Microglia regulate central nervous system myelin growth and integrity

Myelin is required for the function of neuronal axons in the central nervous system, but the mechanisms that support myelin health are unclear. Although macrophages in the central nervous system have been implicated in myelin health1, it is unknown which macrophage populations are involved and which aspects they influence. Here we show that resident microglia are crucial for the maintenance of myelin health in adulthood in both mice and humans. We demonstrate that microglia are dispensable for developmental myelin ensheathment. However, they are required for subsequent regulation of myelin growth and associated cognitive function, and for preservation of myelin integrity by preventing its degeneration. We show that loss of myelin health due to the absence of microglia is associated with the appearance of a myelinating oligodendrocyte state with altered lipid metabolism. Moreover, this mechanism is regulated through disruption of the TGFβ1–TGFβR1 axis. Our findings highlight microglia as promising therapeutic targets for conditions in which myelin growth and integrity are dysregulated, such as in ageing and neurodegenerative disease2,3.

Myelin is required for the function of neuronal axons in the central nervous system, but the mechanisms that support myelin health are unclear. Although macrophages in the central nervous system have been implicated in myelin health 1 , it is unknown which macrophage populations are involved and which aspects they influence. Here we show that resident microglia are crucial for the maintenance of myelin health in adulthood in both mice and humans. We demonstrate that microglia are dispensable for developmental myelin ensheathment. However, they are required for subsequent regulation of myelin growth and associated cognitive function, and for preservation of myelin integrity by preventing its degeneration. We show that loss of myelin health due to the absence of microglia is associated with the appearance of a myelinating oligodendrocyte state with altered lipid metabolism. Moreover, this mechanism is regulated through disruption of the TGFβ1-TGFβR1 axis. Our findings highlight microglia as promising therapeutic targets for conditions in which myelin growth and integrity are dysregulated, such as in ageing and neurodegenerative disease 2,3 .
Myelin ensheathes neuronal axons to ensure their health and rapid propagation of electrical impulses to support central nervous system (CNS) functions, for example, cognition. Learning and memory involve the formation of myelin and require myelin to be of good structural integrity. Myelin layers are compacted to a thickness proportional to the axon diameter 4 ; however, with ageing and in neurodegenerative disease, disruption of these myelin properties occurs through hypermyelination. Enlarged areas of uncompacted myelin (where myelin grows) leads to myelin that is thicker, unravelling and forming protrusions (termed outfoldings), and loss of myelin integrity through degeneration also occurs in these contexts 2,3,5 . These myelin changes lead to impaired cognition in mice and predict poor cognitive performance in aged nonhuman primates and humans [6][7][8][9] . However, the fundamental mechanisms that coordinate appropriate formation, growth and integrity of CNS myelin are unclear. Recent research has implicated a population of CNS-resident macrophages, microglia, in this process.
Myelination and the generation of myelin-forming oligodendrocytes are impaired following microglial depletion through loss of function of the pro-survival colony stimulating factor 1 receptor (CSF1R) 1 . However, this approach also targets CNS-resident border-associated macrophages (including perivascular macrophages) and blood monocytes. Therefore, it is unclear which macrophage populations regulate myelin, and the specific involvement of microglia in myelin formation and health is unknown.

Article
Our results indicate that microglia are dispensable for oligodendrocyte maturation and developmental myelin ensheathment. This finding is in contrast to previous attributions of these functions to microglia following depletion of all CNS macrophage populations.

