Novel App knock-in mouse model shows key features of amyloid pathology and reveals profound metabolic dysregulation of microglia

Genetic mutations underlying familial Alzheimer’s disease (AD) were identified decades ago, but the field is still in search of transformative therapies for patients. While mouse models based on overexpression of mutated transgenes have yielded key insights in mechanisms of disease, those models are subject to artifacts, including random genetic integration of the transgene, ectopic expression and non-physiological protein levels. The genetic engineering of novel mouse models using knock-in approaches addresses some of those limitations. With mounting evidence of the role played by microglia in AD, high-dimensional approaches to phenotype microglia in those models are critical to refine our understanding of the immune response in the brain. We engineered a novel App knock-in mouse model (AppSAA) using homologous recombination to introduce three disease-causing coding mutations (Swedish, Arctic and Austrian) to the mouse App gene. Amyloid-β pathology, neurodegeneration, glial responses, brain metabolism and behavioral phenotypes were characterized in heterozygous and homozygous AppSAA mice at different ages in brain and/ or biofluids. Wild type littermate mice were used as experimental controls. We used in situ imaging technologies to define the whole-brain distribution of amyloid plaques and compare it to other AD mouse models and human brain pathology. To further explore the microglial response to AD relevant pathology, we isolated microglia with fibrillar Aβ content from the brain and performed transcriptomics and metabolomics analyses and in vivo brain imaging to measure energy metabolism and microglial response. Finally, we also characterized the mice in various behavioral assays. Leveraging multi-omics approaches, we discovered profound alteration of diverse lipids and metabolites as well as an exacerbated disease-associated transcriptomic response in microglia with high intracellular Aβ content. The AppSAA knock-in mouse model recapitulates key pathological features of AD such as a progressive accumulation of parenchymal amyloid plaques and vascular amyloid deposits, altered astroglial and microglial responses and elevation of CSF markers of neurodegeneration. Those observations were associated with increased TSPO and FDG-PET brain signals and a hyperactivity phenotype as the animals aged. Our findings demonstrate that fibrillar Aβ in microglia is associated with lipid dyshomeostasis consistent with lysosomal dysfunction and foam cell phenotypes as well as profound immuno-metabolic perturbations, opening new avenues to further investigate metabolic pathways at play in microglia responding to AD-relevant pathogenesis. The in-depth characterization of pathological hallmarks of AD in this novel and open-access mouse model should serve as a resource for the scientific community to investigate disease-relevant biology.


Background
Alzheimer's disease (AD) is a devastating neurological condition. Research using preclinical AD models has been an essential step towards the development of therapies and discovery of disease-related biomarkers. In particular, mouse models recapitulating some aspects of this complex disease are instrumental in gaining insight into the neurobiology of AD [1]. Transgenic mouse models overexpressing human amyloid precursor protein (hAPP) with familial AD (fAD) mutations, sometimes in combination with PSEN1 fAD mutations, have furthered our understanding of disease mechanisms [2][3][4], but also have the potential to introduce artifacts associated with App overexpression. Recently, a new generation of mouse models using knock-in (KI) approaches have been developed to avoid some of the limitations of transgenesis, including random genetic integration, ectopic expression and non-physiological protein levels [5]. In those models, the App gene has been humanized for the Aβ sequence and harbored fAD mutations resulting in progressive accumulation of amyloid deposits. Other pathologies associated with amyloid deposition have been observed, including immune responses and neurodegenerative processes [6,7]. The identification of a plaque-induced gene signature (PIG) partially overlapping with gene expression changes observed in AD brain samples using an App KI model [8] further supports the relevance of these new models and provides incentive to continue their utilization to better define disease mechanisms and test candidate therapeutics.
Over the last decade, convergent findings from human genetics, gene expression studies and preclinical research revealed that microglia, the resident immune cells of the central nervous system (CNS), are likely contributing to AD pathophysiology [9][10][11]. Microglia respond to pathogenic drivers of AD, including amyloid-β (Aβ) and tau, but how microglia may become dysfunctional and contribute to disease remain unclear. Microglia expression profiles from AD and preclinical mouse models have revealed that microglia undergo profound changes in their gene expression profiles [8,9,[12][13][14][15][16][17][18][19]. However, the functional implications of this molecular remodeling are poorly defined. It has been hypothesized that this gene expression program may enable microglia to sequester and/or resorb amyloid deposits and associated neurodegenerative processes. Under these circumstances, the phagocytic activity of microglia is critical and the metabolic demand and burden on microglial degradation pathways are high, potentially leading to immunometabolic alteration [20] and increasing cellular dysfunction over time [21][22][23]. Indeed, alterations of lipid metabolism have been observed in AD brain, and several genes associated with late onset AD (LOAD) risk that control lipid metabolism (e.g., TREM2, APOE) are highly expressed in microglia [10,24].
Here, we describe the engineering and characterization of a novel humanized Aβ App KI mouse model carrying the Swedish, Arctic and Austrian fAD mutations (App SAA ). This model demonstrated elevated Aβ 42/40 ratio in relevant tissues, progressive deposition of amyloid plaques, increased biomarkers of neuroinflammation and neurodegeneration, as well as mild behavioral phenotypes. Leveraging this new model, we discovered extensive lipid and metabolite alterations in microglia, associated with their level of phagocytic activity and proximity to amyloid plaques. Specifically, metabolic profiles of phagocytic microglia were consistent with lysosomal dysfunction (e.g., accumulation of ganglioside GM3) and foam cell phenotypes (e.g., accumulation of neutral lipids like triglycerides). Those findings were corroborated by gene expression changes in phagocytic microglia and highlight the profound reprogramming of microglia as they respond to amyloid pathology. Our highdimensional profiling of microglia opens new avenues to explore the contribution of metabolic pathways in their response to AD pathology, including specific lipids and metabolites of interest.

Generation of App SAA knock-in mouse model
The App SAA knock-in mouse model was engineered by insertion of 6 mutations into the genomic App locus via homologous recombination. Three amino acids APP G676R, APP F681Y and APP R684H were substituted to humanize the mouse Aβ 1-42 region and the following three FAD-linked APP mutations were inserted: KM670/671NL (Swedish), E693G (Arctic) and T714I (Austrian). This mouse model was created on a C57BL/6J background by Ozgene (Australia) using goGermline technology [25].
To generate the targeting vector for the App SAA KI mouse model, six individual fragments (A-F) that cover the targeting area were engineered and introduced into six cloning vectors respectively. Fragment A encoding a 2828bp right homology arm fragment within intron 17 (mouse genomic DNA chr16:84,959,292-84,962,119 GRCm38/mm10) was amplified by PCR from BAC genomic DNA (clones RP23-126H12 and RP23-99P18) with primers P1915_41 and P1915_51. Fragment B encoding a portion of intron 17, exon 17 with the T714I and E693G mutations, a portion of intron 16 (corresponding to mouse genomic DNA region chr16:84,962,120-84,963,051 GRCm38/mm10) and a portion of the hygromycin cassette was synthesized as a gBlock by Integrated DNA Technologies. Fragment C encoding the remainder of the hygromycin cassette was amplified by PCR from an Ozgene in-house cloning vector using primers P1915_74 and P1915_53. Fragment D encoding a 2032bp portion of intron 16 (mouse genomic DNA region chr16:84,963,052-84,965,083 GRCm38/mm10) was synthesized as a gBlock by Integrated DNA Technologies. Fragment E encoding a portion of intron 16, exon 16 with the R684H, F681Y, G676R, and KM670/671NL mutations, and a portion of intron 15 (corresponding to mouse genomic DNA chr16:84,965,084-84,965,993 GRCm38/mm10) was synthesized by GENEWIZ and was housed within a vector that contained the neo cassette on the vector backbone. Fragment F encoding a 5870bp left homology arm within intron 15 (mouse genomic DNA chr16:84,965,994-84,971,863 GRCm38/mm10) was amplified by PCR from BAC genomic DNA (clones RP23-126H12 and RP23-99P18) with primers P1915_46 and P1915_56. The final targeting vector was then generated by the sequential assembly of fragments A-F. Fragment A digested with AatII was ligated into vector B digested with the same enzyme to generate vector AB. Fragment AB digested with AgeI was ligated into fragment C digested with the same enzyme to generate vector ABC. Fragment ABC digested with AscI was ligated into vector D digested with MluI to generate vector ABCD. Fragment ABCD digested with AscI was ligated into vector E digested with the same enzymes to generate vector ABCDE. Fragment ABCDE digested with enzyme MluI was ligated into vector F digested with the same enzyme to generate vector ABCDEF. The targeting vector containing Fragment ABCDEF was linearized with PmeI and electroporated into Bruce ES cells which were derived from C57BL/6-Thy1.1 mice. ES cells were maintained in the medium supplemented with G418 drug at 200ug/ml, and surviving clones were picked after 8 days of drug selection. ES cells were then screened for correct targeting events by qPCR and then confirmed by Southern analysis. For southern analysis, genomic DNA was purified from two ES clones (II_1C11 and II_1G11), digested with SpeI and then detected with 5'probe, 3' probe, Hygro probe, enP probe, and neoP probe. Both clones were tested positive for proper homologous recombination in left and right arms. The ES cell clones (clone II_1C11 and clone II_1G11) that were confirmed to carry the correct homologous recombination events were injected into goGermline blastocysts, and the resulting chimeric mice were crossed to Flp transgenic mice (OzFlp) to excise the neo and hygro selection cassettes. The resulting heterozygous KI mice were backcrossed to C57BL/6J for more than two generations and then intercrossed to obtain homozygous KI mice for further characterization. The App SAA model is available from the Jackson Laboratory as B6(Cg)-Apptm1.1Dnli/J (https:// www. jax. org/ strain/ 034711). Primer sequences and assays for mouse model generation are listed in the Supplementary Table 1.

