Bsx controls pineal complex development

Development of the epithalamus, which comprises the pineal complex (PC) in the dorsal midline of the diencephalon and the flanking paired habenulae, has long fascinated developmental biologists because of both the formation of left-right asymmetrical anatomical structures and light-regulated melatonin synthesis in the pineal system (Concha and Wilson, 2001; Duboc et al., 2015; Roberson and Halpern, 2018; Sapède and Cau, 2013). In mammals, information about light conditions reaches the pineal gland (PG) via the retina and suprachiasmatic nucleus, whereas the PG of poikilothermic vertebrates is directly photoreceptive (Dodt, 1973).

The PC comprises the dorsomedial PG and the adjacent parapineal organ (PP). During zebrafish embryonic development, cells from the anterior part of the pineal anlage migrate away from the midline to form the PP, which, in >95% of wild-type embryos is located on the left side (Concha et al., 2000). Laser ablation studies revealed that unilateral PP formation is necessary for asymmetric development of the habenulae (Gamse et al., 2003). Habenula asymmetry is not only reflected in differential gene expression on the left and right sides, but is also apparent in different efferent projections of the two lobes (reviewed in Duboc et al., 2015). Functional implications of habenula laterality were recently demonstrated through right-isomerized zebrafish showing aberrant behavior (Facchin et al., 2015).

The first transcription factor to be identified as being crucial for proper PG development was Floating head (Flh/Noto). The expression domain of flh in the PG is specified by Wnt along the anterior-posterior axis (Masai et al., 1997), and by Bmp along the dorsoventral axis (Barth et al., 1999). Tbx2b, which was previously implicated in retinal photoreceptor (PhR) cell differentiation (Gross and Dowling, 2005), has been proposed to specify premigratory PP cells in the anterior part of the pineal anlage (Snelson et al., 2008a,b). Subsequently, in an Fgf8a-dependent mechanism, those cells migrate leftward away from the midline to form the PP cell cluster (Regan et al., 2009; Clanton et al., 2013). Pitx2c, a target of Nodal signaling, restricts the number of PP cells (Garric et al., 2014). Most recently, Nr2e3, another transcription factor with a long-established role in retinal PhR cell specification (Akhmedov et al., 2000), was described as a positive regulator of PG cells and a negative regulator of PP cell fate (Khuansuwan et al., 2016).

Although a small structure, the exact identity and number of different cell types within the teleost PC in adult and larval stages is still disputed. Based on morphological features, pineal PhR cells in the adult teleost are difficult to classify into rod-like or cone-like types because of their rudimentary outer segments (Meissl and Yáñez, 1994). In the larval PG, PhR cell morphology is even less well characterized. However, building on distinct gene expression profiles of rod and cone cells in the retina, discrimination between PhR cell types in the early larval PG is frequently based on the expression of genes involved in retinal rod- or cone-specific phototransduction (Lagman et al., 2015). Apart from PhR cells, the teleost PG harbors neurons that form long and extensively arborizing projections to both diencephalic and mesencephalic nuclei (Yáñez et al., 2009). Most recently, cells expressing agouti-related peptide 2 (agrp2) were suggested to be a novel cell type in the larval zebrafish PG (Shainer et al., 2017). Despite their crucial and well-described role in the establishment of left-right asymmetry, functional characterization of PP cells is sparse. However, recent transcriptome analyses suggest that PP cells represent a specialized form of PhR cells (Khuansuwan and Gamse, 2014).

Here, we studied the Brain-specific homeobox (BSX) transcription factor, which is highly conserved among vertebrates (Fig. S1A). Mammalian Bsx transcripts have been demonstrated to be expressed in the developing PG, ventral hypothalamus and telencephalic septum (Cremona et al., 2004). BSX in the hypothalamus is required for normal expression levels of two genes encoding potent orexigenic neuropeptides: agouti-related neuropeptide (Agrp) and neuropeptide Y (Npy) (Sakkou et al., 2007). Data on the role of BSX in PG development and function are sparse: PG hypoplasia in Bsx mutant mice has been described but not characterized in detail (McArthur and Ohtoshi, 2007) and a role for Bsx in pineal PhR cell differentiation has been suggested (D'Autilia et al., 2010).

We detected zebrafish bsx expression in anatomical domains corresponding to those described in mice. We explored Bsx functions in epithalamus development and integrate Bsx into the current regulatory network underlying PC development. Bsx is required for PP cell specification and, thus, is crucial for asymmetric development of the habenulae. In addition, Bsx is required for proper differentiation of PhR cells and neuroendocrine melatonin-releasing pinealocytes.

Performing whole-mount in situ hybridization (WISH), we found bsx in zebrafish embryos to be first expressed during the mid-somitogenesis stages [10 somites, 14 h post fertilization (hpf); Fig. 1A], correlating with late neurulation. The earliest bsx expression domain was detected in the dorsal-most part of the diencephalon in the pineal anlage area (Fig. 1A, arrow and A′), and had become more compact by the 20-somite stage (19 hpf; Fig. 1B,B′) and 24 hpf (Fig. 1C,C′). At ∼30 hpf, bsx expression had become asymmetric (Fig. 1D,D′), indicating that bsx is expressed in both medially located PG cells and leftward migrating PP cells. PC bsx expression persisted during both embryonic and larval development (Fig. 1E-I′).

Another bsx expression domain in the presumptive hypothalamus became apparent during mid-somitogenesis, initially comprising a few scattered cells (10-somite stage; Fig. 1A, arrowhead). bsx-expressing cell clusters formed in the ventral and caudal hypothalamus as well as in the rostral hypothalamus/prethalamus region (Fig. 1D-I). At ∼24 hpf, a few cells started to express bsx in the ventral telencephalon, which, during the second day of development, formed a domain that was likely to be located in the septal region, based on comparison with data from mice (Cremona et al., 2004). bsx expression in the forebrain persisted throughout all the larval stages assessed (until 5 days post fertilization; dpf).

