Actin-based protrusions lead microtubules during stereotyped axon initiation in spinal neurons in vivo

In vitro, developing neurons progress through well-defined stages to form an axon and multiple dendrites. In vivo, neurons are derived from progenitors within a polarised neuroepithelium and it is not clear how axon initiation observed in vitro relates to what occurs in a complex, three-dimensional in vivo environment. Here we show that the position of axon initiation in embryonic zebrafish spinal neurons is extremely consistent across neuronal sub-types. We investigated what mechanisms may regulate axon positioning in vivo and found that microtubule organising centres are located distant from the site of axon initiation in contrast to that observed in vitro, and that microtubule plus-ends are not enriched in the axon during axon initiation. F-actin accumulation precedes axon formation and nascent axons form but are not stabilised in the absence of microtubules. Laminin depletion removes a spatial cue for axon initiation but axon initiation remains robust.


INTRODUCTION
During development neurons polarise to form axonal and dendritic domains. This is essential for circuit formation and for the directional movement of information through the nervous system, but many of the fundamental mechanisms that initiate and build axons and dendrites are not understood.
In vitro, dissociated rodent hippocampal neurons progress through well-defined stages to develop one axon and multiple dendrites ( (Dotti et al., 1988); reviewed in ).
The neuron first extends several neurites, any of which has the potential to become an axon. Neuronal polarisation occurs when one neurite is specified to become the axon and grows longer and faster than the others, which subsequently become dendrites.
The question of which neurite becomes the axon has been studied extensively.
Proteins associated with axon specification (also called axon determinants, including PIP3, LKB1 and the Par complex) move between neurites before becoming persistently localised in one neurite, which subsequently becomes the axon (Shi et al., 2003;Ménager et al., 2004;Yoshimura et al., 2006;Barnes et al., 2007;Shelly et al., 2007). Their downstream signalling pathways reinforce axonal identity and induce cytoskeletal changes. Neuronal polarisation can occur stochastically in the absence of extracellular factors, but environmental cues such as growth and guidance factors (Hilliard and Bargmann, 2006;Wolman et al., 2008;Yi et al., 2010;Cheng et al., 2011;Nakamuta et al., 2011;Shelly et al., 2011) and extracellular matrix proteins (Gomez and Letourneau, 1994;Esch et al., 1999;Ménager et al., 2004;Randlett et al., 2011) can position axogenesis by regulating the localisation of axon determinants. However, one caveat of using this in vitro system is that axon initiation is defined by the growth and specification of an already existing neurite. Indeed, axon growth is required to identify neuronal polarisation and axon initiation in this context (Goslin and Banker, 1989;Yamamoto et al., 2012). As such, it is difficult to determine what is actually required for axonal specification and what simply promotes axonal growth (reviewed in (Barnes and Polleux, 2009)).
In vitro, an increase in intracellular trafficking towards the future axon early in neuronal polarisation is thought to be due to polarisation of the microtubule cytoskeleton and to cause the accumulation of axon determinants at the nascent axon (Bradke and Dotti, 1997). Polarisation of the microtubule array has been suggested to be driven by the asymmetric localisation of microtubule organising centres (MTOCs) -the centrosome and Golgi complex -which nucleate microtubules. Indeed, both the centrosome and Golgi complex have been described to localise to the base of the axon in hippocampal neurons (de Anda et al., 2005), and cerebellar granule cells in vitro (Zmuda and Rivas, 1998). However, other studies have described the centrosome as positioned randomly with respect to the axon in cultured hippocampal neurons (Dotti and Banker, 1991), or to localise to the future axon after it has developed (Gärtner et al., 2012). In vivo, the centrosome of Rohon-Beard neurons in the embryonic zebrafish spinal cord is on the opposite side of the cell to the central axons when they are initiated, but has been reported to then move to the base of the peripheral axon during its initiation (Andersen and Halloran, 2012). The centrosome is not close to the axon in zebrafish retinal ganglion cells (Zolessi et al., 2006) or tegmental hindbrain neurons (Distel et al., 2010), and neurons in Drosophila embryos without centrosomes are able to extend axons apparently normally (Basto et al., 2006).
We have recently shown that the well-defined stages of neuronal polarisation described in vitro do not occur in neurons in the embryonic zebrafish neural tube (Hadjivasiliou et al., 2019). The neuronal soma moves to the basal surface of the neuroepithelium while maintaining an attachment to the apical surface and then extends two long, transient protrusions rostrally and caudally along the basal surface of the neural tube. In contrast to the neurites observed in polarising neurons in culture (Dotti et al., 1988), these protrusions have stereotyped orientations, are transient and are fully retracted along with the apical attachment prior to axon extension. The axon is then initiated directly from the cell body rather than developing from a pre-existing neurite, and well before the appearance of dendrites.
