Single cell transcriptomic proling of neurodegeneration mediated by tau propagation in a novel 3D neuron-astrocyte coculture model

Research into neurodegeneration has been hampered by lack of systems that accurately recapitulate neurodegenerative processes1. Propagation of tau through a ‘prion-like’ process has emerged as an important aspect of neurodegenerative diseases including Alzheimer’s disease2. However, molecular mechanism of tau propagation is still largely unknown and a human 3D cellular model is still lacking. Here, we report development of the AstAD system, which uses human iPSCs to create neuron-astrocyte spheroids that incorporates propagation of toxic tau oligomers3,4. Single cell transcriptomic proling reveals roles for ribosomes, TNF mediated neuroinammation and heat shock proteins (HSP) as major elements of the disease stress response. Treatment with the HSP90 inhibitor PU-H71, which is selective for the dysfunctional HSP epichaperome, demonstrates reduction of pathology and neurodegeneration in the AstAD system5. tau conformers. suggests that tau propagation is an important contributor to multiple human neurodegenerative diseases. Accumulating studies that exogenous tau oligomers are internalized, tracked, and propagated via release across cell junctions in and cell models, spreading disease pathology in a prion-like manner 2,10-18 . Most studies of tau have focused on tau brils, which propagate pathology but elicit little if any neurodegeneration 2,10-18 . In contrast, we have shown that exposure to tau oligomers propagates tau pathology and induces robust neurodegeneration 3,19 . The AstAD system incorporates propagation of tau oligomers to create a highly reproducible human 3D model of tauopathy. The AstAD system enables analysis of neurodegenerative processes while also enabling concomitant analysis of the disease responses of astrocytes, which have been noted to play a large role in tauopathies

To develop AstAD (diagramed in Fig. 1A) human iPSCs (hiPSCs) derived neural progenitor cells (NPCs) were directed into neuronal and astrocytic lineages. The neuronal lineage was created by NGN2 overexpression to produce hiPSC derived neuronal cell (hiNCs), while the astrocytic lineage was generated via small molecule-directed differentiation to produce hiPSC derived astrocytes (hiACs) (Fig.   1A, Suppl. Fig. 1A). The resulting hiNCs and hiACs were then combined into 3D culture termed asteroids (~2000 cells each) and allowed to develop over the course of three weeks (Suppl. Fig. 1B-C) 4 . Analysis of cell type markers showed robust differentiation of neurons that were positive for MAP2 and Tuj1 (Fig. 1B, Suppl. Fig. 1D) and astrocytes that were positive for S100β and GFAP (Fig. 1B, E, Suppl. Fig. 1D). Astrocytes incorporation was necessary for asteroid survival, and astrocytes exhibited processes that were striking for their length and complexity, as well as for the intimate association with developing neuronal processes (Suppl. Fig. 1B, D, Suppl. Vid. 1); the resulting highly arborized astrocytic morphologies resembled that observed in vivo.
Asteroid cellular phenotypes were also characterized by acquiring single cell RNA sequencing (scRNAseq) pro les for 46,420 single cells with an average of 2000 genes per cell from varying experimental conditions across a time course of three weeks (Suppl. Fig. 2). Visualization by uniform manifold approximation and projection (UMAP) demonstrated the separation of cells into trajectory guided clusters (Suppl. Fig. 3A). These clusters were identi ed as the expected cycling neuronal progenitors (CYC), astrocytes precursors (ASC_P), astrocytes (ASC), and two subpopulations of neurons (NEU_A, NEU_B) based on the expression of established cell type speci c genes including TOP2A and NES for cycling progenitors, SLC1A3, VIM, and CLU for astrocytes, and SNAP25, SYT1, MAP2, and STMN2 for neurons (Suppl. Fig. 3A-C). Manual cell typing was supported by the automated cell typing platform SingleR (Suppl. Fig. 3D). Notably, over 21 days of 3D co-culture (21 DIV3D) populations of cycling progenitor decreased while the populations of mature cell types increased (Fig. 2B, D). We proceeded to model tauopathy in asteroid cultures. Oligomeric tau (oTau) was isolated from 9-month old PS19 P301S tau mice by centrifugation as described previously 3,23 (Suppl. Fig. 4A). hiNCs were exposed to oTau (0.04 mg/mL) for 24 hrs after which they were washed and combined with hiACs to generate the self-aggregating three-dimensional asteroid AD model (AstAD) (Suppl. Fig. 4B). Analysis of the AstAD cultures over 21 DIV3D showed rapid development of tau pathology, including tau hyperphosphorylation, misfolding, oligomerization and brilization (Fig. 1). Statistically signi cant increases in hyperphosphorylated tau were evident at positions S202/5 and S262 (Suppl. Fig. 5C-H). Misfolding and oligomerization of tau were observed using the MC1 and TOMA2 antibodies respectively, with each showing similar patterns of evolution ( Fig. 1B-D, Suppl. Fig. 5A-B). Fibrillar tau pathology also evolved, but at a slower rate. Fibrillar tau reactive with the dye thio avine S became evident only at 21 DIV3D (Fig. 1F, G). The striking neuropathology was also associated with neurodegeneration by 21 DIV3D. Prominent neuronal injury was evident by 21 DIV3D with Fluoro Jade B ( Fig. 1 H-I), reduced immunolabeling of MAP2 (Suppl. Fig. 6A-B) and increased LDH release (Suppl. Fig. 6C).
