The zinc-finger transcription factor GLI3 is a regulator of precerebellar neuronal migration

The hindbrain precerebellar system is divided into the mossy fiber (MF) and the climbing fiber (CF) system, based on the type of afferent connection with the cerebellum. Projections of the MF system terminate in cerebellar glomeruli where they form synapses with cerebellar granule cells and Golgi cells. CFs terminate on the dendrites of Purkinje cells. Both, MF and CF fibers also form collateral projections to the deep cerebellar nuclei (Paxinos, 2004). Anatomically, the hindbrain precerebellar system comprises bilaterally symmetrical pairs of nuclei situated in the hindbrain. There are four major MF nuclei: the external cuneate nuclei (ECN) and the lateral reticular nuclei (LRN) in the caudolateral hindbrain; and the pontine nuclei (PN) and the reticulotegmental nuclei (RTN) in the rostroventral hindbrain. The inferior olivary nuclei (ION) in the caudoventral hindbrain are the one pair of CF nuclei (Watson, 2012, schematic in Fig. 1).

Hindbrain precerebellar neurons originate from the rhombic lip (RL), a neuroepithelium in the dorsalmost part of the hindbrain that rims the opening of the fourth ventricle (Ray and Dymecki, 2009; Wingate, 2001). The RL extends along the entire rostrocaudal axis of the hindbrain, from rhombomere 1 to rhombomere 8 (r1-r8) (Liu et al., 2010; Ray and Dymecki, 2009; Storm et al., 2008). Anatomically, it can be divided into the upper RL (uRL in r1) and the lower RL (lRL in r2-r8). The uRL contains the progenitors of cerebellar granule neurons, unipolar brush cells and a subset of deep nuclei neurons (Englund et al., 2006; Machold and Fishell, 2005; Wang et al., 2005). The lRL is further subdivided: progenitors at the r2-r5 level contribute neurons to the cochlear nucleus, whereas the caudal lRL (clRL, r6-r8) gives rise to precerebellar MF neurons (MFNs) and CF neurons (CFNs) (Farago et al., 2006; Ray and Dymecki, 2009).

In the murine hindbrain, MFNs and CFNs emanate from the clRL between embryonic day (E) 10 and E16, and undergo a long-distance migration to reach their final position. MFNs migrate just beneath the pial surface in the anterior and posterior extramural stream (AES and PES). The AES turns rostrally after an initial step of ventrally oriented migration, and moves through r5/r4, passing the vestibulocochlear and facial nerve roots. Once the AES reaches the trigeminal nerve root, it turns ventrally and the majority of AES neurons stop at the ipsilateral side of the ventral midline to form the PN/RTN. Neurons in the PES migrate ventrally, cross the midline and move dorsally to establish the LRN and ECN on the contralateral site. CFNs migrate in an intramural migrating stream (IMS), which is directed ventrally and eventually forms the ION (schematic in Fig. S1). Most CFNs stop at the ventral midline and their cell bodies remain ipsilateral to their progenitor domain. CFNs are the first to reach the ventral midline (around E13.5). The last neurons to settle are the MFNs that form the PN (up to E18.5) (Bloch-Gallego et al., 2005; Kratochwil et al., 2017; Shinohara et al., 2013). The distinctive migratory routes of these neurons are severely altered in the absence of guidance factors such as SLITs, netrin and the chemokine CXCL12 (C-X-C motif chemokine 12) (Di Meglio et al., 2008; Dominici et al., 2018; Geisen et al., 2008; Gilthorpe et al., 2002; Kim and Ackerman, 2011; Kuwako et al., 2010; Marillat et al., 2004; Moreno-Bravo et al., 2018; Zhu et al., 2009). However, it is not completely understood how the multi-step migration process of the precerebellar neurons is regulated.

Here, we show that the zinc-finger transcription factor GLI3, a transcriptional repressor in the Sonic hedgehog signaling pathway, plays an important role in maintaining the organization of the precerebellar migratory streams and the final positioning of precerebellar neurons. In particular, GLI3 is essential for the rostral migration of the r7/r8-derived subset of the AES, and for the formation of a properly sized and organized ION. Using conditional gene inactivation of Gli3, we demonstrate that defects in precerebellar neuron migration and nuclei development are not dependent on a cell-autonomous function of GLI3 in MFN progenitors or neurons, and that inactivation of Gli3 in the central nervous system (CNS) after E10.5 does not alter the formation of the precerebellar system. These results suggest that the abnormal migratory path of precerebellar neurons in Gli3 null mutants could be caused by: (1) early developmental patterning defects in non-clRL-derived hindbrain nuclei and their projections; or (2) aberrant development of tracts originating from non-CNS-derived neuronal clusters, i.e. the cranial ganglia. These structures may normally act as an orienting substrate for the migrating precerebellar neurons. We provide indirect evidence that the descending tract of the trigeminal ganglia (spinal trigeminal tract; sp5) could potentially serve such a role. In controls, the AES is in close contact with the sp5 during its rostral-oriented migration phase. In Gli3-null mutants, the sp5 is defasciculated and the AES is detached from the tract. In contrast, the organization of non-precerebellar hindbrain nuclei appears to be largely unaffected in absence of Gli3. Further investigation will be required to determine whether the sp5 does indeed act as a guiding structure for precerebellar migration, and whether and which hindbrain nuclei and projections influence precerebellar migration.

