Phospholipid Composition of APOE Lipoproteins Affects Microglia Activation in an Isoform-specic Manner

Apolipoprotein E4 (APOE) is the strongest genetic risk factor for Alzheimer’s disease (AD). Our lipidomic analysis identied a common phospholipid signature with a high level of correlation between APOEε3/3 and APOEε4/4 AD postmortem brain samples and native lipoproteins isolated from astrocyte conditioned media of mice expressing human APOE3 or APOE4. Behavioral testing demonstrated that native E3 lipoproteins were more effective than E4 at ameliorating the harmful effects of Aβ on cognition. We posit that APOE isoform-specic differences in the phospholipid composition of native lipoproteins prompt a differential microglial response. Using time-lapse in vivo two-photon imaging we compared the effect of E3 or E4 infused with Aβ and determined that E3 lipoproteins induced a faster microglial migration towards Aβ. To determine how E3 and E4 lipoproteins affect microglial transcriptome in response to Aβ we performed bulk and single cell RNA-seq of WT and Trem2 ko mice. We show that compared to E4, cortical infusion of E3 lipoproteins upregulated a higher proportion of genes associated with an activated immune response accompanied by a downregulation of homeostatic genes. scRNA-seq identied microglia-specic clusters affected by Trem2 deciency suggesting that lack of Trem2 impairs the transition of microglia from homeostatic to an activated state. Compared to E3, E4-expressing microglia showed a reduced Aβ uptake that was additionally aggravated by Trem2 deciency. Together, our ndings have elucidated unique phenotypic and transcriptional differences in the microglial response to Aβ in the presence of E3 or E4 lipoproteins which could impact AD pathogenesis. that the infusion of Aβ with E3 lipoproteins induced a more rapid microglial response than with E4 lipoproteins. Behavioral testing following cortical Aβ infusion demonstrated that both native lipoproteins ameliorated Aβ deleterious effects on cognition; with APOE3 exhibiting a higher ecacy than APOE4. To determine how E3 and E4 lipoproteins affect microglia expression in response to Aβ we performed bulk and single cell RNA-seq (scRNA-seq). Collectively, the data showed that the addition of E3 lipoproteins to Aβ upregulated a higher proportion of DAM genes and is associated with more active transcriptomic response than E4. scRNA-seq identied microglia-specic clusters affected by Trem2 deciency suggesting that lack of Trem2 impairs the transition of microglia from homeostatic to activating state. Compared to E3, E4-expressing microglia showed a reduced Aβ uptake that was additionally aggravated by Trem2 deciency. Together, our ndings have elucidated unique phenotypic and transcriptional differences in the microglial response to Aβ when in the presence of either E3 or E4 lipoproteins. using A and B gradient solvents mM ammonium triethylamine (A - v/v/v) and B -propanol:hexane:water The column was eluted for 0.5 min isocratically at 25% B, then from 0.5 to 6.5 min with a linear gradient from 25% to 40% solvent B, from 6.5–25 min using a linear gradient of 40-55% solvent B, from 25-38 min with a linear gradient of 55-70% solvent B, from 38-48 min using a linear gradient of 70%-100% solvent B, isocratically from 48-55 min at 100% solvent B followed by a return to initial conditions from 55-70 min from 100% to 25% B. The at 25% B for additional 5 min. ion mode a of for the full MS scan in a data-dependent scan range for MS analysis injection and Aβ uptake assay were performed using WT, E3, E4, E3/Trem2 ko and E4/Trem2 ko pups (1-3 days old) as described before 42, 43 . Briey, the cortices and hippocampi were mechanically dissociated using a sterile Pasteur pipette. The dissociated cells were plated in 75 cm 2 asks with DMEM/F12 medium (Thermo Fisher) containing 10% FBS at DIV0. 24 h after plating (DIV1), the media was replaced to remove debris. Microglia were collected by tapping at DIV14 and plated on 0.01% poly-L-lysine (Sigma) coated 12mm circular coverslips in 24-well plates with the density of 60,000 cells/well. 24 h after plating, cells in the treatment groups were treated with 1µM Hi-Lyte TM Fluor 488-labeled Aβ (Anaspec) at 37 °C for 1 h. Following the treatment, cells were washed three times with PBS, xed in 4% PFA and permeabilized with 0.2% Triton X-100 at room temperature. Microglia were then labeled with Anti-IBA1 antibody (1:500) at 4 °C for 18 h and followed by 2 h incubation with horse anti-rabbit 594 secondary antibody and stained with DAPI to visualize nuclei.


Introduction
Late-onset Alzheimer's disease (AD) is age-related dementia characterized by amyloid β (Aβ) plaques, neuro brillary tangles and cognitive decline 1 . Inheritance of Apolipoprotein (APOE) ε4 allele is the major genetic risk factor for late-onset AD, modifying disease risk and progression in an isoform dependent manner (ε2 < ε3 < ε4, with ε4 the highest risk) 2, 3 . Isoform differences in amino acids at position 112 affects lipid binding properties of APOE and at 158 the receptor-binding a nity to APOE receptors 4 . APOE is a major lipid transporter and the structural differences result in different lipid binding properties. Thus, APOE4 preferentially binds lower density lipoproteins (LDL, VLDL) whereas APOE3 and APOE2 high density lipoproteins (HDL) 5 . In a recent study we applied Shotgun Lipidomics to measure the major phospholipid classes and their molecular spices in brains of AD patients comprising of different APOE genotypes 6 . Our data demonstrated that few of these phospholipids were signi cantly lower in APOEε4/c carriers vs APOEε2/c carriers 6 . In the brain, APOE is primarily secreted by astrocytes and to a lesser extent microglia in a lipid-poor form 7 . ABCA1 transporter mediates the transfer of phospholipids and cholesterol to lipid-free APOE thus forming discoidal high-density lipoprotein (HDL) particles (reviewed in 8 ). APOE3 and APOE4 lipoproteins isolated from astrocytes conditioned media (referred to as E3 and E4 lipoproteins) are composed of proteins, cholesterol and phospholipids; and de cient of cholesteryl esters 9 . It has been reported that APOE lipidation state signi cantly impacts Aβ aggregation 10 and Aβ clearance at the blood-brain barrier (BBB) 11 . APOE4 isoform is associated with impaired Aβ uptake, clearance and degradation 12 resulting in higher amyloid load in APP transgenic mice 13,14 . Interestingly, a study by Verghese et al. suggested that APOE in uences Aβ clearance in vivo not through direct binding but via its interaction with other cell surface receptors 15 .
Microglia can have bene cial effect assisting Aβ clearance and by forming a barrier surrounding amyloid plaques but also they could be detrimental, causing chronic neuroin ammation (reviewed in 16,17 ). Recent comprehensive single-cell RNA sequencing (scRNA-seq) analysis in mouse models of neurodegeneration discovered a subset of microglia named disease-associated microglia (DAM) 18,19 . The upregulation of DAM genes is usually accompanied by a decreased expression of homeostatic microglia genes responsible to maintain the normal microglia function of surveillance. Among DAM genes are gene variants associated with increased risk for AD, such as APOE and TREM2. It has been reported that TREM2 can recognize a variety of ligands including APOE, HDL and LDL which could affect microglial interaction with Aβ and APOE 20,21,22 . APOE and TREM2 are essential for the development of microglia barrier around the plaques 23 . Recently, we demonstrated that the lack of TREM2 differentially affects the phenotype and transcriptome of human APOE3 and APOE4 expressing mice 24 . TREM2 de ciency increased plaque growth concurrently to the impairment of microglia barrier, an effect most pronounced at earlier stages of amyloid deposition. Interestingly, lack of Trem2 signi cantly decreased plaqueassociated APOE protein only in APOE4 but not in APOE3 mice in agreement with gene expression data.
In this study, after performing Lipidomic analysis of native E3-and E4-lipoproteins and brain samples from APOEε3/3 and APOEε4/4 AD patients, we identi ed a common phospholipid signature between their major phospholipid classes. We hypothesized that isoform-speci c phospholipid composition of E3 and E4 native lipoproteins would elicit distinct phenotypic and transcriptomic response by microglia. We used in vivo two-photon imaging and determined that the infusion of Aβ with E3 lipoproteins induced a more rapid microglial response than with E4 lipoproteins. Behavioral testing following cortical Aβ infusion demonstrated that both native lipoproteins ameliorated Aβ deleterious effects on cognition; with APOE3 exhibiting a higher e cacy than APOE4. To determine how E3 and E4 lipoproteins affect microglia expression in response to Aβ we performed bulk and single cell RNA-seq (scRNA-seq).
Collectively, the data showed that the addition of E3 lipoproteins to Aβ upregulated a higher proportion of DAM genes and is associated with more active transcriptomic response than E4. scRNA-seq identi ed microglia-speci c clusters affected by Trem2 de ciency suggesting that lack of Trem2 impairs the transition of microglia from homeostatic to activating state. Compared to E3, E4-expressing microglia showed a reduced Aβ uptake that was additionally aggravated by Trem2 de ciency. Together, our ndings have elucidated unique phenotypic and transcriptional differences in the microglial response to Aβ when in the presence of either E3 or E4 lipoproteins.

