Temporal transcriptomic changes in the THY-Tau22 mouse model of tauopathy display cell type- and sex-specific differences

Ali, Muhammad; Garcia, Pierre; Lunkes, Laetitia P.; Sciortino, Alessia; Thomas, Melanie H.; Heurtaux, Tony; Grzyb, Kamil; Halder, Rashi; Skupin, Alexander; Buée, Luc; Blum, David; Buttini, Manuel; Glaab, Enrico

doi:10.1186/s40478-025-02013-z

Research
Open access
Published: 07 May 2025

Temporal transcriptomic changes in the THY-Tau22 mouse model of tauopathy display cell type- and sex-specific differences

Muhammad Ali¹,
Pierre Garcia¹,
Laetitia P. Lunkes¹,
Alessia Sciortino¹,
Melanie H. Thomas¹,
Tony Heurtaux²,
Kamil Grzyb¹,
Rashi Halder¹,
Alexander Skupin¹,
Luc Buée³,
David Blum³,
Manuel Buttini¹ &
…
Enrico Glaab¹

Acta Neuropathologica Communications volume 13, Article number: 93 (2025) Cite this article

482 Accesses
11 Altmetric
Metrics details

Abstract

Background

Tauopathies, including Alzheimer’s disease (AD) and frontotemporal dementia (FTD), display sex-specific differences in prevalence and progression, but the underlying molecular mechanisms remain unclear. Single-cell transcriptomic analysis of animal models can reveal how AD pathology affects different cell types across sex and age.

Objective

To understand sex-specific and sex-dimorphic transcriptomic changes in different cell types and their age-dependence in the THY-Tau22 mouse model of AD-linked tauopathy.

Methods

We applied single-cell RNA sequencing (scRNA-seq) to cortical tissue from male and female THY-Tau22 and wild-type mice at 17 months of age, when they had prominent tau inclusion pathology, and compared the results with corresponding data previously obtained at 7 months of age. Using differential statistical analysis for individual genes, pathways, and gene regulatory networks, we identified sex-specific, sex-dimorphic, and sex-neutral changes, and looked at how they evolved over age. To validate the most robust findings across distinct mouse models and species, the results were compared with cortical scRNA-seq data from the transgenic hAPP-based Tg2576 mouse model and human AD.

Results

We identified several significant sex-specific and sex-dimorphic differentially expressed genes in neurons, microglia, astrocytes and oligodendrocytes, including both cross-sectional changes and alterations from 7 months to 17 months of age. Key pathways affected in a sex-dependent manner across age included neurotransmitter signaling, RNA processing and splicing, stress response pathways, and protein degradation pathways. In addition, network analysis revealed the AD-associated genes Clu, Mbp, Fos and Junb as relevant regulatory hubs. Analysis of age-dependent changes highlighted genes and pathways associated with inflammatory response (Malat1, Cx3cr1), protein homeostasis (Cst3), and myelin maintenance (Plp1, Cldn11, Mal) that showed consistent sex-dependent changes as the THY-Tau22 mice aged. Multiple genes with established implications in AD, including the long non-coding RNA gene Malat1, displayed concordant sex-specific changes in mouse models and human AD.

Conclusions

This study provides a comprehensive single-cell transcriptomic characterization of sex-linked and age-dependent changes in the THY-Tau22 tauopathy model, revealing new insights into the interplay between age-dependent AD-like pathologies and sex. The identified sex-specific changes and their conservation across models and human AD highlight molecular targets for further preclinical investigation of sex-specific therapeutic strategies in AD.

Introduction

Alzheimer’s disease (AD) is a heterogeneous, progressive disorder with varied clinical presentations, rates of progression, and treatment responses. Biological sex is a key factor contributing to this diversity, with women showing an age-adjusted higher prevalence compared to men, comprising nearly two-thirds of individuals living with AD in the United States [1, 2]. Sex differences extend beyond prevalence; women with AD often experience faster cognitive decline and more severe neuropathological changes, while males tend to die faster with AD [3, 4] and have lower brain resilience to tau pathology [5]. Furthermore, sex differences have been observed in treatment effectiveness and risk factor impact [6, 7]. Despite established knowledge of AD hallmarks, amyloid-beta plaques and neurofibrillary tangles, the molecular mechanisms driving sex-specific differences remain poorly understood, presenting a significant barrier to developing personalized therapeutic strategies.

High-throughput single-cell RNA sequencing (scRNA-seq) technologies have significantly improved our ability to investigate complex cellular systems at unprecedented resolution, providing insights into cell type-specific changes and heterogeneity within tissues. In AD research, scRNA-seq has become a key tool for mapping cellular and molecular alterations associated with disease progression. Several studies have used scRNA-seq to investigate AD-related changes in human post-mortem brain tissue and animal models. For instance, Mathys et al. performed single-nucleus RNA-seq on human prefrontal cortex samples [8], revealing cell type-specific transcriptomic changes associated with AD pathology. Similarly, Grubman et al. used single-cell sequencing to identify altered cellular pathways in human AD brain tissue [9]. In animal models, studies such as the one by Keren-Shaul et al. [10] have employed scRNA-seq to characterize microglial states in mouse models of AD-like pathology. Building on this single-cell approach, our previous work has investigated cross-sectional sex-dependent changes at early time points in the THY-Tau22 and Tg2576 models [11, 12]. However, while these studies have provided new insights into AD- and tau-related transcriptomic changes, the potential impact of sex on these alterations at different ages has not been sufficiently addressed.

The THY-Tau22 mouse model, which over-expresses human 4-repeat tau with G272V and P301S mutations that were originally identified in frontotemporal dementia (FTD) patients [13], offers a valuable tool for studying mechanisms relevant to tauopathies, such as FTD and AD [14]. This model recapitulates several key features observed in tauopathies, including progressive cognitive decline, synaptic dysfunction, and tau pathology, without the confounding effects of motor deficits often seen in other transgenic lines [15]. By comprehensively studying scRNA-seq data for THY-Tau22 mice and their wild-type littermates at 17 months, and contrasting them to our previous analysis of 7-month old mice [12], we characterize age- and sex-dependent transcriptomic alterations associated with tau pathology. Our approach identifies sex-specific, sex-dimorphic, and sex-neutral gene expression changes across multiple cell types, including major AD-associated cell populations, such as neurons, astrocytes, microglia, and oligodendrocytes, to better understand cell type-specific responses to tau pathology in both sexes.

Our analysis reveals numerous sex-specific and sex-dimorphic differentially expressed genes across various cell types. We identify key biological pathways affected in a sex-dependent manner, both cross-sectionally and across age groups, including RNA processing and splicing, stress response pathways, neurotransmitter signaling, and protein degradation pathways. Through gene regulatory network analysis, we determine potential regulatory hubs, such as the genes Clu, Mbp, Fos and Junb, that may play central roles in mediating sex-specific responses to tau pathology. Furthermore, our age-dependent analyses identify genes showing consistent and progressive sex-dependent alterations between 7 and 17 months of age. By comparing our findings with cortical scRNA-seq data from human AD and another mouse model, the Tg2576 model of amyloid-beta pathology, we identify overlapping significant genes across different experimental settings, such as Malat1. Overall, this study provides new insights into the interplay between tauopathy and biological sex at the single-cell level, potentially informing future investigations of sex-specific therapeutic strategies for AD and related tauopathies.