Although Bsx expression in both the PG and hypothalamus of mice has been described previously (Cremona et al., 2004), our work revealed asymmetric bsx expression in the zebrafish PC. Only one study had previously addressed the molecular function of Bsx in the PG, revealing strong circadian rhythmicity in expression of bsx in Xenopus laevis larvae, with high expression levels during the night and almost no expression during the day (D'Autilia et al., 2010). However, in zebrafish, we found little or no circadian variation in bsx expression and no significant differences in expression levels when comparing light-entrained embryos with embryos raised in darkness (Fig. S2), suggesting different Bsx functions in amphibians compared with teleosts.

To reveal which cell types of the PC express bsx, we performed double-fluorescent WISH. We found that bsx is co-expressed with exorhodopsin (exorh), a marker of rod-like PhR cells (Fig. 2A-A″), and the gnat2 gene, encoding cone-specific G protein-transducing alpha subunit (Fig. 2B-B″), indicating that bsx is present in both rod- and cone-like PhR cells. A well-studied function of the PG is the release of melatonin during the night. Thus, we assessed the co-expression of bsx and tryptophan hydroxylase 1a (tph1a), encoding a key enzyme in the synthesis of the melatonin precursor serotonin. Co-expression of both genes (Fig. 2C-C″) indicated that melatonin-releasing cells express bsx. We also found bsx transcripts in cells expressing agrp2 (Fig. 2D-D″), which were recently proposed as a novel pineal cell type (Shainer et al., 2017). Colocalization of bsx and gfi1ab transcripts demonstrated bsx expression in not only PG cells, but also PP neurons (Fig. 2E-E″). To test whether bsx is also expressed in developing pineal PNs, we used elavl3 as a marker to label these neurons (Cau et al., 2008). We found bsx to be colocalized with elavl3 (Fig. 2F-F″), indicating that at least a subset of PNs also express bsx. From these experiments, we conclude that cells of each major cell type in the PC express bsx.

Although elavl3-expressing PNs have been described to be located mainly in two lateral fields within the PG (Cau et al., 2008), we noted that additional cell types might preferentially cluster in subregions of the PG (Fig. 2). When analyzing z-projections of confocal image stacks (Fig. S3), we found photoreceptive cells clustered in the anterior part of the PG, whereas agrp2-expressing cells were distributed more evenly, with a slightly stronger accumulation along the midline in the posterior part of the PG. We integrated these observations into a model of PC cellular organization (Fig. 8B).

The pineal anlage is specified along the anterior-posterior axis by Wnt (Masai et al., 1997), and along the dorsoventral axis by Bmp (Barth et al., 1999). Using Tg(hsp70l:Wnt8a-GFP), we overexpressed a Wnt8a-GFP fusion protein by heat-shock treatment from 12 to 13 hpf to induce canonical Wnt signaling and to test whether the pineal bsx expression domain expands throughout the dorsal forebrain. However, when analyzed at 16 hpf, we found bsx expression in the pineal anlage to be unaffected by high levels of Wnt8a (Fig. 3A,B). As a control, and in accordance with the literature (Masai et al., 1997), we confirmed that heat-shock Wnt8a overexpression strongly expanded the flh expression domain anteriorly (Fig. 3C,D). Inhibition of Wnt signaling by the chemical inhibitor IWR-1 exerted no effect on the pineal bsx expression domain (Fig. 3E,F), whereas a substantial downregulation of flh expression was observed (Fig. 3G,H). We conclude that bsx expression in the pineal anlage is not regulated by Wnt signaling.

We analyzed the effect of Bmp overexpression on bsx expression by applying heat-shock from 12 to 13 hpf to Tg(hsp70l:bmp2b) embryos. We found bsx expression in the pineal region at 24 hpf to be expanded not only ventrally, but also along the anterior-posterior axis (Fig. 3I-J′). Conversely, inhibition of Bmp signaling through 12 to 13 hpf by heat-shock-driven overexpression of a dominant-negative Bmp receptor from Tg(hsp70l:dnXla.Bmpr1a-GFP) resulted in a reduction of bsx expression in the pineal anlage at 24 hpf (Fig. 3K-L′).

Given that bsx expression starts around the time when the pineal anlage fuses at the midline (Fig. 1A,A′), we assessed whether neural tube closure is necessary for bsx expression. We found that, in lefty1-injected embryos, in which the neural tube fails to close, bsx was expressed bilaterally in the epithalamus (Fig. S4A,B). This indicates that neural tube closure is not necessary for bsx expression in the pineal anlage. Later in development, left-sided Nodal signals act as positional cues to direct the PP to the left side (Concha et al., 2000; Liang et al., 2000). In accordance with this model, we found sidedness of PP bsx expression to be randomized in ntl/ta mutant embryos, which lack midline structures thought to represent a diffusion barrier for Nodal (Concha et al., 2000; Fig. S4C,D).

To analyze the role of Bsx in the PC, we used Transcription Activator-like Effector Nucleases (TALENs) to establish a genetic bsx loss-of-function mutation. Using a previously described TALEN pair (Sander et al., 2011), we targeted the homeobox of bsx. Sequencing the targeted genomic region in F2 animals revealed an indel (25 bp deletion and 5 bp random insertion) leading to a frame shift after 114 amino acids, and introducing a premature stop codon after 14 additional amino acids. This bsx^m1376 allele was predicted to express a truncated Bsx protein lacking most of its DNA-binding homeodomain (Fig. S1D). We compared the biological activities of bsx wild-type and bsx^m1376 synthetic mRNAs by microinjection at the one-cell stage. We found that overexpression of wild-type bsx mRNA strongly impeded early development, whereas bsx^m1376 mRNA did not interfere with normal development (Fig. S5). Hence, bsx^m1376 is considered a null allele.

