APOE from astrocytes restores Alzheimer's Aβ-pathology and DAM-like responses in APOE deficient microglia

APOE from astrocytes restores Alzheimer’s Aβ-pathology and DAM-like responses in APOE deficient microglia Pranav Preman1,2, Daan Moechars1,2, Emre Fertan3,4, Leen Wolfs1,2, Lutgarde Serneels1,2, Disha Shah1,2, Jochen Lamote5, Suresh Poovathingal1, An Snellinx1,2, Renzo Mancuso6,7, Sriram Balusu1,2, David Klenerman3,4, Amaia M Arranz8,9, Mark Fiers1,2,+, Bart De Strooper1,2,10,+ 1 VIB Center for Brain & Disease Research, Leuven, Belgium. 2 Laboratory for the Research of Neurodegenerative Diseases, Department of Neurosciences, Leuven Brain Institute (LBI), KU Leuven (University of Leuven), Leuven, Belgium. 3 Yusuf Hamied Department of Chemistry, University of Cambridge, Cambridge, UK. 4 UK Dementia Research Institute, University of Cambridge, Cambridge, UK. 5 VIB FACS Expertise Center, Center for Cancer Biology, Leuven, Belgium. 6 Microglia and Inflammation in Neurological Disorders (MIND) Lab, VIB-UAntwerp, Centre for Molecular Neurology, Antwerp, Belgium. 7 Department of Biomedical Sciences, University of Antwerp, Antwerp, Belgium. 8 Laboratory of Humanized Models of Disease, Achucarro Basque Center for Neuroscience, Leioa, Spain. 9 Ikerbasque Basque Foundation for Science, Bilbao, Spain. 10 UK Dementia Research Institute, University College London, London, UK. + Correspondence to Bart De Strooper: bart.destroope[email protected].uk or for data analysis Mark Fiers: [email protected] Final character count (no spaces): 122,340 ABSTRACT The major genetic risk factor for Alzheimer’s disease (AD), APOE4, accelerates beta-amyloid (Aβ) plaque formation, but whether this is caused by APOE expressed in microglia or astrocytes is debated. We express here the human APOE isoforms in astrocytes in an Apoe-deficient AD mouse model. This is not only sufficient to restore the amyloid plaque pathology but also induces the characteristic transcriptional pathological responses in Apoe-deficient microglia surrounding the plaques. We find that both APOE4 and the protective APOE2 from astrocytes increase fibrillar plaque deposition, but differentially affect soluble Aβ aggregates. Microglia and astrocytes show specific alterations in function of APOE genotype expressed in astrocytes. Our experiments indicate a central role of the astrocytes in APOE mediated amyloid plaque pathology and in the induction of associated microglia responses. Keywords: Alzheimer’s disease/Astrocytes/Microglia/APOE/β-amyloid pathology INTRODUCTION The APOE4 variant of apolipoprotein E was identified 30 years ago as the major genetic risk factor for Alzheimer’s Disease (AD) at the population level (Corder et al., 1993; Strittmatter et al., 1993; Bennett et al., 2009) while the APOE2 variant was the first identified protective allele against AD (Corder et al., 1994; Nagy et al., 1995). Apolipoprotein E is a lipid carrier and as such involved directly or indirectly in a spectrum of pathophysiological processes. In AD, its specific role in promoting β-amyloid (Aβ) plaque pathology is, however, well established (Schmechel et al., 1993; Kok et al., 2009; Fleisher et al., 2013; Murphy et al., 2013). While effects on clearance of Aβ (Deane et al., 2008; Castellano et al., 2011; Verghese et al., 2013) have been described, the major role is direct promotion of Aβ fibrillization (Ma et al., 1994; Wisniewski et al., 1994; Castano et al., 1995) as also supported by in vivo experiments (Huynh et al., 2017; Liu, Zhao, et al., 2017). This interpretation is supported by genetic studies in humans. The APOE4 allele is associated with increased amyloid deposition (Polvikoski et al., 1995; Bennett et al., 2009; Hanson et al., 2013; Gonneaud et al., 2016), possibly explaining the early onset of disease in homozygous APOE4 carriers (Blacker et al., 1997; Martins et al., 2005; Sando et al., 2008; Powell et al., 2021). On the other hand, AD mouse models show limited amyloid plaque pathology in the absence of Apoe (Bales et al., 1999; Holtzman, Fagan, et al., 2000; Ulrich et al., 2018). APOE4targeted replacement mice, expressing the human APOE4 allele instead of the mouse Apoe, crossed with different AD mouse models such as PDAPP, APP/ PS1ΔE9 and 5xFAD, accelerated plaque formation compared to APOE3-targeted replacement mice (Holtzman, Bales, et al., 2000; Fagan et al., 2002; Bales et al., 2009; Castellano et al., 2011; Youmans et al., 2012; Rodriguez et al., 2014). This has also been observed with viral-vector mediated over-expression of APOE4 isoforms in AD mouse brains (Dodart et al., 2005; Hudry et al., 2013; Zhao et al., 2016). Logically, it is assumed that the protective effect of APOE2 is the opposite of APOE4. The protective effect is reflected in a low prevalence of APOE2 alleles in AD patient cohorts (Talbot et al., 1994; West, William Rebeck and Hyman, 1994; Reiman et al., 2020) and reduction of Aβ plaque load in APOE2 carriers (Nagy et al., 1995; Polvikoski et al., 1995; Chiang et al., 2010; Kantarci et al., 2012; SerranoPozo et al., 2015; Grothe et al., 2017). Other studies suggest that APOE2 protection may be at the level of tau pathology or even at the level of delay in cognitive deficits, without reduction in Aβ plaque formation (Berlau et al., 2007, 2009). Similarly, in mouse models, the effect of APOE2 on Aβ plaques remains controversial with some studies confirming decreased amyloid deposition compared to APOE3 (Fagan et al., 2002; Castellano et al., 2011; Hudry et al., 2013; Zhao et al., 2016), while others show similar or even increased deposits (Dodart et al., 2005; Bales et al., 2009; Youmans et al., 2012; Rodriguez et al., 2014). Besides, in many studies the APOE2 genotype was not included when reporting APOE4 isoform-specific effects on amyloid plaque formation (Fryer et al., 2005; Kim et al., 2011; BienLy et al., 2012; Liu, Zhao, et al., 2017; Mahan et al., 2022). The cellular source that produces the pathologically relevant APOE is another point of debate. The protein is expressed mainly in the liver. However, APOE is not transferred over the blood brain barrier (Lane-Donovan et al., 2016; Huynh et al., 2019) suggesting that astrocytes are the main contributor of APOE in the brain (Boyles et al., 1985; Zhang et al., 2014). Under stress conditions APOE might be expressed by neurons (Xu et al., 2006), and under these conditions APOE4 impairs synaptic function and aggravates tau pathology (Andrews-Zwilling et al., 2010; Buttini et al., 2010; Knoferle et al., 2014; Koutsodendris et al., 2023). Neuronal APOE, however, does not seem to be associated specifically with amyloid pathology. In contrast, microglia can significantly upregulate APOE expression as part of their DAM (Disease Associated Microglia) response to the amyloid plaques (Keren-Shaul et al., 2017; Krasemann et al., 2017; Sala Frigerio et al., 2019; Chen et al., 2020). However, reconstitution of APOE expression in microglia only in an Apoe knockout background (Mancuso et al., 2024) or microgliaspecific knockout of APOE had no effect on amyloid plaque formation or microglial association with plaques (Henningfield et al., 2022). This suggests that APOE upregulation in microglia, while an important characteristic of DAM cell state induction, is not directly involved in the initial plaque deposition, which as we discussed, seems to be the main AD causing pathological effect of APOE4. Thus, it appears that in the very early phases of AD, when Aβ is believed to seed small soluble aggregates, and no DAM cell state is yet induced in the microglia, astrocytes serve as the main source of APOE in the brain. APOE expression specifically under the GFAP-promoter modified plaque deposition in an isoform-specific manner (Holtzman, Bales, et al., 2000; Fagan et al., 2002; Liu, Zhao, et al., 2017), while astrocyte specific knockout of APOE3/APOE4 led to reduction of Aβ plaque formation (Mahan et al., 2022), in line with this hypothesis. Therefore, we investigated whether it is indeed sufficient to express APOE in astrocytes to induce Aβ plaque formation, and more importantly, whether that is sufficient to induce the DAM microglia response. Interestingly, in our experimental paradigm, APOE2 expression only in astrocytes did not reduce the amyloid plaque formation compared to APOE3, even when APOE4 had significantly higher levels than both. We present transcriptomic changes in astrocytes expressing the different APOE isoforms, that suggest possible impairment of proteostasis and autophagy in APOE4 astrocytes. We provide evidence that microglia lacking endogenous APOE expression, can be stirred in their response to amyloid pathology by APOE secreted by astrocytes. Moreover, we show that the APOE effect of astrocytes on amyloid plaques is largely cell autonomous, as in the absence of microglia, amyloid plaques are still generated. RESULTS Expression of APOE in astrocytes is sufficient to restore fibrillar amyloid plaque pathology In order to study the role of astrocyte-derived APOE on amyloid plaque pathology, we expressed the different human APOE isoforms (further referred to as APOE2, APOE3 and APOE4), along with reporter mCherry protein, specifically in astrocytes in an Apoe-deficient AD mouse model (AppNL-G-F x Apoe-/-), using recombinant AAV2/8 particles (injected ICV at postnatal day 2/3, see Materials and Methods for details) and the astrocyte specific promoter ALDH1L1 (Fig. 1A). AAV particles expressing only mCherry protein were used as controls (further referred to as APOEKO). At 2 months of age, we observed mCherry expression in the cortical regions, along with limited spread in the subcortical area (Fig. S1A). At 6 months of age, we used immunofluorescence to confirm mCherry expression in ALDH1L1+ astrocytes (Fig. S1B). The ALDH1L1 promoter is specifically expressed in astrocytes in mouse brain (Cahoy et al., 2008; Zhang et al., 2014) and has been extensively used to target gene expression in an astrocyte-specific setting (Clarke et al., 2018; Michalovicz et al., 2019; Hasel et al., 2021; Endo et al., 2022; Mahan et al., 2022). Using flow cytometry, we found that around 14% of ACSA2+ astrocytes expressed mCherry (Fig. S1Cii) and around 90% of the mCherry-positive cells were positive for astrocyte marker ACSA2 (Fig. S1Ciii). It is likely that remaining mCherry-positive cells are a subpopulation of astrocytes that are ACSA2-negative. We do not exclude that a small number of nonastrocytic cells are transduced by the viral particles. However, we confirmed by immunofluorescence the lack of mCherry expression in NeuN+ neurons, Apc+ oligodendrocytes and Iba1+ microglia (Fig. S1D). We also confirmed APOE expression in mCherry+ cells (Fig. S1E), demonstrating that we reconstituted human APOE expression in astrocytes. We characterized the amyloid pathology in the transduced mice at 6 months of age. AppNL-G-F mice exhibited a high load of amyloid plaque deposits in cortical regions as observed by X-34 staining (Fig. S2B). This was largely absent when they were crossed with Apoe-/- mice (Figs. 1B, S2A-B). Expression of human APOE isoforms in astrocytes restored the fibrillar Aβ plaque depositions in the AAVtransduced cortical regions (Figs. 1B, S2A). APOE4 and, remarkably, also APOE2, showed significantly increased load of fibrillar plaques compared to APOE3 reconstituted animals (Fig. 1C). We further quantified morphological differences, namely size and compactness (Fig. S3A), of individual Aβ plaques between the groups (see Material and Methods for details). The APOE3 group showed a significant increase in the proportion of medium sized plaques in the range of 20-40 µm diameter, while APOE2 and APOE4 both showed increased abundance of larger plaques (above 40 µm diameter) (Fig. S3B). Plaques in the APOE3 group showed significantly increased compactness across all size ranges (Fig. S3Ci-iii). Therefore, fibrillar plaques formed in the presence of APOE3 tend to be smaller and more compact morphologically compared to plaques formed in the presence of APOE2 and APOE4. We quantified human APOE expression in homogenized brain tissues using a human APOE specific ELISA (Fig. 1D) and qPCR (Fig. S1F) and found no significant overall differences in levels of expression between APOE2, APOE3 and APOE4 groups (Figs. 1D, S1F). Interestingly, we observed that APOE secreted from astrocytes colocalized with X-34 positive fibrillar amyloid plaques (Fig. S2C). Therefore, we wondered whether each