Recycling old drugs: cardiac glycosides as therapeutics to target barrier inflammation of the vasculature, meninges and choroid plexus

Neuroinflammation is a key component of virtually all neurodegenerative diseases; preceding neuronal loss and associating directly with cognitive impairment. Neuroinflammatory signals can originate and be amplified at barrier tissues such as brain vasculature, surrounding meninges and the choroid plexus. We designed a high-throughput screening system to target inflammation in cells of the blood-brain barrier (primary human pericytes and endothelia) and microglia enabling us to target human disease-specific inflammatory modifiers. Screening an FDA-approved drug library we identified digoxin and lanatoside C, members of the cardiac glycoside family as inflammatory modulating drugs that work in blood-brain barrier cells. A novel ex vivo assay of leptomeningeal and choroid plexus explants further confirmed that these drugs maintain their function in 3D cultures of brain border tissues. While current therapeutic strategies for the treatment of neurodegenerative diseases are missing the mark in terms of targets, efficacy and translatability, our innovative approach using in vitro and ex vivo human barrier cells and tissues to target neuroinflammatory pathways is a step forward in drug development and testing, and brings us closer to translatable treatments for human neurodegenerative disease. One Sentence Summary We have identified cardiac glycosides as powerful regulators of neuroinflammatory pathways in brain-barrier tissues such as vasculature, meninges and choroid plexus.

Neuroinflammation can be most easily targeted anatomically at several sites; the 68 cerebrovasculature that reaches the entirety of the CNS bordered luminally by the blood, and 69 abluminally by the cerebrospinal fluid (CSF), the meninges which surround and envelop the 70 brain and spinal cord directly communicating with lymphatic vessels, and the CP that 71 produces CSF and facilitates fluid movement by way of glymphatic flux (11-13). It has been 72 suggested that the brain's immune functions have been concentrated at these border tissues, 73 which may be more sensitive to inflammatory cues than the rest of the CNS (14-16). We have 74 designed our study to test inflammatory modulating drugs on the spectrum of cell types that 75 make up these structures. 76 At the cellular level the neurovascular unit (NVU) is formed by the specialized arrangement 77 of endothelia, pericytes, astrocyte endfeet, and neurons that work together to maintain brain 78 homeostasis. We have demonstrated that pericytes derived from post-mortem human brains 79 display a robust inflammatory response when stimulated by classical inflammatory mediators 80 such as tumor necrosis factor α, interferon γ (IFNγ), interleukin 1β (IL-1β) and 81 lipopolysaccharide (LPS), and are likely to be targets of neuroinflammation in both an 82 autocrine and paracrine fashion (17,18). Importantly, endothelia and pericytes in the NVU 83 contribute to neuroinflammatory responses by chemokine secretion, adhesion molecule-84 mediated leucocyte extravasation, and damage-associated molecular patterns (DAMPS) and 85 pathogen associated molecular patterns (PAMPS) receptor expression (11,(19)(20)(21)(22). 86 Attenuating the infiltration of peripheral immune cells into the brain or preventing 87 inflammatory propagation by barrier cells such as neurovascular cells is likely to be beneficial 88 in limiting CNS immune responses, although few CNS therapies have been based upon this 89 approach. Namely the MS drug natalizumab that targets immune cell entry into the CNS is a 90 good example of this approach (23). In this study in addition to primary human brain 91 neurovascular cells such as pericytes, endothelia cells and microglia, we also made use of 92 explants from barrier tissues such as meninges and CP which are directly involved in 93 inflammation. 94 The human meninges are composed of three membranes (dura, arachnoid, and pia mater) 95 that envelop the brain and spinal cord. The arachnoid and pia mater, collectively termed the 96 leptomeninges, form a semi-permeable membrane to the CSF which fills the subarachnoid 97 space. The leptomeninges are comprised of several cell types, including macrophages, 98 dendritic cells, mast cells, and fibroblasts and are permeated by leptomeningeal arteries (24, 99 25). Furthermore, diverse leucocyte populations reside in the subarachnoid space providing 100 a conduit for immune-brain communication (26-28). Meningeal inflammation often precedes 101 inflammation in the CNS and is present in neurodegenerative diseases, chronic inflammatory 102 conditions, and acute pathogen introduction (29-31). Importantly, meningeal-derived factors 103 can permeate the brain parenchyma, presumably through glymphatic influx, suggesting that 104 modulating meningeal-derived inflammation also represents an appropriate target to prevent 105 inflammatory-mediated CNS insults (32, 33). 