Neurons in the inferior olivary nuclei (IO neurons) send climbing fibers to Purkinje cells to elicit functions of the cerebellum. IO neurons and Purkinje cells are derived from neural progenitors expressing the proneural gene ptf1a. In this study, we found that the homeobox gene gsx2 was co-expressed with ptf1a in IO progenitors in zebrafish. Both gsx2 and ptf1a zebrafish mutants showed a strong reduction or loss of IO neurons. The expression of ptf1a was not affected in gsx2 mutants, and vice versa. In IO progenitors, the ptf1a mutation increased apoptosis whereas the gsx2 mutation did not, suggesting that ptf1a and gsx2 are regulated independently of each other and have distinct roles. The fibroblast growth factors (Fgf) 3 and 8a, and retinoic acid signals negatively and positively, respectively, regulated gsx2 expression and thereby the development of IO neurons. mafba and Hox genes are at least partly involved in the Fgf- and retinoic acid-dependent regulation of IO neuronal development. Our results indicate that gsx2 mediates the rostro-caudal positional signals to specify the identity of IO neurons from ptf1a-expressing neural progenitors.
In an initial step during the formation of neural circuits, neurons are differentiated from their neural/neuronal progenitors. The fate of neurons is thought to be determined by inductive signals that their progenitors receive depending on the position of the neural/neuronal progenitors. By receiving positional signals, neural progenitors express transcription factors, such as basic helix-loop-helix (bHLH) and homeobox transcription factors, that function in the differentiation and specification of neurons. Elucidation of these transcription factors and their regulation is a prerequisite for understanding the mechanism of neural circuit formation. Cerebellar neural circuits provide a good model for revealing positional signal-dependent neurogenesis.
In both mammals and teleosts, such as zebrafish, cerebellar neural circuits are composed of several types of neurons, including granule cells (GCs) and Purkinje cells (PCs) in the cerebellum, as well as neurons in the inferior olivary nuclei (IO neurons) in the caudal hindbrain. PCs receive axons from GCs (parallel fibers) and IO neurons (climbing fibers: CFs), then integrate the input information to elicit the functions of the cerebellum. PCs then send output to the outside of the cerebellum via efferent neurons, which are eurydendroid cells in zebrafish. In zebrafish, some GCs also project their axons to dendrites of crest cells, which are PC-like cells located in the medial octavolateral nucleus (MON) of the dorsal hindbrain (Bae et al., 2009; Hashimoto and Hibi, 2012; Hibi and Shimizu, 2012). These neurons are derived from the hindbrain, which contains seven compartments collectively referred to as the rhombomeres (r). PCs and IO neurons are generated from neural progenitors in the dorsal ventricular zone of the hindbrain. Analyses of mutants and cell lineage tracing for the bHLH-type proneural gene pancreas transcription factor 1a (Ptf1a) in mice revealed that PCs and IO neurons are derived from Ptf1a-expressing progenitors and that Ptf1a is required for the generation of these neurons in mice (Hoshino et al., 2005; Yamada et al., 2007). Lineage tracing with transgenic (Tg) zebrafish also showed that PCs and IO neurons are derived from ptf1a-expressing (ptf1a+) progenitors in zebrafish (Kani et al., 2010). Inhibitory neurons in the dorsal cochlear nuclei (DCN neurons) were also shown to be derived from Ptf1a+ neural progenitor cells in r2-5 in mice (Fujiyama et al., 2009). These reports indicate that, despite being present in the dorsal ventricular zone of all rhombomeres, ptf1a+ progenitors generate different types of neurons depending on their position. At r1, PCs are generated from ptf1a+ progenitors and migrate dorsally, whereas in the caudal hindbrain, IO neurons are generated from ptf1a+ progenitors and migrate ventrally (reviewed by Hashimoto and Hibi, 2012; Hoshino, 2012). However, it is not clear yet how positional signals are interpreted to specify neurons from ptf1a+ progenitors and what molecules control the differentiation and/or specification of IO neurons.
Rostro-caudal patterning of the caudal hindbrain is controlled by gradients of fibroblast growth factors 3 and 8a (Fgf3/8a) and retinoic acid (RA) signals (Dupé and Lumsden, 2001; Gavalas and Krumlauf, 2000; Marin and Charnay, 2000; Maves et al., 2002; Maves and Kimmel, 2005). During early neurogenesis in zebrafish, fgf3 and fgf8a are expressed at r4 and are required for fate determination of r5 and r6 (Maves et al., 2002; Shimizu et al., 2006; Walshe et al., 2002; Wiellette and Sive, 2004). In the aldehyde dehydrogenase 1 family, the A2 gene (aldh1a2) encodes an enzyme that is required for the production of RA and is expressed in forming somites during early neurogenesis (Begemann et al., 2001; Grandel et al., 2002). RA signaling is required for the formation of r5-7 and the anterior spinal cord (Begemann et al., 2001; Grandel et al., 2002; Niederreither et al., 2000; Shimizu et al., 2006). These data suggest that Fgf and RA signal gradients in which Fgf and RA signals are high at r4 and in the caudal-most hindbrain, respectively, play important roles in specifying neurons in the caudal hindbrain. Although Fgf and RA signals regulate the expression of mafba, knox20 and Hox cluster genes (Ghosh et al., 2018; Marin and Charnay, 2000; Walshe et al., 2002), it is largely unknown which genes function downstream of these signals to regulate neuronal specification.
