Abstract

   Alzheimer’s disease (AD) is the most common cause of dementia,
   characterized by memory loss, cognitive decline, personality changes,
   and various neurological symptoms. The role of blood–brain barrier
   (BBB) injury, extracellular matrix (ECM) abnormalities, and
   oligodendrocytes (ODCs) dysfunction in AD has gained increasing
   attention, yet the detailed pathogenesis remains elusive. This study
   integrates single-cell sequencing of AD patients’ cerebrovascular
   system with a genome-wide association analysis. It aims to elucidate
   the associations and potential mechanisms behind pericytes injury, ECM
   disorder, and ODCs dysfunction in AD pathogenesis. Finally, we
   identified that abnormalities in the pericyte PI3K–AKT–FOXO signaling
   pathway may be involved in the pathogenic process of AD. This
   comprehensive approach sheds new light on the complex etiology of AD
   and opens avenues for advanced research into its pathogenesis and
   therapeutic strategies.

   Keywords: Alzheimer’s disease, ScRNA-seq, GWAS, Blood–brain barrier,
   PI3K–AKT–FOXO

   Subject terms: Alzheimer's disease, Neurochemistry

Introduction

   Alzheimer’s disease (AD), the leading cause of dementia, is
   characterized by symptoms such as memory loss, cognitive decline,
   personality changes, along with other neurological manifestations
   including depression, restlessness, anxiety, and aggression^[50]1.
   Various hypotheses have been proposed to explain the pathogenesis of
   AD, including the amyloid-beta (Aβ) hypothesis^[51]2, oxidative stress
   hypothesis^[52]3, and abnormal phosphorylation of tau proteins^[53]4.
   However, the exact pathogenic mechanisms underlying AD remain largely
   undefined.

   Recent studies have increasingly focused on the roles of blood–brain
   barrier (BBB) injury^[54]5–[55]9, extracellular matrix (ECM)
   abnormalities^[56]10,[57]11, and oligodendrocyte (ODC)
   dysfunction^[58]12 in the development of AD. The BBB, comprising
   pericytes, endothelial cells (ECs), astrocytes (ASCs), and the basal
   membrane, forms a critical component of the neurovascular unit
   (NVU)^[59]13,[60]14. Emerging evidence suggests that BBB damage, which
   can occur early in both AD patients and animal models, might contribute
   to AD pathogenesis independently of Aβ and tau
   accumulation^[61]6,[62]15,[63]16. This to some extent explains the
   pericyte loss observed in the brains of AD patients^[64]17,[65]18. The
   extent of BBB impairment is closely linked to cognitive decline,
   highlighting the significance of pericytes in BBB integrity in
   AD^[66]19–[67]21. Multiple explanations exist for the cause of BBB
   damage, including pericytes and ECs cell degeneration, immune cell
   infiltration, and abnormal angiogenesis^[68]5. In recent years, the
   role of pericytes in AD has been extensively studied. Pericytes
   integrate the functions of ECs and ASCs to regulate BBB
   permeability^[69]13,[70]22. Additionally, pericytes can directly clear
   Aβ, so their loss may accelerate the progression of Aβ and tau
   pathology in AD mouse models^[71]23. This may be related to factors
   such as tight junctions^[72]24,[73]25 and transmembrane transport
   mediated by phagocytosis^[74]26. The ECM, a complex network of
   macromolecules including collagen, elastin, fibronectin, laminin, and
   various glycoproteins, is vital for the structural and functional
   integrity of the central nervous system^[75]27. Within the brain, the
   ECM not only supports cellular structures but also plays a pivotal role
   in neuronal health and signaling^[76]28. It should be noted that
   previous studies have identified a subtype of pericytes related to the
   ECM, which are significantly reduced in the brain tissue of AD patients
   and may be associated with BBB damage in AD^[77]29. ODCs are essential
   for myelin sheath formation and overall neuronal support. Recent
   research has underscored the involvement of ODCs in neurodegenerative
   diseases, where their dysfunction can lead to demyelination and
   neuronal damage^[78]30,[79]31. The discovery of disease-associated ODCs
   (DOLs), whose emergence is not directly linked to Aβ and tau
   pathologies^[80]32–[81]35, has opened new avenues for exploring
   mechanisms of AD. Studies have found that modulating the PI3K/AKT
   signaling pathway may alter the production and clearance of amyloid
   beta and tau in AD patients, thereby slowing down the disease
   progression^[82]36. It should be noted that, due to the relatively
   minor proportion of the cerebral microvascular system within the brain
   parenchyma and the complexity inherent in research design, there have
   been no studies to date that conclusively illuminate the
   interconnections between pericyte injury, ECM disruption, and ODCs
   dysfunction in AD.

