Abstract
The human brain vasculature is of vast medical importance: its dysfunction causes disability and death, and the specialized structure it forms—the blood-brain barrier—impedes treatment of nearly all brain disorders. Yet, no molecular atlas of the human brain vasculature exists. Here, we develop Vessel Isolation and Nuclei Extraction for Sequencing (VINE-seq) to profile the major human brain vascular and perivascular cell types through 143,793 single-nucleus transcriptomes from 25 hippocampus and cortex samples of 17 control and Alzheimer’s disease (AD) patients. We identify brain region-enriched pathways and genes divergent between humans and mice, including those involved in disease. We describe the principles of human arteriovenous organization, recapitulating a gradual endothelial and punctuated mural cell continuum; but discover that many zonation and cell-type markers differ between species. We discover two subtypes of human pericytes, marked by solute transport and extracellular matrix (ECM) organization; and define perivascular versus meningeal fibroblast specialization. In AD, we observe a selective vulnerability of ECM-maintaining pericytes and gene expression patterns implicating dysregulated blood flow. With an expanded survey of brain cell types, we find that 30 of the top 45 AD GWAS genes are expressed in the human brain vasculature, confirmed in situ. Vascular GWAS genes map to endothelial protein transport, adaptive immune, and ECM pathways. Many are microglia-specific in mice, suggesting an evolutionary transfer of AD risk to human vascular cells. Our work unravels the molecular basis of the human brain vasculature, informing our understanding of overall brain health, disease, and therapy.
Main Text
Brain health depends on brain vascular health. The brain is one of the most highly perfused organs in the body, necessary to meet its unique metabolic needs1. Brain vascular dysfunction is a major contributor to stroke, the second leading cause of death worldwide2–4. Dysfunction causes serious long-term disability: vascular-specific genes are mutated in congenital neurological disorders5, 6 and age-related vascular impairments are increasingly appreciated in neurodegenerative disease7–11. The brain vasculature moreover forms a special structure, the blood-brain barrier (BBB), that mediates selective movement of molecules between the blood and the brain12–15. While necessary for optimal neuronal function16, 17, the BBB frustrates the pharmacological treatment of nearly all brain disorders15, 18, 19, and extensive efforts are underway to identify targets on the BBB for enhanced drug delivery20–22. These brain vascular properties arise from a complex ecosystem of cells and their interactions17, 23, 24: endothelial cells, adjacent mural smooth muscle cells and pericytes, perivascular immune cells, and surrounding astrocytes that differ across brain regions and vary along an arteriovenous gradient14, 25, 26. Heterogeneity along this gradient produces functionally segmented circulatory, metabolic, and permeability properties6, 27.
Recent studies have characterized the cellular heterogeneity of the human brain in health and disease using single-nucleus RNA sequencing (snRNA-seq)28–33. They have elucidated cell type-specific perturbations in multiple sclerosis, autism, and Alzheimer’s disease; pinpointed which cell types express risk genes identified in genome-wide association studies (GWAS); and nominated biological pathways for further study. Yet, though vascular cell density34, 35 is estimated at 70,000 cells/mm3 (approaching total glia density34, 36), such studies, to our knowledge, have mostly lost these cells during the isolation process for unknown reasons. Pioneering work has profiled the mouse brain vasculature37–42, but it remains unclear how conserved these findings are in humans, given the approximately 96 million years of evolutionary divergence43. Indeed, recent studies have documented species-specific pathways in microglia, notably in disease GWAS loci44; and recent attempts to advance brain-penetrant AAVs into the clinic have stalled because of mouse-specific expression of the cognate endothelial receptor LY6A45–47. Moreover, mouse brain vascular sequencing studies so far have been limited to the non-diseased setting and without regard to brain region heterogeneity.
Given the scientific, medical, and pharmacological importance of the human brain vasculature, we set out to systematically characterize the principal vascular cell types in both the hippocampus and cortex of control and AD patients.
Cells of the human brain vasculature
We hypothesized that unlike parenchymal nuclei, vascular cells and nuclei remain entombed in the basement membranes of blood vessels after typical dounce homogenization and processing of frozen brain tissue for snRNA-seq28–33, 48–52. Such vessel fragments are caught on strainers prior to droplet capture, and vessel fragments that do pass through can yield doublet or hybrid nuclei resulting in unreliable or artificial clusters upon analysis. Thus, we set out to develop methods to first physically isolate brain vessels and then extract discrete nuclei from them. Specifically, after density centrifugation53 and strainer capture54, we tested various enzymatic (e.g., papain, collagenase, trypsin), chemical (e.g., osmolarity, detergents), and physical (e.g., sonication, TissueRuptor) approaches to liberate nuclei. Nearly all resulted in nuclei damage or nuclei devoid of RNA reads. We finally found success adapting a gentle protocol for splenocyte isolation55 (Methods)—and combined it with extensive sucrose and FACS-based cleanup to ensure high-quality data (Fig. 1a, Supplemental Fig. 1).
With our new method, which we call VINE-seq (Vessel Isolation and Nuclei Extraction for Sequencing), we processed 25 samples: the hippocampus of 9 AD and 8 age- and sex-matched controls, as well as the superior frontal cortex from a subset of 8 patients (4 samples per group, Supplemental Table 1). Samples included a range of APOE genotypes (E3/3, E3/4, E4/4). After quality-control (Methods), we obtained 143,793 single nucleus transcriptomes. Visualization in uniform manifold approximation and projection (UMAP) space separated nuclei into distinct clusters, which we mapped to 15 major cell types (Fig. 1b), including all known vascular and perivascular cell types, many not captured before from human brains: endothelial cells (arterial, capillary, venous), smooth muscle cells, pericytes, astrocytes, perivascular macrophages, T cells, and both perivascular and meningeal fibroblasts. The number of cerebrovascular nuclei captured here exceed those in the literature by at least several hundred-fold (Fig. 1b, Supplemental Fig. 2). Canonical markers used to identify cell types (Supplemental Fig. 3-4, Supplemental Table 2) were validated for their predicted vascular localization in situ (Fig. 1c). Expression levels for each gene across cell types are available to browse at https://twc-stanford.shinyapps.io/human_bbb.
The distributions of cell types differed between the hippocampus and frontal cortex (Fig. 1d). Astrocyte and oligodendrocyte progenitor cell (OPC) frequencies were higher in the hippocampus, recapitulating prior cell density studies34, 56. Amongst all cell types, astrocyte transcriptional identity was the most influenced by brain region, forming distinct hippocampus- and cortex-enriched cell subclusters (Supplemental Fig. 4d-g). Pericytes, critical for regulating blood-brain barrier (BBB) function15, 23, 24, were reduced already in control hippocampi relative to cortices. Each vascular cell type exhibited brain region-specific enrichments in genes and pathways (Supplemental Table 3). For example, hippocampal endothelial cells demonstrated greater baseline inflammation, such as IFN-γ signaling, than those in the cortex (Fig. 1e). Such inflammatory signaling has recently been described to inhibit hippocampal neurogenesis57–59, and together with the aforementioned pericyte loss, provides a molecular hypothesis for the particular susceptibility of the hippocampal vasculature to dysfunction in both aging8, 54 and AD55.