In the current study we focused on the stage after basal protrusion retraction to investigate what regulates the initiation of axonal outgrowth in the embryonic spinal neurons. We used time-lapse imaging and observed that the axon is consistently initiated from the basal and ventral side of the soma. Laminin promotes the basal orientation of axon initiation but we were not able to perturb the ventral orientation of axon initiation, which is particularly robust. MTOCs are at the opposite side of the cell at the time of axon initiation and F-actin localises to the future axon initiation site before the accumulation of microtubule plus-ends. Nascent axons can form in the absence of microtubules. This study establishes embryonic zebrafish spinal neurons as a favourable model for investigating axon initiation in vivo and proposes F-actin accumulation as the primary cytoskeletal element for axon initiation.
Blocking was performed for two hours at room temperature in appropriate serum.
Imaging was performed from 16 hpf. Embryos were dechorionated, mounted in lowmelting point agarose (Sigma-Aldrich) and anaesthetised with MS-222 (Sigma-Aldrich/Merck) if required (Alexandre et al., 2010). Confocal imaging was performed on a spinning disk confocal (PerkinElmer, Waltham, U.S.A.) or LSM880 laser scanning confocal (Zeiss, Oberkochen, Germany) with or without Airyscan, using a 20x water immersion objective with numerical aperture of 0.95 or higher. For high resolution imaging we used Zeiss Airyscan acquisition and processing. Lightsheet imaging was performed on a Zeiss LightSheet Z.1 microscope using 10x illumination objectives and 20x water immersion detection objectives. If required, nocodazole (5 mg/mL stock in DMSO) was diluted in fish water to a final concentration of 5 µg/mL. Following treatment, nocodazole was washed out of the imaging chamber with fish water.
Images were acquired as 40-100 um deep z-stacks with or without time-lapse every 2 seconds to 10 minutes over 5-15 hours depending on the experiment. Images and videos in the manuscript result from maximum projections of z-stacks using Fiji (Schindelin et al., 2012) or 3D reconstructions using Volocity (Perkin Elmer).
Surrounding cells were occasionally edited from the field of view using ImageJ or Imaris (Bitplane, Belfast, U.K.) to more clearly show behaviours of the individual cells under investigation.
To analyse organelle position with respect to the cell centroid, the field of view was reoriented so the basal surface was to the right. The cell centroid was determined using the ImageJ 3D Object Counter plugin and the 3D coordinates of the site of axon initiation, centrosome or Golgi complex was determined manually.
Trigonometry was used to calculate the distance and angle of each organelle with respect to the cell centroid and this was analysed using Moore's modification of the Rayleigh statistical test. Trigonometry was also used to calculate the distance between two different organelles within the same cell. These positions were analysed using Moore's test for paired data. Organelle positions in different conditions were compared using Batschelet's alternative to the Hotelling test.
Distances between organelles in different conditions were analysed using Student's t-test except for centrosome-axon distance in Sly -/and lamMO-injected embryos, which were compared to WT using one-way ANOVA. Change in cilium length over time was measured using ImageJ and analysed using non-linear regression to compare each slope to 0 (Prism 8, GraphPad, San Diego, U.S.A.). Protrusion length, width and duration were measured using ImageJ and compared using Student's ttest. Fluorescence intensity analysis was performed using the ImageJ Plot Profile plugin. Distances between neuron and basal surface were compared using the Student's t-test.

