Estrogen deprivation exacerbates Alzheimer’s disease pathology through neuronal CTSS signaling

Abstract

Alzheimer’s disease (AD) exhibits a pronounced sex bias, with women facing disproportionately higher risk and more severe pathology. Postmenopausal estrogen decline is implicated in this vulnerability, yet the molecular mechanisms linking estrogen loss to AD pathogenesis remain incompletely understood. Here, we demonstrate that ovariectomy (OVX) in female 5xFAD mice significantly exacerbates amyloid-β (Aβ) pathology, cognitive deficits, neuroinflammation, and reduces synaptic markers. Pharmacological blockade of estrogen receptor signaling recapitulated these effects, confirming their dependence on estrogen receptor pathways. Single-nucleus RNA sequencing (snRNA-seq) revealed widespread transcriptional reprogramming across brain cell types following estrogen deprivation, with prominent upregulation of the lysosomal protease cathepsin S (Ctss) and the AD risk gene ApoE. Remarkably, partial genetic reduction of CTSS prevented OVX-induced Aβ accumulation, glial activation, and synaptic decline in female 5xFAD mice, establishing CTSS as a critical downstream mediator of estrogen deficiency-driven pathology. Our findings provide mechanistic insight into sex-biased AD vulnerability and identify CTSS as a promising therapeutic target for mitigating AD risk in postmenopausal women.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12974-026-03970-5.

Introduction

AD is a devastating neurodegenerative disorder characterized by the progressive accumulation of Aβ plaques, neurofibrillary tangles, and profound neuroinflammation, ultimately culminating in synaptic loss and cognitive decline [1, 2]. With global prevalence projected to triple by 2050, identifying the drivers of AD susceptibility is a public health imperative [1, 3]. A striking epidemiological hallmark of AD is its pronounced sex bias; women constitute nearly two-thirds of the patient population, a disparity that persists even after accounting for their longer average lifespan [4]. Emerging evidence suggests that the menopause-associated decline in ovarian hormones, particularly 17β-estradiol (E2), acts as a critical biological catalyst for late-onset AD in women [5].

Estrogen exerts multifaceted neuroprotective effects, including the maintenance of synaptic plasticity and the regulation of Aβ metabolism through its receptors, ERα (Esr1) and ERβ (Esr2), which are highly expressed in the hippocampus and cortex [6–9]. Beyond its direct effects on neurons, estrogen is a potent immunomodulator. In the healthy brain, estrogen signaling maintains microglial homeostasis and restrains pro-inflammatory transcriptional programs [10, 11]. Consequently, the abrupt loss of E2 following menopause or OVX in rodents triggers microglial hyperactivation and exacerbates Aβ pathology, suggesting that estrogen deficiency destabilizes the neuroimmune environment [12–14]. However, the specific molecular bridges that translate hormonal loss into chronic neuroinflammatory states—and whether these effects are driven by cell-type-specific signaling—remain incompletely defined.

Recent advances in single-nucleus transcriptomics have highlighted dysregulation of the endolysosomal-proteolytic system as a central driver of AD [15]. Among these, CTSS, a lysosomal cysteine protease, has emerged as a key candidate. While primarily known for antigen presentation, neuronal CTSS can be secreted into the extracellular space, where it cleaves the chemokine CX3CL1, thereby triggering the CX3CR1-dependent recruitment and activation of microglia [16]. Although CTSS has been implicated in Aβ processing and synaptic pruning, its role in the context of sex-specific AD vulnerability has not been explored. We hypothesized that CTSS may serve as the missing link between estrogen depletion and the subsequent neuroinflammatory surge observed in females.

In this study, we utilized ovariectomized 5xFAD mice to model postmenopausal estrogen deprivation and investigate its impact on AD progression. We show that OVX significantly exacerbates Aβ deposition, synaptic degeneration, and memory deficits. Using snRNA-seq, we identified Ctss in neurons as a prominent estrogen-responsive gene that is markedly upregulated upon hormonal loss. Crucially, partial genetic reduction of Ctss effectively reversed the OVX-induced pathological aggravation. Our findings define a previously unrecognized estrogen–CTSS–microglia axis, providing a mechanistic framework for female-biased AD susceptibility and identifying cell-type-specific targets for therapeutic intervention in neuroinflammation.

Materials and methods

Animals

Female 5xFAD transgenic mice, maintained as hemizygotes on a C57BL/6 background, were procured from Jackson Laboratory (MMRRC stock #34840-JAX). Female Ctss KO mice were procured from Cyagen Guangzhou Biosciences (S-KO-01693). To generate experimental cohorts with specific genotypes, we performed the following breeding strategies: male 5xFAD mice were crossed with female Ctss KO (Ctss^–/–) mice to obtain 5xFAD;Ctss⁺/^– offspring. All mice were group-housed (≤ 5 mice per cage) under controlled conditions (12 h light/dark cycle) with ad libitum access to water and standard rodent chow. All animal experimental procedures were approved by the Institutional Animal Care and Use Committee of Chongqing Medical University (IACUC-CQMU-2024-0931).

OVX

Bilateral OVX was performed as described previously [17]. Mice were anesthetized with isoflurane inhalation (induction at 3–4%, maintenance at 1–2%) and placed in the prone position on a surgical platform equipped with a heating pad set to 42 °C. After securing the limbs with surgical tape, the dorsal fur was shaved, and the skin was disinfected with 75% ethanol. Once the alcohol had completely evaporated, a midline dorsal incision was made through the skin and underlying muscle layers using a sterile scalpel to expose the mice’s abdominal cavity. The ovaries and attached oviducts were gently externalized using sterile forceps. Each ovary, along with approximately 3 mm of the adjacent oviduct, was removed using sterile scissors. The remaining oviduct was returned to the abdominal cavity. The muscle and skin layers were sutured sequentially, and the incision site was disinfected with povidone-iodine. After surgery, mice were returned to individual cages and provided with buprenorphine (0.1 mg/mL in drinking water) for postoperative analgesia. The mice were housed individually until they had fully recovered. Sham-operated mice underwent the same surgical procedures, including incision and manipulation of the ovaries; however, the ovaries and oviducts were left intact.

Intraperitoneal injection

Female 5xFAD mice were randomly assigned to the vehicle, tamoxifen, or fulvestrant groups at 2 months of age. Tamoxifen and fulvestrant were administered by intraperitoneal injection once weekly at a dose of 100 mg/kg body weight for 6 months, from 2 to 8 months of age, resulting in 24 injections per mouse. Both tamoxifen and fulvestrant were prepared in corn oil. Mice in the vehicle group received an equal volume of corn oil following the same schedule.

Vaginal cytology

To monitor estrous cyclicity and confirm the success of OVX, vaginal cytology was performed on 5xFAD-SHAM and 5xFAD-OVX mice for 5 consecutive days immediately preceding tissue collection (≈ 6 months of age, ≈ 4 months after surgery). Approximately 0.1 mL of sterile PBS was aspirated with a sterile pipette tip; the tip was gently inserted ~ 1–2 mm into the vaginal opening, and the vaginal canal was flushed 2–3 times. A single drop of the lavage was spread evenly onto a glass slide to form a thin smear, air-dried at room temperature, and stained with hematoxylin and eosin.

Open field test

The open field test (OFT) was used to assess spontaneous locomotion and exploratory behavior. Mice were individually placed into a white acrylic chamber (40 cm × 40 cm × 30 cm; length × width × height) with a smooth, non-reflective floor. The test was conducted in a quiet room with uniform, ambient lighting. Each mouse was allowed to freely explore the arena for 10 min. Mouse locomotion was recorded and analyzed using the TopScanLite system (CleverSys Inc.), which automatically tracked movement trajectories and quantified behavioral parameters. The arena was cleaned with 75% ethanol between trials to eliminate olfactory cues.

Y-maze

Spontaneous alternation behavior was assessed using a Y-maze apparatus consisting of three identical arms (each 30 cm in length, 8 cm in width, and 15 cm in height) arranged at 120° angles. Mice were placed at the end of one arm and allowed to explore the maze freely for 8 min. The sequence and number of arm entries were recorded using TopScanLite. Alternation was defined as entry into all three arms consecutively (e.g., ABC or BCA). Mice were included in the spontaneous alternation analysis only if they made at least 8 arm entries during the session. The percentage of spontaneous alternation was calculated as the ratio of actual alternations to the total possible alternations: % alternation = (number of alternations) / (total arm entries – 2) × 100%. The maze was cleaned with 75% ethanol between trials to eliminate any olfactory cues.

Fear conditioning

Fear conditioning was performed to assess associative learning and memory. Mice were first habituated to the behavioral testing room for at least 4 h. The test consisted of a training session, followed by contextual and cue fear memory tests. During training, mice were placed in a fear-conditioning chamber and allowed to explore for 2 min, followed by two-tone–foot-shock pairings (28 s, 80 dB tone co-terminating with a 2 s, 0.5 mA foot shock), spaced 2 min apart. The total training session lasted for 8 min. Behavior was recorded using the TopScanLite system, and freezing behavior (defined as immobility > 1 s) was quantified. Twenty-four hours after training, contextual fear memory was tested by returning mice to the original conditioning chamber for 5 min without tone or foot shock. After a 1-h interval, cued fear memory was tested in a modified chamber with altered contextual cues, including different walls and floor texture. During the cued fear test, mice were first allowed to explore the modified chamber for 2 min without tone to establish baseline freezing, followed by three conditioned stimulus presentations without foot shock. Freezing percentages were calculated separately for the baseline period and each conditioned stimulus period as the duration of freezing divided by the corresponding phase duration.

Morris water maze

The Morris water maze (MWM) test was used to assess spatial learning and memory in mice. The apparatus consisted of a circular pool (120 cm in diameter, 50 cm in height) filled with opaque water (22 ± 1 °C). A hidden escape platform (10 cm in diameter) was submerged 1 cm below the water surface in a fixed quadrant. During the training phase, the mice underwent four trials per day for five consecutive days. In each trial, mice were released into the pool from one of four randomized start positions and allowed to search for the hidden platform for up to 60 s. Mice that failed to find the platform within 60 s were guided to it and allowed to remain for 15 s. Escape latency and swim paths were recorded using the TopScanLite tracking system. Twenty-four hours after the final training day, a probe test was performed in which the platform was removed from the pool. Mice were allowed to swim freely for 60 s, and the time spent in the target quadrant was recorded as an index of spatial memory.

