Reactive oligodendrocyte progenitor cells (re-)myelinate the regenerating zebrafish spinal cord

Local death of mature myelinating oligodendrocytes is one of the first events that occurs after spinal cord injury (SCI) in mammals (Totoiu and Keirstead, 2005), impeding both the survival and function of denuded axons (Ludwin and Maitland, 1984). This is followed by a secondary damage, which involves excitotoxicity and a deleterious immune response, resulting in a subacute phase of cell death and prolonged loss of oligodendrocytes that develops in a time-dependent manner and persists chronically (Crowe et al., 1997; Guest et al., 2005; Papastefanaki and Matsas, 2015).

Limited re-myelination after insult may occur by recruitment of oligodendrocyte progenitor cells (OPCs), as oligodendrocytes are post-mitotic and do not contribute to re-myelination (Franklin et al., 1997; Keirstead and Blakemore, 1997). Although OPCs are highly dynamic and respond to injury by rapidly migrating and proliferating (Hughes et al., 2013), oligodendrocytes are not sufficiently replaced and the formation of new myelin from newly generated oligodendrocytes is either incomplete or, in the case of chronically demyelinating lesions, insufficient (Keirstead et al., 1998). This has been attributed to progenitor depletion over the course of damage-progression, age-related impairment of OPC recruitment and differentiation, and axonal dystrophy (Raine and Cross, 1989; Sim et al., 2002; Mason et al., 2004; Neumann et al., 2019). Another possibility is that the inhibitory environment of the injury scar perturbs migration and differentiation of OPCs into oligodendrocytes (Fok-Seang, 1995; Plemel, 2013). Interestingly, transplantation of myelinating progenitors (Cao et al., 2005; Keirstead et al., 2005) as well as activation of endogenous progenitor pools (Llorens-Bobadilla et al., 2020) have been shown to enhance myelination capacity and, hence, improve conduction recovery after injury. Given the importance of myelination in the mechanical, trophic and metabolic support of spared and regrown axons (Fünfschilling et al., 2012), this highlights the importance of a holistic understanding of OPC biology towards enhancing re- and de novo myelination, and potentially functional recovery, after injury of the central nervous system (CNS).

In contrast to mammals, adult zebrafish possess a high CNS regenerative capacity and efficiently regenerate injured brain and spinal cord tissue (Becker et al., 1997; Kaslin et al., 2008; Reimer et al., 2008; Kroehne et al., 2011). For the latter, remarkable recovery of function has been documented. Although spinal cord transection results in full paralysis caudal to the lesion site (Becker et al., 2004; Mokalled et al., 2016), zebrafish are able to regain swimming activity by 42 days post lesion (dpl), a process that is dependent on regeneration of axonal tracts across the lesion site (Becker et al., 2004).

Although, the transcription factors as well as the molecular mechanisms governing lineage progression of oligodendroglial cells and myelination are largely conserved between zebrafish and mammals, differences might exist in the dynamics of OPC/oligodendrocyte responses after injury, which may contribute to the different regenerative outcomes. However, to date, a detailed analysis of oligodendroglial cell behaviour after SCI in a vertebrate of high regenerative capacity is lacking.

Here, we have used an adult zebrafish spinal cord transection model (Becker et al., 1997) to provide novel insight into OPC responses in a regenerative context. We find that, after spinal cord transection, myelinating oligodendrocytes are lost close to the injury site. OPCs appear to largely survive the primary and secondary damage, enter an activated state, increase proliferation and differentiate into oligodendrocytes, allowing mature oligodendrocytes to re-establish their population within 2 weeks post-injury. Transcriptome analysis of OPCs after lesion reveals that activated OPCs upregulate genes related to OPC differentiation and maturation into myelinating oligodendrocytes. Hence, adult zebrafish OPCs represent a highly dynamic cell population that contributes to the regeneration response of the injured spinal cord.

To characterize the dynamics of oligodendroglial lineage cell response after injury in the adult zebrafish spinal cord, we first analysed oligodendrocytes using the transgenic Tg(mbp:eGFP) line. In this line, eGFP expression is driven by the myelin basic protein (mbp) promoter, labelling oligodendrocyte cell bodies and myelin sheaths of actively myelinating mature oligodendrocytes (Jung et al., 2010). After spinal cord transection, the number of mbp:eGFP⁺ oligodendrocytes was significantly reduced at 3 and 7 dpl in the spared spinal cord tissue immediately adjacent to the transection site, when compared with sham controls (Fig. 1A,B). Furthermore, myelin sheaths were damaged by the acute insult (Fig. 1A, insets). Thus, mbp:eGFP⁺ oligodendrocytes are lost early after SCI. At 14 dpl and 42 dpl, the number of mbp:eGFP⁺ oligodendrocyte cell bodies returned to pre-injury levels (Fig. 1A,B), coinciding with active axonal regrowth (Kuscha et al., 2012) and recovery of locomotor function, respectively (Becker et al., 2004). The loss of mbp:eGFP⁺ oligodendrocytes both at 3 dpl (not shown) and at 7 dpl occurred in a spatially confined manner, within a 700 µm zone around the transection site [±350 µm along the rostrocaudal (RC) axis] (Fig. 1C; Fig. S1), henceforth referred to as the lesion site (Fig. 1G). Thus, in the zebrafish spinal cord, mbp:eGFP⁺ cell loss and damage is confined to a highly restricted zone in the immediate vicinity of the transection site – consistent with previous reports describing the spatial confinement of reactive neurogenesis proximal to the site of injury in adult zebrafish (Reimer et al., 2008).

