Axon guidance at the spinal cord midline—A live imaging perspective

Abstract During neural circuit formation, axons navigate several choice points to reach their final target. At each one of these intermediate targets, growth cones need to switch responsiveness from attraction to repulsion in order to move on. Molecular mechanisms that allow for the precise timing of surface expression of a new set of receptors that support the switch in responsiveness are difficult to study in vivo. Mostly, mechanisms are inferred from the observation of snapshots of many different growth cones analyzed in different preparations of tissue harvested at distinct time points. However, to really understand the behavior of growth cones at choice points, a single growth cone should be followed arriving at and leaving the intermediate target. Existing ex vivo preparations, like cultures of an “open‐book” preparation of the spinal cord have been successfully used to study floor plate entry and exit, but artifacts prevent the analysis of growth cone behavior at the floor plate exit site. Here, we describe a novel spinal cord preparation that allows for live imaging of individual axons during navigation in their intact environment. When comparing growth cone behavior in our ex vivo system with snapshots from in vivo navigation, we do not see any differences. The possibility to observe the dynamics of single growth cones navigating their intermediate target allows for measuring growth speed, changes in morphology, or aberrant behavior, like stalling and wrong turning. Moreover, observation of the intermediate target—the floor plate—revealed its active participation and interaction with commissural axons during midline crossing.


| INTRODUCTION
Commissural axons in the developing spinal cord have been used for over two decades to learn fundamental molecular mechanisms of axon guidance (Stoeckli, 2018). The dI1 subtype of dorsal commissural interneurons are an excellent model, as their axons have a stereotypical trajectory at the ventral midline, as all axons of this subpopulation cross the floor plate (FP), exit it and turn rostrally along the contralateral border. Thus, dI1 commissural neurons offer an easy read-out for deciphering molecular mechanisms of axon guidance at choice points. Since the first application of lipophilic dye tracing in open-book preparations of rat spinal cords that revealed the normal trajectory of these axons 30 years ago (Bovolenta & Dodd, 1990), this method continues to be used to assess axon guidance at the midline in mouse and chicken embryos. The comparison between axons in open-book preparations of control and experimentally manipulated spinal cords, dissected at specific time points, offered a solid understanding of molecules involved in axonal midline crossing and subsequent turning in higher vertebrates. However, the information about mechanisms that can be extracted from such experiments is limited, as it is deduced from snapshots of axons taken from different animals sacrificed at a specific time point. For this reason, we have established a live-imaging approach that allows for visualization of dI1 axonal behavior at the FP, while axons are in the process of crossing the midline and then turning rostrally. We have chosen the chicken embryo, as it is a very accessible model for studying various developmental processes in intact tissues in vivo and ex vivo (Boubakar et al., 2017;Das & Storey, 2014;Li et al., 2019;Sanders et al., 2013). Thanks to a very stable and reproducible spinal cord culture, we could for the first time characterize the exact timing of midline crossing and the details of rostral turning by dI1 axons in an intact environment in control and experimentally manipulated spinal cords. We could get more insight into growth cone dynamics and morphologies at their choice point, the FP. Furthermore, our ex vivo method also shed new light on the role of the intermediate target, the FP cells, their dynamics, morphology and interaction with commissural axons during midline crossing. Finally, our results on the dynamic behavior of growth cones at the spinal cord midline allowed us to compare axonal behavior at different choice points, such as the optic chiasm, reported previously (Godement et al., 1994;Sretavan & Reichardt, 1993).

| Dissection of intact spinal cords
Intact spinal cords were dissected from HH22 embryos in ice-cold, sterile PBS (Gibco) in a silicon-coated Petri dish with sterile instruments. Embryos were pinned down with their dorsal side down with thin needles (insect pins). Here, special care was taken not to damage or detach meninges surrounding the spinal cord by avoiding too much rostro-caudal and lateral tension. Internal organs and ventral vertebrae were removed to access the spinal cord. Ventral roots were cut off and the spinal cord was carefully extracted from the embryo with forceps, avoiding any excessive bending. Note that dorsal root ganglia were not cut off and all were still attached to the spinal cord. The ventral and dorsal midlines were kept intact throughout dissection.
