Relative Contribution of Amyloid-β Plaque Associated And Plaque Distant Microglia To Alzheimer’s Disease (AD) Progression

Background: Research in recent years rmly established that microglial cells play an important role in the pathogenesis of Alzheimer's disease (AD). In parallel, a series of studies showed that, under both homeostatic and pathological conditions, microglia are a heterogeneous cell population. In AD, amyloid-b (Ab) plaque-associated microglia (PAM) display a clearly distinct phenotype compared to plaque-distant microglia (PCM), suggesting that these two microglia subtypes likely differently contribute to disease progression. Methods: In this study, we combined cell-specic laser capture and RNA-seq analysis to investigate the functional role of both plaque-associated and plaque-distant microglia. Results: First, we established that this approach allows selective isolation of microglia, while preserving spatial information and preventing transcriptome changes induced by classical purication approaches. Then, we identied, in both microglia subpopulations, networks of co-deregulated genes and analyzed their potential functional roles in AD. In addition, we investigated the dynamics of microglia transcriptomic remodeling at early, intermediate and late stages of the disease. Conclusions: Our comprehensive study demonstrates that the proximity of microglia to Ab-plaques dramatically alters the microglial transcriptome and reveals that these changes can have both positive and negative impacts on the surrounding cells. These opposing effects may be driven by local microglia heterogeneity also demonstrated by this study. Our results also suggest that Ab plaque-associated microglia undergo exhaustion in the later stage of the disease. Our approach also allowed to molecularly dene the overlooked plaque-distant microglia. We show, that although the transcriptomic changes are far less striking compared to what is observed in plaque-associated microglia, plaque-distant microglia are not bystanders of the disease. In particular, our results suggest they are involved in Ab oligomer detection and in Ab-plaque initiation, with increased contribution as the disease progresses. P36961). Slides were imaged on an Imager Z1 microscope (Zeiss) equipped with an AxioCam MR R3 camera. Images were acquired with an 20X/0.50 M27 Zeiss Plan-Neouar air objective; 11 images (corresponding to 10 µm-thick optical sections) were acquired. may have both positive and negative impacts on the surrounding tissues: they may represent a physical barrier to prevent small Aβ oligomers spreading, but are also releasing pro-inammatory cytokines that are deleterious for the surrounding cells and are less supportive for synaptic functions. Importantly, our study reveals for the rst time that, although PCM display a homeostatic-like phenotype, those microglia subtype are reactive, engaged in specic biological processes and actively participate to the disease progression. Our data suggest that they may be key for detecting small Aβ oligomers and for initiating plaque formation. Finally, we reveal further molecular heterogeneity in both PAM and PCM. Further work is needed to understand how this very local heterogeneity builds-up, to identify its dynamics, and to determine the consequences for the disease progression. Deciphering these mechanisms will allow to target specic subpopulation of microglia with the ultimate goal of promoting benecial microglial functions and alleviating deleterious ones. (PCM); control microglia (CM); genome wide association studies (GWAS); Disease-associated microglia (DAM); Late response microglia (LRM); Activated response microglia (ARM); Early response microglia (ERM), Interferon-response microglia (IRM); Transiting response microglia (TRM); microglial neurodegenerative phenotype (MGnD); Inammatory Associated Microglia (IAM); plaque-induced gene (PIG); Class II major histocompatibility complex (MHCII+); Fluorescent-activated Cell sorting (FACS); Paraformaldehyde (PFA); Thiazine red (TR); RNA Integrity Numbers (RINs); Relative Log Expression (RLE); differentially expressed genes (DEG); Generalized Linear Models (GLM); Principal Component Analysis (PCA); Weighted gene co-expression network analysis (WGCNA); Edge Percolated Component (EPC), EcCentricity, Maximal Clique Centrality (MCC); Maximum Neighborhood Component (MNC); Gene Ontology (GO); Kyoto Encyclopedia Gene and Genomes (KEGG); Laser capture microdissection (LCM); false discovery rate (FDR); Generalized Linear Model (GLM); Tumor Necrosis Factor (TNF); single-molecule uorescence in situ hybridization (smFISH); reactive oxygen species (ROS).

disease-driven microglia transcriptomic changes [19]. Finally, other studies used laser-capture approaches to analyze the transcriptomic changes in plaque-associated and/or plaque-distant tissues, thus capturing transcriptomic changes arising in heterogeneous cell types [6,20,21]. However, the transcriptomes of PAM and PCM have never been directly established.
To overcome these limitations, we combined cell-speci c laser capture and RNA-seq analysis to investigate the functional roles of both PAM and PCM. First, we established that this approach is well suited to study the transcriptome remodeling of spatially distinct microglia. We then identi ed, in both microglia subpopulations, networks of co-deregulated genes and analyzed their potential functional roles in AD. Moreover, by investigating the microglia transcriptomic remodeling at early, intermediate and late stages of the disease, we were able to highlight the dynamics of these processes. Our comprehensive study con rms that proximity to Aβ-plaques dramatically alters the microglia transcriptome and reveals that these changes can have both positive and negative impacts on the surrounding cellular network. Our approach also allowed us to study the overlooked plaque-distant microglia and to reveal that PCM are not bystanders to the disease progression and may be involved in Aβ oligomer detection and plaque initiation, with their contribution increasing as the disease progresses.

Animals
The APP swe /PSEN1 dE9 : CX3CR1 +/eGFP mice, referred to as APP/PS1-CX3, used in this study were obtained by crossing heterozygous APP swe /PSEN1 dE9 male mice with female CX 3 CR 1 eGFP/eGFP . APP swe /PSEN1 dE9 mice were purchased from the Jackson Laboratories [22,23] and subsequently bred under the C57BL6/J background in the Speci c Pathogen Free animal facility of the Institute for Functional Genomic (IGF, Montpellier, France; Agreement from the Ministry of Agriculture N° D34-172-13). CX 3 CR 1 eGFP/eGFP mice were generous gift from Dan Littman [24] and were also maintained in the IGF facility under the C57BL6/J background. Comprehensive characterization of the APP swe /PSEN1 dE9 : CX3CR1 +/eGFP has been performed previously [25] and revealed that Cx3cr1 haplode ciency has little impact on the disease progression and that APP/PS1-CX3 mice are a useful model to study microglia in AD-like pathology. Mice were housed in a 12 h light-dark schedule with food and water available ad libitum. All experiments followed European Union (Council directive 2010/63/UE) and institutional guidelines for the care and use of laboratory animals. The animal experiment protocols used in this study were approved by the Comité d'Ethique pour l'Expérimentation Animale Languedoc Roussillon (CEEA-LR; APAFiS#5252). Experiments were performed in 4-, 8-and 12-months-old (mo) animals (see Table S1 for details on the mice used).

