The number of sheaths supported by a single oligodendrocyte in the zebrafish spinal cord at 4 days post fertilization (dpf) is widely variable (Almeida et al., 2011). However, it is unclear how this variability arises during development. Oligodendrocytes in the spinal cord myelinate either dorsal or ventral axon tracts and occupy these regions with differing cell densities. There are ~2-fold more myelinating oligodendrocytes in the ventral tracts compared to the dorsal tracts (Figure 1A and B). These domains have different functions, with ascending dorsal axons primarily transmitting sensory information and descending ventral axons transmitting motor information (Thau and Reddy, 2022). We investigated whether some of the variability in oligodendrocyte sheath number comes from quantifying cells from both regions together. Zebrafish embryos were injected at the single-cell stage with a sox10:eGFP-CAAX plasmid to sparsely label the membrane of individual oligodendrocytes. At 4 dpf, sheath number ranged from 4 to 21 per cell (Figure 1C and D), which is in line with previous work (Almeida et al., 2011). Importantly, oligodendrocytes in the dorsal region of the spinal cord had more sheaths on average than oligodendrocytes in the ventral region (Figure 1E). No statistically significant differences in average sheath length or the total sheath length were found when comparing these groups (Figure 1F and G). Oligodendrocytes in the ventral spinal cord that myelinated Mauthner axons were excluded from all analyses in this manuscript. To investigate how the regional difference in sheath number is generated, we compared sheath initiation and loss by oligodendrocytes in the dorsal and ventral regions of the spinal cord.

To establish parameters for quantifying the dynamics of sheath initiation and loss, we sparsely labeled neurons by injecting the pan-neuronal neuroD:tagRFP-CAAX axon reporter into Tg(nkx2.2:eGFP-CAAX), a transgenic myelin reporter line. The earliest stages of eGFP-CAAX-labeled sheath formation on tagRFP-CAAX-labeled axons were captured with extensive time-lapse imaging from 2.5 to 3 dpf (Figure 2A, see Materials and methods for more details). We could consistently identify immature ensheathments that were ~2 µm or longer, surrounding axons with a cylindrical shape (Figure 2B and C). Unexpectedly, each oligodendrocyte repetitively ensheathed the same axonal domain an average of 2–3 times before any stable sheaths were formed (Figure 2D, E [T0’, T15’, T45’, T60’]). We define repetitive ensheathment as multiple rounds of sheath initiation/loss on the same domain of an axon. Each axon domain was defined as the region between the two most lateral ensheathment attempts made by the same oligodendrocyte during a series of repetitive ensheathments. In cases with only a single ensheathment attempt, the axon domain was considered the region directly underneath the sheath. Of the 25 axonal domains analyzed, 14 of them ended up with a stabilized sheath (Figure 2F) and 11 were lost (Figure 2G). The total number of ensheathment attempts on each axonal domain ranged from 1 to 6, and 72% of these domains were ensheathed more than once. This repetitive ensheathment phenomenon was therefore common in our experiment and possibly fundamental to the process of building myelin sheaths.

It is important to note that we performed this time-lapse imaging paradigm using tricaine as anesthesia. This drug is a sodium channel blocker that silences action potentials throughout the CNS. Neuronal activity is important for myelination, and it is possible that the repetitive axonal ensheathment that we observed could result from silencing neuronal activity. We tested this using the same axonal ensheathment imaging paradigm (Figure 3A). We anesthetized each larva in either tricaine or pancuronium bromide, which is a neuromuscular blocking drug that does not dampen neuronal activity in the CNS (https://www.ncbi.nlm.nih.gov/books/NBK538346/). It has also been used previously in zebrafish activity-dependent myelination studies (Hughes and Appel, 2020; Koudelka et al., 2016; Hughes and Appel, 2019). Importantly, there was essentially no difference in the average number of times an oligodendrocyte ensheathed the same axonal domain when comparing these forms of anesthesia (Figure 3B–D). Since we sometimes observed more than one axonal domain on the same axon, we also normalized the data by averaging the total number of ensheathment attempts across all domains of each axon (Figure 3E). However, there was still no difference between the two forms of anesthesia. We also found that 43% of the domains were stabilized (16 of 37) at the end of the imaging paradigm for the tricaine group and 55% were stabilized (16 of 29) for the pancuronium bromide group (Figure 3F). A Fisher’s exact test found no significant difference between these values.

As a follow-up to this work, we performed a quality control experiment to demonstrate that there was a measurable difference in the CNS activity of larvae anesthetized in tricaine vs pancuronium bromide in our experiment. To do this we measured calcium transients in the zebrafish spinal cord using the Tg(elav3:H2B-GCaMP6f) pan-neuronal transgenic line (Dunn et al., 2016). We anesthetized larvae at 2.5 dpf in pancuronium bromide and imaged the spinal cord above the yolk sac extension at 4 Hz (Figure 3—figure supplement 1A). Each larva was then spiked with embryo media containing either tricaine or pancuronium bromide (control) for 5 min. After this, another video was collected on the same cells. There was no significant difference in the number of calcium transients in the pre- and post-videos for the controls, but there was a clear reduction in transients after the tricaine spike (Figure 3—figure supplement 1B-D). Due to the variability in the total number of transients that we observed across all the larvae, we also normalized the data by dividing the number of transients in the post-video by the number of transients in the pre-video to generate a percentage. There was still a clear difference in the percentage of calcium transients in the tricaine spiked group relative to the controls (Figure 3—figure supplement 1E). Collectively, this experiment allows us to more strongly conclude that the repetitive ensheathment phenomenon that we observed in Figures 2 and 3 is independent of neuronal activity.

