The membrane axis of Alzheimer’s nanomedicine

Alzheimer’s disease (AD) is a major neurological disorder impairing its carrier’s cognitive function, memory and lifespan. While the development of AD nanomedicine is still nascent, the field is evolving into a new scientific frontier driven by the diverse physicochemical properties and theranostic potential of nanomaterials and nanocomposites. Characteristic to the AD pathology is the deposition of amyloid plaques and tangles of amyloid beta (A β ) and tau, whose aggregation kinetics may be curbed by nanoparticle inhibitors via sequence-specific targeting or nonspecific interactions with the amyloidogenic proteins. As literature implicates cell membrane as a culprit in AD pathogenesis, here we summarize the membrane axis of AD nanomedicine and present a new rationale that the field development may greatly benefit from harnessing our existing knowledge of A β -membrane interaction, nanoparticle-membrane interaction and A β -nanoparticle interaction. Abstract The development of AD nanomedicine, an emerging frontier in advanced nano-biomedical research, can benefit from our accumulating knowledge of A β -cell membrane, nanoparticle-cell membrane and A β -nanoparticle interactions.


Introduction
Alzheimer's disease (AD) is the primary form of neurological disorder with a global burden of 50 million. Histopathologically, the AD brain post mortem is characterized by the presence of extracellular plaques comprised of amyloid beta (Aβ), lipids, metals (e.g., Zn 2+ , Cu 2+ and Fe 3+ ) as well as proteins. [1] Aβ, ranging from 36-43 in total number of residues, is a proteolytic product of transmembrane amyloid precursor protein (APP), cleaved off by β and γ secretases sequentially. The self-assembly of Aβ into cross-β fibrils, or Aβ amyloidosis in short, has been linked to neurodegeneration and the pathology of AD, as described by the amyloid cascade hypothesis. [2] As Aβ plaques are primarily found in the extracellular space in the central nervous system (CNS), while the peptide itself is derived from APP expanding from extracellular to transmembrane domains (Scheme A), understanding Aβ amyloidosis in the presence of membrane mimetics (e.g., micelles, nanodisks, vesicles, liposomes, amphiphilic polymers and lipid bilayers), naturally, has been a major research undertaking over the past two decades, especially in the biophysical and biochemical communities. [3] Indeed, AD has been considered a membrane disorder, resulting from the interactions of Aβ with lipids, lipid rafts, transmembrane proteins and nicotinic acetylcholine receptors (nAChRs). [4] Furthermore, neurotoxicity in AD has been proposed to arise from the interaction of Aβ with membrane receptors [5] and the generation of membrane pores. [6][7] While membrane mimetics bear limited resemblance to cell membranes, fundamental research with these systems has nonetheless shed a crucial light on the structural transformation of Aβ in physiological environments as well as on their implications for neurotoxicity and AD pathology.
Towards the development of therapeutics against amyloid diseases, much research has been focused on the design and delivery of synthetic small molecules, natural polyphenols, peptidomimetics, monoclonal antibodies and, more recently, nanomaterials, in order to target or sequester the toxic oligomeric and fibrillar peptide species. [8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25] Especially with nanomaterials, their small size and diverse physicochemical properties have often entailed a remarkable inhibitory potential against the aggregation and toxicity of Aβ, alpha synuclein (αS) as well as human islet amyloid polypeptide (IAPP) associated with AD, Parkinson's disease (PD) and type 2 diabetes (T2D), among others. [26] In this review, we summarize the state-of-the-art of Aβ amyloidosis inhibition with nanomaterials by highlighting the central interactions between the peptide, the nanoparticle inhibitors, and the membrane substrates of various degrees of complexity. We present the rationale that, fundamentally, such application can greatly benefit from exploiting our prior and growing knowledge of a) Aβ-membrane interaction (section 2), b) nanoparticlemembrane interaction (section 3), and c) Aβ-nanoparticle interaction (section 4), for facilitating the development of potent AD nanomedicines (section 5) (Scheme B). An outlook on future AD nanomedicine is provided at the end of the article.

Aβ-membrane interaction
In this section, we summarize computational and experimental findings concerning Aβmembrane interaction, the first of the three main constituents of this review that entail important implications for the development of AD nanomedicine.