To investigate potential pineal morphogenesis and differentiation defects in bsx mutant larvae, we made use of otx5 as a marker for PC cells adopting a PhR cell fate (Gamse et al., 2002). otx5 expression was reduced in the PG of bsx mutants and a domain corresponding to the PP was not detected (Fig. 4A,A′ arrowheads). In agreement, we found PP cell-specific gfi1ab expression to be completely absent in bsx mutant larvae (Fig. 4B,B′ arrowheads). In fgf8a and tbx2b mutants, the PP cell cluster was not established on the left side of the brain, and cells expressing PP-specific genes were still present but failed to be displaced from the midline, indicating a migration defect (Regan et al., 2009; Snelson et al., 2008b). By contrast, failure to detect any PP-specific gene expression suggested an early specification defect in bsx mutants. To test this possibility, we followed GFP⁺ cells in Tg(foxd3:GFP), which marked early PP cells forming at the midline in the anterior part of the pineal anlage before 34 hpf, and migrating to the left over the following 12 h. However, in bsx mutant embryos, PP cells expressing GFP did not form at 34 hpf or later (Fig. 4H). The absence of PP cells at this early specification stage was confirmed by the lack of sox1a expression in the PC of bsx mutants at 30 hpf (Fig. S6), when sox1a is a marker of premigratory PP cells (Clanton, 2013; Khuansuwan et al., 2016). In addition, LIM homeobox 2b (lhx2b) expression, which is detected at the onset of PP cell migration (Clanton, 2013), is reduced in bsx mutants (Fig. 4C,C′ arrowheads).

Formation of the PP in the left epithalamus was previously demonstrated to be necessary for the establishment of habenula asymmetry in wild-type larvae: laser ablation of the PP cells before or during migration results in right-isomerized habenulae (Gamse et al., 2003). In line with these results, bsx mutant larvae developed double-right-sided habenulae, as detected by the symmetric expression of potassium channel tetramerization domain-containing 12.1 (kctd12.1) and kctd12.2 genes (Fig. 4D-E′). Habenula right isomerization in bsx mutant embryos was confirmed by the absence of adenylate cyclase-activating polypeptide 1a (adcyap1a) expression, which, in wild-type embryos, is expressed in the left habenula only (Fig. 4F,F′), and by bilateral expression of slc18a3b (vesicular acetylcholine transporter; Fig. 4G,G′), a cholinergic gene that, in wild-type embryos, is expressed at higher levels in the right habenula (Hong et al., 2013).

Taken together, our data suggest an early defect in PP development, subsequently leading to habenula right isomerism. We note that, in addition to PP absence, otx5 and lhx2 expression is also reduced in the PG (Fig. 4A,A′,C,C′), suggesting not only PP, but also PG defects in bsx mutants.

Upon loss of flh expression, pineal neurogenesis is impaired (Masai et al., 1997). We found bsx expression in flh mutants to be substantially downregulated (Fig. 5A,B). This indicates that bsx expression in the PC is, to some extent, dependent on Flh. Conversely, flh expression was normal in bsx mutants (Fig. 5G,H), suggesting that Bsx does not regulate flh. Another transcription factor acting downstream of Flh in PC differentiation, otx5, was downregulated in the PG of bsx mutants (Fig. 4A,A′). Using a previously described morpholino (MO) for otx5 knockdown (Gamse et al., 2002), we found bsx to be expressed normally upon knock down of Otx5 (Fig. S7A-D). These results indicate that otx5 is regulated by Bsx, but not vice versa. As previously shown, otx5 transcript levels are reduced and displaced from the midline in flh mutants (Gamse et al., 2002). We analyzed the interplay between flh and bsx in double mutants. We found that, whereas loss of either Bsx or Flh was sufficient to substantially reduce the expression of otx5 (Fig. 5C-E), loss of both Bsx and Flh was required for complete absence of otx5 transcripts from the PC (Fig. 5F). This indicates that Bsx and Flh provide partially redundant parallel input into otx5 regulation in PhR cell specification.

Next, we assessed a potential connection between Bsx and other transcriptional regulators implicated in PP specification. Expression of nr2e3 in bsx mutants appeared to be unaffected (Fig. 5K,L). Likewise, bsx expression was not substantially altered upon knock down of nr2e3 (Fig. S7H), indicating that Bsx and Nr2e3 do not regulate the expression of the other, but rather act in parallel. Also, expression of tbx2b was normal in bsx mutants (Fig. 5M,N). However, when knocking down tbx2b expression by MO injection, we found no or only very few bsx⁺ PP cells displaced from the medial PG (Fig. S7F). This result is consistent with a previously published report showing both a decreased number and migration failure of PP cells upon knock down of tbx2b (Snelson et al., 2008b).

Next, we analyzed the interplay between bsx and pitx2c, a Nodal target expressed in the left diencephalon, which has been shown to regulate PP cell number (Garric et al., 2014). We found left-side-specific expression of pitx2c to be normal in bsx mutants (Fig. 5O,P) and bsx expression not to be altered in pitx2c morphants (Fig. S7G). Thus, we hypothesize that Bsx and Pitx2c determine PP cell fate largely independent of each other.

Fgf8a signaling is crucial for migration of PP cells to the left (Regan et al., 2009; Clanton et al., 2013). In line with this, we found no bsx expression displaced from the midline in hypomorphic fgf8a^ti282a mutants (Fig. S7J,K). Given that we detected no difference in fgf8a expression between bsx mutants and wild type (Fig. 5Q,R), we conclude that fgf8a expression is independent of Bsx. Thus, both Bsx and Fgf8a appear to be necessary for the establishment of a left-positioned PP, but one factor does not substantially influence the expression of the other. Furthermore, we found that, in bsx mutant embryos, the absence of PP cells was also apparent through loss of bsx expression left of the PG but that bsx expression was not affected in the PG cells themselves (Fig. 5I,J). This indicates that Bsx activity is not required to maintain bsx expression. In conclusion, our data establish Bsx as an important factor in the pineal transcriptional specification network that is necessary for PP cell fate, and is expressed mostly independent of previously described PC transcription factors, except Flh.

Both otx5 and lhx2 have been implicated in PhR cell differentiation (Gamse et al., 2002; Gordon et al., 2013; Roy et al., 2013). Given that the expression of both genes is substantially downregulated in bsx mutants, we analyzed formation of the different pineal cell types using Tg(foxD3:GFP) (Gilmour et al., 2002), which has been suggested to label PP cells, PNs and cone-like PhR cells in the PG (Clanton et al., 2013). We also assessed Arr3a immunoreactivity, which has been described as a cone-specific marker (Larison and Bremiller, 1990). We found no Arr3a immunoreactivity in bsx mutants, and the expression of GFP from Tg(foxD3:GFP) to be significantly reduced in bsx mutants (Fig. 6A,B).