Tissue sample preparation
All solutions mentioned hereon were prepared using RNAse-free buffers Laser microdissection. After induction of deep anesthesia with 2 µg/g pentobarbital (Euthasol Vet, TVM), mice were perfused intracardially with 20 ml of phosphate buffer saline (PBS; Ambion), followed by 20 ml PBS solution containing 20% sucrose (Sigma-Aldrich, S7903). The brain was then removed, immersed overnight in PBS solution containing 30% sucrose and then ash frozen in -40°C Isopentane (Merck, #320404). Brains were stored at -80°C for at least 24 hours and for up to 6 months. 8 μm thick coronal sections were cut using a cryostat (Leica) with chuck and cabin temperatures were maintained at −24 °C. Sections were mounted onto Superfrost slides and then stored at -80°C for up to two days before laser microdissection.
RNAscope: After induction of deep anesthesia with 2 µg/g pentobarbital (Euthasol Vet, TVM) mice were transcardiacally perfused with 10 ml cold PBS. Brains were extracted, xed in 4% paraformaldehyde (PFA, Sigma, P6148) for 2h at room temperature (RT) and post-xed overnight at 4°C in fresh 4% PFA. Tissues were then cryoprotected by successive immersion in PBS solutions containing increasing sucrose (Sigma, S7903) concentrations (i.e. 10%, 20% and 30%). Tissues were transferred from one solution to the next when the brain sank indicating equilibrium between the tissue block and the solution. Brains were included in OCT (TissueTek, #4583), ash frozen in -50°C Isopentane (Merck, #320404), and stored at -80°C for at least 24 hours. 14 μm thick serial coronal sections were cut using a cryostat (Leica), directly mount onto Superfrost slides and stored at -80°C until use.

Laser capture microdissection
Thiazine red staining for Ab plaque detection: Thiazine red (TR) is an analog of naphthol-based azo structures which binds b-pleated sheet structures. Like Thio avin-S, it stains dense core plaques but with maximum emission at 580 nm [26]. On the day of microdissection, slides were removed from freezer, immediately placed for 1 min in 70% Ethanol solution and then stained by immersion for 1 min in 75% Ethanol solution containing 0.165% TR. Excess TR was removed by performing three 15 sec with 75% Ethanol solutions. Dehydration was continued by successive immersion for 1 min in ethanol solutions of increasing concentration (VWR, #20281.310; 95%, 100% and 100%), followed by two immersions of 5 min each in 100% xylene (VWR, #289751.291). Then, slides were allowed to dry in a vacuum bell for at least one hour. Hygrometry in the microdissection room was controlled throughout the procedure and, to ensure preserving the sample quality, slides were used within 3h after removal from the vacuum bell.
Microglial cells (identi ed as GFP expressing cells) and TR staining were visualized at 20x magni cation using the Nikon Eclipse Ti-E epi uorescence microscope which equipped the PixCell IIe Laser Capture Microdissection system (Applied Biosystems/Excilone Elancourt, France). GFP+ cells of the cerebral cortex were laser captured in CapSure HS LCM Caps (Arcturus/Life Technologies). Laser characteristics were set at the smallest "spot" size (ie. 7.5 µm); the power of the Infra-red laser, the number and the durations of the pulses were adjusted for each slide. At the end of the session the captured cells were immediately lysed in the RLT-plus buffer (Qiagen; #1053393) and stored at -80°C. In AD-CX3 mice. Plaque-associated microglia and plaque-distant microglia were isolated from the same mice with TR staining used to discriminate both microglia subtypes. Thus, PAM corresponded to microglia located within 70 µm of the center of a dense core Ab plaque whereas PCM were microglia located further than 100 µm from the center of any dense core Ab plaque ( Figure S1). Distances were chosen based on the average size of the plaques (i.e. 12-20 µm in diameter) and on preliminary observations showing that microglia clustered around plaques were located within 70-80 µm of the center of the plaque (not shown). Additionally, when selecting PCM we avoided microglia that, even in the absence of TR staining, appeared clustered.
RNA extraction and RNA-seq LCM-isolated microglia from at least 4 micro-dissected sections in at least 2 independent sessions were pooled. This totaled about 400 microglial cells for each mouse and experimental condition (Table S1).
RNAs were extracted using the Qiagen RNeasy Plus micro kit (Qiagen, #74034) following a protocol slightly adapted from that of the manufacturer. RNAs were eluted in 16 µl RNAse-DNAse free H 2 O. Total RNA quality was veri ed by extracting RNA from the tissue remaining on the slide after microdissection and determining its integrity using the Agilent 2100 Bioanalyzer (Agilent). All RNAs had RNA Integrity Numbers (RINs) higher than 8.0.
Library preparation and RNA sequencing were performed by the Pro leXpert core facility (Lyon, France). In brief, mRNAs were pre-ampli ed from 200 pg total RNA using the SMART-Seq V4 Ultra Low RNA kit (Clonetec). Library preparation was performed from 500 pg cDNAds using the Nextera kit (Illumina) following manufacturer instructions. Libraries were sequenced using an Illumina NextSeq500 platform and 75 bp single-end sequencing data were obtained with between 28 to 41 million reads per sample.
Perfect trimmed reads were aligned to Mus musculus mm10 reference genome using the TopHat2 software [27]. The featureCounts tool was used to determine the number of reads mapping to each gene [28].