Simulation studies
Although the interaction of Aβ with cellular membrane is likely a key factor in AD pathogenesis, the underlying molecular mechanisms remain largely unknown. While it is challenging to experimentally determine Aβ-membrane interaction with single-molecule resolution, computer simulations are ideally suited for delineating such interaction with molecular details and for providing new insights into the structure and function of complex biomolecular systems. It should be acknowledged that extensive computational studies have been conducted to date concerning the inner workings of Aβ-membrane binding. These studies can be categorized by the states of Aβ [27][28][29][30] (e.g., monomers, oligomers and fibrils), the positioning of Aβ with respect to the membrane [31][32] (e.g., adsorbed on or inserted into the membrane), and the computational methods adopted [33][34] (e.g., all-atom and coarsegrained/CG simulations), as discussed to various extents in previous reviews. [ [35][36][37][38][39] In this section, we summarize the critical findings revealed by simulation studies concerning the role of Aβ-membrane interaction in promoting the peptide aggregation and membrane damage.
Aβ has been shown to strongly interact with membranes in experiments. One of the main questions that computer simulations attempt at resolving is how the cell membrane modulates Aβ aggregation. Towards that goal, the adsorption of Aβ on the lipid bilayer and the early stage of the peptide oligomerization have been simulated. All-atom molecular dynamics (MD) simulations showed that Aβ was attracted from the solution to the surface of both zwitterionic dipalmitoylphosphatidylcholine (DPPC) and anionic dioleylphosphatidylserine (DOPS) bilayers driven by the electrostatic interactions between the charged peptide residues and the lipid headgroups, which accelerated Aβ aggregation by increasing the local peptide concentration and promoted Aβ diffusion by limiting their motion to the two dimensional surface compared with those in solution. [40][41][42] Once an Aβ monomer was bound to the membrane, their overall secondary structure did not change noticeably on the DPPC bilayer compared with their random coil structure in solution, whereas the DOPS bilayer strongly enhanced the helical structure, especially near the Nterminus of the peptide. No significant β-structure was observed except for a small, unstable β-hairpin formed on the anionic DOPS bilayer. However, Aβ bound to DOPS was able to adopt the conformation with a more exposed hydrophobic C-terminus that was more accessible to the adjacent Aβ peptides, which promoted protein-protein interaction and favored the early stage of Aβ aggregation. These results suggested that the structural conversions observed in experiments were not due to the ordering of monomeric Aβ on the bilayer surface but a result of protein-protein interaction that occurred during oligomerization, further enhanced by Aβ binding to the cell membrane. [43] Additional simulations considered the fact that the local pH near an anionic DOPS lipid bilayer was lower than the bulk, giving rise to transient stabilization of the peptide β-structure. [43] This observation implied that the Aβ-membrane interaction at the physiological pH can stabilize Aβ monomers in an aggregation-prone intermediate state. The hypotheses derived from simulations of Aβ monomers interacting with membranes have been validated by the dimerization of Aβ on the lipid membranes, where DOPS lipids promoted strong proteinprotein interaction between Aβ monomers while weakened protein-lipid interaction during the dimerization. [44] For larger Aβ oligomers, the interaction of an Aβ tetramer with the membrane resulted in elongation of the tetramer in the presence of both pure 1-palmitoyl-2oleoyl-sn-glycero-3-phosphocholine (POPC) and cholesterol-rich raft model membranes. Tetramer-raft interactions caused rearrangement of key hydrophobic regions in the tetramer and formation of a rod-like structure indicative of a fibril-seeding aggregate. [45] In addition, the chemical composition of the lipid membrane can also modulate the Aβ-membrane interaction. [46][47] Simulations showed that an increased cholesterol content significantly promoted the binding of Aβ monomers to a POPC lipid bilayer by increasing the bilayer thickness, hydrophobic chain order and surface hydrophobicity, while decreasing the lipid mobility. [48] In the presence of monosialotetrahexosylganglioside (GM1), which is enriched in the outer leaflet of the plasma cell membranes in the CNS, the GM1-Aβ interaction facilitated the formation of β-sheet structures. [49] The molecular mechanism underpinning the toxicity of Aβ is another central topic in numerous simulation studies. It has been increasingly accepted that the cytotoxicity resulting from amyloidosis may be attributed to membrane disruption caused by amyloid aggregation, and the formation of membrane pores by amyloid proteins is regarded as a key event leading to membrane leakage and subsequent cell death. [50][51] Since it is computationally challenging to directly observe the spontaneous formation of amyloid pores by Aβ in membrane simulations, molecular modeling approaches combined with MD simulations are usually adopted. When inserted into the membrane, early-stage Aβ aggregates were postulated to adopt two models that mimicked the pores in membranes -one is the channel model where the peptides were inserted without inclination to the membrane normal axis, and the second one is the β-barrel model where the peptides were inclined with respect to the pore axis (Fig. 1A). [52][53] The dynamic properties of the two possible Aβ channel topologies were examined, as the CNpNC (C and N denote the C-and N-terminal β-strands, respectively, and p stands for the central pore), where the polar/charged N-terminal β-strand faced the water-filled pore and the hydrophobic C-terminal β-strand faced the bilayer, and the NCpCN, where the C-terminal β-strand faced the solvated pore (Fig. 1B). The CNpNC channel preserved the pore and conducted solvent, whereas the NCpCN channel failed to preserve the pore due to the hydrophobic collapse, in good agreement with the images acquired by atomic force microscopy (AFM). [54] Simulations of Aβ ion channels consisted of U-shaped β-strand-turn-β-strand motifs in the lipid bilayer showed that the optimized size of channels presenting a toxic ionic flux ranged between 16-to 24-mer with the subunit organization and dimensions being consistent with the experiments, whereas the smaller (12mer) channel collapsed and the larger (36-mer) channel could not be supported by the bilayer. [55] Importantly, in agreement with AFM, the simulations indicated that the β-sheet channels broke into loosely associated mobile β-sheet subunits [56][57] (Fig. 1C). On the other hand, the β-barrels of Aβ  and Aβ  were also shown to preserve the U-shaped CNpNC topology with a similar subunit organization and dimensions as the channel models [58] (Fig. 1D). Moreover, the β-strands of the barrels adjusted the conformation by decreasing the tilt angle of the N-terminal β-strand and increasing the tilt angle of the Cterminal β-strand. As a result, the 12-mer Aβ barrel produced a well-defined pore, in contrast to the collapsed pore of the channel model, suggesting that the Aβ barrels provided a more optimized pore organization. [58] In addition to the N-terminal truncated Aβ, channels and barrels with L-and D-enantiomers Aβ [59] and Aβ mutants [60] (F19P, F20C and ΔE22) have also been identified in simulations. These models of Aβ channels and barrels in the bilayer have offered valuable insights into the molecular mechanisms of amyloid toxicity and revealed structural information pertinent to drug discovery efforts focused on inhibiting the formation of such toxic channels and pores in the cell membrane.

