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

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.

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.

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.

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).

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.

Brains were homogenized using Qiagen Tissue-Lyser II (28 Hz, 3 min, 3 times) in lysis buffer (Cell Signaling #9803) containing protease inhibitor cocktail (Roche #4693159001) and PhosSTOP (Roche #4906837001) (1 ml lysis buffer for 100 mg tissue). Lysates were then incubated on ice for 10 min followed by centrifugation at 18,660 g for 20 min at 4°. Supernatants were transferred to new tubes for protein concentration determination and loading sample preparation. Protein lysate samples were boiled at 95°C and SDS-PAGE was performed using standard BioRad reagents. For Western blotting, PDVF membranes were incubated overnight at 4°C with the following primary antibodies diluted blocking buffer (Rockland): Mouse anti APP A4 clone 22C11 (Millipore MAB348, 1:1,000), Mouse anti APP clone 6E10 (Biolegend 803001, 1:2,000), Rabbit anti APP CTF (Sigma A8717, 1:4,000), Mouse anti-β-Actin (Sigma A2228, 1:4,000). Membranes were then incubated with the appropriate fluorescently conjugated secondary antibody (1:10,000, Li-Cor) and imaged using a Li-Cor Odyssey CLx system.

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.

Total mouse tau in CSF was measured using an ECL-immunoassay 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 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).

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).

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.

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, TissueVision 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 multi-step 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.

Fresh brains were fixed by immersion in 4% paraformaldehyde at 4°C for 24 hours then transferred to a phosphate buffered saline (PBS) solution with 0.1% sodium azide for storage until ready for processing. Brains were initially transferred to a 30% sucrose solution in PBS for two days before sectioning and then subsequently sectioned coronally on a freezing microtome at a thickness of 40 μm for most stainings except for astrocyte-related stainings (GFAP, GLT-1 and C3), which were performed on 30 μm sections. Sections were stored in cryoprotectant buffer (30% glycerol, 30% ethoxyethanol, and 40% PBS) at -20°C prior to staining. Seven to twelve coronal brain sections (from approximately bregma to 4.8 mm posterior to bregma) were selected for immunostaining for Aβ, Iba1, CD68, AT8, LAMP1, Neurofilament, CD31, alpha smooth muscle actin, GFAP, GLT-1 and C3. Sections were incubated for 1 hour at room temperature in 1x Tris-buffered saline solution containing 0.05% Tween (TBST) and 5% donkey serum, followed by primary antibodies overnight at 4°C. Sections were then washed in TBST and a solution of secondary antibodies was then applied for 1 hour at room temperature. Sections were washed in TBST prior to mounting and cover slipping with Prolong Glass Antifade Mountant solution (Thermo Fisher, P36984↗). Immunofluorescence was performed using the following primary antibodies: rabbit anti-human amyloid beta (IBL America, 18584; 1:500), rat anti-CD68 (BioRad, MCA1957A; 1:500), goat anti-Iba1 (Novus, NB100-1028; 1:500), mouse anti-pTau (Invitrogen, MN1020; 1:500), goat-anti-CD31 (R&D Systems, AF3628, 1:500), mouse-anti-alpha smooth muscle actin Cy3 (Sigma, C6198; 1:500), rat anti-LAMP1 (DSHB, 1D4B; 1:250), mouse anti-Neurofilament (BioLegend, 837801; 1:500), rabbit anti-GFAP (Sigma, G9269, 1:500), guinea pig anti-GLT-1 (Sigma, AB1783, 1:1000) and rat anti-complement component 3 (C3; clone 11H9, Novus Biologicals, NB200-540, 1:50); and the following secondary antibodies: Alexa Fluor 488 Donkey anti-rabbit IgG (Thermo Fisher, A-21206; 1:200), DyLight 550 Donkey anti rat-IgG (Thermo Fisher, SA5-10027, 1:200), Alexa Fluor Plus 555 Donkey anti-mouse IgG (Invitrogen, A32773; 1:200), Alexa Fluor 647 Donkey anti-goat IgG (Thermo Fisher, A-21447; 1:200), Alexa Fluor Plus 647 Donkey anti-mouse IgG (Thermo Fisher, A32787; 1:200), Alexa Fluor 488 donkey anti-mouse IgG (Thermo Fisher, A21202; 1:500), Alexa Fluor 555 goat anti-rat IgG (Thermo Fisher, A21434; 1:500), Alexa Fluor 555 goat anti-guinea pig IgG (Thermo Fisher, A21435; 1:500), Alexa Fluor 555 donkey anti-rabbit IgG (Thermo Fisher, A31572; 1:500).

