Ms4a4a deficiency ameliorates plaque pathology in a mouse model of amyloid accumulation

Abstract

INTRODUCTION

Genome‐wide association studies have identified MS4A4A, a microglia‐enriched gene, as a modulator of Alzheimer's disease (AD) risk. Common variants in MS4A4A affect AD susceptibility, gene expression, triggering receptor expressed on myeloid cells 2 (TREM2) signaling, and microglial transcriptional states, but the gene's functional role remains unclear.

METHODS

Using a novel model, we investigated the impact of Ms4a4a loss in the 5xFAD mouse model of amyloid beta (Aβ) accumulation.

RESULTS

Ms4a4a deficiency reduced steady‐state Aβ levels and shortened its half‐life in brain interstitial fluid. Aged 5xFAD mice lacking Ms4a4a exhibited more compact plaques and lower overall plaque burden. Microglia deficient in Ms4a4a showed a pro‐inflammatory profile and elevated matrix metalloproteinase 9 (MMP‐9) production, which may facilitate Aβ degradation. Notably, human carriers of the AD‐resilient variant rs1582763 near MS4A4A also displayed increased cerebrospinal fluid MMP‐9 levels.

DISCUSSION

Together, we show that Ms4a4a loss enhances Aβ clearance and reduces pathology, suggesting a protective mechanism that may inform microglia‐targeted AD therapies.

Highlights

• We examined the impact of Ms4a4a loss on amyloid beta (Aβ) pathology using a mouse model of Aβ accumulation (5xFAD).

• Ms4a4a loss reduces overall plaque burden and increases plaque compaction.

• Microglia lacking Ms4a4a are more pro‐inflammatory and produce more matrix metalloproteinase 9 (MMP‐9).

• Alzheimer's disease (AD) resilience variant carriers, MS4A4A rs1582763, exhibit significantly elevated levels of cerebrospinal fluid MMP‐9.

• Our findings suggest that reduction of MS4A4A may be a therapeutic approach for AD.

1. BACKGROUND

Genome‐wide association studies have begun to uncover the complex genetic architecture of Alzheimer's disease (AD). AD risk loci expression is enriched in microglia. ¹ , ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ , ⁹ Common variants near the MS4A gene family have been associated with AD risk ⁸ , ¹⁰ and levels of soluble triggering receptor expressed on myeloid cells 2 (sTREM2) in the cerebrospinal fluid (CSF). ¹¹ , ¹² However, the mechanism by which MS4A4A contributes to AD remains unknown.

MS4A4A is a tetraspan protein enriched in myeloid cells, including microglia. ¹³ In the periphery, MS4A4A has been proposed to facilitate signaling through recycling of tyrosine‐protein kinase KIT, promote immune response via co‐localization with TREM2, and contribute to an anti‐inflammatory response. ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ In the human brain, MS4A4A is expressed in a subset of microglia that are transcriptionally enriched in anti‐inflammatory cytokines. ¹⁹ , ²⁰ Other microglia‐enriched AD risk genes, including TREM2, CD33, and PLCG2, have been shown to alter amyloid accumulation by modifying microglia recruitment and phagocytosis of amyloid beta (Aβ). ²¹ , ²² , ²³ , ²⁴ , ²⁵ Yet, microglia play many critical roles in the brain that, when disrupted, may contribute to AD pathophysiology. It is unknown whether AD risk genes consistently impact a subset of microglia functions or whether different AD risk genes act on distinct microglia processes.

Here, we leveraged a novel mouse model to explore a known risk factor for AD. A global Ms4a4a knockout in a model of amyloid pathology (5xFAD) revealed that loss of Ms4a4a decreases basal Aβ concentration and decreases Aβ half‐life in the brain interstitial fluid (ISF) prior to plaque accumulation. Ms4a4a loss results in a reduced plaque burden in aged mice, with plaques in Ms4a4a‐deficient mice being more compact. Interestingly, these phenotypes are not driven by altered microglial recruitment to plaques, as has been described for other AD risk genes like TREM2, but instead, they are driven by a shift to a pro‐inflammatory state and the production of more Aβ‐degrading enzyme, matrix metalloproteinase 9 (MMP‐9). In human subjects, an AD resilience variant near the MS4A4A gene locus (rs1582763) is associated with elevated MMP‐9 levels in the CSF. Together, this study suggests that Ms4a4a deficiency improves Aβ pathology by altering Aβ degradation via MMP‐9.

2. METHODS

2.1. Animals

Animal care and surgical procedures were approved by the animal studies committee of Washington University School of Medicine in accordance with guidelines from the US National Institutes of Health.

Constitutive Ms4a4a ^−/‐ mice were generated by Alector using CRISPR/Cas9 technology. Guide RNAs (gRNAs) were designed to target a non‐coding region preceding exon 1 (ACCAGATCCAGTCCTTGTAG) and a non‐coding region succeeding exon 7 (GGAATAATCTGCCAACTTCC‐TGG; Figure 1A) and injected into zygotes to generate Ms4a4a ^+/− mice.

FIGURE 1.

Generation of 5xFAD Ms4a4a deficient mice. A, Gene targeting strategy for CRISPR/Cas9‐mediated knockout of Ms4a4a. B, Breeding scheme for the generation of an experimental cohort of 5xFAD Ms4a4a‐WT and 5xFAD Ms4a4a‐KO mice. C, Quantification of Ms4a4a levels in 5xFAD 4A‐WT and 5xFAD 4A‐KO mice reveals the absence of Ms4a4a expression in 5xFAD 4A‐KO mice (n = 6 mice per genotype). Welch's t test, ***, P = 0.0005. D, Quantification of Ms4a family gene expression reveals similar expression of Ms4a family genes between 5xFAD 4A‐WT and 5xFAD 4A‐KO mice (n = 6 mice per genotype; two‐way analysis of variance (ANOVA))

To understand the impact of Ms4a4a loss on amyloid pathology, 5xFAD (B6.Cg‐Tg[APPSwFlLon, PSEN1*M146L*L286V[6799Vas/Mmjax, Stock number 034848‐JAX, MMRRC, The Jackson Laboratory) hemizygous mice were crossed with the Ms4a4a ^+/− mice to generate 5xFAD x Ms4a4a ^+/− mice. Next, 5xFAD x Ms4a4a ^±/‐ were crossed with Ms4a4a ^+/− mice to obtain 5xFAD x Ms4a4a ^−/− (5xFAD 4A‐KO) mice and 5xFAD x Ms4a4a ^+/+ (5xFAD 4A‐WT) littermates (Figure 1B). Genotype was confirmed using primers to determine the presence of Ms4a4a and the 5xFAD transgene (Table S1 in supporting information). Loss of Ms4a4a was validated using quantitative polymerase chain reaction (qPCR; Figure 1C; Table S2 in supporting information).

2.2. In vivo microdialysis

Prior to plaque accumulation (8 weeks of age), hippocampal ISF Aβ levels were quantified in 5xFAD 4A‐KO and 5xFAD 4A‐WT littermate controls using microdialysis as previously described. ²⁶ , ²⁷ A 38‐kDa molecular weight cut‐off semi‐permeable membrane was used to allow for molecules smaller than this cut‐off to diffuse into the probe. The probe was flushed with perfusion buffer at a constant rate (1.0 uL/minute), and samples were collected into a refrigerated fraction collector. ISF samples were then assayed by sandwich enzyme‐linked immunosorbent assay (Aβx‐40). A mouse monoclonal anti‐Aβ40 capture antibody (mHJ2) made in house was used in conjunction with a biotinylated central domain detection antibody (mHJ5.1) and streptavidin‐poly‐HRP‐40 (Fitzgerald Industries). ²⁸ Mice were housed in cages that permit free movement and ad libitum food and water during microdialysis. ISF Aβ was sampled every 60 minutes between hours 5 and 12 (after microdialysis probe insertion), and concentrations were averaged to determine the baseline ISF Aβ levels in each mouse. At hour 12, mice were administered the γ‐secretase inhibitor (Compound E; Tocris; 20 mg/kg) intraperitoneally, which enabled determination of the elimination half‐life of ISF Aβ. Aβ half‐life was calculated using first‐order kinetics. Statistical difference was measured in GraphPad Prism 10 using an unpaired Student t test with Welch's correction.