Bioinformatics analyses and Networks
Bioinformatics and statistical analyses were performed in collaboration with the StatABio facility (BioCampus UAR 3426 CNRS -US 09 INSERM -UM) using R software (3.6.0).
Differential gene expression: Samples to be included in the different analyses were selected according to the question addressed. After selection, gene expression normalization was performed using Relative Log Expression (RLE) normalization implemented in the DESeq 2 R package and genes with less than 1 count per million (cpm) in at least 3 out of 4 replicates and at least one condition, were ltered out. To detect the differentially expressed genes (DEG), we applied Generalized Linear Models (GLM) with tagwise dispersion. Both raw-(pv raw ) and adjusted-Pvalues (pv adj ) were computed.
Weighted gene co-expression network analysis (WGCNA): WGCNA R software package was applied to identify co-expression modules among pre-selected genes [30]. In brief, mean connectivity and scale dependency measures were calculated to choose the proper soft power and to reconstruct the network.
Soft threshold power was then evaluated using network analysis functions to preserve more correlated genes based on scale-free topology [31]. Identi cation of the potential modules was performed by applying the module analysis algorithm to the dissimilarity matrix. The minimal number of genes in each module was set to limit the number of unassigned genes. In practice, unassigned genes represented 3-11% of the genes' selection. The extracted modules were labeled with colors, with turquoise being the most abundant, blue the second most abundant and brown the third most abundant modules.
Unassigned genes were placed in the grey module. Eigenvalues, which can be seen as the expression values of "arti cial genes" quantifying the expression variations of the genes within a particular module, were also calculated.
Gene network representations. Gene networks of preselected genes were constructed using speci c applications (Apps) implemented in the Cytoscape software. First, we used the STRING App, with full STRING network and 0.7 con dence cutoff as settings to construct the network. At this stage, isolated genes were removed. Large networks were further divided into sub networks using the MCL tool of the STRING App. To identify hub genes, we then ran the CytoHubba app, employing ve calculation methods: Degree, Edge Percolated Component (EPC), EcCentricity, Maximal Clique Centrality (MCC), and Maximum Neighborhood Component (MNC). The intersecting genes derived using these ve algorithms encode the most highly connected proteins and may represent key candidate genes with important biological regulatory functions. Finally, we used the Omics Visualizer app to display gene expression values changes on the generated networks.
Functional enrichment analysis. To study the biological mechanisms and gene ontology of the selected genes, we used the g:Pro ler software (https://biit.cs.ut.ee/gpro ler/gost). The Gene Ontology (GO)biological processes associated with the selected genes were listed; nodes (GO-biological process) with adjusted p-value less than 5% were reported as important. To avoid overly speci c and general processes, only GO-biological processes with a size between 30 and 300 were considered. When the large number of GO-terms are affected, redundancy between them is high making it di cult to read graphs. To visualize and interpret those results, we used the EnrichmentMap and AutoAnnotate apps in Cytoscape to visualize the GO-terms network [35]. Contrastingly, when too few biological processes were detected, larger GObiological processes (up to 500 genes) were also considered. In these cases, to get further functional insights, Kyoto Encyclopedia Gene and Genomes (KEGG [36]) and Reactome [37] databases were also used to perform pathway enrichment analyses.

Results
We analyzed the repertoire of genes expressed in cortical microglia from both control and AD-mice ( Figure 1A) by combining laser capture microdissection (LCM) and RNA-seq approaches in transgenic mice expressing eGFP under the control of the CX3CR1 promotor [24,25]. Moderate tissue xation preserves uorescence but alters the quantity and quality of recovered RNA, whereas eGFP uorescence is generally low in fresh un xed brain samples [38] (and Data not shown). To overcome these issues, we developed a speci c protocol based on tissue preservation by sucrose perfusion and immersion, rapid freezing in -40°C isopentane, cryo-sectioning and dehydration, which allows eGFP uorescence preservation, amyloid plaque staining as well recovery of RNA in good quality and quantity ( Figure S1A).
We extracted total RNA from 400-600 microglia per animal per experimental condition (Table S1) and performed mRNA sequencing. We detected 13,923 expressed genes, and compared the expression of 133 randomly selected ones (i.e. exhibiting low, medium or high expression levels) in FACs sorted microglia ( [33]). Linear regression revealed a signi cant correlation between gene expression levels in LCM and FACS isolated microglia (r=0.697, p<0.0001), demonstrating that our data are consistent with previously published microglia gene expression pro les ( Figure S1B).
Isolation of microglia from mouse brain tissue with preservation of spatial information Cell isolation through LCM allows preservation of spatial information, but is subject to crosscontamination by surrounding cells whose processes may by be captured together with the cell of interest. To assess the degree of microglial enrichment, we evaluated the expression levels of speci c microglial, astrocytic, oligodendrocytic and neuronal genes in the different LCM samples. Figure 1B shows that microglia speci c genes are about 10 times enriched in the LCM samples compared to the whole cortex tissue, whereas, reversely, other glial cells and neurons speci c genes are strongly depleted.
In CX3CR1 +/eGFP mice, eGFP is expressed in all myeloid cells including in ltrating monocytes that can penetrate brain parenchyma in pathological conditions [39]. To assess the possibility that our LCM samples could be contaminated by in ltrating monocytes, we analyzed the expression of peripheral monocyte/macrophage speci c ( Figure 1B). Interestingly, we demonstrated that the expressions of Cd163 and S100a4 were also depleted in the LCM samples. Cd74 expression was depleted in control microglia (CM), but enriched in PAM and PCM indicating that this gene is upregulated in reactive microglia.
An important drawback in cell isolation is the intrinsic cell activation induced by generating single-cell suspension. This is particularly true for microglia which are inherently reactive cells, for which it has been shown that FACS sorting may alter the analysis of the disease-induced transcriptomic changes [19,40]. In contrast to FACS, in LCM, cells are isolated from their environment after dehydration, which prevents the cell reaction as shown by low expression of immediate early genes in the LCM isolated microglia ( Figure  1C).
The reproducibility of our data was demonstrated by the strong correlation between the biological replicates ( Figure S1C). In addition, principal component analysis on mRNA expression pro les allowed good discrimination between the different microglia subtypes ( Figure 1D). In particular, PAM clearly separate from both CM and PCM. Interestingly, PCM also segregate from CM. However, this analysis did not allow further separation of the samples by age within each sub-population. We then examined, in the different subpopulations, gene expression for markers of previously identi ed microglia gene signatures, namely markers of homeostatic microglia (P2ry12 & Tmem119, Figure 2A), Disease-associated microglia (DAM, Apoe & Ctsd, Figure 2B, [11]), Activated-response microglia (Cst7 & H2-ab1, ARM, Figure 2C, [13]) and Interferon-response microglia (I t2 & I t3, IRM, Figure 2D, [13]). Our results showed that homeostatic microglial gene expressions remained stable in CM and PCM, and only appeared to decrease in late stage PAM. Interestingly, although DAM, ARM and IRM markers are expressed at higher levels in PAM, our results revealed that these genes are also up-regulated in PCM with their expression increasing in an agedependent manner. Notably, expression of DAM and ARM markers remained stable across disease stages in PAM, whereas that of IRM appeared to increase in older PAM ( Figure 2C-D).
Having established that this protocol allows the isolation of spatially distinct sub-populations of microglia with minimal intrinsic perturbation and su cient enrichment, we conducted speci c contrast analyses to identify DEGs. We then performed bioinformatic/biostatistic analyses to infer the pathophysiological role of the different microglia subpopulations in AD progression. The work ow for data analysis is presented in Figure S2.

