Myelin biogenesis is associated with pathological ultrastructure that is resolved by microglia during development

To enable rapid propagation of action potentials, axons are ensheathed by myelin, a multilayered insulating membrane formed by oligodendrocytes. Most of the myelin is generated early in development, in a process thought to be error-free, resulting in the generation of long-lasting stable membrane structures. Here, we explored structural and dynamic changes in CNS myelin during development by combining ultrastructural analysis of mouse optic nerves by serial block face scanning electron microscopy and confocal time-lapse imaging in the zebrafish spinal cord. We found that myelin undergoes extensive ultrastructural changes during early postnatal development. Myelin degeneration profiles were engulfed and phagocytosed by microglia in a phosphatidylserine-dependent manner. In contrast, retractions of entire myelin sheaths occurred independently of microglia and involved uptake of myelin by the oligodendrocyte itself. Our findings show that the generation of myelin early in development is an inaccurate process associated with aberrant ultrastructural features that requires substantial refinement.


INTRODUCTION
electron-dense fragments that likely represent products of myelin phagocytosis (Figure 2A, section 1 7). To confirm this, we co-immunostained tissue sections from mouse P14 and P80 corpus callosum 2 ( Figure 2C) and optic nerve ( Figure 2D) with myelin-basic protein (MBP), IBA1 and LAMP1 to 3 label microglia and late endosomes/lysosomes. About 3.5% of the microglia contained MBP + 4 fragments within LAMP1 + organelles at P10, while around 1.2% or less did at P80 (corpus 5 callosum: P14: median 3.56% (interquartile range (IQR) 3.10-3.94%), P80: median 1.22% (IQR 6 0.93-2.27%), p=0.0286; optic nerve: P14: median 3.48%, (IQR 2.98-5.44%), P80: median 0.49% 7 (IQR 0.31-1.32%), p=0.0286, Mann Whitney test, n = 4 mice). Together, these results show that 8 microglia engulf and degrade myelin in mice under normal conditions. To clarify whether this role 9 is unique to microglia, we performed immunostainings for astrocytes, which also have the capacity 14 Microglia surveil and phagocytose myelin in zebrafish 15 In order to gain insight into how microglia interact with myelin in vivo, we conducted further 16 experiments in the zebrafish spinal cord. This enabled us to follow myelination in a relatively short 17 developmental window by expressing fluorescent reporters and implementing time-lapse imaging. 18 When we imaged transgenic zebrafish larvae expressing mpeg1:EGFP (labelling microglia) and 19 sox10:mRFP (labelling myelin) sheaths starting from the onset of myelination at 3 days post 20 fertilization (dpf), we found that microglia were often elongated and in contact with myelin sheaths,  Figure 3C) and occasionally took breaks with negligible soma displacement over one or several 5 subsequent time frames (average break duration: 72.00 ± 47.44 min, Figure 3D). We were able to 6 identify distinct migrating patterns ( Figure 3E): some microglia migrated from anterior to posterior 7 (or vice versa) in a 'linear' fashion, while others performed a 'circular' movement, returning to 8 approximately their original position within hours. We also observed combinations of these two 9 patterns ('undulating') and microglia that remained at the same position during most of the 10 acquisition ('static'). The observation that some microglia returned to their previous position or 11 remained static over longer periods prompted the question whether myelin sheaths in some 12 oligodendrocyte territories are more intensely screened than others. We therefore analyzed 13 microglial presence in quadrants of the dorsal and ventral spinal cord. Strikingly, we saw that 14 microglial presence strongly deviated from a random distribution, with some quadrants being 15 highly covered and others not at all during the entire acquisition ( Figure 18 We next asked whether microglia phagocytose myelin during their surveillance. Because the 19 sox10:mRFP reporter transiently labels sensory neurons in addition to oligodendrocytes, we 20 analyzed myelin phagocytosis in Tg(mpeg1:EGFP; mbp:mCherry-CAAX) animals. With time-21 lapse imaging, we observed the intake of a dense mbp:mCherry-CAAX + myelin piece by a 22 microglial process towards the microglial soma ( Figure 3M; Video 4). At 3 dpf, no microglia 23 contained myelin fragments, whereas all of them did by 7 dpf (n = 5 fish). Microglia showed 1 increasing volumes of internalized mbp:mCherry-CAAX + fragments between 5 and 7, or 14 days 2 ( Figure 3G). Importantly, microglia did not change their morphology towards an activated 3 phenotype during myelin phagocytosis. Myelin fragments were highly motile inside microglia  Next, we asked which molecular cues on myelin sheaths are recognized by microglia and promote 11 myelin phagocytosis. Myelin is rich in the phospholipid phosphatidylserine (PS), and when 12 exposed on the outer leaflet, it may serve as an 'eat-me' signal for microglia, which contain 13 multiple PS receptors (Prinz, Jung, & Priller, 2019). To determine whether PS is exposed on 14 degenerating myelin sheaths, we used a recently described PS reporter that can be applied in vivo. 15 We took advantage of secreted glycoprotein MFG-E8, which specifically recognizes PS and, when 16 fused to EGFP, can be delivered as a recombinant protein for PS detection (Kranich et al., 2020). 17 Because injections of recombinant protein into the spinal cord may induce damage, we modified 18 this approach by expressing the genetically encoded secreted MFG-E8-EGFP in microglia ( Figure   19 3I,J). When we expressed mpeg1:KalTA4;UAS:MFG-E8-EGFP in transgenic mbp:mCherry-20 CAAX zebrafish larvae and masked MFG-E8-EGFP fluorescence by the mbp:mCherry-CAAX 21 channel, we found that MFG-E8-EGFP localized to myelin sheaths (16.34 ± 6.58%) and to small, 22 extracellular fragments positive for the myelin reporter (in 51.06 ± 16.15%) ( Figure 3K). MFG-23 E8-EGFP-positive spots on myelin sheaths were focal and their inspection in 3D revealed that they 1 marked small protrusions of the myelin sheath. We propose that these distinct irregularities on 2 otherwise normal-appearing sheaths may eventually shed off, resulting in the release of 3 extracellular fragments. These fragments may resemble "shedosomes" that occur during dendritic 4 pruning (Han, Jan, & Jan, 2011) or "axosomes" which form during axon branch removal (Bishop,5 Misgeld, Walsh, Gan, & Lichtman, 2004). 6 We occasionally found microglia engulfing abnormal myelin sheaths that either lost their uniform 7 3D structure, or showed signs of fragmentation or local outfoldings ( Figure 3L). However, myelin 8 abnormalities were remarkably stable which made their removal by microglia difficult to follow 9 with time-lapse imaging of larva beyond 5 dpf. We therefore asked whether myelin pathologies 10 increase in the absence of microglia. For this, we analyzed a double mutant of the csf1 receptor, a 11 duplicated gene in zebrafish, which is essential for microglia proliferation and development. These double mutant fish showed significantly more myelin pathologies in the 10 dpf ventral spinal cord 18 compared to wild-type animals (wt: 5.5 ± 1.72, csf1r DM : 11.27 ± 2.76, p<0.0001, n = 10 (wt) and 19 n = 11 (csf1r DM ) fish; Figure 3N). These pathologies included fully degenerating sheaths and 20 myelin segments that partially lost their 3D structure, making them appear like flaps. However, the 21 increased myelin pathologies in csf1r DM mutants were primarily due to the presence of bulb-like myelin outfoldings ( Figure 3O), resembling 'myelinosomes' previously described in acute 1 demyelinating EAE lesions(Romanelli et al., 2016).