Finally, remaining dorsal tissues were discarded. See Figure 1 for a detailed step-by-step protocol to successfully dissect intact HH22 spinal cords. Note that this procedure can also be applied to older embryos (at least HH24-25). Once intact spinal cords were dissected and cleaned from any remaining dorsal tissues, they were embedded with the ventral side down in a warm (39 C) 100-μl drop of 0.5% low- added to get 0.5% final agarose concentration). Note that the spinal cord should be as straight as possible with the dorso-ventral axis perpendicular to the glass bottom, as any pronounced curvature or tilting of the midline would induce axon guidance artifacts or death of the axons, respectively. To this end, a 12-mm flexiPERM conA ring (Sarstedt) was placed in the center of the culture dish before the agarose drop was added (the drop should not touch the ring). Hence, the medium added to the drop of low-melting agarose could touch the ring all around and therefore stabilize the position of the agarose drop and avoid any movement of the spinal cord during recordings thanks to surface tension (Figure 1(g)). Once the agarose solidified (around 5 min at room temperature), 200 μl of spinal cord medium were added to the drop and the culture could be started.

| Dissection of open-books
Open-book preparations of spinal cords were dissected from HH24 embryos as previously described in a video protocol for HH25-26 embryos (Wilson & Stoeckli, 2012). The first steps were identical to the protocol for intact spinal cord dissection given above (steps a,b in Figure 1). Starting there, the tension along the rostro-caudal axis was increased using the upper and lower needles and meninges were removed with a blade made of fire-polished tungsten wire. Spinal cords were cut transversally at the wing and leg levels and carefully extracted from the embryo with forceps. At this point, the dorsal midline spontaneously opened. Open-book preparations of spinal cords were then plated with the apical side down (Figure 2(g,h)) in the center of a 35-mm Ibidi μ-dish with glass bottom (Ibidi, #81158), precoated with 20 μg/ml poly-L-lysine (Sigma). A homemade, harp-like holder made out of a Teflon ring and thin nylon strings was used to keep the spinal cord in place (Figure 2(g)). Note that the strings were barely touching the open-books but stabilized the flat position of the spinal cord. Then, 100 μl of 0.5% low-melting agarose (see above) were added on top of the spinal cord. Once the agarose solidified (around 5 min at room temperature), 200 μl of spinal cord medium were added to the agarose drop and the culture could be started.
F I G U R E 1 Dissection of intact spinal cords (SCs) from HH22 chicken embryos. (a) HH22 embryos were pinned down with the dorsal side down in a silicon-coated Petri dish in sterile, cold PBS. Internal organs were removed by first cutting the ventral skin along the dashed lines and pinching out the organs with forceps. (b) Then, a laminectomy was performed, that is, the ventral vertebrae were cut along the caudal-rostral axis at the level of the outer SC boundaries and the stripe of bone structure was removed with forceps. (c) The ventral roots exiting the ventral part of the SC and the peripheral processes of the dorsal root ganglia (DRG) were cut in parallel to the SC without cutting off any DRG. (d) The SC was then cut at the level of the wings and legs. (e) The SC with attached DRG was carefully separated from the rest of the embryo with forceps. Here, special care should be given not to bend the SC by stabilizing the tissue with a second forceps. (f) At this point, the dorsal skin and dermomyotome (black arrows) were removed by first inducing an opening with forceps (white asterisk) taking care not to damage the dorsal SC. Then, using forceps, the dorsal skin and dermomyotome were carefully removed all along the caudal-rostral axis. After this step, the dorsal SC should look as clean as the ventral SC with clearly visible midline and no remaining tissues attached (compare dorsal and ventral view). (g) Finally, the intact SC with attached DRG could be embedded as straight as possible in a drop of low-melting agarose-medium mix with the ventral side down. White dashed lines indicate where cuts with small spring scissors should be made

| Live imaging
Live imaging recordings were carried out with an Olympus IX83 inverted microscope equipped with a spinning disk unit (CSU-X1 10,000 rpm, Yokogawa). Cultured spinal cords were kept at 37 C with 5% CO 2 and 95% air in a PeCon cell vivo chamber (PeCon). Temperature and CO 2 -levels were controlled by the cell vivo temperature controller and the CO 2 controller units (PeCon). Spinal cords were incubated for at least 30 min before imaging was started. We acquired 18-40 planes (1.5 μm spacing) of 2 × 2 binned z-stack images every 15 min for 24 h with a ×20 air objective (UPLSAPO ×20/0.75, Olympus) and an Orca-Flash 4.0 camera (Hamamatsu) with the Olympus CellSens Dimension 2.2 software. We performed most of our recordings in the lumbar level of the spinal cord and always took three channels of interest: emission at 488 and 561 nm, as well as bright field.
