Endothelial Leakiness Induced by Amyloid Protein Aggregation

Alzheimer’s disease (AD) is a major cause of dementia debilitating the global ageing population. Current understanding of the AD pathophysiology implicates the aggregation of amyloid beta (Aβ) as causative to neurodegeneration, with tauopathies and neuroinammation considered as other major culprits. Curiously, vascular endothelial barrier dysfunction is strongly associated with Aβ deposition and 80-90% AD subjects also experience cerebral amyloid angiopathy. Here we show amyloid proteins-induced endothelial leakiness (APEL) in human microvascular endothelial monolayers as well as in mouse cerebral vasculature. Using signaling pathway assays and discrete molecular dynamics, we revealed that the angiopathy rst arose from a disruption to vascular endothelial (VE)-cadherin junctions exposed to the nanoparticulates of Aβ oligomers and seeds, preceding the earlier implicated proinammatory and pro-oxidative stressors to endothelial leakiness. These ndings are analogous to nanomaterials-induced endothelial leakiness (NanoEL), a major phenomenon in nanomedicine depicting the paracellular transport of anionic inorganic nanoparticles in the vasculature. As APEL also occurred with the anionic seeds of pathogenic alpha synuclein and functional FapC bacterial amyloid, this study projects a general new paradigm for elucidating the vascular permeation, systemic spread, and cross-seeding of amyloid proteins that underlie the pathogeneses of AD, Parkinson’s, as well as a range of amyloid diseases. monolayers, blood-brain barrier and biopsied and rodent connection with the production of reactive oxygen species (ROS), modied cytoskeletal network, altered tight-junction protein expression, and signaling to pericytes. In this study, we report on amyloid proteins-induced endothelial leakiness (APEL) in human microvascular endothelial cell (HMVEC) monolayers resulting from their exposure to the nanoparticulates of oligomers, protobrils, and sonicated amyloid seeds of Aβ 42 (abbreviated as “seeds” and “Aβ” hereafter), Parkinson’s alpha synuclein (abbreviated as “αS” hereafter) and FapC, a major protein constituent of the extracellular functional amyloid matrix of Pseudomonas aeruginosa. We further demonstrated endothelial leakiness in vivo with mouse cerebral vasculature exposed to Aβ oligomers and seeds. Using signaling pathway assays and steered discrete molecular dynamics (sDMD) simulations, we revealed that APEL arose from the anionic protein nanoparticulates interacting with vascular endothelial (VE)-cadherins and their associated adherens junction machinery. Our in vitro data showed that APEL arose from non-ROS and non-apoptotic events, where Aβ oligomers and seeds instead underwent direct extracellular interactions with VE-cadherins to trigger molecular pathways yielding intercellular gap formation. Through the sDMD simulations, we found that the oligomers remarkably reduced the cadherin dimer stability followed by the amyloid seeds, while exible amyloid monomers did not affect the cadherin integrity. These results converged on a general new phenomenon that the nanoparticulates of anionic peptides and proteins rendered through molecular self-assembly and fragmentation, can remodel the vasculature endothelium preceding the onset of ROS, inammation, cytotoxicity, and cerebral blood ow constriction as suggested by earlier etiological paradigms .


Introduction
Alzheimer's disease (AD) is a primary form of neurological disorder entailing severe adverse effects on the memory, cognition, and life expectancy of the global ageing population. The most in uential paradigm concerning the AD pathology is the amyloid cascade hypothesis proposed by Hardy and Higgins in 1992 1 and its modi cations thereafter, where amyloid beta (Aβ) evolves from disordered monomers to toxic oligomers and amyloid brils through molecular self-assembly, modulated by environmental factors such as pH, temperature, metals, chaperones, and cell membranes 2 . Accordingly, much effort over the past three decades has been made towards inhibiting or clearing the toxic Aβ aggregates, employing small molecules, peptidomimetics, antibodies and, more recently, nanoparticles [3][4][5][6][7][8][9][10][11][12] . A lack of clinical success, however, has shrouded these efforts suggesting the pathophysiology of AD is multifactorial as its triggers 13 .
Indeed, it has now been realized that, in addition to Aβ amyloidogenesis, tauopathies, apolipoprotein E, and neuroimmune activation are all causative to neurodegeneration in AD 13 . The great (80-90%) correlation between AD subjects and patients carrying cerebral amyloid angiopathy (CAA) further suggests an important role of endothelial integrity in the development of AD pathogenesis 14 , also evidenced by observations of cerebral endothelial dysfunction and microvascular injury induced by Aβ [15][16][17] . Intriguingly, while Aβ originates from the proteolytic cleavage of amyloid precursor protein (APP) in endosomal membrane 18 , deposits of Aβ are seen throughout the central nervous system, cerebral blood vessels, cerebrospinal uid, and the plasma 14,19,20 . Aβ-mediated vasoactivity, vascular capillary constriction, blood ow reduction, and paracellular transport have been reported with endothelial monolayers, blood-brain barrier (BBB), and biopsied human and rodent brain tissues 15,17,21−23 , in connection with the production of reactive oxygen species (ROS), modi ed cytoskeletal network, altered tight-junction protein expression, and signaling to pericytes.
In this study, we report on amyloid proteins-induced endothelial leakiness (APEL) in human microvascular endothelial cell (HMVEC) monolayers resulting from their exposure to the nanoparticulates of oligomers, proto brils, and sonicated amyloid seeds of Aβ 42 (abbreviated as "seeds" and "Aβ" hereafter), Parkinson's alpha synuclein (abbreviated as "αS" hereafter) and FapC, a major protein constituent of the extracellular functional amyloid matrix of Pseudomonas aeruginosa. We further demonstrated endothelial leakiness in vivo with mouse cerebral vasculature exposed to Aβ oligomers and seeds. Using signaling pathway assays and steered discrete molecular dynamics (sDMD) simulations, we revealed that APEL arose from the anionic protein nanoparticulates interacting with vascular endothelial (VE)-cadherins and their associated adherens junction machinery. Our in vitro data showed that APEL arose from non-ROS and non-apoptotic events, where Aβ oligomers and seeds instead underwent direct extracellular interactions with VE-cadherins to trigger molecular pathways yielding intercellular gap formation. Through the sDMD simulations, we found that the oligomers remarkably reduced the cadherin dimer stability followed by the amyloid seeds, while exible amyloid monomers did not affect the cadherin integrity. These results converged on a general new phenomenon that the nanoparticulates of anionic peptides and proteins rendered through molecular self-assembly and fragmentation, can remodel the vasculature endothelium preceding the onset of ROS, in ammation, cytotoxicity, and cerebral blood ow constriction as suggested by earlier etiological paradigms 16,21,23−26 .
Phenomenologically as well as mechanistically, APEL is analogous to nanomaterials-induced endothelial leakiness (NanoEL) [27][28][29] , where in the latter, anionic inorganic nanoparticles of less than 100 nm in size open up the endothelial paracellular route by disrupting the VE-cadherin junctions of apposing cells. The observed occurrence of APEL here was highly surprising in that proteins and peptides are biomolecules differing signi cantly from inorganic nanoparticles in origin, structure and function, and the density of proteins/peptides (~1.35 g/m 3 ) is markedly below the density threshold determined for NanoELcompetent inorganic nanoparticles (1.72 g/m 3 ) 30 . Regardless, the ndings of APEL entailed rich biological and medical implications, from the systemic spread of amyloid proteins to their cross-seeding, BBB translocation, and clearance that underpin the pathogeneses of AD, Parkinson's disease (PD) and other amyloid diseases debilitating the brain and other human organs.