Materials and methods

Ethical statement

All experimental procedures were conducted in strict accordance with European guidelines for animal research (FELASA) and received approval from the local Animal Experimentation Ethics Committee as well as oversight and approval from Luxembourg’s Ministry of Agriculture and Ministry of Health (registered under LUPA 2020/23).

Animal model, colony maintenance, and genotyping

We used a genetically engineered tauopathy mouse model, THY-Tau22 Mice (B6.Cg-Tg(Thy1-MAPT)22Schd, MGI:3717232), characterized by the over-expression of a modified human tau protein, harboring the mutations P301S and G272V, originally identified in frontotemporal dementia (FTD) patients [14]. This mutated protein is expressed under the control of the neuronal promoter Thy1.2. As these mice age, they exhibit several hallmarks of AD, including cognitive decline, impaired synaptic function, increased astrocyte reactivity, and accumulation of tau proteins.

For the THY-Tau22 mice, the original breeding pairs were obtained from the University of Lille, France. These mice were then rederived in the mouse facility of the University of Luxembourg and bred by crossing heterozygous males with C57Bl6/J females (RRID: MGI:3028467). The animals were maintained in pathogen-free facilities, with constant access to standard mouse chow (Ssniff, # V 1534 − 300) and water. Genotyping analysis of the mice was performed as described previously [12].

Experimental design and sample collection

For this investigation, we studied the cerebral cortex of THY-Tau22 mice and their wild-type littermates at 17 months of age (data newly obtained for this study) and then compared the results to the same groups of mice at 7 months of age (data obtained from our previous cross-sectional study [12]). This allowed us to distinguish between early and late pathological changes and to identify and interpret age-dependent changes in this model of tauopathy. Figure 1 shows the experimental design and workflow, comparing our new data from 17-month-old mice with previous data from 7-month-old mice (age-dependent comparison), and with cortical tissue data from human AD tissue and the Tg2576 mouse model (cross-model comparison).

We used mice of both sexes, from different litters, heterozygous for the transgene and their wild-type littermates as controls, and sample sizes were optimized for each type of analysis based on their specific requirements. For single-cell transcriptomic analyses, 16 mice in total were profiled (male THY-Tau22 transgenic mice (n = 4), male wildtype littermates (n = 4), female THY-Tau22 transgenic mice (n = 5), and female wildtype littermates (n = 3)). Single-cell RNA sequencing analyses typically require fewer samples (n = 3–5 per group) than traditional bulk analyses due to the high number of cells profiled per sample and the resulting increased statistical power from cell-level measurements. This sample size has been validated in our previous studies with mouse models of neurodegeneration. For behavioral analyses, larger group sizes (11–17 mice per group) were used to account for higher individual variability in behavioral measurements. For neuropathological analyses, we used 8 transgenic mice and 2 wild-type controls per sex. The smaller number of controls in this case was sufficient due to their low variability and our focus on comparing male versus female transgenic mice. Sample sizes for the between-group comparisons of interest were verified using power analysis, confirming a minimum of 80% power to detect between-group changes with a type I error probability of 5% for a minimum detectable effect size of 2 (the G*Power 3.1 software was used for calculations).

After reaching the desired study ages, the animals were humanely euthanized following approved animal welfare protocols using a combination of anesthetic drugs (administered intraperitoneally with ketamine at 150 mg/kg and medetomidine at 1 mg/kg) followed by transcardial perfusion with phosphate-buffered saline (PBS) to clear blood. The brain tissue was carefully extracted and prepared for the subsequent analyses, including histological examination (one brain hemisphere) and advanced single-cell sequencing techniques (using freshly dissected cortex from the other hemisphere for DropSeq analysis).

For scRNA-seq analyses, we used the entire neocortex rather than a specific cortical subregion. This approach was deliberately chosen because: (1) the THY-Tau22 model expresses mutant tau broadly throughout the neocortex under the Thy1.2 promoter, with pathology developing across cortical regions; (2) using the whole neocortex allowed us to capture tau-associated single-cell transcriptomic changes that might occur heterogeneously across different cortical areas, providing a more comprehensive view of pathology; (3) sampling the whole neocortex provided sufficient cell numbers for robust single-cell analysis of all major cell types, including less abundant populations; and (4) the mouse neocortex is, unlike the primate neocortex, smooth and does not allow for a clear distinction between subregions upon extraction of fresh tissue.

To ensure precise sampling and minimize contamination from non-cortical regions, we used a standardized dissection protocol performed by experienced neuroscientists. After euthanasia and transcardial perfusion, the entire neocortex was carefully separated from subcortical structures, hippocampus, and olfactory bulbs using anatomical landmarks. All dissections were performed rapidly, on an iced surface, to minimize RNA degradation and with care to prevent cross-contamination.

Behavioral assessment (Y-maze)

The spatial working memory performance of mice was assessed in the Y-maze. The Y-maze used here takes advantage of the natural exploratory behavior of the mouse. The maze consists of three identical arms with walls (40 cm long, 10 cm high, 10 cm wide), made of white opaque plastic, arranged in a Y-shape, with 120° angles between the arms around a central platform (Noldus, RRID: SCR_004074). Visual shapes (squares, circles, triangles) were depicted on the walls of the maze to allow for visual orientation and memory formation.

The test measures spontaneous alternation behavior (SAB), where the mouse is expected to remember which arm it has previously visited, and explore a new one. This ability to alternate between arms reflects working memory, as the mouse must continuously update its memory of recent choices. Memory performance is measured by calculating the percentage of spontaneous alternations, which reflects the mouse’s ability to remember and alternate between arms. A correct alternation occurs when the mouse consecutively enters each of the three arms in turn without repeating the same arm in a sequence. A high percentage of alternations indicates good working memory, as the mouse can remember which arms it has visited previously. The alternation percentage was calculated as: (number of correct alternations/total number of possible alternations (total number of arm entries minus 2)) ×100.

Seventeen-month-old THY-Tau22 (17 males, 11 females) and wild-type littermates (11 males, 13 females) were tested. The experimenter was blinded to the genotypes of the mice. At the start of the test, each mouse was placed at the end of an arm, facing outwards. The test was conducted in a controlled low-intensity light environment (10 lx). Mice were allowed to freely explore the maze during a single 5-minute session. We tracked and recorded trajectories using Ethovision software from Noldus (RRID: SCR_000441). The total number of visits to each arm and the number of correct alternations were counted manually, and the percentage of alternations was calculated using the formula above. Mice that visited fewer than 10 arms were excluded (one male and one female wild-type).