Recordings of axons after Fzd3 or luciferase knockdown were performed at the thoracic level (see cartoon in Figure 5(g)). For higher magnification recordings, a ×40 silicone oil objective was used (UPLSAPO S ×40/1.25, Olympus) with the same acquisition settings as above.
Images taken every 5-15 min. The quantifications of growth cone behavior, such as splitting in the FP, formation of ventral-dorsal protrusions and protrusions before FP exit, were based on a time resolution of one stack taken every 15 min. The analysis of protrusions before FP exit could be done only on the first axons exiting the FP, because with time it was difficult to visualize the small protrusions correctly due to the increase in the density of axons at the FP exit site.

| Data processing and virtual tracing
Z-stacks and maximum projections of Z-stack movies were evaluated and processed using Fiji/ImageJ (Schindelin et al., 2012). The MtrackJ plugin (Meijering et al., 2012) was used to virtually trace single Math1-positive dI1 commissural axons crossing the FP. This helped to keep track of which axons had already been quantified. The leading edge (and not filopodia) of growth cones was always selected for each F I G U R E 2 Labeling strategy for dI1 interneurons and spinal cord (SC) culture systems. (a-c) In ovo injection and electroporation of a plasmid mix to specifically label dI1 interneurons. (a) The plasmid mix was injected into the central canal of the SC of HH17-18 chicken embryo in ovo, followed by unilateral electroporation. (b) Plasmid constructs injected to target all cells (β-actin::EGFP-F) and dI1 interneurons (Math1::tdTomato-F). en., enhancer; β-glob., β-globin. (c) Immunostaining of a transverse cryosection of a HH22 SC taken from an embryo sacrificed 1 day after electroporation with the plasmids indicated in (b). At this stage, most dI1 growth cones were approaching the FP area, but none of them had crossed it yet (white arrowheads). However, a substantial number of Math1-negative, but EGFP-F-expressing commissural axons of more ventral populations had already crossed the FP at HH22 (arrow). selected with the tracing tool, as growth cones very often slightly changed their directionality and drastically change their shape before turning. For comparable analyses of axonal behavior and speed in the two halves of the FP, we only traced and quantified axons that entered, crossed and exited the FP during the 24-hour imaging period. Axons already in the FP at the beginning of the imaging period were not considered. Overlays of labeled axons with EGFP-F and bright-field channels were used to assess the FP boundaries and midline localization. The virtual tracing tool was also used to extract the local growth speed for each single axon. Local speed is defined here as the average speed calculated for the last 15 min time period. Note that the montage of dI1 commissural axons shown in Movie 2 was generated from z-stacks that were 2D deconvolved (nearest neighbor) using the Olympus CellSens Dimension 2.2 software and assembled with Fiji/ ImageJ. All data acquired with higher magnification (×40 silicone oil objective) were 3D deconvolved using constrained iterative deconvolution of the Olympus CellSens Dimension 2.2 software (five iterations with adaptive PSF and background removal, Olympus).
Maximum projections of live images containing Hoxa1::EGFP-Fpositive cells (channel) were corrected for photo bleaching in Fiji/ ImageJ.

| Temporal-color projections and kymographic analysis
Temporal-color projections were generated using Fiji/ImageJ. Kymograph analysis of axons crossing or exiting the FP as previously described (Medioni et al., 2015) using a region of interest (ROI) selection, the re-slice function and the z-projection of the re-sliced results in Fiji/ImageJ, which allowed following pixel movements within the horizontal axis. The ROI in the FP was selected as a 103 × 51 μm 2 (Figure 7(c,d)) or 103 × 27 μm 2 (Figure 7(e,f)) rectangle.

