In toto imaging of glial JNK signaling during larval zebrafish spinal cord regeneration

Spinal cord injuries (SCIs) in humans and other mammals are debilitating and frequently associated with severe pathologies, such as paralysis, chronic pain, and loss of bowel and bladder function. These permanent consequences arise in part because mammalian spinal cords form a glial scar after injury that protects neurons from injury-induced inflammation, but also acts as a physical barrier to the reconnection of spinal cord neurons (Alunni and Bally-Cuif, 2016; Faulkner et al., 2004; Karimi-Abdolrezaee and Billakanti, 2012; Varadarajan et al., 2022). In contrast, zebrafish are able to regenerate their spinal cords after injury and regain full motor function in both larval and adult contexts (Alper and Dorsky, 2022; Becker et al., 1997). After injury, ependymo-radial glial cells (ERGs) around the central canal of the spinal cord begin to proliferate and give rise to neurons and glia (Reimer et al., 2009, 2008; Ribeiro et al., 2017). These newly generated glia and neurons have been proposed to help establish a tissue bridge across the lesion site that eventually thickens and is remodeled to accommodate growing axons, transduce neural signals, and ultimately restore motor function (Goldshmit et al., 2012; Mokalled et al., 2016; Vandestadt et al., 2021; Vasudevan et al., 2021).

Several recent reports have focused on identifying extracellular factors present in the lesion sites that instruct cellular responses to SCI. Recent studies have found that ERGs require paracrine cues, such as Wnts, Fgfs and Ctgfa (Ccn2a) to enter the cell cycle and generate progeny that migrate or are displaced toward the lesion and help construct a tissue bridge (Briona et al., 2015; Goldshmit et al., 2012, 2018; Mokalled et al., 2016; Strand et al., 2016; Wehner et al., 2017). Regrowing axons also use extracellular guidance cues, such as Syntenin-a (Sdcbp2), Tenascin-c and Contactin 1a, to navigate and cross the lesion site (Schweitzer et al., 2007; Yu and Schachner, 2013; Yu et al., 2011). Transcription factors, such as Sox2, Sox11b and Atf3, have also been reported to play roles in orchestrating pro-regenerative transcriptional responses that lead to cellular events such as proliferation and differentiation into new spinal cord tissue (Guo et al., 2011; Ogai et al., 2014; Wang et al., 2017). A better understanding of the signaling cascades that occur downstream of extracellular cues and lead to pro-regenerative transcriptional responses may provide attractive targets for therapeutic intervention in cases of mammalian SCI.

The c-Jun N-terminal kinase (JNK) signaling pathway is a mitogen-activated protein kinase pathway that has a diverse array of functions, such as regulating apoptosis, cell growth, inflammation and proliferation. The role that JNK signaling plays in any given cell depends on the cellular context in which JNK is activated as well as the type of extracellular cue that is used to activate signaling. For example, JNK signaling is understood to regulate apoptosis downstream of cellular stress and other signals such as TNFa and glutamate (Dhanasekaran and Reddy, 2017). Yet JNK signaling is also able to promote neurogenesis and neuronal migration through cues such as Wnt and DLK (Bengoa-Vergniory et al., 2014; Hirai et al., 2006), and JNK-mediated phosphorylation of JunB has been reported to be necessary for regeneration of amputated fins in zebrafish (Ishida et al., 2010). JNK signaling is an intriguing pathway to study in the context of tissue regeneration, which requires coordination of a wide range of cellular processes from apoptosis to cellular migration to proliferation, differentiation and cell growth.

Larval zebrafish have certain advantages over adults as a model system; notably, their optical transparency is well-suited to imaging regenerating tissue in real time, and they are accessible for pharmacological intervention at large scale. Recent studies have used these properties of zebrafish larvae to directly visualize neuronal recruitment and macrophage activity in the injury sites of lesioned spinal cords (Anguita-Salinas et al., 2019; Goldshmit et al., 2018; Vasudevan et al., 2021). In this study, we employ larvae to perform a pharmacological shelf screen and find a key role for JNK signaling in the formation of a pro-regenerative tissue bridge during spinal cord regeneration. By generating a transgenic line expressing a JNK activity sensor in glial cells, we monitor JNK signaling in vivo and describe its dynamics during the growth and regeneration of the spinal cord. Our study provides a prototype for visualization of molecular signaling events during spinal cord regeneration at new resolution, and implicates a new signaling pathway in this process.

To identify signaling pathways required for spinal cord regeneration, we performed a small shelf screen of compounds targeting classic developmental signaling pathways. We used a GFAP:EGFP transgene to visualize glial bridging in order to assess the effect of each of these compounds on spinal cord regeneration after injury at 3 days post-fertilization (dpf). We calculated a glial bridging index by comparing the width of the thinnest point of the bridge to the width of the unlesioned spinal cord directly rostral to the lesion site (Fig. S1A). Approximately 30 larvae were assessed per treatment. Animals that did not form bridges were considered to have a bridge width and corresponding glial bridging percentage of zero, and these animals were included in the analysis of bridging percentage (Fig. S1B). We also recorded the percentage of animals possessing at least one continuous EGFP⁺ fascicle that spanned the lesion site as an alternative qualitative metric of spinal cord regeneration, consistent with previous studies in larval zebrafish (Wehner et al., 2017).

