Short-range Fgf signalling patterns hindbrain progenitors to induce the neurogenesis-to-oligodendrogenesis switch

The vertebrate nervous system is characterised by its structural complexity that underlies its remarkable emergent properties. Unlike many other organs that comprise repeating tissue units with a limited number of cell types, the nervous system has an exceptionally large number of distinct cell types relative to its size (Tasic et al., 2018). Mechanisms that underlie increasingly refined tissue patterning over a prolonged period are particularly crucial to neural development.

Strategies for increasing heterogeneity among progenitor cells in the developing nervous system involve spatial and temporal regulation of transcription factor expression together with progenitor maintenance. Extracellular signalling molecules, such as morphogens, play an important role in the spatial patterning of the neural tube along both the anterior-posterior (AP) and the dorsal-ventral (DV) axes (Jessell, 2000; Papalopulu and Kintner, 1996; Sasai et al., 2014). The acquisition of cell type identities is dependent on the concentration and/or duration of morphogen signals (Tozer et al., 2013;Greenfeld et al., 2021). A classic example is the DV-patterning of the ventral neural tube by a concentration gradient of Sonic hedgehog (Shh) signalling (Dessaud et al., 2007), which results in the activation of a robust gene regulatory network and subdivision of the progenitor population into precise proportions (Exelby et al., 2021).

The superimposition of spatial and temporal patterning is required for the specification of later-born cell types. This is particularly relevant to gliogenesis, which generally occurs later in development, (Sagner, 2024) and is preceded by neurogenesis (Cepko, 2014; Guillemot, 2007; Miller and Gauthier, 2007; Rossi et al., 2017; Rowitch and Kriegstein, 2010). Extracellular signalling plays an important role in modulating intrinsic temporal programmes among neural progenitors for the timely production of different cell types (Dias et al., 2014; Rossi and Desplan, 2020; Sagner et al., 2021; Shi and Liu, 2011). Notably, it has been found that earlier-born neurons can act as a source of signalling to regulate the production of later-born cell types (Barnabé-Heider et al., 2005; Seuntjens et al., 2009). Such feedback regulatory mechanisms effectively provide temporal checkpoints in tissue development and ensure spatial precision in cell fate acquisition.

Temporal patterning is necessarily concomitant with progenitor maintenance to produce the correct proportion of late-born cell types. The inhibition of neurogenesis is an essential prerequisite that permits gliogenesis, as proneural basic helix-loop-helix (bHLH) transcription factors repress gliogenesis in neural progenitors (Sun et al., 2001; Zhao et al., 2015). Proneural bHLH factors also commit progenitors to neuronal differentiation (Bertrand et al., 2002; Guillemot and Hassan, 2017), which depletes the progenitor pool for later glial production. For example, the pMN progenitors in the spinal cord generate both somatic motor neurons and oligodendrocytes; lowering the activity of proneural neurogenin 2 by upregulation of Olig2 helps maintain pMN progenitors for later oligodendrogenesis (Lee et al., 2005). Similarly, gliogenic transcription factors, such as Sox9 and Nfi, exert their function at least partly through inhibiting neurogenesis (Deneen et al., 2006; Stolt et al., 2003; Vong et al., 2015). Juxtacrine Notch signalling is a key pathway implicated in gliogenesis, which inhibits proneural genes through its downstream effectors the HES/Her transcription factors (Itoh et al., 2003; Kageyama and Ohtsuka, 1999). Notch signalling is often permissive rather than instructive for the acquisition of glial fate by inhibiting neurogenesis (Rowitch and Kriegstein, 2010). For example, forced activation of the Notch pathway in the ventral spinal cord results in an overall increase in oligodendrocytes but has a negligible effect on olig2 expression or in advancing the timing of oligodendrocyte production (Park and Appel, 2003).

Previous studies have identified a role of Fgf20 signalling in patterning proneural gene expression in the hindbrain (Gonzalez-Quevedo et al., 2010; Terriente et al., 2012). Neural development in the hindbrain provides an accessible context for discovering patterning mechanisms required for increasing tissue complexity, as it shares many of the initial conditions of DV-patterning in its posterior counterpart at the spinal cord level (Krumlauf and Wilkinson, 2021). In the zebrafish hindbrain, proneural gene expression is patterned along the AP axis, with a high level of expression flanking the hindbrain boundaries and inhibition in each segment centre (Amoyel et al., 2005). The inhibition of proneural gene expression at the hindbrain boundaries is regulated by Notch signalling as well as Yap/Taz-TEAD-mediated mechanotransduction (Cheng et al., 2004; Voltes et al., 2019), while the inhibition in the segment centre is regulated by Fgf20a signalling emanating from a subset of earlier-born neurons positioned at each centre (Gonzalez-Quevedo et al., 2010). Accompanying the patterning of proneural gene expression, Fgf20 is also responsible for the upregulation of various markers in the segment centre, including gliogenic factors, such as sox9 and members of the meteorin family (Gonzalez-Quevedo et al., 2010; Lee et al., 2010; Nishino et al., 2004; Stolt et al., 2003; Tambalo et al., 2020). Additionally, Fgf signalling has been implicated in the production of oligodendrocytes (Chandran et al., 2003; Esain et al., 2010; Kessaris et al., 2004). In particular, Esain et al. (2010) found that Fgf-receptor (Fgfr) activation is required for the production of oligodendrocyte precursor cells (OPCs) in the zebrafish hindbrain through its regulation of olig2 and sox9 expression. Crucially, the time during which Fgfr activation is required for this process significantly overlaps with the period of Fgf20 signalling (Gonzalez-Quevedo et al., 2010). This suggests a potential role of Fgf20 in mediating the switch from neurogenesis to gliogenesis.

Here, we find that Fgf20 signalling emanating from earlier-born neuronal clusters acts at short range to confer precise spatial and temporal patterning of the hindbrain segments to induce the neurogenesis-to-oligodendrogenesis switch. This signal inhibits neurogenesis by downregulating proneural gene expression and induces oligodendrocytes among competent progenitors in the segment centre. The short-range property of Fgf20 confers precise spatial patterning of multiple proneural genes along both the AP and DV axes. Additionally, we find that the downregulation of early proneural genes is highly sensitive to Fgf pathway activation. Thus, the propensity for neurogenesis among progenitors in the segment centre is defined by the dynamics of Fgf20 signalling. Fgf20 signalling acts instructively as it upregulates specification and differentiation factors of oligodendrocytes among competent cells in the segment centre, and the timing and extent of signalling determines the timing of induction and quantity of oligodendrocytes.