RESEARCH IN CONTEXT

1. Systematic review: Prior literature suggests that common variants in MS4A4A influence Alzheimer's disease (AD) risk, MS4A4A expression, triggering receptor expressed on myeloid cells 2 (TREM2) signaling, and a specific microglial transcriptional state, though the exact role of MS4A4A in AD remains unclear.

2. Interpretation: Microglia lacking Ms4a4a in a mouse model of amyloid accumulation have fewer plaques and an increase in pro‐inflammatory microglia that produce more matrix metalloproteinase 9 (MMP‐9), which promotes degradation of all forms of amyloid beta (Aβ), including Aβ fibrils. Human subjects that carry a variant near MS4A4A (rs1582763) that confers resilience to AD also exhibit significantly elevated levels of MMP‐9 in their cerebrospinal fluid.

3. Future directions: Our findings suggest that loss of Ms4a4a improves Aβ pathology by altering Aβ clearance, offering insights for therapeutic interventions in AD. Further research is necessary to understand the mechanisms by which Ms4a4a impacts additional domains of AD pathophysiology.

2.3. Brain tissue preparation

At 6 months of age, female 5xFAD 4A‐KO and 5xFAD 4A‐WT mice were anesthetized with sodium pentobarbital and perfused with ice cold phosphate‐buffered saline (PBS). Brains were dissected and cut into two hemispheres. The right hemisphere was further dissected to isolate the cortex and hippocampus, which were snap frozen on dry ice and stored at –80°C for biochemical analyses. The left hemisphere was fixed in 4% paraformaldehyde (w/v) for 48 hours followed by 30% sucrose in PBS at 4°C. Coronal sections (40 µm) were cut on a freezing‐sliding microtome, and slices were collected and stored in cryoprotectant solution (0.2 M PBS, 30% sucrose, and 30% ethylene glycol) at –20°C.

2.4. 3,3′‐Diaminobenzidine staining for amyloid plaque quantification

Brain sections from each animal were incubated in 0.3% H₂O₂ in Tris‐buffered saline (TBS) for 10 minutes, then blocked in 3% milk in TBS containing 0.25% Triton‐X100 (TBS‐X) for 30 minutes. To evaluate plaque pathology, tissue was incubated in HJ3.4b antibody (anti‐Aβ‐1‐13; 1.2 µg/mL; a generous gift from the Holtzman lab) diluted in 3% milk in TBS‐X overnight at 4°C. Sections were washed with TBS‐X, then incubated in Vectastain ABC Elite solution (Vector Laboratories, PK‐6100) in TBS for 1 hour at room temperature. Sections were washed with TBS and then incubated in 3,3′‐diaminobenzidine (DAB) solution. Sections were then mounted onto slides and cover slipped with Cytoseal XYL (Electron Microscopy Sciences 18009). Brightfield imaging was performed with a Keyence BZ‐X810 microscope.

Plaque burden, plaque count, individual plaque size, and average plaque size were analyzed using NIH ImageJ software. The hippocampus and cortex were analyzed separately for each mouse. Plaque burden was expressed as a percentage of the total area for each brain region and averaged across two slices per animal. Plaque count was expressed as the number of plaques per mm² averaged across two slices for each animal. Plaque size was expressed in µm² averaged across two slices for each animal. Analysis of plaque distribution was performed by stratifying total plaque coverage based on size in µm² and the frequency of occurrence in 349 um² increments as previously described. ²⁹ Statistical analyses were performed using GraphPad Prism 10. Differences in plaque burden, size, and count were measured using an unpaired Student t test with Welch's correction. Statistical difference in plaque distribution was measured using a one‐way ANOVA.

2.5. Immunofluorescence and X‐34 plaque staining

Brain sections from each animal were briefly washed with PBS and then permeabilized with PBS containing 0.25% Triton‐X100 (PBS‐X) for 30 minutes at room temperature. Sections were then stained with X‐34 (10 mM stock solution in dimethyl sulfoxide) diluted 1:3000 in staining buffer containing 60% PBS and 40% ethanol, pH 10 for 20 minutes at room temperature. Sections were rinsed in a wash buffer containing 60% PBS and 40% ethanol. After X‐34 plaque staining, sections were blocked and permeabilized in PBS‐X containing 10% goat serum for 1 hour at room temperature, then incubated with the following antibodies diluted in PBS‐X containing 1% goat serum overnight at 4°C: rabbit anti‐Iba1 (Wako 019‐19741; 1:500), rat anti‐CD68 (BioRad MCA1957; 1:2000), chicken anti‐GFAP (abcam ab4674; 1:1000), rabbit anti‐BACE1 (abcam ab108394; 1:500), and HJ3.4b anti‐Aβ (a generous gift from the Holtzman lab, 1.2ug/mL). Sections were washed with PBS then incubated in the following secondary antibodies diluted 1:400 in PBS‐X for 1 hour at room temperature: goat anti‐rabbit AF568 (Invitrogen A11011), goat anti‐rat AF647 (Invitrogen A21247), goat anti‐chicken AF647 (Invitrogen A32933), and streptavidin AF647 (Invitrogen S21374). Sections were washed with PBS, mounted on slides, and cover slipped with Fluoromount‐G mounting medium (Invitrogen 00‐4958‐02).

2.6. Immunofluorescent microscopy and quantification

To analyze ionized calcium‐binding adaptor molecule 1 (Iba1) and glial fibrillary acidic protein (GFAP) coverage and intensity in the hippocampus and cortex, images were acquired using a Keyence BZ‐X810 microscope. Laser intensity and exposure times were set for each staining cohort by surveying tissue and selecting appropriate parameters that could remain the same for all samples. While these values varied by antibody, all sections in a staining cohort were imaged under identical conditions at the same magnification.

To determine percent area stained for markers such as Iba1 and GFAP, TIFF image files were opened in ImageJ and converted to 8‐bit grayscale files. Images from the same staining cohort, displaying varying fluorescence intensities, were used to determine an optimal threshold value that could capture staining across all samples. The threshold was then consistently applied to all images in the cohort. Grayscale images were generated for the cortex and hippocampus and quantified as percent area stained using the Analyze Particles function. Two sections per mouse per region were analyzed and averaged.

All confocal images were acquired using a Zeiss LSM980 microscope. To assess plaque composition and microglia distance from plaques for Scholl analysis, z‐stack images of individual plaques were imaged at 20x with 2.2x zoom. To evaluate Iba1, CD68, and beta‐secretase 1 (BACE1) proximity (within 15 µm) to plaques, 20x images were taken.

Confocal images were analyzed in Imaris, in which 3D reconstructions were made. To determine plaque composition, the volumes of HJ3.4 and X‐34 were calculated for each plaque. The X‐34 proportion of the plaque volume was then calculated in Microsoft Excel (X‐34 volume/[X‐34 volume + HJ3.4 volume – colocalized volume])). Imaging and quantification were performed in two brain sections per mouse. Within each brain section, 5 images per brain region (e.g., cortex or hippocampus) were analyzed and averaged (10 images per brain region per mouse).