To further characterize the extent to which the different pineal cell types are affected by Bsx loss, we performed double-immunolabeling of GFP and Elavl3/HuC in Tg(foxD3:GFP) embryos and counted cells (Fig. 6C-D‴, Movie 1). We found that the number of GFP⁺ cells was drastically reduced in bsx mutants (15.1±2.9 compared with wild type 82.7±7.0, mean±s.d.; Fig. 6E,F and Table S3). The number of HuC⁺ cells was also significantly reduced in bsx mutants (27.9±4.4 compared with wild type 46.9±3.3). In wild-type embryos, approximately half of the GFP⁺ cells did not express HuC (37.6±6.2), whereas, in bsx mutants, there are very few such cells (0.7±1.5). That most of the remaining GFP⁺ cells in bsx mutants were HuC⁺ supports the hypothesis that most of the GFP⁺ cells in bsx mutants are PNs. In addition to reduced cell numbers in bsx mutants, we also observed that remaining PNs in bsx mutants did not form lateral clusters but tended to be located only in the caudal-most part of the PG (Fig. 6D′, Movie 1). We observed axonal projections originating from the GFP⁺ cells in bsx mutants, indicating that these cells correspond to PNs (Fig. 6G-H). However, the PN neurites frequently formed supernumerary and morphologically abnormal branches compared with wild type (Fig. 6G-I).

To determine the formation of neuroendocrine pinealocytes, we analyzed PG expression of enzymes involved in melatonin production. We found that expression of arylalkylamine N-acetyltransferase 2 (aanat2) (Fig. 6J,K) and of tph1a and tph2 (Fig. 6L-O; Fig. S8) was completely absent in bsx mutants. This suggests that the PG of bsx mutants is unable to produce both melatonin and its precursor, serotonin. Furthermore, expression of agrp2 was strongly reduced in bsx mutants (Fig. 6P,Q).

Loss or reduction of PG cell types can be caused by apoptosis; therefore, we performed terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) assays at 30 hpf or 72 hpf, but found no evidence of increased apoptosis in the PC of bsx mutants (Fig. S9). Thus, loss of specific cell types could be caused by mis-specification or abnormal differentiation rather than by cell death.

Next, we investigated Bsx activities in cone differentiation. For most of the marker genes analyzed, we did not detect complete elimination of expression, but variation in WISH stain intensity or area (Fig. 7A). To identify subtle, but significant changes, we quantified the staining intensity of our WISH data through background-adjusted measurements of integrated density in the PG (Fig. 7B, Materials and Methods, Fig. S10) of genotyped embryos. We analyzed the expression of the cone-specific G-protein transducin alpha and gamma subunits gnat2 and gngt2a. Interestingly, we found gnat2 expression to be normal in bsx mutants, whereas that of gngt2a was downregulated. We further analyzed the expression of the cone-specific recoverins (rcv) rcv1b, rcv2a and rcv2b and detected a significant downregulation of each of them in bsx mutants. Given that Arr3a, for which we demonstrated the absence of both the protein (Fig. 6A) and transcripts (Fig. 7A) in the PG of bsx mutants, has been shown to be red/green double cone specific (Renninger et al., 2011), we investigated whether cones with different spectral sensitivity are differentially affected upon loss of Bsx. Thus, we analyzed the expression of the UV- and blue-cone-specific arrestin 3b (arr3b) in bsx mutants, and found expression levels to be similar to wild type. Even more strikingly, we found the UV-sensitive opsin 1 (cone pigments), short-wave-sensitive 1 (opn1sw1) to be expressed in bsx mutants, whereas it was not expressed in wild-type PGs in the larval stages. However, when we assessed the expression of rcv1a, which is specific to UV cones and rods (Zang et al., 2015), we observed a significant downregulation in bsx mutants. This result suggests that rod-like PhR cells are also affected in bsx mutants. However, similar to the cone-specific transducin subunits, the rod-specific alpha subunit gnat1 was unchanged in bsx mutants, whereas the gamma subunit gngt1 was downregulated. Remarkably, we found the rod-specific opsins rhodopsin (rho) and exorh to be significantly upregulated in bsx mutants (Fig. 7A,B).

Given that we demonstrated that Bsx acts on otx5, a well-established determinant of PhR cell fate, we tested whether the effects of Bsx loss on pineal PhR cells were secondary effects resulting from downregulation of otx5. Thus, we assessed opn1sw1 expression in embryos injected with a MO against otx5 (Gamse et al., 2002), but were unable to detect opn1sw1 in the PG of those embryos at 2 dpf (Fig. S7L,M). By contrast, bsx mutant embryos expressed opn1sw1 in the PG at 2 dpf (Fig. S5O,P), indicating that Bsx has a function in PhR cell differentiation that cannot solely be explained through the regulation of otx5. Given that Tbx2b specifically promotes UV-cone cell fate in the retina (Alvarez-Delfin et al., 2009), we assessed whether knock down of tbx2b would interfere with ectopic opn1sw1 expression in bsx mutants. We found a significantly reduced pineal opn1sw1 expression in bsx mutants upon knock down of tbx2b (Fig. S11). This was consistent with Tbx2b positively and Bsx negatively regulating opn1sw1 expression.

The ambiguous effects of Bsx loss on cone- or rod-like marker gene expression raise doubts about cell type specificity in the zebrafish PC during early larval stages. Using double-fluorescent WISH, we visualized exorh and rcv2b transcripts, which selectively label rod or cone PhR cells, respectively. To our surprise, instead of a mutually exclusive expression pattern, we found colocalization of exorh and rcv2b transcripts in PG cells of wild-type larvae 78 hpf (Fig. 7C-C″). This suggests that, at the developmental stage analyzed, PG cells do not strictly segregate into cells expressing exclusively rod or cone markers.