106 The CP is a highly vascularised tissue that harbours fenestrated capillaries surrounded by 107 apical facing, CSF-producing epithelial cells and serves as a gateway for immune cells from 108 the circulatory system to the CNS (34). The fact that the CP is a sink for immune cells is of 109 particular importance when it comes to immune cell trafficking in both sterile inflammatory 110 and chronic inflammatory conditions (35,36). Similarly, the CP itself increases secretion of 111 inflammatory messenger molecules into the CSF with age and disease, which are transported 112 throughout the brain through CSF circulation (37, 38). Together this supports the examination 113 of the CP as a target for reducing neuroinflammation. 114 Here, we utilise primary dissociated human brain pericytes and endothelia, as well as explants 115 from the meninges and the CP to screen drugs for anti-inflammatory properties. Drug 116 screening approaches are not typically performed using primary human brain cells due to 117 limited tissue yields, difficulties of human brain cell culture and accessibility. However, rodent 118 immune cells display numerous discrepancies compared to their human counterparts, with 119 respect to both immune functions and pharmacological responses (32, 39), necessitating the 120 use of primary human brain cells for pre-clinical compound screening with an intent for 121 translational drug discovery. Primary human brain pericytes are suitable for high throughput 122 drug screening for candidate compounds with anti-inflammatory functions because unlike 123 many other human brain cell types, pericytes undergo rapid proliferation in vitro, allowing for 124 efficient bulking of these cells for the purposes of high-throughput drug screening. In this 125 study, selected hits from initial screens were validated in both pericyte and endothelial 126 cultures to explore the anti-inflammatory efficacy of these compounds in NVU-associated 127 cells. Next we describe and characterise the inflammatory contribution of an ex vivo model of 128 human leptomeningeal and CP explants using it to investigate lead anti-inflammatory 129 compounds in a complex multicellular system more closely recapitulating the in vivo human 130 brain environment. Finally, we demonstrate the efficacy of two anti-inflammatory 131 compounds digoxin and lanatoside C, in attenuating meningeal inflammatory responses and 132 preventing meningeal-mediated inflammatory propagation. 133 134

Results 135
Identifying cardiac glycosides in a screen using primary pericytes 136 Primary human brain pericytes respond to inflammatory stimuli to induce the expression of 137 chemokines and adhesion molecules, including chemokine (C-C motif) ligand 2 (CCL2) and 138 intercellular adhesion molecule-1 (ICAM-1), respectively (17, 40-42). Due to their involvement 139 in enhancing leucocyte infiltration into the brain, these mediators were selected as candidate 140 proteins to determine the anti-inflammatory potential of the Prestwick compound library 141 (17). Concentration response curves of IL-1β-induced CCL2 and ICAM-1 protein expression, 142 as determined by immunocytochemistry, revealed an EC50 of 0.02 ng/mL for CCL2 and 0.03 143 ng/mL for ICAM-1, thus 0.05 ng/mL was selected as an optimal concentration of IL-1β to 144 determine immune modulation, allowing for identification of compounds which may induce 145 and attenuate inflammatory responses ( Fig. 1A  anti-septic, anti-neoplastic, anti-amoebic, and anti-inflammatory ( Fig. 1E and F). A second 159 confirmatory screen of the 82 hits generated 44 that had reproducible effects. These 44 hits 160 were then screened at three concentrations (0.1, 1, and 10 µM) for the ability to modify IL-161 1β-induced CCL2 or ICAM-1 in pericytes, without causing significant cell loss (Trial 3, fig S1). 162 IL-1β-independent effects were assessed by screening compounds in pericytes with the same 163 paradigm (24 hours of compounds at 10 µM, 24 hours of vehicle only for IL-1β), which 164 revealed no significant changes relating to CCL2 or ICAM-1 expression ( fig. S1). 