Experimental studies
Aβ peptides are produced from APP within the transmembrane region of neurons and synapses, mainly in the forms of Aβ40 and Aβ42, whose aggregation propensities are closely related to the occurrence and progression of AD. While Aβ generation is associated with the membrane lipid environment, Aβ-membrane interaction can induce membrane reorganization, deformation, pore formation, increased permeabilization, and lipid extraction, leading to intracellular ion dyshomeostasis, production of reactive oxygen species (ROS) with an excess of lipid peroxidation, increased lipid susceptibility to oxidative damage [61] and, eventually, neuronal death. As these processes are critical to the AD pathology, understanding Aβ-membrane interaction should directly benefit the development of new AD therapeutics. [62] Accordingly, various types of membrane models with controlled characteristics, bilayer components, thickness and fluidity, have been designed to mimic the plasma membrane and screen their interactions with Aβ (Table 1). All Aβ species displayed a certain level of affinity for the anionic membrane, [63] with the oligomers displaying a greater membrane affinity than the monomers or fibrils. [64] Kayed et al. observed that synthetic Aβ oligomers, other than monomers or fibrils, could increase the conductance of the lipid bilayer and induce a sharp decrease in electron density throughout the lipid membrane, without generating discrete channel or pore formation or ion selectivity. [65] However, other studies revealed that Aβ fibrils could also be embedded in the upper leaflet of the bilayer. [66][67] The carpet model (Fig. 2) was proposed to explain the peripheral association of the bilayer and Aβ, which was accompanied by a general increase in membrane conductance either by membrane thinning or lateral spreading of lipid headgroups. [68][69][70] In addition, significant lipid extraction and disruption to bilayer were induced by oligomeric Aβ and were visualized through AFM and transmission electron microscopy (TEM) (Fig. 3A). [66] The phospholipid molecules extracted from both bilayer leaflets, or first from the outer leaflet and then the inner leaflet of the membrane, may contribute to the formation of Aβ-lipid complexes resulting in a detergent-like lipid removal action (Table 1, Fig. 2). [66,68,71] Thereafter, large pores spanning over the bilayer (~50 nm) were formed, and small Aβ oligomers could insert themselves into the pores that were expanding over time. Moreover, recent studies have revealed that this detergent-like activity of the peptide strongly depends on the structure and size of the Aβ oligomers. [63,72] For example, globular nonfibrillar oligomers caused the largest reduction of lipid diffraction in anionic 1,2-dimyristoyl-snglycero-3-phosphoglycerol (DMPG) lipid monolayers upon insertion, compared to short fibrillar oligomers and monomers. [63] Small-sized Aβ oligomers, corresponding to low molecular-weight Aβ oligomers, could destroy the brain total lipid extract (BTLE) membrane bilayer by lipid extraction. In contrast, the aggregation of large-sized Aβ oligomers was accelerated on the BTLE membrane surface without affecting the membrane integrity. [72] Small-sized Aβ oligomers and globular nonfibrillar oligomers have been considered as the more toxic species in causing neurodegeneration. [73] Alternatively, Aβ could induce channel-like perforation in neuronal cell membranes (Fig. 2), causing an increase in membrane conductance as well as intracellular calcium and ethidium bromide influxes in a concentration-and time-dependent manner. [74][75] Upon addition to large unilamellar vesicles (LUVs) soluble oligomers formed small cation-selective pores immediately, whereas large molecules were allowed to enter the cells with the membrane gradually losing its physical integrity over time. [76] The interaction sites for Aβ peptides within the membrane were localized at the lipid rafts enriched in GM1, cholesterol and phosphatidylcholine (Table 1). [64,[76][77][78] The increment of GM1 distributed in the membrane was reported in the temporal and frontal cortices of the brain suffering from AD, and a specific form of Aβ bound to GM1 may serve as a seed for the formation of toxic amyloid aggregates. [64,[79][80][81] The more hydrophobic Aβ42 oligomers were rapidly sequestered from the brain interstitial fluid in vivo. [64] Compared to the nontoxic monomers, the Aβ42 oligomers could strongly bind to the GM1 on neuronal membranes to induce disruption of lipid bilayers via pore formation and malfunction of raftassociated Ca 2+ channels, leading to Ca 2+ influx into cells and downstream synaptotoxic effects (Fig. 3) [82][83] Hence, blocking the interaction between GM1 and Aβ peptides has been suggested as a strategy for AD drug discovery. [74,84] Cholesterol, as a principal constituent in lipid rafts, not only facilitates Aβ binding to GM1 but also interacts with Aβ directly. [4,67,78] Cholesterol-containing lipid vesicles could promote Aβ42 aggregation, as evidenced by the enhanced primary nucleation rate of Aβ42 by up to 20-fold through a heterogeneous nucleation pathway. GM1 and cholesterol cooperation has been proposed to expedite amyloid pore formation in the cell membrane: amyloid peptides first bind to the negative charged sialic acid in GM1 via electrostatic interaction, then cholesterol interacts with the inserted peptides to finalize the pore formation process. [4,85] A membrane therapy, targeting either cholesterol or GM1 or both, was successfully demonstrated by effectively disrupting the interactions of Aβ with the two types of membrane lipids. [85] In addition to membrane lipids, other cell surface molecules have been considered as possible receptors of Aβ peptides, including cell-prion peptide (PrP C ) and Toll-like receptor 4 (TLR4), the receptor for advanced glycosylation end products (RAGE) and nAChRs (Table 1). [86][87][88][89][90][91][92][93] Aβ oligomers were found to bind with PrP C specifically and the binding complex was related to synaptotoxicity, inhibition of long-term potentiation (LTP), memory impairment, and decreased survival in AD mice models. [88][89] PrP C -binding Aβ oligomers, and even their reactions with phosphorylated τ-protein (p-τ) in neurofibrillary tangles were also found in the human brain. [88,90] Concomitant hippocampal Aβ burden and RAGE upregulation were discovered in mice, and the inhibition of RAGE ameliorated memory impairment. [92] The LTP deficits and neuronal death caused by Aβ were also found through an autocrine/paracrine mechanism due to TLR4 signaling. [91] Longer Aβ protofibrils caused inflammation and increased production of TNF-α and IL-β via TLR4. [94] Therefore, characterization of the binding of membrane surface receptors and Aβ will undoubtedly facilitate the development of novel AD therapies. Furthermore, alternations to lipid rafts and perturbations to membrane signaling may also be considered as promising molecular targets for AD.

Nanoparticle-membrane interaction
In this section, we highlight computational and experimental findings of nanoparticlemembrane interactions, which provide a valuable basis for describing the behavior and fate of AD nanomedicines in cellular environments.