Further assessment of the two axonal ensheathment experiments from Figures 2 and 3 indicates that for the tricaine condition, 72.6% of all axonal domains were ensheathed more than once (Figure 3—figure supplement 2A). Similarly, 65.5% of domains were ensheathed more than once for the pancuronium bromide group. In 100% of these instances, each axonal domain was repetitively ensheathed by the same oligodendrocyte (Figure 3—figure supplement 2B). Additionally, the same cell process performs each ensheathment and seems to maintain continuous contact with the axon ~63–68% of the time (Figure 3—figure supplement 2C and D, see Materials and methods for more details). This suggests that maintaining contact with the axon could be important for sheath formation.

We next tried to better understand how the stabilization of sheaths might result from this repetitive ensheathment process. Due to the similarity in the outcomes for the tricaine and pancuronium bromide conditions in the previous paragraph, we combined all our data together to look at other metrics. We first looked at whether the average number of repetitive ensheathments for an axonal domain is different based on whether a sheath becomes stabilized. However, we found that there was no significant difference for this parameter (Figure 3—figure supplement 2E). We then divided all stabilized domains by the number of total ensheathment attempts and this results in an a relatively low stabilization rate of 20.5% (Figure 3—figure supplement 2F). This means that only one in five ensheathment attempts is stabilized on average by the end of the imaging paradigm at 3 dpf. However, for the instances when the process maintained contact with the axon, 65.1% of these resulted in a stabilized sheath (Figure 3—figure supplement 2G). This suggests that continuous axonal contact might promote sheath stabilization. Collectively, this data supports a model where oligodendrocytes extensively explore the axonal environment to find targets that are primed for sheath formation.

If the repetitive ensheathment of axons is independent of neuronal activity, it is possible that the mechanism is more biophysical in nature. One possibility is that oligodendrocytes measure the size of each axon by wrapping and unwrapping it multiple times. To explore this possibility, we measured the diameters of all the axonal domains in our two repetitive ensheathment experiments from Figures 2 and 3 above. We performed a full-width 1/3 max measurement of the diameter of each axonal domain underneath one of the ensheathments in each series. We then did a linear regression to assess the relationship between the diameter of the axon and the number of times that it was ensheathed. The R² value was 0.01 regardless of whether the ensheathment was stabilized or not. Thus, we found no relationship between these parameters in our data (Figure 3—figure supplement 3A–C).

In the previous axonal ensheathment experiments, we captured the dynamics of sheath initiation from the perspective of the individual processes of an oligodendrocyte. We next sought to understand how these dynamics contribute to the overall sheath number of entire oligodendrocytes. To do this, we modified the ensheathment dynamics imaging paradigm from Figures 2 and 3 (Figure 4A). Embryos were co-injected with our sox10:eGFP-CAAX oligodendrocyte lineage cell membrane reporter and myrf:tagRFP, a cytosolic reporter expressed in myelin-fated oligodendrocytes. This allowed us to identify sparse, double-labeled eGFP⁺RFP⁺ cells in both the dorsal and ventral tracts of the spinal cord at 2.5 dpf. We captured the cellular dynamics of each cell for 15 hr with a 5 min imaging interval. At 3 dpf, larvae were placed back into embryo medium. At 4 dpf, the same larvae were remounted, and a single static image was taken for each of the same cells. From this, the formation and loss of every ensheathment could be tracked for each cell. Additionally, quality control experiments determined that the conditions of this imaging paradigm did not change the final sheath number or average sheath length of these cells (Figure 4—figure supplement 1, see Materials and methods for further details).

To begin answering the question of how sheath initiation and loss is balanced for individual oligodendrocytes, every ensheathment was manually tracked and quantified in each frame of our time-lapse data set. As in published studies (Czopka et al., 2013), we found that oligodendrocytes have an initial phase of immature sheath accumulation (Figure 4B), followed by a phase of sheath stabilization and loss (Figure 4C and Figure 4—figure supplement 2). However, the accumulation phase was very dynamic in our studies, with a significant number of ensheathments eventually lost. For example, only 5 out of 19 ensheathment attempts were stabilized for the outlined region of the cell in Figure 4B (# 10, 13, 14, 18, and 19, Figure 4D). This is consistent with our data in Figure 3—figure supplement 2F. We also readily observed instances of possible repetitive ensheathment (Figure 4E). However, since we do not have axons labeled in this experiment, we cannot definitively conclude that this is what we observed.