Animals
The App SAA mice were maintained on the C57BL/6J genetic background. The App SAA heterozygous KI mice were intercrossed to obtain three genotypes of interest. Other App transgenic lines were used for comparative analysis: 5xFAD (B6SJL-Tg(APPSwFlLon,PSEN1*M146L *L286V)6799Vas/Mmjax, MMRRC Strain #034840-JAX [26]), APP/PS1 (B6.Cg-Tg[APPswe, PSEN1dE9]85Dbo/ Mmjax, MMRRC Strain #034832-JAX [27]), Tg2576 (B6;SJL-Tg[APPSWE]2576Kh a [28]). The App SAA mice were bred either at the Jackson Laboratory (USA, Bar Harbor or Sacramento) or Ozgene (Australia, Perth). The 5xFAD mice were bred at the Jackson Laboratory (USA, Sacramento). The APP/PS1 and Tg2576 mice were studied at Allen Institute and all experimental procedures related to the use of these two lines of mice were approved by the Institutional Animal Care and Use Committee of the Allen Institute for Brain Science, in accordance with NIH guidelines. Before experiments, the App SAA mice were shipped to Denali Therapeutics for an acclimation period of at least 2 weeks. Housing conditions were similar among labs and included standard pellet food and water ad libitum, a 12-h light-dark cycle at 22°C with a maximum of 5 mice per cage. App SAA KI mice used for experiments are listed in the Supplementary Table 2. All mouse husbandry and experimental procedures were reviewed and approved by Denali Institutional Animal Care and Use Committee and were conducted in full compliance with regulatory statutes, Institutional Animal Care and Use Committee policies, and National Institutes of Health guidelines.

Frailty index composite and voluntary running assays
The behavioral assays were conducted at the Neurobehavioral Phenotyping core of the Jackson Laboratory (Bar Harbor, USA). Mice were brought in the testing room at least 60 min before the assessment for habituation. For the Frailty Index composite assay, mice were weighed and evaluated one at a time in a battery of assessment. Except for a few quantitative measures (body weight and body temperature; Braintree Scientific #TH5 Thermalert), all other variables were scored in a scale of 0 to 1 by a trained technician. Characteristics observed include physical, physiological and innate reflex conditions (modified from Whitehead et al. [29]). To measure voluntary running, mice were individually placed in a new cage equipped with a running wheel (Med Associates) and the activity of the animals was recorded during a minimum of 2 days and 3 nights. Data were evaluated for onset of activity (typically correlated with lights off ), distance traveled (rpm) and time spent running (min). To test the activity of the mice in a novel environment, mice were placed inside an open-field area (Omnitech Electronics, Columbus, OH, USA) for 60 min. The enclosure was surrounded by two levels of infrared photo beam sensors enabling recording of the horizontal and vertical activity of the mice.

Repeated open field test
Exploratory locomotor activity was measured in an open field with a video-tracking system (EthoVision; Noldus) at the Gladstone Institute. 16-17-month-old App SAA mice (KI/KI) (n=21, 11 males and 10 females) and wildtype littermate controls (+/+) (n=26, 14 males and 12 females) were used for the analysis. Mice were placed in one of four identical clear plastic circular chambers (12" diameter) for 5 min. The chamber walls were covered with plain gray backing to visually isolate the mice from their surroundings. Distance travelled in the center (0-4.5" diameter) and periphery (4.5-6" diameter) were measured in cm. Mice were tested twice a day (morning and afternoon sessions; 3-4 hours apart) on two consecutive days (Days 1-2; Trials 1-4) and retested two weeks later (Days 15-16; Trials 5-8) in the same chamber. The chambers were cleaned with 70% ethanol between trials. Only male or female mice were tested simultaneously in the four chambers, and to ensure balanced groups throughout the day, male and female mice were tested alternately. Statistical analyses were performed with SPSS (v.27).

Tissue collection
Prior to tissue harvest, animals were deeply anesthetized via intraperitoneal (i.p.) injection of 2.5% Avertin. All collected tissues were snap-frozen on dry ice and stored at -80°C. Blood samples were collected by cardiac puncture into an EDTA tube (Sarstedt Microvette 500 K3E, Ref# 201341102), then slowly inverted 10 times prior to centrifugation at 12,700 rpm for 7 minutes at 4°C to collect plasma. CSF samples were collected by pre-pulled glass capillary tube from the cisterna magna and then transferred to 0.5 mL Protein LoBind Eppendorf tubes (Eppendorf Cat #022431064) for centrifugation at 12,700 rpm for 7 minutes at 4°C, supernatant was then collected. The mice were then perfused intracardially with cold PBS, and brain tissues were sub-dissected to separate the cortical and hippocampal regions.

Aβ extraction and measurement by MSD
Cortical and hippocampal tissues were weighed and homogenized in TBS (140 mM NaCl, 3 mM KCl, 25 mM Tris-HCl, pH 7.4, 5 mM EDTA, 2 mM 1,10-phenanthroline) containing protease inhibitor (Roche,#4693159001). The homogenates were centrifuged at 100,000 g, 4°C for 1 hour, and then the supernatant was collected (TBS-soluble fraction). Pellets were resuspended in 5 M guanidine, 50 m M Tris, pH 8.0, further homogenized, and incubated at room temperature for 3 hours. Samples were then centrifuged at 20,800 g, 4°C, for 20 minutes, and the supernatant was collected (TBS-insoluble GuHCl fraction). To measure Aβ in brain fractions and biofluids, we tested two commercial kits: MSD V-PLEX Aβ Peptide Panel 1 (4G8) Kit (cat#K15199E) and MSD V-PLEX Aβ Peptide Panel 1 (6E10) Kit (cat#K15200E). We found that the 4G8 kit has a good cross-reactivity to mouse and human Aβ. Since the 4G8 kit showed a greater sensitivity than the 6E10 kit to detect mutated human Aβ from 4-month App SAA KI/ KI brains, we used 4G8 kit for Aβ analysis. Samples were diluted by Diluent 35 provided in the kit (GuHCl fractions 1:20; TBS fractions, no dilution; plasma 1:4; CSF 1:26). Plates were read using the SECTOR Imager 2400A.

Tau measurement by MSD
Total mouse tau in CSF was measured using an ECLimmunoassay MSD 96-well mouse total tau assay (Meso Scale Discovery: MSD) according to the manufacturer's instructions. Briefly, CSF samples (5 μL) were diluted 1 to 5 in 10% blocking solution (10% BSA in tris wash buffer) and incubated at room temperature for 1 hour on a shaker plate. Standard samples were also diluted in 10% blocking solution. Plates were washed 4 times with 1XTris washing buffer and then incubated at room temperature with sulfo-tag total tau detection antibody for 1 hour. After a few additional washes, plates were read on the MSD SECTOR 600 Reader. Samples were fit against a 9-point standard curve.

TREM2 measurement by MSD
TREM2 levels in brain homogenates prepared by Cell Signaling lysis buffer (#9803) and plasma were quantified by using MSD technology. The sandwich layout of this TREM2 MSD assay from the bottom to the top is (1) capture-antibody (1 μg/mL; BAF1729, R&D Systems), (2) samples, (3) primary-antibody (10 μg/mL; anti-TREM2 antibody -clone 4D9) [30], and (4) detection-antibody (0.5 μg/mL; SULFO-TAG-labeled anti-human antibody, Meso Scale Discovery). MSD 96-well plates coated with streptavidin were first incubated with 150 μL MSD-Blocker A. For building each layer for the assay, 25 μL of the reagent was added to each well, and the plates were incubated at room temperature on a shaker at 800 rpm for 1hr. Between incubation step, each well was washed with TBST buffer (0.05% Tween 20 in TBS) using ELx406 plate washer (BioTek). Diluted samples (Brain lysates = 1:5; plasma = 1:20) and TREM2 protein standard in assay buffer (25% MSD-Blocker A and 75% TBST) were prepared and aliquoted for running duplicates in the assay. A four-fold serial-dilution standard curve of recombinant murine TREM2 extracellular domain was prepared to include the linear detection range from 62.5 ng/ml to 15.25 pg/mL. For acquiring the MSD units on the MSD Sector Imager S600 reader (MSD), 150μL of 2X MSD read buffer was added to each well after the final TBST wash. TREM2 levels were calculated using the MSD Discovery Workbench software and normalized by each lysate's concentration (brain samples only).

Cytokines
Brain homogenates prepared by Cell Signaling lysis buffer (#9803) were diluted to 5 μg/μl with PBS and plasma samples were prepared at a 2-fold dilution with PBS. Diluted samples were sent to Eve Technologies (Canada) for cytokines and chemokines measurement by using the Mouse Cytokine Array / Chemokine Array 44-Plex (MD44).

