Designing peptide nanoparticles for efficient brain delivery

The targeted delivery of therapeutic compounds to the brain is arguably the most significant open problem in drug delivery today. Nanoparticles (NPs) based on peptides and designed using the emerging principles of molecular engineering show enormous promise in overcoming many of the barriers to brain delivery faced by NPs made of more traditional materials. However, shortcomings in our understanding of peptide self-assembly and blood-brain barrier (BBB) transport mechanisms pose significant obstacles to progress in this area. In this review, we discuss recent work in engineering peptide nanocarriers for the delivery of therapeutic compounds to the brain, from synthesis, to self-assembly, to in vivo studies, as well as discussing in detail the biological hurdles that a nanoparticle must overcome to reach the brain.


Introduction
To successfully deliver its cargo to a target tissue, a nanocarrier must navigate through the body while eliciting minimal immunogenic response and avoiding off-target delivery of cytotoxic compounds. In brain delivery, this challenge is often magnified by the need to achieve transport across the blood-brain barrier (BBB), perhaps the most tightly regulated biological barrier in the human body. Essential to success in this area are novel approaches and materials that combine complex functionality with biocompatibility. However, these two properties are often compromised as complex functions require access to untested chemistries that can be potentially toxic. Avoidance of toxicity is particularly critical in developing systems for the treatment of central nervous system (CNS) disorders, which adds an extra dimension to the toxicological profile: the neurological one.
One way around this is to use the same chemical currency as nature and adapt, functionalise, and to facilitate brain entry [49]. The ligands are conjugated at the NP surface, and this conjugation can be carried out either before (pre-formulation conjugation) or after (post-formulation conjugation) the formation of the NPs [17]. Both pre-and post-conjugation approaches have been applied to decorate polypeptide-based NPs for brain delivery. A fuller exposition of the effects of particular targeting ligands upon delivery efficacy is presented in Section 4.
In the pre-formulation conjugation approach, two methods can be employed to synthesise the polymer-ligand conjugates: "grafting-to" and "grafting-from". In the "grafting-to" method, the targeting ligands are conjugated to polymers by specific reactions with the end group or pendant side groups as a post-polymerisation modification [50]. A number of different reactions have been used for this strategy, including the nucleophilic attack of amino groups to N-hydroxysuccinimide esters (amine-NHS ester chemistry) [51,52], thiol-ene coupling of cysteine residues with alkenecontaining groups such as maleimides (thiol-maleimide chemistry) [14,15,48], and "click" reaction of azide with alkyne to form a 5-membered triazole ring ("click" chemistry) [53,54]. With relatively low requirements on the substrates and reaction conditions, the "grafting-to" method has been widely employed for the conjugation of a plethora of ligands to polypeptides. As an example, Kataoka et al.
In contrast to "grafting-to", the "grafting-from" method is based on the introduction of the targeting moiety at the very first step of polymer synthesis either in the initiator or the monomer.

The Hydrophobic Effect in Peptide Self-Assembly
Amphiphilic molecules aggregate to reduce the interaction of a hydrophobic moiety with the surrounding aqueous phase. The structure that an amphiphile will self-assemble into can be estimated using the packing factor, , which assigns an area, , to the hydrophilic moiety and a length, , and volume, , to the hydrophobic moiety [62][63][64]. This can be seen as the ratio of two areas: , and ; any difference in these two areas inherently leads to some curvature in the structure. This simple model is most applicable to low MW amphiphiles [65][66][67]. A similar trend can be seen in amphiphilic diblock copolymers where , the mass of the hydrophilic block divided by the total mass of polymer, plays the largest single role in determining the morphology of the selfassembled structure [68]. The relation is even more qualitative but, roughly, when , vesicles are formed, when cylindrical micelles are formed, and for spherical micelles are most often seen [63,69]. Kinetic factors and the greater conformational complexity of polymers also play a major role in determining system morphology [70,71]. Despite these limitations, varying is still the most common method for targeting two of the most popular structures for drug delivery applications: micelles and vesicles (Figure 3).
Forming the copolypeptide with one (or both) of the blocks comprising racemic amino acids disrupted helical structure, allowing the helicity of the polypeptide to be systematically varied.