APOE isoform could differentially affect plaque formation within each experimental group. We used ELISA to quantify the plaque associated guanidine soluble Aβ42 levels in the different samples as a quantitative read out for amyloid deposition. Linear regression analysis to examine the relationship between APOE levels and guanidine soluble Aβ42 levels (Fig. 1E), showed a relatively strong positive association between APOE4 and Aβ42 (R2 = 0.61, slope = 0.23, p = 0.007). APOE2 follows a similar trend (R2 = 0.43, slope = 0.11, p = 0.077), while APOE3 levels showed no association (R2 = 0.006, slope = 0.005, p = 0.843). Thus, APOE4 has a strong dose-dependent effect on amyloid fibrillization in this model of plaque deposition. We wondered whether the absence of fibrillar amyloid plaques in the APOEKO group led to an increase in soluble Aβ species in the mouse brains. Using the single molecule pulldown (SiMPull) technique (Sideris et al., 2021), we quantified the number of soluble Aβ aggregates (Figs. 1F-G) eluted by soaking the brain tissue in artificial CSF. APOEKO mouse brains showed a significant increase in the number of soluble Aβ aggregates compared to other experimental groups (Fig. 1G), confirming a shift towards soluble species in the brain parenchyma in the absence of plaque deposition. Interestingly, the APOE4 group which showed a high level of fibrillar plaque deposition, showed low levels of soluble Aβ aggregates (Fig. 1G). APOE2 was intermediary between APOEKO and APOE4. Most remarkably, APOE3, which showed lower amounts of fibrillar plaque accumulation than APOE4 (Fig. 1C), displayed low levels of soluble aggregates similar to APOE4 in this assay (Fig. 1G). Overall, our data confirm and extend the notion that APOE secreted from astrocytes is central in the formation of amyloid plaques in mice. While APOE4 accelerates this process, APOE2 also shows a similar trend, especially when compared to APOE3, indicating that the APOE2 protective effect is not simply the opposite of APOE4’s accelerating effect on plaque formation in AD, at least in the AD mouse model used here. APOE expression modulates astrocyte cell-states in AppNL-G-F mouse brain Our data points to an important role for astrocytes in amyloid plaque formation and a distinct role of the different APOE isoforms in this process at 6 months of age. There are only limited transcriptomic studies available that investigate the effect of the human APOE variants on astrocyte phenotypes in vivo (Serrano-Pozo et al., 2021; Tcw et al., 2022; Lee et al., 2023). We therefore decided to analyze the transcriptome profiles of mCherry+/ACSA2+ astrocytes from mouse brains at 6 months of age using droplet-based scRNAseq. After filtering out non-astrocyte cells and low-quality cells, we retained 18,667 astrocytes across the 4 experimental groups. Clustering analysis revealed 12 astrocyte subpopulations (Fig. 2A) with unique gene signatures (Fig. 2B, Table S1-S2), agreeing with the high level of regional and functional heterogeneity of this cell type in the brain. The clusters divided broadly into two groups based on telencephalon and non-telencephalon identity, as previously described (Zeisel et al., 2018) (Figs. 2A, B). Clusters a1, a3 and a4 show a high expression of Agt, Itih3 and Slc6a11 which are markers for non-telencephalon astrocytes. Clusters a0, a2, a5, a6, a7, a8 and a9 show increased expression of Mfge8, Lhx2 and Ppp1r3g which are markers for telencephalon astrocytes (Fig. S4A). Since we observed changes in plaque pathology according to APOE isoforms in the cortical areas, we focused on only the telencephalon astrocytes for further analysis. Out of these, Cluster a2 was enriched for homeostatic genes of cortical astrocytes such as Epha5, Id3, Car2 (Murai and Pasquale, 2011; Batiuk et al., 2020; Theparambil et al., 2020). Subventricular zone astrocytes were enriched in Cluster a5 (Slc38a1, Hopx, Zic1 and Zic4) (Mizrak et al., 2019; Hasel et al., 2021), hippocampal astrocytes in Cluster a7 (Hopx, Hes5 and Nr2f1) (Hatakeyama et al., 2004; Li et al., 2015; Zweifel et al., 2018; Bertacchi et al., 2020) and Cluster a9 (Hopx, Gfap and Igfbp5) (Ye and D’Ercole, 1998; Bushong et al., 2002; Li et al., 2015), and striatal astrocytes in Cluster a8 (Crym) (Chai et al., 2017; Ollivier et al., 2024). We also identified two reactive populations of astrocytes. Cluster a10 was mainly characterized by downregulation of homeostatic astrocyte genes such as Slc1a2, Glul, Aqp4, and Gja1 (Fig. S4B), and upregulated Gfap, Vim, S100a6 and Meg3 (Figs. 2B, S4B). Cluster a10 showed, in addition, increased expression of plaque induced genes (PIGs) such as Cyba, Gpx4, Igfbp5, S100a6, Cd9, and Gfap, previously proposed to be part of the combined microglia-astrocyte response to amyloid plaques (Chen et al., 2020) (Fig. S4C). Besides, Cluster a11 is characterized by upregulation of interferon genes such as Cxcl10, Ifit3, Iigp1 and Ifitm3, indicating a cell-state of astrocyte associated to inflammation (Hasel et al., 2021) (Fig. S4D). Thus, our APOE reconstitution experiment covered a wide range of different astrocyte subtypes and cell states. Overall, different APOE isoforms do not seem to have a major effect on the clusters apart from the APOEKO condition where Cluster a0 was strongly enriched and Cluster a6 completely absent (Fig. 2C). Functional enrichment (Gene Ontology: Biological Process database) analysis of Cluster a6 showed enrichment for terms related to negative regulation of cellular (GO:0048523) and biological (GO:0048519) processes (Fig. 2D), driven by genes such as Pde10a, Emd, Nr4a2, and Eprs (Table S2). Among the top markers of Cluster a6 are genes involved in inflammatory pathways (Fos, Junb, Nr4a2) (Zenz et al., 2008; Estrada et al., 2020) and cell proliferation and communication (Btg2, Ccn1) (Jun and Lau, 2011; Yuniati et al., 2019; Suzuki et al., 2021) (Fig. 2B, Table S1). Cluster a0 showed enrichment for terms related to sterol (GO:0016126) and cholesterol (GO:0006695) biosynthetic processes (Fig. 2D), driven by genes such as Msmo1, Insig1 and Hmgcr (Table S2). To explore the effect of APOE deficiency on astrocyte transcriptomes further, we performed differential gene expression analysis to compare each experimental group against the APOE3 group (Fig. 3A, Table S3). We found that the highest number of DEGs were in the APOEKO group375 UP and 321 DOWN genes, compared to number of DEGs in APOE2 (p = 4.4e-64) or APOE4 (p = 4e-85). Top upregulated genes in the APOEKO astrocytes such as Ubb, Ubc and Ctsl (Fig. 3B, Table S3), are involved in protein degradation, while Ptgds (Kanekiyo et al., 2007; Kannaian et al., 2019) and Cst3 (Sastre et al., 2004; Mi et al., 2007), are associated with inhibition of Aβ fibrillization, in line with absence of amyloid plaques in APOEKO mice. Other upregulated genes such as Ndrg2 (Feng et al., 2022; Zhang et al., 2023), Gpx4 (Chen et al., 2020) and Hmgb1 (Fan et al., 2016) are known modulators of astrocyte reactivity. We compared next the differential effects of APOE4 vs. APOE3 and APOE2 vs. APOE3 (Fig. 3D). Compared to APOE3, both APOE4 and APOE2 significantly upregulated Pde10a, which was also highly upregulated in Cluster a6 (Fig. 2B) and showed minimal expression in APOEKO astrocytes (Fig. 3B, C). Pde10a belongs to a family of phosphodiesterases and their inhibition has been suggested to modulate AD and other neurodegenerative disorders, potentially activating CREB (cAMP response elementbinding)- dependent synaptic plasticity and memory formation (O’Connor et al., 2004; Puzzo et al., 2009; García-Osta et al., 2012), but these effects were related to expression of Pde10a in neurons, not astrocytes. Both APOE4 and APOE2 astrocytes also upregulated reactive markers such as Neat1 and Cd9 (Chen et al., 2020; Habib et al., 2020; Irwin et al., 2023). APOE4 astrocytes specifically upregulated several mitochondrial genes (mt-Nd1, mt-Cytb, mt-Rnr2). Notably, APOE4 also downregulated several genes involved in autophagy (Actg1, Tax1bp1, Sqstm1, Lamp1, Lamp2) (Fig. 3D) and lipid metabolism (Msmo1, Hmgcs1). On the other hand, APOE2 astrocytes, similar to APOEKO, upregulated genes involved in protein degradation such as Ubb and Uba52, with downregulation of lipid metabolism related genes (Etnppl, Insig1) (Fig. 3D). Both APOE4 and APOE2 differ from APOE3 in the sense that APOE3 is characterized by upregulated interferon response genes such as Ifi27, Lgals3bp, Ifit3 and Ifitm3 (Fig. 3D). APOE3 astrocytes also upregulated genes involved in proteostasis (Hspa8, Hsp90ab1, Hspa5), Sesn3 that reduces reactive oxygen species (ROS) in stress induced state, genes regulating phospholipase enzymes/PI signaling (Acsl3, Pla2g7, Plcd4, Prex2), and synaptic plasticity (Nnat). Thus, different APOE isoforms induce different transcriptomic responses in astrocytes in the AppNL-G-F mouse model, with potential impairment of proteostasis and autophagy in APOE4-expressing astrocytes. We also explored the astrocyte transcriptome for expression level and possible dysregulation of previously identified AD risk genes (Bellenguez et al., 2022) (Fig. S4E). Overall, astrocytes expressed only a small number of these risk genes. Out of these, Clu and Cox7c become highly upregulated in the absence of APOE, while Abca1 and Sort1 get downregulated (Figs. 3B, S4E). Clu has been reported to have the opposite effect of Apoe on amyloid pathology, with overexpression of Clu in AD mice showing decreased plaque deposition (Chen et al., 2021). Notably, we saw decreased expression of Fermt2 and Ctsb in APOE4 astrocytes (Fig. S4E), but the difference was not significant compared to APOE3 group (Table S3). A recent study has shown that downregulation of Fermt2 in astrocytes can lead to reduction of territory size and associated cognitive deficits (Endo et al., 2022). This suggests that APOE expression in an amyloid environment is linked to the modulation of AD risk genes in astrocytes. Overall, our analysis indicates that absence or expression of APOE has profound effects on astrocyte transcriptional features, indicating the multifunctionality of the protein. The differences between APOE2, APOE3 and APOE4 are less outspoken, but cover important functions such as interferon response genes, proteostasis genes, PI signaling, lipid metabolism, and as salient observation the upregulation of Pde10a. Fibrillar plaques induced by astrocyte-derived APOE elicit a reactive response in Apoe-deficient microglia Reactive microglia states associated to amyloid plaques strongly upregulate Apoe expression (KerenShaul et al., 2017; Krasemann et al., 2017; Sala Frigerio et al., 2019). We therefore questioned whether APOE from astrocytes is necessary and sufficient to restore the response of Apoe-deficient microglia to amyloid plaques in our model. We used immunofluorescence to analyze the morphological changes in microglia near the astrocyteinduced amyloid plaques. We indeed observed increased microglial clustering around the fibrillar plaque deposits (Fig. 4A). To confirm this, we segmented individual microglia (Figs. S5A-B) and classified them based on their proximity to amyloid plaques as “plaque-associated” (within X34+ thresholded plaque area and up to 5 µm outside a plaque’s edge) or “non-plaque-associated” (Figs. S5C, D). Plaque-associated microglia clustered together with a significantly lower distance to neighbouring microglia, compared to the well-tiled microglia present away from plaques (Fig. S5E). Plaque-associated microglia also had significantly lower territory size (convex area) (Fig. S5Fi) and total process length (Fig. S5Fii), indicating a less branched, amoeboid morphology, characteristic of reactive microglia. We confirmed that plaque-associated microglia stained positive for phagocytic marker Cd68 (Fig. 4B, S5G) and lysosomal protease Ctsd (Fig. 4C, S5H), characteristic of the DAM cell state (KerenShaul et al., 2017). Thus, Apoe-deficient microglia are still able to adopt a reactive state in response to fibrillar plaques induced by astrocyte-derived APOE. Next, we wondered whether the APOE isoforms influence microglial responses, in terms of number and morphological changes. We observed a significantly higher number of microglia in the AAVtransduced region for APOE4 compared to other groups (Fig. 4Di), but regression analysis demonstrated this was due to the strong positive association (R2 = 0.86, slope = 731.09, p = 3.7e-12) of total microglia numbers with