Results
Common phospholipid signature between APOEε4/4 and APOEε3/3 AD brains and APOE3 and APOE4 native lipoproteins Previously we have shown that there is a signi cant APOE isoform-speci c difference in phospholipid content in the brains of AD patients 6 . In the present study, we compared the phospholipid pro le of native APOE lipoproteins isolated from conditioned media of APOE3 and APOE4 primary astrocytes to those of APOEε4/4 and APOEε3/3 brains samples from AD patients. To measure the most abundant nine phospholipid classes in human brain and the native APOE lipoproteins we used Multi-Dimensional Mass Spectrometry Shotgun Lipidomics (MDMS-SL) 6 . We found that out of nine, ve phospholipid classes (PE, PI, PS, SM and PA) had a statistically higher level in APOEε3/3 vs APOEε4/4 brains (Fig. 1a). Overall, with the exception of PG, APOE-containing lipoproteins had a similar direction of fold change for each of class as the human samples (Fig. 1b). In the lipoprotein data set, we found that PI, PS, SM and LPC were statistically higher in E3 vs E4 lipoproteins (for the abbreviation see the legend of Fig. 1). Lipoproteins derived from E3/Abca1 het and E4/Abca1 het astrocytes 13 were used as negative controls and as expected demonstrated half of the phospholipid content of their wild-type counterparts (Suppl. Fig. 1). As shown on Fig. 1c-d, 103 lipid species were commonly identi ed in the human and lipoprotein samples, most of which were affected in APOE isoform-speci c manner in both datasets and were higher in E3 samples.
Correlation analysis (Fig. 1e) demonstrated a very strong correlation between the phospholipid pro le of human and native lipoprotein samples.
To con rm the results from Shotgun Lipidomics of APOE native particles, we applied liquid chromatography-mass spectrometry (LC-MS) and measured ve of the phospholipid classes reported to comprise the bulk of brain APOE-containing lipoproteins 9 namely: PC, PE, PI, and PS; while CL was used as a control. The tSNE clustering shown on Fig. 1f demonstrated that the phospholipid molecular species from the two APOE isoforms segregated in separate groups forming distinct clusters. Similar to the MDMS-SL results, we found that the negatively charged phospholipids (PI, PE and PS) were signi cantly higher in E3 than in E4 lipoproteins but CL was unchanged ( Fig. 1g-h). Additionally, speci c subspecies of PE and PS (PE36:1, PE36:2, PS36:1, and PS40:6) represented a large portion of the total amount of those classes and were found in signi cantly less abundance in APOE4 lipoproteins. These speci c subspecies may represent potential lipid activation signals which impact receptor binding and could activate signal transduction events.
Since the diameter of lipoproteins is commonly used to represent their lipidation 25 , we used nondenaturing native gel electrophoresis (Fig. 1i) and dynamic light scattering assay (Fig. 1j) to determine the diameter of APOE native particles. The results demonstrated that the size of E3 native lipoproteins peaked at 13 nm while the size of E4 particles at 12.5 nm (Fig. 1j). The native gels (Fig. 1i), localized APOE within three populations of particles, migrating around 12, 9 and 8 nm. In agreement with a previous study on HDL, the measurements allowed us to conclude that particles with a diameter between 12 and 9 nm corresponded to discoidal a-HDL and are lipidated, while particles with a size of 8 nm corresponded to pre-β-HDL and are lipid poor/non-lipidated 26 . SDS NuPAGE (Fig. 1i, bottom panel) con rmed that an equal amount of APOE protein was used for native gels. Lastly, Suppl. Fig. 1k shows that the cholesterol content of E3 lipoproteins was slightly but signi cantly increased compared to E4.

