Perivascular cells induce microglial phagocytic states and synaptic engulfment via SPP1 in mouse models of Alzheimer’s disease

Alzheimer’s disease (AD) is characterized by synaptic loss, which can result from dysfunctional microglial phagocytosis and complement activation. However, what signals drive aberrant microglia-mediated engulfment of synapses in AD is unclear. Here we report that secreted phosphoprotein 1 (SPP1/osteopontin) is upregulated predominantly by perivascular macrophages and, to a lesser extent, by perivascular fibroblasts. Perivascular SPP1 is required for microglia to engulf synapses and upregulate phagocytic markers including C1qa, Grn and Ctsb in presence of amyloid-β oligomers. Absence of Spp1 expression in AD mouse models results in prevention of synaptic loss. Furthermore, single-cell RNA sequencing and putative cell–cell interaction analyses reveal that perivascular SPP1 induces microglial phagocytic states in the hippocampus of a mouse model of AD. Altogether, we suggest a functional role for SPP1 in perivascular cells-to-microglia crosstalk, whereby SPP1 modulates microglia-mediated synaptic engulfment in mouse models of AD.

Alzheimer's disease (AD) is characterized by synaptic loss, which can result from dysfunctional microglial phagocytosis and complement activation. However, what signals drive aberrant microglia-mediated engulfment of synapses in AD is unclear. Here we report that secreted phosphoprotein 1 (SPP1/osteopontin) is upregulated predominantly by perivascular macrophages and, to a lesser extent, by perivascular fibroblasts. Perivascular SPP1 is required for microglia to engulf synapses and upregulate phagocytic markers including C1qa, Grn and Ctsb in presence of amyloid-β oligomers. Absence of Spp1 expression in AD mouse models results in prevention of synaptic loss. Furthermore, single-cell RNA sequencing and putative cell-cell interaction analyses reveal that perivascular SPP1 induces microglial phagocytic states in the hippocampus of a mouse model of AD. Altogether, we suggest a functional role for SPP1 in perivascular cells-to-microglia crosstalk, whereby SPP1 modulates microglia-mediated synaptic engulfment in mouse models of AD.
The integrity of neurons and synapses is critically dependent on brain-resident macrophages, which include microglia and perivascular macrophages (PVMs) that continuously monitor and clear phagocytic targets across the lifespan 1,2 . Microglia, tissue-resident macrophages of brain parenchyma, contribute to circuit refinement through engulfment of synapses and axon tracts during postnatal development, whereas PVMs associate with the perivascular space where they represent the first responders to toxic agents and pathogens that may cross the blood-brain barrier (BBB) into the brain parenchyma [3][4][5][6][7] . Microglia and PVMs both share similar yolk-sac origin and reside on the parenchymal side of the BBB, yet occupy distinct microenvironments and express unique cell-type-specific markers [8][9][10][11][12][13] . Because of the immediate juxtaposition of perivascular space with the brain parenchyma, we reasoned that microglia and PVMs could potentially influence each other to coordinate phagocytosis in response to central nervous system perturbations. For example, in Alzheimer's disease (AD), many Article https://doi.org/10.1038/s41593-023-01257-z increase of Spp1 mRNA expression levels in hippocampal CA1 sections of App NL-F mice compared to WT mice by smFISH in intact tissue, which was further confirmed by qPCR analysis on brain homogenates (Fig. 1g,h). The specificity of SPP1 antibody and Spp1-targeting smFISH probes was validated by absence of signals in Spp1 KO/KO mice (Extended Data Fig. 1c).
Interestingly, Spp1 mRNA expression in the hippocampal CA1 was specifically enriched within the stratum lacunosum-moleculare (SLM) layer and Spp1 + cells displayed an elongated vascular-like pattern (Extended Data Fig. 1d). Costaining with pan-endothelial marker GLUT1 showed cellular zonation of Spp1 mRNA expression adjacent to vascular structures, whereas outside of the endothelial membrane, no Spp1 mRNA was detected (Extended Data Fig. 1d, inset). Using high-resolution confocal imaging and 3D surface rendering, we confirmed the close association of cytosolic SPP1 protein expression with GLUT1 + vasculature (Fig. 1i). Similar to the murine App NL-F hippocampus, we also found a striking presence of SPP1 immunoreactivity along the vasculature in the hippocampus of postmortem AD patient tissues ( Fig. 1j and Supplementary Table 1). Altogether, these data suggest that SPP1 expression is enriched along the vasculature in App NL-F mice and AD patients. Further, we found negligible levels of Aβ oligomers (oAβ) or plaques at this early time point in the parenchyma of App NL-F hippocampus (Extended Data Fig. 1e) 30 . In contrast, however, we observed oAβ deposition along the hippocampal vasculature at 6 months, assessed by the oAβ-specific NAB61 + immunostaining juxtaposed to GLUT1 + vasculature (Extended Data Fig. 1f) 33 . The latter was confirmed using two other anti-Aβ antibodies, the 4G8 and HJ3.4, which recognize Aβ17-24 and Aβ1-13, respectively (Extended Data Fig. 1g) 34 . The vascular oAβ deposition was more pronounced in 15-month App NL-F mice, in contrast to age-matched WT controls in which no positive NAB61 staining was found (Extended Data Fig. 1h). These data suggest early, preplaque vascular Aβ deposition in the App NL-F , coinciding with SPP1 activation in the perivascular space.

SPP1 is expressed by PVMs and fibroblasts
Spp1 expression in 6-month App NL-F hippocampus was remarkably restricted to cells along the perivascular space ( Fig. 1). Given the role of SPP1 as a marker of macrophage subsets in peripheral tissues, we assessed whether Spp1 is expressed by brain-resident macrophages associated with the vascular space, that is Cd163 + CD206 + PVM 35 . Spp1 mRNA expression predominantly colocalized with pan-PVM markers Cd163 and CD206 juxtaposed to GLUT1 + hippocampal vasculature (Fig. 2a,b). In contrast, we failed to observe Spp1 expression in Tmem119 hi P2Y12 + microglia of 6-month App NL-F animals (Extended Data Fig. 2a). Of note, Spp1 expression in microglia was detected only at later stages, that is, in 6E10 + plaque-rich 15-month App NL-F hippocampus (Extended Data Fig. 2b), likely reflecting SPP1 + disease-associated microglia 25,36 . Further, immunophenotypic characterization revealed restricted expression of SPP1 within CD206 + CX3CR1 + cells, and Spp1 + cells were found positive for PVM-specific platelet factor 4 (Pf4) and Cd163 (Extended Data Fig. 2c,d) 11,13,37 . Interestingly, we also found rare Spp1 + Cd163 − Pf4 − cells in the hippocampal perivascular space (Extended Data Fig. 2d). Spp1 mRNA has been suggested by scRNA-seq to be expressed by vascular leptomeningeal cells, also known as PVFs, which are located within parenchymal arteries (Extended Data Fig. 2e) 37 . Indeed, the second cluster of Spp1 + cells coexpressed the pan-PVF marker Pdgfra (encoding CD140a), suggesting PVFs as a second cellular source of SPP1 surrounding GLUT1 + vasculature in the hippocampus, albeit much less abundant compared to PVMs (Extended Data Fig. 2f,c).
To further assess in vivo SPP1 expression, we developed Spp1-IRES-TdTomato (TdT) (Spp1 TdT ) reporter mice, which carry an IRES-TdT cassette in Spp1 exon 7 retaining endogenous SPP1 expression (Extended Data Fig. 3a-c). Flow cytometric analysis of naïve animals revealed the expression of Spp1-TdT within 25.9% of pregated CD45 hi CD206 + PVM, in contrast to CD11b + CD45 int CX3CR1 hi microglia that were almost devoid of TdT expression (0.9% of which were TdT + ) of the identified risk variants point toward defective phagocytic and endolysosomal pathways in microglia and PVMs; however, intercellular and intracellular mechanisms governing this impairment remain unclear 14 . In line, microglia mediate regional synapse loss in AD, but environmental cues that modulate potential microglial functional states are poorly understood 15 .
Here we uncovered secreted phosphoprotein 1 (SPP1/osteopontin), predominantly derived from PVM, as an extrinsic modulator of microglial phagocytosis. We observed that SPP1 is required for activation of complement-initiator C1q and synaptic engulfment by microglia in AD mouse models. SPP1 has been shown in multiple peripheral tissues to regulate phagocytosis by macrophages [16][17][18][19][20] . In AD patients, secreted SPP1 levels are increased in cerebrospinal fluid (CSF) and plasma; further, SPP1 is found on plaque-associated microglia and has been suggested as a conserved disease-associated marker among mice and human [21][22][23][24][25][26][27] . However, the role of SPP1 in the brain or its relevance in AD is unknown. Here we found a region-specific activation of SPP1 in the hippocampal perivascular space of AD mouse models as well as in AD patient tissues. Super-and ultra-structural examination revealed that SPP1 is expressed predominantly by PVMs, and to a lesser extent, by perivascular fibroblasts (PVFs) in the adult hippocampus. In mouse models of AD, we found that perivascular SPP1 is upregulated in a region-specific manner at onset of synaptic elimination by microglia. Genetic ablation of SPP1 ameliorated microglial phagocytic states in AD mouse models as well as C1q activation, leading to the prevention of microglia-synapse engulfment and synapse loss in these mice despite amyloid-β (Aβ) challenge. Using single-cell RNA sequencing (scRNA-seq) and computational ligand-target predictions (NicheNet 28 ), we noted multiple autocrine and paracrine signaling pathways to be potentially modulated by perivascular SPP1, with many of these pathways converging on microglial phagocytic functional states. Altogether, our data suggest a functional crosstalk between PVMs and microglia and propose perivascular SPP1 as an extrinsic signal that modulates synapse phagocytosis by microglia.