Dispersing highly helical P(Lys P ) 100 -b-P(Leu) 20 in water resulted in the formation of micron-sized vesicles and a few, larger, sheet-like structures ( Figure 3A). It is of note that for P(Lys P ) 100 -b-P(Leu) 20 , Studying block copolypeptides in which one block is synthetic yields more 'traditional' selfassembly behaviour, including familiar morphological transitions from micelles to worms to vesicles as the relative size of the hydrophilic group decreases [92][93][94][95][96][97][98][99]. The synthetic block can simplify the interactions between the polymers, add functionality from a well-understood polymer, and act as a macroinitiator for ROP of NCAs. Quadir et al. successfully reported the production of pH-sensitive vesicles using PEG-b-poly(g-propargyl L-glutamate) block copolymers and their use in delivering doxorubicin (DOX) to breast adenocarcinoma cells [100]. Chécot et al. compared varying lengths of both poly(butadiene)-b-poly(Glu) and poly(isoprene)-b-poly(Lys) and found that the copolymers that were comprised of <70 mol% peptide formed pH-sensitive vesicles [97]. They also found that changes in conformation of both polypeptides, induced by pH alterations, led to an increase in NP size [99]. Vesicle and micelle formation can also be seen when the hydrophobic block is synthetic [101]. Li et al. studied the self-assembly of polystyrene-b-poly( -propoargyl--glutamate-g-ethylene oxide) in organic solvent as a function of water concentration [102]. At low water concentrations, micelles were obtained. As water concentration was increased, morphological transitions to worms and vesicles were seen, reminiscent of the early work of Eisenberg et al. on polystyrene-bpoly(acrylic acid) (PS-PAA) polymers [103]. Several groups have used poly(butadiene)-b-P(Lys) copolymers to produce vesicles that exploit the ionisation and change in secondary structure of lysine at low pH to induce vesicle swelling [104,105]. Gebhardt et al. found that stable vesicles of a pH switchable diameter could be formed from PB-b-P(Lys) with the amino acid in a helical conformation at low-and intermediate pH and in a -sheet conformation at high pH [104,106]. The latter observation of the polypeptides forming vesicles in a -sheet conformation is particularly remarkable and may be the product of the deeply non-ergodic nature of these self-assembled structures.
Polypeptides with more complex amino acid sequences can yield more controlled selfassembly behaviour than 'simpler' diblock copolypeptides [27,90,107]. ELPs, initially described by Urry and co-workers [108,109], are based on the pentamer sequence VPGXG, where X represents a guest residue and can be any amino acid except proline. Weitzhandler et al. used recombinant synthesis to produce copolypeptides with an elastin-like block of [XGVPG] x -Y and a resilin-like block (RLP) of G-(QYPSDGRG) y N, where x = 20-80 and y = 40-160 ( Figure 3B, Ref. [25]). Systematic variation of the guest peptide with Ser, alternating Ala and Gly, or Val, (from most to least hydrophilic) in the ELP allowed the hydrophobicity of the ELP to be systematically varied. ELPs exhibit a lower critical solution temperature (LCST) in water, meaning that they become insoluble with increasing temperature and, hence, self-assemble ( Figure 3B). RLPs on the other hand exhibit an upper critical solution temperature, meaning that they become soluble with increasing temperature. This J o u r n a l P r e -p r o o f produces a system in which the tunable geometry and solubility of the two blocks compete against one another to determine self-assembly conditions and assembled morphology. Studying the structures formed in response to a temperature increase allowed for the isolation of structures in which the ELP formed the hydrophobic block. Increasing the size of the corona-forming block in these systems led to a transition from non-spherical micelles to spherical micelles, similar to more traditional amphiphilic block copolymer systems ( Figure 3C) [25,110]. Deviations from traditional behaviour were seen elsewhere, in which the tuning the hydrophobicity of an oligopeptide attached to an extremely large AGVPG ELP led to a micelle-to-worm transition, despite a negligible change in [111]. As will be discussed in Section 4.3, ELP-based micelles have been more successfully translated into applications than the structures discussed so far.
Perhaps the standout examples of molecular engineering in synthetic peptide self-assembly are peptide amphiphiles (PAs, Figure 3D). At the simplest level, these molecules comprise an oligopeptide modified with a hydrophobic, lipid or lipid-like tail, thus combining the structural features of amphiphilic surfactants with the functions of bioactive peptides [112][113][114]. In a landmark work, Hartgerink et al. developed a PA that self-assembles into nanofibers based on a molecular design comprising 4 domains: (1) a hydrophobic alkyl tail, (2) charged amino acids to increase solubility, (3) a β-sheet forming peptide sequence, and (4) small bioactive sequences ( Figure 3D, i) [91]. A reduction in pH was found to drive the self-assembly of the PAs into nanofibers with a uniform diameter of 7 nm, while an increase in pH led to dissolution of the fibers. Extensive variations on the design of Hartgerink et al. can be found in the literature. Generally, a PA's backbone is composed of 8-30 amino acids with a hydrophilic block, formed by polar amino acids, and a hydrophobic block with apolar amino acids or grafted alkyl, acyl, or aryl lipidic tails [115][116][117].
