Gliovascular transcriptional perturbations in Alzheimer's disease reveal molecular mechanisms of blood brain barrier dysfunction

Blood-brain barrier (BBB) impairment is a hallmark of Alzheimer’s disease (AD) and is thought to permit entry of neurotoxic blood-borne factors, trigger inflammation, and reduce cerebral blood flow. Amyloid-β accumulation around cerebral vessels is both a cause and consequence of BBB dysfunction, which occurs early, predicts cognitive decline, and contributes to AD pathogenesis. Despite advances in single-cell and single-nucleus RNA sequencing (sc/snRNAseq) that have revealed AD-associated cell states, most studies have emphasized neurons and abundant glia. The transcriptional changes within the gliovascular unit (GVU)—especially vascular cells (endothelia, pericytes) and astrocytes—and their molecular interactions remain incompletely defined at a systems level. This study addresses that gap by profiling GVU cell types in postmortem AD and control brains, prioritizing ligand–target pairs linking astrocytes and vascular cells, and validating the top candidates functionally and across species to provide mechanistic insight into BBB dysfunction in AD.

Prior sc/snRNAseq studies have characterized cell-type–specific perturbations in AD brains, identifying vulnerable neuronal and glial populations and disease-associated microglia. However, fewer studies have focused on cerebrovascular cells due to their low abundance. Recent snRNAseq analyses have begun to delineate human cerebrovasculature and reported AD-related differentially expressed genes (DEGs) in endothelial cells and pericytes, including studies with vascular enrichment and multiregion sampling. Nevertheless, a systematic interrogation of GVU transcriptional perturbations, prioritization of astrocyte–vascular molecular interactions, and experimental validations linking these interactions to BBB integrity have been lacking. This work builds on that literature by integrating discovery, prioritization via NicheNet, orthogonal validations (qPCR, RNAscope, IHC/IF), external dataset replication, and functional tests in human iPSC-derived pericytes and zebrafish models.

- Human postmortem cohort: Temporal cortex (superior temporal gyrus) tissue from 12 neuropathologic AD and 12 age/sex-matched controls was processed for single-nucleus RNA sequencing (snRNAseq). RNA integrity (RIN>5.5) was required. Nuclei isolation was optimized (homogenization, detergent, debris removal, FANS with anti-HNA), with QC for nuclear purity (RNA/protein markers) and integrity. - snRNAseq: 10x Genomics Chromium 3' v3 libraries were prepared and sequenced (HiSeq4000). Cell Ranger (v3.1.0) aligned reads to GRCh38 including intronic regions. Filtering removed nuclei with high mitochondrial UMIs (>10%), low/high gene/UMI counts, and doublets (Scrublet). After QC, 78,396 nuclei remained. - Clustering and annotation: Seurat v3 integration (FindIntegrationAnchors/IntegrateData), 31 PCs (95% variance), clustering (FindNeighbors/FindClusters), UMAP visualization. Cell types assigned using conserved marker genes, published marker sets, and known markers (e.g., PDGFRB for pericytes, CLDN5/FLT1 for endothelia, AQP4/GFAP for astrocytes). Eight major cell types identified; GVU focus on three vascular clusters (pericytes cl.25, endothelia cl.26, perivascular fibroblasts cl.30) and three astrocytic clusters (cl.8, cl.11, cl.31). Signature genes (≥50% cells, log2FC≥1, Bonferroni p<0.05) and GO enrichment (MSigDB v7.0) were computed. - Differential expression: MAST hurdle models per cluster (genes detected in ≥10% cells; DEGs retained with q<0.05 and |logFC|>0.1, present in ≥20% AD or control cells). GO enrichment performed where sufficient DEGs existed. - Ligand–target prioritization: NicheNet used astrocyte DEGs as ligands and vascular DEGs as targets. Ligands had to have expressed receptors in the target vascular cluster; targets were top 250 regulated genes per prior model. This yielded 26 unique vascular target candidates across clusters. SMAD3 (pericyte target) and VEGFA (astrocyte ligand) were prioritized based on regulation strength, directionality in AD (SMAD3 up; VEGFA down), connectivity, and biology. - Orthogonal validations: Bulk nuclei RNA from the same region/cohort was used for RT-qPCR of selected targets (SMAD3, STAT3, ANGPT2, AHNAK, ECE1, TSC22D3). RNAscope on isolated nuclei co-stained SMAD3 with vascular marker LEF1 and VEGFA with astrocytic marker AGT; positivity quantified via high-content imaging and CellProfiler. Immunofluorescence (GFAP/VEGFA, PDGFRB/SMAD3) performed in an independent small cohort. Phospho-SMAD3 IHC quantified in pericytes (morphological localization) in 10 AD and 10 controls using a custom color deconvolution algorithm. - External replication: Integrated external snRNAseq datasets for pericytes (n=4,730) and astrocytes (n=150,664) across multiple regions/studies. Downsampling applied to avoid overrepresentation. Harmony integration and subclustering; differential expression of SMAD3 (pericytes) and VEGFA (astrocytes) tested using negative binomial GLMMs (glmmTMB) with study/participant random effects and fixed effects for diagnosis, age, sex. - Antemortem blood associations: In MCSA and ADNI cohorts, tested 588 SMAD3-locus variants for association with (a) brain MRI infarcts (binary, GLM adjusted for demographics and PCs) and (b) blood SMAD3 expression (MCSA RNAseq; ADNI microarray probes; LMMs adjusting for covariates with batch as random effect). Random-effects meta-analysis combined results. Whole-brain voxel/surface analyses related blood SMAD3 expression to amyloid PET (18F-Florbetapir) SUVR and cortical thickness (FreeSurfer), with multiple-comparison correction (RFT/FDR). - Human iPSC pericyte assays: Two AD and two control female APOE ε4/ε4 iPSC lines differentiated to pericytes (N2B27-based protocol with BMP4/CHIR, Activin A, PDGF-BB). Differentiation validated by FACS (TRA-1-60↓, PDGFRβ/NG2↑), ICC, and RT-qPCR. Pericytes treated with recombinant VEGF (50/100/200 ng/mL), VEGFR2 (KDR) inhibitor cocktail (SU5416, Tivozanib, ZM 306416), or aggregated Aβ; SMAD3 expression measured by RT-qPCR at 6/12/24 h; linear mixed-effects models assessed treatment effects. - Zebrafish in vivo: Aβ42 injected into adult telencephalon of Tg(her4:DsRed);Tg(fli1a:eGFP) double reporters; FACS of astroglia/vascular cells followed by scRNAseq (GRCz11). Expression of vegfaa (astroglia) and smad3a (pericytes) compared between Aβ and PBS. To inhibit Vegf signaling, Tg(kdrl:GFP) fish treated with Vegfr2 blockers; immunostaining quantified pERK/GFP colocalization (pathway activity), pSMAD3 in endothelia and pericytes, and ZO-1/GFP colocalization (BBB tight junction integrity). Statistics used unpaired t-tests with Welch’s correction; sample sizes ≥4 per group. - Ethics and QA: IRB approvals, de-identified human specimens with consent; zebrafish IACUC approvals; extensive QC, randomization, blinding, and replication strategies detailed.