the level of plaque load, independent of APOE isoform (Fig. 4Dii). Interestingly, this increase in microglia population with plaque load was driven by increased number of plaque-associated microglia (R2 = 0.97, slope = 973.13, p = 2e-16) while the non-plaque-associated population remained relatively stable despite increase in plaque load (R2 = 0.5, slope = -242.04, p = 3.5e-05) (Fig. 4Dii). Additionally, there were no significant changes in microglial morphology (territory size and total process length), for either non-plaque-associated (Fig. S5I) or plaque-associated microglia (Fig. S5J), between the different experimental groups. Therefore, at morphological level, microglial responses to fibrillar plaques do not seem to depend on the isoform of APOE produced by astrocytes but are likely mediated directly by the plaque load. In order to explore the changes in microglial reactivity at the transcriptomic level, we performed droplet-based scRNAseq on CD11b+/CD45-low microglia isolated at 6 months of age from the mouse brains expressing different APOE isoforms in astrocytes. After filtering out non-microglial cells and lowquality cells, we retained 18,569 Apoe-deficient microglia over the four experimental groups (APOEKO, APOE2, APOE3 and APOE4). Unbiased clustering analysis led to the identification of 7 sub-populations of microglia (Figs. S6A-B, Table S4-S5). We did not find major shifts in cell states as observed in other experimental paradigms ((Keren-Shaul et al., 2017; Sala Frigerio et al., 2019) probably because the AAV vector restores amyloid plaque formation only in the area of transduction (Fig. S2A). It is indeed remarkable that the effects of APOE expression in astrocytes on amyloid plaque formation is limited to those brain area where the APOE is expressed, despite APOE being a secreted protein. As we isolated microglia from the whole brain, including the non-transduced regions, only a fraction of the total pool of microglia is exposed to the amyloid plaques. However, we observed interesting shifts in gene signatures by performing differential gene expression analysis (Fig. S7A, Table S6). Among the upregulated genes in the APOEKO group were several genes involved in the cytokine response state (Ccl3, Ccl4, Ccl12, Il1a, Gpr84, Jun) (Fig. S7B). Thus, APOEKO, which shows absence of fibrillar plaques and increased soluble Aβ aggregates, is associated with microglia tending to adopt a cytokine response state. Interestingly, among the different sets of DEGs, the highest intersection was observed between APOEKO and APOE4 (p = 4.2e-78), where 52 genes were commonly upregulated in the two groups (Fig. S7A, S7C, Table S6) including significant overlap (p = 2.8e-11) of a small subset of plaque induced genes (PIGs) (C1qb, C1qc, Cd63, Cd9, Cst3, Fcer1g, Hexb, Ly86) (Chen et al., 2020) (Fig. S7C, Table S6). Their upregulation in the APOEKO group which lack fibrillar plaques, suggests that induction of these plaqueassociated microglial markers possibly precedes fibrillar plaque formation in amyloid models. We compared also the differentially expressed genes in APOE4 vs. APOE3 and APOE2 vs. APOE3 to understand the differential effect of the disease promoting and the disease protective allele versus APOE3 that is considered neutral (Fig. S7D). We observed that compared to APOE2, APOE4 had a significant upregulation of several DAM marker genes such as Cd74, Cst7, Cd63, Cd9, Ccl6, complement cascade genes (C1qa, C1qb, C1qc) and the pro-inflammatory gene Nfkbia (Sousa et al., 2018). This was accompanied by downregulation of homeostatic genes (Cx3cr1, P2ry13, Ccr5, Siglech). Also, Mertk which is necessary for detection and engulfment of amyloid plaques by microglia (Huang et al., 2021), and Inpp5d, a known AD risk factor which has a role in limiting plaque formation (Castranio et al., 2023), are downregulated in APOE4. Interestingly, compared to APOE2 and APOE4, APOE3 upregulated several markers of interferon response in microglia (Ifitm3, Ifit3, Ifi27l2a, Lgals3bp, Stat1) (Figs. S7D). APOE3 also showed an upregulation of antigen presentation genes (H2-K1, H2-D1, H2-Q7). Overall, our data show that Apoe-deficient microglia are able to react to fibrillar plaques mounting a DAM-like response when APOE is provided via astrocytes. Astrocyte-derived APOE is sufficient for fibrillar amyloid plaque pathology Our data suggest two possibilities: either APOE from the astrocytes acts directly on amyloid plaque formation and then microglia respond to the plaques formed, or APOE directly modifies microglia responses and thus they sculpt the amyloid plaques with the help of the APOE delivered by the astrocytes. To address this, we depleted microglia using PLX3397 treatment at 2 months of age before the fibrillar plaques become visible and continued treatment until the mice were 6 months of age (Fig. 5A). Despite strong depletion of microglia (percentage reductions for APOE2: 83%, APOE3: 77%, APOE4: 87%) under those conditions (Figs. 5B, S8A), fibrillar plaque deposition in APOE expressing cortical regions remained substantial, with only APOE4 group showing statistically significant decrease in plaque load upon depletion (Fig. 5C). As depletion of microglia was performed at 2 months and initial amyloid plaque seeding might have occurred already before that time point, we cannot rule out the possibility that microglia played a role in the APOE induced initial seeding of the plaques or that the remaining microglia are sufficient to induce plaque seeds. Since we have earlier seen that APOE levels also influence Aβ plaque deposition (Fig. 1E), we used multiple regression analysis to account for the confounding effect of APOE levels while comparing guanidine-soluble Aβ42 levels between PLX3397-treated and untreated animals in each experimental group (Fig. 5D). It is notable that especially in the APOE3 mice, the APOE levels are increased in the microglia depleted brains (Fig. S8Bii). This increase in APOE levels was present in APOE4 mice as well (Fig. S8Biii), with APOE2 mice showing a similar non-significant trend (Fig. S8Bi). Interestingly, for the APOE3 group, like above (Fig. 1E), we found no significant association between Aβ42 levels and APOE levels and there was no significant shift in mean Aβ42 levels (β = -28.53, SE = 38.31, p = 0.4708) with PLX-treatment. On the other hand, both the APOE2 group (β = -148.2, SE = 32.22, p = 0.0013) and the APOE4 group (β = -357.1, SE = 60.64, p = 0.0002) showed a statistically significant decrease in mean Aβ42 levels with PLX-treatment compared to the non-treated group (Fig. 5D). Therefore, this suggests that while APOE from astrocytes is sufficient for plaque deposition at 6months of age in this model, microglia play a role in modulating Aβ pathology in APOE2 and APOE4 expressing mice. DISCUSSION We investigated the role of APOE in the interplay of astrocytes, microglia and amyloid plaques. The study confirms that APOE derived from astrocytes is sufficient to induce fibrillar Aβ plaques. It was unexpected that the protective APOE2 isoform had similar, albeit less pronounced, effects as pathogenic APOE4 on amyloid plaque formation in our experiments. Thus, its protective role in AD is likely downstream of the amyloid plaque formation, affecting the cellular phase of the disease, i.e. the response of the cells in the brain to amyloid plaques (De Strooper and Karran, 2016). Previous studies have shown that knocking out mouse Apoe leads to a significant reduction of fibrillar plaque formation in different AD mouse models (Bales et al., 1999; Holtzman, Fagan, et al., 2000; Ulrich et al., 2018). We confirm here in the AppNL-G-F mouse that despite the presence of strong aggregation promoting mutations in this particular Aβ sequence, Apoe deletion almost annihilates the generation of fibrillar amyloid plaques, while increasing the soluble Aβ aggregate species (Fig. 1). These species induce a cytokine response (Mancuso et al., 2019) in the Apoe-deficient microglia. Interestingly, Apoedeficient astrocytes upregulate genes associated with inhibition of Aβ fibrillization (Ptgds and Cst3) and regulation of protein degradation (Ubb, Ubc and Ctsl) (Wegiel et al., 2000; Apelt, Ach and Schliebs, 2003; Liu, Hu, et al., 2017; Davis et al., 2021). Overall, loss of APOE expression has clearly multiple effects on the astrocyte-amyloid plaque-microglia interactions, in accordance with the central role of APOE in this important aspect of AD pathogenesis. It has been reported that reduction of fibrillar amyloid plaques via astrocytic-APOE knockout (Xiong et al., 2023), but not microglial-APOE knockout (Henningfield et al., 2022), results in a shift towards formation of cerebral amyloid angiopathy (CAA). However, we did not investigate the formation of CAA in our model and future studies will be needed to understand this aspect of cell-specific APOE on plaque pathology. Restoration of APOE expression in astrocytes alone is sufficient to induce formation of fibrillar plaques and microglia reactivity, extending findings from previous studies (Holtzman, Bales, et al., 2000; Liu, Zhao, et al., 2017; Mahan et al., 2022). In general experiments suggest the order of APOE4 > APOE3 > APOE2 for effects on amyloid deposition (Fagan et al., 2002; Castellano et al., 2011; Hudry et al., 2013; Zhao et al., 2016), and therefore our observation that APOE2 is more similar to APOE4 than APOE3 in that regard in our model, is somewhat puzzling. Nevertheless other studies have also reported that APOE2 expression is associated with similar or even higher plaque load than APOE3 (Dodart et al., 2005; Bales et al., 2009; Youmans et al., 2012; Rodriguez et al., 2014). Interestingly, having an APOE2 allele (Berlau et al., 2007, 2009) or the rare isoform APOE3Ch (Christchurch variant) (ArboledaVelasquez et al., 2019), shields patients against dementia even in the presence of high amyloid plaque load. The protective effects of APOE3Ch downstream of the amyloid plaque formation (on which a relative modest effect was observed) were also seen in an AD amyloid mouse model where overexpression had stronger effect in reducing tau seeding and propagation along with neuronal dystrophy around plaques upon injection of AD-tau brain extract (Chen et al., 2023). No consensus exists whether and how additional pathophysiological functions of APOE2 and APOE4 contribute to the specific risk of AD. A major problem to clarify the role of APOE in Alzheimer’s disease specifically and to define a therapeutic hypothesis to support a translational approach, is the pleiotropic function of APOE. Indeed, APOE4 has been implicated in the proinflammatory state in microglia (Brown et al., 2002; Colton, Brown and Vitek, 2005; Vitek, Brown and Colton, 2009; Lin et al., 2018; Lanfranco et al., 2021; Serrano-Pozo et al., 2021) and astrocytes (Arnaud et al., 2022); endolysosomal dysfunction (Ji et al., 2002; Nuriel et al., 2017; Prasad and Rao, 2018; Xian et al., 2018; Pohlkamp et al., 2021); impaired autophagy (Simonovitch et al., 2016, 2019; Parcon et al., 2018); inducing and promoting spreading of tau pathology, likely mediated by APOE receptors (Brecht et al., 2004; Shi et al., 2017, 2021; Rauch et al., 2020; Wang et al., 2021); neurodegeneration (Agosta et al., 2009; Susanto et al., 2015; Lupton et al., 2016); maintaining integrity of pericytes and blood brain barrier (Bell et al., 2012; Montagne et al., 2020, 2021); and impaired myelination in oligodendrocytes (Blanchard et al., 2022). We suggest that the APOE2 protective effect is largely driven by effects on the cellular reactions downstream of the amyloid plaque formation. We cannot exclude however that the APOE2 promoting effect on amyloid plaques as seen in the current work is model specific, and that expressing APOE2 also in other cell types might induce amyloid plaque reducing effects. It has also been shown that APOE2 expression leads to increased APOE levels in mouse models and human brain (Riddell et al., 2008; Bales et al., 2009; Conejero-Goldberg et al., 2014), which can influence amyloid pathology. This is not recapitulated in our model due to inherent limitation of AAV-mediated overexpression. APOE3 expression in astrocytes in our experiments results in two phenomena: (1) the restoration of amyloid plaque formation at 6 months and (2) the decrease in soluble Aβ aggregates in these animals. At first glance, increased fibrilization might explain decreased levels of soluble Aβ aggregates, but the APOE2 allele causes more amyloid plaques than APOE3 in our model, while it also leaves more soluble Aβ aggregates in the extracellular milieu (Figs. 1C and 1G). In addition, APOE2, but not APOE3, shows a dose dependent relationship between level of expression and amyloid aggregation (Fig. 1E), as APOE4 does. The results thus indicate that the relationship between APOE genotypes and Aβ plaque formation is not simple. For the interpretation of the data, it is important to take into account that APOE is secreted here only by astrocytes but is taken up via LDL receptors into the endosomal compartments of all cell types in the brain, including the microglia and astrocytes. Aβ, when endocytosed, becomes concentrated in the acidic endosomal compartments, which promotes amyloid fibril formation and then secretion (Knauer et al., 1992; Chung et al., 1999; Hu et al., 2009). Therefore, uptake of Aβ bound to APOE might cause increased concentrations of Aβ in these acidic compartments. Thus, in the absence of APOE, clearance of Aβ is lowered, resulting in increased soluble Aβ aggregates in the medium as demonstrated with the APOEKO. When APOE3 is present, Aβ peptides are taken up by astrocytes and microglia, resulting in a decrease of soluble Aβ (Figs 1F-G). This process is receptor were calculated to quantify their morphological changes (Peng et al., 2010; Xiao and Peng, 2013). For counting number of microglia associated to each plaque, 3d z-stacks were flattened to tiffs post microglia segmentation. X-34-positive plaques were thresholded and segmented as described above. Any microglia whose centroid was located within X-34-positive thresholded plaque area and up to 5 µm outside a plaque’s edge was considered as plaque-associated microglia. For each microglia, the mean distance to its three nearest neighbours were also calculated as a measure of clustering. FACS isolation of microglia and astrocytes For sample collection, 6 months old mice were anaesthetized using CO2 chamber and perfused with ice-cold 1x DPBS (Gibco, #14190-144). Each mouse brain, excluding cerebellum and olfactory bulbs were dissected and placed in FACS buffer (1x DPBS, 2% FCS and 2 mM EDTA) + 5 μM Actinomycin D (ActD, Sigma, #A1410-5MG) for transcriptomics. Brains were mechanically and enzymatically dissociated using Miltenyi neural tissue dissociation kit P (Miltenyi, #130-092-628) supplemented with 5 μM ActD. Next, samples were passed through a 70 μm strainer (BD2 Falcon), washed in 10 ml of icecold FACS buffer + 5 μM ActD and spun at 300 rcf for 15 minutes at 4 °C. Note that 5 μM ActD was kept during collection and enzymatic dissociation of the tissue to preserve transcriptional state and prevent artificial induction stress-response genes during tissue dissociation (Perocchi et al., 2007; Marsh et al., 2022). ActD was removed from the myelin removal step to prevent toxicity derived from long-term exposure. Following dissociation, myelin was removed by resuspending pelleted cells into 30% isotonic Percoll (GE Healthcare, #17-5445-02) and centrifuging at 300 rcf for 15 minutes at 4 °C. Accumulating layers of myelin and cellular debris were discarded and Fc receptors were blocked in FcR blocking solution (1:10, Miltenyi, #130-092-575) in cold FACS buffer for 10 minutes at 4°C. Next, cells were washed in 5 ml of FACS buffer and pelleted cells split into two suspensions (one each for astrocyte and microglia sorting), and incubated with the following antibodies: for microgliaPE-CD11b (Miltenyi, #130-113-806, 1:50), BV421-CD45 (BD, #563890, 1:500); for astrocytesAPC-ACSA2 (Miltenyi, #130116-142, 1:200), BV421-CD11b (BD, #562605, 1:500), PE-O4 (Miltenyi, #130-117-357, 1:50); in cold FACS buffer for 30 minutes at 4 °C. TotalSeq™-A cell hashing antibodies (1:500, Biolegend) and viability dye (eFluor 780, Thermo Fisher Scientific, #65-0865-14) were also added to both fractions during the incubation step. After incubation, cells were washed, and the pellet was resuspended in 500 μl of FACS buffer and passed through 35 μm strainer prior sorting. For isolating microglia, cell suspension was loaded into the input chamber of MACSQuant Tyto Cartridge and cells were sorted on MACSQuant Tyto Cell Sorter at 4 °C, based on CD11b-positive and CD45-low populations. For isolating AAVtransduced astrocytes, cells were sorted on Aria Fusion, based on negative gating for O4 and CD11b populations, followed by positive gating for double-positive ACSA2 and mCherry populations. FACS data was analysed using FCS express software. Single cell library preparation and sequencing For single-cell RNA sequencing microglia and astrocytes, 20,000-30,000 cells of each type from each mouse were sorted as described above and diluted to a final concentration of 1000 cells/µl. All the samples were individually hashed using TotalSeq™-A cell hashing antibodies, 3000 cells/animal were pooled and loaded onto the Chromium Next GEM Chip G (PN #2000177). The DNA library preparations were generated following manufacturers’ instructions (CG000204 Chromium Next GEM Single Cell 3’ Reagent Kits v3.1). In parallel the hashtag oligo libraries were prepared according to manufacturers’ instructions (BioLegend - TotalSeq™-A Antibodies and Cell Hashing with 10x Single Cell 3’ Reagent Kit v3 3.1 Protocol) using 16cycles for the index PCR. From the 4 experimental groups (APOEKO, APOE2, AOE3, and APOE4), a total of 8 libraries (APOEKO_astrocytes, APOE2_astrocytes, APOE3_astrocytes, APOE4_astrocytes, APOEKO_microglia, APOE2_microglia, APOE3_microglia, APOE4_microglia), containing 22 biological replicates (n(APOEKO)= 2, n(APOE2)= 3, n(APOE3)=3, n(APOE4)= 3), were sequenced (service provided by BGI Genomics, China) targeting 90% mRNA and 10% hashtag oligo library (50,000 reads/cell) on a DNBSEQ-G400 (MGI Tech) platform with the recommended read lengths by 10X Genomics workflow. Analysis of single-cell RNA sequencing datasets Alignment and analysis pipeline Raw BCL files were aligned to a modified GRCm39 mouse genome, which was appended with the human APOE gene. We used Bedtools 2.27.1 to extract the human APOE sequence and a thousand base buffer from GRCh38 (Quinlan and Hall, 2010). This was then appended to the murine GRCm39 genome in an artificial chromosome. The resulting genome was then filtered using the mkgtf function from Cellranger (6.1.2), in accordance with 10x guidelines (Zheng et al., 2017) and subsequently processed into a functional reference using cellranger mkref. Finally, each of the eight scRNAseq libraries were aligned to a custom genome using Cellranger’s count function which matches each transcript to its corresponding gene using the STAR aligner (Dobin et al., 2013). Raw count matrices were imported in R (v4.3.3) for data analysis. Datasets were analysed using R packages: Seurat (4.4) (Hao et al., 2021), scater (1.30.1) (McCarthy et al., 2017), clustree (0.5.1) (Zappia and Oshlack, 2018), and SuperExactTest (1.1) (Wang, Zhao and Zhang, 2015). Visualizations were done using functions from Seurat, SuperExactTest or ggplot2 (3.5). Quality control of cells and samples In this overexpression model, since APOE is under a constitutive promotor, and hence not a part of the normal transcriptional regulatory networks, we excluded human APOE gene expression from both astrocyte and microglia datasets prior to the following steps. For each library of microglia, low-level quality control was done by first filtering out cells with < 200 genes detected and genes expressed in less than 3 cells, from the raw counts matrix. Subsequent filtering involved three steps. Step 1: Cell hashing information was added to each library using Seurat::HTODemux() function to assign each cell as either singlets, doublets or negatives. Only singlets were retained, as doublets and singlets had abnormally high and low number of reads, respectively, indicative of poor-quality cells. Step 2: Libraries from the 4 experimental groups were merged using Seurat::merge() function, to have in total 23,495 cells (Fig. S9A). Cell clustering was performed using the pipeline described below. Cell type annotation was done using seurat::AddModuleScore() to assign cell scores based on previously identified signatures for different brain cell types (Zeisel et al., 2018). ~9% of cells had non-microglial identity with half of those having macrophage identity. Contaminating cell types were identified (Figs. S9A, B) as macrophages (microglia QC Cluster 4), endothelial cells (microglia QC Cluster 5), mix of astrocyte and microglial cells (microglia QC Cluster 6), monocytes (microglia QC Clusters 7 and 9), and oligodendrocytes (microglia QC Cluster 10). Out of the rest ~10% separated out as a cluster (microglia QC Cluster 2) with low number of reads and gene count (Fig. S9C). Step 3: After filtering out these clusters, each library was split into individual ones and quality control was refined with an additional step using scater::isOutlier() function. Here cells with number of reads or genes or percentage of mitochondrial genes (%mito), outside n median absolute deviation (n(reads) = +/- 4, n(genes) = - 4, n(%mito) = + 5) from library median, were removed. Finally, 18,569 microglia were retained for analysis as good-quality cells. For astrocytes, the same steps as described above with few modifications were executed. Step 1: Cell hashing did not work efficiently for astrocytes, as a good proportion of cells identified as negatives had read counts in the same range as Singlet cells. Thus, we used the other two steps to filter out lowquality cells. Step 2: Libraries from the 4 experimental groups were merged to have in total 32,982 cells (Fig. S9D). After cell type annotation as described above, ~7% of cells had non-astrocyte identity. Contaminating cell types were identified (Figs. S9D, E) as ependymal cells (astrocyte QC Cluster 5), mural cells (astrocyte QC Cluster 7), microglia (astrocyte QC Cluster 9), endothelial cells (astrocyte QC Cluster 10) and a population of undetermined identity (astrocyte QC Cluster 8). Out of the rest, ~31% separated out as clusters (astrocyte QC Clusters 1 and 6) with low number of reads and gene count (Fig. S9F). Step 3: After filtering out these clusters, each libraries were split into individual ones using Seurat::SplitObject() and quality control was refined with an additional step using scater::isOutlier() function. Here cells with number of reads or genes or percentage of mitochondrial genes (%mito), outside n median absolute deviation (n(reads) = +/- 2.5, n(genes) = - 2.5, n(%mito) = + 5) from library median, were removed. Finally,18,667 astrocytes were retained for analysis as good-quality cells. Library Integration and Cell Clustering To remove batch effects between libraries while cell clustering, the Seurat Object was split into individual libraries using Seurat::SplitObject(). Each libraries were individually normalized (Seurat::NormalizeData) and variable genes identified (Seurat::FindVariableFeatures). Genes that are repeatedly variable across libraries were selected for integration (Seurat:: SelectIntegrationFeatures()). Integration anchors were identified (Seurat::FindIntegrationAnchors()) and batch correction was performed (Seurat::IntegrateData()). After scaling and centering genes (Seurat::ScaleData()) in the integrated dataset, Principal Component Analysis (PCA) was performed. n dimensions from the PCA ( n(astrocytes) = 50; n(microglia) = 40) were selected for identifying clusters (Seurat::FindNeighbors() and Seurat::FindClusters)). To avoid over-clustering or under-clustering, multiple resolutions were used for the FindCluster() function and clustree::clustree() was used for selecting an optimum resolution (astrocytes = 0.7; microglia = 0.4). For visualization of clusters, dimensionality reduction by Uniform Manifold Approximation and Projection (UMAP), was performed with the Seurat::RunUMAP() function. Cluster markers for astrocytes and microglia were identified using Seurat::FindAllMarkers() function which calculates differential expression for each cluster against the rest. Astrocyte clusters were scored using seurat::AddModuleScore(), for markers from previously published genesets of astrocyte cell states (Zamanian et al., 2012; Hasel et al., 2021), plaque-induced genes (PIGs) (Chen et al., 2020) and AD risk genes (Bellenguez et al., 2022). Differential Expression For both astrocytes (telencephalon subtype) and microglia, differential expression was