We next compared nuclei transcriptomes between human and mouse endothelial cells and pericytes. Using a strict cutoff (>10x difference, logCPM > 0.5, Supplemental Table 4), we found hundreds of species-enriched genes (Fig. 1f). These include the known mouse-specific endothelial anion transporter Slco1c160 and AAV PHP.eB receptor Ly6a45–47. These also include disease-related genes such as A2M and CASS4, implicated in β-amyloid processing (Fig. 1g)61–64. Several small molecule transporters varied, suggesting species differences in brain metabolism. For instance, the GABA transporter SLC6A12 is enriched in human over mouse pericytes, with implications for GABAergic neurotransmission and associated diseases like epilepsy65. We confirmed the vascular localization of SLC6A12 and other human-enriched genes in human brain tissue at the protein level (Fig. 1g, Supplemental Fig. 5). Several genes of high pharmacological importance mediating small molecule and protein BBB transport vary between species (Supplemental Fig. 6). Finally, this dataset enables study of diseases that involve the human brain vasculature, such as genes relevant to SARS-CoV-2 neuroinvasion66, 67, neurotoxicities associated with cancer immunotherapies68 (e.g., no CD19 expression in human adult brain pericytes), and the cell type etiology of ALS69 (Supplemental Fig. 7). Together, the VINE-seq method introduced here opens the human brain vasculature for molecular study and provides an important data resource for interrogating its diverse cell types.
Organizing principles of human brain endothelial and mural cells
With our capture of large numbers of vascular nuclei (>30x mouse37, >200x human28, 29 prior studies), we sought to comprehensively characterize the molecular basis of endothelial and mural cell organization along the human brain arteriovenous axis. Cellular and molecular changes along this axis have been referred to as zonation15, 37, 41, 70, 71. Beginning with endothelial cells, a UMAP representation resolved the 36,825 captured nuclei into the known vessel segments: arterial and venous clusters were located at opposite ends, separated by a major capillary cluster (Fig. 2a, Supplemental Fig. 8a-b). These clusters were defined by established zonation markers, such as arterial VEGFC and ALPL; capillary MFSD2A and SLC7A5; and venous IL1R1 and NR2F237, 41. While capillaries make up the vast majority (∼90%)27, 72 of the endothelium, our method facilitated robust capture of rarer arterial (at 19%) and venous (at 27%) endothelial cells, likely either because of more efficient strainer retention or nuclei liberation. We also noticed a small endothelial cluster (571 nuclei, ∼0.1%) outside conventional arteriovenous zonation. This cluster expressed genes characteristic of ‘tip’ cells (e.g., PLAUR and LAMB1)73 as well as ‘proteostatic’ heat shock proteins37.
We next ordered and aligned endothelial nuclei along a single one-dimensional Monocle pseudotime74 range to better recapitulate the anatomical arteriovenous axis. As expected, known arterial and venous markers peaked at opposite ends of this range, and capillary markers peaked in between (Fig. 2b). We used the 665 most significantly variable cluster genes to order the endothelial nuclei and observed a distribution of seven gradually changing gene expression patterns, representing arterial, capillary, or venous segments, and combinations thereof (Fig. 2c). We confirmed that the Monocle range represented a cell order matching anatomical arteriovenous zonation by examining data from the Human Protein Atlas75 (Supplemental Fig. 9a). Our patterns recapitulate the gradual/ seamless zonation continuum described in mice—but interestingly, this similar overall continuum arises from significantly different individual/ component zonation markers (Supplemental Table 2, 5). For example, only a minority of the top 100 human arterial, capillary, and venous markers are such in mice (Fig. 2j), even if we expand the denominator of mouse genes compared against to 500 genes.
We thus wondered whether established zonation markers in mice would be conserved in humans. We calculated a score (Methods) measuring each gene’s specificity to a given zonation (e.g., arterial, capillary, venous). Indeed, we observed across all vessel segments a significant number of markers that lost their predictive value between species (Fig. 2d, Supplemental Fig. 10). For example, the blood clotting gene von Willebrand factor (VWF) is largely expressed in mouse venous endothelial cells37. However, in humans, VWF is highly expressed throughout the endothelium, even in small diameter capillaries (Fig. 2d-e). VWF abundance has been tied to increased risk for ischemic stroke76, 77, and its species-specific distribution could be one reason why mouse models of stroke have faced notoriously low translational success rates78, 79.
We next visualized and clustered 34,508 mural cell nuclei in UMAP space, resolving clusters for arterial smooth muscle cells (aSMCs) and arteriolar SMCs (aSMCs)—but interestingly, we also discovered two subclusters of pericytes (Fig. 2f, Supplemental Fig. 8c-e). One pericyte subcluster was enriched for small molecule transmembrane transport activity, which we refer to as T-pericytes (for transport); while the other for extracellular matrix (ECM) formation and regulation, which we refer to as M-pericyte (for matrix). This suggests that function rather than anatomical location is the major driver of pericyte transcriptional identity in humans, and that both capillary and venous (vSMC) pericytes span these two functional clusters (confirmed below, Fig. 2g-h, Supplemental Fig. 9b, d). The existence of an M-pericyte cluster holds interesting implications for small vessel diseases like CADASIL, CARASIL, and Collagen IV deficiencies for which perturbations in the vascular ECM cause disease80, 81. Moreover, human T-pericytes—but not mouse pericytes—express transporters like the GABA transporter 1 SLC6A1 (involved in epilepsy) and glutamate transporter SLC1A3 (Supplemental Fig. 9c), suggesting an evolutionary pressure for expanded solute transport across the human BBB. Because recent mouse pericyte datasets have reported confounding endothelial contamination37, 71, 82, 83, we assessed and found no such contamination in our human pericyte nuclei (Supplemental Fig. 3, 7b). The isolation of nuclei instead of whole cells may aid in minimizing contaminating endothelial fragments.
