A Trem2R47H mouse model without cryptic splicing drives age- and disease-dependent tissue damage and synaptic loss in response to plaques

All experiments involving mice were approved by the UC Irvine Institutional Animal Care and Use Committee and were conducted in compliance with all relevant ethical regulations for animal testing and research. All experiments involving mice comply with the Animal Research: Reporting of in Vivo Experiments (ARRIVE) guidelines.

To generate Trem2^R47H mice (B6(SJL)-Trem2^em1Aduci/J, Jackson Laboratory stock number #034,036), Alt-R Crispr RNA (TMF1342 – gaagcactgggggagacgca) and tracrRNA plus CAS9 protein (HiFi Cas9 nuclease V3, Integrated DNA Technologies (IDT), Coralville, IA) as a ribonucleoprotein (RNP) was microinjected into C57BL/6 J zygotes (Jackson Lab Stock # 000,664) along with a ssODN sequence (TMF 1341–5’-CAAGCCCTCAACACCACGGTGCTGCAGGGCATGGCCGGCCAGTCGTTAAGGGTATCCTGCACTTATGACGCGTTGAAACATTGGGGCAGACATAAGGCCTGGTGTCGGCAGCTGGGTGAGGAGGGCCCATGCCAGCGTGTGGT—3’) to introduce the R47H missense mutation. G0 founder animals containing the desired DNA sequence changes were backcrossed with C57BL/6 J mice and N1 heterozygous mice were sequenced to determine the mutant allele. N1 heterozygous mice were backcrossed to produce N3F1 heterozygotes, which were used to generate animals for subsequent analysis. 5xFAD hemizygous (B6.CgTg(APPSwFlLon,PSEN1*M146L*L286V)6799Vas/Mmjax, Jackson Lab Stock # 34,848, MMRRC) and non-transgenic littermates were produced by natural mating or IVF procedures with C57BL/6 J females. After weaning, they were housed together with littermates and aged until the harvest dates. All animals were bred by the Transgenic Mouse Facility at UCI.

Genomic DNA was extracted from mouse tail biopsies using DirectPCR Lysis Reagent (Viagen Biotech, Los Angeles, CA) and Proteinase K (Roche, Indianapolis, IN). Amplification was performed using a Bio-Rad CFX-96 instrument. For each amplicon, a single PCR product was confirmed by capillary electrophoresis (Fragment Analyzer, AATI / Agilent, Santa Clara, CA) then subjected to Sanger sequencing (Retrogen, San Diego, CA) and analyzed using SeqMan Pro 17.2 (DNASTAR, Madison, WI). Primers for PCR amplification and sequencing for off-target analysis are listed in Supplemental Table). 1

Oligonucleotides for PCR-based genotyping were purchased from IDT. Trem2^R47HNSS genotyping was performed using a common primer set to amplify both Trem2 wildtype allele and Trem2^R47HNSS allele (For 5′-TCAACACCACGGTGCT -3′ and Rev 5′-TGTGTGCTCACCACACG -3′). Two fluorophore labeled-hydrolysis probes which hybridized specific to mouse Trem2 wildtype amplicon (5’-TGCGTCTCCCCCAGTGCTTCAA-3’ + HEX) and Trem2^{R47H NSS} mutation (5’-TATGTCTGCCCCAATGTTTCAACGCG-3’-FAM) were used to detect the allelic ratio in the amplicon. The relative fluorescence from each probe was quantified at the end point of PCR cycles to call the genotype using Allelic Discrimination function using Bio-Rad CFX Maestro software (Bio-Rad, Hercules, CA). For 5xFAD genotyping, a hydrolysis probe which hybridizes to APP(Swe) mutation amplicon was used (For 5′-TGGGTTCAAACAAAGGTGCAA-3′ and Rev 5′-GATGACGATCACTGTCGCTATGAC-3′: APP(Swe) probe 5′-CATTGGACTCATGGTGGGCGGTG-3′.) to detect transgenes. We used endogenous ApoB allele (For 5′-CACGTGGGCTCCAGCATT-3′ and Rev 5′-TCACCAGTCATTTCTGCCTTTG-3′: ApoB probe 5′-CCAATGGTCGGGCACTGCTCAA-3′) to normalize the Ct values.

Eight-week-old C57BL/6 J, Trem2^em1Aduci (i.e. Trem2^{R47H NSS}), Trem2^em1Adiuj (i.e. Trem2^{R47H CSS}), and Trem2^em2Adiuj (i.e. Trem2-KO) mice (Jackson Laboratory stock number: 000664, 034,036, 027,918, 027,197 respectively) were used. Each mouse model consisted of 2 groups of 4 male mice. Control mice were supplied standard chow for 6 weeks while the CPZ groups were fed 0.3% cuprizone chow also for 6 weeks (Envigo, Indianapolis, IN). Mice within the same experiment group (i.e., same genotype and diet) were housed together for the duration of feeding. Weights of individual mouse and chow consumptions of each cage were recorded, and chow was changed every 3 or 4 days to monitor expected weight loss as well as ensuring freshness of cuprizone chow. Brains were collected and fixed in 4% paraformaldehyde for 24 h followed by cryoprotection by immersion in 5% sucrose for 24 h then 30% sucrose for 5 days, all at 4 °C.

Mice were euthanized at 4 and 12 months of age via CO₂ inhalation and transcardially perfused with 1X phosphate buffered saline (PBS). For all studies, brains were removed, and hemispheres separated along the midline. Brain halves were either flash frozen for subsequent biochemical analysis or drop-fixed in 4% paraformaldehyde (PFA (Thermo Fisher Scientific, Waltham, MA)) for immunohistochemical analysis. Fixed half brains were sliced at 40 μm using a Leica SM2000R freezing microtome. All brain hemispheres were processed and coronal brain slices (between -2.78 mm posterior and –3.38 mm posterior to Bregma according to the Allen Mouse Brain Atlas, Reference Atlas version 1, 2008) imaged with a Zeiss Axio Scan Z1 Slidescanner using a 10X 0.45 NA Plan-Apo objective. Images were also acquired using a 20X objective via a Leica TCS SPE-II confocal microscope and quantified using Bitplane Imaris Software.

