Main

Rapid conduction of nerve impulses in vertebrates requires that most large-diameter axons be wrapped by myelin. In the CNS, myelin sheaths are formed by oligodendrocytes, one of the major classes of glial cells1,2. During development, discrete subpopulations of neural precursors give rise to OPCs. OPCs proliferate and migrate from their origins, eventually becoming uniformly distributed throughout the gray and white matter of the brain and spinal cord3. During late embryogenesis and early postnatal life, many OPCs stop dividing and differentiate as oligodendrocytes, which extend fine membrane processes that wrap axons. Each oligodendrocyte can wrap numerous axons, and each axon is ensheathed by periodically spaced processes extending from multiple oligodendrocytes. Uniform and periodic spacing of myelin sheaths on axons is a critical feature of saltatory conduction, but whether spacing is determined by OPCs or positional cues within axons is unknown.

To investigate behavior of OPCs as they migrate and make contact with axons in vivo, we created transgenic zebrafish that express fluorescent proteins in oligodendrocyte-lineage cells and performed time-lapse imaging during normal development and after laser microsurgical removal of nearby cells. Our data establish three key points. First, OPCs constantly remodel numerous filopodium-like processes as they migrate and for many hours before they wrap axons. Second, OPCs often retract processes and change direction of migration after making contact with a neighboring OPC. Third, OPCs can divide and migrate to replace nearby oligodendrocyte-lineage cells removed by laser microsurgery. These results raise the possibility that OPCs survey their environments by filopodial activity and adjust their migration and division according to the density and distribution of nearby OPCs and oligodendrocytes. Density-dependent regulation of migration and division ensures a uniform distribution of oligodendrocytes, facilitating uniform myelination.

In mice, genetically ablated subpopulations of OPCs are rapidly replaced by nearby cells, suggesting that OPCs compete for space4. In humans, OPCs seemingly divide and migrate to replace oligodendrocytes lost from disease or injury5. Our data point toward filopodium-based surveillance a mechanism that regulates myelination and, possibly, remyelination.

Results

Migrating OPCs rapidly remodel filopodium-like processes

The transgenic line Tg(nkx2.2a:megfp) expresses a membrane-tethered enhanced green fluorescent protein (mEGFP) under control of nkx2.2a regulatory sequences carried on a bacterial artificial chromosome (BAC)6. The transgene expresses EGFP in a pattern similar to nkx2.2a RNA, marking a subset of oligodendrocyte-lineage cells (data not shown). To examine spinal cord OPC behaviors, we mounted Tg(nkx2.2a:megfp) embryos on their sides and performed time-lapse imaging using spinning disc confocal microscopy from 36 h post-fertilization (hpf), when OPCs are first specified in ventral spinal cord of zebrafish, to 72 hpf, when axon wrapping is initiated (Fig. 1 and Supplementary Video 1 online). Migrating OPCs formed numerous, highly branched filopodium-like processes that typically, but not exclusively, extended in the direction of migration. The Tg(nkx2.2a:megfp) transgene marks axons that descend from the hindbrain through the dorsal longitudinal fascicle (DLF). OPCs appeared to make contact with these axons and generally stopped their dorsalward movement at the DLF. Occasionally OPCs continued dorsally but returned to the DLF. Once at the DLF, OPC process extension and retraction persisted for several hours until the cells began to wrap axons and some OPCs continued to adjust their positions with longitudinal movements.

Figure 1: OPC migration in zebrafish spinal cord.
figure 1

Frames were captured from a 36-h time-lapse sequence. Images were obtained from the side of the spinal cord of a Tg(nkx2.2a:megfp) larva, with dorsal up and anterior left. The 0-min time point is from approximately the 40-hpf stage. Times shown in upper right corners represent time elapsed after the start of the sequence. Two OPCs are marked by * and #. Axons migrating in the dorsal longitudinal fascicle (DLF) are indicated by arrows and OPC processes by arrowheads. Scale bar equals 22 μm.