Microglia prevent hypermyelination
However, Fire Δ/Δ mice showed abnormal myelin structure indicative of hypermyelination. At P25-P30, there was an increase in myelin   that was outfolding or unravelling in Fire Δ/Δ mice (Fig. 2a-c), which was documented in 44% of sheaths compared with only 14% in Fire +/+ controls. Fire Δ/Δ mice showed enlarged areas of uncompacted myelin (inner tongue) on smaller diameter axons ( Fig. 2d-f). As enlarged inner tongues precluded conventional g ratio analysis, we measured myelin thickness directly and observed increased myelin thickness in Fire Δ/Δ mice preferentially on large diameter axons (Fig. 2g,h). Axon diameter-dependent observations may reflect that myelination of larger diameter axons occurs before that of smaller ones, first involving growth at the inner tongue followed by compaction that thickens the myelin sheath. We assessed the impact of these myelin changes on axonal health in Fire Δ/Δ mice. Although expression of phosphorylated neurofilament was unaffected at 1 month of age, axonal spheroids, which are indicative of impaired axonal transport, were occasionally observed at later ages in <0.1% of myelinated axons (Extended Data Fig. 2f-h). Myelin outfoldings and unravelling persisted in Fire Δ/Δ mice at 3-4 months of age ( Fig. 2i-k), and enlarged inner tongues and increased myelin thickness were observed across all axon diameters compared with Fire +/+ mice ( Fig. 2l-o) and younger Fire Δ/Δ mice (Extended Data Fig. 3). Therefore, hypermyelination occurs in the absence of microglia, which indicates that microglia are required for the regulation of myelin growth. Given that these changes in myelin structure are sufficient to cause cognitive impairment in other models 12,13 , we evaluated cognition in Fire Δ/Δ mice using the Barnes maze spatial learning and memory task (Extended Data Fig. 4a). Both Fire Δ/Δ and Fire +/+ mice became progressively faster at locating the target hole with the underlying escape chamber (primary latency), which indicated spatial learning in these mice (Extended Data Fig. 4b,c). Following removal of the escape chamber, probes 1 h and 3 days later indicated no memory-encoding deficit, as indicated by the percentage of time spent in the target quadrant and the number of nose pokes in and around the target hole (Extended Data Fig. 4d-f). Next, we tested cognitive flexibility, which is learning to adjust thinking from an old to a new situation and is highly dependent on the structural integrity of myelin 6,14,15 . This experiment involves learning to locate an escape hole placed 180° from the original target (Extended Data Fig. 4g-k). Although the time taken to locate the new target was unimpaired in Fire Δ/Δ mice, significantly more errors were made before reaching it (Extended Data Fig. 4h,i), which indicated that these mice have poor cognitive flexibility. Fire Δ/Δ mice did not have confounding anxiety or motor deficits (Extended Data Fig. 4l-r). Therefore, the absence of microglia is associated with impaired cognitive flexibility.
Recent studies have indicated that learning and memory encoding require new oligodendrogenesis, and that long-term consolidation of this information involves increased myelination 16,17 . Therefore, we assessed whether these processes occur in the absence of microglia. The generation of new oligodendrocytes from proliferating progenitor cells was identified through the incorporation of 5-ethynyl-2′-deoxyuridine (EdU) provided during the cognitive testing stage. There was a similar number of newly generated oligodendrocytes in Fire Δ/Δ mice and Fire +/+ mice (Extended Data Fig. 5a-d), which was consistent with the largely unimpaired learning and memory encoding observed in Fire Δ/Δ mice. However, whereas Fire +/+ mice had a significantly increased number of myelinated axons in the corpus callosum 6 weeks after completion of the cognitive task, this did not differ between untrained and trained Fire Δ/Δ mice (Extended Data Fig. 5e-g). These findings suggest that the absence of microglia prevents the increase in myelination that normally occurs with consolidation of new spatial information. Of note, we did not find an association between the number of myelinated axons in a given mouse and its reversal cognitive performance (Extended Data Fig. 5h). This result suggests that either a threshold number of myelinated axons is required for cognitive flexibility or that myelin structural changes may be more relevant for this function.

Microglia prevent demyelination
Assessment of myelin in Fire Δ/Δ mice at 6 months of age showed areas of substantial demyelination (Fig. 3a) and areas of patchy demyelination. This in turn resulted in a significant decrease in the number and proportion of myelinated axons compared with Fire +/+ mice ( Fig. 3b-d).
The patchy nature of demyelination did not result in widespread loss of myelin protein across the corpus callosum (Extended Data Fig. 6a-c). Axonal spheroids were rarely observed (<0.1% of axons). Axons retaining myelin in 6-month-old Fire Δ/Δ mice had reduced inner tongue size and thinner myelin compared with Fire +/+ mice (Fig. 3e,f and Extended Data Fig. 6d) and with 3-4-month-old Fire Δ/Δ mice (Extended Data Fig. 3). Demyelination was not associated with loss of oligodendrocytes at this age or younger (Extended Data Fig. 6e-h). Demyelination was initiated at 4.5 months of age in Fire Δ/Δ mice (Extended Data Fig. 6i-l), and unmyelinated axons were medium-to-large calibre (mean 0.73 µm ± 0.1 s.e.m.) in size (Extended Data Fig. 6l). This result indicated that these axons underwent demyelination first, as medium-to-large diameter axons showed hypermyelination immediately before demyelination at 3-4 months of age (Extended Data Fig. 6k); therefore, hypermyelination may precede demyelination. These findings demonstrate that the lack of microglia is sufficient to induce CNS demyelination with increasing age.

Microglia maintain existing myelin
We next asked whether these changes in myelin growth and integrity reflect disruption of myelin formation or myelin maintenance. To that end, we depleted microglia in mice when developmental myelination is complete (2 months of age onwards) by providing the CSF1R inhibitor PLX5622 in the diet of adult Fire +/+ mice for 1 month. This resulted in >50% reduction of IBA1 + cells at 3 months of age (Extended Data Fig. 7a-c). Compared with mice fed the control diet, microglia depletion from 2 to 3 months of age resulted in enlarged inner tongues and thicker myelin (Extended Data Fig. 7d-g), whereas depletion from 5 to 6 months of age caused patchy demyelination (Extended Data Fig. 7j-l). Oligodendrocyte numbers were unchanged (Extended Data Fig. 7h,i). Therefore, microglia depletion in adult mice mirrored the hypermyelination and myelin degeneration observed at equivalent ages in Fire Δ/Δ mice, which indicated that microglia are required for myelin maintenance once it is already formed.