We observed no measurable effect on bridging at 2 days post-injury (dpi) after manipulations of vitamin D, Notch, retinoic acid, MEK, Sonic Hedgehog, Egf, or ERK signaling pathways (Fig. 1A,B). An AMPK agonist (AICAR) produced a modest 7% increase in the thickness of glial bridges when applied at 15 µM, suggesting that AMPK signaling may play a pro-regenerative role (Fig. 1A). Among all chemicals examined, the most prominent inhibitions of tissue bridging were observed upon treatment with the JNK antagonist SP600125, the Tgfβ receptor antagonist SB505124, and the Fgf receptor antagonist PD173037 at 15 μM each (Fig. 1A,B). We also compared the glial bridging percentage of each treatment without considering animals that failed to form a bridge, observing that Fgfr inhibition, Tgfβ receptor inhibition and JNK inhibition produced glial bridges that were 21%, 22% and 28% thinner than vehicle controls respectively (Fig. S1C). Fgf signaling has previously been implicated in glial bridge formation during spinal cord regeneration, indicating that this screen is sufficiently sensitive to identify signaling pathways that contribute to glial bridging (Goldshmit et al., 2012, 2018; Xu et al., 2022). Given its novelty in spinal cord regeneration, we focused our efforts on JNK signaling as a potential effector.

To interrogate further the role of JNK signaling in spinal cord regeneration, we examined larger numbers (n≈100) of animals at 1 and 2 dpi, again assessing glial bridging in GFAP:EGFP larvae. JNK inhibition reduced the frequency of glial bridge formation by 32% and decreased the width of glial bridges by 50% compared with vehicle-treated controls (Fig. 1C-E). By 2 dpi, 100% of vehicle-treated animals formed glial bridges, with an average bridge diameter ratio of 75%. In contrast, only 70% of JNK-inhibited animals demonstrated glial bridge formation, and those bridges were 60% thinner than vehicle-treated controls (Fig. 1D,E). Bridge diameter was comparable between JNK-inhibited animals at 1 and 2 dpi (Fig. 1D). These observations were consistent even after removal of animals that failed to demonstrate glial bridges (Fig. S1D). Our findings indicate that JNK inhibition leads to a defect in glial bridging. Two possible explanations for this defect are decreased production of bridging glia or a decreased capacity for glia to migrate to the lesion site.

To test whether components of the JNK signaling pathway are present in spinal cord glia, we consulted published single-cell RNA sequencing data collected from uninjured and lesioned larval zebrafish spinal cords (Cavone et al., 2021). Zebrafish possess four JNK proteins: JNK2, JNK3, and two isoforms of JNK1 (JNK1a and JNK1b). Within GFAP⁺ glial cells, JNKs 1a, 1b and 3 are enriched in quiescent ERG progenitors and proliferating ERGs (Fig. S2A). JNK2 is enriched in proliferating ERGs, suggesting that regulation of JNK may play a role in ERG proliferation (Fig. S2A). Additionally, expression of both JNK1a and JNK1b is enriched in GFAP⁺ cells after SCI, further supporting a role for glial JNK signaling in spinal cord regeneration (Fig. S2B).

In zebrafish larvae, regenerating axons can traverse the lesioned spinal cord environment independently of glial bridging (Wehner et al., 2017). We therefore assessed axonal bridging after SCI by whole-mount immunohistology for acetylated α-tubulin. JNK inhibition resulted in axon bridges that were similar in width to control animals at both 1 and 2 dpi (Fig. 1F-H, Fig. S1E). Interestingly, the frequency of bridged injuries was similar at 1 dpi but 32% lower at 2 dpi compared with vehicle controls (Fig. 1G,H). This suggests that although JNK inhibition does not affect the establishment of neural bridges it results in neural bridges that are less stable in the longer term. Thus, JNK signaling is required for glial bridge formation after SCI, whereas neural bridging appears to be less reliant on this pathway.