In silico studies
Theoretical studies of nanoparticle-membrane interaction first began in 2007, when Qiao et al. examined a DPPC bilayer exposed to a pristine fullerene C 60 and its hydroxylated counterpart fullerol C 60 (OH) 20 (Fig. 4A). [97] Using atomistic molecular dynamics simulations the researchers found that the potential of mean force (PMF) favored uptake of the hydrophobic C 60 , while the amphiphilic C 60 (OH) 20 was excluded by the lipid bilayer. However, the hydroxyl groups of the fullerol enabled their H-bonding with the head groups of the DPPC bilayer, rendering their close association with the top leaflet of the artificial membrane. This study provided the first theoretical support to the experimentally observed phenomenon that the toxicity of nanoparticles usually increases with their increasing hydrophobicity. [98][99] Many computational works have ensued, offering new insights into nanoparticle-cell membrane interactions and their biological implications. Using CG simulations Wongekkabut et al. examined the translocation of many C 60 nanoparticles across a dioleoylphosphatidylcholine (DOPC) lipid bilayer, which was shown to be thermodynamically driven and occur on the microsecond timescale. [102] In addition, high concentrations of the fullerene nanoparticles induced changes in both the structure and elasticity of the lipid bilayer. Using dissipative particle dynamics (DPD), Yang and Ma found that the membrane-penetrating capabilities of nanoparticles possessing the shapes of spheres, ellipsoids, rods, discs and pushpin-like were determined by the contact area between the nanoparticles and the lipid bilayer, as well as the local curvature of the nanoparticles at the contact points. [103] Using CG simulations, Shi et al. noted that the cell entry of anisotropic one-dimensional carbon nanotubes favored tip recognition and was accompanied by rotation. [104] In addition, the insertion and lipid extraction capacities of two-dimensional graphene nanosheets interacting with a lipid bilayer of Escherichia coli (E. coli) (outer layer: palmitoyloleoylphosphatidylethanolamine or POPE; inner layer: both POPE and palmitoyloleoylphosphatidylglycerol or POPG) were reported by Tu et al. employing CG MD simulations (Fig. 4B). [100] Utilizing MD simulations, Kraszewski et al. examined the association of C 60 with potassium ion channels in a DOPC lipid bilayer, where the free energies of the binding between the nanoparticle and the KcsA, MthK, and Kv1.2 protein domains of the potassium channels ranged from 0.7-7.6 kcal/mol to offer a quantitative baseline for nanoparticle-ion channel association (Fig. 4C). [101]

Experimental studies
Several experimental studies have reported the dependence of cell/lipid membrane interactions with 0D-2D nanomaterials on the nanoparticle size, hydrophobicity, morphology, charge as well as surface modification. With carbon-based nanomaterials ( Table 2), Chen et al. [105] reported the plasma membrane translocation of C 70 assembled with natural organic matter (NOM) led to their increased cellular uptake in mammalian cells. This effect was not found with plant cells, however, due to the size-exclusion effect afforded by the plant cell wall. With carbon nanotubes (CNTs), their cell internalization was highly dependent on their physical size. Unlike multi-walled CNTs, single-walled CNTs exhibited high cellular uptake and internalization via the energy-dependent/independent endocytotic pathways based on their length (100-200 nm: clathrin dependent; 50-100 nm: clathrin and caveolae dependent, <50 nm: energy independent) ( Table 2). [106] In vitro study on the intracellular transport of graphene quantum dots (GQDs) in Madin-Darby Canine Kidney (MDCK) monolayer cells showed a lipid-raft mediated transcytosis of GQDs across the plasma membrane and the cytoplasm, which was inhibited by a lipid-raft inhibitor methyl-βcyclodextrin. [107] For graphene oxide (GO) nanosheets, quartz crystal microbalance with dissipation monitoring (QCM-D) sensing showed the rupture of positively charged POPC/ POEPC (3:1) liposomes mainly driven by electrostatic interactions. [108] In a cellular membrane environment, exposure to GO nanosheets induced pore formation in both viable and non-viable A549 and Raw264.7 cells initiated by lipid extraction. [109] Using a 2D sandwiched graphene-membrane superstructure, Chen et al. [110] reported the Brownian diffusion pattern of GO across cell membranes. Taking advantage of the formed pores the researchers demonstrated GO as efficient drug nanocarriers against breast cancer cells (Table  2). Exposing large and small GO sheets to neutrophils, furthermore, elicited membrane stripping, prompted the formation of neutrophil extracellular traps (NET), and elevated the production of carbon radicals and oxidized cholesterol species, in a GO-size dependent manner. [111] In addition to bare nanomaterials, negatively charged inorganic TiO 2 , SiO 2 , gold nanoparticles (AuNPs) have been reported to interact and disrupt VE-cadherin-VE-cadherin homophilic interactions in endothelial cells, causing leakiness (termed as "NanoEL") in a dose-and size-dependent manner ( Table 2). [112][113][114] This transient cellular phenomenon provides both a new mechanism for understanding nanotoxicity and a strategy for devising new nanoparticle-based cancer drug delivery and therapies. On the other hand, as with any viable nanomedicine, their prolonged blood circulation is essential for desired drug delivery and efficacy. Zhao et al. [115] reported that the adsorption of large mesoporous silica nanoparticles (600 nm) to silanol rich red blood cell (RBC) membranes induced disruption, nanoparticle internalization and hemolysis, compared with functionalized small mesoporous silica nanoparticles (~100 nm) ( Table 2). Surface-modified AuNPs/AuNRs (gold nanorods) have been studied for their interactions with cell membranes. The electrostatic interactions between positively charged AuNPs and cell membranes exerted a significant effect on the integrity of plasma membranes causing rapid membrane depolarization and intracellular uptake of the nanoparticles. [116] Apart from charge, poly(vinyl alcohol-co-N-vinylamine) (PVA)-coated AuNPs displayed an increased effect of their backbone-dense primary amines on cellular association and the susceptibility of primary amines to protein corona formation. [117] Applying a library of ligand-coated AuNPs ( Table 2) to supported lipid bilayers, Wang et al. [118] identified size and adsorption affinity as two key parameters for ligand exchange at the nanoparticle-cell membrane interface, and determined the role of ligand stability on membrane integrity in a separate study. [119] A wealth of literature over the past decade [120] has implicated the central role of the protein corona for determining the biological behavior and fate of nanoparticles, including the cellular association and subsequent translocation of nanoparticles. The conformational changes of proteins upon binding with nanoparticles have been shown to trigger the activation of macrophage class A scavenger receptor (SR-A)-mediated phagocytosis of the nanoparticles in macrophage cells, compared to the reduced nanoparticle internalization in monocytic cells. [121] On the other hand, incubation of silver nanoparticles with human serum and cell culture medium and subsequent treatment of the nanoparticle-protein complexes with the scavenger receptor BI decreased their toxicity and IL-6 mRNA expression in rat lung epithelial and rat aortic endothelial cells. [122] By modifying protein charge distribution [123] or deglycosylating the coronal proteins, [124] other studies have revealed the importance of Coulombic interactions and immune activation for the cellular recognition and uptake of nanoparticles. In addition, surface chemistry of polystyrene nanoparticles has been found crucial for the rendition of a hard protein corona, thereby influencing the cellular uptake and endocytotic pathways of carboxylated vs. bare polystyrene nanoparticles. [125]