Characterization of amyloid protein aggregates
In our previous works 27, 31 , we found that certain nanoscale structures could interact with the VE-cadherin proteins in adherens junctions, leading to a cascade of intracellular signaling that included actin rearrangements and development of transient intercellular gaps -a phenomenon we termed as NanoEL.
Although our previously used nanoparticles were synthetic and not of a natural origin, the current evidence reiterated that it was their possession of certain physicochemical traits, such as their size range 31 and anionic charge 32 that conferred them with the NanoEL ability, instead of their identity as being a certain material. Considering this, we hypothesized that biological nanoparticulates such as amyloid protein aggregates possessing an optimal size range, stiffness and anionic charge, could potentially induce a similar phenomenon to NanoEL when interacting with the endothelium. Aβ (isoelectric point pI: 5.5) 33 oligomers and seeds were rst chosen due to their relevance in vascular pathology and their signi cance to AD, as well as their polymorphic architectures. Three other types of amyloid proteins were also employed to facilitate a comprehensive understanding of APEL with respect to protein composition, size, and charge. αS, an anionic neuronal protein (pI: 4.67) 34 implicated in the pathology of PD 35,36 , islet amyloid polypeptide (IAPP), a cationic pancreatic polypeptide (pI: 8.8) 37 implicated in the pathology of type 2 diabetes (T2D) 38 , and FapC, an anionic protein of Pseudomonas aeruginosa functional amyloid, were used in our study 39 .
Fibrillization of the four amyloid protein species was performed (described within Experimental Section) and a thio avin T (ThT) uorescence kinetic assay was conducted simultaneously to monitor the amyloid formation processes. Samples at various time points were analyzed via transmission electron microscopy (TEM) to con rm their structural evolution. In Figure 1B (IAPP f and IAPP s ) for IAPP, and 180 h for FapC s according to their ThT assays ( Figure 1F). In Figure 1G, the sizes of the various amyloid protein aggregates were tabulated from the TEM analysis, where the oligomers of the four amyloid proteins ranged ~20-70 nm in length and their seeds ~20-90 nm in length, with their thicknesses below 20 nm.

APEL was observed with several species of amyloid protein aggregates
Subsequently, we investigated the occurrence of endothelial leakiness when HMVECs were treated for 30 min with varied sizes of Aβ (monomers Aβ m , oligomers Aβ o , proto brils Aβ o−p1 and Aβ o−p2 , brils Aβ f and sonicated seeds Aβ s ), αS (monomers αS m, oligomers αS o and seeds αS s ), IAPP (monomers IAPP m , oligomers IAPP o and seeds IAPP s ), and FapC (seeds FapC s ). Confocal uorescence microscopy revealed APEL occurring in the endothelial cell monolayers exposed to the protein nanoparticulates (Figure 2A and B). Notably, IAPP, the only positively charged peptide, did not induce any leakiness in endothelial barrier across all its four forms, including the seeds, which led to leakiness for all other three types of amyloid protein aggregates. This suggested that the characteristic of negative charge, as opposed to positive charge, is a necessary factor for the APEL phenomenon to occur. Within the three anionic amyloid species that gave rise to APEL, their soluble monomers and lengthy amyloid brils were unable to induce leakiness.
Gap area, diameter, and frequency distribution were analyzed from captured images ( Figure 2C and S1, Table S1). In Figure 2C, for the forms of Aβ, the calculated percentage of gap area was insigni cant for Aβ m when compared to the non-leaky control, followed by a signi cant increase for Aβ o , before decreasing when the aggregate length increased from Aβ o−p1 to Aβ f , where gap area % nally became insigni cant compared to control. The subsequent sonication of Aβ f to Aβ s led to a signi cant APEL occurrence again. In frequency distribution, the number of gaps induced by Aβ o was 14.3 ± 3.1×10 2 gaps/mm 2 , which decreased to 5.3 ± 2.8×10 2 gaps/mm 2 for Aβ o−p2 and increased back to 15.8 ± 6.8×10 2 gaps/mm 2 after Aβ s were introduced ( Figure S1). Similarly, αS o and αS s yielded signi cant increases in gap formation, while αS m did not elicit leakiness of signi cance against untreated control ( Figure 2C, S1).
Functional FapC s , tested solely as seeds, also elicited signi cant gap formation. Next, we employed transwell assays to quantify the extent of induced endothelial leakiness with the use of uorescein isothiocyanate conjugated dextran (FITC-dextran) as a probe. In agreement with confocal uorescence microscopy, the transwell assay revealed that when the four forms of Aβ (Aβ m, Aβ o, Aβ f, Aβ s ) were incubated with HMVECs and probed for permeability changes, there were signi cant increases in FITCdextran transport across the endothelial barrier for Aβ o and Aβ s compared to untreated control, while Aβ m and Aβ f did not yield signi cant results ( Figure 2D).
Collectively, APEL bore similarities with NanoEL. The differing identities of the amyloid species (such as human or bacterial origin, pathogenic or functional) did not appear to dictate the occurrence of APEL on endothelial barriers, but rather possession of relevant physicochemical properties, such as suitable size range and negative charge, was necessary. Better performances by the oligomers and sonicated Aβ seeds than brils, along with the short duration of induction (30 min, or likely even faster) in increasing vascular permeability were notable as APEL differed from some currently proposed mechanisms of Aβ-induced permeability. These include: rat brain microvessel endothelial cells subjected to brillar Aβ 42 for durations of 24-72 h exhibited an altered pattern of tight junction protein expression and localization 40 ; murine brain endothelial cells subjected to Aβ 42 for 24 h led to increased permeability, which was proposed to result from Aβ m interacting with receptors for advanced glycation end products (RAGE), leading to downstream disturbance to tight junction protein expression and permeabilization 41 ; and human umbilical vein endothelial cell barrier treated with Aβ 42 for 24 h experienced downregulated junction protein expression and increased permeability 42 . Hence, we were prompted to further elucidate the speci cs of the APEL process.