To investigate whether the loss of mbp:eGFP⁺ expression reflects loss of oligodendrocytes due to cell death, we used the TUNEL assay to quantify dying cells. To exclude the possibility that dying cells are missed due to the distance from the lesion site or their rapid clearance, we examined cell death within 100 µm from the transection site and lesion epicentre (Fig. S2) as early as 4 h post lesion (hpl) (Fig. 1D). Indeed, we could readily detect TUNEL⁺ cells at 4 hpl and at 1 and 2 dpl (Fig. 1E), confirming sufficient sensitivity of the assay. Interestingly, the number of mbp:eGFP⁺/TUNEL⁺ cells remained very low at all investigated time-points, as compared to sham controls (Fig. 1F). This raises the possibility that mbp:eGFP⁺ cells survive the injury. However, we cannot exclude the possibility that the absence of mbp:eGFP⁺/TUNEL⁺ cells is due to an altered transcriptional activity of GFP-reporter expression. Taken together, in the adult zebrafish spinal cord, the mbp:eGFP⁺ oligodendrocyte population recovers after injury, in a process that begins early after the acute insult and continues until the restoration of mbp:eGFP⁺ cell numbers.

Because the number of mbp:eGFP⁺ oligodendrocytes was re-established by 42 dpl in the lesion site, we sought to investigate myelination status at high cellular resolution using transmission electron microscopy (TEM). As the experimental set-up used in this study (full transection) precludes the unambiguous differentiation between re-myelination of previously demyelinated axons, de novo myelination of regrown axons and spared or non-transected axons, we henceforth refer to this process as (re-)myelination.

To examine the effect of SCI at the ultrastructural level, we first quantified the number of axons in the rostral part of the lesion site at 7 and 42 dpl and in sham controls. To avoid bias of small- and/or large-diameter axons, we performed the analysis in two zones of the adult zebrafish spinal cord, at defined dorsal and ventral positions (see schematic in Fig. 2). At 7 dpl, massive tissue disruption was evident in both zones examined, as indicated by the perturbed tissue architecture, the global loss of compact myelin sheaths and the presence of cellular and myelin debris, likely as a result of the primary impact of the transection (Fig. 2A,B, 7 dpl). Furthermore, we found that in both ventral and dorsal areas of the spinal cord, the total number of axons (both non-myelinated and myelinated) was reduced by half, when compared with sham controls. Although tissue integrity was restored at 42 dpl, the number of axons remained lower when compared with sham controls (Fig. 2C). To further monitor myelination dynamics, we also quantified the percentage of myelinated axons after injury (Fig. 2D). In both zones examined, we found that the percentage of myelinated axons was significantly reduced at 7 dpl. Interestingly, although detectable recovery was observed from 7 to 42 dpl, the percentage of myelinated axons remained lower when compared with sham controls at 42 dpl. Moreover, axonal and myelin sheath loss was accompanied by a marked response of the macrophage/microglia that were detected close to the transection site, and were closely associated with myelin and lipid debris, often engulfing them (arrowheads in Fig. 2E).

Taken together, these data show that almost 50% of axonal tracts rostral to the transection site are lost after injury, which persists until at least 42 dpl. Furthermore, although myelin pattern partially recovers, the percentage of myelinated axons does not fully return to pre-injury levels. Given the distinct poor regenerating capacity of ascending fibres and that descending tracts substantially retract after transection and before re-extension (Becker and Becker, 2001; Becker et al., 2005), the axons detectable in the areas examined (see Fig. 2A-D) likely represent regrowth of new axonal segments. This provides tentative evidence for re-myelination of regenerated axonal segments after SCI.

To determine the anatomical quality of myelin formation after injury, we determined the G ratio of axons (diameter of axon/diameter of axon plus myelin), as a readout for myelin thickness (Fig. 2F). Interestingly, G-ratio analysis revealed variability of (re-)myelination patterns among analysed fish (Fig. 2G,H). Specifically, in two out of three fish, the diameter of myelinated axons was higher at 42 dpl when compared with sham controls, while spinal cord transection resulted in lower frequency of medium-sized myelinated axons (Fig. 2I), when compared with sham controls (Fig. 2J). These data suggest that, at 42 dpl, (re-)myelination is not yet complete, or is partial, and that regeneration extent varies between individual fish.

Interestingly, besides the distinct (re-)myelination pattern observed, SCI resulted in a markedly altered tissue architecture in the regenerated spinal cord that persisted at 42 dpl. Severed Mauthner axons did not recover, and some myelin/lipid debris remained close to the lesion site, in spite of the restoration of tissue integrity (Fig. 3A, yellow arrowheads). Moreover, at 42 dpl, a large proportion of the spinal cord parenchyma was covered by vessels (Fig. 3A, 42 dpl), while the area of the ventricular central canal was increased already at 7 dpl and remained significantly increased at 42 dpl (Fig. 3B,C). To determine whether tissue architecture is restored and ventricle size regresses long term, we analysed fish at 1 year post lesion (ypl). Intriguingly, the central canal at the lesion site remained enlarged, when compared with sites distal to the transected area (Fig. 3D,E). Taken together, these data suggest that, although swimming activity recovers by 42 dpl (Becker et al., 2004), anatomical and myelination features of the regenerated zebrafish spinal cord do not fully return to pre-injury levels.