Microglia deficits in humans affect myelin health
After demonstrating that microglia are required for myelin health in mice, we investigated the relevance of these findings in humans. We analysed samples from individuals with the rare leukoencephalopathy adult-onset leukoencephalopathy with axonal spheroids and pigmented glia (ALSP) (Extended Data Table 1). In ALSP, heterozygous CSF1R mutations lead to cognitive dysfunction in association with reduced IBA1 + parenchymal cells, especially in frontal white matter, whereas those in the grey matter are relatively preserved 18 . In comparison to age-matched individuals who died of non-neurological causes, there was a significant decrease in IBA1 + microglia and macrophages in the frontal white matter of individuals with ALSP ( Fig. 4a,b). Moreover, there was a relative increase in the proportion of perivascular macrophages (IBA1 + LYVE1 + ) (Extended Data Fig. 8a-d). Ultrastructural analysis of ALSP white matter revealed myelin outfoldings and unravelling (Fig. 4c), thicker myelin (Fig. 4c,d) and enlarged inner tongues (Fig. 4c,e and Extended Data Fig. 8e-g). Demyelination was also observed and progressively worsened with age (Fig. 4f). Larger axon diameters were noted in ALSP samples compared with unaffected samples (Fig. 4d,e). This result is consistent with axonal swelling being a typical pathological feature of this disorder; however, myelin was still thicker than would be expected of these axon diameters. Extra thick myelin in ALSP was Article associated with myelin unravelling (Fig. 4c), which may indicate early stages of a transition from hypermyelination to demyelination. These findings show that a reduction in white matter microglia in humans is associated with hypermyelination and eventual demyelination.

Microglia suppress oligodendrocyte state
We next sought to determine the cellular and molecular mechanisms that underpin the loss of myelin health in the absence of microglia. To that end, we performed single-cell RNA sequencing of brain samples from Fire Δ/Δ and Fire +/+ mice at 1 month of age (Extended Data Fig. 9a-j). Mature oligodendrocytes were identified through the expression of myelin genes (Plp, Mog, Mag and Mbp) and the absence of expression of markers for other cell types (Extended Data Fig. 9a-d). The oligodendrocytes were clustered into four states (Oligo1 to Oligo4) ( Fig. 5a and Extended Data Fig. 9k,l). Notably, we observed that cluster 1 oligodendrocytes (Oligo1) were almost exclusively found in Fire Δ/Δ mice (Fig. 5b,c) and distinguished by the high expression of genes (Supplementary Table 1 and Fig. 5d) including Serpina3n and C4b (Fig. 5d,e). We confirmed the presence of SERPINA3N + OLIG2 + cells almost exclusively in Fire Δ/Δ mouse white matter, whereas these cells were undetectable in grey matter of either genotype (Fig. 5f,g). Analysis of differentially expressed genes in the Oligo1 cluster revealed that the top canonical pathways were related to lipid synthesis and metabolism. Specifically, ingenuity pathway analysis highlighted the following significant pathways: superpathway of cholesterol biosynthesis (P = 8 × 10 −12 ) and cholesterol biosynthesis I-III (P = 1.23 × 10 −7 ). Analysis using the DAVID bioinformatics resource highlighted the following pathways: lipid biosynthesis (P = 3.8 × 10 −8 ); lipid metabolism (P = 5.1-7.0 × 10 −6 ); and cholesterol metabolism (P = 9.2 × 10 −4 ). Of note, cholesterol is enriched in myelin and required for myelin growth 19 . Lipidomics analysis of Fire Δ/Δ mouse white matter revealed an increase in cholesterol esters and a decrease in triglycerides (Fig. 5h), which was indicative of excess cholesterol and impaired lipid export, respectively. This result is consistent with the observed surplus in myelin membrane formation in Fire Δ/Δ mice. Moreover, dysregulation of genes associated with cholesterol transport has been reported in ALSP 20 .