In mice, the homeobox gene Gsx2 (formerly Gsh2) is expressed in the ventral telencephalon and is involved in fate determination of lateral ganglionic eminence (LGE) progenitor cells, which give rise to striatal neurons and olfactory bulb interneurons (Corbin et al., 2000; Hsieh-Li et al., 1995; Szucsik et al., 1997; Waclaw et al., 2009; Yun et al., 2003). Gsx2, which is also expressed in the ventricular zone of the spinal cord and hindbrain, is involved in the generation of a subset of hindbrain neurons and dorsal interneurons in the spinal cord of mice (Kriks et al., 2005; Mizuguchi et al., 2006; Satou et al., 2013). However, the role of gsx2 in the development of IO neurons has not been elucidated. In this report, we show that gsx2 is co-expressed with ptf1a in neuronal progenitors of IO neurons and is required for the development of IO neurons in zebrafish. gsx2 expression is regulated through transcriptional networks that are activated by Fgf and RA signals in the caudal hindbrain. Our results reveal that gsx2 mediates the rostro-caudal positional signals to control the identity of IO neurons.
gsx2 is co-expressed in IO progenitors that express ptf1a
To reveal the mechanisms that generate IO neurons, we sought genes that are expressed in IO progenitors in the dorso-caudal hindbrain. A previous study with a BAC transgenic fish TgBAC(gsx2:LOXP-Tomato-LOXP-GFP) [hereafter referred to as Tg(gsx2:RFP)] suggested that gsx2 is expressed in the caudal hindbrain (Satou et al., 2013). Thus, we focused on gsx2 and ptf1a, which is also involved in the generation of IO neurons in mice (Yamada et al., 2007). We found that gsx2 was expressed in the dorsal part of the caudal hindbrain and the ventral telencephalon in an early stage [2, 3 days post fertilization (dpf)] of zebrafish larvae (Fig. 1A,B,E,F). As reported previously, ptf1a was expressed in the dorsal ventricular zone of all rhombomeres (Fig. 1C,D,G,H). To examine gsx2-expressing cells in detail, we compared RFP expression in Tg(gsx2:RFP) larvae (gsx2:RFP) with GFP expression in gSAIzGFFM35A;Tg(5xUAS:EGFP) (Fig. 1I-L) and Tg(ptf1a:EGFP) larvae (ptf1a:GFP, Fig. 1M-W). gSAIzGFFM35A is a Gal4 trap line that harbors a Gal4 gene in the exon of mafba (previously known as valentino or kreisler) and can drive UAS-mediated GFP expression at r5 and 6 (referred to as mafbaGFF;UAS:GFP, Fig. 1I,K) (Asakawa and Kawakami, 2018). The gsx2:RFP+ and mafbaGFF;UAS:GFP+ cells did not overlap and were separated adjacently in the caudal hindbrain at an early larval stage (3 dpf, Fig. 1I-L), indicating that gsx2 is specifically expressed in r7 of the hindbrain and the rostral spinal cord. As reported previously (Kani et al., 2010), ptf1a:GFP expression was detected in neural progenitors located in the dorsal ventricular zone of all rhombomeres at an early larval stage (5 dpf, Fig. 1M,N,Q). gsx2:RFP expression was detected in the dorsal ventricular zone of only the caudal hindbrain (Fig. 1M,O,Q). gsx2:RFP and ptf1a:GFP were co-expressed in cells in the dorsal ventricular zone (Fig. 1M,Q). As GFP and RFP (Tomato) are relatively stable, they can be used for lineage tracing in zebrafish (Kani et al., 2010). Although gsx2 and ptf1a were not detected in the ventral hindbrain (Fig. 1E-H), gsx2:RFP and ptf1a:GFP were co-expressed in neurons in the ventral hindbrain in 5-dpf larvae (Fig. 1R-T). The gsx2:RFP+ neurons in the ventral hindbrain projected their axons to the cerebellum (Fig. 1O,P,V,W). To examine whether they are IO neurons, we used the Tg line hspGFFDMC28C;Tg(5xUAS:EGFP) (28C;UAS:GFP), which expresses GFP in IO neurons (Takeuchi et al., 2015). gsx2:RFP and 28C;UAS:GFP were co-expressed in neurons in the ventral part of the caudal hindbrain (Fig. S2). Although there were gsx2:RFP+ 28C;UAS:GFP− cells in this region (Fig. S2), the absence of 28C;UAS:GFP expression may be due to a mosaic expression of UAS:GFP (Akitake et al., 2011). These results indicate that gsx2 is co-expressed with ptf1a in IO progenitors. We also found that gsx2:RFP+ axons from the ventro-lateral part of the caudal hindbrain projected to the ventral part of the cerebellum (Fig. 1O,P). These axons are potentially from the nucleus commissure of Wallenberg, which send mossy fibers that project to GCs, as reported for other teleost species (Xue et al., 2004).
The gsx2 mutation leads to a strong reduction or loss of IO neurons
To reveal the roles of Gsx2 in IO neuronal development, we generated gsx2 mutants using the CRISPR/Cas9 method. The gsx2 mutants harbor a 5 bp (gsx2Δ5) or 8 bp (gsx2Δ8) deletion in exon 1 of the gsx2 gene that introduces a premature stop (Fig. S1). Although gsx2 mRNA was not reduced in the gsx2 mutant larvae (Fig. S3), the Gsx2 proteins of the putative mutants lack the homeodomain (Fig. S1) and the mutations are likely null alleles. In control 5 dpf 28C;UAS:GFP Tg larvae, GFP+ somata were located in the ventral hindbrain and they extended CFs to PCs, which were marked with anti-parvalbumin7 (Pvalb7) antibody (Fig. 2A-D, Fig. S4). In contrast, GFP+ somata were rarely detected and GFP+ CFs were not observed in gsx2 mutants (Fig. 2E-H,M, Fig. S4). In addition to 28C;UAS:GFP, we found that the TgBAC(mnx2b:GFF);Tg(5xUAS:EGFP) line also expressed GFP in a region of IO neurons in addition to motoneurons (mnx2b:GFF;UAS:GFP) (Asakawa et al., 2013) (Fig. 2I,J). mnx2b:GFF;UAS:GFP and gsx2:RFP were co-expressed in neurons in the ventral part of the caudal hindbrain (Fig. S2). Similar to 28C;UAS:GFP+ cells, a strong reduction or loss of mnx2b:GFF;UAS:GFP+ IO neurons was observed in 5 dpf gsx2Δ5 mutant larvae (Fig. 2K,L). We further examined endogenous markers of IO neurons. pou4f1 (also known as brn3a) and a glutamate receptor, metabotropic 5a (grm5a), are expressed in zebrafish IO neurons (Bae et al., 2009; Haug et al., 2013). foxp2 is expressed in IO neurons of mice (Ferland et al., 2003; Fujita and Sugihara, 2012). Whereas the expression of these markers was detected in IO neurons of wild-type larvae, it was not observed in gsx2 mutant larvae (Fig. 2N-S, Fig. S4), indicating a strong reduction or absence of IO neurons in gsx2 mutants. These data reveal that Gsx2 is required for IO neuronal development (Fig. 2T). In addition, mnx2b:GFF;UAS:GFP+ neurons located in the ventro-lateral part of the caudal hindbrain were reduced or absent in gsx2 mutants (Fig. 2I,K).