Neurofilament Light detection
CSF neurofilament light (Nf-L) concentrations were measured using Simoa NF-Light ® (SR-X version, Quanterix 103400) bead-based digital ELISA kits and read on the Quanterix SR-X instrument. CSF samples were diluted 100x with Sample Diluent (Quanterix 102252) before being added to Simoa 96-well microplates (Quanterix 101457). Following kit instructions, Simoa Detector Reagent and Bead Reagent (Quanterix 103159, 102246) were added to the samples before incubating and shaking for 30 mins, 30°C at 800 rpm. After incubation, the sample plate was washed with Simoa Wash Buffer A (Quanterix 103078) on a Simoa Microplate Washer according to Quanterix's two step protocol. After initial washes, SBG Reagent (Quanterix 102250) was added and samples were again incubated at 30°C, 800 rpm for 10 min. The two-step washer protocol was continued, with the sample beads being twice resuspended in Simoa Wash Buffer B (Quanterix 103079) before final aspiration of buffer. Sample Nf-L levels were measured using the NF Light analysis protocol on the Quanterix SR-X instrument and interpolated against a calibration curve provided with the Quanterix assay kit.

Whole brain imaging and quantification of plaque distribution
To label amyloid plaques, mice were administered 3.3 mg/kg of methoxy-X04 (R&D Systems) by intraperitoneal (i.p.) injection. Twenty-four hours later, mice were perfused with PBS and 4% paraformaldehyde (PFA) sequentially and then intact brains were dissected and postfixed in 4% PFA at room temperature for 6 hours, followed by overnight at 4°C. Whole brain fluorescence imaging was performed as described in Whitesell et a l [31] with serial two-photon (STP) tomography (TissueCyte 1,000, Tis-sueVision Inc., Somerville, MA), using 925 nm excitation, a 500 nm dichroic mirror, and a 447/60 bandpass emission filter on the blue channel. One hundred and forty serial block-face images were acquired from each brain at 0.35 μm/pixel lateral resolution with a 100 μm sectioning interval. Automated segmentation of the fluorescent signal from methoxy-X04 labeled plaques and registration to the 3D Allen Mouse Brain Common Coordinate Framework, v3 (CCFv3) [32] were performed as previously described [31]. Briefly, segmented fluorescence output is a full resolution mask that classifies each 0.35 μm × 0.35 μm pixel as either signal or background. An isotropic 3D summary of each brain is constructed by dividing each image series into 10 μm × 10 μm × 10 μm grid voxels. Total signal is computed for each voxel by summing the number of signal positive pixels in that voxel. Each image stack is registered to the CCFv3 in a multistep process using both global affine and local deformable registration. Plaque density for each structure in the reference atlas ontology was calculated by summing voxels from the same structure. To obtain plaque counts within each structure, we used a standard feature labeling algorithm. Adjacent and orthogonally adjacent voxels in the segmentation signal were grouped together as one plaque object. Due to the 100 μm z-sampling interval, our resolution limit for detecting separate plaques in the z-axis was 100 μm. Plaque quantification is reported for one hemisphere per brain, chosen based on image and tissue quality (Supplementary Table 3). All image series were subjected to manual QC checks for completeness and uniformity of raw fluorescence images, minimum fluorescence intensity, and artifacts. Automatic segmentation results were checked for overall quality and false positive signals by overlaying segmentation results for 3-5 single coronal sections with raw fluorescent images from STP imaging.

Multiplexed Fluorescence in situ hybridization and imaging
Fresh-frozen mouse brains embedded in OCT were coronally sectioned at 15 μm onto Superfrost Plus glass slides (Fisher Scientific). Sections were stored at -80°C until RNAscope treatment. The RNAscope multiplex fluorescent v2 kit was used following the manufacturer's protocol for fresh-frozen tissue sections (ACD 323100) except that protease treatment was performed for 15 minutes. Probe sets for Trem2 (ACD 404111-C2) and Tmem119 (ACD 472901-C3) were used with Opal 520 and Opal 570 dyes (Akoya Biosciences FP1487001KT), respectively. Methoxy-X04 (Tocris 4920) was applied to the tissue sections at 100 μM in 1xPBS for 30 min at room temperature immediately following in situ hybridization. Sections were then washed three times for 10min in 1xPBS before mounting in ProLong Gold Antifade Mountant (Ther-moFisher P36930) followed by a coverslip. Sections were imaged using a confocal microscope (Leica SP8; Leica Microsystems, Inc.) with a 20x/.75IMM oil objective.

Microscopy and quantification for immunohistochemistry
Fluorescent stained sections were captured using a Zeiss Axioscan.Z1 slide scanner with a 10x / 0.45 NA objective. A custom macro in Zeiss Zen software was used to quantify signal for Aβ, CD68, Iba1 and AT8 following median smoothing, channel extraction, and local background subtraction to create binary masks of fluorescent signal for each channel. Objects smaller than 10 pixels were excluded as staining artifacts / debris. Individual plaques were dilated to create a region surrounding the plaques to detect microglia adjacent to plaques. The cortex and hippocampus regions of interest (ROIs) were manually outlined for each section and analyzed using the macro to quantify ROI area and measure the number and area of all plaques and microglia, including overlapping area between microglia and the dilated region surrounding and including plaques. The percentage amyloid plaque, Iba1, CD68, or AT8 positivity in the areas of interest were calculated by normalizing the positive pixel area to the quantified tissue area and averaged from 7-12 sections per mouse.
For 3D assessment of plaque architecture and the associated cellular changes, super resolution confocal microscopy of fluorescently stained sections was performed using a scanning confocal microscope (Leica SP8, Leica Microsystems, Inc.) operated in super resolution lightning mode. Images were acquired using a 63x / 1.4 NA oil immersion objective at a pixel size of 50 nm and processed using the Adaptive processing algorithm. Confocal z-stacks of 18-25 μm were acquired for each channel using sequential scan settings. A minimum of four representative fields were captured from within the cortical brain region of App SAA KI/KI (n=3-4) and App SAA +/+ control (n=2) mice.
Microscopic evaluation of cerebral amyloid angiopathy (CAA) pathology was performed on larger fields using a 10x / 0.45 NA objective at a pixel size of 150 nm and z-stacks of 30-40 μm, and at higher resolution using a 20x / 0.75 NA oil immersion objective at a pixel size of 67 μm and z-stacks of 25-35 μm. In both cases, images were processed using the lightning Adaptive processing algorithm.
To assess the 3D morphology of microglia and subcellular localization of methoxy-X04 and Aβ, a minimum of 3 cortical regions per animal were imaged in App SAA KI/KI mice (n=6) using a 40x/1.3 NA oil immersion objective at a pixel size of 50 nm in Lightning mode and processed using the Adaptive processing algorithm. Confocal z-stacks of 15-30 μm with a z-step size of 500 nm were acquired for each channel using sequential scan settings. To reduce noise, images were first resampled to 500 nm isotropic volumes using local mean resampling. The background intensity of each image was estimated using gaussian smoothing with a kernel size of 10 μm isotropic and subtracted from the original image. The methoxy-X04 and Aβ surfaces were estimated by manual thresholding of each image. The surface of Iba1+ microglia was estimated using a bipartite manual threshold, where values below the lowest value were assigned background, values above the high threshold were assigned to the microglial volume, and intermediate values were assigned using random walker diffusive segmentation algorithm with a diffusion constant of 10 [33]. Microglia surfaces were found to often contain several microglia, so the positions of individual microglial soma were then estimated by finding regions where the surface was at least 7 μm in diameter, spaced at least 10 μm from other regions of high surface thickness, and then finding local peak cell thickness using non-maximum suppression. These regions of peak thickness were then used as seeds for a watershed segmentation of the overall microglial surface, where each voxel in the surface was assigned to the closest peak. Both segmentation stages were performed using the algorithms implemented in the Python package scikit-image v0.18.1 [34].
To assess astrogliosis, image acquisition of entire hemispheres, hippocampal subregions, and posterior parietal cortex was performed using the BX-X710 microscope (Keyence) with 4X, 20X or 40X objectives (Nikon). Image analysis was performed with FIJI [35]. Briefly, for GFAP immunoreactivity, images were background-subtracted and thresholded to clearly outline astrocytic cells and the total GFAP-positive area was measured. For C3 immunoreactivity, a GFAP-positive mask was generated and C3 fluorescence intensity was measured in GFAP-positive astrocytes. For GLT-1 immunoreactivity, total fluorescence intensity was measured. Statistical analysis was performed using GraphPad Prism (v9.3.1).

