Myelin dysfunction drives amyloid-β deposition in models of Alzheimer’s disease

The incidence of Alzheimer’s disease (AD), the leading cause of dementia, increases rapidly with age, but why age constitutes the main risk factor is still poorly understood. Brain ageing affects oligodendrocytes and the structural integrity of myelin sheaths1, the latter of which is associated with secondary neuroinflammation2,3. As oligodendrocytes support axonal energy metabolism and neuronal health4–7, we hypothesized that loss of myelin integrity could be an upstream risk factor for neuronal amyloid-β (Aβ) deposition, the central neuropathological hallmark of AD. Here we identify genetic pathways of myelin dysfunction and demyelinating injuries as potent drivers of amyloid deposition in mouse models of AD. Mechanistically, myelin dysfunction causes the accumulation of the Aβ-producing machinery within axonal swellings and increases the cleavage of cortical amyloid precursor protein. Suprisingly, AD mice with dysfunctional myelin lack plaque-corralling microglia despite an overall increase in their numbers. Bulk and single-cell transcriptomics of AD mouse models with myelin defects show that there is a concomitant induction of highly similar but distinct disease-associated microglia signatures specific to myelin damage and amyloid plaques, respectively. Despite successful induction, amyloid disease-associated microglia (DAM) that usually clear amyloid plaques are apparently distracted to nearby myelin damage. Our data suggest a working model whereby age-dependent structural defects of myelin promote Aβ plaque formation directly and indirectly and are therefore an upstream AD risk factor. Improving oligodendrocyte health and myelin integrity could be a promising target to delay development and slow progression of AD.

The incidence of Alzheimer's disease (AD), the leading cause of dementia, increases rapidly with age, but why age constitutes the main risk factor is still poorly understood. Brain ageing affects oligodendrocytes and the structural integrity of myelin sheaths 1 , the latter of which is associated with secondary neuroinflammation 2,3 . As oligodendrocytes support axonal energy metabolism and neuronal health 4-7 , we hypothesized that loss of myelin integrity could be an upstream risk factor for neuronal amyloid-β (Aβ) deposition, the central neuropathological hallmark of AD. Here we identify genetic pathways of myelin dysfunction and demyelinating injuries as potent drivers of amyloid deposition in mouse models of AD. Mechanistically, myelin dysfunction causes the accumulation of the Aβ-producing machinery within axonal swellings and increases the cleavage of cortical amyloid precursor protein. Suprisingly, AD mice with dysfunctional myelin lack plaque-corralling microglia despite an overall increase in their numbers. Bulk and single-cell transcriptomics of AD mouse models with myelin defects show that there is a concomitant induction of highly similar but distinct disease-associated microglia signatures specific to myelin damage and amyloid plaques, respectively. Despite successful induction, amyloid disease-associated microglia (DAM) that usually clear amyloid plaques are apparently distracted to nearby myelin damage. Our data suggest a working model whereby age-dependent structural defects of myelin promote Aβ plaque formation directly and indirectly and are therefore an upstream AD risk factor. Improving oligodendrocyte health and myelin integrity could be a promising target to delay development and slow progression of AD.
The pathology of AD is characterized by the deposition of Aβ plaques and neurofibrillary tangles primarily in the neocortex and hippocampus. According to the amyloid hypothesis of AD, the build-up of Aβ initiates a cascade of harmful events that lead to neuronal dysfunction. More than lifestyle choices and genetic predisposition, old age is the primary risk factor for AD development, but exactly how brain ageing is linked to amyloid deposition is unclear. Myelin, a spirally wrapped and compacted glial membrane, enhances axonal conduction speed, and its non-compacted regions allow oligodendrocytes to provide metabolic support to the neuronal compartment 7 . The specific cellular architecture of myelin makes protein and lipid turnover challenging and slow 8,9 . This, together with the long lifetime of oligodendrocytes 10 , might explain the structural deterioration of myelin with age 1 . We speculated that the breakdown of myelin integrity in the ageing brain acts as a driving force for Aβ deposition. We tested this hypothesis in mouse models of amyloidosis, in which we used genetic and pharmacological manipulation to introduce various degrees of myelin dysfunction.

Myelin decline in patients with AD
2 | Nature | www.nature.com

Article
Microscopy-based evidence of this damage, however, remains scarce. We therefore used immunofluorescence to analyse intracortical myelin integrity in a small group of patients with AD, focusing on the trans-entorhinal area ( Fig. 1 and Extended Data Fig. 1a,b). We observed a decline in intracortical myelin density in autopsy samples from patients with AD that were not limited to the immediate vicinity of plaques. Sites of myelin loss were also associated with increased numbers of IBA1 + microglia (Extended Data Fig. 1a,b). Correlations in human neuropathology cannot determine whether myelin loss is the cause or consequence of neuronal AD pathology such as axonal decay. To clarify this issue, we used two different experimental animal models of AD (5×FAD and APP NLGF ; Extended Data Fig. 1c) to investigate whether loss of myelin can act as an upstream driver of amyloidosis. Specifically, we combined the AD mouse models with mice that develop subtle myelin disintegration that is induced using genetic methods (Extended Data Fig. 1c). Mice with loss of function (null mutants) of the myelin architectural proteins CNP and PLP display minor structural myelin defects: lack of CNP causes the collapse of cytosolic channels within the myelin sheath 15 associated with axonal swellings 16 . PLP-deficient myelin lacks physical stability, impairs axonal energy metabolism 17 and causes axonopathy at higher age 4 . We first compared normal brain ageing of C57BL/6 wild-type (WT) mice with that of Cnp −/− mice and Plp −/y mice on the same background. Immunostaining showed that Cnp −/− and Plp −/y mice have progressive gliosis in grey and white matter between the age of 3 and 6 months. The progressive gliosis was qualitatively similar to aged WT mice (24 months old) but more substantial (Extended Data Fig. 1d,e and Supplementary Fig. 2). Similar to patients with AD, the aged WT mice and more so the myelin-defective mutant mice displayed significant reductions in intracortical myelin content, especially in the upper cortical layers (Extended Data Fig. 1d,e).

Myelin defects exacerbate Aβ deposition
To determine whether such ageing-associated myelin defects can drive amyloid deposition, we crossbred Cnp −/− and Plp −/y mutants with 5×FAD mice and analysed plaque burden in the resulting offspring (Extended Data Fig. 1c). For this analysis, we optimized a Congo red in toto plaque staining and clearing protocol based on the iDISCO technique for light sheet microscopy (LSM) (Supplementary Fig. 3) to determine the amyloid plaque load in the entire brain in an unbiased fashion. When compared with 5×FAD mice, both Cnp −/− ;5×FAD mice and Plp −/y ;5×FAD mice showed marked increases in amyloid plaque load in hippocampal white matter (alveus) and cortex at 6 months of age ( Fig. 2a-d and Extended Data Fig. 2a-d). In both mouse models, the effects were strongest in the alveus, a hippocampal white matter tract: here Aβ was deposited in very small aggregates, which indicated an increased formation of amyloid seeds. Increases in plaque load were observed at 6 months but not yet at 3 months of age (Extended Data Fig. 2e,f) when Cnp −/− mice and Plp −/y mice exhibited only subtle myelin defects and reactive gliosis (Extended Data Fig. 1d). Moreover, 5×FAD control mice and the myelin mutant mice did not show plaque pathology in the alveus at this age (Extended Data Fig. 2g). To exclude the possibility that these effects are specific to the 5×FAD model and App overexpression, we validated our findings in crossbreedings of Cnp −/− mice and the APP NLGF knock-in model, in which three mutations in the human gene encoding amyloid precursor protein (APP) is expressed under the control of the endogenous App locus (Extended Nature | www.nature.com | 3 Data Fig. 2h). We also tested whether aged myelin mutant animals per se (without the human APP transgene) could naturally develop amyloid plaques in their lifetime. No amyloid deposits, however, were detected in 14-month old Plp −/y mice or in 22-month-old long-lived mice in which Plp was conditionally knocked out in the forebrain (Emx cre Plp fl/fl mice) (Extended Data Fig. 2i,j). These results are in line with studies showing that rodent amyloid is relatively resistant to aggregation 18 and that mice overexpressing rodent Aβ do not develop amyloidosis in their lifetime 19 . Secondary myelin alterations have been described in mouse models of AD 20,21 . However, our immunostaining, western blot and electron microscopy (EM) analyses did not detect substantial myelin pathology in 6-month-old 5×FAD mice or APP NLGF mice (Extended Data Fig. 3). Similarly, the subtle myelin pathology that occurs in Cnp −/− mice and Plp −/y mice was not modified by crossbreeding to the 5×FAD genotype (Extended Data Fig. 3i-l). The observed plaque-promoting effects are therefore not simply an altered secondary response of oligodendrocytes or myelin to amyloid deposits.
We also investigated whether myelin dysfunction modifies 5×FAD behavioural deficits by performing behavioural testing in the Y-maze and the elevated plus maze (Extended Data Fig. 3m-p). In both types of mazes, myelin mutant mice presented with hyperactivity that was further exacerbated in Cnp −/− ;5×FAD mice. In the elevated plus maze, Cnp −/− 5×FAD mice preferred the open arms, which is indicative of an unusual lack of anxiety. Statistical analyses confirmed supra-additive effects (interactions) of 5×FAD and especially the Cnp −/− genotype. Myelin defects and amyloid pathology seem to synergistically worsen behavioural deficits indicative of disinhibition, which is a neuropsychiatric symptom that is also present in patients with AD 22 .

