Oligodendrocyte progenitors balance growth with self-repulsion to achieve homeostasis in the adult brain

Adult NG2⁺ cells extend processes with dynamic filopodia

To define the behavior of NG2⁺ cells in the adult brain, we generated BAC transgenic mice that express a membrane anchored form of EGFP (ref. ²⁴) under control of the NG2 promoter (NG2-mEGFP-H mice) (Supplementary Fig. 1), which allowed visualization of the full extent of their complex morphologies (Supplementary Fig. 2). In vivo two-photon imaging through a cranial window revealed that fluorescent NG2⁺ cells were distributed in a highly ordered manner throughout the upper layers of the cortex in these mice (Supplementary Movie 1), with cells occupying non-overlapping domains. NG2⁺ cells extended highly ramified processes studded with numerous filopodia-like protrusions into the surrounding neuropil (Fig. 1a). In time-lapse recordings, these thin protrusions were highly dynamic, continually extending, retracting, branching and altering their trajectory (Fig. 1a-c, Supplementary Movies 2,3) with an average speed of 0.3 ± 0.01 μm/min (185 protrusions on 10 cells in 5 mice). Motile filopodia were present along the length of each process, and even extended from the soma (Fig. 1a,d,e), indicating that NG2⁺ cells actively survey their local environment. This dynamic behavior is similar to that exhibited by microglial cells^{25, 26}, although NG2⁺ cell filopodia lacked the bulbous endings characteristic of microglial processes and moved more slowly (microglia process speed = 1.47 ± 0.1 μm/min; from ref ²⁶).

Despite the active growth of NG2⁺ cell processes, these cells maintained a radial morphology and contacts between processes were rarely observed. Previous studies indicate that contact mediated inhibition can help establish a radial morphology and control cell spacing through homotypic repulsion^{27, 28}. In time-lapse imaging, growing filopodia of NG2⁺ cells always retracted when they contacted a process of the same or an adjacent NG2⁺ cell (Fig. 1f,g; Supplementary Movie 3), indicating that these progenitors engage in continual self-avoidance through contact mediated inhibition.

Previous studies have shown that removal of the skull and implantation of a chronic cranial window can induce reactive changes in glial cells and enhance the dynamics of dendritic spines on neurons²⁹. To determine if the dynamic behavior of NG2⁺ cells was a reactive response to injury, we prepared NG2-mEGFP-H mice with thinned-skull cranial windows³⁰ and performed time-lapse imaging. NG2⁺ cells imaged through the skull had complex morphologies and highly dynamic filopodia were observed along their processes, features comparable to those observed in mice implanted with chronic cranial windows (Supplementary Fig. 3; Supplementary Movie 4). Moreover, mice implanted with chronic cranial windows did not show reactive changes in microglia or increases in the number of proliferating cells relative to controls when in vivo imaging was initiated (Supplementary Figure 4). These results indicate that the extension of dynamic filopodia from NG2⁺ cell processes is a normal feature of these progenitors in the uninjured brain, rather than a reactive response to cranial window implantation.

NG2⁺ cells migrate continuously through the adult cortex

The tips of advancing NG2⁺ cell processes displayed numerous motile filopodia, reminiscent of neuronal growth cones, whereas retracting processes lacked filopodia (Fig. 1h), suggesting that these dynamics may enable cell migration. To determine if process motility is accompanied by somatic translocation, we re-imaged NG2⁺ cells in the same volume of tissue one month later. Although the position of vascular-associated pericytes in these regions was unchanged, the distribution of NG2⁺ cell somata and the orientation of their processes were dramatically altered, such that it was not possible to unambiguously identify the same NG2⁺ cells from the previous imaging session (Fig. 2a).