The ptf1a mutation leads to a strong reduction or loss of IO neurons
We next analyzed the roles of ptf1a in the development of IO neurons by generating ptf1a mutants using the CRISPR/Cas9 method. The ptf1a mutants have a 4 bp (ptf1aΔ4) deletion or 11 bp insertion (ptf1a+11) in exon 1 (Fig. S1). Although the ptf1a mutant larvae retained ptf1a mRNA (Fig. S3), the putative mutant Ptf1a proteins lack the bHLH domain and the mutations were likely null alleles. The ptf1a mutant larvae displayed a complete loss of IO neurons that express GFP in 28C;UAS:GFP or mnx2b:GFF;UAS:GFP Tg lines (Fig. 3E-H,K-M Fig. S4). Furthermore, the IO markers pou4f1, grm5a and foxp2 were not detected in ptf1a mutant larvae (Fig. 3N-S, Fig. S4). These results are consistent with the defective development of IO neurons in Ptf1a mutant mice (Yamada et al., 2007). In addition to IO neurons, the ptf1a mutants showed a reduction or loss of mnx2b:GFF;UA:GFP+ neurons in the ventro-lateral hindbrain as in gsx2 mutants (Fig. 3K,L). Currently, the identity of these neurons is not known, but they are likely derived from gsx2- and ptf1a-expressing neural progenitors. The ptf1a mutants further displayed a reduction – but not loss – of PCs and crest cells, which were positive for Pvalb7, as was shown previously (Bae et al., 2009) (Fig. 3E, Fig. S5). These data suggest that, even though PCs, crest cells and IO neurons are derived from ptf1a-expressing progenitors, ptf1a functions redundantly with other genes during the development of PCs and crest cells, while ptf1a is essential for the development of IO neurons in zebrafish (Fig. 3T).
Non-interdependent regulation and distinct roles of gsx2 and ptf1a in IO progenitors
We examined the relationship between gsx2 and ptf1a in IO neuronal development. The expression of ptf1a and gsx2 was not affected in gsx2 and ptf1a mutant larvae, respectively, during the development of IO neurons (Fig. 4). These results indicate that ptf1a and gsx2 are regulated independently of each other in the caudal hindbrain. We further analyzed apoptosis by immunohistochemistry with an anti-cleaved caspase 3 antibody as Ptf1a mouse mutants show an increase in apoptotic cells in the caudal hindbrain (Yamada et al., 2007). As in Ptf1a mutant mice, the ptf1a zebrafish mutant larvae showed an increase in the number of cleaved-caspase 3+ cells (Fig. 5A-H,Q). In contrast, there was no significant increase in cleaved-caspase 3+ cells in gsx2 mutant larvae (Fig. 5I-P,R). Furthermore, gsx2:RFP expression was retained in cells migrating from the dorsal progenitor domain in gsx2 mutant larvae (Fig. S6), suggesting that gsx2+ progenitors remained undifferentiated, or differentiated into other neurons in gsx2 mutants. Our findings indicate that ptf1a and gsx2 play different roles in IO neuronal development.
To further address the cooperative role between ptf1a and gsx2 in IO neuronal development, we expressed gsx2 in ptf1a-expressing progenitors by using ptf1a:Gal4-VP16 and UAS:gsx2 Tg lines (Fig. 6B,C,E,G). However, the ectopic co-expression of ptf1a and gsx2 did not induce ectopic IO neurons in the rostral hindbrain. These data indicate that co-expression of ptf1a and gsx2 is not sufficient to induce IO neurons in the hindbrain.
gsx2 expression is negatively regulated by Fgf signaling and Mafba
Gradients of Fgf and RA signals are involved in the formation and patterning of the caudal hindbrain (Dupé and Lumsden, 2001; Gavalas and Krumlauf, 2000; Marin and Charnay, 2000; Maves et al., 2002). We inhibited Fgf and RA signals in order to investigate the regulation of gsx2 expression. The inhibition of fgf8a and fgf3 by injection of antisense morpholino (MO) into fgf8a mutants (fgf8ati282a/ti282a;fgf3MO) significantly increased gsx2:RFP expression and the number of 28C;UAS:GFP+ IO neurons, compared with control MO-injected wild-type larvae (Fig. 7A-D,I,K-P,W). Consistent with this finding, inhibition of the Fgf signal by SU5402, which is a chemical inhibitor of Fgf receptor tyrosine kinases (Mohammadi et al., 1997), led to an expansion of gsx2:RFP expression and an increase in the number of 28C;UAS:GFP+ IO neurons (Fig. 7E-H,J,Q-V,X). Stronger Fgf inhibition by SU5402 correlated with an increase in gsx2:RFP expression and IO neurons (Fig. S7). Our findings suggest that the Fgf signal suppressed rostral expansion of gsx2 expression, thereby limiting the gsx2-expressing IO progenitor domain to r7. The expansion of gsx2 expression by inhibition of the Fgf signal likely led to the expansion of IO progenitors and a subsequent increase in IO neurons.
mafbaGFF;UAS:GFP expression was reduced when the Fgf signal was inhibited (Fig. 7G,H). We further analyzed the roles of mafba in gsx2 expression and in the development of IO neurons (Fig. 8). In mafbaGFF/GFF mutant larvae (Asakawa and Kawakami, 2018), rostral gsx2:RFP expression expanded slightly compared with control heterozygotes, and gsx2:RFP+ cells were observed in the mafbaGFF;UASGFP+ region, corresponding to r5/6 (Fig. 8D-F,O). Furthermore, 28C;UAS:GFP+ IO neurons increased significantly in mafbab337/b337 mutant larvae (Fig. 8K-N,P). These data suggest that Mafba is at least partly involved in the Fgf signal-mediated suppression of gsx2 expression and IO neuronal development.