Full size image

Time-lapse imaging showed that OPC process activity was highly dynamic during migration. To quantify process motility, we collected images at 2-min intervals and measured changes in process length. We found that the same process could extend, retract and form new branches within a few minutes (Fig. 2a,b). An OPC could both extend and retract different processes at the same time, but extension and retraction were not necessarily balanced (Fig. 2c). The velocity of extension and retraction were similar, averaging from about 4.2–7.4 μm for each 2-min period for four individual processes, but were highly variable, ranging from <1 μm to >13 μm (Fig. 2d).

Figure 2: OPC processes are rapidly and extensively remodeled.
figure 2

Images were captured from a time-lapse sequence of an approximately 60-hpf Tg(nkx2.2a:megfp) larva; dorsal is up and anterior is left. (a) First and last images of a 30-min time-lapse sequence. Two processes (1 and 2) for each of two cells (a and b) are indicated by arrows. (b) Six-minute image sequence for process b2. Arrow indicates retracting process, and asterisk marks formation of a new branch. (c) Graphs showing process length changes during the 30-min imaging sequence. Arrows and asterisk correspond to those in b. (d) Graphs showing velocity of process extensions and retractions. Extensions are shown as positive numbers and retractions as negative numbers. The average extension ('avg. ext.') and average retraction ('avg. ret.') are indicated for each process. Scale bar equals 15 μm.

Full size image

Time-lapse imaging also showed that although OPCs had a net dorsal movement, they frequently changed direction to move ventrally, anteriorly and posteriorly (Fig. 3a and Supplementary Video 1). OPCs that migrated anteriorly or posteriorly were usually close to DLF axons. OPCs did not migrate at consistent rates but instead moved intermittently. Among the 18 cells we traced, none followed a straight path, and no two paths were the same. The total distance traveled by individual OPCs was variable, ranging from 43–253 μm (Fig. 3b). OPCs occasionally retracted all their processes and divided (Fig. 3c and Supplementary Video 1). These cells were usually close to the DLF and always divided in the longitudinal plane. In every case, sister OPCs subsequently migrated away from each other.

Figure 3: OPC migration is variable in direction and distance.
figure 3

(a) Tracings of migratory paths of 18 OPCs. Each trace is oriented so that ventral is to the bottom and anterior to the left. Arrowheads indicate starting point of each cell. (b) Graph showing total distances traveled by individual OPCs. Distances are split into 25-μm bins, and the number of OPCs in each category is plotted on the y-axis. (c) Images captured from a time-lapse sequence showing an OPC (asterisk) retracting its process, rounding up and dividing in the anteroposterior axis to produce two daughter OPCs. Scale bar for c equals 15 μm.

Full size image

We noticed that process retraction and alteration of migratory paths often occurred after apparent contacts between neighboring OPCs (Fig. 4a and Supplementary Video 1). To examine interactions between neighboring cells more carefully, we marked OPCs with different-color fluorescent proteins by injecting Tg(nkx2.2:megfp) embryos with the plasmid p7.2sox10:mrfp, which drives expression of membrane-tethered monomeric red fluorescent protein (RFP) under control of sox10 regulatory DNA7. Injected plasmids were distributed mosaically, resulting in a mixture of mEGFP+ mRFP+, mEGFP+ mRFP− and mEGFP− mRFP+ OPCs. This demonstrated extensive overlap and interdigitation at the process terminals of neighboring OPCs (Fig. 4b and Supplementary Video 2 online). Overlapping terminals were constantly remodeled and gradually withdrew from each other to occupy largely nonoverlapping territories. Thus, process activity and migration seem to be influenced by signals, perhaps mediated by contact, that pass between OPCs.