SPP1 upregulation at onset of microglia-synapse phagocytosis
To address whether SPP1 is dysregulated at a time point when synapses are vulnerable to microglial engulfment in an AD-relevant context, we used the slow-progressing App NL-F mouse model, where control by the endogenous App promoter allows for physiological cell-type specific and temporal regulation of Aβ production 15,29,30 . We first assessed phagocytic microglia-synapse interactions in the App NL-F hippocampus at the age of 6 months, an age that precedes robust plaque deposition in the hippocampal parenchyma 30 . We quantified internalized level of Homer1-immunoreactive synaptic puncta within CD68 + P2Y12 + microglial lysosomes and observed an approximate sevenfold increase in synaptic engulfment by microglia in App NL-F mice as compared to those of age-and sex-matched wild-type (WT) mice (Fig. 1a,b). We also observed upregulation of C1q, the initiating protein of the classical complement cascade that mediates microglial phagocytosis of synapses 3,15,31 (Fig. 1c,d). C1qa expression was contained within Tmem119 + microglia as assessed by single-molecule fluorescent in situ hybridization (smFISH), confirming microglia as the main producers of C1q in the adult hippocampus (Fig. 1c, insert) 15,32 . We next assessed potential cross-regulation between synaptic engulfment and SPP1, a glycoprotein associated with phagocytosis in peripheral macrophages, in the 6-month App NL-F hippocampus 18 . Using 3D-stimulated emission depletion (3D-τ-STED) super-resolution imaging, we found an approximately threefold increase of punctate SPP1 protein immunoreactivity in the CA1 hippocampus of App NL-F mice as compared to age-and sex-matched WT controls (Fig. 1e,f). The SPP1 upregulation was region-specific, that is, in hippocampus but not cerebellum (Extended Data Fig. 1a,b). In line with increased SPP1 production, we found an approximate threefold Article https://doi.org/10.1038/s41593-023-01257-z ( Fig. 2d-f). Further, only approximately 6% of total live TdT + cells were CD140a + (gene product of Pdgfra), in contrast to 23.9% for CD206 + , highlighting PVMs as a predominant cellular source of SPP1 (Fig. 2e,f). Using high-resolution confocal imaging, we found that the distribution of SPP1-TdT was comparable to the SPP1 immunoreactivity detected by IHC (Fig. 1i), that is, along GLUT1 + vasculature of the hippocampus (Extended Data Fig. 3d). Further, we saw similar cellular localization of SPP1-TdT within CD206 + PVMs in the Spp1 TdT hippocampus (Fig. 2g). Finally, we used correlative light and electron microscopy (CLEM) to target and visualize the ultrastructure and environmental context of SPP1-TdT positive cells in the hippocampus (Fig. 2h, Extended Data Fig. 3e and Supplementary Video 1). CLEM identified SPP1-TdT expressing cells as lysosome-rich PVMs located within the basement membrane of the perivascular space ( Fig. 2h and Extended Data Fig. 3e). CLEM and mRNA in situ localization identified PVMs as a source for SPP1 in mice, which translated to human tissue, where we found enrichment of SPP1 within the perivascular space of AD patients, occasionally overlaying with CD206 + cells (Fig. 2i). Collectively, our data suggest that SPP1 is predominantly expressed by CD206 + Cd163 + Pf4 + PVMs and Pdgfra + /CD140 + PVFs in the mouse hippocampus during onset of microglia-synapse engulfment. Similar to perivascular SPP1 upregulation in mice, we also found perivascular SPP1 expression in CA1 hippocampus of AD patient brains.

SPP1 drives microglial engulfment of synapses in AD context
We next addressed the functional consequences of the SPP1 increase that spatiotemporally coincides with onset of microglia-synapse engulfment. To determine whether SPP1 has a role in microglial phagocytosis of synapses, we first performed intracerebroventricular (ICV) injections of   (Fig. 3a). Concurrently, we observed a robust upregulation of SPP1 in the contralateral hippocampus of WT mice 18 h after ICV injection of oAβ (Fig. 3b). This upregulation coincided with microglial C1q activation and peak engulfment of Homer1-immunoreactive synapses by microglia in contralateral hippocampus of oAβ-injected WT mice (Fig. 3c, d,h). In contrast, oAβ failed to upregulate C1q or induce microglial synaptic engulfment in Spp1 KO/KO mice, suggesting that SPP1 is necessary for microglia-synapse engulfment.
To test whether Spp1 deficiency altered phagocytic signature in microglia in 6-month App NL-F mice, we crossed the Spp1 KO/KO mice to App NL-F mice (App NL-F ·Spp1 KO/KO ). Using smFISH-IHC and 3D reconstruction, we observed upregulation of Grn and Ctsb, encoding for progranulin and Cathepsin B, respectively, key components of the endolysosomal processing machinery in P2Y12 + microglia of App NL-F animals ( Fig. 4a-d) 14 . However, in the App NL-F ·Spp1 KO/KO hippocampus, we found substantially decreased levels of Grn and Ctsb mRNA expression by P2Y12 + microglia . In fact, Grn and Ctsb expression levels were similar to WT and Spp1 KO/KO mice. These results suggest that microglia fail to adopt phagocytic states in App NL-F ·Spp1 KO/KO mice as compared to those of App NL-F mice, despite oAβ production. Further, we observed a sharp reduction of engulfed synapses inside P2Y12 + microglia of App NL-F ·Spp1 KO/KO mice (Fig. 4e,f). These data altogether suggest that the absence of SPP1 prevents microglial phagocytosis as well as synaptic engulfment in response to amyloidosis . We next tested whether microglia in Spp1 KO/KO mice display intrinsic defective phagocytic capacity. We assessed the ability of microglia to engulf synaptosomes in vitro using isolated primary microglia from postnatal WT versus Spp1 KO/KO mice (Extended Data Fig. 4a). We found that engulfment of oAβ-bound synaptosomes was similarly performed by microglia isolated from either genotype (Extended Data Fig. 4b-e). This suggests that diminished microglia-synapse engulfment in Spp1 KO/KO mice in vivo is not due to an inability of microglia to phagocytose or respond to oAβ, but likely results from defective extrinsic SPP1 signaling in the brain. Corroborating this hypothesis, the application of a  23.9 ± 2.9 0.9 ± 0.3 25.9 ± 2.4