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 (ThermoFisher P36930↗) followed by a coverslip. Sections were imaged using a confocal microscope (Leica SP8; Leica Microsystems, Inc.) with a 20x/.75IMM oil objective.

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 sub-cellular 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).

Mice were perfused with cold PBS and cortical and hippocampal tissues were dissected and dissociated into a single cell suspension using the Adult Brain Dissociation Kit (Miltenyi Biotec 130-107-677), according to the manufacturer’s protocol. Cell suspensions were stained with the following antibodies in FACS buffer (1% fatty acid-free BSA + 1mM EDTA in PBS) for 25 minutes on ice: Fixable Viability Stain BV510 (BD Biosciences, 564406), CD11b-BV421 (BioLegend 101251), CD45-APC (BD Biosciences, 559864), ACSA-2-PE (Miltenyi Biotec, 130-102-365), Mouse Fc blocker (anti-mouse CD16/32, BioLegend 101320). Cells were washed twice with FACS buffer and strained through a 100 μm filter before sorting CD11b⁺ microglia and ACSA-2⁺ astrocytes on a FACS Aria III (BD Biosciences) with a 100 μm nozzle. For bulk RNAseq analysis, cells were sorted directly into freshly prepared RLT-plus buffer (Qiagen) containing beta-mercaptoethanol; for lipid extraction, cells were sorted directly into LC/MS grade Methanol with internal standards.

To label Aβ-containing microglia, mice were injected i.p. with 10 mg/kg of methoxy-X04 (R&D Systems). Mice were perfused with PBS 24 hours after injection and cortical and hippocampal tissues were dissected and processed into single-cell suspension for staining as described above except the antibodies used for FACS were: Fixable/viability Dye 780 (Thermo fisher 65-0865-14), Cd11b-PE (BioLegend, 101208), CD45-APC (BD Biosciences, 559864), Mouse FC block 1:50 (BioLegend, 101320). Methoxy-X04 positive and negative microglia (CD11b+) were collected for RNAseq and LC-MS analysis.

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 SeqMatic (Fremont, CA, USA).

Sequencing adapters were trimmed from the raw reads with skewer (version 0.2.2) [36]. Reads were aligned to the mouse genome version GRCm38_p6. A STAR index (version 2.7.1a) [37] and built with the --sjdbOverhang=50 argument. Splice junctions from Gencode gene models (release M17) were provided via the --sjdbGTFfile argument. STAR alignments were generated with the following parameters: --outFilterType BySJout, --quantMode TranscriptomeSAM, --outFilterIntronMotifs RemoveNoncanonicalUnannotated, --outSAMstrandField intronMotif, --outSAMattributes NH HI AS nM MD XS and --outSAMunmapped Within. Alignments were obtained with the following parameters: --readFilesCommand zcat --outFilterType BySJout --outFilterMultimapNmax 20 --alignSJoverhangMin 8 --alignSJDBoverhangMin 1 --outFilterMismatchNmax 999 --outFilterMismatchNoverLmax 0.6 --alignIntronMin 20 --alignIntronMax 1000000 --alignMatesGapMax 1000000 --quantMode GeneCounts --outSAMunmapped Within --outSAMattributes NH HI AS nM MD XS --outSAMstrandField intronMotif --outSAMtype BAM SortedByCoordinate --outBAMcompression 6. Gene level counts were obtained using featureCounts from the subread package (version 1.6.2 )[38] . Gene symbols and biotype information were extracted from the Gencode GTF file.