It was previously hypothesised that fgf20a-expressing neuronal clusters establish a morphogen gradient from the segment centre, and that a high level of signalling in the progenitors is required for the inhibition of neurogenesis (Terriente et al., 2012). A correlation between the positioning of the fgf20a-expressing neuronal clusters and a patterned Fgfr pathway response in the neuroepithelium has been shown (Gonzalez-Quevedo et al., 2010). However, the methodologies used in these studies lack the spatial resolution necessary to test whether a graded response to Fgf20 is established for the neurogenesis inhibition. To address this, we set out to characterise the expression of fgf20 and the downstream target genes, etv5b and spry1 (Gonzalez-Quevedo et al., 2010; Raible and Brand, 2001; Roehl and Nüsslein-Volhard, 2001), using hybridisation chain reaction (HCR) RNA fluorescence in situ hybridization (RNA-FISH), which enables quantitative measurement of gene expression with high spatial resolution (Choi et al., 2010; Trivedi et al., 2018).

Fgf20a-expressing neurons were first detected in the hindbrain segment centre at 16 h post-fertilisation (hpf), with two or three neurons per segment (Fig. S1A,B). The size of the fgf20a-expressing neuronal clusters increased rapidly in the hindbrain from 15-16 hpf onwards (Fig. 1A-C; Fig. S1C). Transverse sections of the segment centre showed that, from 20 hpf onwards, fgf20a-expressing neurons were positioned at a ventral-lateral region of the neural tube (Fig. 1D; Fig. S1B). This is concomitant with the morphological changes of the hindbrain during ventricle formation, which causes the lateralisation of the dorsal neural tube and the ventralisation of the mantle zone in which differentiated neurons reside (Elsen et al., 2008; Lowery and Sive, 2005). We found that the orthologue fgf20b was also expressed in the hindbrain segment centre (Fig. S2A) and was co-expressed with fgf20a in one or two neurons within each of the fgf20a-expressing neuronal clusters (hereafter referred to as fgf20-expressing neurons) (Fig. S2B). Nonetheless, the severity of fgf20a^−/− embryo defects in the patterning of proneural gene expression suggests that Fgf20a has a major role that is not compensated by Fgf20b (Gonzalez-Quevedo et al., 2010).

Despite detection of fgf20-expressing neurons from 16 hpf, etv5b expression was initially broad and only became restricted and upregulated in the segment centre from 20 hpf onwards (Fig. 1B,C; Fig. S1A). Before this stage, etv5b was expressed broadly in rhombomeres (r) 4-6 due to earlier segmental Fgf3/8a signalling (Maves et al., 2002; Walshe et al., 2002; Wiellette and Sive, 2004). Indeed, it was previously shown that suppression of segmental Fgf3 expression is necessary before the onset of Fgf20-signalling activity for the switch in Fgf signalling patterns (Leino et al., 2023). Importantly, analysis of etv5b and spry1 expression revealed that the extent of Fgf signalling in response to Fgf20 was highly restricted along both the AP and DV axes of the segments. In dorsal and lateral views it was seen that the AP extent of etv5b (Fig. 1C′) and spry1 (Fig. 1C″) expression in the neural epithelium aligned with fgf20 expression in the mantle zone. We further analysed expression in transverse sections of the hindbrain neural tube in which the DV axis of the neuroepithelium was revealed. The membrane marker EGFP-CAAX was used to visualise the orientations of the neural progenitors (Fig. 1D). In the neural tube, the neuroepithelial cells were arranged perpendicular to the mantle zone, with close contacts between their basal processes and differentiated neurons. We found that etv5b expression was restricted to the subpopulation of progenitors that are perpendicular to the fgf20-expressing neurons. This suggests that Fgf signalling is activated only in progenitors that are highly proximal to the source of Fgf20. Therefore, Fgf20 signalling may act at short range around the basal region of the neuroepithelium. Indeed, in efnB3b^−/− embryos that have mispositioned fgf20-expressing neurons (Cayuso et al., 2019; Terriente et al., 2012), we observed a corresponding shift in etv5b expression along the DV axis to the position of fgf20-expressing neurons (Fig. S3). Similarly, in fgf20a^−/− embryos, a low level of etv5b expression correlated with the position of fgf20b-expressing neurons, except for r4, where dorsal etv5b expression was adjacent to Fgf3/8a signalling from the anterior pole of the otic vesicle (Figs 1E, 4G; Fig. S2C) (Maier and Whitfield, 2014). Together, these data highlight the limited range of Fgf20 signalling in the zebrafish hindbrain.

A previous study (Kalinina et al., 2009) has shown that some members of the mammalian FGF9/16/20 family, including FGF20, have strong equilibrium bias towards homodimerisation. This acts as an autoinhibitory control of their activity due to occlusion of receptor binding sites in the homodimer, with an increase in the ligand binding affinity to heparin sulphate, which significantly restricts the diffusion range (Harada et al., 2009; Kalinina et al., 2009). Protein sequence alignment of FGF20 and the zebrafish orthologues showed that the residues required for homodimerisation are conserved (Fig. S4). This suggests that the highly restricted signalling range of Fgf20 in the hindbrain could be due to biochemical properties of the ligand.

Next, we investigated whether the short-range nature of Fgf20 signalling is linked to the spatially restricted expression of proneural genes. It was shown previously that expression of the early proneural gene neurog1 and late proneural gene neurod4 is patterned along the AP axis by Fgf20a signalling (Gonzalez-Quevedo et al., 2010). This patterning emerges from an initially homogeneous expression along the AP axis, resulting in high levels of proneural gene expression in areas flanking the hindbrain boundaries, with inhibition in the segment centre (Fig. 2A,B; Fig. S5). The sensitivity of the HCR technique revealed that a low level of neurog1 and neurod4 expression was maintained in the segment centre of wild-type (WT) embryos (Fig. 2C). This pattern failed to emerge in fgf20a^−/− embryos as high levels of neurog1 and neurod4 expression persisted throughout the segment (Fig. 2A-C).