Fig. 2 | SPP1 is expressed by PVMs and fibroblasts. a-c,
Representative images of Spp1 mRNA expression juxtaposed to GLUT1 + vasculature, colocalizing with pan-PVM markers Cd163 (a), CD206 (b) and PVF (Pdgfra + ) (c) in 6-month App NL-F SLM as characterized by smFISH-IHC. Scale bar represents 10 µm. Data are representative of four App NL-F mice examined over two independent experiments. d-f, Representative FACS plots to identify PVMs (CX3CR1 + CD45 + CD11b + CD206 + (d), microglia CX3CR1 high CD45 + CD11b + (e) or gated on total TdT expressing cells isolated from Spp1 TdT hippocampal homogenates and quantification (f)). One datapoint represents one cell type per mouse (PVM, PVF and microglia) pooled from n = 7 mice examined over two independent experiments (f). P values from one-way ANOVA, Kruskal-Wallis test. g, Three-dimensional reconstruction of CD206 + PVMs expressing SPP1-Td along GLUT1 + vessels in SLM from Spp1 TdT mice. Scale bar represents 7 µm. Data are representative of three Spp1 TdT mice examined over two independent experiments. h, Representative single serial section SEM backscatter electron image of a representative SPP1-TdTpositive PVM as identified by CLEM (Upper). SPP1-TdT-positive cell manually pseudocolored red, together with neuropil (pink), astrocytes (lilac), smooth muscle cells (purple), endothelial cells (cyan) and other perivascular cells (green), shown with reduced opacity over the electron microscopy data (lower). Accompanying confocal overlays and correlation images shown in Extended Data  Table 1). Data are shown as mean ± s.e.m. TdT, TdTomato.
Finally, to determine the consequence of Spp1 deficiency on synapse numbers in 6 mo App NL-F animals, we used super-resolution microscopy to analyze presynaptic and postsynaptic markers (Bassoon and Homer1, respectively) in the hippocampal CA1 stratum radiatum of 6 mo App NL-F mice (Fig. 4g,h). While we observed synaptic loss in colocalized Bassoon and Homer1-immunoreactive synaptic puncta density in App NL-F versus WT mice, Spp1 deficiency restored this synapse loss (Fig. 4h). Likewise, we observed that acute oAβ challenge failed to induce synapse loss in Spp1 KO/KO mice as it did in WT mice, suggesting that SPP1 is required for oAβ-induced synapse loss (Extended Data Fig. 5). Altogether, these results suggest that SPP1, likely secreted from PVMs in vivo in response to oAβ vasculature deposition, acts as an extrinsic signal to promote microglial engulfment of synapses and synapse loss in 6-month App NL-F animals.

Spp1 regulates perivascular-microglial interaction networks
To obtain insight into how SPP1 may regulate interactions between perivascular cells, that is PVM and PVF, and parenchymal microglia in the App NL-F hippocampus, we dissected hippocampi from 6-month WT, Spp1 KO/KO , App NL-F , and App NL-F ·Spp1 KO/KO animals and sorted CD45 high CD11 int CD206 + PVMs, CD140a + PVFs and CD45 int CD11 int CX3CR1 high microglia for scRNA-seq analysis using the 10× Genomics platform. After quality control, we ran unsupervised clustering and annotated cell types based on expression of known marker genes (Extended Data Fig. 6a,b and Supplementary Table 2). Microglia expressed exclusive pan-markers such as Sall1, Tmem119 and P2ry12; PVMs showed enrichment in Mrc1, Pf4 and Cd163 (Extended Data Fig. 6b) 11 . Of note, the sorted CD140a + cells also included oligodendrocyte precursors (OPCs), which were discerned according to expression of OPC-specific markers including Lhfpl3, Sox6 and Bcan (Extended Data Fig. 6b). Further, PVFs were positive for Cdh5, Lama1 and Dcn that distinguished them from OPC. Interestingly, we did not observe Spp1 mRNA expression levels in the sorted PVMs (Extended Data Fig. 2d). This conforms with previous scRNA-seq datasets from homogenized and isolated tissues 35 ; however, translatome analysis of PVMs of brains of lipopolysaccharide (LPS) endotoxin challenged Cx3cr1 ccre :Lyve1 ncre :RiboTag mice have demonstrated the presence of Spp1 in Lyve1 + PVMs (Extended Data Fig. 6c) 11,37 . Indeed, comparative RiboTag analysis of challenged split-Cre animals that target PVMs and microglia, respectively 11 , confirmed prominent Spp1 expression in PVMs as compared to microglia (Extended Data Fig. 6d). The discrepancy of detected expression levels between scRNA-seq  One datapoint represents one ROI per mouse hippocampus, with total of n = 3 mice per genotype and time point, examined over one independent experiment. P values from two-way ANOVA. c, Representative images of C1q expression in SLM of 3-month PBS versus oAβ-injected WT mice. Scale bar represents 20 µm. d, Representative 3D reconstructed images showing Homer1 engulfment within CD68 + lysosomes of P2Y12 + microglia from WT mice injected with oAβ. Scale bar represents 5 µm. e, Representative images of C1q expression in Spp1 KO/KO mice injected with PBS versus oAβ. f, Representative 3D reconstructed images showing Homer1 engulfment within CD68 + lysosomes of microglia from Spp1 KO/KO mice injected with oAβ. Scale bar represents 5 µm. g, Quantification of C1q particles (puncta) in WT or Spp1 KO/KO mice treated with either PBS or oAβ, as in c and e. One datapoint represents one ROI, with total of 10 (WT) and 9 (Spp1 KO/KO ) ROIs pooled from n = 3 mice per genotype, examined over two independent experiments. P values from two-way ANOVA. h, Quantification of Homer1 engulfment in WT or Spp1 KO/KO P2Y12 + microglia, ICV treated with either PBS or oAβ, as in d and f. One datapoint represents one individual P2Y12 + microglia with a total of 15-16 cells pooled from n = 2 animals per genotype examined over two independent experiments. P values from two-way ANOVA. Data are shown as mean ± s.e.m. Article https://doi.org/10.1038/s41593-023-01257-z of isolated cells versus the RiboTag approach, which bypasses cell isolation and sorting 38,39 , and the highly sensitive smFISH methods suggest that Spp1 signature within PVMs is highly regulated and closely associated to changes in its microenvironment, as earlier demonstrated for many other microglial transcripts 40 . It further suggests that Spp1 + PVMs likely do not survive isolation and highlights the relevance of studying SPP1 in an intact spatial context.
Next, we used NicheNet on the scRNA-seq datasets to predict ligand-target links between interacting cells 28 . Specifically, we applied NicheNet to investigate how intercellular communication between PVM/PVF (ligand) and microglia (target) is altered by Spp1 deficiency in 6-month App NL-F mice (Fig. 5). We hypothesized SPP1 to function through paracrine and/or autocrine mechanisms, that is, modulating microglial function through direct or indirect signaling within the perivascular space. Among top predicted ligands affected in PVMs and PVFs by Spp1 deficiency in 6-month App NL-F hippocampus is transforming growth factor-beta 1 (TGF-β1), a cytokine critical for microglial development and previously shown to be modulated by SPP1 within fibroblasts 41,42 (Fig. 5a-

Fig. 4 | SPP1 drives microglial engulfment of synapses in AD context. a-d,
Representative image of 3D reconstructed P2Y12 + microglia expressing phagocytic markers Ctsb (a,b) and Grn (c,d) assessed by smFISH-IHC in 6-month WT, Spp1 KO/KO , App NL-F and App NL-F ·Spp1 KO/KO SLM. Scale bar represents 7 µm. Quantification of Ctsb (b) and Grn (d) mRNA levels expression within P2Y12 + microglia. One datapoint represents one individual P2Y12 + microglia with a total of 18 microglia (b) and 18 (WT, Spp1 KO/KO ), 17 (App NL-F ) and 15 (App NL-F ·Spp1 KO/KO ) microglia (d) pooled from n = 3 mice per genotype examined over two independent experiments. P values from one-way ANOVA, Kruskal-Wallis test. e, Representative 3D reconstructed images showing Homer1 engulfment within CD68 + lysosomes of P2Y12 + microglia in 6-month App NL-F versus App NL-F ·Spp1 KO/KO SLM. Scale bar represents 7 µm. f, Quantification of Homer1 engulfment ratio in P2Y12 + microglia of WT versus Spp1 KO/KO versus App NL-F versus App NL-F ·Spp1 KO/KO mice. One datapoint represents one individual P2Y12 + microglia with a total of 23 (WT, App NL-F ), 24 (Spp1 KO/KO ) and 25 (App NL-F ·Spp1 KO/KO ) microglia pooled from n = 3 mice per genotype examined over two independent experiments. P values from one-way ANOVA, Kruskal-Wallis. g, Representative super-resolution images of Homer1 and Bassoon puncta colocalization in 6-month WT, Spp1 KO/KO , App NL-F and App NL-F ·Spp1 KO/KO SLM. Scale bar represents 5 µm. h, Quantification of Homer1/ Bassoon colocalization density normalized to WT or Spp1 KO/KO accordingly. One datapoint represents the average of one animal (3-5 ROIs per animal) with a total of n = 4 animals per genotype. P < 0.0066 (WT) and 0.0831 (Spp1 KO/KO ) from two-way ANOVA, Bonferroni's multiple comparison test. Data are shown as mean ± s.e.m. Article https://doi.org/10.1038/s41593-023-01257-z Metallopeptidase Domain 17 (ADAM17) and calreticulin, the latter being reported as an 'eat-me' signal driving macrophage phagocytosis of apoptotic cells 43 . Further, expression of Itgb5, Itgb1 and Itgav, encoding integrin receptors and subunits of the SPP1 receptor, were affected in microglia of App NL-F ·Spp1 KO/KO versus App NL-F mice, as expected 44 . Other top affected receptors in microglia were Tgfbr1, Tgfbr2 and protein tyrosine phosphatase Ptpn6 (Fig. 5b,c) 45,46 . We further validated these results via smFISH-IHC; as suggested by NicheNet, Tgfbr1 and Itgb5 expression levels were dysregulated in microglia of App NL-F ·Spp1 KO/KO versus App NL-F mice ( Fig. 5e and Extended Data Fig. 6e,f). In addition, we confirmed decreased expression of CD29 (Itgb1) and CD321 (F11r) in hippocampal microglia of App NL-F ·Spp1 KO/KO compared to App NL-F animals by flow cytometry (Fig. 5f,g and Extended Data Fig. 6f). Altogether, NicheNet revealed evidence for paracrine and autocrine crosstalk signals between the perivascular space and microglia in App NL-F mice in the context of SPP1 signaling, with the majority of the top hits implicating potential alterations of microglial phagocytic states.