Axon initiation in vivo is highly stereotyped
To observe neuronal polarisation in vivo, we sparsely and randomly labelled zebrafish embryonic spinal cord cells with a membrane marker and imaged them using time-lapse confocal microscopy from 16 hours post fertilisation (hpf). We previously reported that differentiating neurons in the zebrafish spinal cord go through a distinctive and very stereotyped T-shaped morphology before extending an axon ( Figure 1B; Video 01; (Hadjivasiliou et al., 2019)). To better understand axon initiation, we focussed on the time immediately after retraction of the basal Figure 1. Axon initiation is highly stereotyped and occurs at the baso-ventral aspect of the soma irrespective of subsequent axon trajectory. (A) Diagram to illustrate the three different views -dorsal, transverse and lateralshown in confocal images and 3D reconstructions throughout this paper. (B) Image sequence from confocal time lapse shows the early steps in neuronal differentiation. Two transient basal protrusions are extended along the basal surface of the neural tube (6h 26 to 14h 01; open arrows) and then retracted (15h 31 to 17h 31). The apical attachment is also retracted (17h 31 to 19h 53; e.g. -1h 30, arrowheads) before the axon is extended (20h 38 to 21h 50; closed arrows). Images are maximum projections (dorsal views) and transverse reconstructions (transverse views; insets) from confocal z-stacks. (C) Image sequence from confocal time lapse shows a neuron before, during and after axon initiation. Prior to axon initiation, the neuron extends multiple small, transient pre-axonal protrusions (-2h to -20m; arrowheads). The nascent axon is extended (0h) and maintained for a short period (0h to 30m) before axon growth begins (45m to 2h 20). Arrows show axon tip. Asterisk marks position of the base of the axon on the soma. Images are transverse reconstructions from confocal z-stacks. (D) Plots showing axon position on the soma (e.g. asterisk in Figure 1C  protrusions and apical attachment (e.g. from 19h53 in Figure 1B). At this initial stage neurons had no prominent or long-lasting protrusions but extended small, transient protrusions and filopodia ( Figure 1C -2h to -20m; Video 02). These transient protrusions occurred in many directions at first (e.g. Figure 1C, -1h30) but were gradually restricted to a more defined baso-ventral position on the cell body. Each neuron then extended a single axonal protrusion from this position ( Figure 1C, 0h).
This protrusion was stable and did not grow significantly for approximately 30 minutes ( Figure 1C 0h to 30m; Figure 1 -figure supplement 1) before extending rapidly ( Figure 1C 45m to 2h20). We defined the time of axon initiation as the first time point that showed a persistent protrusion that subsequently matured into an axon with growth cone ( Figure 1C, 0h). We call this persistent protrusion the nascent axon. We defined the site of axon initiation as the position that the nascent axon protrusion emerges from the cell body at this time point ( Figure 1C 0h, asterisk). The position of the nascent axon was analysed with respect to the cell centroid and found to be highly biased towards the baso-ventral quadrant of the soma ( Figure 1D).
Several different projection neuron subtypes arise during the first few hours of neurogenesis in the embryonic zebrafish spinal cord (Bernhardt et al., 1990;Hale et al., 2001). They can be distinguished by the dorso-ventral position of the soma within the spinal cord together with their axon trajectory ( Figure 1E). Our cell labelling method randomly targeted all of the early embryonic neuronal subtypes reported previously (Figure 1 -figure supplement 2A; (Bernhardt et al., 1990;Hale et al., 2001)). Of the 86 neurons that we analysed for the site of axon initiation ( Figure 1D . We find the site of nascent axon formation was baso-ventral for all of these neuronal subtypes ( Figure  1H-M). The only neuronal subtype whose axons are not circumferential are dorsal lateral ascending (DoLA) neurons, which project their axons rostrally towards the hindbrain ( Figure 1E; (Bernhardt et al., 1990)). Surprisingly however, DoLA neurons also had a baso-ventral location for their nascent axon ( Figure 1F transverse view 0h, G; Video 03). Time lapse imaging showed that, after basolateral nascent axon formation, the growth cone turned and grew rostrally to establish its characteristic axon trajectory ( Figure 1F lateral view 4h; Video 03).
Altogether, this confirms and quantifies our previous observation that axon initiation occurs directly from the neuronal soma (Hadjivasiliou et al., 2019). The first persistent axonal protrusion, which we define as the nascent axon, is formed at an extremely stereotyped position such that all analysed neuronal subtypes exhibited a strong baso-ventral bias of the position of nascent axon outgrowth. That this is consistent regardless of the subsequent axonal trajectory shows that axon initiation in the zebrafish spinal cord is a separate process that can be decoupled from axonal growth and guidance.

Pard3 and Cdh2 are not detected at the growth cone
Several polarity proteins localise to the growth cone in cell culture and have been implicated in axon specification, including Pard3 (Shi 2003) and Cdh2 (previously Ncadherin; (Gärtner et al., 2012)). To investigate the localisation of Pard3 during axon initiation we used a zebrafish Pard3-GFP reporter line (TgBAC(Pard3-GFP)). Time- We performed similar experiments using a zebrafish reporter line for Cdh2 (Tg(cdh2:cdh2-tFT)). We observed clear Cdh2 rings at the apical surface as expected (Symonds et al., 2020)

Centrosome behaviour prior to axon initiation
Previous studies have suggested that centrosome position is important for positioning the axon (de Anda et al., 2005;Andersen and Halloran, 2012). To get a comprehensive analysis of the centrosome leading up to and during axon initiation we monitored centrosome position from its initial location at the apical surface of the neuroepithelium through to the time of nascent axon establishment. We first focussed on the time when the neuron detaches from the apical surface (see Figure   1B, 17h31 to 19h 53). In ex ovo chick neural tube slices the centrosome remains at the apical surface during neuronal differentiation until abscission of the apical processes, when the centrosome is retracted back to the soma along with the apical process. The retracting process abscises from the apical endfoot, which is left behind at the apical surface of the neural tube along with the cilium (Das and Storey, 2014).