TGFβ1 signalling regulates myelin health
To determine how the absence of microglia contributes to these findings, we assessed the predicted upstream regulators of genes in the Oligo1 cluster. TGFβ1 was identified as a prime candidate (P = 7.7 × 10 −13 ; Supplementary Table 2), as it is predominantly expressed by microglia in both mouse and human brain (https://www. brainrnaseq.org) 21,22 , and it is known to influence lipid metabolism 23 . Accordingly, TGFβ1 levels were significantly downregulated in Fire Δ/Δ mouse white matter (Fig. 6a). We next assessed the capacity of oligodendrocytes to respond to TGFβ1. Although there were abundant TGFβR1 + OLIG2 + cells in Fire +/+ mice, the number and percentage of these cells were significantly reduced in Fire Δ/Δ mice ( Fig. 6b-d). These findings led us to ask whether elimination of TGFβR1 signalling in oligodendrocytes is sufficient to cause myelin pathology. As Tgfb1 knockout in the CNS results in a confounding decrease in microglia number, loss of microglia homeostasis and monocyte infiltration 24 , we utilized a conditional knockout of Tgfbr1 in mature oligodendrocytes (Plp creERT ;Tgfbr1 fl/fl ). Following tamoxifen administration from P14 to P18 (Fig. 6e), TGFβR1 expression by oligodendrocyte lineage cells was significantly reduced in Plp creERT ;Tgfbr1 fl/fl mice at 1 month of age compared with controls (Extended Data Fig. 10a,b). There was no significant impact on the number of myelinated axons (Extended Data Fig. 10c). However, by P28, conditional knockout mice had enlarged  inner tongues on smaller diameter axons ( Fig. 6f-h and Extended Data Fig. 10d) and thicker myelin on larger diameter axons ( Fig. 6f,g,i and Extended Data Fig. 10e) relative to tamoxifen-treated floxed and wild-type controls. This result mimics the observations in Fire Δ/Δ mice at 1 month of age. We next asked whether stimulating TGFβR1 signalling in Fire Δ/Δ mice could rescue myelin health. In the absence of sufficient TGFβR1 expression by oligodendrocytes in Fire Δ/Δ mice, we used a small-molecule activator of downstream TGFβ signalling, SRI-011381 hydrochloride. This approach bypasses the need to stimulate the receptor by activating the SMAD2-SMAD3 pathway 25 . We administered SRI-011381 hydrochloride to Fire Δ/Δ mice from 2 months of age to observe the potential impact on the significant myelin pathology observed by 3 months of age (Fig. 6j). SRI-011381 had no influence on myelinated axon number (Extended Data Fig. 10f) but significantly reduced inner tongue thickness ( Fig. 6k-m and Extended Data Fig. 10g) and myelin thickness (Fig. 6k,l,n and Extended Data Fig. 10h) compared with vehicle-treated Fire Δ/Δ mice. Following SRI-011381 hydrochloride treatment, these parameters overlapped with those in age-matched Fire +/+ mice (Fig. 6m,n). This result suggested that disrupted TGFβR1 signalling is the primary mechanism by which myelin is dysregulated in Fire Δ/Δ mice. Altogether, these findings reveal the importance of the TGFβ1-TGFβR1 axis in microglia-oligodendrocyte communication for the regulation of myelin health.

Discussion
Here we identified the requirement for microglia in the maintenance of healthy CNS myelin. Our use of a new transgenic model in which microglia are lacking while other CNS macrophages are present revealed that microglia are not required for developmental oligodendrocyte maturation or myelin ensheathment. Rather, microglia preserve the structural integrity of myelin. We demonstrated the role of microglia in limiting hypermyelination and preventing demyelination of existing myelin sheaths in adulthood. Our results complement recent work implicating microglia in the regulation of myelin sheath number during myelin formation in embryonic development 26 and the inhibition of myelination of regenerated axons after optic nerve injury 27 . This raises the question of whether other macrophage populations, such as perivascular and/or peripheral macrophages, influence developmental myelination or whether other glial cell types, such as astrocytes, have a compensatory role in the absence of microglia. Our work indicates that a threshold number of microglia is needed to maintain myelin health, as even just a 50-60% decrease in white matter microglia in mice or humans is associated with loss of myelin integrity. Altogether, these findings suggest prudence in the current trials of CSF1R inhibitors to deplete microglia in cancer  Fire +/+
We associated structural changes in myelin in the absence of microglia with poor cognitive flexibility, along with impaired de novo myelination that normally underpins long-term memory consolidation 17,28 . This builds on previous work revealing that microglia dysregulation (in response to chemotherapy) is sufficient to disrupt myelin structure and cognitive function 9,13 , as we have uncovered the requirement for healthy microglia in the prevention of these pathologies. Given the close relationship between myelin structure and neuronal activity 29 , our findings also raise the possibility of a role for microglia in influencing adaptive myelination to reinforce cognitive circuits. However, understanding the impact of the absence of microglia on neuronal activity and synaptic health requires further investigation. In addition, our work has important implications for understanding cellular networks that contribute to cognitive decline with ageing, in which there is prominent hypermyelination 3,12 , demyelination and impaired new myelination 5,28 , alongside microglia dysfunction 21,30,31 . This may also be relevant to dementia-associated neurodegenerative disease, given that in a mouse model of Alzheimer's disease, there are opposite changes in gene modules primarily associated with microglia (and astrocytes) compared with those related to oligodendrocytes and myelination 32 . That study