Discussion
Microglia-synapse engulfment has been demonstrated to be relevant to synaptic loss and dysfunction in a wide variety of neurologic diseases across the lifespan, including neurodevelopmental and  Itga6  Igf1r  Epha2  Dag1  Dab2  Csf2rb  Cd151  Cd47  TIr2  Tgfbr2  Tgfbr1  F11r  AcvrI1  Bmpr1a  Itgb1  Itgb5  Itga9  Ncstn  Eng  Itgav  Itgb3  Lrp5  II7r  Lrp1  AIox5  Ptch1  II6ra  Notch1  Sort1  Tnfrsf1a  Tnfrsf1b   Btg2  Ccnd1  Itgb1  Ptpn6  Cd34 Vasp Calr Percent expressed 0 25 50 75 Average expression regulates perivascular-microglial interaction networks. a, Expression level of selected ligands expressed by cell types known to express Spp1 (PVF and PVM), by cell type and genotype. Radius of dot is proportional to the percentage of cells expressing the gene; color is the scaled gene expression level. b, Predicted receptor genes for ligands represented in a, which show differential expression in microglia (receiver cells). Color represents the predicted interaction potential. c, Predicted target genes downstream of receptors identified in b, which show differential expression in microglia (receiver cells Article https://doi.org/10.1038/s41593-023-01257-z neuropsychiatric disorders, acute injury and virus-induced cognitive impairment, and multiple sclerosis and neurodegeneration 15,29,[47][48][49][50][51][52][53][54][55][56] . Genetic studies in sporadic AD further highlight the importance of understanding microglial phagocytosis 14 . Therefore, our study demonstrating a functional role for SPP1 in microglia-synapse phagocytosis will likely have broad relevance to understanding neuroimmune mechanisms of synapse vulnerability in AD and other neurologic diseases involving synaptopathy. Here we report a role for perivascular SPP1 as an upstream mediator of microglia-synapse engulfment in both genetic (App NL-F ) and acute oAβ challenge mouse models. SPP1 is a multifunctional glycoprotein that was originally identified as a proinflammatory cytokine secreted by T cells and later found to be expressed in distinct tissue-resident macrophages linked with active clearance of apoptotic cells, chemotaxis and macrophage migration 18,57 . In the brain, SPP1 expression appears to be highly regulated in a spatiotemporal and cell type-specific manner, depending on context, age and brain region. In the perinatal and prenatal brain, Spp1 is expressed by microglia and associates with axon tracts of the corpus callosum, in contrast to the adult brain where SPP1 expression is thought to be restricted to glutaminergic and GABAergic neurons in adult hindbrain 11,37,[58][59][60] . Using various imaging tools including smFISH-IHC and CLEM in Spp1 TdT mice, we demonstrate that SPP1 expression in the hippocampus of healthy adult WT animals is restricted to the perivascular space, that is, Pf4, CD163 + and CD206 + PVM and to a much lesser extent, the closely neighboring PVFs. The finding that PVMs act as a major source of SPP1 upon oAβ challenge is consistent with the enrichment of ribosome-attached Spp1 transcripts in Lyve1 + PVMs compared to Sall1 + microglia and is consistent with our earlier study that shows Spp1 upregulation in Lyve1 + PVMs upon LPS challenge in adult mice 11 . Further, we found PVMs as the predominant cellular origin for SPP1 activation in early App models coinciding with Aβ aggregation along the vasculature. Of note, a recent study demonstrated early Spp1 upregulation in spinal cord PVFs in a mouse model of amyotrophic lateral sclerosis, preceding microglial activation and onset of motor neuron loss 61 . This suggests that SPP1 activation might represent a conserved molecular response to perturbed perivascular homeostasis, potentially beyond Aβ pathology.
We further recapitulated the perivascular localization of SPP1 in postmortem human brains of AD patients. These results are in line with a recent single-nucleus transcriptomic study in vessel-associated structures of the human cortex that found SPP1 expression in perivascular myeloid cells 62 . Our study thus raises an intriguing question of what could possibly trigger SPP1 elevation within the perivascular space. In AD patients, SPP1 levels are elevated in CSF 21-23 , which flows via the perivascular space toward the subarachnoid space. The cellular origin or the functional relevance of SPP1 increase, however, has been unclear. PVMs are distributed along the perivascular space and may have a critical role as sentinels in amyloidosis (our data here and ref. 63 ). Clearance of Aβ across the BBB represents an important homeostatic function, the impairment of which has been linked to exacerbated vascular as well as parenchymal deposition of Aβ 64,65 . Blood vessels therefore may represent a frequent, early and vulnerable site for Aβ deposition. In support of this, vascular pathology is commonly observed in AD, afflicting approximately 80% of patients 66 . Our findings of oAβ deposition along the vasculature in early stages of amyloidosis in App NL-F mice further support this view. Indeed, it was postulated that early failure of perivascular drainage could contribute to parenchymal Aβ deposition and associated neuronal toxicity 67 . Furthermore, SPP1 has been linked to the clearance of vascular Aβ deposits in mice 68 . Altogether, these and our studies highlight Aβ deposition along the vasculature as a possible trigger for SPP1 upregulation in PVMs and PVFs. Future studies allowing cell-type specific fate mapping and mutagenesis of SPP1 in hippocampal PVM and PVF subsets will be needed to provide insight into the origin and fate of SPP1-expressing perivascular subsets.
Our observation that SPP1 signaling promotes synaptic engulfment by microglia in the hippocampus of AD models suggests that SPP1 functions to regulate phagocytic cell states. Indeed, in the absence of Spp1, microglia fail to upregulate key phagocytic and AD-relevant genes such as progranulin and Ctsb in 6-month App NL-F mice. Of note, phagocytosis induced by microglial progranulin as well as complement proteins including C3 has been considered protective against Aβ plaque load and neuronal loss in plaque-rich mouse models of AD. In line with our data, this suggests that the timing of phagocytic activity in different stages of disease might be critical to determine the protective versus detrimental nature of the outcome 15,69,70 . Another interesting question raised is how SPP1 modulates microglial phagocytosis in the presence of Aβ pathology. SPP1 functions as a secreted or intracellular isoform 71 . Based on our data, a direct effect of intracellular SPP1 on microglia function seems unlikely in our experimental setting. First, we found that microglia-intrinsic Spp1 signaling is not required for the engulfment of Aβ-treated synaptosomes, where primary microglia derived from WT and Spp1 KO/KO mice exhibit similar engulfment capacities of synaptosomes in the presence of extracellular SPP1. This is consistent with work in primary gut macrophages from Spp1-deficient mice, whereby low doses of recombinant SPP1 restored phagocytic function toward opsonized bacteria, akin to what we observed for primary microglia and synaptosomes 72 . Second, we did not observe mRNA or protein SPP1 expression localized inside hippocampal microglia at the age of 6 months, yet the same cells fail to engulf synapses in App NL-F ·Spp1 KO/KO mice or Spp1 KO/KO mice challenged with synaptotoxic Aβ. Third, we found increased protein levels of SPP1-immunoreactive puncta within the SLM parenchyma using super-resolution imaging, suggesting that the observed punctate signals are mostly extracellular. Altogether, these results suggest that SPP1 is secreted by PVMs and PVFs and influences microglia phagocytosis.
Precise mechanisms of how microglial engulfment is initiated by SPP1 are to be investigated. SPP1 could opsonize neuronal material for example damaged synapses for microglial phagocytosis 73 . It could also engage directly with the canonical receptors, α v β 3 , which are expressed on microglia 37 , to trigger downstream signaling pathways. Indeed, using NicheNet, we observed Itgb3 and Itgav, which encode α v β 3 , to be downregulated in microglia of App NL-F ·Spp1 KO/KO mice. Our NicheNet analysis further shows reduced expression levels of both Tgfb1 in PVMs and its receptor Tgfbr2 in microglia from App NL-F ·Spp1 KO/KO mice, suggesting that SPP1 potentially coordinates perivascular-microglial crosstalk via autocrine TGF-β signaling. Indeed, SPP1 has been identified to promote autocrine TGF-β signaling in fibroblasts in a mouse model of muscular dystrophy, thereby promoting fibrosis by neighboring fibroblasts and macrophages 41 . Further, TGF-β is a major determinant of microglia maturation and homeostasis, and Tgfbr2 deficient microglia demonstrate dysregulated expression of transcripts related to phagosome formation and immune activation 42,45 . In contrast to decreased Tgfbr2, we observed increased Tgfbr1 signaling in microglia of App NL-F ·Spp1 KO/KO versus App NL-F animals. It is tempting to speculate that this paradoxical down-and upregulation reflects compensatory mechanisms to counteract microglia phagocytosis induced by SPP1. Finally, TGF-β1 has been shown to be increased in CSF and perivascular space of AD patients, where its levels positively correlate with Aβ deposition along blood vessels, and TGF-β1-overexpressing mice show enhanced microglial phagocytosis and plaque amelioration in a complement-dependent manner 74,75 . Further studies are needed to gain better insight into the role of TGF-β signaling in governing autocrine and paracrine PVM-microglial crosstalk in AD. Other predicted Spp1-specific pathways in microglia of App NL-F mice include Calr, encoding for calreticulin, a multifunctional chaperone protein that has been described as an 'eat-me' signal 43 . Altogether, our results suggest multiple mechanisms by which microglial phagocytosis could be modulated by perivascular SPP1 signaling. Determining molecular mechanisms of how these pathways may contribute to synaptic engulfment will Article https://doi.org/10.1038/s41593-023-01257-z be important to better understand multicellular events and crosstalk between perivascular space and the brain parenchyma.
Altogether, our study nominates perivascular SPP1 as a modulator for microglial phagocytosis, impacting microglia-synapse interactions and synaptic homeostasis. In peripheral tissues, insight into stromamacrophage interactions demonstrates intricate and functionally relevant immune crosstalk that underlies critical tissue remodeling and homeostasis 76 . Likewise, our results highlight a potential role for perivascular-microglia interactions in the brain, and the presence of an SPP1-producing perivascular niche may offer opportunities to specifically manipulate and block microglia-mediated synaptic engulfment.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41593-023-01257-z.