In some cases, PAs include a linker segment in between these two blocks, such as 2 kDa PEG or glycine, and a bioactive sequence that recognises targets in cells or tissues [113]. The interplay between hydrophobic, electrostatic, π-π interactions, and hydrogen bonding drives the selfassembly of PAs into wide array of well-defined nanostructures, including, micelles, vesicles, nanotubes, nanofibers, and nanosheets ( Figure 3D, ii) [112,113]. As will be seen throughout Section 4, these structures show great promise in treating a number of brain disorders, however the rules describing the structures and given PA will self-assemble into remain poorly understood.
Therapeutics can then be encapsulated within these NPs via electrostatics, preferential partitioning, or chemical attachment. Indeed therapeutics can even form the building blocks of polyplexes, such as in the pegylated dendrigraft poly(Lys) (DGL-PEG) NPs of the Jiang group that appear throughout the Section 4 [119]. Making polyplex NPs from diblock copolymers, in which one block is a polypeptide and another block is an uncharged synthetic block such as PEG, adds geometrical constraints that can produce spheroidal and hollow spheroidal structures similar to those produced by simple amphiphiles. The extensive body of work from the Kataoka group on the self-assembly of negatively charged PEG-b-P(Asp) and positively charged PEG-P(Asp-XX), where XX refers to a cationic modification to the aspartic acid block, is particularly significant here (Figure 4). Koide et al. first reported that these pairs of oppositely charged block copolymers undergo controlled self-assembly into hollow vesicular structures referred to as polyion complex polymersomes (PICsomes) [120].
Anraku et al. later found that, remarkably, the size of the self-assembled vesicles varied linearly with polymer concentration [121]. Wibowo et al. reported that the geometry of the PICsomes depended on both the temperature of self-assembly and the relative size of the PEG block [122], however no quantitative relations between these factors could be established. As with other polyplex NPs the materials from which PICsomes are made can be readily varied, such as the siRNA and PEG-b-P([N-(5-aminopentyl]-, -aspartamide] (PEG-b-P(Asp-AP)) PICsomes of Kim et al ( Figure 4B) [123].
Micelle-like peptide polyplexes are also often reported [57]; however, this name is possibly a misnomer as the core of these structures is hydrophilic and it is not clear whether the polymer blocks are as highly segregated as one would expect in a true micelle. Semantics aside, these peptide polyplex 'micelles' can encapsulate a range of drugs and the structures themselves are highly amenable to modification. This tunability makes them ideal for fine-grained studies of the effects of functional group density upon NP interactions with biological tissues [57]. As we will see in Section 4, peptide polyplex NPs have a demonstrated efficacy in brain delivery and benefit from being extremely simple to produce. However, a lack of understanding of the underlying physical chemistry that governs their self-assembly means that systematically engineering polyplex NPs is still a process of guess work.

Rational Design is Possibly Not the Best Approach to Engineer Peptide Self-Assembly
The majority of efforts to apply a reductionist approach to peptide self-assembly fail. Indeed, multiple recent commentaries have argued that evolutionary methods, both in silico and experimental, may be more effective ways to study the self-assembly of complex molecules ( Figure   5) [124,125]. Simulations-aided materials discovery is the oldest and most obvious of these approaches. Frederix et al. used coarse-grained molecular dynamics simulations to map out the selfassembly behaviour of all possible di-and tripeptides formed from the 20 proteinogenic peptides ( Figure 5A) [126,127]. Pleasingly, good agreement with experiment was observed, and only a weak correlation was observed between the hydrophobicity of a tripeptide and its propensity to self- between peptides, as well as detailed investigation of the effects of peptide terminus acetylation, which is widely observed to impact peptide assembly [131,132].

Figure 5. Evolutionary and computational approaches to peptide design.