Euthanasia and tissue collection

Mice were deeply anesthetized with Avertin (400 mg/kg, Target Mol, T0807), and the thoracic cavity was opened to expose the heart. Approximately 200 µL of blood was collected by cardiac puncture using a sterile 1 mL syringe. Samples were kept at 4 °C for 2 h and centrifuged at 400 g for 10 min. Serum was collected and stored at − 80 °C until use. Then, a perfusion needle was inserted into the left ventricle using a peristaltic pump, and the right atrium was incised to allow efflux. Systemic perfusion was performed with 30 mL of ice-cold PBS, followed by immediate brain removal and hemispheric dissection. The left hemisphere was post-fixed in 4% paraformaldehyde (PFA) for histological analysis, whereas the right hemisphere was flash-frozen in liquid nitrogen for biochemical analyses. The left cerebral hemisphere was carefully dissected and post-fixed in 4% PFA for 24 h. Subsequently, the tissue was dehydrated in 30% sucrose solution at 4 °C for 3 days. Following dehydration, the brains were embedded in optimal cutting temperature (OCT) (SAKURA, 4583) compounds and snap-frozen. Serial coronal brain sections (40-µm-thickness) were cut at one-in-six intervals using a cryostat (RWD, FS800A), spanning from Bregma − 0.95 mm to Bregma − 3.40 mm. Sections were stored in a cryoprotective solution (40% PBS, 30% glycerol, and 30% ethylene glycol) at -20 °C until processing.

Antibodies

The information of primary antibodies used was as follows: rabbit anti-IBA1 (FUJIFILM Wako Chemicals, 019-19741, RRID: AB_839504, 1:500 for immunofluorescence(IF)), goat anti-IBA1 (Novus, NB100-1028, RRID: AB_3148646, 1:500 for IF, 1:1000 for WB), rabbit anti-GFAP (Abcam, ab7260, RRID: AB_305808, 1:500 for IF, 1:1000 for WB), rat anti-CD68 (BIO-RAD, MCA1957, RRID: AB_3100585, 1:500 for IF), rat anti-LAMP1 (DSHB, clone 1D4B, RRID: AB_2134500, 1:500 for IF), rabbit anti-PU.1 (CST, 2258 S, RRID: AB_2186909, 1:500 for IF), rabbit anti-Amyloid Precursor Protein, C-Terminal (Sigma-Aldrich, A8717, RRID: AB_258409, 1:1000 for WB), mouse anti-PSD-95 (NeuroMab, 75− 028, RRID: AB_2292909, 1:500 for IF, 1:1000 for WB), rabbit anti-Synaptophysin (Abclonal, A6344, RRID: AB_2766946, 1:1000 for WB), rabbit anti Bassoon (Abcam, ab82958, RRID: AB_1860018, 1:500 for IF), rabbit anti-β-Actin (Proteintech, 66009-1-Ig, RRID: AB_2687938, 1:10000 for WB), rabbit anti-β Tubulin (Proteintech, 10094-1-AP, RRID: AB_2210695, 1:5000 for WB), rabbit anti-α Tubulin (Abclonal, A6830, RRID: AB_2863541, 1:10000 for WB) and mouse anti-GAPDH (Proteintech, 60004-1-Ig, RRID: AB_2107436, 1:10000 for WB).

The secondary antibodies used were as follows: donkey anti-rabbit IgG (H + L) with Alexa Fluor 488 (Invitrogen, A-21206, RRID: AB_2535729), donkey anti-rabbit IgG (H + L) with Alexa Fluor 568 (Invitrogen, A10042, RRID: AB_2534017), donkey anti-rabbit IgG (H + L) with Alexa Fluor 647 (Invitrogen, A-31573, RRID: AB_2536183), donkey anti-goat IgG (H + L) with Alexa Fluor 488 (Invitrogen, A-11055, RRID: AB_2534102), donkey anti-goat IgG (H + L) with Alexa Fluor 568 (Invitrogen, A-11057, RRID: AB_2534104), and donkey anti-rat IgG (H + L) with Alexa Fluor 488 (Invitrogen, A48269, RRID: AB_2893137), all diluted by 10% donkey serum in PBS 1:500 for IF. HRP-conjugated goat anti-rabbit IgG (H + L) (Proteintech, SA00001-2, RRID: AB_2722564), HRP-conjugated goat anti-mouse IgG (H + L) (Proteintech, SA00001-1, RRID: AB_2722565), all diluted in 5% BSA in TBST at 1:8000 for WB.

Immunostaining

Immunostaining was performed as previously described [18]. Briefly, the sections were permeabilized with 0.3% Triton X-100 and blocked with 10% donkey serum at room temperature for 1 h. The sections were then incubated with primary antibodies at 4 °C overnight. After three washes with PBS, the sections were incubated with the appropriate fluorophore-conjugated secondary antibody for 2 h. Nuclei were counterstained with DAPI (Thermo Fisher Scientific, 62248).

X34 staining

X34 staining for amyloid plaques and subsequent quantification were performed as previously described [19], and the analysis was conducted by an investigator blinded to the animals’ treatment conditions. For each mouse, ten brain sections evenly distributed between Bregma − 0.95 mm and − 3.40 mm were analyzed, and the mean of these sections was used as a mouse-level measure of Aβ burden. The sections were washed thrice with PBS (5 min each) and permeabilized with 0.3% Triton X-100 in PBS for 30 min. The sections were then incubated in 40% ethanol and 60% PBS containing dissolved X34 (MERCK, SML1954) for 20 min. This was followed by washing with X34 buffer (40% ethanol in PBS) and two additional washes with phosphate-buffered saline (PBS). The sections were subsequently mounted and sealed using Fluor mount-G (SouthernBiotech, 0100-01) and stored in the dark at 4 °C until ready for imaging.

Plaque quantification was performed as previously described [20] with slight modifications. Briefly, images were acquired using a spinning-disk confocal microscope and processed using the Olympus Viewer software. The images were then converted to 8-bit grayscale using the NIH ImageJ software to facilitate analysis. A threshold was applied to grayscale images to enhance plaque visualization while minimizing background noise. After thresholding, each object was manually reviewed to confirm its identity. For each mouse, ten brain sections were analyzed. Plaque quantification was performed in both the cortex and hippocampus. These regions of interest (ROI) were carefully outlined in each image using ImageJ; only plaques with an area greater than 5.5 µm² were included in the quantification. The total area of the brain sections or ROIs was measured using ImageJ to provide accurate scaling and enable comparisons across samples.

Image acquisition and analysis

Confocal images were acquired using a spinning disk confocal microscope (Olympus) and processed using the OlyVia viewer software. For each animal, one representative coronal brain section at a matched anatomical level (Bregma − 1.5 mm to − 2.5 mm) was selected for analysis based on consistent structural landmarks. For cortical analyses, the ROI encompassed the entire dorsal cortex, including the primary somatosensory cortex and the adjacent motor areas. The entire hippocampal formation was included in the hippocampal analyses. ROI boundaries were manually delineated in ImageJ based on DAPI counterstaining and anatomical reference. Each quantified image reflected marker expression across the entire ROI (not subsampled fields), ensuring standardized and comprehensive comparisons between animals. Markers including IBA1, GFAP, CD68, LAMP1, and BACE1 were quantified using marker-specific threshold settings. Synaptophysin quantification was performed by calculating the average fluorescence intensity across the entire ROI. All imaging and quantification were performed by an investigator blinded to the group identity.

Aβ measurement

Brain samples were prepared and measured as previously described [19, 21]. To extract Aβ from different fractions, brain tissue was sequentially homogenized in cold PBS and 5 M guanidine buffer, both containing 1x protease inhibitor mixture. The samples were diluted 10-fold with the sample diluent (Suzhou AstraBio Technology Co., Ltd.) and analyzed using AsT-Sc-Lite, an automated single-molecule detection system (Suzhou AstraBio Technology Co., Ltd.), according to the manufacturer’s instructions. The procedure included the following: (i) Sample loading: A 25 µL sample was mixed with Reagent 1 (0.2 mg/mL magnetic beads coated with Aβ capture antibody and protecting reagents) and incubated for 6 min. (ii) Antibody binding: Reagent 2 (Aβ detection antibody labeled with single-molecule fluorescent probes) was added, mixed, and incubated for 4 min at 40 °C. (iii) Magnetic separation: Magnetic beads were captured on the flow cell surface using a magnet. Unbound fluorophores were removed using wash buffer. (iv) Fluorescent imaging and analysis: The single-molecule signals were analyzed by the machine, and protein concentrations were calculated using a pre-prepared standard curve.

Serum estradiol measurement

Serum estradiol levels were quantified using an Estradiol Parameter Assay Kit (R&D Systems, KGE014) according to the manufacturer’s instructions. Absorbance at 450 nm was measured using a multimode microplate reader (Thermo Fisher Scientific, VL0000D0). A four-parameter logistic (4-PL) standard curve was generated, and estradiol concentrations were calculated accordingly.

Primary microglial cultures

The isolation and culture of primary mouse microglia were performed as previously described [22]. Briefly, primary mouse microglia were cultured from postnatal days 1–3 (P1-P3) using a previously established protocol. The cortices and hippocampi were dissected in ice-cold calcium- and magnesium-free Hank’s balanced salt solution (HBSS) to remove the meninges. The brain tissue was then washed with HBSS and subjected to enzymatic digestion using 0.25% trypsin-EDTA for 20 min at 37 °C, with gentle agitation every 5 min to facilitate tissue dissociation. The enzymatic reaction was stopped by adding a trypsin inhibitor (0.6 mg/mL) dissolved in DMEM, supplemented with 10% FBS and 1% P/S. The tissue was mechanically dissociated by trituration using pipettes, and the resulting cell mixture was filtered through a 70-µm cell strainer to remove any remaining clumps. The cells were pelleted at 900 × g for 5 min and resuspended in microglia growth medium consisting of DMEM, 10% FBS, 1% P/S, and 5 ng/mL granulocyte-macrophage colony-stimulating factor (GM-CSF, PeproTech, AF-300-03). The cells were then plated onto poly-L-lysine-coated T75 flasks. After 8 h, the medium was replaced and renewed every 3 days. Upon reaching confluence, microglia were detached by agitating the flasks at 200 rpm for 30 min. The supernatant was collected, centrifuged at 900 × g for 10 min, and the cell pellet was resuspended in 1 mL of medium and seeded into 12-well plates at a density of 1 × 10⁵ cells per well for further experiments. Primary microglia were treated with vehicle or 200 nM E2 for three days. The fresh medium containing the respective treatments was replenished every 3 days.