RNA isolation, RT-qPCR, and SMARTSeq library preparation
RNA from bulk mouse brain tissue was extracted using the RNeasy Plus Mini Kit (Qiagen) and resuspended in nuclease-free water. For RT-qPCR, 3 μl of RNA was transcribed into cDNA using SuperScript IV (Invitrogen). App gene expression level was assessed using Taqman probe (FAM-Mm01344172_m1) on a QuantStudio 6 Flex (Applied Biosystems) and normalized to Gapdh.
For RNA-seq analysis of FACS sorted cells, 200 live cells were sorted directly into 11.5 microliters of Clontech SMARTseq 10X reaction buffer for reverse transcription. cDNA synthesis was performed with the Clontech SMARTSeq v4 3′ DE kit (Takara Bio USA, Inc. 635040). Reverse transcription was followed by 14 cycles of cDNA amplification. cDNA was purified with 0.8X volume of SPRIselect beads (Beckman Coulter B23318), quantified on a Bioanalyzer 2100 System with a High Sensitivity DNA chip (Agilent 5067-4626). 300 pg of cDNA from each sample was used as input for library preparation with the Nextera XT DNA Library Prep Kit (Illumina FC-131-1096). Fragmentation and adaptors insertion were performed by tagmentation, followed by 12 cycles of PCR amplification. The final libraries were purified using 0.8X SPRIselect beads. Library quantity and quality were assessed with Qubit 4 Fluorometer with the 1X dsDNA HS Assay Kit (Invitrogen Q33231) and on a Bioanalyzer 2100 System with a High Sensitivity DNA chip. Sequencing reads were generated on an Illumina NovaSeq 6000 instrument (100 bp single end) by Seq-Matic (Fremont, CA, USA).

RNA-seq data analysis
Following alignment and expression quantitation, lowly expressed genes were removed, and differential expression analysis was performed using the limma/ voom with sample weighting framework [39,40]. Lowly expressed genes are those that did not have more than ten reads assigned to them in at least as many samples as the minimum replicate size per experiment (six samples in GSE158152 reporting on total microglia analysis and ten in GSE158153 reporting on methoxy-X04 (+) vs methoxy-X04 (-) microglia). Technical replicate libraries from GSE158152 (identified by the animal_id column) were summed together prior to analysis. Linear models were fit against the covariate(s) of interest (genotype and/ or methoxy-X04-status) with "take-down day" and "sex" encoded as batch covariates. Because the direct comparison of the methoxy-X04 (+) vs methoxy-X04 (-) profiles in App SAA KI/KI mice utilized repeated measures from the same animal, "subject_id" was fit as a random effect using the duplicateCorrelation function from limma [41]. Empirical Bayes moderated t-statistics and p-values were computed relative to a 1.2 fold-change cutoff using treat [42]. Significantly differentially expressed genes were then defined as those having an FDR <= 10%.
Single sample gene set activity scores were calculated in Fig. 3b by taking a weighted average of the expression of the genes in each gene set using the eigenWeightedMean function from the sparrow Bioconductor package (v1.0.1; DOI: 10.18129/B9.bioc. sparrow). Weights correspond to the loadings of each gene on the first principal component of the meancentered gene-by-sample matrix for each gene set. Activity scores for 5xFAD mouse profiles were taken from the microarray data reported in Ulland et al. [43] (GSE65067).
Gene set enrichment analyses were performed via the fgseaMultiLevel function in the fgsea R package using the moderated t-statistic as the gene ranking statistic [44]. Gene sets used for testing were taken from the Biological Process collection of the Gene Ontology database, the KEGG database, as well as a custom set of genes compiled from the literature that are enumerated in Supplemental Table 4 [45][46][47]. Gene set enrichment scores shown Fig. 4i were calculated by averaging the moderated t-statistics of the genes in the leading edge of the gene set. Because the leading edge can consist of different genes for each comparison, we defined the genes used in this calculation as the universe of the genes that appeared in the leading edge of the gene set across all comparisons shown in these figures. The gene sets presented in Fig. 4i along with the genes that make up their leading edge are listed in Supplementary  Table 4. The methoxy-X04 (+) specific gene signature was extracted from Supplementary Table 4 (Specific X04+ DEGs) from Grubman et al. [13] by selecting only those genes with an estimated log2FC >= 1.
All software versions for the RNA-seq analysis correspond to Bioconductor release version 3.14 [48].

FACS lipid extraction
Lipid extraction was performed using Matyash liquidliquid extraction protocol with the following modifications. Briefly, 50,000 cells were sorted directly into 400 μl of methanol containing surrogate internal standards and kept on ice. To each tube, 200-400 μl of water was added to achieve final volume of 800 μl. These samples were vortexed for 5 min at room temperature. To these samples, 800 μl of tert-Butyl methyl ether (MTBE) was added and vortexed for an additional 5 min at room temperature. Samples were then centrifuged at 21,000 x g for 10 min at 4°C. Following centrifugation, 700 μl of upper organic layer was collected and dried under constant stream of N 2 gas. Dried samples were reconstituted in 100 μl of MS-grade methanol for further analysis by LC-MS/MS.
Electrospray ionization was performed using the following settings: curtain gas at 30 psi; collision gas at 8 psi; ion spray voltage at 5500 V (positive mode) or -4500 V (negative mode); temperature at 250°C (positive mode) or 600°C (negative mode); ion source Gas 1 at 55 psi; ion source Gas 2 at 60 psi; entrance potential at 10 V (positive mode) or -10 V (negative mode); and collision cell exit potential at 12.5 V (positive mode) or -15.0 V (negative mode). Data acquisition was performed in multiple reaction monitoring mode (MRM) with the collision energy (CE) values reported in Supplementary Table 5. Lipids were quantified as area normalized to specific non-endogenous internal standards as reported in Supplementary Table 5. Quantification was performed using MultiQuant 3.02 (Sciex). Lipids were normalized to protein amount. Protein concentration was measured using the bicinchoninic acid (BCA) assay (Pierce, Rockford, IL, USA).

LCMS analysis of polar metabolites Positive mode
Metabolites analyses were performed liquid chromatography (UHPLC Nexera X2) coupled to electrospray mass spectrometry (QTRAP 6500+, Sciex). For each analysis, 5 μL of sample was injected on a BEH amide 1. min at 95% B. The following source settings were applied: curtain gas at 30 psi; collision gas was set at at 8 psi; ion spray voltage at 5500 V; temperature at 600°C; ion source Gas 1 at 50 psi; ion source Gas 2 at 60 psi; entrance potential at 10 V; and collision cell exit potential at 12.5 V. Data acquisition was performed in multiple reaction monitoring mode (MRM) with the collision energy (CE) values reported in Supplementary Table 6. Quantification was performed using MultiQuant 3.02 (Sciex).

Negative mode
Metabolites analyses were performed liquid chromatography (ACQUITY I-Class Plus UPLC FL, Waters Corp) coupled to electrospray mass spectrometry (XEVO TQ-S Micro, Waters Corp). For each analysis, 5 μL of sample was injected on an Agilent InfinityLab Poroshell 120 HILIC-Z P 2.7 μm, 2.1×50 mm (Agilent Technologies Inc., Santa Clara, CA USA); using a flow rate of 0.5 mL/ min at 25°C. Mobile phase A consisted of water with 10 mM ammonium acetate + 5 μM medronic acid (Agilent), pH 9. Mobile phase B consisted of acetonitrile:water 9:1 with 10 mM ammonium acetate + 5 μM medronic acid (Agilent), pH 9. The gradient was programmed as follows: 0.0-1.0 min at 95% B; 1.0-6.0 min to 50% B; 6.0-6.5 min to 95% B; and 6.5-10.0 min at 95% B. Electrospray ionization was performed in negative ion mode. The following source settings were applied: capillary voltage at 1.9 kV; source temperature at 150°C; desolvation temperature at 600°C; desolvation gas flow at 1000 L/hr; cone gas flow at 50 L/hr; cone voltage at 20 V; nebulizer gas at 7 bar. Data acquisition was performed in multiple reaction monitoring mode (MRM) with the collision energy (CE) values reported in Supplementary Table 6. Quantification was performed using Skyline (v19.1;University of Washington).

In vivo PET imaging and quantification
All rodent PET procedures followed an established standardized protocol for radiochemistry, acquisition times and post-processing [49], which was transferred to a novel PET/MRI system. In brief, [ 18 F]GE-180 TSPO-PET and [ 18 F]FDG-PET was used to measure cerebral microglial activity and glucose metabolism respectively. For TSPO-PET imaging, App SAA mice (KI/KI) (n=21, 14 males and 7 females) and aged-matched wildtype littermate controls (+/+) (n=30, 6 males and 24 females) were analyzed from 5 to 20 months of age. For FDG-PET imaging, another cohort of App SAA mice (KI/KI) (n=14, 7 males and 7 females) and aged-matched wildtype littermate controls (+/+) (n=27, 6 males and 21 females) were studied from 5 to 20 months of age.
All mice were scanned with a 3T Mediso nanoScan PET/MR scanner (Mediso Ltd, Hungary) with a FDG-PET and TSPO-PET values deriving from a predefined cortical volume of interest [51] were extracted and analyzed as a function of age in direct comparison of App SAA and wildtype mice.