165 Hits were narrowed down to 10 compounds that consistently modified either IL-1β-induced 166 CCL2 or ICAM-1 expression by immunocytochemistry without significant cell reduction (as 167 assessed by total cell counts). To further investigate the anti-inflammatory ability of hit 168 compounds, their effect on secretions of a larger panel of inflammatory chemokines, 169 cytokines, and adhesion molecules previously identified to be secreted by pericytes in 170 response to IL-1β was determined by cytometric bead array (CBA) (43). Concentrations 171 determined from Trial 3 were used to treat pericytes as above ( fig. S1). Analysis of cytokine 172 secretion showed that both cardiac glycosides digoxin and lanatoside C had the most 173 substantial effects on soluble ICAM-1 (sICAM-1) and fractalkine (CX3CL1) (a known microglial 174 ligand) (44). In particular, digoxin inhibited IL-1β-induced secretion of CCL2, sICAM-1, soluble 175 vascular cell adhesion molecule-1 (sVCAM-1), and CX3CL1 but increased secretion of 176 interleukin-6 (IL-6) more than 6-fold over IL-1β alone (Fig. 1G, fig S2). A generalized pipeline 177 for refinement of the hit compound list leading to digoxin as a lead compound is presented in 178 Fig. 1H. Digoxin is indicated for the treatment of symptomatic heart failure, and atrial 179 fibrillation, thus we were surprised to see anti-inflammatory effects on vascular cells. 180 Lanatoside C, another cardiac glycoside found in the Prestwick library, demonstrates 181 structural similarity with digoxin (45). Little is known concerning inflammatory outputs by 182 cardiac glycosides on brain vascular cells, and due to the ability of digoxin to most effectively 183 limit pericyte-mediated inflammatory responses, digoxin and the closely-related lanatoside C 184 where chosen as lead compounds for further analysis. 185 Anti-inflammatory effects of digoxin and lanatoside C in primary human brain pericytes 186 Pericytes express several proteins that facilitate leukocyte extravasation and polarise 187 surrounding immunologically-active cells to pro-or anti-inflammatory phenotypes (41, 46). 188 Secretion data from pericytes pre-treated with digoxin and lanatoside C were consistent with 189 an inhibitory effect on IL-1β-induced soluble adhesion molecule (sICAM-1 and sVCAM-1) and 190 CX3CL1 secretion, whilst IL-6 was increased and interkeukin-8 (IL-8) displayed no change ( these compounds may be acting on multiple pathways to reduce inflammatory responses. 223 224 In mixed glial cultures, while neither digoxin or lanatoside C reduced IL-1β-induced NFκB 225 translocation in microglia ( Fig. 3A to D), cytokine secretions were consistent with pericyte 226 cultures ( Fig. 3E to M). This is most likely due to the low percentage of microglia present in 227 the predominantly pericyte cultures at early passage (7.261 %, ± 4.709 PU.1 positive cells). 228 This suggests that digoxin and lanatoside C do not modulate NFκB translocation-dependent 229 inflammatory effects in human microglia. 230 231 Anti-inflammatory effects of digoxin and lanatoside C in primary human brain endothelia 232 Brain endothelia lining the cerebral blood vessels are the first point of contact for therapeutic 233 drugs in the systemic circulation and express numerous active efflux transporters which 234 complicates CNS drug delivery. Additionally, brain endothelia play an important role in 235 neuroinflammatory responses generated from both the periphery and the brain parenchyma 236 by mediating recruitment, adhesion, and transcellular/paracellular immune cell trafficking 237 across the BBB through expression of cellular adhesion molecules (ICAM-1, VCAM-1) and 238 chemokines (IL-8, CCL2) (50-52). As p-glycoprotein substrates, an intact BBB would likely 239 exclude digoxin and lanatoside C entrance to parenchymal brain cells, although this barrier is 240 often weakened during inflammatory states or neurological diseases and could facilitate 241 cerebral penetrance. In order to further assess anti-inflammatory effects of digoxin and 242 lanatoside C, particularly with respect to leucocyte extravasation in the cell type most likely 243 to be exposed in vivo, inflammatory responses of primary adult human brain endothelia were 244 assessed using the aforementioned paradigm described for pericytes. 