Mice
APPswe/PS1ΔE9 double-transgenic mice were sourced from the Model Animal Research Center of Nanjing University (Nanjing, China). These mice are derived from the B6.Cg-Tg (APPswe/PS1ΔE9) 85Dbo/Mmjax line, originally developed at The Jackson Laboratory (JAX#034832). Age-matched C57BL/6J wild-type (WT) littermates served as controls. Mice were housed under specific pathogen-free (SPF) conditions in individually ventilated cages (IVC) at 23°C with 50–60% relative humidity and maintained under a controlled light-dark cycle. Mice aged 21–28 days were separated from their parents for genotyping. Genomic DNA extracted from ear biopsy samples was subjected to PCR amplification using primers specific for APP and PS1 genes. All experimental protocols were conducted in accordance with guidelines approved by the Animal Use and Care Committee of the Shenzhen Peking University-The Hong Kong University of Science and Technology Medical Center (SPHMC), under protocol number 2022-014. Only male mice were included in the study. All efforts were made to minimize animal suffering and reduce the number of animals used.
Astrocytes isolation
Astrocytes were isolated from WT and AD model mice from 6-month-old. Tissue from each brain was pooled to constitute a single experimental unit. Before sample collection, mice were deeply anesthetized. The entire brain was decapitated and immediately placed in cold HBSS (Thermo Fisher, 14170138) devoid of calcium and magnesium ions, and kept on ice for subsequent processing. Enzymatic cell dissociation was performed using an Adult Brain Dissociation Kit (Miltenyi Biotec, 130-107-677) on program 37C_ABDK_01 of the gentleMACS™ Octo Dissociator with Heaters. Afterward, the cell suspension was washed with HBSS and the astrocytes were magnetically labeled with anti-ACSA-2 microbeads (Miltenyi Biotec,130-097-678), according to the manufacturer’s protocol. The resulting cell were passed through LS columns (Miltenyi Biotec, 130-042-401) placed on an OctoMACS™ manual separator and collect the ACSA-2-positive cell fraction.
Immunostaining and imaging
For fresh specimens, mice were anesthetized with pentobarbital and then perfused with ice-cold PBS until the irrigation fluid was completely clear. This was followed by a 10-minute perfusion with ice-cold 4% paraformaldehyde (PFA). The mouse brain tissues were subsequently fixed in 4% PFA for 12 hours, dehydrated in sucrose, and embedded in optimal cutting temperature compound for coronal sectioning. Sections were cut to a thickness of 20 µm.
For slicing after MALDI-MSI data acquisition, excess matrix was removed by washing in 100% ethanol (2×5 min), 75% ethanol (1×5 min) and 50% ethanol (1×5 min), after which this MSI-analyzed-tissue-section was fixed using 4% PFA for 10 min22 .
The sections were blocked and permeabilized for 1 hour in PBS containing 4% bovine serum albumin (BSA) and 0.4% Triton X-100. Anti-GFAP antibody (1:400, BOSTER, BA0056), Purified anti-β-Amyloid (1:200, 6E10, Biolegend, 803001) were incubated overnight at 4℃, followed by corresponding fluorescent-labelled secondary antibodies (Invitrogen) for 1h at room temperature and 1×DAPI for 5 min. Images were captured with Zeiss Celldiscoverer 7, OLYMPUS VS200 or 3DHISTECH Pannoramic MIDI.
RNA sequencing (RNA-seq) analysis
The RNA-seq dataset analyzed for this work was generated as part of another study23, where it is described in detail and are accessible via the GEO series accession number GSE137028. Briefly, the dataset includes samples from both WT and AD model mice at ages of 2, 4, 6, 9, and 12 months. Triplicate samples (n = 3) were used for all genotypes and time points. Astrocytes were isolated from fresh cerebral tissues of these mice and subjected to RNA sequencing. Sequencing was performed on an Illumina NovaSeq platform and 150-bp paired-end reads were generated. A reference genome index was constructed and the paired-end reads were aligned using Hisat2 v2.0.524. Feature Counts v1.5.0-p325 was employed to enumerate the reads aligned to each gene. Subsequently, the expression level of each gene was quantified as fragments per kilobase of transcript per million mapped reads (FPKM).