To determine microglial distance from plaques using Scholl analysis, the volumes of X‐34 and Iba1 were calculated for each plaque. Each image contained only one plaque. Iba1 measurements were classified based on distance from the plaque in 15 µm increments. Microglial volume within each bin was normalized to plaque volume within an image. Imaging and quantification were performed in two brain sections per mouse. Within each brain section, two to four images per brain region (e.g., cortex or hippocampus) were analyzed and averaged. GraphPad Prism 10 was used to identify and exclude outliers. Statistical difference was measured using a two‐way ANOVA. Comparisons were made between genotypes within a bin and between bins within a genotype. Iba1, CD68, and BACE1 measurements within 15 µm of a plaque were classified as “proximal to a plaque.” The volumes of all reconstructions determined to be “proximal to a plaque” were added and normalized to the total plaque volume within an image. Two sections per mouse with one image per brain region per section were analyzed and averaged. Statistical difference was measured in GraphPad Prism 10 using an unpaired Student t test with Welch's correction (plaque compaction and Iba1, CD68, or BACE1 proximity to plaque).

2.7. Serial protein extraction

Frozen cortical tissue was weighed and homogenized using a previously described three‐step serial protein extraction protocol. ³⁰ Tissue was first homogenized in reassembly buffer (RAB; 20 mg tissue per 400 uL buffer; GBiosciences 786‐91) supplemented with protease inhibitor cocktail (PI; Sigma P8340; 1:500) and phosphatase inhibitor (PPI; PhosSTOP, Roche 04906837001) with the Red Lysis Kit (1.5 mL tubes, Next Advance REDE1) using a bullet blender. Samples were centrifuged at 10,000 × g for 5 minutes at 4°C to pellet the RAB‐insoluble fraction, then the supernatant was ultracentrifuged at 50,000 × g for 20 minutes at 4°C using an MLA‐130 rotor in an ultracentrifuge (Beckman) to obtain the RAB‐soluble protein fraction. The pellet remaining in the beaded tube was then homogenized in radioimmunoprecipitation assay (RIPA) buffer (150 mM NaCl, 50 mM Tris, 0.5% deoxycholic acid, 1% Triton X‐100, 0.1% SDS, 5 mM EDTA, 20 mM NaF, and 1 mM Na₃VO₄, pH 8) supplemented with PI and PPI using a bullet blender. Samples were centrifuged at 10,000 × g for 5 minutes at 4°C to pellet the RIPA‐insoluble fraction, then the supernatant was sonicated at 10% amplitude for 30 seconds and ultracentrifuged at 50,000 × g for 30 minutes at 4°C to obtain the RIPA‐soluble protein fraction. The pellet remaining in the beaded tube was then homogenized in 70% formic acid (FA) using a bullet blender then centrifuged at 10,000 × g for 5 minutes at 4°C. The supernatant was sonicated at 10% amplitude for 30 seconds and ultracentrifuged at 50,000 × g for 20 minutes at 4°C to obtain the FA‐soluble protein fraction. The FA‐soluble fraction was then neutralized with 1 M Tris supplemented with PI and PPI by making a 1:20 dilution. All protein fractions were stored at –80°C for biochemical analyses.

2.8. Immunoblotting

To evaluate total protein concentration of mouse brain RAB‐ and RIPA‐soluble protein fractions, a bicinchoninic acid assay (PierceBCA Protein Assay Kit, ThermoFisher Scientific 23225) was performed following manufacturer's recommendations. To perform sodiumdodecylsulfate polyacrylamide gel electrophoresis, 5 ug of protein was mixed with 4 × Laemmli sample buffer (Bio‐Rad 161‐0747) and 10% β‐mercaptoethanol and heated at 95°C for 10 minutes. A 4% to 12% bis‐tris gel (NuPAGE) was used to assess MMP‐9, neprilysin (NEP), insulin‐degrading enzyme (IDE), and cathepsin B levels. A 16.5% tris‐tricine gel was used to assess levels of full‐length amyloid precursor protein (APP) and C‐terminal fragments (CTF). Proteins were transferred to a polyvinylidene fluoride membrane. Membranes were blocked in 5% milk in PBS supplemented with 0.1% Tween 20 (PBS‐T) for 1 hour at room temperature then probed with the following antibodies diluted in 5% milk in PBS‐T overnight at 4°C: rabbit anti‐beta Amyloid (CT695; ThermoFisher Scientific 51‐2700; 1:400), rabbit anti‐MMP‐9 (abcam ab38898; 1:1000), rabbit anti‐neprilysin (abcam ab5458; 1:1000), rabbit anti‐insulin degrading enzyme (abcam ab133561; 1:1000), rabbit anti‐Cathepsin B (Cell Signaling Technology 3383; 1:1000), and rabbit anti‐β‐Actin (Cell Signaling Technology 4970; 1:8000). Membranes were then washed and incubated with one of the following secondary antibodies diluted in 5% milk in PBS‐T for 1 hour at room temperature: mouse anti‐rabbit horseradish peroxidase (HRP; Jackson ImmunoResearch 211‐032‐171; 1:5000) and goat anti‐rabbit HRP (Cell Signaling Technology 7074; 1:5000). After washing, membranes were developed using Lumigen ECL Ultra (Lumigen TMA‐100) on a Bio‐Rad Chemidoc Imaging System. Images were analyzed using Bio‐Rad Image Lab Software. Statistical differences were measured in GraphPad Prism 10 using an unpaired Student t test with Welch's correction.

2.9. qPCR

Frozen cortical tissue was weighed then homogenized in QIAzol Lysis Reagent (20 mg tissue/ 700 uL QIAzol; Qiagen 79306) using a handheld tissue grinder. QIAzol samples were then subjected to chloroform extraction followed by mixing and centrifugation at 12,000 × g for 15 minutes at 4°C. RNA was extracted from the aqueous layer using a RNeasy Mini Kit (Qiagen 74106) according to manufacturer's instructions. RNA concentration was measured on a Nanodrop spectrophotometer, then cDNA was made using a High‐Capacity cDNA Reverse Transcription Kit (Applied Biosystems 4268814). Real‐time qPCR was performed with a Taqman gene expression assay or primers (Table S2) with iTaq Universal SYBR Green Supermix (BioRad 1725121) using a QuantStudio 12k Flex Real‐Time PCR System (ThermoFisher Scientific). Measurements were normalized to Gapdh for analysis. Statistical differences were measured in GraphPad Prism 10 using a two‐way ANOVA.

2.10. CSF marker association with rs1582763 and rs6591591

Informed consent was approved by the Washington University School of Medicine in St. Louis Institutional Review Board and Ethics Committee. All data were analyzed anonymously.

CSF measurements of MMP‐9 were generated as part of the SOMAscan7k proteomics panel. ³¹ The dataset consisted of approximately equal numbers of neurologically healthy controls, individuals with AD, and individuals with other forms of dementia (frontotemporal dementia, Lewy body dementia, and Parkinson's disease). Each individual had genomic data generated using either whole‐genome sequencing or array‐based genotyping. In total, 3506 CSF samples from eight cohorts were measured using either the SOMAscan7k or SOMAscan5k panel, after filtering based on quality control or genetic relatedness and selection for individuals of European ancestry.

The dataset underwent stringent quality control at both the protein and individual level to ensure high‐quality data. After quality control, protein measurements were converted to the log(10) scale and were converted to z scores with a mean of zero and standard deviation of 1 using the scale() R function.