245 Further, inflammatory secretions in the subarachnoid space in the ventricular space by the CP 267 can penetrate the brain to alter cerebral functioning, likely as a consequence of glymphatic 268 influx, whereby neurovascular cells including pericytes, endothelia, and astrocytes will also 269 be subjected to cytokine presence (13, 60). In order to further investigate potential anti-270 inflammatory functions of digoxin and lanatoside C, we first established a method allowing 271 for the ex vivo cultures of human leptomeninges and CP to study inflammation, and potential 272 drug interventions in a culture system more appropriately recapitulating the in vivo 273 microenvironment than can be achieved by standard in vitro cultures. CD68 in both LME and CPE after more than one month in culture ( fig S6). Whilst the 284 inflammatory secretome and cytokine-specific responses of brain pericytes and endothelia 285 have been studied to some extent, significantly less is known regarding this response in the 286 leptomeninges. Secretions from LME in response LPS, IL-1β, and IFNγ closely mirrored what 287 we saw in pericytes and endothelia with adhesion molecule and chemokines responsible for 288 immune cell recruitment largely being induced by IL-1β, and LPS, and a lesser response with 289 IFNγ (22) (Fig 5B to G, fig S7A). Thus, we used cytokine profiler arrays, which allows for 290 simultaneous detection of 102 different soluble inflammatory mediators in response to the 291 aforementioned immunogenic stimuli (LPS, IL-1β, and IFNγ). As a likely reflection of their cell 292 heterogeneity, the LME demonstrated an extensive basal secretion of inflammatory 293 mediators which was differentially induced by LPS, IL-1β, and IFNγ (Fig. 5H, fig S8A). 294 Importantly, these mediators had no effect on cell viability ( fig. S5F). Notably basal detection 295 of several immune mediators was observed from the LME, in particular angiogenin, 296 angiopoetin-2, chitinase 3-like, C-X-C motif chemokine 5 (CXCL5), growth/differentiation 297 factor 15 (GDF-15), insulin-like growth factor protein-2 (IGFBP-2), IL-8, osteopontin, and high 298 basal levels of serpin E1. The most drastic change in secretion in response to cytokine 299 treatment was interferon γ-induced protein-10 (IP-10/CXCL10), which is induced by both IFNγ 300 and LPS, as well as GM-CSF -by IL-1β, and IL-6 by IFNγ and IL-1β. 301 We sought to investigate whether digoxin and lanatoside C could reduce this pro-302 inflammatory response. As was observed for pericytes and endothelia, digoxin and lanatoside 303 C reduced IL-1β-induced inflammatory mediators in the leptomeninges including CCL2, 304 sICAM-1, sVCAM-1 and CX3CL1 (Fig 6A-I, fig S7B). Using the proteome profilers to interrogate 305 the extent of inhibition by digoxin or lanatoside C we found that both drugs decreased IL-1β 306 -induced secretion of G-CSF, GM-CSF, C-X-C motif chemokine 11 (CXCL11), platelet-derived 307 growth factor AA (PDGF-AA), and sVCAM-1, and they increased secretion of IL-1β dependent 308 IL-6, macrophage inflammatory protein 1-alpha (MIP1α) and TNFα consistently across three 309 LME cases (Fig 6J, fig S7B). 310 Using the same paradigm in choroid plexus explants (CPE) we first examined the cellular 311 composition and found the highly vascularized tissue to be harbouring similar cell types to 312 the LME, with the addition of transthyretin-positive epithelia (Fig 7A and B). Similar to the 313 LME, the CPE secreted a range of cytokines in response to pro-inflammatory stimulation albeit 314 to a lesser extent than the LME (Fig 7C-L, fig S8 and S9). Just like the LME under basal 315 conditions, the CPE had undetectable levels of G-CSF, and GM-CSF, very low expression of 316 RANTES, sICAM-1, and noticeable expression of chitinase 3-like, GDF-15, IGFBP-2, IL-8, 317 osteopontin and serpin E1. Expectedly CPE cultures responded to cytokine treatment by 318 changing their secretome. Digoxin and lanatoside C pre-treatment resulted in reduced IL-1β-319 responses such as inhibition of sICAM-1, sVCAM-1 and CX3CL1 secretion by CBA (Fig 8 A-I). 320 Proteome profilers of CPE conditioned media in contrast to LME show an increase in IL-1β 321 induced secretion of G-CSF, GM-CSF, and CXCL5, but reduced VCAM-1, angiogenin, 322 osteopontin, and interleukin 1 receptor-like 1 (ST2) by digoxin and lanatoside (Fig 8). 323

Discussion 324