For the singular value decomposition (SVD) analysis, gene expression levels were quantified using FPKM values and ranked based on the median absolute deviation (MAD) to identify the top 10,000 genes. After selection, the data underwent a logarithmic transformation (log10) followed by Z-score normalization across columns. This processing resulted in a matrix, designated as Matrix A, encompassing expression data for 9,528 genes across 30 samples. These samples represented AD and WT groups across five time points (2, 4, 6, 9, and 12 months), with three replicates each. The SVD theorem26 states that the matrix A can be written as
$$\:A=U\varSigma\:{V}^{T}$$
,
where U (a 9528×9528 matrix) and V (a 30×30 matrix) are orthogonal and Σ (a 9528×30 matrix) is a diagonal matrix containing the singular values. These r singular values are the square roots of the non-zero positive eigenvalues of both AAT and ATA, sorted in descending order of magnitude. The different columns in the ΣVT matrix correspond to the times at which the corresponding expression data are measured, each row in ΣVT matrix represents an adjusted feature mode, capturing distinct patterns of gene expression influenced by the magnitude of the corresponding singular values. We termed the first row of ΣVT as mode 1. The contribution of each gene to mode 1 is calculated by the squared values of the first column of the left singular vector \(\:{c}_{j}={\left({U}_{j,1}\right)}^{2}\). Based on this gene ranking, we conducted a Gene Set Enrichment Analysis (GSEA)27 within Gene Ontology Biological Processes (GO-BP)28.
For the Short Time-series Expression Miner (STEM)29 analysis, gene expression levels were quantified using FPKM values and ranked based on the MAD to identify the top 10,000 genes. After selection, calculating the ratios of the average expression levels at each time point (2, 4, 6, 9, and 12 months) for the AD group relative to the corresponding WT group averages. These ratios were then transformed using a base-2 logarithm. The processed data were imported into the STEM software for analysis. The parameters set for the analysis included a maximum number of model profiles at 20 and the clustering method as the STEM clustering method, with all other settings remaining at their default values.
Differentially expressed genes (DEGs) between specified subgroups were identified utilizing the DESeq230 package, with a significance threshold set at an adjusted p-value of less than 0.05 and an absolute log2 fold change (log2FC) greater than 0.5. Enrichment analysis for GO-BP 28 and Kyoto Encyclopedia of Genes and Genomes (KEGG)31 pathways was conducted using the clusterProfiler32 package. Pathway enrichment results were visualized using the R package ggplot233 or Cytoscape34. In Cytoscape, the enrichmentMap plugin was employed to construct a similarity-based network of enriched biological processes. GO terms were represented as nodes, with edges drawn based on a combined similarity coefficient (Jaccard + Overlap score > 0.375). The Markov Cluster Algorithm (MCL) was used to annotate clusters of enriched GO terms. The genes in profile 5 were intersected with the downregulated DEGs at 6 months using the VennDiagram35 package.
Quantitative RT-PCR validation of selected genes
Astrocytes isolated from the brain tissues of 6-month-old AD and WT mice were used for RNA extraction. RNA was extracted following the instructions of the RNA-easy Kit (Vazyme, R701-01). The obtained RNA was reverse transcribed into cDNA using a reverse transcription kit (Promega, 1725124). Quantitative PCR (qPCR) was performed in triplicate in 96-well plates using a qPCR machine (LC480, Roche) and the SYBR Green I Master mixture (Roche, 4887352001) to detect amplification products. The following thermocycling protocol was used: initial denaturation at 95°C for 10 minutes, followed by 40 cycles of 95°C for 15 seconds and 60°C for 1 minute, with a final cycle at 25°C for 15 seconds. Relative quantification of mRNA expression was performed using the comparative cycle threshold (Ct) method, calculated as the ratio of the gene of interest to Gapdh. Gene expression levels were quantified using the 2−ΔΔCt method. All primers were designed using Primer Bank36, and the primer sequences are listed in Additional file: Table S3.
Protein extraction and mass-spectrometry
Protein from astrocytes of 6-month-old AD and WT mice (n = 3) was extracted and digested using the iST kit (PreOmics) following the manufacturer's protocol. Briefly, the process involved lysing the cells, digesting the proteins enzymatically, and preparing the peptides for subsequent analysis. Briefly, add LYSE reagent with each astrocyte sample and incubate for 10 mins at 95°C. Centrifuge to collect supernatant. Following lysis, proceed to enzymatic digestion by adding DIGEST and RESUSPEND agents, incubate for 60 mins at 37°C, then halt the reaction with STOP solution. Purification involves twice centrifugation steps with wash solutions, followed by twice elution steps into a fresh tube. Finally, dry the purified peptide for 4D-label free liquid chromatography-mass spectrometry (LC-MS) analysis.