Statistical analysis
Data have either been expressed as means ± SEM or as indicated in graphs. Statistical analysis of data was performed in GraphPad Prism 8 or as indicated otherwise for whole brain plaque analysis, RNAseq and LC-MS data. Analysis was performed using t test or with oneway analysis of variance (ANOVA) with Dunnett multiple comparison as indicated in figure legends. Criterion for differences to be considered significant was P < 0.05.
Comparison of LC/MS results from FACS-isolated microglia from App SAA +/+, App SAA KI/+ and App SAA KI/KI mice: Integrated peak areas were divided by the area of pre-assigned internal limma standards, and the resulting ratios were log2 transformed. Analytes detected in less than 75% of the samples were removed. To identify and account for unwanted sources of variation, two surrogate variables (SV) were identified with the sva R packag e [52]. Genotypes were compared by fitting the following linear model using the limma R packag e [53]: log 2 (abundance) ~ genotype + batch + sex + SV. Sample weights were incorporated vi' limma's arrayWeights function. Analytes with an estimated absolute fold difference of more than 20% at a false discovery rate of less than 10% were deemed significant. To visualize adjusted abundances in the heatmap, the batch, sex and SV covariates were regressed out with limma's removeBatchEffect function. LC/MS results from methoxy-positive and -negative microglia were analyzed in a similar way, but as no significant surrogate variables were identified, median scaling was applied to the log 2 transformed ratios of each sample instead and the following linear model was fit using limma: log 2 (abundance) ~ genotype + batch + sex.

Data availability
RNA-seq datasets have been deposited online in the Gene Expression Omnibus repository as SuperSeries GSE158156, which is composed of the SubSeries GSE158152 and GSE158153

Generation of the App SAA KI mouse model and characterization of APP metabolism
We used a knock-in (KI) strategy to humanize the Aβ sequence of the murine App gene and introduced three fAD mutations -Swedish (KM670/671NL) [54], Arctic (E693G) [55] and Austrian (T714I) [56] -using homologous recombination ( Supplementary Fig. 1a). To determine whether introduction of the fAD and humanization mutations disrupts App expression, we examined App mRNA levels in App SAA KI/KI (homozygous), App SAA KI/+ (heterozygous) and App SAA +/+ (wild type) mice at 2 months of age. RT-qPCR analysis showed normal levels of App mRNA in KI/KI and KI/+ brains (Fig. 1a). Western blotting analysis with an antibody recognizing the N-terminus of mouse and human APP (22C11) revealed similar full-length APP protein levels across the three genotypes (Fig. 1b). Immunoblotting with an antibody recognizing amino acids 1-16 of human Aβ (6E10) detected the humanized Aβ sequence in the full-length APP and C-terminal APP fragments (APP-CTF) in brains of KI/+ animals, and we observed elevated levels in the KI/KI background (Fig. 1b), as expected based on enhanced cleavage due to the Swedish mutation [57,58]. The elevation of APP-CTF was further validated by antibody A8717 which specifically recognizes the C-terminus of APP (Fig. 1b). These results demonstrate that App SAA does not affect full-length APP expression but alters its cleavage products.
We then used an immunoassay with the 4G8 antibody to measure Aβ 40 and Aβ 42 in App SAA mice. The Aβ 42/40 ratio was increased in soluble and insoluble brain fractions, CSF, and plasma from App SAA KI/KI at 2 and 4 months of age (Fig. 1c, d and Supplementary Fig. 1be). At 2 months of age, before amyloid plaques were detected (Fig. 2a), we observed a reduction of Aβ 40 levels in brain fractions, CSF and plasma from App SAA KI/ KI, leading to an increased Aβ 42/40 ratio ( Supplementary  Fig. 1b-e). At 4 months of age, an increase of Aβ 42 levels in insoluble brain fraction further increased the Aβ 42/40 ratio in the App SAA KI/KI (Fig. 1c). The reduction of Aβ 40 levels in brain and biofluids in young App SAA KI/KI mice could be induced by the Austrian fAD mutation, which has been reported to impact APP cleavage by γ-secretase [56]. Alternatively, it is possible that the 4G8 antibody used in this assay has reduced binding to humanized Aβ (in App SAA KI/KI) as compared to murine Aβ (in App SAA +/+), leading to an apparent decrease in Aβ 40 . However, since 4G8 antibody similarly binds Aβ 40 and Aβ 42 , the calculated Aβ 42/40 ratio for both App SAA +/+ controls and App SAA KI/KI mice are independent of the affinity of 4G8 for humanized versus murine Aβ.

Progression of amyloid-plaque pathology and associated neurodegeneration
Methoxy-X04 is a blood-brain permeable fluorescent dye, derived from congo red, that is widely used to stain amyloid plaques containing fibrillar forms of Aβ [59,60]. To analyze the spatial distribution of Aβ deposition at the whole brain level, we injected methoxy-X04 intraperitoneally in KI/KI, KI/+ and +/+ mice from 2 to 8 months of age and euthanized animals 24 hours post injection for two-photon serial imaging [31]. Amyloid deposition was detected in App SAA KI/KI from 4 months of age and the total brain density of Aβ plaques increased in an age-dependent manner at 4 and 8 months of age in this genotype (Fig. 2a). In contrast, no methoxy-X04 positive plaques were observed in App-SAA KI/+ from 4 to 8 months. Aβ deposition in App SAA KI/KI was comparable to that reported for commonly used transgenic models, such as the APP/PS1 and the 5xFAD lines (Fig. 2a). At 8 months of age, Aβ plaques were detected in multiple brain regions, but the highest density of pathology was found in cortical and hippocampal regions (Fig. 2b, c). Comparing the temporal and spatial plaque deposition patterns in App SAA KI/KI with progression of amyloid pathology in human AD [31,61] revealed remarkable similarities in plaque distribution ( Supplementary Fig. 2). Immunostaining with an anti-amyloid antibody confirmed the age-dependent amyloid pathology in the App SAA KI/KI brains (Fig. 2d,  f; Supplementary Fig. 3a, b) and identified plaques in the App SAA KI/+ brains at 16 months of age (Supplementary Fig. 5a). No sex effect was detected in amyloid burden in the cortex (2-way ANOVA: p=0.49, 4 males and 2 females) or the hippocampus (2-way ANOVA: p=0.31, 4 males and 2 females) in 8 month-old App-SAA KI/KI mice (Fig. 2f ), but since sample sizes were relatively small in this analysis, a potential sex effect cannot be firmly excluded. Moreover, anti-Aβ and methoxy-X04 co-stainings revealed a lower percentage of methoxy-X04 positive plaques in the App SAA KI/KI brains compared to that in 5xFAD brains at 6-8 months of age ( Supplementary Fig. 4a, b), indicating that the plaques in App SAA KI/KI brains are less dense than in the 5xFAD brains. We also identified vascular Aβ deposition in App SAA KI/KI at 8 and 16 months of age, particularly in leptomeningeal (pial) vessels at the brain surface, demonstrating the presence of cerebral amyloid angiopathy (CAA) in this model (Fig. 2d, Supplementary Fig. 5b-e).
Swollen, dystrophic neural processes surrounding amyloid plaques are a common pathological feature of AD [62]. To determine if amyloid plaques were associated with this type of neurodegenerative processes in App SAA KI/KI, we analyzed three markers of dystrophic neurites Fig. 1 Alterations of APP cleavage and Aβ levels in App SAA knock-in mice. a RT-qPCR analysis of brain RNA shows normal App mRNA level in the App SAA homozygous mice (KI/KI), heterozygous mice (KI/+) and wild-type mice (+/+) at 2-month-old. b Western blotting analysis of brain lysates at 2-month-old shows normal expression of full-length APP proteins in all three genotypes and KI gene dosage-dependent increase of human full-length APP proteins and APP-CTF in KI/KI and KI/+ mice. β-Actin was used as internal loading control. c Measurement of Aβ40 and Aβ42 in brain insoluble and soluble fractions extracted from animals at 4 months of age. Level of insoluble Aβ42 is increased in KI/KI homogenates and level of insoluble Aβ40 is reduced in KI/KI and KI/+ homogenates relative to wild-type control. The insoluble Aβ42/Aβ40 ratio is significantly enhanced in KI/KI brains. Level of soluble Aβ42 is unchanged in KI/KI homogenates and level of soluble Aβ40 is reduced in KI/KI and KI/+ homogenates relative to wild-type control. The soluble Aβ42/Aβ40 ratio is significantly increased in KI/KI brains. d Measurement of Aβ40 and Aβ42 in CSF and plasma collected from mice at 4 months of age. Level of Aβ42 is unchanged in KI/KI CSF while level of Aβ40 is reduced in KI/KI and KI/+ CSF relative to wild-type control. The CSF Aβ42/Aβ40 ratio is significantly increased in KI/KI CSF. Level of Aβ42 is unchanged in KI/KI plasma while level of Aβ40 is reduced in KI/KI and KI/+ plasma relative to wild-type control. The plasma Aβ42/Aβ40 ratio is significantly increased in KI/KI. Graphs are box and whisker plots and P values: one-way ANOVA with Dunnett's multiple comparison test, each group compared to the App SAA +/+ control group; **P < 0.01, and ***P < 0.001. Sample size: 3-6 per genotype. Xia et al. Molecular Neurodegeneration (2022) 17:41 by immunohistochemistry at 8 months of age: phosphorylated tau (pTau), neurofilament (axonal marker) and LAMP1 (lysosomal marker). We observed numerous AT8-positive pTau puncta (Fig. 2e, g), abnormal shape of neurites (swollen, expanded, dystrophic) and accumulation of LAMP1 (Fig. 2e) around plaques, which are all prominent features of dystrophic neurites.
To further explore if amyloid pathology caused elevation of biomarkers of neurodegeneration in App SAA KI mice, we measured total tau and neurofilament light chain (Nf-L) in CSF. We found that total tau levels were elevated in App SAA KI/KI relative to the App SAA +/+ controls at 8 months of age (Fig. 2h). We also observed age-dependent increase of CSF Nf-L in both App SAA KI/ KI and App SAA +/+ from 2 to 23 months of age (Fig. 2i). Levels of Nf-L were significantly higher in App SAA KI/ KI mice at 18 and 23 months of age when compared to App SAA +/+ (Fig. 2i). No sex difference was detected in Nf-L levels across ages (ANOVA, sex variable, p=0.49). Altogether, our data demonstrate that the App SAA KI mouse model develops progressive deposition of neuritic plaques that is associated with an elevation of neurodegeneration biomarkers.
To investigate microglial responses, we stained brain sections for Iba1 and CD68 proteins in the cortex and the hippocampus from App SAA KI/KI at 8 months of age. We found that microglia density was increased in the vicinity of amyloid plaques with an enrichment in CD68-positive microglia (Fig. 2e, j-l and Supplementary Fig. 6a-d), indicating clustering of responsive microglia around plaques. Additionally, levels of TREM2 and various cytokines were elevated in brain homogenates from App SAA KI/ KI at 8 months of age (Fig. 2m, n), showing profound immune responses.
One of the hallmarks of AD and related animal models is reactive astrogliosis, a process characterized by complex molecular, morphological, and functional changes in astrocytes [63]. To assess astrocytic changes in 18-month-old App SAA KI/KI mice, we first analyzed the expression of glial fibrillary acidic protein (GFAP), a commonly used marker of astrocytic responses in disease. We found robust and widespread increases in GFAP immunoreactivity in App SAA KI/KI mice as compared to age matched App SAA +/+ controls, especially in the neocortex and hippocampus (Supplementary Fig. 7a, d). We also detected strong polarization of GFAP-immunoreactive astrocytes around Aβ plaques in App SAA KI/ KI mice. Next, we assessed the expression of GLT-1 (also known as Slc2a1, or excitatory amino acid transporter 2 (EAAT2)), a major glutamate transporter mainly expressed by astrocytes and essential for synaptic glutamate homeostasis [64,65]. Decreased GLT-1 levels have been previously reported in postmortem AD brain tissue [66][67][68], suggesting that AD pathology involves dysregulation in excitatory transmission. In agreement with previous reports, we found decreased levels of GLT-1 immunoreactivity in the hippocampus of App SAA KI/KI mice ( Supplementary Fig. 7b, e). Lastly, we assessed the expression of C3, a complement factor implicated in AD pathology and upregulated in astrocytes in response to microglial release of proinflammatory factors [69,70]. Similar to GFAP, astrocytic C3 levels were increased in the hippocampus, neocortex, and other brain regions of App SAA KI/KI mice, and these increases seemed highest near Aβ plaques ( Supplementary Fig. 7c, f ). Together, these results suggest that App SAA KI/KI mice have profound alterations in astrocytes that may contribute to disease progression in this model.