Myelin thickness (μm)
Myelin thickness (μm) Article also identified that the oligodendrocyte-associated module is initially upregulated followed by a downregulation in microenvironments with the highest β-amyloid accumulation. Whether this represents initial hypermyelination followed by demyelination remains to be determined. The worsening myelin pathology we observed with age in response to microglia depletion points to an increasing dependence on healthy microglia for myelin integrity with ageing. Microglia dysfunction may therefore initiate myelin damage with ageing and in neurodegenerative disorders. Previous work has implicated the peripheral immune system or primary oligodendrocyte dysregulation in inducing demyelination.
Here we propose that the contribution of microglia now also needs to be considered. Our data suggest that hypermyelination may precede demyelination, which raises the question of whether this sequence of events underpins myelin damage in ageing and neurodegenerative disease. Notably, we identified parallels in dysregulated cellular profiles and molecular mechanisms between microglia-deficient mice and other neurological injury models. These results implicate microglia as crucial regulators of myelin pathology in these contexts. For instance, our data show that microglia normally suppress the appearance of a dysregulated oligodendrocyte state (expressing Serpina3n and C4b), similar to that recently documented in mouse models of demyelination, ageing and Alzheimer's disease [33][34][35][36][37] . This implies that in pathological contexts, microglia dysregulation may permit these oligodendrocytes to appear. We have made data-mining of this (and other) oligodendrocyte populations in the Fire Δ/Δ mouse model publicly available on the following website: https://annawilliams.shinyapps.io/shinyApp_oli-gos_VM/. The functional consequences of the appearance of these oligodendrocytes, and the molecular mechanisms involved in their function, have hitherto been unclear. Our association of these oligodendrocytes with an altered lipid profile (for example, increased cholesterol esters) is consistent with the hypermyelination in Fire Δ/Δ mice. This may shed light on potential pathological mechanisms in human neurological conditions in which cholesterol esters are increased, such as Huntington's disease and multiple sclerosis 38,39 . The finding that microglia regulate the lipid profile of mature myelinating oligodendrocytes complements their recently discovered role in promoting oligodendrocyte progenitor maturation during myelin regeneration through the supply of a cholesterol pathway intermediate 40 . Our pathway analysis of these oligodendrocytes identified a dysregulated TGFβ-TGFβR1 axis, which we showed is a mediator of myelin pathology. Transcriptomic analyses of ALSP brain samples has indicated dysregulation of the TGFβ pathway specifically in the white matter 20 , and an upregulation of TGFB1 in the least affected white matter region 41 . Signalling downstream of TGFβ receptors is also reduced in ageing and neurodegenerative disease 42,43 .
We previously discovered that a subset of microglia expressing a TGFβ superfamily member, activin-A, regulates remyelination efficiency 44,45 . Taken together with the transcriptomic heterogeneity of microglia during development, homeostasis, demyelination, remyelination and ageing 30,44,46-50 , we are now poised to ask whether specific microglia states are required to regulate myelin growth and integrity. Recently, studies have identified microglia states associated with white matter, with roles in phagocytosis of dying cells in development 50 or myelin debris in ageing 30 , and we found that a shift in functional microglial states underpins their capacity to support remyelination 44,46 . Whether altered heterogeneity with ageing and disease is associated with a loss in supportive microglia states that then contributes to progressive myelin pathology needs to be investigated. Altogether, our study uncovered the role of microglia in preserving myelin health and integrity in adulthood, and highlights microglia as key therapeutic targets in the context of disrupted myelin in ageing and disease.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-05534-y. Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

Animals
All experiments were performed under project licences approved by the UK Home Office and issued under the Animals (Scientific Procedures) Act. This study used Csf1r FireΔ/Δ mice, wild-type controls from Csf1r FireΔ/+ crossings, Plp creERT mice ( Jackson Laboratories) and Tgfbr1 fl/fl mice (provided by S. Karlsson, Lund University). Recombination was induced in Plp creERT ;Tgfbr1 fl/fl mice through the administration of 4-hydroxytamoxifen (100 mg kg -1 , intraperitoneally; Sigma-Aldrich) dissolved in ethanol and corn oil (1:9) mixture for 5 consecutive days from P14 to P18, then killed at P28. All animals were housed at a maximum number of 6 animals per cage in a 12 h light-dark cycle with unrestricted access to food and water. For animal experiments, the sample size was determined by power analysis calculated by two-sided 95% confidence interval through the normal approximation method using OpenEpi software (Openepi.com), and reached >80% power for all experiments. Both males and females were used throughout the study, except for open-field experiments, for which only male mice were used. Animals were randomized to time points analysed. ARRIVE2 guidelines were followed in providing details of experiments, quantifications and reporting.

Genotyping
Genomic DNA was extracted from ear biopsy tissue using a Wizard SV genomic purification system (Promega) according to the manufacturer's instructions. Csf1r FireΔ/Δ mice were genotyped using PCR strategies as previously described 10 . Genotyping of Plp creERT ;Tgfbr1 fl/fl mice was performed with genomic DNA extracted from the tail, as previously described for Plp creERT mice 51 and Tgfbr1 fl/fl mice 52 .
Sections were imaged on a Leica SPE or Zeiss LSM 510 confocal microscope. Cell counts were calculated from a measured area based on assumption of circularity using Fiji/ImageJ (Fiji.sc), with three regions of interest quantified per section. Colocalization analysis of SOX9 and GFAP was performed using Imaris software v.9.7.

Electron microscopy of mouse tissue
Mice were intracardially perfused with 4% PFA (w/v) and 2% glutaraldehyde (v/v; TAAB Laboratories) in 0.1 M phosphate buffer. Tissue was post-fixed overnight at 4 °C and transferred to 1% glutaraldehyde (v/v) until embedding. Tissue sections (1 mm) were post-fixed in 1% osmium tetroxide and dehydrated before processing into araldite resin blocks. Next, 1 µm microtome-cut sections were stained with a 1% toluidine blue/2% sodium borate solution before bright-field imaging using a Zeiss Axio microscope. Ultrathin sections (60 nm) were cut from corpus callosum samples, stained with uranyl acetate and lead citrate, and grids imaged on a JEOL transmission electron microscope. Axon diameter, myelin and inner tongue thickness were calculated from a measured area based on assumption of circularity using Fiji/ ImageJ (Fiji.sc) (diameter = 2 × √[area/π]), with 100-200 axons per animal analysed.