Aβ-nanoparticle interaction
The interactions between amyloidogenic human β2-microglobulin and a host of nanoparticles (copolymer particles, cerium oxide particles, quantum dots, and CNTs) were first examined by Linse et al. in 2007, [126] around the same time as the protein corona concept was put forth by Dawson and Linse et al. [127] Since then, amyloidosis inhibition with nanomaterials has evolved into a major effort concerning the aggregation, toxicity and mitigation of a range of amyloid proteins and peptides. In the following section, we highlight recent computational and experimental findings of Aβ-nanoparticle interactions, a foundation for the development of AD nanomedicines.

Simulation studies
Due to their versatile physiochemical properties, nanoparticles are promising amyloidosis inhibitors. [26,128] Accurate and efficient modelling of Aβ-nanoparticle interaction can significantly contribute to our understanding of the bio-nano interfacial phenomena and provide a guidance to the design of anti-amyloid nanomedicines. Extensive simulations have been conducted to date to characterize the interactions between Aβ and various types of nanoparticles, including soft polymeric nanoparticles (e.g., dendrimers and hydrogels), carbon-based nanoparticles (e.g., fullerenes, carbon nanotubes, graphene oxides and carbon quantum dots), and metallic nanoparticles (e.g., gold nanoparticles), among others. [129][130] Aβ showed a strong affinity for PEGylated nanoparticles due to the combination of both hydrophilic and hydrophobic interactions between the PEG shell and the peptide, which enabled the nanoparticle to encapsulate Aβ monomers and oligomers. [131] Carbon-based nanoparticles have been widely used in various biomedical applications such as biosensing, imaging and drug delivery. [132][133] C 60 disrupted Aβ fibrillization by destructing the βhelical twist, destabilizing the D23-K28 salt bridge and hydrophobic interactions between the Aβ peptides. [134] 1,2-(dimethoxymethano) fullerene, a water-soluble fullerene derivative, inhibited the dimerization of Aβ by binding with the peptide aromatic residues (F4 and Y10), the central hydrophobic region (L17-A21) and the C-terminal hydrophobic regions (I31-V40). [135] Aβ peptide was found to adsorb on the surface of single-walled CNT rapidly, during which the hydrophobic core of Aβ initiated the adsorption process and subsequently brought the peptide N-terminal region in contact with the CNT. Once adsorbed onto the CNT, the peptide propensity of collapsing between the central hydrophobic core region and the hydrophobic patches near the peptide C-terminus was reduced significantly, indicating that CNT could disrupt Aβ aggregation. [136] Driven by the strong hydrophobic interaction between Aβ and CNTs, Aβ (25)(26)(27)(28)(29)(30)(31)(32)(33)(34)(35) peptides with mixed antiparallel-parallel βstrands assembled into a β-barrel wrapping around the CNTs. [137] Similarly, carbon quantum dots (CQDs) exhibited strong binding with Aβ, especially with the peptide Nterminal region. Upon adsorption onto the CQD surface, Aβ peptides preferred to bind with CQDs instead of other peptides, resulting in a reduction of the inter-peptide contacts and subsequent fibrillization inhibition. [138] In addition, graphene nanosheets were shown to effectively break and clear the pre-formed fibrillar structures of Aβ by extraction of the peptides, which was attributed to the exceptionally strong dispersion interactions between the peptides and graphene that were further enhanced by their strong π-π stacking. [139] AuNPs are often employed as theranostic agents in drug delivery and photothermal therapy due to their relatively low toxicity and surface chemistry for functionalization. [140] By parameterizing interactions between AuNPs and proteins in MD simulations using density functional theory (DFT) calculations, [141] bare AuNPs facilitated the adsorbed Aβ to adopt the fibril-prone conformations, suggesting that bare AuNPs could promote the peptide fibrillization. [142][143] Other simulations with negatively charged AuNPs showed that the fibrillization process of Aβ was actually inhibited due to the electrostatic interactions between Lys 16 and Lys 28 of the peptide with the nanoparticle surface charge. [144] Based on the experimental and simulation studies, it can be concluded that nanoparticles can exhibit both promotion and inhibition effects on the aggregation of Aβ under different conditions. [145] To understand these seemingly conflicting effects of nanoparticles on Aβ aggregation, CG molecular dynamics simulations have been conducted to probe the effects of the strength of nonspecific nanoparticle-protein attraction and the relative concentration between protein and nanoparticles (Fig. 5). [146] It was revealed that the aggregation of Aβ on the nanoparticle surface was initially promoted with increasing nanoparticle-peptide attraction due to increased local peptide concentration on the nanoparticle surface and destabilization of the peptide folded state (Fig. 5A&C). However, further increase of nanoparticle-peptide attraction decreased the stability of amyloid fibrils and reduced their lateral diffusion on the nanoparticle surface necessary for peptide conformational changes and self-association, thus prohibiting amyloid aggregation (Fig. 5B&D). Moreover, it was found that the amyloid aggregation was also regulated by the relative concentration between protein and nanoparticles (Fig. 5E&F). Specifically, with a high nanoparticle/protein ratio, nanoparticles with intrinsic promotion effects may inhibit amyloid aggregation by depleting the proteins in solution while having a low concentration of the proteins on each nanoparticle's surface. These results offered a molecular basis for delineating the contrasting and seemingly conflicting effects of nanoparticles on amyloid aggregation, [26] which can be utilized for tailoring the design of AD nanomedicines. [146] Most of the existing simulations so far have only considered pristine or bare nanoparticles. However, upon entry into biological systems such as the plasma, the nanoparticle surfaces will be covered by various sugars, lipids and proteins, which leads to the formation the biocorona. [147] As a result, it is the corona-coated nanoparticles rather than the pristine or bare nanoparticles that interact with Aβ in vivo. Therefore, it is of practical importance to investigate the effects of the protein corona on Aβ-nanoparticle interaction in the future. In a recent work by Javed et al., the interaction between Aβ and casein coated-gold nanoparticles (βCas AuNPs) was examined. [24] The results showed that Aβ monomer became more helical and an Aβ oligomer could not grow into an extended β-sheet structure upon binding with the intrinsically disordered βCas with high conformational flexibility. Moreover, there was no overlapping between the binding sites of βCas with Aβ and those with the AuNPs, suggesting that the chaperone-like βCas in the AuNP corona could trap Aβ monomers in the aggregation-incompetent forms and cap Aβ fibrils from elongation.