Neuroinflammation is present in almost every neurological disorder and contributes to 325 disease pathogenesis by precipitating neuronal loss and BBB dysfunction (62, 63). As such, 326 the identification of therapeutic compounds that effectively attenuate neuroinflammation 327 has the potential to be beneficial in a diverse range of neurodegenerative diseases, 328 neuropsychiatric disorders, and acute brain injuries. Unfortunately, compounds that were 329 beneficial in preclinical models of neurological disease, including anti-inflammatory therapies, 330 have largely failed to effectively translate to clinical use (64, 65). Whilst the failure of anti-331 inflammatory interventions may be a function of inappropriate timing in the treatment 332 regime, the lack of translational therapeutics could be equally attributed to the frequent use 333 of rodent models in neurological drug discovery, in which responses may poorly reflect those 334 observed in humans. Failure may also be associated with targeting of general inflammatory 335 pathways irrelevant to neurodegenerative disease, suggesting that targeting distinct CNS 336 neuroinflammation pathways may prove more successful. 337 To address this issue, we utilised primary human brain pericytes to screen for anti-338 inflammatory drugs in a library consisting of >1280 FDA approved compounds. These cells 339 have the obvious benefit of being of human origin, therefore negating any species differences 340 which complicate observed responses. Additionally, unlike human microglia or astrocytes, 341 these cells are highly proliferative in vitro, allowing for efficient bulking and bio-banking of 342 these cultures to generate sufficient yields to undertake large screens (66). Due to its 343 ubiquitous presence in neurological disease, we sought to target neuroinflammation in this 344 screen in particular because together, endothelia and pericytes are the major mediators of 345 leucocyte extravasation during neurodegenerative diseases, particularly in AD and MS, 346 through CCL2-directed chemotaxis and ICAM-1 and VCAM-1-mediated adhesion (52, 67-69). 347 However, this approach would be equally suited to identify mitogenic enhancers, autophagy 348 regulators, phagocytic inducers or pro-survival drugs, utilising differing stimulation 349 paradigms. 350 The strength of the current screening approach resides in the fact that compounds underwent 351 several rounds of analysis in primary human brain cells derived from diverse healthy and 352 diseased post mortem and biopsy samples. The drug effects were consistent across different 353 donors, emphasizing the capacity of this screening platform as a tool to identify modifiers of 354 human-specific targets taking into account the inherent human variation from the first stages 355 of drug discovery. With recent advancements in the ability to isolate and culture cells from 356 the human brain, in addition to deriving these cells from patient induced pluripotent stem 357 cells, these screening approaches can be expanded to human microglia, astrocytes, and even 358 neurons (70-72). 359 Whilst the use of non-human or immortalized cell lines has potential issues with such high-360 throughput screens, so too does the use of in vitro primary cultures themselves. Pericytes and 361 endothelia utilised here were subjected to several rounds of passaging in order to obtain 362 sufficient yields, for which phenotypic drift can occur (73, 74). It is therefore unclear whether 363 cells accurately reflect the phenotype observed in vivo. In order to analyse the anti-364 inflammatory effects of digoxin and lanatoside C in a more complex, multi-cellular model 365 without the complications of dissociated and passaged cells , a protocol to simply and 366 effectively isolate leptomeningeal and choroid plexus explant cultures was used (75). These 367 explants display excellent viability, they are easily accessible from the post-mortem brain, or 368 from neurosurgical specimens and their ease of handling makes them an attractive model. 369 The advancements in organotypic slice cultures could soon allow for coculture methods of 370 LME and CPE with human neurons to define cellular interactions in disease conditions (76). 371 The explants displayed an extensive inflammatory secretome, including cytokines, 372 chemokines, and adhesion molecules. Of equal importance, the anti-inflammatory effects 373 observed using in vitro monocultures of pericytes and endothelial cells were largely consistent 374 with our ex vivo leptomeningeal and CPE cultures, suggesting the appropriateness of both 375 model systems in identifying immune modulating compounds. The ability of these 376 compounds to attenuate inflammatory responses in several distinct cell types is promising 377 with the ability to effectively attenuate global neuroinflammation, as this response is not 378 restricted to one particular region. Furthermore, neither digoxin nor lanatoside C completely 379 blocked inflammatory responses, as can be deleterious, with a basal level of inflammatory 380 capabilities important for homeostasis and defence responses (77). 