Gli3 is expressed in the developing dorsal neural tube and plays important roles in the development of dorsal brain areas (Blaess et al., 2006, 2008; Haddad-Tóvolli et al., 2012; Hui et al., 1994; Theil et al., 1999; Willaredt et al., 2008). To investigate Gli3 expression in the developing precerebellar system, we performed RNA in situ hybridization for Gli3 and analyzed β-gal expression in Gli3-lacZ knock-in mice (Gli3^lacZ/+, Garcia et al., 2010). In the E9.5 hindbrain, Gli3 was expressed in the entire dorsal area (alar plate), including the clRL (Fig. S1A-C; Ray and Dymecki, 2009). By E10.5, Gli3 was excluded from the dorsal-most hindbrain and no longer overlapped with the Atoh1-positive MF progenitor domain. The adjacent, Atoh1-negative CF domain continued to express Gli3 (Fig. S1D-H). At E9.5 and E10.5, Gli3 was also expressed in the forming trigeminal ganglia (Fig. S1A,E). The Barhl1 (BarH like 1)-positive MF migratory streams and nuclei were negative for Gli3 at E14.5 and E18.5 (Fig. S1I-Q; Li et al., 2004). At these stages, Gli3 was primarily expressed in the Er81 (Etv1)-positive subset of the ION (Fig. S1I,J,O,P,R-W; Xiang et al., 1996; de Diego et al., 2002; Zhu and Guthrie, 2002; Hashimoto et al., 2012) and in a few non-precerebellar hindbrain structures, including the superior olivary complex, the ventral nucleus of the lateral lemniscus and the solitary nucleus (Fig. S1X-BB and data not shown).

To examine whether Gli3 plays a role in the development of the precerebellar system, we analyzed MF and CF nuclei in mice homozygous for the Gli3 extratoe (xt) allele. The Gli3 xt allele contains a large deletion that includes the 3′ end of the Gli3 gene, resulting in a loss-of-function phenotype for GLI3. Gli3^xt/xt mice have multiple defects in organ and brain development, and die at birth (Hui and Joyner, 1993; Johnson, 1967). At E18.5, the Barhl1-expressing PN and RTN were not properly formed in the Gli3^xt/xt hindbrain (Fig. 1A-F). Barhl1-expressing neurons were located close to the pial surface and were dispersed in discontinuous clusters along the ventral midline of r3-r8 in Gli3^xt/xt mutants (Fig. 1A-N, Fig. S2). The distribution of the cells varied in different mutants: in about half of the analyzed brains, cells accumulated at the midline from r3-r8, with most cells positioned in caudal rhombomeres (n=11/20). More than one quarter of the brains (n=7/20) had nuclei of variable size at both the PN/RTN and ION level, with cells sporadically distributed along the rostrocaudal axis. The remaining mutants had no discernible PN/RTN (n=2/20) (Fig. 1C,D, Fig. S2). The clusters at ectopic caudal positions in Gli3^xt/xt brains were organized similarly to the PN/RTN in controls: cells accumulated at both sites of the ventral midline but were separated by a small gap (coronal sections, n=15/15 at rostral levels, n=11/15 at r6-r8) (Fig. 1C,F; Fig. S2F-H,J-L). Barhl1-expressing cells in an extramural layer in the lateral hindbrain at r4-r6 levels of some Gli3^xt/xt mutants were likely remnants of the AES (Fig. 1F, Fig. S2G).

At the level of r7/8, Barhl1-positive cells clustered where the LRN normally forms, but the presumptive LRN was reduced in size in the Gli3^xt/xt hindbrain and some Barhl1-positive cells remained extramurally (Fig. 1L,N). Barhl1 and Er81 double-positive ECN cells did not cluster dorsally but were dispersed along the dorsolateral pial surface in the Gli3^xt/xt hindbrain (Fig. 1G-N). Monitoring the location of labeled cells derived from Atoh1 (atonal homolog 1)-expressing MF progenitors in Atoh1-Cre, Gli3^xt/+ and Atoh1-Cre, Gli3^xt/xt mice that also contained a Rosa reporter allele (R26^lacZ/+ or R26^EYFP/+) confirmed the altered distribution of MFN in Gli3^xt/xt mutants (Fig. 1O,P; Fig. S2M,N).

To examine whether clusters of Barhl1-positive cells located along the ventral midline of r4-8 were PN/RTN neurons, we analyzed the expression of Lhx2 (LIM homeobox protein 2), Brn3.2 (Pou4f2) and Barhl2 (Fig. 1Q-FF), three genes that are not expressed in ECN/LRN neurons at E18.5 (data not shown; Mo et al., 2004). Barhl1-positive ventral clusters in r7/8 of the Gli3^xt/xt hindbrain expressed Lhx2, Brn3.2 and Barhl2, confirming their PN identity (Fig. 1Z,BB,DD,FF). Brn3.2 and Barhl2 are weakly expressed in RTN neurons (Fig. 1U,W). Interestingly, Barhl1-positive clusters in r4 of Gli3^xt/xt hindbrains expressed only low levels of Brn3.2 and Barhl2, suggesting that they have a RTN identity (Fig. 1V,X).

Analysis of the expression of the transcription factors Er81 and FOXP2 (forkhead box P2, which labels all ION neurons, Fujita and Sugihara, 2012; Zhu and Guthrie, 2002) showed that the ION was decreased in size and that the rostral ION in r7 was not established in E18.5 Gli3^xt/xt hindbrains compared with controls. The layered organization of the remaining ION was altered in Gli3^xt/xt mutants (Fig. S3A-H). In contrast to the variations in the clustering of ectopic MFN neurons, the ION phenotype was comparable in all analyzed mutants (n=8/8).

To investigate whether neurons in the altered precerebellar system in the Gli3^xt/xt are still able to establish connections to the cerebellum, we performed DiI labeling experiments in which DiI crystals were placed into the E18.5 cerebellum (Fig. S4A-H). This analysis showed that at least some ectopic MFN in the caudal hindbrain retain their ability to connect to the cerebellum (Fig. S4F). In contrast, we did not detect any DiI labeling in the rostrally located MFN clusters or in the ION in the Gli3^xt/xt brains, suggesting that MF and CF are not properly established (Fig. S4D,F). Whether the apparent loss of ION-cerebellum connections is a primary defect in ION/MF neurons could not be clarified, as it might be a consequence of the malformed cerebellum in Gli3^xt/xt brains (Blaess et al., 2008).