To better understand the impact of these ensheathment dynamics, we needed to average the levels of sheath initiation and loss across all cells in our data set and establish a graphic method for presenting the results. However, the number of hours over which each cell accumulates sheaths ranged from ~4 to 8 hr (Figure 5A), making it difficult to combine data from multiple cells for quantitative comparison. Nevertheless, we identified a definable transition point for each cell from when they were accumulating new immature sheaths to when they were stabilizing them. Thus, once an oligodendrocyte accumulated a peak (or maximum) number of immature sheaths, every cell slowed down and stopped forming new sheaths. Given this common dynamic of accumulation and plateau, it was therefore possible to normalize the timing of sheath accumulation in each video by defining the frame at which each cell had accumulated its peak number of sheaths as time zero for that cell (Figure 5B). Using this normalization for 19 individual dorsal cells and 18 individual ventral cells, we found that the dorsal cells accumulated a higher peak number of ensheathments compared to the ventral cells (Figure 5B). Additionally, ~95% of all ensheathments attempts occurred before reaching this peak for both groups (Figure 5C and Figure 5D, vertical dashed line). This was an average of ~100 ensheathment attempts for dorsal cells and ~74 for ventral cells. Interestingly, ~75% of these ensheathments were also lost before reaching peak for both groups (Figure 5E and F). Collectively, this data shows that oligodendrocytes initiate a remarkable number of immature ensheathments, with only a small percentage maintained by the end of the accumulation phase.

It is very striking that all oligodendrocytes exhibited a similar rate of sheath loss during the accumulation phase (Figure 5), yet dorsal cells end up with more sheaths than ventral cells. Dorsal oligodendrocytes exhibited more ensheathment attempts, accumulated a higher peak number of immature ensheathments, and maintained more sheaths at the 4 dpf time point compared to ventral cells (Figure 6A–E). Dorsal oligodendrocytes also lost more sheaths during the stabilization phase compared to ventral cells (Figure 6F). However, both populations stabilized a similar percentage of ensheathments throughout the accumulation, stabilization, and combined phases of our experimental paradigm (Figure 6G–I). Thus, the levels of sheath initiation for each cell are very heterogeneous (ranging from ~20 to 190), but all oligodendrocytes exhibit a similar overall rate of sheath stabilization (~10–20%). To be confident with our quantification of this stabilization rate, we needed to be sure that these cells did not produce any new sheaths from 3 to 4 dpf during the sheath stabilization phase. We designed a modified ensheathment dynamics imaging paradigm to test this and found that oligodendrocytes made no new sheaths during this time interval (Figure 6—figure supplement 1, see Materials and methods for further details). Collectively, we conclude that dorsal cells end up with more sheaths at 4 dpf compared to ventral cells due to increased sheath initiation.

We predict that this consistent sheath stabilization rate, irrespective of sheath initiation, could be a fundamentally important parameter for understanding oligodendrocyte ensheathment behavior. Indeed, a simple linear regression comparing the total number of ensheathment attempts with the final number of sheaths for each cell established high correlation of these parameters (R² value of ~0.73 for dorsal cells and ~0.83 for ventral cells, Figure 6J). Thus, for each stabilized ensheathment, there is a relatively predictable number of ensheathments that are lost. Altogether, these results strongly suggest that while both sheath initiation and loss vary among oligodendrocytes, they are proportionately regulated across all oligodendrocytes.

The repetitive ensheathment of axons is orchestrated by a dynamic series of morphological transitions that are energetically expensive for oligodendrocytes. Thus, it seems plausible that this process is fundamental for the proper accumulation and stabilization of sheaths. The endocytic recycling pathway is responsible for the reuse or turnover of membrane and membrane proteins in a cell (depicted in Figure 7A). We therefore decided to determine the importance of these early sheath initiation dynamics by disrupting the endocytic recycling pathway.

The endocytic recycling pathway is controlled by several Rab-GTPase proteins that act as master regulators of membrane trafficking (reviewed in Pfeffer, 2017). Rab5, Rab7, and Rab11 regulate the early endosome, late endosome/lysosome, and recycling endosome, respectively (Figure 7A). The cargo of endocytic vesicles is sorted at the early endosome (Rab5) and is then trafficked back to the plasma membrane or to the recycling endosome (Rab11) for further sorting. Alternatively, molecular cargo will be degraded as early endosomes mature into late endosomes/lysosomes (Rab7) (Grant and Donaldson, 2009; Langemeyer et al., 2018).

We initially analyzed whether the Rab5, Rab7, or Rab11 proteins localized to immature myelin sheaths. Expression constructs were generated to specifically label Rabs in oligodendrocytes, using myrf regulatory DNA to drive eGFP-Rab fusion proteins: myrf:eGFP-RAB5C, myrf:eGFP-RAB7A, and myrf:eGFP-RAB11A. We transiently expressed each Rab fusion protein with sox10:mScarlet-CAAX to label oligodendrocyte membranes. Images of early myelinating oligodendrocytes expressing these Rab proteins were collected at 2.5 dpf (Figure 7B and C), and the number of Rab+ puncta present in each immature ensheathment was counted and normalized by the length of each sheath. All three types of endosomes were present within immature ensheathments (Figure 7D). This data suggests that Rab5, Rab7, and Rab11 could all potentially play a role in sheath formation.