Genotypes for each individual at rs1582763 and rs6591561 (GRCh38 coordinates chr11:60254475:G:A and chr11:60302703:A:G) were extracted from the full matrix along with z score–normalized measurements of MMP‐9 (SOMAscan aptamer ID X2579.17). To control for potential confounding variables, MMP‐9 protein levels were residualized using the lm() function in R, in which z score levels of the aptamer were treated as the response variable and age, sex, the first 10 genetic principal components, and dummy variables combining both cohort and genotyping array (e.g., ADNI_OmniEx) were treated as the terms of the model. The residuals were then extracted, and test statistics for the difference in the residuals for each of the three genotypes (A/A, A/G, G/G) were calculated using a two‐sided Wilcoxon test as implemented in the wilcox_test() R function in the rstatix package (version 0.7.2). The violin plots were generated using the ggplot2 (version 3.5.1) and ggpubr (version 0.6.0) R packages.

2.11. Tissue dissociation for flow cytometry

At 6 months of age, 5xFAD 4A‐KO and 5xFAD 4A‐WT mice were anesthetized with isoflurane and retro‐orbitally injected with an APC/Cy7‐CD45 antibody (100 uL; 3 mg/mL). Five minutes after injection, mice were anesthetized with sodium pentobarbital then perfused with ice cold PBS. The brain was harvested, and the cerebellum and olfactory bulbs were removed. Brains were temporarily stored on ice in Dounce Buffer (1X HBSS, 5 mM HEPES, 0.5% glucose).

Tissue was processed using a previously described method. ³² All procedures were performed on ice. Using a razor blade, brains were chopped to roughly 1 mm³ pieces then transferred to a 5 mL Dounce homogenizer containing 5 mL of Dounce Buffer. Tissue was homogenized with 6 to 10 full strokes until the solution appeared to be homogenous then passed through a 40 µm strainer into 50 mL tubes. Samples were centrifuged and resuspended in 30% Percoll with 2 mL of 1X Dulbecco's phosphate‐buffered saline carefully overlaid. Samples were centrifuged at 400 × g for 20 minutes with the brake set to 0. The myelin layer and supernatant were removed, and the remaining cell pellet was resuspended in RPMI 1640 and stored on ice until staining for flow cytometry.

2.12. Flow cytometry

Cells were incubated with Zombie Aqua dye (Fixable Viability kit, BioLegend 423102; 1:1000) for 20 minutes in 1X PBS then washed with fluorescence‐activated cell sorting (FACS) buffer (1X PBS, 2% FBS). Samples were incubated with Mouse BD Fc Block (anti‐CD16/CD32; BD Biosciences 553142; 1:10) for 10 minutes then washed with FACS buffer. Cells were then stained with Pacific Blue‐CD45 (BioLegend 103126; 1:200), PE/Cy7‐CD11b (BioLegend 101216; 1:200), BV605‐CD86 (BD Biosciences 563055; 1:100), FITC‐Cxcr4 (BD Biosciences 551967; 1:100) for 20 minutes then washed with FACS buffer. Cells were examined using a BD FACSSymphony A1 Cell Analyzer with 250,000 events collected per sample. Data were analyzed using FlowJo V10.10.0 software.

3. RESULTS

3.1. Ms4a4a deficiency in a mouse model of amyloid accumulation

A common variant located in an intergenic region near MS4A4A (rs1582763) is associated with reduced AD risk ¹⁰ , ¹¹ , ³³ and delayed age at onset, ³⁴ while a second, independent variant (rs6591561; p.M159V) is associated with increased AD risk ¹⁰ , ¹¹ , ³³ and accelerated age at onset. ³⁴ However, the impact of these variants on MS4A4A expression and function is poorly understood. To begin to understand how the AD risk gene, Ms4a4a, contributes to AD pathophysiology, gRNAs were designed to target intronic regions 5′ to exon 1 and 3′ relative to the stop codon in exon 7 of Ms4a4a (Figure 1A). Mice with one copy of Ms4a4a deleted globally (Ms4a4a ^+/−) were crossed with 5xFAD mice, ³⁵ resulting in the generation of our experimental cohort of 5xFAD Ms4a4a ^+/+ (5xFAD 4A‐WT) and 5xFAD Ms4a4a ^−/‐ (5xFAD 4A‐KO) mice (Figure 1B). qPCR of brain tissue demonstrated that Ms4a4a mRNA was absent from 6‐month‐old 5xFAD 4A‐KO mice (Figure 1C). Ms4a4a is a member of the Ms4a super family on chromosome 19 that includes Ms4a4b, Ms4a4c, Ms4a4d, Ms4a2, Ms4a6b, Ms4a6c, Ms4a6d, and Ms4a7. ³⁶ , ³⁷ To determine whether Ms4a4a deletion impacts expression of other Ms4a family members, transcript levels of Ms4a4b, Ms4a4c, Ms4a4d, Ms4a2, Ms4a6b, Ms4a6c, Ms4a6d, and Ms4a7 were measured by qPCR in brain tissue from 5xFAD 4A‐WT and 5xFAD 4A‐KO mice (Figure 1D). Ms4a family genes were expressed at similar levels in 5xFAD 4A‐WT and 5xFAD 4A‐KO mice (Figure 1D). Thus, Ms4a4a deletion does not impact the expression of other Ms4a family members, and subsequent phenotypes are driven by the loss of Ms4a4a rather than broad dysregulation of the Ms4a locus.

3.2. Ms4a4a loss decreases soluble Aβ half‐life in 5xFAD mice

Aβ, the primary component of amyloid plaques, is generated when APP undergoes a series of enzymatic cleavage events. ³⁸ , ³⁹ Aβ is primarily generated by neurons and released into the brain ISF, which can be monitored in mice using in vivo microdialysis. ²⁶ , ⁴⁰ To determine whether loss of Ms4a4a alters soluble Aβ prior to plaque accumulation, ⁴¹ in vivo microdialysis was performed in 8‐week‐old 5xFAD 4A‐KO and 5xFAD 4A‐WT mice (Figure 2A). Steady‐state Aβ levels were significantly reduced in 5xFAD 4A‐KO ISF compared to 5xFAD 4A‐WT ISF controls (Figure 2B,C; P = 4 × 10⁻⁴). To determine whether Ms4a4a loss alters the ISF Aβ elimination rate (half‐life), 5xFAD 4A‐KO and 5xFAD 4A‐WT mice were treated with a 𝛾‐secretase inhibitor (Compound E) to prevent production of new Aβ. ISF was then sampled hourly to calculate the Aβ elimination rate. ISF Aβ elimination was significantly faster in 5xFAD 4A‐KO (0.87 hours) compared to 5xFAD 4A‐WT (1.16 hours) mice (Figure 2B,D; P = 0.03). Thus, in the absence of Ms4a4a, steady‐state Aβ and Aβ half‐life are reduced prior to pathology.

FIGURE 2.

Ms4a4a deficiency decreases basal Aβ levels and Aβ half‐life in the brain ISF of 5xFAD mice. A, Schematic depicting experimental timeline for in vivo microdialysis. B, Aβ levels in ISF sampled over 12 hours. C, Quantification of basal Aβ levels in brain ISF. Welch's t test, ***, P = 4 × 10⁻⁴. D, Quantification of Aβ half‐life in brain ISF after treatment with a gamma‐secretase inhibitor. Welch's t test, *, P = 0.03. B–D, Data represent 5xFAD 4A‐WT n = 9 (4 male, 5 female); 5xFAD 4A‐KO n = 8 (6 male, 2 female); 8‐week‐old mice. E, Immunoblot of RIPA‐soluble protein isolated from 5xFAD 4A‐WT and 5xFAD 4A‐KO mouse cortices. APP (CT695) reveals full‐length APP and APP C‐terminal fragments (CTF; C99 and C83). F, Quantification of full‐length APP and CTF (C99 and C83). Two‐way ANOVA. G, Quantitative polymerase chain reaction for genes involved in APP processing reveals similar expression between 5xFAD 4A‐WT and 5xFAD 4A‐KO mice. Two‐way ANOVA. E–G, Data represents 5xFAD 4A‐WT n = 4; 5xFAD 4A‐KO n = 4; 6‐week‐old female mice. Aβ, amyloid beta; ANOVA, analysis of variance; APP, amyloid precursor protein; CTF, C‐terminal fragments; ISF, interstitial fluid; RIPA, radioimmunoprecipitation assay

To determine whether decreased ISF Aβ levels were due to altered APP processing, we measured levels of full‐length APP and its C‐terminal fragments (CTF; C99 and C83) by immunoblotting the RIPA‐soluble protein fraction (cytoplasmic and membrane proteins) from brains of 6‐week‐old mice (Figure 2E). We found that full‐length APP and CTFs remained unaltered in the absence of Ms4a4a (Figure 2F). To further assess whether APP processing was altered by Ms4a4a loss, we measured expression of genes that encode proteins involved in this process: hAPP, hPSEN1, mAdam17, mBace1, mNcstn, mAph1b, and mPen2. Expression of APP processing genes were similar with loss of Ms4a4a (Figure 2G). Thus, prior to pathology, Ms4a4a loss leads to reduced Aβ levels without altering APP processing.