In the clRL, the dorsal Atoh1-expressing domain contains MF progenitors whereas the ventral clRL domain gives rise to CF neurons and is divided into Ngn1- (neurogenin 1), Ascl1- (achaete-scute complex homolog 1) and Ptf1a- (pancreas specific transcription factor 1a) positive subdomains (Fig. 2A, Kim et al., 2008; Ray and Dymecki, 2009; Rodriguez and Dymecki, 2000; Storm et al., 2008; Yamada et al., 2007). As GLI3 plays a role in dorsoventral fate specification in the neural tube (Persson et al., 2002), we analyzed whether the different transcription factor subdomains in the clRL are altered in E10.5 Gli3^xt/xt embryos, but did not detect an obvious difference in the size and organization of these domains when compared with controls (Fig. 2B-J,O and data not shown). Expression of Lmx1a, a roof plate marker, was also comparable in control and mutant embryos (Fig. 2K,L).

WNT1 is a regulator of progenitor proliferation and, depending on the developmental stage and the rostrocaudal hindbrain level, Wnt1 expression fully or partially overlaps with the MF and CF progenitor domains (Gibson et al., 2010; Rodriguez and Dymecki, 2000; Ulloa and Martí, 2010). Comparison of the Wnt1-positive domain in control and mutant hindbrain at E10.5 did not reveal a difference in domain size or in Wnt1 expression level (Fig. 2M-O). The number of proliferating cells in S phase (BrdU-positive cells after 1 h BrdU pulse) in the Atoh1- and Ngn1-positive progenitor domain at E10.5 was not significantly different between control and mutant (Fig. 2P). In summary, MF or CF progenitor domains are established normally in absence of Gli3 function and changes in progenitor fate or proliferative potential are unlikely to be the underlying cause for the aberrant organization of precerebellar nuclei in Gli3^xt/xt mutants.

As MF and CF progenitor domains are not altered in Gli3^xt/xt mutant embryos, the aberrant development of precerebellar nuclei is likely caused by alterations in the corresponding migratory streams. We first investigated the IMS by analyzing Er81- and Brn3.2-positive CF neurons (r7/r8, Fig. S3I-X). In control brains, the IMS was present at E12.5, reached the ventral midline by E13.5 and gave rise to the typical layered ION structures at E16.5 (Fig. S3I-P). In the Gli3^xt/xt hindbrain, the IMS was barely discernible at E12.5 and the Brn3.2/Er81-expressing cells reached the ventral midline only at E14.5 (Fig. S3Q-X). In contrast to control animals, in which Er81-positive/Brn3.2-negative cells at the tip of the IMS were trailed by a population of Er81-positive /Brn3.2-positive cells at E12.5 and E13.5 (Fig. S3I-X), we could only detect an Er81/Brn3.2 double-positive cell population in Gli3^xt/x mutants at E13.5 (Fig. S3Q-X). In addition to the delayed migration and the altered composition of the IMS in Gli3^xt/x mutants, the forming ION domain was reduced in size, in particular in r7, and layering was abnormal in the remaining ION (Fig. S3Q-Y). Cell death is unlikely to be the primary cause for the reduced size of the ION, as the number of cleaved caspase 3-expressing CFNs was not increased in E14.5 Gli3^xt/xt embryos when compared with controls (data not shown).

We next assessed the development of the PES in r7/8. In control embryos, the Barhl1-positive PES started to form at E12.5, reached the midline by E13.5, and established the LRN and ECN by E16.5 (Fig. 3A-D; Kawauchi et al., 2006). In Gli3^xt/xt mutant embryos, the PES reached the ventral midline only at E14.5 and Barhl1-expressing cells persisted extramurally in the E16.5 Gli3^xt/xt hindbrain (Fig. 3E-H). These results imply a delay in cell migration for both the IMS and PES in Gli3^xt/xt embryos.

The AES could be visualized as a compact, extramural accumulation of Barhl1-positive cells in the dorsolateral hindbrain (r4-r6) of E13.5-E16.5 control brains (Fig. 3I-L; Geisen et al., 2008). In Gli3^xt/xt mutant embryos, Barhl1-positive cells started to accumulate extramurally in r4-r6 at E14.5 but were more dispersed than in controls. Between E15.5 and E16.5, extramural Barhl1-expressing cells reached the ventral midline and accumulated in ectopic caudal positions (Fig. 3M-P). Whole-mount immunostaining for BARHL1 on E13.5 and E16.5 Gli3^xt/xt and control brains confirmed the delayed onset and disorganization of the MFN migratory streams in Gli3^xt/xt hindbrains (Fig. 3U-X′; Nichols and Bruce, 2006; Geisen et al., 2008).

The ectopic position of PN/RTN neurons in r7/8 of the E18.5 Gli3^xt/xt hindbrain suggests that some AES cells migrate ventrally instead of turning rostrally. To examine this, we analyzed Barhl2 expression, as it is only expressed in AES, but not PES, neurons (Fig. 3Q,S). In Gli3^xt/xt mutants, Barhl2 was weakly expressed in the Barhl1-positive area in r4-r6 and extended towards the ventral-lateral hindbrain in r7/8 (Fig. 3R,T). These data suggest that a subset of AES cells fails to turn rostrally and instead continues on a ventral path in Gli3^xt/xt mutants, leading to ectopic accumulation of PN/RTN neurons at the ventral midline of the caudal hindbrain.