In vitro Aβ APEL occurs independently of toxicity phenomena or endocytic process
We isolated one species of the amyloid proteins that resulted in signi cant APEL, Aβ, for subsequent characterization of the APEL process. The two forms that induced the most signi cant fold changes in leakiness ( Figure 2D), Aβ o and Aβ s , were identi ed and selected. Aβ-induced APEL was rst characterized through a transwell assay, where HMVECs were treated with Aβ o and Aβ s at two concentrations (20 µM and 40 µM) for 30 min. We measured an increase in FITC-dextran transport (and therefore, leakiness) with increasing concentrations of both Aβ forms ( Figure 3A). Next, these APEL-relevant concentrations of Aβ were employed in examination of common toxicity effects that could lead to intercellular gap formation, which were important due to Aβ's established roles in neurotoxicity when interacting with brain endothelial cells 1 . ROS is known to lead to cell shrinkage through apoptosis 43 , and free radical oxidative stress is a critical pathological effect by Aβ 44,45 . We detected no signi cant increase in ROS production under the different Aβ treatments for as late as 2 h, beyond the timepoint (30 min) when APEL had occurred ( Figure 3B). In complement, as seen in Figure 3C, prior treatment with ROS scavenger N-acetyl cysteine (NAC) did not signi cantly reduce the degree of Aβ-induced APEL despite ROS scavengers being reported as protective against Aβ-induced oxidative stress 46 . In addition, apoptosis, a known contributor to cell shrinkage 47 , was also reported as a toxic effect induced by Aβ structures from oligomers to brils, involving caspases such as 3, 8 and 9 48-50 . Firstly, no decline in cell viability was detected under Aβinduced APEL conditions ( Figure 3D). Subsequently, as shown in Figure 3E, we detected no activated forms of important apoptotic enzymes, namely caspase 9 (an initiator caspase), caspase 3 (an effector caspase), and poly(ADP-ribose) polymerase (PARP; a downstream cleavage target) during Aβ-induced APEL conditions 51,52 . Furthermore, endocytosis was found to not signi cantly contribute to Aβ-induced APEL as the mixture of endocytic inhibitors methyl-β-cyclodextrin (MβCD) and monodansylcadaverine (MDC) could not signi cantly reduce leakiness induced in the HMVEC barrier ( Figure 3F). Collectively, Aβinduced APEL was found to be not signi cantly attributable to endothelial toxic effects typical to Aβ, most likely due to the short duration (30 min) of cell exposure to Aβ, and was triggered extracellularly, bearing strong similarities to NanoEL 53 .

In vitro Aβ-induced APEL involves extracellular interactions with VE-cadherin
To validate the extracellular triggering of the APEL process and potential involvement of interactions of Aβ with adherens junction protein VE-cadherin, we further performed a competitive inhibition assay with Aβ s . A pre-treatment of BV-6 antibody blocked the extracellular domain of VE-cadherin before the later incubation with Aβ s . Concurrently, a similar set of groups instead involving pre-treatment with NAC was employed. NAC was a pre-treatment not anticipated to decrease the overall APEL extent, based on our ndings in Figure 3B and C. As captured in confocal uorescence microscopy in Figure 4A and then analyzed semi-quantitatively in Figure 4B, the percentage of gap area formation was signi cantly reduced under the additional BV-6 pre-treatment (BV-6+Aβ s ) compared to only Aβ s . Interestingly, only adding BV-6 also led to some signi cant gap formation in HMVEC monolayers compared against the untreated control. However, it is important to note that there were no signi cant differences between BV-6+Aβ s group and only BV-6 group, which suggested BV-6 had negated the effects exerted by Aβ s . Taken together, extracellular interactions with VE-cadherin were not only important but essential for the occurrence of APEL. In the parallel series, the degree of gap formation under Aβ s was not signi cantly reduced by pre-treatment with NAC. Comparison between control, NAC-only and NAC+Aβ s groups a rmed that NAC played no role in inducing APEL and the APEL effect was exerted by Aβ s . Gap frequency distribution analysis ( Figure S2) supported similar overall conclusions. The Aβ s -only group had a frequency of 21.7 ± 2.4×10 2 gaps/mm 2 , while BV-6 + Aβ s exhibited a signi cant decrease in gap frequency with 10.3 ± 5.0×10 2 gaps/mm 2 . As anticipated, NAC+ Aβ s group yielded no signi cant decrease in gap frequency compared to Aβ s -only group, at 18.7 ± 5.3×10 2 gaps/mm 2 .
Next, we employed a co-localization assay where cells were treated with ThT-labeled Aβ s and VE-cadherin was later labeled during immuno uorescence staining. Under confocal uorescence microscopy, it was observed that bright yellow dots were yielded, indicating ThT-labeled Aβ s (in green channel) and high VEcadherin expression levels (in red channel) were co-localized ( Figure 4C). The yellow dots were located along the perimeters of HMVECs, where the cell junctions were found, and most were just next to or inside the intercellular gaps. This assay further suggested that interactions took place between Aβ s and VEcadherins during the APEL process.