Two-way ANOVA with sex and genotype as factors was performed, followed by Sidak post hoc test to analyze the results using GraphPad Prism software (v10.3.0) (RRID: SCR_002798). P-values less than 0.05 were considered significant.

Histology and immunohistochemistry

Staining procedures followed standard protocols [16] on randomly selected free-floating 50 µM thick sections (2 sections/mouse for THY-Tau22). Sections from 7 to 8 mice per sex were used for THY-Tau22 mice, and sections from two wild-type littermates per sex served as negative controls. Prior to staining, epitope exposure was achieved by immersing sections in 50% vol/vol formic acid (Sigma, 33015) for 5 min. The sections were then washed three times for 10 min each in PBS containing 0.1% Triton X-100 (Sigma, T8787). For permeabilization, sections were incubated in PBS with 3% H₂O₂ (Sigma, 31642) vol/vol and 1.5% Triton X-100 vol/vol for 30 min. After three 10-minute washes, sections were incubated in 5% BSA (Sigma, A8022) wt/vol in wash buffer for 1 h to avoid unspecific antibody binding, before being incubated overnight at RT with AT8 primary antibody (Fisher, MN1060 RRID: AB_223652) diluted 1/1000 in PBS, 0.3% TritonX100, 2% BSA to detect hyperphosphorylated tau. The following day, sections were washed three times in PBS and incubated with anti-mouse secondary antibody (Alexa 555-conjugated, Invitrogen, A21422, RRID: AB_141822) for 2 h at RT. Fluorescence-stained sections were washed three times for 10 min each in PBS, mounted on Superfrost plus slides (Thermoscientific, Waltham, MA) and air dried overnight.

To reveal beta-pleated structures, AT8-stained dried sections on mounted slides were circled with DakoPen (Merck, Z377821), and incubated for 25 min in 0.3% thioflavin-S (Thio-S, Sigma, T1892) in a 50:50 ethanol: water mixture. Staining was differentiated in a 50%/50% ethanol/water mixture for 10 min, and sections were rinsed twice for 1 min in water. The sections were then air dried and coverslipped with Fluoromount G mounting medium (ThermoFisher, 00-4958-02).

After drying for 24 h, sections were viewed on a Zeiss AxioImager1 microscope using an Alexa 488 detection filter for Thio-S and an Alexa555 detection filter for AT8 under a 20x objective. AT8- and Thio-S-positive neuronal profiles were manually quantified separately in the cortex and hippocampal CA1 using a cell counter, and the average number/section of AT8- and Thio-S-positive neuronal profiles was calculated for each mouse. Statistical comparisons were made using a two-tailed Student’s t-test after confirming the normality of the data, and correlation analysis was performed using Pearson’s R and GraphPad Prism software (v10.3.0) (RRID: SCR_002798).

Single-Cell RNA sequencing

For single cell gene expression studies, fresh cortical tissue was dissociated with the Adult - Brain Dissociation kit (Miltenyi Biotec, #130-107-677), following the manufacturer’s protocol.

The dissociated cells were filtered through a 40 μm strainer, assessed for viability (> 80% required to proceed), and adjusted to a concentration of 120 cells/µl. Single-cell suspensions were then processed using a microfluidics-based approach following a modified Drop-seq protocol [11]. Briefly, cells were co-encapsulated with barcoded beads (ChemGenes Corp., 180 beads/µl in lysis buffer) in droplets generated using custom PDMS microfluidics devices fabricated by soft lithography with an aspect depth of 90 μm. Droplet generation was performed at flow rates of 2.5 ml/h and 11 ml/h for the cell/bead suspensions and carrier oil, respectively. After droplet breakage, captured mRNAs were reverse transcribed using Maxima H-RT in RT buffer (1× Maxima RT buffer, 4% Ficoll PM-400, 1 µM dNTPs, 1 U/ml RNase Inhibitor, 2.5 µM Template Switch Oligo), followed by exonuclease treatment and PCR amplification (4 cycles at 98 °C/65°C/72°C, 9 cycles at 98 °C/67°C/72°C). Libraries were purified using 0.6× AMPure XP beads, tagmented using Nextera XT, and sequenced on an Illumina NextSeq 500 with paired-end reads (20 bp for cell barcodes/UMIs and 60 bp for transcript sequences). Raw sequencing data was processed using the standard Drop-seq computational pipeline as detailed in previous work [11].

Sequencing data Pre-Processing

The scRNA-seq data was processed in the R statistical programming environment (version 4.4.0, RRID: SCR_001905) using the dedicated software package Seurat (version 5.1.0, RRID: SCR_007322). The data underwent quality control filtering using established metrics to ensure robust downstream analysis. Specifically, we filtered out potential technical artifacts by removing outlier cells according to three criteria: (1) cells with fewer than 200 unique feature counts, which likely represent empty droplets; (2) cells with more than 7,000 unique feature counts, which may indicate cell doublets or multiplets; and (3) cells with greater than 5% mitochondrial gene content, which often signifies stressed, damaged, or apoptotic cells. These quality control thresholds were selected according to the distribution of these metrics across all samples and established best practices in single-cell analysis.

All samples passed these quality control assessments and were retained for subsequent analysis. Following filtration, the data were normalized and scaled using the SCTransform function in Seurat, which employs a regularized negative binomial regression model to normalize cellular gene expression values while controlling for technical confounders. We then identified the 2,500 most variable genes across the dataset using the FindVariableFeatures function, prioritizing genes likely to be most informative for distinguishing between different cell types and states. These selected features were scaled using the ScaleData function with default parameters to prepare the data for dimensional reduction and clustering. The final quality-controlled dataset comprised 28,190 cells and 17,681 genes, which formed the basis for all subsequent analyses.

Bioinformatic analysis

Cell type identification and clustering

Before clustering the data for cell type annotation, dimensionality reduction was performed using principal component analysis (PCA) followed by Uniform Manifold Approximation and Projection (UMAP) for visualization [17]. The optimal number of principal components was determined through two complementary approaches: (1) the elbow method implemented in Seurat v5.1.0 FindElbowPoint function and (2) a quantitative cumulative variance analysis where we selected components until the percent change in variation between consecutive PCs fell below 0.1%. This approach identified 19 principal components as optimal for downstream analysis.

Cell clustering was performed using the shared nearest neighbor (SNN) graph-based clustering algorithm (Seurat v5.1.0 FindClusters function) [18] with resolution parameter 0.06, selected by maximizing the average silhouette width across multiple resolutions (0.001-1.0) using the cluster R package (v2.1.6). This optimization identified 9 distinct clusters as optimal, with an average silhouette width of 0.463. The FindNeighbors function was run with dims = 1:19 and default parameters. All visualizations were generated using ggplot2 v3.5.1.