381 The utilisation of an FDA-approved compound library is advantageous as compounds have 382 already undergone safety trials in humans, easing the repurposing of drugs for other 383 disorders. As such, potential information about efficacy of such compounds can often be 384 gleaned from retrospective analysis of patient cohorts taking these compounds for other 385 indications, therefore expensive and timely clinical human safety trials are not necessary. 386 Further, extensive data are already available on their pharmacological properties (45, 72, 78). 387 Digoxin and lanatoside C are structurally very similar and well-known for their inhibitory 388 effects on the Na + /K + -ATPase pump (79). Both act similarly in cancer cell lines to inhibit 389 proliferation having actions on tumour necrosis factor-related apoptosis-inducing ligand 390 (TRAIL), Src and protein kinase C δ (PKCδ) pathways. Digoxin is approximately three times 391 more potent than lanatoside C, which is consistent with the data presented here (45, 80-82). 392 Currently, evidence for digoxin and lanatoside C in anti-inflammatory mechanisms is limited, 393 with suggested effects on leukocyte infiltration and extravasation through inhibition of 394 cytokine/chemokine secretion and reduced NFκB expression (83). Although the efficacy of 395 other cardiac glycosides as anti-inflammatory compounds in brain barrier tissues has not been 396 investigated, cardiac glycosides demonstrate a large range of chemical diversity and 397 absorption, distribution, metabolism, elimination and toxicity (ADMET) properties (79). 398 Nonetheless, we have shown that these compounds can act on multiple inflammatory 399 pathways in brain barrier cells including pericytes, endothelia, microglia, and in meningeal 400 and choroid plexus explants. Therefore they have strong potential as anti-inflammatory drugs 401 in the context of neuroinflammation. 402 Neurodegenerative diseases are often a result of several diverging dysregulated pathways, 403 including inflammatory responses, protein aggregation, BBB breakdown, and inappropriate 404 clearance mechanisms. As such, the identification of multi-target, multi-action drugs 405 (targeting both neuroinflammatory pathways and blood-brain barrier cells) will prove more 406 effective than single target therapeutics (7). Moreover, accessibility to vascular cells via 407 perivascular spaces, and tissues such as the meninges and CP through the brain boundary 408 regions and CSF compartments should be considered in drug delivery strategies. Here we 409 demonstrate the utility of repurposed drug screens in human brain cells and identify drugs 410 for use as therapeutic agents in attacking the neurodegenerative cascade in cells and tissues 411 that can be more easily reached. Taken together, our described pipeline represents a 412 promising approach for neuroinflammatory and neurodegenerative drug development. 413

Study Design 415
This study was designed to identify chemical compounds that could modulate inflammatory 416 responses in human brain vascular cells. Using post-mortem brain tissue of healthy and 417 disease patients, as well as epilepsy biopsy tissue we screened 1280 FDA-approved 418 compounds against an IL-1β-induced inflammatory response. Candidates identified from 419 the initial screen by CCL2 and ICAM-1 expression in pericytes were forwarded for secretion 420 analysis, and transcription factor activity in pericytes, endothelial cells, and mixed glial 421 cultures. Meningeal and choroid plexus explants were tested as an ex vivo model of 422 neuroinflammation. 423

Tissue source 424
Biopsy human brain tissue was obtained from the Neurological Foundation Human Brain 425 Bank, with informed written consent, from the middle temporal gyrus (MTG) of patients 426 undergoing surgery for drug-resistant epilepsy. Post-mortem leptomeninges were obtained 427 from regions overlying the MTG, and CPE were derived from the lateral ventricle from one 428 hemisphere from neurologically normal individuals, or those with various neurological 429 diseases (Table S1). All brain tissue collection and processing protocols were approved by the 430 Northern Regional Ethics Committee, (AKL/88/025/AM09 New Zealand) for biopsy tissue, and 431 the University of Auckland Human Participants Ethics Committee (Ref # 011654, New Zealand) 432 for the post-mortem brain tissue. All methods were carried out in accordance with the 433 approved guidelines. 434

Primary human brain mixed glial cultures 435