Biological functions and Master gene regulators in PAM
In the APP/PS1 model, dense amyloid plaques begin to appear in cortical areas at about 4-months of age [25]. However, at this age, plaques are very sparse and it was not technically feasible to isolate PAM.
Thus, to identify gene deregulation in PAM versus CM microglia, we restricted the analysis to 8-mo and 12-mo samples.
PCA revealed a clear distinction between CM and PAM, and statistical analysis identi ed 1851 DEGs (false discovery rate (FDR) <0.05), about two-thirds of which were up-regulated ( Figure 3A and data notshown). Previous studies established microglia reaction signatures in different pathological conditions. This includes the DAM, ARM and IRM signatures previously mentioned [11,13], but also signatures for In ammatory Associated Microglia (IAM) which represent a large set of DEGs in in ammatory conditions [34] and for the Reactome a smaller set of 86 genes deregulated in different acute and neurodegenerative conditions [33]. Figure 3B shows that, compared to control microglia, PAM DEGs were signi cantly enriched for these different pathological microglia gene signatures. The Sensome gene signature represents a set of membrane-associated proteins and receptors that are selectively expressed in microglia and that help them sense changes in their environment [32] (re ned in [33]). This signature was also signi cantly affected in PAM. Using a spatial transcriptomic approach, Chen et al. recently identi ed a plaque-induced gene (PIG) network mainly involving microglial and astroglial genes [20]. As expected, PAM signature was strongly enriched with PIGs genes, actually 52 of the 57 PIGs genes were deregulated in PAM ( Figure 3B).
Genes with similar expression patterns (co-expressions) are likely to have similar functions and can be grouped into modules by WGCNA [30]. We performed WGCNA analysis to identify gene modules among the DEGs, and GO based enrichment analyses to extract the hypothetical biological functions for each of these modules (see Figure S2 and Materials and methods for details). Among the 1851 DEGs, we identi ed two distinct modules signi cantly correlated with the Microglia-subtype trait. The largest module (i.e. turquoise module) included 1639 genes mainly up-regulated in PAM ( Figure S3A and Table  S2A). These genes were primarily associated with in ammation related biological processes, including Cell activation & proliferation, Immune response, Cytotoxicity, Exocytosis, Chemotaxism, Antigen presentation, etc. ( Figure 3C; Table S2B). Alterations in Cell morphology was another signi cantly affected biological function. The second module (i.e. blue module) was smaller and contained 212 genes mainly down-regulated in PAM ( Figure S3B; Table S2A). This module mainly related to Synaptic transmission associated biological processes ( Figure 3D; Table S2C).
In gene networks or subnetworks, hub genes (i.e. most highly connected genes) represent master regulators that are likely to play essential roles in controlling the biological response. We used speci c applications in Cytoscape (see materials and methods) to rst construct the genes' network of the two WGCNA modules, and second to identify the most connected sub-networks and their potential hub genes.
In the Turquoise module, the 1074 most highly connected genes were separated into subnetworks using the MCC cluster tool. The ten larger clusters are detailed in Table S3, with hub genes highlighted in dark green. The three larger subnetworks are also shown in Figure S3B-D, with hub genes in yellow. In the largest subnetwork, App and Penk which are respectively up-and down-regulated in PAM appeared to play orchestrating and redundant roles for controlling chemotaxis and endopeptidase activities ( Figure  S3B). The second subnetwork included genes that control cell shape and antigen processing. Hub genes of this network were the GTPases Rac1, Rhoa and Rhog which belong to, respectively, the Ras and Rho super-families ( Figure S3C). Rab5c is another small GTPase involved in controlling receptors endocytosis, vesicle tra cking, and endo-lysosomal pathways [41] and which played a central role in the third largest subnetwork ( Figure S3D). In the blue module, we identi ed a single network of 48 highly connected genes ( Figure S3F). This gene network was associated with control of the synaptic vesicle cycle (Table S3); hub genes were Syt1, Vamp2 and Snap25 which represent key proteins for neurotransmitter release.
Next, we addressed the question of the extent to which age affected the transcriptomic changes observed in PAM. To meet this goal, RNA-seq data from the 8-mo and 12-mo samples were reanalyzed using a Generalized Linear Model (GLM) model, with Microglia-subtype (PAM vs CM) and Age (8-mo vs 12-mo) as between samples' factors ( Figure S2). Thus, we identi ed 723 DEG in the Microglia-subtype:Age interaction (raw p-value<0.05). Among them, we restricted our analysis to the 179 genes that were signi cantly deregulated in PAM vs CM ( Figure S4A). WGCNA analysis further identi ed three genes modules (turquoise, blue and brown) that were signi cantly correlated with the Microglia-subtype trait. The largest module (i.e. turquoise module) included 96 genes that related to (1) Wound healing, Cell projection organization and PKB signaling biological processes; (2) Bacterial invasion and Phagosome KEGG pathways and (3) Ephrin signaling KEGG pathways ( Figure S4C). Interestingly, these genes were more strongly over-expressed in 8-mo versus 12-mo PAM and were slightly over-expressed in 12-mo CM suggesting that the normally occurring overexpression of these genes during normal ageing was accelerated in PAM ( Figure S4B). The blue module regrouped genes that were mostly down-regulated in PAM and tended to be less expressed / more down-regulated in 8-mo PAM. Moreover, these genes were also down-regulated in 12-mo CM. Genes of this module did not relate to any speci c GO biological processes, but were associated with MAP kinases, ErbB signaling and mitophagy KEGG pathways ( Figure  S4C). Finally, genes of the brown module are upregulated in PAM but down-regulated in older CM ( Figure  S4B). These genes are associated to hypoxia related and cell adhesion pathways ( Figure S4C).
Overall, our comparison of PAM versus CM in the APP/PS1 model shows that PAM exhibit profound transcriptomic changes which drive an increased in ammatory reaction, support morphological changes and contribute to the degradation of synaptic support functions. Although location at the proximity of Abplaques is the most important driver for transcriptomic changes, Age contributes, but to a lesser extent, to the observed alterations.