In vitro Aβ-induced APEL involves VE-cadherin signaling and actin remodeling
To further elucidate the molecular mechanisms behind Aβ-induced APEL, we examined the effects of APEL on VE-cadherin signaling pathway. Canonically, VE-cadherin signaling includes phosphorylation at two important residues tyrosine 658 (Y658) and tyrosine 731 (Y731), which leads to downstream reduced interactions with p120 and β-catenin, with actin rearrangement and appearance of intercellular retractions as notable endpoints [54][55][56][57] . Our previous works with NanoEL revealed that a similar cascade could also be triggered by synthetic nanoparticles with certain physicochemical parameters, despite their exogenous origin 31,53 . In a similar manner, we observed increased phosphorylation of Y658 and Y731 residues of VEcadherin under Aβ o treatment, where the degree of phosphorylation increased with the concentration of Aβ ( Figure 5A). Src-kinase inhibitor, PP1, which inhibits the kinase responsible for phosphorylation of the residues, was also employed as a pre-treatment, where it was observed that the degree of phosphorylation under every Aβ o treatment decreased compared to their relevant control but was incompletely inhibited ( Figure 5A-C). Likewise, Aβ s treatment resulted in increased phosphorylation of the two residues and application of PP1 pre-treatment modulated an increase in phosphorylation ( Figure 5D-F). Complementarily, in our parallel transwell assay, PP1 pre-treatment also resulted in a signi cant reduction of induced leakiness under the treatment of either Aβ nanoparticulates, when compared to their non-PP1 pre-treated counterparts ( Figure 5E). This further evidenced the involvement of activated VEcadherin signaling during the APEL phenomenon. To validate the involvement of actin remodeling in the APEL event, we also employed a Rho-associated, coiled-coil containing protein kinase (ROCK) inhibitor, Y27632, which disrupts normal actin functioning through destabilizing focal adhesions and stress bers 58 . The transwell assay revealed that pre-treatment with Y27632, which interfered with normal actin remodeling, led to suppressed induction of leakiness by Aβ o or Aβ s compared to the respective untreated groups ( Figure 5F). Together, these results revealed the role of VE-cadherin signaling and actin remodeling as part of the molecular mechanisms involved during APEL.

Aβ-induced APEL with ex vivo blood vessels and in vitro& in vivo BBB models
In addition to in vitro investigation, an ex vivo assay was further conducted to determine vascular leakiness in swine vessels after the treatment of Aβ o or Aβ s at 200 or 400 µM ( Figure 6A, B). Permeability of Evans blue dye (EBD) at concentrations of 200 µM and 400 µM for both Aβ species had a signi cant increase compared to the respective untreated control. The highest degree of permeabilization was observed at 400 µM for both species. Subsequently, the Aβ species were investigated for their effects on the permeability of the BBB. Human brain endothelial cells hCMEC/D3 were employed as a BBB model in the transwell assay format and exposed to 0-40 µM of Aβ s . An increase in the penetration of FITC-dextran was observed with increasing concentration of Aβ s , with 40 µM of Aβ s attaining the highest permeability compared to control of 0 µM, while 5 µM had a comparatively insigni cant difference ( Figure 6C).
Considering the signi cant induction of permeability by Aβ species in the in vitro BBB model, we further determined the effects of Aβ in an in vivo setting. Swiss mice were intravenously injected with a solution of EBD containing either Aβ s or Aβ o , or solely EBD. Fluorescence imaging of mice brains post-sacri ce revealed increased the permeability across BBB for both Aβ s and Aβ o treatments compared to EBD-only treatment, with the Aβ s group attaining the higher increase in permeability ( Figure 6D). Furthermore, transverse brain sections were also derived from mice after similar injected treatments for immuno uorescence analysis after staining with anti-A11 antibody ( Figure 6E). Presence of Aβ structures resembling oligomers were found in both Aβ s and Aβ o groups, as inferred from the anti-A11 antibody staining. Comparatively, no Aβ structures were discovered in the EBD-only treated control. Together, the in vivo ndings revealed an increased permeability in the brains of mice after treatment with Aβ species, as corroborated by a concurrent presence of Aβ structures within brains.