To investigate whether Rab5, Rab7, or Rab11 are important for regulating sheath number in oligodendrocytes, we performed a sheath analysis in cells expressing dominant-negative Rab mutants (Clark et al., 2011; Zhang et al., 2007). Rab proteins are GTPases that are active when bound to GTP and inactive when bound to GDP (Müller and Goody, 2018). Thus, the rab5C^S36N, rab7A^T22N, and rab11A^S25N dominant-negative point mutations maintain the Rab protein in a GDP-bound ‘off’ state. However, since these mutant proteins still bind GEF proteins that normally activate Rabs, they reduce the level of GEFs available to activate endogenous Rabs, thereby decreasing the overall activity of the endogenous Rab protein.

We injected myrf:tagRFP-rab5C^S36N, myrf:tagRFP-rab7A^T22N, and myrf:tagRFP-rab11A^S25N fusion constructs into embryos alongside sox10:eGFP-CAAX to label oligodendrocyte membranes. We captured static images of ventral cells only at 4 dpf and quantified myelin sheath number and sheath length. We found that the Rab5DN and Rab11DN mutants both decreased the average sheath number per cell without changing average sheath length (Figure 8A–D). Thus, expression of these mutants apparently had no general negative impact on oligodendrocyte development, but they likely regulate a specific aspect of the ensheathment process. On the other hand, the Rab7DN mutant had little impact on sheath number or sheath length (Figure 8A–D), which suggests that the phenotype we observe for the Rab5DN and Rab11DN mutants is specific to the recycling pathway and is not a generic effect of expressing a dominant-negative Rab mutant.

We hypothesize that the repetitive ensheathment of axons is required to facilitate optimal sheath accumulation and stabilization and is regulated by the endocytic recycling pathway. We therefore predict that disrupting endocytic recycling will reduce these dynamics and diminish overall sheath accumulation. Both Rab5 and Rab11 are involved in the recycling component of the endocytic pathway and their dominant-negative mutants both decreased the overall sheath number per cell. Therefore, we decided to specifically investigate Rab5 in depth. We analyzed cells expressing the Rab5DN mutant in the ventral spinal cord tracts using the same ‘oligodendrocyte ensheathment dynamics paradigm’ from Figures 4—6 above. Oligodendrocytes expressing either myrf:tagRFP-RAB5C or myrf:tagRFP-rab5C^S36N were imaged and compared to the ventral control data from Figures 4—6.

Unexpectedly, the over-expression of the Rab5DN mutant did not change the dynamics of the accumulation phase. These cells formed and lost the same number of ensheathments during this phase and accumulated ensheathments to a similar peak number compared to the controls (Figure 9A–C, Figure 9—figure supplement 1A–F). Instead, the Rab5DN mutant specifically impacted the stabilization phase (which starts after cells reach peak sheath accumulation). We found a strong trend that the mutant increased the net number of sheaths that were lost from the peak sheath number out to 4 dpf although it was not statistically significant (Figure 9C–F). Consistent with this trend, cells over-expressing the Rab5DN mutant stabilized a lower percentage of sheaths in the stabilization phase specifically (Figure 9G–I). A simple linear regression comparing the total ensheathment attempts to the final sheath number showed that the slope of the mutant regression line was significantly lower than for the control group (Figure 9J). Collectively, we conclude that the dominant-negative Rab5 mutant expressed from the myrf driver negatively impacts the longitudinal sheath stabilization rate of oligodendrocytes but has little or no impact on initial sheath accumulation.

The Institutional Animal Care and Use Committee at the University of Colorado School of Medicine approved all animal work (#00419). This group follows the U.S. National Research Council’s Guide for the Care and Use of Laboratory Animals and the U.S. Public Health Service’s Policy on Humane Care and Use of Laboratory Animals. Zebrafish larvae were raised at 28.5°C in embryo medium and were staged as dpf. Transgenic lines used in this study were Tg(nkx2.2a:EGFP-CAAX), Tg(mbp:eGFP-CAAX), and Tg(mbp:tagRFP) (Kirby et al., 2006; Kucenas et al., 2008; Preston et al., 2019; Brown et al., 2021). The Tg(elav3:H2B-GCaMP6f) line was a kind gift from Dr. David Schoppik (Dunn et al., 2016). This line expresses GCaMP6f fused to the histone H2B and is thus localized to the nucleus. All other constructs were introduced by transient transgenesis to achieve sparse labeling for single-cell analysis in wild-type ABs. All experiments and analyses were performed blind to gender since sex is determined at later stages than those studied here.

Multi-site gateway cloning was used to produce the following plasmids: pEXPR-sox10:eGFP-CAAX, pEXPR-sox10:mScarlet-CAAX (gift from Bruce Appel’s lab), pEXPR-neuroD:tagRFP-CAAX, and pEXPR-mbp:eGFP. The first three plasmids contain a cysteine-aliphatic amino acid-X (CAAX) prenylation motif that targets the fluorescent protein to the cell membrane. The following entry plasmids were used to build these constructs: p5E-sox10 (Mathews and Appel, 2016), p5E-neuroD (gift from Bruce Appel’s lab), p5E-mbp (Brown et al., 2021), pME-eGFP-CAAX, pME-tagRFP-CAAX, p3E-eGFP, p3E-polyA, pDEST-Tol2 (no transgenesis marker).