   This study evaluated the expression of AD-related GWAS risk genes in
   the brain vascular system and conducted correlation analysis to infer
   AD-related cell subgroups. Based on this, a deeper exploration of the
   mechanisms was carried out, speculating on the potential process from
   BBB damage to ECM imbalance, and then to ODCs’ functional abnormalities
   in AD. Additionally, the PI3K–AKT–FOXO pathway was identified as a
   potential mechanism leading to pericytes dysfunction.

Methods

   This study used data from publicly available GEO and GWAS catalog
   databases. These data sources had obtained informed consent from the
   research participants beforehand, and according to the statements of
   the database providers, the collection and use of these data complied
   with all applicable ethical standards. Therefore, this study did not
   require additional ethical approval.

ScRNA-seq analysis process

   We obtained the scRNA-seq data ([83]GSE163577) (including the cortex
   and hippocampus) of 9 AD patients and 8 matched normal controls from
   the GEO public database. In this study, the research subjects were
   divided into two groups: the AD and the normal group. Scanpy (version
   1.8.2)^[84]37 was used to remove low-quality cells with more than 5%
   mitochondrial genes, less than 200 genes, or more than 5000 genes.
   BBKNN (version 1.5.1)^[85]38 was used for batch effect removal in each
   sample. After standardization and normalization of the combined data,
   scanpy was used to determine the first 2000 highly variable genes in
   the whole data set, after which dimension reduction and clustering were
   successively carried out. The resolution was set to 0.3, and the
   principal component was set to 15. SCENIC (version 0.11.2) analysis was
   used for transcription factor prediction^[86]39. The cell types were
   annotated according to the expression of marker genes: ODC (MBP), ASCs
   (GFAP, AQP4), pe (PDGFRB), ECs (CLDN5), MiCs (PTPRC), OPCs (DSCAM),
   GlNs (SYT1)^[87]29. The marker genes in each cell type were calculated
   using the Wilcoxon test in Scanpy, and then the top 500 with the most
   significant p-values were selected as the cell-type-specific gene set.
   The list of marker genes is available in Supplement Table [88]1.
   Differential genes were calculated using the find markers command in
   Scanpy, where log2FC > 0.25 and p < 0.05 were considered significant.

MetaXcan identifies AD risk and protective genes

   We obtained the GWAS summary data (GCST007511) from the GWAS Catalog
   ([89]https://www.ebi.ac.uk/gwas/summary-statistics). This research
   included 21,982 AD patients and 41,944 controls from Europe. Detailed
   information can be accessed on GEO Accession viewer (GEO Accession
   viewer (nih.gov). The risk and protective genes of AD were predicted by
   MetaXcan (version 0.6.2)^[90]40. MetaXcan utilizes gene expression
   prediction models based on tissue specificity, combined with summary
   statistics from GWAS. It calculates the association strength between
   gene expression levels and phenotypes, effectively integrating
   information from multiple tissues and phenotypes to discover potential
   causal candidate genes. In this study, genes with a z-score greater
   than 2.5 were considered risk genes for AD, while genes with a z-score
   less than − 2.5 were considered protective genes for AD.

ScDRS identifies disease-related cell types

   The comprehensive scores of preset gene sets in single cells were
   calculated by single-cell disease relevance score (scDRS) (version
   1.0.2)^[91]41. ScDRS is a method that associates individual cells from
   scRNA-seq data with disease GWAS. It can identify cells with
   overexpressed disease-associated genes, thereby discovering
   disease-relevant cell subpopulations and mechanisms. This comprises
   three steps: (1) scDRS generates disease-related gene sets; (2) scDRS
   calculates the original disease score for each cell; (3) after gene set
   and cell normalization, scDRS calculates each cell’s association
   p-value by comparing the cell’s normalized disease-related score with
   the empirical distribution of the normalized control score of all gene
   sets and all cells.