To monitor OPC-specific responses after SCI in zebrafish, we examined the Tg(olig2:eGFP) line, in which eGFP is driven by the olig2 promoter (Shin et al., 2003) (Fig. 4A). In the adult zebrafish spinal cord, olig2:eGFP labels OPCs in the parenchyma and a subset of ventricular radial glia that can be distinguished from parenchymal OPCs based on their size, morphology and expression of radial glia markers (Fig. S3) (Kroehne et al., 2017). We found that the number of parenchymal olig2:eGFP⁺ cells was transiently reduced at 7 dpl compared to sham controls, but returned to pre-injury levels by 42 dpl (Fig. 4B). Interestingly, however, OPCs reacted rapidly after the lesion, as indicated by marked changes in their cellular morphology. Under physiological conditions, the majority of parenchymal olig2:eGFP⁺ cells displayed a round cell body with thin bipolar processes, thereafter referred to as ‘non-reactive’ (Fig. 4A, insets and enlargements). In contrast, at 3 dpl, we observed that 92.5±1.6% of parenchymal olig2:eGFP⁺ cells were activated, as indicated by a marked increase in their cell size and the extension of long multi-branched processes (Fig. 4A, insets and enlargements; Fig. 4C), thereafter referred to as ‘reactive’. The percentage of reactive parenchymal olig2:eGFP⁺ cells remained significantly increased at 7 dpl, while the morphology of reactive OPCs returned to sham control levels by 42 dpl (Fig. 4A, insets and enlargements; Fig. 4C). Hence, these data indicate that zebrafish OPCs are a highly dynamic population that readily senses and responds to injury while maintaining its pool (Fig. 4D).

To exclude the possibility that (1) OPCs are dying at very early time-points after the injury and (2) the transient, yet significant, reduction of OPCs at 7 dpl is due to cell death, we performed the TUNEL assay and monitored cell death at 4 hpl, 1 dpl, 3 dpl and 7 dpl (Fig. 4E). Consistent with our analysis in the Tg(mbp:eGFP) line (see Fig. 1E), we readily detected an increase in the number of dying cells over baseline levels early after SCI (Fig. 4F). Interestingly, this transient increase was accompanied by a significant increase of immune cells at 7 dpl, as indicated by staining with the pan-leukocyte marker L-plastin, that could contribute to the rapid clearance of dying cells (Fig. S4). In contrast, the number of olig2:eGFP⁺/TUNEL⁺ cells remained very low, with less than one cell per section at all investigated time-points, although a significant increase over baseline levels was detectable at 7 dpl (Fig. 4G). However, given the total number of OPCs at 7 dpl, this corresponds to less than 2% of the total OPC population. Taken together, these data are consistent with a scenario in which adult zebrafish OPCs vastly survive the primary insult and do not undergo cell death after lesion.

As the number of olig2:eGFP⁺ OPCs and mbp:eGFP⁺ oligodendrocytes cells returned to control levels by 42 dpl, we examined the induction of acute OPC proliferation after SCI. To detect cycling cells and determine spatial changes in cell proliferation, we used an antibody to PCNA that labels proliferating cells (Fig. 5A) and quantified the number of olig2:eGFP⁺/PCNA⁺ cells starting from 700 µm rostral to 700 µm caudal to the lesion site. We detected a significant increase in the number of parenchymal olig2:eGFP⁺/PCNA⁺ cells within a 700 µm zone around the transection site (±350 µm along the RC axis) compared to sham controls, in which proliferation levels were lower (Fig. 5B). This suggested an immediate proliferative reaction of OPCs in the lesion site, in response to the acute insult.

As indicated previously, the majority of the olig2:eGFP⁺ cells detectable at 7 dpl displayed a reactive morphology. We therefore sought to determine whether proliferation is a property linked exclusively to reactive OPCs, or whether non-reactive cells are as likely to proliferate as reactive cells. As sham controls have a much higher proportion of non-reactive OPCs, we examined proliferating olig2:eGFP⁺/PCNA⁺ cells in the sham control group. We found olig2:eGFP⁺/PCNA⁺ cells in both populations of non-reactive and reactive OPCs (Fig. 5C). This suggests that proliferation is not specifically linked to features of cell activation, and that the morphological changes observed in reactive cells might serve functions other than exclusively proliferation (Fig. 5E). For example, reactive cells might not only proliferate to replenish the population, but also migrate in from outside the 700 µm zone, such that their morphological changes might reflect migratory cytoskeletal rearrangements. In order to investigate such a scenario, we examined the spatial distribution of OPCs by quantifying the number of olig2:eGFP⁺ cells along the RC axis (Fig. 5D). At 7 dpl, the number of olig2:eGFP⁺ cells was found to be unaltered along the RC axis, suggesting that OPC total numbers are not altered by immigration of distal cells.