Cell type annotation was conducted using the ScType algorithm [19] with marker genes sourced from the Cell Marker database (RRID: SCR_018503) [20]. Cell identities were assigned based on cell type-specific expression patterns, using a scoring approach that weights markers by their specificity (marker sensitivity threshold = 0.5). Cluster identity was assigned to the highest-scoring cell type when the cumulative score exceeded 25% of the total cell count in that cluster. Finally, the validity of the determined cell type identities was further corroborated by comparing the most significant differentially expressed genes (DEGs) within each cluster against established markers from the database PanglaoDB (RRID: SCR_022580) [21].

Differential gene expression analysis

For our analysis of differential gene expression, we used a Poisson regression model, implemented in the Seurat v5.1.0 package [22]. This approach was applied to identify DEGs between genotypes with both sex-dependent patterns and shared patterns across the sexes. Specifically, we categorized DEGs into three primary classes according to their expression patterns across sexes and genotypes: (1) Sex-neutral DEGs: Genes exhibiting significant changes (FDR < 0.05, min. absolute log fold change: 0.5) between transgenic and wild-type animals in both sexes, with the same sign of the log fold change; (2) Sex-specific DEGs: Genes showing significant differential expression (FDR < 0.05, min. absolute log fold change: 0.5) in one sex exclusively, while showing no trend toward even nominal significance in the opposite sex (nominal P-value > 0.5; this conservative cut-off for non-significance was chosen to avoid spurious detection of sex-specificity due to stochastic variation of p-value estimates around commonly used significance thresholds); (3) Sex-dimorphic DEGs: Genes demonstrating significant changes in both sexes (FDR < 0.05, min. absolute log fold change: 0.5), but with opposing directions of change.

These three types of DEGs were assessed and compared in three different analyses: (1) cross-sectional examination of DEGs at 17 months of age, (2) comparison of these DEGs with those identified at 7 months of age, which have been discussed in more detail in a dedicated publication focused on early-stage changes [12], and (3) age-dependent assessment of expression changes between the two time points, contrasting transgenic and wildtype animals while accounting for sex-neutral, sex-specific, and sex-dimorphic effects. We build on our previous analysis of cross-sectional changes at 7 months of age [12] here to enable temporal comparisons.

Furthermore, we determined genes exhibiting sex differences in the magnitude, but not direction, of expression changes through an analysis of sex-genotype interactions. The resulting genes with more subtle sex-dependent expression patterns are presented as additional, lower-confidence candidates for sex-related DEGs in the Supplementary Materials, while our main discussion is centered on the three primary categories (sex-neutral, sex-specific, and sex-dimorphic DEGs).

To identify both cell type-specific gene expression changes and common changes, we performed independent analyses for specific cell types and across all cell types integrated (pseudobulk analysis). While we filtered out consistently low-expressed genes before analysis, some genes showed minimal expression in just one condition in either bulk or cell type-specific analyses. For these cases, the estimated log fold changes and p-values can reach extreme values. Although these results correctly identify differentially expressed genes and preserve their relative rankings, they may overestimate both the magnitude of change and statistical confidence. Our interpretation therefore emphasizes the broader patterns of differential expression and relative gene rankings rather than focusing on specific numerical estimates.

To contextualize our findings and validate them across species and models, we compared our results with human AD post-mortem brain data [9] and single-cell data from the Tg2576 mouse model [11]. Specifically, these single-time-point datasets from Tg2576 mice (24 weeks) and human AD post-mortem samples were compared against the THY-Tau22 data for both 7 months of age [12] and 17 months of age (collected for this study) to identify robust changes shared across models, time points and species. For this purpose, mouse and human gene symbols were standardized using the HGNChelper R package (version 0.8.1). Visualization of the gene overlap across different studies in Venn diagrams was performed using the ggVennDiagram R package (version 1.5.2) [23].

Gene set enrichment and pathway analysis

To explore the effect of sex-dependent transcriptomic alterations on cellular pathways, we applied an over-representation analysis using the enrichGO function from the R package clusterProfiler (version 4.12, RRID: SCR_016884) [24] and annotations from the Gene Ontology (RRID: SCR_010326) database. The Gene Ontology enrichment analysis was conducted using two different subsets: biological processes (BP) and molecular functions (MF). We sourced the mouse gene annotations from the org.Mm.eg.db package (version 3.19.1, RRID: SCR_002811). Enrichment significance was determined using an adjusted p-value cutoff of 0.05. Finally, we visualized the enrichment results using enrichplot (version 1.24.0) and cowplot (version 1.1.3, RRID: SCR_018081).

Gene regulatory network (GRN) analysis

To investigate coordinated changes in gene activity within regulatory sub-networks, we constructed gene regulatory networks (GRNs) using a dedicated network reconstruction algorithm [25] in combination with the GeneGo MetaCore™ knowledge database (RRID: SCR_008125). This database integrates curated, experimentally validated gene-gene interactions derived from the scientific literature. For this analysis, we relied only on the retrieved interactions with regulatory effects and defined directionality, focusing on categories such as “Regulation”, “Transcriptional Regulation”, “Influence on Expression”, “Binding”, and “co-Regulation of Transcription”. To perform the network analysis, we only used statistical significance (FDR < 0.05) as a filter without requiring a minimum effect size, as this approach can detect coordinated changes across multiple genes that might be missed when examining only large individual gene changes.

For genotype-specific GRN reconstruction, we employed a condition-specific algorithm [25] that refines the initial network based on the binary gene expression profiles (increased expression = 1, decreased expression = 0) of transgenic and wild-type phenotypes. This method uses a genetic algorithm-based optimization to exclude interactions from the initial knowledge-based network that do not align with the observed expression states. Additionally, for interactions in the MetaCore database lacking activation or inhibition annotations, the algorithm inferred these regulatory effects from the expression data and the network topology [25].

Network perturbation analysis

While the GRN analysis already highlights key regulatory genes that influence the expression of the DEGs, a more refined ranking of these regulators can be achieved by simulating the effects of modifying their activity. To this end, we performed an algorithmic network perturbation analysis [25], which evaluates the ability of candidate regulators (perturbagens) to counteract downstream gene expression changes. Regulators with higher perturbation scores have the potential to partially reverse pathological expression patterns in a significant subset of their target genes. Consequently, the top-ranked regulators from this analysis represent promising candidates for initial drug target discovery, laying the ground for further in-silico druggability assessments and subsequent experimental validation.

Results

Exploratory activity, cognitive impairment, and neuronal Tau inclusions in 17-month-old THY-Tau22 mice

To confirm that the THY-Tau22 mice of our colony had an AD-like phenotype at 17 months of age, we tested their cognition with the Y-maze, and tested for the presence of tau inclusions. Specifically, we first used the Y-maze to test exploratory activity and spatial working memory. By analyzing the number of entries into the arms of the Y-maze, a measure of exploratory activity, we found a genotype effect (F(1,46) = 5.46; p = 0.02), no significant sex effect, but a significant interaction between the two (F(1,46) = 12,82; p = 0008). Post hoc analysis indicated that wild-type males explored a similar number of arms as their female counterparts. However, THY-Tau22 males showed a significantly less exploratory activity compared to their sex-matched wild-type littermates (WT: entries = 28.95; THY-Tau22: entries = 20.03; p = 0.0004) and to THY-Tau22 females (male entries = 20.03; female entries = 26.09; p = 0.023) (Fig. 2A).