| Immunohistochemistry
Spinal cords dissected from HH22 embryos or intact spinal cords that were cultured for 1 day ex vivo were fixed 1 h at room temperature with 4% paraformaldehyde in PBS, washed three times for 5 min each with PBS and cryopreserved for at least 24 h at 4 C in 25% sucrose in PBS. After mounting in O.C.T. compound (Tissue-Tek) and freezing the spinal cords, 25-μm thick cryosections were collected using a cryostat.

F I G U R E 3
The mean intensity was measured in the FP using a circle ROI with a diameter of 100 μm. For each embryo, this intensity was measured from two to three consecutive sections per level (lumbar and thoracic), averaged and normalized to the average intensity of the lumbar level. This was performed for four intact spinal cords and open-book preparations that were fixed after 24 h in culture.
For all embryos and levels, the acquisition settings were exactly the same. Images were taken with an Olympus BX61 upright microscope and a ×20 water objective (UAPO W/340 ×20/0.70, Olympus) and an Orca-R 2 camera (Hamamatsu) with the Olympus CellSens Dimension 2.2 software.

| Statistics and figures assembly
Statistical analyses were carried out with GraphPad Prism 7.02 software. All data were assessed for normality (normal distribution) using the D'Agostino and Pearson omnibus K2 normality test and visual assessment of the normal quantile-quantile plot before choosing an appropriate (parametric or nonparametric) statistical test. p-Values of the simple linear regression shown in Figure 7(i,j) demonstrate whether the slope is significantly different to zero and the dashed lines represent the 95% confidence intervals. staining (magenta asterisks, Figure 4(e)). The same staining also F I G U R E 4 Patterning and morphology of cultured intact spinal cords were conserved after 1 day ex vivo. After intact HH22 spinal cords were cultured and imaged for 1 day ex vivo, they were fixed and transverse cryosections were immunostained for different dorsal and ventral patterning markers, and counterstained with Hoechst. (a) The dI1 interneuron marker Lhx2 confirmed that these neurons were still localized in the most dorsal part of the spinal cord, as expected (black arrowheads). (b) Islet-1 was used as a marker for dorsal root ganglia (DRG) neurons (magenta asterisks), dI3 interneurons (black arrows) and motoneurons (black arrowheads). All of them maintained the appropriate position: clustered DRG neurons adjacent to the spinal cord; dI3 interneurons localized ventrally of dI1 interneurons; motoneurons on both sides of the ventral spinal cord. (c) Nkx2.2 staining was used to reveal the ventral population of V3 progenitors that are just next to the FP and form the typical inverted V-shape (black arrow). (d) Finally, FP cells forming the intermediate target for dI1 axons were visualized with Hnf3β staining. They were localized at the ventral midline of the spinal cord as expected (black arrow). (e) Intact spinal cord cultured for 24 h maintained morphology and localization of cell types as illustrated by staining for BEN: motor neurons (black arrowheads) and their ventral roots (magenta asterisks, the dorsal roots (black asterisks), and the dorsal funiculi (black arrows) formed by DRG afferents. (f) BEN staining also revealed that the FP maintained its triangular shape (dashed lines) and its bulky basal segment (white arrow). (g) Same staining of open-book preparations, which were cultured for the same amount of time: the motor column (black arrowhead) is stained for BEN but ventral roots and DRG afferents were removed during dissection.  site, the high incidence of aberrant axon growth into the contralateral half of the spinal cord is not acceptable. We found that 91 ± 6% of dI1 axons overshot, whereas 8 ± 6% turned rostrally, and 1 ± 1% turned caudally at the contralateral side (mean ± SD, N(embryos) = 4, n(axons) = 392). In our system, none of them overshot, and 98 ± 2% turned rostrally and 2 ± 2% caudally at the contralateral side (e,f) and 4(h)) and the fact that diffusible guidance cues are not well retained in the tissue, as exemplified by Shh that lost its caudal (lumbar) to rostral (thoracic) gradient after 24 h in culture ( Figure 5(d-f,g,i)).