For label-free LC-MS analysis, samples were loaded onto a 25 cm C18 analytical column (Ionopticks) at a flow rate of 300 nL/min and subjected to gradient elution using solvents A (H2O-FA, 99.9:0.1 v/v) and B (ACN-FA, 99.9:0.1 v/v). The gradient conditions were set as follows: from 2–22% B over 75 min, 22–37% B from 75 to 80 min, 37–80% B from 80 to 85 min, and held at 80% B for the final 5 min. The QE mass spectrometer (Thermo, USA) operated with a capillary voltage of 1.5 kV, drying gas at 180°C and a flow rate of 3.0 L/min. The mass spectrometer scanned from 100 to 1700 m/z, with collision energy settings adjustable between 20 and 59 eV to optimize ion fragmentation.
For the database search conducted using MaxQuant, utilizing the Andromeda search engine for label-free quantification (LFQ) analysis. The search was controlled with a false discovery rate (FDR) of 0.01 to limit the identification of false positives. Enzymatic digestion was modeled with trypsin allowing for up to two missed cleavages. A tolerance of 20 parts per million (ppm) for both MS and MS/MS. Fixed and variable modifications were defined as carbamidomethyl (C) and oxidation (M) respectively. The database utilized was the 'uniprot-reviewed_yes + taxonomy_10090.fasta', incorporating sequences specifically reviewed for mouse. A decoy database with a reverse pattern was used to assess the rate of false identifications.
Proteomics data analysis
For data preprocess, proteins absent in more than three samples were excluded using the 50% rule. The limma37 package was used for normalization and differential expression analysis between the WT and AD groups, setting a threshold of p < 0.05 and an absolute log2FC greater than 0.3 to identify differentially expressed proteins (DEPs). Gene regulation at the RNA and protein levels was compared using the VennDiagram35 package. Pathway analysis was performed using KEGG31 pathways as described in the RNA-seq analysis section.
Sample preparation for matrix-assisted laser desorption/ionization mass spectrometry imaging (MALDI-MSI)
For 6-month-old AD and WT mice (n = 3) anesthetized and killed by cervical dislocation. 10% carboxymethyl cellulose (Sigma) was used to embed brain tissue in the cassettes. The cassettes were then immediately submerged in an isopentane-dry ice bath until the medium solidified and turned white.
Tissues were stored at -80°C prior to analysis. One hour before use, they were transferred to -20°C, and sections were then prepared using a Leica CM1950 cryostat (Leica Microsystems GmbH). Tissue sections, 20 µm thick, were thaw-mounted onto ITO-coated microscopic slides (Bruker) for MALDI-MSI analysis. Adjacent sections were collected for immunofluorescence staining. The mounted tissue sections were dried for 30 minutes in a desiccator before matrix application, and subsequently vacuum-sealed and stored at -80°C.
The matrix consisted of 9-AA dissolved in 70% ethanol for MALDI-2 was prepared and applied to the conductive slides bearing tissue sections using a TM-Sprayer (HTX Technologies). The parameters of the matrix application set in the TM-sprayer were as follows: spray nozzle velocity (1200 mm/min), track spacing (3 mm), flow rate (0.12 mL/min), spray nozzle temperature (90°C), and nitrogen gas pressure (10 psi).
Mass spectrometry imaging
Metabolites in the samples were imaged using a timsTOF fleX MALDI 2 (Bruker) equipped with a 10 kHz smartbeam 3D laser. MALDI2 MSI was operated in positive ion mode in full scan mode for m/z 100–1200. MALDI2 MSI was performed at 20 µm spatial resolution. The frequency of the laser was set to 10,000 Hz, with 60% the laser energy and 400 laser shots per pixel.
MALDI-MSI Data Analysis
Raw imaging data were processed using SCiLS™ Lab 2021 (Bruker Daltonics), yield two datasets: the first detailing mass spectrometry data for each spot, including m/z values and signal intensities, and the second listing the X and Y coordinates of each spot. Metabolite identification was achieved by matching the accuracy of m/z values (< 10 ppm) against an in-house database and the Bruker Library MS-Metabobase 3.0 database.
A SeuratObject was subsequently constructed utilizing the Seurat38 package. Data normalization employed a dual strategy involving both 9-AA reference compound normalization and total ion count (TIC) normalization. To correct for batch effects in the analysis, we applied the ComBat method from the sva39 package.