- Cell-type landscape: From 87,493 nuclei (78,396 post-QC) across 24 donors (12 AD, 12 controls), three vascular clusters were identified and annotated as pericytes (cl.25; n=926), endothelia (cl.26; n=739), and perivascular fibroblasts (cl.30; n=488), all expressing BBB-specific LEF1. - Differential expression (vascular): Pericytes were most perturbed in AD with 220 DEGs (156 up, 64 down), versus endothelia 44 (34 up, 10 down) and perivascular fibroblasts 14 (8 up, 6 down). Four genes (INO80D, LINGO1, RASGEF1B, SLC26A3) were upregulated across all three vascular clusters. In pericytes, growth-factor signaling genes (FLT1, SMAD3, STAT3) were upregulated, cytoskeletal genes (DMD, MYO1B) downregulated. Endothelia showed upregulation of angiogenesis-related genes (ANGPT2, INSR) and STAT3. - Differential expression (astrocytes): Astrocyte clusters exhibited widespread perturbations: cl.8 (696 DEGs: 312 up, 384 down), cl.11 (822 DEGs: 573 up, 249 down), cl.31 (328 DEGs: 139 up, 189 down). Upregulated terms included cytoskeleton and differentiation; downregulated terms included cell signaling, neurogenesis, and cilia/motility. ~23% of astrocytic DEGs were shared across clusters. - Ligand–target prioritization: NicheNet identified 26 unique vascular targets regulated by 40 astrocytic ligands. SMAD3 (pericyte target) had the strongest astrocytic connectivity and was significantly upregulated in AD pericytes (β=0.47, q=3.42e-05). VEGFA, predominantly expressed in astrocytes, was significantly downregulated in AD astrocyte cl.8 and was a top ligand predicted to regulate pericytic SMAD3. - Orthogonal validations: qPCR on bulk nuclei confirmed higher expression in AD for SMAD3, STAT3, AHNAK, ECE1, TSC22D3 (ANGPT2 trended, p=0.066). RNAscope co-localized SMAD3 with LEF1+ vascular nuclei and VEGFA with AGT+ astrocytic nuclei; immunofluorescence showed SMAD3 in PDGFRB+ pericytes and VEGFA in GFAP+ astrocytes. Phospho-SMAD3 immunoreactivity was significantly higher in AD pericytes vs controls (p<0.01; n=10 per group). - External replication: In integrated pericyte datasets (n=4,730; two subclusters), SMAD3 was significantly upregulated in AD in multiple cohorts and in the combined analysis for the predominant subcluster. In integrated astrocyte datasets (n=150,664; 14 subclusters), VEGFA was significantly downregulated in AD in multiple cohorts and in combined analyses for the largest subclusters; no cohort showed VEGFA upregulation. - Blood associations: Meta-analysis across ADNI and MCSA identified six intronic SMAD3 variants (rs71400360, rs12904527, rs12909923, rs71400361, rs35779650, rs28564777) nominally associated (p<0.05) with higher blood SMAD3 expression and lower odds of MRI-detected infarcts. Higher blood SMAD3 levels were associated with less brain amyloid deposition and greater cortical thickness (less atrophy), especially in temporal, parietal, and frontal lobes (multiple-comparison corrected p<0.05). - iPSC pericyte functional assays: Recombinant VEGF reduced SMAD3 expression in a treatment-duration dependent manner with a significant decrease at 24 h (p=1.17×10^-3); VEGFR2 inhibitor cocktail increased SMAD3 expression at 6/12/24 h; Aβ treatment did not significantly change SMAD3. - Zebrafish in vivo: Astroglial vegfaa expression decreased after Aβ42 injection; pericyte smad3a showed a non-significant upward trend. Vegfr2 blockade reduced pERK/GFP colocalization (confirming pathway inhibition), increased the percentage of pSMAD3+ endothelia and pericytes, and decreased ZO-1/GFP colocalization, indicating BBB impairment. pERK and ZO-1 colocalization measures were positively correlated. - Overall: AD brains show extensive GVU transcriptome changes, with pericytic SMAD3 upregulation and astrocytic VEGFA downregulation forming a prioritized, validated ligand–target pair linked to BBB dysfunction across human tissue, blood biomarkers, iPSC pericytes, and zebrafish models.