Mice
All experiments were performed in accordance with the UK Animal (Scientific Procedures) Act, 1986 and following local ethical advice. Experimental procedures were approved by the UK Home Office and ethical approval was granted through consultation with veterinary staff at University College London (UCL). For all experiments, C57BL/6J (WT) mice were obtained from Charles River UK. Spp1 KO/KO (B6.129S6(Cg)-Spp1 tm1Blh /J; stock 4936) and CX 3 CR-1 GFP (B6.129P2(Cg)-Cx3cr1 tm1Litt /J; stock 5582) mice were obtained from The Jackson Laboratory. App NL-F/NL-F mice were obtained from F. A. Edwards (Department of Neuroscience, Physiology & Pharmacology, UCL, UK) 30 . Cx3cr1 ccre :Sall1 ncre (split-Cre) R26-LSL-tdTom:Rpl22 HA (RiboTag) and Cx3cr1 ccre :Lyve1 ncre R26-LSL-tdTom:Rpl22 HA homozygous animals were maintained in specific pathogen-free conditions and handled according to protocols approved by the Weizmann Institute Animal Care Committee as per international guidelines. All animals were housed under temperature-controlled (temperature, 23.1 °C; humidity, 30-60%) and pathogen-free conditions with 12 h light/12 h dark cycle with an ad libitum supply of food and water. Both male and female age-matched mice were used in this study: WT males and females, Spp1 KO/KO males, CX 3 CR-1 GFP males, Spp1 tm1(tdTomato)Msasn males, App NL-F/NL-F males and App NL-F ·Spp1 KO/KO males.

Spp1 reporter mouse model development
All animal work was approved by the Jackson Laboratory Animal Care and Use Committee and adhered to the standards of the Guide for the Care and Use of Laboratory Animals set forth by the NIH. The Spp1 tm1(tdTomato)Msasn mouse allele was generated using direct delivery of CRISPR-Cas9 reagents to mouse zygotes. An IRES-tdT construct was introduced in the mouse Spp1 gene (Ensembl Gene UID, ENS-MUSG00000029304; Extended Data Fig. 3a). Analysis of genomic DNA sequence surrounding the target region, using the Benchling (www.benchling.com) guide RNA (gRNA) design tool, identified a gRNA sequence (AACAAGAAAAAGTGTTAGTG) with a suitable target endonuclease site at the stop codon of exon 7 of the mouse Spp1 locus. Streptococcus pyogenes Cas9 (SpCas9) V3 protein and gRNA were purchased as part of the Alt-R CRISPR-Cas9 system using the crRNA:tracrRNA duplex format as the gRNA species (IDT). Alt-R CRISPR-Cas9 crRNAs (1072532, IDT) were synthesized using the gRNA sequences specified in the DESIGN section and hybridized with the Alt-R tracrRNA (1072534, IDT) as per the manufacturer's instructions. A plasmid construct with a 2 kB 5′ homology arm ending 62 bp past the Spp1 stop codon in exon 7, IRES, tdT coding sequence, bGH poly(A) signal and 1.5 kB 3′ homology arm was synthesized by Genscript (Extended Data Fig. 3b). To prepare the gene editing reagent for electroporation, SpCas9:gRNA ribonucleoprotein complexes were formed by incubating Alt-R-SpCas9 V3 (1081059, IDT) and gRNA duplexes for 20 min at room temperature in embryo tested Tris-EDTA (TE) buffer (pH 7.5). Fertilized mouse embryos were generated via natural mating and cultured as described previously 77 . C57BL/6J (Stock 000664, The Jackson Laboratory) donor female mice (3-4 weeks of age) were superovulated by administration of 5 IU of pregnant mare serum gonadotrophin via intraperitoneal (ip) injection (HOR-272, ProSpec) followed 47 h later by 5 IU (ip) human chorionic gonadotrophin (hCG) (HOR-250, ProSpec). Immediately postadministration of hCG, the female was mated 1:1 with a C57BL/6J stud male and 22 h later checked for the presence of a copulation plug. Female mice displaying a copulation plug were sacrificed, the oviducts excised and embryos collected. Electroporation was performed as described in ref. 77 . In brief, zygotes were treated with the acidic Tyrode's solution (T1788, Millapore-Sigma) for 10 s and washed extensively in prewarmed M2 media (M7167, Millapore-Sigma). Zygotes were then placed in 10 µl drops of Opti-MEM media (3198570, Thermo Fisher Scientific-Gibco). Ten microliters of gene editing reagent solution, including the SpCas9/ gRNA and the ssDO, was mixed with the Opti-MEM drops with the embryos and deposited into a 1 mm electroporation cuvette (45-0124, Harvard Apparatus). Electroporation was performed in an ECM830 Square Wave Electroporation System (45-0661, BTX). The electroporation setting was a 1 ms pulse duration and two pulses with 100 ms pulse interval at 30 V. Following the electroporation, a prewarmed 100 µl aliquot of M2 media was deposited into the cuvette with a sterile plastic pipette to recover the embryos. The zygotes were removed from the cuvette and washed in prewarmed M2 media. Embryos were immediately transferred into B6Qsi5F1 pseudopregnant female mice, an F1 hybrid strain produced by breeding C57BL/6J female mice with the inbred Quackenbush Swiss line 5 mouse strain 78 .
Founders were selected that were positive by short-range PCR assays, had appropriate sequence across the homology arm junctions, were negative for the plasmid backbone and had correct sequence of the inserted construct. These were bred to C57BL/6J.
Once the line was established, mice were genotyped using a Taqman qPCR protocol run on a real-time PCR instrument (Roche LightCycler480). Forward primers for the wild-type allele (AAACA-CAGTTCCTTACTTTGCAT producing a 94 bp product) and the mutant allele (AGGATTGGGAAGACAATAGCA in the bGH poly A producing an 84 bp product) were combined with a common reverse primer (CACT-GAACTGAGAAATGAGCAGT) using an annealing temperature of 60 °C. The wild-type probe (5′ HEX fluorophore label) used was TGTTAGT-GAGGGTTAAGCAGGAATA and the knock-in probe (5′ FAM fluorophore label) used was ATGCGGTGGGCTCTATGG, each utilizing a black hole quencher on their 3′ end. These were run with an EndPoint protocol, imaging negative (quenched) fluorescent values at the completion of the cycling protocol.

