IntroductionAlzheimer’s disease (AD) is a complex progressive neurodegenerative disorder and the most common cause of dementia, affecting millions of people worldwide1. AD is characterized by hallmark pathological features, including extracellular amyloid-β (Aβ) plaques, intracellular tau neurofibrillary tangles, neuroinflammation, and widespread synaptic and neuronal loss2. Despite decades of intensive research, the precise mechanisms driving the onset and progression of AD remain elusive, including whether intracellular or extracellular Aβ causally contributes to the disease beyond being a hallmark3, and effective therapeutic options are limited. One of the greatest challenges in AD research lies in uncovering the molecular and cellular changes that arise during the preclinical phase, a prolonged asymptomatic stage during which pathological features, such as Aβ deposition, develop4. This phase, which can span several decades, represents a critical window for therapeutic intervention aimed at preventing or delaying the onset of clinical symptoms. However, studying preclinical AD, presents several challenges, particularly when working with human subjects, including difficulties in identifying at-risk individuals, limited access to relevant tissues, the presence of confounding factors (e.g., aging and lifestyle), variability in disease progression, and the subtlety of molecular changes at early stages of pathology. Overcoming these barriers is critical for understanding the early mechanisms driving AD and enabling the development of effective strategies to combat later symptoms.To address these challenges, a range of genetically modified animal models, particularly transgenic mice, have been developed5. While no single model fully replicates the complexity of AD, each successfully mimics key features of the condition making their detailed characterization crucial for understanding AD mechanisms. A prominent model is the AppNL-G-F mouse6 which carries a knock-in of the human amyloid precursor protein (APP) with three specific mutations: the Swedish mutation (KM670/671NL), which increases the overall production of Aβ40 and Aβ42; the Arctic mutation (E693G), which promotes the formation of insoluble Aβ fibrils; and the Beyreuther/Iberian (I716F) mutation, which raises the ratio of Aβ42 to Aβ40. The AppNL-G-F model exhibits early-onset amyloid pathology starting at 2 months (~ 8 weeks) of age, with neuropathology progressing to saturation by 7 months (~ 28 weeks). Cognitive impairments typically manifest by 6 months (~ 24 weeks)7,8. This predictable disease timeline makes the AppNL-G-F model useful for studying molecular changes across all stages of pathology, particularly the preclinical stages. Importantly, due to the early onset of disease in this model the confounding effects of advanced age are minimized, offering a robust platform for investigating early molecular events leading to amyloidosis and for identifying blood-based biomarkers of brain dysregulation before the onset of clinical symptoms. However, despite its promise, the AppNL-G-F model has not yet been used to examine epigenetic remodeling before and after amyloid onset. This gap limits our understanding of both the model itself and the epigenetic mechanisms driving and contributing to AD pathology, ultimately hindering efforts to identify early epigenetic biomarkers and therapeutic targets.In this study, we leveraged a combination of single-nucleus and bulk epigenetic and transcriptomic data from a cohort of AppNL-G-F mice and their wild-type littermates to characterize molecular differences across three ages, ranging from postnatal week 3 (an early pre-symptomatic stage prior to detectable Aβ pathology6) to postnatal week 24 (representing advanced pathology with near-saturation of Aβ and the onset of cognitive decline)6. Our study specifically focuses on the hippocampus, a brain region critical for memory, learning, and spatial navigation9 which is known to exhibit early and pronounced vulnerability in AD, with significant tissue loss and connectivity disruptions10. Our findings reveal substantial early epigenetic and gene expression dysregulation in the hippocampus of AppNL-G-F mice prior to the onset of severe amyloidosis. Importantly, we also identified early DNA methylation signals in peripheral blood of the AppNL-G-F mice that warrant further study as potential blood-based biomarkers for preclinical AD, paving the way for development of blood-based diagnostic tools in preclinical AD.ResultsHippocampal cell composition remains largely stable during progression of amyloid pathologyTo investigate epigenetic and transcriptomic changes preceding and following onset of severe amyloidosis in the brain and blood of the AppNL-G-F mouse model, we collected