One representative brain slice from each mouse of the same experimental group (i.e. same genotype, age, and sex) was stained in the same well. Free-floating sections were washed three times with 1X PBS (1 × 10 min and 2 × 5 min) and for Thioflavin-S (Thio-S) staining, 10 min incubation in 0.5% Thio-S (1892; Sigma-Aldrich) diluted in 50% ethanol. Sections were then washed 2X for 5 min each in 50% ethanol and one 10-min wash in 1xPBS. For Amylo-Glo staining, following the PBS washes, free-floating brain slices were washed in 70% ethanol for 5 min and rinsed in deionized water for 2 min before being immersed for 10 min in Amylo-Glo RTD Amyloid Plaque Staining Reagent (1:100; TR-200-AG; Biosensis, Thebarton, South Australia) diluted in 0.9% saline solution. Afterwards, sections were washed in 0.9% saline solution for 5 min then rinsed in deionized water for 15 s before proceeding with a standard indirect immunohistochemical protocol. From the incubation period for both Thio-S and Amylo-Glo onwards, sections were kept under foil or in the dark. Sections were immersed in normal blocking serum solution (5% normal goat serum with 0.2% Triton X-100 in 1X PBS) for 1 h before overnight incubation at 4 °C in primary antibodies diluted in normal blocking serum solution.

Brain sections were stained following a standard indirect technique as described [37, 38] with the following primary antibodies against: ionized calcium-binding adapter molecule 1 (IBA1; 1:2000; 019–19,741; Wako, Osaka, Japan), Aβ_1-16 (6E10; 1:2000; 8,030,001; BioLegend, San Diego, CA), glial fibrillary acidic protein (GFAP; 1:1000; AB134436↗; Abcam, Cambridge, MA), S100β (1:200; AB41548; Abcam), lysosome-associated membrane protein 1 (LAMP1; 1:200; AB25245, Abcam), neurofilament light chain (NfL; 1:200; 171 002; Synaptic Systems, Germany), CD74 (1:500; 151,002; BioLegend), CD11c (1:100.; 50–112-2633; eBioscience), myelin basic protein (MBP; 1:200; MAB386; (Millipore Sigma, Billerica, MA), Synaptophysin (1:1000; S5768; Sigma-Aldrich), PSD-95 (1:500; ab18258; Abcam), Bassoon (BSN; 1:250; AB82958; Abcam), OC (1:1000; AB2286; Sigma-Aldrich). Brain sections were mounted on microscope slides, dried overnight and desiccated for one hour prior to standard Luxol Fast Blue (LFB) staining protocols [39].

High-resolution fluorescence images were obtained using a Leica TCS SPE-II confocal microscope and LAS-X software. For confocal imaging, one field of view (FOV) per brain region was captured per mouse using the Allen Brain Atlas to capture comparable brain regions.

For super-resolution imaging, images of the CA1 stratum radiatum (CA1-SR) hippocampal region of WT, Trem2^{R47H NSS}, 5xFAD and 5xFAD/Trem2^{R47H NSS} animals at 4 and 12 months of age were acquired with identical conditions. Super-Resolution Lattice Structured Illumination Microscopy (Lattice-SIM) was performed using an Elyra 7 microscope system (Carl Zeiss, White Plains, NY). Samples were imaged using a 63 × 1.4NA Plan-Apo objective lens and Immersol 518F (23°C) immersion oil. Images were collected as z-stacks (110 nm step interval, within a depth of 5–8 µm, covering an area of 64 × 64 µm) and for each focal plane, 9 phase images were acquired. Images were then processed using ZEN SIM² (weak fixed option selected) on the ZEN (black edition) software. Two images per mouse/genotype/age/sex were acquired.

Preparation of samples and quantification of Aβ was performed as described [37, 38]. Micro-dissected hippocampal and cortical regions of each mouse were flash-frozen and processed for biochemical analysis. Samples were pulverized using a Bessman Tissue Pulverizer. Pulverized hippocampal tissue separated for biochemical analysis was homogenized in 150µL of Tissue Protein Extraction Reagent (TPER; Life Technologies, Grand Island, NY), while cortical tissue was homogenized in 1000µL/150 mg of TPER. This composition of TPER includes 25 mM bicine and 150 mM sodium chloride (pH 7.6) to efficiently solubilize proteins within brain tissue following homogenization. Together with protease (Roche) and phosphatase inhibitors (Sigma-Aldrich), the homogenized samples were centrifuged at 100,000 g for 1 h at 4 °C to generate TPER-soluble fractions. For formic acid-fractions, pellets from TPER-soluble fractions were homogenized in 70% formic acid: 75µL for hippocampal tissue or half of used TPER volume for cortical tissue. Afterwards, samples were centrifuged again at 100,000 g for 1 h at 4 °C. Protein in the insoluble fraction of micro-dissected hippocampal and cortical tissue were normalized to its respective brain region weight, while protein in soluble fractions were normalized to the protein concentration determined via Bradford Protein Assay. Formic acid neutralization buffer was used to adjust pH prior to running ELISAs.

Quantitative biochemical analyses of human Aβ soluble and insoluble fraction levels were acquired using the V-PLEX Aβ Peptide Panel 1 (6E10) (K15200G-1; Meso Scale Discovery, Rockville, MD). Quantitative biochemical analysis of neurofilament-light chain (NfL) in plasma was performed using the R-Plex Human Neurofilament L Assay (K1517XR-2; Meso Scale Discovery).

Confocal images of each brain region were quantified automatically using the spots module within Imaris v9.7 then normalized to the area of the field-of-view (FOV). Amyloid burden was assessed by measuring both the total Thio-S⁺ plaque number normalized to FOV area and their volume via the surfaces module in Imaris. Similarly, volumetric measurements (i.e., Thio-S⁺ plaque volume, IBA1⁺ microglia volume, etc.) were acquired automatically utilizing the surfaces module with confocal images of each brain region. Quantitative comparisons between experimental groups were carried out in sections stained simultaneously.