Pathway analysis and construction gene module

   Enrichr (version 3.0) was used to perform enrichment analysis of
   differential genes to identify gene sets associated with biological
   processes and pathways. Enrichr is an online tool that can perform
   enrichment analysis of gene lists using various reference gene set
   databases^[92]42. Details of the differential gene can be found in
   Supplement Tables [93]3–[94]5. Gene set variation analysis (GSVA,
   version 1.48.3) is a method to calculate enrichment scores for a given
   gene set, to assess the activity level of that gene set in selected
   samples^[95]43. Then differential analysis is performed by Wilcox test.
   We used the default parameter settings for GSVA. The STRING database
   (version 12.0)^[96]44 was used to extract the protein–protein
   interaction network data of the differential genes. The network data
   from STRING were then analyzed using the iGraph package (version
   1.5.1)^[97]45 in R (version 4.2.0). A network analysis of the extracted
   network data employed the random walk method to highlight subnetworks
   or neighborhoods. We set the walk length to 3, the walk times to 1000,
   and the neighborhood size to 10.

Results

ScRNA-seq and GWAS summary data preprocessing

   We acquired scRNA-seq data from four cortical and nine hippocampal
   samples collected from nine AD patients and four cortical and eight
   hippocampal samples from eight normal controls. After batch removal,
   double cell removal, low-quality cell filtering, dimension reduction,
   and clustering, 264,987 single cells were obtained. Display cell
   distribution using UMAP plots (Fig. [98]1A). The purpose of UMAP plots
   is to group similar cells together and separate different cells, so as
   to display cell clustering and differentiation on two-dimensional or
   three-dimensional graphs^[99]46. Cell types were annotated on the basis
   of marker gene expression (Fig. [100]1C), and seven cell types were
   obtained: ASCs (15.5%), ECs (32.0%), glutaminergic neurons (GlNs)
   (22.6%), microglia cells (MiCs) (4.2%), ODCs (24.1%), oligodendrocyte
   progenitor cells (OPCs) (1.7%), and pericytes (20.1%) (Fig. [101]1B). A
   volcano map revealed the differentially expressed genes in each cell
   type, and the results indicate that the top five upregulated and
   downregulated genes were significantly different in each cell type
   (Fig. [102]1D). These results suggest that different cell types may
   exhibit distinct patterns of differential gene expression and play
   different roles in the onset of AD.

Fig. 1.

   [103]Fig. 1
   [104]Open in a new tab

   (A) UMAP displaying the integration results of multiple scRNA samples
   with good consistency; (B) UMAP showing cell annotation results,
   divided into seven types: astrocytes (ASCs), endothelial cells (ECs),
   glutamatergic neurons (GlNs), microglia cells (MiCs), oligodendrocytes
   (ODCs), oligodendrocyte precursor cells (OPCs), and pericytes; (C) The
   cell types express marker genes: ODCs (MBP), ASCs (GFAP, AQP4),
   pericytes (PDGFRB), ECs (CLDN5), MiCs (PTPRC), OPCs (DSCAM), and GLNs
   (SYT1); Expression of marker genes across all cell groups; (D) Volcano
   plot showing the differentially expressed genes for each cell group,
   highlighting the top five upregulated and downregulated genes.

   The GWAS summary data of AD incidence included 21,982 cases and 41,944
   controls. MetaXcan analysis predicted the risk and protective genes
   associated with AD, and yielded 125 risk genes and 120 protective
   genes. See Supplement Table [105]6 for detailed results. The genes were
   organized into separate gene sets to characterize their risk and
   protective effects on AD. These sets included known common AD risk
   genes such as ACE, CLU, CR1, KAT8, HLA-DQB1, HLA-DRB1, LACTB, STX6,
   PLEKHA1, MTCH2, C1QTNF4, MADD, and CD2AP^[106]47–[107]52.

Identifying disease-associated cell types

   ScDRS can comprehensively score a given gene set at the single-cell
   level to determine the overall expression of that gene set in a single
   cell. To explore the scores of AD risk and protective gene sets in the
   scRNA-seq data from AD patients, we used scDRS for evaluation and
   showed the outcome in UMAP, which indicated that the AD risk gene set
   had a high score in ODCs (Fig. [108]2A). The AD protective gene set was
   high mainly in ASCs, pericytes, and ODCs (Fig. [109]2B).

Fig. 2.