A requirement for oTau in the seeding process was demonstrated by immunodepleting oTau, which prevented any subsequent tau pathology or neurodegeneration (Suppl. Fig. 4D). The role of seeded oTau was further explored by exposing hiNCs to FITC labeled oTau. hiNC neurons exposed to FITC labeled oTau was present in >75% of hiNCs 24 hrs after seeding, however the vast majority of the seeded tau was degraded by 5 DIV3D (Suppl. Fig. 4B-D), and remaining FITC showing no correlation with accumulating tau pathology (Suppl. Fig. 4D). These results indicate that tau pathology developing after 5DIV3D must have been generated largely from endogenous tau sources.
Astrocytes also showed striking pathophysiological responses reminiscent of that observed in the human brain. The astrocytic markers S100β and GFAP showed statistically signi cant increases in reactivity beginning at 14 DIV3D and became prominent at the 21 DIV3D time point (Fig. 1B, E, Suppl. Fig. 5E, I). Astrocytic processes in the oTau AstAD cultures showed more intimate localization with neurons than in the control asteroid cultures (Fig. 1B, Suppl. Vid. 1). Taken together, these results indicate that over 21 DIV3D, the AstAD system develops a range of responses analogous to observed in the brains of human subjects with tau-mediated neurodegeneration.
A comparative analysis of control and AstAD scRNA-seq pro les across 21 DIV3D revealed shared and cell type speci c transcriptional responses to oTau induced pathology ( Fig. 2A). UMAP projections demonstrate changes in the relative proportion of cell types between control and AstAD over the course of growth in culture (Fig.2B, D). The neuronal cluster NEU_A (containing excitatory and inhibitory neuronal markers) and astrocytic cluster ASC increased in abundance, while the NEU_B cluster (containing only excitatory neuronal markers) decreased (Fig. 2B, D).
Differential gene expression analysis between the control and AstAD conditions within these clusters identi ed striking transcriptional changes over the time course. Notably, transcriptional changes were evident in both the ASC and NEU groups by 7 DIV3D, suggesting a high degree of interaction between the cell types (Fig. 2C). The AstAD system exhibited regulation of ribosomal transcripts, heat shock protein (HSP), chaperone transcripts, and TNF neuroin ammatory associated intermediate early response (IER) transcripts (Fig. 2C, Suppl. Fig. 7). Unexpectedly, an increase in ribosomal transcripts including RPS26 and RPS28 (green) dominated the early response at 7 DIV3D, suggesting a robust stress response (Fig.  2C). In contrast, suppression was observed for HSP (red) and TNF (purple) associated transcripts (Fig.  2C). This was reversed with increasing oTau pathology at 21 DIV3D, in parallel with increases in stress response transcripts including the TNF neuroin ammatory JUN, FOS, EGR1, and IER5 and HSP transcripts HSPA1B, DNAJB1, HSPB1, HSPH1, HSP90AB1, and HSP90AA1 (Fig. 2C). By taking the average expression of all genes in a gene set for each cell we performed a comparative module analysis (see Methods) and demonstrated the observed transcriptional changes at 21 DIV3D in AstAD NEU_A align with transcripts upregulated in AD human brain 24,25 , while AstAD ASC changes have a strong correlation with reactive astrocyte markers 26 and AD speci c disease associated astrocytes 27 (Fig. 2E).