Biopsy human brain tissue was obtained from the MTG of patients with drug-resistant 436 epilepsy and mixed glial cultures, containing microglia, astrocytes, and pericytes, were 437 generated as described previously (71). Mixed glial cultures were maintained in complete 438 media (DMEM/F12 with 10% FBS and 1% PSG (penicillin 100 U/mL, streptomycin 100 µg/mL, 439 L-glutamine 0.29 mg/mL)) at 37 °C with 5% CO2 until confluent. Flasks were trypsinized with 440 0.25% Trypsin-1mM EDTA and scraped to detach firmly adherent microglia. Viable cells were 441 counted based on trypan blue exclusion and 5,000 cells/well were seeded into 96-well plates 442 in complete media and used for experimentation after 1-3 days. All experiments performed 443 on mixed glial cultures were at passage two. 444

Primary human brain pericyte culture 445
To generate pure pericyte cultures, mixed glial cultures were sub-cultured up to passage four 446 in order to eliminate non-proliferative microglia and astrocytes as described previously (71,447 84). Pericyte cultures were maintained in complete media at 37 °C with 5% CO2. Viable cells 448 were counted based on trypan blue exclusion and 5,000 cells/well were seeded into 96-well 449 plates in complete media and used for experimentation after 1-3 days. All experiments 450 performed on pericytes were at passages 4-9. 451 Primary human brain endothelial culture 452 Primary human endothelial cells were isolated from brain microvessels as described 453 previously ( were then washed once more with 3% BSA in PBS and imaged as described above. 523

Primary human leptomeningeal and choroid plexus explant culture 524
Leptomeninges were removed by gross dissection from the brain overlying the MTG, from 525 autopsy tissue of neurologically normal or pathologically confirmed cases of 526 neurodegenerative disorders (AD, PD, HD, and FTD), as previously described (86). 527 Leptomeningeal tissue was washed in complete media in sterile petri dishes and dissected 528 into pieces ~ 2 mm 3 . Leptomeningeal explants were placed in individual wells in 500 μL of 529 complete media in a 24-well plate allowing them to remain in suspension. Media changes 530 were performed twice a week. Explants that failed to alter media acidification, as determined 531 by a phenol red colour change, were deemed non-viable and discarded. Explants were 532 cultured for at least a week before using for experiments to allow equilibration but remained 533 viable for up to four months after initial isolation. For functional studies, individual explants 534 were placed in 100 μL of media within a 96-well plate. 535

Explant viability and histological processing 536
Cell death was determined using the ReadyProbes™ cell viability imaging kit (Invitrogen). 537 NucBlue live reagent and NucGreen dead reagent stains were diluted into media (2 drops per 538 mL blue, 1 drop per mL green), and added to cells in culture 30 minutes prior to endpoint and 539 incubated at 37 o C for 30 minutes. Z-stacks (200-250 slices, 5 μm step) of explants were then 540 acquired using the ImageXpress high content imaging system (Molecular Devices). 2-D 541 projections were used for analysis of positively labelled nuclei with the Custom Module Editor 542 (image analysis pipeline described in detail in supplementary data ( fig. S5). 543 Immunohistochemical staining on formalin fixed, paraffin embedded explants, sectioned at 544 7µm was done as previously described (87). Antigen retrieval with TRIS-EDTA (pH 9) was 545 performed, followed by blocking with 10% donkey serum prior to antibody incubation. 546 Sections were incubated with primary antibodies (for dilutions see Table S2) overnight at 4 547 o C, rinsed in PBS then incubated with secondary antibodies and Hoechst 33258 (SIGMA) for 3 548 hours at room temperature in the dark. Sections were imaged using the 20X objective on the 549 Metasystems V-slide scanning microscope and stitched to generate images of whole explants. 550 Control sections where the primary antibody was omitted showed no immunoreactivity. The 551 control experiments showed that the secondary antibodies did not cross-react with each 552 other. All confocal recordings were done using an FV1000 confocal microscope (Olympus) 553 with a 40X oil immersion lens (NA 1.00). 554

Statistical Analysis 555
All cell culture experiments were performed three separate times with cells from different 556 donors. Data were normalized to vehicle controls were indicated in figure legends. Statistical 557 test were performed using Graphpad Prism software, one-way, or two-way ANOVA with 558 Tukey's post hoc analysis. Data are presented as the mean ± SEM. 559