Aβ interactions with VE-cadherin dimers characterized in silico
To understand the disruption of VE-cadherin dimer induced by amyloid proteins at the molecular level, we employed all-atom discrete molecular dynamics (DMD) and steered DMD (sDMD) simulations. We rst performed binding simulations of VE-cadherin with Aβ species. First, we employed a EC1 cadherin dimer from the full-length of a VE-cadherin dimer to e ciently mimic the trans interactions ( Figure 7A). From our recent study, we successfully demonstrated that the EC1 dimer was a suitable model to evaluate the dimer stability in the presence and absence of gold nanoparticles (AuNPs) 53 . Next, three different forms of Aβ monomer (Aβ m ), oligomer (Aβ o ), and tetramer seed (Aβ s ) were prepared to assess their interactions with the cadherin dimer and how the binding disrupted the dimer ( Figure 7B and Experimental Section). The three different forms of Aβ were randomly located near the EC1 dimer, and 40 independent binding DMD simulations for 50 ns were performed. For the binding simulations, we computed the binding frequencies of the peptide species with the EC1 dimer. We observed that each form of Aβ entailed distinct binding behaviors with the EC1 dimer ( Figure 7C). Our recent study revealed that an AuNP coated with citric acids preferred to bind the turn region of the dimer mostly consisted of cationic amino acids 53 . Similarly, we observed that the Aβ o highly bound to the turn regions of the EC1 dimer, however, the Aβ s mostly bound to the C-terminus of the dimer while the Aβ m bound to the entire region of the dimer. The colored binding frequencies on the surface of the EC1 dimer detailed the binding behaviors of the Aβ species with the EC1 dimer ( Figure 7D). Such different binding behaviors can also be veri ed by computing the binding frequency of the EC1 dimer with the Aβ species ( Figure S3A). We gured out that the rst 10 residues of Aβ o and middle 15 residues (15-30) of Aβ s mostly interacted with the EC1 dimer, where all amino acids of the Aβ m highly bound to the EC1 dimer. Except for the monomer, the preferred binding sites of the Aβ o and Aβ s with the EC1 dimer were different from each other due to the distinct conformations of the Aβ nanoparticulates ( Figure S3B).
After the binding DMD simulations, all-atom sDMD simulations were carried out to understand the EC1 dimer stability in the presence and absence of the Aβ species. For the sDMD simulations, one of the EC1 domains was immobilized and the other side of the domain stayed exible. Constant forces in the range of 0-60 pN with 10 pN of windows were applied to the exible domain of the EC1 dimer toward to the EC2 domain of the VE-cadherin dimer ( Figure 7E). Subsequently, we performed 70 independent sDMD simulations with randomized initial velocities assigned according to Maxwell-Boltzmann distribution and each sDMD simulation lasted for 100 ns. Then, we evaluated the dimer stability with and without Aβ species by violin plots as a function of rst mean dissociation time and applied forces after the sDMD simulations ( Figure 7G-I). The rst mean dissociation time was determined when the number of contacts at the dimer interfaces was reduced to zero. Here, we note that the 100 ns of the dissociation time means that the EC1 dimer stayed associated during the sDMD simulations. We observed that the Aβ o signi cantly increased the probability of early cadherin dimer dissociation, followed by the seed.
Speci cally, Aβ o elicited the highest cadherin dimer dissociation under the low force range (0-30 pN).
However, the effect of Aβ m on the dimer disruption was negligible for all applied forces. Representative trajectories of the Aβ o -and Aβ s -EC1 complexes re ected the early dissociation of the cadherin dimer ( Figure 7J and K). Aβ m was detached from the dimer at an early stage or moved around during the sDMD simulations due to the exible nature of the monomer ( Figure S4). To con rm the reduced cadherin dimer stability, we measured the RMSF of the exible domain of the EC1 dimer ( Figure 7F). In a recent study 53 , we showed that a reduced entropy disrupted the inherent function of a cadherin dimer. Our calculated RMSF results indicated that the Aβ o and Aβ s considerably reduced the overall RMSF, while Aβ m slightly reduced the RMSF for some amino acids. Although the Aβ species and the EC1 dimer in the simulations were not entirely the same sizes as in the experiments due to the high computational costs, the overall tendencies of the violin plots agreed well with the experimental data (Figures 3, 5, 6). Taken together, our computational results demonstrated Aβ-induced cadherin disruption and further revealed their corresponding dissociation mechanisms on the molecular scale.

Conclusion
Molecular assembly of proteins and peptides is, in essence, a remarkable nanotechnology employed by biological systems to render functional and pathogenic outcomes. Within the framework of AD, the selfassembly of Aβ m into oligomers, proto brils and amyloid brils entails rich pathogenic implications that are not fully understood, as re ected by a lack of success in clinical trials targeting protein amyloid aggregation. While CAA and compromised BBB usually accompany the symptoms of dementia 14 , the exact cause of cerebral vascular damage and their relationship with the development of AD remain unclear, with ROS production, in ammation, and the physical breakdown of tight junctions implicated as causative instigated by Aβ, Tau and apolipoprotein E, among others 14,15,17,21−23 .
In this study, we discovered that the oligomers and seeds but not monomers or brils of anionic amyloidogenic proteins, either pathogenic (i.e., Aβ and αS) or functional (i.e., FapC), ubiquitously elicited APEL in HMVECs whose characteristics were reminiscent of NanoEL, a biological phenomenon entailed by inorganic nanoparticles of certain size (<100 nm) and charge (anionic or near neutral) 29 . Speci cally, our signaling pathway and transwell assays as well as uorescence imaging, complemented with atomistic DMD and sDMD simulations, revealed disruption to the VE-cadherin machinery introduced by Aβ o and Aβ s . In contrast, Aβ m , Aβ f and all three forms of cationic IAPP were incompetent in inciting APEL.
These ndings suggested that, aside from size and charge to t within the nite paracellular space and evade endocytosis, the stiffness of amyloid proteins, elevated in the oligomers and proto brils than the monomers due to β-sheets stacking through brillization ( Figures S5&S6), could also play a role in the occurrence of APEL.
Notably, our signaling pathway assays revealed that APEL was independent of ROS production, cytotoxicity and endocytosis but mediated by the protein nanoparticulates engaged with the extracellular domains of VE-cadherins, triggering intracellular actin network reorganization as well as intercellular gap formation (Figures 3-5). These latter ndings were consistent with that of NanoEL but distinguished sharply from the existing literature linking vasculature damage and AD 13,59 , both in terms of the mechanisms and the spatiotemporal characteristics of intercellular gap formation. In light of the bodily distributions of the vasculature and amyloid proteins, the mechanism of amyloid protein aggregates navigating through the paracellular space of the endothelium appears broadly applicable to elucidating the coexistence of AD and PD pathologies with vasculature damage 15,17,59,60 , the selective neuronal pathology incited by systemic administration of αS 61 , and the curious cross-talks between the amyloidogeneses pertinent to AD, PD, prion diseases, and T2D 12,62−64 .