In addition to neurog1, other early proneural genes, such as ascl1a and ascl1b, were expressed in the dorsal neural tube of the zebrafish hindbrain (Fig. 2D) (Thisse and Thisse, 2005; Amoyel et al., 2005; Jászai et al., 2003). We observed that the combination of the early proneural genes ascl1a, ascl1b and neurog1 spatially overlap with the late proneural gene neurod4 along both the AP and DV axes (Fig. S5C), consistent with neurod4 acting downstream of these early proneural genes in this context (Bertrand et al., 2002). The most dorsal domain of neurod4 expression in the segment centre persisted after downregulation of neurog1 in the ventral segment centre, suggesting that there is DV patterning of neurogenesis (Fig. S5C) (Gonzalez-Quevedo et al., 2010). We therefore wondered whether early proneural genes are regulated along the DV axis by signalling from the ventrally-located fgf20-expressing neurons.

We first examined ascl1a and ascl1b expression at 42 hpf, when the patterning of proneural gene expression is fully established. Ascl1b was expressed broadly along the DV axis of the segment. In WT embryos, it was patterned along the AP axis, except for the most dorsal extent of the expression (Fig. 2E). We found that the AP-patterning of ascl1b expression failed to emerge in fgf20a^−/− embryos, with high levels of expression maintained in the segment centre. In contrast, ascl1a was expressed in the dorsal hindbrain in two narrow and discrete expression domains (Fig. 2E,F). Importantly, in WT embryos, the AP-patterning of ascl1a expression occurred only for the ventral domain, and this patterning was lost in fgf20a^−/− embryos, where ectopic expression was detected in the segment centre (Fig. 2F). Expression analysis of the proneural genes and etv5b at 36 hpf revealed high spatial complementarity between Fgf20 signalling and ascl1a/b (Fig. 2G). Crucially, the extent of Fgf20 signalling in the segment centre, as indicated by etv5b expression, did not surpass the ventral edge of the dorsal ascl1a domain (Fig. 2H). This suggests that Fgf20 signalling patterns proneural gene expression along both the AP and DV axes and the short-range nature of the signal confers the spatial precision of the patterning.

A prediction of the correlation between spatial expression of Fgf20 and inhibition of proneural gene expression is that ectopic activation of Fgf signalling will lead to a downregulation of neurog1 expression. To address this, we used a transgenic zebrafish line [Tg(hsp70:ca-xfgfr1)] that consists of a heat-shock promotor controlling constitutively active Xenopus Fgfr1 (Marques et al., 2008). We performed a time-course analysis of neurog1 and neurod4 after ca-xfgfr1 induction over the course of 32-36 hpf, the period during which hindbrain neurogenesis occurs rapidly (Lyons et al., 2003) (Fig. 3A-D). Heat-shock at 30-31 hpf induced ubiquitous overactivation within 1 h post heat-shock, with the highest level detected at 3 h. This then began to be down-regulated by 5 h (Fig. 3B; Fig. S8). By 3 h post heat-shock, we observed significant downregulation of neurog1 expression, and thus expression of neurog1 is sensitive to Fgf signalling levels (Fig. 3B,C; Fig. S8). By 5 h post heat-shock, neurog1 expression returned to the level seen in WT embryos. In contrast, neurod4 had a dampened and delayed response to ca-xfgfr1 induction (Fig. 3D; Fig. S8). These results suggest that Fgf signalling downregulates proneural gene expression in neural progenitors, and that expression of the early proneural gene neurog1 is more sensitive than the late proneural gene neurod4.

We assessed the ability of neural progenitors in the hindbrain to respond to Fgf20 signalling by generating a transgenic line [Tg(hsp70:fgf20a;fgf20a^−/−)] with heat-shock promotor controlling Fgf20a expression in the fgf20a^−/− background. The ubiquitous overexpression of fgf20a at 24-25 hpf resulted in upregulation of etv5b and spry1 in all hindbrain neural progenitors, including hindbrain boundary cells (Fig. S6). This result suggests that all hindbrain neural progenitors are receptive to Fgf20 signalling. Similar to the overactivation of Fgfr in Tg(hsp70:ca-xfgfr1), the ubiquitous expression of fgf20a led to significant downregulation of proneural gene expression (Fig. 3E). Together, these results suggest that sustained Fgf20 signalling is required to downregulate proneural gene expression in the segment centre. Neuronal organisation following a loss of proneural gene patterning, either from the overexpression of fgf20a or loss-of-function of fgf20a, was markedly disorganised compared with WT embryos (Fig. S7).

The Fgf20-dependent downregulation of proneural gene expression is accompanied by the upregulation of several markers in the segment centre (Gonzalez-Quevedo et al., 2010; Tambalo et al., 2020), including those associated with gliogenesis, such as sox9b and metrnla (Fig. S15) (Lee et al., 2010; Nishino et al., 2004; Stolt et al., 2003). This points to the possibility that a function of Fgf20 signalling in the hindbrain is to regulate the transition from neurogenesis to gliogenesis. The study of gliogenesis has been challenging, as canonical glial markers are often also expressed in pluripotent neural stem cells (Than-Trong and Bally-Cuif, 2015). For example, glial fibrillary acidic protein (gfap) labelled a subset of progenitors committed to neuronal differentiation (Fig. S9). We also examined the expression of glutamine synthetase (GS), a marker of mature astroglia, and did not detect a difference between WT and fgf20a^−/− embryos (Fig. S10). However, as previously suggested, oligodendrogenesis could be a target of Fgf20 signalling (Chandran et al., 2003; Esain et al., 2010; Kessaris et al., 2004).

Olig2 is a key transcription factor that regulates the specification of both oligodendrocytes and somatic motor neurons (Li et al., 2011; Park et al., 2002; Rowitch et al., 2002). OPCs can be identified with specific markers such as Sox10, which is regulated downstream of Olig2 and required for the terminal differentiation of OPCs and regulation of myelin genes (Claus Stolt et al., 2002; Küspert et al., 2011; Liu et al., 2007; Pozniak et al., 2010). Previous studies have shown that sox9, a key factor in OPC specification, is downstream of Fgf20 signalling (Gonzalez-Quevedo et al., 2010; Stolt et al., 2003). Furthermore, Esain et al. (2010) found that Fgfr activation is required to specify OPCs and Alcama⁺ abducens motor neurons in the zebrafish hindbrain by promoting olig2 and sox9 expression (Zannino and Appel, 2009). Crucially, the timing of when the Fgfr pathway is required for this process overlaps with the period of Fgf20 signalling. We therefore hypothesised that Fgf20 is the ligand responsible for regulating olig2 expression in the segment centre, and thus the production of OPCs and motor neuron subtypes from olig2⁺ progenitors.

We found that olig2 expression is reduced but not abolished in fgf20a^−/− embryos. At 20 hpf, olig2 was expressed homogeneously in a ventral layer 1-2 cells thick in r5 and r6 (Fig. S11A). This suggests that Fgf20a signalling is not required for the initiation of olig2 expression in the ventral hindbrain. We investigated whether Fgf20b contributes to the remaining olig2 expression observed in fgf20a^−/−. Multiplexed HCR RNA-FISH in fgf20a^−/− for olig2, fgf20b and etv5b showed that olig2 expression did not correlate with the region of Fgf20b signalling (Fig. 4A). Instead, olig2 was initiated within a ventral region of Fgfr pathway activation, indicated by expression of etv5b, independent of Fgf20b signalling (Fig. 4A). olig2 expression in fgf20a^−/− embryos can be further inhibited by treatment with the FGFR tyrosine kinase inhibitor SU5402 (Fig. S14). Together, these results suggest that the initiation of olig2 expression in the ventral hindbrain is largely independent of Fgf20 signalling. Nonetheless, we found that there are altered levels, pattern and dynamics of olig2 expression in fgf20a^−/− embryos. In a time course analysis, we found that olig2 expression was upregulated at 20-36 hpf in both WT and fgf20a^−/− embryos (Fig. 4B-D). Quantification of olig2 in r5 and r6 shows that the expression levels in fgf20a^−/− embryos were significantly lower than in WT embryos at all time points assessed (Fig. 4B-D; Fig. S11B), suggesting that Fgf20 signalling promotes olig2 expression. In addition to the ventral expression, a distinct olig2⁺ region occurred in the dorsal hindbrain from around 28 hpf onwards. Unlike the ventral expression, this was unaffected in fgf20a^−/− embryos or by fgf20a overexpression (Fig. 6D; Fig. S11D), and thus Fgf20 modulated olig2 expression exclusively in the ventral hindbrain. Crucially, analysis of the expression dynamics in r5 and r6 of WT embryos revealed a rapid, 6.5-fold increase in olig2 upregulation over the course of 2 h, between 26 and 28 hpf in WT embryos (Fig. 4F). This occurred ∼2 h after the peak of Fgf20 signalling (Fig. 4E; Fig. S11C). In contrast, olig2 expression in fgf20a^−/− embryos lacked such rapid upregulation, with a maximum of 2.2-fold increase between 26 and 28 hpf. This suggests that Fgf20 signalling upregulates ventral olig2 expression in r5 and r6 (Fig. 4G).

In addition, olig2 expression occurred in the segment centre of r4 and r7 by 36 hpf and 26 hpf, respectively (Fig. 4B,H; Fig. S11E), after the initiation of Fgf20 signalling. Furthermore, no olig2 expression was detected in r4 of fgf20a^−/− embryos, while only low level expression was detected in r7 by 33-36 hpf – a 10 h delay and 3-fold less than the expression level in WT embryos (Fig. 4B,H; Fig. S11E). This suggests that Fgf20 signalling is essential for the timely expression of olig2 in ventral r4 and r7. Interestingly, we observed an increase in the upregulation of olig2 expression in r5 and r6 of fgf20a^−/− embryos between 33 and 36 hpf (Fig. 4D,F), whereas in WT embryos olig2 expression stabilised from 28 hpf onwards. This suggests a compensatory mechanism in fgf20a^−/− embryos to restore olig2 expression levels. Nonetheless, the later upregulation did not result in a well-defined expression pattern in the segment centre, unlike WT embryos (Fig. 4B). Interestingly, we observed that olig2 was ectopically upregulated at the r5/6 boundary in fgf20a^−/− embryos by 36 hpf (Fig. S11E), suggesting that boundary cells may have a role in the compensatory mechanism.

Finally, we assessed whether the observed increase in olig2 expression is an indirect consequence of Fgf20-dependent proliferation (Kosaka et al., 2006; Lahti et al., 2011; Ma et al., 2009; Raballo et al., 2000). We performed EdU-assay (Fig. S12A-C), phospho-histone H3 (Fig. S12D) and Mki67 staining (Fig. S12E-G) in the hindbrain from 20-42 hpf. We found that proliferation was not patterned within that period, and no difference in proliferation was observed between WT, fgf20a^−/− and embryos with fgf20a overexpression.

Given the role of Fgf20 in modulating olig2 expression, we next asked to what extent it regulates the timing and production of oligodendrocytes and alcama⁺ motor neurons in the hindbrain. Alcama⁺ motor neurons in the zebrafish hindbrain are generated from olig2⁺ precursors at ∼36 hpf (Zannino and Appel, 2009). The timing of Fgf20 signalling is compatible with it being the primary source of Fgf signalling necessary for the production of these motor neurons (Esain et al., 2010). However, quantification of alcama in embryos with normal (WT), low (fgf20a^−/−) and high [Tg(hsp70:fgf20a;fgf20a^−/−)] levels of Fgf20 signalling showed that Fgf20 signalling has only a small, though significant, effect on the production of alcama⁺ motor neurons (Fig. S13A). In contrast, inhibition of Fgfr activation at 20-30 hpf with SU5402 resulted in a major reduction in the quantity of alcama⁺ motor neurons (Fig. S14). Timed pulses of inhibition in fgf20a^−/− embryos from 16 to 24 h shows that the key window of Fgfr pathway activation that is crucial for alcama⁺ motor neurons is 20-24 hpf (Fig. S13B). This suggests that although Fgf20 signalling has some influence on the production of alcama⁺ motor neurons, Fgfr pathway activation independent of Fgf20 is responsible for their specification and production.