We next investigated whether a specific subset of AES cells fails to turn rostrally in Gli3^xt/xt embryos. AES and PN can be topographically divided into three subsets based on a Hox code that reflects the origin of the cells in r6, r7 or r8 (Di Meglio et al., 2013, schematic in Fig. 4). To examine how the organization of the AES and of the PN/RTN are affected in Gli3^xt/xt hindbrain, we analyzed Hoxb3, Hoxb4 and Hoxb5 expression at E14.5 and E18.5, and Lhx2 expression at E14.5. At E14.5, Lhx2 is most strongly expressed in the Hoxb3-only population in the AES and, in contrast to the Hox genes, it is absent from the PES (Fig. 4A-B,K,L; Fig. S5). In E14.5 Gli3^xt/xt mutants, both Lhx2 and Hoxb3 were present in the dispersed AES in r5/r6, whereas Hoxb4 and Hoxb5 were not expressed (Fig. 4F-J, Fig. S5). By contrast, in r7/8, where the nascent AES extends ectopically into the ventral hindbrain in the mutant embryos (compare with Fig. 3G,R), Hoxb3, Hoxb4 and Hoxb5 were present. Weak Lhx2 was also detected in a few AES cells (Fig. 4P-T′, Fig. S5). At E18.5, the PN in r4-r6 was negative for Hoxb4/5 in the mutant hindbrain (Fig. S6E-H), whereas the ectopic PN neurons in r7/8 expressed all three Hox genes (Fig. S6M-P). Hoxb3 was expressed broadly in the ectopic PN at this level, whereas Hoxb4 and Hoxb5 were expressed in subdomains, resembling the Hox expression pattern in the PN of controls (Fig. S6A-D,M-P). In summary, these data suggest that the AES is split in Gli3^xt/xt mutants: the r6-derived part (Hoxb3/Lhx2 positive) partially retains its ability to migrate rostrally, whereas the r7/8-derived part (Hoxb3/Hoxb4 or Hoxb3/Hoxb4/Hoxb5 positive) follows a direct route towards the ventral midline (Fig. 4U).

Interfering with the function of guidance factors such as SLIT proteins (SLIT1-SLIT3), netrin and the chemokine CXCL12 or their receptors leads to aberrant migration of precerebellar neurons, resulting in phenotypes that appear similar to the ones observed in Gli3^xt/xt mutants (Di Meglio et al., 2008; Dominici et al., 2018; Geisen et al., 2008; Gilthorpe et al., 2002; Kim and Ackerman, 2011; Kuwako et al., 2010; Marillat et al., 2004; Moreno-Bravo et al., 2018; Zhu et al., 2009). As the SLIT receptors ROBO1/2 have recently been shown to function non-cell-autonomously in precerebellar migration (Dominici et al., 2018), we focused our analysis on the expression of netrin (Ntn1) and Cxcl12 and their respective receptors between E13.5 and E15.5 (Fig. 5, Fig. 8 and data not shown). Expression of Ntn1 in the floor plate and the ventricular neuroepithelium (E13.5-E14.5), and of its receptors Unc5c (only expressed in the PES at E13.5 and E14.5) (Fig. 5A-D,E-H and data not shown) and DCC (in AES and PES, E14.5-E15.5) was not altered in the Gli3^xt/xt hindbrain (Fig. 8I-P and data not shown). The netrin receptor UNC5B is restricted to the dorsal AES (Di Meglio et al., 2013). In our hands, Unc5b expression was weak even in the compact AES in controls, making it difficult to determine whether Unc5b was expressed in the dispersed migratory stream in Gli3^xt/xt mutants (data not shown).

ROBO3 forms a complex with DCC, which has been shown to promote ventral migration of precerebellar neurons (Di Meglio et al., 2008; Marillat et al., 2004; Zelina et al., 2014). Robo3 was prominently expressed in the AES of control and Gli3^xt/xt mutants between E14.5 and E16.5, and in the AES and PES at E13.5 (Fig. 5I-L, Fig. S7 and data not shown). In IMS cells, Robo3 is presumably downregulated after their leading processes cross the midline (Di Meglio et al., 2008; Marillat et al., 2004). Indeed, we could not detect Robo3 expression in the forming ION of control or Gli3^xt/xt embryos at E13.5 and subsequent stages (Fig. 5K,L, Fig. S7 and data not shown), suggesting that midline crossing of ION projections is not affected in Gli3^xt/xt mutant embryos.

The CXCL12 receptor Cxcr4 was expressed in the AES in E14.5 control brains. In Gli3^xt/xt mutants, Cxcr4 was expressed only in the AES in r7/8, but was downregulated in the rostrally migrating AES cells (Fig. 5M-P), suggesting that the rostral migration of the AES is independent of Cxcr4 in the absence of Gli3. As the inactivation of Cxcr4 results in disrupted marginal migration (Zhu et al., 2009), the reduced Cxcr4 expression in the AES may explain the partial delamination of AES neurons into the intramural space observed at E16.5 and E18.5 (compare with Fig. S2F-H). The ligand Cxcl12 was expressed in the meninges and in the ventral cochlear nucleus in control and Gli3^xt/xt mutant brains at E15.5 (Fig. 5Q,R; Zhu et al., 2009). In summary, these data indicate that alterations in the expression of guidance cues and receptors known to regulate precerebellar migration are unlikely to be the primary cause for the altered migratory streams in Gli3^xt/xt mutant brains.

Gli3 is only expressed in MF progenitors in the clRL before E10.5 and is not present in the AES or PES (Fig. S1), suggesting that Gli3 regulates migration of MFNs non-cell-autonomously. To investigate this, we inactivated Gli3 specifically in cells that express Atoh1 (Fig. 6A). In this mouse model (Atoh1-Gli3 cko, genotype Atoh1-Cre, Gli3^xt/flox), Cre-mediated recombination and thus inactivation of Gli3 occurs specifically in MF and other RL progenitors around E10.5 (Fig. 6B-E,K, Fig. S8). In Atoh1-Gli3 cko brains, the location and size of MF nuclei at E18.5, and the organization of the AES and PES at E14.5 was comparable with controls. Moreover, both streams maintained their proper topographical organization as assessed by Hox gene expression (Fig. 6Q-Z). In summary, these data support a non-cell-autonomous role of Gli3 in MFN migration.

Based on the continuous expression of Gli3 in a subdomain of the ION throughout embryonic development, Gli3 might have a cell-autonomous role in the assembly of the ION (Fig. S1). To examine this possibility, we generated a CNS-specific Gli3 conditional knockout using Nestin-Cre (termed Nes-Gli3 cko; Fig. 7A). In this mouse model, recombination in the CNS, including the entire hindbrain, starts around E10 (Fig. 7B; Blaess et al., 2006; Graus-Porta et al., 2001). We did not detect any phenotypes in MF or CF streams or nuclei in these mutants at E14.5 or E18.5 (Fig. 7C-J, Fig. S9 and data not shown). These results demonstrate: (1) that the normal development of the ION does not depend on GLI3 function after E10.5; and (2) that the overall development of the precerebellar system is not affected by inactivation of Gli3 in the CNS after E10.5.