More frequent imaging revealed that NG2⁺ cells continually reoriented their processes and moved an average distance of 2.1 ± 0.1 μm per day (318 NG2⁺ cells in 5 mice) (Fig. 2b-d; Supplementary Movies 5,6), in contrast to pericytes that remained largely stationary (Fig. 2e,f). Individual NG2⁺ cells exhibited heterogeneous behaviors, with some cells moving at a continuous rate during the two-week imaging period, while others interrupted periods of relative stability with large movements of > 30 μm over a period of a few days (Fig. 2f). Long-range cues guide the migration of NG2⁺ cells during development³¹, raising the possibility that similar gradients influence their movement in the adult CNS. However, analysis of the trajectories of all NG2⁺ cells in a 0.06 mm³ cortical volume revealed that there was no directional bias in their movement (Fig. 2g), suggesting that their behavior is controlled by local interactions rather than large-scale gradients of attractive or repulsive cues.

Constant turnover of NG2⁺ cells in the adult brain

In addition to the constant reorganization of NG2⁺ cell position in the cortex, there were dynamic changes in the overall population, due to the proliferation, differentiation, and death of individual NG2⁺ cells (Fig. 3a-c). Cell death was characterized by hypertrophy, retraction of processes and abrupt disappearance of the cell (Supplementary Fig. 5), whereas differentiation into oligodendrocytes was accompanied by a gradual decrease in fluorescence, transformation of their highly ramified processes to long, unbranched processes, and expression of myelin proteins (Supplementary Fig. 6). NG2⁺ cells did not differentiate into astrocytes or neurons in this region, providing further evidence that these glial cells serve as lineage restricted progenitors in the normal CNS⁷; however, if EGFP is degraded more rapidly during transdifferentiation, adoption of these alternate fates would be difficult to assess using this approach.

In contrast to the segregation of principal neurons into distinct cortical lamina, NG2⁺ cells were evenly distributed among cortical layers, and the core behaviors (proliferation, differentiation, and death) were not biased to a particular lamina (Supplementary Fig. 7). On each day, 3 % of NG2⁺ cells were engaged in these dynamic behaviors (Fig. 3d), which expanded to include 36 % of the population over two weeks. Notably, cell loss through death or differentiation (1.2 ± 0.1 %/day) was balanced by cell addition through proliferation (1.5 ± 0.1 %/day, P = 0.09), allowing the density of NG2⁺ cells to remain constant (Fig. 3e), indicating that this progenitor pool is constantly reorganized by the continual loss and addition of new cells.

NG2⁺ cell density is maintained through local proliferation

If homeostatic control of NG2⁺ cell density is governed by local interactions, then differentiation or death should be associated with division of a neighboring cell. We observed that 98 % of differentiating and 76 % of dying NG2⁺ cells were associated with proliferation of a neighboring NG2⁺ cell (< 50 μm away), whereas only 43 % of stable NG2⁺ cells were associated with a local proliferation (P < 0.005 for both death and differentiation) (Fig. 4a-c). Moreover, the probability of local proliferation in instances of death or differentiation was highest at the time of cell loss, but did not vary with time for control NG2⁺ cells (Fig. 4d). Neighboring NG2⁺ cells rapidly invaded the territory of differentiating cells (Fig. 5a,b,e), suggesting that the homotypic repulsive cues that prevent overgrowth are down regulated with differentiation. In contrast, dying NG2⁺ cells maintained distinct, though progressively smaller, territories until removal (Fig. 5c,d,e). The gradual decrease in the territory of dying cells (Supplementary Fig. 5), occasionally resulted in repositioning of neighboring NG2⁺ cells without cell proliferation, accounting for the somewhat lower incidence of neighbor proliferation in cases of death.

To further assess whether cell loss is sufficient to induce local proliferation, we ablated individual NG2⁺ cells in vivo using focal laser illumination and examined the response of neighboring NG2⁺ cells. Following cell removal, neighboring NG2⁺ cells reoriented their processes and invaded the former territory of the ablated cell, providing further evidence that NG2⁺ cell domains are maintained by self-avoidance (Fig. 6a-d; Percent territory invasion: Control, 6 ± 1 %, n = 10 cells; Cell ablation, 52 ± 10 %, n = 12 cells; *P < 0.001, Student’s t test). Moreover, removal of several NG2⁺ cells within the imaging field triggered reorientation, cell migration, and proliferation of NG2⁺ cells in the vicinity, resulting in rapid restoration of their density (Fig. 6e-h; Number of proliferating cells: Control 1 ± 0.4 cells, Cell ablation 5 ± 0.3 cells; volume analyzed = 0.02 mm³; from 4 control and 3 ablation mice; *P < 0.001, Student’s t test). These data indicate that NG2⁺ cells maintain a constant density though a local homeostatic mechanism, in which cell loss results in release from growth inhibition and rapid replacement through division of a neighboring cell.