Next, we examined the effects of the amount of Fgf20 signalling on the production of OPCs. In WT embryos, Sox10⁺ OPCs were detectable in the posterior hindbrain from ∼42 hpf (Fig. 5A). We found that the number of OPCs in fgf20a^−/− embryos was 7-fold less than in WT embryos (Fig. 5B). Remarkably, overexpression of fgf20a throughout the hindbrain at 24-30 or 24-33 hpf resulted in a 9-fold increase in the number of OPCs (Fig. 5B). Furthermore, the overexpression of fgf20a also resulted in precocious specification of OPCs by 32 hpf (Fig. 5A). These results suggest that the number of Fgf20-expressing cells influences both the quantity and timing of OPC specification. Parallel to the gliogenic effect, in alignment with earlier observations on proneural gene expression, fgf20a overexpression resulted in a substantial reduction in neuronal production, as exemplified by the marked decrease in thickness of the mantle zone (Fig. 5C). This points to the overarching pro-gliogenic functions of Fgf20 signalling in the developing hindbrain.

To examine whether the effect of Fgf20 on OPCs had the corresponding effects on oligodendrocyte production, we next quantified the number of oligodendrocytes at 96 hpf using the marker proteolipid protein 1a (plp1a) (Aggarwal et al., 2011) (Fig. 5D). Indeed, we found that the number of plp1a⁺ oligodendrocytes in fgf20a^−/− was half that of WT embryos, while in embryos overexpressing fgf20a we observed twice as many (Fig. 5E). This shows that the role of Fgf20 signalling in OPC specification is important for later plp1a⁺ oligodendrocyte production. Thus, the extent of Fgf20 signalling determines the number of oligodendrocytes produced in the hindbrain.

We next used HCR RNA-FISH to probe how Fgf20 signalling influences the spatiotemporal dynamics of sox10 and olig2, markers that are essential for OPC specification. We examined their expression in WT, fgf20a^−/− and Tg(hsp70:fgf20a;fgf20a^−/−) embryos at 36 and 42 hpf, time points when alcama⁺ motor neurons are produced and Sox10⁺ OPCs start to be detected, respectively (Fig. 6A,B). In WT embryos, sox10 expression was detected in the segment centres by 36 hpf, within a subset of olig2⁺ progenitors (Fig. 6A). Despite the observation that olig2 levels were highest in r5 and r6 at this stage, sox10 was most strongly expressed in r7 and co-expressed with olig2. This reflects the segmental differences in the lineage propensity for OPCs and the heterogeneity in cell fates among olig2⁺ progenitors in r5 and r6. In contrast, sox10 was not expressed in r4-r6 of fgf20a^−/− at 36 hpf, while a low level of expression was detected in r7. This is congruent with the observation in WT embryos, where r7 appeared to have a higher propensity for sox10 expression. By 42 hpf, a low level of sox10 expression was detected in a subset of olig2⁺ progenitors in r5-7 of fgf20a^−/−, but continued to be absent in r4 (Fig. 6C). This suggests that Fgf20 signalling is necessary for the timely expression of sox10 in olig2⁺ progenitors in the ventral hindbrain.

Congruent with the earlier observation from immunostaining of Sox10, ubiquitous overexpression of fgf20a from 20 to 36 hpf resulted in the precocious specification and differentiation of sox10⁺ OPCs among olig2⁺ progenitors (Fig. 6A). A shorter, but later, pulse of fgf20a overexpression of 6 h from 36-42 hpf resulted in major upregulation of both olig2 and sox10 expression among olig2⁺ progenitors (Fig. 6C,D), suggesting that fgf20a overexpression beyond the normal period of requirement of the signalling is sufficient to induce olig2 and sox10 expression in competent cells. Therefore, Fgf20 signalling has an instructive role on the dynamics of a gene regulatory network involving olig2 and sox10 (Liu et al., 2007). Together with the inhibition of proneural gene expression, this suggests that Fgf20 signalling induces a switch from neurogenesis to oligodendrogenesis in the developing hindbrain (Fig. 7).

In this study, we provide an example in the vertebrate nervous system in which precise tissue patterning and the induction of late-born cell types are regulated by a short-range signal emanating from earlier-born neurons. Such mechanisms provide temporal checkpoints and are particularly valuable for neural development at later stages where refined spatial and temporal patterning of pre-patterned progenitor pools is needed to further increase cell type diversity. In the zebrafish hindbrain, fgf20-expressing neuronal clusters pattern progenitors spatially and temporally to induce the neurogenesis-to-oligodendrogenesis switch in the centre of segments. The signal inhibits neurogenesis in the segment centre through downregulation of proneural genes, while inducing oligodendrogenesis among competent cells in that region through upregulation of OPC specification factors olig2 and sox9 (Gonzalez-Quevedo et al., 2010), and differentiation factor sox10. We show that Fgf20 is instructive for oligodendrogenesis as the amount of signalling defines the timing of induction and quantity of oligodendrocytes produced.

Whereas other members of the Fgf family act at long-range and form gradients in the nervous system (Yu et al., 2009), Fgf20-signalling does not appear to pattern hindbrain segments as a morphogen. The striking spatial correlation between the position of Fgf20-expressing neurons and progenitors with Fgfr pathway activation suggests that Fgf20 signalling acts at a short-range in the hindbrain. We have shown by ubiquitous overexpression of the ligand that all hindbrain progenitors are able to respond to Fgf20 signalling. The expression pattern of the reporter genes indicate that the signalling is likely to be restricted to the basal region of the progenitors, which are proximal to the signalling source. It is unclear whether Fgf receptors are localised at a specific region along the apical-basal axis of the progenitors. Nonetheless, it has been shown previously that the high tendency of Fgf20 to homodimerise is a dominating factor in determining the signalling activity and diffusion range (Harada et al., 2009; Kalinina et al., 2009; Liu et al., 2017). The homodimerisation acts as an auto-inhibitory mechanism by occluding the receptor binding site and restricting the diffusion range by increasing affinity for heparan sulphate proteoglycans (HSPGs). This is unique to the ligands of the Fgf9/16/20 subfamily, in particular Fgf9 and Fgf20. These properties provide a potential explanation for the highly restricted spatial expression of Fgf20-regulated genes. It also points to the possibility that Fgf20 is a weak signal, as the monomeric form is required for receptor binding.