The pEXPR-mbp:eGFP plasmid was linearized using SalI and SnaBI to remove both the mbp driver and eGFP insert. An myrf driver element was PCR amplified from p5E-myrf (gift from Bruce Appel’s lab) and was ligated into the linearized vector with an eGFP insert flanked with BamHI and AgeI sites using NEBuilder HiFi assembly cloning. The resulting pEXPR-myrf:eGFP plasmid was confirmed by diagnostic digest and sequencing and was used as a base vector for building the remaining over-expression constructs.

Plasmids encoding the RAB5C, RAB7A, and RAB11A zebrafish coding sequences (Addgene, RAB5C=80518, RAB7A=80522, and RAB11A=80529) were sequenced, and the RAB7A and RAB11A sequences each had a point mutation (D63G and Q166R, respectively) relative to the NCBI protein sequences (RAB7A=NM_200928.1, RAB11A=NM_001007359.1) and relative to other previous work in zebrafish (Clark et al., 2011). These mutations were corrected using a combination of mutagenesis by PCR-driven overlap extension (Heckman and Pease, 2007) and NEBuilder HiFi assembly cloning.

The RAB5C, RAB7A, and RAB11A sequences were then cloned directly into the pEXPR-myrf:eGFP plasmid as follows. This plasmid was linearized using BamHI and AgeI to remove the eGFP insert. Each RAB sequence was PCR amplified using the following primer sequences (additional overhang sequences were added to each primer for NEBuilder HiFi assembly-based cloning that are not shown here):

The RAB PCR fragments were then ligated into the linearized plasmid, along with either tagRFP or eGFP PCR fragments, using NEBuilder HiFi to generate the following fusion constructs (a kozak sequence, GCCACC, was added directly 5’ of the translational start site for all constructs):

We also made a set of dominant-negative point mutations for each of the RAB genes (rab5C^S36N, rab7A^T22N, rab11A^S25N) that have been characterized in zebrafish previously (Clark et al., 2011). We generated the rab5C^S36N mutant through a combination of mutagenesis by PCR-driven overlap extension (Heckman and Pease, 2007) and NEBuilder HiFi assembly cloning. Primer sequences:

Both final PCR fragments were then ligated into the same linearized pEXPR:myrf plasmid described above, along with a tagRFP PCR fragment, using NEBuilder HiFi to generate the following fusion plasmid (a kozak sequence, GCCACC, is directly 5’ of the translational start site):

The rab7A^T22N and rab11A^S25N mutations were made by Keyclone Technologies (San Marcos, CA, USA). Resulting plasmids:

All plasmids were confirmed by diagnostic restriction digest and sequencing.

Plasmids were injected into zebrafish embryos at the single-cell stage with Tol2 mRNA to achieve transient transgenesis and sparse labeling. On the desired day, the injected larvae were mounted laterally in 0.8% low-melt agarose with 140 µg/mL (0.014%) of tricaine methanesulfonate (Syndel) as anesthesia. All imaging was performed live using a Nikon A1R resonance scanning confocal microscope and a 40× Apochromat long-working distance water immersion objective with a 1.15 NA (Nikon MRD77410). A temperature-controlled stage maintained at 28.5°C was used for all time-lapse experiments. All imaging was done in the spinal cord of each larva above the yolk-sac extension. Oligodendrocytes in the ventral spinal cord that myelinated Mauthner axons were excluded from all analyses in this work.

In the axonal ensheathment activity experiment in Figure 3, we compared tricaine and pancuronium bromide (Sigma, P1918) forms of anesthesia. At 2.5 dpf, larvae were put down with tricaine and a tiny slit was made at the tip of each tail using a razor blade. Some of these larvae were then mounted in agarose containing the same concentration of tricaine used in all other experiments (no activity group) and the rest of the larvae were transferred into embryo media containing 0.45 mg/mL pancuronium bromide and without tricaine for 5 min. These larvae were then mounted in agarose containing the same concentration of pancuronium bromide (activity group). All other general imaging parameters were kept the same for this experiment.

For the calcium activity experiment, Tg(elav3:H2B-GCaMP6f) larvae were all put down in pancuronium bromide exactly as above. (See this experimental section below for additional details of how the tricaine spike was performed.)

Tg(mbp:eGFP-CAAX) and Tg(mbp:tagRFP) transgenic lines that express membrane-tethered eGFP and cytosolic tagRFP in myelinating oligodendrocytes respectively were crossed and embryos were grown to 4 dpf. Both sides of the spinal cord for each larva were imaged laterally above the yolk-sac extension with a 1× optical zoom (0.31 µm XY pixel size), a 0.3 µm z-step size (Nyquist), and 32× line averaging. Cell counts were performed in Imaris (version 9.8). Each image was cropped to separate dorsal and ventral regions. A Gaussian filter was then applied with a 0.311 µm width to smooth out noise. Following this, a local-background subtraction was performed with an estimated cell body size of 7 µm. A threshold for fluorescent intensity was then applied to each image manually. Finally, a 0.5 µm water-shed filter was used to separate contacting cell bodies and Imaris counted the number of cells in each image. Cell bodies occupying the area in between the dorsal and ventral axon tracts were manually removed (these cells were in the minority).