   [110]Fig. 2
   [111]Open in a new tab

   ScDRS identifies disease-related cell types. (A) The AD risk gene set
   scored the highest in oligodendrocytes (ODCs); (B) The AD protective
   gene set scored higher in pericytes, astrocytes (ASCs), and ODCs; (C)
   The marker gene sets for ODCs, pericytes, and ASCs all exhibited good
   specificity; (D) Correlation analysis revealed that ODCs were
   positively correlated with the risk of AD; (E) ODCs were negatively
   correlated with the protective effect against AD, whereas ASCs and
   pericytes showed a positive correlation with the protective effect
   against AD. The calculation of differential genes was based on the
   Wilcoxon test method. The correlation analysis employed a Pearson
   correlation coefficient-based method to evaluate the associations
   between specific subgroups and various traits. This was then followed
   by calculating the Pearson correlation coefficient using feature scores
   and spatial scores, and finally, the p-value was computed by comparing
   this value with the Pearson correlation coefficient of the control
   scores. The color scale represents the scores of gene sets in the
   cells, transitioning from blue to red to indicate scores from low to
   high. (p < 0.05 was considered statistically significant.
   ***p < 0.001).

   The marker gene sets in ODCs, ASCs, and pericytes had good specificity
   (Fig. [112]2C–E). Next, the scores of the marker gene set and AD risk
   and protective gene sets in cells were calculated separately prior to
   correlation analysis. The results show that ODCs were significantly
   positively correlated with AD risk (Fig. [113]2F), whereas there was a
   significant positive correlation between pericytes/ASCs and the
   protective effect on AD (Fig. [114]2G). Although the AD protective gene
   set was highly expressed in ODCs, its functional direction was opposite
   to that of pericytes/ASCs (Fig. [115]2G). Correlation analysis did not
   find any significant association with the aforementioned gene sets in
   other cell types. Further statistical analysis of the proportion of
   cells showed that ODCs increased significantly in the AD (AD group:
   0.36 ± 0.12; normal group: 0.14 ± 0.01, Fig. [116]3A), whereas
   pericytes decreased significantly (AD group: 0.25 ± 0.10; normal group:
   0.13 ± 0.07, Fig. [117]3B). There was no statistical difference in the
   proportion of ASCs between the AD and normal groups (AD group:
   0.19 ± 0.07; normal group: 0.14 ± 0.08, Fig. [118]3C). Detailed
   results, such as mean and standard deviation, refer to Supplement Table
   [119]7. In conclusion, by evaluating the expression of AD-related GWAS
   risk genes in cells, we have once again confirmed a significant
   association between ODCs, pericytes, ASCs, and the onset of AD from the
   perspective of risk genes.

Fig. 3.

   [120]Fig. 3
   [121]Open in a new tab

   Cell proportion analysis and ECM pathway enrichment analysis. (A)
   Oligodendrocytes (ODCs) increased significantly in AD; (B,D) Pericytes
   and endothelial cells (ECs) significantly decreased in AD; (C) There
   was no statistical difference in astrocytes (ASCs) between the AD and
   normal groups. Each point in the scatter plot represents an individual
   sample. Statistical significance was determined using the t-test. The
   annotations in the figure represent the mean value plus or minus one
   standard deviation, indicating the range of variation around each data
   point, and are labeled below each graph. (E) Gene Set Variation
   Analysis (GSVA) showed that ECM pathway activity in AD was
   significantly downregulated in all cell types. Wilcoxon test. The upper
   and lower boundaries of the boxplot represent the upper quartile (Q3)
   and lower quartile (Q1) of the data, respectively. The horizontal line
   inside the box represents the median (Q2). (***p < 0.001).

   To evaluate whether the above findings also exist in the hippocampus
   and cortex, we performed the same analysis on the hippocampus and
   cortex separately. The results showed that ODCs in the cortex had no
   significant correlation with the protective effect, while pericytes and
   ASCs in the hippocampus showed stronger correlation with the protective
   gene set (eFigure [122]1 in supplementary). Cell proportion analysis
   found that, unlike the overall analysis, the proportion of ASCs in the
   hippocampus of AD group increased significantly, suggesting that the
   proliferation of ASCs in the hippocampal of AD patients was more
   prominent (eFigure [123]2 in supplementary). In summary, the results of
   this study suggest that both the hippocampus and cortex are related to
   the pathogenesis of AD, but the hippocampus may play a more important
   role.