APOE lipoproteins rescue Aβ-induced memory impairments
Prior studies demonstrated that the injection of oligomeric Aβ into brain caused signi cant memory de cits 27,28 . First we tested the effect of native E3 and E4 lipoproteins on Aβ aggregation as before 29 . As visible from Supp. Fig. 2a-c, Aβ incubation with either E3 or E4 native lipoproteins greatly reduced the formation of oligomers. Next, to examine if the addition of native APOE lipoproteins to Aβ ameliorate Aβinduced memory de cits, Aβ+/-APOE (referred to as AbE3 or AbE4) were infused into the cortex of young WT mice followed by behavioral testing as shown on Fig. 2a-d. For both tests, Ab oligomers signi cantly worsened cognitive performance compared to their corresponding negative control ( Fig. 2b and d). In contrast, injections of both AbE3 or AbE4 signi cantly ameliorated the effects of Ab on cognitive performance. Importantly, the cognitive performance in mice infused with AβE4 remained signi cantly worse compared to their AβE3 counterparts when tested by novel object recognition. For both test, behavior controls showed no signi cant differences between the groups (Suppl. Fig. 2d-f). The conclusion is that both E3 and E4 native lipoproteins decrease Aβ aggregation and ameliorate its deleterious effects on cognitive performance, with E3 demonstrating a better e cacy.
To investigate the effect of AβE3 and AβE4 infusions on transcriptome, we injected them in the cortex and performed RNA-seq using sorted microglia and neurons. There were no differentially expressed genes (DEG) when comparing AβE3 vs AβE4 neuronal transcriptomes (Fig. 2e). In contrast, there were 1792 differentially expressed genes (DEG) in microglia: 722 upregulated in mice injected with AβE3 and 1072 genes upregulated in AβE4 injected mice (Fig. 2f, Suppl. Table 2). As shown on Fig. 2g-h, upregulated in AβE3 vs AβE4 microglia were many microglia-speci c genes (Tmem119 and Trem2) as well as genes and biological processes associated with microglia activation such as cytokines (Ccl4) and catepsins (Ctsd).
We also examined the morphology of microglia surrounding the injection site using IBA1 staining and Imaris lament module to trace microglial projections ( Fig. 2i-j). The heatmap of microglial projections around the infusion site demonstrated that microglia in the AβE3 group have signi cantly increased branch length and branch points (bar graph insets) and the projections were oriented more toward the infusion site (the area between heat maps) when compared to the AβE4 group. We also performed macrophage speci c F4/80 immunohistochemistry analysis. Fig. 2k-m shows an increased F4/80/Adgre1 immunostaining around the infusion site in mice injected with AβE3 compared to AβE4, which is consistent with the overall reduction in Adgre1 gene expression seen in AβE4 microglia. Thus, the results from gene expression and microglia morphology suggest that AβE3 induce stronger microglial response when compared to AβE4.
To determine if the presence of native E3 and E4 lipoproteins affects microglia function in vivo, we used Cx3cr1 GFP mice that express EGFP protein in microglia and peripheral monuclear cells and uorescentlylabeled Ab. First we tested the lipoproteins impact on Aβ uptake of microglia by injecting AbE3 or AbE4 into the cortex and performing ow cytometry at 4 and 24 h post-injection ( Fig. 3a-c, Suppl. Fig. 3a-d).
The cell suspension was separated on microglia without Ab (green signal, single+) and microglia that has engulfed Ab (green + red signal, dual+). As shown on Fig. 3b-c, E3 lipoproteins were more e cient than E4 in facilitating Ab uptake by microglia at 24 h, but not at 4 h after the injection.
To further examine if APOE native lipoproteins have an impact on the microglial response to infusion of Aβ, we employed in vivo two-photon imaging in Cx3cr1 GFP mice. For each animal, ZT-stack images were analyzed to quantify the microglial migration towards the infusion site and microglial coverage of the infusion site over time (Fig. 3d-f and Suppl. Fig. 3e-f). The representative images ( Fig. 3d and Suppl. processes towards the AβE3 infusion over the rst 40 minutes. Once the processes reached the AβE3, microglia were able to form lamellipodia that covered the AβE3 throughout the remainder of the experiment (40-80 min). In contrast, microglia demonstrated a decreased capacity to extend processes towards the AβE4 as well as encapsulate AβE4 infusion. The results demonstrate that the infusion of Aβ together with E3 lipoproteins, unlike with E4, induces a more rapid response by microglia to isolate Aβ that suggests a protective mechanism diminishing the spread of Aβ.
E3 and E4 lipoproteins differentially affect microglia transcriptome in response to Aβ infusion Next, we performed RNA-seq on two FACS sorted microglial populations (dual+ and single+ microglia) from Cx3cr1 GFP mice injected with AbE3 or AbE4 (see Fig. 4a for the design). We assume that dual+ cells are in an active stage compared to their single+ counterparts which are either resting or in transition from homeostatic to active state. The goal was to examine how APOE isoform and the post-injection interval affects this transition. tSNE plots demonstrate that the expression pro les of single+ and dual+ microglia cluster into sharply de ned groups for both time points (Fig. 4b). Suppl. Fig. 4a-b demonstrates that numerous overlapping DEGs were commonly upregulated for the two time points in dual+ vs single+ microglia. For each single+ and dual+ populations, we identi ed genes with a similar level of expression at both time points (unchanged, grey), a group with higher expression at 4h that declined in time (upregulated at 4h, blue), and a third group (red) with higher expression at 24h (Suppl. Fig. 4c-f). Unchanged in dual+ cells were genes activated from the start with continuous expression and related to interferon signaling (I 204) or lipid transport (Lpl). Early responding genes upregulated in dual+ at 4 vs 24 h were associated with in ammatory response, chemokine-mediated signaling and chemotaxis (Suppl. Fig. 4c-d, blue). Genes that were upregulated at the later time point in dual+ microglia (late responding) were associated with protein folding, ribosome biogenesis and glycolysis consistent with the increased energy demands of processing the engulfed Ab (Suppl. Fig. 4c-d, red color).
As seen from the tSNE plot on Fig. 4b, APOE isoform-speci c effect on microglial transcriptomes was more pronounced at 4 h. The examination of DEGs in dual+ vs single+ microglia in both treatments revealed a higher number of total DEGs and speci cally a higher number of upregulated genes in dual+ AbE3 microglia than in dual+ AbE4 microglia ( Fig. 4c-d, 2760, and 1532 respectively, pie charts dark purple and dark green). For both treatments, there were many shared/common DEGs upregulated in dual+ vs single+ ( Fig. 4e-f, grey boxes). In both AβE3 and AβE4 dual+ microglia, there was an overall upregulation of DAM genes (186 genes) and downregulation of the homeostatic genes (212 genes). Commonly upregulated in dual+ microglia were DAM genes (Apoe, Axl) and downregulated classical homeostatic genes (P2ry12, and Tmem119) indicating a pro le characteristic for microglia in an active state ( Fig. 4g-h). Commonly upregulated in dual+ were also receptors (Clec4 cluster, Marco, Axl and Gas6), cytokines (9 genes) and chemokines (12 genes) and genes related to phagocytosis and lysosomes (Cd68). Surprisingly, Trem2 as well as few other DAM genes (Csf3r, Cebpa and Ctss) had a higher expression in single+ microglia (respectively downregulated in dual+) in both treatments and at both time points. This suggest that Trem2 is more associated with transition than fully activated state. Uniquely upregulated DAM genes for AβE3 were genes associated with lysosome/autophagy process (Lamp1 and Bloc1s1), glycosylation (St6gal1), which was a glycosyltransferase implicated by GWAS with AD 30 , cytokines (Ccl6) and receptors (Clec4e and Clec12a). DAM genes uniquely upregulated in AβE4 were Capg (binds PIP2 and affects actin dynamic in macrophages 31 ), Eef2 and Eef1b2 (regulate translation 19 ) as well as Ly9 (lymphocyte antigen 9 32 ). Overall, there was a higher number of DAM genes upregulated in dual+ AbE3 (310) than in dual+ AbE4 (198) (Fig. 4g-h) suggesting that E3 could initiate a stronger response by microglia than E4.

