Analysis of changes in intercellular communications in Alzheimer’s disease reveals conserved changes in glutamatergic transmission in mice and humans

Bartas, Katrina; Nguyen, Megan; Zhao, Wei; Hui, May; Nie, Qing; Beier, Kevin T.

doi:10.1038/s41598-025-10795-4

Download PDF

Article
Open access
Published: 19 July 2025

Analysis of changes in intercellular communications in Alzheimer’s disease reveals conserved changes in glutamatergic transmission in mice and humans

Katrina Bartas¹,
Megan Nguyen²,
Wei Zhao³,
May Hui²,
Qing Nie^3,4,5 &
…
Kevin T. Beier^2,6,7,8

Scientific Reports volume 15, Article number: 26248 (2025) Cite this article

4581 Accesses
3 Citations
10 Altmetric
Metrics details

Subjects

Abstract

Analysis of system-wide cellular communication changes in Alzheimer’s disease (AD) has recently been enabled by single nucleus RNA sequencing (snRNA-seq) and new computational methods. Here, we combined these to analyze data from postmortem human tissue from the entorhinal cortex of people with AD and compared our findings to those from multiomic data from the 5xFAD amyloidogenic mouse model at two different time points. Using the cellular communication inference tool CellChat we found that disease-related changes were largely related to neuronal excitability as well as synaptic communication, with specific signaling pathways including BMP, EGF, and EPHA, and relatively poor conservation of glial-related changes during disease. Further analysis using the neuron-specific NeuronChat revealed changes relating to metabotropic glutamate receptors as well as neuronal adhesion molecules including neurexins and neuroligins. Our results that cellular processes relating to excitotoxicity are the best conserved between 5xFAD mice and AD suggest that excitotoxicity is the main common feature between pathogenesis in 5xFAD mice and people with AD.

Mettl3 regulates the pathogenesis of Alzheimer’s disease via fine-tuning Lingo2

Article 02 April 2025

Hippocampal disruptions of synaptic and astrocyte metabolism are primary events of early amyloid pathology in the 5xFAD mouse model of Alzheimer’s disease

Article Open access 16 October 2021

Systematic phenotyping and characterization of the 5xFAD mouse model of Alzheimer’s disease

Article Open access 15 October 2021

Introduction

Alzheimer’s disease (AD) is a complex disease characterized by systemic changes in brain function^1,2. While the pathological hallmarks of AD are the accumulation of amyloid plaques and tau tangles in the brain, changes in cellular function occur in neurons and glia throughout the course of disease that contribute to the development of age-associated cognitive impairment^3,4. Understanding the root molecular causes of disease is thus required for the development of effective therapeutics to combat the development of AD. Our ability to understand the molecular features of disease across the lifespan has been greatly accelerated by the recent development of single-cell and single-nucleus RNA sequencing (snRNA-seq) technologies. These methods have provided an atlas of gene expression throughout the course of disease, particularly in late-stage AD in brain regions such as the frontal cortex, entorhinal cortex (ENT), and the hippocampus^5,6,7,8. However, a key limitation of these experiments is that they focus only on the cell-autonomous features of the disease; that is, they examine cells in isolation. While a reductionist approach is necessary to define the core features of disease, understanding how the molecular features of disease, as revealed by snRNA-seq, influence cellular communication is essential towards uncovering the root and progression of AD.

Fortunately, in recent years a variety of computational methods have been developed that enable the inference of cellular communication through coordinated changes in receptor-ligand interactions in different cell types. These include CellPhoneDB⁹ and CellChat¹⁰among others. These methods have been widely used in a variety of tissue types to enrich single cell sequencing datasets and provide a wealth of information about modes of biological communication within tissues including in the AD brain. However, as these studies focused on glia and did not analyze data from neurons, they include only data from selected cell types^{11,12,13,14,15}. While these methods’ utility in part is due to their ability to be generalized to different tissues, they also focus exclusively on receptor-ligand interactions that are encoded by proteins. While this is an acceptable definition for most tissue types, its use is more limited in the brain, where cells principally communicate through chemical transmission of small-molecule neurotransmitters, including glutamate and GABA, that are not encoded by proteins. To fill this gap, NeuronChat¹⁶ was developed to infer the intercellular communications mediated by small-molecule neurotransmitters. As such, if we want to understand how communication systems in the brain change during disease, an integrated understanding of how both general communication and neuron-specific modes of communication change during disease is required.

Here we focus on the ENT, as it is a core region in the hippocampal circuit and thought to be a main contributor to AD-associated memory and cognitive deficits^17,18,19,20. For these studies we used the 5xFAD mouse model, a commonly studied amyloidogenic model used in ~ 10% of all AD studies employing rodent models^21,22. We sequenced the ENT from 5xFAD and littermate control mice at 2 months of age, before memory deficits are apparent, and at 8 months of age, when behavioral deficits are widespread^22,23. For comparison of gene expression changes across species we utilized publicly available snRNA-seq data from postmortem ENT tissue from people with AD and those without cognitive impairments, and collected similar data for mild cognitive impairment (MCI) to provide a more comprehensive examination of AD progression in mice and human postmortem tissue across disease progression.

Methods

Mouse and human tissue

Hemizygous 5xFAD mice (mutations in APP: Swedish, Florida, and London; mutations in PSEN1: M146L and L286V) on a C57BL/6J background (Jackson Laboratory, Stock #34848) were utilized. These were obtained from the Jackson Laboratory. Wildtype littermates on the same background served as control animals. A total of 8 mice were used in this study, and all were female. Mice were euthanized at either 2 or 8 months of age depending on the time point of the experiment. Mice were housed on a 12-hour light–dark cycle with food and water ad libitum. Before brain dissection and brain isolation, mice were deeply anesthetized with pentobarbital and decapitated to enable isolation of the ENT tissue. All procedures were approved by Institutional Animal Care and Use Committees of the University of California, Irvine. All procedures were done in accordance with the federal regulations and guidelines on animal experimentation (National Institutes of Health Offices of Laboratory Animal Welfare Policy). Human postmortem brain tissue samples were obtained from UC Irvine’s Alzheimer’s Disease Research Center (ADRC) under UCI’s Institutional Review Board (IRB) and approved by IRB. All human experiments were performed in accordance with relevant guidance and regulations. Results from the study are reported in accordance with ARRIVE guidelines. One sample was utilized in this study from a female individual with MCI. Characteristics of this sample and the donor are as follows: postmortem interval (PMI) of 6.03, age of 85 years, clinical diagnosis of MCI, tangle stage 2, plaque stage A, and APOE genotype of 2/2.

Sample collection and single-nucleus sequencing in mice

Single-nucleus ATAC-seq and RNA-seq were conducted using the 10x Multiome kit from 5xFAD or wild-type mice, both on the C57BL/6J background, from 2 months or 8 months of age. Animals received an intraperitoneal injection of a lethal dose of pentobarbital at a dose of 25 mg/kg, and euthanized by decapitation. Brain tissue was immediately extracted, and ENT was freshly dissected out of each sample. From this point onwards, both mouse and human tissue were treated the same. To isolate the nuclei, tissue was homogenized using Lysis Buffer and manually disrupted using pestles, followed by a 4-minute incubation period. Samples were centrifuged at 500 g for 5 min at 4 °C, and the supernatant was discarded. Subsequently, 1 mL of Nuclei Wash and Resuspension buffer (NWR) was added and incubated for 5 min. The pellet was carefully resuspended using a P1000 pipette and centrifuged again at 500 g for 5 min at 4 °C. This wash-resuspension-centrifugation process was repeated twice more. After the third centrifugation, the supernatant was discarded, and 200 µL of NWR buffer + 20 µL 7AAD (1:100) was added and passed through a 40 μm filter directly into a tube used for flow cytometry sorting. Cells were sorted at the UC Irvine Institute for Immunology Flow Cytometry Facility to remove debris and distinguish viable nuclei. Following sorting, the samples underwent a permeabilization process, which began with centrifugation at 500 g for 5 min at 4 °C. Subsequently, 100 µL of 0.1x Lysis buffer was added, followed by resuspension and a 2-minute incubation. Next, 1 mL of Washing Buffer was added, and the sample was centrifuged again for 5 min under the same conditions. Then, 500 µL of Diluted Nuclei Buffer was added and carefully mixed with a pipette. Finally, the resulting sample was spun down and submitted to the UC Irvine’s Genomics Research and Technology Hub for further processing.

Single-nucleus sequencing alignment and QC in mice

Alignment of the ATAC-seq and gene expression libraries to the mouse reference genome (mm10, v2020-A) was done using Cell Ranger ARC (v2.0.1, 10x Genomics). Alignment was performed on the High-Performance Community Computing Cluster (HPC3) at UC Irvine. Further processing was done in R largely using Signac²⁴ (v1.7.0), which is integrated with Seurat²⁵ (v4.3.0), following best practices for single-cell sequencing²⁶. Nuclei with the following QC metrics in the RNA-seq data were filtered out as they were outliers that were likely doublets, debris, or damaged: over 20% of reads aligned to mitochondrial genes, more than 70,000 or fewer than 200 UMIs, or more than 10,000 or fewer than 75 unique genes detected (Figure S1A-B). Mitochondrial gene percentage was used in this way as expression of these genes has previously been demonstrated to be indicative of low quality, or damaged cells²⁷. Nuclei with the following metrics were further filtered out from the ATAC-seq data to prioritize only high-quality nuclei within the ATAC modality as well: fewer than 10% reads in peaks, nucleosome signal score of greater than 3, transcription start site enrichment of less than 2, over 5% of reads being in a blacklist region, or counts in peaks of under 1,000 or over 70,000, as these were either outliers, or indicative of technical defects such as amplification biases²⁸.

Following QC, there were a total of 35,188 high quality nuclei remaining: 8,286 from 2-month-old wild-type mice, 15,126 from 2-month-old 5xFAD mice, 6,431 from 8-month-old wild-type mice, and 5,345 from 8-month-old 5xFAD mice.

Single-nucleus sequencing alignment and QC, human postmortem samples

Previously published snRNA-seq data collected from the ENT of AD and non-cognitively impaired, or control, postmortem human tissue was downloaded from the gene expression omnibus^5,29accession numbers GSE160936 and GSE138852, and re-processed using Seurat²⁵. To most closely match the mouse data, only female samples were utilized, meaning 4 samples per dataset were re-analyzed, for a total of 8 samples: 4 AD, and 4 non-cognitively impaired (2 per condition from each study, which represents all of the data from female individuals). Additional MCI data was obtained from postmortem tissue that was cut by and obtained from the UCI ADRC; this was sequenced at UCI’s GRT Hub and resulting fastqs were aligned to the NCBI Homo Sapiens Annotation Release 109 using the Parse Biosciences computational pipeline (v1.0.5p). This corresponds to the same genome, GRCh38, to which publicly available data was aligned. The resulting gene by cell counts matrices were QCed individually by sample and subsequently integrated. First, data were normalized and variable features were found using Seurat’s FindVariableFeatures function with default parameters. Based on these, integration anchors were selected using the SelectIntegrationFeatures and FindIntegrationAnchors commands sequentially with default parameters. Finally, these anchors were fed into Seurat’s IntegrateData function with default parameters. Despite differences in technology and processing, nuclei from control, MCI, and AD samples appeared to be well mixed after integration, with cell types from every condition being intermingled in 2D space after further processing with the following commands: ScaleData, RunPCA, RunHarmony³⁰, RunUMAP³¹, FindNeighbors, and FindClusters³².