In line with this, staining for laminin revealed that the meninges surrounding the spinal cord were intact in our culture system (black arrowheads, Figure 4 -Cohen et al., 1999;Wright et al., 2012;Zisman et al., 2007). The laminin staining also revealed that the basal lamina was intact at the commissure of cultured intact spinal cord (white arrowheads, Figure 4 (j)), whereas it was deformed (white arrow) and discontinued in cultured open-books (white arrowheads, Figure 4(l)). We also assessed possible apoptosis in cultured intact spinal cords by staining for cleaved caspase. We found very little apoptosis in the ventral spinal cord (arrowheads, Figure 4(m,n)) and DRGs in our system (black asterisks, Figure 4(m)). In contrast, cultured open-books showed massive  Figure 4(o,p)) in the region of the motor columns that might be caused in part by the complete loss of ventral roots in this system (black arrowhead, Figure 4 (g)). A combination or all of these marked differences observed in cultured open-books are likely to be responsible for the strong artifactual behavior of dI1 axons at the contralateral FP border. In contrast, our ex vivo method of culturing intact spinal cords offers a highly stable intact system in which dI1 commissural axons are behaving as expected based on what is known from in vivo studies with the advantage that individual axons can be followed during midline crossing and subsequent turning along the contralateral FP border.

| Characterization of the timing of midline crossing by dI1 commissural axons
The time it takes commissural axons to cross the FP has been estimated but could not be measured exactly (Stoeckli, 2018;Zou, 2012).
However, timing is an issue, because axons have to change their responsiveness to FP-derived guidance cues, like the Slits, Shh, or Wnt proteins, by expressing appropriate receptors in a precisely regulated manner (Bourikas et al., 2005;Domanitskaya et al., 2010;Long et al., 2004;Lyuksyutova et al., 2003;Philipp et al., 2012;. With our method, we could track single dI1  Figure 3(g)). We were therefore able to ask, how long dI1 axons needed for FP crossing and their subsequent rostral turn (Figure 7(a)). On average, dI1 commissural axons took 5.6 ± 1.4 h to cross the entire FP and 1.4 ± 1.0 h to turn and initiate the rostral growth at the FP exit site. Thus, in total, they needed 6.9 ± 1.8 h from entering the FP to the initiation of their rostral growth (mean ± SD, Figure 7(b) and Table 1). There was no significant difference between  (Figure 7(e,f)), it was not sensitive enough to detect more subtle changes in growth speed.
Hence, we used a virtual tracing tool to follow the movement of the leading edge of each growth cone at each time point (Figure 7(g)).
With this tool, we could extract the local growth speed for each axon at a specific time point (defined as the average speed during the previous 15 min). It turned out that the large majority of them had a fluctuating growth pattern with random acceleration-deceleration pulses that could be observed in early as well as late crossing axons (Figure 7 (h), Movie 5, Figure 8). Another interesting observation was made when we compared the times of crossing the FP and the initiation of rostral growth after turning. There was no significant difference in the time of FP crossing of later versus earlier crossing axons (p = .0503; Figure 7(i)). However, the time dI1 axons took to turn rostrally at the exit site was significantly reduced over time (Figure 7(j)). The latter observation suggests that commissural axons that already turned anteriorly at the contralateral FP exit site might help the following ones to turn more rapidly. Our method offers new opportunities for further investigations of possible collaborations between axons at choice points.

| Characterization of the dI1 growth cone morphology at choice points
Another aspect that we considered was growth cone morphology.
The growth cone plays a central role in axon guidance, as it explores the environment for guidance cues and translates this information into the directionality of growth (De Ramon Francàs et al., 2017;Stoeckli, 2018). In agreement with published live imaging studies from the visual system, where growth cone morphology was recorded at the chiasm (Godement et al., 1994;Sretavan & Reichardt, 1993), we found changes in growth cone morphology during midline crossing in the spinal cord. We observed that dI1 growth cones in the FP appeared to have a thin and elongated shape in the direction of growth. At the FP exit site, they transiently enlarged (arrowheads, Figure 9(d), Movie 2). We measured the average growth cone area in each segment of interest and confirmed that growth cones at the exit site of the FP were indeed significantly larger than the ones within the FP or after the turn (Figure 9(a,b), Table 1). There was no significant change in the average growth cone area between the first and second half of the FP (Figure 9(b), Table 1). The changes of growth cone shape were in line with previous reports on chicken and rat commissural axons in vivo (Bovolenta & Dodd, 1990;Yaginuma et al., 1991) and our data on Math1-positive axons in vivo (Figure 9(c), Table 1, Figure 6(b)). The possibility to follow individual axons over time allowed us to make novel observations of their behavior at the FP exit site. One hundred percent of the observed growth cones extended long filopodia in both rostral and caudal direction just before turning (N(embryos) = 7, n(axons) = 305, Figure 9(d), Movies 6 and 7).