Experimental Section/methods
Aβ, IAPP and αS sample preparations: 1 mg of Aβ (AnaSpec Inc., purity ≥ 95%) and IAPP (AnaSpec Inc., purity ≥ 95%) were treated with 1 mL of hexa uoro-2-propanol (HFIP, Sigma Aldrich, USA) for 3 h at room temperature to break down the pre-existing aggregates. Then the solutions were aliquoted and freeze-dried for future use. The dried Aβ was dissolved in 0.1% NH 4 OH and further diluted in MilliQ H 2 O to a required stock concentration. The dried IAPP, lyophilized αS (AlexoTech) and FapC were respectively dissolved in MilliQ H 2 O.
To acquire different species of the amyloid proteins, the aqueous solutions of Aβ, IAPP, αS and FapC were incubate at 37°C and allowed to brillate for different times according to a thio avin T (ThT) kinetic assay. The seeds were derived from sonicated mature brils of different amyloid proteins using a Vibra-Cell™ Ultrasonic VCX 750 sonicator equipped with a 3 mm microtip. 2 min sonication at 20% of the maximum output power of the sonicator was applied. Transmission electron microscopy TEM images of all amyloid proteins at different time points according to the ThT result were acquired using a Tecnai F20 electron microscope (200 kV). 10 µL of each sample was placed onto glow discharged formvar/carbon coated copper grids (400 mesh, Proscitech) and blotted after 1 min incubation. The sample-coated grids were negatively stained with 5 µL of 1% uranyl acetate (UA) for 30 s. The grids were blotted and left to air dry before imaging.

Thio avin T kinetic assay
Attenuated total re ection-Fourier transform infrared (ATR-FTIR) spectroscopy: FTIR spectra (1580-1700 cm −1 ) that contained the amide I regions of each amyloid protein at different stages of brillization (monomeric/oligomeric/proto brillar/ brillar/seeds) were acquired with an IRTracer-100 (Shimadzu) spectrometer, which was equipped with a He-Ne laser and an MCT detector (Hg-Cd-Te). The MCT detector was constantly being cooled with liquid nitrogen. Regarding sample preparation, the peptide samples were prepared as described above and incubated at 37°C at 100 µM concentration. Concentration for the brillar FapC solution was 50 µM. The collection of the oligomeric/proto brillar or brillar species included a 100 µL acquisition and further lyophilization using a freeze-drier at appropriate timepoints added to the top of the reservoir. Air-drying then followed using a heat gun. Samples containing monomeric species were in the powder form during spectrum acquisition. Data acquisition included the absorbance (%) measurement mode with 512 number of scans. All spectra were acquired with 4 cm −1 resolution and signals were processed with the Happ-Genzel apodization function. Peak deconvolution that generated the Gaussian band pro le for each selected peak on each amide I region was performed with the built-in PeakDeconvolution application through Origin Software (Origin Lab). Deconvoluted band regions were assigned to each type of secondary structure following previous reports 66 .

Cell culture
Human dermal microvascular endothelial cells (HMVECs) were obtained from Lonza (Switzerland) and cultured in complete EndoGRO-MV-VEGF growth medium (Merck Millipore, USA). hCMEC/D3 human brain endothelial cells were derived from Merck Millipore (USA) and cultured in complete endothelial cell medium (Sciencell, USA). All cell culture was performed under a standard culture condition (37°C, 5% CO 2 ). In all experiments, endothelial cells were seeded at an initial density of 40,000 cells/cm 2 and were cultured to form a con uent monolayer before subsequent treatments. Immuno uorescence staining of in vitro samples: HMVECs were grown on 8-well chamber slides to reach con uent monolayers. After being treated with different amyloid proteins for 30 min, cells were gently washed with Hank's balanced salt solution (HBSS) and xed by 4% paraformaldehyde for 15 min, followed by permeabilizing and blocking with 0.1% saponin and 5% horse serum in PBS/azide for 1 hour at room temperature. Thereafter, xed cells were washed three times with PBS and incubated overnight at 4°C with primary rabbit anti-VE-cadherin antibody (Abcam) at 1:400 dilution with 5% horse serum in PBS/azide. Then the cells were washed with PBS and incubated with secondary donkey anti-rabbit Alexa Fluor 594 antibody (1:500, Abcam) and Phalloidin-iFluor 488 (1:1000, Abcam) in PBS/azide solution for 2 h at room temperature. After nuclei staining with Hoechst 33342 (Sigma Aldrich, USA) at 1:2000 dilution for 5 min, the cells were imaged with Leica SP8 lightening confocal microscope (Leica, Germany) and semiquantitative analysis was performed using ImageJ.

Treatments prior to exposure to Aβ samples
In assays that involve pre-treatments, the cultured HMVECs were exposed to the respective treatments, which were supplemented into complete EndoGRO-MV-VEGF growth medium, for a duration of 1 h. For experiments with antioxidants, the pre-treatment involved N-acetyl cysteine (NAC, 10 µM; Sigma Aldrich, USA), which was used to alleviate ROS production. For experiments about endocytosis, the inhibitors monodansyl cadaverine (MDC, 10 µM; Sigma Aldrich, USA) and methyl β-cyclodextrin (MβCD, 5 mM; Sigma Aldrich, USA) were used as a cocktail to inhibit endocytosis. For experiments related to cell signaling, the Rho-associated protein kinase inhibitor Y-27632 (Sigma Aldrich, USA) and Src family tyrosine kinase inhibitor PP1 (Sigma Aldrich, USA) were employed at 10 µM in complete growth medium.
After the pre-treatment, the medium was replaced with fresh growth medium now containing respective Aβ species and the same antioxidant/inhibitor(s). Growth medium treatment that contained the antioxidant/inhibitor(s) but no Aβ species acted as negative control. The usage of BV-6 antibody as a pretreatment was described under "competitive inhibition assay" instead.