NG2⁺ cells directly differentiate into oligodendrocytes

NG2⁺ cells in the adult CNS can be generated from SVZ progenitors³², which may have a greater capacity for proliferation and differentiation than cells with a longer residence time. To determine if repopulation is accomplished by a subset of highly proliferative, self-renewing NG2⁺ cells³³, we followed the behavior of proliferating NG2⁺ cells by repeatedly imaging the same cortical volume for many weeks. During this period, the majority of proliferating NG2⁺ cells divided only once, and for cells that divided multiple times, cell cycle length was variable (average cell cycle: 18.5 ± 1 days, n = 97 cells) (Fig. 7a-c), as expected if extrinsic factors (i.e. death/differentiation of a neighbor) determine when NG2⁺ cells divide. Indeed, cells that underwent multiple divisions were located in regions where there was substantial turnover/differentiation. Moreover, lineage trees of highly proliferative cells revealed instances where both sister cells underwent multiple divisions, as well as instances where one sister cell either died or differentiated and the other remained stable (Fig. 7d). These findings, and the consistent proliferation of immediate neighbors following cell loss, suggest that homeostasis is achieved by endowing all NG2⁺ cells with the capacity to divide⁷, rather than by seeding the brain with a subset of highly proliferative cells.

New neurons are formed in the CNS through symmetric or asymmetric division of progenitors³⁴; however, it is unclear whether oligodendrocytes are formed through similar mechanisms or by direct differentiation of NG2⁺ cells²⁰. Lineage tracing through in vivo imaging revealed that differentiation was rarely preceded by cell division (7/107 differentiation events) (Fig. 7e), indicating that oligodendrocytes in the adult CNS are formed primarily through direct differentiation of these progenitors. Thus, proliferation of NG2⁺ cells in vivo reflects a homeostatic response to cell depletion rather than oligodendrogenesis.

NG2⁺ cells participate in glial scar formation

NG2⁺ cell proliferation is often enhanced after injury and the NG2 proteoglycan accumulates at lesion sites where it can inhibit axon outgrowth^{18, 35}. However, the reason for their enhanced proliferation is unknown, and the participation of NG2⁺ cells in scar formation has not been established, as NG2 also is expressed by macrophages and pericytes in injured tissue^{36, 37}. To assess the behavior of NG2⁺ glial cells to CNS injury, we induced focal laser lesions²⁵ in the cortex of adult NG2-mEGFP-H mice. Immediately following injury, NG2⁺ cells adjacent to the lesion extended processes towards the lesion site and eventually surrounded the site of injury (Fig. 8a-d, Supplementary Movie 7). Although their response to focal injury was similar to microglia²⁵, NG2⁺ cells extended their processes more slowly than microglia to the lesion site (Fig. 8b). As a consequence, scars acquired a layered structure consisting of microglial cell processes surrounded by NG2⁺ cell processes (Supplementary Fig. 8). Reorientation and extension of NG2⁺ cell processes after injury was followed by migration of these cells to the lesion site within several weeks, and eventual removal of these cells as the scar resolved (Fig. 8e,f and Supplementary Movie 8). As predicted by their homeostatic behavior to cell loss, depletion of NG2⁺ cells from the area surrounding the lesion was accompanied by proliferation of neighboring NG2⁺ cells (Fig. 8g), with the number of proliferating cells matching the number lost (before lesion, 18 ± 2 NG2⁺ cells; after lesion, 16 ± 1 NG2⁺ cells, n = 4 mice, P = 0.5) (Fig. 8h). These findings demonstrate that NG2⁺ cells contribute to glial scar formation, and provide an explanation for the enhanced proliferation of NG2⁺ cells that occurs following many types of CNS trauma.