We too found that the centrosome in zebrafish spinal cord cells is retracted along with the apical process ( Figure 2A). However, unlike the chick spinal cord, most zebrafish spinal neurons do not show any apical abscission events ( Figure 2B; Video 04). We next used a zebrafish transgenic line with GFP-tagged cilia (Tg(arl13b-GFP)) and observed cilia in the zebrafish spinal cord during a period when we know that many neurons will be retracting their apical processes. We saw many examples of Cilium length did not change (P > 0.05 for n = 6/7 cilia; non-linear regression). (F) Distance between centrosome and cilium in Tg(arl13b-GFP) embryos fixed and processed for immunohistochemistry against GFP to label the cilium and γ-tubulin to label the centrosome. No difference was found in the distance between the two organelles when close to the apical surface or away from the midline (n = 50 cells per condition; P = 0.8279; midline mean = 0.4956, s.d = 0.1125; away from midline mean = 0.5004, s.d = 0.1077; Student's unpaired t-test).
cilia moving from the apical surface to close to the basal surface ( Figure 2C), reminiscent of apical retraction. In several instances we could follow a particular cilium continually for up to 45 minutes before it left the apical surface and then moved basally (n = 13 cilia). Cilium length did not change either while at the apical surface or while moving towards the basal surface ( Figure 2E). These data show that in zebrafish the cilia are retained by most spinal neurons during apical retraction rather than being abscised and regrown. Finally, we observed centrosomes and cilia at the same time by labelling centrosomes in the zebrafish cilium line. Time lapse videos showed the centrosome and cilia stay in close proximity both at the apical surface and as they moved together towards the basal surface ( Figure 2D; Video 05).
Immunohistochemistry also showed cilia and centrosomes close together in several positions along the apico-basal axis of the spinal cord, and close proximity was maintained no matter where along they were along this axis ( Figure 2F). Altogether this data shows that the centrosome and cilia do not physically dissociate but are retracted together within the apical process during the large majority of neuronal differentiation events in the zebrafish spinal cord. Since it travels in the apical pole of the retracting process, the centrosome locates to the apical pole of the neuronal soma at the end of this phase of differentiation.

The centrosome is located on the opposite side of the cell to the nascent axon
As apical retraction was completed the centrosome was close to the apical pole of the neuronal soma. To assess whether proximity of the centrosome is involved in axon initiation we next analysed centrosome position in neurons during establishment of the nascent axon. Time lapse imaging showed that the centrosome was not close to the position of the nascent axon ( Figure 3A 0m; Video 06). When the nascent axon is first identifiable, the mean distance between the centrosome and nascent axon was 10.1 µm ± 3.3 ( Figure 3C). To put this into perspective, the mean diameter of these cells' nuclei was 7.8 µm ( Figure 3C dotted line, s.d. = 0.7, n = 5 cells). To quantify the spatial relationship between centrosome and nascent axon we analysed their (A) Image sequence from confocal time lapse shows a neuron labelled with membrane (grey) and centrosome (green) markers before (-16m), during (0m) and after axon initiation (15m, 31m). The centrosome (open arrows) is located on the opposite side of the cell to the nascent axon (closed arrows). Images are transverse reconstructions from confocal z-stacks. (B) Image sequence from confocal time lapse shows a neuron labelled with membrane (grey) and Golgi complex (magenta) markers before (-17m), during (0m) and after axon initiation (17m, 37m). The Golgi complex (open arrows) is located on the opposite side of the cell to the nascent axon (closed arrows). Images are transverse reconstructions from confocal z-stacks. Finally, we used immunohistochemistry to investigate γ-tubulin location. γ-tubulin is highly likely to be required for microtubule nucleation in cells and so marks any potential MTOC (Moritz and Agard, 2001). We could only find obvious γ-tubulin accumulation at one concentrated point in each neuron that appeared to correspond with the centrosome (Figure 3 -figure supplement 2). Along with analysis of centrosome and Golgi positioning, this suggests that there is no potential MTOC close to the site of axon initiation.