gsx2 expression is positively regulated by the RA-Hox cascade
We inhibited the RA signal by a MO against aldh1a2, which recapitulates the phenotypes of an aldh1a2 mutant (Begemann et al., 2001), or an Aldh1a2 inhibitor diethylaminobezaldehyde (DEAB) (Russo, 1997). Inhibition of the RA signal with both MO and DEAB resulted in reduced gsx2:RFP expression but increased mafbaGFF;UAS:GFP expression (r5/6, Fig. 9A-L). It also reduced the number of 28C;UAS:GFP+ IO neurons (Fig. 9M-T,Y,Z) and expression of the IO neuronal marker pou4f1 (Fig. 9U-X). The reduction was stronger in DEAB-treated larvae than in aldh1a2 morphants, consistent with a previous report showing that DEAB-treated larvae showed more severe phenotypes than aldh1a2 mutants (Begemann et al., 2004). Our data indicate that the RA signal positively controls gsx2 expression in the caudal hindbrain, and is also involved in IO neuronal development. The reduction in gsx2 expression by inhibition of the RA signal likely led to the reduction of IO progenitors and the subsequent reduction of IO neurons.
The RA signal controls the expression of Hox genes in the caudal hindbrain (Ghosh et al., 2018; Marin and Charnay, 2000). To address the roles of Hox genes in IO neuronal development, we examined gsx2 expression and IO neurons in larvae defective of Pbx2 and Pbx4, which are co-factors in the rostral Hox protein (Waskiewicz et al., 2002). The pbx2/pbx4 morphant larvae recapitulate the phenotypes of maternal zygotic pbx2 mutants in which pbx4 inhibition was mediated by MO, and in which r2-6 acquired the r1 state (Maves et al., 2009). In pbx2/4 morphant larvae, ectopic PCs were observed caudally but crest cells were present normally (Fig. 10G). In contrast, gsx2:RFP expression, the number of 28C;UAS:GFP+ IO neurons (determined by their location in the caudal hindbrain and their axons) and pou4f1 expression were strongly reduced or absent in pbx2/4-deficient larvae (Fig. 10C,D,G,H,J,K,L). To address the roles of Hox genes, we further ectopically expressed hoxb4a as it is expressed in r7 and in the spinal cord. Injection of hoxb4a mRNA expanded gsx2:RFP expression and increased 28C;UAS:GFP+ IO neurons in the caudal hindbrain (Fig. 10O-R). Furthermore, injection of hoxb4a mRNA at least partly rescued gsx2:RFP expression and 28C;UAS:GFP+ IO neurons in aldh1a2 morphants or DEAB-treated larvae (Fig. 10S-V). Although hoxb4a might not be the only Hox gene regulating caudal-most hindbrain identity, our data suggest that the RA-Hox cascade plays a major role in controlling gsx2 expression and subsequent IO neuronal development.
Expression of gsx2 and ptf1a in IO progenitors located in rhombomere 7
gsx2 and ptf1a were expressed in the dorsal ventricular zone of r7 (Fig. 1). Tracing with the Tg lines revealed that gsx2- and ptf1a-expressing cells (gsx2:RFP+ and ptf1a:GFP+ cells) gave rise to IO neurons (Fig. 1). Consistent with this, single-cell RNA sequencing analysis revealed that gsx2 and ptf1a were co-expressed in dorsomedial progenitor cells of the hindbrain (Tambalo et al., 2020). Furthermore, the gsx2 and ptf1a mutations abrogated IO neuronal development (Figs 2 and 3). These data indicate that gsx2/ptf1a-expressing cells are the neural progenitors of IO neurons. Previous lineage tracing with the Cre-loxP system in mice revealed that IO neurons were derived from Ptf1a-expressing neural progenitors (Yamada et al., 2007). Avian grafting studies suggested that IO neural progenitors were located in r6-8 (Ambrosiani et al., 1996; Cambronero and Puelles, 2000; Kawauchi et al., 2006). Our data clearly indicate that, in zebrafish, IO neurons are derived from gsx2/ptf1a-expressing neural progenitors in r7.
Role of Ptf1a in the development of hindbrain neurons
ptf1a (Ptf1a in mice) is expressed in neural progenitors in the ventricular zone of the entire dorsal hindbrain in both mice and zebrafish (Fujiyama et al., 2009; Hoshino et al., 2005; Kani et al., 2010; Yamada et al., 2007). Zebrafish ptf1a mutants showed a reduction in the number of PCs and crest cells, and the loss of IO neurons (Fig. 3, Fig. S5). In mice, Ptf1a mutants showed a complete loss of PCs (and other cerebellar GABAergic interneurons), DCN neurons and IO neurons (Fujiyama et al., 2009; Hoshino et al., 2005; Yamada et al., 2007). Although the mice and zebrafish phenotypes are slightly different, our data and those of others suggest that ptf1a-expressing progenitors give rise to PCs and IO neurons in both species, and also contribute to the development of DCN neurons in mice and of crest cells in zebrafish (Fig. 11).
A milder reduction of PCs in zebrafish suggest that other genes compensate for the lack of Ptf1a. In mice, bHLH-type proneural genes neurogenin 1 and 2 (Neurog1 and Neurog2) and Ascl1 are expressed in the ventricular zone of the dorsal hindbrain (Kim et al., 2008; Landsberg et al., 2005; Lundell et al., 2009; Zordan et al., 2008). In mice, Neurog2 mutants show a reduction in number and defective dendritogenesis of PCs (Florio et al., 2012), while the Ascl1 mutation leads to a reduction of cerebellar GABAergic interneurons and oligodendrocytes (Grimaldi et al., 2009; Sudarov et al., 2011), suggesting the involvement of other proneural genes in the development of PCs and other interneurons in the cerebellum. Zebrafish has neurog1, ascl1a and ascl1b but not neurog2 (ZFIN, https://zfin.org). Some of these proneural genes may redundantly function with ptf1a in the development of PCs. Unlike PCs and crest cells, IO neurons were completely absent in ptf1a mutants (Fig. 3), indicating that Ptf1a is indispensable for IO neuronal development in both mice and zebrafish.