This dataset was further used for uniform manifold approximation and projection (UMAP) analyses40. UMAP reduction results (cell.embeddings) were then used as input data for k-means clustering, specifying the number of clusters (centers) as five and visualized using ggplot233 package. Adjusted to four clusters to better reflect underlying patterns. The differential abundance of metabolites between clusters were analyzed using FindAllMarkers function in Seurat38, with criteria of average log2FC greater than 0.5 and pct.1 greater than 0.7.
Regions of interest (ROIs) were designated by identifying areas with accumulations of reactive astrocytes in the immunofluorescence-stained sections from MALDI-MSI analysis and their adjacent sections. Using a QuPath to SCiLS™ Lab software plugin in QuPath41, the delineated ROI files were then transferred into SCiLS™ Lab, where the position, contours, and names of the regions were displayed. Subsequently, this information was exported to previously constructed SeuratObject. Spots that were both within interested k-means cluster and belonged to the ROIs were identified. Differential expression analysis between spots within the ROIs and those outside it was performed using the FindMarkers function. Metabolic markers were identified based on criteria of an absolute log2FC greater than 0.5 and a p-value less than 0.05.
For integrated analysis, genes that are downregulated in astrocytes at the RNA level at 6 months of age, along with their fold changes, and the fold changes of all metabolites when comparing the ROIs to the unselected regions of cluster 1 were used as input for the Joint Pathway Analysis function in the MetaboAnalyst42 tool. This analysis was conducted to elucidate the interactions at the pathway level between transcriptomics and metabolomics, results were displayed as enriched KEGG pathways.
Western blot
Three 6-month-old mice were anesthetized and euthanized using the same method as previously described. The entire right hemisphere of the mouse brain and the left cortex and hippocampus were dissected. Tissues were lysed by adding a protein lysis buffer containing RIPA (Beyotime), 10% protease-inhibitor cocktail (Roche), and 1% phosphatase inhibitor (Thermo Fisher), followed by homogenization using a tissue lyser (IKA) in an ice-water mixture. Lysates containing 1 × SDS-PAGE Sample Loading Buffer (Beyotime) were denatured for 10 min at 95°C. Samples were separated on 12% SDS-PAGE and then transferred to PVDF membranes. Membranes were blocked with 5% skimmed milk (BBI life science) in tris-buffered saline tween (TBST) and incubated with primary antibodies at 4°C overnight and then washed and incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies for 1 h at room temperature. Finally, membranes were visualized with ECL substrate (Millipore). The primary and secondary antibodies used in this study were as follows: Perilipin-2/ADFP (1:500, Novus Biologicals, NB110-40877), GAPDH (1:1000, CST, 2118S), and HRP-AffiniPure Goat Anti-Rabbit IgG (H + L) (1:10000, Jackson ImmunoResearch, 111–035 − 003).
Single-nucleus RNA sequencing (snRNA-seq) analysis
The snRNA-seq data utilized in this study were obtained from the Allen Institute for Brain Science and are available through the Seattle Alzheimer’s Disease Brain Cell Atlas (SEA-AD)43 database (https://registry.opendata.aws/allen-sea-ad-atlas). The database contains sequencing samples from the temporal neocortex of 84 donors, who were elderly with ages at death ranging from a minimum of 65 to an average of 88 years. Based on quantitative neuropathological assessments, these individuals were classified into different stages of Alzheimer's disease neuropathologic change (ADNC): 9 donors as Not AD, 12 as Low, 21 as Intermediate, and 42 as High ADNC. For the purposes of this research, data about astrocytes of 9 donors without Alzheimer's neuropathologic changes and 12 with low levels of neuropathologic changes were downloaded for analysis.
For data preprocessing, the downloaded counts table and metadata were used to construct a SeuratObject utilizing the Seurat38 package. Quality control was then applied with criteria specifying that each cell must have more than 2,000 but fewer than 7,500 RNA sequencing features, and mitochondrial gene content less than 4%. Data normalization was carried out using the LogNormalize method with a scaling factor set at 10,000. Principal component analysis (PCA) was performed based on the top 2,000 most variable genes, followed by UMAP analysis using the top 30 principal components. Clustering was executed at a resolution of 0.5.