2
Oligodendrocytes retract myelin sheaths independently of microglia and degrade myelin 3 membrane fragments 4 We also wondered whether myelin ingestion by microglia could be attributed to myelin sheath (R 2 =0.0256, p=0.4453, n = 5 fish, Figure 4C). Furthermore, retractions did not significantly differ 15 between microglia-depleted csf1r DM and wild-type larvae (wt: 42.25 ± 7.50, csf1r DM : 57.75 ± 15.90, 16 p=0.1429, Figure 4D; Video 5). Inspecting retractions more closely revealed that myelin sheaths 17 were entirely withdrawn into oligodendrocyte cell bodies and no visible fragments were split off 18 to the exterior. However, bright fragments appeared and shuttled within the oligodendrocyte soma 19 in the course of retractions ( Figure 4E; Video 5). We stained Tg(mbp:mCherry-CAAX) larvae with 20 Lysotracker and analyzed the colocalization of these bright membrane fragments with lysosomes. 21 We found that major portions of the fragments, but not the rest of the oligodendrocyte membrane, 22 colocalized with lysosomes ( Figure 4F). Intriguingly, we also found Lysotracker staining of myelin 23 sheath pathologies, but not of normal-appearing adjacent sheaths or total myelin in the same image 1 ( Figure 4G, H), raising the possibility that some of the pathology is resolved by the oligodendrocyte 2 itself. We next assessed whether myelin membrane fragments occur inside of the oligodendrocyte 3 soma beyond the zebrafish model. For this, we co-immunostained mouse tissue sections of the 4 optic nerve and the brain for MBP, the oligodendrocyte marker OLIG2, and LAMP1. In the corpus 5 callosum, we found lysosome-associated MBP + fragments in 0.53% (0.44-1.00%, median and IQR) 6 of the OLIG2 + oligodendrocytes at P14, but only 0.20% (0.09-0.31%, median and IQR) at P80 7 (p=0.0286, Fig. 4I). Similarly, 1.92% (1.08-2.29%, median and IQR) of the oligodendrocytes in 8 the optic nerve contained MBP + fragments at P14, but only 0.30% (0.00-0.40%, median and IQR) 9 did at P80 (p=0.0571, Figure 4I).  Counter tool in Fiji.

16
The number of cells with internalized myelin particles, labeled by anti-MBP and colocalizing with 17 anti-LAMP1, was quantified and normalized by the total numbers of cells in an area of 0.6 mm 2 .

18
Tracking of microglia motility in zebrafish 19 Tracking of microglia soma displacement was done with the Manual Tracking plugin in Fiji (F.  Colocalization of LysoTracker staining with myelin pathology 18 Myelin pathologies and parts of adjacent myelin sheaths were manually segmented in Imaris. 19 Additionally, an automatic surface was created for the 552 nm-channel, representing total myelin Statistical analysis was performed in GraphPad Prism. All samples were tested for normality.

10
Comparison of two independent groups was carried out using student's t test (for normally 11 distributed data) and Mann-Whitney U test (for non-normally distributed data). Groups with small 12 sample sizes that were visibly non-normally distributed were tested by Mann-Whitney U test.

13
Comparisons between three or more groups were done by one-way ANOVA, followed by Tukey's  The authors declare no competing interests.