Both centrosome and Golgi move to the base of the axon after its initiation
Some previous studies have shown that the centrosome and Golgi complex are located at the base of axons in cell culture (de Anda et al., 2005;Stiess et al., 2010) and the centrosome moves towards the basal side of the cell during peripheral axon extension in Rohon-Beard neurons (see Figure 3 -figure supplement 1; (Andersen and Halloran, 2012)). We hypothesised that the proximity of these MTOCs to the axon may reflect axon growth rather than axon initiation, so we analysed the position of the centrosome and Golgi complex during axon pathfinding, between six and twelve hours after establishment of the nascent axon ( Figure 3D, E). We found that both organelles had moved close to the base of the axon ( Figure 3C). The position of all of these organelles was at the baso-ventral side of the cell and was not random ( Figure 3H, I). Paired analysis of the positions of the centrosome or Golgi complex and base of the axon in the same cell showed that these were not different in most cases ( Figure 3H, I). Altogether our results show MTOCs are closely related to the base of the axon during axon growth but not during axon initiation.

Actin accumulation precedes enrichment of growing microtubule plus-ends during nascent axon formation
Although the location of MTOCs is not close to the site of nascent axon initiation it is still possible that microtubules are involved in specifying the nascent axon.
Alternatively actin accumulation may precede microtubules to specify this site, as has been observed for neurite initiation (Zhang 2016). To identify the earliest cytoskeletal component we investigated the localisation of various cytoskeletal markers in individual neurons before, during and after axon initiation.
A constitutively active version of the kinesin 1 motor domain, Kif5c560, is trafficked specifically on axonal microtubules and is an early axonal marker in vitro and in vivo (Jacobson et al., 2006;Randlett et al., 2011). We found that Kif5c560-YFP localised basally in neuroepithelial cells ( Figure 4A). In neurons it was fairly evenly localised To compare the timing of actin accumulation and the invasion of growing microtubules into the nascent axon we analysed the distribution of filamentous actin (F-actin) and EB3 simultaneously. EB3 is a microtubule plus-end binding protein that labels the tips of all growing microtubules. We mosaically expressed EB3-GFP mRNA to observe growing microtubules along with lifeact-Ruby mRNA to label Factin-based protrusions before, during and after axon initiation. We found that EB3 was localised throughout the cell before axon initiation ( Figure 4C -30m to -10m). In contrast to Kif5c560, however, most EB3 was still located in the soma during nascent axon establishment and only a few microtubule plus-ends were localised within the nascent axon itself ( Figure 4C 0h; Video 08). EB3 was subsequently enriched in the axonal growth cone during axon growth ( Figure 4C  Video 10; n = 14/15 cells). In many cases lifeactRuby was localised baso-ventrally more than 30 minutes before nascent axon protrusion ( Figure 4F; n = 10/15 cells).
Thus, F-actin accumulates at the site of the nascent axon before it is stabilized and is the earliest cytoskeletal sign of specification of nascent axon formation.

Nascent axons form in the absence of microtubules
The finding that F-actin accumulation consistently preceded the localisation of an axon-specific microtubule motor such as kif5c560 suggests F-actin accumulation may be the primary cytoskeletal element necessary for axon initiation. To test this we examined axon initiation in the absence of microtubules. We labelled cells for membrane and F-actin and bathed embryos in 5 µg/mL nocodazole for 60-180 minutes to depolymerize microtubules (Head et al., 1985;Jordan and Wilson, 1998;Gallo and Letourneau, 1999). By 30 minutes EB3-GFP labelled comets that label growing microtubules have disappeared from cells ( We focussed on 54 neurons that had retracted their apical and basal processes and had not yet extended a nascent axon (see Figure 1B  (A) Image sequence from confocal time lapse of a neuron that does not extend a nascent axon during nocodazole treatment labelled with a membrane marker (grey) and lifeact-Ruby (green) before (-10m) and during (0m to 170m) nocodazole treatment. Protrusions present before nocodazole addition are retracted upon nocodazole treatment (20m). Short, transient non-axonal filopodia are extended during nocodazole treatment (arrowheads). Images are transverse reconstructions from confocal z-stacks. (B) Image sequence from confocal time lapse of a neuron that extends a nascent axon during nocodazole treatment labelled with a membrane marker (grey) and lifeact-Ruby (green) before (-10m) and during (0h to 170m) nocodazole treatment. Small protrusions present before nocodazole addition (-10m) are retracted upon nocodazole treatment (0m). A nascent axon-like protrusion (long, broad, long-lived) is extended during nocodazole treatment (110m to 170m, arrows). These results show microtubules are not required for the initiation of a nascent axon and, together with our results showing that F-actin localisation preceded that of microtubule markers, suggests that F-actin accumulation is the key cytoskeletal element for nascent axon initiation. The retraction of nascent axons that existed before nocodazole treatment suggests microtubules are important stabilising nascent axons in addition to their previously suggested role in axon maintenance and growth (Letourneau and Ressler, 1984;Hahn et al., 2019).