Single-nucleus sequencing integration, dimensionality reduction, and clustering, mice

To integrate 5xFAD and control mouse brain transcriptomic data, sctransform-based normalization was used to normalize, scale, and find variable features within each dataset³³. Mitochondrial mapping percentage, a potential confounder which is typically more indicative of cellular stress or differences in processing than biological differences, was regressed out, and the Gamma-Poisson Generalized Linear Model method³⁴ was used, otherwise default sctransform parameters were used. Similar to integration for the human data, 3,000 variable features across all datasets were selected as potential integration features using the FindVariableFeatures command. These were then further narrowed down with the SelectIntegrationFeatures command in Seurat followed by the FindIntegrationAnchors command, both using default parameters. Finally, the datasets were integrated using the IntegrateData command in Seurat, with anchor features being set to those that were commonly variable in our list of 3,000 features, the normalize method specified as sctransform, and the other parameters set to their default values. This resulted in one Signac object containing all mouse data, including the 2-month-old and 8-month-old wild-type and 5xFAD mice. After integration, the standard Seurat workflow for dimensionality reduction, visualization, and clustering was run sequentially using the following commands: RunPCA, RunUMAP, FindNeighbors, and FindClusters. For RunUMAP and FindNeighbors the first 30 principal components were used, and for RunUMAP the number of nearest neighbors parameter was set to 30. The resolution parameter for FindClusters was set to 1.0.

Annotation of cell types

Cell types in our mouse data were predicted by transferring annotation labels from the Allen Brain Atlas’ transcriptomic data^35,36 to our datasets, and this predicted annotation was manually checked and refined using previously established marker genes^37,38,39. Oligodendrocyte markers used were Mobp, Il33, and Mog, OPC markers were Cspg4, Tnr, and Pdgfra, astrocyte markers were Aqp4, Slc1a2, and Gja1, microglia markers were Csf1r, Ctss, and C1qb, GABAergic neuron markers were Gad1 and Gad2, glutamatergic neuron markers were Slc17a7 and Slc17a6, and endothelial cell markers were Bsg, Flt1, and Vwf. The same marker genes were used to manually annotate the data from postmortem human samples. Clusters of fewer than 100 nuclei, that did not express known marker genes, or that expressed marker genes of multiple cells types, indicating a mixed population, were removed.

snATAC-seq data processing, mice

Following integration and annotation, we re-processed the ATAC data. As peaks had previously been called on each sample individually by Cell Ranger ARC, they needed to be redefined so that common peaks would be identified across all mouse brain samples. We also aimed to call peaks by cell type to identify peaks in rare cell types that were potentially missed in aggregate. On the integrated data we ran the CallPeaks function in Signac, which uses MACS2 to call peaks⁴⁰. We performed this peak calling by cell type, annotated according to the RNA-seq data, and these peak sets were then combined. Peaks found to be in genomic blacklist regions²⁸ or on nonstandard chromosomes were removed.

To visualize the ATAC data we performed the standard Signac workflow with the following commands: RunTFIDF, FindTopFeatures, RunSVD, and RunUMAP. We then combined this representation of the data with the transcriptomic data visualization. This representation of both RNA and ATAC data was created using the FindMultiModalNeighbors function in Signac, followed by running UMAP on the weighted nearest neighbor (WNN) graph.

Differential gene expression analysis

Before analysis, counts were normalized for all cell types together using the NormalizeData command in Seurat with default parameters. Differential gene expression was then performed on each cell type cluster (oligodendrocytes, astrocytes, microglia, OPCs, endothelial cells, glutamatergic neurons, and GABAergic neurons) using the R package presto⁴¹. Genes were considered significantly different if the adjusted p-value (Bonferroni correction) was less than 0.05. Comparisons were made between wild-type and 5xFAD mice at 2 months of age, wild-type and 5xFAD mice at 8 months of age, tissue from people with MCI and without cognitive impairment, and AD and those without cognitive impairment.

To further investigate differential gene expression in genes of interest and variance within our data, we applied jackknife resampling to our differential gene expression analyses to obtain 95% confidence intervals (CIs) for log-fold change values, and averages of these as well as adjusted p-values. This was done in R, with the same R package presto⁴¹ being used for the same comparisons. For each comparison for each cell type cluster, 100 iterations of subsampling of cells were performed. In each iteration, a random portion of 80% of the data was utilized rather than the entire dataset, differential gene expression analyses performed, and results saved. After 100 iterations, results were aggregated, and averages and confidence intervals calculated. Results were similar to standard differential gene expression analyses and are reported in Supplementary Table 3.

CellChat and NeuronChat analysis

CellChat (v1.6.1)¹⁰ and NeuronChat (v1.0.0)¹⁶ were used to infer communication between cell types. First, CellChat and NeuronChat’s standard workflows were run on each condition within the snRNA-seq data individually: 2-month-old wild-type mice, 2-month-old 5xFAD mice, 8-month-old wild-type mice, and 8-month-old 5xFAD mice. Separately, human samples were also put through this pipeline: control, MCI, and AD. Comparative analyses were then performed between wild-type and 5xFAD mice at 2 months of age and 8 months of age. MCI and cognitively normal tissue samples, and AD and cognitively normal tissue samples were also compared.

hdWGCNA, mouse and human comparison

To detect modules of genes that co-vary in a cell type-specific manner, high dimensional weighted gene co-expression network analysis (hdWGCNA) was performed using the hdWGCNA package⁴². hdWGCNA adapts the well-established WGCNA method⁴³ for snRNA-seq data. This analysis was performed on cell types where there were at least 1,000 nuclei in the mouse dataset: microglia, oligodendrocytes, OPCs, astrocytes, GABAergic neurons, and glutamatergic neurons. Modules were detected in each cluster within the mouse snRNA-seq data, projected onto the human data, and then module quality and preservation were evaluated. Modules found to be of high quality and highly preserved across species (zsummary.qual > 10 and zsummary.pres > 10) were further investigated. Differential module eigengene analysis was performed to analyze differences in expression of each module between control and disease groups, as demonstrated in the hdWGCNA tutorial and done in previous work⁴⁴.

This analysis was performed, and results visualized using EnrichR (v3.178)^45,46 and hdWGCNA functions RunEnrichr, GetEnrichrTable, and EnrichrBarPlot sequentially. Enrichment analysis was done using the 2021 versions of the Gene Ontology databases for Cellular Communication, Molecular Function, and Biological Process^47,48.

DIRECT-NET analysis and visualization, mice

DIRECT-NET⁴⁹ was applied to the mouse 10x Genomics Multiome data to detect regulatory links specific to either wild-type or 5xFAD mice at 2 and 8 months of age and to construct gene regulatory networks implicated in AD. Promoters for each gene in the mouse genome were defined to be the 1,000 basepairs before the transcription start site of a gene. The DIRECT-NET tutorial was then followed, with analyses being applied to each genotype and time point individually first and then compared. First, cis-regulatory elements (CREs) were inferred following the DIRECT-NET tutorial for genes of interest based on previous analyses, including Gria4, Grin2b, and Grm1. Links were detected by genotype by subsetting the data and then running the Run_DIRECT_NET function on each genotype at each age. Signac’s coverage plot function was used to visualize these links. DIRECT-NET categorizes links as either high, low, or medium confidence based on importance score, a measure of a feature’s impact on the XGBoost model accuracy, with a high importance score indicating a high-confidence (HC) link. Links with high importance scores were more likely than those with lower scores to be validated by interactions detected in HiC data⁴⁹. Because of this finding, only HC links were plotted and utilized in our analyses, with links specific to 5xFAD mice highlighted in pink, links specific to control mice highlighted in blue, and links apparent in both genotypes highlighted in purple.

We then proceeded to construct gene regulatory networks based on genes upregulated in 5xFAD mice for both the 2- and 8-month time points. Following the DIRECT-NET tutorial, we used presto⁵⁰ to determine genes that were upregulated in 5xFAD mice according to auROC analyses (auc > 0.5) of snRNA-seq data. We excluded genes corresponding to ribosomal proteins and those without known functions, then evaluated which of the remaining genes overlapped with regions of the genome that were more accessible in 5xFAD mice based on the accompanying snATAC-seq data. CREs relating to both upregulated and accessible genes were then used to generate links to known transcription factors using the generate_peak_TF_links command in DIRECT-NET. These links were then constructed into a network and visualized using the Python package networkx⁵¹.

Results

Sample selection and quality control for mouse and human AD samples

One significant advantage of using rodent models of AD is that AD pathology can be studied across the lifespan, including at pre-symptomatic time points. We chose two time points for this study: 2 months, a pre-symptomatic time point when plaques begin to form in select regions in the brain, and 8 months, after significant plaque accumulation has occurred and cognitive deficits have emerged^22,23. We performed a multiomic assessment (single nucleus RNA-seq plus single nucleus ATAC-seq) of cells from the ENT of 5xFAD and age-matched controls from each age group (Fig. 1A). After alignment and quality control assessments of data, we had 23,412 nuclei for the 2-month time point and 11,776 nuclei for the 8-month time point. All expected cell types were captured in each dataset, including excitatory neurons, inhibitory neurons, astrocytes, microglia, oligodendrocytes, and oligodendrocyte precursor cells (OPCs; Fig. 1B, C). Cell types were similarly represented across time points and genotypes (Fig. S1C-E). To confirm that our integrated multiomic strategy worked as expected, we assessed the correspondence between the snRNA-seq and snATAC-seq datasets. Marker genes such as Slc17a7, Mog, and C1qb were more accessible and more highly expressed in glutamatergic neurons, oligodendrocytes, and microglia, respectively, than in other cell types (Fig. 1C-F), consistent with known gene expression patterns. Accessibility of other marker genes for each cell type are shown in coverage plots in Fig. S2.

To explore the diversity of neuron types identified, we assessed the number of distinct subtypes of glutamatergic and GABAergic neurons within our mouse dataset. We performed unsupervised clustering using the smart local moving (SLM) algorithm and found 5 inhibitory (In) and 12 excitatory (Ex) neuronal subclusters with distinct transcriptional profiles (Fig. S3A). We also transferred labels from a well-annotated dataset from the Allen Brain Institute and assessed gene expression differences (Fig. S3B,C). Using this annotation method, we identified the following cell types: GABAergic subcluster 1 (In 1) was defined by expression of the vasal intestinal peptide (Vip), In 2 by synuclein gamma (Sncg), In 3 by lysosomal-associated membrane protein 5 (Lamp5), In 4 by parvalbumin (Pvalb), and In 5 by somatostatin (Sst). Excitatory neurons were classified by layer (depth in cortex, L2-6) and connectivity patterns (intratelencephalic, IT; corticothalamic, CT; pyramidal tract, PT; near projecting, NP).

To assess how our mouse results compared with those from human postmortem brain tissue, we utilized data from previously published studies from ENT tissue in humans with and without AD. We also performed snRNA-seq of postmortem brain tissue at the MCI timepoint to acquire data from an earlier stage of disease. After merging and processing the human data, we observed the same cell types as we found in our mouse data, though in different quantitative proportions (Fig. 1G). Annotation of cell types in the human data was based on the same marker genes as used in the mouse data (Fig. 1C,H). As we identified fewer neurons in the human samples, we did not perform further annotation of neurons beyond inhibitory and excitatory.

Differential gene expression and high dimensional weighted gene correlation network analysis (hdWGCNA) indicate a greater conservation of neuron-related gene expression changes relative to glial-related gene expression changes between mouse models and human AD

We next explored differences in gene expression between 5xFAD versus control mice, and human AD/MCI postmortem samples versus samples from non-cognitively impaired human postmortem brain tissue. Results of differential gene expression analysis for each cell type and each of the four comparisons (5xFAD versus control at 2 months, 5xFAD versus control at 8 months, MCI versus control, and AD versus control) including fold-change and adjusted p-value are reported in Supplementary Table 1.