In addition to PCs, crest cells in MONs were reduced in ptf1a mutant larvae (Fig. S5). In mice, inhibitory neurons in DCNs are derived from Ptf1a-expressing neural progenitors in the r2-5 ventricular zone (Fujiyama et al., 2009). Although zebrafish do not have DCNs, there is a similarity between DCN and MON: they both receive input from hair cells (in the cochlear organ in mice or via the lateral line system in teleosts) (Yamamoto and Ito, 2005). The DCN and MON have a neural-circuit structure similar to the cerebellum and are thus called cerebellum-like structures (Bell et al., 2008; Oertel and Young, 2004). Although crest cells are excitatory, our data suggest that both crest cells and DCN inhibitory neurons are derived from ptf1a-expressing progenitors. It is tempting to speculate that the DCN and MON were derived from a common ancestral origin during evolution.
Role of Gsx2 in the development of IO neurons
The gsx2 mutants showed a strong reduction or loss of IO neurons (Fig. 2), indicating that Gsx2 is required for IO neuronal development. Previous studies in mice showed that Gsx2 is expressed in the ventricular zone of the ventral telencephalon, hindbrain and spinal cord, and that it is required for the development of striatal neurons in the telencephalon, olfactory interneurons in the olfactory bulb and dorsal interneurons in the spinal cord (Corbin et al., 2000; Kriks et al., 2005; Szucsik et al., 1997). Therefore, although Gsx2 is involved in the development of different types of neurons, it plays a similar role in the specification of neurons from their progenitors in the ventricular zone of these brain regions.
In some gsx2 mutants, a small number of IO neurons remained, and projected CFs to PCs (Fig. 2), suggesting the presence of genes that redundantly function with gsx2 in IO progenitors. gsx1, which is a homologue of gsx2, is expressed in the hindbrain (r1-7) and spinal cord of zebrafish (Begemann et al., 2001). In mice, Gsx1 and Gsx2 redundantly function in LGE patterning and in the specification of dorsal interneurons of the spinal cord (Mizuguchi et al., 2006; Yun et al., 2003). Although the expression domains of gsx1 and gsx2 are different in the spinal cord of zebrafish (Satou et al., 2013), it is possible that gsx1 is co-expressed with gsx2 and cooperatively functions in the development of some IO neurons. A previous study showed the presence of IOs in the hindbrain of Gsx2 mutant mice, although it is not clear whether IO neurons were reduced (Szucsik et al., 1997), suggesting that the IO phenotype of Gsx2 mouse mutants are milder than in the phenotype of the gsx2 zebrafish mutant. Expansion of Gsx1 expression in Gsx2 mutants was reported for the development of mouse LGE (Toresson and Campbell, 2001). Gsx1-mediated compensation of Gsx2 deficiency might function at lower levels in zebrafish than in mice. Future studies on gsx1 expression and gsx1 mutants will clarify this issue.
Different roles of Ptf1a and Gsx2 in IO neuronal development
The expression of ptf1a and gsx2 in the caudal hindbrain was independent during early neurogenesis (Fig. 4). Whereas the ptf1a mutants showed an increase in apoptosis in the caudal hindbrain, the gsx2 mutants did not (Fig. 5), suggesting different roles of ptf1a and gsx2 in IO neuronal development. The phenotype of zebrafish ptf1a mutants is consistent with that of mouse Ptf1a mutants, which also showed increased apoptosis in the IO region (Yamada et al., 2007). These data imply that a major population of IO progenitors that did not differentiate to neurons died by apoptosis in the absence of ptf1a in both mice and zebrafish. However, some cells derived from Ptf1a-expressing progenitors (Ptf1a-lineage cells) become precerebellar neurons that send mossy fibers in Ptf1a mutant mice (Yamada et al., 2007), suggesting a role of Ptf1a in fate determination of IO neurons in mice. However, in zebrafish, the ptf1a/gsx2-lineage cells (ptf1a:GFP+ and gsx2:RFP+ cells) were present in the dorsal hindbrain and did not migrate ventrally to where mossy fiber neurons are supposed to be located in ptf1a mutants (Figs S5 and S6). Therefore, even if some cells derived from IO progenitors undergo a change in fate, they are likely a small population in zebrafish.
In contrast to the ptf1a mutants, gsx2-lineage cells migrated ventrally from the ventricular zone and some of them differentiated to neurons that extended axons to the cerebellum region in gsx2 mutants (Fig. S6). These neurons are potentially the nucleus commissure of Wallenberg. However, these fibers were also reduced in gsx2 mutants (Fig. S6). In addition, mnx2b:GFF;UAS:GFP+ neurons located laterally to IOs were reduced in gsx2 and ptf1a mutants (Figs 2 and 3). These data suggest that the development of these neurons depends on gsx2. We currently do not know what type(s) of neurons these gsx2-lineage cells were differentiated to in the absence of Gsx2. Nevertheless, our findings suggest that ptf1a is involved in neuronal differentiation, whereas gsx2 is involved in the specification of IO neurons. Ptf1a and Gsx2 cooperatively control the development of IO neurons. Ascl1 and Gsx1/2 coordinately control the specification of dorsal glutamatergic sensory neurons in mice (Mizuguchi et al., 2006). Thus, cooperation between proneural gene(s) and gsx-family genes is a common mechanism that coordinately controls the differentiation and specification of a subset of neurons from their progenitors.
We further demonstrated that the co-expression of ptf1a and gsx2 was not sufficient to induce IO neurons (Fig. 6). Other factors may function with Ptf1a and Gsx2 in IO neuronal development. One of the candidate co-factors is the bHLH-type transcription factor Olig3, which cooperates with Ptf1a in IO neuronal development in mice (Storm et al., 2009). Future studies will reveal a set of transcription factors that sufficiently initiates the genetic program of IO neuronal development.
gsx2 mediates Fgf and RA signals to specify IO neuronal fate
Inhibition of the signal of fgf3 and fgf8a, which are known to be expressed in r4 and are involved in the specification of r5/6 (Maves et al., 2002; Walshe et al., 2002; Wiellette and Sive, 2004), led to reduced mafba expression, an expansion of gsx2 expression, and an increase in the number of IO neurons (Fig. 7). The mafba mutants showed a slight increase in gsx2 expression and in the number of IO neurons (Fig. 8). Fgf signal-mediated mafba expression has previously been reported for chicken (Marin and Charnay, 2000) and zebrafish (Ghosh et al., 2018). Collectively, these data indicate that the Fgf signal suppresses rostral expansion of gsx2 expression in the caudal hindbrain and thereby limits the IO progenitor domain to r7; Mafba is at least partly involved in this regulation (Fig. 11). Fgf8a is involved in the formation of the cerebellum (Reifers et al., 1998). In fgf3/fgf8a-defective larvae, the number of IO neurons increased but PCs were absent (Fig. 7). Intriguingly, IO neurons project their axons rostrally to the optic tectum region even in the absence of PCs, which are the targets of CFs (Fig. 7), indicating the presence of a PC-independent axon guidance mechanism for CFs. Future studies will clarify this mechanism. As MafB directly activates hoxb3 expression in mice (Manzanares et al., 1997), the mafba-hoxb3 cascade may play a role in repressing gsx2 expression and IO neuronal identity.