Acute demyelination drives Aβ deposition
To confirm that myelin dysfunction is an upstream driver of plaque pathology, we tested the effect of acute demyelination. Young adult 5×FAD mice were fed a cuprizone-containing diet for 4 weeks followed by a 4-week recovery period, after which the plaque load was determined by LSM. This method leads to substantial demyelination in the medial corpus callosum and the alveus accompanied by microgliosis followed by complete remyelination after 4 weeks (Extended Data Fig. 4a,b). Interpreting the outcome of this experiment was complicated, however, because the copper-chelating properties of cuprizone interfere with plaque core formation, which depends on copper 23 . Indeed, the in toto LSM results showed that cuprizone treatment seemed to ameliorate hippocampal and cortical Aβ pathology (Extended Data Fig. 5c). However, the amyloidosis-driving effect of demyelination predominated the inhibition by copper chelation in substantially demyelinated areas, such as the hippocampal alveus. In this brain region, less-compacted Aβ plaques in cuprizone-treated 5×FAD mice were well stained with anti-Aβ antibodies, which revealed a large increase in small amyloid aggregates (Fig. 2e).
Independent evidence to support our working model was provided by results from 5×FAD mice in which experimental autoimmune encephalomyelitis (EAE) was induced. We immunized young 5×FAD animals with myelin oligodendrocyte glycoprotein (MOG) peptide and analysed  Article their brains and spinal cords 4 weeks later ( Fig. 2f and Extended Data Fig. 4d). Results from a previous study 24 of aged J20 and Tg2576 mouse models of AD showed that EAE reportedly reduced brain plaque load. However, we did not find a difference in amyloid deposition in young 5×FAD mice with EAE (Extended Data Fig. 4e) due to the lack of demyelination pathology in the brain (Extended Data Fig. 4f-h). By contrast, in the spinal cord, where demyelinating EAE pathology substantially develops (Extended Data Fig. 4i), we identified small, atypical amyloid aggregates in the peri-lesion environment, which were absent from control 5×FAD mice (Fig. 2f). We verified the presence of aggregated amyloid by staining spinal cord sections with the β-sheet dye methoxy-04 (Me-04) (Fig. 2f). Of note, in spinal cord from WT animals with EAE, no such Me-04 + material was found, which rules out the possibility of nonspecific detection of lipid deposits in demyelinated lesions (Fig. 2f). Spinal cord grey matter plaque deposition, as seen later in 5×FAD mice with disease progression, could not be detected in 14-week-old 5×FAD mice or in EAE 5×FAD mice (Extended Data Fig. 4i,j). Taken together, myelin defects-both chronic and acute-drive amyloid deposition in mouse models of AD, which identifies dysfunctional myelin as an upstream risk factor for amyloid deposition.

Lack of myelin ameliorates Aβ deposition
Compared to other apes, humans show a disproportional enlargement of prefrontal white matter 25 . This raises the question of whether the extent of cortical myelination per se could have a causal role in human AD. We therefore wondered what the impact of the near complete absence of cortical myelin would have on the course of amyloidosis in 5×FAD mice. To this end, we generated a line of forebrain-specific shiverer mice in which cortical axons are largely unmyelinated (Emx cre Mbp fl/fl ; forebrain shiverer mice) and crossbred them to 5×FAD mice (Extended Data Fig. 5a). At 3 months of age, forebrain shiverer;5×FAD mice were strongly protected against amyloid deposition in both the hippocampus and the cortex (Extended Data Fig. 5b-d). However, at 6 months of age, this effect was largely lost, which indicated that there is a delay in plaque formation in the absence of myelin (Extended Data Fig. 5e). At 6 months of age, plaque burden in the alveus was increased in forebrain shiverer;5×FAD mice, quantitatively similar to Cnp −/− and Plp −/y crossbred mice. Qualitatively, however, alveus plaques in forebrain shiverer;5×FAD mice differed in their morphology and distribution (Extended Data Fig. 5f), which pointed towards a different mechanism of increased plaque burden in these models. We conclude that proper myelin ensheathment and healthy myelin have an initially inhibitory effect on plaque formation.

Myelin dysfunction affects APP metabolism
We next investigated how myelin defects mechanistically drive amyloidosis. In theory, the defects could either promote APP processing and Aβ generation or interfere with Aβ removal (or a combination of both). We first investigated APP metabolism in Cnp −/− ;5×FAD mice. Axonal swellings that stain positive for APP are prominent features of ischaemia, brain injury and myelin disorders, and we speculated that these swellings contribute to the generation of Aβ in myelin mutant animals. Stalling axonal transport can enhance amyloid production, probably by increasing BACE1 and APP interactions in axonally transported vesicles [26][27][28] , and axons seem to be important sites of amyloid secretion [29][30][31] . In Cnp −/− mice and Cnp −/− ;5×FAD crossbred mice, numerous myelin damage-associated swellings of axonal origin were observed (Extended Data Fig. 6a-d). In 5×FAD control mice, almost no such swellings were found and axonal injury was limited to the immediate vicinity of the plaque and arranged in a circular pattern (termed corona) (Extended Data Fig. 6a-d). High-pressure freezing EM analyses showed that myelin-damage-associated swellings were enriched in vesicular structures that were probably endosomal or lysosomal in origin (Fig. 3a), which are thought to be primary production sites of Aβ 32,33 . Using antibodies against APP processing enzymes and multiple APP-and Aβ-specific antibodies (Fig. 3b), we found that axonal swellings in the brains of Cnp −/− ;5×FAD mice accumulated β-secretase and γ-secretase, co-stain for APP and BACE1 and consequently stain positive for β-and γ-cleaved APP fragments and Aβ (Fig. 3c-e and Extended Data Fig. 6e-g). Immunostaining also showed increased levels of soluble APPβ in cortical tissue in Cnp −/− ;5×FAD mice. In this brain region, similar to white matter, numerous myelin-damage-associated swellings were observed in addition to plaque-associated swellings (Extended Data Fig. 6h). Similar observations were made in Plp −/y ;5×FAD mice, which were used as an independent model of myelin dysfunction (Extended Data Fig. 6i-k). We further corroborated these findings by western blot analyses of white matter and the cortex, which showed increased levels of BACE1 (which was significant in white matter) ( Fig. 3f and Extended Data Fig. 6l) in 6-month-old Cnp −/− ;5×FAD mice. When we investigated APP processing by immunoblotting, cortical but not white matter APP metabolism was shifted towards an increased abundance of APP carboxy-terminal fragments (CTFs), but without changes to full-length APP abundance or the α/β CTF ratio ( Fig. 3g and Extended Data Fig. 6m-o). It is tempting to speculate that axonally generated CTFs in the white matter could be retrogradely transported and accumulate over time in neuronal cell bodies of the cortex, as western blotting showed that these areas are enriched in CTFs. Together, these findings suggest that myelin defects affect APP metabolism and Aβ generation.