Importantly, when analyzing the percentage of alternations, which measures cognitive performance, we found effects of both sex (F(1,46) = 5.051; p = 0.0295) and genotype (F(1,46) = 13.67; p = 0.0006). Post hoc analysis suggested that male and female wildtype littermates had similar percentages of alternations. However, while the percentage of alternations in THY-Tau22 males compared to their sex-matched wild-type controls did not reach significance (p = 0.077), this decrease was significant in transgenic THY-Tau22 females compared to their wild-type controls (p = 0.029) (Fig. 2B).

We quantified intraneuronal tau inclusions by counting the number of Thio-S-positive and AT8-positive neuronal profiles in the cortex and CA1 region of the hippocampus in THY-Tau22 mice of both sexes. We found no signal in these two brain regions in wild-type controls of either sex (not shown). Furthermore, we found no significant difference between the sexes in either the number of Thio-S-positive or that of AT8-positive neuronal profiles in the cortex (Fig. 3A) or hippocampus of THY-Tau22 mice (Fig. 3B). The number of Thio-S-positive or AT8-positive neuronal profiles in the cortex and the hippocampus did not correlate significantly with the percentage of alternation in either sex (Fig. 3, panels A and B; last column).

Characterization of cell types and marker genes

Our scRNA-seq analysis identified 9 different cell types in the mouse cortex samples for the 17 months time point (Suppl. Table ST1). The most abundant cell types were microglia, oligodendrocytes, astrocytes, and endothelial cells, each with over 5,000 cells profiled for both time points. Neurons and oligodendrocyte precursor cells (OPCs) were also well represented (over 1k cells for the 17 months time point, compared to ~ 0.8k cells for the 7 months time point). Smaller populations of macrophages, mural cells, and ependymal cells were detected, although the low cell counts for these types (less than 600 cells each) limited our ability to robustly characterize subpopulations. Given these abundance profiles, we focused our downstream analyses of transcriptomic changes on the major glial and vascular cell types. The cell type clusters were visualized using the UMAP dimensionality reduction, with annotated results shown in Fig. 4 (17 month time point; Suppl. Fig. S1 shows the corresponding results for the 7-month time point, adapted from our previous study [12]). A heatmap visualizing cell-type-specific marker gene expression across identified cell type clusters is provided in Suppl. Fig. S2.

Sex-dependent and sex-neutral gene-level transcriptomic changes

Global transcriptomic alterations across all cell types (THY-Tau22, 17 months of age)

In the global gene-level analysis of differential expression of THY-Tau22 mice vs. wild-type littermates across all cell types combined (also known as pseudobulk analysis) at 17 months of age, we identified 15 male-specific DEGs, 1 female-specific DEG, 21 sex-dimorphic DEGs (showing changes in both sexes but with opposing directions), and 10 sex-neutral DEGs. A detailed list of these DEGs is provided in Suppl. Table ST2.

Male mice showed a stronger transcriptional response to tau pathology, as evidenced by the higher number of male-specific DEGs. Furthermore, several sex-dimorphic DEGs (34) were observed, where genes changed in opposite directions between males and females.

Below, we briefly describe the top significant DEGs with known functional annotations in each category (male-specific, female-specific, sex-dimorphic, and sex-neutral DEGs). A summary of these genes and their alteration statistics in males and females is provided in Table 1, and a dot plot visualization of the sex-dependent changes is shown in Fig. 5.

Table 1 Most significant sex-dependent differentially expressed genes between THY-Tau22 and wild-type mice at 17 months of age

Full size table

Male-Specific DEGs:

1.
Scd2 (FDR = 2.54E-20, logFC = 1.49): Stearoyl-CoA desaturase 2, involved in fatty acid metabolism. Its male-specific increased expression highlights sex-specific alterations in lipid metabolism and membrane fluidity [26].
2.
Trf (FDR = 1.09E-19, logFC = 8.98): Transferrin, essential for iron transport and homeostasis. Its strong male-specific overexpression indicates pronounced sex-specific differences in iron metabolism and transport [27].
3.
Zc3h13 (FDR = 8.00E-16, logFC=-0.83): Zinc finger CCCH-type containing protein 13, involved in RNA methylation and processing, pointing to decreased expression of RNA modification pathways in males [28].

Female-Specific DEGs (only one identified):

Hsp90aa1 (FDR = 0.005, logFC = 0.65): Heat shock protein significantly over-expressed in females but non-significantly underexpressed in males, indicating sex-specific differences in protein folding and stress response pathways and matching with unbalanced expression of human HSP90AA1 in the entorhinal cortex previously reported for Aβ pathology [29].

Sex-Dimorphic DEGs:

1.
Cst3 (up in males, FDR < 2.50E-274; down in females, FDR = 1.37E-44): Cystatin C, an inhibitor of cysteine proteases and major AD risk factor [30]. Cystatin C has been shown to bind to Aβ and inhibit Aβ oligomerization and amyloid fibril formation [30].
2.
Mbp (up in males, FDR = 1.67E-135; down in females, FDR = 2.42E-12): Myelin basic protein, involved in myelin maintenance and repair, showing strong opposing gene expression changes between sexes. This protein has been reported to associate with amyloid plaques in the cortex of AD brains [31].
3.
Apod (up in males, FDR = 9.63E-124; down in females, FDR = 1.47E-4): Apolipoprotein D, involved in lipid metabolism and neuroprotection, showing highly significant divergent expression changes.

Sex-Neutral DEGs:

1.
Fth1 (up in both sexes, male FDR < 1.67E-135, female FDR = 1.97E-09): Ferritin heavy chain, a regulator of iron homeostasis.
2.
Malat1 (down in both sexes, male FDR < 1.67E-135, female FDR = 1.96E-21): Long non-coding RNA involved in gene expression regulation and splicing.
3.
Plp1 (down in both sexes, male FDR = 2.50E-274, female FDR = 5.18E-66): Proteolipid protein 1, essential for myelin structure.

Overall, differential expression analysis identified several key patterns associated with tauopathy. We observe strong sex-dependent patterns in protein homeostasis and processing (Cst3), myelin-related gene expression (Mbp), and lipid metabolism (Apod), where males and females show opposing responses. These sex-dimorphic changes are complemented by male-specific alterations in lipid metabolism (Scd2), iron transport (Trf), and RNA processing (Zc3h13), as well as female-specific changes in stress response pathways (Hsp90aa1). However, some core disease processes are shared between sexes in the bulk level analysis, as shown by similar changes in both males and females for genes involved in iron homeostasis (Fth1), transcriptional regulation (Malat1), and myelin structure (Plp1).