For synaptic quantification, the total number of Bassoon or PSD95 puncta was quantified using the spots function on Imaris. Co-localization of pre- and post-synaptic puncta (Bassoon-PSD95) was determined using the spots function on Imaris and Matlab software to determine the total number of colocalized spots (defined as ≤ 200 nm distance). Results were normalized to the total volume of each image, to correct for any difference in the depth of imaging.

Hippocampal slices were prepared from WT, homozygous Trem2^{R47H NSS}, hemizygous 5xFAD, 5xFAD/ Trem2^{R47H NSS} mice (1–2 slice recordings/mouse for 5–6 mice/sex/genotype totaling to n = 8–12 slices/sex/genotype) at 4 and 12 months of age. Hippocampal slice preparation and long-term potentiation (LTP) recording was performed as described [37, 38]. Following isoflurane anesthesia, mice were decapitated, and the brain was quickly removed and submerged in ice-cold, oxygenated dissection medium containing (in mM): 124 NaCl, 3 KCl, 1.25 KH₂PO₄, 5 MgSO₄, 0 CaCl₂, 26 NaHCO₃, and 10 glucose. Coronal hippocampal slices (340 µm) were prepared using a Leica VT1000S vibrating tissue slicer before being transferred to an Interface recording containing preheated artificial cerebrospinal fluid (aCSF) of the following composition (in mM): 124 NaCl, 3 KCl, 1.25 KH₂PO₄, 1.5 MgSO4, 2.5 CaCl₂, 26 NaHCO₃, and 10 glucose and maintained at 31 ± 1 °C. Slices were continuously perfused with this solution at a rate of 1.75–2.00 ml/min while the surface of the slices were exposed to warm, humidified 95% O₂ / 5% CO₂. Recordings began following at least 2 h of incubation. Field excitatory postsynaptic potentials (fEPSPs) were recorded from CA1b striatum radiatum using a single glass pipette filled with 2 M NaCl (2–3 MΩ) in response to orthodromic stimulation (twisted nichrome wire, 65 µm diameter) of Schafer collateral-commissural projections in CA1 striatum radiatum. Pulses were administered at 0.05 Hz using a current that elicited a 50% maximal response. Paired-pulse facilitation was measured at 40, 100, and 200 s intervals prior to setting baseline. After establishing a 20-min stable baseline, the orthodromic stimulated pathway was used to induce LTP by delivering 5 ‘theta’ bursts, with each burst consisting of four pulses at 100 Hz and the bursts themselves separated by 200 ms (i.e., theta burst stimulation (TBS)). The stimulation intensity was not increased during TBS. Data were collected and digitized by NAC 2.0 Neurodata Acquisition System (Theta Burst Corp., Irvine, CA).

RNA sequencing was performed as described [37]. Frozen tissues were lysed and homogenized using TissueLyser II (Qiagen, Germantown, MD). Total RNAs were extracted using RNeasy Mini Kit (Qiagen) and RNase-Free DNase Set (Qiagen) on a QIAcube (Qiagen) liquid handling platform. RNA integrity number (RIN) was measured by Qubit RNA IQ Assay (Life Technologies) and samples with RIN > = 7.0 were kept for cDNA synthesis. cDNA synthesis and amplification were performed followed by Smart-seq2 [40] standard protocol.

Libraries were constructed using the DNA Prep Kit (Illumina, San Diego, CA) on an epMotion 5070 TMX (Eppendorf, Enfield, CT) automated pipetting system. The 4 months libraries were constructed manually and the 12 months libraries by the epMotion 5070 TMX (Eppendorf) automated pipetting system using the DNA Prep Kit (Illumina).

Libraries were base-pair selected based on Agilent 2100 Bioanalyzer profiles and normalized determined by KAPA Library Quantification Kit (Roche). The libraries were built from 3 to 5 different mice per genotype, sex and tissue (hippocampus) across 2 different timepoints (4 and 12 months). For the cuprizone experiment, whole brains were used for 4 males per genotype per condition (control or CPZ) at 3 months of age. The 4-month-old mouse libraries were sequenced using paired-end 43 bp mode on Illumina NextSeq500 platform with > 14 million reads per sample. The 12-month-old mouse libraries were sequenced using paired-end 101 bp mode on Illumina NextSeq2000 platform with > 28 million reads per sample. Sequences were aligned to the mouse genome (mm10) and annotation was done using GENCODE v21. Reads were mapped with STAR v.2.7.3a and RSEM (v.1.3.3) was used for quantification of gene expression.

Nanopore libraries were built from three mice per genotype for C57BL/6 J, Trem2^{R47H CSS} homozygotes and Trem2^{R47H NSS} homozygotes. Each library was constructed from 200 fmol Smartseq2 cDNA using Ligation Sequencing Kit (SQK-LSK114, Oxford Nanopore Technology, (ONT) Lexington, MA) and NEBNext® Companion Module for ONT Ligation Sequencing (E7180S). The Short Fragment Buffer (SFB) from the Ligation kit was used during the wash steps. Each 10 fmol library was loaded on one R10.4.1 flow cell (FLO-MIN114, ONT) and sequenced using a GridIONx5 Mk1B (ONT) with live base calling using Guppy (v6.2.7) run in super accurate mode.

Every reported n represents the number of independent biological replicates. The sample sizes are similar with those found in prior studies conducted by MODEL-AD and were not predetermined using statistical methods [37, 38]. Electrophysiology, immunohistochemical, and biochemical data were analyzed using Student’s t-test or two-way ANOVA via Prism v.9 (GraphPad, La Jolla, CA). Tukey’s post-hoc tests were utilized to examine biologically relevant interactions from the two-way ANOVA. Where sex-differences are apparent, a Student’s t-test was used within genotype group. Outlier tests were performed via Prism v.9 where relevant and any datapoints removed from the analyses acknowledged in the relevant figure legend. *p ≤ 0.05, ** p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001. Statistical trends were accepted at p < 0.10 (^#). Data are presented as raw means and standard error of the mean (SEM).