Our study identifies the highly conserved Bsx homeodomain transcription factor as a major regulator of PC development in zebrafish, affecting specification of the pineal anlage, including the PP, cell lineage decisions, morphogenesis and differentiation in the pineal system (Fig. 8A). bsx expression starts in the pineal anlage around the 10-somite stage, and, thus, slightly later than that of flh and tbx2b (both at the 6-somite stage; Snelson et al., 2008b) or nr2e3 (8-somite stage; Khuansuwan et al., 2016). In line with flh being expressed earlier than bsx, Flh contributes to activation of bsx expression, even though, in some pineal anlage cells, Flh is not strictly required for bsx to be expressed. Loss of Bsx activity in turn does not affect flh expression. Analysis of flh and bsx double-mutant embryos revealed that Flh and Bsx act in parallel in specification of the pineal anlage, because only double mutants lacked otx5 expression completely, suggesting an absence of PhR cells. The transcription factor Flh can be grouped together with Bsx into the ANTP class of homeodomain proteins (Ferrier, 2016), and might share target specificity, providing a potential molecular basis for partial redundancy of both transcription factors.

Our data further suggest that, in the control of flh and bsx expression, two signaling pathways converge: Wnt and Bmp, which together pattern the dorsal diencephalon and specify the pineal anlage. Wnt signaling induces flh and determines the extent of the pineal anlage in the dorsal diencephalon (Masai et al., 1997). Surprisingly, despite the vast expansion of the flh domain upon ectopic Wnt signaling, we did not observe a concomitant expansion of the bsx expression domain in the dorsal diencephalon. By contrast, we found the bsx expression domains to be specified by Bmp signaling levels. Bmp was previously described in early dorsoventral patterning of the neural plate (Barth et al., 1999), as well as in later aspects of PG development, when it promotes PhR cell fate (Quillien et al., 2011). However, our data revealed that Bmp also has a role in pineal anlage specification shortly after neural tube closure. This supports a model in which Wnt and Bmp signaling provide the spatial coordinates to control Flh and Bsx activity, respectively. Given that Wnt activity has been reported to expand the pineal anlage anteriorly (Masai et al., 1997), whereas we see a Bmp-mediated expansion of the bsx domain in a ventral direction, we hypothesize that surface ectoderm-derived Bmp specifies the ventral extent of the pineal anlage, whereas Wnt signaling specifies the anterior extent.

The generation of pineal asymmetry crucially depends on Nodal signaling, which controls unilateral expression of pitx2c in the left diencephalon and PP development (Liang et al., 2000; Concha et al., 2000; Garric et al., 2014). Although Nodal is not required for bsx expression in the pineal anlage, when we abolished Nodal asymmetry in ntl/ta mutants, sidedness of unilateral bsx expression in the PP primordium was randomized. Our loss-of-function experiments revealed that bsx and pitx2c expression do not regulate each other. However, we found a complete absence of the PP in bsx mutants, as judged by otx5, gfi1ab, lhx2 and sox1a expression, revealing a strict requirement for bsx in PP development. fgf8a expression, which appears to instruct PP cell fate (Clanton et al., 2013) as well as PP cell migration away from the midline (Regan et al., 2009), is expressed normally in bsx mutants, as is tbx2b, which instructs PP formation (Snelson et al., 2008b). Thus, we conclude that Bsx is strictly required for PP cell specification in the pineal primordium. In line with the established role of the PP in development of habenular asymmetry (Gamse et al., 2003), the habenulae in bsx mutant embryos were right isomerized. Complex behavioral phenotypes of right-isomerized zebrafish recently received attention (Facchin et al., 2015), for which bsx mutant zebrafish could constitute a valuable genetic model.

Our analysis of confocal stacks of labeled cell types in the PG suggested that PG cell types are spatially clustered: although it is well established that PNs are located in lateral positions in the caudal two-thirds of the PG, our data suggest that PhR cells cluster in the rostral median PG (Fig. 8B). In bsx mutants, this organization was disturbed, and the PNs clustered at the caudal margin of the PG, suggesting that Bsx is also required for proper spatial organization of different cell types.

Our data further revealed that Bsx controls cell differentiation of multiple cell types in the PG. Melatonin is a prominent PG neuroendocrine hormone with circadian expression in zebrafish (Gothilf et al., 1999). Different from circadian bsx expression reported in Xenopus larval PGs (D'Autilia et al., 2010), we could not detect circadian modulation of bsx expression in the zebrafish PG. However, bsx mutant PGs did not express tph1a, tph2 or aanat2, and, thus, are devoid of serotonin- and melatonin-producing PG cells. Loss of melatonin was recently reported to disturb sleep patterns in zebrafish (Gandhi et al., 2015), and bsx mutants could provide a model to assess melatonin and potential pineal-specific serotonin functions. Furthermore, we found Bsx to regulate agrp2 expression in the PC, suggesting that these bsx-expressing cell lineages might share Bsx transcriptional targets with hypothalamic cell lineages in which BSX regulates expression of the orexigenic neuropeptide Agrp (Sakkou et al., 2007).

The severe effects of loss of Bsx on cell lineages were also reflected by a more than fivefold reduction in GFP-expressing cells in Tg(foxd3:GFP), as well as a reduction in HuC⁺ PN numbers by approximately half. Although the remaining PNs did form neurites, these projections showed aberrant branching morphology compared with wild-type embryos. We do not know the cause for the projection phenotype, but the abnormal caudomedial clustering of PNs in bsx mutants could lead to different neurite exit points from the PG and different axon navigation environments.