Myelin defects alter microglia responses
Glial cells play important parts in the clearance of myelin debris and amyloid peptides, and microglia form barriers around amyloid plaques 34 . Thus, we investigated in vitro and in vivo how these cells react to amyloid when additionally challenged with defective myelin. Prior exposure of bone-marrow-derived macrophages-a commonly used in vitro model to study AD-related phagocyte function-to myelin debris strongly inhibited amyloid phagocytosis (Extended Data Fig. 7a,b). In vivo, cortical microglia failed to cluster around amyloid plaques in Cnp −/− ;5×FAD mice and Plp −/y 5×FAD mice despite substantial increases in microglia numbers in both white and grey matter in these models ( Fig. 4a-d and Extended Data Fig. 7c-k). Notably, we observed the same effect in Cnp −/− ;APP NLGF crossbred mice (Extended Data Fig. 7h). An equivalent analysis of this phenotype was not possible in the primary demyelination models owing to substantial microglia and macrophage infiltration and because the primary effects in these models were not observed in cortical tissue (Extended Data Fig. 7l,m).
Similar defects in microglial plaque-corralling have been previously described in Trem2 loss-of-function scenarios 34-36 . We therefore investigated whether myelin dysfunction interferes with the upregulation of Trem2 and the induction of the disease-associated microglia (DAM) signature that is associated with a plaque-corralling phenotype 37 . To this end, we performed magnetic activated cell sorting (MACS) of microglia and bulk RNA sequencing (RNA-seq) ( previously reported DAM signature genes (Clec7a, Gpnmb, Apoe, Spp1, Axl and Itgax) and genes of the Ms4a cluster as highly upregulated in Cnp −/− ;5×FAD mice (Extended Data Fig. 8f-h). Such a signature was previously described to be a feature of highly phagocytic microglia in early development and for brain border macrophages 38 . Trem2 and Tyrobp, however, showed unaltered induction levels (Fig. 4f). As Ms4a genes can regulate the level of soluble TREM2 and modify AD risk 39 , we used immunoblot analysis to rule out the possibility that myelin dysfunction affects TREM2 protein levels or cleavage through increased Ms4a expression (Extended Data Fig. 9a). Functional enrichment   Each violin plot represents a single animal/biological replicate/n. n = 5 per group. Black lines represent medians. In total, 2,017 individual cortical plaques were analysed in 5×FAD brain slices and 2,190 in Cnp −/− ;5×FAD brain slices. Middle, graph shows distribution of medians. Lines indicate means. Statistical analysis: two-sided, unpaired Student's t-test (P values are indicated on the graphs) on biological replicate data. Right, two representative plaques for each genotype. c, Experimental setup for microglia bulk RNA-seq. Microglia were isolated from hemispheres of 6-month-old WT, Cnp −/− , 5×FAD and Cnp −/− ;5×FAD animals and subjected to RNA-seq (biological replicates, n = 4 for each genotype). d, PCA was used for evaluating relative distances between normalized RNA transcripts per kilobase million (TPM) profiles. Dots represent biological replicates. Principal component 1 (PC1) explained 80.1% of data variability and strongly reflected Cnp −/− -dominated microglia transcriptome changes. e, Heatmap of top genes contributing to PC1 variability. f, Normalized expression level for selected genes (homeostatic marker P2ry12, DAM signature Trem2, Tyrobp, Axl and Clec7a, and differential regulated genes in Cnp −/− ;5×FAD mice (Apoc1, ApoE, Ms4a7 and Mmp12)). Bars represent means, dots represent mice/ biological replicates/n (n = 4 per group). g, ApoE enrichment in microglia in white matter from Cnp −/− ;5×FAD mice. Immunofluorescence staining of ApoE and IBA1 to mark microglia. The same observations were made in three independent samples per mice per group. h, Left, microscopy analysis of plaque-associated ApoE in cortical plaques in 5×FAD mice and Cnp −/− ;5xFAD mice pseudo-coloured according to a rainbow lookup table. Right, violin plots of mean ApoE fluorescence intensity per plaque. Each violin plot represents a single animal. Black lines indicate medians. Statistical analysis: unpaired, two-tailed Student's t-test on biological replicate data. n = 3 per genotype, 703 plaques for 5×FAD, 846 plaques for Cnp −/− ;5×FAD in total.
To better understand the DAM-like microglial signature induced by myelin defects, we further performed single-nuclei RNA-seq (snRNA-seq) on brain tissue from Cnp −/− ;5×FAD mice and respective controls ( Fig. 5a and Extended Data Fig. 9b-d). Cluster analysis identified five major microglia and macrophage subpopulations, including two distinct clusters (clusters 4 and 5) with high expression of DAM marker genes 37 (Trem2, Lpl and Spp1) (Fig. 5b-d). Cluster 4 was almost exclusively derived from cells with the 5×FAD genotype, whereas cluster 5 consisted of cells from the Cnp −/− background (Fig. 5e). We therefore termed these two clusters myelin-DAM and amyloid-DAM, respectively. Differential expression analysis between myelin-DAM and amyloid-DAM revealed upregulation of lipid-metabolism-related genes (Apoe, Abca1 and Apobec1) and genes of the Ms4a cluster in myelin-DAM, which corroborated findings from our bulk RNA-seq experiment ( Fig. 5f and Extended Data Fig. 9e). Genes specifically upregulated in amyloid-DAM included Cst3, Ctnna3 (which encodes alpha-T-catenin) and Gpc5 (which encodes the heparan sulfate proteoglycan glypican 5). Despite a low detection rate, Cst7 was also specifically enriched in amyloid-DAM (P values are indicated in the graphs) on biological replicate data. j, Scheme illustrating how myelin dysfunction acts as an AD risk factor. Upstream myelin defects cause microglia engagement and axonal transport problems. The two downstream pathologies are probably interrelated (dashed arrow). Axonal problems lead to endosome and lysosome accumulation and enhanced amyloid production. Simultaneously, microglia become increasingly engaged with defective myelin, which reduces their interaction with amyloid plaques. Both processes contribute to the enhanced deposition of amyloid. (Fig. 5g). Notably, when comparing microglia from Cnp −/− ;5×FAD mice with 5×FAD mice within the amyloid-DAM cluster, we still observed upregulation of Apoe in Cnp −/− ;5×FAD microglia (Extended Data Fig. 9g,h). Analyses of DEGs shared between DAM clusters and genotype comparisons within DAM clusters revealed that both amyloid-DAM and myelin-DAM of the Cnp −/− ;5×FAD crossbred mice retained signature features of the other DAM cluster, which highlights the intermediate phenotype of microglia in this mouse model (Extended Data Fig. 10h,i).