A) Appropriate coarsegraining and force-field parameterisation allows for the systematic study of the self-assembly behaviour of a broad palette of short oligopeptides. Reproduced with permission [127]. Copyright 2014 Springer Nature. B) Studying a solution containing a mixture of dipeptides in the presence of a relatively non-specific protease allows the system to probe its self-assembly free energy landscape without external intervention. Reproduced with permission [133]. Copyright 2016 Springer Nature. C) Home-made experimental setups driven by a genetic algorithm allow for the optimisation of nanoparticle synthesis on the timescale of days without human intervention. Reproduced under the terms of the CC-BY 4.0 license (https://creativecommons.org/licenses/by/4.0/) [134]. Copyright 2020 the Authors.
Self-assembly studies of large ensembles of longer peptides are beyond the reach of current simulations approaches and will likely remain so for some time; experimental approaches suffer from no such limitation. A dynamic combinatorial library approach, in which peptide combinations can be tested in situ and promising candidates either amplified or isolated are a powerful method for studying large numbers of permutations of peptides [135][136][137]. Pappas et al. applied these methods to study the self-assembly of unprotected molecules up to 8 residues long ( Figure 5B) [133]. Mixtures of Leu, Phe, Trp, Ser, and Asp dipeptides were investigated in the presence of thermolysin, a relatively non-specific protease that does not cleave terminal peptide bonds. As such, dipeptides were stable in the presence of thermolysin, while oligopeptides were continuously formed and broken apart. Self-assembly into more thermodynamically stable structures retarded thermolysin-catalysed hydrolysis, meaning that oligopeptides with a greater propensity for selfassembly were inherently amplified with this technique. The system is naturally sensitive to initial conditions; different combinations of input dipeptides resulted in convergence upon different final structures. Diphenylalanine was found to evolve from a free-flowing tubular dipeptide assembly into a gelled dispersion of short rods comprising 80% F 6 after 15 days. By contrast, ditryptophan exhibit limited molecular transformation, with only around 20% of W 2 converting into larger oligomers.
Combinations of FS and FD converged up to a single, highly amplified FDFSFDFS oligomer, reminiscent of the alternating sequences of charged/polar and hydrophobic motifs associated with -sheet formation [138,139].
The method of Pappas et al. only allows peptide self-assembly to be optimised towards a single goal: thermodynamic stability. It is not obvious how one would target a specific function or structure this way, nor are the most useful structures necessarily the most thermodynamically stable. Autonomous exploration of parameter space using a genetic algorithm allows self-assembly to be engineered towards an arbitrary goal. Furthermore, to our knowledge, it is an entirely neglected area within peptide NP self-assembly, although it has long been used in related disciplines [125,[140][141][142][143]. To illustrate the usefulness of this technique, we turn to work from the inorganic NP community. Salley et al. used an autonomous experimental platform made entirely from commercially or freely available parts to study gold NP synthesis ( Figure 5C) [134]. With each iteration, a genetic algorithm sought reaction conditions that maximised the value of a user-defined fitness function. Within 10 generations, often significantly fewer, the setup had converged upon a J o u r n a l P r e -p r o o f synthesis method for rods and spheres of low polydispersity and was even capable of being set an arbitrary goal, yielding novel, faceted structures. The versatility of this method is such that it could be readily and affordably applied to peptide NP self-assembly. Peptide NPs could then be optimised for any number of goals such as particle shape, size or polydispersity, target-receptor binding, or all of the above.