Single cell sequencing informs on distinct microglial populations in response to Aβ injection
To test how Trem2 de ciency affects acute microglial response to AβE3 or AβE4 infused into the cortex of WT and Trem2 ko mice we used scRNA-seq (see Fig. 5a for the design). We collected a total of 49,817 sequenced cells with 2,228 median genes expressed and 5,234 median unique molecular identi ers (UMIs) per cell. Non-microglial cells were identi ed using cell type-speci c genes and excluded from the analysis (Fig. 5b-c, Suppl. Fig. 5) and microglial cells were re-clustered to catalog transcriptionally distinct subpopulations. Thus, we re-clustered 36,244 microglial cells into ve clusters and performed trajectory analysis to identify distinct activation states in response to the infusion ( Fig. 5d and Suppl. Fig. 6). We determined three potential trajectories (clusters 2, 3, and 4) from clusters 1a and 1b (Fig. 5d).
To characterize heterogeneity within microglia, we identi ed cluster-speci c genes that are differentially expressed between clusters (Fig. 5e). As seen in Fig. 5e, the microglia in cluster 1a and 1b had increased expression of homeostatic signature genes such as P2ry12, Tmem119, and Siglech 19,33 . However, cluster 1b showed an increase from the baseline (cluster 1a) of the expression of genes related to microglial activation such as Apoe, Cst7, Trem2, and Tyrobp. Thus, we characterized cluster 1b as a transitional state. Interestingly, in WT mice Trem2 had the highest expression in cluster 1b suggesting its involvement in the transition from homeostatic to active state. Genes associated with activated microglial were upregulated in clusters 2, 3, and 4 and accompanied by downregulation of the homeostatic microglial genes. This suggests that clusters 2, 3, and 4 form in response to infusion of AβE3 or AβE4 and present the active microglia state. Microglia in cluster 2 showed upregulation of genes related to chemotaxis (Ccl12, Ccl3, Ccl4, Cxcl16, Spp1, and Lgals3), in ammatory cytokine (Mif), lipoprotein lipase (Lpl) and ribosomal components. Microglia in cluster 3 showed upregulation of pro-in ammatory genes, especially a set of genes important in NF-B signaling such as Tnf, Il1b, and Ptgs2, as well as immediate early genes (Fos, Jun, and Egr1) and cytokines (Ccl3 and Ccl4). In cluster 4, genes related to the interferon signaling pathway including I tm3, I 27l2a, I t1, Irf7 as well as Apoe were highly expressed with top enriched GO term "immune system process" (Fig. 5e, Suppl. Table 3). In both genotypes, the highest proportion of cells was found in cluster 1b (transitional state) and cluster 2 (active state). We observed an increased proportion of cells in cluster 1b and reduced in cluster 2 in Trem2-de cient microglia compared to WT (Fig. 5f). This suggests that Trem2-de ciency impairs the transition of microglia from homeostatic to activating state particularly from cluster 1b to cluster 2.
scRNA-seq identi es activated gene clusters downregulated in Trem2 ko mice following Aβ infusion To identify differential transcripts, we compared microglia from Trem2 ko and WT mice and observed a higher number of DEGs in clusters 2, 3, and 4 associated with active microglia (Fig. 6a, Suppl. Table 4). Overall, there was reduced expression of immune response genes in Trem2 de cient microglia compared to WT microglia ( Fig. 6b-d). In clusters 2, 3, and 4, DAM genes comprised 30-44% of the total number of differentially expressed genes that were upregulated in WT vs Trem2 ko and this was accompanied by the same decrease of homeostatic microglia genes ( Fig. 6b-d) suggesting that microglia from Trem2 ko mice cannot adequately respond to the AβE3/AβE4 infusion challenge. In Cluster 2 top downregulated genes in Trem2 ko microglia were Lpl, Spp1, Cd9 and opposed by upregulation of canonical homeostatic genes such as P2ry12 (Fig. 6b). Clusters 3 (Fig. 6c) and Cluster 4 ( Fig. 6d) exhibited the largest number of down regulated genes in Trem2 ko microglia and also showed the highest mismatch between WT and Trem2 ko microglia in terms of homeostatic/DAM gene expression (see the bar graphs in Fig. 6c and d).
We validated the scRNA-seq data using uorescent in situ hybridization (FISH) for Tmem119 and Adgre1 and spatially visualized their expression near the infusion site. We selected Tmem119 as a canonical microglia-speci c gene that was not affected by Trem2 de ciency and Adgre1 (F4/80) was selected because it was signi cantly downregulated in Trem2 ko microglia and we showed to be increased near the infusion site (see Fig. 2k-i). Similarly, to the scRNA-seq data we found a signi cant reduction in gene expression of Adgre1 in Trem2 ko mice and no effect on Tmem119 level ( Fig. 6e-g). Thus, FISH performed near the infusion site con rms the result for selected genes in the scRNA-seq data sets.

Trem2 de ciency and APOE isoform affects microglia response to Aβ
We next examined if the response to AβE3 and AβE4 infusion is APOE isoform-and/or Trem2-dependent.
We compared microglia from AβE3 and AβE4 injected WT and Trem2 ko mice separately within clusters 2, 3 and 4 which correspond to active microglia ( Fig. 7a-c). We determined that in WT mice, clusters 2 and 3 contained a higher number of DEGs as well as DAM genes in AβE3 treated mice than in AβE4 and the opposite was detected for cluster 4. For all clusters, while AβE3 injection upregulated a comparable number DEGs in Trem2-de cient microglia as in WT, AβE4 injection resulted in signi cantly less DEGs in Trem2 ko microglia ( Fig. 7a-c). In cluster 2, Trem2 ko mice injected with AβE4 had a higher expression of classic homeostatic genes such as P2ry12 and P2ry13 vs their E3 counterparts (Fig. 7a, Trem2 ko plot) and this was not observed in WT microglia (Fig. 7a, WT plot). On the other hand, there were many overlapping DEGs commonly upregulated in WT and Trem2 ko mice injected with AbE3 such as Spp1 (Fig.  7a). In cluster 4, there were more homeostatic genes (Cx3cr1, Csf1r) upregulated in WT-AβE3 mice than in WT-AbE4 mice (Fig. 7c) and the opposite was observed for DAM genes that were upregulated in WT-AbE4 mice (I 27l2a, Eef1g). However, for Cluster 4 we identi ed very few DEGs in Trem2 ko mice. Overall, these results suggest that Trem2 de ciency may further exaggerate E4 lipoprotein insu ciency to activate microglia.
To test how Trem2 de ciency and APOE native particles affect Aβ uptake we used primary microglia from mice expressing human E3 and E4; and compared to their Trem2 ko counterparts (E3/Trem2 ko , or E4/Trem2 ko ). As seen on Fig. 7d-f, both APOE4 isoform and Trem2 status signi cantly decreased Aβ uptake. As visible, the overall reduction of Aβ uptake by E4/Trem2 ko microglia was reduced 5-fold compared to E3 expressing microglia. Interestingly, Trem2 de ciency signi cantly affected the morphology of the microglia as its shape become more circular with less processes (Fig. 7f). These results con rm that E3 is more successful than E4 at promoting Aβ uptake and that Trem2 deletion further aggravates E4 mediated ine ciency.