3.3. Ms4a4a loss reduces plaque burden in 5xFAD mice

The concentration of Aβ in the ISF of young mice has been shown to be highly correlated with the extent of plaque burden in aged mice. ⁴² Given our findings that young 5xFAD 4A‐KO mice produced significantly less ISF Aβ than 5xFAD 4A‐WT mice, we hypothesized that 5xFAD 4A‐KO mice would display less amyloid pathology with age. To test this hypothesis, 5xFAD 4A‐WT and 5xFAD 4A‐KO mice were sacrificed at 6 months of age, and brain sections were stained with the HJ3.4 antibody to label total Aβ (Figure 3A). Plaque burden, as defined by the area of HJ3.4 coverage, was significantly reduced in the hippocampus and cortex of 5xFAD 4A‐KO mice compared to 5xFAD 4A‐WT controls (Figure 3B; P = 0.01; Figure S1 in supporting information). The average plaque size was also significantly reduced in the hippocampus of 5xFAD 4A‐KO mice (Figure 3C; P = 0.05), while the number of plaques per 1000 µm² remained similar between the two groups (Figure 3D; P = 0.08). Amyloid plaques vary in size, and smaller non‐fibrillar plaques are more toxic. ⁴³ To determine the impact of Ms4a4a loss on small plaques, plaque frequency was plotted in 349 µm² increment bins. 5xFAD 4A‐KO mice exhibited significantly fewer smaller plaques (hippocampus: 602–1301 µm²; cortex: 600–1649 µm²; Figure 3E, Figure S1). Together, we show that Ms4a4a regulates total plaque burden and plaque size.

FIGURE 3.

Loss of Ms4a4a reduces plaque burden and increases plaque compaction in 5xFAD mice. A, Representative images of immunohistochemistry for Aβ plaques (HJ3.4; scale bar = 1000 µm). B, Quantification of HJ3.4 percent area within the hippocampus. Welch's t test, *, P = 0.02. C, Quantification of average plaque size within the hippocampus (µm²). Welch's t test, *, P = 0.05. D, Quantification of average number of plaques/1000µm² within the hippocampus. Welch's t test, P = 0.08. E, Quantification of the frequency of plaques binned based on size in µm² within the hippocampus. Two‐way ANOVA, ****, P = < 0.0001. B–E, Each dot represents the mean of two brain sections from one animal. F, Representative immunofluorescence confocal images of fibrillar Aβ plaque core (X‐34; blue) and total Aβ (HJ3.4; white) in the hippocampus. Scale bar = 20 µm. G, Quantification of X‐34 proportion relative to total plaque volume. Welch's t test; *, P = 0.02. Each dot represents the mean of two brain sections from one animal. H, Immunoblot assessing expression of full‐length APP, C99, and C83 in the RIPA‐soluble protein fraction in 5xFAD 4A‐WT (white bar) and 5xFAD 4A‐KO (red bar) mice (CT‐695). I, Quantification of full‐length APP, C99, and C83 levels in RIPA‐soluble protein fraction. Two‐way ANOVA, *, P = 0.022. J, Quantification of APP processing gene expression between 5xFAD 4A‐WT (white bar) and 5xFAD 4A‐KO (red bar) mice. Two‐way ANOVA, * P = 0.034. A–H, Data are representative of 5xFAD 4A‐WT: n = 6; 5xFAD 4A‐KO: n = 7; 6‐month‐old female mice. I–J, Data are representative of 5xFAD 4A‐WT: n = 6; 5xFAD 4A‐KO: n = 6; 6‐month‐old female mice. Aβ, amyloid beta; ANOVA, analysis of variance; APP, amyloid precursor protein; RIPA, radioimmunoprecipitation assay

In addition to size, Aβ plaque composition plays a role in plaque toxicity. ⁴⁴ Aβ plaques are composed of a fibrillar core surrounded by monomeric and oligomeric Aβ. ⁴⁵ A plaque with a dense fibrillar core (i.e., more compact) has a lower affinity for soluble Aβ, limiting the formation of neurotoxic regions surrounding plaques. ⁴⁴ We examined whether there was a change in β‐sheet rich, dense fibrillar plaques using the X‐34 dye. Fibrillar plaque burden, size, and count were similar in 5xFAD 4A‐WT and 5xFAD 4A‐KO mice within the hippocampus and cortex (Figure S2 in supporting information). To assess the impact of Ms4a4a loss on plaque composition, brain sections were co‐stained with X‐34 (β‐sheet‐rich, dense core) and HJ3.4 (total Aβ; Figure 3F). Calculating the proportion of X‐34 volume relative to total plaque volume revealed that plaques were significantly more compact in the hippocampus of 5xFAD 4A‐KO mice than their 5xFAD 4A‐WT counterparts (P = 0.02; Figure 3G). In the cortex, however, X‐34‐positive plaques were similar in 5xFAD 4A‐KO and 5xFAD 4A‐WT (P = 0.12; Figure S3 in supporting information). Neuritic dystrophy, measured by BACE1 immunofluorescence, in proximity to X‐34 positive plaques (within 15 µm) was similar in the hippocampus and cortex of 5xFAD 4A‐KO and 5xFAD 4A‐WT mice (Figure S4 in supporting information). Thus, Ms4a4a regulates plaque composition in the hippocampus of 5xFAD mice.

We next sought to determine whether Ms4a4a loss alters APP processing in aged 5xFAD mice. Total full‐length APP levels and APP processing contribute to plaque pathology. ⁴⁶ Full‐length APP and CTFs were measured by immunoblotting the RIPA‐soluble protein fraction from the brains of 6‐month‐old mice (Figure 3H). Full‐length APP protein levels were similar in 5xFAD 4A‐KO and 5xFAD 4A‐WT mice (Figure 3I). We observed a modest reduction in human APP mRNA in 5xFAD 4A‐KO mice compared to 5xFAD 4A‐WT mice (Figure 3J; P = 0.056). In aged mice, Ms4a4a loss resulted in a significant decrease in CTFα (C83) levels (Figure 3I; P = 0.022), which is produced when APP is cleaved by α‐secretase along the non‐amyloidogenic pathway. CTFβ (C99) levels were similarly reduced, though to a lesser extent (Figure 3I; P = 0.064). CTFβ is generated after APP is cleaved in the amyloidogenic pathway by γ‐secretase. A reduction of both CTFs could point to more efficient APP processing or degradation. Among genes encoding proteins involved in APP processing, human PSEN1 expression, a component of γ‐secretase, was significantly reduced in 5xFAD 4A‐KO mice compared to 5xFAD 4A‐WT mice (Figure 3J; P = 0.034). However, other γ‐secretase components, mNcstn, mAph1b, and mPen2, were unchanged in the absence of Ms4a4a (Figure 3J). Similarly, α‐secretase (mAdam17) and β‐secretase (mBace1) were also unchanged (Figure 3J). Together, we find that Ms4a4a loss in aged 5xFAD mice results in less CTFs, pointing to more efficient APP processing and/or degradation.