We observed 16 ± 8% of the growth cones to transiently split just before turning rostrally, similar to dorsal root ganglia central afferents in the mouse dorsal root entry zone before bifurcating (Dumoulin et al., 2018) (mean ± SD, N(embryos) = 7, n(axons) = 305, Movie 8). All these features are present in vivo, as similar growth cone morphologies were found in fixed HH24-25 spinal cords (Figure 9(e)).  where growth cones need to read longitudinal gradients to initiate the rostral turn after exiting the FP (Pignata et al., 2019;Stoeckli, 2018).
Intriguingly, we detected that just before exiting the FP, dI1 growth cones very often sent a long protrusion into the FP (arrow, Figure 10 F I G U R E 9 Live imaging of intact spinal cords revealed dI1 growth cone morphologies at chosen time points. (a) Schematic depicting where the growth cone area of individual dI1 axons was measured for (b) and (c). (b) Average growth cone areas were measured from 24-h time-lapse recordings of dI1 axons crossing the midline (N embryos = 7; n growth cones = 127). No significant difference in the area of growth cones was found between the first and second half of the floor plate (FP). However, growth cones were significantly larger at the exit site but then again reduced in size after having turned rostrally (paired Friedman test with Dunn's multiple-comparisons test). (c) Average growth cone areas were measured in vivo from fixed HH23-25 spinal cords (N embryos = 8; n growth cones = 285 (first half), 153 (second half), 68 (exit), and 102 (after turn). The relationship between the average growth cone area and the position in the FP corroborated results using the ex vivo culture system shown in (b) (unpaired Kruskal-Wallis test with Dunn's multiple-comparisons test). (d,e) Examples corroborating the similarities in growth cone morphology ex vivo and in vivo in the FP (black arrowheads), at the exit site (green arrowheads) and after rostral turn (magenta arrowheads). At the exit site, growth cones were spiky with always some filopodia pointing caudally just before rostral turn, a feature that was also observed in vivo (black arrows). ipsi, ipsilateral; contra, contralateral; r, rostral; c, caudal. Error bars represent SD. p < .0001 (****), p < .01 (**), and p ≥ .05 (ns) for all tests. Scale bars: 10 μm growth cones transiently split while crossing the FP (mean ± SD, N (embryos) = 4, n(axons) = 116, asterisks in Movie 13). The splitting created two more or less equal branches (black arrows, Figure 10(k), Movie 16), but only one persisted and grew straight to the contralateral side, while the other one was retracted (black asterisks, Figure 10 (k), Movie 16). Also this behavior was supported by snapshots from in vivo behavior of dI1 growth cones (arrows, Figure 10(l)). Taken together, ex vivo live imaging combined with high magnification analysis of growth cone dynamics allowed us to characterize the behavior dI1 growth cones at choice points in more detail.
3.6 | Live imaging unraveled the dynamics and morphologies of FP cells during midline crossing The orientation of dI1 growth cones as well as their behavior during FP crossing suggested that they have to squeeze their way between the basal feet of FP cells which are attached to the basal lamina (Yaginuma et al., 1991;Yoshioka & Tanaka, 1989). Moreover, very little was known about the morphology of FP cells during axonal midline crossing and their potential active contribution in this process has never been addressed (Campbell & Peterson, 1993; Yaginuma et al., 1991;Yoshioka & Tanaka, 1989). Therefore, we examined the behavior and morphology of FP cells during midline crossing in our ex vivo system. We electroporated spinal cords at HH17-18 after injection of a plasmid encoding EGFP-F under the FP-specific Hoxa1 enhancer for expression of the membrane-bound fluorescent protein in FP cells (Li & Lufkin, 2000;Wilson & Stoeckli, 2011;Zisman et al., 2007) (Figure 11(a-d)). With this we were able to see Hoxa1:: EGFP-F-positive bulky FP basal feet in the commissure in vivo (white arrowheads, Figure 11(c)) as well as their thin morphology and orientation (white arrows) that seemed to be tightly aligned with dI1 growth cones crossing the midline (white arrowheads, Figure 11(e)).