Discussion
In this study, we identi ed a common phospholipid signature between major phospholipid classes of native E3 and E4 lipoproteins and brain samples from APOEε3/3 and APOEε4/4 AD patients. Our results support the hypothesis that native E3 lipoproteins trigger a more effective microglial response to Aβ than E4 in ameliorating Aβ deleterious effect. We posit that the divergence in phospholipid content of E3 and E4 native lipoproteins affects Aβ binding to microglia receptors in isoform-speci c manner and consequently have a differential impact on microglia transcriptome (Fig. 8). Consequently, E3 induces a faster microglial response than E4 characterized by rapid process extension allowing Aβ to be isolated and engulfed. Lastly, behavioral testing following Aβ injections demonstrated that E3 native lipoproteins exhibited a better protection than E4 in diminishing Aβ effects on memory.
The quantitative assessment of the phospholipid composition of E3 and E4 native lipoproteins and brain samples from APOEε3/3 and APOEε4/4 AD patients demonstrated a signi cant APOE isoform-speci c difference in phospholipid pro les. Interestingly, our analysis demonstrates a very strong correlation of the phospholipid pro le of human and native lipoprotein samples (Fig. 1e). It should be noted that the human brain samples include lipids that are structural part of cellular membranes as well as lipoproteins in interstitial uid. Regardless that the source of the phospholipids was not entirely the same, it was remarkable to discover such high level of correlation between the abundance and the direction of fold change between the phospholipids of human samples and the native lipoproteins isolated from astrocyte conditioned media. As the major lipid carrier in the brain APOE lipoproteins play an important role in the chaperoning of cholesterol and phospholipids between cells and serving as ligand for multiple immune receptors such as TREM2. It has been shown that TREM2 signaling is activated by phospholipids such as phosphatidylinositol and phosphatidylcholine 34 and TREM2 binds APOE as well as LDL and HDL 22 . Our lipidomic results demonstrate that four major phospholipid classes including neutral (PC and PI) and negatively charged (PE and PS) are signi cantly reduced in E4 lipoproteins when compared to E3 but not cardiolipin. Sensing lipids is important for microglia activation and affects functions such as chemotaxis and phagocytosis (reviewed in 35 ). Increasing evidence suggests that phospholipid levels decrease with aging as well we during AD progression 36 , making the initial de cit we have identi ed in E4 particles even more detrimental with the progression of the pathology. The reduced lipidation of E4 compared to E3 native particles may impact Aβ aggregation which in turn could affect cognitive performance and microglia activation state 37 . Numerous studies demonstrated APOE binds to Ab with high a nity and most agree that there is an isoform-speci c effect on Ab aggregation E4>E3>E2 38,39 . Furthermore, studies have suggested that E3 forms more stable complex with Ab than E4 39 however in the present study we did not examine the stability of Ab-APOE complex. Our data demonstrated that both E3 and E4 native lipoproteins precluded the formation of Ab oligomers (Suppl. Fig. 2a-b). However, we cannot exclude a possible presence of occasional Ab oligomers and potentially an isoform-speci c effect on Ab oligomerization that could be detected using quantitative methods with higher resolution. In con rmation of such possibility is the fact that E3 was more effective than E4 in rescuing cognitive de cits caused by Ab oligomers (Fig. 2a-d).
In terms of APOE effect on microglia we found that E3 native lipoproteins are more effective than E4 in triggering microglia response to Ab injections. The transcriptomic pro le of microglia isolated from AβE3 injected mice have a higher enrichment of gene expression networks associated with innate immune response compared to microglia from AβE4 injected. To address the transcriptional changes in phagocytic and non-phagocytic microglia we compared dual+ vs single+ at 4-and 24-hours postinjection. Transcriptionally, the dual+ populations are strikingly different from single+ at both 4-and 24hours post-injection, representing an activated state with upregulation of DAM genes and downregulation of homeostatic genes. We further sub-divided dual+ population into early responders (4h; Ccl9, I t1, and Ctsb), genes with continuous expression (I 204, Lpl, and Il10), or late responders (24h; Adgre1, Mif, and Ccl2). By separating of microglia on dual+ and single+ we conclude that that the APOE-isoform-speci c effects are more pronounced at earlier points of the injury that may re ect involvement in the initial recognition of Ab+APOE complexes by microglia. Furthermore, in Cx3cr1 mice (Fig. 4), AβE3 injections were associated with more active transcriptomic pro le comprised by differential increase of DAM compared to AβE4 injections.
These transcriptional changes are accompanied by phenotypic changes in microglial morphology exhibited by increased process orientation towards the infusion site in AβE3 vs AβE4 injected mice and an increase in microglia which colocalized with Aβ. We con rmed this using in vivo two-photon imaging, demonstrating that within the rst hour of infusion microglia exposed to AβE3 extended processes toward Aβ faster than AβE4. Thus, in brains injected with AβE3, microglia can cover and envelop Aβ with better e ciency that microglia from AβE4 injected brains (Fig. 3). Finally, in vivo and in vitro experiments show that E3-native particles are more effective than E4 at promoting Aβ uptake ( Fig. 3 and 7). Our interpretation of these ndings is that E3 is better at mobilizing microglia to form a protective barrier surrounding the neuronal damage caused by the Aβ injection and restrict its spread by increasing Aβ uptake.
Recently we reported that microglia-plaque coverage of small plaques is more e cient at restricting plaque-growth than of larger plaques 24 , and this effect is modulated by both APOE isoform as well as TREM2 status. This emphasizes the importance of the early response by microglia to amyloid and implicates the phenotypic response seen in the rst 24 hours in the AβE3 mice as critical to the long-term protection from amyloid buildup and eventual neuronal damage. Studies addressing the microglia phenotypic response in TREM2 de cient mice have shown impaired ligand binding, reduced cell clustering to amyloid, reduced plaque compaction, and an impaired activation pro le 19,21,24,34,40 . Using scRNA-seq we identify clusters 2, 3 and 4 as infusion-responsive clusters characterized by upregulation of DAM genes and downregulation of homeostatic microglial genes, a pro le that has been reported in neurodegenerative diseases 19,41 . In these clusters, Trem2 ko microglia when compared to WT microglia show a signi cant downregulation of DAM genes and upregulation of homeostatic genes (Fig. 6). Finally, within the activated clusters we nd less difference between APOE isoforms in Trem2 ko mice possibly resulting from the impaired ability of Trem2 ko microglia to recognize ligands.
Collectively our ndings have identi ed that in the presence of Aβ, E3 lipoproteins are more effective at providing rapid transcriptional and phenotypic response than E4 lipoproteins that can counteract Aβ harmful effects.

Lead Contact
Further information and requests for resources and reagents should be directed to and will be addressed by the Lead Contact, Rada Koldamova (radak@pitt.edu).

Materials Availability
This study did not generate new unique reagents.

Data and Code Availability
The RNA-seq expression data has been deposited in the GEO database under the accession number: To Be Determined.

Experimental models
All animal experiments were performed in accordance with the NIH Guide for Care and Use of Animals and approved by the University of Pittsburgh Institutional Animal Care and Use Committee. The following mouse lines were bred to generated experimental mice: Wild-type C57BL/6J (WT; JAX), human APOE3 and APOE4 targeted replacement mice (E3 & E4; Taconic), Trem2 em2ADiuj /J mice (Trem2 ko ; JAX) and 129P2(Cg)-Cx3cr1 tm1Litt /J (Cx3cr1 GFP ; JAX). Human APOE3 and APOE4 targeted replacement mice were bred to Abca1 tm1Jdm /J mice (Abca1 ko ; JAX) to generate APOE3/Abca1 het and APOE4/Abca1 het (E3 het & E4 het ). Likewise, human APOE3 and APOE4 targeted replacement mice were bred to Trem2 em2ADiuj /J mice (Trem2 ko ; JAX) to generate APOE3/Trem2 ko and APOE4/Trem2 ko (E3/Trem2 ko & E4/Trem2 ko ). All experimental mice were on the C57BL/6 genetic background, kept on a 12 h light-dark cycle with ad libitum access to food and water and randomly assigned to an experimental group. All reagents were purchased from Fisher Scienti c unless documented otherwise.

AD brain samples
Human samples (Suppl. Table 1) were provided by the University of Pittsburgh Alzheimer's Disease Research Center (ADRC) brain bank and the Sanders-Brown Center on Aging at the University of Kentucky.
Braak staging was performed on Bielschowsky-stained sections and APOE allelic polymorphism determined by a PCR-based assay. Gray matter samples of APOEε3/3 and APOEε4/4 genotypes from the right inferior parietal lobule were dissected and used for Multi-Dimensional Mass Spectrometry Shotgun Lipidomics (MDMS-SL). Age and Postmortem intervals (PMI) matching was con rmed by t test.

Native APOE generation and characterization
Cultures of primary astrocytes were established from one-day-old E3, E4, E3 het and E4 het targeted replacement pups as previously shown 42,43 . Brie y, the cortices and hippocampi were mechanically dissociated using a sterile pasteur pipette and cultivated in DMEM/F12 medium supplemented with 10% bovine growth serum, l-glutamine and antibiotics. Cells were cultured on poly-d-lysine (100 μg/mL) coated T75 Costar asks (Corning). Con uent primary astrocytes were incubated with treatment medium (Neurobasal medium with antibiotics and glutamine as above minus serum) for 48 h. Conditioned medium were collected, immediately ltered through 0.22 μm lter and concentrated using Amicon® Ultra Centrifugal Filters (10kD cutoff). Native APOE lipoproteins were washed 3 times with cold PBS and their concentration measured by quantitative Western Blotting (NUPAGE) using commercial human APOE (Meridian Life Science) as a standard. Native APOE particles were resolved on Novex™ 4-20% Tris-Glycine gels using an Amersham™ HMW calibration kit as native ladder (GE Healthcare). The size of APOE lipoprotein particles were measured by dynamic light scattering with Zetasizer Nano Z (Malvern). APOE lipoprotein cholesterol content was assessed with the Cholesterol Quantitation Kit (Sigma) according to the manufacturer's instructions.