Biological functions and master gene regulators in PCM
Because amyloid plaques relate to one of the most prominent features of the disease, studies on the role of microglia in Alzheimer's disease have often focused on PAM. Quite the reverse, PCM whose morphology is very similar to that of CM are generally overlooked ( Figure S5A). However, these cells are also part of the pathological environment and we reasoned that they are likely to also contribute to the disease progression.
To investigate whether speci c biological functions were altered in PCM versus control CM, we identi ed genes signi cantly deregulated between the two conditions irrespective of age. PCA discriminated PCM from CM according to the second dimension ( Figure S5B), and statistical analysis identi ed 102 DEGs (FDR <0.05), the great majority (87/102) of which were up-regulated ( Figure 4A, Table S4A). Interestingly, as for PAM, deregulated PCM DEGs were very signi cantly enriched for the different pathological microglia gene signatures (i.e. IAM, DAM, ARM, MGnD, IRM signatures), the Reactome and the Sensome signatures ( Figure 4B). More surprisingly, PCM's DEGs were also highly enriched for PIG gene network. However, this may be explained by the fact that in Chen et al. [20] study, amyloid load was quanti ed based on 6E10 immunostaining which labels more diffuse Ab plaques. GO analyses also revealed that these genes are associated with immune related functions, including Tumor Necrosis Factor (TNF) and Cytokine production, Immune response and Antigen presentation ( Figure 4C; Table S4B). Cellular reaction in this microglia subtype was also demonstrated by deregulation of functions linked to Cell differentiation and Myeloid activation. Among those DEGs, WGCNA analysis identi ed 2 distinct modules of co-deregulated genes (Table S4A). The largest one contained the vast majority of the DEGs (91/102) and corresponded to genes that showed an age-dependent upregulation in PCM ( Figure 4D). The second module was limited to only 7 genes, including App, which by construction is over-expressed in the APP/PS1, and could not be related to a speci c biological function (data not shown). On the other hand, gene network analysis identi ed a cluster of 49 highly connected genes that are strongly associated with the Lysosome (p= 4.1x10 -11 ), the Antigen processing & presentation (p= 2.4x10 -10 ), and the Phagosome (p= 7.4x10 -6 ) KEGG pathways ( Figure 4E). Cd68, Ctsd, H2-aa and C3ar1 represented hub genes within this network.
As shown in Figure 4A, gene expression changes were quite variable in PCM with a general trend for higher deregulation in microglia isolated from older mice. Additionally, although the expression changes were more similar within 4-mo and 12-mo samples, the inter-individual variation appeared greater in microglia isolated at in 8-mo mice. To address whether age affected the transcriptomic changes observed in PCM, we rst identi ed 1334 genes deregulated in PCM versus CM (raw p-value<0.05), and then searched among them which ones are also deregulated in the Microglia-subtype:Age interaction (raw p-value<0.05) ( Figure S5C). We thus identi ed 595 genes whose expression changed in PCM in an agedepend manner. WGCNA analysis further identi ed two gene modules. The largest one (turquoise module) was signi cantly correlated with the Microglia-subtype factor. Genes of this module were upregulated in the intermediate and late stage of the disease ( Figure S5D, upper panel) and related to in ammatory processes, notably Cytokine production, Antigen presentation, Myeloid cell activation and the Phagosome KEGG pathway ( Figure S5E). Interestingly, these genes showed opposite regulation in CM being less expressed in 12-mo compared to younger cells. The second module (blue module) was signi cantly correlated with the Age factor and contained genes whose expression were down-regulated in an age-dependent manner speci cally in PCM ( Figure S5D, lower panel). Genes of the blue module related to Lipid oxidation, Organelle transport and Synaptic transmission biological processes ( Figure  S5E).
On the whole, these results demonstrated that although PCM are not associated to Ab plaques, and display homeostatic-like morphology, they exhibit age-dependent transcriptome alterations. These alterations are associated with important microglial functions that are typical of microglial reaction.
To what extent do PAM and PCM differ?
To further investigate the extent to which PAM and PCM differ at the transcriptomic level, we searched for genes signi cantly deregulated between the two microglia subtypes. Considering both the 8-mo and 12mo samples, we identi ed 551 DEGs (FDR < 0.05), of which 80% (446/551) were up-regulated in both PAM vs PCM ( Figure 5A). WGCNA analysis identi ed a single module of 497 genes, which was signi cantly correlated with the Microglia-subtype trait (r=0.95; p<2.10 -8 ) and more highly expressed in PAM ( Figure 5B, Table S5A). These 497 genes were associated with in ammation related biological processes ( Figure 5C; Table S5B) and, at least in part, overlapped with those deregulated in PCM vs CM, indicating that PCM present an intermediate reactive state between CM and PAM. However, some biological functions were speci c to PAM, including Chemotaxism, Cell proliferation, Cell architecture and ROS production. By comparing the three lists of DEGs (i.e. PAM vs CM, PCM vs CM and PAM vs PCM), we also identi ed 11 genes that were deregulated in PCM only ( Figure 5D). Globally, these genes showed signi cantly greater expression in PCM compared to CM, whereas in PAM their expression was either not different or lower (i.e. 1 gene, Efnb3) than in controls (Table S6). This latter result suggests that PCM are also engaged in speci c functions compared to PAM. However, these small panel of genes could not be associated to any speci c biological processes (not shown).
Among the genes deregulated in PAM versus PCM, we then identi ed 96 genes that were changed in Microglia-subtype:Age interaction (raw p-value<0.05) ( Figure S6A). WGCNA analysis re ned this list to 91 co-expressed genes that were more highly expressed in PAM compared to PCM. These genes were enriched for biological processes associated with Actin lament organization, Cell migration & differentiation, Peptidase activity ( Figure S6B and Table S7) and for the KEGG Chemokine signaling pathway (Table S7). They showed opposite age-dependent regulation in PCM and PAM, and thus globally appeared less up-regulated in 12-mo PAM ( Figure S6C and Figure S6D).
As a whole, these results indicate that although PAM and PCM share common signaling pathways, they are also engaged in speci c biological functions. Our data also reveal different age-dependent regulations in PCM and PAM.
To further explore the relative contribution of PCM and PAM to AD, we tested whether AD risk genes were enriched among the PAM and PCM DEGs. To that purpose, we used a list of genes from a recent and extensive GWAS study, converting the human ID genes for their murine orthologs [42]. Recent studies on polygenic risk scores have shown that genes with even small signi cance in GWAS carry information with regard to the risk of AD [43], thus we tested for enrichment in GWAS genes at different cutoffs ( Figure 5E). At all cutoffs (p-values ranging from 10 -6 to 10 -2 ), PAM DEGs were signi cantly enriched for GWAS-associated AD genes thus suggesting that PAM play a key role in AD pathogenesis. In contrast, PCM DEGs were signi cantly enriched for GWAS AD genes only at the lowest cutoffs. This suggests that PCM can contribute to AD pathogenesis, although, to a lower extent than PAM.