ECM dysregulation is involved in the pathogenesis of AD

   To explore the specific pathogenesis of AD, we performed pathway
   enrichment analysis of total differentially expressed genes between AD
   and normal groups. The results show that the downregulated
   differentially expressed genes of AD were mainly enriched in the ECM,
   homeostasis, integrin cell surface interaction, and immune-related
   signaling pathways (Fig. [124]4A) that are involved in organizing
   extracellular structures and regulating cell proliferation and
   differentiation as well as cytokine response, vesicle transport, BBB
   transport, and other biological processes (Fig. [125]4B). The
   upregulated differentially expressed genes of AD are involved mainly in
   nervous system development and neuron projection morphogenesis (eFigure
   [126]3A,B in supplementary). The GSVA comparison analysis revealed that
   the activity of the ECM pathway in the AD was significantly
   downregulated along with the activities of adhesion, endocytosis,
   antigen processing and presentation, and cytokine–cytokine receptor
   interaction, which are involved mainly in information and substance
   exchange and other biological processes (Fig. [127]4C). We wondered
   whether an abnormal ECM was involved in the pathogenesis of AD and used
   GSVA to further explore the activation of the ECM in each cell types.
   The results show that the activity of the ECM pathway was significantly
   downregulated in all cell types (Fig. [128]3E), consistent with the
   expected results. Additionally, we conducted an analysis of ECM pathway
   alterations in the hippocampus and cortex separately. The results
   indicate that nearly all cell types in both the hippocampus and cortex
   of the AD group exhibit an imbalance in the ECM, with the hippocampus
   showing more pronounced effects (eFigure [129]2I,J in supplementary).
   In conclusion, an abnormal ECM pathway may be a cause of AD
   pathogenesis.

Fig. 4.

   [130]Fig. 4
   [131]Open in a new tab

   Pathway analysis of downregulated differential genes in AD. (A) Pathway
   enrichment analysis showed an abnormal ECM pathway in AD; (B)
   Biological process analysis indicated abnormalities in the ECM,
   vascular transport, and angiogenesis in AD. (C) GSVA analysis showed
   that the ECM was significantly downregulated in AD. (p < 0.05 was
   considered statistically significant.)

ODC hyperplasia and dysfunction caused by an abnormal ECM

   To explore the role of ODCs in the pathogenesis of AD, pathway
   enrichment analysis was conducted on ODCs’ differentially expressed
   genes, and the results show that the downregulated differentially
   expressed genes of AD were enriched mainly in the ECM, adhesion, tight
   junction, MAPK, and other pathways (Fig. [132]5A). It is involved in
   biological processes such as extracellular structure organization,
   response to cytokine, vesicle transport, and the regulation of ECs
   migration (Fig. [133]5B). The upregulated differentially expressed
   genes of AD are involved mainly in nervous system development,
   axonogenesis, and the regulation of neuron death (eFigure [134]3C,D in
   supplementary). GSVA was further used to evaluate differences in the
   activity of pathways in ODCs between the AD and normal groups, and the
   results show that the activity of the ECM pathway in the AD was
   significantly downregulated as was the activity of pathways associated
   with adhesion, cytokine–cytokine receptor interaction, and the MAPK
   pathway. To better display the differential results, we use t-value to
   draw box plots to show the enrichment level of pathways (Fig. [135]5C).
   Overall, we speculate that ODC dysfunction caused by an abnormal ECM
   pathway may be a pathogenic mechanism of AD.

Fig. 5.

   [136]Fig. 5
   [137]Open in a new tab

   Differential gene pathway enrichment analysis. (A/B) Pathway and
   biological process enrichment analysis revealed that downregulated
   genes in ODCs were enriched in pathways such as the ECM, vascular
   transport, response to unfold protein, ferroptosis, and angiogenesis;
   (C) GSVA showed that ECM pathway activity was significantly
   downregulated in AD. Bar chart display the distribution of t-values
   across samples for different gene sets, used to assess their enrichment
   in various samples. Higher t-values indicate more significant
   enrichment. (p < 0.05 was considered statistically significant.)