Transwell insert assays
In general, HMVECs were cultured on transwell inserts (polycarbonate membrane, 0.4 µm pore diameter; Corning Costar, USA) in a 24-well plate until the formation of a con uent layer (seeded at density of 40,000 cells/cm 2 , 2 days). HMVECs were treated with different types and concentrations of the amyloid proteins that were supplemented into complete EndoGRO-MV-VEGF growth medium, for the stated durations. Subsequently, Aβ treatments were removed and the wells were washed with PBS. Fresh media with FITC-dextran (1 mg/mL, 40 kDa; Sigma Aldrich, USA) were then added to quantitatively determine permeability in endothelial barriers. After 30 min exposure to FITC-dextran treatment, solutions in the lower compartment were sampled (100 µL) and their uorescence was recorded by a microplate reader (Hidex, Finland) at wavelengths of 490/520 nm (excitation/emission). The degree of FITC-dextran transport was de ned as uorescence reading from a treated group normalized by its corresponding untreated control.
Competitive inhibition assay: Con uent HMVECs cultured on 8 well chamber slides were incubated with VE-cadherin BV-6 antibody (1:100; Millipore) in medium containing 3% fetal bovine serum for 1 h. VEcadherin BV-6 antibody speci cally recognizes the extracellular domain of VE-cadherin. Cells were washed with HBSS and subsequently incubated with Aβs for 30 min. Then, immuno uorescence staining was performed to observe the leakiness.
Association of Aβs and VE-cadherin assay Page 15/27 Con uent HMVECs were incubated with ThT-labeled Aβ s for 30 min, which were derived from the brils in the ThT assay. After that, excess Aβ s were removed, and the cells were washed with HBSS.
Immuno uorescence staining was performed to observe the location of ThT-labeled Aβ s and VE-cadherin.
Immunoblotting: HMVECs were cultured in 6 cm cell culture dishes and subjected to the pre-treatment of PP1 inhibitor in fresh cell medium or fresh medium only. Subsequently, HMVECs were exposed to different concentrations of Aβ o and Aβ s for 1 h. Following the exposure, each dish was washed thrice with chilled PBS and lysed with Laemmli sample buffer (63 mM Tris-HCl pH 6.8, 2% sodium dodecyl sulphate (SDS), 10% glycerol, 1% 2-mercaptoethanol and 0.0005% bromophenol blue). The cell lysis buffer solution had been supplemented with a cocktail of 1% protease and phosphatase inhibitors (Sigma Aldrich, USA). Gel electrophoresis was conducted with the derived cell lysates using 10% resolving polyacrylamide gels (Mini Protean, Bio-Rad, USA) and proteins transferred onto nitrocellulose membranes (Sigma Aldrich, USA). The membranes were blocked with 5% bovine serum albumin (BSA) solution for 1 h and incubated with a solution of the relevant primary antibody overnight at 4°C. Subsequently, membranes were washed thrice and then incubated in a solution of relevant horseradish peroxidase (HRP)-conjugated secondary antibody for 1 h. Membranes were exposed to Immobilon Western Chemiluminescent HRP substrate kit (Merck, USA) and protein bands were captured visually through a chemiluminescence imaging system (Syngene, UK). Expression levels of proteins in images were analyzed semi-quantitatively through ImageJ software and normalized against respective control group's protein band within each image. In all washing steps, as well as preparation of blocking and antibody solutions, Tween 20 detergent (TBST; composed of: 150 mM NaCl, 20 mM Tris-HCl, 0.1% Tween 20) was used. Primary antibodies were used at dilution of 1:1000 and secondary HRP-conjugated antibodies were employed at 1:2500. The complete list of antibodies utilized for immunoblotting are provided in Table S2.