Finally, to examine the cellular dynamics of the OPC population, we monitored OPC turnover at the whole-tissue level. To examine whether acute proliferation contributes to the re-establishment of the OPC population after injury, we injected EdU to label proliferating cells in the S-phase of the cell cycle, and examined the whole-mount spinal cord using tissue clearing (Susaki et al., 2014) (Fig. 5F,G). As expected, we observed that in the unlesioned spinal cord of sham controls, proliferation levels were very low, with parenchymal olig2:eGFP⁺/EdU⁺ cells representing less than 1% of the total olig2:eGFP⁺ population in the parenchyma (Fig. 5H, sham control). In contrast, at 14 dpl, the number of EdU⁺ cells was markedly increased, both in the parenchyma and around the ventricular central canal (Fig. 5H, 14 dpl). Parenchymal olig2:eGFP⁺ cells significantly increased proliferation compared to sham controls, with almost 10% of the total population retaining EdU (Fig. 5I). This indicates that, in the regenerating zebrafish spinal cord, OPCs are able to replenish their population after injury by increasing proliferation.

Our data suggest that less than 2% of OPCs are dying at 7 dpl, and that the population of mbp:GFP⁺ oligodendrocytes is re-established at 14 dpl. Therefore, a possible scenario might be that the transient decrease of OPCs at 7 dpl is due to their differentiation into oligodendrocytes. Indeed, immunostaining against Claudin K (which labels mature oligodendrocytes and myelin sheaths) and Neurofilament (which labels axonal filaments) in Tg(olig2:eGFP) fish at 7 dpl, revealed that some reactive OPCs are expressing Claudin K, while being closely apposed to axons, indicating their differentiation into pre-myelinating oligodendrocytes (Fig. 6A).

Next, we analysed the generation of new oligodendrocytes after SCI in the Tg(mbp:eGFP) line, after injection of EdU (Fig. 6B). We found that, at 14 dpl, the number of newly generated mbp:eGFP⁺/EdU⁺ cells was significantly higher when compared with sham controls (Fig. 6C,D). This suggests that, in the regenerating adult zebrafish spinal cord, OPCs differentiate into mature oligodendrocytes, and new mbp:eGFP⁺ oligodendrocytes are generated (Fig. 6E).

The molecular signature associated with OPC responses in a regenerative context is currently unknown. Because the morphology of OPCs has been previously described to define their maturation state (Huang et al., 2011), we hypothesized that reactive OPCs might become pre-myelinating, and thus eventually contribute to (re-)myelination of spared and regrown axonal segments.

To determine the molecular signature of the initiation of myelination, we transcriptionally profiled OPCs isolated from the zebrafish spinal cord at 7 dpl, when OPC activation features are markedly detectable and reactive proliferation is at its peak. As SCI results in a spatially restricted damage and cellular responses are strictly confined within 700 µm around the transection site, we dissociated, FAC-sorted and analysed the transcriptome of olig2:eGFP⁺ OPCs from ∼0.5 mm of spinal cord tissue (Fig. 7A; Fig. S5, Fig. S6A-C). Comparing lesioned (reactive OPCs) with sham control (non-reactive) samples, we identified a total of 1755 differentially expressed genes (DEGs). Of these, 1124 genes were upregulated and 631 genes were downregulated. Gene ontology (GO) analysis resulted in the classification of DEGs into six major categories, including developmental, growth and metabolic processes (Fig. 7B). As it was initially hypothesized that reactive OPCs might contribute to (re-)myelination, the DEGs were analysed based on their functional annotations, using GO terms related to the process of myelination. Reactive OPCs upregulated a plethora of genes involved in migration, cell adhesion and cytoskeletal remodelling, as well as genes related to cell cycle and differentiation, lipid metabolism, protein and vesicular transport. Importantly, reactive OPCs upregulated genes expressed in mature oligodendrocytes and related to compact myelin formation (Fig. 7C,D). Interestingly, the upregulation of differentiation-related genes in reactive OPCs was also concomitant with the upregulation of several genes associated with immune system function, such as tnfrsf11a, cxcl14, il4r.1 and stat6 (Fig. S7). Intriguingly, single-cell RNAseq showed that reactive OPCs fall into distinct clusters (Fig. S6D,E), suggesting functional heterogeneity within the OPC population after SCI.

To verify our RNASeq dataset, we performed fluorescent in situ hybridization in Tg(olig2:eGFP) fish at 7 dpl. We selected four genes (apoeb, ctgfa, sema3c and tnc) based on their high counts. Quantification of the number of parenchymal OPCs expressing the respective genes confirmed their upregulation after SCI, validating our transcriptome data (Fig. 8). Taken together, these data suggest that, in the adult zebrafish spinal cord at 7 dpl, OPCs initiate transcriptional programmes related to differentiation, membrane formation and, thus, oligodendrocyte maturation.