Laminin provides a basal cue for axon initiation
Finally, we investigated external factors that may be responsible for directing nascent axon formation. As axon initiation occurs from the baso-ventral side of the soma, we hypothesised that the nascent axon is extended adjacent to the neuroepithelial basal surface. We investigated this by randomly labelling cells in a utrophin reporter line that marks the neuroepithelial basal surface (Tg(actb1:utr-mCherry); Figure 6A). We used time-lapse imaging with Airyscan acquisition and processing to observe the position of the neuronal soma, the base of the nascent axon and the tip of the nascent axon protrusion with respect to the neuroepithelial basal surface at the time of axon initiation ( Figure 6C). Fluorescence intensity analysis showed that all three regions of the neuron were within a few microns of the edge of the neuroepithelium (Figure 6C, E). One neuronal subtype in the zebrafish spinal cord derives from division of a nonapical progenitor (Vsx1 progenitors) close to the basal surface of the neural tube. The Vsx1 progenitors are rare and undergo a final mitosis producing two neurons that rapidly extend axons (McIntosh et al., 2017). We observed two cases of non-apical progenitor divisions in which mitotic cleavage resulted in one daughter adjacent to the basal surface and the other apparently not in contact with the basal surface ( Figure 6B). Interestingly, the daughter without contact with the basal surface initiated its axon from the ventral but not basal side of the soma while the daughter adjacent to the basal surface initiated its axon baso-ventrally along the basal surface as expected ( Figure 6B 0m; n = 2/2 divisions).
This led us to hypothesise that the basal surface of the neural tube may provide a directional cue for nascent axon specification in spinal cord neurons. The extracellular matrix protein laminin, a component of the basal lamina, can influence neuronal polarity and promote axon outgrowth in vitro (Esch et al., 1999), and orient axon outgrowth in zebrafish retinal ganglion cells in vivo (Randlett et al., 2011), so it is a good candidate to provide a directional cue. We used utrophin reporter line embryos that had been injected with LamininC1 morpholino (lamMO; (Parsons et al., 2002)) to ask whether the presence of laminin influenced the site of axon initiation. Fluorescence intensity analysis showed that in the Laminin-depleted embryos the site of axon initiation and the tip of the nascent axon protrusion were significantly further away from the edge of the basal lamina than in control embryos ( Figure 6D, E). However the position of the soma was not affected, indicating that this is not due to incorrect positioning of the neuron ( Figure 6E).
To examine the role of laminin in neuronal polarity more closely we analysed the site of axon initiation in the Sly/LamC1 zebrafish mutant line, which have no detectable laminin expression in the basal lamina at this embryonic stage ( Figure 6F, compare to Figure 1C; Video 12). There was no difference in the position of nascent axon initiation between controls and Sly -/embryos when analysed from the dorsal perspective but, analysis from the transverse perspective showed the nascent was less basal and more ventral compared to controls ( Figure 6F 0m, G, H). LamMOinjected embryos had a similar but more severe phenotype than Sly -/embryos. The basal bias of the site of the nascent axon was lost in cells in lamMO-injected embryos and the ventral bias was increased compared to controls ( Sly -/dorsal view 0.1 < P < 0.2, transverse view 0.1 < P < 0.2; lamMO dorsal view 0.05 < P < 0.2, transverse view 0.5 < P). Thus, although the centrosome tends to be opposite the nascent axon site in wildtype, changes in axon position resulting from Laminin depletion do not appear to significantly alter centrosome position.