During early neurogenesis, aldh1a2 is expressed in forming somites in mouse, chicken and zebrafish (Begemann et al., 2001; Berggren et al., 1999; Blentic et al., 2003; Grandel et al., 2002; Niederreither et al., 1997). The treatment of mouse embryos with RA increased IO neurons (Yamamoto et al., 2005) but defective RA signals led to a reduced caudal hindbrain in zebrafish (Begemann et al., 2004, 2001; Grandel et al., 2002). We showed that the inhibition of RA signals resulted in reduced gsx2 expression and IO neuronal development (Fig. 9). These data indicate that the RA signal positively controls gsx2 expression and, subsequently, IO neuronal development. Hox genes function downstream of RA signals (reviewed by Nolte et al., 2019). Defects in the RA signal in zebrafish result in a strong reduction or absence of hoxb4a in the caudal hindbrain (Begemann et al., 2004, 2001; Grandel et al., 2002). hoxb4a enhancer-driven YFP expression was detected in IO neurons (Ma et al., 2009; Punnamoottil et al., 2008). hoxb4 is a direct target of the RA receptor RAR/RXR complex in chickens (Gould et al., 1998). These data suggest that the RA signal controls caudal hindbrain identity through the expression of Hox genes, including hoxb4a, in the caudal hindbrain. As Hox paralogues function redundantly, it is difficult to address the roles of individual Hox genes. We demonstrated that the loss of Pbx2 and Pbx4, which function as co-factors for rostral Hox proteins harboring a YPWM motif (Johnson et al., 1995; Mann and Chan, 1996; Shanmugam et al., 1997), led to a strong reduction of gsx2 expression and IO neurons (Fig. 10). Furthermore, misexpression of Hoxb4a, which harbors a YPWM motif, elicited the expansion of gsx2 expression (Fig. 10). Expression of Hoxb4a at least partly rescued gsx2:RFP expression and 28C;UAS:GFP+ IO neurons in the aldh1a2 morphant or in DEAB-treated larvae (Fig. 10). Our findings suggest that r7-expressed Hox genes, such as hoxb4a, function directly downstream of the RA signal to control gsx2 expression and IO neuronal development. pbx2/pbx4-deficient larvae show defects in the function of most rostral Hox proteins and the loss of r2-6 identity (Waskiewicz et al., 2002). Although the number of PCs increased, the number of crest cells was normal in pbx2/pbx4-deficient larvae (Fig. 10), suggesting that the specification of PCs and crest cells does not rely on Pbx-interacting rostral Hox proteins. The RA-Hox cascade controls the specification of IO neurons from ptf1a-expressing progenitors by regulating gsx2 expression (Fig. 11). Our study did not address whether MafBa and/or Hox proteins directly regulate gsx2 expression in the caudal hindbrain. The zebrafish gsx2 gene has an ∼800 bp element on the 3′ side of the open reading frame that displays partial sequence homology with the mouse Gsx2 enhancer that can drive transgene expression in the mouse ventral telencephalon and caudal hindbrain (K.S., unpublished). Although the zebrafish element contains putative binding motifs of MafBa and Hox proteins, further analysis is required to understand the regulation of gsx2 expression in the caudal hindbrain.
ptf1a-expressing neural progenitors give rise to three different neuron PCs, crest cells and IO neurons following positional signals. Gsx2 mediates Fgf and RA positional signals to specify IO neurons. Future studies will reveal genetic programs for the specification of PCs and crest cells. The identification of targets for Gsx2 and Ptf1a will clarify the mechanisms underlying the development and functionality of IO neurons.
MATERIALS AND METHODS
The animal work in this study was approved by the Nagoya University Animal Experiment Committee and was conducted in accordance with the Regulations on Animal Experiments at Nagoya University. Wild-type zebrafish with the Oregon AB genetic background were used. For immunohistochemistry and whole-mount in situ hybridization, larvae were treated with 0.003% 1-phenyl-2-thiourea (PTU) to inhibit the formation of pigmentation. Tg(UAS:EGFP) (nkuasgfp1aTg), Tg(ptf1a:EGFP), TgBAC(mnx2b:GFF), TgBAC(gsx2:LOXP-Tomato-LOXP-GFP) and TgBAC(ptf1a:GAL4-VP16) have been previously reported and characterized (Asakawa et al., 2013, 2008; Parsons et al., 2009; Pisharath et al., 2007; Satou et al., 2013). hspGFFDMC28C is a Gal4 trap line that drives UAS-dependent expression in IO neurons (Takeuchi et al., 2015). gSAlzGFFM35A is a mafba mutant in which the mafba exon is disrupted by the Gal4 gene trap, and is also known as mafbankgsaizgffm35aGt (Asakawa and Kawakami, 2018). The Tg(UAS-hsp70l;gsx2BLRP-P2A-BirA-P2A-mCherry) line, which harbors a transgene containing gsx2, biotin ligase recognition peptide (BLRP), 2A peptide sequences of porcine teschovirus-1 (P2A), biotin ligase (BirA) and mCherry genes, will be described elsewhere. The mafbab337 and fgf8ati282a mutants, known previously as valentino and acerebellar, respectively, have been described elsewhere (Moens et al., 1998; Reifers et al., 1998). The allele names of the gsx2Δ5, gsx2Δ8, ptf1aΔ4 and ptf1a+11 mutants established in this study are designated as gsx2nub32, gsx2nub33, ptf1anub34 and ptf1anub35, respectively, in ZFIN (https://zfin.org). The zebrafish were maintained at 28°C under a 14 h light and 10 h dark cycle. Embryos and larvae were maintained in embryonic medium (EM) (Westerfield, 2000).