In this study, we provide previously unreported insights into oligodendroglial lineage cell biology after SCI in a regenerating vertebrate. Using a spinal cord transection paradigm, we show that the initial cellular processes orchestrating injury responses in adult zebrafish resemble those seen in mammals (Papastefanaki and Matsas, 2015). We find that >60% of oligodendrocytes are lost within 7 dpl. Intriguingly, within the same time-frame we did not detect significant death of oligodendrocytes. This raises the possibility that, in the zebrafish spinal cord, oligodendrocytes are injury resistant. However, owing to technical limitations we cannot draw definite conclusions. These include the possibility that mbp:eGFP⁺/TUNEL⁺ oligodendrocytes have been missed due to (1) GFP expression downregulation in the Tg(mbp:eGFP) reporter line and/or (2) the segregation of membrane-tethered signal occurring during myelin formation by oligodendrocytes (Aggarwal et al., 2011). Similarly, we cannot exclude the possibility that pre-existing oligodendrocytes might de-differentiate to become (re-)myelinating OPCs, owing to lack of a functional Cre-based genetic lineage tracing. A relatively weak expression level of Cre is required in an inducible CreER^T2 system in order to avoid unconditional recombination (Kroehne et al., 2011). Moreover, as OPC differentiation is induced by tamoxifen (Gonzalez et al., 2016), this would interfere with the usage of the CreER^T2 system and its tamoxifen inducer. In the absence of such a marking technique, it is inherently difficult to distinguish between a possible contribution from spared oligodendrocytes that are able to re-myelinate (Duncan et al., 2018b; Yeung et al., 2019; Bacmeister et al., 2020), and spared oligodendrocytes that de-differentiated and initiated secondary (re-)myelination.

Similar to what has been shown in a mouse model of demyelination (Romanelli et al., 2016), at the ultrastructural level the primary impact causes a clear disruption of tissue architecture, axonal and myelin sheath loss, accompanied by a pronounced inflammatory response of activated macrophage/microglia. This is in accordance with studies in mammals, where the initial injury affects blood vessels, including blood-brain barrier (BBB) disruption, leading to the immediate infiltration of peripherally derived immune cells to the injury site (Habgood et al., 2007). Surprisingly, the data acquired in this study suggest that, in contrast to mammals, zebrafish oligodendrocyte numbers return to pre-injury levels by 42 dpl, the time-point of functional recovery (Becker et al., 1997). Interestingly, by this time, total axon number and percentage of myelinated axons, have not returned to pre-injury levels. In addition, the majority of myelinated axons is insulated by myelin sheaths of lower compaction that resemble newly forming myelin sheaths during development (Bunge et al., 1961). Taken together, our data suggest that in the adult zebrafish spinal cord, functional recovery coincides with partially recovered axonal profiles and myelination. This is consistent with previously published data in zebrafish (Kuscha et al., 2012) and other vertebrates, where thinner myelin sheaths were shown to stabilize axonal survival (Duncan et al., 2017) and ensure proper signal conduction (Smith et al., 1979). Our results suggest that, at the point of functional restoration, (re-)myelination is not complete, and argue in favour of the possibility that neuronal activity (Mensch et al., 2015) and/or individual axon subtypes (Almeida et al., 2011; Koudelka et al., 2016) actively regulate myelin sheath formation and remodelling (Auer et al., 2018). This is further supported by the observed increase in axon diameter of myelinated axons at 42 dpl, as, similarly, axon calibre size has been previously shown to trigger myelination (Goebbels et al., 2017). Accordingly, a longer recovery and myelin refinement period might be required to monitor the extent of complete (re-)myelination after SCI in adult zebrafish. This is particularly interesting given the restricted time-window in which new myelin sheaths are formed by individual oligodendrocytes during development (Czopka et al., 2013). Full-thickness myelin sheaths have been observed to form in the spinal cord of larvae, in the optic nerve of juvenile zebrafish (Münzel et al., 2014; Karttunen et al., 2017) and in the spinal cord of adult mice (Powers et al., 2013). Future studies need to examine whether (1) the number of myelin sheaths per oligodendrocyte and/or (2) the myelin sheath length is altered, compared to pre-injury levels. Interestingly, using a larval zebrafish model of demyelination, Neely et al. recently showed that the myelination pattern of newly generated oligodendrocytes involves more myelin sheaths as compared to before demyelination, and shorter average sheath length that may, however, grow over time (Neely et al., 2020 preprint). These data challenge the notion that functional repair requires complete re-myelination and argue in favour of a necessary threshold of axonal support that subsequently ensures locomotor function, as was recently also reported for a mouse model of SCI (Duncan et al., 2018a).