Hedgehog signalling does not provide a ventral cue for axon initiation
Our results show that laminin biases the nascent axon formation site towards the basal side of the neuronal soma, but the ventral bias was maintained in the absence of laminin. As such, we attempted to identify a dorso-ventral cue that promoted the ventral positioning of the nascent axon. Zebrafish embryos were exposed to inhibitors of various signalling pathways or proteins that are known to have a dorsoventral gradient in the spinal cord, that are known to be axon guidance cues, or that have been shown to influence neuronal polarity in other systems: BMPs (Lee et al., 1998;Augsburger et al., 1999), FGF (Walicke et al., 1986), Hedgehog (Charron et al., 2003), and septins (Boubakar et al., 2017) (not shown). Of these, only inhibition of Hedgehog signalling via cyclopamine (100 µM; (Taipale et al., 2000;Chen et al., 2002)) showed potentially promising results (not shown). To test this we analysed axon initiation in Smu b641-/embryos, which have a mutation in the Smoothened protein that is required for both canonical and non-canonical Hedgehog signalling

DISCUSSION
We have investigated the earliest steps in axon formation in spinal projection neurons in vivo. Our key findings are: -an accumulation of F-actin is an early molecular indicator of axon initiation, precedes the generation of a stable nascent axonal protrusion and precedes microtubule accumulation -nascent axons are still formed in the presence of the microtubule disruptor nocodozole -axon initiation is extremely stereotyped across different spinal neuron subtypes irrespective of subsequent axon guidance -MTOCs, the centrosome and Golgi apparatus, are located on the opposite side of the cell to the site of axon initiation but move to the base of the axon during axon growth -Laminin is not required for axon initiation or early growth but is a positional cue for axon initiation.
The polarisation of individual neurons into axonal and dendritic compartments is critical for correct nervous system development. The mechanisms that may define axon initiation have been studied for several decades but many of the previous works are compromised by studying this process in neurons growing in vitro that lack the complex 3-dimensional architecture and molecular environment of in vivo settings. Additionally, some in vitro studies may describe the repolarisation of neurons that had previously polarised in vivo rather than neurons polarising for the first time (reviewed in (Barnes and Polleux, 2009)). Further, axon initiation in vitro is the differentiation of a pre-existing neurite to become an axon (Dotti et al., 1988), making it difficult to differentiate between axon initiation and axonal growth (Jiang and Rao, 2005;Barnes and Polleux, 2009). For some existing models of neuronal differentiation in vivo it can be difficult to disentangle axon initiation from neurite growth and neuronal migration (Barnes and Polleux, 2009). Our study of spinal neurons in vivo overcomes these reservations. The cell bodies of neurons in the zebrafish spinal cord move to the basal surface of the neuroepithelium before delaminating from the apical surface and remain established at the basal surface for several hours before axon initiation (Hadjivasiliou et al., 2019), thus the remodelling of cell polarity in this system is not complicated by polarity changes related to cell migration. Axon formation occurs directly from the cell body, so can be clearly identified and separated from axon growth, and we find it occurs from a stereotyped position in all of the spinal neuron subtypes that we investigated, independently of the subsequent direction of axon projection (except for Rohon-Beard neurons, which elaborate three axons). As such, the zebrafish embryonic spinal cord provides a complex in vivo system where we can definitively separate axon initiation from both axonal growth and neuronal migration. We define the first phase of axon initiation as the nascent axon -this is a protrusion that is beginning to take on the characteristics of an axon and would normally become an axon when stabilised by microtubules.
Our principle finding is that the earliest indication of axon initiation is a biased accumulation of F-actin in the baso-ventral quadrant of the neuron cell body. This first coincides with unstable protrusions from the baso-ventral soma and then with a stable protrusion that we term the nascent axon ( Figure 4). Although microtubules rapidly invade the nascent axon, we find nascent axons can still be formed in the absence of microtubules ( Figure 5). Thus, nascent axon formation from spinal neurons in vivo is an F-actin-based protrusion that forms directly from a specific location on the cell body. The generation of a single stereotyped axonal protrusion from spinal neurons in vivo is very different from the axon selection process from multiple random neurites seen in the in vitro neuronal polarisation models (e.g. (Dotti et al., 1988), and reviewed in (Barnes and Polleux, 2009)) and this very likely reflects the difference in complexity of environmental cues in these two cases. The very stereotyped location of nascent axon formation from the baso-ventral quadrant of spinal neurons suggests this process is strongly influenced by local environmental cues in the early neural tube and we uncover that one of those cues is Laminin ( Figure 6), an extracellular protein abundant at the basal surface of the neural tube.
Laminin may play a common role in the differentiation early born neurons as it has previously been shown to influence retinal ganglion cell axon initiation in the retina (Randlett et al., 2011) and axonal growth in the primary sensory Rohon-Beard neurons in zebrafish neural tube (Andersen and Halloran, 2012).
