1. Introduction
Alzheimer's disease (AD), a severe neurodegenerative disorder, is the most common
type of dementia, affecting 57.4 million patients worldwide in 2019 and will reach
152.8 million by 2050 (GBD 2019 Dementia Forecasting Collaborators, 2022). However,
there is currently no effective treatment for AD (Gao et al., 2016). Part of the reason
is the lacking understanding of the high cellular heterogeneity in AD brains, which
is closely associated with the onset, progression, and pathological process of AD
(Lau et al., 2020). Single-cell RNA sequencing (scRNA-seq) is extremely effective
in interpreting the cellular heterogeneity of AD (Habib et al., 2020). Advances in
scRNA-seq technologies in the past decade have generated a large amount of AD scRNA-seq
data, but studies integrated these multi-source data are rare, which greatly limited
non-bioinformatics users' access to such public data. Therefore, it is essential to
establish a database that collects AD scRNA-seq data.
Here, we collected 17 AD or mild cognitive impairment (MCI) scRNA-seq projects from
the Gene Expression Omnibus (GEO) and the Synapse databases, covering 21 data-sets
with 359 samples, 10 brain regions, 16 major brain cell types, and 1,564,825 cells
to explore cellular heterogeneity, cell communication, and cell trajectory in brain
tissues with AD/MCI pathology. The results were integrated into a database named SCAD-Brain
(scRNA-seq analysis for AD Brain, https://www.bioinform.cn/SCAD/). The users can access,
reuse, and analyze the data by visiting the website, to perform cell marker analysis,
gene expression analysis, pathway enrichment analysis, cell communication analysis,
and cell trajectory analysis for multiple brain regions from patients with AD/MCI
and AD mouse models. SCAD-Brain provides a user-friendly platform to explore and visualize
scRNA-seq data of AD brain, assist in experiments design, verify hypotheses, and especially
investigate the cellular heterogeneity of AD brain.
2. Methods
2.1. Data collection and quality control
Raw data of AD brain scRNA-seq from the GEO database and Synapse database were collected
in October 2021. For data collection through the GEO database, medical subject headings
of AD and scRNA-seq were collected from the MeSH database to build a search formula:
[(Alzheimer Dementia) OR (Alzheimer Dementias) OR (Dementia, Alzheimer) OR (Alzheimer's
Disease) OR (Dementia, Senile) OR (Senile Dementia) OR (Dementia, Alzheimer Type)
OR (Alzheimer Type Dementia) OR [Alzheimer Type Dementia (ATD)] OR [Dementia, Alzheimer-Type
(ATD)] OR [Dementia, Alzheimer-Type (ATD)] OR(Alzheimer Type Senile Dementia) OR (Primary
Senile Degenerative Dementia) OR (Dementia, Primary Senile Degenerative) OR (Alzheimer
Sclerosis) OR (Sclerosis, Alzheimer) OR (Alzheimer Syndrome) OR (Alzheimer's Diseases)
OR (Alzheimer Diseases) OR (Alzheimers Diseases) OR (Senile Dementia, Alzheimer Type)
OR (Acute Confusional Senile Dementia) OR (Senile Dementia, Acute Confusional) OR
(Dementia, Presenile) OR (Presenile Dementia) OR (Alzheimer Disease, Late Onset) OR
(Late Onset Alzheimer Disease) OR (Alzheimer's Disease, Focal Onset) OR (Focal Onset
Alzheimer's Disease) OR [Familial Alzheimer Disease (FAD)] OR [Alzheimer Disease,
Familial (FAD)] OR [Familial Alzheimer Diseases (FAD)] OR (Alzheimer Disease, Early
Onset) OR (Early Onset Alzheimer Disease) OR (Presenile Alzheimer Dementia) AND [(single
cell) OR (single-cell) OR (single-nucleus) OR (single nucleus)], to search datasets
on the GEO database. For data collection through the Synapse database, AD scRNA-seq
data were collected from the AD Knowledge Portal, which is a repository for multi-omic
data on Alzheimer's disease and aging. The information of each sample was manually
confirmed according to the metadata and the related published studies. Irrelevant
samples (e.g., non-AD or ambiguous disease or AD with genes knocked out) were discarded.
After the above procedures, a total of 21 data-sets with 359 samples were collected
from public data portals. Raw scRNA-seq data were downloaded from the GEO database
and unpacked through the SRA Toolkit (version 2.11.1). The synapseclient and synapseutils
python packages were used to download raw scRNA-seq data from the Synapse database.