Establishment of gsx2 and ptf1a mutants by the CRISPR/Cas9 system
The gRNA targets were designed by the web software ZiFit Targeter 4.2 (http://zifit.partners.org/ZiFiT/) (Hwang et al., 2013; Mali et al., 2013). To generate gRNAs, the following oligonucleotides were used: 5′-TAGGCGGAATTCCACTGCTCAA-3′ and 5′-AAACTTGAGCAGTGGAATTCCG-3′ for gsx2Δ5; 5′-TAGGTCTTCCGGGACGGGCAGA-3′ and 5′-AAACTCTGCCCGTCCCGGAAG-3′ for gsx2Δ8; 5′-TAGGAAGAGGCGGAGGCGCATG-3′ and 5′-AAACCATGCGCCTCCGCCTCTT-3′ for ptf1aΔ4; and 5′-TAGGCGTCAAGCTGCCAACGTC-3′ and 5′-AAACGACGTTGGCAGCTTGACG-3′ for ptf1a+11. gRNA and Cas9 mRNA syntheses were performed as previously reported (Nimura et al., 2019). A solution containing 25 ng/µl gRNA and 100 ng/µl Cas9 mRNA or 1000 ng/µl Cas9 protein (Toolgen) was injected into one-cell-stage embryos using a pneumatic microinjector (PV830, WPI). The insertion and/or deletion (indel) mutations on the target region were detected by a heteroduplex mobility assay (Ota et al., 2013). The mutations were confirmed by sequencing after subcloning the target regions amplified from the mutant genome into pTAC-2 (BioDynamics Laboratory).
To detect indel mutations, the following primers were used: 5′-GTGCGTATCCTCACACATCCACTCT-3′ and 5′-TGCCATCCTCTGGCAGAACG-3′ to detect the gsx2Δ5 mutation; 5′-CATTGGCATGCACTCTCCCG-3′ and 5′-TGAGGATACGCACAGCGGAC-3′ to detect the gsx2Δ8 mutation; 5′-ACCTCAGAGCTGTCCCCTCACAGA-3′ and 5′-GGCAGCTTGACGCAACTGTT-3′ to detect the ptf1aΔ4 mutation; and 5′-GGAGGCGCATGAGGTCTGAAGT-3′ and 5′-GCAGTCCCTCGAAAGCATCG-3′ to detect the ptf1a+11 mutation. To identify gSAIzGFFM35A fish, 5′-CGAGGTAGGAGAAGGGCTGT-3′ and 5′-CTGGAGCGTTTGATGGATACAG-3′ primers were used to amplify about 200-bp PCR products from wild-type but not gSAIzGFFM35A genomic DNA. As Tg(gsx2:loxp-DsRed-loxp-GFP) are BAC Tg fish and contain a region of the gsx2 exon1, we could not distinguish the wild-type allele of gsx2 from the endogenous locus or the transgene in this fish with the primers described above. To detect wild-type and mutant alleles of gsx2 in gsx2Δ5;Tg(gsx2: LOXP-Tomato-LOXP-GFP) fish, the following primers were used: 5′-CCATCAGCATCTCGCTCAG-3′ and 5′-CAGTGAAGCCTTGTCCCTCG-3′. After PCR amplification, PCR products were digested with XhoI. The wild-type gsx2 allele gave rise to two bands, 21 and 257 bp fragments, while the mutant allele displayed a single band as it cannot be digested with XhoI. To detect the fgf8ati282a mutation, 5′-CAGGAGGGGGAAACTGATTGTCTAG-3′ and 5′-CCCTTTCTAGGTGGGATTCTTCTC-3′ primers were used. After PCR amplification, PCR products were digested with XbaI. The wild-type PCR products could not be digested, whereas the mutant PCR products gave rise to 130 and 20 bp DNA fragments.
Injection of antisense morpholino oligonucleotides
fgf3, aldh1a2, pbx2 and pbx4 MOs were translation-blocking antisense MOs that were previously reported (Begemann et al., 2001; Maves et al., 2007; Wiellette and Sive, 2004). 0.3 mM fgf3 MO, 0.1 mM aldh1a2 MO, 0.03 mM pbx2 MO2, 0.06 mM pbx2 MO3, 0.06 mM pbx4 MO1 or 0.06 mM pbx4 MO2 (1 nl) were injected into one-cell-stage embryos. As a control, we used a standard control MO: 5′-CCTCTTACCTCAGTTACAATTTATA-3′. These MOs were obtained from Gene Tool.
Treatment with chemical inhibitors
DEAB (Wako) and SU5402 (Calbiochem) were dissolved in DMSO at 100 mM and 20 mM, respectively. Embryos were treated with 0.25 µM DEAB or 1.0, 2.5, 5.0 and 200 µM SU5402 in 0.003% PTU/EM from 6 to 22 h post fertilization (hpf).