To visualize JNK activity in the larval spinal cord, we employed a kinase translocation reporter (KTR) (Regot et al., 2014). KTRs represent a class of ratiometric reporters used to visualize kinase dynamics in in vivo contexts such as zebrafish scale regeneration and mouse blastocyst lineage determination (De Simone et al., 2021; Simon et al., 2020). The JNK KTR consists of a fragment of cJun fused to a nuclear export sequence and the fluorescent protein Venus (Fig. 2A). The cJun protein fragment acts as a docking site for JNK for phosphorylation of the nuclear export sequence, which in turn shuttles the protein out of the nucleus when JNK activity is high. Conversely, the KTR is sequestered in the nucleus when JNK activity is low (Fig. 2A) (Regot et al., 2014). By comparing the ratio of nuclear to cytoplasmic Venus fluorescence, one can quantify relative JNK activity in living organisms at single-cell resolution (Fig. 2B, Fig. S3, Movie 1). Given the JNK-mediated defect in glial bridging we observed, we established a transgenic line expressing this KTR under the control of the GFAP promoter (GFAP:cJun-NES-Venus^pd390) to monitor JNK signaling dynamics in zebrafish glia. Correspondingly, to visualize glial nuclei in vivo and enable quantification, we established a transgenic line expressing a histone-tagged mCherry under the control of the GFAP promoter (GFAP:H2A-mCherry^pd391). The combination of GFAP:cJun-NES-Venus and GFAP:H2A-mCherry transgenic lines will subsequently be referred to as JNK-KTR.

KTRs have been employed to measure kinase activity both in vitro and in vivo in cells with large cytoplasmic area around the nucleus, such as scales, muscle, and imaginal disk epithelial cells in Drosophila (De Simone et al., 2021; Mayr et al., 2018; Regot et al., 2014; Yuen et al., 2022). However, ERGs possess a small volume of cytoplasm in the soma and long, thin processes that we expected to present additional challenges to quantification of KTR activity. Therefore, to test whether this line accurately and sensitively reflects JNK activity, we imaged JNK-KTR larvae under the effects of either the JNK inhibitor SP600125 or anisomycin, which blocks translation globally and acts as a JNK agonist. We quantified JNK activity by comparing the average Venus fluorescence within the nuclear area (defined by a mask of GFAP:H2A-mCherry fluorescence) and the average Venus fluorescence directly surrounding the nucleus in the soma (defined by a dilation of the nuclear mask and subsequent subtraction of the original nuclear area) (Fig. 2B).

First, we examined JNK-KTR under the effects of SP600125 to test whether JNK-KTR is responsive to induced changes in JNK activity. JNK-KTR larvae treated with JNK antagonist for 3 h at 72 hours post-fertilization (hpf) showed a measurable downward shift in the distribution of single-cell KTR values and significantly lower mean KTR values across the entire tissue (Fig. 2C-E; additional examples in Fig. S4A and Movie 2). We noted that some cells retained high JNK activity even after JNK inhibition, suggesting that the tissue may not respond uniformly to inhibitor treatment. To test whether SP600125 treatment was inhibiting JNK signaling in cells with high baseline JNK activity as well as in cells with low baseline activity, we calculated the ratio of cells with high JNK activity (defined as KTR value >1.0) to the total number of GFAP⁺ cells before and after SP600125 treatment. We found that fewer cells exhibit high JNK activity after SP600125 treatment (Fig. 2F). Thus, JNK-KTR is broadly sensitive to decreases in JNK signaling and is sufficient for quantification of JNK signaling with high spatiotemporal resolution in ERGs in the larval zebrafish spinal cord.

To test whether activation of JNK-KTR was detectable, we treated larvae with anisomycin, which has been shown to stimulate JNK signaling in zebrafish larvae (Yuan et al., 2021). We treated 96 hpf embryos for 10 h and collected measurements at 3 and 10 h post treatment to test whether the sensor is sensitive to rapid changes in JNK activity, and to attempt to measure the theoretical maximum value of the sensor, respectively. Anisomycin treatment shifted the distribution of single-cell KTR values upwards and increased the mean of KTR values across the entire tissue at both time points (Fig. 2G-I; additional examples in Fig. S4B). Additionally, anisomycin treatment increased the percentage of cells with high JNK activity, indicating that this increase in activity is global and not driven by subpopulations of GFAP⁺ cells (Fig. 2J).

Across antagonist and agonist experiments, 90% of KTR values measured between 0.4 and 1.7; thus, we chose those values as the functional range of JNK-KTR activity and limited analysis in all experiments to nuclei within that range. It is important to note that this dynamic range of the sensor may not cover the entire range of cellular JNK activities; in addition, the relationship between JNK activity and the KTR read-out may not be linear. Nevertheless, this is a typical limitation of live kinase sensors, and we are able to take full advantage of relative measurements by comparing JNK signaling within the same animal and even within the same populations of cells over time by in vivo imaging (Movies 1 and 2). We conclude that KTR technology can be applied in the larval zebrafish spinal cord to visualize dynamics of JNK kinase activity, and that JNK signaling can be measured in whole populations of glial cells in the larval zebrafish spinal cord.