2.2. Contents of data projects
The AD brain scRNA-seq data projects we included in this study are: SRP290191 (Huang
et al., 2021), SRP238096 (Sierksma et al., 2020), SRP190819 (Otero-Garcia et al.,
2022), SRP254025 (Leng et al., 2021), SRP298630 (Yang et al., 2022), SRP243446 (Habib
et al., 2020), SRP252065(Alsema et al., 2020), SRP282056 (Lau et al., 2020), SRP282467
(Xu et al., 2021), SRP124513 (Mathys et al., 2017), SRP305673(Safaiyan et al., 2021),
syn18485175 (Mathys et al., 2019), syn24168322, syn16780177, syn12514624, syn21670836,
and syn23763501. There are seven mice studies. The SRP290191 and SRP238096 projects
explored the response of the microglia system to amyloid-β in pooled cortices (5xFAD,
also known as APP/PS1 mice model) and hippocampus (APPswe/PSEN1L166P and THY-Tau22
mice models), respectively. The SRP243446 project focuses on the disease-associated
astrocytes population in the 5xFAD mice model. The SRP282467 project elucidated single
cell expression of hippocampal tissues from the J20 AD mouse model. The SRP124513
and SRP305673 projects focus on microglia function in AD mice hippocampus (CK-p25,
of severe neurodegeneration which develops AD-like pathology in an inducible and temporally
predictable manner) and white/gray matter (Trem2KO mouse), respectively. The syn23763501
project performed scRNA-seq for the whole brain of the 5xFAD mouse model. There are
ten human studies. The SRP190819 project demonstrates the cell-type-specific responses
to neurofibrillary tangles susceptibility in AD prefrontal cortex. The SRP254025 project
focuses on selectively vulnerable neurons in AD caudal entorhinal cortex and the superior
frontal gyrus. The SRP298630 project concentrates on the vasculature system in the
AD hippocampus and superior frontal cortex. The SRP252065 project focuses on microglial
subpopulations in AD superior parietal lobe. The SRP282056 project focuses on angiogenic
endothelial cells and neuroprotective glia in AD prefrontal cortex. The syn24168322,
syn16780177, syn12514624, syn18485175, and syn21670836 projects provide scRNA-seq
data from the dorsolateral prefrontal cortex. The 17 data projects were additionally
separated into 21 datasets according to species and brain tissues: SRP243446H_Hippocampus_mice,
SRP243446C_Prefrontalcortex_mice, syn21670836_DorsolateralPrefrontalCortex_human,
syn24168322_DorsolateralPrefrontalCortex_human, syn18485175_DorsolateralPrefrontalCortex_human,
syn23763501_WholeBrain_mice, syn16780177_DorsolateralPrefrontalCortex_human, syn12514624_DorsolateralPrefrontalCortex_human,
SRP124513_Hippocampus_mice, SRP190819_PrefrontalCortex_human, SRP238096APP_Hippocampus_mice,
SRP238096TAU_Hippocampus_mice, SRP252065_SuperiorParietallLobe_human, SRP254025C_CaudalEntorhinalCortex_human,
SRP254025G_SuperiorFrontalGyrus_human, SRP282056_PrefrotnalCortex_human, SRP282467_Hippocampus_mice,
SRP290191_PooledvCortices_mice, SRP298630H_Hippocampus_human, SRP298630S_superiorFrontalCortex_human,
SRP305673_WhiteGreyMatter_mice.
2.3. Data processing
For 10x single cell data, CellRanger software (v.6.1.2) (10x Genomics) was used to
obtain gene counts by aligning reads to reference genomes (the hg38/GRCh38 for human
and mm10/GRCm38 for mouse). Samples with less than 300 cells detected were excluded.
For Smart-seq2 single cell data, the Trimmomatic (Bolger et al., 2014) (version 0.39)
with default parameters was used to remove adapter sequence and reads trimming, the
FastQC (version v0.11.9) was used for quality control (QC), the Hisat2 v2.2.1 (Kim
et al., 2019) was employed to align clean reads to mouse reference genome GRCm38,
the samtools v0.11.5 (Danecek et al., 2021), and the FeatureCount (Liao et al., 2014)
were used to obtain the scRNA expression matrix.
Seurat (Stuart et al., 2019) (version 4.1.0) R package was used to analyze the single
cell expression matrix. In detail, quality control of cells was conducted: (1) The
percentage of transcripts that map to mitochondrial genes in each cell was calculated;
(2) The cells that with >5% mitochondrial genes expressed were discarded; (3) The
cells that expressed 200–2,500 genes were remained. Then, cell-level normalization
was performed for gene expression by the NormalizeData function in the Seurat R package
with default parameters. Highly variable features (HVGs) were identified based on
mean gene expression and gene expression variance. Principle component analysis (PCA)
and uniform manifold approximation and projection for dimension reduction (UMAP) were
accomplished based on HVGs to reduce dimension. After that, clusters of cells were
identified by a shared nearest neighbor (SNN) modularity optimization based original
Louvain algorithm. Then, cell type clusters were annotated by cross-checking the specifically
expressed genes of each cell type, pre-existing cell annotations from published studies,
and known cell markers from the CellMarker databases (Zhang et al., 2019).