The    (Alther et al., 2016;Lyuksyutova et al., 2003). We used our ex vivo culture system to visualize Math1::EGFP-F-positive dI1 axons expressing a microRNA for Fzd3 (miFzd3, Figure 12(b,c)). This allowed us to follow in real time how dI1 growth cones were turning caudally instead of rostrally at the contralateral FP border (black arrowheads,   (Pignata et al., 2019). Our ex vivo culture system is highly reproducible and generates a manageable amount of data compared to live imaging using light sheet-based microscopy, for example (Liu et al., 2018).
The comparison of growth cones at the FP (this study) and at the optic chiasm (Godement et al., 1994;Sretavan & Reichardt, 1993) reveals both common and different behaviors. As seen at the chiasm (Godement et al., 1994), growth cones exhibited a saltatory growth At the chiasm, crossing axons did not change their morphology significantly, whereas those that changed their trajectory to turn along the ipsilateral pathway displayed a dramatic shape change just before the turn. Therefore, this resembles the behavior at the FP exit site, where a new direction is adopted by turning into the longitudinal axis.
Our comparative analyses demonstrate that midline crossing of dI1 axons in our ex vivo system was very similar to what happens in vivo (Figures 3-6). Therefore, our ex vivo system can be used to monitor and assess axonal behavior at choice points. We could detect that dI1 growth cones took on average 5.6 h to cross the entire FP and that they did so in a pulsed manner (black arrowheads, Figure 13).
We could also measure that they needed on average 1.4 h to initiate their rostral growth, and that the first axons exiting the FP took longer than the followers (Figure 13(a)). In total they needed almost 7 h from entering the FP to making the decision to turn rostrally (blue arrowhead, Figure 13(a)). This is enough time for growth cones to change their responsiveness to specific guidance cues for crossing and exiting the FP as well as for turning rostrally due to changes in receptor expression regulated at the posttranslational, translational, and even transcriptional level (Nawabi et al., 2010;Philipp et al., 2012;Pignata et al., 2019;Preitner et al., 2016;Stoeckli, 2018;. The growth cone is the decision center where axon guidance instructions are transduced to the cytoskeleton (Vitriol & Zheng, 2012). With our newly developed ex vivo system, dynamic changes in dI1 commissural growth cone morphology and behavior at the midline can be observed in real time. Growth cones were thin and elongated in the FP with their major extension in the dorso-ventral axis (black arrowheads, Figure 13(a,c)). At the FP exit site, they showed a 90 rotation to be enlarged and active in the longitudinal axis (green arrowhead and black arrow, Figure 13(a), black arrowhead, Figure 13(d)). The fact that dI1 growth cones sent a long filopodium into the FP, toward the FP cell soma area, while crossing it and just before exiting it, suggests that they might need to read signals from this area in order to move on and exit the FP (orange arrows in Figure 13(c,d)). The extension of long filopodia just before FP exit and rostral turning suggests that actin polymerization might be required to sense repulsive cues-for instance, SlitN and Shh, respectively-and transduce the signal into the growth cone, as suggested for F I G U R E 1 3 Cartoon depicting the midline crossing characteristics of dI1 axons based on data extracted from our ex vivo culture system. (a) On average, it took dI1 axons 5.6 h to cross the midline. Growth cones showed a random pulsed growth and had a thin shape in the growth direction (black arrowheads). At the floor plate (FP) exit site, dI1 growth cones were first enlarged (green arrowhead), then extended filopodia along the longitudinal axis (black arrow) right before turning rostrally (blue arrowhead). After arriving at the exit site of the FP, it took dI1 axons on average about 1.4 h to turn rostrally. In fact, the first-exiting dI1 axons took longer to turn rostrally than the late exiting ones. (b-d) Live imaging of intact spinal cords ex vivo using a high magnification objective shed light on dI1 growth cone orientation, FP morphology and dynamics during midline crossing. (b) While dI1 growth cones (black arrowhead) approached the FP, basal feet of lateral FP cells sent protrusions toward them and eventually interacted with them (black arrows). (c) When dI1 growth cones crossed the FP (black arrowhead), their dorsoventral orientation aligned perfectly with the orientation of basal feet of medial FP cells. While basal feet of FP cells sent protrusions in axonal growth direction (black arrows), dI1 growth cones sent long filopodia in direction of the apical FP, toward the FP cell soma (orange arrow). (d) Just before exiting the FP, dI1 growth cones showed dorso-ventral activity with a long protrusion growing toward the FP soma (orange arrow) area followed by a 90 change in their orientation to become flattened in the dorso-ventral axis and enlarged in the longitudinal axis (black arrowhead). Ent, entry; Mid, midline; Ex, exit; r, rostral; c, caudal; ipsi, ipsilateral; contra, contralateral Slit-induced growth cone collapse in vitro (McConnell et al., 2016).