To study mural cell zonation, we similarly compared the distribution of known mural cell transcripts across the Monocle range with corresponding protein expression in situ (Fig. 2g, Supplemental Fig. 9b, d). We used the 799 most significantly variable cluster genes to order all 36,825 mural cell nuclei and observed the expected order of aSMC markers on one end (e.g., ACTA2, TAGLN), followed by aaSMC (e.g., CTNNA3, SLIT3); and pericyte markers on the other end (e.g., ABCC9, PTN). Recapitulating the mural cell pattern described in mice37—and as opposed to the gradual zonation pattern in endothelial cells—, we observe in human mural cells an abrupt transition between SMCs and pericytes: one set of transcripts are expressed highly in aSMCs and aaSMCs but at low levels in pericytes, while another set of transcripts exhibits the opposite pattern (Fig. 2g). As expected from their clustering by functional pathways, the two pericyte subclusters did not segregate along the Monocle range and localized across both large and small diameter vessels in situ, suggesting that they intercalate throughout the capillary and venous vasculature (Fig. 2h). Moreover, previously reported vSMC markers37 were expressed in both pericyte clusters (Supplemental Fig. 9d).
Given species-specific differences across all brain cell types, we find that only a minority of the top mouse SMC and pericyte markers retain their predictive value in humans (Fig 2j, Supplemental Fig. 9d). Because the proteins encoded by zonation marker genes perform a variety of important functions at defined arteriovenous locations, species-specific endothelial and mural cell differences likely reflect fundamental differences in brain vascular properties that can now be tested to inform translational studies.
Molecular definitions for perivascular and meningeal fibroblasts
Complex barrier structures maintain brain homeostasis84. Cooperating with the vascular BBB, the recently (re)discovered meningeal lymphatics plays important roles in waste clearance and neuroimmune surveillance85–88. Using annotations from recent mouse studies42, 89, 90, we noticed our capture of fibroblast-like cells from both the meningeal and vascular barriers provided the opportunity to directly compare these populations for insights into their specialized functions (Fig. 3a-e). First, we noticed that fibroblasts transcriptionally segregated according to anatomical location: vascular versus meningeal (Fig. 3a-b), but also separated according to the layers of the meninges (Fig. 3a, d). No strong differences were observed between hippocampal and cortical-derived fibroblasts (Supplemental Fig. 11b), suggesting that the micro- but not macro-environment shapes brain fibroblast identity.
Pathway enrichment analysis of marker genes demonstrated a strong divergence in fibroblast functions by anatomical location (Fig. 3b, e): perivascular fibroblast-like cells showed enriched expression for ECM structural components or its modifiers and receptors (e.g., “TGF- β regulation of the ECM), while meningeal fibroblasts enriched for solute transporters. This suggests that perivascular but not meningeal fibroblasts form fibrotic scars after brain injury91, 92. Closer comparison of differentially expressed genes between fibroblast populations (Fig. 3e) revealed a remarkable polarization of solute influx and efflux pumps: meningeal fibroblasts specifically expressed SLC influx solute transporters, while perivascular fibroblasts exclusively expressed ABC efflux pumps (Fig. 3f). Perivascular fibroblasts reside in the Virchow-Robin space, and thus like meningeal fibroblasts, come into contact with the cerebrospinal fluid (CSF). This cooperative circuit of polarized transporters suggests fibroblast regulation of solute exchange between the brain and CSF. A gradient of polarized influx/ efflux within a shared CSF compartment also provides evidence for convective rather than diffusive fluid flow via the recently described ‘glymphatic’ system85.
Because perivascular fibroblast-like cells reside in close proximity to other vascular cells captured, we used our single-cell data to infer cell-cell communication pathways93, 94. This corroborated fibroblast-like cells as major recipients of TGF-β signaling (Supplemental Fig. 11c-d). Overstimulation of TGF-β signaling in the brain vasculature promotes ECM basement membrane thickening and triggers neuropathology, though the effector cell type has been unknown95, 96. This analysis nominates perivascular fibroblast-like cells. Cell-cell communication analysis also predicted signaling between capillaries and fibroblast-like cells, despite their localization in mice exclusively around arterial and venous vasculature37. Using fibroblast-specific genes conserved in mice and humans, we surprisingly found fibroblast-like cell marker co-localization around human capillaries (<10 μM diameter) as well as larger vessels in situ (Fig. 3h, Supplemental Fig. 12). Fibroblast or fibroblast-derived protein localization in capillaries poses interesting questions for vessel development and maintenance. As in endothelial and mural cells, perivascular fibroblast-like cell markers varied by species (Fig. 3j). Together, these data provide a first characterization of human brain fibroblast diversity, revealing the molecular basis of their anatomical specialization and a cooperative circuit for CSF solute exchange.
Vascular cell-type specific perturbations in Alzheimer’s disease
Alzheimer’s disease (AD) is a progressive neurodegenerative disorder culminating in severe impairment of memory, cognition, and executive functions97, 98. These impairments arise from complex perturbations in cell composition and gene expression99–101. We thus sought to profile changes in the AD human brain vasculature at single-cell resolution. We defined our AD patient group by clinical diagnosis and confirmed via immunohistochemistry the presence of β-amyloid plaques in the hippocampus and cortex (Supplemental Fig. 13).
Recent studies have identified context-dependent, disease associated glial subpopulations28–30, 67, 102. We did not observe new vascular cell subclusters emerging with AD (Fig. 4a, Supplemental Fig. 8). However, in contrast to parenchymal cells28, 29, we found a strong loss of brain vascular nuclei—across endothelial, SMC, pericyte, and fibroblast-like cells—with AD (Fig. 4a-b). This is consistent with reports of focal cerebrovascular damage in AD27, 103, but suggests that vascular loss is widespread across cells types throughout the arteriovenous axis. Interestingly, among pericytes, only M-pericytes involved in ECM organization declined (Fig. 4b). This selective, disease-associated susceptibility provides a molecular hypothesis for the physical BBB breakdown and vascular ECM disruption reported in AD104. Finally, because the observed reductions in vascular nuclei could also reflect an increased fragility during isolation, we stained for and confirmed our findings in situ (Fig. 4c). In short, instead of a new AD-associated subpopulation emerging, we find in the vasculature an AD-associated disappearance of specific cell type subpopulations.
We next systematically examined cell type-specific gene expression changes in AD (Methods). Across the vascular cell types captured in sufficient numbers for statistical power, we identified 463 unique differentially expressed genes using more stringent thresholds (DEGs, Methods, Fig. 4d, Supplemental Table 6). Overall, mural cells exhibited the strongest changes, with other cell types showing a signature of gene repression: 61-78% of DEGs were downregulated (Fig. 4d). DEGs were robustly detected across different levels of expression (Supplemental Table 6). The vast majority of DEGs were cell type- (Fig. 4e-f) and zonation-specific, suggesting a heterogenous response to AD pathology across the vasculature. Intriguingly, several DEGs are risk genes implicated in recent AD and small vessel disease GWAS studies (Fig. 4f). At the pathway level, DEGs in mural cells and fibroblast-like cells implicated dysregulated vasoconstriction and compromised blood flow (Fig. 4g). This provides a molecular basis for the cerebral hypoperfusion discernible in MRI-based imaging of living AD patients105, 106. Interestingly, DEGs in pericytes and SMCs resembled those found in CADASIL and CARASIL107–110 (Fig. 4h), rare hereditary diseases also marked by impaired blood flow, cognitive deficits, repeated strokes, and dementia.