As the conditional gene inactivation of Gli3 suggests a non-cell-autonomous role for GLI3 in precerebellar migration, we investigated whether defective precerebellar development might be a consequence of overall altered hindbrain development. It has previously been suggested that specific cell clusters, such as the facial motor nucleus, may act as guideposts for the AES (Geisen et al., 2008). Analysis of Isl1 (islet 1) expression showed that the hindbrain motor nuclei, including the facial motor nucleus, were properly established in the Gli3^xt/xt hindbrain at E14.5 (Fig. S10A and data not shown). The principal trigeminal nucleus (Pr5) and spinal trigeminal nucleus (Sp5) are derived from the dorsal Gli3-expressing hindbrain neuroepithelium and Atoh1-expressing RL. These nuclei, which are located along the migratory pathway of the AES or the AES and PES, respectively, are Lhx2/Lhx9/Pax2 positive (Pr5) or Lhx2/Lhx9/Pax2/Gbx2 positive (caudal Sp5). The oral Sp5 is only Lhx2 positive. Based on these markers, the Pr5 and the caudal and oral Sp5 were established in Gli3^xt/xt embryos (Fig. S10B-E and data not shown). The Pax2-positive median raphe nucleus and the Gbx2-positive gigantocellular reticular nucleus were analyzed as examples of ventrally derived nuclei and these were also formed in Gli3^xt/xt hindbrain (Fig. S10D,E and data not shown). To assess whether GLI3 might play a general role in development of lRL derived-neurons, we analyzed the cochlear extramural stream (Robo3/Lmx1a positive, Barhl2 negative) and ventral cochlear nuclei (Mafb positive), which are generated in the rostral lRL in r2-r5, in control and Gli3^xt/xt embryos. We found that these RL-derived structures were formed in Gli3^xt/xt embryos (Fig. S10F-J and data not shown). Thus, severe defects in hindbrain development appeared to be restricted to clRL-derivatives in the Gli3^xt/xt mutants. Although we cannot fully exclude the possibility that there are subtle changes in other non-precerebellar hindbrain nuclei and their projections, these results suggest that alterations in non-precerebellar hindbrain nuclei may not be the primary cause of the precerebellar defects.

AES neurons pass the roots of cranial nerves during their rostral-ventral directed movement (Nichols and Bruce, 2006; Kratochwil et al., 2017) and AES and ION neurons have been shown to migrate along cranial nerve roots to the periphery in absence of interactions with their preferred guidance cue netrin in the pial surface (Moreno-Bravo et al., 2018; Yung et al., 2018). This prompted us to examine whether alterations in cranial ganglia and their projections may contribute to the defects in the precerebellar system of Gli3^xt/xt embryos. Using ISL1 immunostaining, we found that the trigeminal ganglia were expanded caudally in E14.5 Gli3^xt/xt mutants and that cell death (cleaved caspase 3-positive cells) was significantly reduced in Gli3^xt/xt trigeminal ganglia when compared with controls (Fig. 8A-H). The size of other cranial ganglia was not obviously altered in Gli3^xt/xt mutants (data not shown). DiI labeling of the trigeminal ganglia at E14.5 and whole-mount immunostaining of E18.5 brains for peripherin (PRPH), an intermediate filament that is primarily expressed in neurons of the peripheral nervous system, showed that the central descending sensory tract of the trigeminal ganglion, the trigeminal spinal tract (sp5), was highly disorganized in Gli3^xt/xt mutants (Fig. S11), suggesting that loss of Gli3 function affects the development of the trigeminal ganglia and its central projections.

The sp5 terminates in the spinal trigeminal nucleus (Sp5), which extends from the caudal hindbrain into the upper cervical spinal cord. To gain insight into whether the sp5 could potentially serve as a guidance structure for migrating precerebellar neurons, we performed immunostaining for PRPH to label cranial tracts and for DCC to label MFNs in E14.5 Gli3^xt/xt and control embryos (Fig. 8Q). In controls, the forming AES was located dorsally to the sp5 in its initial migration phase in r7/8 (Fig. 8O). During its rostrally oriented migration, the AES was positioned just laterally to the sp5, and cells in the ventral AES were in close contact with the tract (Fig. 8K,M; Movie 1). In r3/4, where the AES turns ventrally to reach the ventral midline, the AES was again separated from the sp5 (Fig. 8I). In Gli3^xt/xt mutants, analysis of the PRPH immunostaining confirmed that the sp5 was disorganized and less fasciculated than in controls. DCC immunostaining revealed that the AES was fragmented into small clusters in Gli3^xt/xt mutants, as previously shown with in situ hybridization for Barhl1 (compare with Fig. 4F-H) and that the AES was largely detached from the sp5 in the mutant embryos (Fig. 8L,N,P). These data show that AES neurons normally migrate in close apposition to the sp5 during their rostrally oriented movement and that the loss of Gli3 results in the defasciculation of sp5 and an apparent detachment of AES neurons from the tract. Together, these results suggest that the disorganization of the sp5 might be one factor contributing to the defects in AES migration in Gli3^xt/xt embryos.