Multi-Dimensional Mass Spectrometry Shotgun Lipidomics (MDMS-SL)
The MDMS-SL assay 44,45,46 was performed to determine differences in the lipid composition of the native E3 and E4 particles isolated from primary astrocyte cultures and human AD brain samples described above. APOE particles isolated from E3 het and E4 het primary astrocytes, exhibiting diminished APOE lipidation states, were used as negative controls. Quantitative analysis was performed on a triplequadrupole mass spectrometer (Thermo Fisher Scienti c) equipped with an automated nanospray apparatus NanoMate and Xcalibur system. Internal standards for the quanti cation of individual molecular species of the major lipid classes were added to each sample at the start of the extraction procedure. Lipid extraction was performed by the methyl-tert-butyl ether (MTBE) method with resuspension in chloroform/methanol (1:1 v/v) solution with nitrogen ush. Identi cation and quanti cation of all reported lipid molecular species were performed using an in-house automated software program 45

Liquid Chromatography-Mass Spectrometry (LCMS) Lipidomics
Phospholipids were extracted and separated as before 47 . Brie y, MS analysis of phospholipids was performed on a Q-Exactive hybrid-quadrupole-orbitrap mass spectrometer (Thermo Fisher Scienti c) as previously described 47 using a normal phase column (Luna 3 µm Silica 100Å, 150 x 2.0 mm, (Phenomenex)) at a ow rate of 0.2 mL/min on a Dionex Ultimate 3000 HPLC system (maintained at 35°C). The analysis was performed using A and B gradient solvents containing 10 mM ammonium acetate and 0.5% triethylamine (A -propanol:hexane:water (285:215:5, v/v/v) and Bpropanol:hexane:water (285:215:40, v/v/v)). The column was eluted for 0.5 min isocratically at 25% B, then from 0.5 to 6.5 min with a linear gradient from 25% to 40% solvent B, from 6.5-25 min using a linear gradient of 40-55% solvent B, from 25-38 min with a linear gradient of 55-70% solvent B, from 38-48 min using a linear gradient of 70%-100% solvent B, then isocratically from 48-55 min at 100% solvent B followed by a return to initial conditions from 55-70 min from 100% to 25% B. The column was then equilibrated at 25% B for an additional 5 min. Analysis was performed in negative ion mode at a resolution of 140,000 for the full MS scan in a data-dependent mode. The scan range for MS analysis was 400-1800 m/z with a maximum injection time of 128 ms using 1 microscan. An isolation window of 1.0 Da was set for the MS and MS2 scans. Capillary spray voltage was set at 3.5 kV and capillary temperature was 320 o C. The S-lens Rf level was set to 60. The phospholipid content for each APOE isoform was normalized to the amount of APOE protein as determined by SDS NuPAGE. The analysis of the resulting LCMS data was performed and visualized in R (v3. 6 were used for all Aβ oligomerizations and injections as before 27 . Under a fume hood, 0.1 mg Aβ peptide was dissolved in ice cold 1,1,1,3,3,3-Hexa uoro-2-Propanol (HFIP, Fluka) then vortex for a few seconds.
The solution was dried with a gentle stream of nitrogen to obtain a peptide lm at the bottom of the vial. Prior to use, the lm was re-suspended in anhydrous dimethyl sulfoxide (DMSO) to form a 5 mM solution, sonicated in a water bath for 10 min and diluted in sterile phosphate-buffered saline (PBS) to a nal concentration of 100 μM. HiLyte™ Fluor 488-or 555-Aβ or Aβ 42 peptide and E3 or E4 native particles were combined, vortexed, held under oligomer forming conditions (room temperature) for 24 h at a nal concentration of 50 μM Aβ and 5 μM APOE and stored at -20° C until use (abbreviated AβE3 or AβE4).
The same molar concentration (10:1) of scrambled Aβ (AnaSpec) was dissolved in vehicle, combined with isolated E3 or E4 particles, held under oligomer forming conditions and stored at -20° C until use (abbreviated scrAβE3 or scrAβE4). scrAβE3 or scrAβE4 was utilized as negative controls throughout the subsequent experiments. Furthermore, Aβ 42 peptide and scrambled Aβ was incubated with PBS vehicle as negative control for APOE particles.

Electron Microscopy
For electron microscopy, 5 μL of each sample was placed on a freshly glow-discharged carbon-coated grid, adsorbed for 2 min and excess solution was blotted using lter paper 27 . The grid was washed with deionized water before staining with 5 μL of freshly ltered uranyl acetate solution (1%, w/v) for 15 s.
Excess stain was blotted and the grid was allowed to air-dry. Grids were imaged on a Tecnai T12 microscope (FEI) operating at 120 kV and ×30,000 magni cation and equipped with an UltraScan 1000 CCD camera (Gatan) with post column magni cation of 1.4x.

Guide cannula implantation
To examine the cognitive effects in WT male mice (3 mo.), Aβ oligomers co-incubated with native APOE lipid particles were infused directly into the brain through implanted guide cannulas 27 . Following anesthesia with iso urane, the head was shaven and sterilized with two separate povidone-iodinealcohol washes. A 50% mixture of bupivacaine and lidocaine was applied to the surgical site and ophthalmic ointment applied to the eyes. The head was leveled in a stereotaxic frame and an incision made exposing the dorsal aspect of the skull. Two holes were drilled into the skull (coordinates: AP = -2.46 mm, L =±1.50 mm) and 26-gauge guide cannulas (Plastics One) were lowered to a depth of 1.0 mm.
Cannulas were xed to the skull with acrylic dental cement attached to two bone anchoring screws and the surgical opening sutured closed. After suturing, animals were administered buprenorphine and sterile saline, body temperature was maintained until sterna recumbence and animals were allowed to recover for 8 days prior to the start of behavioral testing.

Behavioral testing
All stages of behavioral testing were performed at the same time of the day (during the light phase), ensuring 24 h between each phase of testing. Fear conditioning was started 24 h after the completion of novel object recognition. Prior to behavioral testing, mice were placed into individual containers, taken to the behavioral testing room and handled for 3 min for three consecutive days to reduce anxiety. Thirty minutes before a training stage, animals received an infusion of Aβ oligomer co-incubated with either E3 or E4 (Fig. 2a-d). Scrambled Aβ co-incubated with either E3 or E4 was used as a negative control. Mice were randomly assigned to either the AβE3, AβE4, scrAβE3 or scrAβE4 group. The dummy cannulas were removed and infusion cannulas, attached to microsyringe pump by polyethylene tubing, were placed in the guide cannula. Aβ oligomer ( nal volume of 1 μL per hemisphere) was infused over 1 min, the cannulas were left in place for 1 min to allow for diffusion of the sample and nally dummy cannulas replaced. Each animal received 5 infusions of Aβ oligomer starting on day 2 of novel object recognition until the completion of behavioral testing (day 6). Between behavioral trials, the paradigms were cleaned with 70% ethanol to eliminate any olfactory cues. Performance was recorded and scored using ANY-maze software (Stoelting Co.) during all phases of testing.

Novel Object Recognition
Novel object recognition (NOR) was performed over three consecutive days as previously described 24 . On Day 1, habituation phase, each animal was allowed to freely explore an open arena (40 cm X 40 cm X 30 cm white plastic box) for two 5 min trials with a 5 min inter-trial interval. On Day 2, familiarization phase, each animal was returned to the arena containing two identical objects (tower of LEGO® bricks) located in opposite diagonal corners for two 5 min trials separated with a 5 min inter-trial interval. On Day 3, testing phase, the animal was returned to the arena with two objects in the same positions as previously, but one object was replaced with a novel object (metal bolt and nut of similar size). Mice were allowed to explore the two objects for 10 min. The exploration of both objects was de ned as the mouse sni ng or interacting while facing an object within 3 cm. Mice were consistently placed into the middle of the arena facing the posterior wall to prevent any object preference. The percent exploration, an indicator of recognition memory, was determined by dividing the time exploring the novel object by the total time exploring both objects. Locomotor activity was assessed by measuring the total distance traveled in the open eld during day 1 of testing.