Cell type-specific transcriptomic alterations

The cell type-specific analyses revealed varying patterns of differential expression across cell populations. Due to the limited number of cells in smaller clusters, which reduce the statistical robustness of differential expression analysis, we focused our investigation on the five major cell types with sufficient representation: microglia, astrocytes, neurons, oligodendrocytes, and endothelial cells. Bar plot and volcano plot representations of the top five most significant sex-dimorphic and sex-specific DEGs with known gene symbols for these major cell types are presented in Fig. 6 and Suppl. Fig. S3, respectively. An overview of the DEG counts per category (female-specific, male-specific, sex-dimorphic, or sex-neutral) for five main brain cell types (microglia, astrocytes, neurons, oligodendrocytes, and endothelial cells) is shown in Suppl. Fig. S4.

Microglial cells showed the highest number of DEGs, with 14 male-specific DEGs, 1 female-specific DEGs, 6 sex-dimorphic DEGs, and 9 sex-neutral DEGs. Other cell types with notable changes included oligodendrocytes (11 male-specific, 7 sex-dimorphic, 3 sex-neutral), astrocytes (9 male-specific, 1 female-specific, 12 sex-dimorphic, 2 sex-neutral) and neurons (7 male-specific, 11 sex-dimorphic, 4 sex-neutral). The complete breakdown of all DEGs for all cell types is shown in Suppl. Table ST3.

Overall, our analysis revealed a pronounced sex-specific pattern in transcriptional responses to tau pathology across all major cell types at 17 months. Male cells exhibited both a broader and stronger response, with male-specific significant expression changes consistently outnumbering female-specific alterations.

Considering the functional annotations of the identified DEGs, our single-cell transcriptomic analysis also highlighted distinct patterns across the major brain cell types for both sex-neutral and sex-dependent changes (see the summary for top-ranked genes in Table 2).

In microglia, we found strong male-specific upregulation of inflammatory and immune response genes (Malat1: logFC = 24.90, FDR = 1.82E-266; H2-D1: logFC = 0.88, FDR = 2.61E-44; Tyrobp: logFC = 5.14, FDR = 3.84E-36), while females showed specific upregulation of stress response in line with the pseudobulk analysis results (Hsp90ab1: logFC = 2.42, FDR = 8.07E-4). Similarly, astrocytes exhibited male-specific changes in protein homeostasis and barrier function (Clu: logFC = 6.16, FDR = 1.42E-14; Cldn10: logFC = 0.79, FDR = 2.44E-14; Cpe: logFC = 2.39, FDR = 6.82E-14) and female-specific alterations in RNA processing (Pnisr: logFC = 1.88, FDR = 1.35E-4).

Neurons showed male-specific changes in particular in mitochondrial and matrix proteins (mt-Cytb: logFC = 15.80, FDR = 8.75E-29; Sparcl1: logFC = 7.19, FDR = 1.58E-26; Snhg11: logFC = 2.50, FDR = 2.37E-10). In oligodendrocytes, we observed strong male-specific changes in carbonic anhydrase and mitochondrial complex I gene members (Car2: logFC = 7.15, FDR = 4.30E-90; mt-Nd1: logFC = 14.84, FDR = 1.57E-38; mt-Nd2: logFC = 10.84, FDR = 1.11E-31), while endothelial cells showed male-specific changes in transcriptional regulation (Id3: logFC = 1.07, FDR = 1.82E-15) and female-specific changes in chemokine signaling (Cxcl12: logFC=-1.34, FDR = 1.10E-3).

Strong sex-dimorphic expression patterns were observed across all cell types, including Cst3 in microglia, mt-Rnr2 in astrocytes, Plp1 in neurons (all up in males/down in females, see Suppl. Tab. ST3), Malat1 in oligodendrocytes, and mt-Rnr2 in endothelial cells (all down in males/up in females, see Suppl. Tab. ST3). Finally, we also identified significant sex-neutral transgene-associated patterns, such as Apoe overexpression in microglia, Malat1 underexpression in astrocytes, Meg3 underexpression in neurons, and mt-Rnr1 underexpression in oligodendrocytes and endothelial cells (see Suppl. Tab. ST3).

To identify additional sex differences affecting only the magnitude of expression changes (rather than the direction or presence/absence of significant changes), we also performed a statistical test for sex-genotype interactions using the software edgeR. This analysis identified numerous significant interactions (Suppl. Tab. ST4), including many genes that were also detected in our primary analysis using Seurat. For example, Malat1 showed strong interaction effects across multiple cell types (e.g., with an adjusted p-value < 2.4E-31 in microglia), confirming its sex-dependent dysregulation. The interaction analysis also revealed additional candidate genes that display more subtle sex differences in only the magnitude of expression changes between males and females. Examples include mt-Rnr1 in astrocytes (interaction adj. p-value = 1.1E-28) and Aldoc in astrocytes (interaction adj. p-value = 1.7E-20). While these additional results represent lower-confidence candidates for sex-dependent regulation compared to genes showing opposing significant changes or sex-specific effects, they may still contribute to the overall sex-dependent manifestation of tau pathology.

Overall, these cell type-specific analyses reveal pronounced sex differences in the response to tau pathology across all major brain cell types, with males showing stronger inflammatory and metabolic alterations (particularly in microglia and oligodendrocytes), whereas females exhibit changes in stress response and RNA processing. While certain pathways show robust sex-specific or sex-dimorphic regulation (such as inflammatory responses in microglia and myelin-related genes in oligodendrocytes), other key AD-related pathways (including Apoe signaling and mitochondrial pathways) show consistent dysregulation across both sexes depending on the cell type. This suggests that tau pathology affects both sex-dependent and sex-independent molecular mechanisms in a partly cell type-specific manner.

Table 2 Cell type-specific sex-dependent gene expression changes across major brain cell types in THY-Tau22 mice at 17 months of age

Full size table

Shared gene-level alterations across mouse models, ages, and human AD

We compared the differential expression analysis results for THY-Tau22 mice at 17 months of age against an earlier time point for the same model, a distinct mouse model of Aβ pathology, and cortical gene expression data in human AD. Specifically, we computed overlaps between the following gene sets: (1) DEGs for THY-Tau22 vs. wildtype mice at 17 months of age, (1) DEGs for THY-Tau22 vs. wildtype at 7 months of age, (2) DEGs for the Tg2576 mouse model of Aβ pathology vs. wildtype (24-weeks old) [11], and (3) DEGs in post mortem human cortical tissue from AD patients vs. controls [9]. The analysis focused on the four major cell types consistently present across all datasets: neurons, astrocytes, microglial cells, and oligodendrocytes.