GLI3 regulates early patterning events in spinal cord, cortex, dorsal midbrain and cerebellum (Blaess et al., 2008; Persson et al., 2002; Theil et al., 1999). In the Gli3^xt/xt spinal cord, progenitor domains at intermediate levels of the dorsoventral axis (known as p1 and dl6), which express Nkx6-2, Dbx1 or Dbx2, are expanded dorsally when compared with controls. In contrast, more dorsal progenitor domains (known as dl5 and dl4) are decreased in size (Persson et al., 2002). In the Gli3^xt/xt hindbrain, only a Nkx6-1-expressing domain is expanded and only in r1-r3 (Lebel et al., 2007). Alterations in the most dorsal domains (dl1-dl3 in spinal cord, RL in hindbrain) have not yet been described in Gli3^xt/xt mutants. We show that the size of clRL domains are not changed in Gli3^xt/xt mutants and that Gli3 expression is downregulated in the most dorsal part of the clRL domain after E9.5. Moreover, we could not detect any severe alterations in the organization of non-precerebellar caudal hindbrain nuclei in Gli3^xt/xt mutants. Even the lRL-derived cochlear nucleus was established correctly in the mutants. Based on these data and on previous analysis of the Gli3^xt/xt hindbrain (Lebel et al., 2007), we propose that indirect effects of patterning defects in the caudal hindbrain of Gli3^xt/xt mutants are not the most likely explanation for the alterations in precerebellar development. Nevertheless, we cannot exclude that our analysis missed more subtle defects in the organization of hindbrain nuclei or alterations in their projections in the Gli3^xt/xt brain, which might be of consequence for the migration of precerebellar neurons. Finally, as recently proposed by Dominici et al. (2018), severe developmental defects, such as those occurring in Gli3-null mutants, might have unspecific consequences on neuronal development, including the development of the hindbrain precerebellar system.

Using conditional gene inactivation, we demonstrate that GLI3 is unlikely to have a cell-autonomous role in the migration of MFNs. This result is consistent with the observation that Gli3 is largely excluded from the MF progenitor domain after E9.5 and is not expressed in migrating MFNs. Notably, recent data show that even guidance factors that are important regulators of precerebellar migration do not act directly on the migrating neurons (Dominici et al., 2018). In mutants without the guidance factor genes Slit1 and Slit2 or the SLIT receptor genes Robo1 and Robo2 the AES is not well defined, Barhl1-positive cells accumulate in ectopic positions at the ventral midline in r4-r8 and the ION is abnormally formed (Di Meglio et al., 2008; Dominici et al., 2018; Geisen et al., 2008), a phenotype closely resembling the one in Gli3-null mutants. However, specific inactivation of Robo1/Robo2 in precerebellar progenitors does not affect precerebellar migration and nuclei formation. Inactivation of Slit1 and Slit2 in the floor plate and in the facial motor nucleus, structures postulated to be important sources of SLIT proteins for migrating precerebellar neurons, results in only a mild phenotype (Dominici et al., 2018). Thus, it is currently unclear which primary defect in Slit/Robo-null mutants leads to aberrant precerebellar migration.

Two recent studies show that NTN1 is not just expressed in the hindbrain floor plate, but is also secreted at the basal lamina by neural progenitor endfeet. Inactivation of netrin results in subsets of AES and IMS cells moving ectopically from the CNS into the PNS (Moreno-Bravo et al., 2018; Yamauchi et al., 2017; Yung et al., 2018). In Gli3^xt/xt mutants, precerebellar neurons stay within the CNS, indicating that netrin-mediated guidance is not affected. However, the fact that precerebellar neurons migrate along cranial nerve roots in absence of netrin suggests that projections from the cranial ganglia may, in principle, serve as corridors for precerebellar migration.

Based on our results, rostrally migrating AES neurons might interact with the sp5 during this phase of their migration. Thus, the altered development of the trigeminal ganglia and its descending central sensory tract, sp5, might contribute to the altered migration of AES neurons in Gli3^xt/xt mutants by disrupting this potential interaction. Central axons of the trigeminal ganglia appear as early as E12.5 in rat, ascending and descending branches are established at E13.5 and the descending branches, which form the sp5, reach the caudal hindbrain 1 day later (Miyahara et al., 2003). Thus, the sp5 is already established by the time the AES starts to form and is ideally positioned to function as a tract for rostrally migrating AES neurons.

We show that AES neurons cluster in close apposition to the sp5 during their rostral migration. Cells in the ventral part of the AES, corresponding to the Hoxb3/4/5-expressing AES cells, appear to be in direct contact with the sp5. In Gli3-null mutants, the ventral Hoxb3/4/5-expressing AES cells fail to switch from a ventrally oriented migration to a rostrally oriented migration and the AES appears to be detached from a defasciculated sp5. An influence of the sp5 on IMS or PES migratory streams, which are also altered in Gli3^xt/xt mutant mice, is less evident. Instead, the defects in the LRN/ECN formation could be a consequence of the accumulation of ectopic PN/RTN cells in the caudal hindbrain in Gli3^xt/xt mutant embryos, whereas the changes in the IMS and ION might be primarily related to a reduction in their cell number.

Cranial ganglia including the trigeminal ganglia are derived from both ectodermal placodes and cranial neural crest (Steventon et al., 2014). The disorganization of the sp5 might be linked to a caudal expansion of the trigeminal ganglia in Gli3-null mutants. An increased size in cranial sensory ganglia has already been noted in the original description of the extratoe phenotype (Johnson, 1967). Highly elevated SHH signaling leads to severely altered development of trigeminal and facial nerves. This effect may be mediated through the negative regulation of WNT signaling by SHH (Kurosaka et al., 2015). Whether the loss of GLI3 and its effect on cranial nerve organization might also be caused by altered WNT signaling remains to be investigated, but data from the spinal cord demonstrate that GLI3 inhibitory activity negatively regulates the canonical WNT signaling pathway by interacting with β-catenin (Ulloa et al., 2007).

In summary, we identify a previously uncharacterized function of the zinc-finger transcription factor GLI3 in regulating precerebellar migration and trigeminal ganglia development. We also show that GLI3 likely acts in an indirect manner on precerebellar migration. Together with the recently published studies that revise the role of netrin and SLIT/ROBO proteins in precerebellar migration (Dominici et al., 2018; Moreno-Bravo et al., 2018; Yung et al., 2018), our results imply that the regulation of precerebellar migration is even more intricate than anticipated. We propose that, in order to fully unravel the mechanisms that guide these migratory streams in the hindbrain, it will be necessary to consider hindbrain fiber tracts as a potential source for guidance cues or as a substrate for migrating precerebellar neurons.