Multi-dimensional phenotyping of the microglial response
The presence of microglia clusters around amyloid plaques and the elevation of cytokines suggested that microglia transitioned to a responsive state. To further characterize this response, we isolated CD11b + microglia from App SAA KI/KI, KI/+ and +/+ mice at 8 months of age by FACS (Fig. 3a) and analyzed their mRNA, metabolite, and lipid profiles as previously described in nonamyloid models [22].
We first confirmed the enrichment and purity of isolated microglia using RNA-seq data to compare the (See figure on next page.) Fig. 2 Amyloid plaque pathology, biomarkers of neurodegeneration and neuroinflammation. a Amyloid plaques were measured by segmentation and registration to the CCFv3 atlas post-methoxy-X04 labeling. N = 4-30 mice/ group. b Methoxy-X04 positive plaque density across brain regions in the App SAA KI/KI. Hippocampal F. = formation. N = 4-6 mice/ group. c 3D heatmaps show brain distribution of methoxy-X04 positive plaques in App SAA KI/KI at 8-month-old. Except panel i, all panels from d-n show data from 8-month-old mice. d Representative images of brain sections show Aβ plaque immunoreactivity in App SAA KI/KI but not in KI/+ mice. Leptomeningeal cerebral amyloid angiopathy is indicated by arrows. e Confocal images from App SAA KI/KI stained for plaques, microglia markers (Iba1, CD68), dystrophic neurites (AT8), neurofilament (Nf ) and lysosomal marker (LAMP1). Far right column provides a magnified view of the inset in merged images. Scale bars = 10 μm. f-g Quantification of brain areas covered by Aβ plaques (f) and AT8 area (g). N = 3-6 mice/ genotype. h Total tau levels in CSF. N = 6-11 mice/ genotype. i Higher level of CSF Nf-L in aged App SAA KI/KI relative to +/+ mice. N = 4-33 mice/ group. Age: F (4, 111) = 61.74, P < 0.0001. Genotype: F (1, 111) = 33.46, P < 0.0001. j-l Quantification of Iba1 (j), CD68 (k), percentage of the Iba1 and CD68 signals overlapping with plaques (l). N = 3-6 mice/ genotype. m TREM2 levels measured in brain homogenates. N = 10 mice/ genotype. n Differential abundance (log2) of cytokines measured in brain lysates. N = 10 mice/ genotype. Bars represent 95% confidence intervals of fold change. Graphs from f-m are box and whisker plots. P values: (f, j, k) one-way ANOVA with Dunnett's multiple comparison test, (g, h, m) t-test. Each genotype compared to App SAA +/+; **P < 0.01, ***P < 0.001 and ****P < 0.0001 expression levels of previously reported marker genes from different cell types in the mouse brain [71] (Supplementary Fig. 8a). At 8 months of age, the microglial transcriptome from App SAA KI/+ showed little difference when compared to App SAA +/+ controls (Fig. 3b).
Microglia from App SAA KI/KI mice, however, contained more than six hundred differentially expressed genes (DEGs) vs. App SAA +/+ controls (absolute log2 fold change [FC] > 1.2, false discovery rate [FDR] < 10%), with an upregulation of disease-associated microglia  Multi-omics analysis of microglia from App SAA KI/KI and control mice at 8 months of age. a FACS gating strategy for isolating microglia from whole brains of App SAA homozygous mice (KI/KI), heterozygous mice (KI/+) and wild-type mice (+/+) at 8 months of age for transcriptomics, lipidomics, and metabolomics analysis. b Volcano plots showing log2 fold change of gene expression between App SAA heterozygous (left) and homozygous (right) knock-in compared to WT; dark grey, fuchsia, and blue indicate DEGs (FDR <= 10%; see methods), DAM, and homeostatic genes, respectively. N = 6 mice per genotype. c Single sample activity scores for curated gene sets in App SAA heterozygous, App SAA homozygous, and 5xFAD. Intensities indicate log2 fold change of aggregated gene set score (row) per mouse (column) with respect to mean gene set score from matched WT mice. N = 5-6 mice per genotype. d Volcano plot showing log2 fold change of lipids between App SAA homozygous mice compared to WT. FDR <= 10%; absolute fold change > 20%. N = 6 mice per genotype. e Heatmap of top 10 lipids significantly altered by genotype in isolated microglia; columns represent individual mice. Intensities indicate log2 fold change of lipids with respect to mean abundance in WT mice. N = 6 mice per genotype (DAM) genes (Fig. 3b). Increased activity of other gene sets related to microglial state and function, such as cholesterol metabolism, glycolysis, phagocytic and lysosomal function resembled those observed in microglia from the 5xFAD transgenic or demyelination models [14,22,43] (Fig. 3c), demonstrating that some aspects of the microglia signature are conserved across diverse APP mouse models, including the App SAA KI mouse model.