Results of previous studies of mice with the Trem2^R47H missense mutation introduced via CRISPR suggested that it acts as a near-complete loss of function, recapitulating phenotypes seen in Trem2 knock-out (KO) mice [34, 36]. However, subsequent analyses of Trem2 expression and splicing in these models identified the unexpected generation of a cryptic splice site and subsequent reduction of Trem2 expression due to the synonymous codon changes introduced as part of the CRISPR repair template [35]. Given the importance of a Trem2^R47H mouse model that more accurately recapitulates human cases, we designed alternative CRISPR repair templates, guided in part by Cheng et al., to introduce the R47H mutation into C57BL/6 J mice [44]. Confirmation of the desired sequence change (CGC – > CAT; arginine – > histidine) and synonymous silent codon changes are shown in Fig. 1a. To analyze expression and isoform usage of Trem2 from the Trem2^{R47H NSS} allele, we extracted RNA from whole brains of 15 week old wild-type and Trem2^{R47H NSS} mice, as well as from a cohort of Trem2^{R47H CSS} mice with the identified Cryptic Splice Site (CSS) and reduced expression [35] and conducted Nanopore long-read RNA-seq to examine the association of exons within individual transcripts. We identified three known annotated Trem2 isoforms – Trem2-201 encoding transmembrane TREM2, Trem2-202 encoding secreted TREM2, and Trem2-203 which contains retained introns (Fig. 1b, Supplemental Fig. 1). However, a novel truncated exon 2 containing isoform was abundant in the Trem2^{R47H CSS} brain and was not present in Trem2^{R47H NSS} mice. We designated our model as Trem2^{R47H NSS} (Normal Splice Site, NSS; available at The Jackson Laboratory—stock #034,036). Furthermore, reads of the major isoform, Trem2-201, are reduced in the Trem2^{R47H CSS} brains (> 50% reduction—wild-type mean TPM = 43.7, NSS = 33.8, CSS = 11.03; Fig. 1c).

To further validate the Trem2^{R47H NSS} model we conducted an off-target analyses of any other putative cut sites that might have been targeted by CRISPR-Cas9 during generation of the Trem2^{R47H NSS} allele. Trem2^{R47H NSS} founder mice were backcrossed with wild type C57BL/6 J animals for three generations before use to generate animals for this study, making it unlikely that a mutation caused by an off-target effect of CRISPR/Cas9 would be present on a chromosome other than chromosome 17, i.e., the location of Trem2 (C57BL/6 J; Chr17; 48.6 Mb, GRCm39, Ensembl release 108). Potential CRISPR/Cas9 off-target sites with up to four mismatches using crRNA TMF1342 were screened for using Cas-OFFinder (http://www.rgenome.net/cas-offinder/↗; [45]). Eleven potential off-target sites on mouse chromosome 17 were identified (Supplemental Table 2). Two potential off-target sites were within coding exons of genes, six were within introns, while the remaining three were within intergenic regions. To screen for evidence of CRISPR/Cas9 RNP activity at each site, DNA from a wild-type and homozygous Trem2^{R47H NSS} mouse was amplified by PCR using the primers listed (Supplementary Table 1) then sequenced across the potential off-target region at each locus (Supplemental Fig. 2a-k). None of the 11 potential off-target sites showed a difference in sequence between WT and homozygous Trem2^{R47H NSS} mice. Moreover, no significant change in gene expression was seen in any of these 11 genes via RNA-seq from whole brain extracted RNA (Supplemental Table 2). We also explored transcript isoform production from these genes through long-read RNA-seq and found no change (e.g. data for Zpf945 shown in Supplemental Fig. 2l, m).

Having confirmed correct expression, isoform usage, and an absence of off-target genomic effects, we next assessed the impact of the Trem2^{R47H NSS} variant on microglial function and tested if it acts as a loss of function allele using a cuprizone model of demyelination [46]. Both myelin and fibrillar Aβ act as TREM2 ligands and induce a distinctive gene expression profile in microglia known as a Disease Associated Microglia (DAM) signature, which can be explored with cuprizone. For comparison, we included cohorts of Trem2^{R47H CSS} mice, as well as Trem2 KO mice. These 3 groups and a wild-type group, were treated with cuprizone (0.3%) or control chow for 6 weeks (Fig. 1d), then were euthanized at 15 weeks and brains examined by histology and bulk RNA-seq. Only male mice were used for the cuprizone study due to the influence of estrogen and progesterone on myelination, as well as evidence of the treatment disrupting estrous cycle in mice [47, 48]. RNA was extracted from half brains, and bulk RNA-seq conducted, with a principal component analysis (PCA) plot for all samples shown in Supplemental Fig. 3a. Trem2 expression values were plotted, showing that Trem2^{R47H NSS} mice have similar Trem2 expression levels to wild-type mice consistent with results of the long-read RNA-seq analysis, and that both Trem2^{R47H CSS} and Trem2 KO mice have reduced expression (Fig. 1e). Notably, with cuprizone treatment, Trem2 levels increased similarly in wild-type and Trem2^{R47H NSS} mice, but to a lesser extent in Trem2^{R47H CSS} mice.

Luxol Fast Blue staining for myelin confirmed demyelination in all cuprizone treated mice with no overt difference in the extent of demyelination between the four groups (Fig. 1f, g). Plasma neurofilament light-chain (NfL) reflects axonal damage in the brain [49]. Interestingly, cuprizone treatment/demyelination did not increase plasma NfL in either WT mice, or the Trem2 variants/KO, although a substantial elevation of plasma NfL was present in Trem2 KO mice independent of cuprizone treatment (Fig. 1h). Immunofluorescence analysis for microglia (IBA1) and DAM marker CD11c (encoded by Itgax) reveals extensive microgliosis in the corpus callosum of cuprizone treated wild-type, Trem2^{R47H NSS}, and Trem2^{R47H CSS} mice, and to a lesser extent Trem2 KO mice (Fig. 1i, j). While CD11c expression is induced in cuprizone treated wild-type, Trem2^{R47H NSS}, and Trem2^{R47H CSS} mice, no induction is observed in the Trem2 KO mice (Fig. 1k).