APOE4 carriers have been reported to exhibit accelerated BBB breakdown before cognitive impairment11, 111, though the underlying mechanisms are unclear in humans. With patient APOE genotypes, we performed similar DEG analysis, finding dramatic interferon inflammation in the endothelium of APOE4 carriers (Fig. 4i, Supplemental Fig. 14, Supplemental Table 7). Finally, we sought to understand the overlap between human vascular AD DEGs and those in mouse models of AD. Such models have facilitated mechanistic study of β-amyloid pathology, but recent reports describe significant species differences in various cell types, such as microglia30, 44, 102. We isolated brain endothelial cells from 12-14 month old Thy1-hAPPLon,Swe mice (and littermate wild-type controls)112 that present prominent amyloidosis, neuroinflammation, and synapto-dendritic degeneration—and processed them for single-cell sequencing. Surprisingly, we observed minimal overlap between human AD and mouse hAPP DEGs in brain endothelial cells (Fig. 4j).
AD pathology begins and spreads via a strikingly consistent regional pattern98, 113. We thus assessed the impact of AD on brain regional vascular specialization (Fig. 1e). Within controls, we found greater vascular density in the cortex compared to hippocampus (Fig. 4k), reflecting either regional baseline differences or hippocampal deficits with normal aging114. We found these regional differences erased in AD patients (Fig. 4k). Likewise, by comparing the number of DEGs between the cortex and hippocampus of the same patients, we noticed a global loss of brain regional specialization across vascular cell types in AD patients (Fig. 4l)— suggesting impairments in brain region-specific vascular function. Together, these findings show that AD patients exhibit heterogeneous cell type-, zonation-, region-, and species-specific perturbations across the brain vasculature that require dedicated isolation and single-cell approaches to profile.
AD GWAS disease variants enriched in the human brain vasculature
A major goal of biomedical research is to identify genes that cause or contribute to disease. GWAs studies have shed insight into the molecular pathways contributing to AD115, 116, though the cell type context in which GWAS genes are expressed was long unknown. Recent snRNA-seq and other cell type-resolved studies have strongly implicated microglia as the major AD GWAS-expressing cell type28–30, 64, 116–121. We wondered, however, whether the unintended depletion of brain vascular cells in conventional preparations may have prematurely dismissed such evidence in brain vascular cells. We curated recent AD GWAS studies117–119, 122, 123 to identify and order the top 45 risk genes. With our more comprehensive survey of brain cell types, we calculated the cell type proportional expression for each GWAS gene using Expression Weighted Cell Type Enrichment (EWCE)124. We indeed observed among brain parenchymal cells a specific myeloid signature for top AD GWAS genes such as TREM2, MS4A6A, CR1, and SPI1 that are now the subject of intense mechanistic study (Fig. 5a, right).
Intriguingly, we noticed that several GWAS genes were strongly expressed in human brain vascular and perivascular cell types (Fig. 5a, left, Supplemental Fig. 15). This included the two GWAS genes previously implicated in the mouse vasculature, PICALM and CD2AP64, 122. But this also included other surprising genes, such as the immune-related PLCG2 and HLA-DRB1/5 in arterial cells, the endocytic INPP5D and USP6NL in capillaries, and ECM-related ADAMTS1, ADAMTS4, FERMT2, and AGRN in SMCs and pericytes (Fig. 5a). Within pericytes, expression varied across M- and T-pericyte subtypes (Supplemental Fig. 16a). APOE, often linked to myeloid cells and astrocytes, was robustly expressed in human SMCs and meningeal fibroblasts. Remarkably, several GWAS genes like ABCA7 and CLNK were enriched in perivascular T cells. Consistent with our findings, an independent dataset shows minimal expression of these genes in parenchymal brain cell types (Supplemental Fig. 16b). Likewise, several GWAS genes like ABCA1, FHL2, HESX1, and IL34 were enriched in perivascular and meningeal fibroblasts. Importantly, we confirmed our findings via immunohistochemical staining. We observed vascular localization for 12 proteins encoded by predicted-vascular GWAS genes, such as CASS4, FERMT2, ACE, PLCG2, and FHL2 (Fig. 5b). Most GWAS genes exhibited minor expression differences between the hippocampus and cortex (Supplemental Fig. 16c). In total, at least 30 of the top 45 AD GWAS genes are enriched in cells of the human brain vasculature (not including those solely in perivascular macrophages). Their distribution across all vascular cell types suggests that vascular and perivascular involvement in AD pathology may be more thorough and complex than anticipated.
The human brain vascular expression of putatively parenchymal—especially myeloid— AD GWAS genes made us wonder whether these genes are expressed in different cell types between mice and humans. We thus examined the expression of human vascular GWAS genes in mouse datasets90. Indeed, many genes like APOE, CASS4, INPP5D, and HLA-DRB1 were predominately expressed in microglia in mice but then also exhibited vascular expression in humans (Fig. 5c, Supplemental Fig. 16d). Corroborating this, nearly every top GWAS gene expressed in BECs exhibited greater expression in humans than in mice (Fig. 5c). Together, these data suggest a partial evolutionary transfer of AD risk genes and pathways from microglia to the vasculature from mice to humans, with implications for translational studies.
We next broadened our scope to a previously compiled list of hundreds of GWAS genes for AD and AD-related traits (Supplemental Fig. 17a-b)29. We observed robust expression across vascular and perivascular cell types (Supplemental Fig. 17a-b). Using EWCE analysis, we found that many risk genes were expressed cell type-specifically (Supplemental Fig. 17). For each gene, we assigned the cell type with the strongest expression, discovering surprisingly, that endothelial cells harbored the most AD-related GWAS genes, followed by microglia/ macrophages (Fig. 5d, Supplemental Table 8). Within BECs, AD-related GWAS genes (Fig. 5a-b) enriched for protein endo- and transcytosis components, such as receptor and clathrin vesicle components (Fig. 5d). We recently demonstrated a decline in BEC clathrin-mediated transcytosis15 with age, suggesting one mechanism by which aging and GWAS genes converge to impair β-amyloid clearance and increase AD risk. In total, over half of all AD-related GWAS genes mapped to vascular or perivascular cell types (383 of 651).
As with top AD GWAS genes, we observed enhanced human over mouse expression of AD-related genes in both BECs and pericytes (Fig. 5e). Importantly, this human-enhanced expression is not observed for the whole transcriptome. Together, these data provide a more comprehensive understanding of the cell types contributing to AD risk. We suggest that a vascular-microglia axis underlies the genetic risk for AD via shared protein clearance (BEC-microglia) and inflammatory pathways (BEC-T cell-microglia) (Fig. 5f), and that this axis is evolutionarily expanded in humans.