Methoxy-X04-positive microglia exhibit an exacerbated gene expression and metabolic response
Next, we asked whether proximity to amyloid plaques affected microglia gene expression, lipid, and metabolite profiles. We hypothesized that the microglia in proximity to amyloid plaques may have more phagocytic activity and show higher intracellular Aβ levels than the microglia distal to plaques. We labelled fibrillar Aβ in App SAA KI/KI by peripheral injection of methoxy-X04 and quantified the proximity of methoxy-X04-positive (+) microglia to amyloid plaques via super-resolution microscopy (Fig. 4a). We confirmed that most methoxy-X04 (+) microglia were found around amyloid plaques (Fig. 4b), consistent with the notion that those microglia phagocytosed Aβ.
To further characterize the relationship between microglial phagocytic activity, gene, and metabolite profiles, we sorted methoxy-X04 (-) and (+) microglia by FACS prior to conducting RNA-seq, lipidomic, and metabolomic analyses (Fig. 4c-d). We found that about 15% of microglia are methoxy-X04 (+) (Fig. 4e) in App-SAA KI/KI at 8 months of age. This figure is comparable to what was reported in the 5xFAD model at 4 months of age [75]. Direct comparison of methoxy-X04 (+) vs.
Gene set enrichment analysis (GSEA) revealed approximately twice as many significant pathways (FDR <= 10%) in the methoxy-X04 (+) relative to the methoxy-X04 (-) microglia when these populations were compared against WT. Most notably, methoxy-X04 (+) cells exhibited higher enrichment for genes regulating innate immune function as well as metabolic functions that appear critical for supporting lipid clearance and metabolism (Fig  4i). As anticipated, these cells also showed an enrichment in genes associated with regulation of Aβ clearance and amyloid plaques (PIGs) (Fig 4i). Finally, the upregulation of the signature genes (Spp1, Tmem163, Apoe, Ctsb, Hif1a, among others) from the Mic1 microglia sub-cluster in human AD [19] were enriched in the phagocytic microglia from the App SAA homozygous mice (p-value < 1e-4, hypergeometric test ; Fig 4i), suggesting that this mouse model successfully recapitulates a subset of the microglial response found in human disease. Mirroring our transcriptomic analyses, lipid alterations were exacerbated in App SAA methoxy-X04 (+) microglia relative to methoxy-X04 (-) and WT microglia, as shown for instance for GM3(d36:1) and TG 20:4/36:3 species (Fig. 5a, Supplementary Tables 7 and 8). In total, 48 of the 172 assayed lipids showed higher and 32 displayed lower abundance in App SAA methoxy-X04 (+) compared to WT microglia (>= 20% change with FDR < 10%). Some lipid changes appeared to be specific to methoxy-X04 (+) microglia, including an increase in cholesteryl ester CE 22:6 and a decrease in other CE species, such as CE 20:4 (Fig. 5a). Finally, analyses of 101 polar metabolites in sorted microglia revealed 2 analytes with increased and 28 with decreased abundance in App SAA methoxy-X04 (+) compared to WT microglia (>= 20% change with FDR < 10%; Supplementary Tables 7 and 8). For example, the polyamine spermine accumulated specifically in methoxy-X04 (+) microglia (Fig. 5b-c). Therefore, our data lends further support to the notion that phagocytic microglia undergo profound cellular alterations, including changes in lysosomal activity, lipid dyshomeostasis, and other metabolic changes.

App SAA mice show enhanced TSPO and FDG-PET signals in brain
Based on the amyloid-related response of microglia in the App SAA KI model, we sought to determine if mice show elevated levels of the 18KDa translocator protein TSPO, an imaging marker that is associated with microglia biology [20,76,77] and elevated in AD patients [78,79]. We compared the TSPO positron emission tomography (PET) signal of App SAA +/+ controls and App SAA Heatmap of (a) lipids and (b) metabolites significantly altered by genotype and/or presence of methoxy-X04 dye in microglia isolated from App SAA KI/KI and App SAA +/+ mouse brain. n=10, columns represent individual mice. c Representative plots showing genotype and methoxy-X04 effects on key lipids and metabolites from FACS-isolated microglia. n=10. Statistical analysis was performed by fitting linear models using limma; FDR was calculated according to Benjamini's and Hochberg's method. All panels from a-c are showing data from 8-month-old mice KI/KI at different ages. App SAA KI/KI mice had a significantly higher TSPO-PET signal in the cortex at 12 and 20 months of age when compared to App SAA +/+ mice (Fig. 6a, c). To evaluate if this increase in TSPO expression was associated with changes in brain metabolism, we measured glucose uptake by 2-deoxy-2-[ 18 F] Fig. 6 Elevated TSPO-and FDG-PET cortical signals in App SAA mice. a-b Axial slices upon an MRI template show TSPO-PET and FDG-PET uptake patterns of WT and App SAA mice at different age groups. c App SAA KI/KI mice had a significantly higher TSPO-PET signal at 12 months and 20 months of age when compared to App SAA +/+ control mice. d App SAA KI/KI mice had a significantly higher glucose uptake at 12 and 20 months of age when compared to App SAA +/+ control mice. P values: t-test; *P < 0.05, **P < 0.01, and ****P < 0.0001. Xia et al. Molecular Neurodegeneration (2022) 17:41 fluoro-d-glucose (FDG)-PET. The cortical signal of FDG-PET was also elevated in App SAA KI/KI compared to App-SAA +/+ across the studied lifespan (Fig. 6b). App SAA KI/ KI had a significantly higher glucose uptake at 12 and 20 months of age when compared to control mice (Fig. 6d). TSPO-PET and FDG-PET quantification were strongly associated in App SAA KI/KI (R=0.709, p=0.007) but not in App SAA +/+ (R=0.171, p=0.577), suggesting that the elevation of glucose metabolism could be partially driven by responsive microglia in this model, as recently demonstrated in a distinct mouse model [20].

Characterization of behavioral phenotypes in App SAA KI model
APP transgenic mouse models often display behavioral alterations, including locomotor hyperactivity, habituation deficits to a novel environment, emotional alterations and learning and memory deficit s [80,81]. The emergence of behavioral abnormalities in App KI mouse models indicates that accumulation of Aβ without APP overexpression is sufficient to trigger age-dependent brain dysfunction [7,82,83].  Table 9). Longitudinal monitoring of running wheel activity in their home cages revealed an age-dependent increase in spontaneous activity (Fig. 7ac). Females showed a more pronounced phenotype compared to males, with a significant hyperactivity phenotype detected at 8 months of age ( Fig. 7a-c). To determine whether App SAA KI/KI show any deficit in habituation to a novel environment, we tested an independent cohort of mice at 16-17-month-old in a repeated version of the open field (Fig. 7d). Mice were tested twice a day for two days (trials 1-4) and retested two weeks later (trials 5-8) to assess locomotor activity and short-and long-term habituation [84].