Functional enrichment analysis con rmed a strong ribosomal response at 7 DIV3D that weakened through 14 and 21 DIV3D (Fig. 2F, H). Pathways associated with neuron cell death, TNF neuroin ammation, and HSPs were enriched among NEU_A upregulated genes at 21 DIV3D (Fig. 2F, H). While ASC shared the TNF pathway enrichment, they also demonstrated a more robust cytokine, HSP, and cell activation stress response suggestive of astrogliosis (Fig. 2F, H). Ontology module analysis highlights these responses in the TNFa via NFKB, transcription from RNA polymerase II promoter in response to stress, positive regulation of neuron death, and the HSP dominated response to topologically incorrect protein pathways at 21 DIV3D (Fig. 2G, Suppl. Fig. 8). Importantly, comparative functional enrichment reveals concordant responses between AstAD and single-nuclei transcriptomic pro les of postmortem AD brain, in particular the HSP response indicating activation of similar pathophysiological cascades 25,27,28 (Suppl. Fig. 9).
The inability of neurons to cope with accumulating tau pathology is hypothesized to stimulate formation of a dysfunctional epichaperome (high molecular weight complexes composed of HSP90 and HSP70 chaperones) and render neurons prone to degeneration 29 . The strong transcriptional response of HSPs in AstAD prompted us to test whether improving chaperone function by dispersing the dysfunctional epichaperome could delay disease progression. The AstAD cultures were treated with the epichaperome speci c HSP90 inhibitor PU-H71 (1 µM, 3 days) that has been shown to disrupt the epichaperome complex ( Fig. 3A) 5 . Subsequent analysis of pathology in AstAD showed a striking reduction in tau pathology and a corresponding reduction of neurodegeneration at 21 DIV3D (Fig. 3). We observed a signi cant reduction in tau phosphorylation (pS202/5, pS262), misfolding (MC1 antibody), oligomerization (TOMA2 antibody) and tangle formation (ThioS) (Fig. 3B-E). Fluoro Jade B reactivity also decreased, indicating a decrease in neuronal cell death ( Fig. 3G-H). This reduction in neuronal stress and death was also accompanied by a decrease in astrogliosis, marked by decreased S100β and GFAP staining (Fig. 3D, F).
Analysis of the transcriptional responses by scRNA-seq supported the putative bene t observed for PU-H17, with the transcriptional pro le shifting to that observed early in the disease process, perhaps prior to a point where the system was overwhelmed by oTau ( Fig. 4). We observed a reduction in transcriptional changes associated with neuronal cell stress and death, including suppression of the TNF, HSP, neuron cell death, and transcriptional stress genes and modules ( Fig. 4B-E, Suppl. Fig. 8, 10). Interestingly, PU-H71 initiated a striking re-expression of ribosomal transcripts at week 3 similar to that observed in week 1 in neurons (Fig. 4B, D-E). PU-H71 ablated the heat shock protein 70 response in astrocytes, perhaps indicating that the basal level of chaperones had become su cient to cope with any oTau present (Suppl. Fig. 10A). These data suggest that dispersing the dysfunctional epichaperome was su cient to reduce oTau induced neuropathology and neurodegeneration in AstAD.
We have presented a novel method for modeling tau pathology in a human iPSC derived co-culture of neurons and astrocytes in a rapid and reproducible manner. oTau seeding in hiNCs before combination with the hiACs allows for a speci c analysis of how neuronal tau pathology interacts with surrounding astrocytes. This study provides the rst single cell transcriptomic study of an organoid based tauopathy model. The astrocytic response to neuronal injury was surprisingly rapid and synchronized. Astrocytes and neurons both mounted robust ribosomal responses, consistent with an adaptation to stress. Astrocytes were notable for mounting a more robust HSP response, perhaps contributing to their underlying resilience and enabling regulation of neuronal response 30,31 . Treatment with PU-H71 is known to disrupt the dysfunctional epichaperome, producing more functional molecular chaperones (HSP70 and HSP90) 5 . We saw a corresponding normalization of toxic TNF and neuronal death transcriptional signatures, as well as a reactivation of the ribosomal stress response, suggesting that this can be bene cial. It is notable that this was a short treatment with PU-H71, indicating that epichaperome inhibition could be a feasible option for treating the development of tauopathies at varying stages of progression.