Behavioural testing
Experimenters were blinded to genotype during behavioural testing and data analyses. All experiments were performed in a behaviour testing room maintained at a constant temperature of 20 °C. The open-field test was performed on male mice at 4-8 weeks and 11-13 months of age to assess locomotor activity and anxiety-associated behaviours. Handling was carried out 3-4 days before testing. Mice were placed in the open field (47 × 47 cm) to freely explore the arena for 10 min. Equipment was cleaned with 70% ethanol between each test to remove odours. The total ambulatory distance travelled (in metres) and the time spent in the edges (9 cm from the wall) and centre (29 × 29 cm) were automatically quantified using the video tracking software Any-Maze (Stoelting Europe, v.6.3). The Barnes maze test was performed in adult mice 2-4 months of age to assess spatial learning, memory and cognitive flexibility. The maze consisted of one white circular platform with 20 circular holes around the outside edge, 91.5 cm in diameter and 115 cm in height (San Diego Instruments). A dark escape chamber was attached to one of the holes, and the location of the escape chamber remained constant for each mouse but was shifted 90° clockwise between consecutive mice to avoid carryover of olfactory cues. Lamps and overhead lights (450 lux) were used to light the maze. Once the trial started, an aversive white noise stimulus at 85 dB was played until the mouse entered the escape chamber. Visual cues were present on the curtains and walls around the maze. Animals were retained within a white holding cylinder (diameter of 10.5 cm) at the beginning of each trial. The maze and escape chamber were cleaned with ethanol between each trial to avoid carryover of olfactory cues between animals. All trials were recorded using the video-based automated tracking software Any-Maze (Stoelting Europe, v.4.99). Before testing, mice were handled for 3-5 min per day for 6-7 days by the experimenter. Animals were brought into the testing room and placed in the holding cylinder to acclimate to the testing environment for 10 s for 2 days before habituation. Mice were habituated to the maze and escape chamber 1 day before the start of the learning phase, whereby each mouse was placed in the holding cylinder for 10 s then allowed to freely explore the maze with no aversive stimuli for 3 min. Mice were then guided to the escape chamber and retained inside for 1 min. During the learning phase (T1-T6), mice were trained to locate the escape chamber over 6 consecutive days with 2 trials per day (1 h inter-trial interval); data per mouse were averaged per day. If the mouse failed to enter the escape chamber during the 3-min trial period, the experimenter guided it to the chamber. Spatial learning was assessed by the total time taken to locate the escape chamber in each trial (primary latency; defined by the head entering the chamber), and spatial working memory was assessed by the number of errors made before locating the escape chamber (primary errors; defined by the nose deliberately entering a hole with some extension of the head and neck or hindpaws). The total distance travelled and speed during the trials were additionally measured. Exclusion criteria were defined before data analysis as follows: mice must enter a minimum of three quadrants of the maze within two of the first five trials and must enter the escape chamber during the first three trials. One wild-type and one knockout mouse were excluded from analysis owing to refusal to enter the chamber. At 1 h and 3 days following the last trial, probe tests were performed whereby the mouse was allowed to explore the maze for 1 min with the escape chamber removed. The time spent in the target quadrant of the maze and the number of nose pokes in each hole were recorded to assess memory of the escape chamber location. To assess cognitive flexibility, mice underwent the reversal learning phase (R1-R3), whereby the escape box was moved to 180° from the original location, and measurements were taken as described above. A probe test was also performed 3 days after the final reversal trial. The median age of mice assessed that were trained and untrained were comparable between genotypes at the time of euthanasia for immunofluorescence or electron microscopy analysis: untrained Fire +/+ and Fire Δ/Δ mice were 118 days old, trained Fire +/+ mice were 119 days old and trained Fire Δ/Δ mice were 120 days old.

EdU incorporation
EdU was dissolved in the drinking water at 0.2 mg ml -1 for a period of 14 days from the end of trial day 1 until the end of the experiment. The water was exchanged every other day, and intake was monitored to assess whether consistent volumes were consumed.

Microglia depletion in adulthood
The CSF1R inhibitor PLX5622 (Chemgood, C-1521) was formulated into chow at a concentration of 1,200 ppm (Research Diets, D11100404i) and fed to 2-month-old and 5-month-old wild-type (Fire +/+ ) mice for 1 month and euthanized as described above.

Measuring TGFβ1 levels by ELISA
Following perfusion with PBS, the corpus callosum was dissected from 2 mm coronal sections of Fire Δ/Δ and wild-type brains and snap-frozen in liquid nitrogen. Samples were homogenized in RIPA buffer (Millipore, 20-188) containing phosphatase and protease inhibitors (Sigma-Aldrich, 4906845001 and 11836170001). Corpus callosum lysate samples were activated and TGFβ1 protein levels were measured by ELISA (BioLegend, 436707) according to the manufacturer's instructions for serum and plasma samples at a final dilution of 1:10. BCA assays were performed to measure total protein according to the manufacturer's instructions (Thermo Fisher Scientific, 23225), and values were normalized to this for each sample.