Discussion
This study demonstrates that a mouse model expressing physiologically relevant levels of APP bearing familial AD mutations recapitulates several key aspects of AD biology. The App SAA KI mouse model shows progressive accumulation of amyloid pathology, increased levels of biomarkers of neurodegeneration, and robust astroglial and microglial responses. Deep phenotyping of microglia by multi-omics approach demonstrates that phagocytic microglia in the vicinity of amyloid plaques display profound gene expression and metabolic dysregulation in this model. Given the large contribution of microglial genes to AD risk, and the many microglial-directed therapies currently in clinical development [10] this model represents an opportunity to gain novel insight into the role of microglial biology in the context of amyloid pathology and offers a translationally relevant model system for evaluating the impact of microglial-directed therapies in a disease context. As anticipated based on the diffuse amyloid deposits seen in AD patients carrying the Austrian or the Arctic mutation [85], a large proportion of the amyloid plaques in the App SAA mice were also diffuse with or without a dense core structure. The fibrillar form of Aβ combined with a diffuse halo of amyloid was associated with key features of neuritic plaques [86,87] in the App SAA mice, including dystrophic neurites, accumulation of phosphorylated tau and enlarged LAMP1-positive lysosomes. In response to the progressive accumulation of amyloid plaques, the App SAA mice developed a robust immune response that was characterized by elevated levels of multiple neuroinflammatory cytokines and TREM2 in brain lysates, an increased density of responsive microglia at the vicinity of amyloid plaques, profound expression changes in genes belonging to inflammatory and DAM pathways in microglia, and an increase of TSPO-PET signal in the brain.
One of the fundamental questions in the field is whether microglia are affected by the amyloid plaque environment and if so, whether these changes in cell state are beneficial (e.g., by facilitating inflammation resolution) or pathogenic (e.g., by preventing inflammation resolution or promoting neurotoxicity). To begin Fig. 7 Hyperactivity phenotype in App SAA KI/KI mice. a-c Spontaneous activity was monitored in home-cage using running wheels. The average time running is represented for males (a) and females (b) at 4, 8, 12 and 18 months of age for the three genotypes (App SAA +/+, KI/+ and KI/KI). Graphs are box and whisker plots. P values: one-way ANOVA with Dunnett's multiple comparison test. App SAA KI/KI compared to the App SAA +/+ control group; ***P < 0.001 and ****P < 0.0001. c Illustration of time spent running averaged per genotype in female mice at 18 months of age. Data are represented as mean ± SEM. d 16-17-month-old male and female App SAA mice (KI/KI) and wildtype littermate controls (+/+) were tested in the open field for 5 min twice a day for two days (Day 1-2; Trials 1-4) and retested two weeks later (Day 15-16; Trials 5-8). Top: Distance travelled (cm) in the center and periphery of a circular arena. Bottom: P values for the repeated three-way ANOVA with genotype (KI/KI vs +/+), sex (male vs female), and location (center vs periphery) as fixed factors and their interactions. Bars represent mean ± SEM. Number of mice indicated in parentheses to address this question, we profiled methoxy-X04 (+) and (-) microglia derived from the App SAA brain using LCMS and assessed impact of Aβ phagocytosis on their lipidome and metabolome. This approach revealed significant alterations that are consistent with cellular dysfunction. One of the most striking manifestations is accumulation of ganglioside GM3 in methoxy-X04 (+) microglia, which might be indicative of a lysosomal dysfunction since gangliosides are glycosphingolipids that typically accumulate in lysosomal storage disorders (LSDs). In fact, lipidomic profiling of sorted microglia from independent LSD models, such as the Ids KO model of MPS-II (Hunter syndrome) and a Grn KO model of GRN-associated frontotemporal dementia, showed similar increases in GM3 [88,89]. GM3 accumulation has previously been described in the entorhinal cortex of AD patients and the forebrain of AD transgenic mouse models, suggesting that it might be broadly associated with AD pathology [72]. One can speculate that microglial GM3 accumulation may result from phagocytosis of Aβ, which can potentially interfere with normal lysosomal degradative capacity and decrease lysosomal catabolism of gangliosides.
Another prominent phenotype is accumulation of neutral lipids, such as triglycerides and to a lower extent, some CE species. While these lipids are generally found in cytoplasmic lipid droplets where they can store fatty acids and cholesterol while minimizing their lipotoxicity, accumulation of these lipids may also reflect metabolic imbalance and has been associated with pro-inflammatory states and neurotoxicity in other models [10,22,90,91]. Remarkably, these biochemical findings are consistent with Dr. Alzheimer's original observations of "adipose inclusions" in glial cells neighboring amyloid plaque s [92]. Consistent with these early seminal observations, xenografted human iPSC-derived microglia (xMG) in an AD mouse model background showed similarities with atherosclerotic "foam" macrophages, including increased occurrence of lipid droplets in amyloid plaque-bound xMGs [73]. Taken together with a growing amount of human genetic data linking various genes mediating lipid metabolism to AD (e.g., APOE4, ABCA7, CLU, TREM2), these published data as well as lipidomic data from our study lends strong support to the notion dysregulation of microglial lipid metabolism contributes to AD pathogenesis.
One of the most striking observations from our metabolomic study is selective accumulation of the polyamine spermine in methoxy-X04 (+) microglia from the App SAA KI mice. Spermine has been shown to modulate various cellular functions, including gene expression, ion channel function, response to oxidative stress, autophagy, metabolism and immune response. Spermine is critical for normal cell function as genetic ablation of spermine synthase (SMS) has been shown to cause oxidative stress and lysosomal impairment [93]. Additionally, spermine binds and neutralizes Aβ [94] and is upregulated in the PSAPP mouse model [95], suggesting that increase in spermine levels in methoxy-X04 (+) microglia may be a compensatory response to prevent the deleterious effects of Aβ. Alternatively, since spermine can be actively transported out of the lysosomes through the lysosome-associated transporter ATP13A2 and can accumulate in these organelles [96], spermine accumulation may reflect lysosomal dysfunction, as speculated for GM3. While our lipidomic and metabolic observations of phagocytic microglia in App SAA KI provide evidence of pathway dysregulation, it will be important to assess whether manipulation of these pathways through targeted molecular genetic and/ or pharmacological approaches in App SAA mice (e.g., via targeting GM3, TG/CE, or spermine metabolism) confer benefits or exacerbate specific disease-related phenotypes.
Microglia expression profiles from preclinical mouse models of AD have shown that microglia rewire their transcriptome in a manner that does not fit the canonical M1/M2 macrophage activation paradigm [8,9,12,14,15,[17][18][19]. A recent meta-analysis of single-cell RNA-seq studies of microglia from these models has postulated that the microglial diversity observed can be characterized as some combination of five broad states: Homeostatic, DAM-like, IFN-response, MHC-II, and a proliferating state [97]. Expression profiles of microglia from App SAA mice show that they exhibit a response similar to what has been seen in 5xFAD mice, with a strong upregulation of the DAM-like signature. Concordant with recent findings in 5xFAD mice reported by Grubman et al. [13], transcriptional analysis of methoxy-X04 (+) microglia in App SAA mice identified a number of acutely upregulated genes and biological processes involved in the regulation of amyloid beta clearance (Lrpap1, Tnf, Trem2), lipid metabolism (Apoe, Olr1, Fabp5, Lpl, Ch25h), and lysosomal function (Gla, Galns, Ctsd, Cst7). The transcriptional program found in the methoxy-X04 (+) microglia of App SAA mice appears to be a hyper-activated version of the methoxy-X04 (-) population when both are compared to microglia from WT mice, which stands in slight contrast to the qualitatively different response between the methoxy-X04 (+) and (-) populations from 5xFAD mice as reported in Grubman et al. [13] (Supplementary Fig. 8c-e).
In addition to neuroinflammation, the App SAA mice show elevated levels of biomarkers of neurodegeneration in the CSF. Neurofilament light chain (Nf-L) has received a lot of attention in the last few years and has been studied across various neurological conditions to evaluate the extent of neuronal damage [98]. In AD, Nf-L progressively increases in CSF and plasma and has been correlated with brain atrophy and cognitive decline [99,100]. Interestingly, we observed a progressive increase of Nf-L levels in the CSF in the App SAA mice, starting with trends at 8 months of age. Despite the increase of Nf-L in CSF from control animals with age, we detected a greater separation of Nf-L levels over time, suggesting that neuronal damage increased as pathology worsens in this model. The extent of Nf-L elevation in this model (~1.5-2 fold) was within the range of what has been reported in patients with mild-cognitive impairment and AD [101,102] and contrasted with the profound elevation of Nf-L levels in CSF reported in the 5xFAD and APP/PS1 mouse models (~10-20 fold) [101,102], indicating that App SAA mice may better recapitulate the rate of neurodegeneration estimated in the human condition than the 5xFAD or APP/PS1 mouse models. The excessive levels of Nf-L Fig. 8 Summary of disease-relevant biology in the App SAA mouse model increase in those models are likely caused by the accumulation of fAD mutations in APP and PSEN1 in the overexpressed transgenes which result in early and fast accumulation of amyloid plaques. Consistent with the Nf-L increase, we also found that total tau was elevated at 8 months of age, again within range previously reported in patients [103]. As Nf-L levels have been associated with hypometabolism in AD [104], we measured FDG-PET signal in the App SAA mice. In contrast to human FDG hypometabolism reported in the later phase of A D [105][106][107][108][109], we observed a cortical FDG hypermetabolism in the App SAA mouse model. Similar hypermetabolism was also found in other amyloid models and probably reflects enhanced microglial and astroglial glucose uptake [20,110]. In addition, it is possible that the neuronal loss in those models is not substantial enough to drive an overall reduction in brain. Despite the mouse to human discrepancy in FDG-PET signal, our data demonstrate increase of biomarkers of neurodegeneration as amyloid pathology accumulates in the App SAA mouse model.
We also observed the emergence of behavioral abnormalities in the App SAA KI mice, who displayed a robust alteration in spontaneous activities. Locomotor hyperactivity is commonly observed in transgenic mouse models of AD [81]. Although locomotor hyperactivity has been suggested to be linked to APP overexpression in APP/ TTA transgenic mice [111], our results indicate that fAD mutations without APP overexpression are sufficient to induce locomotor hyperactivity and habituation deficits. Importantly, we found that the hyperactivity in App SAA mice was experience-dependent and emerged as wildtype littermate mice habituate to the environment, particularly during delayed trials, potentially indicating mild impairment in context-dependent habituation, as previously described in other models [84,112].
To the extent that we could, we included both males and females, different ages and the three genotypes of interest in this study. Early in the project, we made the decision to use wild-type littermate mice as controls. While this experimental approach provides many advantages, we also recognized that the combination of the 3 fAD mutations (Swedish, Arctic and Austrian) in the App SAA KI line unfortunately precludes us from determining the relative contribution of each fAD mutation on the generation of Aβ peptides in this model. Generation of humanized Aβ mouse lines with inclusion of a single fAD mutation, and/or dual combination of those mutations (e.g., Swedish and Austrian, Swedish and Arctic, and Arctic and Austrian), would be necessary to rigorously explore the contribution of those mutations on Aβ production and stability. We have ongoing efforts to generate those new lines and will make them also available to the field. A recently generated mouse model expressing humanized Aβ (hAβ-KI) did not form amyloid plaques or show robust immune/glial responses and had only minor signs of neurodegeneration detected at 22 months of age [113]. In addition, a transcriptomics analysis of bulk brain tissue in the hAβ-KI showed that only a very few genes were differentially regulated compared to wild-type control mice, contrasting with the profound gene expression changes reported in microglia from App SAA KI mice. Surprisingly, the hAβ-KI mouse model showed behavioral and synaptic deficits after 10-14 months of age, suggesting that humanization of Aβ per se may be sufficient to cause minor behavioral abnormalities, at least in the background strain studied (mixed C57BL/6NTac and C57BL/6J background) [113]. At this stage, we thus cannot parse out the potential contribution of Aβ humanization and fAD mutations on behavioral phenotype in the App SAA KI mice. Altogether, our observations indicate that biological changes in the App SAA KI mouse model were largely driven by the fAD mutations inserted on the App gene, which is consistent with the notion that amyloid plaques and associated neurodegeneration are the main pathological driver in App KI models.

Conclusions
Our in-depth analysis of the App SAA knock-in mouse model confirms emergence of disease-relevant biology and progressive accumulation of pathological hallmarks of AD (Fig. 8). We identified extensive lipid and metabolite alterations in microglia, depending on their stage of phagocytic activity and proximity to amyloid plaques. Based on these findings, this new mouse model provides a helpful tool to investigate multiple relevant aspects of AD biology. We have therefore made it available to the broad scientific community to facilitate AD research and therapeutic development.