Pericyte disorders and BBB injury lead to ECM dysregulation in AD

   To explore the possible causes of ECM imbalance in AD, we performed
   pathway enrichment analysis of pericytes’ differentially expressed
   genes, and the results show that the downregulated differentially
   expressed genes of AD were enriched mainly in the PI3K–AKT, VEGF, and
   adhesion pathways (Fig. [138]6A) and were involved in biological
   processes such as BBB transport, vesicle transport, ECM organization,
   and cytokine response (Fig. [139]6B). The upregulated differentially
   expressed genes of AD are involved mainly in myelin development, axon
   protein localization, ODC differentiation, and other biological
   processes (supplement Fig. [140]1E/F). Further GSVA comparative
   analysis showed that the activity of the PI3K–AKT pathway was
   significantly downregulated in the AD along with the activities of the
   FOXO, cell cycle, VEGF, cytokine–cytokine receptor interaction, and
   MAPK pathways (Fig. [141]6C). Modules were constructed for the
   differentially expressed genes in downregulated pathways in GSVA, and
   the PI3K–AKT pathway was found to be in the central position
   (Fig. [142]7A). It should be noted that abnormalities in the PI3K–AKT
   pathway are not evident in other cell types (not shown). In conclusion,
   we speculate that an abnormal PI3K–AKT pathway may be a pathogenic
   pathway for pericyte dysfunction in AD.

Fig. 6.

   [143]Fig. 6
   [144]Open in a new tab

   Differential gene pathway analysis. (A) Pathway enrichment analysis
   showed that pericytes’ downregulated genes in AD were enriched in the
   PI3K–AKT pathway; (B) Enrichment analysis for downregulated genes
   primarily identifies significant enrichment in pathways associated with
   vascular transport, transport across the blood–brain barrier (BBB), and
   the extracellular matrix (ECM). (C) GSVA showed that PI3K–AKT pathway
   activity was significantly downregulated in pericytes of AD.

Fig. 7.

   [145]Fig. 7
   [146]Open in a new tab

   (A) The PI3K–AKT pathway was central in the construction of the
   pericytes differential gene module; (B) Key genes related to the
   PI3K–AKT pathway were significantly downregulated in the pericytes of
   AD patients; (C) SCENIC transcription factor prediction analysis showed
   that the activity of FOXO1/4/6 was significantly upregulated in the
   pericytes of AD; (D) Schematic diagram of the PI3K–AKT–FOXO pathway;
   (−p: dephosphorylation process; +p: phosphorylation process) (E) GSVA
   analysis of hippocampal and cortical pericytes revealed differences in
   pathway activity between the AD and normal groups; (F) GSVA analysis
   indicated that hippocampal immune-related pathway activity in AD was
   significantly higher than in the cortex, whereas the activity levels of
   the PI3K–AKT, FOXO, and cell cycle pathways were significantly lower
   than those in the cortex. (p < 0.05 was considered statistically
   significant.)

   The PI3K–AKT pathway is involved in cell development, metabolism,
   proliferation, survival, and angiogenesis, and phosphorylated AKT
   (p-AKT) plays a key role in regulating the activity of the pathway.
   PI3K activates the PI3K–AKT pathway by binding to the
   phosphatidylinositol-dependent kinase (PDK) family to induce AKT to the
   cell membrane for phosphorylation^[147]53,[148]54. In further exploring
   the specific mechanism of the abnormal PI3K–AKT pathway in the
   pericytes of AD, we found that the expression of PDK1/3/4 in the
   pericytes of the AD was significantly downregulated (Fig. [149]7B). The
   expression levels of genes inhibiting AKT dephosphorylation, such as
   PPP2R1B and PPP2R2D, were also significantly downregulated
   (Fig. [150]7B), which decreased the p-AKT level and inhibited the
   PI3K–AKT pathway. The FOXO family is the downstream regulator of AKT. A
   SCENIC analysis found that pericytes FOXO1/4/6 transcription factor
   activity was significantly upregulated in the AD (Fig. [151]7C). The
   schematic diagram of the mechanism is shown in Fig. [152]7D. These
   results indicate that an abnormal PI3K–AKT–FOXO pathway of pericyte may
   be a pathogenic cause of AD.