An interesting observation of our study is that adult zebrafish OPCs are an apparently robust population that survives SCI. Although a significant reduction of OPC numbers was detectable at 7 dpl, coinciding with a significant increase in TUNEL⁺ OPCs, this decrease corresponded to less than 2% of the total OPC population. This is an intriguing finding, given the overt loss of OPCs that occurs after CNS injury in mammals, where at 7 dpl, nearly 50% of the NG2⁺/Olig2⁺ cells were positive for TUNEL, indicating the vulnerability of mouse NG2⁺/Olig2⁺ cells to degeneration (Payne et al., 2013). Our observation is consistent with data from larval zebrafish, in which no olig2:GFP⁺/TUNEL⁺ cells were detectable after SCI (Ohnmacht et al., 2016). Consistently, at 14 dpl, the number of newly generated oligodendrocytes was significantly increased compared to sham controls. Together, our data suggest that proliferating OPCs need to successfully progress along their lineage to give rise to mature oligodendrocytes, as was recently also observed in a focal demyelinating SCI model in rodents (Foerster et al., 2020 preprint). Interestingly, by 14 dpl, we could detect only three newly generated oligodendrocytes – as a result of progenitor proliferation – as indicated by their retention of EdU (Fig. 6D). However, at the same time-point, we have detected a total of ∼170 oligodendrocytes and, hence, the re-establishment of the oligodendrocyte population (Fig. 1B). We therefore speculate that oligodendrocytes become re-established not only through proliferation, but also directly from reactive OPCs. In this model, a fraction of reactive OPCs might be able to differentiate into oligodendrocytes when demand arises, e.g. by 7 dpl, and without proliferation. This possibility is supported by (1) our observation that myelin-related proteins, such as Claudin K, are expressed in reactive OPCs (Fig. 6A), (2) our RNAseq data showing upregulation of myelin differentiation-related genes in reactive OPCs (Fig. 7) and (3) the number of mbp:GFP⁺/EdU⁺ cells at 14 dpl (Fig. 6D). In combination with our single-cell analysis, these data are consistent with the existence of functionally distinct subgroups of zebrafish OPCs (Tsata et al., 2019; Marisca et al., 2020) that might respond differently to injury, as previously reported in rodents (Káradóttir et al., 2008).

We find that OPCs enter an activated state and proliferate, despite the concurrent marked inflammation. This points to a coupling between inflammatory response and OPC activation, also observable in the upregulation of several immune system-related genes in reactive OPCs. This is in accordance with previous studies in the regenerating zebrafish brain and spinal cord, where, in contrast to mammals, inflammation is required for the activation of radial glial progenitor cells and axon regeneration (Kyritsis et al., 2012; Tsarouchas et al., 2018). Consistent with such a possibility, the engulfing of myelin/membrane/lipid debris by activated macrophages/microglia at 7 dpl points to the efficient clearance close to the lesion site of degenerating cellular components and/or cellular debris, which might otherwise hinder regeneration (Neumann et al., 2009).

We report the first transcriptome dataset of DEGs in reactive OPCs after transection of the adult zebrafish spinal cord, showing that reactive OPCs systemically express genes related to maturation and myelination. Already by 3 dpl, OPCs have transformed from small, round, bipolar cells to enlarged, highly branched cells, exhibiting changes that, during development, correlate with OPC migration or responses to maturation stimuli and, potentially, their lineage progression to differentiating oligodendrocytes (Ono et al., 2001; Pedraza et al., 2014). Similar morphological changes have also been observed during differentiation of primary adult zebrafish OPCs from the spinal cord during co-culture with human motor neurons (Kroehne et al., 2017). Moreover, in vitro analysis of activated OPCs in adult mice has previously revealed an activation-dependent increase in migration and differentiation rates (Moyon et al., 2015). Indeed, our RNAseq data revealed that reactive OPCs upregulate genes related to distinct phases of OPC differentiation, such as migration, cytoskeletal remodelling in axon guidance and membrane formation or transport, all key processes of active myelination. Given the fact that the morphological differences of OPCs are already apparent by 3 dpl, these transcriptional changes presumably have started even earlier. Thus, zebrafish OPCs seem to readily undergo crucial functional changes, unlike mammalian OPCs, where the similar myelination genes are upregulated later and only after OPCs are successfully recruited to the injury site (Huang et al., 2011).

An important factor impeding axonal regeneration in mammals is growth inhibition by the extracellular matrix (ECM), which is also secreted by presumed OPCs (NG2⁺ cells) (Filous et al., 2014). Recent work in zebrafish has shown that ECM components in the spinal lesion site are permissive and required for axonal regrowth (Wehner et al., 2017), thus creating a regeneration-promoting environment when compared with mammals (Cigliola et al., 2020). Our RNAseq analysis revealed that reactive OPCs upregulate a plethora of genes coding for ECM and ECM-remodelling enzymes, including ctgfa, sema3c and tnc, offering the possibility that OPCs promote regeneration by secreting growth-permissive ECM in the zebrafish lesion site. Together, the genes identified here represent a tantalizing starting point to further test potential targets for the activation of endogenous progenitor cells in non-regenerating environments (Llorens-Bobadilla et al., 2020).

In summary, several lines of evidence obtained in this study support the scenario that in the regenerating zebrafish spinal cord, OPCs contribute to (re-)myelination. First, OPCs enter an activated state and efficiently proliferate despite the marked inflammation. Second, OPCs upregulate several genes involved in oligodendrocyte maturation and differentiation. Third, both activation and increased proliferation result in OPCs becoming pre-myelinating, as well as in formation of new mature oligodendrocytes. Our findings support a model in which the efficient OPC response following SCI may be sufficient to provide the level of (re-)myelination needed, thus contributing to the restoration of the injury-induced functional deficits (Fig. 9). Zebrafish OPCs therefore offer an excellent system for studying successful regeneration responses of the oligodendrocyte lineage in a vertebrate of high regenerative capacity, with the potential to developing novel therapeutic strategies for the treatment of spinal cord injuries also in mammals.