Materials And Methods oTau Processing
Generation of S1p fraction: Frozen hippocampus and cortical tissues of 9-month old PS19 mice were weighed (100mg-250mg) and put in Beckman Centrifuge Tube, polycarbonate thick wall (cat # 362305). A 10 × volume of homogenization buffer was used to homogenize brain tissue with Hsaio TBS buffer (50 mM Tris, pH 8.0, 274 mM NaCl, 5 mM KCl) supplemented with protease and phosphatase inhibitor cocktails (Roche, cat#05892791001 and cat#04906837001), as described previously 3,23 . Brie y, the homogenate was centrifuged at 48,300 g for 20 min at 4 °C. The supernatant was then centrifuged a second time at 186,340 g at 4°C for 40 min. The TBS-extractable pellet (S1p) fraction was resuspended in a 4x volume of TE buffer relative to the starting weight of the tissue homogenate, aliquoted and frozen at -80°C. S1p fraction quanti cation: Immuno-depletion of tau from S1p fraction: Tau aggregates in S1p fractions were eliminated from the fractions by a direct immuno-precipitation kit (Pierce, cat# 26148). Brie y, rst tau-5 antibody was coupled to AminoLink plus Coupling Resin, and the fractions were pre-cleared using the Control Agarose Resin with all the materials provided by the kit. The sample was added to the antibody-coupled resin in the spin column and incubated in the column for overnight at 4°C on a gentle rotator. The column was centrifuged, and the ow-through saved for further experimentation. After 3 washes with IP buffer, the spin column was placed into a new collection tube, and tau plus antibodies were eluted from the resin.
The eluate was analyzed for presence of tau. hiNC oTau Treatment: hiNCs were selectively exposed to 0.04 mg/mL oTau by direct administration in cell culture media for 24 hours before incorporation into asteroid culture.
PU-H71 Treatment: Asteroids were treated with 1 uM PU-H71 by direct administration in cell culture media for 72 hours before timepoint collection.

Sample Collection
Asteroid Fixation: At time of collection asteroids were transferred to a 1.5 mL Protein LoBind Eppendorf (Eppendorf 022-43-108-1) and allowed to settle. The supernatant was discarded and asteroids xed in 4°C 4% PFA in 1X PBS for 15 minutes, rotating at room temperature. After xation, asteroids were washed 3x for 5 minutes each with 4°C 1X PBS, rotating at room temperature. Samples were stored in 1X PBS at 4 °C.
Conditioned Media Collection: 50 uL replicates of conditioned cell culture media from replicate asteroids were collected in at bottomed 96 well plates and frozen at -20 °C.
Thio avin S staining: The fresh made Thio avin S (ThioS) solution was prepared by dissolving 1g of ThioS (Millipore Sigma, Cat# T1892) in 100ml 80% ethanol and was kept stirring for overnight at 4 °C before ltered for nal use. The asteroids to be stained were washed sequentially in 70% and 80% ethanol, 1 min each, prior to incubating in ThioS/80% ethanol solution for 15 min. Asteroids were then sequentially washed in 80% and 70% ethanol, 1 min each, followed by two rinses in PBS. Asteroids were mounted in Prolong Gold antifade reagent and stored in the dark until imaging.
Flouro Jade B staining: The Flouro jade B reagent was purchased from EMD Millipore (Cat# AG310-30MG) and the staining protocol was followed as instructed by the manufacture. Brie y, the staining solution was prepared from a 0.01% stock solution for Fluoro-Jade B that was made by adding 10 mg of the dye powder to 100 mL of distilled water. To make up 100 mL of staining solution, 4 mL of the stock solution was added to 96 mL of 0.1% acetic acid vehicle. This results in a nal dye concentration of 0.0004%. The stock solution, when stored in the refrigerator was stable for months, whereas the staining solution was typically prepared within 10 minutes of use and was not reused. Before staining, the asteroids were rinsed in distilled water and were then treated with 0.06% KMnO4 solution for 15 min. Then the asteroids were stained with FluoroJade B working solution for 30 min followed by being washed with PBS 5 min twice. Asteroids were mounted in Prolong Gold antifade reagent and stored in the dark until imaging.
LDH Cytotoxicity Assay: The CytoTox 96 Non-Radioactive Cytotoxicity Assay was performed as per manufacturer's instructions using 50 uL conditioned media replicates to measure lactate dehydrogenase (LDH) release (Promega G1780). Absorbance readings at 490 nm were taken on a SpectraMaxM5plate reader with SoftMax Pro 7.1 software.