Generation of NG2-mEGFP transgenic mice

A mouse bacterial artificial chromosome (BAC) clone containing the NG2 (Cspg4) gene (RP23-309G21) was purchased from BACPAC Resources Center and modified by homologous recombination⁵¹. To target EGFP to the plasma membrane, the first 26 amino acids of Lck, a Src family tyrosine kinase which contains two palmitoylation domains and a myristoylation domain (gift from Drs. M. Dailey and S.Green, University of Iowa)²⁴, was fused to the N-terminus of EGFP in pEGFP-N1 (Clontech). A cDNA encoding this membrane anchored EGFP (mEGFP) and subsequent rabbit β-globin polyA signal were placed at the translational initiator (ATG) of the NG2 (Cspg4) gene with two flanking homology arms (500 bps for each arm). The modified BAC was linearized by NotI digestion, injected into the pronucleus of mouse embryos in the JHU Transgenic Core, and implanted into pseudopregnant females. Of the three lines of NG2-mEGFP mice generated, two lines were selected for further studies.

In vivo two-photon microscopy

All experiments were performed in strict accordance with protocols approved by the Animal Care and use Committee at Johns Hopkins University. To prepare cranial windows, NG2-mEGFP-H mice (2–3 months old) were anesthetized by interperitoneal injection of ketamine (100 mg/kg b.w.) / xylazine (10 mg/kg b.w.); body temperature was maintained at 37°C with a thermostat-controlled heating plate. The skin over the right cerebral hemisphere was retracted and the skull cleaned. A metal plate with a center hole was attached to the skull with dental cement (C&B-Metabond) to allow for head stabilization. A ~400 μm diameter region of skull over somatosensory cortex (−0.5 to −2 mm post bregma and 1 to 3.5 mm lateral) was either thinned (~20 μm thickness) or removed using a high-speed dental drill^{30, 52}. For cranial windows, a piece of cover glass (VWR, No. 1) was placed within the craniotomy and sealed with dental cement. In vivo imaging sessions began immediately after surgery (thinned-skull preparation) or after a minimum of 3 weeks (chronic cranial window). Although craniotomy can induce gliosis²⁹, these small cranial windows did not induce reactive changes in microglial cells or NG2⁺ cells 3 weeks following implantation (Supplementary Fig. 13). Mice were anesthetized with isoflurane (1.0–1.5%; mixed with 0.5 liters/min O₂) during imaging. Images were collected using a Zeiss LSM 710 microscope equipped with a GaAsP detector using a mode-locked Ti:sapphire laser (Coherent Ultra II) tuned to 920 nm. The average power at the sample during imaging was <50 mW. Vascular landmarks (EGFP⁺ pericytes) were utilized to identify the same imaging area on subsequent days.

Two-photon laser-induced lesions and individual cell ablation

Small lesions were induced in cortical gray matter by illuminating with the Ti:sapphire laser for several seconds (780 nm; ~150 mW, 1–3 s)²⁵. This protocol induced a ~ 20 μm area of damage visible as an autofluorescent sphere. To ablate individual NG2⁺ cells in mice with thinned-skull windows, cell somata were subjected to repeated, short-duration laser irradiation (780 nm; ~150 mW, 100-500 ms, 1–6 repetitions)⁵³. Successful ablations were preceded by slight swelling of the soma that was followed by cell disappearance within 1–8 hours. Laser power, pulse duration, and pulse repetition were varied depending on the depth of the targeted cell and thickness of the overlying thinned-skull.

Image processing and analysis

Image stacks and times series were analyzed using Fiji/ImageJ⁵⁴. All analysis was performed on unprocessed images. For presentation in figures, image brightness and contrast levels were adjusted for clarity. Filopodial images were additionally processed with a Gaussian filter (radius = 2 pixels) to remove detector noise. For pseudo-color display of individual cells, the 3D cell volume was defined plane-by-plane for each time point and a custom Fji/ImageJ script was used to segment and/or colorize the cell.