Individual sheaths were visualized by labeling oligodendrocytes with sox10:eGFP-CAAX by transient transgenesis. These cells were imaged at 4 dpf with a 2× optical zoom (0.16 µm XY pixel size), a 0.3 µm z-step size (Nyquist), and 32× line averaging. Cells were chosen that were not too crowded for accurate quantification. Except for Figure 1D, dorsal and ventral cells were never combined for this analysis. In each image every sheath was identified and counted by inspecting individual optical sections. The lengths of each sheath were measured from maximum intensity projections using the simple neurite tracer plugin in Fiji. From this data we calculated sheath number, average sheath length, and the total sheath length per cell (calculated by adding up the length of all sheaths supported by each cell). The same images and ventral control data from this analysis were used and presented in Figures 1 and 8, and Figure 4—figure supplement 1.

Criteria for quantifying ensheathment attempts using labeled axons were established. Axon membranes were sparsely labeled with the neuroD:tagRFP-CAAX plasmid by transient transgenesis in the Tg(nkx2.2:eGFP-CAAX) stable line. At 2.5 dpf larvae were anesthetized in tricaine and mounted, and time-lapse imaging was performed for 15–18 hr with an imaging interval of 5 min. A 2× optical zoom (0.16 µm XY pixel size), a 0.5 µm z-step size, and 16× line averaging were used. Imaging was done in the spinal cord above the yolk sac extension and focused on dorsal and midline axons, since the Tg(nkx2.2:eGFP-CAAX) line was too crowded in the ventral region for visualizing ensheathment dynamics. Regions in each video with potential axonal ensheathments were cropped and corrected for drift using the 3D drift correct plugin in Fiji. Individual optical sections were inspected and a volume projection in Imaris was analyzed for each axonal ensheathment. From this, we developed quantification criteria. Ensheathments had to be ~2 µm or longer and cylindrical in shape around the axon, that is, the axon fluorescent signal had to go through the center of the oligodendrocyte signal, and it had to be more than ~75% of the way around the axon. Once an ensheathment had formed, it had to shrink to below 2 µm in length and be less than half-way around the axon to be considered lost.

Repetitive ensheathments were multiple rounds of sheath initiation/loss on the same domain of an axon. Each axon domain was the region between the two most lateral ensheathment attempts that took place during a series of repetitive ensheathments. In cases with only a single ensheathment attempt, the axon domain was the region directly underneath the sheath. These domains ranged from ~2 to 20 µm in length. Some axons had more than one domain that was ensheathed. These domains were considered separate because they were ensheathed by different oligodendrocytes or because a single oligodendrocyte had more than one sheath on the same axon at the same time. Axonal domains were considered destabilized if each ensheathment was lost by the end of the time-lapse period and stable if they remained. We required that a minimum of 3 hr of video be acquired after the final ensheathment attempt to be quantified.

For the activity axonal ensheathment experiment in Figure 3, Tg(nkx2.2:eGFP-CAAX) larvae were anesthetized and mounted in either tricaine or pancuronium bromide as described in the ‘Injections, anesthesia, mounting, and general imaging parameters’ section above. Using multi-well glass-bottom dishes (Greiner Bio-One, 627871), we imaged larvae from each group in each experiment exactly as described earlier in this section, and the videos were all processed and quantified exactly as was described.

We present data in Figure 3—figure supplement 2C, D showing how often the same process from the same oligodendrocyte performs each repetitive ensheathment. We also report on how often these processes seem to remain in contact with the axon. Except for one instance in this data set (64 repetitively ensheathed domains), if the same process performed each repetitive ensheathment it also remained in contact with the axon. For instances where different processes performed the repetitive ensheathments, this involved complete process retraction before a different process started the next ensheathment in the series. About 15–20% of processes could not be sufficiently tracked throughout the entire video due to periods of crowding from other processes. The activity for these instances was categorized as ‘unclear’. It is important to note that our imaging interval for these experiments was 5 min, which is not fast enough to fully capture all oligodendrocyte process dynamics. Analysis of this data set therefore could have missed some of these process movements.

Tg(elav3:H2B-GCaMP6f) larvae were mounted at 2.5 dpf in pancuronium bromide as described in the ‘Injections, anesthesia, mounting, and general imaging parameters’ section above. We mounted two to three larvae in each well of a multi-well glass-bottom dish (Greiner Bio-One, 627871). This allowed us to pick a single larva per well to do a spike experiment on and then multiple spike experiments could be performed per dish. The same region of the spinal cord above the posterior portion of the yolk-sac extension was imaged. We used a 1× optical zoom (0.31 µm XY pixel size) and 8× line averaging. A 5 min, single optical-section, time-lapse video was collected at 4 Hz (4 fps) to get a baseline of calcium transients in each larva. Each well had 500 µL of embryo media with 0.45 mg/mL pancuronium bromide (same concentration as the agar) for the initial round of imaging. After this, each well was spiked with 500 µL of embryo media containing either pancuronium bromide (control group) or tricaine. For the control group the embryo media contained the same concentration of pancuronium bromide (0.45 mg/mL) and the embryo media for the tricaine spike group was at a 2× concentration (140 µg/mL final concentration as was used for all other experiments). All larvae were incubated for 5 min after the spike. Then, a final round of imaging was performed on the same cells in the same field of view for measuring changes in the overall number of transients.

