mammalian MA. Our analysis also revealed the existence of distinct subdomains that overlap and surround one of the MA nuclei, the MB (or mammillary nucleus). Although a few transcription factors have been identified that are required for MA development, almost nothing was known about how the MA neurons are specified. In this study, we identified molecular requirements for the specification of neurons that produce two important hypothalamic hormones, Vip and Uts1. We show that these two neuron types are specified through coordinated activities among the transcription factors Otp, Fezf2, Sim1a and Foxb1.2, which define distinct subdomains of the MB. Our results also highlight the important dual roles of Fezf2 for both early development of the posterior hypothalamus and late specification of Uts1 neurons.

Fezf2 is a key transcription factor controlling multiple aspects of forebrain development. One of the highly conserved functions of Fezf2 is the regulation of forebrain regionalization by maintaining anterior forebrain fate (Jeong et al., 2007; Scholpp and Lumsden, 2010; Staudt and Houart, 2007; Toro and Varga, 2007). By contrast, although Fezf2 is expressed early in the hypothalamus, its function in the posterior hypothalamus was not known. We showed that fezf2 expression is limited to the posterior half of the nkx2.1a domain at 1 dpf, perfectly abutting but not overlapping with the anterior shh domain, suggesting a role for Fezf2 in the posterior hypothalamus. Indeed, our results show that, in fezf2 morphants, the posterior hypothalamic regional marker emx2 was severely reduced.

Since Wnt signaling influences regionalization within the hypothalamic subdomains, promoting posterior fate at the expense of anterior fate, we tested whether the effect of Fezf2 on emx2 could be mediated by Wnt signaling. It has been shown that the expression of exogenous wnt8b in the presumptive hypothalamus results in the expansion of emx2 (Kapsimali et al., 2004). Consistent with this overexpression phenotype, we observed a strong reduction of emx2 in wnt8b morphants. However, our results show that although fezf2 regulates wnt8b and emx2 expression, its effect on emx2 cannot be via wnt8b because we observed the expansion and not the reduction of wnt8b in fezf2 morphants. Previous studies have shown that wnt8b acts upstream of fezf2 in the neural plate (Hashimoto et al., 2000) as well as in controlling DA cell number (Russek-Blum et al., 2008). Our results reveal that Fezf2 can in turn regulate wnt8b expression, revealing a novel feedback control by Fezf2. Nodal signaling and inhibition of Hedgehog signaling are important for establishing the posterior-ventral hypothalamus (Mathieu et al., 2002). shh expression is indeed reduced in fezf2 morphants (Jeong et al., 2007). Therefore, it is possible that the loss of emx2 expression that we observe in fezf2 morphants might involve early alterations of Hedgehog signaling.

To the best of our knowledge, this is the first study to identify transcription factor requirements for Uts1- and Vip-positive neuron development. Uts1 is the teleost ortholog of mammalian urocortin 1 (UCN1) (Vaughan et al., 1995). The urocortins are structurally related to corticotropin-releasing hormone (CRH) and consist of UCN1 (or simply UCN), UCN2 and UCN3 (Hsu and Hsueh, 2001; Lewis et al., 2001; Reyes et al., 2001; Vaughan et al., 1995). UCN1 is broadly distributed in the brain and in the periphery and has been implicated in a wide range of functions including appetite suppression, immunomodulation and stress adaptation (Oki and Sasano, 2004). Although UCN1 is not normally expressed in the MA (according to the mouse expression atlas, http://www.brain-map.org), in colchicine-treated rats it is detected in the lateral mammillary and supramammillary nuclei and posterior hypothalamic area, suggesting that these are sites of low levels of Ucn1 expression in mammals (Bittencourt et al., 1999). In adult zebrafish, uts1 expression is found within the corpus mamillare (Alderman and Bernier, 2007).

Similar to Uts1, Vip is expressed broadly and has pleiotropic regulatory functions. In the central nervous system, Vip expression is found in the suprachiasmatic nucleus and the paraventricular nucleus of the hypothalamus, where it influences circadian clocks and pituitary adrenocorticotropic hormone secretion, respectively (Ceccatelli et al., 1989; Hökfelt et al., 1987; Nussdorfer and Malendowicz, 1998; Vosko et al., 2007). Although Vip immunoreactivity and Vip binding sites have been found in the mammillary nuclei (Saper, 2000; Sarrieau et al., 1994), VIP expression itself in the MA is not reported in the mouse expression atlas, suggesting that the expression of Vip in this region in zebrafish might not be an evolutionarily conserved feature.