performed between the 4 experimental groups (each of APOEKO, APOE2 and APOE4 compared to APOE3), using Seurat::FindMarkers(). Genes expressed in at least 5% of cells were calculated. For calculating p-values, MAST was used which is a GLM-framework that treats cellular detection rate as a covariate. Since there were differences in the gender ratio between the groups, gender identity was regressed out during differential expression using latent.vars parameter. Post-analysis, genes with their adjusted P value < 0.05 (post-hoc, Bonferroni correction) and | log(fold-change) | > 0.2 were considered as significant. SuperExactTest (Wang, Zhao and Zhang, 2015) was used to calculate statistical significance (Tables S3, S6) and visualize overlap of differentially expressed genes while comparing each group with APOE3. Size of background gene population (n) for SuperExactTest was collected from total number of genes in the Seurat object for astrocytes and microglia library (n(astrocytes)=25321, n(microglia)=23819). Volcano plots and quadrant plots using ggplot2::ggplot() and dot plots using Seurat::DotPlot() were used to visualize differentially expressed genes between the groups. Gene Ontology (GO) Enrichment Analysis For astrocytes and microglia, cluster markers identified, as mentioned above, was used to perform functional enrichment analysis. This was done using g:Profiler web interface (https://biit.cs.ut.ee/gprofiler/gost), where the cluster markers were queried against GO biological process database, with g:SCS multiple testing correction method applying significance threshold of 0.05 (Kolberg et al., 2023). Bar plots using ggplot2::ggplot() was used to visualize the top 10 significantly enriched terms for astrocyte clusters a0 and a6. Statistical analysis Number of samples included in each experiment is included in the results section/figure legends. Sample size estimation was based on pilot experiments. All statistical comparisons were performed in Graphpad Prism (v10.0.2) or R (v4.3.3). Statistical test used for each experiment is reported in the figure legends. For analysis where individual plaques or microglia were plotted as data points, mixed effects models were used to account for the variability between individual mice and Q-Q plots were used to assess the normality of the residuals, using the lme4 package for mixed effects models and the “estimated marginal means” package in R. For comparison of ASM values in Figs. 2 and 5, the linear mixed-effects model was used with mouse ID and diameter of plaques as confounding factors. ASM response variable was also log-transformed to improve the normality of the residuals. Differential abundance of plaque-associated microglia around individual plaques in Fig. 4 was tested using binomial generalized linear mixed-effects model with mouse ID as confounding factor. Morphological differences between plaque-associated microglia and non-plaque-associated microglia in Fig S3 was tested using linear mixed-effects model with mouse ID as confounding factor. Differences in total number of differentially expressed genes between groups (compared to APOE3) was tested using Fisher’s exact test with Bonferroni correction of p-values. Enrichment of plaque-induced genes (PIGs) (Chen et al., 2020) in intersection of differentially expressed genes in microglia dataset was tested using hypergeometric test with Bonferroni correction of p-values. DATA AVAILABILITY Raw data used for this research article is available upon request to the corresponding authors. Single cell RNA sequencing data generated in this study are available at Gene Expression Omnibus (GEO) database with accession number GSE252454. ACKNOWLEDGMENTS We thank Veronique Hendrickx and Amber Claes for help with the mouse colonies and Dries T’Syen, Katrien Horré and Katleen Craessaerts for technical assistance. Mouse experiments were supported by CBD Mouse Expertise Unit (KU Leuven and VIB). Confocal microscopy was performed in the VIB Bio Imaging Core (LiMoNe). Part of the figure schematics was created with BioRender (biorender.com). FUNDING SOURCE This project received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 Research and Innovation Programme (grant agreement no. ERC-834682 CELLPHASE_AD). This work was also supported by funding from UKRI and the Medical Research Council (MR/Y014847/1) via the Dementia Research Institute. Further support was given by the Flanders Institute for Biotechnology (VIB vzw), a Methusalem grant from KU Leuven and the Flemish Government, the Fonds voor Wetenschappelijk Onderzoek, KU Leuven, The Queen Elisabeth Medical Foundation for Neurosciences, the Opening the Future campaign of the Leuven Universitair Fonds, The Belgian Alzheimer Research Foundation (SAO-FRA) and the Alzheimer’s Association USA. B.D.S. holds the BaxVanluffelen Chair for Alzheimer’s Disease. D.K. is funded by UK Dementia Research Institute, which receives its funding from DRI Ltd. funded by the UK Medical Research Council and the Royal Society. A.M.A. is funded by the Ministerio de Ciencia e Innovación under grant no. MCIN/AEI/10.13039/501100011033 (PID2021-125443OB-100 also by FEDER Una manera de hacer Europa and RYC2020-029494-I by FSE invierte en tu futuro), the Alzheimer’s Association (AARG-21850389), and the Basque Government (PIBA-2020-1-0030). AUTHOR CONTRIBUTIONS P.P., A.M.A. and B.D.S. conceived and designed the study and planned the experiments. M.F. checked all the bioinformatics data and their interpretations. P.P. performed experiments and analyzed data together with B.D.S. and A.M.A.. D.M. provided bioinformatic support and analyzed data together with M.F. E.F. performed SimPull experiments, under the supervision of D.K.. L.W., J.L., and S.P., assisted with single cell RNA-sequencing experiments. L.S. generated the mouse lines. D.S. assisted with generation of AAV vectors. A.S. assisted with immunohistochemistry experiments. R.M. and S.B. assisted with designing and execution of microglia depletion studies. All authors discussed the results and commented on the manuscript CONFLICT OF INTEREST B.D.S. is or has been a consultant for Eli Lilly, Biogen, Janssen Pharmaceutica, Eisai, AbbVie and other companies. B.D.S is also a scientific founder of Augustine Therapeutics and a scientific founder and stockholder of Muna Therapeutics. FIGURE LEGENDS Main Figures Figure 1. Expression of APOE in astrocytes is sufficient to restore fibrillar Aβ plaque pathology. (A) AAV2/8 vector was used to express human APOE under the ALDH1L1 promoter. AAV particles were injected via ICV in AppNL-G-F x Apoe-/- neonatal mice on postnatal days two or three (P2/P3). Samples were collected for analysis at 6 months of age. (B) IF images of AAV-transduced cortical regions (see Fig. S2A) at 6 months of age. mCherry (red) shows the transduced astrocytes. X34 (white) shows the fibrillar plaque deposits. Scale bar: 100 µm. (C) Bar plots showing (i) number of X-34+ plaques per field of view (FOV, 20x magnification) and (ii) area of X-34+ plaques (as fraction of total area in FOV, 20x magnification), in the AAV-transduced cortical region. Data points show mean value of 3 fields of view per mouse (n= 6-10 mice per group). (D) Bar plot showing total APOE levels in brain homogenates measured by MSD-ELISA. Data points show mean value for 2 technical replicates per mouse (n= 8-10 mice per group). ns = non-significant. (E) Scatter plot showing linear regression between total APOE levels (x-axis) and corresponding guanidine-soluble Aβ42 levels (y-axis), measured by MSD-ELISA. Data points show mean value of 2 technical replicates per mouse (n= 8-10 mice per group). (F) TIRF images of mab 6E10-positive Aβ aggregates (white) captured from soaked brain fraction using SimPull technique. Scale bar: 10 µm (G) Quantification of mAb 6E10-positive Aβ aggregates. Data points show mean value for 3 technical replicates per mouse (n= 4 mice per group, 9 field of views per technical replicate). Colour legend for experimental groups in (D), (E) and (G) are indicated in (C). Statistical tests: Data presented as mean ± SEM (C), (D), and (G). One-way ANOVA and Tukey’s multiple comparison test (C), (D), and (G). Significance shown for pairwise comparisons *p < 0.05; **p < 0.01; ***p < 0.001, ****p < 0.0001 Figure 2. APOE expression and astrocyte cell-states in AppNL-G-F mouse brain. (A) UMAP plot showing18,667 astrocytes (mCherry+/Acsa2+) sorted from 6 months old mouse brains from the four experimental groups (n= 2 or 3 mice per group). Different sub-populations identified have been assigned Cluster numbers. Dotted lines separate the clusters with telencephalon astrocyte signatures. (B) Dot plot showing the top 10 differentially expressed genes in each cluster. Colour scale indicates normalized expression level, scaled per gene (z-score). Dot size indicates percentage of cells, in each cluster, expressing the gene. (C) Bar plot showing the proportion of astrocytes from each experimental group present in the different clusters. (D) Bar plot showing top 10 significantly enriched GO Biological process terms for Clusters a0 and a6, ordered according to -log10(adjusted p-value). Statistical tests: g:SCS multiple testing correction method (Kolberg et al., 2023) in (D) applying significance threshold of 0.05. Figure 3. APOE isoforms modulate astrocyte transcriptome in AppNL-G-F mouse brain. (A) Upset plot showing number of differentially expressed genes (UP or DOWN) in telencephalon astrocytes from each experimental group, compared to APOE3 group. Bar plot shows the number of differentially expressed genes (DEGs) common between different DE analyses (overlapping sets indicated by black dots in column below the bars). Bars are colored according to statistical significance of intersection of DEGs. (B) Volcano plot showing differentially expressed genes between APOEKO telencephalon astrocytes and APOE3 expressing telencephalon astrocytes. Datapoints for significant genes are coloured (Red for UP and blue for DOWN). Significance assigned based on | Log2(Fold Change) | > 0.2 and adjusted p-value <0.05. (C) Violin plot showing normalized gene expression for Pde10a in telencephalon astrocytes from the experimental groups. (D) Quadrant plot comparing differential expression of genes in telencephalon astrocytes in APOE2 vs APOE3 (along x-axis) and in APOE4 vs APOE3 (along y-axis). Colours in legend key indicate significance of genes upor downregulatedin APOE2 or APOE4 or both. Significance of differentially expressed genes based on | Log2(Fold Change) | > 0.2 and adjusted p-value < 0.05). Pearson’s correlation, R= 0.5. Statistical tests: MAST differential expression test in (A), (B) and (D), p-values were adjusted with Bonferroni correction based on the total number of genes in the dataset. SuperExactTest (Wang, Zhao and Zhang, 2015) in (A). Figure 4. Apoe-deficient microglia mount reactive responses to fibrillar amyloid plaques. (A) IF images of AAV transduced cortical regions (see Fig. S2A), at 6 months of age. mCherry (red) shows the transduced astrocytes. X-34 staining (white) shows the fibrillar plaque deposits. Co-staining with antiIba1 antibody shows microglial cells (green). Scale bar: 100 µm. (B) IF images of AAV-APOE4 transduced cortical regions (see Fig. S2A) at 6 months of age. X-34 staining (white) shows the fibrillar plaque deposits. Co-staining with anti-Iba1 antibody shows microglial cells (green) and anti-Cd68 antibody shows phagocytic structures (red) inside microglia surrounding plaques. Scale bar: 100 µm. (C) IF images of AAV-APOE4 transduced cortical regions at 6 months of age. X-34 staining (white) shows the fibrillar plaque deposits. Co-staining with anti-Iba1 antibody shows microglia (green) and anti-CtsD antibody shows lysosomal structures (red) inside clustered microglia surrounding plaques. Scale bar: 100 µm. (D) (i) Bar plot showing total number of microglia per field of view (FOV, 40x magnification) in AAV transduced regions. Data points show mean value of 3 fields of view per mouse (n= 6-7 mice per group). (ii) Scatter plot showing linear regression between area of X-34+ staining (as fraction of total area in FOV) and total number of microglia (solid triangle), or number of plaque-associated microglia (solid circle), or number of non-plaque-associated microglia (hollow circle), for each experimental group. Data points show mean value 3 fields of view per mouse (n= 6-7 mice per group). Shape legends for different categories of microglia in (i) and (ii) are indicated. One-way ANOVA and Tukey’s multiple comparison test in (Di). Significance shown for pairwise comparisons **p < 0.01; ***p < 0.001, ****p < 0.0001 Figure 5. Astrocyte-derived APOE initiates, and microglia modify, fibrillar Aβ plaque pathology. (A) Experimental design illustrating PLX3397 treatment for microglial depletion in AppNL-G-F x Apoe-/- mice. Treatment started at 2 months of age by mixing PLX3397 (600mg/kg) in chow and samples were collected for analysis at 6 months of age. (B) IF images of AAV transduced cortical regions (see Fig. S2A) at 6 months of age from control (top row) and PLX3397-treated (bottom row) groups. mCherry (red) shows the transduced astrocytes. X-34 staining (white) shows the fibrillar plaque deposits in mouse brain cortex. Co-staining with anti-Iba1 antibody shows microglial cells (green). Scale bar: 100 µm. (C) Bar plots showing number of X-34+ plaques per field of view (FOV, 20x magnification) and area of X34+ plaques (as fraction of total area in FOV, 20x magnification), in the AAV-transduced cortical region for (i) APOE2, (ii) APOE3 and (iii) APOE4 groups and their respective PLX3397 treated groups. Data points show mean value of 3 fields of view per mouse (n= 4-10 mice per group). Data points for control non-treated groups are the same as shown in Fig. 1C (D) Scatter plot showing linear regression between total APOE levels (x-axis) and corresponding guanidine-soluble Aβ42 levels (y-axis), measured by MSD-ELISA for PLX3397-treated and control mice from (i) APOE2, (ii) APOE3 and (iii) APOE4 groups. Data points show mean value for each mouse (n= 4-10 mice per group. 