Rod-, red/green- and blue cone-specific opsins have been reported for the teleost PG (Ekström and Meissl, 2003). However, it remains unclear whether this corresponds to different PhR cell types or to a single cell type with mixed PhR properties. We found a putative rod-specific marker (gngT1) to be downregulated in bsx mutants, whereas others (rho and exorh) were upregulated. Similarly, we found several supposedly cone-specific markers (gngT2a, rcv1b, rcv2a and rcv2b) to be downregulated in bsx mutants, whereas, strikingly, the cone-specific UV-sensitive opsin opn1sw1 was strongly expressed in bsx mutant but not wild-type embryos. In terms of spectral sensitivity, we could not find a clear tendency of Bsx influencing short-wave- or longer-wave-specific PhR cells. However, when we arranged the PhR cell markers according to their functional role in the phototransduction cascade, we detected a pattern consistent with a role of Bsx or its downstream effectors in downregulating opsin expression, but promoting expression of recoverins, as well as of specific transducins and arr3a. Thus, we hypothesize that Bsx in PhR cell differentiation is involved in regulatory circuits encoding PhR cell compensation strategies, balancing the relative expression of photopigments and downstream components of the phototransduction cascade. In line with this hypothesis, we assume abnormalities in response kinetics and light adaptation of pineal PhR cells in bsx mutants.

The observation that rod- or cone-like markers were affected differentially in bsx mutants prompted us to test the identity of pineal PhR cells. We found cells in which expression of the rod-specific exorh was colocalized with cone-specific rcv2b. If co-expression of these and other markers are confirmed in future studies, the current model of clearly distinct cone-like and rod-like PhR cells in the larval PG needs to be re-examined. Although morphologically it is difficult to classify pineal PhR cells according to a canonical rod/cone discrimination, two clearly distinct types of PhR cells can be detected in the adult teleost PG based on electrophysiological properties (Ekström and Meissl, 2003). Therefore, co-expression in the embryo could be a transient phenomenon and correspond to PhR progenitor cells with the plasticity to develop into cone- or rod-like PhR cells during postembryonic stages. The idea of cone-to-rod and rod-to-cone ‘transmutations’ was proposed more than 80 years ago (Walls, 1934), and was recently supported by findings in snakes (Schott et al., 2016; Simões et al., 2016). Thus, early developmental plasticity could contribute to evolutionary dynamics.

Bsx has crucial roles in distinct steps of PC development, including early PC specification, PP formation and PhR cell differentiation. These roles might be conserved, in part, in neuroendocrine mammalian pinealocytes, notwithstanding their loss of photoreceptive functions. Furthermore, Bsx in the PC is crucial for asymmetric differentiation of the zebrafish habenulae, structures for which asymmetries were recently demonstrated both in volume (Ahumada-Galleguillos et al., 2017) and connectivity (Hétu et al., 2016) also in humans. Thus, bsx mutant zebrafish could be developed as a model to study fundamental questions in neurobiology, including brain asymmetry, circuits balancing expression of photopigment and phototransduction components, and brain function without melatonin and pineal serotonin.

Zebrafish embryos were obtained through natural breeding of ABTL zebrafish. All embryos were kept in 3 g/l Red Sea salt (Red Sea) and 28.5°C under standard conditions in a 14 h light/10 h dark cycle and staged according to Kimmel et al. (1995). To prevent pigmentation, embryos were treated with 0.2 mM N-phenylthiourea (Sigma-Aldrich).

The following mutant or transgenic lines were used: bsx^m1376 (this study), ntl^b160 (Halpern et al., 1993), flhⁿ¹ (Talbot et al., 1995), ace/fgf8a^ti282a (Reifers et al., 1998), Tg(foxD3:GFP)^zf15 (Gilmour et al., 2002), Tg(hsp70l:Bmp2b)^fr13 (Chocron et al., 2007), Tg(hsp70l:dnXlaBmpr1a-GFP)^w30 (Pyati et al., 2005) and Tg(hsp70l:Wnt8a-GFP)^w34 (Weidinger et al., 2005).

All experiments were carried out in accordance with the German Animal Welfare Act.

Heat shock was performed by placing petri dishes with zebrafish embryos in a 39°C water bath for 1 h.

For cDNA cloning, total RNA was extracted from embryos of the desired staged, homogenized using a syringe (100 Sterican, Ø 0.60×30 mm, Braun), and suspended in 600 µl RLT Lysis Buffer from the RNAeasy Mini Kit (Qiagen), complemented with 10 µl 2-mercaptoethanol. Lysate was then filtered on a QIAshredder column (Qiagen) before RNA extraction was continued using the RNAeasy Mini Kit (Qiagen) following the manufacturer's protocol. Total RNA was then used as template for cDNA synthesis using random primers and the Superscript III RT-PCR system (Thermo Fisher Scientific). Coding-sequence fragments were PCR amplified from cDNA using MyTaq DNA Polymerase (Bioline). Amplicons were analyzed on standard gel electrophoresis for correct size and then subcloned into pCRII-TOPO (Invitrogen). All the primer sequences used are shown in Table S1. Correct insertions into vector backbone were confirmed by sequencing. For all plasmids used as template for in vitro transcription of antisense RNA probes, see Table S2.

The full-length wild-type bsx coding sequence was amplified from cDNA using the primers shown for bsx in Table S1. For Gateway cloning, the forward primer (annealing part in italic) was flanked with an attB1 site (bold): 5′-GGGGACAAGTTTGTACAAAAAAGCAGGCTTGATATGAATCTGAACTACACGTCTCCG-3′. The reverse primer (annealing part in italic) was flanked with an attB2R site (bold): 5′-GGGGACCACTTTGTACAAGAAAGCTGGGTCTACTAGAGTAAATGTTCAGGAGAGCAA-3′. A BP recombination reaction was performed using the Gateway BP Clonase II Enzyme Mix (Thermo Fisher Scientific) to generate the Gateway middle entry plasmid pME-bsx. pME-bsx was then recombined in a Multisite Gateway 4-1-2-3 reaction using LR Clonase II Plus Enzyme (Thermo Fisher Scientific) together with the following three plasmids: p5E-CMV/SP6, p3E-polyA and pDestTol2pA2 (all from the Tol2Kit, Kwan et al., 2007) to generate a CMV/SP6:bsx-polyA-pDest construct. Full-length m1376 bsx mRNA was amplified from cDNA of embryos resulting from a bsx^+/m1376 in-cross using the primers shown for bsx in Table S1 and subcloned into pCRII-TOPO (Invitrogen). The bsx^m1376 fragment was excised from an identified clone using XhoI and SpeI and subcloned into a pCS2+ vector opened with XhoI and XbaI to generate a bsx^m1376-pCS2+ construct. Sequencing revealed no or only silent mutations (potential polymorphisms) within the reading frame of wild-type bsx and bsx^m1376, which harbors the TALEN-induced mutation.