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 mean-centered 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 [45–47]. 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].

Lipid extraction was performed using Matyash liquid-liquid 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 N₂ gas. Dried samples were reconstituted in 100 μl of MS-grade methanol for further analysis by LC-MS/MS.

Lipid analyses were performed by 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 C18 1.7 μm, 2.1×100 mm column (Waters) using a flow rate of 0.25 mL/min at 55°C. Electrospray ionization was performed in positive and negative ion modes. For positive ionization mode, mobile phase A consisted of 60:40 acetonitrile/water (v/v) with 10 mM ammonium formate + 0.1% formic acid; mobile phase B consisted of 90:10 isopropyl alcohol/acetonitrile (v/v) with 10 mM ammonium formate + 0.1% formic acid. For negative ionization mode, mobile phase A consisted of 60:40 acetonitrile/water (v/v) with 10 mM ammonium acetate + 0.1% acetic acid; mobile phase B consisted of 90:10 isopropyl alcohol/acetonitrile (v/v) with 10 mM ammonium acetate + 0.1% acetic acid. The gradient was programmed as follows: 0.0-8.0 min from 45% B to 99% B, 8.0-9.0 min at 99% B, 9.0-9.1 min to 45% B, and 9.1-10.0 min at 45% B.

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. Lipids were quantified as area normalized to specific non-endogenous internal standards as reported in Supplementary Table. 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). 5 5

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, [¹⁸F]GE-180 TSPO-PET and [¹⁸F]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 triple-mouse imaging chamber. A 15-minute anatomical T1 MR scan was performed at 15 min after [¹⁸F]FDG injection or at 45 min after [¹⁸F]GE-180 injection (head receive coil, matrix size 96 × 96 × 22, voxel size 0.24 × 0.24 × 0.80 mm³, repetition time 677 ms, echo time 28.56 ms, flip angle 90°). PET emission was recorded at 30-60 min p.i. ([¹⁸F] FDG) or at 60-90 min p.i. ([¹⁸F]GE-180). PET list-mode data within 400-600 keV energy window were reconstructed using a 3D iterative algorithm (Tera-Tomo 3D, Mediso Ltd, Hungary) with the following parameters: matrix size 55 × 62 × 187 mm³, voxel size 0.3 × 0.3 × 0.3 mm³, 8 iterations, 6 subsets. Decay, random, and attenuation correction were applied. The T1 image was used to create a body-air material map for the attenuation correction.

Normalization of injected activity was performed by the previously validated myocardium correction method [50] for [¹⁸F]GE-180 TSPO-PET and by standardized uptake value (SUV) normalization for [¹⁸F]FDG-PET.

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.

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 one-way 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]: log₂(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 log₂ transformed ratios of each sample instead and the following linear model was fit using limma: log₂(abundance) ~ genotype + batch + sex.

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↗

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 Aβ₄₀ and Aβ₄₂ 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. 1b-e). At 2 months of age, before amyloid plaques were detected (Fig. 2a), we observed a reduction of Aβ₄₀ 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 Aβ₄₂ levels in insoluble brain fraction further increased the Aβ_42/40 ratio in the App^SAA KI/KI (Fig. 1c). The reduction of Aβ₄₀ 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 Aβ₄₀. However, since 4G8 antibody similarly binds Aβ₄₀ and Aβ₄₂, 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.

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 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 [66–68], 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.

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 non-amyloid models [22].

We first confirmed the enrichment and purity of isolated microglia using RNA-seq data to compare the 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 (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.