The production of a short-range signal from a localised source – comprising no more than 8-10 cells arranged in a cluster – is an effective way of achieving precise patterning within the spatial confines of pre-patterned tissues. This is exemplified by the highly restricted expression of proneural genes along both the AP and DV axes of hindbrain segments regulated by the fgf20-expressing neuronal clusters. This, in turn, impacts the number and proportion of cell types produced, as well as the structural organisation of the tissue. Intriguingly, it has also been suggested that, in the mouse neocortex, earlier-born neurons secrete Fgf9 to regulate the timing of the switch from neurogenesis to gliogenesis (Seuntjens et al., 2009). This points to the possibility of a common strategy of vertebrate nervous systems in using short-range Fgf signalling to regulate developmental processes that would benefit from high spatial precision. Nonetheless, it is unclear whether the fgf9-expressing neurons regulate tissue patterning in the mouse neocortex, as occurs in the hindbrain. The mechanism for the onset of Fgf20 expression in selective nascent neurons in the hindbrain segment centre is not known. Sip1 is a potential candidate, as Seuntjens et al. (2009) have shown that this factor is upstream of fgf9 expression in the earlier-born neurons of the mouse neocortex.

Another potential significance of short-range localised signals in regulating neuroepithelial development is its ability to influence the timing of cell type formation when factoring in the involvement of interkinetic nuclear migration (INM). Indeed, it has been described in the zebrafish retina (Del Bene et al., 2008) that INM is crucial for modulating the rate of neurogenesis across the progenitor populations by bringing the cell bodies closer to or away from the apically located Notch signalling. An interesting avenue for future investigation is the potential influence of the rate of INM of hindbrain progenitors on the temporal dynamics of Fgf20-mediated gliogenesis.

This study has provided evidence that a key function of Fgf20 signalling in the hindbrain is to induce glial cell fate. We found that Fgf20 is an instructive signal for oligodendrogenesis as it controls both the timing and the proportion of oligodendrocytes produced. The commitment of both OPCs and motor neurons from olig2⁺ can be attributed to the upregulation of olig2 (Sagner et al., 2018; Scott et al., 2021; Wegener et al., 2015). It has been suggested, through single cell RNA-sequencing, that the maturation of OPCs is coupled with an increase in the level of olig2 (Scott et al., 2021). Furthermore, gain-of-function of Olig2 in OPCs can result in precocious differentiation of these cells (Wegener et al., 2015), likely through Olig2 regulation of sox10 expression (Küspert et al., 2011; Pozniak et al., 2010). In the zebrafish hindbrain, the initiation of olig2 expression in the ventral hindbrain does not require Fgf20 signalling. However, Fgf20 is necessary for further upregulating olig2 expression and the timely induction of OPCs. This process appears to be delayed in the loss-of-function of Fgf20, whereas overexpression of Fgf20 leads to precocious specification of OPCs. Expression of olig2 in Fgf20 loss-of-function mutants has dynamics that suggests a compensatory mechanism is in place to restore the expression level. Nonetheless, the number of oligodendrocytes is ultimately significantly reduced, potentially due to the constraint of a specification time window (Economou et al., 2022). This highlights the role of Fgf20 signalling in ensuring the induction of sufficient OPCs within a time window to attain the correct proportion of oligodendrocytes.

The ubiquitous overexpression of Fgf20 in the hindbrain only induces ectopic oligodendrogenesis among competent progenitors in a ventral domain. In particular, a short burst of Fgf20 overexpression leads to overexpression of olig2 and sox10 exclusively in ventral olig2⁺ progenitors. This suggests that Fgf20 may be acting on progenitors pre-patterned by Shh signalling along the DV axis. The interplay between Fgf and Shh signalling pathway in oligodendrogenesis has been previously explored. Kessaris et al. (2004) demonstrated in vitro that although Fgf can induce OPCs independent of Shh, ligand-independent Fgfr activation is required for Shh-mediated OPC induction. It has been shown that in the chick spinal cord Fgfr activity is required cell-autonomously in Shh-induced olig2⁺ progenitors to maintain high level Olig2 expression for the commitment of OPCs (Farreny et al., 2017). In amniotes, Fgf can promote olig2 expression and OPC induction independent of Shh in the dorsal neural tube (Chandran et al., 2003; Fogarty et al., 2005; Naruse et al., 2006; Vallstedt et al., 2005). However, the olig2-expressing domains in the dorsal hindbrain are not affected by Fgf20 overexpression.

We observed that the level of Fgf20 signalling affects alcama⁺ motor neuron production with the same directionality as OPC induction, but with a much lesser magnitude of change. Although in this study we have not investigated the role of Fgf20 specifically on bipotential olig2⁺ progenitors (Zannino and Appel, 2009), one potential explanation for the mild effects of Fgf20 levels on the production of alcama⁺ motor neurons is the opposing effect of the signal in promoting olig2 expression while downregulating proneural gene expression. An increase in olig2 expression could lead to an increase in the number of alcama⁺ motor neurons, as olig2 promotes proneural gene expression (Mizuguchi et al., 2001; Novitch et al., 2001). However, this effect could be cancelled out by Fgf20-mediated downregulation of proneural genes. Nonetheless, this highlights the gliogenic nature of the signal. Indeed, as shown in previous studies, Fgf20 upregulates the gliogenic factor sox9 in the segment centre (Gonzalez-Quevedo et al., 2010). Sox9 has key roles in the specification of astroglia and oligodendrocytes (Poché et al., 2008; Stolt et al., 2003) and has been implicated in the neurogenesis-to-gliogenesis transition (Vong et al., 2015). Crucially, in mouse, sox9 overexpression can invert the usual sequence of olig2 upregulation followed by sox10 (Vogel et al., 2020). The present study suggests that Fgf20 acts as a key regulator of the GRN for oligodendrogenesis in the ventral hindbrain to promote the production of oligodendrocytes. Future studies that explore further glial markers will help further understanding of the role of Fgf20 in regulating the neurogenesis-to-gliogenesis transition. As etv5b expression is upregulated by ERK1/2 downstream of Fgfr activation, it will also be interesting to determine whether ERK1/2 and/or other targets of Fgf signalling are involved in the switch to gliogenesis.