Discussion
We report here 143,793 single-cell, genome-wide quantitative transcriptomes from the human brain vasculature in health and AD. We use these transcriptomes to molecularly define the principal vascular cell types; their differences by brain region and species; the organizational principles of endothelial, mural, and fibroblast-like cells; a selective loss of M-pericytes and the transcriptomic perturbations contributing to clinical AD dementia; and the unexpected expression of AD GWAS genes across the human brain vasculature. We subsequently confirm these findings in situ at the protein level. Single-cell resolution was necessary for these findings, which would have been obscured in the average profiles generated via bulk RNA-sequencing.
How do human vascular GWAS genes fit into established AD pathways? Current understanding implicates β-amyloid metabolism, cholesterol/ lipid dysfunction, innate immunity, and endocytosis121, 125. Vascular GWAS gene expression confirms these pathways and expands the set of cell types involved, such as β-amyloid endocytosis and clearance via BEC clathrin-mediated transport; and adaptive in addition to innate immunity via perivascular T cells (Fig. 5f). We propose that the dramatic expansion of the human brain, brain activity, and activity byproducts (like β-amyloid126) necessitates enhanced neuroimmune surveillance and clearance mechanisms. In this model, microglia are still frontline participants in AD pathogenesis. But more so than in mice, human vascular and perivascular cells partake. For example, the clearance functions of microglia can become overwhelmed127, diverting debris clearance to BECs. This is supported by recent studies finding microglial depletion results in cerebral amyloid angiopathy128. But unlike microglia129, vascular cells are unable to proliferate efficiently130. Thus, constant vascular exposure to β-amyloid triggers dysfunction via cell loss and impaired blood and CSF flow131 (Fig. 4). Recent work identified a human-unique CD8 TEMRA (CD45RA+CD27−) population clonally expanded in AD CSF132. Thus, it is possible that perivascular/ meningeal T cell GWAS hits contribute to inflammatory pathology in ways not seen in mice. Together, we suggest an intertwined microglia-vascular axis expanded in humans, with vascular cells playing an auxiliary role via shared endocytosis and inflammatory pathways. We note though the likelihood of additional vascular contributions, as evidenced by SMC, pericyte, and both perivascular and meningeal fibroblast-enriched GWAS genes of unclear function.
Given the evolutionary divergence between mice and humans, these data expand by orders of magnitude the number of arteriovenous markers in the human brain that can now be used for vessel type identification, to assess the fidelity of in vitro human cell and organoid cultures, and to reliably deconvolute and enhance the utility of hundreds of publicly available bulk brain RNA-seq datasets (Supplemental Table 2). Mechanistic studies in mouse models can now be combined with this dataset to systematically pinpoint the cell types and genes mediating core vascular functions such as cerebral autoregulation and BBB permeability that are often seen perturbed in disease by clinical imaging10. Despite recent landmark studies on human brain parenchymal cells116, many AD and other disease risk variants remain unmapped. Because such variants enrich in gene expression-regulating enhancer regions133 that undergo accelerated evolution134, they may exert their influence through human brain vascular cells. With further optimization, this method should be compatible with ATAC-, ChIP-, and PLAC-seq assays135–137 to address this possibility.
Our work opens several translational opportunities. This dataset informs ongoing efforts to develop ‘brain shuttles’ and other modalities to better deliver therapeutics to treat human brain disorders15, 21. This method facilitates study of the brain vasculature across a variety of disease conditions, such as stroke, multiple sclerosis, and even COVID-1967. Together, the field now has a near complete catalog of cell types in the human brain, which can be integrated into ongoing efforts such as the Human Cell Atlas138. As with recent high-profile snRNA-seq studies of the brain parenchyma28–32, 139, it will now be important to distinguish which of the vascular transcriptional perturbations observed in disease are responsive versus driving, clarify their links to various clinical and pathologic traits, and dissect the exact mechanisms by which vascular-expressed AD GWAS genes confer greater disease risk. Overall, the VINE-seq method introduced here and the ensuing single-cell data (https://twc-stanford.shinyapps.io/human_bbb) provide a blueprint for finally studying the molecular makeup of the human brain vasculature, promising further discoveries in health, disease, and therapy.
Author contributions
A.C.Y. and T.W.-C. conceptualized the study. A.C.Y. devised the isolation method. M.W.M. provided and A.C.Y. organized tissue samples. D.L.P. and A.C.Y. performed tissue dissociations. N.S., R.T.V., D.G., K.C., and A.C.Y. prepared libraries for sequencing. R.T.V., F.K., A.K., C.A.M., M.B.C., R.P., A.S., N.K., and A.C.Y. performed computational analysis. D.P.L, C.A.M., M.A., D.G., E.Y.W., J.L., and A.C.Y. performed immunohistochemical stains. P.M.L. developed the searchable web interface (Shiny app). C.A.M. and A.C.Y. drew diagrams. A.C.Y. wrote the manuscript with input from all authors. A.C.Y. and T.W.-C. supervised the study.
Competing interests
T.W.-C. is a founder and scientific advisor of Alkahest Inc.