Volcano plots of brain gene expression in cuprizone treated vs. control mice across the 4 groups reveal that animals in wild-type, Trem2^{R47H NSS}, and Trem2^{R47H CSS} groups all show clear upregulation of inflammatory genes in response to cuprizone that is markedly suppressed in the Trem2 KO group (Fig. 2a). We selected differentially expressed genes (DEGs) from the Trem2^{R47H NSS} mice (FDR < 0.05 for cuprizone vs. control) and created a heatmap to compare the response to that in the other 3 groups (Fig. 2b). Upregulated genes are all associated with inflammation and are mostly shared with the wild-type and Trem2^{R47H CSS} groups and include DAM genes such as Apoe, Clec7a, and Itgax (Fig. 2c), consistent with CD11c histology (Fig. 1k). Some inflammatory genes are also significantly upregulated in Trem2 KO mice suggesting their induction is Trem2 independent and includes more classical inflammation related genes such as C1qa, Hmox1, Tyrobp, and Trem2 itself (Fig. 2d and Supplemental Fig. 3b). Downregulated genes in Trem2^{R47H NSS} include a unique set not altered in wild-type, Trem2^{R47H CSS}, or Trem2 KO groups, and are associated with dopaminergic signaling in the striatum (Fig. 2e). Shared downregulated genes are associated with myelin and oligodendrocytes, including Cldn11, Mal, Mobp, Opalin, and Plp1 (Fig. 2f), suggesting that demyelination induced by cuprizone is not dependent upon the Trem2-dependent inflammation. Comparisons of DEGs between the three genotypes compared to wild-type in the absence of cuprizone treatment are shown as volcano plots (Supplemental Fig. 3c). Notably, Trem2^{R47H CSS} brains show more DEGs compared to wild-type mice than Trem2^{R47H NSS} (Supplemental Fig. 3d). Likewise, exploration of the number of DEGs induced by cuprizone in each of the four genotypes further revealed an extensive number of genes induced in the Trem2^{R47H CSS} brains that are not induced by cuprizone in wild-type, Trem2^{R47H NSS} or Trem2 KO brains (Supplemental Fig. 3e; shown as a heatmap in Supplemental Fig. 3f).

To further explore gene expression changes across the groups, we analyzed functional networks of correlated genes (weighted correlation gene network analysis (WGCNA)) and identified two modules (White and Lightcoral) associated with cuprizone treatment (Fig. 2g). White module eigengene values were increased with cuprizone to similar extents in Trem2^{R47H NSS} and Trem2^{R47H CSS} mice, but significantly decreased in Trem2 KO compared to wild-type mice (Fig. 2h), with gene ontology being associated with inflammation (Fig. 2i). Lightcoral module eigengene values were reduced in both Trem2^{R47H NSS} and Trem2 KO mice with cuprizone but not in wild-type or Trem2^{R47H CSS} mice (Fig. 2j), and gene ontology associated with myelination and lipid metabolism (Fig. 2k), suggesting that the presence of Trem2^{R47H NSS} induces the appropriate inflammatory response to demyelination yet exacerbates its effects on myelination gene expression compared to wild-type Trem2. As myelin acts as a TREM2 ligand, inducing similar gene expression changes in microglia as exposure to Aβ plaques, we then compared all modules to AMP-AD identified modules that define gene expression changes in AD [50], for significant gene overlap (Supplemental Fig. 4). Notably, the White module (i.e., microglia/inflammation) strongly overlaps with the immune response and cytokine signaling AMP-AD modules, while the Lightcoral module (i.e., myelin and lipid metabolism related) strongly overlaps with the myelination AMP-AD modules. In addition to these two modules, the Lightgray module has significant overlap with the neuronal-related AMP-AD modules and is significantly inversely correlated to gene expression in Trem2^{R47H NSS}, but not Trem2^{R47H CSS} or Trem2 KO brains (Fig. 2g). Collectively, these results show that Trem2^{R47H NSS} mice show appropriate expression of Trem2 transcripts, and that the Trem2^{R47H NSS} allele does not function as a null allele as assessed by cuprizone challenge.

To investigate the contributions of Trem2^R47H to the pathogenesis of AD, we bred Trem2^{R47H NSS} mice with 5xFAD mice on a congenic B6J background to generate 4 groups: (i) wild-type, (ii) homozygous Trem2^{R47H NSS}, (iii) hemizygous 5xFAD, and (iv) hemizygous 5xFAD; homozygous Trem2^{R47H NSS}. We used Trem2^{R47H NSS} homozygotes rather than the heterozygous TREM2^R47H allele found in human AD individuals, to exacerbate phenotypes associated with the R47H mutation. Hereafter, for simplicity we refer to the Trem2^{R47H NSS} genotype as Trem2^R47H. Mice were aged to 4 and 12 months and analyzed. Four-month-old 5xFAD mice are in their rapid plaque growth stage, which then plateaus throughout the brain by ~ 8–12 months depending on brain region [37]. While plaque densities are similar for both 5xFAD and 5xFAD/Trem2^R47H 4-month-old mice when collapsed cross the sexes (Fig. 3c, e), homozygosity for Trem2*R47H induces a robust sex difference in the manifestation of Thio-S⁺ plaques in both the cortex and subiculum, with lower plaque load in male 5xFAD/Trem2^R47H vs. 5xFAD mice (Supplemental Fig. 5b, n). Similar sex differences for plaque densities have been reported for Trem2 KO mice crossed with 5xFAD mice [51]. Furthermore, female 5xFAD/Trem2^R47H mice exhibit increased plaque volume in the cortex compared to the age-matched 5xFAD mice (Supplemental Fig. 5e). Plaques in the subiculum of 4-month-old 5xFAD/Trem2^R47H mice are smaller with reduced mean plaque intensity compared to 5xFAD mice (Fig. 3f, g). Differences in the effects of Trem2^R47H on plaque volume between the cortex (where they are larger) and subiculum (where they are smaller) may be disease stage dependent as pathology begins earlier in the subiculum than the cortex, where at this timepoint plaques are only just beginning to form. Immunostaining for Aβ fibrillary oligomers using the conformation specific antibody OC, revealed less OC⁺ material, which includes more diffuse material in addition to the dense cores, in 5xFAD/Trem2^R47H compared to 5xFAD mice in the subiculum at 4 months (Fig. 3h, i). Quantification of Thio-S⁺ plaques in the visual cortex at the 12-month time point shows no differences (Fig. 3k-l), but there is a trend towards increased plaque volume in male compared to female 5xFAD/Trem2^R47H mice (Supplemental Fig. 5h). In the subiculum, 5xFAD/Trem2^R47H mice have higher plaque counts with comparable sizes (Fig. 3m-n). However, increased mean Thio-S⁺ plaque intensity in 5xFAD/Trem2^R47H mice is seen, with no difference in OC⁺ plaques (Fig. 3o-q). In the cortex, however, there is no difference in either mean plaque intensity or total OC⁺ volume at both 4- and 12-months (Supplemental Fig. 6).