Mice were housed at a maximum of five animals per cage with a standard 12 h light/dark cycle and given food and water ad libitum. Experiments were performed in compliance with the guidelines for the welfare of animals issued by the Federal Government of Germany and the directives of the European Union. Gli3^Xt (Hui and Joyner, 1993), Gli3^flox (Blaess et al., 2008), Gli3^lacZ (Garcia et al., 2010), Nestin-Cre (Tronche et al., 1999), Atoh1-Cre (Matei et al., 2005), Rosa26-stop-lox-stop lacZ (R26^lacZ/+, Soriano, 1999) and Rosa26-stop-lox-stop EYFP (R26^EYFP/+, Srinivas et al., 2001) were genotyped as described. Noon of the day of the vaginal plug was assigned as embryonic day 0.5 (E0.5). The following conditional knock-out (cko) mice were generated: Nes-Gli3 cko (Nestin-Cre, Gli3^xt/flox) and Atoh1-Gli3 cko (Atoh1-Cre, Gli3^xt/flox). We have previously established that mice homozygous for the Gli3^Xt allele (Gli3^Xt/Xt) and mice homozygous for the recombined floxed allele (Gli3^rec/rec) display comparable brain phenotypes, indicating that both alleles are Gli3-null alleles (Blaess et al., 2008). A few Gli3^xt/xt mutants displayed exencephaly; these mutants were excluded from the analysis.

For BrdU (bromodeoxyuridine; Merck) labeling of embryos, pregnant dams were injected intraperitoneally with 100 μg BrdU/g body weight and sacrificed 1 h later.

Embryos or embryonic brains were dissected and collected in ice-cold 0.1 M phosphate-buffered saline (PBS), and embryonic tail samples were collected separately for DNA extraction and genotyping. The embryos were fixed overnight at 4°C in 4% paraformaldehyde (PFA) in 0.1 M PBS. After rinsing twice in 0.1 M PBS for 30 min, the embryos were either treated in an ascending sucrose series (15 and 30% in 0.1 M PBS) and mounted in Tissue Tek freezing medium (Sakura) or dehydrated and processed for paraffin wax embedding. X-gal staining of 14 μm cryosections and immunostaining of 14 μm cryosections or 7 μm paraffin sections were performed as described (Blaess et al., 2011). The following primary antibodies, all of which have been validated in previous publications, were used: mouse anti-BrdU (3D4, BD Biosciences; Lot: 25098; Sudarov et al., 2011) at 1:50; mouse anti-cleaved caspase 3 (5AE1, Cell Signaling Technologies; Blaess et al., 2006 and antibody validation on manufacturer's website) at 1:200; goat anti-DCC (A-20, sc-6535, Santa Cruz Biotechnology; Lot L1610; Dominici et al., 2018) at 1:400; goat anti-FOXP2 (S262, EB05226, Everest Biotech; Lot S2G2; Fujita and Sugihara, 2012) at 1:200; rat anti-GFP (04404-84, Nacalai Tesque; Lot: M8A3283; Sudarov et al., 2011) at 1:2000; rabbit anti-GFP (A11122, Life Technologies; Lot 1828014 and 178991; Sudarov et al., 2011) at 1:500; mouse anti-ISL1 (39.3F7, DSHB; Roy et al., 2012) at 1:50; goat anti-NGN1 (A-20, sc-19231, Santa Cruz Biotechnology; Lot: A1209; Di Giovannantonio et al., 2014) at 1:200; and rabbit anti-PRPH (AB1530, EMD Millipore; Lot 2972430; Ter-Avetisyan et al., 2018) at 1:200. The following secondary antibodies were used: Cy3-conjugated donkey anti-rabbit, anti-mouse or anti-goat (Jackson ImmunoResearch Laboratories) at 1:200; Alexa 488-conjugated donkey anti-rabbit or anti-mouse IgG (ThermoFisher Scientific) at 1:500; Alexa 488-conjugated goat anti-rabbit and anti-mouse, and Alexa 546-conjugated goat anti-rabbit or anti-guinea pig (ThermoFisher Scientific) all at 1:1000. For immunofluorescent detection of transcription factors, sections were treated with 1 mM EDTA at 65°C for 10 min and incubated with biotinylated anti-mouse or anti-goat secondary antibodies (Jackson ImmunoResearch Laboratories) at 1:200, followed by Cy3-conjugated streptavidin (Jackson ImmunoResearch Laboratories) at 1:1000. For BrdU immunostaining, the sections were fixed in 4% PFA for 10 min at room temperature, treated with 2 N HCL for 30 min at 37°C and quickly rinsed in boric acid buffer, followed by a standard immunostaining protocol (Blaess et al., 2011).

RNA in situ hybridization was essentially performed as previously described (Blaess et al., 2011). Frozen sections were fixed in 4% PFA and washed in PBS. Paraffin sections were deparaffinized, rehydrated, treated with proteinase K (Roche) and then acetylated. After washing in H₂O, frozen and paraffin wax-embedded sections were dehydrated in different concentrations of ethanol (70%, 80%, 95% and 100%) and incubated in chloroform to de-fat the sections. After hybridization with the cRNA probes and immunodetection of digoxigenin with alkaline phosphatase-conjugated antibody (Roche), sections were incubated with BM purple (Roche). The color reaction was stopped using Tris-EDTA buffer. The following in situ probes were used: Ascl1 (Sudarov et al., 2011); Atoh1 (Corrales et al., 2004); Barhl1 (Bulfone et al., 2000; IMAGE clone ID 335997, accession number W36375); Barhl2 (Saba et al., 2003); Brn3.2 (de Diego et al., 2002); Cxcl12 and Cxcr4 (Tissir et al., 2004); Er81 and Gli3 (Corrales et al., 2004); Hoxb3, Hoxb4 and Hoxb5 (a gift from Robb Krumlauf, Stowers Institute for Medical Research, MO, USA); Isl1 (Blaess et al., 2008); Lhx2 (Rétaux et al., 1999); Lmx1a (Blaess et al., 2011); Ntn1 (Hammond et al., 2009); Ngn1 (Liu et al., 2008); Pax2 (Asano and Gruss, 1992); Ptf1a (bp 1023-1490 of cDNA); Robo1 and Robo2 (Bagri et al., 2002); Robo3 (Marillat et al., 2004); Slit1 and Slit2 (Bagri et al., 2002); Slit3 (Geisen et al., 2008); Unc5b (Lu et al., 2004); and Unc5c (Przyborski et al., 1998).