Single Cell RNA Sequencing Sample Preparation and Sequencing: 30 asteroid per condition were pooled in a 1.5 mL Protein LoBind Eppendorf (Eppendorf 022-43-108-1) and allowed to settle. The supernatant was carefully discarded and a single cell suspension was produced by incubation in 500 uL digestion buffer (Accutase TM with 80 U/mL Protector RNase Inhibitor (Sigma-Aldrich 03335402001)) for 1 hour at 37 °C with gentle pipette mixing every 10 minutes. At the end of the incubation the single cell suspension was washed with 500 uL wash buffer (0.02% BSA in 1X PBS with 80 U/mL Protector RNase Inhibitor) and passed through a 20 uM lter (MACS, Miltenyi Biotec 130-101-812) to a fresh 2 mL Protein LoBind Eppendorf. An additional 1 mL of wash buffer was then passed through the same lter for a total single cell suspension of 2 mL. The samples were centrifuged at 300 g for 5 minutes at 4 °C followed by another was in 1 mL wash buffer. After another centrifugation the supernatant was discarded and the single cell pellet gently resuspended in 50 uL wash buffer. Cells were counted in quadruplicate on the Cellometer K2 with AOPI and processed through the single cell RNA-sequencing pipeline from 10X Genomics, 3' Version 3 (10X Genomic Chromium).
Brie y, the single cell suspension was mixed with RT reaction mix to target a 8000 cell recovery and 75 ul was loaded onto a chromium micro uidics chip with 40 μL of barcoded beads and 280 μL of partitioning oil. The chip was run on the chromium controller, encapsulating a single cell and barcoded bead within individual oil droplets. Reverse transcription was performed within these individual oil droplets to produce barcoded cDNA. cDNA was then isolated by Silane DynaBeads (Thermo Fisher Scienti c, Dynabeads MyONE Silane, Cat# 37002D) before PCR ampli cation. Ampli ed cDNA cleanup and size selection was performed using SPRIselect beads (Beckman-Coulter, SPRIselect, Cat# B23317) and cDNA quality was assessed by the High-Sensitivity DNA assay (on the Agilent 2100 BioAnalyzer (Agilent, High-Sensitivity DNA Kit, Cat# 5067-4626). Sequencing libraries were then prepared according to 10X speci cation, including fragmentation, sequencing adaptor ligation, and sample index PCR. Between each of these steps, library cleanup and size selection was performed by SPRIselect beads. Final cDNA library quality was assessed by the Agilent BioAnalyzer High-Sensitivity DNA assay and the Qubit High-Sensitivity DNA assay and quality-con rmed libraries were sequenced on Illumina's NextSeq 500 platform to a depth of 200 million paired-end reads.

Data Analysis
Images Analysis: Images were captured by Carl Zeiss confocal LSM700. The immuno-uorescence stained DAPI-positive cells in each image of asteroids were quanti ed by Image J with function of automatically cell counting. The staining intensity in immuno-uorescence labeled asteroids were measured by ImageJ. The intensity of MC1, TOMA2, CP13, AT8, ThioS and Fluoro-jade B were normalized by DAPI numbers. Schematics were created with BioRender.com.
GraphPad Prism Statistical analysis :Statistical analyses and gures artwork were performed using GraphPad Prism version 9.00 for Windows with two sided α of 0.05. All group data are expressed as mean ± SEM. Colum means were compared using one-way ANOVA with treatment as the independent variable. And group means were compared using two-way ANOVA with factors on oTau treatment and time points, respectively. When ANOVA showed a signi cant difference, pair wise comparisons between group means were examined by Tukey's, Dunnett or uncorrected Fisher's LSD multiple comparison test.
Signi cance was de ned when p< 0.05. LDH assay data analysis was performed in with a paired t-test.
Single Cell RNA Sequencing Data Analysis CellRanger Pipeline: CellRanger version 3.1.0 (10X Genomics) was used to combine and process the raw Illumina NextSeq 500 RNA sequencing data. First each sequencing library was demultiplexed by sample index to generate FASTQ les for paired-end reads using the CellRanger mkfastq pipeline. FASTQ les were then passed to the CellRanger count pipeline, which used STAR aligner 32 to align reads to the human reference genome (GRCh38). The CellRanger aggr pipeline was then used to equalize the aligned molecule_info.h5 sample libraries across sequencing depths (by each sample cell being down-sampled to have the same con dently mapped reads per cell) and aggregated together to generate the gene-cell barcode matrix.