Methods
Isolation of vascular nuclei from frozen post-mortem brain tissue
Post-mortem fresh-frozen hippocampus and superior frontal cortex tissue were obtained from the Stanford/ VA/ NIA Aging Clinical Research Center (ACRC) with approval from local ethics committees and patient consent. Group characteristics are presented in Supplemental Table 1. Note, patients were grouped by clinical diagnosis, with two of the control patients exhibiting amyloid beta plaque staining in the hippocampus, though not to a sufficient degree for an expert pathologist to diagnose Alzheimer’s disease by histopathological criteria. Clinical instead of pathologic diagnosis was chosen because of potentially vascular contributions to AD independent of the well-known hallmarks of AD, β-amyloid and tau pathophysiology7. All procedures were carried out on ice in a 4°C cold room as rapidly as possible. 0.3 grams or more of brain tissue was thawed on ice for 5 minutes with 5 ml of nuclei buffer (NB): 1% BSA containing 0.2 U μl−1 RNase inhibitor (Takara, 2313A) and EDTA-free Protease Inhibitor Cocktail (Roche, 11873580001). Tissue was quickly minced and homogenized with 7 ml glass douncers (357424, Wheaton) until no visible chunks of debris remained. Similar to before54, homogenates were transferred into 50 ml tubes containing 35 ml of chilled 32% dextran (D8821, Sigma) in HBSS. Samples were vigorously mixed before centrifugation at 4,400g for 20 minutes with no brake. After centrifugation, samples separate into a top myelin layer, middle parenchymal layer, and vascular-enriched pellet. The myelin layer was aspirated, tips changed, and the parenchymal layer carefully removed without disturbing the pellet. Pellets were resuspended in 8 ml of 32% dextran, transferred to 15 ml falcon tubes, and centrifuged again. Vascular-enriched pellets were gently resuspended in 1ml of NB and added to pre-wetted 40 μm strainer sitting atop 50 ml falcon tubes. From here diverging from prior protocol, strainers were washed with 10 ml of cold 0.32 M sucrose in PBS and 90 ml of PBS until flow through the strainers was unimpeded to deplete contaminating parenchymal cells from trapped microvessels. At this step, retained microvessels turn white in color, indicating removal of circulating blood cells. Strainers were switched to new collection 50 ml falcon tubes. Various techniques were tested and optimized to extract vascular cells from the isolated microvessels (e.g., enzymatic digestion, TisssueRuptor, sonication, etc.), but nearly all resulted in loss of nuclei integrity or low nuclei complexity (<50 median genes/ nuclei). Eventually, adapting a method for the isolation of murine splenocytes proved successful: vascular fragments were mashed four times through the cell strainer using the plunger end of a 3 ml syringe, with intermittent elution via 10 ml of 0.32 sucrose and 40 ml of PBS. Liberated vascular cells were pelleted at 500g for 10 minutes and resuspended in 1.5 ml of EZ Prep Lysis Buffer (Sigma, NUC101) spiked with 0.2 U μl−1 RNase inhibitor (Takara, 2313A) and EDTA-free Protease Inhibitor Cocktail (Roche, 11873580001). Nuclei were homogenized with 2 ml glass douncers (D8938, Sigma) 20 times with pestle B (pestle A optional). Spiked EZ lysis buffer was added to samples up to 4 ml and incubated on ice for 5 minutes before pelleting at 500g for 6 minutes. This incubation step was repeated. Debris was depleted via a sucrose gradient before flow cytometry isolation of nuclei. Briefly, pelleted nuclei were resuspended in 0.5ml of NB before the addition of 0.9 ml of 2.2 M sucrose in PBS. This mixture was layered atop 0.5 ml of 2.2 M and samples were centrifuged at 14,000g for 45 minutes at 4°C, with no brake. Pellets were aspirated in 1ml of NB, filtered through a 40 μm strainer (Flowmi), transferred to FACS tubes, stained with Hoechst 3342 (1:2000, Thermo) and rabbit monoclonal anti-NeuN Alexa Fluor® 647 (1:500, Abcam, ab190565), and nuclei collected on a SH800S Cell Sorter into chilled tubes containing 1 ml of NB without protease inhibitor. In pilot runs, we noticed the cytometer overestimated nuclei counts by ∼3.4x, and thus we sorted ∼34,000 nuclei to target ∼10,000 nuclei per sample. Sorted samples were inspected for lack of debris on a brightfield microscope. We note that an iodixanol gradient52 can substitute for the 2.2 M sucrose, but that unfortunately with either gradient, flow sorting is required—unlike parenchymal myelin debris, vascular debris is not sufficiently removed by gradient centrifugation alone. Vascular debris will confound downstream cDNA traces with higher background and low molecular weight peaks. We are happy to share the detailed protocol widely but note that since high-quality human postmortem brain tissue is difficult to obtain, tissue would be limited in quantities to share widely.
Droplet-based snRNA-sequencing
For droplet-based snRNA-seq, libraries were prepared using the Chromium Single Cell 3ʹ v3 according to the manufacturer’s protocol (10x Genomics), targeting 10,000 nuclei per sample after flow sorting (Sony SH800S Cell Sorter). 15 PCR cycles were applied to generate cDNA before 16 cycles for final library generation. Generated snRNA-seq libraries were sequenced on S4 lanes of a NovaSeq 6000 (150 cycles, Novogene).
snRNA-seq quality control
Gene counts were obtained by aligning reads to the hg38 genome (refdata-gex-GRCh38-2020-A) using CellRanger software (v.4.0.0) (10x Genomics). To account for unspliced nuclear transcripts, reads mapping to pre-mRNA were counted. As previously published, a cut-off value of 200 unique molecular identifiers (UMIs) was used to select single nuclei for further analysis28. As initial reference, the entire dataset was projected onto two-dimensional space using Uniform Manifold Approximation and Projection (UMAP) on the top 30 principal components142. Three approaches were combined for strict quality control: (1) outliers with a high ratio of mitochondrial (>5%, <200 features) relative to endogenous RNAs and homotypic doublets (> 5000 features) were removed in Seurat143; (2) after scTransform normalization and integration, doublets and multiplets were filtered out using DoubletFinder144; and (3) after DoubletFinder, nuclei were manually inspected using known cell type-specific marker genes, with nuclei expressing more than one cell type-specific marker further filtered144. For example, BEC nuclei containing any reads for the following cell type markers were subsequently filtered: MOBP, MBP, MOG, SLC38A11, LAMA2, PDGFRB, GFAP, SLC1A2, and AQP4. We note that the vascular nuclei in prior human single cell datasets exhibit contamination with other cell type-specific gene markers, potentially confounding downstream analysis. After applying these filtering steps, the dataset contained 143,793 high-quality, single nuclei.
Cell annotations & differential gene expression analysis
Seurat’s Integration function was used to align data with default settings. Genes were projected into principal component (PC) space using the principal component analysis (RunPCA). The first 30 dimensions were used as inputs into Seurat’s FindNeighbors, FindClusters (at 0.2 resolution) and RunUMAP functions. Briefly, a shared-nearest-neighbor graph was constructed based on the Euclidean distance metric in PC space, and cells were clustered using the Louvain method. RunUMAP functions with default settings was used to calculate 2-dimensional UMAP coordinates and search for distinct cell populations. Positive differential expression of each cluster against all other clusters (MAST) was used to identify marker genes for each cluster145. We annotated cell-types using previously published marker genes28, 30, 32, 146. For brain endothelial cells, zonation specificity scores for each gene were calculated separately for arterial, capillary, and venous segments as in the following example for a given gene in capillaries:
Differential gene expression of genes comparing Alzheimer’s disease, ApoE4, and control samples—or comparing cell type sub-cluster markers—was done using the MAST145 algorithm, which implements a two-part hurdle model. Seurat natural log (fold change) > 0.5 (absolute value), adjusted P value (Bonferroni correction) < 0.01, and expression in greater than 10% of cells in both comparison groups were required to consider a gene differentially expressed for subcluster analysis and natural log (fold change) > 0.3 (absolute value), adjusted P value (Bonferroni correction) < 0.01, and expression in greater than 10% of cells in both comparison groups for Alzheimer’s disease and ApoE4 comparisons, both more stringent than the default Seurat settings. We incorporated age, gender, and batch as covariates in our model. A more lenient threshold of the above but with natural log (fold change) > 0.2 (absolute value) was used for brain region (i.e., hippocampus vs cortex. Biological pathway and gene ontology enrichment analysis was performed using Enrichr147 or Metascape141 with input species set to Homo sapiens141. UpSet plots were generated using identified differentially expressed genes as inputs using the R package UpSetR148. Diagrams were created with BioRender.