Figure 4: OPC processes retract after contact.
figure 4

(a) Images captured from time-lapse sequence showing two newly divided sister OPCs (asterisks). At 0 min in the sequence, one OPC extended a secondary process (arrow) soon after making contact with a process from an OPC outside the field of view (arrowhead). At 7 min, the secondary process had extended. By 14 min, the primary process had retracted, and the OPC began to migrate behind the secondary process. (b) Images captured from time-lapse sequence showing interactions of three OPCs. At 0 min in the sequence, a mEGFP+ mRFP− OPC (green asterisk) extended a process (green arrow) that interdigitated with a process from a mEGFP+ mRFP+ OPC (yellow asterisk and yellow arrow). These processes withdrew from each other over time, while a mEGFP− mRFP+ OPC extended a process (red asterisk and red arrow) and migrated dorsally to fill the vacated space. Scale bar equals 15 μm.

Full size image

OPCs divide and migrate to replace nearby ablated cells

Oligodendrocytes might be regularly spaced because OPC process activity functions as a surveillance mechanism to recognize the presence or absence of nearby oligodendrocyte-lineage cells. Accordingly, OPCs should divide and migrate to fill territories unoccupied by other OPCs, producing a uniform distribution. We tested this using a laser to ablate oligodendrocyte-lineage cells expressing a cytosolic EGFP driven by olig2 regulatory DNA in Tg(olig2:egfp) larvae8. We ablated cells in 4-d-post-fertilization (dpf) larvae, 1 d after initiation of axon wrapping, because little OPC division and migration occurs normally at that time (see below). A small amount of cell debris was evident immediately after ablation (Fig. 5a,b) showing that the targeted cells were effectively killed. We confirmed that oligodendrocyte-lineage cells were eliminated by ablating cells on one side of the spinal cord and labeling transverse sections with Sox10 antibody (Fig. 5c). In one series of experiments, ablated sides had 12 EGFP+ Sox10+ cells, compared with 54 on control sides (n = 3 larvae). By contrast, EGFP+ cells did not die when we ablated neural cells next to them, suggesting that the laser did not create extensive damage outside the targeted cell (data not shown). Next, we ablated all oligodendrocyte-lineage cells within a five-hemisegment region of the dorsal trunk spinal cord. One day later, the number of EGFP+ cells within the ablated region was about 50% of neighboring control hemisegments (Fig. 5d). The number of OPCs in the ablated region continued to increase, reaching about 70% of the number of neighboring control hemisegments by 4 d after ablation. When we ablated a single neural cell next to each OPC and oligodendrocyte within a four-hemisegment region at 4 dpf (n = 13 larvae), EGFP+ cell number did not differ from control hemisegments during the following 2 d, indicating that OPCs did not proliferate in response to general cell damage. We then reasoned that ablated oligodendrocyte-lineage cells could be replaced by OPCs from nearby unablated regions of spinal cord, by ventral neural precursors or by dorsal neural precursors that were recently shown to produce a small number of OPCs in mouse9,10,11,12. To distinguish between these possibilities, we performed time-lapse imaging after laser ablation. Within 3 h, EGFP+ OPCs began to migrate into the ablated region from nearby anterior, posterior and ventral regions (Fig. 5e and Supplementary Video 3 online). At the same time, a larger percentage of OPCs divided compared with control larvae. During time-lapse sequences of control larvae, 16% of dorsal EGFP+ oligodendrocyte-lineage cells migrated and divided (two larvae, imaged for 16 and 21.5 h, 18.75 h average), whereas in ablated larvae, 44% of the cells migrated and divided (three larvae, imaged at 11.5, 15 and 20 h; 15.5 h average). In time-lapse imaged experimental larvae, all EGFP+ cells that occupied the ablated region originated as nearby OPCs and did not arise de novo from ventral or dorsal precursors (Fig. 5e).