We quantified Aβ40 and Aβ42 in detergent soluble and insoluble fractions from microdissected hippocampi and cortices. At 4 months of age, increased soluble Aβ40 and 42 are found in the 5xFAD/Trem2^R47H hippocampus (Fig. 4h) and a trending reduction in soluble Aβ42 in the cortex (Fig. 4f), but no difference in the insoluble fraction in either brain region (Fig. 4a-d), compared to 5xFAD mice. By 12 months of age, Aβ is increased in both brain regions and fractions, with increased soluble Aβ42 in the hippocampus and increased insoluble Aβ40 and 42 in the cortex of 5xFAD/Trem2^R47H mice (Fig. 4i-p) compared to 5xFAD mice. Collectively, these results show that Trem2^R47H impacts the level of both plaque and Aβ in a brain region and age-specific manner.

Given the expression of Trem2 by microglia in the brain, we looked for changes in microglial densities and morphologies in both a non-pathologic and early pathologic (i.e., cortical regions of 5xFAD mice without overt dense core plaque load) state. Morphological analyses of cortical IBA1⁺ microglia showed increased process length but decreased diameter in Trem2^R47H compared to WT mice, while these differences were less apparent in the 5xFAD hemizygous background (Fig. 5a, b; images and sample analyses shown in Supplemental Fig. 7). We next imaged microglia and plaques in the visual cortex and subiculum (Fig. 5c, d). At 4-months old, as expected, the sex difference observed in plaque load of 5xFAD/Trem2^R47H mice is reflected in cortical microglial density with females showing higher microglia and plaque densities (Fig. 5e; data separated by sex included in Supplemental Fig. 8). In the subiculum, where Thio-S⁺ plaques are abundant, IBA1⁺ microglial density and size are increased in 5xFAD and 5xFAD/Trem2^R47H compared to wild-type mice (Fig. 5g, h). However, IBA1⁺ cell densities and volumes of 5xFAD/Trem2^R47H are lower than 5xFAD mice (Fig. 5g, h). Notably, in both brain regions, Trem2^R47H mice exhibit lower IBA1⁺ cell volume compared to WT, indicating that the Trem2^R47H variant elicits a plaque-independent effect on microglia morphology (Fig. 5f, h). A lack of plaque-microglia interaction is found in 5xFAD/Trem2^R47H mice as shown through quantification of Thio-S and IBA1 colocalization in the subiculum, with a further impairment in female 5xFAD/Trem2^R47H mice compared to males (Fig. 5o). However, the decreased microglia volume in Trem2^R47H mice and 5xFAD/Trem2^R47H compared to WT and 5xFAD, respectively, is absent at 12-month (Fig. 5 j, l), as is the impairment in plaque-microglia interaction, suggesting age/disease-dependent changes in the Trem2^R47H variant effect on microglia morphology and function (Fig. 5m-p).

We quantified astrocytes in the visual cortex and subiculum and found that the Trem2^R47H variant has minimal effects on the astrocytic response to plaques at both 4- and 12-months (Supplemental Fig. 9).

Microglia form a protective barrier around plaques while contributing to their compaction and growth [21]. Given the initial impairment in microglial-plaque interactions in 5xFAD/Trem2^R47H mice we next explored the halo of dystrophic neurites that develops around dense-core plaques [52, 53] which can be visualized with lysosomal-associated membrane protein 1 (LAMP1, Fig. 6a, b). Normalization of LAMP1 volume to plaque volume reveals an increase in dystrophic neurites per plaque area in 5xFAD/Trem2^R47H mice at 4 months of age (Fig. 6c). Consistent with the restoration of microglial-plaque interactions at 12 months of age, no difference in dystrophic neurites per plaque is seen between 5xFAD/Trem2^R47H and 5xFAD mice at this age (Fig. 6d). Twelve-month-old 5xFAD females have more dystrophic neurites than male (Fig. 6d). Interestingly, LAMP1⁺ dystrophic neurites exhibit a more dissipated morphology at 12-month compared to 4-month, consistent with disrupted axonal transport at later age/disease stages [54].