Live-cell imaging

Primary microglia were seeded into 96-well plates at a density of 1 × 10⁴ cells/well in phenol red-free DMEM supplemented with 10% charcoal-stripped fetal bovine serum, 1% penicillin–streptomycin, and 5 ng/mL GM-CSF. Cells were treated with 200 nM E2 or an equivalent volume of absolute ethanol (vehicle control). Twelve hours before the phagocytosis assay, the culture medium was replaced with fresh medium containing either E2 or vehicle to allow adaptation to treatment. For the phagocytosis assay, pHrodo™ Red Zymosan BioParticles™ (Invitrogen, P35364) were added to the culture medium at a final concentration of 50 µg/mL. The old medium in each well was aspirated and replaced with 100 µL of fresh E2- or vehicle-containing medium supplemented with pHrodo dye. The 96-well plates were placed in an IncuCyte live-cell imaging system (Sartorius, S3), and images were acquired at 10× magnification from five fields per well using both bright-field and red-fluorescence channels. Images were captured every 20 min for a total duration of 1 h. Phagocytic activity was quantified as the percentage of pHrodo-positive cells among all cells.

Bulk RNA sequencing (RNA-seq)

Primary microglia were isolated and cultured as described above [22]. Total RNA was extracted using the Monarch Total RNA Miniprep Kit, following the manufacturer’s instructions. RNA purity and concentration were measured using a NanoPhotometer (IMPLEN, N50), and RNA integrity was assessed using an Agilent 2100 Bioanalyzer (Agilent Technologies). Samples with an RNA integrity number (RIN) ≥ 7.0 were used for library preparation. Ribosomal RNA was depleted using the NEBNext rRNA Depletion Kit v2 (New England Biolabs, E7400L), followed by Revert Aid First Strand cDNA Synthesis Kit (Thermo Fisher, K1622). Sequencing libraries were prepared using the NEBNext Ultra RNA Library Prep Kit for Illumina (New England Biolabs, E7770), including end repair, adaptor ligation, and polymerase chain reaction (PCR) amplification. The library concentration was quantified using a Qubit 2.0 fluorometer (Invitrogen, Q32866), and the size distribution was confirmed using a Bioanalyzer 2100. Paired-end sequencing (2 × 150 bp) was performed using an Illumina NovaSeq 6000 platform. Raw reads were assessed using FastQC (version 0.10.1), and adapters and low-quality bases were trimmed using Trimmomatic [23]. Clean reads were aligned to the mouse reference genome (mm10) using HISAT2 [24]. Gene-level counts were generated using featureCounts [25], and differential expression analysis was performed using DESeq2 (version 1.38.3) in R [26]. Genes with an adjusted p-value (padj) < 0.01 were considered to be significantly differentially expressed. Functional enrichment analysis was performed on the identified DEGs using the R package richR (https://github.com/hurlab/richR) based on the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway database and the Gene Ontology (GO) database to identify overrepresented biological functions.

SnRNA-seq library preparation and sequencing

SnRNA-seq was performed on mouse hippocampal tissues using the 10x Genomics Chromium platform. Briefly, freshly dissected hippocampal tissues were placed on ice and homogenized in ice-cold lysis buffer (10 mM Tris-HCl, pH 8.0, 10 mM NaCl, 3 mM MgCl₂, 0.1% Nonidet P-40, 1 U/µL RNase inhibitor) using a dounce homogenizer. Homogenates were filtered through a 40 μm strainer and centrifuged at 500 × g for 5 min at 4 °C. Nuclei were resuspended in nuclei wash buffer (PBS, 1% BSA, 0.2 U/µL RNase inhibitor) and counted using trypan blue exclusion under a hemocytometer. Single-nucleus suspensions with > 90% viability were loaded onto a Chromium Controller (10x Genomics) for GEM generation and barcoding using the Chromium Single Cell 3′ Reagent Kits v3.1. Reverse transcription, cDNA amplification, and library preparation were performed according to the manufacturer’s instructions. Sequencing was conducted on an Illumina NovaSeq 6000 platform to achieve a depth of approximately 50,000–100,000 reads per nucleus. Raw base-calling files were processed using Cell Ranger (v6.1.1, 10x Genomics) with the pre-mRNA reference transcriptome (mm10) to include intronic reads. Downstream quality control, normalization, dimensionality reduction, clustering, and differential expression analyses were performed using Seurat (v5.0.1) in R.

snRNA-seq data processing

Cells were excluded based on the following criteria: <400 genes detected, > 7,000 genes detected, > 10% mitochondrial UMIs fraction. Raw counts were log-normalized for each cell using the LogNormalize method with the Seurat ‘NormalizeData’ function. Then, the 2,000 most variable genes were identified by the ‘FindVariableFeatures’ function. Gene expression values were standardized across all cells using the ‘ScaleData’ function. Batch effect correction was implemented using the Harmony R package (v1.2.1) [27]. Uniform Manifold Approximation and Projection (UMAP) was applied to harmony-corrected embeddings using Seurat’s ‘RunUMAP’ function with the first 30 principal components [28]. Graph-based clustering was performed using the Louvain algorithm (FindClusters function, resolution = 0.1) on a k-nearest-neighbor graph constructed with FindNeighbors. Cluster-defining marker genes were identified using FindAllMarkers. Cell clusters were annotated based on the expression of canonical marker genes. Within annotated cell types, DEGs between OVX and SHAM groups were identified using FindMarkers. KEGG pathway enrichment analysis was conducted using richR (https://github.com/hurlab/richR).

Western blotting

Brain tissues from two randomly selected mice were pooled to generate one sample (four samples per group) and homogenized in RIPA buffer, which contained (in mM) 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 2 mM EDTA, 1 mM PMSF, 50 mM sodium fluoride, 1 mM sodium vanadate, 1 mM DTT with 1% sodium deoxycholate, 1% SDS, and 1× complete protease inhibitor and phos-STOP phosphatase inhibitor cocktails. Total protein concentration was determined using a BCA protein assay kit (Thermo Fisher, #23225) according to the manufacturer’s instructions. Twenty-five micrograms of total protein were resolved by electrophoresis on a 4%-20% SurePAGE™ gel in Tris/MOPS/SDS running buffer (GenScript, C34382110). The proteins were then transferred to a PVDF membrane using the Trans-Blot Turbo Transfer System (Bio-Rad). Following transfer, the membranes were incubated with rapid blocking buffer (MCE, HY-K1027) for 15 min at room temperature. Primary antibodies were then applied and incubated overnight at 4 °C. After extensive washing, the membranes were incubated with HRP peroxidase-conjugated secondary antibodies in blocking buffer for 1 h at room temperature. Immunoreactive bands were visualized using a VILBER imaging system (VILBER, FUSION-FX.6EDGE/SPECTRA 7), and band density was quantified using the ImageJ software. The protein densities of interest were normalized to the loading control. For sequential blots of loading control proteins, the primary and secondary antibodies were removed by incubating the membranes in stripping buffer (Beyotime P0025N) at room temperature for 10 min on a shaker. After the stripping buffer was removed, the blots were rinsed with deionized water for 1–2 min, followed by three washes in Tris-buffered saline containing 0.25% Tween 20, each lasting 10 min. The membranes were then processed as described previously.

RT-qPCR

Total RNA was isolated from brain tissue or cultured microglia using the Monarch Total RNA Miniprep Kit (NEB, T2010S) according to the manufacturer’s instructions. Complementary DNA (cDNA) was synthesized from RNA using the Revert Aid First Strand cDNA Synthesis Kit (Thermo Fisher, K1622). Quantitative PCR (qPCR) was performed using a LightCycler 480 Instrument II (Roche) with SYBR Green detection (Roche, #04887352001). Relative expression was quantified using the comparative Ct method (ΔΔCt), with glyceraldehyde 3-phosphate dehydrogenase (Gapdh) serving as the internal reference gene. Primers for individual genes were as follows: APP (forward: 5’-CCTTC TCGTT CCTGA CAAGT GC-3’, reverse: 5’-GGCAG CAACA TGCCG TAGTC AT -3’); Bace1 (forward: 5’-TGCTG CCATC ACTGA ATCGG AC-3’, reverse: 5’-GGAAT GTGGG TCTGCT TCACC A-3’); Bace2 (forward: 5’-GATTG GTGCG ACCGT GATGG AA-3’, reverse: 5’-GTTGC TGGCT ATGTC TTCCG TG-3’); Psen1 (forward:5’-GAGAC TGGAA CACAA CCATA GCC-3’, reverse: 5’-AGAAC ACGAG CCCGA AGGTG AT-3’); Psen2 (forward: 5’-CTGGT GTTCA TCAAG TACCT GC C-3’, reverse: 5’-TTCTC TCCTG GGCAG TTTCC AC-3’); PSEN1 (forward: 5’-GCAGT ATCCT CGCTG GTGAA GA-3’, reverse: 5’-CAGGC TATGG TTGTG TTCCA GTC-3’) and GAPDH (forward: 5’-CATCA CTGCC ACCCA GAAGA CTG-3’, reverse: 5’-ATGCC AGTGA GCTTC CCGTT CAG-3’).