Figure 5: OPCs divide and migrate to replace ablated cells.
figure 5

(a,b) Images of spinal cord of a Tg(olig2:egfp) larva before and after laser ablation. Arrow and arrowhead mark ablated and nonablated OPCs, respectively. Dorsal is up. (c) Transverse section of Tg(olig2:egfp) larva in which OPCs were ablated in the dorsal left spinal cord. Arrowheads mark EGFP+ Sox10+ OPCs in ventral and dorsal right spinal cord. (d) Quantification of dorsal oligodendrocyte-lineage cells after ablation at 4 dpf (0 d after ablation) and each day until 4 d after ablation. White and black bars show average number of dorsal EGFP+ cells in control hemisegments 4–5 and 11–12, respectively, which border ablated hemisegments 6–10, represented by gray bars. Error bars indicate s.e.m. For ablated versus control hemisegment cell counts, P < 0.05 for 0–2 d after ablation and P > 0.05 for 3 and 4 d after ablation. (e) Time-lapse images after ablation. Dorsal is up. Two OPCs on either side of the ablated region are marked by asterisks. Each OPC migrated into the ablated region and divided a single time. Sister OPCs are marked by asterisks of the same color. Scale bar equals 48 μm. (f) Time-lapse images after ablation of dorsal OPCs in a 3-dpf Tg(nkx2.2a:megfp) larva. The positions of the ablated cells are outlined in the first frame. Dorsal and ventral OPCs that extended processes and migrated into the ablation site are marked by asterisks and arrows. Scale bar equals 24 μm.

Full size image

To examine more closely the behavior of OPCs after ablation of nearby cells, we performed experiments using Tg(nkx2.2a:megfp) larvae. Because it is difficult to see cell bodies at 4 dpf using this transgene, we ablated cells between 50–60 hpf, before axon wrapping normally begins. Fewer than 2 h after ablation, nearby OPCs in both dorsal and ventral spinal cord began to remodel their processes so that the majority extended toward the ablation site (Fig. 5f and Supplementary Video 4 online). Within 5 h, the site was filled with OPCs that migrated longitudinally from dorsal spinal cord and dorsally from ventral spinal cord. Within 12 h, newly migrated OPCs began to wrap axons. Together with the above data, these observations show that OPCs can rapidly adjust their filopodium-like processes, migratory activities and rate of division in response to loss of nearby oligodendrocyte-lineage cells.

Discussion

Uniform myelination requires uniform distribution of oligodendrocytes in sufficient number to wrap the entire length of all target axons. Because oligodendrocyte-lineage cells arise as proliferative progenitors from spatially restricted subsets of neural precursors, key influences of oligodendrocyte distribution include OPC migration and division13. Secreted growth factors regulate both OPC motility and division rate, but the mechanisms by which oligodendrocytes are uniformly distributed are unclear14. As oligodendrocyte-lineage cells develop, they produce numerous fine membrane processes that eventually wrap axons1,2. Our data provide evidence that OPC process activity influences oligodendrocyte spacing.

One of the most notable features of the development of oligodendrocytes is their formation of extensive membrane processes. Distinct process morphologies are evident in fixed tissues and cell culture, and these seem to represent progressive oligodendrocyte maturation from premyelinating to myelinating states1,2. Process remodeling has been inferred previously from examination of antibody-labeled oligodendrocyte-lineage cells in fixed thick sections of mouse brains15 and observed directly by time-lapse imaging of cultured oligodendrocytes16. However, these activities are associated with apparently post-mitotic, nonmigratory cells and are assumed to reflect a mechanism for axon identification and contact. By contrast to the elaborate morphologies of more mature cells, migrating OPCs in cell culture or cultured tissue slices usually appear to have unipolar or bipolar morphologies17,18,19,20. Consequently, we were surprised by the number and dynamic activity of OPC processes during migration in zebrafish. Our imaging shows that zebrafish OPCs can be unipolar or bipolar but that these are very transient states, and the cells frequently form more numerous processes that project in different directions. Process extension, retraction and higher-order branching are interrupted only when OPCs divide and otherwise occur continuously until the cells settle into their final positions and begin to wrap axons. Thus, OPC process activity may be more extensive than previously appreciated. Although morphological differences in OPCs could reflect species differences, we think an alternative explanation is that various labeling methods produce different impressions of OPC shape and behavior. In fact, from examination of the Tg(olig2:egfp) transgenic line, which expresses a cytosolic form of EGFP, we had previously concluded that migrating OPCs are bipolar in zebrafish8. By examining in this study a new line that expresses a membrane-localized EGFP, we have been able to see processes that were not visible with cytosolic EGFP.