2 technical replicates per mouse). Data points for control non-treated groups are the same as shown in Fig. 1E. Statistical tests: Data represented as mean ± SEM in (C). Unpaired t-test in (C). p-values are indicated. Supplementary Figures Figure S1. AAV-mediated transduction enables expression of APOE specifically in astrocytes in vivo. (A) IF image of AAV transduced mouse brain coronal sections at 2 months of age showing distribution of transduced astrocytes (mCherry in red) in cortex and subcortical areas (in AppNL-G-F x Rag2-/- x Apoe- /- mice. Scale bar: 500 µm. (B) IF image of AAV transduced cortical region (see Fig. S2A) of 6 months old AppNL-G-F x Apoe-/- mouse brain, showing mCherry (red) co-localizing with astrocyte marker Aldh1l1 (green). Image below shows a zoomed (5x) view of merged image indicated by the inset box of white dashed-lines. Scale bar: 100 µm. (C) Contour plot from flow cytometry analysis showing (i) density distribution of mCherry+ and Acsa2+ astrocytes, split to show (ii) proportion of mCherry+ cells in total Acsa2+ cells, and (iii) proportion of Acsa2+ astrocytes in total mCherry+ cells, in 6 months old AppNL-GF x Apoe-/- mouse brain. Colour scale indicates count density for each contour bin. (D) IF images of AAV transduced cortical region (see Fig. S2A) of 6 months old AppNL-G-F x Apoe-/- mouse brain. mCherry (red) shows the transduced cells. Co-staining with anti-Iba1 antibody showing microglia (cyan) (Note that this is the same image as S1B and co-staining was done with Aldh1l1); with anti-Apc antibody showing oligodendrocytes cells (green); with anti-NeuN antibody showing neurons (white) (Note that Apc and NeuN panels are shown from the same image as these markers were co-stained in the same slide). Images below show zoomed in (5x) views indicated by the inset box of white dashed-lines. Scale bar: 100 µm. (E) IF images of AAV transduced cortical region (see Fig. S2A) of 6 months old AppNL-G-F x Rag2- /- x Apoe-/- mouse brain, showing APOE3 (green) co-localizing with mCherry+ (red). Scale bar: 100 µm. (F) Bar plot showing human APOE mRNA levels at 6 months of age in mouse brain homogenates measured by semiquantitative real-time PCR. Data points show mean value for 3 technical replicates per mouse (n= 3 mice per group). ns = non-significant. Statistical tests: Data presented as mean ± SEM in (F). One-way ANOVA and Tukey’s multiple comparison test (F). Significance shown for pairwise comparisons: ****p < 0.0001 Figure S2. Astrocyte-derived APOE leads to fibrillar Aβ plaque formation in vivo. (A) IF images showing AAV transduced (red) regions in the cortex and the associated distribution of X-34+ fibrillar plaque deposits (white) in 6 months old AppNL-G-F x Rag2-/- x Apoe-/- mice. Scale bar: 500 µm (B) IF images showing distribution of X-34+ fibrillar plaque deposits (white) and Iba1+ microglia clustering (green) in cortical region in 6 months old AppNL-G-F x Rag2-/- mice (top panel). Absence of fibrillar plaque deposits and clustered microglia in cortical region of 6 months old AppNL-G-F x Rag2-/- x Apoe-/- mouse brain (bottom panel). Scale bar: 500 µm. (C) IF images of AAV transduced cortical region of 6 months old AppNL-G-F x Apoe-/- mouse brain, showing APOE3 (green) co-localizing with mCherry+ (red) astrocytes and X-34+ (white) fibrillar plaque deposits. Scale bar: 100 µm. Figure S3. Astrocyte-derived APOE modulates the size and compactness of fibrillar Aβ plaques. (A) Composite tile showing individual X-34+ fibrillar plaques, with increasing plaque size (measured in diameter) along the x-axis and increasing compactness (measured in Angular Second Moment (ASM) value) along the y-axis. Higher ASM values correlate with higher compactness. (B) Stacked bar plot showing proportion of X-34+ fibrillar plaques from APOE2, APOE3 and APOE4 groups in different size categories (diameter in µm) along the x-axis. Lower limit for plaque size set to 5 µm in diameter. Size categories used: small (5-20 µm diameter), medium (20-40 µm diameter) and large (above 40 µm diameter). (C) Box plot showing ASM values as an estimate of plaque compactness in APOE2, APOE3 and APOE4 groups. Size categories used: (i) small (5-20 µm diameter), (ii) medium (20-40 µm diameter) and (iii) large (above 40 µm diameter). Data points show value for individual plaques (n= 8-10 mice per group. 3 FOV per mouse). Statistical tests: Data presented as median and interquartile range ± values within 1.5 times the interquartile range (C). Linear mixed effects model with Tukey’s HSD test in (C) done at mouse sample level. Significance shown for pairwise comparisons **p < 0.01; ***p < 0.001, ****p < 0.0001 Figure S4. APOE expression modulates astrocyte cell-states in AppNL-G-F mouse brain. (A) UMAP plot (same as Figure 6a), showing expression of markers for non-telencephalon (Agt, Slc6a11, Slac6a9) and telencephalon (Mfge8, Lhx2, Ppp1r3g) astrocytes. Colour scale indicates normalized gene expression. (B) Violin plot showing normalized module score in each cluster, for marker gene set from previously identified reactive astrocyte cell-states. JLZ_PanPan-reactive astrocytes, JLZ_A1A1 astrocytes, JLZ_A2A2 astrocytes from (Zamanian et al., 2012), PH_Clust4Cluster 4, PH_ClustCluster 8 from (Hasel et al., 2021). (C) Violin plot showing normalized gene expression in each cluster, for homeostatic astrocyte markers (Slc1a2, Glul, Aqp4, Gja1) and reactive astrocyte markers (Gfap, Vim, S100a6, Meg3) (D) Dot plot showing expression of previously identified Plaque Induced Genes (PIGs) (Chen et al., 2020) in each cluster. Colour scale indicates normalized expression level, scaled per gene (z-score). Dot size indicates percentage of cells, in each group, expressing the gene. (E) Dot plot showing expression of previously identified AD risk genes, split by experimental groups. Colour scale indicates normalized expression level, scaled per gene (z-score). Dot size indicates percentage of cells, in each group, expressing the gene. Squares around dots mark statistically significant genes (expressed in more than 25% of cells in each group and with adjusted p-value < 0.05) based on differential expression against APOE3 group. Green squares indicate genes with | Log2(Fold Change) | > 0.2. Red squares indicate genes with | Log2(Fold Change) | < 0.2. Statistical tests: MAST differential expression test in (E), pvalues were adjusted with Bonferroni correction based on the total number of genes in the dataset. Figure S5. Apoe-deficient microglia mount reactive responses to fibrillar amyloid plaques. (A) Representation of thresholding (first step) and segmentation of microglia cell bodies, based on antiIba1 immunostaining, using a combination of Cellpose (second step: top panel) and Labkit (second step: bottom panel) algorithms. See Materials and Methods for details. (B) Representation of thresholding (first step) and tracing (second step) of primary and secondary processes of segmented microglial cell. Sum of the length of primary and secondary process calculated as “total process length”. Area of bounding box (in white) calculated as “convex area” for territory size. (C) Number of microglia segmented and quantified from AAV transduced cortical regions (40x magnification), at 6 months of age, from the experimental groups, further classified as plaque-associated (within X34+ thresholded plaque area and up to 5 µm outside a plaque’s edge) or non-plaque-associated. (n= 6-7 mice per group. 3 fields of view per mouse). (D) Bar plot showing proportion of total number of microglia from all the experimental groups (Non-plaque-associated microglia = 0.74; Plaque-associated microglia = 0.26). (E) Box plot showing mean distance between each microglia to its nearest three neighbours, for microglial cells classified as plaque-associated or non-plaque-associated. Data points show individual microglia from all experimental groups. (F) Box plots showing (i) territory size (convex area) occupied by, and (ii) total process length of, microglia classified as plaque-associated or non-plaque-associated. Data points show individual microglia from all experimental groups. Texture legends for microglia classification in (D), (E) and (F) are indicated. (G) IF images of AAV-APOE2 and AAV-APOE3 transduced cortical regions (see Fig. S2A) at 6 months of age. X-34 staining (white) shows the fibrillar plaque deposits. Co-staining with anti-Iba1 antibody shows microglial cells (green) and anti-Cd68 antibody shows phagocytic structures (red) inside microglia surrounding plaques. Scale bar: 100 µm. (H) IF images of AAV-APOE2 and AAV-APOE3 transduced cortical regions (see Fig. S2A) at 6 months of age. X-34 staining (white) shows the fibrillar plaque deposits. Co-staining with anti-Iba1 antibody shows microglia (green) and anti-CtsD antibody shows lysosomal structures (red) inside clustered microglia surrounding plaques. Scale bar: 100 µm. (I) Box plots showing (i) territory size (convex area) occupied by, and (ii) total process lengths of, non-plaque-associated microglial cells compared between the experimental groups. Data points show individual microglia from each experimental group. (J) Box plots showing (i) territory size (convex area) occupied by, and (ii) total process lengths of, plaqueassociated microglial cells compared between experimental groups. Data points show individual microglia from each experimental group. Colours legends for experimental groups in (G) and (H) are indicated. Statistical tests: Data represented as median and interquartile range ± values within 1.5 times the interquartile range in (D), (E), (F), (I) and (J). Linear mixed effects model in (C-E). Linear mixed effects model with Tukey’s HSD test in (F), (I) and (J). Significance shown for pairwise comparisons ****p < 0.0001 Figure S6. Astrocyte-derived APOE influences the expression of microglial cell state markers. (A) UMAP plot showing 18,569 APOE-deficient microglia (Cd11b+/Cd45-low) sorted from 6 months old mouse brains from the four experimental groups (n= 2 or 3 mice per group). Different sub-populations identified through unbiased clustering have been assigned cluster numbers. (B) Dot plot showing the top 10 differentially expressed genes in each cluster. Colour scale indicates normalized expression level, scaled per gene (z-score). Dot size indicates percentage of cells, in each cluster, expressing the gene. (C) Bar plot showing the proportion of microglia from each experimental group present in the different clusters. Figure S7. Astrocyte-derived APOE modulates gene expression in microglia. (A) Upset plot showing number of differentially expressed genes (UP or DOWN) in microglia from each experimental group, compared to APOE3 group. Bar plot shows the number of differentially expressed genes (DEGs) common between different DE