Computing the intersections of DEGs for each of these cell types revealed limited overlap across the datasets (Fig. 7). No genes showed significant changes across all four datasets in microglia, neurons, or astrocytes. The overlap between the 7- and 17-month THY-Tau22 data was small across all cell types, with 2–8 shared DEGs between these time points depending on the cell type. In oligodendrocytes, only two genes (Malat1 and Mbp) were differentially expressed across all datasets (Fig. 7b). Malat1, a long non-coding RNA, displayed an overall dominant pattern of male-specific underexpression in oligodendrocytes, though the sex-dependencies varied across the datasets. While it showed male-specific underexpression in human AD oligodendrocytes and sex-dimorphic expression in THY-Tau22 mice at 17 months of age (down in males, and up in females), it appeared as sex-neutral, underexpressed in oligodendrocytes in 7 months THY-Tau22 and Tg2576 models. Similarly, the myelin basic protein encoding gene Mbp showed dominant male-specific changes in oligodendrocytes, with consistent male-specific underexpression in the Tg2576 mouse model and in human AD; however, different patterns were observed in the ThyTau22 model for this gene, with a relative female-specific overexpression at the 7 months time point and significant sex-dimorphic changes at the 17 months time point (overexpression in males, and underexpression in females).

Hsp90ab1, a member of the heat shock protein 90 family with chaperone activity, displayed significant changes in microglia with varying patterns across three datasets: female-specific in 17 months THY-Tau22 (logFC = 2.5, FDR = 8.10E-04), male-specific in human AD data (logFC = 0.3, FDR = 8.25E-05), and sex-neutral in the Tg2576 model (logFC_m= -0.22, FDR_m=2.47E-07, logFC_f= -0.16, FDR_f=4.80E-03; Fig. 7a). The only other gene with significant alterations in the same cell type across a minimum of three datasets was Atp1b1, encoding a subunit of a sodium/potassium-transporting ATPase. This gene displays female–specific changes in human AD (logFC= -0.37, FDR = 9.95E-03) and THY-Tau22 mice (logFC= -0.51, FDR = 2.70E-06), and sex-dimorphic changes in the 17 months THY-Tau22 AD model (logFC_m= -0.76, FDR_m=6.35E-06, logFC_f=5.31, FDR_f=9.78E-22; Fig. 7c).

Additional significant patterns emerged when comparing smaller subsets of these datasets (see the Venn diagram in Fig. 7, detailed summary statistics in Suppl. Tab. ST18, and the overview of intersection set sizes in Suppl. Fig. S5). The well-established AD risk factor gene Apoe displayed sex-neutral microglial over-expression in both the 7 month and 17 months THY-Tau22 datasets. Finally, Malat1’s alteration patterns also extended beyond oligodendrocytes, e.g., showing consistent underexpression in astrocytes and OPCs as well, suggesting its broad involvement in multiple cell types relevant to AD pathology.

Sex-dependent pathway alterations in single cell types

We investigated cell type-specific pathway alterations by performing separate gene set enrichment analyses for Gene Ontology (GO) terms within each major cell population (a global, pseudobulk analysis of pathway alterations across all cell types is presented in the Suppl. Materials, see also Suppl. Figures S6-S9 and Suppl. Tables ST6-ST7). Analyses were performed separately for male-specific and sex-dimorphic DEGs for each cell type (for female-specific DEGs, the number of globally significant DEGs was insufficient for enrichment analysis). Below, we report the results for the largest cell populations for the cross-sectional analysis of THY-Tau22 mice vs. wild-type mice at 17 months of age.

Sex-dependent pathway alterations (THY-Tau22 mice, 17 months of age)

Microglial cells

Sex-dimorphic DEGs (6 genes) showed enrichment in synaptic and inflammatory response pathways (Suppl. Table ST8). Among male-specific DEGs in microglia (14 genes), significantly enriched biological processes included the regulation of neuron apoptosis, amyloid-beta formation, and regulation of myeloid cell differentiation (adjusted p < 0.05, see the dot plot visualization in Suppl. Fig. S10-S13). Molecular functions showed strong enrichment for tau protein binding, ubiquitin protein ligase binding, and peptide binding.

Oligodendrocytes

Male-specific DEGs in oligodendrocytes (11 genes) showed significant enrichment in myelination, chemokine production, and regulation of amyloid fibril formation pathways (see the dot plot visualization in Suppl. Fig. S14-S17). Molecular functions were enriched for structural constituents of myelin sheath and ferric iron binding. Sex-dimorphic DEGs (7 genes) showed enrichment in lipid metabolism and membrane organization pathways (Suppl. Table ST9).

Astrocytes

Male-specific DEGs in astrocytes (9 genes) showed significant enrichment in pathways related to negative regulation of amyloid beta formation, central nervous system (CNS) myelination, and axon ensheathment in CNS. Molecular functions included tau protein binding, amyloid protein binding, and functions related to myelin sheath structural components. While the single female-specific DEG (Pnisr, encoding a serine and arginine rich splicing regulatory protein) was insufficient for pathway analysis, it has previously been implicated in dysfunctional lactate metabolism in human AD [32]. Sex-dimorphic DEGs (12 genes) showed enrichment in cellular stress response pathways (Suppl. Fig. S18-S21 and Suppl. Table ST10).

Neurons

In neurons, sex-dimorphic DEGs (11 genes) were strongly enriched in transmembrane transport pathways and sodium ion homeostasis, suggesting sex-dependent alterations in neuronal communication (Suppl. Table ST11, Suppl. Fig. S22-S25). Male-specific DEGs (7 genes) were enriched in symporter activity and glutamate binding related molecular functions.

Gene regulatory network analysis of sex-dependent changes

To characterize the regulatory mechanisms involved in sex-dependent gene expression changes in THY-Tau22 mice, we performed differential gene regulatory network (GRN) analyses (see Methods). Beyond detecting potential disease-relevant regulatory mechanisms, this enabled us to determine key regulatory genes involved in controlling sex-dependent molecular disease phenotypes. The differential GRN inference approach was applied separately to the male-specific and sex-dimorphic DEGs in order to reconstruct phenotype-specific networks for the cross-sectional analysis of THY-Tau22 mice vs. wild-type mice at 17 months of age. Because the number of DEGs was only sufficient to build GRNs for the global analysis integrating all cell types, and only for male-specific and sex-dimorphic DEGs, we only report global rather than cell type-specific GRN analyses for these sex-dependent DEGs.

Gene regulatory networks (THY-Tau22, 17 months of age)

Male-specific differential GRNs

For male-specific DEGs, two GRNs were built, reflecting the regulatory sub-networks controlling these genes in transgenic mice (see Fig. 8) and wild-type controls respectively. The THY-Tau22 sub-network involves 28 gene nodes and 32 interactions, while the WT network includes 32 genes and 35 interactions. Among the upstream transcription factors in this GRN, the known Alzheimer risk factor gene clusterin (Clu) [33] was identified as a regulator gene that showed significant male-specific increased expression and involvement in regulating multiple downstream genes.