Monocle trajectory analysis
Monocle was used to generate the pseudotime trajectory analysis in brain endothelial and mural cells74. Cells were clustered in Seurat and cluster markers used as input into Monocle to infer arteriovenous relationships within endothelial cells and pericytes. Specifically, UMAP embeddings and cell sub-clusters generated from Seurat were converted to a cell_data_set object using SeuratWrappers (v.0.2.0) and then used as input to perform trajectory graph learning and pseudo-time measurement through Independent Component Analysis (ICA) with Monocle. Cluster marker genes identified in Seurat were used to generate a pseudotime route and plotted using the ‘plot_pseudotime_heatmap’ function.
Cell-cell communication
Cell-cell interactions based on the expression of known ligand-receptor pairs in different cell types were inferred using CellChatDB93 (v.0.02). Briefly, we followed the official workflow and loaded the normalized counts into CellChat and applied the preprocessing functions identifyOverExpressedGenes, identifyOverExpressedInteractions, and projectData with standard parameters set. As database we selected the Secreted Signaling pathways and used the pre-compiled human Protein-Protein-Interactions as a priori network information. For the main analyses the core functions computeCommunProb, computeCommunProbPathway, and aggregateNet were applied using standard parameters and fixed randomization seeds. Finally, to determine the senders and receivers in the network the function netAnalysis_signalingRole was applied on the netP data slot.
Mouse wild-type and APP T41B BEC single-cell and nuclei sequencing
Whole cell isolation from the CNS followed previously described methods20, 40, 58. Briefly, cortices and hippocampi were microdissected, minced, and digested using the Neural Dissociation Kit (Miltenyi). Suspensions were filtered through a 100 µm strainer and myelin removed by centrifugation in 0.9 M sucrose. The remaining myelin-depleted cell suspension was blocked for ten minutes with Fc preblock (CD16/ CD32, BD 553141) on ice and stained for 20 minutes with antibodies to distinguish brain endothelial cells (CD31+/ CD45-). Brain endothelial cells from 12-14 month old Thy1-hAPPLon,Swe mice and littermate wild-type control112 mice (pool of 4-6 mice per group) were sorted into PBS with 0.1% BSA. Nuclei isolation from 4-6 month-old mouse hippocampi followed protocols adapted from previous studies28–30, 52, 149. Briefly, tissue was homogenized using a glass douncer in 2 ml of ice-cold EZ PREP buffer (Sigma, N3408) and incubated on ice for 5 min. Centrifuged nuclei were resuspended in 1% BSA in PBS with 0.2 U μl−1 RNase inhibitor and filtered through a 40 μm cell strainer. Cells or nuclei were immediately counted using a Neubauer haemocytometer and loaded on a Chromium Single-Cell Instrument (10x Genomics, Pleasanton, CA, USA) to generate single-cell GEMs. The 10x-Genomics v3 libraries were prepared as per the manufacturer’s instructions. Libraries were sequenced on an Illumina NextSeq 550 (paired-end; read 1: 28 cycles; i7 index: 8 cycles, i5 index: 0 cycles; read 2: 91 cycles). De-multiplexing was performed using the Cellranger toolkit (v3.0.0) “cellranger mkfastq” command and the “cellranger count” command for alignment to the mouse transcriptome, cell barcode partitioning, collapsing unique-molecular identifier (UMI) to transcripts, and gene-level quantification. ∼70% sequencing saturation (>20,000 reads per cell) was achieved, for a median of ∼2,000 genes detected per cell and ∼16,500 genes detected in total. Downstream analysis using the Seurat package (v3)150 was performed as previously described37, applying standard algorithms for cell filtration, feature selection, and dimensionality reduction. Samples with fewer than 1,000 and more than 4,000 unique feature counts, samples with more than 15% mitochondrial RNA, samples with more than 15% small subunit ribosomal genes (Rps), and counts of more than 10,000 were excluded from the analysis. Genes were projected into principal component (PC) space using the principal component analysis (RunPCA). The first 30 dimensions were used as inputs into Seurat’s FindNeighbors and RunTsne functions. Briefly, a shared-nearest-neighbor graph was constructed based on the Euclidean distance metric in PC space, and cells were clustered using the Louvain method. RunTsne functions with default settings was used to calculate 2-dimensional tSNE coordinates and search for distinct cell populations. Cells and clusters were then visualized using 3-D t-distributed Stochastic Neighbor embedding on the same distance metric. Differential gene expression analysis was done by applying the Model-based Analysis of Single-cell Transcriptomics (MAST). Significant differentially expressed genes in Thy1-hAPPLon,Swe BECs were called by Log (fold change) > 0.15 (absolute value), adjusted P value (Bonferroni correction) < 0.01. This lowered Log (fold change) was to ensure our claims of limited overlap with human AD BECs were robust.
GWAS analysis
For calculation of proportional cell type-specific gene expression, we followed the expression weighted cell type enrichment (EWCE) method described by Skene et al.124, and used previously on human snRNA-seq data29. For Alzheimer’s disease (AD) analysis, we compiled a list of top GWAS risk genes from Lambert et al.117, Kunkle et al.118, and Jansen et al.119, sorted descending by approximate P-value. Each gene’s expression sums to 1 across the cell types, with each heatmap cell showing the fraction of total gene expression as determined from EWCE analysis. The set of 720 AD and AD-related trait GWAS genes were obtained from Grubman, et al.29, and using EWCE analysis, the strongest expressing cell type was determined for each gene. Note that the original list was slightly parsed to 720, as several genes were not detected as expressed in our dataset.
For analysis across CNS diseases, from the GWAS catalog151, we obtained GWAS risk genes for neurological disorders [Alzheimer’s disease (AD), amyotrophic lateral sclerosis (ALS), brain aging, multiple system atrophy (MSA), multiple sclerosis (MS), Parkinson’s disease (PD), and narcolepsy], psychiatric disorders [Attention deficit hyperactivity disorder (ADHD), autism, bipolar disorder, depression, psychosis, post-traumatic stress disorder (PTSD), and schizophrenia], and neurobehavior traits [Anxiety, suicidality, insomnia, neuroticism, risk behavior, intelligence, and cognitive function]. We removed gene duplicates and GWAS loci either not reported or in intergenic regions and used a P < 9 × 10−6 to identify significant associations29. Then, since GWAS signals can point to multiple candidate genes within the same locus, we focused on the ‘Reported Gene(s)’ (genes reported as associated by the authors of each GWAS study). Following gene symbol extraction, we curated the gene set by (1) removing unknown or outdated gene names using the HGNChelper package (v.0.8.6), (2) converting remaining Ensembl gene IDs to actual gene names using the packages ensembldb (v.2.10.0) and EnsDb.Hsapiens.v86 (v.2.99.0), and (3) removing any remaining duplicates. For each disease, we allocated each of its GWAS risk genes to the cell type that proportionally expressed it most (EWCE analysis), before tallying this number in both counts and as a percentage of the diseases total number of GWAS risk genes. Finally, a statistical enrichment of each overlap against background was calculated using a hypergeometric test with the total background size set equal to the number of unique genes (21,306).