Wild-type Danio rerio, fgf20a^−/− (Whitehead et al., 2005), efnb3b^−/− (Cayuso et al., 2019), Tg(hsp70:ca-xfgfr1) (Marques et al., 2008) and Tg(hsp70:fgf20a;fgf20a^−/−) were used in this study. Fgf20a^−/− line (fgf20a^zp3) contains a point mutation in fgf20a resulting in a Y148S amino acid substitution at the β-trefoil region which renders the protein null in function (Whitehead et al., 2005). All mutant embryos used in this study were obtained from homozygous in-crosses. Genotyping of the parents of fgf20a^−/− was performed using a Derived Cleaved Amplified Polymorphic Sequences (dCAPS) assay (https://zebrafish.org/fish/pdf/pcr/zp3.pdf). Genotyping of the parents of Tg(hsp70:fgf20a;fgf20a^−/−) embryos was performed by fluorescent microscope screening for cmlc2:EGFP expression in progeny embryos. Genotyping of the parents of Tg(hsp70:ca-xfgfr1) was performed by fluorescent microscope screening for acr:RFP expression in progeny embryos. The homozygosity and efficacy of the transgenic lines were further verified by NBT/BCIP in situ hybridisation for etv5b expression in heat-shocked progeny embryos. Embryos were incubated at 31°C, 28.5°C, 25°C or 22.5°C to regulate the rate of development. They were staged according to the Zebrafish Developmental Staging Series on ZFIN. Embryos at 16-22 hpf were staged according to somite numbers; embryos at 24-42 hpf were staged according to morphological and physiological features. Pigmentation was inhibited by incubation in embryo medium containing 0.003% PTU (1-phenyl 2-thiourea) from 24 hpf.

pDestTol2-hsp70:Fgf20a-P2A-H2B-Citrine;cmlc2:EGFP was created by replacing ca-Bmpr1 coding sequence from pDestTol2-hsp70:ca-Bmpr1a;cmlc2:EGFP using the restriction enzyme KpnI (Bielen and Houart, 2012) by a 2010 bp sequence consisting of Fgf20a(stop codon removed)-P2A-H2BCitrine, assembled synthetically using Invitrogen GeneArt. The Tg(hsp70:fgf20a;fgf20a^−/−) line was generated by injecting the DNA construct pDestTol2-hsp70:Fgf20a-P2A-H2B-Citrine;cmlc2:EGFP (10 ng/μl working concentration) with Tol2 transposase RNA (30 ng/μl working concentration) into one-cell stage embryos from the fgf20a^−/− line. The injection volume was 1.5-2 nl. H2B-Citrine is not expressed in the resulting transgenic line.

mRNA of EGFP-CAAX was synthesized from pCS2-EGFP-CAAX plasmid, linearised with XmnI and transcribed with SP6 polymerase. Then 30 ng/μl of working concentration of mRNA was used for injection into zebrafish embryos at the one- to four-cell stages. Embryos were fixed in 4% paraformaldehyde at the desired stages and stored in 100% methanol at −20°C before proceeding to HCR RNA-FISH.

All heat-shocks were performed at 39°C for 1 h in a Petri dish. E3 medium was preheated to 39°C before the heat-shock procedure. PTU was added to the medium for treating embryos beyond 24 hpf. At the end of the heat-shock, the E3 medium in the dish was replaced with E3 medium at room temperature (25°C) and transferred into a 28.5°C incubator. For the protocol that requires prolonged overexpression of the transgenes, repeated heat-shocks were performed. In experiments using Tg(hsp70:caxfgfr1), heat-shock was performed every 3 h. In experiments using Tg(hsp70:fgf20a;fgf20a^−/−), heat-shock was performed every 2 h.

In situ hybridisation and colour development with NBT/BCIP (Roche) were performed as previously described (Xu et al., 1994). The DIG riboprobes for glula mRNA were synthesized from PCR products amplified from cDNA of embryos at 30 hpf and 48 hpf, using reverse primers with a T7 promoter site (5′gaaatTAATACGACTCACTATAGg3′). Primers were: forward 5′-GAGATCACTTGTGGGTAGCTC-3′, reverse 5′-GAAATTAATACGACTCACTATAGGGGAGGGAAATTCAGTCCAGTAA-3′.

Embryos were fixed in 4% paraformaldehyde at room temperature for 3 h and stored in 100% methanol at −20°C for at least 24 h before HCR RNA-FISH or immunostaining. HCR RNA-FISH was performed according to the whole-mount zebrafish embryos and larvae protocol available on the website of Molecular Instruments (molecularinstruments.com). Following rehydration with a graded methanol wash of 75%, 50% and 25% diluted in PBST (PBS with 0.1% Tween 20), proteinease K treatment (10 μg/ml) was performed on embryos at 24 hpf or above. The duration of the treatment ranged from 4 to 30 min, dependent on the stage of the embryos. Post-fixation was performed with 4% paraformaldehyde at room temperature for 20 min.

The following HCR RNA-FISH probe sets, designed and synthesised by Molecular Instruments, were used in the study: fgf20a-B1, fgf20b-B5, etv5b-B2, spry1-B5, neurog1-B3, neurod4-B5, ascl1a-B1, ascl1b-B4, olig2-B3, sema3gb-B4, plp1a-B4, sox10-B5, alcama-B1, mki67-B1, metrnla-B3. All probe sets were 20 bases in length and generated based on Danio rerio mRNA sequences. HCR amplifiers corresponding to the initiator sequence with Alexa Fluor 488, 514, 546 and 647 were used for detection. Embryos were incubated with the probe for 14 h at the detection stage, then with the fluorescent HCR amplifiers for 12 h at the detection stage.

In experiments with double staining with HCR RNA-FISH and antibodies, immunofluorescence staining was performed after HCR RNA-ISH in 5× SSCT (saline-sodium citrate buffer with 0.1% Tween 20) instead of PBS/PBST. Blocking solution with 5× SSC/0.1% Tween 20 containing 5% goat serum and 1% DMSO were used. Samples were washed and stored in 5× SSC/0.1% Tween 20 before imaging.