Dynamic process activity could serve as a mechanism for OPCs to survey their environments for extracellular cues. Consistent with this idea, structures similar to axonal growth cones, which are associated with signal reception, often form at the process tips of cultured oligodendrocyte-lineage cells18,19,21, and they are evident in our in vivo time-lapse sequences. Numerous secreted and contact-mediated signals are implicated in regulation of OPC migration and proliferation3,22. For example, Netrin-1, a secreted molecule that can either attract or repel axons23, can repel OPCs in vitro24,25,26. In mouse embryos, ventral midline cells close to the origin of OPCs express Netrin-1, whereas OPCs express the Netrin receptors DCC and Unc5h1 (ref. 25), and OPCs do not disperse from the ventral spinal cord of mutant mice lacking functional Netrin-1 or DCC20,25. Secreted platelet-derived growth factor-A (PDGF-A) also promotes OPC migration, as well as proliferation in vitro19,27,28,29,30,31. Transgenic mice that overexpress PDGF-A produce excess OPCs32, and PDGF-A–deficient mice form greatly reduced numbers of OPCs and oligodendrocytes32,33, establishing a critical role for this growth factor. Oligodendrocyte-lineage cells express distinct integrins (cell surface receptors for extracellular matrix ligands) as they mature34,35, and these have been functionally linked with migration, proliferation and differentiation17,36. Integrins can influence growth factor signaling and consequently might help regulate transition of OPC migratory, proliferative and differentiative behaviors in response to broadly distributed signals14. However, mechanisms that ensure uniform spacing of oligodendrocytes, leading to uniform and complete myelination, remain unknown.

Observations from our time-lapse imaging raise the possibility that the spacing of oligodendrocytes is influenced by contact-mediated signaling between OPCs as they migrate and divide to populate the neural tube. Dorsally migrating OPC paths are not straight, as would be predicted if they were simply moving down a Netrin chemorepellent gradient, but instead are highly variable, with OPCs frequently pausing and changing direction to move anteriorly, posteriorly and even ventrally. OPCs migrating in cell culture and tissue slices show similar stop-and-go movements18,20. We found that OPC processes often seem to withdraw after contact with a nearby process, suggesting that they repel one another, and that contact frequently precedes a change in the trajectory of OPC migration. Consequently, during development, OPCs might divide and migrate to fill space not occupied by other OPCs, eventually achieving a uniform distribution.

When subsets of OPCs are genetically ablated in mice, they are rapidly replaced, suggesting that OPCs compete for space4. Similarly, experimental demyelination in rodents is followed by elevated rates of OPC division37,38, and demyelinated lesions of individuals with multiple sclerosis appear to be populated by immature oligodendrocyte-lineage cells39,40,41,42,43. Because OPC number seems to be finely tuned by the level of growth factor32,44, loss of OPCs might increase available PDGF, leading to an elevated proliferative response by remaining OPCs44. Additional regulatory mechanisms might involve contact between OPCs. OPCs cultured at different densities give rise to similar numbers of oligodendrocyte-lineage cells after several days because cells at low density divide more often than those at high density45. OPC proliferation in this system is not limited by PDGF but instead seems to be restricted by cell contact. By subtly changing oligodendrocyte density in vivo using cell ablation, we have found that nearby OPCs divide and migrate to rapidly fill the vacated space. This is consistent with the possibility that OPC migration and division are normally limited by contact inhibition, although we cannot rule out the possibility that ablation releases an attractive and mitogenic signal. Contact inhibition of proliferation via, for example, activation of receptor tyrosine phosphatases by molecules on the surface of adjacent cells, is an important mechanism of growth control46, but molecules that might mediate it in OPCs have not been described.