Article
Overall, our single-nucleus transcriptome analysis indicated that microglia respond in a similar but distinct manner to myelin dysfunction and the amyloid plaques they encounter. In the Cnp −/− ;5×FAD model, in which the two pathologies co-exist, both microglia states are adequately induced (despite modifications to the signatures). Our imaging data, by contrast, showed an almost complete loss of plaque-corralling microglia, which were previously identified as the in situ equivalent to DAM 37 . We wondered whether myelin damage would engage a large fraction of microglia in myelin clearance, thereby distracting even successfully induced amyloid-DAM away from plaques. Immunostaining for three markers showed that a large number of microglia and plaque-diverted microglia in Cnp −/− ;5×FAD mice seemed to be reactively engaged in myelin phagocytosis ( Fig. 5h and Extended Data Fig. 9j). Cst7 in situ hybridization-a marker for amyloid-DAM-showed that amyloid-DAM in Cnp −/− ;5×FAD mice were frequently found distal to the plaque, which was in contrast to 5×FAD control mice (Fig. 5i). We propose that such distraction of microglia leads to a faster build-up of amyloid in the brain. Moreover, factors secreted by activated but myelin-engaged microglia (such as APOE) may further fuel plaque seeding.
We next investigated whether altered microglia phenotypes also contribute to amelioration of plaque burden in forebrain shiverer;5×FAD mice (which lack myelin in the forebrain). To this end, we performed snRNA-seq analysis of microglia states in forebrain shiverer mice and contrasted those with the microglia signature found in Cnp −/− animals at 3 months of age (Extended Data Fig. 10a). In forebrain shiverer mice, we failed to detect activated microglia or myelin-DAM compared with microglia from Cnp −/− mice (Extended Data Fig. 10b-d). Similarly, immunostaining showed that microglia from forebrain shiverer;5×FAD mice remained in an homeostatic state and displayed a highly ramified morphology. However, we detected an increase in microglia cell numbers (Extended Data Fig. 10e). This could convey a certain protection against plaque build-up early in disease as more microglia could potentially clear amyloid. Also, contrary to mice with dysfunctional myelin, the lack of myelin in forebrain shiverer;5×FAD mice did not lead to axonal swellings or alterations in APP metabolism (Extended Data Fig. 10f,g).
Our findings and experimental models (myelin defects on top of an AD background) are highly relevant to the human disease as they mimic the state of the old 'comorbid' brain. That is, microglia would see multiple pathological stimuli simultaneously-such as age-related myelin breakdown and amyloid-similar to microglia in Cnp −/− 5×FAD mice. A re-analysis of published human AD snRNA-seq datasets showed that activated microglia-in contrast to pure amyloid-DAM in 5×FAD miceshowed upregulation of MS4A cluster genes, thereby demonstrating a myelin clearance-related signature (Extended Data Fig. 10h-o).

Ageing myelin as a risk factor for AD
Only a few studies have experimentally investigated oligodendrocytes and myelin in amyloid mouse models of AD 20,21,[41][42][43] . Nearly all have focused on the effect of amyloid plaques and Aβ on oligodendrocytes and myelin, partially describing secondary demyelination or hypomyelination. Similarly, single cell RNA-seq analysis of mouse models of AD demonstrated signature alterations in oligodendrocytes, specifically nearby amyloid plaques 44,45 . RNA-seq studies using human AD autopsy samples have identified myelin-related transcripts among the top altered gene clusters in patients with AD 35,46 . These findings, however, probably reflect the downstream effects of overt Aβ and tau pathology on oligodendrocytes in the latest stages of disease. In our study, we specifically asked whether myelin defects are upstream of amyloid deposition.
Based on our findings, we propose a working model for AD (Fig. 5j) in which myelin dysfunction in the ageing forebrain (modelled by the premature ageing of white matter in myelin mutant mice) causes microglia to become engaged. This microglial activity interferes with the ability to clear Aβ deposits and prevent plaque formation. Simultaneously, ageing myelin loses its axon-supportive functions, leading to axonal distress, which in turn increases neuronal BACE1 and APP CTF levels suggestive of enhanced Aβ production. It is tempting to speculate that other well-known AD risk factors, such as traumatic brain injury or cardiovascular diseases, also convey risk by negatively affecting myelin health or at least through similar mechanisms (that is, distracting microglia).
The identified plaque-promoting factors, axonal distress and microgliosis are tightly linked downstream phenomena of myelin injury, which makes it difficult to study them in isolation. We have previously shown that depletion of microglia in Cnp −/− mice can reduce the number of axonal swellings 47 , which indicates that axonal distress and APP metabolism changes could be downstream of microglia activation. However, nonspecific microglia activation induced by lipopolysaccharide does not have the same amyloid-promoting effect as myelin dysfunction, as increases, decreases or no changes in amyloid burden have been reported 48-51 . By contrast, myelin defects seem to be unique drivers of microglia activation in the ageing brain (as previously shown 3 ), which challenge microglia with the uptake and digestion of large amounts of lipids and highly compacted membranes.
An important clinical implication of our findings is the co-existence of AD and multiple sclerosis (MS). Based on our findings in EAE and cuprizone-treated AD mice, we would suspect an increased rate of AD as a comorbidity of MS. A recent study 52 has shown that patients with MS have a higher risk of receiving an AD or dementia diagnosis. However, given the very limited dataset, further epidemiological studies concerning the comorbidity of MS and AD are needed 53 .
A role of myelin deterioration in the progression of AD has been previously proposed in light of correlative macroscopic-scale brain imaging studies 54 , but they lacked any evidence of causality. Our experimental findings in genetic models of AD now provide a causal link with a molecular underpinning and are in line with the prevailing amyloid hypothesis and the role of neuroinflammation in disease progression. Moreover, we can position the well-documented loss of myelin integrity in the ageing primate brain as an upstream initiator of AD pathology, which might help to explain why age is the major AD risk factor. Our working model is further supported by previous classical neuropathological observations 55 , in which a conspicuous temporal relationship with CNS myelination was reported, which occurs latest in those brain regions that are the first to develop AD pathology. Mechanistically, cortical ensheathment by oligodendrocytes is thin and may lack the distinct myelinic channel architecture important for glial metabolic support found in subcortical tracts that appear resistant to AD pathology.
Our discovery that age-dependent loss of myelin integrity can be a driver or risk factor of amyloid deposition changes our view on the role of oligodendrocytes in AD-from passive bystanders to active contributors of disease progression. If further corroborated in humans, promoting myelin health should be considered as a therapeutic target for the delay or prevention of AD.

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/s41586-023-06120-6. 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 licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence 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 licence, visit http://creativecommons.org/licenses/by/4.0/.

Human tissue analysis
Selection of patients was performed from a pool of approximately 400 individuals, in which an autopsy with neuropathological evaluation was performed between 2018 and 2019 as a matter of routine procedure following death at the Leipzig University Hospital. Ethics oversight was performed by the Ethics Board of the Leipzig University Hospital. In the individual contracts that govern medical treatment, all patients included in the study provided written consent following hospital admission to the scientific use of tissue removed and stored after any biopsy or during autopsy.
Selection of patients was performed according to exclusion and inclusion criteria. Samples were anonymized and processed in a blinded manner. Selected patients were of mixed age and sex, aged between >60 years and <90 years of age, had a clinical history of dementia and a National Institute on Aging-Alzheimer's Association (NIA-AA) score in neuropathological assessment between A2B3C2 and A3B2C2 (inclusion criteria) and did not suffer from another severe neurological disorder (exclusion criteria). In addition, individuals of the same age range without any clinical or neuropathological record of neurological disease were selected. No other criteria besides the described characteristics were applied. In total, paraffin-embedded CNS tissue from three patients with moderate to pronounced AD neuropathological changes according to the NIA-AA (see above) and from three unaffected individuals were histologically evaluated. In our analysis, we used biopsy samples from the medial temporal lobe containing the hippocampal formation. We performed histological assessment for intracortical myelin content on the human tissue provided. For this, we sectioned paraffin blocks (5 µm) and stained paraffin sections for CNP, PLP, IBA1 and amyloid plaques simultaneously (see the section 'Paraffin slices and histological stainings' for details). Quantification of microgliosis (IBA1) and myelin levels (CNP and PLP) were performed in the trans-entorhinal cortex by performing thresholding, and the percentage of positive area was determined using Fiji in equally sized ROIs in the transentorhinal cortex.