The study addressed the gap in understanding molecular crosstalk within the gliovascular unit contributing to BBB dysfunction in AD. Single-nucleus profiling revealed that pericytes are the most transcriptionally perturbed vascular cell type, with enrichment in signaling pathways, while astrocytes exhibit broad shifts impacting cytoskeletal and neurogenic processes. Integrative ligand–target analyses highlighted pericytic SMAD3 as a key, upregulated target with strong connectivity to astrocytic ligands, notably VEGFA, which was downregulated in AD astrocytes. These reciprocal changes suggest disrupted astrocyte-to-pericyte signaling. Orthogonal validations confirmed cell-type localization and directionality of SMAD3 and VEGFA changes, and phospho-SMAD3 increases in AD pericytes support activation of SMAD3 signaling in vivo. Replication across external snRNAseq datasets demonstrated the generalizability of pericyte SMAD3 upregulation and astrocyte VEGFA downregulation across multiple brain regions and studies. Antemortem analyses linked genetic and expression variation at the SMAD3 locus to brain vascular disease (infarcts) and to amyloid and neurodegeneration markers, implicating peripheral SMAD3 as a biomarker candidate. Functional experiments established an inverse relationship between VEGF signaling and pericytic SMAD3 levels: VEGF decreases, whereas VEGFR2 inhibition increases, SMAD3 expression in human pericytes; in zebrafish, reduced Vegf signaling increased pSMAD3 in vascular cells and impaired BBB tight junction integrity. Together, these findings support a mechanistic model in which AD-related reductions in astrocytic VEGFA signaling elevate pericytic SMAD3 expression and activity, contributing to BBB disintegrity. The work advances the field by providing a systems-level, cross-validated framework linking GVU transcriptional perturbations to functional BBB outcomes and by nominating SMAD3–VEGFA as a therapeutic and biomarker axis.

This study provides a comprehensive, multi-tiered analysis of gliovascular perturbations in AD. It identifies pericytes as the most transcriptionally altered vascular cell type, prioritizes the pericytic SMAD3–astrocytic VEGFA axis as a key perturbed ligand–target pair, and validates these findings across human postmortem tissues, external single-nucleus datasets, peripheral blood, human iPSC-derived pericytes, and zebrafish models. The results support a model in which reduced astrocytic VEGFA signaling elevates pericytic SMAD3 expression and signaling, contributing to BBB disintegrity in AD. These data highlight SMAD3 and VEGFA as candidate targets and biomarkers for BBB-related pathology in AD. Future studies should investigate upstream regulators and receptors mediating SMAD3 changes, dissect cell-type specific causal roles, extend analyses to additional brain regions and disease stages, and test therapeutic modulation of the SMAD3–VEGF pathway on BBB integrity and AD progression.

- Scope of interactions: Analyses focused on astrocyte–vascular interactions; potential ligand–target relationships with neurons, oligodendrocytes, and OPCs were not systematically interrogated. - Target selection: Experimental validation centered on a single prioritized pair (VEGFA–SMAD3); other predicted pairs remain to be tested. - Mechanistic details: The specific receptor complexes and binding partners mediating VEGFA’s regulation of pericytic SMAD3 in human brain remain to be identified. - Cohort and region: Discovery used a modest-sized cohort (n=24) and one brain region (temporal cortex) from late-stage AD, limiting power and generalizability. - Pericyte rarity: Low abundance of pericytes in zebrafish limited statistical power for smad3a analyses. - Causality: Antemortem blood associations are observational and do not establish causality; directionality of SMAD3 signaling effects may be context-dependent (protective vs detrimental). - Heterogeneity: Differences across external datasets (sampling, enrichment, APOE distributions, regions) required downsampling/integration strategies and may introduce residual heterogeneity.