3.4. Ms4a4a loss does not alter microglial recruitment in 5xFAD mice

Ms4a4a is enriched in myeloid lineage cells, including microglia. ⁸ , ¹⁹ , ⁴⁷ In the central nervous system (CNS), microglia are responsible for phagocytosing and degrading Aβ, clearing it from the parenchyma, and limiting amyloid accumulation. ⁴⁸ , ⁴⁹ , ⁵⁰ , ⁵¹ , ⁵² , ⁵³ Modulation of microglia‐enriched AD risk genes such as Trem2, CD33, Inpp5d, and Plcg2 have been shown to regulate microglial engulfment of Aβ plaques. ²¹ , ²² , ²³ , ²⁴ , ²⁵ Thus, we sought to determine whether Ms4a4a loss results in altered microglial recruitment and reactivity around plaques in 5xFAD mice. Brain sections from 6‐month‐old 5xFAD 4A‐WT and 5xFAD 4A‐KO mice were stained for Aβ plaques (X‐34), total microglia (Iba1), and activated microglia (CD68; Figure 4A, hippocampus; Figure S5A in supporting information, cortex). Iba1‐positive microglial recruitment within 15 µm of a X‐34‐positive plaque was similar between 5xFAD 4A‐WT and 5xFAD 4A‐KO mice (P = 0.65; Figure 4B, hippocampus; Figure S5B, cortex). CD68‐positive microglia, representing activated microglia, within 15 µm of a X‐34‐positive plaque were also similar between 5xFAD 4A‐WT and 5xFAD 4A‐KO mice (P = 0.64; Figure 4C, hippocampus; Figure S5C, cortex). To evaluate the impact of Ms4a4a loss on microglia more distal to amyloid plaques, we carried out a Scholl analysis to assess microglial volume within 15 µm increments of a X‐34‐positive plaque (Figure 4D). Iba1‐positive microglial volumes within 0 to 15 µm, 15 to 30 µm, 30 to 45 µm, or > 45 µm from X‐34‐positive plaques were similar between 5xFAD 4A‐WT and 5xFAD 4A‐KO mice (Figure 4D,E, hippocampus; Figure S5D,E, cortex). Thus, Ms4a4a does not regulate microglial recruitment to plaques.

FIGURE 4.

Evaluating the impact of Ms4a4a loss on microgliosis in 5xFAD mice. A, Representative confocal images of Aβ plaques (X‐34; blue), total microglia (Iba1; red), and a reactive microglia marker (CD68; white; scale bar = 20 µm) in the hippocampus. B, Quantification of Iba1‐positive microglial volume within 15 µm of plaques in 5xFAD 4A‐WT and 5xFAD 4A‐KO mice in the hippocampus. Welch's t test; P = 0.30. C, Quantification of CD68 volume within 15 µm of plaques in 5xFAD 4A‐WT and 5xFAD 4A‐KO mice. Welch's t test; P = 0.12. D, Representative confocal image of Aβ plaques (X‐34; blue) and microglia (Iba1; red; scale bar = 20 µm) with Imaris microglial distance mask (blue: 0–15 µm; purple: 15–30 µm; green: 30–45 µm; red: 45 µm +) in the hippocampus. E, Quantification of microglial volume within 15 µm increments from plaques in the hippocampus. Two‐way ANOVA. Data are representative of 5xFAD 4A‐WT: n = 6; 5xFAD 4A‐KO: n = 7. F, Diagram. Mice were retro‐orbitally injected with an APC/Cy7 anti‐CD45 antibody to label circulating immune cells, then brains were extracted and homogenized. Flow cytometry was performed. Microglia were gated and then analyzed for CD86 and Cxcr4 expression. G, Representative histogram and quantification of CD86 frequency of expression in microglia from 5xFAD 4A‐WT (gray) and 5xFAD 4A‐KO (red) mice. Welch's t test, *, P = 0.0108. H, Representative histogram and quantification of Cxcr4^hi frequency of expression in microglia from 5xFAD 4A‐WT (gray) and 5xFAD 4A‐KO (red) mice. Welch's t test, *, P = 0.049. Data are representative of 5xFAD 4A‐WT: n = 8; 5xFAD 4A‐KO: n = 5; 6‐month‐old female mice. Aβ, amyloid beta; ANOVA, analysis of variance

To determine whether Ms4a4a loss impacts overall microgliosis or astrogliosis, we measured Iba1 and GFAP across the brain. Microgliosis, measured as the percent area of Iba1 staining in the hippocampus and cortex, was similar in 5xFAD 4A‐WT and 5xFAD 4A‐KO mice (Figure S6A–E in supporting information). Total Iba1 intensity, a proxy for microglial activation, was also similar in the hippocampus and cortex of 5xFAD 4A‐WT and 5xFAD 4A‐KO mice (Figure S6A–E). Similarly, the degree of astrogliosis, measured as the percent area and intensity of GFAP in the hippocampus or cortex, was similar in 5xFAD 4A‐WT and 5xFAD 4A‐KO mice (Figure S6F–J). Thus, Ms4a4a does not alter gliosis.

Activation and recruitment to plaques represent only a subset of microglia functions. To determine whether Ms4a4a loss alters the inflammatory state of microglia, we measured CD86 and Cxcr4 expression in microglia from the whole brain of 6‐month‐old 5xFAD 4A‐WT and 5xFAD 4A‐KO mice (Figure 4F). Briefly, an APC/Cy7‐CD45 antibody was retro‐orbitally injected in 5xFAD 4A‐WT and 5xFAD 4A‐KO mice to label any circulating immune cells in the brain. Brains were then homogenized and FACS analyses were performed. APC/Cy7‐CD45‐positive circulating cells were excluded, followed by the identification of brain‐resident cells expressing an intermediate level of CD45 (termed: CD45‐mid). CD45‐mid cells are also CD11b positive, suggesting these cells are microglia (Figure 4F). The frequency of CD45‐mid/CD11b‐positive microglia expressing CD86, a pro‐inflammatory marker, was significantly increased with loss of Ms4a4a (Figure 4G; P = 0.01). We also found that in the absence of Ms4a4a, there were more microglia expressing high levels of Cxcr4, a chemokine receptor involved in inflammation (Figure 4H; P = 0.05). Thus, Ms4a4a deficiency impacts the inflammatory state of microglia in 5xFAD mice.