For the single-cell metabolic pathway analysis, the SCPA44 R package was utilized, offering a novel graph-based test to define pathway activity. Normalized data were derived from astrocytes with low ADNC and no ADNC. A comprehensive set of 243 metabolic pathways was compiled from the Hallmark45, KEGG31, and Reactome46 databases. Pathway comparisons were executed using the compare_pathways() function in SCPA, setting downsample = 5000 to manage data volume. The fatty acid metabolism score for each cell was calculated based on the expression levels of 157 genes from the "HALLMARK_FATTY_ACID_METABOLISM" pathway.
$$\:G=\left\{{g}_{1},{g}_{2},\dots\:,{g}_{n}\right\},\:\text{w}\text{i}\text{t}\text{h}\:n=157$$
$$\:S\left({c}_{i}\right)=\sum\:_{j=1}^{n}E({g}_{j},{c}_{i})$$
where \(\:G\) represents the genes in the pathway, \(\:{g}_{j}\) represents a specific gene within the pathway, and \(\:S\left({c}_{i}\right)\) denotes the fatty acid metabolism score of cell \(\:{c}_{i}\).
Analysis of dietary fatty acids and cognitive function
The National Health and Nutrition Examination Survey (NHANES) is a cross-sectional survey, designed to provide a representative sample of the US non-institutionalized civilian population47. For these analyses, the two cycles (2011/2012, 2013/2014) with information on saturated fatty acids (SFA), monounsaturated fatty acids (MUFA), and polyunsaturated fatty acids (PUFA) intake as well as cognitive performance measures, were used to conduct a cross-sectional analysis in the combined set of participants.
Total estimated dietary SFA, MUFA, and PUFA intake (grams, gm), was averaged over the two recall periods (if only the first day was available, that value was used instead of an average). Additionally, participants' intake from dietary supplements during the same periods was considered, with average intakes over the two days calculated when data was available. Total intake of SFA, MUFA, and PUFA was then derived by summing dietary and supplemental sources. Finally, the proportion of each type of fatty acid relative to the total fatty acids was calculated and labeled accordingly as SFAT, MFAT, and PFAT.
For 2011–2014, cognitive testing was performed for participants aged 60 years and older. There were three tests administered: the CERAD Word Learning sub-test (CERAD W-L) to assess immediate and delayed recall of new verbal information (memory sub-domain); the Animal Fluency test to assess categorical verbal fluency (component of executive function); and the Digit Symbol Substitution Test (DSST) to assess processing speed, sustained attention, and working memory.
The analysis encompassed a pooled dataset of 2,502 individuals aged 60 years and older from two survey cycles. Each participant had completed assessments for cognitive performance and provided detailed data on SFA, MUFA, and PUFA, as well as information on critical confounding factors including age, sex, body mass index (BMI), smoking status, alcohol consumption, family income (poverty income ratio, PIR), and educational attainment. Confounding factors are controlled by incorporating them as covariates within the regression model.
Data analysis was carried out using the survey48 R package. The Kruskal-Wallis test was first used to explore the relationship between quartile-based categories of fatty acid intake and cognitive function scores across four tests. Subsequently, linear regression models were employed to examine the relationships between fatty acid intake proportions and cognitive measures. Fatty acid intake was considered both as quartiles and as a continuous variable in the analysis.
Statistical analysis
For each computational analysis undertaken using previously published codes or resources, the statistical methodologies were adopted as specified in the respective publications and detailed in the corresponding sections of the methods. For bulk-RNA sequencing data, differential expression analysis was conducted using the `DESeq()` function with default settings from the DESeq230 package. In proteomics analysis, the `eBayes()` function from the limma37 package was applied, also under default settings. Spatial metabolomics data were analyzed using the `FindMarkers` or `FindAllMarkers` functions with the Wilcoxon test (`test.use = 'wilcox'`) from the Seurat38 package. snRNA-seq data were analyzed using the `compare_seurat()` function within the SCPA44 package. In vitro experiments involved statistical comparisons across multiple experimental groups using one-way analysis of variance (ANOVA) complemented by Dunnett's multiple comparison test, while the two groups’ comparisons were handled through unpaired t-tests.
Data and code availability
The proteomics data generated in this study have been deposited in iProX (IPX0009463000) and are publicly available. The spot and spectra data related to MALDI-MSI have been deposited in Figshare (https://doi.org/10.6084/m9.figshare.26778319.v1) and are publicly available. All original code has been deposited in GitHub (https://github.com/Jie-Z-159/AD-astrocytes-multiomics.git) and is publicly available.