Experimental studies
The oligomers are the major toxic species of Aβ amyloidogenesis, inducing disruptions to membrane structure and cellular ion homeostasis as well as inflammation, immunity and apoptosis due to their interactions with cell membranes, proteins, chaperones and small ligands. [26,148] The assembly of Aβ into pathological seeds, including the oligomers and subsequent tertiary aggregates such as protofibrils and amyloids, constitutes the primary neuropathological hallmark of AD. Thus, the design of AD therapeutics is usually focused on inhibiting and disrupting the Aβ aggregation pathway or accelerating Aβ elimination in the patient brain. To achieve this, a range of inorganic, polymeric, carbon-based and biomimetic nanoparticles and nanocomposites have been developed for potential AD treatment ( Table 3, Fig. 6).
AuNPs have been recognized as an excellent anti-amyloid agent. A systematic study on the effects of AuNP size, shape and surface charge on Aβ aggregation was performed on a total brain lipid-based supported lipid bilayer. [149] Larger AuNPs induced large amorphous Aβ aggregates on the lipid bilayer with a shortened lag phase of aggregation, whereas smaller AuNPs rendered protofibrillar Aβ structures. Naturally, positively charged AuNPs were more strongly attracted to the anionic Aβ, giving rise to the formation of small amorphous aggregates of fewer β-sheets and more random coil structures, which also elicited significant toxicity to SH-SY5Y neuroblastoma cells. In contrast, negatively charged AuNPs acted as nucleation seeds to accelerate the formation of larger Aβ aggregates by gathering more peptides and AuNPs together. Hence, anionic or neutrally charged nanoparticles appeared to be more effective candidates against the amyloidosis of Aβ, whereas cationic nanoparticles were more prone to protein fouling and membrane damage. Furthermore, a comparison of different shaped AuNPs revealed that the longer gold nanorods elicited a greater affinity for Aβ to inhibit its fibrillization, whereas all the facets of gold nanocubes were accessible for interacting with Aβ to produce a fibrillar network. [149] The mitigation potential of βCas AuNPs against Aβ toxicity and Alzheimer's-like symptoms was demonstrated in vitro and in a zebrafish model. [24] βCas, a milk protein adsorbed onto the surface of AuNPs, facilitated the binding of the nanoparticles with Aβ monomers/oligomers. The small βCas AuNPs (5-8 nm size) administrated to the bloodstream were able to cross the blood-brain barrier (BBB) and subsequently sequester toxic Aβ42 in the brain of larval and adult zebrafish through a nonspecific, chaperone-like mechanism (Fig. 7A&B). The mobility and cognitive function of adult zebrafish were recovered after the treatment of βCas AuNPs (Fig. 7C&D). In addition to AuNPs, nanoceria, iron oxide nanoparticles, silicon dioxide nanoparticles, silver nanoparticles as well as transition-metal nanomaterials such as molybdenum disulfide and black phosphorus have been shown as promising biocompatible nanomedicines for sequestering toxic amyloid species and alleviating their cell toxicity (Table 3). [150][151][152][153][154][155][156][157] Polymeric nanocomposites possess versatile physicochemical properties, which can be utilized to render nanoparticle-biomolecular composites for biomedical applications. [158] By altering their lipid concentration and composition, polymethacrylate-copolymer (PMA) encased lipidnanodiscs (~10 nm) effectively regulated Aβ fibrillization to exhibit less neurotoxicity. [159] Wan et al. developed high generations of phenyl-modified carboxylterminated polyamidoamine (PAMAM) dendrimers, with the capability of inhibiting Aβ42 aggregation and altering the ultrastructure of Aβ42 aggregates via the hydrophobic bindingelectrostatic repulsion (HyBER) theory. [160] Aβ-targeting ligand conjugation on nanoparticles is an efficient approach for disrupting the amyloidogenic process. Aβ-binding peptide KLVFF integrated with polymeric nanocomposites (NC-KLVFF) remarkably eliminated toxic Aβ aggregates, regained endocranial microglia's capability to phagocytose the peptide, and protected hippocampal neurons against apoptosis in AD mice (Fig. 8). [22] In addition, the inhibitory potency of biomimetic nanomaterials against amyloidosis has been recently demonstrated, further displaying satisfactory biocompatibility and biodegradability (Table 3). [161][162] A nanostructure αNAP-GM1-rHDL, comprised of GM1-modified reconstituted high-density lipoprotein (GM1-rHDL) and a neuroprotective peptide NAPVSIPQ (NAP), was found effective in protecting neurons from Aβ42 oligomers/ glutamic acid-induced cytotoxicity in vitro. [163] The biomimetic nanocomposite also showed the ablity to reduce Aβ deposition, ameliorate neurologic changes, and rescue memory loss efficiently in AD mice. Exploiting the membrane affinity of Aβ, GM1-rHDL displayed an antibody-like strong binding affinity for Aβ, thereby facilitating Aβ degradation and Aβ efflux across the BBB. The efficient loading of NAP further enabled the nanoparticulate drug delivery system for achieving AD therapy. [163] Smart nanoparticles, namely dcHGT NPs, were designed to not only inhibit and reduce Aβ aggregation but also simultaneously regulate acetylcholine imbalance. [164] Comprised of clioquinol (metal-ion chelating agent) and donepezil (acetylcholinesterase inhibitor), transcriptional activator protein, GM1 and human serum albumin (HSA), dcHGT NPs significantly inhibited Aβ aggregation while relieving acetylcholine-related inflammation in microglial cells. Furthermore, dcHGT NPs reduced Aβ deposition, ameliorated morphological neuron changes, rescued memory deficits, and significantly improved acetylcholine regulation in vivo.