Although nascent axons can be produced in the absence of microtubules, they are not stable and retract ( Figure 5B) demonstrating a requirement for microtubules to stabilize this protrusion. Perhaps in this respect microtubules cooperate with dynamic actin in similar ways as they do to facilitate turning in neuronal growth cones (reviewed in (Geraldo and Gordon-Weeks, 2009)), and to promote axon specification (Bradke and Dotti, 1999;Geraldo et al., 2008;Witte et al., 2008;Zhao et al., 2017) and neurite initiation in vitro (Dent et al., 2007;Flynn et al., 2012). Increased numbers of microtubules and enriched microtubule plus-ends could play an important role in anterograde transport (reviewed in (Schelski and Bradke, 2017)).
However, in spinal neurons in vivo we see few EB3-labelled growing microtubule plus-ends in both pre-axonal protrusions and in the nascent axon, illustrating that there are few growing microtubules until axonal growth commences.
Consistent with our observation that microtubules are not required for nascent axon formation, we also show neither the centrosome nor the Golgi complex is close to the site of axon initiation in spinal neurons. Although several studies suggest the centrosome and Golgi complex are close to the base of the neurite that becomes the axon (Zmuda and Rivas, 1998;de Anda et al., 2005), our results support previous observations that show centrosome proximity is not required for axon initiation (Dotti and Banker, 1991;Zolessi et al., 2006;Distel et al., 2010;Gärtner et al., 2012).
There are several potential explanations for discrepancies between these findings.
There may be innate differences in cytoskeletal organisation between different neuronal subtypes or species; differences in substrate properties may alter cytoskeletal organisation, as has been shown for migrating cells (Pouthas et al., 2008); or, alternatively, studies showing MTOCs close to the base of the axon may be looking after axon initiation. This is supported by our observation that both the centrosome and Golgi complex move close to the base of the axon during pathfinding and that the centrosome is not close to the site of axon initiation in zebrafish retinal ganglion cells, in which axon initiation can also be easily identified (Zolessi et al., 2006). The centrosome has previously been described as being associated with peripheral axon formation in Rohon-Beard neurons and to be important for its growth (Andersen and Halloran, 2012), but we find that the centrosome is not close to the base of the axon at the time of initiation of any Rohon-Beard axon, including the peripheral axon. Combining these findings and others (e.g. (Stiess et al., 2010)) suggests that centrosome proximity is not required for axon initiation but is required for axon growth, at least at early stages.
Although not close to the site of axon initiation the centrosome is not positioned randomly in spinal neurons; instead, it is consistently opposite the site of axon initiation. As the centrosome is situated apically while the new neuron is still attached to the apical surface and is retracted into the neuronal cell body upon delamination, it may be that its medial position in the soma at axon initiation is simply related to the location of the retracting apical process. Interestingly we found no evidence that apical abscission is required for apical process retraction in the zebrafish spinal cord. This is in contrast to chick and mouse but in agreement with some observations in the zebrafish retina (Zolessi et al., 2006;Das and Storey, 2014;Lepanto et al., 2016). Although the centrosome is not close to the site of nascent axon initiation, it could still play a role in its stabilisation and transition to a growing axon. A small number of microtubules do enter the nascent axon and it is possible that only a few stable microtubules are required for the next steps in axon differentiation. Interestingly, modelling has shown that stochastic microtubule dynamics can lead to stabilisation of the longest microtubules (Seetapun and Odde, 2010), suggesting a method by which the distant centrosome may stabilise the nascent axon on the opposite site of the cell where only the longest microtubules can reach.
The position of the nascent axon is influenced by the extracellular matrix protein laminin at the basal surface of the neural tube, as the loss of laminin leads to the loss of the basal bias to the nascent axon position. In the retina, laminin stabilises newly initiated axons and promotes axonal growth (Randlett et al., 2011). The neurons that we observed were among the earliest that differentiated, meaning that they were almost always already adjacent to the laminin-rich basal surface when extending an axon. It would be interesting to compare this with later-born neurons, which would have earlier-born neurons between them and the basal surface. Nonetheless, we show that axon initiation, stabilisation and growth can occur robustly in the spinal cord without laminin. Although we were unable to find evidence of a dorsoventrally oriented cue that was required for ventral directional bias of the nascent axon, we can't rule out that one exists. An alternative to a molecular cue that directs the ventral bias of the nascent axon could be the overall physical architecture of the cells in the early neural tube. All neuroepithelial cells and neurons have a curved morphology in transverse sections, with the lateral poles of both neuroepithelial progenitors and neurons curving ventrally as they approach the basal surface (see Figure 1B, transverse inserts). It seems possible this morphological organisation of cells could provide a 3-dimensional physical substrate or orientation that encourages ventral growth of nascent axon protrusions.