All videos were processed and quantified post blinding. ROIs for the neuronal cell bodies in each field of view were generated in Fiji. The ‘despeckle’ filter was first used to clean up salt-and-pepper noise. A standard deviation time projection was made to create a visual of all the cell bodies in each video. Huang’s threshold was then applied manually to each projection to binarize the signal of each cell body. We used the ‘analyze particles’ function to automate the generation of ROIs around each cell. The ‘water-shed’ function was then used to separate cell bodies that were touching. Each set of ROIs were then applied to the raw video file and additional ROIs were added manually as needed that were missed by the segmentation work-flow. The mean intensity within each ROI was collected for every frame of the video and was transferred to excel for further analysis. Two ROIs were placed manually in each video for determining background levels. All the intensities in every frame for both background ROIs were averaged, and this single value was subtracted from every other mean intensity value that was collected. Calcium transients were determined with a moving baseline. Ten frames of video prior to each transient were averaged to set the baseline. The fluorescence intensity had to be 3 standard deviations above this baseline for a minimum of three frames to be counted. Within complex calcium transients with multiple peaks, transients were counted as being separate if the signal returned to below 3 standard deviations before going back up again. The total number of calcium transients within each video was quantified and after unblinding, the pre- and post-calcium transient numbers were compared.

The diameters of all the domains in both axonal ensheathment experiments (Figures 2 and 3) were measured by selecting the first video frame of a single ensheathment for each axonal domain; a single optical section was then isolated for measurement in Fiji. Two different line scans with a 3-pixel width were drawn manually across each axon at the ensheathment site. The axon fluorescence intensity data was transferred to excel for further analysis. Two 8×4-pixel ROIs were also drawn to measure the background intensity. After background subtraction, the full-width 1/3 maximum was determined for each line scan manually and the two values from each domain were averaged as the representative diameter of each axonal domain.

Oligodendrocytes were sparsely double-labeled with our pEXPR-sox10:eGFP-CAAX plasmid and one of the following plasmids using transient transgenesis: pEXPR-myrf:tagRFP, pEXPR-myrf:tagRFP-RAB5C, or pEXPR-myrf:tagRFP- rab5C^S36N. The myrf-driven plasmids identified cells that were fated to make myelin. At 2.5 dpf, larvae were anesthetized in tricaine and mounted, and eGFP+/RFP+ progenitor cells were chosen for imaging above the yolk-sac extension in the spinal cord. Time-lapse imaging was performed for 15 hr with an imaging interval of 5 min. A 2× optical zoom (0.16 µm XY pixel size), a 0.75 µm z-step size, and 16× line averaging were used. After the time-lapse imaging period was over at 3 dpf, the larvae were removed from the mounting agar and were put back into embryo medium in separate wells of a 24-well plate. At 4 dpf, a final static image was taken for each of the same cells to conclude the imaging paradigm. To be considered for analysis, we had to obtain at least 30 min of imaging before each cell started the ensheathment process. Additionally, each cell had to stop making sheaths for at least 90 min before the end of the imaging interval.

A max projection of each video was background-subtracted and corrected for drift using Fiji’s 3D drift correct plugin. Sheath initiation and loss were counted manually in every frame using the multi-point counter in ImageJ based on the ensheathment criteria we established in the previous section. Although the counter ROIs for this were placed within each max projection, every sheath initiation/loss event was determined by inspecting individual optical sections and by looking at a volume projection in Imaris for each frame. At the final 4 dpf time point, all sheaths were counted, and the lengths were measured using the simple neurite tracer plugin in Fiji.

A quality control experiment determined that the conditions of this imaging paradigm did not significantly change the average sheath number or sheath length across a population of cells. Static images of both dorsal and ventral oligodendrocytes were collected at 4 dpf from zebrafish larvae that had been anesthetized in tricaine, mounted in agar, and housed in the time-lapse imaging chamber for the 15 hr experiment from 2.5 to 3 dpf. However, these larvae were not imaged during this time. We compared the sheath number and lengths for these cells (agar_tricaine condition) to our sheath number counts from Figure 1C (standard condition) and to cells that went through the full ensheathment dynamics imaging paradigm (time-lapse condition) (Figure 4—figure supplement 1A–D).

A separate quality control experiment determined that oligodendrocytes did not make any new ensheathments during the stabilization phase from 3 to 4 dpf. Cells were imaged using a modified ensheathment dynamics imaging paradigm. First, from 2.5 to 3 dpf, the imaging interval was 30 min rather than 5 min. At 3 dpf, the larvae were removed from the tricaine and imaging agar and were allowed to recover in embryo media for 30 min. After this the larvae were remounted, and the same cells were imaged for an additional 23 hr with an imaging interval of 15 min. We tracked sheath initiation and loss from 3 to 4 dpf for five dorsal cells and three ventral cells. None of these eight oligodendrocytes produced any new sheaths from 3 to 4 dpf (Figure 6—figure supplement 1).