In this study, we showed that the numbers of both Vip- and Uts1-positive neurons are reduced in fezf2 morphants, demonstrating that Fezf2 is required for their development. What might be the mechanism by which Fezf2 controls Vip- and Uts1-positive neuron development? In Vip-positive neurons, Fezf2 present in neighboring cells might non-cell-autonomously regulate Vip-positive neuron development. Such a non-cell-autonomous effect of Fezf2 has been proposed for 5-HT- and DA-positive neurons, as Fezf2 is not expressed in these cells either (Levkowitz et al., 2003). Alternatively, Fezf2 might affect the early development of VIP progenitor cells. We consider the latter possibility to be more likely because Fezf2 is expressed broadly in the posterior hypothalamus at early stages and downregulates otp expression. In fact, our fezf2 PM experiments show that the effect of Fezf2 on vip as well as on otp takes place before 1 dpf. Further, Fezf2 overexpression at 34 hpf had no effect on the number of Vip-positive cells. Since Fezf2 overexpression was not spatially restricted, a non-cell-autonomous effect of Fezf2 should have been apparent in such manipulations.

In contrast to Vip, Fezf2 is expressed in Uts1-positive cells. Fezf2 loss-of-function led to Uts1 reduction and Fezf2 overexpression led to an increase in cell number. Here we believe it likely that Fezf2 directly regulates late aspects of Uts1 specification. In Fezf2 overexpression embryos we did not observe changes in the expression of several hypothalamic transcription factors (data not shown), making it unlikely that Fezf2 overexpression leads to gross patterning defects. The relatively mild effect of Fezf2 overexpression on Uts1-positive cell number is likely to be due to the fact that heat shock was performed at 2.5 dpf. Much of the hypothalamic differentiation program has already been initiated at this stage, making it difficult to induce significant changes. Furthermore, the expression level of Fezf2 in our overexpression experiments is likely to be significantly lower than that in the Fezf2 overexpression system reported previously, which uses the Gal4-UAS system for strong transgene induction (Jeong et al., 2007). That we nonetheless detected an effect in the number of Uts1-positive cells suggests that these cells might be exquisitely sensitive to the level of Fezf2.

Several mechanisms have been suggested to explain how great neuronal diversity can be achieved with a small number of common factors. For example, in the developing spinal cord, feed-forward regulatory loops segregate the fate of motoneurons and interneurons although they are specified by related LIM complexes (Lee et al., 2008). In the Drosophila central nervous system, a temporal cascade of transcription factors limits progenitor competence over time, and can act in conjunction with subtemporal genes to create diverse cell types (Baumgardt et al., 2009). By contrast, although several transcription factors have been identified as being involved in hypothalamic neuronal development, it is not understood how these factors interact to generate different neuron types. Our study identified that interplay among fezf2, otp, foxb1.2 and sim1a is crucial to define different domains within the MA and generate distinct neuronal subtypes. Strikingly, Otp and Sim1a are transcription factors that regulate the development of the anterior hypothalamic nucleus, such as the paraventricular nucleus in mouse, as well as the homologous PO in zebrafish. Here we showed that Otp regulates fezf2 differently in the PO and the MA, thereby identifying one possible mechanism by which Otp exerts different functions in distinct hypothalamic regions.

Our findings provide important mechanistic insight into the generation of neuronal diversity in the hypothalamus. First, early regulators such as Fezf2 are redeployed later to generate different neuronal subtypes. Second, common transcription factor sets are used in distinct hypothalamic regions in a different regulatory context.

We thank Gonzalo Alvarez-Bolado for helpful discussions and comments on the manuscript; Jose Arturo Guiterrez-Triana, Colette Maurer, Ingrid Lohmann and Detlev Arendt for critical reading of the manuscript; Matthias Carl, Laure Bally-Cuif, Wolfgang Driever and Hans-Martin Pogoda for providing probes or MOs; Günes Ozhan-Kizil and Rodrigo De Marco for helpful advice; Regina Singer for technical assistance; and Gabi Shoeman, Regina Singer and Angelika Schoell for expert fish care.