As a benchmark, we first focused on several genes known to be associated with disease-associated astrocytes (DAAs) that emerge during the development of AD: Clusterin (Clu), Apolipoprotein E (Apoe), and Cystatin C (Cst3)^52,53. DAAs have previously been identified in the hippocampus and cortex of both postmortem human tissue from people with AD and 5xFAD mice^52,53. Genes upregulated in DAAs are associated with amyloid clearance, aging, and the complement system⁵². Unexpectedly, at 2 months of age in 5xFAD mice, DAA genes were downregulated compared to controls (Fig. 2A). Interestingly, we see a very similar downregulation of APOE and CST3 in the human MCI brain (Fig. 2B). However, in 5xFAD mice at 8 months of age the DAA genes Clu, Apoe, and Cst3 were significantly upregulated, consistent with previous reports in mice and humans that these genes are enriched in DAAs or upregulated in AD in subclusters of astrocytes (Fig. 2C)^52,54. We observed a significant upregulation of each of these genes in the human AD brain (Fig. 2D). These data together indicate that gene expression patterns occurring in the earliest stages of disease may not mirror those that occur at later stages, and in some cases may trend in the opposite direction. This in turn may indicate that gene expression changes during AD pathogenesis may not progress in a linear fashion and could be dynamically regulated during disease.

While assessing changes in the expression of individual genes is useful to assess specific changes, an alternative approach where interactions between genes are considered may better represent the overall patterns of expression changes during disease progression. Previously, weighted gene co-expression network analysis (WGCNA⁴³) was applied to bulk RNA-seq data to detect such gene networks, which allowed for further analysis and comparison of these gene modules. This approach was recently modified for use with single-cell data (hdWGCNA⁴²). Here we applied hdWGCNA to each cell population in the mouse dataset (oligodendrocytes, astrocytes, microglia, OPCs, GABAergic neurons, and glutamatergic neurons). We then projected these cell type-specific mouse gene modules onto the corresponding human gene expression data and quantified module preservation across species and module quality⁵⁵. Only high-quality modules, meaning reproducible and robust, and highly preserved modules with a similar network architecture in both species were further examined. This quality and preservation analysis was done to focus our efforts on modules whose expression change was not due to chance, and was not unique to the mouse model to increase the likelihood that our results are relevant to human AD.

Interestingly, we found that in microglia and astrocytes, two cell types commonly studied in the context of AD in rodent models including 5xFAD^21,52,56,57no modules were both of high quality and highly preserved (Fig. S4A-B), indicating that patterns of gene expression changes were distinct in 5xFAD mice and human AD for these cell types. In oligodendrocytes just one module, Oligo-M3, was both highly preserved and of high quality (Fig. S4C). To evaluate whether expression of genes in this module was significantly different between genotypes, we used differential module eigengene (DME) analysis implemented in hdWGCNA and found that the module expression was significantly upregulated in 2-month-old 5xFAD mice and in postmortem brain tissue samples from both human MCI and AD compared to samples from non-cognitively impaired donors (Fig. S4D, F-G). In 8-month-old 5xFAD mice a similar effect was seen but it was not statistically significant (Fig. S4E). Similarly, in GABAergic neurons one module, GABA-M3, was both highly preserved and of high quality (Fig. S5A). This module was not significantly differentially expressed between samples from 5xFAD and control mice at either age, nor postmortem brain samples from MCI or AD tissue donors and controls (Fig. S5B-E), suggesting that the differences in expression were relatively mild. In OPCs, one module, OPC-M1, was both highly preserved and of high quality (Fig. S5F). The upregulation of genes in this module was not significant at 2 months (Fig. S5G), however, this module was downregulated in MCI (Fig. S5H) and upregulated in human AD and in 5xFAD mice (Fig. 2E-F). We further investigated what genes were expressed in OPC-M1 and to what pathways they may contribute. OPCs primarily differentiate into oligodendrocytes, but there is also evidence that they can become neurons or astrocytes under certain conditions^58,59,60,61. OPCs are involved in biological processes including myelination⁶² as well as synaptic remodeling and pruning^63,64,65. Based on gene ontology (GO) analysis, genes in OPC-M1 were most highly associated with cyclic guanosine monophosphate (cGMP) and cyclic adenosine monophosphate (cAMP) binding (Fig. 2G). cGMP and cAMP are both well-known small molecule second messengers for a variety of signal transduction pathways and have both been shown to modulate synaptic plasticity^66,67,68. Additionally, it has previously been suggested that cAMP and cGMP alter metabolism of amyloid-β or stimulate its production^69,70,71which is especially relevant given amyloid-β’s abundance in the 5xFAD mouse brain.

Overall, the best module quality and conservation was observed in glutamatergic neurons, where we observed 2 modules, Glut-M1 and Glut-M3, that were both of high quality and highly preserved (Fig. S5I). DME analysis indicated that only Glut-M1 was significantly differentially expressed – upregulated – in both 5xFAD mice at both time points and postmortem human AD brains relative to their respective controls (Fig. 2H-I, Fig. S5J). Differences were not significant in MCI (Fig. S2K). GO analysis of this gene module highlighted biological processes including voltage-gated calcium channel activity (Fig. 2J), G protein-coupled receptors (GPCRs; Fig. 2K), and actin filament network formation (Fig. 2L), all processes that are important in synaptic regulation^72,73,74. Interestingly, though the GABA-M3 module noted previously was not significantly differentially expressed, GO analysis of GABA-M3 also yielded terms associated with calcium transport and glutamate receptors (Fig. S5L), indicating a common set of neuronal changes relating to cellular excitability. Together, these results suggest that gene expression patterns related to neuron-specific signaling processes, including synaptic modulation by OPCs, intrinsic cellular excitability, and glutamatergic signaling are better preserved between 5xFAD mice and human AD brains than modules specific to glial cells, and that these neuronal processes may play a central role in disease pathophysiology.

Changes in inter-cellular communication networks during AD development

While single cell sequencing data enable the study of gene expression and gene module regulation within defined cell types, one key limitation is that these results only focus on cell-intrinsic changes. This provides only a partial picture of disease, as the brain consists of numerous cell types that operate in a dynamic fashion during both health and disease, and appropriate inter-cellular communication is necessary for proper brain function. For example, as glial cells play critical roles in synaptic function and remodeling which occur in neurons, focusing only on the transcriptome of glial cells in isolation, and not considering how glial changes impact communication with neurons, may mask their true importance in disease. To gain a deeper understanding of the evolution of inter-cellular interactions within the ENT, we utilized the computational package CellChat¹⁰. This method infers changes in cell-cell communication networks based on known cellular communication pathways. CellChat determines potentially important signaling pathways by identifying in which cell populations known ligands and corresponding receptor genes are expressed, including multimeric proteins, as well as other proteins that may be involved. Based on expression of these genes, we can infer modes of inter-cellular communication, which can be visualized as a directed network where nodes are cell types and edges are connectivity probabilities. These networks can inform us of likely communication pathways between identified cell types, and how the strength of these predicted interactions change across conditions and/or disease.

We first assessed the overall inferred level of incoming and outgoing signaling for each cell type, based on snRNA-seq data. In both 2- and 8-month-old control mice, the majority of inter-cellular signaling – both outgoing and incoming – was predicted to either arise from or be received by OPCs or neurons, with microglia and endothelial cells predicted to contribute the least to inter-cellular signaling (Fig. 3A-B). Notably, only relatively small differences were detected in 5xFAD mice relative to controls when comparing data from 2- and 8-month-old animals, primarily in astrocytes and oligodendrocytes (Fig. 3A-B), indicating only subtle changes in inferred inter-cellular communication occur over time in 5xFAD mice. In contrast, while we observed similar elevations in predicted signaling in astrocytes and oligodendrocytes in human MCI and AD tissue samples, changes in predicted signaling in glutamate and GABA cells was much more pronounced, with more outgoing and incoming signaling predicted in each cell type in MCI and AD than controls (Fig. 3C, S6A). These data together indicate that 5xFAD mice may exhibit similar patterns of signaling changes as in human AD, though to a lesser extent (Fig. 3A-C, S6A).

We next identified molecular signaling pathways that showed similar or different changes between 5xFAD mice and human AD samples (Fig. S6B). Pathways that exhibited the most similar disease-related changes across species were Ephrin-A (EPHA), Claudin (CLDN), Epidermal Growth Factor (EGF), and Bone morphogenetic protein (BMP). Predicted signaling through EPHA pathways was highly conserved between mice and humans across time points, and there was a higher level of predicted signaling in the EPHA pathway in 5xFAD mice at both timepoints as well as human MCI and AD compared to controls (Fig. 3D, S6B-D). Glutamatergic neurons showed the largest increase in predicted EPHA signaling in the disease state relative to controls, an effect that was more pronounced in human MCI and AD than 5xFAD mice (Fig. 3D, S6B-D). While the EPHA signaling pathway includes the genes encoding the receptor tyrosine kinases Ephrin1-8 (Epha1-8) and ligands Ephrin 1–5 (Efna1-5), this elevation in inferred signaling was mostly driven by expression changes in Epha4 and Epha7, which were significantly upregulated in glutamatergic neurons in AD compared to controls (EPHA4, logFC = 0.695, adjusted p-value = 0.000557; EPHA7, logFC = 0.386, adjusted p-value = 0.00468, Supplementary Table 1). Interestingly, a loss of EphA4 was previously shown to improve aspects of social memory⁷⁵ and synaptic plasticity in the hippocampus⁷⁶suggesting that an upregulation in EphA4, as observed here in AD-related disease, may impair properties of neuronal signaling. Interestingly, other studies have also linked EphA4 to amyloid-β, though the nature of this link is not clear and may be bidirectional. For example, while some studies suggested that Epha4 regulates amyloid-β⁷⁷, others indicated that amyloid-β regulates Epha4⁸⁰. While our results do not provide an answer to this discrepancy, they support the observation that as amyloid-β accumulates, Epha4 expression increases.

Notably, while glutamatergic neurons showed the largest predicted differences in EPHA signaling, the Epha3, Epha4, Epha5, and Epha7 genes were also present in the high-quality and highly preserved OPC-M1 discussed previously (Supplementary Table 2). In addition, the two GO terms previously mentioned as highly associated with OPC-M1 – cAMP and cGMP signaling – are engaged in the EPHA signaling pathway⁷⁸and are upregulated, consistent with an elevation in signaling through EPHA. Therefore, an upregulation in EPHA signaling may impact multiple cell types and contribute to neuronal dysfunction during the development of AD-related disease.

In the CLDN signaling pathway, only one gene was implicated, the tight junction-associated Cldn11, which was significantly upregulated in both 8-month-old 5xFAD mice and postmortem human AD tissue samples (Fig. 3E-F). We found that the upregulation of Cldn11 in oligodendrocytes is opposite of that found in a previous study performed in human postmortem AD samples, where a downregulation was observed⁷⁹. Notably, that study was performed using tissue from the precuneus region in the brain rather than the ENT, and therefore, one possibility is that changes may be brain region-dependent. Indeed, in another region of the brain, the frontal cortex, Cldn11 was found to be significantly upregulated in the oligodendrocytes of postmortem samples from individuals with AD compared to aged controls⁸⁰. Supporting this interpretation, another study using the amyloidogenic APP-NL-G-F⁸¹ mice found that a gene module containing Cldn11 was either upregulated or downregulated depending on the brain region under study⁸².