Hippocampal pericyte damage and immune infiltration are more significant in
AD

   Given that hippocampal atrophy is a typical pathological feature of
   AD^[153]55, we wondered whether we would find different pathological
   mechanisms between the hippocampal and cortical pericyte samples in AD.
   Therefore, we conducted a comparative GSVA analysis of the hippocampal
   and cortical pericytes of the AD and normal groups. We found that the
   activities of the cytokine–cytokine receptor interaction, antigen
   processing and presentation, ECM, and other pathways in the hippocampus
   and cortex were significantly downregulated. Most of the key pathways,
   such as the PI3K–AKT, FOXO, endocytosis, and spliceosome pathway, were
   more significantly downregulated in the hippocampus (Fig. [154]7E). We
   further compared differences in the activation levels of pericyte
   pathways in the hippocampal and cortical regions of AD patients by GSVA
   and found that the activities of the PI3K–AKT pathway, endocytosis
   pathway, tight junction pathway, and adhesion pathway were
   significantly decreased in the pericytes of AD patients, whereas
   immune-related signaling pathways, such as the antigen processing and
   presentation, and autoimmune were significantly activated
   (Fig. [155]7F). These findings suggest that AD patients have more
   significant pericyte damage and immune levels in the hippocampus.

Abnormal ECs are involved in the disruption of BBB function in AD

   Although scDRS did not find a significant association between ECs and
   AD, given their important role in the BBB and their significant
   reduction in AD (Fig. [156]3D and eFigure [157]2D,H in supplementary),
   it suggests that ECs may be involved in BBB damage in AD. The cell
   proportion analysis results are as follows: AD group: 0.21 ± 0.13;
   normal group: 0.41 ± 0.1. Interestingly, although the GSVA results
   showed a significant downregulation of pathways such as tight junction,
   focal adhersion, adherens junction, and endocytosis in the hippocampal
   region of the AD group, they showed a significant activation of
   pathways such as tight junction, adherens junction, and endocytosis
   (only including ECs) in the cortical (eFigure [158]4 in supplementary),
   which are essential for the BBB barrier and transport functions of
   pericytes and endothelial cells^[159]56,[160]57. In summary, these
   results suggest that the BBB in the hippocampal of AD may suffer more
   severe damage than in the cortex.

Discussion

   Sn-RNA approaches help reveal molecular disturbances in AD, but most of
   these studies focus on abundant brain cell types^[161]58–[162]60.
   Previous research has already identified significant associations
   between gene expression patterns of different cell types and AD risk
   genes, including ECs, pericytes, ASCs, and
   MiCs^[163]51,[164]61–[165]63. Luquez et al. and Murdock et al. have
   provided detailed summaries of gene expression patterns and potential
   pathogenic processes in different cell types in AD^[166]64,[167]65.
   Similarly, our study utilized scDRS to assess the expression of AD risk
   genes in the brain vascular single-cell atlas and conducted correlation
   analysis, identifying cell types significantly associated with AD risk
   genes, including ODCs, pericytes, and ASCs. In contrast to the
   expression-weighted cell type enrichment (EWCE) approach commonly used
   in previous studies, which analyzes associations between specific risk
   genes and cell types, scDRS implements a comprehensive weighting
   strategy for all genes in individual cells, providing a better
   assessment of the cumulative risk level of each cell in multi-gene
   diseases. First, we found that the PI3K–AKT–FOXO pathway is a potential
   mechanism for pericyte damage^[168]66, which can lead to BBB disruption
   and further result in ECM imbalance^[169]10,[170]11 and immune
   infiltration^[171]67,[172]68. Our results partially explain how the
   PI3K/AKT signaling pathway may alter the production and clearance of
   amyloid beta and tau in AD patients, thereby slowing down disease
   progression^[173]36. Additionally, our results show that the
   downregulation of the pericyte FOXO pathway, specifically the
   downregulation of p-AKT, activated the transcriptional regulatory
   activity of the downstream FOXO family and inhibited FOXO. The FOXO
   pathway is involved in regulating cell proliferation, and its
   activation can promote cell apoptosis and inhibit the cell
   cycle^[174]69. Finally, we observed that pericytes in the hippocampal
   region of AD patients exhibit more severe functional impairment and
   immune infiltration compared to those in the cortical region. This
   supports the notion that AD may be an autoimmune
   disease^[175]67,[176]68.

   Although previous studies have revealed the selective susceptibility of
   pericytes in AD, which is also confirmed by our results, their
   explanations often focus on describing risk genes and lack deeper
   insights. However, we found that an abnormal PI3K–AKT–FOXO pathway may
   be the underlying cause of pericytes injury in AD. Additionally,
   several explanations are offered for the loss of pericytes in AD
   patients, such as PDGFRβ^[177]70and PI3Kβ^[178]71 signal loss as well
   as APOE-mediated activation of cyclophilin A (CypA) and matrix
   metalloproteinase-9 (MMP9)^[179]7. Pericytes also play a role in
   regulating blood flow. The dysfunction and decrease of pericytes reduce
   the regulating effect of pericytes on cerebral blood flow, offering a
   possible explanation for the decreased cerebral blood flow in AD
   patients^[180]72.