Brain dissociation and cell sorting for single-cell RNA sequencing
Brains were collected from 6-7-week-old female mice at the same time of day for each animal. Mice were culled by cervical dislocation and brains were dissected, with the olfactory bulbs and cerebellum removed. Hippocampi from both hemispheres and the remainder of the left hemisphere (without the hippocampus) were collected in ice-cold HBSS (without Ca 2+ and Mg 2+ ; 14175-053, Gibco) with 5% trehalose (T0167, Sigma Aldrich) and 30 µM actinomycin D (A1410, Sigma Aldrich) and were finely minced using a 22A scalpel. Brains were digested using the an Adult Brain Dissociation kit (130-107-677, Miltenyi Biotec) with the following modifications: (1) tissues were dissociated as described in the "manual dissociation" section of the Neural Tissue Dissociation kit protocol (130-092-628, Miltenyi Biotec); (2) enzymatic digestions were performed at 35 °C; (3) half the concentration of enzyme P was used; (4) actinomycin D was used to limit dissociation-induced transcriptional changes; (5) 5% trehalose was added in all buffers to increase cellular viability; (6) cell clusters were removed by filtration through pre-moistened 70 µm (352350, Falcon) and 40 µm (352340; Falcon) cell strainers; (7) erythrocyte and myelin debris removal steps were omitted during dissociation steps; and (8) all centrifugations were performed at 200g at 4 °C. After dissociation, cells were collected in PBS with 0.2% BSA before being sorted on a Sony SH800 cell sorter. Gates were chosen based on forward and side scatter to exclude myelin debris, erythrocytes and doublets. Non-viable cells were excluded based on DRAQ7 and/or DAPI staining (DRAQ7 high and/or DAPI high cells were classified as non-viable). After confirming the viability of cells after sorting, using trypan blue and a haemocytometer, single cells were processed through the Chromium Single Cell Platform using a Chromium Next GEM Single Cell 3′ GEM Library and Gel Bead kit (v.3.1 chemistry, PN-1000121, 10x genomics) and a Chromium Next GEM Chip G kit (PN-1000120) and processed following the manufacturer's instructions. Libraries were sequenced using a NovaSeq 6000 sequencing system (PE150 (HiSeq), Illumina).
Pre-processing of sequencing data and single-cell RNA sequencing analysis Alignment to the reference genome, feature counting and cell calling were performed following the 10x Genomics CellRanger (v.5.0.0) pipeline, using the default mm10 genome supplied by 10x Genomics (https://cf.10xgenomics.com/supp/cell-exp/refdata-gex-mm10-2020-A.tar.gz). From the output, the filtered matrices were used for downstream analyses. Pre-processing was performed on the University of Edinburgh's compute cluster Eddie. The analysis was performed with R v.4.1.1. Full details to replicate the analysis pipelines described below can be found in code scripts available on GitHub (https://github. com/Anna-Williams/Veronique-Firemice). SingleCellEXperiment v.1.14.1 was used to handle the single-cell experiment objects in R. Cells were filtered using dataset-specific parameters on the basis of genes and unique molecular identifiers (UMIs) per cell, the ratio between these two parameters and the percentage of mitochondrial gene reads per cell. Thresholds were computed with the isOutlier function from scater (v.1.20.1), as batch 6 was of poorer quality than the other batches, with outlier values, the subset argument was used. Only genes that were detected in at least two cells were kept. Using scran (v.1.20.1), the data were normalized by deconvolution, and the top 15% highly variable genes were selected. Following principal component analysis (PCA), 25 principal components (PCs) were kept for downstream analysis (cut-off selected by examination of an Elbowplot). Nonlinear dimensional dimension representation and t-distributed stochastic neighbour embedding (t-SNE) and gene expression variance explained by batch (computed with scater) revealed the need for batch correction. Batch correction was performed using mutual nearest neighbours with fastMNN, batchelor (v.1.8.0). Finally, a graph-based clustering approach was used to cluster the cells using the clusterCells function from scran, with k = 60. Clusters with the highest expression of oligodendrocyte markers (Plp1, Mog, Mag and Mbp) and that did not express other cell type markers (for example, astrocyte, oligodendrocyte progenitor cell or microglia markers) were subsetted to be analysed separately. Cell and gene quality control were further adjusted, setting a stricter minimum UMI count threshold (5,000 UMIs) and maximum percentage of mitochondrial gene reads per cell (10%). A small cluster of cells of lower quality based on the new thresholds was also excluded from the analysis (Extended Data Fig.  10). Ultimately, we included a total of 19,506 genes and 13,583 cells. The normalization, feature selection, dimensional reduction and batch correction were repeated with the subset dataset as described above. Clustering was performed at different resolutions, and after examination with clustree (v.0.4.4), k = 100 was selected and then merged into four clusters ( Fig. 6a and Extended Data Fig. 10). Differential gene expression between cluster 1 (specific to the Fire Δ/Δ mice) and the mean expression of all other cells was performed using FindMarkers from Seurat (v. 4