bsx TALEN mRNA was transcribed from TAL3034 (Addgene plasmid #41208, deposited by Keith Joung) and TAL3035 (Addgene plasmid #41209, deposited by Keith Joung) (Sander et al., 2011). Then, 1 µg of AgeI-linearized plasmids were used as a template for mRNA transcription with the mMessage mMachine T7 Transcription kit (Ambion/Thermo Fisher Scientific). Fertilized ABTL eggs were injected with bsx TALEN mRNA (200 pg each, for left and right TALEN, per embryo). Twelve injected embryos at 24 hpf were pooled and lysed in Lysis Buffer (10 mM Tris-HCl pH 8, 30 mM KCl, 30 mM NaCl, 1 mM EDTA, 0.5% Tween20, 0.5% Nonidet P-40) containing 1 mg/ml Proteinase K (Sigma-Aldrich). XhoI (NEB) digest of a PCR fragment covering the TALEN target region (forward: 5′-ATTGCAAAAGGAATGCAGATG-3′; reverse: 5′-ATTGTCGTCCAGCGTGTATCT-3′; Fig. S1B) confirmed the occurrence of indels disrupting the XhoI recognition site. Genotyping of further generations was performed using the same PCR and XhoI digest after lysing fin clips or embryonic tails in 50 mM NaOH for 45 min at 95°C, then neutralizing lysis with 1/10 volume 1 M Tris-HCl pH 7.5, before 2 µl of lysed tissue were used for PCR. Then, 15 µl of the PCR reaction was used for XhoI digest and restriction-length polymorphism was visualized by agarose gel electrophoresis (Fig. S1C).

All MOs used in this study have previously been described and were ordered from Gene Tools. MOs were diluted in water and a volume of 1-2 nl was injected into the yolk of embryos at the one-cell stage. For details, see Table 1. Control embryos were injected with 1-2 nl of standard control MO (5′-CCTCTTACCTCAGTTACAATTTATA-3′).

Functional knock down of Otx5 was confirmed by downregulation of irbp in the PG (Gamse et al., 2002, Fig. S5C,D). Splice-blocking activity of the nr2e3 MO was confirmed by RT-PCR on 24 hpf total RNA, revealing the presence of a 300 bp band in knockdown embryos as opposed to a 210 bp band in controls (Fig. S7I) when using the following primers on cDNA: 5′-GGACTTCTTTGTAAAGTGTGTTCTGAC-3′ and 5′-TCATTCCAGCTTGAAGGCATTTC-3′. Functional knock down of Tbx2b was confirmed by absence of the PP cell cluster in the majority of injected embryos (Fig. S7F).

lefty1 (zf-lefty1-pCS2+; Thisse and Thisse, 1999), gfp (pG1-GFP; gift from Gudrun Aspöck) and bsx^m1376 mRNA (bsx^m1376-pCS2+, this study) were synthesized from NotI (NEB) linearized plasmids, and wild-type bsx mRNA (CMV/SP6:bsx-polyA-pDest, this study) from a KpnI (NEB) linearized plasmid, using the Ambion mMessage mMachine Transcription kit, T7 or SP6, according to the manufacturer's protocol.

WISH was carried out as described (Thisse et al., 1993) with the following modifications. Digoxigenin (DIG)-labeled RNA antisense probes were transcribed from the plasmids in Table S2 using T7, SP6 or T3 RNA polymerase (Thermo Scientific Scientific) following the manufacturer's protocol, and were used full length (without alkaline hydrolysis). After dehydration, embryos were subjected to Proteinase K (10 µg/ml in PBST) digestion at room temperature: 3 min for 16 hpf, 5 min for 20 hpf, 10 min for 24 hpf, 15 min for 30 hpf, 25 min for 2 dpf, 45 min for 3 dpf, and 60 min for 4 dpf embryos. The embryos were then post-fixed in 4% PFA/PBST for 30 min. PFA was washed out five times for 5 min in PBST, and embryos were prehybridized in Hybridization Mix [5 mg/ml Torula RNA (Sigma-Aldrich), 50 µg/ml heparin (Sigma-Aldrich), 0.1% Tween 20 (AppliChem) in 50% formamide (AppliChem)/5× SSC] for at least 2 h at 65°C, then incubated overnight in Hybridization Mix containing the probe. The following day, embryos were washed for 20 min each at 65°C: twice in 50% formamide/2× SSCT, twice in 25% formamide/2× SSCT, twice in 2× SSCT, and twice in 0.2× SSCT. They were transferred to PBST and washed twice in PBST at room temperature (5 min each). Embryos were blocked for at least 2 h at room temperature in Blocking Solution [2 mg/ml BSA (AppliChem), 5% sheep serum, in PBST], and incubated with anti-DIG-AP Fab Fragments (Roche) 1:4000 overnight at 4°C. Embryos were then washed six times for 20 min each in PBST and three times for 10 min each in NTMT [0.1 M Tris-HCl (pH 9.5), 0.1 M NaCl, 0.05 M MgCl₂, 0.1% Tween 20]. Staining reaction was in NTMT containing NBT (45 µg/ml) and BCIP (20 µg/ml). After staining, embryos were rinsed in 1 mM EDTA/PBST and post-fixed in 4% PFA for 1 h or overnight at 4°C. Then, embryos were rinsed again in PBST and transferred to 1 mM EDTA/80% glycerol/20% PBST for storage and documentation.

Double-fluorescent WISH was carried out as previously described (Ronneberger et al., 2012).

Apoptotic cells were visualized through a TUNEL assay using the ApopTag peroxidase in situ apoptosis-detection kit (Merck S7100), as previously described (Ryu et al., 2005).