We first used JNK-KTR animals to investigate the spatiotemporal dynamics of JNK signaling in ERGs in the spinal cords of developing zebrafish. The GFAP promoter directs strong expression in the developing spinal cord beginning at 24 hpf (Bernardos and Raymond, 2006). We chose to assess JNK activity every 24 h from 48 to 144 hpf, as we found nuclei in the spinal cord to be too densely packed to segment properly with our imaging protocol prior to 48 hpf (Fig. 3A; additional examples in Fig. S5A). We found that generally over the course of development, GFAP:H2A-mCherry⁺ cell numbers decreased in our imaging frame, which we attribute both to glia differentiating from GFAP⁺ cells to GFAP⁻ cells and to the physical expansion of the tissue, which lowers the density of GFAP⁺ cells within the imaging frame (Fig. 3A). The mean KTR value across the tissue peaked at 72 hpf before dropping at 96-144 hpf (Fig. 3B). Similarly, the larval spinal cord contained the highest proportion of cells with high JNK activity at 3 dpf, suggesting this trend is due to an increase in JNK signaling throughout the tissue and not an increase in JNK signaling in individual cell populations (Fig. 3C). Accordingly, we found no significant change in JNK activity along the anteroposterior (AP) axis at any time point from 48 to 144 hpf (Fig. 3D, Figs S5B and S6A).

Whereas AP bias was undetectable in the developing spinal cord, we observed a measurable dorsoventral (DV) bias in glial JNK activity during spinal cord development, with 5% and 20% higher ventral values at 72 and 96 hpf, respectively (Fig. 3E,F). Both of these time points also demonstrated a smaller local maximum in JNK activity located more dorsally in the tissue. Interestingly, we did not observe a DV bias in JNK activity at 48 and 144 hpf and observed a dorsal bias at 120 hpf (Figs S5C and S6B). This suggests that JNK signaling may be transiently patterned dorsoventrally in the developing spinal cord. Secreted factors known to establish the DV pattern of the developing spinal cord are Shh, secreted by the ventral floorplate, and BMPs and Wnts, secreted by the dorsal roofplate (Lee and Jessell, 1999). We posit that JNK signaling in the developing spinal cord might be organized by one or more of these or other ligands that influence DV fate decisions during spinal cord development.

To characterize the spatiotemporal dynamics of JNK signaling after a SCI, we performed spinal cord transection at 72 hpf. We found that the mean glial JNK activity value across all cells in the image frame was 10% higher at 2 dpi compared with age-matched controls (Fig. 4A,B; additional examples in Fig. S7A). Additionally, the percentage of cells with high JNK activity was 20% higher at 2 dpi, suggesting this increase is due to nascent induction of JNK signaling in cells that would otherwise be inactive, rather than solely boosting JNK activity in already active cells (Fig. 4C).

To identify spatial changes in JNK activity compared with uninjured animals, we examined JNK activity at 1, 2 and 3 dpi binned along AP and DV axes (Fig. 4D-G, Fig. S7B,C). Although we did not identify any difference in global JNK activity at 1 dpi compared with uninjured controls, we nevertheless observed significant differences in both AP and DV distributions of JNK activity at this time point compared with the distribution of JNK activity in controls (Fig. 4D,H). Most notably, the ventral bias in JNK activity that we measured at 4 dpf is diminished in age-matched 1 dpi spinal cords (Fig. 4D, Fig. S5B). This indicates that the normal developmental signaling milieu in the larval spinal cord is likely disrupted by SCI and regeneration. This result supports previous findings by other groups suggesting that spinal cord regeneration in larval zebrafish is a distinct process from spinal cord development and does not necessarily recapitulate developmental cell behaviors (Alper and Dorsky, 2022).

At 2 and 3 dpi, JNK activity values showed a small local maximum (calculated at 5% compared with the lowest measured bin average at both time points) anterior to the lesion (Fig. 4F,G, Fig. S7B). The AP distribution of JNK activity at 2 dpi differed significantly from the distribution in age-matched uninjured controls (Fig. 4I), and the local maximum in JNK activity at 2 dpi was the point with the greatest difference between the two curves, suggesting that the injury-induced increase in JNK activity is not uniform across the tissue. Of note, JNK activity was not increased at the lesion site itself, but rather in the rostral stump directly anterior to the lesion.

To test whether the increase in JNK activity that we observed near the lesion at 2 dpi was a local increase and not a result of a global increase in JNK signaling, we imaged 2 dpi larvae and 5 dpf controls with a wider image frame (Fig. 5A). We again saw a peak in JNK signaling only near the lesion site and observed that the imaged region distal to the lesion has JNK activity more similar to uninjured controls (Fig. 5B,C). The area distal from the lesion demonstrated no difference in JNK activity at any position along the DV axis, and was not significantly different from the DV distribution of JNK activity in uninjured controls (Fig. 5D). In contrast, the area proximal to the lesion had significantly higher JNK activity on the dorsal side of the spinal cord and differed significantly from the DV pattern of JNK activity in uninjured controls (Fig. 5E). At 2 dpi, the cellular processes largely contributing to spinal cord regeneration are proliferation and production of new cells to bridge the lesion site and begin to remodel the spinal cord (Anguita-Salinas et al., 2019). The observation that JNK signaling is highest while these processes are most active suggests JNK signaling may be involved in the production of new cells during spinal cord regeneration.