3.5. Loss of Ms4a4a increases MMP‐9 secretion

Aβ clearance occurs by several mechanisms. Beyond uptake and degradation by activated astrocytes and microglia, soluble and fibrillar forms of Aβ can be enzymatically degraded by extracellular proteases. ⁵⁴ A number of proteolytic enzymes have been shown to degrade Aβ, including IDE, ⁵⁵ , ⁵⁶ , ⁵⁷ NEP, ⁵⁸ , ⁵⁹ MMP‐9, ⁶⁰ , ⁶¹ , ⁶² and cathepsin B. ⁶³ , ⁶⁴ IDE is enriched in astrocytes and only degrades monomeric Aβ, while NEP is expressed in neurons where it degrades Aβ monomers and small oligomers. ⁵⁶ , ⁶⁰ , ⁶³ , ⁶⁵ , ⁶⁶ , ⁶⁷ , ⁶⁸ MMP‐9 and cathepsin B are enriched in microglia and digest monomers, oligomers, and fibrils. ⁵⁷ , ⁶⁰ , ⁶³ , ⁶⁶ , ⁶⁷ , ⁶⁸

To determine whether Ms4a4a loss impacts Aβ‐degrading enzymes in 5xFAD mice, we measured levels of Aβ‐degrading enzymes by immunoblotting brain homogenates from 6‐month‐old mice (Figure 5). MMP‐9 protein levels were significantly increased in the RAB‐soluble fraction (enriched for secreted proteins) of 5xFAD 4A‐KO brains compared to 5xFAD 4A‐WT mice (P = 2.7 × 10⁻³; Figure 5A,B). MMP‐9 protein levels were also significantly increased in RAB‐soluble protein fractions of 6‐week‐old 5xFAD 4A‐KO mice, prior to plaque accumulation (Figure S7 in supporting information; P = 0.015). IDE, NEP, and cathepsin B protein levels, however, were similar in 6‐month‐old 5xFAD 4A‐KO and 5xFAD 4A‐WT mice (P = 0.87, P = 0.14, P = 0.80, respectively; Figure 5A,B). We next sought to determine whether Aβ‐degrading enzymes were altered in the RIPA‐soluble fraction, which contains membrane‐bound proteins. While MMP‐9, IDE, and NEP levels remain similar (P = 0.98, P = 0.43, P = 0.87, respectively), cathepsin B levels were significantly reduced in 5xFAD 4A‐KO mice compared to 5xFAD 4A‐WT mice (Figure 5C,D; P = 0.02, respectively). The mRNA expression of these enzymes was similar in 5xFAD 4A‐KO and 5xFAD 4A‐WT mice, suggesting that Ms4a4a acts on Aβ‐degrading enzymes at the protein level (Figure 5E). Thus, Ms4a4a alters microglia‐enriched Aβ‐degrading enzymes MMP‐9 and cathepsin B, enzymes that degrade Aβ oligomers and fibrils.

FIGURE 5.

MMP‐9 secretion is increased in 5xFAD mice lacking Ms4a4a. A, Immunoblot assessing expression of Aβ‐degrading enzymes in RAB‐soluble protein fraction in 5xFAD 4A‐WT and 5xFAD 4A‐KO mice. B, Quantification of Aβ‐degrading enzyme levels in RAB‐soluble protein fraction. Two‐way ANOVA; **, P = 2.7 × 10⁻³). C, Immunoblot of Aβ‐degrading enzymes in the RIPA‐soluble protein fraction in 5xFAD 4A‐WT and 5xFAD 4A‐KO mice. D, Quantification of Aβ‐degrading enzyme levels in RIPA‐soluble protein fraction. Two‐way ANOVA; *, P = 0.02. E, Quantification of amyloid‐degrading enzyme gene expression reveals similar expression between 5xFAD 4A‐WT and 5xFAD 4A‐KO mice (two‐way ANOVA). Data are representative of 5xFAD 4A‐WT: n = 6; 5xFAD 4A‐KO: n = 7; 6‐month‐old female mice. Aβ, amyloid beta; ANOVA, analysis of variance; MMP‐9, matrix metalloproteinase 9; RAB, reassembly buffer; RIPA, radioimmunoprecipitation assay

3.6. AD resilience allele in MS4A4A is associated with increased CSF MMP‐9 levels

Together, our findings that Ms4a4a loss reduces plaque burden and promotes formation of more compact plaques, point to a protective phenotype in 5xFAD mice. The minor allele (A; protective) of a common variant near the MS4A4A locus (rs1582763) has been associated with reduced risk for AD. ¹¹ , ⁶⁹ To determine whether minor allele carriers of rs1582763 mimic features of the protective phenotypes in the mouse model, we analyzed a cohort of 3506 subjects with genetic and CSF proteomic data. ³¹ We found that minor allele (A; protective) carriers had significantly higher levels of CSF MMP‐9 than major allele (G) carriers (Figure 6A). Interestingly, there is an independent common variant in MS4A4A (rs6591561) that confers increased risk for AD. ¹¹ , ³⁴ Minor allele carriers (G; risk) of the AD risk variant in MS4A4A exhibit significantly lower levels of CSF MMP‐9 (Figure 6B). Thus, MS4A4A may contribute to AD resilience via modulation of the Aβ‐degrading enzyme MMP‐9.

FIGURE 6.

MS4A4A variants associated with altered MMP‐9 levels in human CSF. A, Violin plot of MMP‐9 levels based on rs1582763 genotype. A, minor allele. Minor allele is associated with reduced AD risk. B, Violin plot of MMP‐9 levels based on rs6591561 genotype. G, minor allele. Minor allele is associated with increased AD risk. P values were obtained using a two‐way Wilcoxon rank‐sum test. AD, Alzheimer's disease; CSF, cerebrospinal fluid; MMP‐9, matrix metalloproteinase 9

4. DISCUSSION

Common variants in or near MS4A4A, a gene implicated in peripheral immune response, are associated with AD risk; however, the role of MS4A4A in AD pathogenesis and microglial biology remains poorly understood. ⁵ , ⁸ , ¹⁰ Our study shows that loss of Ms4a4a improves pathology in a mouse model of amyloid accumulation. Ms4a4a deficiency results in lower basal Aβ levels and a shorter Aβ half‐life in the ISF of young mice. Prior to amyloid accumulation, Ms4a4a deficiency does not impact APP processing but alters MMP‐9 levels, which contribute to Aβ degradation. This, in turn, contributes to fewer small plaques and increased plaque compaction in aged mice. With age and plaque burden, Ms4a4a loss is associated with reduced APP CTFs, which could point to a role for Ms4a4a in APP processing and/or clearance of CTFs. Ms4a4a loss enhances the inflammatory state of microglia without altering recruitment to plaques or overall gliosis. Ms4a4a also increases the microglia‐enriched, Aβ‐fibril‐degrading enzyme MMP‐9 and decreases cathepsin B in 5xFAD 4A‐KO mice. Consistent with our discovery that Ms4a4a loss is protective in the amyloid phase, carriers of a variant that reduces AD risk near MS4A4A, rs1582763, have elevated CSF MMP‐9 levels, while carriers of an independent variant in MS4A4A that increases AD risk, rs6591561, have reduced CSF MMP‐9 levels. Together, these results suggest that MS4A4A plays a role in mediating amyloid pathobiology by enhancing Aβ clearance.

Plaque accumulation in aged mice corresponds with Aβ concentration in younger mice. ⁴² We show that Ms4a4a is a regulator of ISF Aβ levels and Aβ half‐life in a mouse model of plaque pathology. Loss of Ms4a4a in 5xFAD mice results in a decrease in ISF Aβ basal levels and half‐life. The accumulation of aggregation‐prone, clearance‐resistant fibrillar Aβ is believed to drive plaque formation and AD, and decreased plaque burden is believed to be protective in disease pathogenesis. ³⁹ , ⁷⁰ In line with this, we show that Ms4a4a deficiency results in decreased plaque burden and reduced average plaque size in 6‐month‐old mice. The reduction in plaque burden is driven by changes in the proportion of small plaques observed in the brains of Ms4a4a‐deficient 5xFAD mice. This finding is consistent with a number of other microglia‐enriched AD risk and resilience genes, including CD33 and Inpp5d. ²¹ , ²⁵ The observed effect of reduced plaque size as well as fewer smaller plaques points to potential mechanisms including altered plaque aggregation and compaction, differential clearance mechanisms of plaque types, or microglia phagocytosis and degradation of plaques. Anti‐Aβ immunotherapies have been shown to reduce plaque load with a preferential impact on small, diffuse plaques, producing a similar effect to what we observed with Ms4a4a loss. ⁷¹ Thus, our findings suggest that loss of Ms4a4a is protective against amyloid accumulation. We also find that plaques in the hippocampus of Ms4a4a‐deficient 5xFAD mice are more “compact” with a dense, fibrillar core surrounded by less non‐fibrillar Aβ. Non‐fibrillar Aβ is thought to be more toxic, as it has greater potential to seed new plaques and serves as a site of increased neuritic dystrophy. ⁴⁴ , ⁷² Thus, our mouse model of Ms4a4a deficiency in the context of plaque pathology appears to capture aspects of resilience in human subjects carrying a common variant near MS4A4A (rs1582763). ¹¹ , ¹⁹ , ²⁰