Dynamic process activity, surveillance and cell-cell interaction might be general features of vertebrate glial cells. In vivo time-lapse imaging demonstrates that resting microglia constantly extend and retract fine processes that can be rapidly oriented toward nearby lesions47. Astrocytes also rapidly remodel processes48, which might mediate dynamic interactions with synapses and blood vessels. Processes of neighboring microglia seem to repel one another47, and processes of astrocytes terminate close to one another but do not intermingle49. Thus, glial process activity might serve both to mediate cell-cell interactions that influence glial cell number and spacing and to facilitate glial cell functions.

Methods

Fish breeding and maintenance.

Embryos were produced by pairwise matings, raised at 28.5 °C in egg water or embryo medium (EM) and staged according to hours post-fertilization (hpf), days post-fertilization (dpf) and morphological criteria. Tg(olig2:egfp) (ref. 8) and Tg(nkx2.2a:megfp) (ref. 6) fish were used for this study.

Plasmid construction and DNA injection.

To construct p7.2sox10:mrfp, a 7.2-kb genomic fragment of sox10 (ref. 7) was subcloned into the XbaI and SalI sites of a modified pBluescript II plasmid that carried I-SceI recognition sequences flanking both ends of the multiple cloning site. cDNA encoding membrane-localized red fluorescent protein (mRFP) upstream of an SV40 polyadenylation signal was then cut from pCS2:mrfp using EcoRI and NotI and inserted downstream of the sox10 sequence. p7.2sox10:mrfp plasmid was injected at the one-cell stage with I-SceI (New England Biolabs) as described previously50.

Time-lapse photomicroscopy.

At 24 hpf, manually dechorionated embryos were transferred to EM containing 0.003% phenyl-thiourea to prevent formation of dark pigment and were then returned to a 28.5 °C incubator. Larvae used for imaging were anesthetized using 3-aminobenzoic acid ethyl ester, immersed in 0.8% low–melting temperature agarose and mounted on their sides in glass-bottomed 35-mm Petri dishes. Time-lapse images were captured using 20 × dry (NA = 0.75) or 40 × oil-immersion (NA = 1.3) objectives mounted on a motorized Zeiss Axiovert 200 microscope equipped with a PerkinElmer ERS spinning disk confocal system and heated stage and chamber to maintain embryos at 28.5 °C. Z image stacks were collected every 2–5 min, and three-dimensional data sets were compiled using Sorenson 3 video compression and exported using QuickTime to create movies.

Laser microsurgical cell ablation.

Anesthetized larvae were mounted on their sides in 2% methylcellulose on bridged slides and viewed using a 40 × water-immersion objective of a Zeiss Axioskop microscope. Hemisegment positions were determined using adjacent somite boundaries as landmarks. Dorsally migrated EGFP+ or mEGFP+ oligodendrocyte-lineage cells were targeted by their fluorescence and ablated using approximately 5-s pulses of 440-nm light generated by a Photonics Micropoint Laser System. Effective ablation was demonstrated by cellular debris visible using differential interference contrast optics. In some experiments, Tg(olig2:egfp) larvae were placed individually in multiwell plates containing EM and periodically remounted, and the number of dorsal EGFP+ cells within ablated hemisegments and flanking control hemisegments was determined. For time-lapse recordings, larvae were mounted and imaged as described above after ablations.

Quantification of process activity and cell migration.

To measure OPC migration trajectory and distance, Z image stacks were compiled into composite images using Volocity. The center of each cell was identified and a line drawn to the corresponding cell at the next time point. Because soma position did not change between each time point, trajectories reflect instances where the soma moved by at least one cell diameter (∼9 μm). The length of the combined line segments corresponds to the total distance each OPC migrated. To describe process activity, process length was measured at 2-min intervals.

Note: Supplementary information is available on the Nature Neuroscience website.