Filopodial analysis

Filopodia on NG2⁺ cell processes (≥ 50 μm in length) were analyzed frame-by-frame using a custom Fiji/ImageJ script to quantify filopodial length, motility, density, and lifetime. Filopodia were considered dynamic if the length varied more than 2 μm during the imaging period. For analysis of filopodial density in fixed tissue, protrusions with a length at least twice the width and a minimum length of 2 μm were classified as filopodia.

Cell tracking

NG2⁺ cells were tracked in 3D using custom Fiji/ImageJ scripts by defining soma position at each time-point, recording XYZ coordinates and categorizing “cellular behavior” (e.g. proliferation, differentiation, etc.). NG2⁺ cells were classified as proliferating if, 1) two cells appeared in the same location where there was one cell on the previous time-point, 2) the somata were separated by less than 50 μm 3) the processes of recently divided cells were oriented in opposing directions, and 4) the somata lacked major cell processes at the point closest to the sister cell. The 3D volume around each proliferating cell was examined to exclude neighboring NG2⁺ cells that migrated into the field. NG2⁺ cells were classified as differentiating if, 1) cells exhibited a progressive decrease in EGFP fluorescence, 2) cells changed from having radial, highly branched processes, to long, unbranched processes characteristic of oligodendrocytes, and 3) the processes of neighboring NG2⁺ cells invaded the territory of the differentiating cell. NG2⁺ cells were classified as dying if cells exhibited a, 1) shortening and/or blebbing of processes, 2) reduction in territory size, 3) increase in brightness of the cell soma, and 4) complete disappearance of EGFP fluorescence in < 3 days without movement from the imaging area. NG2⁺ cells that lost EGFP fluorescence but could not be definitively identified as differentiated or dying were classified as “Unsure.”

NG2⁺ cell territory invasion

To calculate territory invasion following death or differentiation, a convex hull was computed plane-by-plane for each time-point around the differentiating/dying cell, the cell masked and EGFP⁺ processes contained within the convex hull autothresholded using the Fiji/ImageJ IsoData algorithm. A maximum projection of 3D volume defined by the convex hull was generated and the area of processes for each time-point determined. To calculate territory invasion following NG2⁺ cell ablation, the 3D volume of the ablated cell was merged with the image of the following time-point to identify “invading” processes. For control cells, the cell in the second time-point was masked and the region of the 3D volume of the cell in the first time-point was overlaid onto the second time-point. A maximum projection of 3D volume of the invading processes was autothresholded using the Fiji/ImageJ IsoData algorithm and the extent of territory invasion calculated by normalizing the area occupied by “invading” processes at the last time-point to the area occupied by the cell at the first time-point.

Analysis of local proliferation

NG2⁺ cells that underwent proliferation, differentiation, or death were identified during the imaging period. For each differentiating cell, the time-point at which the cell displayed a premyelinating oligodendrocyte morphology⁵⁵, and neighboring cells had begun invading that cell’s territory, was marked as day 0. For each cell that died, the time-point that the cell displayed altered morphology (increased brightness of cell soma and reduced territory size), immediately preceding disappearance, was marked as day 0. After identification of a NG2⁺ cell that underwent differentiation/death, immediately adjacent NG2⁺ cells were examined for cell division preceding and following the event. The day of proliferation was marked as the time-point at which the two sister cells had undergone cytokinesis and had no adjoining processes. Cells were excluded from analysis if the entire differentiation/death process did not occur during the imaging period or if not all adjacent cells were located within the field of view. For controls, we analyzed cells that did not undergo proliferation, differentiation, or cell death for the entire imaging period and did not have more than 1 cell division within 100 μm.

Analysis of the response to focal injury

To follow appearance of NG2⁺ cells at lesions, EGFP⁺ processes entering a 75 μm zone surrounding lesion sites (area X) from a larger 150 μm zone (area Y) were measured as a function of time (Fig. 4a)²⁵. To calculate the area surrounding the lesion occupied by NG2⁺ cell processes, images were autothresholded using the Fiji/ImageJ IsoData algorithm and the number of pixels was assessed. The change in pixel number over time (R_x(t)) compared to the initial value of pixels (R (0)) represented NG2⁺ x cell processes accumulation. To control for variations in the number NG2⁺ cells surrounding the lesion site in different experiments, accumulation of NG2⁺ cell processes was calculated relative to the number of processes in the outer area Y immediately following induction of the lesion (R_y(0)). The NG2⁺ cell response index at any time-point, (R(t)), is given by the equation:

To quantify response speed, we measured the length of NG2⁺ cell processes within 150 μm of the lesion site over time. To determine the leading and trailing processes of NG2⁺ cells, the coordinate system was rotated so that the y-axis passed through the lesion site at the top and the soma of the cell was located at the origin (lesion site = 90° from the cell of interest). The angle at which processes extended from the soma determined if they were on the leading edge (0–180°) or trailing edge (181–360°) of the NG2⁺ cell.

To determine the magnitude and direction of NG2⁺ cell movements with respect to the lesion site, the coordinate system was rotated such that the x-axis passed through the lesion site at the right and the start point of the cell was located at the origin (lesion site = 0° from the start point). Next, the angle θ between the displacement vector of NG2⁺ cell movement and the x-axis (displacement vector between the start point and the lesion site) was determined using:

where a_i is the displacement vector for the cell, i and a_L is the displacement vector between the start point of the cell and the lesion site, respectively, a_i • a_L is the dot product between the two vectors, and $\mid a\mid =\sqrt{a•a}$. The sign of the angle was determined by calculating the determinant of the two vectors using:

where the Cartesian coordinates of the lesion site, the start position, and the final position of the NG2⁺ cell of interest are defined as a, b, c respectively.

Immunohistochemistry

Mice were administered an overdose of anesthesia (pentobarbital; 100 mg/kg, i.p.) perfused transcardially with 4% formaldehyde (PFA in 0.1M phosphate buffer [pH 7.4]), their brains extracted and postfixed in 4% formaldehyde for 4 hours or overnight at 4°C, then cryoprotected in a 30% sucrose solution (in PBS, pH 7.4) at 4°C for up to 36 hours. Brains were frozen in TissueTek, sectioned (coronal, bregma 0.2 to –1.9 mm) at 30–50 μm thick, and incubated free-floating for 1–2 hr at room temperature (RT) in a blocking solution (5% normal donkey serum, 0.3% Triton X-100 in PBS, pH 7.4). Sections were incubated with primary antibodies (Supplementary Table 1) suspended in blocking solution overnight at 4°C on an orbital shaker. For Olig2 immunostaining, brain sections were incubated in LAB solution (Polysciences) for 10 min prior to blocking. For BrdU immunostaining, sections were treated with 2N HCl at 37°C for 30 min., then neutralized with 0.1 M sodium borate buffer (pH 8.5) prior to incubation with primary antibodies. After washing in blocking solution, sections were incubated with fluorescently conjugated secondary antibodies (Supplementary Table 2) for two hours at RT then mounted on slides with Aqua Poly/Mount (Polysciences). Images were acquired using either an epifluorescence microscope (Zeiss Axio-imager M1) and Axiovision software (Zeiss) or a confocal laser-scanning microscope (Zeiss LSM 510 Meta). For cell proliferation analysis, mice were provided with BrdU-containing drinking water (1 mg/ml supplemented with 1% sucrose) and received two injections of BrdU (50 mg/kg, i.p.) per day, at least 8 hr apart, for one week prior to perfusion. To analyze the effects of cranial window implantation, burr holes were drilled <1 mm outside the crainiotomy area and 1 μl of dextran-conjugated rhodamine was injected into the brain to label the position of the cranial window just prior to sacrifice.

Statistical analysis

Statistical analyses were performed in OriginPro (OriginLab), Excel (Microsoft), or Fiji/ImageJ. Sample sizes were chosen according to the standard practice in the field. Normality of the data was tested with the Shapiro-Wilk test. Significance was determined using the Mann-Whitney test, unpaired or paired two-tailed Student’s t test, or one-way ANOVA with Tukey post-hoc test, as noted. All direction statistics were performed using the modification by Moore to the Rayleigh’s test⁵⁶, implemented in Chemotaxis and Migration Tool (ImageJ plug-in, Ibidi). All data are presented as mean ± standard error of the mean (s.e.m.).