ICV injections of S26C oAβ
S26C Aβ(1-40) 2 dimers were purchased from Phoenix Pharmaceuticals 15 . Adult (2-3 months) mice were anesthetized with 4% isoflurane followed by maintenance at 1.5-2% isoflurane during surgery. Surgery was performed after head fixing in a stereotaxic frame (World Precision Instruments). Marcaine (0.025%) was applied locally, the skull was exposed by a single incision along the midline and a unilateral craniotomy was drilled with a 0.9-bit drill burr (Hager and Meisinger). Pulled long-shaft borosilicate pipettes (Drummond Scientific) were backfilled with mineral oil before loading with oAβ (1 ng µl −1 ) or sterile PBS vehicle. Four microliters total volume was injected into the right lateral ventricle (stereotaxic coordinates in millimeters from Paxinos and Franklin's The Mouse Brain in Stereotaxic Coordinates, Fourth Edition; AP: −0.40, ML: 1.00, DV: −2.50) using a Nanofil 10 ml syringe (World Precision Instruments) connected to an UltraMicroPump-3 (World Precision Instruments) at a flow rate of 400 nl min −1 . The pipette was left in place for 5 min after complete substance injection and slowly withdrawn to avoid backflow along the pipette track. The incision on the scalp was closed with Vetbond tissue adhesive (3 M). Subcutaneous carprofen (Carprieve, 5 mg g −1 body weight) and buprenorphine (Vetergesic, 0.1 mg g −1 body weight) diluted in 0.9% saline were administered perioperatively. Animals received carprofen (33.33 mg ml −1 ) Article https://doi.org/10.1038/s41593-023-01257-z in their drinking water until tissue collecting. The left hemisphere, contralateral to the injection site, was analyzed.

Immunohistochemistry
Deeply anesthetized mice were transcardially perfused with 25-30 ml of filtered PBS, followed by 20 ml of 4% ice-cold ultrapure PFA (Generon, 18814-20). Brains were removed from the skull and fixed in 4% PFA (Generon) for 24 h at 4 °C. For cryoprotection, brains were rinsed in PBS for at least 2 h to remove excess PFA and placed in 30% sucrose for 48 h before embedding in optimal cutting compound. Fresh cryosections were blocked with 5% BSA, 0.2% Triton X-100 and 5% Donkey serum in PBS at room temperature for 90 min before incubation with primary antibodies at 4 °C overnight. All primary antibodies were diluted in a blocking buffer. After 4X PBS washes for 15 min each, sections were incubated with secondary antibodies in blocking buffer for 2 h at room temperature and then washed again with PBS. All secondary antibody aliquots were centrifugated at 15,000g for 15 min before being diluted in blocking buffer. The sections were then incubated for 5 min in DAPI diluted in PBS before being mounted either in Prolong Gold Antifade Mounting Medium (Thermo Fisher Scientific, P36930) for confocal imaging or in Prolong Glass Antifade Mountant (P36982) for STED imaging. For immunostaining of C1q and synapses, 30 µm free-floating tissue sections were washed in PBS, followed by pretreatment in 1% Triton X-100 for 20. Sections were then blocked in 20% NGS, 1% BSA and 0.3% Triton, in PBS for 2 h followed by standard primary antibody incubation overnight at 4 °C. Sections are washed in 0.3% Triton X-100 in PBS for 30 min followed by secondary incubation for 4 h at room temperature and wash for 30 min before mounting.

Postmortem human brain tissue
Brains were donated to the Queen Square Brain Bank (QSBB) for neurological disorders (UCL Queen Square Institute of Neurology). All tissue samples were donated with full, informed consent. Accompanying clinical and demographic data of all cases used in this study were stored electronically in compliance with the 1998 data protection act and are summarized in Supplementary Table 1. Ethical approval for the study was obtained from the NHS research ethics committee and in accordance with the human tissue authority's code of practice and standards under license number 12,198, with an approved material transfer agreement. The cohort included pathologically diagnosed cases of AD (n = 6) and neurologically normal controls (n = 6). The level of AD pathology in all cases was assessed using current diagnostic consensus criteria 79,80 . The APOE genotype was also determined for each case as previously described 81 . Information regarding sex is included in Supplementary  Table 1; however, sex was not taken into consideration when including patient samples. Findings did not apply to only one sex .

Immunohistochemistry on postmortem human brain
Slides with 8 µm mounted tissue sections from the frontal cortex were incubated at 60 °C overnight. Sections were deparaffinized in xylene and rehydrated in decreasing grades of alcohol. Slides were incubated in methanol/hydrogen peroxide (0.3%) solution for 10 min to block endogenous peroxidase activity. For heat-induced antigen retrieval, slides were then transferred to a boiling solution of 0.1 M citrate buffer (pH 6.0) and pressure cooked at maximum pressure for 10 min. Nonspecific binding was blocked by incubating slides in 10% nonfat milk for 30 min at room temperature. Sections were incubated in anti-SPP1 antibody for 1 h at room temperature. After three gentle 5 min washes in tris-buffered saline with tween (TBS-T); slides were incubated for 45 min in biotinylated goat anti-mouse IgG secondary antibody (Vector Laboratories, BA-9200, 1:200). Slides were washed and incubated in avidin-biotin complex (ABC; Vector Laboratories) for signal amplification. The slides were then washed for a final time and 3,3′-diaminobenzidine was used as the chromogen and counterstained in Mayer's hematoxylin (BDH). Finally, slides were dehydrated in increasing grades of alcohol (70%, 90% and 100% industrial methylated spirits), cleared in xylene and mounted.
Double immunofluorescence staining was carried out on the frontal cortex. Sections were prepared as detailed above up to the incubation of the anti-SPP1 antibody, secondary biotinylated goat anti-mouse and ABC. Antibody binding was visualized using a TSA Cyanine 3 amplification kit (PerkinElmer), which was applied to sections for 20 min at room temperature. After TBS-T washing, sections were incubated with anti-CD206 antibody for 1 h at room temperature and species-appropriate Alexa Fluor 658 secondary antibodies (Invitrogen, 1:1,000) for 2 h at room temperature to visualize the antibody. Sections were washed a final three times in TBS-T and mounted using Vectashield antifade mounting medium (Vector Laboratories).

Image acquisition
Images were acquired using a Zeiss LSM800 confocal microscope (×40 objective, 1.3NA oil, ×20 objective 0.8-NA and 63×0.8-NA oil). Settings were kept constant for all sections in the same comparison group.
Step size was determined using the optimal interval adjustment on the Zen blue software with a stack size of 10-14 µm for all on-slide IHC experiments and 6-10 µm for RNAScope experiments.

Secreted SPP1 fluorescence intensity analysis
For quantification of secreted SPP1, Triton X-100 was excluded from the blocking buffer to retain signals of secreted and membrane-bound protein immune reactivity. To quantify SPP1 protein expression, images were processed in Fiji ImageJ (NIH) 82 . An automated ImageJ macro was created to analyze the signal intensity of each slice in the z stack and select the plane with the highest signal intensity. Background was subtracted with a rolling ball radius of 10 pixels. Images were thresholded with consistent thresholding parameters and made into binary images based on intensity. As pixel intensity information has been translated into area, the particle analysis function was used to quantify the total immune-reactive area.