AD nanomedicines and cell membranes
This section describes different routes for the transport of anti-AD nanomedicines across the BBB and how Aβ-membrane affinity may be exploited in the context of nanomedicine design. Nanomedicines that mimic the physicochemical properties of cell membranes may be crafted by either surface modification with lipid moieties or by cloaking them with cellular membranes. Such membrane-philic nanomedicines may hold the key for the effective BBB delivery of anti-AD therapies.

Nanoparticle-BBB interaction
The BBB is a highly selective semipermeable endothelial layer reinforced by tight junctions. The BBB separates the brain parenchyma from the lumens of blood vessels and protects the CNS from the entrance of xenobiotics from the periphery. The mechanisms a nanoparticle may employ to cross the BBB include passive diffusion of the nanoparticle through cellular membranes, hitchhiking the luminal surface transporters of the BBB for essential solutes, or via the endocytotic pathways. [169] In addition to the BBB being a major hurdle to AD nanomedicines, hematic proteins and immunity cells in the blood may further sequester nanomedicines from circulation. Among the myriad types of nanomaterials which have been tested over the years, inorganic nanoparticles, dendrimers, liposomes, polymeric nanoparticles and nanogels have been the most employed for the delivery of therapeutic agents and probes to the brain. [170][171] Recently, cell-mediated transport has been demonstrated as a promising strategy, where biomolecular-nanoparticle complexes were entrapped inside immune cells to reach the diseased tissues by chemotaxis. [169] Chemotactic drug delivery further enabled the formulation of a self-propelling system to shuttle across tissues driven by a glucose concentration gradient. [172] Alternatively, nanoparticles may be modified to mimic activated leukocytes, or functionalized with polymers or cellular membrane derivatives. These strategies disguise nanoparticles to freely circulate through the blood and effectively translocate across the BBB. Furthermore, the gaps between tight junctions may be transiently widened by ultrasound, or by downregulating the expression of occludin and claudin-5 that are components of the tight junctions, [173][174] thereby allowing the paracellular passage of nanoparticles.