Sex-dimorphic differential GRNs

For sex-dimorphic DEGs, the THY-Tau22-specific network contains 22 nodes and 40 interactions (see Fig. 9), while the WT network covers 20 nodes and 37 interactions. Central regulatory genes identified include MBP, FOS, and JUNB, which display significant sex-dimorphic expression and regulate substantial portions of the network.

Network perturbation analysis (THY-Tau22, 17 months of age)

To identify potential targets for therapeutic disease modification among the regulatory genes in the created GRNs, we performed a network perturbation analysis [25]. The analysis revealed several gene regulators with the potential to modulate and reverse many downstream pathological changes in the network.

In the male-specific GRN, the clusterin gene (Clu) was identified as a central regulator with the potential to normalize the expression of 10 downstream DEGs. Another regulator, Rps9, showed the capacity to modulate 8 downstream DEGs. When targeted in combination, Clu and Rps9 have the potential for reversing changes in 12 downstream genes (Suppl. Table 16).

For sex-dimorphic networks, both Hspa5 and Fos individually showed a more modest perturbation potential, each capable of modulating 3 downstream DEGs (Suppl. Table S17). However, targeting the combination of Hspa5 with Junb provides a synergistic effect, reversing sex-dependent changes in 10 downstream genes. Visualizations of the condition-specific gene regulatory networks highlighting key regulatory transcription factors and their interactions with direct downstream target genes are shown in Fig. 8 (male-specific GRN) and 9 (sex-dimorphic GRN).

Overall, these analyses reveal a hierarchical organization of sex-dependent gene expression changes in THY-Tau22 mice at 17 months of age, with a few key regulatory genes controlling larger sets of downstream genes. The main identified regulatory hubs include Clu for male-specific changes and Hspa5/Junb for sex-dimorphic changes.

Age-dependent cell-type and sex-linked transcriptomic changes in THY-Tau22 mice

Age-dependent gene expression alterations in THY-Tau22 mice

The analysis of age-dependent and sex-linked alterations in the THY-Tau22 model revealed extensive cell type-specific patterns of gene expression changes between 7 and 17 months of age. Across nine distinct cell populations, a total of 1,057 genes showed significant disease-associated temporal changes that were either sex-dimorphic or sex-neutral. For detecting more subtle sex-specific changes between age groups, statistical power was insufficient in this analysis.

Among all cell types, microglial cells exhibited the most pronounced temporal changes, with 45 sex-dimorphic and 299 sex-neutral DEGs. Other major cell populations also demonstrated considerable age-dependent changes, though with varying patterns of sex dependence. Oligodendrocytes showed 24 sex-dimorphic and 117 sex-neutral DEGs, while astrocytes had 28 sex-dimorphic and 92 sex-neutral DEGs. Endothelial cells displayed 17 sex-dimorphic and 112 sex-neutral DEGs, suggesting significant vascular involvement in disease progression. Neurons showed only 2 sex-dimorphic DEGs but 96 sex-neutral DEGs, and similarly, OPCs only exhibited few sex-dimorphic DEGs (4), but many sex-neutral DEGs (201).

Overall, while microglia, oligodendrocytes, astrocytes, and endothelial cells displayed a high number of sex-dimorphic changes, only few sex-dimorphic DEGs could be detected in neurons and OPCs.

The smaller cell populations in the dataset showed more limited significant temporal changes, likely due in part to their lower representation in the samples. Macrophages displayed 3 sex-dimorphic and 9 sex-neutral DEGs, while mural cells and ependymal cells showed only sex-neutral changes (7 and 1 DEGs, respectively). Across all cell types, sex-neutral age-dependent changes predominated over sex-dimorphic ones. The complete gene-level statistics for these age-dependent alterations across all covered cell types are provided in Suppl. Table ST5.

Considering the most significant DEGs with known annotations in the individual cell types, in microglia, sex-dimorphic changes were led by Malat1 (down in males/up in females, FDR = 1.73E-140/<2.11E-280). Malat1 is a long non-coding RNA involved in inflammation already identified as significant in the cross-sectional analyses across different time points in THY-Tau22 and in the data for the Tg2576 mouse model and human AD (see section on ‘Shared gene-level alterations across mouse models, ages, and human AD’). Further significant age-dependent DEGs included Cx3cr1 (down in males/up in females, FDR = 1.05E-74/3.77E-22), an essential microglial receptor involved in neuron-microglia communication, and B2m (up in males/down in females, FDR = 2.01E-74/4.03E-84), involved in immune response and in aging-related processes as a systemic pro-aging factor [34]. Sex-neutral changes included consistent alterations in Ctsd (down in both sexes, FDR = 1.54E-236/<2.11E-280), a cathepsin involved in lysosomal function, in Cst3 (down in both sexes, FDR = 6.63E-213/<2.11E-280), a key regulator of protein homeostasis already identified as significant in the analysis of THY-Tau22 mice at 17 months of age, and in Hexb (down in both sexes, FDR = 6.27E-185/<2.11E-280), a beta-hexosaminidase subunit important for lysosomal function.

In astrocytes, sex-dimorphic patterns again included strong changes in Malat1 (down in males/up in females, FDR = 4.31E-184/<2.11E-280), Mt-Nd2 (up in males/down in females, FDR = 5.68E-67/2.41E-21), indicating mitochondrial dysfunction, and Ccdc88a (down in males/up in females, FDR = 8.00E-47/1.61E-2), an actin-binding protein involved in cell migration. Sex-neutral changes included Atp1a2 (down in both sexes, FDR = 1.58E-259/2.30E-81), encoding a Na+/K + ATPase essential for ion transport, Mt-Nd5 (up in both sexes, FDR = 9.85E-245/2.09E-7), indicating mitochondrial stress similar to Mt-Nd2, and Plpp3 (down in both sexes, FDR = 2.87E-122/6.63E-87), a phosphatase involved in lipid metabolism.

In oligodendrocytes, sex-dimorphic changes were again observed for Malat1 (down in males/up in females, FDR < 1.58E-259/<2.11E-280), Fth1 (up in males/down in females, FDR = 9.28E-245/4.43E-82), indicating altered iron homeostasis, and Car2 (up in males/down in females, FDR = 1.44E-58/9.54E-15), involved in myelin maintenance. Sex-neutral changes featured Plp1 (down in both sexes, FDR = 2.82E-97/7.52E-12), essential for myelin structure, Cldn11 (down in both sexes, FDR = 1.02E-93/5.05E-170), important for myelin barrier function, and Mal (down in both sexes, FDR = 5.58E-93/1.66E-29), a further gene involved in myelin maintenance. In general, oligodendrocytes showed pronounced changes, in particular in myelin-related genes.

Age-dependent pathway alterations in THY-Tau22 mice

The analysis of sex-dimorphic pathway alterations between THY-Tau22 mice at 7 and 17 months of age revealed diverse patterns of enrichment across major brain cell types, with both shared and unique features characterizing the aging response in different cellular populations.