In conclusion, we find that JNK signaling in GFAP⁺ cells in the developing spinal cord undergoes whole-population and regional changes during development. JNK activity peaks at 72 hpf and transiently establishes a ventrally biased pattern until 96 hpf. Spinal cord transection disrupts this DV bias and induces JNK activity in dorsally located glia near the lesion site at 2 dpi. These experiments reveal and map dynamic regulation of JNK activity during stages of spinal cord development and regeneration.

To test whether JNK signaling features and requirements involve control of glial proliferation, we generated a pcna-GFP fusion allele by inserting a monomeric EGFP (mGFP) cassette directly upstream of the stop codon of the proliferating cell nuclear antigen (pcna) gene (pcna^mGFPpd392) using CRISPR/Cas9-mediated homologous recombination (Yao et al., 2017; Zacharias et al., 2002). PCNA is a protein clamp that acts as a homotrimer to encircle replicating DNA, tethering polymerases and other proteins to the site of DNA synthesis during DNA repair and DNA replication prior to mitosis (Ellison and Stillman, 2003). After crossing a pcna^mGFP allele into the GFAP:H2A-mCherry background, we were able to identify PCNA-mGFP fluorescence in nuclei of both GFAP:H2A-mCherry⁺ (Fig. 6A) and GFAP:H2A-mCherry⁻ (Fig. 6B) cells prior to mitosis (Movie 3). Specifically, we observed speckles of strong fluorescence where PCNA-mGFP aggregates around replicating DNA in S phase, uniform nuclear fluorescence in both G1 and G2 phases, and fluorescent signals exiting the nucleus and entering the cytoplasm as the nuclear envelope breaks down in M phase (Fig. 6A,B). These fluorescence patterns are similar to PCNA patterns previously observed in dividing cells in in vitro and in vivo studies using similar fusion transgenes (Essers et al., 2005; Leonhardt et al., 2000; Leung et al., 2011; Zerjatke et al., 2017).

We performed SCI on GFAP:H2A-mCherry; pcna^mGFP larvae and treated them with SP600125 to measure SCI-induced glial cycling after JNK inhibition. Because pcna^mGFP is expressed globally in all cycling cells in zebrafish larvae, we generated a mask of the H2A-mCherry signal and digitally subtracted any PCNA-mGFP signal in GFAP⁻ cells both for visualization purposes and to ensure that we limited our cycling measurement to GFAP⁺ glial cells within the spinal cord (Fig. S8A). After JNK inhibition, PCNA-mGFP fluorescence was broadly decreased in GFAP:H2A-mCherry⁺ nuclei at 1, 2 and 3 dpi (Fig. 6C-F).

To describe the spatial occurrence of glial cycling after SCI, we measured the total mGFP fluorescence at each point along the AP axis. Because pcna is most highly expressed in G1/S phase and decreases upon cell cycle exit, PCNA-mGFP fluorescence levels should correlate spatially with the areas of highest cycling (Santos et al., 2014; Thacker et al., 2003). In control animals, PCNA-mGFP fluorescence primarily localized to areas near the lesion at 1 and 2 dpi, and progressively decreased distal to the lesion. At 3 dpi, PCNA-mGFP fluorescence marked the newly regenerated junction between the two severed ends of the spinal cord (Fig. 6E). JNK inhibition changed the localization of PCNA-mGFP fluorescence within the tissue. PCNA-mGFP⁺/GFAP:H2A-mCherry⁺ cells were sparsely found in the areas proximal to the lesion at 1 and 2 dpi as in controls, but this pattern of proliferating cells did not converge at the lesion site at 3 dpi as in controls (Fig. 6F). Notably, by 2 dpi we observed sparse GFAP:H2A-mCherry⁺ nuclei in the lesion site of animals under JNK inhibition even with the decrease in cycling, indicating that glia can still access the lesion site by processes such as migration or tissue displacement (Fig. S8B).

To calculate a glial cycling index at each time point during regeneration, we manually counted GFAP:H2A-mCherry⁺ cells in S phase based on the PCNA-mGFP fluorescence pattern. Consistent with spatial mGFP fluorescence measurements, the proportion of cycling glia in vehicle-treated animals increased from 1 to 3 dpi, peaking at ∼50% of GFAP⁺ cells in the imaging window at 3 dpi. In contrast, JNK inhibition resulted in much fewer cycling glia at each time point, peaking at ∼10% of GFAP⁺ cells at 3 dpi (Fig. 6D). To assess cycling with a different marker that does not reflect DNA synthesis or repair, we performed whole-mount immunostaining for phospho-histone 3 (H3P) in GFAP:H2A-mCherry larvae. H3P is only present in M phase when cells have condensed chromatin (Elmaci et al., 2018). We observed a ∼90% decrease in the number of cells co-expressing H3P and mCherry at 2 dpi in JNK-inhibited larvae (Fig. 6G,H). We conclude that JNK signaling is required for glial cell cycling after SCI, and that even transient inhibition of JNK signaling is sufficient to cause lingering effects on the spatiotemporal dynamics of injury-induced cycling.