In vivo microglial engulfment analysis
Engulfment analysis was performed as previously described 15 . Thirty micrometers of free-floating tissue sections were immunostained with IBA1, CD68 and Homer1. For each mouse, four to six regions of interest within CA1 SLM were acquired using a 63 × 1.4-NA objective on Nature Neuroscience Article https://doi.org/10.1038/s41593-023-01257-z a Zeiss 800 microscope. Next, 60-80 z-stack planes were taken with 0.27-µm spacing and raw images were processed in Imaris (Bitplane) for analysis, after background subtraction of Homer1 channel. P2Y12 + cells and CD68 lysosomes were surface rendered with 0.25-µm and 0.1-µm smoothing, respectively. A mask was applied within P2Y12 + CD68 + reconstructed lysosomes for Homer1, and the percentage engulfment of Homer1 within lysosomes was calculated using the following formula: volume of engulfed material (Homer1 within CD68)/total microglial volume × 100.

Super-resolution imaging and synapse analysis
Super-resolution synapse images were acquired on a Zeiss LSM 880 microscope with Airyscan detector using a 63x, 1.4-NA oil immersion Plan-Apochromat objective (theoretical maximum resolution-140 nm lateral, 350 nm axial). A zoom factor of 1.8× and a 37 × 37 × 144 nm voxel pixel size. The Airyscan detector was aligned before imaging each new slide. Three regions of interest of 1.15 µm were acquired in the center of the hippocampal CA1 stratum radiatum for each brain section. Super-resolution synapse images were processed in Zen Black using 3D Airyscan Processing. Imaris software was used for presynaptic and postsynaptic puncta detection, using intensity centers to maximize the detection of immunoreactive spots. Presynaptic and postsynaptic spots were colocalized using a MATLAB colocalization script (Colocalize Spots XTension), using a colocalization distance of 0.25 µm between spot centers. Synapse density is shown as App NL-F mice normalized to WT mice, and App NL-F ·Spp1 KO/KO normalized to Spp1 KO/KO mice.

RNA isolation, reverse transcription and RT-qPCR
Mice were anesthetized and perfused with ice-cold PBS. Next, fresh cortex and hippocampus were dissected and homogenized with TissueLyser II (QIAGEN) using the mRNA easy mini kit (QIAGEN) as described by the manufacturer. Next, RNA purity and concentration were assessed by Nanodrop. mRNA was converted to cDNA using the qScript cDNA SuperMix reverse transcription kit as described by the manufacturer (95048, Quantabio). For RT-qPCR, 12 ng of cDNA was loaded in triplicates per gene in a total volume of 20 µl using the SYBR green PCR master mix as described by the manufacturer (4309155, Thermo Fisher Scientific). The reaction was run using a LightCycler 96 Instrument (Roche) with white 96-well plates (04729692001, Roche). Triplicate Ct values were averaged and data are shown as respective to the geomean of three housekeeping genes (Actb, Gapdh and Rpl32) using the Ct delta method . Primers purchased from IDT were used at a concentration of 200 nM. Actb, forward-CATTGCTGACAG-GATGCAGAAGG, reverse-TGCTGGAAGGTGGACAGTGAGG; Gapdh, forward-CATCACTGCCACCCAGAAGACTG, reverse-ATGCCAGT-GAGCTTCCCGTTCAG; Rpl32, forward-ATCAGGCACCAGTCAGAC-CGAT, reverse-GTTGCTCCCATAACCGATGTTGG. Spp1, forward-ATC TCACCATTCGGATGAGTCT, reverse-TGTAGGGACGATTGGAGTGAAA

Microglia isolation for flow cytometry and FACS sort
PBS-perfused brains were quickly isolated from the skull and the hippocampus was dissected on ice using chilled instruments. For scRNA sequencing, brains were perfused with inhibitor cocktail including Actinomycin D (5 µg ml −1 ) and triptolide (10 µM). Next, single-cell suspension was prepared using the Adult Brain Dissociation kit from Miltenyi Biotec (Bergisch Gladbach, Germany), according manufacturer's instructions and ref. 25 . Afterwards, cell suspension was filtered through a 70 µM cell strainer before mixing with debris removal solution. Cells were centrifuged (300g for 10 min), washed with ice-cold FACS buffer (PBS, 2% FBS, 0.78 mM EDTA) and incubated for 30 min at 4 °C with FACS buffer containing Fc block (BD Biosciences) and primary antibody mix. Flow cytometry data were analyzed using FACSDiva software 4.0 and FlowJo 10 software (Treestar).

Primary microglia isolation and culturing
For generation of microglial primary cultures, P0 WT or Spp1 KO/KO mice were decapitated, the brain was dissected from the skull and meninges were removed in ice-cold HBSS with 5% FBS. Eight to ten mice were pooled per culture preparation. Tissue was homogenized first with 2 ml pipettes (15 strokes) in 15 ml falcon tube and subsequently transferred to a prewet 50 ml tube with 70 µM strainer. The 15 ml tube was washed with HBSS and then put through filter to ensure all tissues were collected. The supernatant was removed and the cell pellet was resuspended in ice-cold 35% isotonic percoll. The interface was carefully created with HBSS. The samples were centrifuged for 40 min at 4 °C at 2,800g with no break and with slow acceleration and deceleration. The myelin layer and supernatant layers were aspirated, and the cell pellet was washed in HBSS. Cells were centrifuged and resuspended in 1 ml microglial media (DMEM F12 (Gibco), 5% FBS (Gibco), 1% pen-strep (Gibco), 50 ng ml −1 CSF1 416-ML-010/CF (RnD Systems), 50 ng ml −1 TGFb1 7666-MB-005/CF (RnD Systems) and 100 ng ml −1 CX3CL1 472-FF-025/CF (RnD Systems)) for cell counting. Cells were used within 7-10 d of plating.

Isolation of synaptosomes and conjugation to pHrodo
Three to five 8-week-old animals were pooled per experiment. In brief, mice were intracardiac perfused with 10 ml ice-cold PBS. The hippocampi and cortices were dissected on ice. Tissue was weighed and homogenized in five volumes of sucrose homogenization buffer (5 mM HEPES pH 7.4, 320 mM sucrose and 1 mM EDTA) using a Dounce homogenizer with 15-20 strokes. The homogenate was centrifuged at 3,000g for 10 min at 4 °C and the supernatant was saved as total homogenate fraction (THF). The THF was centrifuged again at 14,000g for 12 min at 4 °C and supernatant was saved as cytosolic fraction. The pellet was carefully resuspended in 550 µl of Krebs-Ringer buffer (KRB: 10 mM HEPES, pH 7.4, 140 mM NaCl, 5 mM KCl, 5 mM glucose and 1 mM EDTA) and 450 µl of Percoll solution (for a final concentration of 45%). The solution was mixed by gently inverting the tube and an interface was slowly created with 400 µl of KRB. After centrifugation at 14,000g for Article https://doi.org /10.1038/s41593-023-01257-z In vitro synaptosome engulfment assay Primary mouse microglia were treated with 1 µg of pHrodo-conjugated S26C oAβ-treated synaptosomes. Plates were then placed in a CD7 with the incubator at 37 °C and 5% CO 2 . Fluorescent (594 nm and 647 nm) and brightfield (oblique and phase) images were acquired at a ×20 objective (×0.5) at intervals of 3-5 min. A three-slice z stack was taken at a 1.5-µm interval to ensure that imaging was within focus throughout the imaging session; however, one plane was used for analysis. For analysis, the z-profile axis was plotted for respective pHrodos on ImageJ with respect to time. Fluorescence intensity at t = 0 was subtracted from subsequent time frames.

3D-τ-STED/STED-FLIM
For visualization of secreted SPP1 in mouse and human tissue, tissue sections were imaged with Leica STELLARIS 8 STED microscope using the ×100 objective (1.4 NA oil) (Leica Microsystems). Tissue sections were imaged at least 24 h after being coverslipped with mounting medium to avoid discrepancies in fluorescence lifetime within each section. STED microscope with Fluorescence Lifetime Imaging (STED-FLIM) was used to visualize secreted SPP1 in the extracellular space and fluorescence lifetime information was used to gate fluorescence signals. Alignment between STED laser and excitation laser was performed before each imaging session. A 775 nm STED laser was used to generate a doughnut beam to silence the peripheral fluorophores from Alexa Flour 647 photoexcitation to achieve subdiffractional resolution. For all STED images, pixel size was limited to at most 50 nm. STED laser intensity was set at 20%. All images were taken with a step size of 0.15 µm. Fluorescence signal was time-gated from −0.5 ms to 4.5 ms. For analysis, raw images were processed by thresholding, followed by binary image transformation to match ROI with intensity. Finally, fluorescence intensity was measured by the particle analysis function in ImageJ (NIH).