hippocampus and blood from 30 AppNL-G-F and 30 wild-type littermates. The cohort included equal numbers of male and female, sampled at postnatal week 3 (weaning), 8, and 24 (n = 5 per sex/genotype/age combination; Fig. 1A). All assays conducted in this study (Fig. 1B) were performed on tissue from the same 60 individual animals. We selected postnatal week 3 (W3) to represent an early pre-symptomatic stage, prior to detectable Aβ pathology6, enabling investigation of initial molecular changes preceding severe amyloidosis in the hippocampus and blood. Postnatal week 8 (W8) represents an intermediate stage associated with early progression of Aβ pathology in the AppNL-G-F model, while postnatal week 24 (W24) represents advanced pathology, approaching Aβ saturation and coinciding with onset of cognitive decline6 (Fig. 1A).Fig. 1The AppNL-G-F mouse model allows studying Aβ pathology along a defined timeline (A) The reported timeline of Aβ pathology in the AppNL-G-F model and timepoints investigated in this study. (B) A summary of all assays conducted in this study.Full size imageChanges in brain cell composition, including neuron loss and gliosis, have been reported in AD11. To investigate changes in brain cell-composition and chromatin accessibility in the AppNL-G-F mice before and during progression of Aβ pathology, we used hippocampus single-cell ATAC-seq (snATAC-seq) to study AppNL-G-F (n = 5 per sex) and wild-type littermates (n = 5 per sex) at W3, W8 and W24. To improve cell yields in our snATAC-seq assay, nuclei from biological replicates were pooled to create one pool per sex/genotype/age combination yielding a total of 12 pools, comprising 14,907 cells. Given the lack of reported sex differences in amyloidosis patterns in the AppNL-G-F model6 and to improve statistical robustness, male and female pools were treated as replicates within each age and genotype combination. To identify and characterize cell types within our snATAC-seq dataset, we conducted unsupervised clustering and identified 21 cell clusters (Fig. 2A). Cross-reference of cluster-specific accessible chromatin regions with known marker genes (Fig. 2B; Supplementary Fig. 1)11,12 identified the following cell-types: microglia (n = 615 cells), oligodendrocyte (n = 1,729), precursor/immature oligodendrocyte (n = 349), endothelial (n = 127), astrocyte (n = 771), Cajal-Retzius cells (n = 63), excitatory neurons (n = 8,070) and inhibitory neurons (n = 1,362) (Fig. 2B; Supplementary Table 1). Although most clusters were successfully annotated, four (C8, C9, C10 and C19) could not be confidently classified due to ambiguous or inconsistent chromatin accessibility patterns (Fig. 2B; Methods). These clusters, which comprise a total of 471 cells, likely represent rare or transitional cell types (C8 and C10), as well as technical noise (C9 and C19) and were therefore excluded from downstream analysis.Fig. 2snATAC-seq reveal cell-specific changes in chromatin accessibility in early vs. late Aβ pathology. (A) Uniform manifold approximation and projection (UMAP) dimensionality reduction after iterative LSI of snATAC–seq data from 12 sample pools. Each dot represents a single nucleus (n = 14,907), colored by its corresponding cluster (left) or cell type (right). Bar plot shows the number of cells per cluster, with corresponding cluster colors and assigned cell type (B) Genomic tracks display chromatin accessibility at a subset of marker genes used to annotate cell types in this study. (C) Estimated average hippocampus cell composition is shown for each time point and genotype. The n refers to the number of nuclei recovered after quality control filtering. Cell composition for WT at W8 is not presented due to low nuclei count (n = 88) (D) MA plots of differential snATAC-seq peaks reveal chromatin accessibility differences in excitatory neurons during early pathology (W3), while in late pathology (W24), these differences are exclusive to inhibitory neurons. (E) Significant gene ontology (GO) terms associated with differentially accessible regions in inhibitory neurons at W24.Full size imageWe used cell distribution across clusters to estimate cellular composition of the hippocampus across ages and genotypes (Fig. 2C; Supplementary Table 1). Across all samples, excitatory neurons were the most prevalent cell type (61.0% ± 0.44; mean ± SEM), followed by inhibitory neurons in most cases (12.6% ± 0.73%; mean ± SEM). Due to low cell recovery from wild-type mice at W8 ( 0.05; Supplementary Table 1).Chromatin accessibility changes are detected in inhibitory neurons later in amyloidosisTo identify condition-specific changes in chromatin accessibility, we first performed marker peak analysis and identified snATAC-seq peaks unique to each age/genotype/cell-type condition (q