Statistical analysis

Statistical analyses were performed using GraphPad Prism (version 10.0; GraphPad Software). Data are presented as mean ± SEM. Outliers were identified and excluded using the ROUT method (Q = 1%). The choice of statistical test was based on the data distribution and the homogeneity of variance. For comparisons between two independent groups, an unpaired two-tailed Student’s t-test was used when the assumptions of normality and equal variances were met; Welch’s t-test was applied when variances were unequal despite normality; and the Mann–Whitney U test was used when data violated normality assumptions. For multiple comparisons between two groups across several variables, multiple unpaired t-tests were performed with Welch’s correction, and significance was determined based on false discovery rate (FDR)-adjusted q-values using the two-stage step-up method of Benjamini, Krieger, and Yekutieli (Q = 5%). One-way ANOVA was used to compare more than two groups. Effect sizes were calculated as Cohen’s d or η² and are reported in the figure legends. No statistical analysis was conducted to determine the sample size. The number of biological replicates is specified in the figure legends. Statistical significance was defined as P < 0.05 for single comparisons and q < 0.05 for multiple comparisons. Significance levels are denoted as follows: * P or q < 0.05, ** < 0.01, *** < 0.001, and **** < 0.0001.

Results

OVX exacerbates cognitive deficits in female 5xFAD mice

To investigate the impact of estrogen depletion on AD progression, OVX was performed on 2-month-old female 5xFAD mice, with sham-operated (SHAM) mice as controls (Fig. 1A). Body weight was monitored throughout the experimental period, and OVX mice showed significantly accelerated weight gain compared with SHAM mice by 30 days post-surgery, with the difference persisting through 90 days and partially declining by 120 days (6 months) (Figure S1A). ELISA analysis confirmed a marked reduction in E2 levels in the OVX group (Fig. 1B). Macroscopic observation revealed that SHAM mice had intact and distended oviducts, whereas OVX mice exhibited pronounced oviduct atrophy (Figure S1B), and quantitative analysis showed significantly reduced oviduct weight (Figure S1C). Vaginal smear staining demonstrated that 5xFAD-SHAM mice showed complete estrous cycles, whereas OVX mice remained in an anestrous state (Figure S1D), confirming the successful establishment of an estrogen-deficient state. We performed a series of behavioral tests to evaluate the effects of OVX on cognitive and emotional functions in 5xFAD mice. In the open field test, OVX mice spent less time and traveled a shorter distance in the center zone compared with SHAM mice, while total locomotor activity was unaffected (Fig. 1C–F). In the Y-maze test, OVX mice exhibited a significant reduction in spontaneous alternation percentage without changes in total arm entries (Fig. 1G, S2A). In the fear-conditioning test, OVX and SHAM mice exhibited comparable freezing behavior across all training stages (Fig. 1H). However, OVX mice displayed significantly reduced freezing during the contextual fear memory test (Fig. 1I), indicating impaired hippocampus-dependent contextual memory. In contrast, cued fear memory performance was comparable between groups (Fig. 1J), suggesting that amygdala-dependent cued fear memory remained largely intact following OVX. In the Morris water maze, both groups showed progressive decreases in escape latency during training (Figures S2B-C), with no differences in average swim speed or platform crossing number (Figures S2D–E). However, in the probe trial, OVX mice spent significantly less time in the target quadrant than SHAM mice (Figure S2F). Collectively, these findings demonstrate that estrogen deprivation significantly exacerbates cognitive deficits and anxiety-like behavior in female 5xFAD mice, supporting a critical role for estrogen in maintaining cognitive function and emotional regulation in the context of AD pathology.

Fig. 1.

OVX in female 5xFAD mice reduces estradiol levels and exacerbates behavioral deficits. A Schematic of the experimental timeline. Female 5xFAD mice underwent bilateral OVX or sham surgery at 2 months of age. Behavioral tests were conducted at 5.5 months of age, followed by tissue collection at 6 months of age. Images were created with BioRender.com. B E2 levels were significantly reduced in OVX mice compared with sham-operated controls. N = 8 mice per group, t₍₁₄₎ = 5.734, P < 0.0001, Cohen’s d = 2.90, C Representative locomotor heatmaps from the OFT for 5xFAD-SHAM and 5xFAD-OVX mice. D Quantification of the total distance in the OFT. E Quantification of the center distance in the OFT. F Quantification of the center time in the OFT. D-F, N = 8 mice per group, (D) t₍₁₄₎ = 1.292, P = 0.2174, Cohen’s d = 0.63; (E) t₍₁₄₎ = 2.169, P = 0.0478, Cohen’s d = 1.10; (F) t₍₁₄₎ = 2.895, P = 0.0118, Cohen’s d = 1.45, two-tailed unpaired t-test. G OVX 5xFAD mice exhibited reduced spontaneous alternation in the Y-maze test. N = 8 mice per group, t₍₁₄₎ = 2.775, P = 0.0149, Cohen’s d = 1.34, two-tailed unpaired t-test. H-J In the contextual fear conditioning test, OVX 5xFAD mice displayed reduced freezing behavior across different phases: acquisition (H), context recall (I), and cue recall (J). N = 8 mice per group, acquisition: Stage 1 t_(13.98) = 0.6393, q = 0.920301, Stage 2 t_(11.53) = 0.1140, q = 0.920301, Stage 3 t_(12.92) = 0.3511, q = 0.920301, Stage 4 t_(11.50) = 1.389, q = 0.771681, two-way ANOVA. context recall: t₍₁₄₎ = 2.779, P = 0.0148, Cohen’s d = 1.43, two tailed unpaired t-test. cue recall: F (3, 42) = 1.211, P = 0.3177, BL, q = 0.7318, CS1, q = 0.9974, CS2, q = 0.9828, CS3, q = 0.6224, two-way ANOVA. Data are mean ± SEM. ^*P < 0.05, ^****P < 0.0001

OVX and pharmacological blockade of estrogen signaling worsen amyloid-β pathology in female 5xFAD mice

Next, we evaluated the effect of OVX on Aβ deposition in 5xFAD mice using X34 staining to visualize amyloid plaques. Compared with SHAM controls, OVX mice exhibited a significantly increased amyloid plaque load in both hippocampus and cortex (Fig. 2A–B), suggesting that estrogen deficiency may accelerate Aβ accumulation in the brain. Although the average plaque size showed a trend toward increasing, the difference did not reach statistical significance (Fig. 2C), whereas the number of plaques was significantly higher in OVX mice (Fig. 2D). Single-molecule detection analysis further revealed that levels of soluble and insoluble Aβ40 and Aβ42 were significantly higher in OVX mice than in SHAM mice (Fig. 2E–F). To determine whether estrogen deprivation alters the relative generation of the more aggregation-prone Aβ42, we calculated the Aβ42/Aβ40 ratio in both fractions. The soluble Aβ42/Aβ40 ratio was unchanged between groups (Fig. 2G), whereas the insoluble ratio was modestly reduced in OVX mice (Fig. 2H). The unchanged soluble ratio indicates that estrogen deprivation increases total Aβ without shifting amyloidogenic processing toward Aβ42, consistent with our observation that APP expression and processing—including α-/β-CTF levels and the expression of App, Bace1/2, and Psen1/2—were unaltered by OVX (Figure S3). To further examine the role of estrogen receptor signaling in Aβ accumulation, we chronically administered estrogen receptor antagonists tamoxifen or fulvestrant to 5xFAD mice and assessed plaque burden at 8 months of age. X34 staining showed that both the tamoxifen- and fulvestrant-treated groups had significantly higher amyloid plaque loads than the vehicle group (Fig. 2I–J), indicating that prolonged disruption of estrogen receptor signaling may directly promote Aβ deposition. Together, these results demonstrate that estrogen deprivation exacerbates Aβ pathology through estrogen receptor-dependent mechanisms, establishing a causal link between loss of estrogen signaling and accelerated amyloid accumulation in female AD mice.

Fig. 2.

Estrogen deprivation aggravates amyloid pathology in 5xFAD mice. A Images showing increased amyloid plaque burden in the 5xFAD-OVX mouse brain at 6 months; scale bar = 500 μm. B-D OVX increased amyloid plaque burden (B) and plaque density (D), with no change in average plaque size (C) in 5xFAD female mice. N = 8 mice per group. Plaque burden t₍₁₄₎ = 3.515, P = 0.0034, Cohen’s d = 1.76; plaque density: t₍₁₄₎ = 2.856, P = 0.0127, Cohen’s d = 1.43; plaque size: t₍₁₄₎ = 1.837, P = 0.0876, Cohen’s d = 0.92, two-tailed unpaired t-test. E-F Levels of soluble and insoluble Aβ_1-40 and Aβ_1-42 in brain lysates measured using a single-molecule detection system. OVX increased soluble Aβ_1-40/Aβ_1-42 and insoluble Aβ_1-40/Aβ_1-42 levels, Soluble Aβ_1-40: t_(11.04) = 3.281, q = 0.00766, Cohen’s d = 1.64; Soluble Aβ_1-42: t_(7.907) = 3.735, q = 0.00766, Cohen’s d = 1.87; Insoluble Aβ_1-40: t_(13.67) = 4.942, q = 0.000488, Cohen’s d = 2.47; Insoluble Aβ_1-42: t_(13.88) = 4.254, q = 0.000859, Cohen’s d = 2.13, multiple unpaired t-test with welch correction. G-H Aβ42/Aβ40 ratio in the soluble (G) and insoluble (H) fractions. The soluble ratio did not differ between SHAM and OVX mice, whereas the insoluble ratio was reduced in OVX mice. N = 8 mice per group, soluble P = 0.0830, Cohen’s d = 0.39, two-tailed Mann-Whitney test, insoluble P = 0.0011, Cohen’s d = 1.40, two-tailed Mann-Whitney test. I Images showing increased amyloid plaque burden in 5xFAD mice treated with Tamoxifen and Fulvestrant at 9.6 months, scale bar = 500 μm. J Quantification of amyloid plaque burden in 5xFAD mice treated with Tamoxifen and Fulvestrant. N = 4 mice per group. F (2, 9) = 25.17, η²=0.8483, Tamoxifen vs. vehicle: q = 0.0011, Fulvestrant vs. vehicle: q = 0.0002, One-way ANOVA. Data are mean ± SEM. ^*P < 0.05, ^**P < 0.01, ^***P < 0.001