Predicted EGF signaling was also different in both 5xFAD mice and postmortem human AD samples relative to controls. Samples from control mice and non-cognitively impaired postmortem brain samples exhibited predicted EGF signaling from OPCs to astrocytes (Fig. 3G-H). However, no EGF signaling was predicted in 5xFAD mice at 8 months of age, or tissue samples from human MCI or AD. Notably, there was still predicted EGF signaling in 2-month-old 5xFAD mice, consistent with these animals representing an early disease stage (Fig. S6B, E). Regarding the potential function of EGF, a previous study showed that EGF treatment prevented cognitive impairment in mice, which is consistent with the hypothesis that a loss of EGF signaling contributes to AD⁸³. Therefore, our data suggest that signaling through the EGF pathway is reduced in AD, consistent with the protective effect of restoring EGF signaling in diseased brains.

Finally, BMP signaling was also predicted to exhibit similar changes in 5xFAD mice and human AD brains (Fig. 3I-J, Fig. S6F-G). While no significant BMP signaling was predicted in control or 5xFAD mice at 2 months of age, BMP signaling was predicted to occur at 8 months of age, and slightly more so in controls than 5xFAD mice (Fig. S6F-G). The difference was mostly manifest as a reduction in astrocyte-astrocyte signaling during disease progression (Fig. S6F-G). Similarly, in human postmortem brain samples, BMP signaling was predicted to be stronger in controls compared to tissue samples from humans with either MCI or AD (Fig. 3I-J, S6B). In this case, changes in astrocyte-astrocyte signaling were still predicted, though the most substantial change was predicted to occur in OPCs, in which predicted signaling was reduced in AD (Fig. 3I-J). Genes implicated in these changes include the serine/threonine receptor tyrosine kinases Bmpr2, Bmpr1b, and Bmpr1a, and ligand Bmp7 (Fig. S6H). While BMP genes encoding the BMP ligands Bmp4 and Bmp6 have previously been linked to AD and were shown to be upregulated in other mouse models of AD, expression of these two genes was not significantly different in controls vs. disease samples in our mouse data⁸⁴ (Supplementary Table 1). In the human data, Bmp7 expression was downregulated in AD compared to control samples in astrocytes (logFC = -0.217, adjusted p-value = 8.94E-78, Supplementary Table 1). Notably, Bmp7, in contrast to Bmp4 and Bmp6, was shown to have a potentially protective effect in AD⁸⁵. Thus, the reduction in Bmp7 signaling parallels the observed reduction of EGF signaling in AD, as both are thought to be forms of neurotropic signaling and protective against disease. These results thus point to a potential loss of trophic signaling as a common feature of AD pathogenesis in mice and humans.

In addition to pathways where similar inferences were made between mice and humans, we also detected some pathways that had contrasting predicted changes between species. For example, the receptor-type tyrosine-protein phosphatase mu (PTPRM) had higher predicted signaling in brains from people with AD but lower in 5xFAD mice, the glycoprotein prosaposin (PSAP) was higher in AD but lower in 5xFAD mice, and the growth factor platelet-derived growth factor (PDGF) was lower in AD and higher in 8-month-old 5xFAD mice relative to controls (Fig. S6B). PTPRM has been suggested to play a role synapse formation⁸⁶and though it has not been widely studied in AD, a previous Genome-Wide Association Study identified two SNPs within the PTPRM gene associated with dementia⁸⁷. As these SNPs would not be present in the 5xFAD model, this may be an explanation for the discrepancy between species. In contrast, PSAP, a lysosomal trafficking protein⁸⁸has been proposed as a potential biomarker for preclinical AD⁸⁹and a drug targeting PSAP reduced neuronal loss and decreased inflammation in a mouse disease model⁹⁰. The PDGF signaling pathway has also long been linked to AD⁹¹with decreases in PDGF signaling being associated with disruption of the blood brain barrier and eventual neurodegeneration⁹². It is possible that these differences across species reflect differences in disease staging, or they may reflect larger differences in disease between the rodent 5xFAD model and human AD.

NeuronChat predicts a conserved increase in glutamatergic transmission in the 5xFAD rodent model and human AD

Our CellChat analysis provided a set of inter-cellular signaling processes that change during disease, data which implicated changes in signaling between both neurons and glia. This provides a global picture of how inter-cellular signaling may be disrupted in the ENT during the development of AD. One limitation of the CellChat package and database is that it only considers classic ligand-receptor interactions between cells where both ligand and receptor are expressed proteins. However, most neuron-neuron and some neuron-glia communication occurs via small molecule neurotransmitters such as glutamate and GABA. Therefore, this analysis may miss a major component of disease-related changes that develop during AD. To explore potential changes in neurotransmitter-based communication, we applied the newly developed computational package NeuronChat, which utilizes a similar analysis method to CellChat but is adapted for neurotransmitter-based communication¹⁶. A focus on disease in neurons is warranted, as of the five hdWGCNA gene modules found to be both of high quality and highly preserved between 5xFAD mice and human AD, three were found in neuronal populations, and one other in OPCs which have neuron-like features. This raises the possibility that exploring changes in inter-neuronal signaling in 5xFAD mice may have the most relevance for the human AD brain.

We used NeuronChat to compute all the predicted changes in neurotransmitter-based communication in cells from mice and human postmortem brain samples (Fig. 4A-E). Changes were observed in 5xFAD mice: at 2 months of age, there was an increase in signaling to oligodendrocytes and endothelial cells (Fig. S6I-J). In 8-month-old 5xFAD mice there was an increase in neurotransmitter-based communications, occurring approximately equally in both GABAergic and glutamatergic cells (Fig. 4A-B, E). In human postmortem AD brain samples, most changes in signaling were predicted to occur from glutamatergic neurons, with a general increase in predicted communication from glutamatergic neurons to both GABA and glutamatergic neurons in the AD brain (Fig. 4C-D). Several predicted increased interactions were observed in both 5xFAD mice at 8 months of age and human AD, including between glutamate and AMPA-type glutamate receptors (Glu-Gria1, Glu-Gria2, Glu-Gria4), NMDA-type glutamate receptors (Glu-Grin2b and Glu-Grin1), kainite-type glutamate receptors (Glu-Grik2), metabotropic glutamate receptors (Glu-Grm1, Glu-Grm5, Glu-Grm8), and the neuronal adhesion molecules neurexin and neuroligin (Nrxn3-Nlgn1; Fig. 4E). Notably, no changes in GABAergic signaling pathways were predicted. In addition, genes mediating interactions – Gria2, Grin1, Grin2b, Grm1, and Grm5w – were also present in the preserved hdWGCNA module Glut-M1 (Supplementary Table 2), indicating that expression of genes for these receptors varied across cells in a coordinated fashion across species. When considering genes implicated in glutamate production rather than receptors, we found that the gene glutaminase (Gls) was significantly upregulated in glutamatergic neurons in 8-month-old 5xFAD mice and tissue samples from human AD brains (logFC = 0.0594, adjusted p-value = 4.79E-05, and logFC = 0.427, adjusted p-value = 4.49E-14, Supplementary Table 1). Gls was downregulated in 2-month-old 5xFAD mice (logFC = -0.177, adjusted p-value = 7.45E-71, Supplementary Table 1), and not significantly altered in MCI. The vesicular glutamate transporter 1 (Slc17a7) was also upregulated in glutamatergic neurons from both 8-month-old 5xFAD mice and human AD brain samples (logFC = 0.0999, adjusted p-value = 2.95E-10, and logFC = 0.271, adjusted p-value = 0.0464, Supplementary Table 1). Similarly to Gls, it was downregulated in 2-month-old 5xFAD mice (logFC = -0.0777, adjusted p-value = 5.46E-13, Supplementary Table 1), and not significantly differentially expressed in MCI.

Related to increased glutamatergic signaling in AD, there is a long-held hypothesis in AD that excitotoxicity mediated by excessive glutamatergic signaling is a key contributor to disease pathogenesis^93,94,95. Our combined observations that (1) most predicted changes in neurotransmitter-based interactions occurred via glutamatergic signaling, (2) glutamatergic neurons had two gene expression modules that were both high quality and highly preserved, more than any other cell type, (3) the EPHA signaling pathway is predicted to strongly impact glutamatergic neurons all support this hypothesis. As many of the genes relevant to glutamatergic signaling (Gria2, Grin1, Grin2b, Grm1, and Grm5) were part of Glut-M1 as identified by hdWGCNA, we were curious what other genes may also be in this module. As expected from the GO analysis shown in Fig. 2J-L, many ion channel genes were found that contribute to intrinsic neuronal excitability, including sodium ion channel genes Scn1a, Scn2a, Scn3a, Scn3b, Scn8a, and calcium ion channel genes Cacna1e, Cacna1b, Cacna1c, Cacna2d1, and Cacng3 (Supplementary Table 2). Of these genes, Cacna1b and Cacna1c were significantly upregulated in human AD, and Cacna1c was significantly upregulated in 5xFAD at 8 months of age (Supplementary Table 1), positioning this gene to play a conserved and potentially key role in AD. Upregulation of these genes would likely lead to an elevation in intrinsic neuronal excitability, though it is not clear if these changes may be drivers of pathogenesis, or compensatory responses to disease development. Cacna1c has previously been linked to bipolar disorder and depression⁹⁶and more recently to AD using exploratory bioinformatic analyses⁹⁷. In mice, an increase in Cacna1c expression has been negatively correlated with object recognition memory⁹⁸. Interestingly, its expression in reactive astrocytes has also been associated with plaque formation, providing a potential mechanistic role for Cacna1c in AD pathogenesis⁹⁹.

We next looked more closely at one particular pathway, the glutamate to metabotropic glutamate receptor 1 (Glu-Grm1, a G_q-coupled GPCR that activates phospholipase C), because (1) Glu-Grm1 was within the highly conserved and high-quality Glut-M1 module, (2) it was identified as differentially regulated in NeuronChat analyses, and (3) it was upregulated in glutamatergic neurons in 5xFAD mice at 8 months of age (logFC = 0.161, adjusted p-value = 4.49E-14, Supplementary Table 1). Unsurprisingly, in both 8-month-old 5xFAD mice and human postmortem brain samples from people with MCI or AD, glutamatergic neurons were the predicted sender of this Glu-Grm1 interaction in all cases (Fig. 4F-H). To define which neuron types may be involved in the predicted Glu-Grm1 interaction, we examined communication in specific annotated neuronal cell clusters of 8-month-old 5xFAD mice (Fig. 4I). Every excitatory cluster (subtypes 1–12) exhibited at least some predicted outbound communication via Glu-Grm1, as expected, and inbound glutamate was also widespread, except for in subtype 10 in both 5xFAD and control mice. Interestingly, for GABAergic interneurons, inbound Glu-Grm1 interactions were only predicted onto some subtypes; for example, GABAergic subclusters inhibitory 3, characterized by Lamp5 expression (Fig. S3B-C), and inhibitory 4, characterized by Pvalb expression (Fig. S3B-C), did not have any significant predicted Grm1-Glu communication (Fig. 4I), which reflects low expression of Grm1 in these subclusters. These results suggest that elevations in glutamatergic signaling may differentially impact select subtypes of glutamatergic and GABAergic inhibitory neurons in the ENT.