Slowing of Aβ clearance contributes to late‐onset AD. ⁷³ Thus, increased clearance of Aβ can result in decreased plaque accumulation, thereby decreasing AD pathophysiology. Microglia, the CNS cell type enriched in Ms4a4a expression, ¹³ play an elemental role in phagocytosing and degrading Aβ. ⁴⁸ , ⁴⁹ , ⁵⁰ , ⁵¹ , ⁵² , ⁵³ Loss of Ms4a4a fails to alter microglial recruitment in proximity to plaques, as has been described for other AD risk genes like TREM2. ²² Instead, we discovered that expression of the pro‐inflammatory markers CD86 and Cxcr4 is increased in Ms4a4a‐deficient microglia in 5xFAD mice. These observations are in line with previous findings which showed that stimulation of peripheral macrophages with interleukin 4 promotes an MS4A4A‐mediated anti‐inflammatory, tissue‐repair response and that MS4A4A promotes M2 (anti‐inflammatory) macrophage polarization in the context of tumor growth. ¹⁷ , ⁷⁴ Together, these findings suggest that loss of  Ms4a4a function results in a more pro‐inflammatory polarization of microglia, which promotes plaque clearance.

Aβ clearance is also regulated by Aβ‐degrading enzymes. We found elevated levels of MMP‐9, a microglia‐enriched, ⁶⁸ zinc‐dependent metalloprotease, in Ms4a4a‐deficient 5xFAD mice. MMP‐9 has been shown to degrade not only soluble, but also fibrillar Aβ in vitro and in situ, with increased activity around compact plaques, generating non‐toxic, non‐fibrillogenic protein fragments. ⁶⁰ , ⁶¹ , ⁶² , ⁶³ Overexpression of MMP‐9 in 5xFAD mice was found to decrease oligomeric Aβ levels. ⁷⁵ We propose that the increase of MMP‐9 in 5xFAD Ms4a4a‐KO mice contributes to the reduced plaque pathology observed, consistent with the role of MMP‐9 in the clearance of oligomeric and fibrillar Aβ. ⁵ , ³⁹ Additional studies will be required to understand the impact of Ms4a4a loss on MMP‐9 activity. We discovered that human carriers of a protective variant near MS4A4A (rs1582763; A) have significantly elevated levels of CSF MMP‐9. Rs1582763 is associated with reduced AD risk and delayed age‐at‐onset, thus offering protection against AD. ⁵ , ¹⁰ , ¹¹ , ³³ Carriers of an MS4A4A variant associated with increased AD risk (rs6591561) ⁵ , ¹⁰ , ¹¹ , ³³ exhibited the opposite effect, whereby CSF MMP‐9 levels were significantly reduced in minor allele carriers (G). MMP‐9 may have utility as a biomarker of engagement of MS4A4A‐driven pathways in future therapeutic studies. Together, we demonstrate that the protective effects of Ms4a4a loss in the 5xFAD mouse model are replicated in individuals with an AD protective allele.

Studies in models of amyloid accumulation deficient in microglia‐enriched AD risk genes have shown a robust impact on microglial function and reactivity. Microglia are the primary cell type in the brain that express Ms4a4a. ¹³ Loss of Trem2 in mouse models of amyloid accumulation (5xFAD and APP/PS1) results in exacerbated plaque pathology, more diffuse plaques, increased neuritic dystrophy, and reduced microglial recruitment toward plaques. ²² , ²³ , ⁷⁶ , ⁷⁷ CD33 inhibits phagocytosis of Aβ42 in vitro, and CD33 deficiency in a mouse model of amyloid accumulation results in reduced plaque burden and increased microglial recruitment to plaques. ²¹ In Ms4a4a‐deficient 5xFAD mice, we identify additional mechanisms by which microglia function is altered in a manner that modifies amyloid pathology. Ms4a4a loss promotes microglia‐mediated clearance of Aβ prior to amyloid deposition and APP‐CTFs after plaques accumulate likely via MMP‐9. Additionally, loss of Ms4a4a in 5xFAD mice cause a global shift toward a pro‐inflammatory microglial state. Yet, Ms4a4a loss does not alter microglial recruitment or reactivity to plaques as is found with the loss of other microglia‐enriched AD risk genes. Placing our findings in the context of the growing literature of functional characterization of AD risk genes suggests there are multiple pathways by which AD risk genes alter microglia behavior in ways that contribute to disease. Thus, there are multiple pathways to target therapeutically.

Common variants associated with AD risk occur outside of (rs1582763) or within (rs6591561; p.M159V) MS4A4A. However, the impact of these variants on MS4A4A expression remains unresolved. Our discovery that Ms4a4a loss is associated with protective phenotypes in an animal model may begin to point to mechanisms of action of these genetic variants. This includes pursuing therapeutic strategies that lower MS4A4A levels such as small molecules, antibodies, or antisense oligonucleotides. However, additional work is required to resolve these questions.

5. CONCLUSIONS

We discovered that loss of Ms4a4a in the 5xFAD background results in decreased plaque pathology and reflects features of resilience against AD pathogenesis. This work highlights the therapeutic potential of targeting MS4A4A, and perhaps modulating the expression of MMP‐9, to promote plaque clearance and mitigate not only plaque formation, but also the development of subsequent AD pathologies.

AUTHOR CONTRIBUTIONS

Designed experiments: Emma P. Danhash, Celeste M. Karch. Performed and analyzed experiments: Emma P. Danhash, Anthony C. Verbeck, Daniel Western, Andrea S. Díaz‐Pacheco, Grant Galasso, Shih‐Feng You, Collin J. Nadarajah, Savannah Tiemann Powles, Guangming Huang, Emma Starr, Nadia Miller, Erik S. Musiek, Jasmin Herz, Abhirami K. Iyer, John Cirrito, Carlos Cruchaga, Celeste M. Karch. Provided funding: Celeste M. Karch. Wrote the manuscript: Emma P. Danhash, Celeste M. Karch. Revised and approved manuscript: Emma P. Danhash, Anthony C. Verbeck, Daniel Western, Andrea S. Díaz‐Pacheco, Grant Galasso, Shih‐Feng You, Collin J. Nadarajah, Savannah Tiemann Powles, Guangming Huang, Emma Starr, Nadia Miller, Erik S. Musiek, Jasmin Herz, Abhirami K. Iyer, John Cirrito, Carlos Cruchaga, Celeste M. Karch.

CONFLICT OF INTEREST STATEMENT

The authors declare that they have no competing interests. Author disclosures are available in the supporting information.

CONSENT STATEMENT

All participants provided informed consent for their data to be used in this study. The study was approved by the institutional review board of Washington University School of Medicine in St. Louis.

Supporting information

SUPPORTING INFORMATION

SUPPORTING INFORMATION

SUPPORTING INFORMATION

SUPPORTING INFORMATION

SUPPORTING INFORMATION

SUPPORTING INFORMATION

SUPPORTING INFORMATION

SUPPORTING INFORMATION

SUPPORTING INFORMATION

SUPPORTING INFORMATION

Danhash EP, Verbeck AC, Western D, et al. Ms4a4a deficiency ameliorates plaque pathology in a mouse model of amyloid accumulation. Alzheimer's Dement. 2025;21:e70580. 10.1002/alz.70580