OVX exacerbated neuroinflammation and lysosomal dysfunction in female 5xFAD mice

We next examined whether OVX influenced glial reactivity in 5xFAD mice. Immunofluorescence staining for IBA1 revealed a significant increase in IBA1-positive microglial area in both the hippocampus and cortex of OVX mice compared with SHAM controls (Fig. 3A–C). Similarly, GFAP immunostaining showed a larger GFAP-positive astrocytic area in OVX mice in both regions (Fig. 3D–F). Subsequent western blot analyses confirmed elevated IBA1 (Fig. 3G–H) and GFAP (Fig. 3I–J) protein levels in cortical lysates of OVX female 5xFAD mice, in agreement with the immunofluorescence staining results. Double labeling with X-34 and IBA1 demonstrated that the number of plaque-associated microglia was significantly higher in OVX mice (Fig. 3K–L). We next examined LAMP1, a lysosomal marker, and found a marked increase in LAMP1-positive dystrophic neurites in the hippocampus and cortex of OVX mice (Fig. 3M–O). Although increased LAMP1 expression may reflect either enhanced microglial phagolysosomal activity or lysosomal accumulation within dystrophic neurites, previous studies have shown that plaque-associated LAMP1 predominantly localizes to dystrophic neurites and represents dysfunctional lysosomal structures surrounding amyloid plaques [29]. Therefore, in the context of the increased amyloid plaque burden observed in OVX mice, the elevated peri-plaque LAMP1 signal is more likely to reflect aggravated neuritic lysosomal pathology rather than enhanced amyloid clearance. Moreover, co-staining for CD68 and IBA1 revealed that the proportion of CD68⁺IBA1⁺ microglia surrounding amyloid plaques was significantly elevated in OVX mice (Fig. 3P–Q). Together, these findings suggest that OVX in female 5xFAD mice leads to a coordinated exacerbation of amyloid-associated neuroinflammation—encompassing both astrogliosis and microgliosis—and is concurrent with significant lysosomal dysfunction.

Fig. 3.

Estrogen deprivation enhances glial activation and plaque-associated microglial responses in 5xFAD mice. A Confocal images showing increased IBA1 immunofluorescence in the brains of OVX 5xFAD female mice. Representative coronal brain section at Bregma − 1.70 mm. Scale bar, 500 μm. B-C Quantification of the IBA1⁺ area in the hippocampus (B) and cortex (C) of 5xFAD-SHAM and 5xFAD-OVX female mice. OVX increased the IBA1⁺ area in both regions. N = 8 mice per group. Hippocampus: t₍₁₄₎ = 4.023, P = 0.0013, Cohen’s d = 2.01, two-tailed unpaired t test; Cortex: t₍₁₄₎ = 4.597, P = 0.0004, Cohen’s d = 2.30, two-tailed unpaired t-test. D Confocal images showing increased GFAP immunofluorescence in the brains of OVX 5xFAD female mice. Representative coronal brain section at Bregma − 1.82 mm. Scale bar, 500 μm. E-F Quantification of the GFAP⁺ area in the hippocampus (E) and cortex (F) of 5xFAD-SHAM and 5xFAD-OVX female mice. OVX increased the GFAP⁺ area in both regions. N = 8 mice per group. Hippocampus: t₍₁₄₎ = 2.497, P = 0.0256, Cohen’s d = 1.25, two-tailed unpaired t-test; Cortex: t₍₁₄₎ = 6.021, P < 0.0001, Cohen’s d = 3.01, two-tailed unpaired t-test. G-H Western blotting showing increased IBA1 protein levels in the brain lysates of OVX 5xFAD female mice. N = 4 mice per group, t₍₆₎ = 2.801, P = 0.0311, Cohen’s d = 1.93, normalized to α-tubulin, expressed as fold change relative to the SHAM group, two-tailed unpaired t test. I-J Western blots showing increased GFAP protein levels in brain lysates of OVX 5xFAD female mice. N = 4 mice per group. t₍₆₎ = 9.152, P < 0.0001, Cohen’s d = 6.47, normalized to GAPDH, expressed as fold change relative to the SHAM group, two-tailed unpaired t test. K Confocal images showing an increased number of plaque-associated microglia in OVX 5xFAD female mice. Scale bar, 50 μm. L Quantification of plaque-associated microglia in 5xFAD-SHAM and 5xFAD-OVX female mice. For each animal, all amyloid plaques with an area of 190–210 μm² were included in the analysis. Microglia were defined as plaque-associated if their soma was located within a 40 μm radius from the center of each plaque. The number of plaque-associated microglia was averaged across all qualifying plaques to generate a single value for each animal. OVX increased the number of plaque-associated microglia in the hippocampus. N = 8 mice per group, t_(9.383) = 2.318, P = 0.0445, Cohen’s d = 1.16, two-tailed unpaired t-test with Welch’s correction. M Confocal images showing increased LAMP1 and X34 staining in the brains of OVX 5xFAD female mice. Representative coronal brain section at Bregma − 2.24 mm. Scale bar, 500 μm. N-O Quantification of the LAMP1⁺ area in the hippocampus and cortex of OVX versus SHAM 5xFAD female mice. N = 8 mice per group. Hippocampus: t₍₁₄₎ = 3.927, P = 0.0018, Cohen’s d = 1.91, two-tailed unpaired t-test; Cortex: t₍₁₄₎ = 3.541, P = 0.0033, Cohen’s d = 2.04, two-tailed unpaired t-test; P Confocal images showing increased CD68 and IBA1 staining in the brain of OVX 5xFAD female mice. Representative coronal brain section at Bregma − 2.46 mm. Scale bar, 500 μm. Q Quantification of the CD68⁺ and IBA1⁺ areas in the hippocampus of OVX versus SHAM 5xFAD female mice. N = 8 mice per group. t₍₁₄₎ = 5.451, P = 0.0001, Cohen’s d = 2.80, two-tailed unpaired t-test. Data are mean ± SEM. ^*P < 0.05, ^**P < 0.01, ^***P < 0.001, ^****P < 0.0001

Estrogen deprivation reduces synaptic density and synaptic markers in female 5xFAD mice

Synaptic loss is a key pathological correlate of cognitive decline in AD [30]. To determine whether estrogen deprivation compromises synaptic integrity, we assessed synaptic density by immunostaining for the postsynaptic marker PSD95 and the presynaptic active-zone marker Bassoon, and quantified their colocalized puncta as a measure of structural synapses (Fig. 4A). The density of PSD95/Bassoon co-localized puncta was significantly reduced in the hippocampus of OVX mice compared with SHAM controls (Fig. 4B). Consistent with this, western blot analysis of the brain lysates revealed significantly lower levels of both the postsynaptic protein PSD95 (Fig. 4C–D) and the presynaptic vesicle protein synaptophysin (SYP; Fig. 4C-E) in OVX mice. Together, these results demonstrate that estrogen deprivation reduces synaptic density and synaptic protein levels in female 5xFAD mice, providing a structural correlate for the observed cognitive deficits.

Fig. 4.

OVX decreases synaptic puncta density and synaptic protein levels in female 5xFAD mice. A Representative confocal images of PSD95 and Bassoon immunostaining in the hippocampus of 5xFAD-SHAM and 5xFAD-OVX mice; co-localized puncta (circles) represent structural synapses. Scale bar, 5 μm. B Quantification of PSD95/Bassoon co-localized puncta, expressed as a percentage of the SHAM group. N = 8 mice per group, P = 0.0003, t₍₁₄₎ = 4.830, Cohen’s d = 2.30, two-tailed unpaired t-test. C Representative western blots of PSD95 and synaptophysin (SYP) protein levels in the brain lysates of 5xFAD-SHAM and 5xFAD-OVX mice. D-E Quantification of PSD95 (D) and SYP (E) protein levels, normalized to α-Tubulin and expressed as fold change relative to SHAM. N = 4 mice per group; PSD95: P = 0.0245, t₍₆₎ = 2.983, Cohen’s d = 2.10, two-tailed unpaired t-test; SYP: P = 0.0286, Cohen’s d = 3.39, two-tailed Mann-Whitney test. Data are mean ± SEM. ^*P < 0.05, ^**P < 0.01, ^***P < 0.001

Estradiol treatment alters the transcriptomic profile of primary microglia and enhances phagocytic capacity

To investigate the mechanisms underlying the increased Aβ pathology observed in OVX female 5xFAD mice, we first examined gene expression involved in Aβ production by quantitative PCR. These included human APP and PSEN1, as well as the mouse genes App, Psen1, Psen2, Bace1, and Bace2. OVX did not alter the expression of any of these genes (Figure S3A). We further analyzed APP processing by Western blot and found that OVX did not affect the levels of full-length APP or its cleavage products, including α-CTF and β-CTF (Figures S3B-E), suggesting that OVX-induced gliosis is not associated with altered APP expression or processing. Given that microglial phagocytosis of Aβ is a critical mechanism for Aβ clearance, we next investigated the effects of E2 on microglial function. We isolated primary microglia from P1-P3 C57BL/6 mice and treated them with E2 or vehicle control, followed by bulk RNA sequencing. Differential expression analysis identified 239 genes significantly altered by E2 treatment, with 160 upregulated and 79 downregulated (Fig. 5A). KEGG pathway enrichment analysis revealed that these differentially expressed genes were predominantly enriched in immune-related pathways, including phagosome, efferocytosis, infectious disease responses, and immune signaling pathways (TNF, MAPK, JAK-STAT), as well as metabolic pathways, including lipid and amino acid metabolism (Fig. 5B). Given the prominent enrichment of phagocytosis-related pathways, we examined the expression of genes within the phagosome pathway. Hierarchical clustering showed that E2 treatment upregulated several key phagocytosis-associated genes, including Eea1, Scarb1, Calr, and Fcgr1 (Fig. 5C), suggesting enhanced phagocytic machinery in E2-treated microglia. To functionally validate these transcriptomic changes, we performed pHrodo phagocytosis assays and found that E2 treatment significantly increased microglial phagocytic capacity compared to vehicle controls (Fig. 5D-E). Collectively, these results demonstrate that E2 treatment enhances the phagocytic function of primary microglia by upregulating transcription of phagocytosis-related genes, suggesting an important role for estradiol in regulating microglial clearance capacity.

Fig. 5.