Larval zebrafish provide useful models for executing chemical and genetic screens. Knockout screens have been employed with targeting gRNAs to provide insight into specific genes that play roles in spinal cord regeneration (Keatinge et al., 2021; Klatt Shaw et al., 2021). Pharmacological screens, by contrast, allow for manipulation of activities of multiple related proteins simultaneously in a relatively non-invasive manner. Here, a limited shelf screen has revealed that the JNK inhibitor SP600125, an ATP-competitive inhibitor of JNKs 1, 2 and 3, limits glial cell proliferation and tissue bridging during spinal cord regeneration in larval zebrafish. We also introduced a JNK translocation reporter and identified context-specific dynamics of JNK signaling as visualized in whole populations of GFAP⁺ spinal cord glial cells during development and regeneration at single-cell resolution.

JNK signaling is highly active in the CNS and plays an important role in regulating protein function in a number of developmental and homeostatic contexts. Phosphorylated JNK is concentrated in the growth cones of growing axons, and is crucial for proper axonal pathfinding both during development and during axon regrowth after axotomy (Oliva et al., 2006; Qu et al., 2013). Additionally, JNK signaling is known to regulate programmed cell death of motor neurons in the developing spinal cord, and inhibition of JNK signaling after SCI in mice has a neuroprotective effect, highlighting the varied roles JNK is known to have in neurons within the CNS (Messina et al., 1996; Repici et al., 2012; Sun et al., 2005). We were not able to observe a measurable defect in neural bridge formation at 1 or 2 days after SCI during global JNK inhibition. This result may point to secondary mechanisms of re-innervation across a spinal cord lesion beyond the direct growth of axons. Previous work has reported that early stages of spinal cord repair in zebrafish are characterized by the recruitment and migration of a functionally mature neuron population to the lesion site that precedes stem cell-derived neurogenesis (Vandestadt et al., 2021). Further research is warranted into the difference between these modes of circuit reformation, as it remains unclear whether bridging by precursor neurons alone is sufficient to produce functional sensorimotor recovery in the regenerated spinal cord.

In contrast to JNK signaling in neurons, the role of JNK signaling in glia of the CNS has been less well characterized. In Drosophila, JNK signaling has in glia has been linked to cell engulfment and the clearance of neuronal debris after nervous system injury (Purice et al., 2017). In mice, JNKs 1 and 3 have been implicated in regulating the balance between neural progenitor maintenance and neurogenesis in the adult hippocampus (Castro-Torres et al., 2019). Here, we report a role for JNK signaling in the proliferation of ERGs after SCI. JNK signaling is induced locally at the lesion site in a pattern distinct from the surrounding developmental context, suggesting that the extracellular cues driving this aspect of the injury response are specific to regeneration. It remains to be seen whether the downstream effects of JNK signaling in this context are consistent with known roles for JNK signaling in glia, or also represent a role that is regeneration preferential or specific.

We find that JNK signaling can be quantified in GFAP⁺ glial cells in the same animal over the course of multiple days with a KTR sensor, eliminating the aspect of variability among animals. In general, techniques to measure kinase activity can be divided into two classes: single fluorophore-based sensors and sensors that use multiple fluorophores (known as fluorescence resonance energy transfer, or FRET) (Maryu et al., 2018). One of the major advantages of KTRs over kinase activity reporters of other classes is that, because they only rely on a single fluorophore (not including the nuclear marker), they can be readily multiplexed by incorporating differently colored fluorophores (Maryu et al., 2016; Regot et al., 2014). This can allow simultaneous assessment of discrete kinase pathways within individual cells, and can identify interplay of different kinase activities in during spinal cord development and regeneration. It had been unclear to us prior to this work whether tissues that are densely packed with cells with a low ratio of cytoplasm to nuclear volume, like spinal cord glial tissue, would be adequate subjects for this type of imaging. However, even in this challenging context, we found we could achieve proper segmentation of the nucleus from the cytoplasm to enable the extraction of quantitative information in toto. Other KTRs have also been developed that report the signaling activity of kinase pathways, such as Erk, AKT, p38 and PKA (Maryu et al., 2016; Regot et al., 2014). It is likely that these other KTRs may also be used to investigate kinase signaling dynamics in larval zebrafish spinal cord tissues.