Mouse strains and husbandry
Animal experiments were conducted in compliance with German animal welfare practices and approved by the local authorities (Landesamt für Verbraucherschutz und Lebensmittelsicherheit, Niedersachsen). Mice were group-housed in the local animal facility of the Max Planck Institute for Multidisciplinary Sciences under a 12-h dark-12-h light cycle and fed ad libitum. Both sexes were used throughout the study. The following original mouse strains were used to generate crossbreeedings: 5×FAD (ref. 56 For analysis, either littermate controls were used or a corresponding control line from the initial F 1 generation was generated. The age of animals is given in the respective figure legend. All animals were maintained on a C57BL/6 background. Genotyping was performed on clips derived from earmarking according to standard protocols (see references for original mouse strains). The genotype was confirmed by regenotyping on a tail biopsy following euthanasia of the animal at the end of the respective experiment.

Demyelination models
As demyelination models, we used cuprizone and EAE, and experiments were conducted as previously described 64 . For cuprizone-mediated demyelination, male 14-week-old 5×FAD mice were fed a powder diet containing 0.2% (w/w) cuprizone (Sigma-Aldrich) for 4 weeks followed by a 4-week recovery period on normal pelleted food without cuprizone supplementation. Control 5×FAD mice received a standard diet without cuprizone. Animals were perfused at 18 weeks of age, and brain tissue was analysed by LSM and epifluorescence microscopy. For EAE experiments, 10-week-old 5×FAD mice were immunized against myelin oligodendrocyte glycoprotein (MOG) by subcutaneously injecting 200 mg MOG(33-35) peptide in complete Freund's adjuvant (Mycobacterium tuberculosis at 3.75 mg ml -1 ; BD) followed by injecting 500 ng of pertussis toxin (Sigma) at day 1 and 3 after EAE induction (d.p. i. 1 and 3). Animals were checked on a daily basis, and a neurological disease score was determined according to the following parameters: 0, normal; 0.5, loss of tail tip tone; 1, loss of tail tone; 1.5, ataxia and mild walking deficits (slip off the grid); 2, mild hind limb weakness, severe gait ataxia and twist of the tail causing rotation of the whole body; 2.5, moderate hind limb weakness and inability to grip the grid with the hind paw but ability to stay on an upright tilted grid; 3, mild paraparesis and falls from an upright tilted grid; 3.5, paraparesis of the hind limbs (legs strongly affected but move clearly); 4, paralysis of the hind limbs and weakness in the forelimbs; 4.5, forelimbs paralysed; 5, moribund or dead. Animals were perfused at 28 d.p.i. at 14 weeks of age. All immunized 5×FAD animals developed EAE. Brain and spinal cords were analysed by LSM and epifluorescence microscopy.

Mouse behavioural testing
Hindlimb clasping was assessed by suspending the mouse on its tail for 5 s, carefully observing movement of the hindlimbs and scoring movement impairments according to a score from 0-4 as follows: 0, no impairment, hindlimbs normally spread and moving; 1, one hindlimb shows slightly less mobility; 2, both hindlimbs show less mobility; 3, both hindlimbs show reduced mobility and reduced spread; 4, both hindlimbs show severely reduced movement and severely reduced spread. Videos were recorded. Animals were tested in the elevated plus (EPM) and Y-maze (YM) on consecutive days. In the EPM experiment, animals were allowed to freely explore an elevated cross-shaped platform with two opposing enclosed arms and two opposing open arms for 5 min without the experimenter present. In the YM experiment, animals were allowed to explore a Y-shaped maze with enclosed arms for 8 min without human interference. Videos were recorded and analysed using the Bioobserve Viewer behavioural analysis setup with automated tracking. Zones (open arms, closed arms, centre) and object detection settings were optimized according to the maze type used. After the run, positional data, track length and full track images were exported. For the EPM experiment, the time spent in both open arms was summed and plotted. For the YM experiment, the order of arm entries was recorded, and the number of correct triads (consecutive visit of the three different arms) determined. The alternation index was calculated according to the formula alternation index = number trials/ number of arm entries - 2. For a successful visit to an arm of the maze, animals had to enter an arm with their full body (excluding the tail).
For statistical analysis of the behavioural data, we report the results of several different type 3 analysis of variance (ANOVA) tests that calculated the main effects for the 5×FAD mice and myelin mutant mice as well as their interaction. All analyses were conducted in R (v.4.04). The ANOVA tests were computed using the afex package (afex v.0.28-1) 65 . Post hoc testing was performed using Tukey's multiple comparison test in Prism 8.

Tissue preparation
For microscopy analysis, animals were deeply anaesthetized or euthanised by CO 2 asphyxiation and subsequently transcardially perfused with ice-cold Hank's buffered salt solution and 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer, pH 7.4. Brains and spinal cords were removed and post-fixed overnight in 4% PFA and phosphate buffer. Brains were washed once in PBS pH 7.4 and stored in PBS at 4 °C until further use. For biochemical analysis of full brains, animals were killed by cervical dislocation, brain and spinal cord were extracted and fresh-frozen on dry ice. Tissue was stored at −80 °C until further use. For microdissection of subcortical white matter and cortical tissue, animals were killed by cervical dislocation and the brain was quickly removed and submerged in ice-cold Hank's buffered salt solution. The brain was inserted into a custom-made brain matrix and the brain was sliced coronally in about 1-mm-thick slices. Brain slices were spread onto an ice-cold glass plate, and cortex and subcortical white matter were separated and excised from each individual brain slices. Tissue was immediately frozen on dry ice and stored at −80 °C until further use.

Whole tissue staining and clearing for LSM
Fixed hemibrains and spinal cords were pretreated and permeabilized following a modified iDISCO protocol 66 . In brief, tissue samples were dehydrated with an ascending concentration of methanol in PBS (50% once, 80% once, 100% twice, 1 h each). Tissue samples were then bleached with a 1:1:4 ratio of H 2 O 2 :DMSO:methanol overnight at 4 °C. Further dehydration was followed by 100% methanol incubation at 4 °C (30 min), −20 °C (3 h) and overnight storage at 4 °C. Samples were incubated the following day in methanol with 20% DMSO before subjecting them to gradual rehydration in a descending methanol in PBS series (80% once, 50% once, 0% once, 1 h each). We then washed the tissue samples with a detergent mixture of 0.2% Triton X-100 in PBS (twice, 1 h) and permeabilized them overnight at 37 °C in PBS with 0.2% Triton X-100, 20% DMSO and 0.3 M glycine. After permeabilization, tissue samples were either stained with the Congo red dye (Sigma Aldrich) for β-sheet structures within amyloid plaques or immunolabelled with antibodies of interest. For Congo red staining, tissue samples were washed with PBS and 0.2% Tween-20, 10 mg ml -1 heparin, 5 mM sodium azide (PTwH; twice, 1 h) before immersing them for 3 days at 37 °C in 0.005% w/v Congo red (100× stock solution in 50% ethanol). For immunolabelling, following tissue permeabilization and glycine treatment, samples were blocked in PBS with 0.2% Triton X-100, 10% DMSO and 6% goat serum (GS) for 3 days followed by washing in PTwH (twice, 1 h) and incubation in primary antibodies with the appropriate dilution factors (1:500, rabbit, anti-IBA1, Wako) in PTwH with 0.2% Triton X-100, 5% DMSO and 3% GS for 14 days at 37 °C. Following completion of labelling with primary antibody, tissue samples were washed in PTwH (6 times, 10 min) and stored overnight at 37 °C. For secondary antibody labelling, tissue samples were again incubated in secondary antibodies with the appropriate dilution factors (anti-rabbit Dylight 633, 1:500, Thermo-Fisher) in PTwH with 3% GS for 7 days at 37 °C. Before clearing, spinal cords were fixed in 1.5% w/v Phytagel in water. Dyed or immunolabelled tissue samples were washed in PTwH (3 times, 10 min each) before rehydration through an ascending concentration of methanol in PBS (20% once, 40% once, 60% once, 80% once, 100% once, 1 h each) and delipidation in a 1:2 mixture of methanol:dichloromethane (once, 1 h 40 min). Last, samples were cleared by immersing them in ethyl cinnamate (Eci, Sigma-Aldrich) until transparent. All incubation steps were carried out at constant medium-speed rotation at the indicated temperatures. Samples were stored at room temperature in Eci until imaging.