AD nanomedicines exploiting cell membranes and the BBB
Although the pathophysiological mechanisms and pharmacological targets of AD remain under debate, there is a consensus that the primary cause of neuronal damage is induced by the membrane-association of small Aβ oligomers (n<10) that have a β-sheet rich structure with a tertiary hairpin. [175][176][177] The membrane affinity of the Aβ monomers is far less than the oligomers or a mixture of the monomers and oligomers. [178] The neuronal membranes also contribute to the structural transition of the Aβ monomers to oligomers. Indeed, the oligomers formed in buffer conditions and in the presence of cell membranes show different secondary structures. Kakio et al. used amide I and II band deconvolution of FTIR and demonstrated that the Aβ oligomers formed in buffer possessed in-register parallel β-sheets with strong interchain interactions. However, the Aβ oligomers formed in the presence of raft-like membranes presented anti-parallel β-sheets with weak interchain interactions. [179] One explanation for this difference is based on the fact that, in buffer conditions, the hydrophobic amyloidogenic region of Aβ is consumed in self-assembly, while in close proximity to cell membranes the peptide hydrophobic region is engaged in interactions with lipid rafts or gangliosides. [180] Exploiting this affinity-based interactions, AD nanomedicines have been devised by ascribing them with membrane-like surfaces, which can sequester the Aβ oligomers and prevent their interactions with cell membranes. In this context, coating of lipid molecules on the surface of metal nanoparticles provided a promising strategy to shut-down Aβ fibrillization. Suga et al. synthesized AuNPs of 100 nm in diameter that were coated with 1,2-dimylistoyl-sn-glycero-3-phosphocholine (DMPC) and 1,2-dimyristoyl-sn-glycero-3-phospho-L-serine (DMPS). [181] The AuNPs were kept in their monodisperse form by the addition of NaCl at physiological concentrations. Exposure of the AuNPs to Aβ resulted in the displacement of NaCl and deposition of an Aβ corona on the nanoparticle surface. The Aβ species further triggered agglomeration of the AuNPs via their concurrent interactions with multiple AuNPs. Similarly, Bhowmik et al. coated silver nanoparticles (AgNPs) of 60 nm in size with a lipid bilayer of POPC, POPG (1-palmitoyl-2oleoyl-sn-glycero-3 phosphoglycerol) and cholesterol. [182] These lipid-coated AgNPs sequestered Aβ oligomers onto their surfaces to reveal conformational changes of the membrane-bound Aβ oligomers. Raman spectroscopy indicated that the β-turns in the membrane-bound Aβ oligomers were flanked by anti-parallel β-sheets. Exploiting the membrane affinity of Aβ, Song et al. designed apolipoprotein E (ApoE3) containing highdensity lipoproteins (HDL) nanodiscs that captured both the Aβ40 and Aβ42 monomers as well as oligomers. [162] Based on TEM, the Aβ40 oligomers were found to bind with the lateral side of the nanodiscs via ApoE3. The Aβ42 oligomers, on the other hand, presented diffuse binding with nanodiscs to render larger nanostructures. These ApoE3-HDL nanodiscs crossed the BBB at 0.4% ID/g (injected dose per gram of tissue), suppressed the microgliosis and supported the immunological clearance of the Aβ monomers and oligomers from the brain of an AD mouse model. This ApoE3-HDL nanosystem was further utilized as a drug carrier against Aβ fibrillization. α-Mangostin, a polyphenolic compound, was loaded inside the ApoE3-HDL nanocarriers to translocate across the BBB (Fig. 9). [161] These nanocarriers, once loaded with α-Mangostin (ANC-α-M), targeted Aβ with an even greater affinity and accumulated specifically around the Aβ plaques in the AD mouse model. In addition to the membrane-mimicking nanoparticles, another delivery strategy which can benefit from the Aβ-membrane affinity is cell membrane-based delivery systems, [183] which use the whole cell or shredded cell exosomes. Another alternative is to coat polylacticglycolic acid (PLGA) nanoparticles and AuNPs with cell membranes. [184][185] Specifically, purified red blood cells (RBCs) were extruded together with the nanoparticles multiple times through a polycarbonate membrane, rendering a 7-8 nm coating layer of RBC membranes onto the NPs. The RBC membranes were in the right-side-out orientation that provided a negative zeta potential (−15 mV) and immune compatibility to the nanoparticles.
A general criterion for nanoparticles to permeate across the BBB is that the nanoparticles should be in the size range of 5-200 nm and possess neutral to negative charge. [186] In the context of AD nanomedicine, three types of organic nanoparticles, i.e., dendrimers, polymeric nanoparticles and liposomes, have demonstrated their BBB translocation capacity. Lightweight, hyperbranched dendrimers displayed a native inhibitory activity against Aβ fibrillization. [187] They also exhibited a neuronal anti-inflammatory potential that can be utilized to develop a suitable drug delivery system against AD. [188] Moscariello et al. used PAMAM dendrimer conjugated with streptavidin to study their mode of BBB translocation in a porcine-based in vitro model. [189] The dendrimers navigated from the luminal to abluminal side of blood vessels via endocytosis (Fig. 10). The intercellular tight junctions remained intact indicating no paracellular transport.
In addition to dendritic nanoparticles, tween 80 coated poly(n-butylcyanoacrylate) nanoparticles were able to deliver an anti-AD drug. [186,190] Specifically, the coating of tween 80 facilitated anchorage of plasma ApoE3 on the surface of the nanoparticles to enable their transcytosis across the BBB. Liposomes, after conjugation with cell-penetrating peptides or ApoE3, exhibited significant BBB permeation largely based on transcytosis. [191][192] In the case of inorganic nanoparticles, small (5-20 nm) sized AuNPs capped with casein, HDL or transferrin antibodies have demonstrated BBB translocation in vitro and in vivo. [24,[193][194] Overall, the use of cellular membrane mimicking or specific ligands can aid in the BBB permeation of anti-AD nanomedicines and their subsequent affinity-based interactions with toxic Aβ species.

Conclusion and future prospect
AD is a primary cause of death for the global ageing population. While the pathobiological origin of the disease remains debatable and AD treatment remains a formidable challenge in modern medicine, it has been generally accepted that neurodegeneration in AD involves the central process of Aβ amyloidosis, exacerbated by its downstream events in tau accumulation and neuroinflammation (e.g., dysregulation of immune activation and ApoE expression in the brain, etc.). [195] As the scope of AD physiopathology is vast, while the field of AD nanomedicine is nascent, in this review we have focused our attention on the structural origin and conformational susceptibility of the amyloidogenic peptide Aβ, and summarized the interplay between the peptide, its membrane environments (i.e., membrane mimetics, cell membranes, the BBB, and more), as well as its nanoparticulate inhibitors in silico, in vitro and in vivo. We wish to demonstrate through this review that our knowledge accumulated over the past two decades concerning Aβ-membrane, nanoparticle-membrane, and Aβ-nanoparticle interactions is highly relevant and beneficial to our understanding and manipulation of the membrane axis of Aβ amyloidogenesis. Briefly, the association of Aβ with cell membranes is understood to trigger conformational changes of the peptide to give rise to the toxic oligomers, which may subsequently induce membrane poration, lipid extraction, ROS production and cell death. In comparison, the adsorption of nanoparticles onto cell membranes may enable the translocation or permeation of the nanoparticles via endocytosis or NanoEL, which may elicit toxicity in the host depending on the size, shape, surface charge and functionalization of the nanoparticles as well as their protein corona. The interaction between Aβ and nanoparticles, on the other hand, may inhibit or accelerate the amyloid aggregation of the peptide and ascribe an amyloid protein corona on the nanoparticle surface. [147] In vivo, the peptide aggregates themselves may also acquire a dynamic protein corona depending on their changing hydrophobicity, mediated by their nonspecific interactions with the myriad biomolecular species in the host media. [196][197][198] While much remains to be understood concerning the interplay between cell membrane, Aβ, and nanoparticle inhibitors, an effective strategy is to elevate Aβ-nanoparticle interaction over Aβ/nanoparticle-membrane interaction by tailoring the nanoparticle surface chemistry to trap the peptide in its monomeric state, or sequester the toxic peptide aggregates via the assembly of functional-pathogenic protein coronae. [199] It is our hope that advanced nanobiomedical research may address many unanswered questions in this interdisciplinary area to facilitate the development of future AD nanomedicines.

DC
Direct current

MSN and RBC membranes
Binding of silanol-rich surface of MSNs with phosphatidyl choline-rich RBC membrane 2011 [115] Surface modified (Au based)