In addition to glutamatergic signaling, we also found differences in predicted interactions via neuronal cell adhesion molecules. For example, we observed in NeuronChat analyses a predicted increase in communication via Neurexin-3 and Neuroligin-1 in human MCI and AD, though this was not seen in mice (Nrxn3-Nlgn1, Fig. 4E). Similar to what we found in data from human postmortem brain samples, expression of Nlgn1 was significantly upregulated in glutamatergic neurons in 5xFAD mice at 8 months of age (log2FC = 0.163, adjusted p-value = 7.66E-16, Supplementary Table 1), and Nrxn3 was upregulated in astrocytes, though this was not statistically significant (log2FC = 0.148, adjusted p-value = 0.357, Supplementary Table 1). Neuroligins are post-synaptic cell-adhesion molecules on excitatory synapses¹⁰⁰ and expression of Nlgn1 has previously been shown to induce increases in glutamatergic synapse density and activity¹⁰¹. We also observed an increase in the expression of neuroligin-related binding proteins in the GO term from Glut-M1 (Fig. 2J). Thus, the observed upregulation of Nlgn1 is consistent with an elevation in excitatory synaptic communication, which would be expected to contribute to excitotoxicity. Nrxn3 is a pre-synaptic protein which can bind to neuroligins, among other proteins, and has been found to be downregulated in postmortem brain samples from people with AD, though these samples were from the hippocampus¹⁰². Indeed, the function of Nrxn3 is likely brain region-dependent¹⁰³. Taken together, our results indicate that increased glutamatergic signaling in AD occurs via changes in multiple cell types including glutamatergic neurons but also GABAergic neurons, OPCs, and astrocytes, consistent with a system-wide disease state.

Disease-associated epigenetic regulation reflects an elevation in glutamatergic communication

The observed systems-wide remodeling of communication between both neurons and glia in the brain are likely dependent on epigenetic modifications that occur during disease. In our rodent sequencing experiments, we performed a multiomic assessment (combined snRNA-seq and snATAC-seq) which enables us to link chromatin accessibility to gene expression across all cell types.

To assess the relationship between chromatin accessibility and gene expression, we utilized DIRECT-NET, which identifies both cis-regulatory and trans-regulatory elements within cellular DNA. Cis-regulatory elements are detected using a machine learning method that links peaks within the snATAC-seq data to gene expression values reflected in the snRNA-seq data, whereas trans-regulatory elements are detected by matching sequences within peaks to known transcription factors. DIRECT-NET was run on data from 5xFAD and control mice from 2-month-old and 8-month-old groups, run separately. This was done so that if links between peaks to specific genes were only found in one genotype or age, we could isolate in which conditions these links were identified. We focused our initial analyses on genes identified in our NeuronChat analyses. Results showed some links specific to 5xFAD, some specific to control mice, and some present in both genotypes (Fig. 5A-C). We found that genes encoding glutamate receptors such as the AMPAR gene Gria4 and NMDAR gene Grin2b had many high confidence (HC) links to peaks in the ATAC-seq data, but none in coding regions of the genome outside of the gene of interest itself (Fig. 5A-B). In contrast, Gria4 had 2 HC links outside of the Gria4 gene locus, which may indicate that these regions are important in the expression of Gria4, though they were not specific to either WT or 5xFAD mice, so we did not pursue these regions further. Grin2b had more links to loci external to coding regions with some being specific to 5xFAD or control mice.

Of particular interest was the metabotropic glutamate receptor Grm1, as in mice at 8 months of age, Grm1 was upregulated in 5xFAD mice relative to controls (logFC = 0.161, adjusted p-value = 4.49E-14, Supplementary Table 1) and had multiple linked peaks specific to each genotype. In addition to non-coding loci, the expression of this gene was linked to expression of multiple other genes. For example, we observed a link specific in 5xFAD mice from Grm1 to the transcription start site of the gene encoding the E3 ubiquitin-protein ligase SHPRH (Shprh; Fig. 5C). To our knowledge, Shprh has not previously been linked to AD, though it is known to be highly expressed in the human brain and a potential tumor suppressor gene¹⁰⁴. There was also a link to the transcription start site of the gene encoding the E3 ubiquitin ligase Fbxo30, which was specific to control mice and not observed in 5xFAD mice. Interestingly, Fbxo30 has previously been implicated in the positive regulation of BMP signaling¹⁰⁵which we noted in our CellChat analyses in Fig. 3 and S6. While Glu-Grm1 interactions identified by NeuronChat were limited to glutamatergic neurons and certain GABAergic neurons, changes in BMP signaling were predicted to occur between astrocytes and other cell types (e.g., Fig. 3I-J, S6F-G). This finding linking Grm1 expression to genes that control BMP signaling suggests that this may be another route, albeit indirect, by which neurons and glial cells may interact.

We next proceeded to expand our multiomic assessment to any gene that was both upregulated in 5xFAD mice according to our snRNA-seq data, and any loci differentially accessible according to our snATAC-seq data. In 2-month-old mice, DIRECT-NET analyses resulted in a small gene regulatory network (GRN) which included transcription factors Zinc finger protein 219 (Zfp219) and Bcl2 modifying factor (Bmf; Fig. 5D). This network’s small size was largely because at 2 months of age, very few loci were significantly differentially accessible between control and 5xFAD mice, which is expected given the early time point in disease progression. To characterize genes within the network we performed a GO analysis and found, unsurprisingly given the role of Zfp219 as a transcription factor, that genes in the network were highly associated with transcription and DNA binding (Fig. 5E-G).

Results from this same analysis method but at 8 months of age yielded a much larger GRN (Fig. 5H), consistent with the later disease stage, resulting in larger and more substantial changes in chromatin accessibility and gene expression regulation. Overall, this GRN contained 4 transcription factors and 89 genes. GO analysis on this set of genes showed not only association with regulation of transcription and DNA binding, as expected, but also association with ion channel activity (Fig. 5I-K). Specifically, sodium channel activity and voltage-gated sodium channel activity were among the top 3 terms relating to molecular function (Fig. 5K), another indication that regulation of ion channels may be a principal feature of disease, both at the gene expression and chromatin regulation levels. Notably, one of these transcription factors within the network, SRY-box transcription factor 11 (Sox11), has previously been implicated in controlling intrinsic neuronal excitability, with its expression being activity-regulated¹⁰⁶, raising the possibility that this gene may play an important role in regulating the elevation of intrinsic neuronal excitability that occurs in AD.

Discussion

Here, we show that while inferred signaling changes during AD are widespread between different cell types in the ENT in both 5xFAD mice and human AD, changes in glutamatergic signaling are the most prominent and shared feature of disease in mice and postmortem human samples. This conclusion is supported by several lines of evidence. First, two of the five total gene expression modules identified using hdWGCNA that were both of high quality and highly conserved were found in glutamatergic neurons, while a third was found in GABAergic neurons, and a fourth in OPCs, which all have processes related to neuronal excitability and/or neuronal synapses (Fig. 2, S5). In addition, we found using CellChat that the major signaling changes occurred in, glutamatergic cells, GABAergic cells, and OPCs both early and late in disease (Fig. 3). Changes in predicted EPHA signaling in AD mostly occurred in glutamatergic neurons (Fig. 3D), but multiple genes associated with EPHA signaling were also observed in the OPC gene expression module M1. This module also contained the GO terms cAMP- and cGMP-binding, which is critical for EPHA signaling (Fig. 2G, Supplementary Table 2). Lastly, our analysis focusing on neuron-specific modes of communication using NeuronChat indicated an elevation in glutamatergic signaling as the main feature of AD pathogenesis in both mice and humans, including the up-regulation of Slc17a7, Gls, and Grm1 (Fig. 4, Supplementary Table 1). Regulation of Grm1 was linked to Fbxo30, which in turn regulates BMP signaling. As both Grm1 and BMP signaling processes are reduced in AD, this may be a mechanism by which changes in glutamatergic signaling coordinate with other cell populations to trigger wider modifications in signaling between cell types.

CellChat and Neuronchat implicate tissue-wide changes in cellular communication during AD

The advent of single cell RNA sequencing technologies has enabled an unprecedented definition of the molecular changes that occur in a variety of disease processes, including in AD. Indeed, there has been a substantial increase in the number of recent publications detailing the molecular changes that occur during AD in a variety of rodent models and human postmortem tissue samples, at multiple time points and in multiple brain regions^{5,6,8,29,44,107–110}. However, one core limitation of these technologies is that they only explore cell-autonomous changes that occur during disease, and do not consider interactions between cells. Thus, traditional sequencing analyses provide only a partial picture of the disease, making it difficult to explore tissue- and systems-level questions. A previous study utilized CellChat to explore the changes in cellular communication that occur between non-neuronal cells¹¹¹; however, how this relates to neuron-neuron and neuron-glia interactions is only just starting to be characterized in humans¹¹²and cross-species comparisons are yet to explored. Here we define the gene expression changes in putative receptor-ligand interactions that are conserved between mouse and humans, and focus on neuron-specific changes in communication, identifying changes in glutamatergic signaling across multiple cell types and signaling pathways. For example, the major cell types that are predicted to engage in inter-cellular signaling are glutamatergic neurons, GABAergic neurons, and OPCs (Fig. 3A-C). Given that these cells together contain 4 of the 5 gene expression modules that are of high quality and well-preserved between mouse and human, these findings further support the need to understand how changes in communication between these cells contribute to disease progression. In the future, our findings will be further enhanced by the application of spatial transcriptomic technologies, which will allow us to add a spatial dimension to the changes in cellular communication that we observed here. We have recently developed several computational packages designed to integrate snRNA-seq and spatial transcriptomic datasets, including COMMOT¹¹³, which should further allow us to link changes in gene expression to spatial patterns of cellular communication, as well as pathology in situ including amyloid plaques and tau tangles.

One of the common changes between the 5xFAD model and postmortem human AD brains was decreased BMP signaling. This was seen early in the 5xFAD mouse model at 2 months of age and continued through 8 months. In data from human postmortem samples, expression was reduced in both MCI and AD. Bmp7, which has not been the focus of many previous studies in the context of AD¹¹⁴stood out as being both highly expressed and significantly differentially expressed across conditions, particularly in astrocytes and OPCs (Fig. S6H, Supplementary Table 1). BMP signaling has many previously established functions, including neurogenesis and astrogliogenesis¹¹⁴and Bmp6 has even been shown to localize to amyloid-β plaques in the hippocampus (Bmp7 was not tested)⁸⁴. Bmp7 specifically has been suggested to be neuroprotective, but of particular interest to us, it has been shown to reduce the negative effect of excess glutamate¹¹⁵. The fact that it is downregulated early in 5xFAD mice and in human AD suggests a reduced ability to respond to increased glutamate load, even before disease is severe.

Further cementing the importance of excitotoxicity were our findings regarding the EGF signaling pathway, which was predicted to decrease in 5xFAD mice at 8 months of age compared to controls, and in MCI and AD compared to controls. While EGF signaling has a multitude of functions, it, like Bmp7, has also been suggested to be neuroprotective, specifically against excess glutamate¹¹⁶. Most prominently, findings from NeuronChat indicated that increased glutamatergic signaling was a shared feature between 5xFAD and postmortem AD human tissue samples. While the glutamate receptor Grm1 was significantly upregulated in 5xFAD mice, this was not the case in AD (Supplementary Table 1), despite the NeuronChat results predicting increased signaling via Grm1. These observations suggest that perhaps in AD, elevations in the production of glutamate, rather than its receptors, may be driving increased signaling. In addition, we found that the major changes in epigenetic regulation that occur in the 5xFAD model are related to cellular excitability and regulation of synaptic inputs. Indeed, genes that were identified as being part of the disrupted gene regulatory network in 5xFAD mice at 8 months of age largely related to sodium channels, which are a key contributor to excitotoxicity.