In toto imaging of whole tissues and analysis and visualization
Cleared hemibrains and spinal cords were imaged in toto with a light sheet microscope setup (UltraMicroscope II, LaVision Biotec) equipped with a ×2 objective lens, a zoom body and a corrected dipping cap. Samples were imaged submerged in a sample chamber containing Eci. For all hemibrain imaging, hemibrains were mounted medial-side down on the sample holder to acquire sagittal images. Images were acquired using ImspectorPro (v.7.124, LaVision Biotech) software in the mosaic acquisition mode with the following settings: 5 µm light sheet thickness, 20% sheet width, 0.154 sheet numerical aperture, 4 µm z-step size, 1,000 × 1,600 pixels field of view, 4 × 4 tiling, dual light sheet illumination and 100 ms camera exposure time. Red fluorescence was recorded with 561 nm laser excitation at 80% laser power and a 585/40 nm emission filter. Far-red fluorescence was recorded with 640 nm laser excitation at 30% laser power and a 680/30 nm emission filter. Autofluorescence was recorded with 488 nm laser excitation at 50% and 525/20 nm emission filter. Images were imported into Vision4D (v.3.2; Arivis) and stitched using the tile sorter setup. For some, images were imported and stitched using Imaris Importer and Stitcher (v.9.1; Bitplane). Rendered whole hemibrains were then processed, and three main ROIs were manually annotated based on the sagittal Allen mouse brain atlas, namely the isocortex, the hippocampus and the alveus. All ROIs were first manually traced in 2D planes to automatically extrapolate the 3D ROIs. Cortical and hippocampal annotations were cropped with a medial cutoff of approximately 0.4 mm and a lateral cutoff of 4.4 mm, which would span the dorsal isocortex and the entire hippocampal formation of one hemibrain. The lateral cutoff for the alveus ROI is the plane where the hippocampal formation appears in 2D. Next, we segmented amyloid plaques within the ROIs. For 3-month-old 5×FAD brain data, we typically used automated intensity thresholding. For 6-month-old 5×FAD data, plaques were segmented using the blob finder algorithm in Vision4D with the following parameters: 20 µm object size, 10-15% probability threshold and 0% split sensitivity. Once segmentation was performed, stringent noise removal was performed by deleting objects with voxel sizes <10 from the object table. Objects were then carefully reviewed, and any additional noise, which might include but are not restricted to blood vessels and nuclei, were manually removed from the dataset. For quantification of amyloid plaques in the APP NLGF brains that typically stained much weaker for Congo red, plaque segmentation was performed using the machine-learning segmenter in Vision4D. Object parameters and ROI volumes were extracted for further quantification. For plaque visualization in the different ROIs, objects are represented in centroids and colour-coded according to location. The following different plaque burden parameters were assessed: plaque count (number of plaques in ROI), total plaque volume (sum of all individual plaque volumes), plaque size (average volume of plaques), plaque density (number of plaques normalized to ROI volume), percentage volume (ROI volume covered by sum of plaque volume).

In situ hybridization
We performed RNAscope Fluorescent Multiplex assays (ACD Bio) according to the manufacturer's instructions for both paraffin-embedded tissue and cryo-tissue sections. In brief, paraffin-fixed slices underwent pretreatment consisting of deparaffinization steps at 60 °C for 1 h, followed by incubation in 100% xylol (5 min, twice) and 100% ethanol (2 min, twice). Slides were then incubated with hydrogen peroxide at 40 °C for 15 min. After a wash with dH 2 O, slides were boiled in the target retrieval reagent for 10 min and washed with dH 2 O (15 s, once) and with 100% ethanol (3 min, once). After drying, a hydrophobic barrier was drawn and protease digestion was performed by incubating slides with RNAscope Protease Plus for 15 min at 40 °C. Before applying the probes, slides were washed with dH 2 O (2 min, twice). The following RNAscope probes were used: Mm-CST7-C3 (498711-C3) and Mm-Ms4a7-C2 (314601-C2). The probe was hybridized for 2 h at 40 °C. Slides were stored overnight in a 5×SSC buffer. The following day, signal amplification was performed by applying amplification reagents provided in the kit. This was followed by incubation with HRP corresponding to the channel probe for 15 min at 40 °C and visualization was performed using the OpalTM 690 fluorophore for 30 min at 40 °C. After the final wash, we proceeded with immunohistochemical co-staining. Blocking of the slides was done with 10% GS in PBS for 1 h at room temperature. After blocking, incubation in primary antibody solution (PBS and 10% GS) was done overnight at 4 °C. The following primary antibodies were used: anti-IBA1 (rabbit, Wako; 1:1,000) and anti-Aβ-6E10 (mouse, BioLegend; 1:1,000). Ms4a7 RNAscope was performed on cryo-sections with IBA1 co-labelling. Cst7 RNAscope was performed on paraffin sections with 6E10 and IBA1 co-labelling. Slides were washed in PBS (5 min, 3 times) and labelled with corresponding fluorescent secondary antibody solution (PBS and 10% GS) for 2 h at room temperature. The following fluorescently conjugated secondary antibodies were used: anti-mouse Dylight488 (donkey, Thermo-Fisher; 1:1,000) and anti-rabbit Dylight555 (donkey, Thermo-Fisher; 1:1,000). Finally, slides were washed in PBS (5 min, 3 times) and mounted with Aqua PolyMount mounting medium (PolySciences). Imaging was performed using a Zeiss Observer Z1 microscope equipped with Plan-Apochromat ×20/08 objective. Quantification of Cst7 RNAscope was performed by identifying plaque-corralling microglia and plaque-distant microglia (without obvious plaque interaction), and the amount of Cst7 puncta within these microglia was counted.