E13.5, E16.5 and E18.5 brains were dissected, fixed in 4% PFA overnight at 4°C and rinsed three times in PBS. The tissue was then dehydrated, bleached, rehydrated and immunolabeled as described by Renier et al. (2014; idisco.info/idisco-protocol/update-history/). Brains were incubated with antibodies for 24-48 h with rabbit anti-BARHL1 (1:500, HPA004809, Sigma Aldrich, Lot: A114004; Dominici et al., 2018) or rabbit anti-PRPH (1:200, AB1530, EMD Millipore, see above for details) and biotin anti-rabbit (1:200; 711065152, Jackson ImmunoResearch) antibodies. Colorimetric detection was achieved using the HRP Vectastain Elite ABC Kit (PK-6100; Vector Laboratories) and the DAB Peroxidase Substrate (SK-4100; Vector Laboratories).

Embryonic brains of control and Gli3^xt/xt mice were removed from the skull with the trigeminal ganglia attached and were fixed overnight with 4% PFA at 4°C. Brains were rinsed three times in PBS to remove residual PFA. Several small crystals of 1,1′-Dioctadecyl-3,3,3′,3′-tetramethylindocarbocyanine perchlorate (DiI, D282; Molecular Probes) or 3,3′-Dioctadecyloxacarbocyanine perchlorate (DiO, D275; Molecular Probes) were attached to the tip of a glass needle. Crystals were placed medially and laterally on the exposed area of one cerebellar hemisphere to retrogradely label precerebellar inputs or on the trigeminal ganglia to anterogradely label the sp5 tract. In a few embryos, DiO was used instead of the DiI to label the trigeminal ganglion. The brains were first kept in 1% PFA at 37°C for 4 days and then at room temperature for 1 month in the dark. The brains were washed in PBS, embedded in 3% Ultra Pure LMP Agarose (15517-022, Life Technologies) and cut at 80 μm with a vibratome. The sections were counterstained with Hoechst, mounted with Fluoromount (F4680, Sigma Aldrich) and imaged. To visualize the distribution of the different precerebellar neurons, adjacent sections were immunostained for BARHL1 (1:500, HPA004809, Sigma Aldrich) or FOXP2 (1:200, EB05226; Everest Biotech).

Bright-field images were acquired with a Leica DM1000 microscope or an inverted microscope (Axio Observer, Zeiss). Images of immunofluorescent samples and DiI/DiO labeled samples were acquired with an inverted fluorescence microscope (Axio Observer, Zeiss) using 10×, 20×, 40× or 63× objectives (EC Plan-Neofluar, Zeiss). Structured illumination (ApoTome, Zeiss) was used to acquire images with the 20×, 40× and 63× objectives. The 63× objective images were obtained as maximum intensity projections of z-stack mosaics. The Zeiss Mosaix software was used to assemble images if more than one image was acquired to cover a region of interest. 3D-projection was assembled in ImageJ.

To define rhombomere levels in the developing hindbrain, morphological features such as the location of the cochlear nuclei, precerebellar nuclei, otic vesicles, various motor nuclei and cranial nerves (Farago et al., 2006; Geisen et al., 2008; Guthrie, 2007; Nichols and Bruce, 2006; Paxinos et al., 2006, 2012; Schambra, 2008) were taken into consideration. In addition, rhombomere levels were assessed by analyzing the expression of Hoxb3, Hoxb4 and Hoxb5 (Philippidou and Dasen, 2013).

The area of domains positive for progenitor markers were measured in E10.5 hindbrains (n≥3 brains per genotype, in sections of five levels for each hindbrain; results of only one level are shown in Fig. 2). The progenitor domain areas were normalized to the area of hindbrain tissue measured at the same level. Areas were quantified using ImageJ.

The area of domains positive for Er81 or Brn3.2 were measured in E15.5 hindbrains (n≥3 brains per genotype, in sections of seven levels for each hindbrain). The area of the expression domains was normalized to the area of hindbrain tissue measured at the same level. Areas were quantified using ImageJ.

BrdU-positive cells were counted within the progenitor domains expressing Atoh1 or Ngn1 at r6/7. The expression of the two progenitor markers was analyzed on adjacent sections. The number of BrdU-positive cells was normalized to the area of the respective progenitor domain (n≥3 brains per genotype, three levels per hindbrain, results of only one level are shown in Fig. 2). Areas were quantified using ImageJ.

At the level of r3-5, ISL1- and cleaved caspase 3-positive cells were counted (n≥3 brains per genotype). Area size or cell counts are presented for the different levels. For the cleaved caspase 3 quantification, numbers of cleaved caspase 3-positive cells and ISL1-positive cells at the different levels were added up and the number of cleaved caspase 3-positive cells was normalized to the number of ISL1-positive cells. Areas were quantified using ImageJ.

Analysis was carried out non-blinded, as the mutant phenotype was obvious. There was no randomization in the groups and no statistical method was used to predetermine sample sizes. Statistical analysis of histological data was performed using Student's t-test, parametric one-way analysis of variance (ANOVA) with a post-hoc Tukey or non-parametric Kruskal–Wallis Test (Origin 8, Origin lab and Prism7, Graphpad). Levene's test was used to assess equality of variances. Significance levels were set at P<0.05. The values are represented as mean±s.d.

ISL1 antibody was obtained through the Developmental Studies Hybridoma Bank under the auspices of the National Institute of Child Health and Human Development and maintained by The University of Iowa (Iowa City, IA). The authors thank Susan Ackerman, Alain Chédotal, Robert Edwards, Anne Eichmann, André Goffinet, Alexandra Joyner, Robb Krumlauf, Mengshen Qiu, Sylvie Rétaux, Filippo Rijli, Tetsuichiro Saito, Marc Tessier-Lavigne and Marion Wassef for in situ probes; Stephan Baader and Ulrich Schüller for the Atoh1-Cre mice; and Martin Jansen for technical support.