Estradiol enhances microglial phagocytic function through transcriptional upregulation of phagocytosis-related genes. A Quantitative bulk RNA sequencing analysis of primary microglia treated with E2 and a volcano plot illustrating downregulated and upregulated genes. (P-value < 0.01). N = 3 mice per group. B KEGG pathway enrichment analysis of differentially expressed genes. C Heatmap of differentially expressed genes enriched in the phagocytosis pathway. D Live-cell imaging of pHrodo uptake by primary microglia following E2 or vehicle treatment. Scale bar = 20 μm. N = 6 per group from six independent experiments. E Quantification of phagocytosis efficiency in primary microglia treated with vehicle or E2, t₍₁₀₎ = 2.427, P = 0.0357, Cohen’s d = 1.39, two-tailed unpaired t-test. Data are mean ± SEM. ^*P < 0.05

Single-nucleus transcriptomics reveals cell type-specific transcriptional reprogramming in response to estrogen deprivation

The hippocampus is a core region of AD pathology and underlies the learning and memory functions that were impaired in our behavioral tests following estrogen deprivation. We therefore performed snRNA-seq on hippocampal tissue from 5xFAD-OVX and 5xFAD-SHAM mice to elucidate the molecular mechanisms underlying estrogen deprivation-induced AD pathology. Higher-resolution clustering resolved 15 transcriptionally distinct subclusters based on canonical marker gene expression (Fig. 6A-B, S4B), revealing the cellular heterogeneity within the hippocampus of these AD model mice. Unbiased clustering and cell type annotation identified five major brain cell populations: neurons, oligodendrocytes, microglia, oligodendrocyte precursor cells (OPCs), and astrocytes (Fig. 6C, S4A). Focusing on neuronal populations, we further delineated excitatory and inhibitory subtypes, which exhibited comparable proportions in the OVX and SHAM groups (Fig. 6D-E). Within the microglial compartment, we identified both homeostatic microglia and disease-associated microglia (DAM), with the latter population markedly expanded in OVX mice compared to controls (Figures S4C-D), consistent with enhanced neuroinflammatory responses following estrogen depletion. Differential gene expression analysis revealed extensive transcriptional reprogramming across multiple cell types in response to OVX. Excitatory neurons exhibited the most profound transcriptional changes, with 514 genes upregulated and 1,807 genes downregulated (Fig. 6F). Microglia showed 28 upregulated and 1 downregulated gene (Figure S4E), oligodendrocytes displayed 68 upregulated and 8 downregulated genes (Figure S4F), and inhibitory neurons had 26 upregulated and 1 downregulated gene (Figure S4G). Gene Ontology (GO) enrichment analysis of upregulated genes in excitatory neurons revealed significant enrichment of biological processes related to synapse organization and regulation, postsynaptic organization, dendritic morphogenesis and development, neuron projection organization, synaptic vesicle cycle, and cognition-related pathways (Fig. 6G), suggesting compensatory transcriptional responses to maintain synaptic integrity and cognitive function under estrogen-depleted conditions. To validate the increased Apoe expression observed in our snRNA-seq data (Figure S5A), we performed immunofluorescence staining and western blot analysis. Immunofluorescence confirmed elevated ApoE protein levels specifically in NeuN-positive neurons of OVX mice compared to SHAM controls (Figures S5B-C), and western blot analysis further demonstrated significant upregulation of ApoE protein in hippocampus lysates from OVX mice (Figures S5D-E), corroborating the transcriptomic findings. Collectively, these comprehensive transcriptomic analyses reveal that estrogen deprivation triggers widespread cellular and molecular reprogramming in the AD brain, with particularly prominent transcriptional changes in excitatory neurons that affect synaptic and cognitive pathways, as well as upregulation of AD-related genes that may contribute to disease pathology.

Fig. 6.

Single-nucleus transcriptomics reveals widespread transcriptional reprogramming in estrogen-deprived 5xFAD hippocampus with prominent alterations in excitatory neurons. A UMAP projection of single-nucleus transcriptomes from 5xFAD SHAM and OVX mouse brains. Single-nucleus transcriptomes were projected onto a two-dimensional UMAP space and grouped into 15 clusters. B Dot plot showing the expression patterns of canonical marker genes across the identified clusters in the SHAM and OVX 5xFAD mouse brains. C Cell type composition of brain samples from SHAM and OVX 5xFAD mice revealed by single-nucleus RNA sequencing. D UMAP visualization of excitatory and inhibitory neuronal subtypes in SHAM and OVX 5xFAD mouse brains. E Proportions of excitatory and inhibitory neurons in the SHAM and OVX 5xFAD mouse brains. F Volcano plot showing differentially expressed genes (DEGs) in excitatory neurons comparing OVX and SHAM 5xFAD mice. G Bubble plot displaying Gene Ontology (GO) enrichment results of differentially expressed genes (DEGs) in excitatory neurons from OVX compared to SHAM 5xFAD mice

Partial genetic reduction of Ctss reverses OVX-induced exacerbation of Aβ pathology in female 5xFAD mice

Our snRNA-seq analysis revealed a prominent upregulation of Ctss expression in the hippocampal tissue of 5xFAD-OVX mice compared with 5xFAD-SHAM controls (Fig. 7A). To validate this transcriptomic finding, we performed western blot analysis on hippocampal lysates and confirmed that CTSS protein levels were significantly elevated in OVX mice(Fig. 7B-C), corroborating the RNA-seq results. Given the established role of CTSS in neuroinflammation and lysosomaldysfunction, and its robust upregulation following estrogen deprivation, we hypothesized that CTSS may serve as acritical mediator of the exacerbation of OVX-induced AD pathology. To test this hypothesis, we generated 5xFAD;Ctss^+/-mice, in which Ctss was heterozygously deleted, by crossing 5xFAD mice with Ctss^-/-mice. These mice were subjected to either OVX or SHAM surgery at 2 months of age, and pathological assessments were performed at 6 months (Fig.7D). Western blot analysis confirmed that CTSS protein levels were reduced to approximately 50% of control levels in Ctss^+/-mice (Figures S6A-B), validating the efficiency of heterozygous Ctss deletion. Given our previous findingthat OVX increased Aβ plaque deposition in 5xFAD mice (Fig. 2A), we asked whether CTSS reduction could reversethis OVX-induced exacerbation of amyloid pathology. To test this, we performed X34 staining in 5xFAD-SHAM,5xFAD;Ctss^+/-–SHAM, and 5xFAD;Ctss^+/-–OVX mice. In contrast to the increase in plaque burden produced by OVX in CTSSintact5xFAD mice (Fig. 2), OVX did not increase plaque burden when CTSS was genetically reduced; 5xFAD;Ctss^+/-–OVX mice did not differ from 5xFAD;Ctss^+/-–SHAM controls (Fig. 7E). Quantitative analysis confirmed that 5xFAD;Ctss^+/-–OVX mice showed no significant differences from 5xFAD;Ctss^+/-–SHAM controls in terms of amyloid plaque load (Fig.7F), average plaque size (Fig. 7G), or plaque number (Fig. 7H), demonstrating that CTSS reduction prevents OVX induced Aβ accumulation. Collectively, these results demonstrate that genetic reduction of CTSS prevents OVX-inducedamyloid accumulation, supporting CTSS as a key downstream mediator of estrogen deprivation–inducedpathology and a promising therapeutic target for mitigating AD risk in postmenopausal women.

Fig. 7.

Genetic reduction of CTSS prevents OVX-induced amyloid accumulation in 5xFAD mice. A UMAP plot showing Ctss expressionin excitatory neurons of SHAM and OVX 5xFAD mice. B-C Western blots showing increased CTSS protein levels in the brain lysates of OVX 5xFAD female mice. N = 4 mice per group, t (6) = 5.024, P = 0.0024, Cohen’s d = 3.49, normalized to α-Tubulin, expressed as foldchange relative to the SHAM group, two-tailed unpaired t-test. D Schematic illustration of the experimental design for 5xFAD;Ctss heterozygous knockout mice. Images were created with BioRender.com E Representative images of X34 staining showing Aβ pathologyin 5xFAD-SHAM, 5xFAD; Ctss⁺/⁻-SHAM, and 5xFAD; Ctss⁺/⁻-OVX mice. Scale bar, 500 μm. F–H Quantification of Aβ plaque burden (F), average plaque size (G), and plaque number (H). All three amyloid plaque measures were significantly lower in 5xFAD; Ctss⁺/⁻-SHAM mice than in 5xFAD-SHAM mice, whereas no significant differences were observed between 5xFAD; Ctss⁺/⁻-SHAM and 5xFAD;Ctss⁺/⁻-OVX mice. N = 4 mice per group, Aβ plaque burden: F (2,9) = 15.94, P = 0.0011, 5xFAD-SHAM v.s 5xFAD; Ctss⁺/⁻-SHAM: Adj-P = 0.0023; 5xFAD; Ctss⁺/⁻-SHAM v.s 5xFAD; Ctss⁺/⁻-OVX: Adj-P = 0.8555, one-way ANOVA test; Aβ plaque size: F (2,9) = 7.779, P= 0.0109, 5xFAD-SHAM v.s 5xFAD; Ctss⁺/⁻-SHAM: Adj-P = 0.0119; 5xFAD; Ctss⁺/⁻-SHAM v.s 5xFAD; Ctss⁺/⁻-OVX: Adj-P = 0.9639,one-way ANOVA test; Aβ plaque number: F (2,9) = 6.247, P = 0.0199, 5xFAD-SHAM v.s 5xFAD; Ctss⁺/⁻-SHAM: Adj-P = 0.0367, 5xFAD;Ctss⁺/⁻-SHAM v.s 5xFAD; Ctss⁺/⁻-OVX: Adj-P = 0.8626, one-way ANOVA test. Data are presented as mean ± SEM, *P<0.05,**P<0.01

Genetic reduction of Ctss prevents OVX-induced neuroinflammation and lysosomal dysfunction in female 5xFAD mice

We previously observed that OVX increased microglial and astrocytic activation and elevated the LAMP1⁺ lysosomal area in CTSS-intact 5xFAD mice (Fig. 3). We therefore asked whether genetic reduction of CTSS could prevent these changes. Using 5xFAD;Ctss⁺/⁻-SHAM mice as the reference group, we found that, compared with 5xFAD;Ctss⁺/⁻-SHAM mice, 5xFAD-SHAM mice (with intact CTSS) showed a significantly larger LAMP1⁺ area (Fig. 8A–B), IBA1⁺ microglial area (Fig. 8C–D), and GFAP⁺ astrocytic area (Fig. 8E–F), indicating that partial reduction of CTSS lowers lysosomal and glial reactivity in the 5xFAD brain. Importantly, OVX did not increase any of these measures when CTSS was genetically reduced: 5xFAD;Ctss⁺/⁻-OVX mice showed no significant difference from 5xFAD;Ctss⁺/⁻-SHAM mice in LAMP1⁺ (Fig. 8A–B), IBA1⁺ (Fig. 8C–D), or GFAP⁺ area (Fig. 8E–F), in contrast to the increases produced by OVX in CTSS-intact 5xFAD mice (Fig. 3). Western blot analysis of the brain lysates confirmed this pattern at the protein level, both GFAP and IBA1 were markedly lower in the two Ctss⁺/⁻ groups than in 5xFAD-SHAM mice, with no difference between 5xFAD;Ctss⁺/⁻-OVX and 5xFAD;Ctss⁺/⁻-SHAM mice (Fig. 8G–J). Together, these results indicate that partial genetic reduction of Ctss attenuates lysosomal dysfunction and glial activation and prevents their exacerbation by estrogen deprivation.