Epifluorescence and brightfield microscopy
Epifluorescence microscopy was performed on a Zeiss Observer Z1 microscope equipped with Plan-Apochromat ×20/08 and Fluar ×2.5/0.12 objectives, a Colibri 5 LED light source (630 nm, 555 nm, 475 nm and 385 nm excitation), 96 HE BFP, 90 HE DAPI, GFP, Cy3, Cy5, 38 GFP, 43 DsRed, 50 Cy5 Zeiss filter sets, a Axiocam MrM and a SMC900 motorized stage. For whole-brain-slice microscopy, a preview scan at ×2.5 magnification was taken, and focus support points were distributed and manually set for imaging at ×20 magnification in ZEN imaging software (ZEN 2011 v.2.0, Zeiss). Tiled images were stitched in ZEN. For visualization, pseudocolours (cyan, magenta, yellow) were assigned to different channels, the intensity was adjusted and images were exported into ZEN. Brightfield microscopy of DAB-stained slices was performed on a Zeiss Axiophot Imager.Z1 equipped with a Achroplan ×4/0.1, Plan-Fluar ×20/0.75 and Plan Neofluar ×40/0.75 objectives and a AxioCam-Mrc camera. For whole-brain-slice microscopy, a preview scan at ×4 magnification was taken, and focus support points were distributed and manually set for imaging at ×40 magnification in ZEN imaging software (Zeiss). Tiled images were stitched in ZEN. Brightness and contrast of RGB images were adjusted and images exported into ZEN.
2D microscopy quantification 2D image analysis was performed in Fiji (v.1.53c) 68 and in ZEN. For quantification of amyloid load, thresholding was performed to segment Aβ deposits (either stained by Me-04 or Aβ plaques), and the positive area was calculated in the respective ROI. For analysis of the plaque-corralling phenotype in the 5×FAD model, in IBA1 + Aβ + co-stainings, individual plaques were segmented using thresholding, and the IBA1 + area in each individual plaque was calculated using a Fiji macro-script. IBA1 coverage was expressed as a percentage. As plaques in the APP NLGF line are not as compacted as plaques in the 5×FAD model and therefore commonly appear 'fragmented', the same type of analysis could not be reliably performed in APP NLGF mice (overestimation of single plaque structures). For both 5×FAD mice and APP NLGF mice, the number of plaque-associated microglia per plaque (50 representative plaques with diameter >20 µm for each animal) was manually counted. For quantification of ApoE levels in plaques, ApoE + plaques were segmented, and the raw mean fluorescence per plaque was calculated. For quantification of microgliosis in Cnp −/− ;5×FAD, Plp −/y ;5×FAD and Emx cre Mbp fl/fl experimental cohorts, the amount of IBA1 + DAPI + cells in the ROI (corpus callosum or cortex) was counted and normalized to the area examined. Additionally, the IBA1 + area fraction was determined in Fiji by performing thresholding. The number of axonal swellings positive for APP-processing, APP and Aβ antibodies was determined by counting these structures in the respective ROI (fimbria) and normalization to the area examined.

EM analyses
Fixed brain tissues were immersed in ice-cold PBS and cut into a coronal section with 300 µm thickness using a vibratome (Leica VT1200). Tissue containing the medial corpus callosum and cingulate cortex with adjacent tissue was punched with a 2 mm diameter punching tool and embedded in Epon using a standard Epon-embedding protocol including 4 h of incubation in 2% OsO 4 (EM TP, Leica). Samples were placed on a Parafilm-covered plate and covered with an Epon-filled gelatin capsule. Polymerization was performed at 60 °C for 24 h. Samples were trimmed, and semi-thin 500 nm sections were cut on a Leica UC7 ultramicrotome (Leica) equipped with a Histo diamond knife (Diatome). Semi-thin sections were stained with Azure II for 1 min to stain lipid-rich areas to confirm the ROI. Next, 60 nm ultrathin sections were cut on the same ultramicrotome equipped with Diatome diamond knife, ultra 35°. Ultrathin sections were transferred to Formvar-coated copper mesh grids (Science Services) and air dried. The sections were contrasted with Uranyless for 30 min (Electron Microscopy Sciences) and washed five times with ddH 2 O. Samples were air dried before further storage. At least 10 images for each area of interest (magnification of ×4,000 for corpus callosum, ×7,000 for cortex) of ultrathin sections were acquired using a Zeiss EM900. Non-overlapping transmission EM micrographs were used to quantify the numbers of myelinated axons and g-ratios. Analyses of the images were done using Fiji. The number of myelinated axons was quantified using the cell counter plugin by counting all myelinated axons in the field of view of five pictures for each animal. For g-ratio analysis, at least 150 fibres per animal were randomly selected. The area of the axon and the entire myelinated fibre were measured and used to calculate the diameter. To determine the g-ratio, the axonal diameter was divided by the diameter of the myelinated fibre. In the results, the g-ratio is plotted against the axon calibre.

High-pressure freezing EM
Sample preparation by high-pressure freezing and freeze substitution was performed as previously described 69 . In brief, optic nerves from 6-month-old Cnp −/− mice and control WT mice were freshly dissected, immersed in 20% PVP in PBS and placed into HPF sample carriers (Wohlwend). After freezing using a HPM100 high-pressure freezer (Leica Microsystems), samples were embedded in Epon after freeze substitution using 0.1% tannic acid in acetone followed by 2% OsO 4 and 0.1% uranyl acetate in acetone. After polymerization, samples were sectioned using a UC7 ultramicrotome (Leica Microsystems) and imaged with a LEO912 transmission electron microscope (Carl Zeiss Microscopy) using an on-axis 2k CCD camera (TRS).

MACS of microglia and bulk RNA-seq
Microglia were isolated from mouse hemibrains (excluding cerebellum and olfactory bulb) by MACS. Dissected tissues were enzymatically and mechanically dissociated using a Miltenyi Biotec adult brain dissociation kit according to manufacturer's protocol. Before microglial isolation using CD11b microglia microbeads and LS columns (Miltenyi Biotec), astrocytes (ACSA-2 microbeads) and oligodendrocytes (O4 microbeads) were removed to enhance the purity of the microglial population. Isolated microglia were directly eluted into RLT lysis buffer, and RNA was isolated using a RNeasy Micro kit (Qiagen). In total, n = 4 replicates were used for each genotype (WT, Cnp −/− , 5×FAD and Cnp −/− ;5×FAD). RNA extracted from sorted mouse brain hemisphere microglia was eluted in 30 µl nuclease-free water and subjected to 50 bp single-end mRNA sequencing using HiSeq 4000 (Illumina). Raw sequencing data were first evaluated using FASTQC (v.0.72) for quality, then aligned against the reference mouse genome GRCm38 using STAR (v.2.5.2b-2) 70 with default parameters. Gene raw counts of each sample were extracted using featureCounts (v.1.6.3) 71 from aligned profiles for differential gene expression analysis using DESeq2 (v.1.26.0) 72 and converted to TPM value for sample distance calculation and visualization, as well as for gene expression pattern analysis. For differential gene expression analysis, each pair of genotype was calculated separately, and statistics results were summarized (Supplementary Table 1). Gene targets with adjusted P value < 0.05 were considered as significantly regulated. Using normalized TPM profiles, samples were embedded by PCA to assess distances.
For significantly regulated genes between each pair of genotype, functional and pathway enrichment analysis using over-representation analysis approach was performed using R (v.3.6.3) with the gprofiler2 package (v.0.2.0) 73 , in which gene set enrichment analysis was performed using the WebGestalt interface 74 . DEGs obtained from all paired comparisons were aggregated and clustered based on their expression patterns across genotypes, whereby clusters were identified using k-means (k = 10) and expression levels were represented by scaled TPM values. GO (biological process) enrichment analysis was performed for each DEG cluster using gprofiler2, and significant terms enriched from Article all clusters were input for similarity analysis using simplifyEnrichment package (v.1.5.2) 75 built under R (v4.1.0). To view potentially gene-gene networks of significantly regulated targets (adjusted P < 0.01) between Cnp −/− ;5×FAD mice and 5×FAD mice, STRING (v.11.5) analysis was applied with the highest confidence (0.900) of interaction score.

In vitro phagocytosis assay
An Aβ phagocytosis assay was performed on bone-marrow derived macrophages (BMDMs) to assess phagocytosis capabilities of phagocytes after myelin pretreatment. BMDMs were isolated and cultured as previously described 64 . For experiments, cells were detached by accutase treatment and seeded onto poly(l-lysine)-coated, HCl washed 12-mm-diameter glass coverslips (density of 18,000 cells per coverslip). Cells were left untreated (control) or treated with purified myelin membranes (purification by sucrose-gradient ultracentrifugation) at 2.5 µg ml -1 for 12 h. Cells were subsequently treated with HiLyte488-labelled aggregated, synthetic Aβ 42 (Anaspec, AS-60479-01) at 1 µg µl -1 for 4 h. Aggregation of Aβ 42 was induced by previous incubation of Aβ 42 at 100 µg ml -1 in DMEM at 37 °C for 24 h after vortexing the solution for 15 s. Cells were fixed in 4% PFA and PBS and subjected to fluorescence labelling with DAPI and Lectin-Dylight 649 (Lycopersicon Esculentum Tomato; Fisher Scientific, NC1093063). Lectin-Dylight 649 was incubated for 2 h at room temperature in PBS. Optical sections were acquired with a confocal laser-scanning microscope (Zeiss LSM 900 AiryScan) using a LD C-Apochromat ×40/1.1 water objective (zoom factor of 1.0). The z-step in z-stacks was kept at 0.8 µm. Images were acquired in bidirectional mode (Aryscan Mode SR:4.1 2D Auto) and the LSM scan speed was kept at 7. Laser intensity was kept at 2% 640 nm, 0.1% 561 nm, 0.5% 488 nm and 0.2% 405 nm. Image resolution was selected to be 1,000 × 1,000 pixels at 16 bit. Acquired images were maximum-intensity projected along the z-axis, and the amount of phagocytosed Aβ 42 was determined by thresholding in the respective channel using Fiji. This percentage positive area was normalized to the area covered by cells (as determined by lectin labelling). Two to three images per treated coverslip were analysed.