Validation in brain tissue
Laser microdissection can be used to isolate discrete cells from complex environments while preserving

Discussion
Microglia reaction in AD was rst evidenced using bulk RNA-seq studies performed on puri ed microglia [6,7]. The molecular heterogeneity of microglia in this pathological context was then studied using top-down approaches (typically scRNA-seq). With these approaches, microglia subtypes are rst identi ed based on transcriptomic similarities, their potential functions are then inferred based on geneontology analyses, and nally their location in the tissue are assessed retrospectively based on the expression of a handful of markers [11][12][13]44]. Based on markers that are up-regulated in PAM, other studies isolated PAM and PCM using FACS, and performed in depth characterization of these two populations [16,18].
Here, to decrypt alterations in microglial cells that are or not associated to Ab-plaques, we used an unbiased bottom-up approach. We combined laser microdissection and RNA-seq to study transcriptome remodeling in these two spatially de ned microglia subpopulations. Further we investigated the evolution of these alterations during the progression of the disease, from the early stage, when the plaques barely form, to the late stage, when Ab-plaque load stabilizes. We con rmed that dense amyloid plaques drive striking transcriptomic alterations, leading to reactive microglia that display strong in ammatory phenotypes and are less supportive to neuronal functions. Besides, our study also provides the rst in depth characterization of plaque-distant microglia (PCM), highlighting that although this microglia subtype does not show major morphological alterations, it exhibits, from the early stages, an increased in ammatory phenotype that progresses with the development of the pathology. Thus, our study reveals that PAM and PCM are both involved in AD progression, and engaged in functions which are only partially overlapping.
Combination of LCM and RNA-seq provide a unique way to decipher the respective roles of discrete microglia subpopulations We have previously shown that Cx3cr1 haplode ciency does not alter disease progression in the APP/PS1 model [25]. Here we demonstrate that quick TR staining and laser microdissection procedures preserves RNA quality. By isolating no less that 1'600-2'000 microglia cells from 4 mice per experimental conditions (32 mice in total), we were able to accurately analyze the transcriptome of spatially distinct microglial populations.
Although LCM does not allow to reach the level of purity of FACS, our data reveal that the large majority of cells analyzed are microglia. First, expression of different brain cell speci c markers revealed a 10-fold enrichment of microglial markers with parallel depletion in astrocyte, neuronal and oligodendrocyte markers. Second, in-situ hybridization of speci c genes of interest con rmed their microglial localization.
In CX3CR1 +/gfp mice, all myeloid cells express GFP, thus we cannot exclude that some of the microdissected cells are in fact in ltrated monocytes or peri-vascular macrophages. However, contamination by these speci c cell types is likely to be low since (1) they tend to express lower level of GFP [45], and (2) we preferentially selected high GFP expressing cells for microdissection. Consistent with a strong enrichment of microglia in the LCM-isolated cells, we showed a high correlation between gene expression in LCM and FACS isolated GFP+ cells from CX3CR1 +/gfp mice. We also showed reduced expression of macrophage and monocyte markers in the LCM isolated cells. This nding is consistent with (1) the transcriptional remodeling observed in tissue surrounding Ab-plaques [21] and (2) recent studies in both mouse and human tissues showing that microglia are the only myeloid cells present at the vicinity of Aβ deposits [46,47], and endorses that the cells we analyzed were indeed microglia.
So far, most of our understanding of microglial gene expression changes in AD has come from RNAseq [6, 7] and scRNA-seq studies [11-13, 18, 44]. Yet, a major issue in transcriptional pro ling of dissociated cells, including scRNA-seq, is the evoked transcriptional perturbations that may occur during tissue dissociation, and which may bias the true detection of disease induced transcriptional changes [19,40]. In LCM approach, the tissue is preserved throughout the procedure and cells are isolated from dehydrated slices which prevents any procedure driven transcriptional perturbations. Accordingly, here, we showed very low expression of immediate early genes in the isolated microglia. Although our technical approaches slightly differ, our results agree with those of Merienne et al. [48] who showed that combining LCM with RNA-seq analyses in reporter mice represents a useful approach to decrypt cell-type speci c gene expression patterns. Our results further expand the value of the approach as here we show that short staining can be added to the procedure to allow isolation and transcriptome analysis of spatially distinct cell subtypes.
AD reactive microglia and DAM/ARM have been associated with reduced expression of microglial homeostatic genes and loss of homeostatic functions [5,6,11,13]. In contrast, our results show that, across age/disease progression, gene expression of homeostatic microglial genes stays relatively stable in CM, PCM and PAM, with only small down-regulation observed in 12-mo PAM. These results are consistent with qPCR analysis of whole cortical tissue, which shows slight down-regulation of the homeostatic genes Pr2y12 and Tgfb only in 12-mo samples (reanalysis of data in [25] and data not shown). Our results also agree with a recent study that revealed that downregulation of homeostatic microglia genes only occurs in mouse models with advanced neurodegeneration [49]. Altogether, these results support that impairment of homeostatic functions occurs speci cally in PAM at the late stage of the disease, and suggest that, in previous studies, brain dissociation procedures may have altered transcriptomic data.