KTRs can also be multiplexed with distinct classes of fluorescent reporters to correlate kinase activity with other biological processes in a variety of cell types. Using the GFAP:H2A-mCherry line, we imaged the developing spinal cord at high frame rates and could visualize subcellular processes, such as interkinetic nuclear migration within ERG processes and tissue retraction, immediately after SCI (Movies 1, 3, 4). It should be possible to refine our toolset to generate quantitative relationships between kinase activity and intricate cellular/subcellular behaviors such as these. Additionally, GCaMPs form a class of calcium sensor that has been widely used to report neuronal activity in living animals, including zebrafish (Del Bene et al., 2010; Muto and Kawakami, 2016; Muto et al., 2013). Motor neurons and other neural subtypes are present in the spinal cord and assembled into functional circuits as early as 96 hpf, and they possess small volumes of cytoplasm in their soma surrounding their nucleus, similar to glia (Borla et al., 2002). We expect it would be possible to image kinase activity dynamics alongside neural activity in living zebrafish embryos. Combinations of KTRs with cell cycle reporters, such as the FUCCI system, would allow for monitoring of kinase pathways that regulate production of new cell populations during tissue growth and regeneration. Genetically encoded transcriptional reporters are also available in the zebrafish, allowing direct visualization of both kinase activity and downstream transcriptional responses to stimuli within the same cell. We thus expect our study to help motivate new applications and studies of signaling in whole cell populations during spinal cord development and regeneration.

Wild-type and transgenic fish used in this study were of the outbred Casper strain to facilitate in vivo imaging in larvae. Embryos were raised in 28°C egg water until 5 dpf and then transferred to the aquarium system, unless imaging at later time points was required, in which case larvae were euthanized after use. Clutchmates were used for all experiments, and matings were timed to ensure imaging and experimental interventions were performed at consistent times during development. All experiments with fluorescent reporters were performed in hemizygous animals. All experiments with animals were approved by the Animal Care and Use Committee at Duke University. Newly generated strains are described below:

The sequence for cJun-NES-Venus was sourced in a Gateway entry vector pENTR-JNKKTRClover (Addgene plasmid #59139, deposited by Markus Covert). The final construct was generated via multi-fragment Gateway cloning in combination with P5E-GFAP, P3E-pA and pDestTI2. The GFAP:cJun-NES-Venus construct was co-injected into one-cell-stage Casper embryos with I-SceI at a concentration of 100 ng/ml. Three founders were isolated and propagated, with one line used for this study. The allele designation for GFAP:cJun-NES-Venus is pd390.

The sequence for H2A-mCherry was amplified via PCR and a-tailed using Taq polymerase, then subcloned into the pCR8/GW/TOPO vector via TA cloning. The final construct was generated via multi-fragment Gateway cloning in combination with P5E-GFAP, P3E-pA and pDestT2. The GFAP:H2A-mCherry construct was co-injected into one-cell-stage Casper embryos with I-SceI at a concentration of 100 ng/ml. Two founders were isolated and propagated, with one line used for this study. The allele designation for GFAP:H2A-mCherry is pd391.

The pcna^mGFP knock-in allele was generated using CRISPR/Cas9-mediated homologous recombination as described with modifications (Yao et al., 2017). Briefly, two gRNAs targeting the pcna stop codon region (5′-CTCTTCTGCTTGGATTTAGGAGG-3′) and GFP sequence (5′-GGCGAGGGCGATGCCACCTACGG-3′) were designed and synthesized as previously described (Han et al., 2019). To generate the template construct, the mGFP sequences were amplified from the cyp24a1 targeting vector and silent mutated to avoid targeting by the GFP gRNA (Han et al., 2019). The homologous arm sequences (793 bp left and 800 bp right) were amplified from zebrafish genomic DNA. Then, the left arm, mutated mGFP and right arm sequences were flanked with two opposite GFP gRNA targeting sequences and cloned into an assembly vector using Golden Gate cloning to ensure mGFP was in-frame with pcna. Finally, the template construct (at 25 ng/µl) was co-injected with Cas9 protein (at 250 ng/µl), pcna-gRNA and GFP-gRNA (both at 25 ng/µl) into one-cell-stage embryos. The founders were outcrossed and screened by GFP expression and F1 embryos were PCR verified and Sanger sequenced to ensure precise insertion. The allele designation for pcna^mGFP is pd392.

Zebrafish larvae were anesthetized in 0.01% Tricaine (Sigma-Aldrich, T0377-100G) and embedded laterally in 1% low-melt agarose on a plate of cured 1% standard agarose. Once the low-melt agarose solidified, egg water with 0.5× Tricaine was added to the top of the dish to keep the embryos hydrated. Images were acquired using a Zeiss 880 confocal microscope with a 20× water immersion lens at room temperature. Using this setup, single embryos could be successfully imaged for up to 24 h at a time, replacing media as necessary to account for evaporation.