Neuronal excitability and excitotoxicity as the key conserved features of AD in mice and humans

We were surprised to find that no gene expression modules in microglia or astrocytes were both of high quality and well-preserved between 5xFAD mice and human AD, but that at least one gene expression module from each of the other cell types analyzed with hdWGCNA were significantly different (Fig. S4A-C, S5A, F, I). This means that the coordinated gene expression changes that occur in microglia and astrocytes during disease development in 5xFAD mice are distinct from those that occur in human AD. If this is the case, it means that processes occurring in 5xFAD mice may not translate well to those occurring in humans. However, before making broad conclusions, several limitations to our study should be considered. First, our study only includes samples from the mouse and human ENT; thus, it is not clear if our findings may extend to other brain regions. Second, we only examined two time points in the mouse, at 2 and 8 months of age. It is possible that if we examined more time points or different time points that we may observe better conservation in some of these groups. Third, levels of gene expression detected using snRNA-seq may not directly translate to protein levels, and the relationship between RNA and protein may be cell type-dependent. It is important to note other significant limitations to this study: 5xFAD mice do not model tau pathology, a significant limitation when comparing this model to human disease, and the human samples that were utilized could not be harmonized due to batch effects and limited information about the brain donors in previous studies. While differences between postmortem samples from individuals with and without cognitive impairment were significant, the possibility that they were influenced by technical effects, differences in genotype, or other confounding variables cannot be eliminated. Results concerning MCI should be interpreted with particular caution, as our dataset only included one sample for this condition.

Despite these limitations, at face value, our results suggest that the set of features of human AD that are best modeled in the 5xFAD model may be changes in neuronal signaling processes and intrinsic excitability, rather than glial cell processes. It is perhaps not surprising that only certain features of human disease are well-modeled in rodents. For example, AD is a complex disease that includes, among other features, progressive aggregation of amyloid and hyperphosphorylated tau. However, most mouse models of AD pathogenesis only model one core feature of AD, for example amyloidogenesis in the 5xFAD mice. Therefore, molecular and cellular responses related to amyloid plaque aggregation and clearance in human AD may be well-recapitulated in the 5xFAD mouse, whereas molecular and cellular responses related to tau may not be. This, however, also raises the more general problem of how useful rodent models of AD are in studying human AD. No rodent model will perfectly capture all the dimensions of human AD, but perhaps we only need them to recapitulate the essential features of AD for the models to be sufficiently useful for therapeutic development. Indeed, the excitotoxicity hypothesis of AD has existed for many decades, though it has somewhat fallen out of favor, replaced in large part by the amyloid hypothesis, as well as involvement of other processes, such as the immune system^117,118. It is possible that all of these are correct and interrelated; perhaps amyloid deposition triggers engagement of glia and inflammatory processes, which then effect changes in neuronal signaling that then drive excitotoxicity. If this is true, it is even more critically important to understand how changes in inter-cellular communication between glia and neurons may contribute to disease development and severity. Furthermore, it is important to leverage the strengths of the mouse lines that are used, and acknowledge their limitations, so that we can best focus on pathways and features of disease that are conserved between the mouse and human to maximize the translatability of our findings.

Rigor in RNA sequencing studies

Reproducibility in science is of critical importance. Lack of rigor in terms of underpowered studies (e.g., in behavior, drug trials, snRNA-seq, multi-omics, etc.) can limit the translatability of research findings. We note that the rigor of our work is in line with other previously published studies^{119,120,121,122,123}and we utilized field-standard statistical approaches to analyze our results. In addition, to provide some additional rigor for our work, we used jackknife resampling to obtain confidence intervals for the log-fold change values and average adjusted p-values over 100 iterations of DGE analysis (e.g., Supplemental Table 3). Regardless, the standards for ‘omics studies, in particular snRNA-seq, have evolved as a more sophisticated understanding of appropriate statistical comparisons that reduce the false discovery rate (for example, see¹²⁴). Given our current understanding of the strengths and limitations of snRNA-seq and mutli-omics studies, a critical view should be considered for all studies utilizing these approaches, both past and present. As such, our study may be underpowered, as practical issues prevented us from including more samples that would have increased the power of our work. The number of distinct samples included, particularly in human tissue that is likely more variable than tissue from inbred rodents, is an important factor towards generating accurate and reproducible results, Therefore, while our results provide interesting hypotheses for fundamental biological processes that contribute to AD, as with all similar studies these should be viewed as hypotheses that require further validation. Acknowledging the limitations to such studies is critical towards a nuanced and careful analysis of ‘omics data in AD. Given the sheer volume of ‘omics studies in AD, the sample size should be viewed as a critical factor towards the confidence in interpreting the results of the study.

Data availability

Data Availability: The datasets generated and analyzed during the current study are available in Gene Expression Omnibus (GEO) under GEO accession number GSE296091.