Ab plaques drive prominent alterations in PAM transcriptome
Being located close to one of the earliest and most prominent feature of AD pathology, PAM have been the subject many studies. Yet, they have mainly been investigated using either low throughput techniques (i.e. mainly imaging based) or indirect (i.e. top-down) medium / large throughput approaches (for review [50]). By combining direct isolation of PAM through LCM and RNA-seq analysis of their transcriptome, our approach provides a comprehensive understanding of the molecular remodeling in this spatially identi ed microglia subpopulation. First, we showed that microglial proximity to amyloid plaques is the most important factor in microglia transcriptomic remodeling in this AD mouse model.
These results are consistent with previous studies showing (1) enrichment, within LCM isolated plaques tissues samples, of DEG identi ed in aging TgCRND8 AD mice cortical brain homogenates [21], and (2) strong transcriptomic alterations in Ab-phagocytosing microglia [18]. It is also in line with a recent spatialtranscriptomic study that revealed gradual co-expression of PIG genes as function of Ab accumulation [20].
More speci cally, extending previous ndings [6, 16, 51], we showed that PAM DEGs are enriched for functions / GO-terms linked to in ammation and immune related pathways. In line with these ndings, we showed that PAM DEGs are enriched for the microglial reactome signature [33]. In addition to their roles in neuroin ammation processes, PAM have been associated with more neuroprotective functions in relation with Amyloid-b processing. Indeed, they have been proposed to play key roles in Ab encapsulation and in plaque compaction [50][51][52][53]. Consistent with PAM clustering around plaques to form a physical barrier, we identi ed DEGs associated with (i) chemotaxis and cell migration and (ii) cell morphology (for example actin polymerization, actin organization, lamellipodium assembly). These later changes are also consistent with morphological changes commonly observed in PAM. We also identi ed enrichment for GO-terms linked to APP processing, phagocytosis, lysosome organization, which are consistent with an involvement of PAM in Ab clearance [5,18]. In addition to the above functions that were mainly associated with up-regulated genes, we also identi ed a module of co-repressed genes which relates, at the cellular level, to synapse establishment and functioning, and at the behavioral level with learning and memory. These microglia expressed genes may represent interesting targets to restore neuronal functioning. Together with our nding that the microglial Sensome signature [32] is also strongly altered in PAM, these results suggest that PAM lack to provide the appropriate support for correct synapse and neuronal functioning. These data are consistent with results showing lower response of PAM to damage signals [54].
PAM have been associated with, and sometimes confounded by, microglia subtypes identi ed by scRNAseq, i.e. DAM, ARM [11,13] or by differential clustering, i.e. MGND [5]. Our direct analysis of PAM, con rmed that PAM DEGs are strongly enriched for genes of the ARM, DAM and MGnD signatures.
Consistent with their close proximity to Ab plaques, we also found a strong enrichment of PAM DEGs for the PIG signature [20]. Our data also showed a strong enrichment for the IRM signature among the PAM DEGs suggesting that IRM are also present at proximity of the Ab plaques. This remarkable phenotypic heterogeneity of microglia even in a spatially restricted area, i.e. at proximity of Ab plaques, was further evidenced by our in-situ data showing that genes enriched in PAM are expressed at varying levels in different PAM of the same plaque. The origins of this local diversity remain to be established and may include paracrine regulation, cell memory mechanisms potentially involving epigenetic marks, and/or contact duration with the plaques.
Our study also provides information relative to the temporal changes in the microglial transcriptomic remodeling in PAM. Thus, we identi ed three subsets of genes that are deregulated in PAM in an age dependent manner. In particular, we identi ed a small subset of genes that are up-regulated in PAM independently of age but decreased in aged CM. These genes are associated with the Syndecans Reactome pathway. Syndecans are known to bind to extracellular matrix molecules [55]. Maintenance of high level of these genes throughout the disease progression may represent a protective mechanism aimed at maintaining Ab plaque compaction to prevent diffusion of smaller and more toxic oligomeric forms of Ab. These genes are also associated with the KEGG Hif signaling pathway. HIF is associated with reactive oxygen species (ROS) formation, which suggests that aged PAM retain their ability to regulate ROS production. Moreover, Hif1a signature has been associated with the Abphagocytosing microglia subtype [18]. We also identi ed two other subsets of genes, which contain genes that are respectively globally up-regulated and down-regulated in PAM but showed opposite deregulation in older versus CM and PAM. Such opposite deregulations in older CM and PAM are consistent with ageing and AD sharing similar molecular mechanisms, while AD representing an accelerated ageing. Additionally, our nding that subsets of genes exhibit less deregulation in older compared to younger PAM suggest that constant stimulation of microglia by external stimuli such as Ab plaques may, on the long term, lead to cell exhaustion / "burn out". This observation is consistent with results from PET studies which showed longitudinal decline of microglial reaction in some subsets of patients [56,57].