Ex vivo vascular leakiness assay
For the ex vivo vascular leakiness assay, swine vessels were obtained from a local slaughterhouse in Chongqing. Three pigs were used, the blood vessels of each pig were taken out and used for one experiment in 3 groups, and the experiment was repeated 3 times for three pigs. Brie y, blood vessels of the coronary artery were cut transversely into individual pieces and placed in a commercial transwell chamber after removal of its original membranes. The blood vessel areas we took could surround the entire internal space of the transwell inserts. Aβ species of 200 µM and 400 µM were added to the custom-made swine vessel transwell device and incubated for 6 h. After the exposure, the Aβ-containing solution was discarded and then Evans blue dye (EBD, 100 mM) was added to each well for an additional 1 h. During the experiment, transwell inserts were placed in a 24-well plate, and then the 24-well plate was placed at 37°C for static culture. Finally, the uorescence from the lower compartment of the transwell was acquired at 624 nm with a microplate reader. Readout from the negative control group was used for normalization.
In vivo leakiness assay of Aβ s and Aβ o : All in vivo mice experiments were approved by the Southwest University Animal Care and Use Committee. Experiments were carried out according to the NIH's guidelines for the care and use of laboratory animals. Adult, 10 weeks old, male Swiss mice were obtained from Ensiweier Biotechnology Co, Ltd. (Chongqing, China). Mice were supplied with free access of food and water and were kept at standard temperature in a humid environment with a light/dark cycle of 12 h. Mice received once intravenous injection 100 µg Aβ s -containing 10 mM Evans blue dye solution or 100 µg Aβ o -containing 10 mM Evans blue dye solution. The control mice received once intravenous injection of 10 mM EBD solution. After 24 h, the mice were sacri ced to obtain the brains for imaging by using NEWTON 7.0 Imaging System. Immuno uorescence staining of mice brain samples: Mice experiments were approved by the Southwest University Animal Care and Use Committee. Adult, 10 weeks old, male Swiss mice were supplied with free access of food and water were kept at standard temperature in a humid environment with a light/dark cycle of 12 h. Mice received one intravenous injection 100 µg Aβ s -containing 10 mM EBD solution or 100 µg Aβ o -containing 10 mM EBD solution. The control mice received once intravenous injection of 10 mM EBD solution. After 24 h, the mice were sacri ced to obtain the brains. The brains were xed in 4% paraformaldehyde solution and removed after 48 h. They were then embedded with optimal cutting temperature (OCT) compound and sliced in a transverse section by a freezing microtome. Then, slices were incubated with 10% skimmed milk for 2 h, then incubated with anti-A11 antibody (1:250 dilution) for 3 h. Alexa Fluor 488-conjugated secondary antibody (1:250 dilution) was used for uorescence detection. DAPI was used for the visualization of nuclei and then imaged with a super-resolution laser confocal microscope (Nikon, N-SIME).
DMD simulations for amyloid and EC1 cadherin dimer binding: All-atom discrete molecular dynamics (DMD) simulation with implicit solvent models was used for this study to characterize the Aβ induced VEcadherin dissociation. DMD is a unique category of molecular dynamics (MD) with signi cantly enhanced sampling e ciency, which has been widely applied to biomolecular studies such as protein folding, peptide aggregation 67-69 , and understanding the protein structure and dynamics 70,71 . The details of the DMD algorithm were described in our previous studies 67-69 . From our recent study, it has been demonstrated that the EC1 dimer from full-length VE-cadherin was suitable model to characterize the trans interaction mimicking the cadherin dimer coming from two opposing cells. Therefore, the EC1 cadherin dimer adapted from the cryo-EM model of EC12 cadherin dimer was considered (PDB ID: 3PPE 72 ) for the current study. For constructing the molecular model of EC1 dimer for this study, the bond constraints for calcium ion sites (i.e., residues Glu11, Asp62, Glu64, Asp96, and Asp99) and Gō-potential were applied to the domain-swapped region of the EC1 dimer, respectively ( Figure 7A). Speci cally, the weak contact energy of 0.4 kcal/mol (~0.67 K B T) was assigned to C β atoms of contacting residues. Next, we prepared three different forms of Aβ 42 monomer (Aβ m ), oligomer (Aβ o ), and tetramer seed (Aβ s ). The atomic models of Aβ m (PDB ID: 1IYT 73 ) and Aβ s (PDB ID: 5OQV 74 ) were used, and Aβ o was brought from our recent work 75 . To consider the Aβ s model, the seed was relaxed and equilibrated for 50 ns with the application of Gō-constraints between each peptide. Same contact energy of Gō-potential applied on EC1 dimer was assigned to the inter-peptides of Aβ s . Each amyloid peptide was randomly located away from the EC1 dimer at least 12 Å away in a 150 nm 3 cubic box and counter ions were distributed to neutralize the net charge. To avoid a biased potential energy, different initial velocity was applied and 40 independent DMD simulations each of 50 ns (an accumulative 2.0 µs DMD simulations) were performed. 50 fs/step of the unit simulation time and 1 kcal/mol of corresponding energy were employed and a temperature of 300 K was maintained with Anderson's thermostat. After the binding simulations, we computed the binding frequencies of the amyloid proteins with the EC1 dimer from the last 20 ns of binding simulations. To calculate the binding frequency, we assigned 0.65 nm of cutoff distance to get an atomistic contact between the EC1 cadherin dimer and amyloid proteins.   Scale bar = 20 µm. Red: VE-cadherin, green: actin, blue: nuclei. Scale bar = 20μm. (C) Gap area percentages were analyzed via ImageJ according to the images related to panels A and B. (D) Transwell assay quantitatively revealed occurrence of endothelial leakiness after 30 min incubation with the different forms of Aβ. Thrombin (3 U/mL) acted as positive control of induced leakiness. Data are shown as mean ± SD (n=3), analyzed via one-way ANOVA using GraphPad Prism 8, *represents P < 0.05, ** represents P < 0.01, *** represents P < 0.001 and **** represents P < 0.0001 between the compared groups. when compared against counterparts without inhibitor treatment (mixture of 5 mM MβCD and 10 µM MDC, from 1 h prior). Results shown as mean ± SD (n=3), analyzed via two-way ANOVA with Tukey's multiple comparison tests in GraphPad Prism 8. * represents P < 0.05, ** represents P < 0.01, *** represents P < 0.001 and **** represents P < 0.0001 statistical signi cance. "n.s." represents no statistical signi cance. Data are shown as mean ± SD (n=3), analyzed via one-way ANOVA using GraphPad Prism 8, *represents P < 0.5, ** represents P < 0.01, *** represents P < 0.001 and **** represents P < 0.0001 between the compared groups. (C) Co-localization of ThT-labeled Aβ s and VE-cadherin was observed in HMVECs near their cell junctions, revealing association between them. Red: VE-cadherin, green: ThT-labeled Aβ s . Scale bar: 20μm. reduced activation under PP1 pre-treatment. Data are mean ± SD, n=3 (biologically independent samples, representative blots presented here), analyzed via two-way ANOVA with Tukey's multiple comparison tests. * represents P < 0.05, ** represents P < 0.01, *** represents P < 0.001 and **** represents P < 0.0001 statistical signi cance. Referring to comparisons between groups with PP1 against their counterparts without PP1, # represents P < 0.05, ## represents P < 0.01, ### represents P < 0.001. "n.s." represents no statistical signi cance. (G) Transwell assay a rmed the signi cant reduction of Aβ APEL (Aβ o or Aβ s , 30 min) due to pre-treatment with PP1 (10 µM, 1 h), when compared to respective counterparts without PP1. (H) Aβ APEL involved actin remodeling. Pre-treatment with a RhoA kinase (ROCK) inhibitor, Y27632 (10 µM, 1 h) led to signi cantly decreased leakiness under Aβ o and Aβ s (30 min). Results presented are shown as mean ± SD (n=3), analyzed via two-way ANOVA with Tukey's multiple comparison tests. * represents P < 0.05, ** represents P < 0.01 *** represents P < 0.001 and **** represents P < 0.0001 statistical signi cance.

Figure 6
Ex vivo characterization and in vitro and in vivo blood-brain barrier models of Aβ-induced endothelial and vascular leakiness. Induced leakiness was observed in swine vessels under the exposure of (A) Aβ o and (B) Aβ s . The ex vivo penetration of Evans blue dye (EBD) in swine vessels was directly relational to the concentration of Aβ species (0, 200 and 400 µM). (C) Transwell assay on hCMEC/D3 brain endothelial cells treated with Aβ s revealed increased leakiness with increased concentrations. hCMEC/D3 were exposed to Aβ s from 0-40 µM, for the duration of 30 min. Penetration of FITC-dextran was used to measure permeability across cells. Results are presented as mean ± SD (n=3), analyzed via two-way ANOVA with Tukey's multiple comparison tests. * represents P < 0.05, ** represents P < 0.01 *** represents P < 0.001 and **** represents P < 0.0001 statistical signi cance. (D) In vivo experiment involving injection of Aβ s or Aβ o (100 μg) into mice revealed the increased leakiness across the blood-brain barrier through

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. APELSIFinal.docx