Larvae were anesthetized at 72 hpf in 0.01% Tricaine and transferred to a clean Petri dish lid, which was pre-scored with a needle as described by Briona and Dorsky (2014). Water was aspirated from the surface until the larvae lay flat and an incision was made with a 25-gauge needle directly above the notochord. For anatomical consistency, the bulge of the yolk sac was used as a landmark to orient the injury site on the AP axis. Injuries were performed in either GFAP:EGFP or GFAP:H2A-mCherry animals, and the respective green or red fluorescence was used to visually confirm full transection on a dissecting scope. Animals with notochord injuries were identified by visual examination of the notochord on a dissecting scope and excluded from analysis, as these fish consistently fail to form tissue bridges after injury. Bridging index was calculated by dividing the width of the bridge at the thinnest point by the width of the uninjured spinal cord directly anterior to the lesion.

Pharmacological treatments were performed at 3 h post-injury to allow the epithelial wound to close without chemical perturbation. Larvae were placed in egg water containing the respective concentration of pharmacological agent for 24 h in darkness. After 24 h of treatment, larvae were placed back in fresh egg water and subjected to a standard photocycle for the duration of the experiment. A list of pharmacological agents and concentrations used is provided in Table S1.

Image processing and data analysis were performed in Fiji (ImageJ) and custom-written MATLAB (MathWorks) R2022b software adapted from De Simone et al. (2021). Discrete z-stacks covering contiguous parts of the spinal cord were stitched into a single z-stack using the built-in ImageJ plugin based on image cross-correlation (Preibisch et al., 2009). Then, z-stacks were rotated to position the spinal cord as parallel as possible to the x-axis, using H2A-mCherry nuclear signal as reference. For the purpose of nuclei segmentation and visualization, this nuclear channel was equalized by contrast-limited adaptive histogram equalization.

For JNK activity quantification, nuclei within the entire z-stack were segmented by TGMM software using the equalized H2A-mCherry signal (Amat et al., 2014). A mask corresponding to each nucleus was drawn using nuclei segmentation, then dilated by three pixel lengths, and the difference of the dilated mask and the nuclear mask was taken to define the cytoplasmic mask. Average cJun-NES-Venus fluorescence was then calculated in both nuclear and cytoplasmic regions. JNK activity was measured as the ratio of cytoplasmic to nuclear average cJun-NES-Venus fluorescence levels.

Because ERGs have a large nucleus compared with the volume of their soma, these dilated masks sometimes included extracellular space. To account for this, any pixels with a Venus fluorescence value of less than 10 arbitrary units (AU) were excluded from the area of the cytoplasmic mask, as we did not observe any areas of nuclear-adjacent ERG cytoplasm with a fluorescence value of less than 20 AU in any randomly sampled example cells. This step also served to discard cells that maintained H2A-mCherry fluorescence but no longer expressed GFAP because of differentiation from GFAP⁺ precursors, as H2A-mCherry is a stable fluorophore that persists in GFAP⁻ daughter cells of GFAP⁺ progenitors. Additionally, to account for occasional undersegmentation of nuclei of differing sizes, we excluded any pixels in the cytoplasmic mask with an mCherry fluorescence value of greater than 10 AU. Both of these computational steps decreased the measured cytoplasmic area, so to ensure that our results were not being biased by poorly segmented cells, we excluded any cells with a modified cytoplasmic area that was less than half of the original cytoplasmic area from analysis.

Larvae were euthanized in 0.9% 2-phenoxyethanol solution and fixed in freshly prepared 4% paraformaldehyde for whole-mount staining. Whole-mount embryos were mounted on slides in glycerol and imaged using a Zeiss LSM 700 confocal microscope. Primary antibodies used in this study were mouse anti-acetylated alpha tubulin (Sigma-Aldrich, T6793; 1:1000) and rabbit anti phospho-histone H3 (Cell Signaling Technologies, 9701S; 1:200). Secondary antibodies used were Alexa Fluor 546 goat anti-mouse and Alexa Fluor 633 goat anti-rabbit antibodies (Thermo Fisher Scientific, A-11003 and A-21071).

Nonparametric statistical tests were chosen in cases in which the data were not expected to conform to a normal distribution. Bridge thickness in Fig. 1 resembles a bimodal distribution, as animals that failed to form a bridge form a cluster of data at zero. For this reason, a Kruskal–Wallis ANOVA with Dunn's multiple comparisons test was performed to compare the data for the shelf screen in Fig. 1A, and a two-tailed Mann–Whitney test was used for the pairwise comparisons in Fig. 1D,G. The data in Fig. 2D,H represent measures of individual cells within a single animal. Because these data are a population of discrete cells and not replicates of a given condition, we chose to compare these data with a two-tailed Mann–Whitney test rather than a parametric test. All other data were compared with standard parametric analyses, with Welch's correction in cases with unequal variance between conditions. All error bars represent s.d. unless otherwise stated.

We thank Jim Burris, Larry Frauen, Colin Dolan, Kelly Scanlon and Daniel Stutts for zebrafish care; and Ilaria Merutka, Dillon King and Marissa Gutenberg for comments on the manuscript.

The peer review history is available online at https://journals.biologists.com/dev/lookup/doi/10.1242/dev.202076.reviewer-comments.pdf.