Fig. 8.

Partial CTSS reduction lowers glial activation and lysosomal markers and prevents their exacerbation by OVX in 5xFAD mice. A Representative confocal image of X34 and LAMP1 staining X-34 and LAMP1 immunofluorescence in the brain of 5xFAD-SHAM, 5xFAD;Ctss⁺/⁻-SHAM, and 5xFAD; Ctss⁺/⁻-OVX mice. Scale bar, 500μm. B Quantification of LAMP1⁺ area. N = 4 mice per group; F (2,9) =9.203, P = 0.0067, 5xFAD-SHAM v.s 5xFAD; Ctss⁺/⁻-SHAM: Adj-P = 0.0049, 5xFAD; Ctss⁺/⁻-SHAM v.s 5xFAD; Ctss⁺/⁻-OVX: Adj-P =0.5527, one-way ANOVA test. C Representative confocal images of IBA1 staining in the brains of the three groups; scale bar, 500μm. D Quantification of IBA1⁺ area. N = 4 mice per group; F (2,9) = 6.064, P = 0.0215, 5xFAD-SHAM v.s 5xFAD; Ctss⁺/⁻-SHAM: Adj-P =0.0173, 5xFAD; Ctss⁺/⁻-SHAM v.s 5xFAD; Ctss⁺/⁻-OVX: Adj-P = 0.7603, one-way ANOVA test. E Representative confocal images of GFAP staining in the brains of the three groups. Scale bar, 500μm. F Quantification of GFAP+ area. N = 4 mice per group. F (2,9) =10.40, P = 0.0046, 5xFAD-SHAM v.s 5xFAD; Ctss⁺/⁻-SHAM: Adj-P < 0.0085, 5xFAD; Ctss⁺/⁻-SHAM v.s 5xFAD; Ctss⁺/⁻-OVX: Adj-P =0.8950, one-way ANOVA test. G Representative western blots of GFAP in the brain lysates of the three groups. H Quantification ofGFAP protein levels, normalized to GAPDH and expressed as fold change relative to the 5xFAD-SHAM group. N = 4 mice per group; F(2,9) = 247.3, P < 0.0001, 5xFAD-SHAM v.s 5xFAD; Ctss⁺/⁻-SHAM: Adj-P < 0.0001, 5xFAD; Ctss⁺/⁻-SHAM v.s 5xFAD; Ctss⁺/⁻-OVX: Adj-P = 0.6992, one-way ANOVA test. Ι Representative western blots of IBA1 in the brain lysates of the three groups. J Quantification of IBA1 protein levels, normalized to α-Tubulin and expressed as fold change relative to the 5xFAD-SHAM group. N = 4 mice per group;4F (2,9) = 4.929, P = 0.0358, 5xFAD-SHAM v.s 5xFAD; Ctss⁺/⁻-SHAM: Adj-P = 0.321, 5xFAD; Ctss⁺/⁻-SHAM v.s 5xFAD; Ctss⁺/⁻-OVX:P = 0.8986, one-way ANOVA test. Data are presented as mean ± SEM. *P<0.05, **P<0.01, ****P<0.0001

Genetic reduction of Ctss protects against OVX-induced synaptic and cognitive impairment in female 5xFAD mice

Having shown that OVX reduces synaptic density and synaptic markers in CTSS-intact 5xFAD mice (Fig. 4), we next asked whether reducing CTSS could prevent these changes, using 5xFAD;Ctss⁺/⁻-SHAM mice as the reference group. At the structural level, 5xFAD-SHAM mice showed a lower density of PSD95/Bassoon co-localized puncta than 5xFAD;Ctss⁺/⁻-SHAM mice (Fig. 9A–B), and western blot analysis revealed a similar pattern for PSD95 and SYP protein levels (Fig. 9C–E), indicating that partial reduction of CTSS increases synaptic markers in the 5xFAD brain—consistent with the established role of CTSS in synaptic pruning [31]. Critically, OVX did not reduce these measures when Ctss was genetically lowered: 5xFAD;Ctss⁺/⁻-OVX mice showed no significant difference from 5xFAD;Ctss⁺/⁻-SHAM mice in PSD95/Bassoon puncta density (Fig. 9A–B) or in PSD95 and SYP levels (Fig. 9C–E), in contrast to the reduction produced by OVX in CTSS-intact mice (Fig. 4). Consistent with this synaptic protection, 5xFAD;Ctss⁺/⁻-OVX mice showed no behavioral deficit relative to 5xFAD;Ctss⁺/⁻-SHAM mice in Y-maze spontaneous alternation (Figure S7A) or in fear conditioning across all phases (Fig. S7B-D). Together, these results indicate that partial genetic reduction of CTSS prevents the synaptic and cognitive impairment induced by estrogen deprivation, identifying CTSS as a key downstream mediator of OVX-induced synaptic decline.

Fig. 9.

Genetic reduction of Ctss protects against OVX-induced synaptic decline in female 5xFAD mice. A Representative confocalimages of PSD95 and Bassoon immunostaining in the hippocampus of 5xFAD-SHAM, 5xFAD; Ctss⁺/⁻-SHAM, and 5xFAD; Ctss⁺/⁻-OVX mice; co-localized puncta (circles) denote structural synapses. Scale bar, 5 μm. B Quantification of PSD95/Bassoon co-localizedpuncta, expressed as a percentage of the 5xFAD-SHAM group. N = 4 mice per group; F (2,9) = 13.69, P = 0.0019, 5xFAD-SHAM v.s5xFAD; Ctss⁺/⁻-SHAM: Adj-P = 0.0015, 5xFAD; Ctss⁺/⁻-SHAM v.s 5xFAD; Ctss⁺/⁻-OVX, Adj-P = 0.5773, one-way ANOVA test. C Representative western blots of PSD95 and SYP in the brain lysates of the three groups. D-E Quantification of PSD95 (D) and SYP(E) protein levels, normalized to α-Tubulin and expressed as fold change relative to the 5xFAD-SHAM group. N = 4 mice per group; F(2,9) = 53.56, P < 0.0001, 5xFAD-SHAM v.s 5xFAD; Ctss⁺/⁻-SHAM: Adj-P = 0.0014, 5xFAD; Ctss⁺/⁻-SHAM v.s 5xFAD; Ctss⁺/⁻-OVX,Adj-P = 0.0005, one-way ANOVA test. Data are mean ± SEM. *P < 0.05, **P < 0.01, ***P < 0.001

Discussion

AD exhibits marked sex differences, with women showing a higher lifetime risk and greater pathological burden than men [4, 32, 33], and accumulating evidence indicates that estrogen signaling is an important determinant of this difference, particularly during aging and the menopausal transition [4, 34]. In the present study, we demonstrate that estrogen deprivation aggravates amyloid deposition, neuroinflammation, lysosomal abnormalities, synaptic impairment, and behavioral deficits in female 5xFAD mice, providing direct experimental support for the idea that estrogen is not merely associated with, but functionally involved in, limiting AD-like pathological progression in the female brain. A particular strength of our approach is the use of multiple complementary models: in addition to OVX, chronic pharmacological blockade of estrogen receptor signaling with tamoxifen or fulvestrant similarly increased amyloid burden, indicating that the observed aggravation is closely linked to disruption of estrogen receptor-dependent signaling rather than to a nonspecific consequence of surgical intervention. This multi-model validation strengthens the internal consistency of our conclusions and enhances their translational relevance, particularly given the widespread clinical use of selective estrogen receptor modulators [35]. Together with prior studies showing that estrogen can modulate Aβ metabolism, synaptic plasticity, and neuroimmune responses [36–39], our results support a model in which intact estrogen signaling helps preserve a less amyloidogenic and less inflammatory brain state, such that estrogen loss creates a cerebral environment more permissive to Aβ accumulation and inflammatory amplification [4, 32–39].

Although our study was not designed to identify the specific estrogen receptor responsible, our data allow some informed speculation. Estradiol signals through the classical nuclear receptors ERα and ERβ and through the membrane receptor GPER1 [40]. Notably, both pharmacological agents we used—tamoxifen and fulvestrant—antagonize ERα and ERβ yet have been reported to retain agonist activity at GPER1 [41], and both reproduced the amyloid-exacerbating effect of OVX. This pattern suggests that the protective action of estrogen is mediated predominantly by the classical ERα and/or ERβ receptors rather than by GPER1, since activation of GPER1 by these compounds did not confer protection. Among the classical receptors, ERβ is particularly relevant to the microglial mechanism we describe: ERβ is enriched in microglia, and its activation has recently been shown to limit amyloid pathology through microglia-mediated amyloid scavenging without altering APP processing [42], closely paralleling our findings. ERα, which is also highly expressed in the hippocampus and exerts broad neuroprotective and anti-inflammatory effects [43], is likely to contribute as well, particularly to the observed neuronal and synaptic alterations. We therefore favor a model in which the effects of estrogen deprivation are mediated mainly by loss of ERα/ERβ signaling—with ERβ being especially pertinent to the microglial phenotype—while a contribution of GPER1 cannot be formally excluded. Definitive attribution will require receptor-selective agonists (e.g., PPT for ERα, DPN for ERβ, and G1 for GPER1) or cell-type-specific receptor deletion, both of which we identify as important directions for future work.