Further investigations using our ex vivo culture system will be required to understand the role of cytoskeletal dynamics in axonal navigation of the intermediate target.
Our method also suggested a probable active contribution of FP cells to axon guidance that goes beyond providing axon guidance cues, as we found the cells of the intermediate target to be very dynamic and to extend protrusions in directions of the arriving axons, or to actively engage with axons in the FP. Thus, it seems that the intermediate target is much more than a passive by-stander and provider of attractive and repulsive axon guidance and cell adhesion molecules. We characterized the FP cell morphologies in detail in the medial as well as the lateral FP. Basal feet appeared to be enlarged and oriented parallel to commissural growth cones (Figure 13(c)). This was in line with previous reports in the chicken and mouse embryos (Campbell & Peterson, 1993;Ducuing et al., 2020;Yaginuma et al., 1991;Yoshioka & Tanaka, 1989).
The lateral FP basal feet sent protrusions (black arrows) toward dI1 growth cones approaching the FP and eventually interacted with them (black arrowhead, Figure 13(b)). This intriguing observation led us to speculate whether these protrusions might be cytonemes. Cytonemes are long protrusions known to spread and deliver morphogens, such as Wnts and Shh, to neighboring or more distant cells (González-Méndez et al., 2019;Sanders et al., 2013;Stanganello & Scholpp, 2016). Given the fact that Shh is involved in guiding precrossing commissural axons toward the FP and that Shh and Wnts are both involved in guiding postcrossing axons toward the brain at the contralateral FP border, it is tempting to speculate that these protrusions might deliver such signals to the growth cones at choice points (Avilés et al., 2013). Moreover, we could appreciate how much the axons and their growth cones were intermingled within the medial FP basal feet which also formed long dynamic protrusions within the commissure (black arrowheads, Figure 13(c)). In sum, the combination of our live imaging approach with a FP-specific marker will give the opportunity to further characterize the behavior of intermediate target cells with regard to axon guidance at choice points.
Ultimately, our method will be useful to get more insights into molecular mechanisms of axon guidance at a choice point, when combined with in ovo RNAi for specific gene knockdowns either in the neurons or in their environment, as exemplified with Fzd3 knockdown experiments ( Figure 12) (Andermatt et al., 2014;Baeriswyl et al., 2020;Pekarik et al., 2003). Similarly, pharmacological blockers will permit to screen for components required downstream of growth cone receptors to transduce guidance signals. Usually such experiments are conducted in vitro with cultured neurons growing axons in a very artificial environment. Thus, our method offers the advantages of an in vitro experiment in an intact complex "in vivo-like" environment. The use of specific reporters will also allow for the assessment of dynamic changes of second messengers or the actin cytoskeleton in growth cones, for example (Nichols & Smith, 2019;Nicol et al., 2011). Moreover, the use of other sets of enhancers and promoters might offer the possibility to study the dynamics of midline crossing in other subtypes of commissural neurons in the spinal cord and in the brain (Hadas et al., 2014;Kohl et al., 2012).