All animal experiments were performed in compliance with animal welfare legislation and were approved by the ethical approval committee of the Regierungspräsidium Dresden, Germany: TVV 53/2015 and TVV 35/2017. All efforts were made to minimize animal suffering and the number of animals used.

Fish were kept and bred in our fish facility according to standard methods (Brand et al., 2002). We used Tg(olig2:eGFP) (Shin et al., 2003) and Tg(mbp:eGFP) zebrafish (Jung et al., 2010).

Adult fish were anaesthetized by immersion in 0.25% w/vol Tricaine (Sigma-Aldrich) and the spinal cord was transected under visual control 5 mm caudal to the brainstem-spinal cord junction, as previously described (Becker et al., 1997). Sham-lesioned fish were treated equally, except that the spinal cord was left intact. Fish were allowed to survive spinal cord injury for up to 42 days. Recovery of swimming activity was confirmed by visual evaluation of all fish that were further processed for the 42 dpl time point.

Contrasted ultrathin sections were analysed on a FEI Morgagni D268 (FEI, Eindhoven, The Netherlands) or a Jeol JEM1400 Plus at 80 kV acceleration voltage. The number of axons and G-ratio were manually analysed in Fiji. Images were processed and assembled using Fiji, Adobe Photoshop CS6 and Illustrator CS6. Microsoft Excel was used to process the measured data. For full details, see supplementary Materials and Methods.

To detect cell death, the ApopTag Red In Situ Apoptosis Detection Kit (Chemicon) was used in 12-μm-thick cryosections, according to manufacturer's instructions. For full details, see supplementary Materials and Methods.

For full details on tissue preparation for immunohistochemistry and antibodies used, see supplementary Materials and Methods.

For fluorescent in situ hybridization combined with immunohistochemistry, 50-µm-thick floating vibratome sections were treated for 12 min with 5 μg/ml proteinase K (Invitrogen) followed by re-fixation for 15 min in 4% PFA-PBS. In situ hybridization with digoxigenin-labelled probes was performed as described in detail elsewhere (Wehner et al., 2017). Colour reaction was performed using SIGMAFAST Fast Red (Sigma-Aldrich) substrate. Following termination of the staining reaction by washing sections in PBS containing 0.1% Tween, sections were co-stained using anti-GFP antibody and DAPI to visualize nuclei, mounted in 75% glycerol-PBS and imaged using a Zeiss LSM 780 confocal microscope. Information on in situ hybridization probes, including primer sequences used for molecular cloning, and antibodies is provided in the supplementary Materials and Methods and Table S1.

The 5-ethynyl-2′-deoxyuridine (EdU) nucleoside analogue of thymidine (C10340, Invitrogen) was injected intraperitoneally in adult zebrafish at a concentration of 1.25 mg/ml in PBS and detected at 14 dpl with the Click-iT EdU Alexa Fluor Imaging Kit (Thermo Scientific) according to manufacturer's instructions. For full details, see the supplementary Materials and Methods.

Tissue was cleared according to the CUBIC-based clearance technique (Susaki et al., 2014). For full details, see the supplementary Materials and Methods.

Images were acquired using either an upright Axio Imager.Z2 apotome microscope with a Plan-APOCHROMAT 20×/0.8 NA objective, an upright Leica TCS-SP5 I confocal microscope or an upright Zeiss LSM 780 microscope with an HC PL APO CS 20×/0.7 NA and HCX PL APO 63×/1.2 NA objectives. Images were acquired as z-stacks with Zen or Leica software. Cleared tissue images were acquired in tile mode using an upright Zeiss LSM 780 microscope with an LCI Plan Apochromat 25×/0.8 NA water immersion objective. To minimize channel crosstalk, images were acquired sequentially. Images were processed and assembled using Fiji, Imaris, Adobe Photoshop CS6 and Illustrator CS6.

All cell counts were analysed within 700 µm of the transection site (350 µm rostral and 350 µm caudal) unless otherwise stated. Cells in vibratome sections were manually counted stereologically with the Cell Counter Plugin of Fiji to 50 µm section thickness. Automated cell counting in whole-mount cleared tissue (Fig. 5F-I) was performed in Imaris using the spot detection tool, after 3D-reconstruction. For the TUNEL assay, 12-μm-thick cryosections were taken within 100 µm from the transection site and lesion epicentre (see Fig. S2). Cell profiles were manually counted with the Cell Counter Plugin of Fiji. Every third section was analysed to exclude overrepresentation of the same cell in consecutive counts. Microsoft Excel was used to process the measured data. Graphs are shown as scatter plots with bars depicting the mean and error bars depict s.e.m. (Figs 1, 2C,D, 3, 4, 5, 6 and 8) or box-and-whisker plots (boxes depict the quartiles, whiskers depict 1.5× inter-quartile-range, outliers are plotted as dots) (Fig. 2H,I). Mean refers to the average of at least four counted slices/sections within the lesion or transection site per lesioned fish or at least four counted slices/sections taken from a similar level along the RC axis, in sham control fish. To analyse significance, P values were determined with GraphPad Prism. Details on the statistical test used in each study are given in the figure legends: ****P<0.0001, ***P≤0.001, **P≤0.01 and *P≤0.05; P>0.05 was not considered significant.