Contextual and Cued Fear Conditioning
Contextual and Cued Fear Conditioning (CCFC) was performed over three consecutive days as previously described 24 . On Day 1, training phase, mice were placed in a conditioning chamber (Stoelting Co.) for 5.5 min. The rst 2 mins were silent, allowing the mouse to acclimate to the chamber; this was followed by a 30 sec tone (2,800 Hz; Intensity 85 dB, conditioned stimulus (CS)) ending in a 2 sec foot shock (0.7 mA, unconditioned stimulus (US)) through the oor of the conditioning chamber. The process was repeated one more time and ended with 30 sec of re-acclimation. On Day 2, contextual phase, mice were placed in the same conditioning chamber for 5 min with no tone or shock administered, to measure contextual fear conditioning. On Day 3, the gray walls of the chamber were replaced with black and white striped walls to introduce a novel environment for assessing cued fear conditioning. Mice were placed in the conditioning chamber for 5 min. After the rst 2 min of silence, the tone was administered for 3 min, to measure cued fear conditioning. Freezing time was de ned as the absence of movement except for respiration and calculated as percent freezing of the total time in the chamber during each phase of testing.

Cortical infusion
WT male mice (3 mo.) received cortical infusions of un-labeled AβE3 or AβE4. In a separate cohort of Cx3cr1 GFP , WT and Trem2 ko male mice (3 mo.), HiLyte™ Fluor 555-labeled Aβ combined with native E3 and E4 particles was infused into the cortex. Mice were randomly assigned to either the scrambled Aβ, Aβ alone, AβE3 or AβE4 group. For the bilateral infusion, a 28-gauge infusion cannula (Plastics One) was connected to a 10 µL glass syringe (Hamilton) with vinyl tubing and placed in a micro syringe pump.
Following anesthesia with iso urane, the head was shaven and sterilized with two separate povidoneiodine-alcohol washes. A 50% mixture of bupivacaine and lidocaine was applied to the surgical site and ophthalmic ointment applied to the eyes. The head was leveled in a stereotaxic frame and an incision made, exposing the dorsal aspect of the skull. Four holes were drilled into the skull (coordinates: AP= -1 and -2.5 mm, L= +/-2.0 mm) and the infusion cannula was lowered into the cortex (DV= -1.0 mm). At each infusion site, 2 µL of Aβ preparation was infused at a rate of 0.5 µL/min and the cannula remained in place for 4 min following the infusion. Following the infusions, the surgical opening was sutured closed, animals were administered buprenorphine and sterile saline and placed on a heating pad until fully recovered.
Animal tissue processing Sections were washed and transferred into secondary donkey anti-rabbit Alexa 594 antibody (Invitrogen) for 1 h, before being washed, mounted on superfrost plus slides and coverslipped. Fluorescent confocal images of the infusion site were taken using a Nikon A1 confocal microscope at 60x magni cation with 1.0 μm step size. Four sequential images were captured anking either side of the infusion site. Analysis was performed as in Marsh et al. 48 with modi cations. The FilamentTracer module (Imaris, version 7.1.1, Bitplane) was utilized to trace processes of cortical IBA1 positive microglia and determine process length and number of branch points as indicators of microglial morphology complexity. Using the 3D mapped lament tracings, we oriented all microglia so their central axis was positioned facing the infusion site and generated a heatmap depicting the location of microglia from an experimental group utilizing the ImageJ -heatmap from image stack plugin (National Institutes of Health). The number of overlapping microglial processes increases the pixel saturation metric. The percent of all processes which occupy space in each of the 4 quadrants was used to determine the percent coverage of that quadrant.
A second series of 6 brain sections around the infusion site was used for F4/80 immunostaining. First, we quenched endogenous peroxidases with 0.3% hydrogen peroxide, tissues were blocked in 3% normal goat serum (Vector), then blocked for endogenous avidin and biotin. Sections were incubated in F4/80 antibody (Abcam) overnight at 4°C. Sections were washed and transferred into secondary biotinylated anti-rat antibody (Vector) for 90 min before being washed and subsequently developed using the Vector ABC kit and DAB substrate kit (Vector). Sections were mounted onto superfrost plus slides and coverslipped. Bright-eld images were taken using a Nikon Eclipse 90i microscope at 20x magni cation encompassing the infusion site. Image intensity threshold was established to detect the F4/80 staining compared to background using NIS Elements software (Nikon Instruments Inc.) and values were represented as the area of staining normalized to total image area or percentage of area covered.

Magnetic-activated cell sorting (MACS)
For WT mice infused with unlabeled Aβ, we isolated microglial and neuronal cellular populations utilizing MACS column-based protocols according to manufacturer instructions (Miltenyi Biotec). First, the tissue was dissociated utilizing the Neural Tissue Dissociation kit (Miltenyi Biotec), a gentle two-step enzymatic dissociation (papain and trypsin) that yields a high number of viable cells. Brie y, the cortical tissue was enzymatically and mechanically lysed at 37°C for 35 min to obtain a cell suspension. Then the suspension was passed through a 40 µm cell strainer to remove debris and ensure a single-cell suspension. After a 10 min centrifugation at 300 g, the pellet was gently resuspended with ice-cold PBS + 0.5% BSA buffer and incubate for 15 min at 4°C with Myelin Removal Beads II (Miltenyi Biotec). After washing with PBS + 0.5% BSA and 10 min centrifugation at 300 g, the pellet was resuspended with PBS + 0.5% BSA. The single-cell suspension was applied to a column placed on a magnetic stand (LS column and magnets from Miltenyi Biotec) to deplete the myelin fragments by magnetic separation and washed with PBS + 0.5% BSA. Finally, the cells were centrifuged for 5 min at 300 g and the pellet was resuspended with 1 mL of PBS + 0.5% BSA. After obtaining a myelin-free single-cell suspension, the microglia were labeled with anti-CD11b immunomagnetic beads and isolated using magnetic columns (Miltenyi Biotec). To isolate the neurons, the cell suspension depleted of microglia cells was incubated with a cocktail of antibodies from the Neuron Isolation Kit (Miltenyi Biotec). Non-neuronal cells like astrocytes, oligodendrocytes, endothelial cells or broblasts were magnetically labeled and depleted through the magnetic columns; therefore, neurons were isolated by negative selection. To con rm the purity of isolated microglia, we performed RT-qPCR with P2ry12 and Tmem119, microglia-speci c markers. To further ensure microglia and neuron purity when using MACS, following sequencing and alignment, all genes with low expression (average raw read count < 30.9) were removed from further analysis.