Seurat Object Filtration: Subsequent ltering, normalization, and scaling of data was performed using Seurat version 3.2.2 33,34 . The Seurat object was created with a min.cells of 3 and a min.features of 200. Cells with less than 200 and greater than 5000 detected genes or greater than 20% mitochondrial counts were ltered out. Samples were subset to a max.cells.per.ident of 4642. Gene counts for each cell were normalized by total expression, multiplied by a scale factor of 10,000 and transformed to log scale. PCA based on the highly variable genes detected (dispersion of 2) was performed for dimension reduction and the top 20 principal components (PCs) were selected. We clustered cells based on graph-based methods (KNN and Louvain community detection method) implemented in Seurat. Clusters were visualized using uniform manifold approximation and projection (UMAP) 35 .
Cluster Cell Type Identi cation: To identify neuronal cell type subpopulations, we performed differential expression analysis using the Wilcoxon rank-sum test implemented in Seurat between previously de ned clusters with a min.pct or 0.1, logfc.threshold of 0.25, and a pseudocount of 1E4 . This identi ed top expressing genes for each cluster, which were then considered alongside the feature expression of canonical gene cell type markers to conclude cell type cluster identi cation.
DE analysis: Differential expression analysis was performed for each cell type between control and AstAD samples using the Wilcoxon rank-sum test implemented in Seurat with a min.pct of 0.1, a logfc.threshold of 0.1, and a pseudocount of 1E4. A multiple comparison correction was performed using the Benjamin & Hochberg FDR method to produce an adjusted p-value 36 . Differentially expressed genes were evaluated according to their log fold change (greater than log2(.25)) and adjusted p values (0.05). All gures were generated using the ggplot2 R package and associated EnhancedVolcano R package 37,38 .
Functional enrichment analysis: Functional geneset enrichment analysis of the signi cant differential genes between control and AstAD samples was performed using the R implemented GPro ler2 39 . The enrichment analysis was run as an ordered query (ordered by log2FC) using a threshold of 0.05 and using Benjamin & Hochberg FDR for multiple testing correction 36 . Only genes in the Seurat dataset were considered by using a custom domain scope. A custom source GMT, gp__zSEF_sD9Q_d1M, was used. It includes all Hallmark gene sets, curated gene sets, and ontology gene sets from the Molecular Signatures Database (MsigDB) v7.2 40,41 . The enrichment analysis was assessed and visualized by a heatmap of signi cance (−log10(p value)) of the top 20 enriched pathways per sample comparison. Comparative functional geneset enrichment analysis between AstAD and published datasets was performed with the same gpro ler2 settings, but as a non-ordered query. Gene input was as follows: From AstAD dataset, genes upregulated in NEU_A, NEU_B, and ASC at 21 DIV3D (n=129, p <0.05, log fold change > 0.25). From Grubman et al. 2020 25 genes upregulated in AD DEG 5 and 7 (n = 109, log fold change > 1 and FDR < 0.01).From Mathys et al. 2020 28 , genes upregulated in late-pathology cases that are common to ≥5 cell types (n=16, log fold change > 1 and FDR < 0.01)). The enrichment analysis was assessed and visualized by a heatmap of signi cance (−log10(p value)) of the top 10 enriched pathways per data set. All heatmaps were generated using the ComplexHeatmap R package and color scale generated using dependent R package circilize 42 . Additional visualization of signi cant (p< 0.01) enriched pathways was performed using Cytoscape EnrichmentMap 43,44 with an edge cutoff of 0.375. Gene sets in EnrichmentMap cluster by similarity, and annotates of shared gene set features were added manually using Cytoscape implemented AutoAnnotate.
Module analysis: The scaled expression per cell of literature curated and MsigDB ontology genesets (see Source Table 1. Genesets) was compared between control and AstAD samples by computing the mean expression using colMeans and performing a t.test across comparison pairs using stat_compare_means.
Figures were generated using the ggplot2 R package 37 .
Data Availability: Raw and processed scRNA-seq data are available from GEO under accession GSE165587. Processed scRNA-seq datasets are available on Single Cell Portal, including the cell barcodes, UMAP coordinates, and other available characteristics. The processed scRNA-seq data is available at https://singlecell.broadinstitute.org/single_cell/study/SCP1271/asteroid1. The source data underlying Fig. 2, 4 and Suppl. Fig. 4C, 7-10 are provided in the Source Data les as follows: Source data of differential gene expression and celltype_markers is available in supplemental le asteroid1_source le_deg and source data of gpro ler2 functional geneset enrichment, Source Table 1, Source Table 2, and Source Table 3 are available in supplemental le asteroid1_source le_gpro.
Code Availability: The original R scripts for Seurat processing are available on github [https://github.com/satijalab/seurat). All custom code to reproduce the analyses and gures reported in this paper are available on github (https://github.com/ChristineLab/asteroid1).