2.4. Statistical analysis
The clusterProfiler (Wu et al., 2021) R package was implemented to conduct the Kyoto
Encyclopedia of Genes and Genomes (KEGG) (Kanehisa and Goto, 2000) and Gene Ontology
(GO) (Gene Ontology Consortium, 2015) enrichment analysis. The GO enrichment analysis
includes three pathway types: cellular component (CC), biology pathway (BP), and molecular
function (MF). The Wilcoxon rank sum test was utilized to calculate differentially
expressed genes (DEGs) between AD and normal groups. The CellChat (Jin et al., 2021)
R package was utilized for cell communication analysis. The Monocle (Qiu et al., 2017)
R package was utilized for cell trajectory analysis. Detailed methods are shown on
the help page of the SCAD-Brain web site.
2.5. Web construction
R/Shiny was applied to construct the user interface and backend of SCAD-Brain. MongoDB
database was utilized to store SCAD-Brain data. R packages like DT (Xie et al., 2023),
ggplot2 (Wickham, 2016), dplyr, and tidyr were adopted to actualize the data process
and data visualization on the web server.
3. SCAD-Brain provides services for the reuse and interpretation of scRNA-seq datasets
from AD brain tissues
3.1. Summary of datasets collected by SCAD-Brain
The 17 scRNA-Seq projects we collected cover 21 datasets, two species (human and mouse),
359 samples, 10 brain regions (pooled cortices, hippocampal, prefrontal cortex, caudal
entorhinal cortex, superior frontal gyrus, superior frontal cortex, superior parietal
lobe, white matter, gray matter, dorsolateral prefrontal cortex), 16 major cell types
(oligodendrocyte, astrocyte, microglial, oligodendrocyte precursor cell, endothelial,
neuron, T and NK cell, monocyte, B cell, proliferating cell, pericyte, smooth muscle
cell, fibroblast, ependymal), and 1,564,825 cells (Figure 1A). Among the 21 datasets
(see method part: Contents of data projects), 12 are human studies and nine are mice
studies (5xFAD, J20, or APPswe/PSEN1L166P mouse model). Among the 359 samples, 227
are from the human brain and 132 are from mice brains. At present, the hippocampal,
dorsolateral prefrontal cortex, and prefrontal cortex are the brain regions that get
the most attention in AD research. Additionally, among all scRNA-Seq data in our study,
microglial, oligodendrocyte, neuron, and astrocyte cell types have the largest amount
of cells.
Figure 1
Work flow of SCAD-Brain. (A) Statistics of datasets that used in our study. (B) Functional
modules provided by the SCAD-Brain web site. (C) Representative outputs of the SCAD-Brain.
3.2. Functional analyses provided by SCAD-Brain
SCAD-Brain provides a flexible web-based service for the analysis and visualization
of collected scRNA-seq data sets from human and mice brains with AD pathology. SCAD-Brain
provides four functional analysis types (Figure 1B), including “Cell marker”, “Comparison
between cell types”, “DEGs between AD and normal”, “Cell communication”, and “Cell
trajectory”. The typical outputs are shown in Figure 1C. Detailed descriptions are
shown below.
3.2.1. “Cell markers” module
The “Cell markers” module conducts cell marker investigation and pathway enrichment
analysis, including four panels: (1) The “Cell annotation” panel provides cell type
annotation for all samples and cell type composition for each sample; (2) The “Cell
markers” panel summarizes marker genes for each cell type; (3) The “Search a gene”
panel presents the gene expression in each cell type; (4) The “KEGG”, “GO-BP”, “GO-CC”,
and “GO-MF” panels provide KEGG and GO pathway enrichment for cell markers.
3.2.2. “Comparison between cell types” module
The “Comparison between cell types” module calculates DEGs between any two cell types,
including three panels: (1) The “DEG” panel performs differential expression analysis
between two selected cell types; (2) The “Search a gene” panel visualizes the distribution
of gene expression in the two selected cell types; (3) The “KEGG” and “GO” panels
perform pathway enrichment analysis for DEGs using the Gene Set Enrichment Analysis
(GSEA) method, based on the KEGG and GO databases, respectively.