Targeted liquid chromatography and mass spectrometry (LCMS) analysis of 176 lipids in FACS-isolated microglia identified 4 significantly decreased and 16 significantly increased analytes between App^SAA KI/KI and +/+ mice (>= 20% change with FDR < 10%) (Supplementary Tables 5 and 6). Ganglioside GM3 (d36:1), various species of triglyceride (TG) (including species of arachidonate-containing TG) as well as a variety of phospholipid species (i.e., PG, PS and PI) were more abundant in App^SAA KI/KI microglia (Fig. 3d-e). Conversely, lysolipid lysophosphatidylinositol LPI 18:1 and coenzyme Q10 were less abundant in App^SAA KI/KI microglia (Fig. 3d-e). Together, these observations indicate broad changes in microglia lipid metabolism in the App^SAA KI mouse model, reminiscent of findings in human AD brains [72] and other AD mouse models [73, 74].

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. (-) microglia from App^SAA KI/KI identified more than 800 DEGs, with methoxy-X04 (+) microglia showing increased DAM activation (Fig. 4f), as well as increased expression of Gusb, Prdx6, Ctsa, Arpc1b, Grn among other genes recently reported in the plaque-induced gene (PIG) signature [8] (Supplementary Fig. 8b). Higher levels of Trem2 transcripts were detected in methoxy-X04 positive amyloid plaques by in situ hybridization (Fig. 4g) in App^SAA homozygous mice at 8 months of age, consistent with an enrichment of responsive microglia clustering around amyloid plaques (Fig. 2k-l). Methoxy-X04 (-) microglia from App^SAA mice also showed significant changes compared to methoxy-X04 (-) microglia from WT control mice, but the effects were not as strong as the ones measured in App^SAA methoxy-X04 (+) microglia (Fig. 4h; Supplementary Fig. 8b).

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.

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 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- [¹⁸F]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].

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]. To determine whether the App^SAA KI mouse model developed similar behavior alterations, we first evaluated a cohort of mice from 4 to 18 months of age using the frailty scoring, activity monitoring in their home cage and an open field. App^SAA KI/KI and KI/+ did not show deficits in frailty index at any of the tested ages (Supplementary Table 9), indicating that this model does not develop major general health issues. App^SAA KI/KI mice displayed a robust hyperactivity phenotype at 18 months of age in the open field (Supplementary Table 9). Longitudinal monitoring of running wheel activity in their home cages revealed an age-dependent increase in spontaneous activity (Fig. 7a-c). 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]. Compared to sex-matched wildtype littermate mice, App^SAA KI/KI had increased locomotor activity (hyperactivity), which became more prominent as wildtype mice habituated to the open field over the repeated testing (habituation deficits), and App^SAA KI/KI showed stronger center preference (disinhibition). Overall, there were prominent effects in genotype (p = 0.004), sex (p = 0.008), and location (p <0.001), indicating that App^SAA KI/KI displayed altered behavior; males and females regardless of genotype behaved differently; and there was a distinct preference for location (center). Genotype interacted with location (p < 0.001), but not with sex (p = 0.439), indicating that App^SAA KI/KI behaved differently in the center (higher activity) than in the periphery (decreased activity) relative to wildtype littermate controls, and that male and female App^SAA KI/KI displayed similar behavioral alterations relatively to sex-matched wildtype mice. Notably, male App^SAA KI/KI displayed prominent hyperactivity during the first (Trials 5 and 7) but not the second (Trials 6 and 8) trials of the retest days, whereas female App^SAA mice displayed a more consistent hyperactivity throughout all trials relative to sex-matched wildtype controls. These results indicate that aged male and female App^SAA KI/KI develop robust hyperactivity, habituation deficits, and disinhibition alterations.

Not applicable

Not applicable

D.X., S.L., T.S., M.C., J.S., E.T., J.D., M.P., N.E.K, J.Y.Z., S.L.D., T.K.E., C.C.L., S.D., C.H., H.H., R.C., E.Y., H.S., S.T.M., R.L., K.L., R.G.T., K.S.L., J.W.L., G.D.P. and P.E.S are paid employees and shareholders of Denali Therapeutic Inc. J.D.W. and J.A.H are currently employed by Cajal Neuroscience.

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↗