Neurofilament light chain (NfL) is emerging as a clinically useful plasma biomarker for damage occurring in the brain, including in AD where it tracks with cortical thinning and cognitive decline [55–57], while in mouse models of AD it correlates with plaque load [38]. We measured plasma NfL as a surrogate marker of brain damage and found it increased in 5xFAD mice compared to WT mice at 4 months of age, with further increase at 12-months of age, consistent with plaque load (Fig. 6e). The Trem2^R47H variant increased plasma NfL in 5xFAD at both ages (Fig. 6e), in line with the hypothesis that TREM2 dysfunction exacerbates neuronal damage. We also measured NfL levels in the detergent soluble and insoluble fractions of microdissected cortices (Fig. 6f, g). NfL in the soluble fraction is not increased in 5xFAD mice compared to WT at either 4 or 12 months of age but is elevated by the presence of Trem2^R47H (Fig. 6f). NfL in the insoluble fraction aligns with levels in the plasma, with increases seen in 5xFAD compared to wild-type mice at 4 months and further increases at 12 months (Fig. 6g). As with plasma, NfL levels further trend higher in 5xFAD/Trem2^R47H mice. Thus, the presence of Trem2^R47H induces changes in NfL, and further exacerbates plaque-induced increases in both detergent insoluble and plasma NfL. To identify the cellular source of NfL, we immunostained for dystrophic neurites (LAMP1) and NfL. Large spherical structures of NfL are seen in the vicinity of plaques and are absent from wild-type and Trem2^R47H mice, where staining is observed only in axonal fibers (Fig. 6a, b). Similar bead-like NfL⁺ spheroids were reported in ischemia-affected human and mouse tissues as sign of axonal damage [58]. Notably, these NfL spheroids colocalize with dystrophic neurites associated with plaques (Fig. 6h, k). Quantification of NfL⁺ structures showed a decreased spheroid number in 5xFAD/Trem2^R47H compared to 5xFAD, while having similar size at 4-month (Fig. 6i, j). At 12-month, no difference in NfL⁺ spheroid number or size is observed between 5xFAD and 5xFAD/Trem2^R47H brains. However, in 5xFAD, there is a sex difference with females having more NfL compared to males, which is also observed in LAMP1⁺ dystrophic neurite amount (Fig. 6d; data separated by sex included in Supplemental Fig. 10). Collectively, these findings reveal associations between plaques and dystrophic neurites with NfL accumulation and its transition to the insoluble fraction and plasma NfL.

Due to the increased dystrophic neurites induced by the R47H variant, we investigated short- and long-term synaptic plasticity in WT, Trem2^R47H, 5xFAD, and 5xFAD/Trem2^R47H hippocampi via theta burst-induced pair-pulse facilitation (PPF) and long-term potentiation (LTP) in acute hippocampal slices. Consistent with our previous findings [37], 5xFAD have impaired LTP at 4 months of age. Remarkably, this impairment is suppressed by the presence of Trem2^R47H (Fig. 7a, b), with no change in PPF (Fig. 7c). Consistent with a lack of LTP impairment in 5xFAD/Trem2^R47H mice, immunostaining of pre-synaptic (Bassoon) and post-synaptic (PSD-95) elements at the CA1 stratum radiatum (CA1-SR) using super-resolution structured illumination microscopy revealed a decrease in co-localized pre- and post-synaptic puncta in 5xFAD animals compared to wild-type mice, which is not seen in 5xFAD/Trem2^R47H mice compared to Trem2^R47H mice (Fig. 7d, e). At 12-months of age, LTP deficits are also seen in Trem2^R47H, as well as 5xFAD, and, to a lesser degree, 5xFAD/Trem2^R47H, compared to WT animals (Fig. 7f, g). Notably, PPF responses in 12-month-old mice show a decrease in presynaptic plasticity in both 5xFAD and 5xFAD/Trem2^R47H compared to their non-5xFAD controls at 40 ms and 100 ms stimulus intervals, although, the effect in 5xFAD/Trem2^R47H is absent at 200 ms interval (Fig. 7h). Quantification of CA1-SR synapses at 12 months recapitulates the LTP results—reduction of synaptic co-localization in Trem2^R47H and 5xFAD compared to WT with a trending rescue in the 5xFAD/Trem2^R47H vs 5xFAD (Fig. 7i, j). Postsynaptic elements are also decreased in 12-month-old Trem2^R47H compared to WT, along with an increase in 5xFAD/Trem2^R47H vs 5xFAD, which reflect the observed fEPSP (Fig. 7g) and mean potentiation data (Supplementary Fig. 11 l). Lastly, basic synaptic transmission was assessed by examining the input–output relationship of the synaptic response in all four groups of mice (Supplemental Fig. 11). As stimulus intensities increased, the amplitude of the fiber volley is significantly reduced in slices from 4 months 5xFAD, Trem2^R47H and 5xFAD/Trem2^R47H mice relative to WT controls indicating reductions in synaptic transmission (Supplemental Fig. 11a-c). By 12 months of age, the amplitude of the fiber volley over a range of stimulus intensities is indistinguishable from control WT. However, a slight reduction in the slope of the fEPSP is measured in all three experimental groups with respect to controls (Supplemental Fig. 11 g-i). These results suggest that 5xFAD mice may have a reduction in afferent activation, but also reveals deficits in synaptic transmission in Trem2^R47H and 5xFAD/Trem2^R47H mice that are not fully explained by long-term plasticity changes.

To assess gene expression changes with age, sex and genotype, we performed RNA-seq from microdissected hippocampi from 4- and 12-month-old WT, Trem2^R47H, 5xFAD, and 5xFAD/Trem2^R47H mice. PCA plots show clustering of 4 vs 12-month samples accounting for most of the variance between samples (Supplemental Fig. 12a). Volcano plots illustrate changes in gene expression in Trem2^R47H relative to WT mice, 5xFAD relative to WT, and 5xFAD/ Trem2^R47H relative to 5xFAD mice at both 4- and 12- month timepoints (Fig. 8a). Differentially expressed genes (DEGs) in 5xFAD mice vs. WT mice at both ages are mostly upregulated genes and represent the strong inflammatory response seen in these mice, and include DAM genes such as Cst7, Itgax, Clec7a, as well as Trem2. Identified Trem2-dependent and -independent induction of inflammatory genes from the cuprizone experiments are shown as a heatmap in Supplemental Fig. 12b and c. DEGs between 5xFAD/Trem2^R47H vs 5xFAD mice at 12 months are plotted as a heatmap (Fig. 8b; FDR < 0.1, no FC cutoff). Notably, this subset consists of downregulated circadian related genes, Ciart and Dbp, compared to 5xFAD mice, but upregulation of many genes associated with interferon signaling, such as Ifi47, Ifit1-3, and Gbp2, 6, and 7 (Fig. 8b). These interferon-related genes are also involved in a highly connected gene network (Fig. 8c).