Adult fish were terminally anaesthetized in 0.1% w/vol Tricaine (Sigma Aldrich) in E3 solution with 10⁻⁵% v/v Methylene Blue. The skin and musculature were removed dorsally until the spinal cord was exposed and carefully removed with fine forceps. Extracted tissue was placed on a microscope scale slide (474026, Carl Zeiss); starting from the lesion site, a ∼0.5 mm piece was cut from the rostral part with a scalpel. For sham control group, a ∼0.5 mm tissue was dissected at the same area of the spinal cord along the rostral to caudal axis of the fish. Nine spinal cords (∼0.5 mm pieces) were used per group and placed in 1 ml of sterile Hanks' buffered salt solution (HBSS) (Gibco). Before tissue dissociation, spinal cords were kept on ice continuously to avoid cell degradation. Tissue dissociation, live cell staining and FACS of cells was carried out with modifications using an unpublished protocol developed by the Brand lab (D. Freudenreich, A. Weber and M. Brand., unpublished and Lange et al., 2020). Briefly, excised tissue was dissociated with the Neural Tissue Dissociation Kit (130-092-628, Miltenyi Bioscience) by incubating for 15 min at 37°C in dissociation buffer. Tissue digestion was stopped by the addition of 30 µl of Papain inhibitor, triturated with 10 strokes of a wide-tipped fire-polished Pasteur pipette and incubated at 37°C for 10 min. After two more trituration steps with 10 strokes of a middle and small-tipped fire-polished Pasteur pipette, respectively, and incubation for 10 min at 37°C, cell suspension was applied to a 20 µm cell strainer (BD Biosciences), mounted in a 15 ml falcon tube. After washing with 10 ml of sterile HBSS, the cell suspension was pelleted by centrifugation at 300 g for 10 min at room temperature. The supernatant was discarded and the pellet was re-suspended in 500 μl fresh sterile HBSS. To stain for viable cells, 1 µl of 2 mM Calcein Blue-AM (C1429, Invitrogen) was added to the cell suspension. The cell suspension was protected from light and incubated for 10 min at room temperature until fluorescence-activated cell sorting was carried out.

Adult zebrafish OPCs were sorted using a BD FACSAria sorter. For the detection of GFP a 488 nm excitation laser and a 530/30 bandpass filter were used. Calcein Blue-AM was detected after 405 nm excitation using a 450/40 nm bandpass emission filter. Cellular events were defined by forward- and side-scatter profile, and by the GFP signal of the transgenic line. From this selection, all events that showed incorporation of the Calcein Blue-AM were finally gated and sorted.

RNAseq was based on Smart-seq2 sensitive full-length transcriptome profiling and modified from Picelli et al. (2013). Briefly, either one or 30 GFP⁺ cells from Tg(olig2:eGFP⁺) fish were FAC sorted into single wells of a 96-well plate containing 2 μl of nuclease-free water with 0.2% (v/v) Triton-X 100 and 4 U murine RNase Inhibitor (NEB), spun down and frozen at −80°C. After thawing the samples, 2 μl of a primer mix was added. RNA was then denatured for 3 min at 72°C and the reverse transcription was performed at 42°C for 90 min after filling up to 10 μl with reverse transcription buffer mix. The reverse transcriptase was inactivated at 70°C for 15 min and the cDNA was amplified using Kapa HiFi HotStart Readymix (Peqlab) at a final 1× concentration and 0.1 μM UP-primer (UP-primer:AAGCAGTGGTATCAACGCAGAGT). The amplified cDNA was then purified using 1× volume of hydrophobic Sera-Mag SpeedBeads (GE Healthcare) and DNA was eluted in 12 μl nuclease-free water. The concentration of the samples was measured with a Tecan plate reader Infinite 200 pro in 384 well black flat bottom low volume plates (Corning) using AccuBlue Broad range chemistry (Biotium). For library preparation, 700 pg cDNA in 2 μl were mixed with 0.5 μl Tagment DNA Enzyme, 2.5 μl Tagment DNA Buffer (Nextera, Illumina) and tagmented at 55°C for 5 min. Subsequently, Illumina indices were added during PCR with 1× concentrated KAPA Hifi HotStart Ready Mix and 0.7 μM dual indexing primers. After PCR, libraries were quantified with AccuBlue Broad range chemistry, equimolarly pooled and purified twice with 1× volume Sera-Mag SpeedBeads. This was followed by Illumina sequencing on a Nextseq500, aiming at an average sequencing depth of 0.5 million reads per cell. Gene ontology was performed with DAVID 6.8 Functional Annotation Tool (NIAID/NIH).

We thank Claudia Froeb for maintaining zebrafish colonies and outstanding technical support, Dr René Bernitz (CRTD, Dresden) for the tnc in situ hybridization probe and Catherina G. Becker for critical reading of the manuscript. This work was supported by the Flow Cytometry Facility, the Light Microscopy Facility, the Electron Microscopy Facility, the Deep Sequencing Facility and the Zebrafish Facility at CRTD of Technische Universität Dresden. The Sächsisches Ministerium für Wissenschaft und Kunst supported this project through grants to the light microscopy and electron microscopy facilities at CMCB, TU Dresden (European Fund for Regional Development - EFRE).