Our myrf:eGFP-RAB5C, -RAB11A, and -RAB7A fusion constructs were transiently expressed alongside sox10:mScarlet-CAAX to label the membrane of individual oligodendrocytes. At 2.5 dpf, larvae were anesthetized in tricaine, mounted, and cells that were in the early stages of the ensheathment process were chosen for imaging. Static snapshots of a mix of both dorsal and ventral cells were taken using a 3× optical zoom (0.1 µm XY pixel size, Nyquist), a 0.75 µm z-step size, and 16–32× line averaging. A larger z-step size than Nyquist (0.3 µm) was used to minimize motion artifacts during imaging. Images were pre-screened by viewing the membrane marker channel only and individual sheaths with high-quality imaging were cropped for further processing. The number of endosomal puncta in each image was quantified using a spots analysis in Imaris (version 9.2). We used an estimated XY diameter of 0.5 µm to detect Rab5 and 0.4 µm to detect Rab7 and Rab11. After a local-background subtraction, a threshold for fluorescent intensity was applied to each image manually. Imaris counted the number of puncta, and we normalized these numbers to the length of each sheath (in µm), measured using the Imaris filament tracer. Only puncta within a sheath were counted based on overlap with the membrane marker. Puncta that were present in adjacent processes or other structures were manually excluded from each image.

All statistics and plots were done using GraphPad Prism (version 9). We used the non-parametric Mann-Whitney test for all unpaired comparisons with no assumption of normality. We used the non-parametric Wilcoxon test for all paired comparisons with no assumption of normality. For groups of three or more, global significance was first assessed using the non-parametric Kruskal-Wallis test. This was followed up by Dunn’s multiple comparison tests if the global p-value was <0.05 (unless otherwise stated in each figure legend). Global p-values are reported when not significant. Non-parametric statistical tests were chosen since each data set had groups that were not normally distributed and/or had unequal variances. A Fisher’s exact test was used when testing categorical data because our sample sizes were less than 100. Simple linear regressions were performed for comparing the association of two variables with each other. Individual data points are shown for all plots with the central dashed lines representing the average of the group. We considered p < 0.05 the threshold for statistical significance and we provide exact p-values for all analyses.

The sample sizes for the cell counts in Figure 1, the repetitive axonal ensheathment analyses in Figures 2 and 3, the Rab+ endosomal quantification in Figure 7, and the modified ensheathment dynamics data in Figure 6—figure supplement 1 were not pre-determined. The sample sizes for the sheath analyses performed in Figures 1 and 8 and Figure 4—figure supplement 1 were determined based on previous practices and are meant to show the full distribution for this type of data set. The control, Rab5WT, and Rab5DN ‘sheath number per cell’ data in Figure 8 was used to do a power analysis (GPower Version 3.1) to determine the appropriate sample sizes for the oligodendrocyte ensheathment dynamics data in Figures 4—6 and Figure 9, and Figure 4—figure supplement 1 (time-lapse group) and Figure 9—figure supplement 1. We performed a power analysis using the parametric ‘ANOVA: Fixed effects, omnibus, one-way’ stats test in GPower with three groups (control, Rab5WT, Rab5DN) with an alpha error probability of 0.05 and 80% power. The effect size was calculated within GPower. From this, GPower calculated needing 15 cells per group. Since sheath number was not normally distributed, we had to assess statistical significance for these experiments using a non-parametric test. We therefore increased the sample size by 20% (n=15 increased to n=18) to account for the decreased power of non-parametric tests. The same time-lapse ventral control data set was used in Figures 5, 6 and 9, and Figure 4—figure supplement 1 because this data is very difficult to collect, and the analysis was very time intensive. Therefore, we also used the results of this power analysis to determine the same sample size of the dorsal group of cells analyzed in Figures 4—6 and Figure 4—figure supplement 1 (time-lapse group).

All representative images in each figure were from the associated quantified data set. These images were adjusted for brightness and contrast for clarity of presentation. No quantifiable data was excluded from any analysis. Results/phenotypes that were replicated are present within this manuscript. All data in each figure was pooled from a minimum of two independent rounds of fish crossing and injections. Data for each individual oligodendrocyte in this study is considered a biological replicate because only one cell was imaged from each larva unless otherwise noted in each figure legend. Each individual axon in the axonal ensheathment dynamics experiments was considered a biological replicate. Individual larvae were biological replicates for cell counting and for the calcium imaging experiment. Data points collected from the same cell (different sheath lengths from the same oligodendrocyte, multiple domains on the same axon, rab puncta counts in individual sheaths) are considered as technical replicates in this study. All analyses were performed blind using a custom Fiji script called ‘blind-files’ (source code can be found at https://github.com/Macklin-Lab/imagej-microscopy-scripts, copy archived at George, 2019).