To quantify signals after WISH, z-stack images were recorded of embryos with low differential interference contrast (DIC) contrast settings. Minimum intensity projections of 25 z-planes (1 µm each) encompassing the PG were made with ImageJ. These pictures were then converted to 8-bit gray-scale and inverted. Integrated Density was then calculated for an area surrounding the PG (80 µm circle for Fig. 7 and Fig. S10) or the whole PC (80 µm×120 µm for Fig. S2), and a larger area surrounding the PC (100 µm diameter or 96 µm×138 µm). A threshold value was determined by dividing the area difference by the Integrated Density difference and the picture histogram was clipped at the lower end to this threshold value. After this background subtraction, the Integrated Density of the smaller area was calculated from the adjusted picture (see also Fig. S10 for the described workflow), and is shown in bar or line charts (Figs 7, S2 and S11).

Immunohistochemistry was performed as previously described (Ronneberger et al., 2012). The following primary antibodies were used: rabbit anti-GFP polyclonal (Thermo Fisher Scientific, #A-1112, 1:1000 dilution), chicken anti-GFP polyclonal (Thermo Fisher Scientific, #A-10262, 1:1000 dilution), mouse anti-GFP monoclonal (Thermo Fisher Scientific, #A-11120, 1:1000 dilution), mouse anti-GFP monoclonal Living Colors Jl-8 (ClonTech/Takara Bio USA, Inc, #632381, 1:1000 dilution), Arrestin3a/zpr-1/Fret43 (Larison and Bremiller, 1990, ZIRC, ZFIN ID: ZDB-ATB-081002-43, 1:2 dilution of predilution) and HuC/HuD monoclonal (16A11, Thermo Fisher Scientific, #A-21271, 1:1000 dilution). As secondary antibodies, we used goat anti-rabbit IgG Alexa 488 (Thermo Fisher Scientific, #A-11070, 1:1000 dilution), goat anti-rabbit IgG Alexa 546 (Thermo Fisher Scientific, #A-11035, 1:1000 dilution), goat anti-chicken IgY Alexa 488 (Thermo Fisher Scientific, #A-11039, 1:1000 dilution), goat anti-chicken IgY Alexa 555 (Thermo Fisher Scientific, #A-21437, 1:1000 dilution), goat anti-mouse IgG Alexa 488 (#A-11001, 1:1000 dilution) and F(ab′)2-goat anti-mouse IgG Alexa 555 (Thermo Fisher Scientific, #A-21425, 1:1000 dilution). For nuclear staining, embryos after immunohistochemistry were incubated in TOTO-3 iodide (Invitrogen, 1:1000 dilution) overnight at 4°C, then washed in PBST and transferred to 80% glycerol/PBST for imaging.

After standard DIG/alkaline phosphatase-based WISH, embryos were mounted for brightfield microscopy in 80% glycerol, 20% PBST, 1 mM EDTA. Pictures were taken with AxioCam CC1 on an AxioPlan2 microscope with a PLAN-NEOFLUAR 20×/0.5 or 10×/0.3 objective and DIC optics using the AxioVs40 Software (all Zeiss). From image stacks (1 µm step size), minimum intensity projections of all z-planes encompassing the PG were made using ImageJ.

Time-lapse imaging of Tg(foxD3:GFP)^zf15 embryos was made using Examiner.D1 and ZEN blue software (all Zeiss). Embryos after fluorescent WISH and immunohistochemistry were recorded using a Zeiss LSM 510 or LSM 880. All figures were assembled using Adobe Photoshop CS4 or CS6. When linear adjustment of levels was made, histograms were clipped to the same values for experimental and control images.

To treat embryos with the Wnt inhibitor IWR-1 (Sigma-Aldrich), 75 µl of a 2 mM IWR-1/DMSO stock solution or DMSO alone was added to 30 ml of E3 medium. Embryos were then kept in this 5 µM IWR-1 working concentration or 0.25% DMSO as control starting from 2 hpf until fixed for analysis.

All charts in this manuscript were generated using GraphPad Prism 6 with the following exceptions: line chart in Fig. S2 (MS Excel 2016), bar chart in Fig. S5 (MS Excel 2016), Venn diagrams in Fig. 6 (Venn Diagram Plotter, PNNL, Apache License 2.0, https://omics.pnl.gov/software/venn-diagram-plotter). Sample sizes used in this study did not allow sensible testing for normal distribution. Thus, for pairwise comparison of two groups, we used the distribution-independent Wilcoxon–Mann–Whitney test, calculated with GraphPad Prism 6. Statistical significance is indicated as follows: *P≤0.05; **P≤0.01; ***P≤0.001; ****P≤0.0001; ns=not significant. If applicable, exact P-values and sample sizes (n) are given. For the experiment shown in Fig. S5, embryos from four groups were assigned to three phenotypic classes. We generated contingency tables and tested whether all values met the minimum expectation requirements for a chi-square goodness of fit test introduced by Yarnold in 1970, who proposed in a contingency table of three or more classes s, the minimum expectation can be as small as 5r/s, with r denoting the number of expectations less than five (Yarnold, 1970). Applying this rule, we determined a minimum expectation of 3.33 and found all values in our contingency table to be above 3.33. Thus, we applied chi-square goodness of fit test using GraphPad Prism 6 to determine the P-values given in the legend of Fig. S5.

Thanks to Keith Joung for providing the bsx TALEN plasmids, Martha Manoli for cloning the bsx plasmid used for probe synthesis, Jochen Holzschuh for cloning the tph2 plasmid used for probe synthesis, and Nadja Bischoff, Jana Ernesti, Lena Tittel and Janna Becker for technical assistance with genotyping. Many thanks to Sabine Götter for excellent fish care, Simone Königsmann and Daniela Reuter-Schmitt for technical assistance, and Logan Doil for proof-reading of the manuscript. Thanks to Heiko Löhr and Mathias Hammerschmidt for critical discussions and providing Tg(hsp70l:Bmp2b) zebrafish. Thanks to Wiebke Herzog for providing flh mutant zebrafish. Thanks to Yoav Gothilf for providing the aanat2 plasmid for probe synthesis.