Total RNA was isolated from 5 or 12 dpf zebrafish larvae using TRI Reagent (Molecular Research Center). cDNA was generated using ReverTra Ace (Toyobo). DNA fragments for gsx2 or foxp2 were amplified from the 5-dpf cDNA library using the primers 5′-GACTCTTTGATTATCAAGGATCCCG-3′ (F) and 5′-CGTCTTCTGAGCGCGGATAAT-3′ (R) for gsx2, and 5′-CCATGGAGGATAATGGGATG-3′ (F) and 5′-TGAGGTAAATTTGGGGGTGA-3′ (R) for foxp2, and subcloned to pGEMT-easy (Promega) and pTA2 (Toyobo), respectively (pGEMTe-gsx2 and pTA2-foxp2). Three DNA fragments for grm5a (grm5a-A, grm5a-B and grm5a-C) were amplified from zebrafish genomic DNA with the primers 5′-CACTTTTCTCCGTCCACCAT-3′ (F) and 5′-GCGATCTGGGGAATATTGAA-3′ (R) for grm5a-A, 5′-ACCTTCAGTGGGGAGATCCT-3′ (F) and 5′-GATGAAGAGTGCCACGATGA-3′ (R) for grm5a-B, and 5′-CGGCCATTATCAAACCATTC-3′ (F) and 5′-TGACGGTAGGATGGTGAACA-3′ (R) for grm5a-C, and subcloned into pTAC2 (pTAC2-grm5a-A/B/C). Three DNA fragments for pou4f1 (pou4f1-A, pou4f1-B and pou4f1-C) were amplified from zebrafish genomic DNA with the primers 5′- GCGATGAGCTGAGATGAGAG-3′ (F) and 5′-AGCTCGAGTGCAGAGTTGTG-3′ (R) for pou4f1-A, 5′-GGGTAAGAGTCACCCGTTCA-3′ (F) and 5′-AGTCCGTTGTTGACGAGTCC-3′ (R) for pou4f1-B, and 5′- CAATTAACGACTCGGACACG-3′ (F) and 5′-TCAGCTATAGCCGCGATTTT-3′ for pou4f1-C, and subcloned to pTAC2 (pTAC2-pou4f1-A/B/C). The full-length cDNA for hoxb4a was amplified from the 12 hpf cDNA library with the following primers: 5′-GGAATTCATGGCCATGAGTTCCTATTT-3′ (F) and 5′-GCTCTAGACTATAGACTTGGCGGAGGTCC-3′ (R). The DNA fragment was digested with EcoRI and XbaI, and subcloned into pCS2+MT (pCS2+MT-hoxb4a).
In situ hybridization
Whole mount in situ hybridization was performed as previously reported (Bae et al., 2009). Detection of ptf1a and pou4f1 riboprobes has been described previously (Bae et al., 2009; Kani et al., 2010). A digoxigenin (DIG)-labeled riboprobe was made from pTAC2 -grm5a-A/B/C, pTA2-foxp2, pGEMTe-gsx2 and pTAC2-pou4f1-A/B/C, using SP6, T3 or T7 RNA polymerase after digestion with restriction enzymes. For grm5a and pou4f1, a mixture of three riboprobes was used (grm5a in Figs 2 and 3; pou4f1 in Fig. S3). To detect transcripts in the IO neurons of 5 dpf larvae, the head regions rostral to the spinal cord of larvae were cut and used for in situ hybridization. To stain larval heads, they were hybridized overnight at 65°C for 72 h and incubated overnight with 1/2000 alkaline-phosphatase conjugated with anti-DIG Fab fragment (Roche) at 4°C for 3 nights. BM-purple signals were acquired using an AxioPlan-2 microscope equipped with an AxioCam CCD camera (Zeiss).
Immunohistochemistry and counting IO neurons
For immunohistochemistry, anti-parvalbumin7 (1/1000, mouse monoclonal, ascites) (Bae et al., 2009), anti-GFP (1/1000, rat, Nacalai Tesque, Cat# 04404-84), anti-DsRed (1/1000, rabbit, Clontech Laboratories, 632496) and anti-cleaved caspase 3 (1/500, rabbit, Cell Signaling Technology, 9661) were used. Alexa Fluor 488 goat anti-rat IgG (H+L, Thermo Fisher Scientific, A-11006), CF488A goat anti-rat IgG (H+L, Biotium, 20023), CF488A goat anti-mouse IgG (H+L, Biotium, 20018-1), CF568 goat anti-mouse IgG (H+L, Biotium, 20301-1) and CF568 goat anti-rabbit IgG (H+L, Biotium, 20103) were used as the secondary antibodies. Larvae were immunostained as described previously (Bae et al., 2009). Some fixed samples were optically cleared with SeeDB reagent as previously reported (Ke et al., 2013; Ke and Imai, 2014). An LSM700 confocal laser-scanning microscope was used to obtain fluorescence images. To count IO neurons, confocal optical sections of the ventral region of the caudal hindbrain (120 μm×120 μm×52 μm) were obtained from immunostained hspGFFDMC28C;Tg(5xUAS:EGFP) larvae. As 28C;UAS:GFP+ cells that were located in this region extended their axons rostrally, they were considered as IO neurons and the number of GFP+ cells were counted manually.
To make hoxb4a mRNA for expression, pCS2+MT-hoxb4a was linearized with NotI and transcribed with SP6 RNA polymerase in the presence of a G(5′)ppp(5′)G RNA cap structure analog (New England BioLabs). One-cell-stage embryos were injected with 25 pg of hoxb4a mRNA.
Data were analyzed for statistical significance using Student's t-test, Welch's t-test and one-way ANOVA using Graphpad PRISM (ver. 5.1).
We thank Shin-ichi Higashijima and the National Bioresource Project for providing the transgenic zebrafish, Makoto Kobayashi for the mafbab337 mutant, Masato Kinoshita and Feng Zhang for the hSpCas9 plasmids, and Kuniyo Kondoh and Yumiko Takayanagi for managing fish mating and care. We also thank the members of the Hibi Laboratory, Hanako Hagio and Naoyuki Yamamoto, for helpful discussion.
Conceptualization: M.H.; Methodology: T.I., M.T., M.S., M.H.; Formal analysis: T.I.; Investigation: T.I.; Resources: K.A., K.S., K.K.; Data curation: T.S., M.H.; Writing - original draft: T.I., M.H.; Writing - review & editing: M.H.; Supervision: M.H.; Funding acquisition: M.T., T.S., M.H.
This work was supported by the Japan Society for the Promotion of Science KAKENHI (JP15H04376 and JP18H02448 to M.H., JP18K06333 to T.S., and JP16K18548 to. M.T.), Core Research for Evolutional Science and Technology Japan Science and Technology Agency (JPMJCR1753 to M.H.), the National Bioresource Project, National Bioresource Project/Genome Information Upgrading Program, and National Bioresource Project/Fundamental Technologies Upgrading Program from Japan Agency for Medical Research and Development (19km0210087j0003, 18km0210151j0001 and 19km0210168j0001 to K.K.)
Peer review history
The peer review history is available online at https://dev.biologists.org/lookup/doi/10.1242/dev.190603.reviewer-comments.pdf
The authors declare no competing or financial interests.