The following primary antibodies were used in this study: anti-HuC/D (mouse monoclonal; Invitrogen A-21271; 1:400); anti-Gfap (rabbit polyclonal; Genetex GTX128741; 1:200); anti-EphA4 (rabbit polyclonal; Irving et al., 1996; 1:500); anti-phospho-histone H3 (Ser10) (rabbit polyclonal; Sigma-Aldrich 06-570; 1:200); anti-Sox3 (rabbit polyclonal; Genetex GTX132494; 1:300); anti-Sox10 (rabbit polyclonal; Genetex GTX128374; 1:200). Anti-HuC/D, anti-Gfap and anti-phospho-histone H3 were found to be compatible with HCR RNA-FISH, and anti-EphA4 and anti-Sox3 to be incompatible with HCR RNA-FISH. The following secondary antibodies were used for detection: Alexa Fluor 488 goat anti-rabbit (Invitrogen, A11008; 1:400); Alexa Fluor 488 goat anti-mouse (Invitrogen, A11029; 1:400); Alexa Fluor 546 goat anti-mouse (Invitrogen, A11030; 1:400); Alexa Fluor 546 goat anti-rabbit (Invitrogen, A11071; 1:400); Alexa Fluor 647 goat anti-rabbit (Invitrogen, A31634; 1:400). We used 0.1% Tween 20 in the blocking and washing buffer of all the aforementioned antibody stainings, except anti-Sox10, for which 0.1% Triton X-100 was used. For anti-Sox10 staining, an additional acetone fixation/permeabilisation step was performed after proteinase K treatment. Embryos were placed in 100% acetone at −20°C for 10 mins for anti-Sox10 staining and rinsed in H₂O before and after the treatment.

The Click-iT EdU Cell Proliferation Kit for Imaging with Alexa Fluor 647 dye (Invitrogen, C10340) was used for detection of cells in their S-phase. To label cells with EdU, ∼25 dechorionated embryos were incubated in 1 ml of ice-cold E2 medium with 500 μM of EdU/10% DMSO in a 2 ml microcentrifuge tube on ice for 20 min. The incubation was performed on an orbital shaker at very low rpm. Embryos were transferred to E2 medium at 28.5°C for 15 min for recovery followed by fixation with 4% paraformaldehyde for 3 h at room temperature. Fixed embryos were stored in PBST at 4°C for the later detection of EdU. Permeabilisation was performed before EdU detection. Embryos were placed in 100% acetone at −20°C for 10 min and they were rinsed with H₂O before and after this treatment. This was followed by 1 h of washing in PBS/1%Triton X-100/1% DMSO. For EdU detection, 500 μl of Click-iT reaction cocktail was added to each tube for 30 min incubation at room temperature, followed by two washes in PBST. Immunofluorescence staining was performed after the EdU detection.

Embryos were dechorionated and treated at the specified stages with 6 μM SU5402 (Tocris; 3300). Preliminary data suggested that this concentration is sufficient to inhibit the Fgfr-pathway, indicated by the lack of etv5b expression by in situ hybridisation. Concentrations above 10 μM are toxic to the embryos. We included 0.003% PTU in treatments of embryos beyond 24 hpf. Embryos were kept in 3 ml E2 embryo medium in low-binding six-well plates during the treatment. For long treatments, the medium was changed every 6-8 h. At least two experiments were performed with 12-15 biological repeats (embryos). All replicates were visually screened on confocal microscopes, with a subset imaged with z-stack of the entire volume of the hindbrain.

Hindbrain samples were flat-mounted on slides in 70% glycerol. Lateral sections (LS) and transverse sections (TS) were generated by imaging manually dissected embryos to expose the desired axis. Unless specified, images were generated from whole-mount samples. Fluorescent-labelled samples were imaged with a 30×/1.05 NA silicone oil immersion objective (Olympus UPLSAPO30XS) on an upright or inverted Olympus FV3000 confocal laser scanning microscope. Images were acquired as 12-bit z-stacks of 1024×1024 resolution, z-step size of 1.5 μm and three times averaging. As silicon oil (n=1.405) has a refractive index closer to 70% glycerol (n=1.428) than the typical immersion oil (n=1.518), spherical aberration-based axial distortion is minimised (Diel et al., 2020). This enables better resolution and deeper 3D imaging. Chromogenic in situ hybridisation samples were imaged with 20×/0.75 NA objective on a Zeiss Axioplan2. All images are representative of imaging, biological and experimental replicates.

All confocal z-stack images shown were visualised and generated using the rendering tools ‘Ortho Slicer’ or ‘Oblique Slicer’ in Imaris software; resulting in maximum intensity projections from a selected subset of the image data. The quantifications of the confocal z-stack images were performed using FIJI software with semi-automated macro scripts and then further processed and plotted in Python. The image analysis pipeline is as follows. (1) Background subtraction – background value was measured by manually selecting a region within the tissue on a representative slice, followed by subtraction of this value in all slices of the z-stack image. It was verified that this measurement was consistent across slices. (2) Sum projection – generating an image with pixel values that are the sum of all pixels with the same xy coordinates in the z-stack image. (3) Measure pixel intensity – this is performed either by measuring the average pixel intensity in an area of the image defined by a rectangular area, or as a line profile along the AP axis (y-axis) of the hindbrain using the ‘Plot Profile’ function in FIJI. In the former method, total grey values of the measured region are generated by the multiplication of the average pixel intensity and the area. The latter method results in a profile plot of grey values along the y-axis that are the average pixel intensity of the x-axis of the sum projected z-stack image. The line thickness is manually set to include only the regions with expression. The latter method is only suitable for expression patterns without significant variation along the medial-lateral axis, and it emphasizes the difference in expression profile along the AP axis. To generate the mean profile plot with standard error of the mean (s.e.m.) from multiple datasets, the datasets were first aligned by trimming the length and re-scaled to a representative sample using Python. The mean profile plot and s.e.m. is then generated by calculating the moving average in an interval size of 1% of the length of the dataset. The result data is denoted as ‘z-sum profile grey value’ in the plots. Statistical significance was determined using Mann–Whitney U-test in Python with SciPy.

Box plots (or box-and-whiskers plots) were used to visualise and summarise the distribution of the quantification data from HCR signal. The box ranges from the lower (Q1) to the upper (Q3) quartile. The difference between Q3 and Q1 is the interquartile range (IQR), which describes the spread of the data. A line across the box marks the median. The whiskers are lines extending from Q1 and Q3 to the minimum and maximum data point within Q1−1.5×IQR and Q3+1.5×IQR, respectively. Data points beyond that range are defined as outliers.

We thank Monica Tambalo, Jordi Cayuso, Qiling Xu and Megan Addison for their advice on experimental techniques, Corinne Houart for heat-shock promoter and EGFP-CAAX constructs, Caroline Hill for resources and continuous support, Edgar Herrera Delgado for useful discussions on image analysis and advice on statistical analysis, and François Guillemot and Scott Wilcockson for helpful comments on the manuscript. We thank the Crick Advanced Light Microscopy and the aquatics unit of the Crick Biological Research Facility for their consistent and excellent support.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.204256.reviewer-comments.pdf