analyses (overlapping sets indicated by black dots in column below the bars). Bars are colored according to statistical significance of intersection of DEGs. (B) Volcano plot showing differentially expressed genes between APOEKO telencephalon astrocytes and APOE3 expressing telencephalon astrocytes. Datapoints for significant genes are coloured (Red for UP and blue for DOWN). Significance assigned based on | Log2(Fold Change) | > 0.2 and adjusted p-value <0.05. Genes discussed in text are indicated in red. (C) Dot plot showing expression of the 52 genes (Figure 2A) commonly upregulated in APOEKO and APOE4 microglia, split by experimental groups. Colour scale indicates normalized expression level, scaled per gene (z-score). Dot size indicates percentage of cells, in each group, expressing the gene. Plaque-induced genes (PIGs) (Chen et al., 2020) are coloured in red on y-axis. (D) Quadrant plot comparing differential expression of genes in microglia in APOE2 vs APOE3 (along x-axis) and in APOE4 vs APOE3 (along y-axis). Colours in legend key indicate statistical significance of genes upor downregulated inAPOE2 or APOE4 or both. . Significance of differentially expressed genes based on | Log2(Fold Change) | > 0.2 and adjusted p-value < 0.05). Pearson’s correlation, R= 0.52. Statistical tests: MAST differential expression test in (B), (C) and (D), pvalues were adjusted with Bonferroni correction based on the total number of genes in the dataset. Figure S8. APOE levels and microglia depletion. (A) Box plots showing area of Iba1+ cells (as fraction of total area in FOV) in AAV transduced cortical regions at 6 months of age, from control and PLX3397treated (shaded) groups in (i) APOE2, (ii) APOE3, (iii) APOE4. Percentage reductions are: APOE2: 83%, APOE3: 77%, APOE4: 87%. Data points show mean value for 3 FOV per mouse (n= 4-10 mice per group). (B) Bar plots showing total APOE levels in brain homogenate from control and PLX3397-treated (shaded) groups in (i) APOE2, (ii) APOE3, (iii) APOE4, measured by MSD-ELISA. Data points show mean value for 2 technical replicates per mouse (n= 4-10 mice per group). Statistical tests: Data represented as median and interquartile range ± values within 1.5 times the interquartile range in (A); mean ± SEM in (B). Non-parametric Wilcoxon test in (a). Unpaired t-test in (b). *p < 0.05; **p < 0.01; ***p < 0.001 Figure S9: Quality control of microglia and astrocyte single cell libraries. (A) UMAP plot showing 23,495 APOE-deficient microglia (Cd11b+/Cd45-low) sorted from 6 months old mouse brains from the four experimental groups (n= 2 or 3 mice per group). (B) Dot plot showing the top 10 differentially expressed genes in each QC cluster of microglia library. Colour scale indicates normalized expression level, scaled by gene (z-score). Dot size indicates percentage of cells, in each cluster, expressing the gene. Microglia QC Clusters 0, 1, 2, 3 and 8 have microglial identity. Contaminating cell types identified are macrophages (Microglia QC Cluster 4), endothelial cells (Microglia QC Cluster 5), mix of astrocyte and microglial cells (Microglia QC Cluster 6), monocytes (Microglia QC Clusters 7 and 9) and oligodendrocytes (QC Cluster 10). (C) Violin plot showing (i) nUMI (total count of RNA transcripts captured), (ii) nGene (total number of genes), (iii) mito (percentage of mitochondrial genes), (iv) ribo (percentage of ribosomal genes), in each microglial QC cluster. (D) UMAP plot showing 32,982 astrocytes (mCherry+/Acsa2+) sorted from 6 months old mouse brains from the four experimental groups (n= 2 or 3 mice per group). (E) Dot plot showing the top 10 differentially expressed genes in each QC cluster of astrocyte library. Colour scale indicates normalized expression level, scaled by gene (z-score). Dot size indicates percentage of cells, in each cluster, expressing the gene. Astrocyte QC Clusters 0, 1, 2, 3, 4 and 6 have astrocyte identity. Contaminating cell types identified are ependymal cells (Astrocyte QC Cluster 5), mural cells (Astrocyte QC Cluster 7), microglia (Astrocyte QC Cluster 9), endothelial cells (Astrocyte QC Cluster 10) and a population of undetermined identity (Astrocyte QC Cluster 8). (F) Violin plot showing (i) nUMI (total count of RNA transcripts captured), (ii) nGene (total number of genes), (iii) mito (percentage of mitochondrial genes), (iv) ribo (percentage of ribosomal genes), in each astrocyte QC cluster. Table S1. List of markers identified for astrocyte clusters from single cell transcriptomic analysis. Table S2. List of GO:Biological Process enriched terms for astrocyte clusters from functional enrichment analysis. Table S3. List of differentially expressed genes (DEGs) (against APOE3) in astrocytes from each experimental group, with details of statistical test of overlap of DEGs between different groups. Table S4. List of markers identified for microglia clusters from single cell transcriptomic analysis. Marsh, S.E. et al. (2022) ‘Dissection of artifactual and confounding glial signatures by single-cell sequencing of mouse and human brain’, Nature Neuroscience, 25(3), pp. 306–316. Available at: https://doi.org/10.1038/s41593-022-01022-8. Martins, C. a. R. et al. (2005) ‘APOE alleles predict the rate of cognitive decline in Alzheimer disease: A nonlinear model’, Neurology, 65(12), pp. 1888–1893. Available at: https://doi.org/10.1212/01.wnl.0000188871.74093.12. McCarthy, D.J. et al. (2017) ‘Scater: pre-processing, quality control, normalization and visualization of single-cell RNA-seq data in R’, Bioinformatics, 33(8), pp. 1179–1186. Available at: https://doi.org/10.1093/bioinformatics/btw777. Mi, W. et al. (2007) ‘Cystatin C inhibits amyloid-beta deposition in Alzheimer’s disease mouse models’, Nature Genetics, 39(12), pp. 1440–1442. Available at: https://doi.org/10.1038/ng.2007.29. Michalovicz, L.T. et al. (2019) ‘Astrocyte-specific transcriptome analysis using the ALDH1L1 bacTRAP mouse reveals novel biomarkers of astrogliosis in response to neurotoxicity’, Journal of Neurochemistry, 150(4), pp. 420–440. Available at: https://doi.org/10.1111/jnc.14800. Mizrak, D. et al. (2019) ‘Single-Cell Analysis of Regional Differences in Adult V-SVZ Neural Stem Cell Lineages’, Cell Reports, 26(2), pp. 394-406.e5. Available at: https://doi.org/10.1016/j.celrep.2018.12.044. Montagne, A. et al. (2020) ‘APOE4 leads to blood–brain barrier dysfunction predicting cognitive decline’, Nature, 581(7806), pp. 71–76. Available at: https://doi.org/10.1038/s41586-020-2247-3. Montagne, A. et al. (2021) ‘APOE4 accelerates advanced-stage vascular and neurodegenerative disorder in old Alzheimer’s mice via cyclophilin A independently of amyloid-β’, Nature Aging, 1(6), pp. 506–520. Available at: https://doi.org/10.1038/s43587-021-00073-z. Murai, K.K. and Pasquale, E.B. (2011) ‘Eph receptors and ephrins in neuron–astrocyte communication at synapses’, Glia, 59(11), pp. 1567–1578. Available at: https://doi.org/10.1002/glia.21226. Murphy, K.R. et al. (2013) ‘Mapping the effects of ApoE4, age and cognitive status on 18F-florbetapir PET measured regional cortical patterns of beta-amyloid density and growth’, NeuroImage, 78, pp. 474– 480. Available at: https://doi.org/10.1016/j.neuroimage.2013.04.048. Nagy, Z.S. et al. (1995) ‘Influence of the apolipoprotein E genotype on amyloid deposition and neurofibrillary tangle formation in Alzheimer’s disease’, Neuroscience, 69(3), pp. 757–761. Available at: https://doi.org/10.1016/0306-4522(95)00331-C. Naslavsky, M.S. et al. (2022) ‘Global and local ancestry modulate APOE association with Alzheimer’s neuropathology and cognitive outcomes in an admixed sample’, Molecular Psychiatry, 27(11), pp. 4800–4808. Available at: https://doi.org/10.1038/s41380-022-01729-x. Nuriel, T. et al. (2017) ‘The Endosomal-Lysosomal Pathway Is Dysregulated by APOE4 Expression in Vivo’, Frontiers in Neuroscience, 11, p. 702. Available at: https://doi.org/10.3389/fnins.2017.00702. O’Connor, V. et al. (2004) ‘Differential Amplification of Intron-containing Transcripts Reveals Long Term Potentiation-associated Up-regulation of Specific Pde10A Phosphodiesterase Splice Variants *’, Journal of Biological Chemistry, 279(16), pp. 15841–15849. Available at: https://doi.org/10.1074/jbc.M312500200. Ollivier, M. et al. (2024) ‘Crym-positive striatal astrocytes gate perseverative behaviour’, Nature, 627(8003), pp. 358–366. Available at: https://doi.org/10.1038/s41586-024-07138-0. Parcon, P.A. et al. (2018) ‘Apolipoprotein E4 inhibits autophagy gene products through direct, specific binding to CLEAR motifs’, Alzheimer’s & Dementia: The Journal of the Alzheimer’s Association, 14(2), pp. 230– 242. Available at: https://doi.org/10.1016/j.jalz.2017.07.754. Peng, H. et al. (2010) ‘V3D enables real-time 3D visualization and quantitative analysis of large-scale biological image data sets’, Nature Biotechnology, 28(4), pp. 348–353. Available at: https://doi.org/10.1038/nbt.1612. Perocchi, F. et al. (2007) ‘Antisense artifacts in transcriptome microarray experiments are resolved by actinomycin D’, Nucleic Acids Research, 35(19), p. e128. Available at: https://doi.org/10.1093/nar/gkm683. Pohlkamp, T. et al. (2021) ‘NHE6 depletion corrects ApoE4-mediated synaptic impairments and reduces amyloid plaque load’, eLife, 10, p. e72034. Available at: https://doi.org/10.7554/eLife.72034. Polvikoski, T. et al. (1995) ‘Apolipoprotein E, dementia, and cortical deposition of beta-amyloid protein’, The New England Journal of Medicine, 333(19), pp. 1242–1247. Available at: https://doi.org/10.1056/NEJM199511093331902. Powell, D.S. et al. (2021) ‘The Relationship of APOE ε4, Race, and Sex on the Age of Onset and Risk of Dementia’, Frontiers in Neurology, 12. Available at: https://www.frontiersin.org/articles/10.3389/fneur.2021.735036 (Accessed: 29 August 2023). Prasad, H. and Rao, R. (2018) ‘Amyloid clearance defect in ApoE4 astrocytes is reversed by epigenetic correction of endosomal pH’, Proceedings of the National Academy of Sciences of the United States of America, 115(28), pp. E6640–E6649. Available at: https://doi.org/10.1073/pnas.1801612115. Puzzo, D. et al. (2009) ‘Phosphodiesterase 5 inhibition improves synaptic function, memory, and amyloid-beta load in an Alzheimer’s disease mouse model’, The Journal of Neuroscience: The Official Journal of the Society for Neuroscience, 29(25), pp. 8075–8086. Available at: https://doi.org/10.1523/JNEUROSCI.0864-09.2009. Quinlan, A.R. and Hall, I.M. (2010) ‘BEDTools: a flexible suite of utilities for comparing genomic features’, Bioinformatics, 26(6), pp. 841–842. Available at: https://doi.org/10.1093/bioinformatics/btq033. Rauch, J.N. et al. (2020) ‘LRP1 is a master regulator of tau uptake and spread’, Nature, 580(7803), pp. 381–385. Available at: https://doi.org/10.1038/s41586-020-2156-5. Reiman, E.M. et al. (2020) ‘Exceptionally low likelihood of Alzheimer’s dementia in APOE2 homozygotes from a 5,000-person neuropathological study’, Nature Communications, 11(1), p. 667. Available at: https://doi.org/10.1038/s41467-019-14279-8. Riddell, D.R. et al. (2008) ‘Impact of Apolipoprotein E (ApoE) Polymorphism on Brain ApoE Levels’, The Journal of Neuroscience, 28(45), pp. 11445–11453. Available at: https://doi.org/10.1523/JNEUROSCI.197208.2008. Rodriguez, G.A. et al. (2014) ‘Human APOE4 increases microglia reactivity at Aβ plaques in a mouse model of Aβ deposition’, Journal of Neuroinflammation, 11, p. 111. Available at: https://doi.org/10.1186/17422094-11-111. Saito, T. et al. (2014) ‘Single App knock-in mouse models of Alzheimer’s disease’, Nature Neuroscience, 17(5), pp. 661–663. Available at: https://doi.org/10.1038/nn.3697. Sala Frigerio, C. et al. (2019) ‘The Major Risk Factors for Alzheimer’s Disease: Age, Sex, and Genes Modulate the Microglia Response to Aβ Plaques’, Cell Reports, 27(4), pp. 1293-1306.e6. Available at: https://doi.org/10.1016/j.celrep.2019.03.099. Sando, S.B. et al. (2008) ‘APOE ε4 lowers age at onset and is a high risk factor for Alzheimer’s disease; A case control study from central Norway’, BMC Neurology, 8(1), p. 9. Available at: https://doi.org/10.1186/1471-2377-8-9. Sastre, M. et al. (2004) ‘Binding of cystatin C to Alzheimer’s amyloid beta inhibits in vitro amyloid fibril formation’, Neurobiology of Aging, 25(8), pp. 1033–1043. Available at: https://doi.org/10.1016/j.neurobiolaging.2003.11.006. Schindelin, J. et al. (2012) ‘Fiji: an open-source platform for biological-image analysis’, Nature Methods, 9(7), pp. 