Volume correlative light and electron microscopy with array tomography
Mice were perfusion fixed with 4% PFA (EM grade) in ice-cold PBS as described above, the brain was dissected and left in 4% PFA in PBS overnight at 4 °C. The following day, coronal vibratome slices (100 µm) of the brain were collected and slices containing clear cross sections of the hippocampus were manually trimmed to minimally contain the hippocampus and ensure that the tissue piece was asymmetric. Low-resolution maps of the entire hippocampus tissue piece were taken using a confocal Zeiss LSM800 and ×10 lens with montaging for gross mapping and identification of regions of interest. High-resolution (63 × 1.4 NA) confocal stacks were taken of regions of interest. Slices were fixed further with 2% formaldehyde/1.5% glutaraldehyde in 0.1 M sodium cacodylate, before being processed for volume EM using a modified protocol based on the NCMIR protocol 83 . Briefly, tissue slices were incubated in 1% osmium tetroxide/1.5% potassium ferricyanide for 1 h at 4 °C, before being washed and left in 0.1 M sodium cacodylate in a fridge overnight. The following day, tissue was incubated in 1% thiocarbohydrazide for 15 min at 60 °C, 2% osmium tetroxide for 30 min, 1% uranyl acetate for 30 min and Walton's lead aspartate for 30 min at 60 °C, with numerous distilled water washes between each step. The samples were consequently dehydrated through an ethanol series, embedded in Epon resin and baked overnight at 60 °C. Using the maps acquired by light microscopy, the block was trimmed down to the approximate region of interest and serial sections were collected on ITO-coated coverslips using an ultramicrotome (Leica) and diamond knife (Diatome). Coverslips were mounted on SEM stubs using carbon stickies and silver DAG and array tomography serial back scattered electron images (5 nm pixels) were acquired for each CLEM cell (n = 4) using Atlas 5 software and a Gemini 300 SEM (Zeiss) operating in high vacuum, at 4.5 kV with tandem decel operating at 3 kV. Serial images were registered using TrakEM2 (ref. 84 ) in Fiji (NIH), aligned with the confocal images in Photoshop (Adobe) using nuclei as unbiased fiducials and 2/3D reconstructed using Amira (Thermo Fisher Scientific).

scRNA-seq library preparation, expression and NicheNet analysis
For scRNA-seq, we pooled tissue from four mice for each genotype, to be able to sort rare cell populations. Indeed, from one such preparation, we could isolate 5,000 microglia but only limited PVMs (200) and PVFs (350) on average per genotype. For each cell type and each genotype, we prepared a single scRNA-seq library, and we validated the findings using orthogonal methods such as smFISH and flow cytometry.
Single-cell libraries were prepared using 10X Genomics Chromium Next GEM Single Cell 3′ kit v3.1, according to the manufacturer's instructions. Cells were loaded on the chromium chip according to Supplementary Table 2; because sorted PVMs and fibroblasts were in low abundance, we loaded the whole sample, while for microglia cells we aimed for 5,000 cells per sample. Libraries were sequenced using an SP flowcell on an Illumina NovaSeq S6000 instrument, with sequencing parameters recommended by 10X Genomics, aiming for 40,000 reads per cell. Raw FASTQ files were preprocessed using 10X Genomics CellRanger (version 6.0) and the mm10 mouse reference genome. Filtered gene expression matrices obtained from CellRanger were imported in R (version 4.1.2) and analyzed using the Seurat (version 4.0.6) and NicheNet (version 1.0.0) packages 28 . Cells with more than 10% of reads aligning to mitochondrial transcripts were deemed to be low quality or damaged and removed from the dataset. Cells remaining after each filtering step are reported in Supplementary Table 2. Identification of high variable genes, PCA and clustering were performed using Seurat functions. Cluster annotation was performed considering the expression of a set of known cell type markers (Extended Data Fig. 6). Differential expression analysis between pairs of conditions was performed with Seurat's FindMarkers function using the Wilcoxon Rank Sum test. Comparisons between selected cell populations were performed using NicheNet functions and standard workflow.

Ribosome immunoprecipitation (IP)
Brains were extracted from mice and homogenized in ice-cold homogenization buffer (50 mM Tris, pH 7.4, 100 mM KCl, 12 mM MgCl 2 , 1% NP-40, 1 mM DTT, 1:100 protease inhibitor (Sigma-Aldrich), 200 units per ml RNasin (Promega) and 0.1 mg ml −1 cycloheximide (Sigma-Aldrich) in RNase-free DDW) 10% wt/vol with a Dounce homogenizer (Sigma-Aldrich) until the suspension was homogeneous. Cell debris was removed by transferring 1 ml of homogenate and centrifuging at 10,000g and 4 °C for 10 min. Next, 10 µl of supernatant was removed for 'input' analysis and 10 µg of anti-HA.7 antibody (H9658, Sigma-Aldrich) or 10 µg of mouse monoclonal IgG1 antibody (Merck, PP100) was added to the supernatant, followed by 4 h of incubation with slow rotation in a cold room at 4 °C. Afterward, Dynabeads Protein G (Thermo Fisher Scientific) were equilibrated to homogenization buffer by washing three times. Samples were then incubated for 4 h with antibody and afterward beads were added for overnight incubation at 4 °C. Afterward, samples were washed three times with high-salt buffer (50 mM Tris, pH 7.4, 300 mM KCl, 12 mM MgCl 2 , 1% NP-40, 1 mM DTT, 1:200 protease inhibitor, 100 units per ml RNasin and 0.1 mg ml −1 cycloheximide in RNase-free DDW) for 5 min. At the end of the washes, beads were magnetized and excess buffer was removed, 100 µl Lysis Buffer was added to the beads and RNA was extracted with Dynabeads mRNA Direct purification kit (Thermo Fisher Scientific). RNA was eluted in 6 µl H 2 O and taken for RNA sequencing.
Bulk RNA sequencing for ribosome IP samples mRNA was captured with Dynabeads oligo(dT) (Life Technologies) according to the manufacturer's guidelines. Qubit fluorometer Article https://doi.org/10.1038/s41593-023-01257-z (Life Technologies) technology was used to calculate library concentration and mean molecule size was determined with a 2200 TapeStation instrument. Next, libraries were sequenced using Illumina NextSeq-500. Raw reads were mapped to the genome (NCBI37/ mm10) using hisat (version 0.1.6). Only reads with unique mapping were considered for further analysis. Gene expression levels of genes of interest (Lyve1, Cx3cr1, Sall1 and Spp1) were calculated using the User-friendly Transcriptome Analysis Pipeline 85 . Next, DESeq2 R-package was used for normalization and differential expression analysis and pValue was adjusted by multiple gene testing <0.05.

Statistics
All statistical analysis was performed in Prism (GraphPad Software, Version 9.3.1). Two groups were compared using two-tailed unpaired Student's t-test. For oAβ injections experiments, two-tailed paired Student's t-test was used to normalize variances of injections on different days or in different oAβ batches. To compare more than two groups (WT, Spp1 KO/KO , App NL-F/NL-F , App NL-F ·Spp1 KO/KO mice or PVMs, microglia and PVFs), one-way ANOVA with Kruskal-Wallis post hoc test was used. To compare multiple variables (WT and Spp1 KO/KO injected with S26C oAβ), two-way ANOVA with Tukey's post hoc test was used. Data in graphs are presented as mean ± s.e.m. Data distribution was assumed to be normal but this was not formally tested. Statistical methods were not used to predetermine study sizes but were based on similar experiments previously published. Experiments were blinded to the genotype of the animal as well as the treatment of the animal. Experiments involving human sections were blinded to the demographics of the patients. Analysis of flow cytometry data of Spp1 TdT mice was performed in an unblinded fashion, for compensation purposes. No data were excluded for analysis.

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

Data availability
The scRNA-seq data generated in this study are deposited at Array-Express (BioStudies, Annotare 2.0) under accession number E-MTAB-11918. The RiboTag-based RNA-seq data generated in this study are deposited at ArrayExpress (BioStudies, Annotare 2.0) under accession number E-MTAB-12515. Source data are provided within this paper. Raw data used in this study and requests for resources and reagents are available from the corresponding author upon request. Source data are provided with this paper. Corresponding author(s): Dr Soyon Hong Last updated by author(s): Dec 8, 2022 Reporting Summary Nature Portfolio wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Portfolio policies, see our Editorial Policies and the Editorial Policy Checklist.

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

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