3.2.3. “DEGs between AD and normal” module
The “DEGs between AD and normal” module includes three submodules: “AD and control”,
“AD and MCI”, and “Braak stages”. Each submodule has three panels: (1) The “DEGs”
panel computes the DEGs between groups; (2) The “Search a gene” panel presents the
gene expression in each cell type between groups. (3) The “KEGG” and “GO” panels perform
pathway enrichment analysis of DEGs using the GSEA method, based on KEGG and GO databases,
respectively.
3.2.4. “Cell communication” module
The “Cell communication” module analyzes ligand-receptor (L-R) interactions among
cell types, including five panels: (1) The “Cell interaction” panel summarizes the
number and strength of interactions among cell types; (2) The “Functional and structural”
panel identifies structurally and functionally similar pathway signals; (3) The “Information
flow” panel analyzes the strength of outgoing and incoming signals of each cell type;
(4) The “L-R” panel summarizes the communication probabilities of ligand-receptor
pairs among cell types; (5) The “Expression of signaling pathway” panel analyzed the
expression of genes in enriched signaling pathway between AD and control.
3.2.5. “Cell trajectory” module
The “Cell trajectory” module performs cell differentiation and development trajectory
analysis, including two panels: (1) The “Trajectories” panel displays cell trajectories
by cell type, cell state, and pseudo-time; (2) The “Gene expression along pseudo-time”
panel calculates the gene expression variation according to pseudo-time.
3.3. Comparing with the existing AD scRNA-seq databases
As far as we know, there are three databases providing analysis and visualization
of scRNA-seq data from AD brain tissues, the TACA (Zhou et al., 2022), the SC2Disease
(Zhao et al., 2021), and the scREAD (Jiang et al., 2020). The TACA database was designed
to perform various differential expression comparisons to identify cell type-specific
gene expression alterations, cell-cell interactions, and drug screening. The SC2Disease
database aims to provide a comprehensive and accurate resource of gene expression
profiles in various cell types for 22 diseases. However, the data provided by SC2Disease
for AD only includes one data set from one brain region (prefrontal cortex). On the
other hand, scREAD, a database designed to perform scRNA-seq data analysis for AD
that covers multiple brain regions, is currently unavailable due to maintenance issues.
Compared to TACA, SC2Disease, and scREAD, SCAD-Brain covers 10 brain regions and 14
major cell types of AD and adds the following two unique features: (1) comparison
between two cell types; (2) cell trajectory analysis. Importantly, although some functions
provided by SCAD-Brain are overlapped with these databases, SCAD-Brain provides richer
visualization formats. All advancements of SCAD-Brain compared with TACA, SC2Disease,
and scREAD are summarized in Table 1.
Table 1
Comparisons among SCAD-Brain, SC2Disease, scREAD, and TACA.
SCAD-Brain
scREAD (lost of maintenance)
SC2Disease
TACA
Number of AD related datasets
21
17
1
26
Number of AD brain regions
10
8
1
11
Number of AD brain major cell types
16
7
5
21
Cell Annotation
√
√
√
√
Cell Markers
√
√
Pathway enrichment of cell markers
√
√
Comparison between two cell types
√
Single gene visualization
√
√
DEGs between AD and normal
√
√
√
√
DEGs between stages
√
√
Pathway enrichment of DEGs
√
√
Cell communication analysis
√
√
Trajectory analysis
√
4. Conclusion
We collected and analyzed scRNA-Seq datasets of 10 brain regions from human and mouse
models with AD pathology. Additionally, we introduced SCAD-Brain, a database to perform
cell marker analysis, GSEA, gene expression comparison between cell types, differential
expression analysis between AD and normal or between stages, and cell communication
analysis, based on the AD scRNA-seq data we collected. SCAD-Brain is a powerful platform
and resource to explore and visualize the cellular heterogeneity of AD brain at single-cell
resolution, assist in experimental design and hypothesis testing, and allow users
to unravel biological pathways, cell interactions, cell trajectories, and DEGs underlying
neurodegeneration in specific cell populations. We will energetically maintain SCAD-Brain
and update it with new data and methods.
Data availability statement
The original contributions presented in the study are included in the article/supplementary
material, further inquiries can be directed to the corresponding authors.
Author contributions
X-WL, T-TD, and J-YC: study design, data collection, web construction, data interpretation
and writing for review and editing, and conceptualization. S-YP: data collection and
web construction. YZ: study design, writing for review and editing, and conceptualization.
F-FH: study design, funding, writing for review and editing, and conceptualization.
All authors agree to be accountable for the content of the work and worked collectively
to develop the protocols and methods described in this article.