676–682. Available at: https://doi.org/10.1038/nmeth.2019. Schmechel, D.E. et al. (1993) ‘Increased amyloid beta-peptide deposition in cerebral cortex as a consequence of apolipoprotein E genotype in late-onset Alzheimer disease.’, Proceedings of the National Academy of Sciences, 90(20), pp. 9649–9653. Available at: https://doi.org/10.1073/pnas.90.20.9649. Serrano-Pozo, A. et al. (2015) ‘APOEε2 is associated with milder clinical and pathological Alzheimer disease’, Annals of Neurology, 77(6), pp. 917–929. Available at: https://doi.org/10.1002/ana.24369. Serrano-Pozo, A. et al. (2021) ‘Effect of APOE alleles on the glial transcriptome in normal aging and Alzheimer’s disease’, Nature aging, 1(10), pp. 919–931. Available at: https://doi.org/10.1038/s43587-021-00123-6. Shi, Y. et al. (2017) ‘ApoE4 markedly exacerbates tau-mediated neurodegeneration in a mouse model of tauopathy’, Nature, 549(7673), pp. 523–527. Available at: https://doi.org/10.1038/nature24016. Shi, Y. et al. (2021) ‘Overexpressing low-density lipoprotein receptor reduces tau-associated neurodegeneration in relation to apoE-linked mechanisms’, Neuron, 109(15), pp. 2413-2426.e7. Available at: https://doi.org/10.1016/j.neuron.2021.05.034. Sideris, D.I. et al. (2021) ‘Soluble amyloid beta-containing aggregates are present throughout the brain at early stages of Alzheimer’s disease’, Brain Communications, 3(3), p. fcab147. Available at: https://doi.org/10.1093/braincomms/fcab147. Simonovitch, S. et al. (2016) ‘Impaired Autophagy in APOE4 Astrocytes’, Journal of Alzheimer’s Disease, 51(3), pp. 915–927. Available at: https://doi.org/10.3233/JAD-151101. Simonovitch, S. et al. (2019) ‘The Effects of APOE4 on Mitochondrial Dynamics and Proteins in vivo’, Journal of Alzheimer’s disease: JAD, 70(3), pp. 861–875. Available at: https://doi.org/10.3233/JAD-190074. Sousa, C. et al. (2018) ‘Single‐cell transcriptomics reveals distinct inflammation‐induced microglia signatures’, EMBO reports, 19(11), p. e46171. Available at: https://doi.org/10.15252/embr.201846171. Spangenberg, E.E. et al. (2016) ‘Eliminating microglia in Alzheimer’s mice prevents neuronal loss without modulating amyloid-β pathology’, Brain: A Journal of Neurology, 139(Pt 4), pp. 1265–1281. Available at: https://doi.org/10.1093/brain/aww016. Stephen, T.L. et al. (2019) ‘APOE genotype and sex affect microglial interactions with plaques in Alzheimer’s disease mice’, Acta Neuropathologica Communications, 7(1), p. 82. Available at: https://doi.org/10.1186/s40478-019-0729-z. Stringer, C. et al. (2021) ‘Cellpose: a generalist algorithm for cellular segmentation’, Nature Methods, 18(1), pp. 100–106. Available at: https://doi.org/10.1038/s41592-020-01018-x. Strittmatter, W.J. et al. (1993) ‘Apolipoprotein E: high-avidity binding to beta-amyloid and increased frequency of type 4 allele in late-onset familial Alzheimer disease.’, Proceedings of the National Academy of Sciences, 90(5), pp. 1977–1981. Available at: https://doi.org/10.1073/pnas.90.5.1977. Sun, J.-H. et al. (2015) ‘Genetics of Vascular Dementia: Systematic Review and Meta-Analysis’, Journal of Alzheimer’s disease: JAD, 46(3), pp. 611–629. Available at: https://doi.org/10.3233/JAD-143102. Susanto, T.A.K. et al. (2015) ‘Cognition, Brain Atrophy, and Cerebrospinal Fluid Biomarkers Changes from Preclinical to Dementia Stage of Alzheimer’s Disease and the Influence of Apolipoprotein E’, Journal of Alzheimer’s Disease, 45(1), pp. 253–268. Available at: https://doi.org/10.3233/JAD-142451. Suzuki, K. et al. (2021) ‘Deletion of B-cell translocation gene 2 (BTG2) alters the responses of glial cells in white matter to chronic cerebral hypoperfusion’, Journal of Neuroinflammation, 18(1), p. 86. Available at: https://doi.org/10.1186/s12974-021-02135-w. Talbot, C. et al. (1994) ‘Protection against Alzheimer’s disease with apoE ∈2’, The Lancet, 343(8910), pp. 1432– 1433. Available at: https://doi.org/10.1016/S0140-6736(94)92557-7. Tcw, J. et al. (2022) ‘Cholesterol and matrisome pathways dysregulated in astrocytes and microglia’, Cell, 185(13), pp. 2213-2233.e25. Available at: https://doi.org/10.1016/j.cell.2022.05.017. Theparambil, S.M. et al. (2020) ‘Astrocytes regulate brain extracellular pH via a neuronal activity-dependent bicarbonate shuttle’, Nature Communications, 11(1), p. 5073. Available at: https://doi.org/10.1038/s41467-020-18756-3. Thomas, R. et al. (2016) ‘Epidermal growth factor prevents APOE4 and amyloid-beta-induced cognitive and cerebrovascular deficits in female mice’, Acta Neuropathologica Communications, 4(1), p. 111. Available at: https://doi.org/10.1186/s40478-016-0387-3. Tropea, T.F. et al. (2018) ‘APOE, thought disorder, and SPARE-AD predict cognitive decline in established Parkinson’s disease’, Movement Disorders: Official Journal of the Movement Disorder Society, 33(2), pp. 289–297. Available at: https://doi.org/10.1002/mds.27204. Tsuang, D. et al. (2013) ‘APOE ε4 increases risk for dementia in pure synucleinopathies’, JAMA neurology, 70(2), pp. 223–228. Available at: https://doi.org/10.1001/jamaneurol.2013.600. Ulrich, J.D. et al. (2018) ‘ApoE facilitates the microglial response to amyloid plaque pathology’, Journal of Experimental Medicine, 215(4), pp. 1047–1058. Available at: https://doi.org/10.1084/jem.20171265. Van Der Walt, S. et al. (2014) ‘scikit-image: image processing in Python’, PeerJ, 2, p. e453. Available at: https://doi.org/10.7717/peerj.453. Verghese, P.B. et al. (2013) ‘ApoE influences amyloid-β (Aβ) clearance despite minimal apoE/Aβ association in physiological conditions’, Proceedings of the National Academy of Sciences, 110(19), pp. E1807–E1816. Available at: https://doi.org/10.1073/pnas.1220484110. Vitek, M.P., Brown, C.M. and Colton, C.A. (2009) ‘APOE genotype-specific differences in the innate immune response’, Neurobiology of Aging, 30(9), pp. 1350–1360. Available at: https://doi.org/10.1016/j.neurobiolaging.2007.11.014. Wahrle, S.E. et al. (2005) ‘Deletion of Abca1 increases Abeta deposition in the PDAPP transgenic mouse model of Alzheimer disease’, The Journal of Biological Chemistry, 280(52), pp. 43236–43242. Available at: https://doi.org/10.1074/jbc.M508780200. Wang, C. et al. (2021) ‘Selective removal of astrocytic APOE4 strongly protects against tau-mediated neurodegeneration and decreases synaptic phagocytosis by microglia’, Neuron, 109(10), pp. 16571674.e7. Available at: https://doi.org/10.1016/j.neuron.2021.03.024. Wang, M., Zhao, Y. and Zhang, B. (2015) ‘Efficient Test and Visualization of Multi-Set Intersections’, Scientific Reports, 5(1), p. 16923. Available at: https://doi.org/10.1038/srep16923. Wegiel, J. et al. (2000) ‘Microglia cells are the driving force in fibrillar plaque formation, whereas astrocytes are a leading factor in plague degradation’, Acta Neuropathologica, 100(4), pp. 356–364. Available at: https://doi.org/10.1007/s004010000199. Wennberg, A.M. et al. (2018) ‘Association of Apolipoprotein E ε4 With Transactive Response DNA-Binding Protein 43’, JAMA neurology, 75(11), pp. 1347–1354. Available at: https://doi.org/10.1001/jamaneurol.2018.3139. West, H.L., William Rebeck, G. and Hyman, B.T. (1994) ‘Frequency of the apolipoprotein E ε2 allele is diminished in sporadic Alzheimer disease’, Neuroscience Letters, 175(1), pp. 46–48. Available at: https://doi.org/10.1016/0304-3940(94)91074-X. Wisniewski, T. et al. (1994) ‘Acceleration of Alzheimer’s fibril formation by apolipoprotein E in vitro’, The American Journal of Pathology, 145(5), pp. 1030–1035. Xia, Z. et al. (2024) ‘Co-aggregation with Apolipoprotein E modulates the function of Amyloid-β in Alzheimer’s disease’, Nature Communications, 15, p. 4695. Available at: https://doi.org/10.1038/s41467-02449028-z. Xian, X. et al. (2018) ‘Reversal of ApoE4-induced recycling block as a novel prevention approach for Alzheimer’s disease’, eLife, 7, p. e40048. Available at: https://doi.org/10.7554/eLife.40048. Xiao, H. and Peng, H. (2013) ‘APP2: automatic tracing of 3D neuron morphology based on hierarchical pruning of a gray-weighted image distance-tree’, Bioinformatics, 29(11), pp. 1448–1454. Available at: https://doi.org/10.1093/bioinformatics/btt170. Xiong, M. et al. (2021) ‘APOE immunotherapy reduces cerebral amyloid angiopathy and amyloid plaques while improving cerebrovascular function’, Science translational medicine, 13(581), p. eabd7522. Available at: https://doi.org/10.1126/scitranslmed.abd7522. Xiong, M. et al. (2023) ‘Astrocytic APOE4 removal confers cerebrovascular protection despite increased cerebral amyloid angiopathy’, Molecular Neurodegeneration, 18(1), p. 17. Available at: https://doi.org/10.1186/s13024-023-00610-x. Xu, Q. et al. (2006) ‘Profile and regulation of apolipoprotein E (ApoE) expression in the CNS in mice with targeting of green fluorescent protein gene to the ApoE locus’, The Journal of Neuroscience: The Official Journal of the Society for Neuroscience, 26(19), pp. 4985–4994. Available at: https://doi.org/10.1523/JNEUROSCI.5476-05.2006. Yang, H.-S. et al. (2018) ‘Evaluation of TDP-43 proteinopathy and hippocampal sclerosis in relation to APOE ε4 haplotype status: a community-based cohort study’, The Lancet. Neurology, 17(9), pp. 773–781. Available at: https://doi.org/10.1016/S1474-4422(18)30251-5. Ye, P. and D’Ercole, J. (1998) ‘Insulin-Like Growth Factor I (IGF-I) Regulates IGF Binding Protein-5 Gene Expression in the Brain*’, Endocrinology, 139(1), pp. 65–71. Available at: https://doi.org/10.1210/endo.139.1.5676. Yin, Z., Rosenzweig, N., et al. (2023) ‘APOE4 impairs the microglial response in Alzheimer’s disease by inducing TGFβ-mediated checkpoints’, Nature Immunology, 24(11), pp. 1839–1853. Available at: https://doi.org/10.1038/s41590-023-01627-6. Yin, Z., Herron, S., et al. (2023) ‘Identification of a protective microglial state mediated by miR-155 and interferon-γ signaling in a mouse model of Alzheimer’s disease’, Nature Neuroscience, 26(7), pp. 1196– 1207. Available at: https://doi.org/10.1038/s41593-023-01355-y. Youmans, K.L. et al. (2012) ‘APOE4-specific Changes in Aβ Accumulation in a New Transgenic Mouse Model of Alzheimer Disease*’, Journal of Biological Chemistry, 287(50), pp. 41774–41786. Available at: https://doi.org/10.1074/jbc.M112.407957. Yuniati, L. et al. (2019) ‘Tumor suppressors BTG1 and BTG2: Beyond growth control’, Journal of Cellular Physiology, 234(5), pp. 5379–5389. Available at: https://doi.org/10.1002/jcp.27407. Zamanian, J.L. et al. (2012) ‘Genomic analysis of reactive astrogliosis’, The Journal of Neuroscience: The Official Journal of the Society for Neuroscience, 32(18), pp. 6391–6410. Available at: https://doi.org/10.1523/JNEUROSCI.6221-11.2012. Zappia, L. and Oshlack, A. (2018) ‘Clustering trees: a visualization for evaluating clusterings at multiple resolutions’, GigaScience, 7(7), p. giy083. Available at: https://doi.org/10.1093/gigascience/giy083. Zeisel, A. et al. (2018) ‘Molecular Architecture of the Mouse Nervous System’, Cell, 174(4), pp. 999-1014.e22. Available at: https://doi.org/10.1016/j.cell.2018.06.021. Zenz, R. et al. (2008) ‘Activator protein 1 (Fos/Jun) functions in inflammatory bone and skin disease’, Arthritis Research & Therapy, 10(1), p. 201. Available at: https://doi.org/10.1186/ar2338. Zhang, Q. et al. (2023) ‘NDRG2 regulates the formation of reactive astrocyte-derived progenitor cells via Notch signaling pathway after brain traumatic injury in rats’, Frontiers in Molecular Neuroscience, 16, p. 1149683. Available at: https://doi.org/10.3389/fnmol.2023.1149683. Zhang, Y. et al. (2014) ‘An RNA-sequencing transcriptome and splicing database of glia, neurons, and vascular cells of the cerebral cortex’, The Journal of Neuroscience: The Official Journal of the Society for Neuroscience, 34(36), pp. 11929–11947. Available at: https://doi.org/10.1523/JNEUROSCI.186014.2014. Zhao, L. et al. (2016) ‘Intracerebral adeno-associated virus gene delivery of apolipoprotein E2 markedly reduces brain amyloid pathology in Alzheimer’s disease mouse models’, Neurobiology of Aging, 44, pp. 159– 172. Available at: https://doi.org/10.1016/j.neurobiolaging.2016.04.020. Zheng, G.X.Y. et al. (2017) ‘Massively parallel digital transcriptional profiling of single cells’, Nature Communications, 8(1), p. 14049. Available at: https://doi.org/10.1038/ncomms14049. Zweifel, S. et al. (2018) ‘HOPX Defines Heterogeneity of Postnatal Subventricular Zone Neural Stem Cells’, Stem Cell Reports, 11(3), pp. 770–783. Available at: https://doi.org/10.1016/j.stemcr.2018.08.006.