References

Galton, C. J., Patterson, K., Xuereb, J. H. & Hodges, J. R. Atypical and typical presentations of Alzheimer’s Disease: A clinical, neuropsychological, neuroimaging and pathological study of 13 cases. Brain 123 (2000).
Busche, M. A. & Konnerth, A. Impairments of neural circuit function in Alzheimer’s disease. Philos. Trans. R. Soc. B 371. https://doi.org/10.1098/rstb.2015.0429 (2016).
Busche, M. A. & Hyman, B. T. Synergy between amyloid-β and tau in Alzheimer’s disease. Nat. Neurosci. 23, 1183–1193. https://doi.org/10.1038/s41593-020-0687-6 (2020).
De Strooper, B. & Karran, E. The cellular phase of Alzheimer’s Disease. Cell 164, 603–615. https://doi.org/10.1016/j.cell.2015.12.056 (2016).
Grubman, A. et al. A single-cell atlas of entorhinal cortex from individuals with alzheimer’s disease reveals cell-type-specific gene expression regulation. Nat. Neurosci. 22, 2087–2097 (2019).
Article CAS PubMed Google Scholar
Mathys, H. et al. Single-cell transcriptomic analysis of alzheimer’s disease. Nature 570, 332–337 (2019).
Article ADS CAS PubMed PubMed Central Google Scholar
Chen, S. et al. Spatially resolved transcriptomics reveals genes associated with the vulnerability of middle Temporal gyrus in alzheimer’s disease. Acta Neuropathol. Commun 10 (2022).
Gabitto, M. I. et al. Integrated multimodal cell atlas of Alzheimer’s disease. bioRxiv. https://doi.org/10.1101/2023.05.08.539485 (2024).
Efremova, M., Vento-Tormo, M., Teichmann, S. A. & Vento-Tormo, R. CellPhoneDB: inferring cell–cell communication from combined expression of multi-subunit ligand–receptor complexes. Nat. Protoc. 15, 1484–1506 (2020).
Article CAS PubMed Google Scholar
Jin, S. et al. Inference and analysis of cell-cell communication using cellchat. Nat.. Commun. 12, (2021).
Chen, Y. et al. Exercise modifies the transcriptional regulatory features of monocytes in alzheimer’s patients: A Multi-Omics integration analysis based on single cell technology. Front. Aging Neurosci. 14 (2022).
Kiani Shabestari, S. et al. Absence of microglia promotes diverse pathologies and early lethality in alzheimer’s disease mice. Cell Rep 39 (2022).
Zhao, S., Ye, B., Chi, H., Cheng, C. & Liu, J. Identification of peripheral blood immune infiltration signatures and construction of monocyte-associated signatures in ovarian cancer and alzheimer’s disease using single-cell sequencing. Heliyon 9, (2023).
Li, H. et al. The role of signaling crosstalk of microglia in hippocampus on progression of ageing and alzheimer’s disease. J. Pharm. Anal. 13, 788–805 (2023).
Article PubMed PubMed Central Google Scholar
Dang, Y. et al. FTH1- and SAT1-induced astrocytic ferroptosis is involved in alzheimer’s disease: Evidence from single-cell transcriptomic analysis. Pharmaceuticals 15 (2022).
Zhao, W., Johnston, K. G., Ren, H., Xu, X. & Nie, Q. Inferring neuron-neuron communications from single-cell transcriptomics through neuronchat. Nat Commun 14, (2023).
Igarashi, K. M. Entorhinal cortex dysfunction in Alzheimer’s disease. Trends in Neurosciences46 124–136 Preprint at (2023). https://doi.org/10.1016/j.tins.2022.11.006
Geula, C. Abnormalities of Neural Circuitry in Alzheimer’s Disease Hippocampus and Cortical Cholinergic Innervation (1998). https://www.neurology.org
van Hoesen, G. W., Hyman, B. T. & Damasio, A. R. Entorhinal cortex pathology in Alzheimer’s disease. Hippocampus 1, 1–8 (1991). https://doi.org/10.1002/hipo.450010102
Di Paola, M. et al. Episodic memory impairment in patients with alzheimer’s disease is correlated with entorhinal cortex atrophy: A voxel-based morphometry study. J. Neurol. 254, 774–781 (2007).
Article PubMed Google Scholar
Forner, S. et al. Systematic phenotyping and characterization of the 5xFAD mouse model of alzheimer’s disease. Sci Data 8 (2021).
Oakley, H. et al. Intraneuronal β-amyloid aggregates, neurodegeneration, and neuron loss in Transgenic mice with five Familial alzheimer’s disease mutations: potential factors in amyloid plaque formation. J. Neurosci. 26, 10129–10140 (2006).
Article CAS PubMed PubMed Central Google Scholar
Ohno, M. et al. Temporal memory deficits in alzheimer ’ s mouse models: rescue by genetic deletion of BACE1. Eur. J. Neurosci. 23, 251–260 (2006).
Article PubMed Google Scholar
Stuart, T., Srivastava, A., Madad, S., Lareau, C. A. & Satija, R. Single-cell chromatin state analysis with Signac. Nat. Methods. 18, 1333–1341 (2021).
Article CAS PubMed PubMed Central Google Scholar
Hao, Y. et al. Integrated analysis of multimodal single-cell data. Cell 184, 3573–3587e29 (2021).
Article CAS PubMed PubMed Central Google Scholar
Luecken, M. D. & Theis, F. J. Current best practices in single-cell RNA‐seq analysis: a tutorial. Mol Syst. Biol 15 (2019).
Ilicic, T. et al. Classification of low quality cells from single-cell RNA-seq data. Genome Biol. 17, 29 (2016).
Article PubMed PubMed Central Google Scholar
Amemiya, H. M., Kundaje, A. & Boyle, A. P. The ENCODE blacklist: identification of problematic regions of the genome. Sci. Rep. 9, 9354 (2019).
Article ADS PubMed PubMed Central Google Scholar
Smith, A. M. et al. Diverse human astrocyte and microglial transcriptional responses to alzheimer’s pathology. Acta Neuropathol. 143, 75–91 (2022).
Article CAS PubMed Google Scholar
Korsunsky, I. et al. Fast, sensitive and accurate integration of single-cell data with harmony. Nat. Methods. 16, 1289–1296 (2019).
Article CAS PubMed PubMed Central Google Scholar
McInnes, L., Healy, J., Saul, N. & Großberger, L. UMAP: uniform manifold approximation and projection. J. Open. Source Softw. 3, 861 (2018).
Article Google Scholar
Waltman, L. & van Eck, N. J. A smart local moving algorithm for large-scale modularity-based community detection. Eur. Phys. J. B. 86, 471 (2013).
Article ADS Google Scholar
Hafemeister, C. & Satija, R. Normalization and variance stabilization of single-cell RNA-seq data using regularized negative binomial regression. Genome Biol. 20 (2019).
Ahlmann-Eltze, C., Huber, W. & glmGamPoi Fitting Gamma-Poisson generalized linear models on single cell count data. Bioinformatics 36, 5701–5702 (2020).
Article CAS PubMed Central Google Scholar
Yao, Z. et al. A taxonomy of transcriptomic cell types across the isocortex and hippocampal formation. Cell 184, 3222–3241e26 (2021).
Article CAS PubMed PubMed Central Google Scholar
Hodge, R. D. et al. Conserved cell types with divergent features in human versus mouse cortex. Nature 573, 61–68 (2019).
Article ADS CAS PubMed PubMed Central Google Scholar
Bennett, M. L. et al. New tools for studying microglia in the mouse and human CNS. Proc. Natl. Acad. Sci. U S A. 113, E1738–E1746 (2016).
Article CAS PubMed PubMed Central Google Scholar
McKenzie, A. T. et al. Brain cell type specific gene expression and Co-expression network architectures. Sci. Rep. 8 (2018).
Keirstead, H., Levine, J. & Blakemore, W. Response of the oligodendrocyte progenitor cell population (defined by NG2 labelling) to demyelination of the adult spinal cord. Glia 22, 161–170 (1998).
Article CAS PubMed Google Scholar
Zhang, Y. et al. Model-based analysis of ChIP-Seq (MACS). Genome Biol. 9 (2008).
Korsunsky, I., Nathan, A., Millard, N. & Raychaudhuri, S. Presto scales Wilcoxon and auroc analyses to millions of observations. BioRxiv https://doi.org/10.1101/653253 (2019).
Article Google Scholar
Morabito, S., Reese, F., Rahimzadeh, N., Miyoshi, E. & Swarup, V. HdWGCNA identifies co-expression networks in high-dimensional transcriptomics data. Cell Rep. Methods 3 (2023).
Langfelder, P. & Horvath, S. WGCNA: An R package for weighted correlation network analysis. BMC Bioinform. 9 (2008).
Morabito, S. et al. Single-nucleus chromatin accessibility and transcriptomic characterization of alzheimer’s disease. Nat. Genet. 53, 1143–1155 (2021).
Article CAS PubMed PubMed Central Google Scholar
Chen, E. Y. et al. Enrichr: Interactive and Collaborative HTML5 Gene List Enrichment Analysis Tool (2013). http://amp.pharm.mssm.edu/Enrichr
Kuleshov, M. V. et al. Enrichr: a comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res. 44, W90–W97 (2016).
Article CAS PubMed PubMed Central Google Scholar
Aleksander, S. A. et al. The gene ontology knowledgebase in 2023. Genetics 224 (2023).
Ashburner, M. et al. Gene Ontology: Tool for the Unification of Biology The Gene Ontology Consortium* (2000). http://www.flybase.bio.indiana.edu
Zhang, L., Zhang, J. & Nie, Q. DIRECT-NET: an efficient method to discover cis-regulatory elements and construct regulatory networks from single-cell multiomics data. Sci. Adv. 8, 7393 (2022).
Article ADS Google Scholar
Korsunsky, I., Nathan, A., Millard, N. & Raychaudhuri, S. Presto scales Wilcoxon and auroc analyses to millions of observations. BioRxiv (2019).
Hagberg hagberg, A. A., -Los, Schult, D. A. & Swart swart, P. J. Exploring Network Structure, Dynamics, and Function Using NetworkX (2008). http://conference.scipy.org/proceedings/SciPy2008/paper_2
Habib, N. et al. Disease-associated astrocytes in alzheimer’s disease and aging. Nat. Neurosci. 23, 701–706 (2020).
Article CAS PubMed PubMed Central Google Scholar
Xu, J. et al. Multimodal single-cell/nucleus RNA sequencing data analysis uncovers molecular networks between disease-associated microglia and astrocytes with implications for drug repurposing in alzheimer’s disease. Genome Res. 31, 1900–1912 (2021).
Article CAS PubMed PubMed Central Google Scholar
Sadick, J. S. et al. Astrocytes and oligodendrocytes undergo subtype-specific transcriptional changes in alzheimer’s disease. Neuron 110, 1788–1805e10 (2022).
Article CAS PubMed PubMed Central Google Scholar
Langfelder, P., Luo, R., Oldham, M. C. & Horvath, S. Is my network module preserved and reproducible? PLoS Comput. Biol 7, (2011).
Mishra, P. et al. Rescue of alzheimer’s disease phenotype in a mouse model by transplantation of wild-type hematopoietic stem and progenitor cells. Cell Rep. 42.(2023).
Scoyni, F. et al. Alzheimer’s disease-induced phagocytic microglia express a specific profile of coding and non-coding RNAs. Alzheimer’s Dement. https://doi.org/10.1002/alz.13502 (2023).
Article Google Scholar
Raft, M. C., Miller, R. H. & Noble, M. A Glial Progenitor Cell That Develops in Vitro into an Astrocyte or an Oligodendrocyte Depending on Culture Medium (1983).
Kondo, T. & Raff, M. Oligodendrocyte precursor cells reprogrammed to become multipotential CNS stem cells. Sci. (1979). 289, 1754–1757 (2000).
CAS Google Scholar
Rivers, L. E. et al. PDGFRA/NG2 glia generate myelinating oligodendrocytes and piriform projection neurons in adult mice. Nat. Neurosci. 11, 1392–1401 (2008).
Article CAS PubMed Google Scholar
Zhu, X., Bergles, D. E. & Nishiyama, A. NG2 cells generate both oligodendrocytes and Gray matter astrocytes. Development 135, 145–157 (2008).
Article CAS PubMed Google Scholar
Zou, P., Wu, C., Liu, T. C. Y., Duan, R. & Yang, L. Oligodendrocyte progenitor cells in Alzheimer’s disease: from physiology to pathology. Transl. Neurodegener https://doi.org/10.1186/s40035-023-00385-7 (2023).
Buchanan, J. A. et al. Oligodendrocyte precursor cells ingest axons in the mouse neocortex. Proc Natl. Acad. Sci. U S A 119 (2022).
Auguste, Y. S. S. et al. Oligodendrocyte precursor cells engulf synapses during circuit remodeling in mice. Nat. Neurosci. 25, 1273–1278 (2022).
Article CAS PubMed PubMed Central Google Scholar
Akay, L. A., Effenberger, A. H. & Tsai, L. H. Cell of all trades: oligodendrocyte precursor cells in synaptic, vascular, and immune function (2021). https://doi.org/10.1101/gad.344218
Shahoha, M., Cohen, R., Ben-Simon, Y. & Ashery, U. cAMP-Dependent synaptic plasticity at the hippocampal mossy fiber terminal. Front. Synapt. Neurosci. https://doi.org/10.3389/fnsyn.2022.861215 (2022).
Wong, S. T. et al. Calcium-stimulated adenylyl cyclase activity is critical for hippocampus-dependent long-term memory and late phase LTP. Neuron 23 (1999).
Fieblinger, T., Perez-Alvarez, A., Lamothe‐Molina, P. J., Gee, C. E. & Oertner, T. G. Presynaptic cGMP sets synaptic strength in the striatum and is important for motor learning. EMBO Rep 23 (2022).
Ricciarelli, R. et al. A novel mechanism for Cyclic adenosine Monophosphate-Mediated memory formation: role of amyloid Beta. https://doi.org/10.1002/ana
Canepa, E. et al. Cyclic adenosine monophosphate as an endogenous modulator of the amyloid-β precursor protein metabolism. IUBMB Life. 65, 127–133 (2013).
Article CAS PubMed Google Scholar
Calcagno, E. et al. Investigating the amyloid-beta enhancing effect of cGMP in neuro2a cells. Mech. Ageing Dev. 166, 1–5 (2017).
Article CAS PubMed Google Scholar
Gentile, J. E., Carrizales, M. G. & Koleske, A. J. Control of synapse structure and function by actin and its regulators. Cells. https://doi.org/10.3390/cells11040603 (2022).
Gerber, K. J., Squires, K. E. & Hepler, J. R. Roles for regulator of g protein signaling proteins in synaptic signaling and plasticity. Mol. Pharmacol. 89, 273–286. https://doi.org/10.1124/mol.115.102210 (2016).
Catterall, W. A. & Few, A. P. Calcium Channel Regulation and Presynaptic Plasticity. Neuron 59, 882–901. https://doi.org/10.1016/j.neuron.2008.09.005 (2008).
Poppe, L. et al. EphA4 loss improves social memory performance and alters dendritic spine morphology without changes in amyloid pathology in a mouse model of alzheimer’s disease. Alzheimers Res. Ther. 11 (2019).
Fu, A. K. Y. et al. Blockade of EphA4 signaling ameliorates hippocampal synaptic dysfunctions in mouse models of alzheimer’s disease. Proc. Natl. Acad. Sci. U S A. 111, 9959–9964 (2014).
Article ADS CAS PubMed PubMed Central Google Scholar
Tamura, K., Chiu, Y. W., Shiohara, A., Hori, Y. & Tomita, T. EphA4 regulates Aβ production via BACE1 expression in neurons. FASEB J. 34, 16383–16396 (2020).
Article CAS PubMed Google Scholar
Baudet, S., Bécret, J. & Nicol, X. Approaches to manipulate ephrin-a:Epha forward signaling pathway. Pharmaceuticals 13, 1–25 (2020).
Article Google Scholar
Sobue, A. et al. Microglial gene signature reveals loss of homeostatic microglia associated with neurodegeneration of alzheimer’s disease. Acta Neuropathol. Commun 9 (2021).
Romanitan, M. O. et al. Altered expression of Claudin family proteins in alzheimer’s disease and vascular dementia brains. J. Cell. Mol. Med. 14, 1088–1100 (2010).
PubMed Google Scholar
Saito, T. et al. Single app knock-in mouse models of alzheimer’s disease. Nat. Neurosci. 17, 661–663 (2014).
Article CAS PubMed Google Scholar
Chen, W. T. et al. Spatial transcriptomics and in situ sequencing to study alzheimer’s disease. Cell 182, 976–991e19 (2020).
Article CAS PubMed Google Scholar
Thomas, R. et al. Epidermal growth factor prevents APOE4 and amyloid-beta-induced cognitive and cerebrovascular deficits in female mice. Acta Neuropathol. Commun. 4, 111 (2016).
Article PubMed PubMed Central Google Scholar
Crews, L. et al. Increased BMP6 levels in the brains of alzheimer’s disease patients and APP Transgenic mice are accompanied by impaired neurogenesis. J. Neurosci. 30, 12252–12262 (2010).
Article CAS PubMed PubMed Central Google Scholar
Sun, L. et al. Protective effects of bone morphogenetic protein 7 against amyloid-beta induced neurotoxicity in PC12 cells. Neuroscience 184, 151–163 (2011).
Article CAS PubMed Google Scholar
Mo, X. et al. PTPRM is critical for synapse formation regulated by zinc ion. Front Mol. Neurosci 15, (2022).
Cao, X., Dong, Y., Xing, L. & Zhang, X. A. Genome-wide association study of dementia using the electronic medical record. BioMedInformatics3, 141–149 (2023).
Zhou, X., Sullivan, P. M., Sun, L. & Hu, F. The interaction between progranulin and prosaposin is mediated by granulins and the linker region between Saposin B and C. J. Neurochem. 143, 236–243 (2017).
Article CAS PubMed PubMed Central Google Scholar
Andersson, A. et al. Development of parallel reaction monitoring assays for cerebrospinal fluid proteins associated with alzheimer’s disease. Clin. Chim. Acta. 494, 79–93 (2019).
Article CAS PubMed Google Scholar
Callizot, N. et al. AZP a new promising treatment for Alzheimer’s 2019 and related diseases. Sci. Rep. 11 (2021).
Masliah, E., Mallory, M., Alford, M., Deteresa, R. & Saitoh, T. PDGF Is Associated With Neuronal and Glial Alterations of Alzheimer’s Disease 16 (1995).
Quaegebeur, A., Segura, I. & Carmeliet, P. Pericytes: Blood-Brain Barrier Safeguards against Neurodegeneration? Neuron 68, 321–323. https://doi.org/10.1016/j.neuron.2010.10.024 (2010).
Maragos, W. F., Greenamyre, J. T., Penney, J. B. & Young, A. B. Glutamate Dysfunction in Alzheimer’s Disease: An Hypothesis (1987).
Conway, M. E. Alzheimer’s disease: targeting the glutamatergic system. Biogerontology 21, 257–274. https://doi.org/10.1007/s10522-020-09860-4(2020).
Hynd, M. R., Scott, H. L. & Dodd, P. R. Glutamate-mediated excitotoxicity and neurodegeneration in Alzheimer’s disease. Neurochemistry International 45, 583–595 (2004). https://doi.org/10.1016/j.neuint.2004.03.007 (2004).
Bhat, S. et al. CACNA1C (Ca v1.2) in the pathophysiology of psychiatric disease. Progr. Neurobiol. 99, 1–14. https://doi.org/10.1016/j.pneurobio.2012.06.001 (2012).
Bocharova, A. et al. Association and gene–gene interactions study of late-onset alzheimer’s disease in the Russian population. Genes (Basel) 12 (2021).
Zanos, P. et al. Sex-dependent modulation of age-related cognitive decline by the L-type calcium channel gene Cacna1c (Cav1.2). Eur. J. Neurosci. 42, 2499–2507 (2015).
Article PubMed PubMed Central Google Scholar
Daschil, N. et al. CaV1.2 calcium channel expression in reactive astrocytes is associated with the formation of amyloid-β plaques in an alzheimer’s disease mouse model. J. Alzheimer’s Disease. 37, 439–451 (2013).
Article CAS Google Scholar
Song, J. Y., Ichtchenko, K., Su¨dhofsu¨dhof, T. C. & Brose, N. Neuroligin 1 Is a Postsynaptic Cell-Adhesion Molecule of Excitatory Synapses. vol. 96 (1999). www.pnas.org
Chubykin, A. A. et al. Activity-Dependent validation of excitatory versus inhibitory synapses by Neuroligin-1 versus Neuroligin-2. Neuron 54, 919–931 (2007).
Article CAS PubMed PubMed Central Google Scholar
Zheng, J. J. et al. Low expression of aging-related NRXN3 is associated with alzheimer disease: A systematic review and meta-analysis. Medicine (United States) 97 (2018).
Aoto, J., Földy, C., Ilcus, S. M. C., Tabuchi, K. & Südhof, T. C. Distinct circuit-dependent functions of presynaptic neurexin-3 at GABAergic and glutamatergic synapses. Nat. Neurosci. 18, 997–1007 (2015).
Article CAS PubMed PubMed Central Google Scholar
Verduci, L., Tarcitano, E., Strano, S., Yarden, Y. & Blandino, G. CircRNAs: role in human diseases and potential use as biomarkers. Cell Death Dis. https://doi.org/10.1038/s41419-021-03743-3 (2021).
Cheng, X. et al. F-box protein FBXO30 mediates retinoic acid receptor γ ubiquitination and regulates BMP signaling in neural tube defects. Cell Death Dis 10, (2019).
Von Wittgenstein, J. et al. Sox11 is an Activity-Regulated gene with Dentate-Gyrus-Specific expression upon general neural activation. Cereb. Cortex. 30, 3731–3743 (2020).
Article Google Scholar
Mathys, H. et al. Single-cell atlas reveals correlates of high cognitive function, dementia, and resilience to alzheimer’s disease pathology. Cell 186, 4365–4385e27 (2023).
Article CAS PubMed PubMed Central Google Scholar
Li, X. W. et al. SCAD-Brain: a public database of single cell RNA-seq data in human and mouse brains with alzheimer’s disease. Front Aging Neurosci 15 (2023).
Park, H. et al. Single-cell RNA-sequencing identifies disease-associated oligodendrocytes in male APP NL-G-F and 5XFAD mice. Nat Commun 14 (2023).
Brase, L. et al. Single-nucleus RNA-sequencing of autosomal dominant alzheimer disease and risk variant carriers. Nat Commun 14 (2023).
Xu, X., Zhang, C., Tang, G., Wang, N. & Feng, Y. Single-cell transcriptome profiling highlights the role of APP in blood vessels in assessing the risk of patients with proliferative diabetic retinopathy developing alzheimer’s disease. Front Cell. Dev. Biol 11 (2023).
Liu, A., Fernandes, B. S., Citu, C. & Zhao, Z. Unraveling the intercellular communication disruption and key pathways in alzheimer’s disease: An integrative study of single-nucleus transcriptomes and genetic association. Alzheimers Res. Ther 16 (2024).
Cang, Z. et al. Screening cell–cell communication in Spatial transcriptomics via collective optimal transport. Nat. Methods. 20, 218–228 (2023).
Article CAS PubMed PubMed Central Google Scholar
Jensen, G. S., Leon-Palmer, N. E. & Townsend, K. L. Bone morphogenetic proteins (BMPs) in the central regulation of energy balance and adult neural plasticity. Metabolism. https://doi.org/10.1016/j.metabol.2021.154837 (2021).
Esquenazi, S., Monnerie, H., Kaplan, P. & Le Roux, P. BMP-7 and excess glutamate: opposing effects on dendrite growth from cerebral cortical neurons in vitro. Exp. Neurol. 176, 41–54 (2002).
Article CAS PubMed Google Scholar
Abe, K. & Saito, H. Protective effect of epidermal growth factor on glutamate neurotoxicity in cultured cerebellar neurons. Neurosci. Res. 14 (1992).
Chen, X. & Holtzman, D. M. Emerging roles of innate and adaptive immunity in Alzheimer’s disease. Immunity 55, 2236–2254. https://doi.org/10.1016/j.immuni.2022.10.016 (2022).
Wu, K. M. et al. The role of the immune system in Alzheimer’s disease. Ageing Res. Rev. https://doi.org/10.1016/j.arr.2021.101409 (2021).
Allen, W. E., Blosser, T. R., Sullivan, Z. A., Dulac, C. & Zhuang, X. Molecular and Spatial signatures of mouse brain aging at single-cell resolution. Cell 186, 194–208e18 (2023).
Article CAS PubMed Google Scholar
Ochocka, N. et al. Single-cell RNA sequencing reveals functional heterogeneity of glioma-associated brain macrophages. Nat. Commun 12 (2021).
Sauler, M. et al. Characterization of the COPD alveolar niche using single-cell RNA sequencing. Nat. Commun. 13 (2022).
Saraiva, L. R. et al. Hierarchical Deconstruction of mouse olfactory sensory neurons: from whole mucosa to single-cell RNA-seq. Sci. Rep. 5 (2015).
Lukassen, S., Bosch, E., Ekici, A. B. & Winterpacht, A. Characterization of germ cell differentiation in the male mouse through single-cell RNA sequencing. Sci. Rep. 8 (2018).
Mukamel, E. & Yu, Z. False positives in study of memory-related gene expression. Nature 642, E1–E3 (2025).
Article CAS PubMed Google Scholar