The magnitude of the effect of ovarian hormone loss on amyloid pathology varies across studies and models. In the APP^NL−G−F knock-in model, for example, OVX increased soluble cortical Aβ and insoluble hippocampal Aβ but otherwise had relatively limited effects on pathology [42]. The more pronounced exacerbation observed in the present study likely reflects model- and design-related factors. 5xFAD carries five familial AD mutations and aggressively overexpresses mutant APP/PSEN1, producing rapid, high-burden amyloidosis that may be especially sensitive to further reductions in clearance capacity, whereas knock-in models such as APP^NL−G−F express APP at near-physiological levels and develop pathology more gradually. Differences in the duration of estrogen deprivation, the age at assessment, and the specific endpoints examined are likely to contribute further to this variability. Importantly, despite these differences in magnitude, the underlying mechanism converges: as in our study, a recent study reported that estrogen signaling does not alter APP processing and instead acts through microglia-mediated amyloid clearance [42]. This concordance reinforces our conclusion that estrogen deprivation aggravates amyloid pathology primarily by impairing microglial clearance rather than by enhancing Aβ production.

Our biochemical analyses indicate that estrogen deprivation increases the overall Aβ burden without altering amyloidogenic processing. Both Aβ40 and Aβ42 rose in the soluble and insoluble fractions, and the soluble Aβ42/Aβ40 ratio—the fraction most representative of newly produced peptide—was unchanged, in line with the unaltered expression and processing of APP and the β-/γ-secretase machinery. These observations argue against a shift in γ-secretase cleavage specificity and instead point to impaired clearance as the principal driver of Aβ accumulation, consistent with the reduced microglial phagocytic capacity observed following estrogen withdrawal. The modest reduction in the insoluble Aβ42/Aβ40 ratio more likely reflects the composition of the deposited pool than a change in production: as total amyloid burden increases, Aβ40 is progressively incorporated into compact/dense-core and vascular deposits [44, 45], lowering the relative proportion of Aβ42 within the insoluble fraction. Notably, the absolute levels of both Aβ40 and Aβ42 were elevated in OVX mice, so this compositional shift occurs within an overall markedly increased amyloid load and does not indicate reduced Aβ42 pathology.

Our comprehensive snRNA-seq analysis provides an important molecular framework for understanding how estrogen deprivation reshapes the AD brain. Rather than revealing a single isolated alteration, the transcriptomic data indicate broad cellular reprogramming across multiple hippocampal cell populations, with especially prominent changes in excitatory neurons. The enrichment of pathways related to synaptic organization, dendritic structure, and cognition suggests that estrogen loss triggers widespread neuronal stress responses and possibly compensatory programs. In this respect, the transcriptomic analysis is valuable not only for candidate discovery but also for showing that estrogen deprivation induces a coordinated pathological shift in the hippocampus. This interpretation is consistent with previous single-cell and single-nucleus transcriptomic studies in AD, which have highlighted cell-type-specific molecular remodeling as a central feature of disease progression [15, 46]. In our dataset, the upregulation of Ctss and Apoe was particularly notable because it links estrogen deprivation to pathways implicated in proteostasis, neuroinflammation, and AD susceptibility [47–54].

Our finding that estrogen deprivation increases neuronal ApoE may at first appear to conflict with reports that estrogen up-regulates ApoE expression [55, 56]. However, the regulation of Apoe by estrogen is neither unidirectional nor context-independent. Notably, Wang et al. demonstrated that the two estrogen receptors exert opposing effects in the hippocampus—ERα activation increases, whereas ERβ activation decreases Apoe expression—so the net effect of estrogen on Apoe depends on the local balance of receptor signaling [56]. Furthermore, the classic evidence for estrogen-induced Apoe up-regulation derives largely from peripheral or plasma Apoe in healthy, non-pathological contexts [55], whereas our measurements reflect neuronal Apoe within an AD brain undergoing marked amyloid accumulation and neuroinflammation. Because neurons up-regulate Apoe as part of a reactive response to injury, the increase we observe is most parsimoniously interpreted as a downstream consequence of the aggravated pathology and glial activation that follow estrogen loss, rather than a direct transcriptional de-repression of the Apoe gene. The cell-type-, region-, and disease-state-specific nature of Apoe regulation thus reconciles our observations with the prior literature.

The consequences of elevated ApoE also differ between mice and humans. Mice express a single Apoe gene and lack the human ApoE2/ApoE3/ApoE4 isoform polymorphism that dominates human AD risk. In murine amyloid models, Apoe is a critical facilitator of fibrillar Aβ deposition—genetic reduction of Apoe markedly decreases plaque burden—so an increase in Apoe is expected to promote, rather than mitigate, amyloid accumulation, consistent with the worsened pathology we observed after OVX [48, 49]. In humans, by contrast, risk is dominated by isoform identity (ApoE4 conferring risk and ApoE2 protection) rather than expression level alone; the dose-dependent, pro-amyloidogenic effect of elevated murine Apoe should therefore be extrapolated to the human condition with caution—a limitation of the model that we now acknowledge.

Among the molecules altered after OVX, CTSS emerged as the most compelling mechanistic candidate. CTSS is a lysosomal cysteine protease with established roles in antigen processing, proteolytic regulation, and microglial function [57–59]. Increasing evidence also suggests that dysregulated cathepsin signaling contributes to neurodegeneration and inflammatory brain disease [60–62]. In our study, Ctss was upregulated after estrogen deprivation, and partial genetic reduction of CTSS markedly attenuated OVX-induced increases in amyloid burden, glial activation, lysosomal abnormalities, and synaptic protein alterations. These findings position CTSS as an important downstream effector of estrogen deprivation-associated pathology. At the same time, our data do not suggest that CTSS is the sole mechanism involved; rather, they support the idea that CTSS occupies a key node linking estrogen loss to maladaptive neuroimmune and proteostatic responses. This interpretation is strengthened by recent evidence showing that neuronal CTSS can drive neuroinflammation and cognitive decline through CX3CL1–CX3CR1 and JAK2–STAT3 signaling in aging and AD models [16]. Our findings also fit with the broader concept that lysosomal dysfunction is a central feature of AD pathogenesis [63–68]. From a translational perspective, identifying CTSS as a druggable downstream target is particularly notable because it raises the possibility of intervening in estrogen-loss-associated pathology without directly manipulating systemic hormone levels.

Limitations

This study has several limitations. First, the 5xFAD model robustly recapitulates amyloid pathology but does not fully capture the complexity of sporadic late-onset AD in humans. Second, our transcriptomic analysis was performed on hippocampal tissue; therefore, the identified molecular changes should not be overgeneralized to the entire brain. Third, although our genetic data demonstrate that partial reduction of CTSS is sufficient to attenuate OVX-induced pathological exacerbation, the precise cell type(s) in which CTSS acts and the downstream pathways linking CTSS to lysosomal dysfunction, gliosis, and amyloid accumulation remain to be defined. Finally, while our findings support a mechanistic link between estrogen deprivation and AD-related pathological worsening, additional studies will be needed to determine how these observations relate to natural reproductive aging and menopause in humans. Despite these limitations, the present work supports a model in which estrogen deprivation promotes coordinated molecular and cellular changes that amplify AD-like pathology, with CTSS emerging as an important downstream mediator in this process.

Conclusions

In summary, our study provides comprehensive evidence that estrogen deprivation exacerbates AD pathology through multiple mechanisms, with CTSS emerging as a critical mediator. The transcriptional reprogramming induced by estrogen loss affects multiple cell types and pathways, with particularly striking changes in neuronal synaptic genes, ApoE expression, and lysosomal proteases. The complete rescue of OVX-induced pathology by CTSS reduction establishes a causal role for this protease in linking estrogen signaling to AD progression and identifies a novel therapeutic target that could be exploited to reduce AD risk in postmenopausal women. These findings advance our understanding of sex differences in AD, provide mechanistic insights into the consequences of menopause, and open new avenues for therapeutic intervention that could benefit the millions of women at elevated risk for this devastating disease.

Authors’ contributions

R.Y.: Resources, Investigation, Methodology, Validation, Visualization, Writing-original draft, Writing-review & editing. L.S.: Methodology, Investigation, Validation, Visualization. F. H.: Investigation, Methodology, Validation, Data curation. Y. H.: Methodology, Investigation, Validation. M. Y.: Data curation. Q.W.: Methodology, Investigation. S.L.: Resources, Supervision. K. G.: Data curation, Software, Investigation, Validation. H. Z.: Conceptualization, Resources, Validation, Supervision, Funding acquisition, Writing-original draft, Writing – review & editing.

Funding

This work was supported by the National Natural Science Foundation of China (Grant No. 82271472), the Lingang Laboratory (Grant No. LG-GG-202401-ADAD060100 and LG-GG-202401-ADA060200), the Science and Technology Research Program of Chongqing Municipal Education Commission (KJQN202200479), the Natural Science Foundation of Chongqing (CSTB2022NSCQ-LZX0033), and CQMU Program for Youth Innovation in Future Medicine (W0158).

Data availability

The bulk RNA-seq and snRNA-seq data generated in this study were deposited in the GEO database under the accession numbersGSE325380 (RNA-seq) and GSE325656 (snRNA-seq). All datasets used and/or analyzed during the current study are available fromthe corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

All animal experimental procedures were conducted in accordance with the National Institutes of Health Guidelines for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committee of Chongqing Medical University (IACUC-CQMU-2024-0931). All cell lines used in this study were obtained from commercial sources.

Consent for publication

All authors have reviewed and approved the manuscript.

Competing interests

The authors declare no competing interests.