Fluorescence-activated cell sorting (FACS) and ow cytometry
For the Cx3cr1 GFP mice infused with HiLyte™ Fluor 555-labeled Aβ, microglia were isolated utilizing FACS. After removing the cerebellum, subcortical area and olfactory bulbs, cortical tissue within 1 mm of either side of the infusion site was processed into a single-cell suspension using the Neural Tissue Dissociation kit followed by the Myelin Removal Beads II (Miltenyi Biotec) as described above. For WT and TREM2 ko mice, after the myelin removal steps the pellet were resuspended with 100 µl of PBS and the microglia cells were labeled with 5 µl of anti-mouse/human CD11b antibody (clone M1/70) conjugated with Alexa Fluor Ò 647 (BioLegend) for 20 min on ice. All cells were then washed and prepared with 1 mL of PBS + 0.5% BSA for FACS sorting.
A bio-contained BD FACSAria™ III sorter (BD Biosciences) was used to sort microglia cells for the two experimental populations. First, live cells were separated from the debris according to their forward scatter and side scatter properties and a second gate was used on individual cells only. The GFP uorescence was detected with a 525 nm lter (488 nm laser) for the microglia isolated from the Cx3cr1 GFP mice, the CD11b/APC uorescence was collected with a 668 nm lter (647 nm laser) for the WT and TREM2 ko mice and the 555-labeled Aβ uorescence was detected with a 613 nm lter (555 nm laser). 2,000 microglia with high GFP signal but low 555-labeled Aβ signal, were considered unexposed to Ab and termed single-positive cells (single+). 2,000 microglia cells with a high signal for both labels (GFP and 555-labeled Ab) were considered exposed to Ab and termed dual-positive cells (dual+). The single+ and dual+ microglial populations were sorted into 1.5 mL tubes containing 350 µL of RLT lysis buffer from the RNeasy Micro kit (Qiagen) and 3.5 µL of 2-Mercaptoethanol. After the FACS sorting, the biocontained BD FACSAria™ III sorter was used with the same gates to count and estimate the percentages of the total microglia population and single+ and dual+ microglia population in a total of 1,000,000 live cells.
RNA isolation and RNA-sequencing RNA was isolated from the MACS-isolated microglia and neurons and FACS single+ and dual+ microglia following the manufacturer instructions for the RNeasy Mini kit and RNeasy Micro kit (Qiagen) respectively. After RNA isolation, concentration was measured with a Qubit 3.0 Fluorometer (Thermo sher) before quality assessment with the 2100 Bioanalyzer instrument (Agilent Technologies). Sequencing libraries for MACS-isolated cells were generated using mRNA Library Prep Reagent set (Illumina) as previously described 49 . The total RNA from sorted cells was fragmented and converted into cDNA using the SMART-Seq® v4 Ultra® Low Input RNA Kit for Sequencing (Takara Bio). Brie y, a minimum of 10 pg of puri ed total RNA was used to perform a rst-strand cDNA synthesis in a PCR clean workstation. After ampli cation by LD PCR and puri cation using the Agencourt AMPure XP beads (Beckman Coulter), 1 μL of the ampli ed cDNA was used for validation using the Agilent 2100 Bioanalyzer on a High Sensitivity DNA chip. 1 ng of the cDNA was used with the Nextera DNA Library generated to visualize patterns of similarity between groups. To generate the t-SNE, normalized data from all genes in the analysis were submitted and the perplexity set as high as possible given the total number of samples in the analysis with the theta set to 0.5.

Single-cell isolation and library creation
Separate biological replicates from AβE3 or AβE4 infused WT and Trem2 ko mice were used for single cell RNA-seq (scRNA-seq) and infused as described above. 24 h after the injections mice were perfused with PBS. After the brain extraction, cortical tissue around the injection site were excised using adult mouse brain matrix at 1 mm before and after the injection site. The brain tissue was then dissociated according to manufacturer instructions (Miltenyi Biotec) as described above.

Two-photon imaging and statistical analysis
We used Cx3cr1 GFP male mice (3 mo.) to determine the microglial response of AβE3 or AβE4 utilizing twophoton imaging 52 . Brie y, the animals were anesthetized I.P. with a cocktail of ketamine/xylazine (75/10 mg/kg) and placed in a stereotaxic frame. Body temperature was maintained at 37.6°C. The skin above the skull was removed and a well of 2 cm diameter was built over both hemispheres with a light-curable cement (Composite Flowable) to hold a saline immersion for the microscope objective. The craniotomy was conducted with a high-speed dental drill over the parietal cortex bilaterally. The craniotomy site was regularly ushed with saline to wash out bone fragments and prevent thermal damage. The exposed cortex was injected with Hi-Lyte™ Fluor 555-labeled Aβ (Anaspec) combined with native E3 or E4 particles pre-incubated for 24 h. The injection was performed at an approximate depth of 300 µm below the surface of the cortex with a borosilicate glass capillary tube (WPI) pulled with a two-stage vertical pipette puller (Narishige) to a tip diameter of approximately 10 µm and coupled with exible tubing to a picospritzer microinjection dispense system (Parker). Mice were randomly assigned to either the AβE3 or AβE4 group. After the injection, the mice were moved under the two-photon microscope within 5 minutes and anesthesia was maintained with 40 mg/ml/h ketamine.
In vivo imaging was conducted with an Ultima IV two-photon laser scanning microscope (Bruker) with an InSight DS Laser system (Spectra-Physics) tuned at a wavelength of 920 nm. Two channels were collected red (595/50) and green (525/50 nm) to visualize the 555-labeled Aβ and microglia, respectively.
Images were collected with 16x Nikon objective lens on Prairie View software using a time series of 11 slices with 1024 x 1024 matrix size, dwell time 4.4 microseconds and 3 µm step size. The effective inplane resolution was 0.4 μm per pixel. The Z-stack images were collected in Galvo acquisition mode such that each Z-stack was collected every minute for a total duration of 80 min.
For each animal, Z-stack images were analyzed to quantify the microglial approach and microglial injection site coverage over time. Initially, the data was loaded in ImageJ (National Institutes of Health) and a rigid-body co-registration with the StackReg plugin was used to account for global temporal drifts of the eld of view. The Z-stacks were transformed to sum-based XY projections by linear interpolation using the 3D Project function, allowing for time series analysis of speci c regions of interest. We generated the panels illustrating cell displacement from those datasets by assigning red to the beginning of the time series (t=0) and green to subsequent time points (10 min or 80 min). Then, the two colorcoded images were merged in ImageJ to reveal spatial details, with red showing the original location, green for the new cellular location after a particular segment of time has elapsed (0-10 min or 0-80 min) and yellow at the regions where the two channels overlap. The 555-labeled Aβ infusion site was shown in white to display the increasing coverage as cell processes approach. The time series were also saved as movies presented in the supplementary material.
The datasets were then converted to mat structures and all subsequent analyses were performed in MATLAB (Math Works), using standard image processing and statistical functions. Brie y, the red channel showing the infusion site was thresholded using the Otsu's method with the multithresh Matlab function. The resulting images were binarized and small features outside of the infusion site were removed using the bwareaopen function. The mask was then dilated to smooth the borders and account for the lower contrast of the diffusion gradient around the borders. The percent coverage by the microglial processes over time was then computed as the increase of the green channel intensity above 1.5 times the standard deviation baseline of the green channel within the binary mask.
The mean distance of the cell processes to the infusion site over time was computed by eroding the binary mask described above and computing the Euclidean distance transformation of the full binary image using the bwdist function. Then, each pixel in the green channel over 1.5 standard deviations of the intensity was mapped with that transformation and assigned a number that is the distance between that pixel and the edge of the infusion site forming progressive concentric level sets with increasing proximity to the infusion site. The change of this Euclidean distance over time was converted from pixels to microns using the effective in-plane resolution of the imaging data and plotted as a time series.
Primary microglia cell culture Primary microglial culture and Aβ uptake assay were performed using WT, E3, E4, E3/Trem2 ko and E4/Trem2 ko pups (1-3 days old) as described before 42,43 . Brie y, the cortices and hippocampi were mechanically dissociated using a sterile Pasteur pipette. The dissociated cells were plated in 75 cm 2 asks with DMEM/F12 medium (Thermo Fisher) containing 10% FBS at DIV0. 24 h after plating (DIV1), the media was replaced to remove debris. Microglia were collected by tapping at DIV14 and plated on 0.01% poly-L-lysine (Sigma) coated 12mm circular coverslips in 24-well plates with the density of 60,000 cells/well. 24 h after plating, cells in the treatment groups were treated with 1µM Hi-Lyte TM Fluor 488labeled Aβ (Anaspec) at 37 °C for 1 h. Following the treatment, cells were washed three times with PBS, xed in 4% PFA and permeabilized with 0.2% Triton X-100 at room temperature. Microglia were then labeled with Anti-IBA1 antibody (1:500) at 4 °C for 18 h and followed by 2 h incubation with horse antirabbit 594 secondary antibody and stained with DAPI to visualize nuclei.