Download references

Acknowledgements

This work utilized the infrastructure for high-performance and high-throughput computing, research data storage and analysis, and scientific software tool integration built, operated, and updated by the Research Cyberinfrastructure Center (RCIC) at the University of California, Irvine (UCI). The RCIC provides cluster-based systems, application software, and scalable storage to directly support the UCI research community, https://rcic.uci.edu. This work was made possible, in part, through access to the following: the Genomics Research and Technology Hub Shared Resource of the Cancer Center Support Grant (P30CA-062203), the Single Cell Analysis Core shared resource of Complexity, Cooperation and Community in Cancer (U54CA217378), the Genomics-Bioinformatics Core of the Skin Biology Resource Based Center @ UCI (P30AR075047) at the University of California, Irvine and NIH shared instrumentation grants 1S10RR025496-01, 1S10OD010794-01, and 1S10OD021718-01. This work was funded by the Alzheimer’s Association (AARG-NTF-20-685694), New Vision Research (CCAD2020-002), Brightfocus Foundation (A2022031S), NIH (R00DA041445, DP2AG067666, R01NS130044, R01DA056599, 1R01DA054374), Tobacco Related Disease Research Program (T31KT1437, T31P1426), and One Mind (OM-5596678) to KTB, NIH T32GM008620 and NIH F30DA056215 to MH, an NSF grant, DMS1763272 and a grant from the Simons Foundation (594598) to QN, and NIH T32 GM136624 to KB.

Author information

Authors and Affiliations

Program in Mathematical, Computational, and Systems Biology, University of California, Irvine, Irvine, CA, 92617, USA
Katrina Bartas
Department of Physiology and Biophysics, University of California, Irvine, Irvine, CA, 92617, USA
Megan Nguyen, May Hui & Kevin T. Beier
Department of Mathematics, NSF-Simons Center for Multiscale Cell Fate Research, University of California, Irvine, Irvine, CA, 92617, USA
Wei Zhao & Qing Nie
The Center for Neural Circuit Mapping, University of California, Irvine, Irvine, CA, 92617, USA
Qing Nie
Department of Developmental and Cell Biology, University of California, Irvine, Irvine, CA, 92617, USA
Qing Nie
Department of Biomedical Engineering, University of California, Irvine, Irvine, CA, 92617, USA
Kevin T. Beier
Department of Neurobiology and Behavior, University of California, Irvine, Irvine, CA, 92617, USA
Kevin T. Beier
Department of Pharmaceutical Sciences, University of California, Irvine, Irvine, CA, 92617, USA
Kevin T. Beier

Authors

Katrina Bartas
View author publications
Search author on:PubMed Google Scholar
Megan Nguyen
View author publications
Search author on:PubMed Google Scholar
Wei Zhao
View author publications
Search author on:PubMed Google Scholar
May Hui
View author publications
Search author on:PubMed Google Scholar
Qing Nie
View author publications
Search author on:PubMed Google Scholar
Kevin T. Beier
View author publications
Search author on:PubMed Google Scholar

Contributions

KB and KTB performed all analyses for all data: MN and WZ assisted KB with data analysis; MH made figures with assistance from KB and KTB; QN provided support and guidance; KTB organized and led the project; KTB wrote the manuscript, with input from all authors.

Corresponding author

Correspondence to Kevin T. Beier.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1

Supplementary Material 2

Supplementary Material 3

Supplementary Material 4

Supplementary Material 5

Supplementary Material 6

Supplementary Material 7

Supplementary Material 8

Supplementary Material 9

Supplementary Material 10

Supplementary Material 11

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

About this article

Cite this article

Bartas, K., Nguyen, M., Zhao, W. et al. Analysis of changes in intercellular communications in Alzheimer’s disease reveals conserved changes in glutamatergic transmission in mice and humans. Sci Rep 15, 26248 (2025). https://doi.org/10.1038/s41598-025-10795-4

Download citation

Received: 21 November 2024
Accepted: 07 July 2025
Published: 19 July 2025
Version of record: 19 July 2025
DOI: https://doi.org/10.1038/s41598-025-10795-4

This article is cited by

Sex-related upregulation of bone morphogenetic protein signaling inhibits adult neurogenesis in APPNL−G−F alzheimer’s disease model mice
- Xingyu Su
- Rina Takayanagi
- Toshio Ohshima
Biology of Sex Differences (2025)