PCM, morphologically intact but nevertheless reactive
In contrast to PAM, PCM have been overlooked in AD studies, and, when studied, this subpopulation was generally compared to PAM rather than CM. Here, the use of LCM provided us with the unique opportunity to comprehensively characterize PCM in AD.
We de ned PCM as high GFP expressing cells located more than 100 µm away from any TR+ plaque. These criteria are similar to those used by Rothman et al. [21] to identify brain tissues associated with plaques. Because thiazine-red stains dense core Ab plaques, we cannot exclude the possibility that some of the microdissected PCM were associated to more diffuse Ab deposits that are TR-but would have been 6E10+. However, during microdissection we paid attention to target isolated microglia and did not consider microglia showing any sort of clustering. Thus, PCM likely relate to microglia that are in contact with soluble or oligomeric Ab species or even no Ab species at all. One important observation is that although PCM overall morphology is not changed compared to control microglia (present data and [14]), this microglia subtype displayed a time-dependent increase in gene deregulation that correlates with disease progression. This dynamic deregulation mirrors the global increase of the neuroin ammatory status observed in the cortex of both mice and humans as the disease progress [25,57], and parallels the increase in soluble Ab fractions in this AD mouse line (data not shown). By showing speci c deregulations in PCM, we demonstrated that this microglia subtype is not a by-stander but rather plays signi cant roles in AD progression.
Our results strongly suggest that, in addition to densely aggregated Ab species, microglia also react to Ab small oligomers or eventually monomers. Microglial detection and reaction to small oligomers are highly relevant to AD progression as these oligomers are more bioactive at synapses and drive stronger microglia reaction [58]. Our results are also consistent with data correlating tissue transcriptomic changes to 6E10 staining and showing a gradual increase in the network connectivity of PIG [20].
Enrichment of PIG in the PCM signature supports that the expression of the so called "plaque induced genes" are not restricted to dense Ab plaques. Similarly, our results show that genes deregulated in PCM are also enriched for the DAM / ARM / MGnD and IRM microglia signatures. This was further con rmed by our in-situ results showing expression of Cst7 and Clec7a, two ARMs markers, and Cybb in PCM. Thus, our data revealed that although DAM / ARM / MGnD and IRM are further represented around Ab plaques, these microglia subtypes are likely to be involved in disease progression more globally. Our data provide a new and comprehensive overview of the potential role of PCM in AD progression and identi ed several hub/key genes. Indeed, functional and gene network analysis of genes deregulated in PCM revealed that these cells are reactive, involved in the immune response notably by producing cytokines. In particular, we show that PCM are involved in antigen presentation and processing pathways. Gene network analysis identi ed several hub genes in this microglia subtype, namely H2aa, Cd68 and Ctsd. H2aa is part of the MHC class II protein complex and has been involved in antigen presentation [59]. In control microglia H2aa is virtually not expressed, but interestingly, we showed that it is already overexpressed in 4-mo PCM demonstrating that these cells may be recruited at the very early stage of the disease. Cd68 and Ctsd are involved in phagocytosis and in lysosomal functions thus suggesting that not only PCM can detect Ab but also react and switch-on clearing mechanisms. The microglial Axl and Merkt TAM receptor tyrosine kinases have recently been shown to be essential mediators of Ab plaques recognition and engulfment and it has been proposed that TAM-driven phagocytosis promotes rather than inhibits densecore plaque development [52]. Our data showed that Axl is also up-regulated in PCM in the early/intermediate stage of the disease. This suggest that this receptor may not only be involved in plaque compaction but may also be involved in plaque formation with both processes representing tentative neuroprotective mechanisms aimed at sequestrating small, more toxic, Ab oligomers. More broadly, this supports that PCM might play an important role in early Ab plaque formation.
Our transcriptomic data also revealed that PCM display an in ammatory pro le that build-up as the pathology progresses. Importantly, we showed that PCM reactive phenotype starts from the early stages when Ab load is small and plaque density low (i.e. Less than 1 plaque / mm 2 re-analysis from [25].

Conclusions
Overall, here, we showed that LCM combined with RNA-seq allow to analyze transcriptome remodeling in spatially distinct cells. Our data con rm and extend previous studies on the role of PAM in AD progression; they offer a comprehensive view of the molecular changes in these cells. As a whole, our results show that PAM may have both positive and negative impacts on the surrounding tissues: they may represent a physical barrier to prevent small Aβ oligomers spreading, but are also releasing proin ammatory cytokines that are deleterious for the surrounding cells and are less supportive for synaptic functions. Importantly, our study reveals for the rst time that, although PCM display a homeostatic-like phenotype, those microglia subtype are reactive, engaged in speci c biological processes and actively participate to the disease progression. Our data suggest that they may be key for detecting small Aβ oligomers and for initiating plaque formation. Finally, we reveal further molecular heterogeneity in both PAM and PCM. Further work is needed to understand how this very local heterogeneity builds-up, to identify its dynamics, and to determine the consequences for the disease progression. Deciphering these mechanisms will allow to target speci c subpopulation of microglia with the ultimate goal of promoting bene cial microglial functions and alleviating deleterious ones.

Abbreviations
A-L H-G performed RNAscope experiments and prepared the gures; CM developed the LCM-based microglia protocol and performed laser microdissection; MP and CR performed the WGCNA analyses; NL contributed to the laser microdissection; JL helped design the stud; CR made major contribution towards implementing the LCM-based microglia protocol and performed RNAseq experiments. FR funded part of the study, was a major contributor in designing the study and in writing the manuscript. HH: designed and funded the study; performed the bioinformatic analyses and write the manuscript.
All authors read and approved the nal manuscript  parenchymal microglia (PCM, oranges) and plaque associated microglia (PAM, purples). Shades code for age: 4-mo, light color; 8-mo, median color; and 12mo, dark color.

Figure 3
Microglial gene expression remodeling in plaques associated microglia (A) Heatmap of differentially expressed genes in PAM (violets) versus CM (greens) along aging. The scaled expression value (row Z score) is shown in a blue-red color scheme with red indicating higher expression, and blue lower expression. The full data is available in Table S2A