Immunohistochemistry
Fresh-frozen control and AD human brain tissue (hippocampus and superior frontal cortex) adjacent to tissue processed for snRNA-seq was subjected to immunohistochemistry (IHC). 10 µm sections mounted on SuperFrost Plus glass slides were fixed with 4% paraformaldehyde (Electron Microscopy Services, 15714S) diluted in PBS at 4°C for 15 minutes before dehydration via an ethanol series or air drying. Sections were blocked in TBS++ (TBS + 3% donkey serum (130787, Jackson ImmunoResearch) + 0.25% Triton X-100 (T8787, Sigma-Aldrich)) for 1.5 hours at room temperature. Sections were incubated with primary antibodies at 4°C overnight: mouse monoclonal anti-CD31 (1:100, JC70A, Dako), rabbit polyclonal anti-VWF (1:100, GA527, Dako), rabbit polyclonal anti-SLC39A10 (1:100, HPA066087, Atlas Antibodies), rabbit polyclonal anti-ALPL (1:100, HPA007105, Atlas Antibodies), rabbit polyclonal anti-A2M (1:100, HPA002265, Atlas Antibodies), rabbit monoclonal anti-β-Amyloid (1:500, clone D54D2 XP, CST), and mouse monoclonal anti-Actin, α-Smooth Muscle - Cy3 (1:100, clone 1A4, Sigma). Sections were washed, stained with Alexa Fluor-conjugated secondary antibodies (1:250) and Hoechst 33342 (1:2000, H3570, Thermo), mounted and coverslipped with ProLong Gold (Life Technologies) or VECTASHIELD (Vector Laboratories before imaging on a confocal laser scanning microscope (Zeiss LSM880). Age-related autofluorescence was quenched prior to mounting with Sudan Black B, as before15, 58. National Institutes of Health ImageJ software was used to quantify the percentage of vasculature (CD31) or the predicted DEG SLC39A10 among CD31+ vasculature, following previously described protocols15, 152, 153. All analyses were performed by a blinded observer.
Data Availability
Raw sequencing data is deposited under NCBI GEO: GSE163577. Data is also available to explore via an interactive web browser: https://twc-stanford.shinyapps.io/human_bbb.
Supplemental Figures
Acknowledgments
We thank T. Iram, E. Tapp, N. Lu, M. Haney, O. Hahn, M.J. Estrada, and other members of the Wyss-Coray lab for feedback and support; Hansruedi Mathys, David A. Bennett, and participants in the CSHL BBB 2021 meeting for valuable advice; and H. Zhang and K. Dickey for laboratory management. This work was funded by the NOMIS Foundation (T.W.-C.), the National Institute on Aging (T32-AG0047126 to A.C.Y., 1RF1AG059694 to T.W.-C), Nan Fung Life Sciences (T.W.-C.), the Bertarelli Brain Rejuvenation Sequencing Cluster (an initiative of the Stanford Wu Tsai Neurosciences Institute), and the Stanford Alzheimer’s Disease Research Center (P30 AG066515). A.C.Y was supported by a Siebel Scholarship. F.K. and A.K. are a part of the CORSAAR study supported by the State of Saarland, the Saarland University, and the Rolf M. Schwiete Stiftung. This study was supported by the AHA-Allen Initiative in Brain Health and Cognitive Impairment: 19PABHI34580007. The statements in this work are solely the responsibility of the authors and do not necessarily represent the views of the American Heart Association (AHA) or the Paul G. Allen Frontiers Group.
References
- 1.↵
- 2.↵
- 3.
- 4.↵
- 5.↵
- 6.↵
- 7.↵
- 8.↵
- 9.
- 10.↵
- 11.↵
- 12.↵
- 13.
- 14.↵
- 15.↵
- 16.↵
- 17.↵
- 18.↵
- 19.↵
- 20.↵
- 21.↵
- 22.↵
- 23.↵
- 24.↵
- 25.↵
- 26.↵
- 27.↵
- 28.↵
- 29.↵
- 30.↵
- 31.
- 32.↵
- 33.↵
- 34.↵
- 35.↵
- 36.↵
- 37.↵
- 38.
- 39.
- 40.↵
- 41.↵
- 42.↵
- 43.↵
- 44.↵
- 45.↵
- 46.
- 47.↵
- 48.↵
- 49.
- 50.
- 51.
- 52.↵
- 53.↵
- 54.↵
- 55.↵
- 56.↵
- 57.↵
- 58.↵
- 59.↵
- 60.↵
- 61.↵
- 62.
- 63.
- 64.↵
- 65.↵
- 66.↵
- 67.↵
- 68.↵
- 69.↵
- 70.↵
- 71.↵
- 72.↵
- 73.↵
- 74.↵
- 75.↵
- 76.↵
- 77.↵
- 78.↵
- 79.↵
- 80.↵
- 81.↵
- 82.↵
- 83.↵
- 84.↵
- 85.↵
- 86.
- 87.
- 88.↵
- 89.↵
- 90.↵
- 91.↵
- 92.↵
- 93.↵
- 94.↵
- 95.↵
- 96.↵
- 97.↵
- 98.↵
- 99.↵
- 100.
- 101.↵
- 102.↵
- 103.↵
- 104.↵
- 105.↵
- 106.↵
- 107.↵
- 108.
- 109.
- 110.↵
- 111.↵
- 112.↵
- 113.↵
- 114.↵
- 115.↵
- 116.↵
- 117.↵
- 118.↵
- 119.↵
- 120.
- 121.↵
- 122.↵
- 123.↵
- 124.↵
- 125.↵
- 126.↵
- 127.↵
- 128.↵
- 129.↵
- 130.↵
- 131.↵
- 132.↵
- 133.↵
- 134.↵
- 135.↵
- 136.
- 137.↵
- 138.↵
- 139.↵
- 140.↵
- 141.↵
- 142.↵
- 143.↵
- 144.↵
- 145.↵
- 146.↵
- 147.↵
- 148.↵
- 149.↵
- 150.↵
- 151.↵
- 152.↵
- 153.↵
- 154.↵
- 155.↵