To focus on the inflammatory response, we first selected homeostatic and disease-associated microglia genes and displayed them as a heatmap (Fig. 8d). This revealed a suppression of these genes in 4-month-old 5xFAD/Trem2^R47H hippocampi compared to 5xFAD hippocampi that was absent at 12 months of age. To further explore gene expression changes across the groups, we analyzed functional networks of correlated genes (WGCNA; module – trait relationships shown in Supplemental Fig. 12d) and identified a module containing inflammation associated genes (Fig. 8f; Darkgrey). We plotted eigengene values of this inflammation module for all groups (Fig. 8e), revealing an increase in 4-month-old 5xFAD mice compared to both WT and Trem2^R47H mice, corresponding to the inflammatory response to the plaques in those mice. Notably, no such increase in eigengene value is observed in 4-month-old 5xFAD/Trem2^R47H mice, showing a suppression of inflammation at the 4-month timepoint compared to 5xFAD mice, mirroring the LTP and synaptic deficits seen in these mice. However, by 12 months of age, robust increases in the inflammation module eigengene values are seen in both 5xFAD and 5xFAD/Trem2^R47H mice, with the presence of Trem2*R47H inducing an even higher increase (Fig. 8e). This further shows the initial suppression of inflammation induced by the Trem2*R47H variant dissipates with time/age/disease progression, consistent with the results of histology. We next compared all modules with the AMP-AD identified modules that define gene expression changes in AD brains [50]. The Darkgrey inflammatory module shares a significant and substantial overlap with the immune response and cytokine signaling modules (Supplemental Fig. 13a), which we established is enhanced in 12-month-old 5xFAD/Trem2^R47H compared to 5xFAD mice (Fig. 8e). The Lightgrey module has significant overlap with the AMP-AD neuronal systems modules and plotting of the eigengene values reveals a significant decrease in this module in 12-month-old 5xFAD/Trem2^R47H compared to 5xFAD mice (Supplemental Fig. 13b). Finally, the Snow module shows significant overlap with the AMP-AD myelination related modules and plotting of the eigengene values reveals a clear effect of Trem2^R47H on this module in 12-month-old mice, regardless of 5xFAD genotype (Supplemental Fig. 13d). Collectively, these results show that the presence of the Trem2^R47H variant induces a gene expression signature that better matches the spectrum of gene expression changes in human AD brains.

Given sex-specific transcriptional changes previously reported in TREM2^R47H variant-carrying AD patients and mice expressing TREM2^R47H cDNA crossed with PS19 mice, we explored gene expression changes in male and female mice comparing 5xFAD/Trem2^R47H with 5xFAD at 12 months [18]. Interestingly, 5xFAD/Trem2^R47H female mice show a unique gene expression signature compared to the age-matched 5xFAD mice (Fig. 8 g-h), which is not seen in males. Moreover, most of the genes downregulated in 5xFAD/Trem2^R47H females are highly upregulated in WT and less so in 5xFAD females, yet this effect is not observed in Trem2^R47H, suggesting the presence of the Trem2^R47H variant dampens the expression of these genes in non-disease and further exacerbates in a disease model (Fig. 8h). Gene ontology analyses suggests these genes are involved in dendrite development, chemical synaptic transmission, and pathways such as circadian rhythm and axonal guidance (Fig. 8i-j).

Given the selective downregulation of Itgax and CD74 at 4-months-of-age between 5xFAD/Trem2^R47H and 5xFAD mice (Fig. 8a), we performed immunofluorescence for both markers (Supplemental Fig. 14a, b), alongside microglia and plaques. CD11c (Itgax) and CD74 staining were seen in a subset of plaque-associated microglia, with greater numbers at 12- vs. 4-months of age. Concordant with gene expression data, CD11c and CD74 staining was reduced at 4-months of age (Supplemental Fig. 14e, f). However, at 12 months old, despite not reaching FDR of 0.1, Itgax TPM level, when examined independently and separate from other genes, shows a significant decrease in 5xFAD/Trem2^R47H compared to 5xFAD (Supplemental Fig. 14g), which is also observed via immunohistochemistry (Supplemental Fig. 14i). CD74 staining (Supplemental Fig. 14j) is also reflective of CD74 TPM level at 12 months (Supplemental Fig. 14h) with no difference between 5xFAD and 5xFAD/Trem2^R47H.

Collectively, these results show that the Trem2^R47H variant mediates an initial suppression of inflammation in response to plaques in 5xFAD mice, which is reversed and exacerbated with age/disease progression. Furthermore, the Trem2^R47H variant generates a unique interferon-related gene expression signature in 5xFAD mice.

Not applicable.

All experiments involving mice were approved by the UC Irvine Institutional Animal Care and Use Committee and were conducted in compliance with all relevant ethical regulations for animal testing and research. All experiments involving mice comply with the Animal Research: Reporting of in Vivo Experiments (ARRIVE-10) guidelines.

Not applicable.

KNG is a member of the advisory board of Ashvattha Therapeutics.

Protocols, data, and results are available via the AD Knowledge Portal (https://adknowledgeportal.synapse.org↗). The AD Knowledge Portal is a platform for accessing data, analyses, and tools generated by the Accelerating Medicines Partnership (AMP-AD) Target Discovery Program and other National Institute on Aging (NIA)-supported programs to enable open-science practices and accelerate translational learning. The data, analyses and tools are shared early in the research cycle without a publication embargo on secondary use. Data is available for general research use according to the following requirements for data access and data attribution (https://adknowledgeportal.org/DataAccess/Instructions↗). Data can be accessed in an interactive matter at UCI Mouse Mind Explorer (admodelexplorer.org).

The Trem2^{R47H NSS} model is available from The Jackson Laboratory (Stock #034,036) without restrictions on its use by both academic and commercial users. The content is solely the responsibility of the authors and does not necessarily represent the official view of the National Institutes of Health.