Isolation of nuclei and single-nuclei transcriptome sequencing
We generated two snRNA-seq datasets for the analysis of microglia states induced by myelin dysfunction and amyloidosis. For the first dataset, cortex and corpus callosum from 3-month-old WT, forebrain shiverer mice (Foxg1 cre Mbp fl/fl ), Cnp −/− mice and WT controls were microdissected, and tissues collected from two animals were pooled for each replicate. For the second dataset, brain hemispheres (without cerebellum and olfactory bulb) were extracted from 6-month-old WT, Cnp −/− , 5×FAD and Cnp −/− ;5×FAD mice. Two replicates per genotype were sequenced. Nuclei were isolated according to previously published methods 76 . In brief, frozen tissue was transferred into 2 ml of pre-chilled homogenization buffer (320 mM sucrose, 0.1% NP40, 0.1 mM EDTA, 5 mM CaCl 2 , 3 mM Mg(Ac) 2 , 10 mM pH 7.8 Tris, 167 µM β-mercaptoethanol and 1× protease inhibitor (Roche)). Tissue was carefully homogenized and filtered through a 80 µm strainer and further centrifuged for 1 min at 100 .r.c.f. For each sample, 400 µl supernatant was collected into a pre-chilled 2 ml low-binding Eppendorf tube, followed by adding 400 µl 50% iodixanol solution (in 1× homogenization buffer containing 480 mM sucrose) to reach a 25% iodixanol concentration. By layering 600 µl of 29% iodixanol underneath the 25% iodixanol mixture, then 600 µl of 35% iodixanol underneath the 29% iodixanol layer, two clear interfaces between different concentrations of buffers were created, and the tube was centrifuged for 20 min at 3,000 r.c.f. After centrifugation, nuclei were collected from the band between the 29% and the 35% iodixanol layers and transferred to a fresh pre-chilled tube. Isolated nuclei were washed and resuspended in cold resuspension buffer (1× PBA, 1% BSA and 0.1 U µl -1 RNase inhibitor) and further subjected to single-nuclei transcriptome libraries using a chromium single cell 3′ reagent kit according to the manufacturer's instructions (10x Genomics). The constructed libraries were sequenced using Novaseq 6000 (Illumina). Raw snRNA-seq data were collected in Fastq format. Data generated from the 3-month time point was aligned to reference genome premRNA using CellRanger toolkit (v.3.0.2; 10x Genomics). The 6-month time point snRNA-seq data were first aligned to the reference mm10 genome version index from 10x Genomics (refdata-gex-mm10-2020-A) using CellRanger toolkit (v.6.1.2; 10x Genomics), and aligned profiles were further processed using CellBender (v.0.2.0) for removal of ambient transcripts.
Matrices containing a unique molecular identifier (UMI) count of each gene in each nuclei were extracted for all samples by filtering out nuclei with <200 detected genes and <500 total transcripts, as well as nuclei with outlier level transcript quantity or a gene detection rate identified according to individual sample sequencing depth (Supplementary Table 1). Genes expressed in fewer than three cells were excluded from further analysis. Filtered expression matrices were combined based on data time point (3-month and 6-month), and the UMI of each nucleus were normalized towards its total UMI count with a scale factor of 10,000 and then log transformed.

Dimensionality reduction, clustering analysis and cell type annotation
The normalized UMI matrix for 3-month and 6-month time point data were mainly analysed using the R package Seurat (v.4.1.1) 77,78 . Highly variable genes were calculated and scaled to support linear dimensionality reduction using PCA. For all cells sequenced from 3-month-old animals, the first 50 PCs were used for further neighbouring embedding using UMAP 79 , as well as for performing the clustering analysis with a resolution of 0.5 using K-nearest neighbour algorithm. Cells sequenced from 6-month-old animals underwent the same neighbouring embedding protocol using the first 50 PCs and proceeded with clustering analysis with a resolution of 0.8. Cluster marker genes in both datasets were calculated using both default Wilcoxon test and the MAST algorithm 80 to carefully determine cluster cell-type annotations. Clusters with undefined cell identities were removed from further analysis. To perform cell-type-specific analysis for microglia, the corresponding cell population from each dataset was first subset and reduced for its dimensionality using PCA. Similarly, selected top PCs were used for UMAP embedding and clustering analysis, with cluster marker genes calculated using the default Wilcoxon test. Specific parameters used for analysing microglia can be found in Supplementary Table 1.
External human snRNA-seq data microglia subset re-analysis Previously published human snRNA-seq data-snRNAseqAD_TREM2 (ref. 35) and snRNAseqPFC_BA10 (ref. 46)-were obtained from AD Knowledge Portal (https://adknowledgeportal.org) following application for data access. Aligned data profiles were downloaded and pre-processed with quality control and removal of outlier cells, referring to parameters that were used in the respective original publications. Cells from each dataset were analysed using Seurat (v.4.1.2). Microglia and macrophage cell populations were identified by a set of marker genes including CX3CR1, P2RY12, TMEM119, HEXB, TREM2 and SPP1 and further extracted for subset analysis. Microglia subset expression matrices from both datasets underwent data normalization, high variable gene calculation, linear reduction of dimensionality and eventually neighbouring embedding using UMAP. Unbiased clustering analysis was performed at a resolution of 0.5 in both subsets. Cluster marker genes were calculated using default setups compiled in the Seurat analysis pipeline. Subpopulations that were identified as doublets, low-quality cells or contaminations were removed from further analysis and visualizations.

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

Data availability
All raw sequencing data and raw and processed count matrices have been uploaded to the Gene expression Omnibus 81 under the SuperSeries accession number GSE178304 (microglia bulk RNA-seq in GSE178296; 6-month-old snRNA-seq data in GSE208683; 3-month-old snRNA-seq data in GSE178295). External datasets recruited for analysis were accessed through accession numbers listed in the Methods. Source data are provided with this paper.

Code availability
Packages involved in data analysis pipelines are listed in the Methods. The code used for bulk and single-nuclei transcriptome sequencing is available on GitHub at https://github.com/TSun-tech/AD_MyelinMutant.git. The Fiji macro script for analysis of plaque-corralling microglia has also been deposited into GitHub at https://github.com/TSun-tech/ AD_MyelinMutant.git. More detailed information is available upon contacting the corresponding authors.

Reporting Summary
Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist.

Statistics
For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.

n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly The statistical test(s) used AND whether they are one-or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of all covariates tested A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals) For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted The code used for bulk and single-nuclei transcriptome sequencing as well as FIJI macro to analyse plaque-coralling is available on GitHub https://github.com/TSun-tech/AD_MyelinMutant.git For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Portfolio guidelines for submitting code & software for further information.

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A description of any restrictions on data availability -For clinical datasets or third party data, please ensure that the statement adheres to our policy All raw sequencing data, as well as raw and processed counts matrices have been uploaded to the Gene expression Omnibus (GEO)101 under the SuperSeries