Immunofluorescence staining of human tissue
Post-mortem tissue from individuals with ALSP and from unaffected individuals who had died of non-neurological causes (Extended Data Table 1) were obtained with full ethical approval from the Queen Square Brain Bank for Neurological Disorders, UCL Queen Square Institute of Neurology, and their use was in accord with the terms of the informed consents for use of donor tissue and information. Ethical approval was granted to the Queen Square Brain Bank for the use of tissue by the National Health Service Health Research Authority through the London Central Research Ethics Committee. Diagnosis of ALSP was confirmed on the basis of mutations in CSF1R and by neuropathological means, and clinical history was provided by Z. Jaunmuktane (University College London). Formalin-fixed paraffin embedded tissue blocks were cut to 10 µm thickness. Sections were placed in the oven at 60 °C for 10 min and deparaffinized by a series of washes in HistoClear (2× 10 min) and ethanol (100% twice, 95%, 70%, 50%; 5 min each). Following washes in PBS, slides were placed in a pressure cooker in Vector unmasking solution for 20 min, cooled, washed once and blocked for 1 h with 5% normal horse serum (Gibco) and 0.3% Triton X-100 (Fisher Scientific) in PBS. Tissue was incubated with primary antibodies in a humid chamber overnight. Sections were then washed in PBS, and fluorescently conjugated secondary antibodies were applied for 2 h at room temperature in a humid chamber (1:500, Life Technologies-Molecular Probes). Following washes in PBS and then water, the sections were counterstained with Hoechst and washed with distilled water for 20 min. Slides were then coverslipped with Fluoromount-G (Southern Biotech). Primary antibodies included IBA1 (Abcam, 1:500; ab5076) and LYVE1 (Abcam, 1:100; ab14917). Entire tissue sections were imaged using a Zeiss Axi-oScan Z.1 SlideScanner and Zeiss Zen2 software (blue edition). Three fields of 150 × 150 µm were counted per region of interest in each case and counts were multiplied to determine the density of immunopositive cells per mm 2 .
Electron microscopy of human tissue ALSP tissue was acquired from the Department of Neuropathology at Charité-Universitätsmedizin Berlin, and tissue from unaffected individuals were obtained from the Medical Research Council Edinburgh Brain and Tissue Bank, approved by the respective ethical review boards. ALSP tissue samples of the white matter were fixed in 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer for 48 h at 4 °C. Samples were post-fixed in 1% osmium tetroxide in 0.05 M sodium cacodylate buffer for 3 h, dehydrated in graded acetone series including en bloc staining with 1% uranyl acetate and 0.1% phosphotungstic acid in the 70% acetone step for 60 min and embedded in araldite resin. For tissue from unaffected individuals, ultrathin sections were stained with uranyl acetate and lead citrate and imaged with a Zeiss P902 electron microscope. For ALSP tissue, ultrathin sections with virtual absence of limiting artefacts were prepared and entirely digitized using a Zeiss Gemini 300 scanning electron microscope with a scanning transmission electron microscopy. In brief, we used 29 kV acceleration voltage, 5 nm pixel size and 1.5 µs beam dwell time for digitization and Fiji/TrakEM2 for stitching to allow for in-depth analysis with QuPath 0.3.0.

Statistics and reproducibility
All manual cell counts were performed in a blinded manner. Data are presented as the mean ± s.e.m. All micrographs are representative images of the result, which were independently repeated a minimum of three times; exact n values are presented in the corresponding graphs of quantification. Before statistical testing, data were assessed for normality of distribution using Shapiro-Wilk test. Statistical tests included two-tailed Student's t-test for normally distributed data or Mann-Whitney test for nonparametric data, two-way analysis of variance (ANOVA) with Sidak's multiple comparisons test for comparing more than two groups, and a one-sample t-test for comparison of log 2 (fold change) to a value of 0. For proportion graphs, one-way ANOVA with Tukey's multiple comparisons test was used. For Barnes maze analysis of primary latency and errors, repeated measures two-way ANOVA with Sidak's multiple comparisons test was used. Before testing for inter-group differences in the probe test data, each genotype was tested against chance (25%) using a one sample t-test. Slopes of myelin and inner tongue thickness versus axon diameter were compared using simple linear regression analysis, and for comparison of more than two groups, Kruskal-Wallis with Dunn's multiple comparisons test. A P value of <0.05 was considered significant. Data handling and statistical processing were performed using Microsoft Excel and GraphPad Prism Software.

Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Data availability
Raw single-cell RNA sequence datasets have been deposited in the Gene Expression Omnibus with accession code GSE215440. Analysed oligodendrocyte sequencing data are perusable on the following shiny application: https://annawilliams.shinyapps.io/shinyApp_oligos_VM/. Full details to replicate the analysis pipelines can be found in code scripts available on GitHub (https://github.com/Anna-Williams/ Veronique-Firemice). Alignment to the reference genome, feature counting and cell calling were performed following the 10x Genomics CellRanger (v.5.0.0) pipeline, using the default mm10 genome supplied by 10x Genomics (https://cf.10xgenomics.com/supp/cell-exp/ refdata-gex-mm10-2020-A.tar.gz). Source data are provided with this paper.

Code availability
Full details to replicate the analysis pipelines described below can be found in code scripts available on GitHub (https://github.com/ Anna-Williams/Veronique-Firemice).