   Our results show that there is a severe imbalance in the ECM and
   activation of immune responses in the brains of patients with AD. As an
   important part of the extracellular environment, the ECM is involved in
   the morphological and structural support of cells and the survival,
   differentiation, development, and migration of nerve cells. It has been
   shown to be highly related to the immune system, and its disorder is
   the direct or indirect cause of most chronic diseases^[181]73–[182]75.
   AD risk genes are all important components of the ECM, such as ADAMTS1,
   ADAMTS4, FERMT2, and AGRN^[183]61. Additionally, it has been proved
   that disorders of HA, heparin sulfate proteoglycan (HSPG), chondroitin
   sulfate proteoglycan (CSPG), and other ECMs are closely related to the
   onset of AD and are associated with damage to the
   BBB^[184]74,[185]76–[186]78. Although previous studies identified a
   subtype of pericytes related to the ECM and associated with BBB damage,
   we found that ECM imbalance occurs in all cell types, not specifically
   in pericytes. We believe this is likely a consequence rather than the
   cause of BBB damage^[187]29.

   We found that AD risk genes are highly expressed in ODCs, suggesting
   that ODCs are involved in the pathogenesis of AD. We observed ECM
   imbalance in ODCs in AD patients, accompanied by downregulation of
   pathways involved in vesicle transport and ECs migration regulation,
   and upregulation of pathways regulating neuronal death. ODCs, as
   myelin-forming cells, are protective of neurons and can be sensitive to
   and respond to changes in the external environment. Recent studies have
   identified a class of DOLs in postmortem examinations of AD patients
   and animal models whose cell biosynthesis and lipid processing are
   enhanced and whose numbers increase significantly with disease
   progression^[188]33,[189]59. The proliferation of DOLs can be induced
   in abnormal microenvironments, such as those affected by immune
   inflammation, cytokines, and damage-related molecular patterns^[190]35,
   which further demonstrates the importance of the ECM’s homeostasis of
   ODCs to maintain its normal function and to some extent explains the
   proliferation of ODCs. Given the substantial evidence suggesting that
   BBB damage may be the initiating factor in
   AD^[191]5,[192]6,[193]15,[194]16,[195]18,[196]23,[197]79, we speculate
   that the ECM mediates BBB damage leading to the dysfunction of ODCs.

   However, it’s important to address its limitations for a comprehensive
   understanding. Although sc-RNA studies have provided unprecedented
   resolution for investigating cellular vulnerability in AD, significant
   questions remain about how neurons become dysregulated and how
   interactions between various cell types shape their vulnerability or
   resistance in the disease. Moreover, the selective vulnerability of
   cell types may vary across different studies, making it necessary to
   perform comprehensive reanalysis of multiple datasets. While these
   tools are powerful for hypothesis generation and preliminary analysis,
   more direct evidence may be needed. This approach, while useful for
   understanding broader mechanistic themes, may overlook critical nuances
   at the molecular or cellular level that could be pivotal in AD
   pathology. The complexity of neurodegenerative diseases like AD often
   requires a multi-dimensional analysis that goes beyond pathway-level
   insights to include gene expression profiles, protein interactions, and
   possibly even epigenetic factors. To address these limitations,
   subsequent research should aim to incorporate experimental validation
   of key findings, particularly those related to protein molecules
   implicated in AD. This could involve in vitro studies using cell lines
   or patient-derived cells, as well as in vivo studies in animal models.
   Such experimental approaches would provide a more solid foundation for
   the bioinformatic predictions and help translate these findings into
   practical clinical applications.

Conclusion

   This study revealed the association and potential pathogenesis of BBB
   injury, ECM disorder, and ODCs dysfunction in the pathogenesis of AD,
   providing an important direction for further exploration of the
   pathogenesis and treatment of AD.

Supplementary Information

   [198]Supplementary Tables. ^(8.5MB, xlsx)
   [199]Supplementary Figures.^ (1.2MB, pdf)

Acknowledgements