Transport to and Across the BBB
Negotiating the biological barriers that regulate transport of molecules through the body and into the brain is one of the biggest open problems in drug delivery today. The CNS functions are maintained by the meticulous coordination of the activity of multiple cells within a neurovascular unit (NVU), including vascular cells (endothelial cells, pericytes, and smooth muscle cells), glia (astrocytes, oligodendrocytes, and microglia) and neurons [144][145][146]. Within the NVU, the continuous non-fenestrated endothelium lining on the brain vasculature forms the BBB, which acts as a barrier that separates blood and brain compartments, and strictly regulates blood-to-brain and brain-to-blood transport of molecules [145]. BECs are tightly connected by tight junction proteins, such as claudin-3, -5, and -12, occludin, and ZO-1, -2, and -3, which impair exchange of molecules through the paracellular route (i.e., transport through the intercellular space between cells) between blood and brain. BECs are sheathed by other cells of the NVU, including pericytes and vascular smooth cells, and astrocyte end-feet [147,148]. Pericytes shelter 60-70% of the basal endothelial surface, while astrocyte end-feet reach up to 99% of the surface, overlapping pericytes and contributing to the barrier properties of the BECs [147][148][149]. In contrast to the peripheral endothelium, BECs exhibit a low rate of transcytosis regulated by specific proteins [150,151] which, together with the sealed cell-to-cell contacts of the endothelium, restrict the entry of blood-derived molecules into the brain. Apart from small lipophilic molecules (< 400 Da) with a polar surface area (< 60-70 Å) and weak hydrogen-bonding potential (< 6 hydrogen bonds), which may cross the brain endothelium by diffusion [49], other molecules require specialised carriers or receptors in the apical surface of BECs to facilitate their transport across the BBB. In the following section, we describe carrier-, receptor-and adsorptive-mediated transport across the BBB (Figure 6), specifically for shuttling of peptides and their use by the community to transport peptide-based NPs. Water-soluble molecules are transported through the intercellular space between the BECs (paracellular transport). Carrier proteins on the apical membrane surface of BECs facilitate the transport of specific substrates, such as glucose and small peptides (carrier-mediated transport, CMT). Adsorptive-mediated transport (AMT) appears to be triggered by interaction of polycationic peptides with anionic components at the surface of the BECs. Receptor-mediated transport (RMT) facilitates the transport of a variety of macromolecules (such as, peptides and proteins) by binding to a specific receptor, followed by intracellular trafficking via endosomal and, possibly, tubular sorting, and exocytosis.

Carrier-mediated Transport
Carrier-mediated transport (CMT) refers to transport mediated by a membrane carrier protein.
PTS-1 and -2 facilitate an efflux of Tyr-MIF-1/enkephalins and AVP from brain to blood, respectively.
PTS-3 mediates unidirectional transport of peptide T analogue from blood to brain [157,159], while PTS-4 transports LHRH bidirectionally [158]. Although the transport of these small neuropeptides J o u r n a l P r e -p r o o f was extensively investigated in the 1980s, a comprehensive study is still missing to fully characterise PTS at the BECs. Apart from PTS, the organic anion transporting polypeptide-2 (OATP-2; SLC21A3) is strongly expressed at the BBB in humans, and findings suggested that OATP-2 mediates transport of opioid peptides, such as [D-penicillamine-2,5] enkephalin and deltorphin (II), across the BBB [160].
Glutathione is a tripeptide (CEG) and, similar to other small peptides, a saturable and specific transporter has been identified for transport across BECs [161,162]. Among these carrier proteins expressed at BECs, glutathione has been explored for drug delivery into the brain (G-Technology ® , a PEGylated liposomal formulation for delivery of systemically administered therapeutics, 2BBB, Leiden, Netherlands). Although not fully characterised, the glutathione transporter is the main endogenous carrier protein considered for CMT of peptides across BECs [163]. Apart from these carriers, which mediate the transport of small peptides, other carrier proteins have been widely studied to allow the BBB crossing by means of ligand-installed peptide-based NPs, including GLUT1, choline and nicotinic acetylcholine receptor (nAchR) ( Table 1). of glycaemic modulation, in which fasting was used as an external trigger to provoke rapid glycaemic increase to then trigger GLUT1 to undergo transcytosis [57]. In this work, PIC micelles self-assembled from the oppositely charged block copolymers PEG-P(Asp) and PEG-poly([5-aminopentyl]-α,βaspartamide) ( 30 nm diameter) were functionalised with varying densities of glucose (% glucose = 10, 25 and 50) by introducing different amounts of Gluc(6)-PEG-P(Asp) in the self-assembly.

J o u r n a l P r e -p r o o f
Interestingly, glucose density within the NPs played a key role in directing NPs towards different brain cells. NPs bearing 25% glucose were mostly found in neurons and microglia while NPs containing the highest density (50%) accumulated within the brain capillary walls. In a more recent study, the authors delivered antisense oligonucleotides (ASO) to the brain following this on-off switch glycaemic control strategy [12]. PEG micelles of 45 nm in diameter consisting of positively charged PEG-P(Lys) were modified with 3-mercaptopropyl amidine, 2-thiolaneimine, and negatively charged ASO. Different glucose amounts, ranging from 0 to 103 ligands per particle, were produced by using Gluc (6)-PEG as initiator for the ROP of Lys(TFA)-NCA. PIC micelles exhibited half-lives of circulation in blood of > 80 minutes regardless of glucose density. However, NPs bearing 52 units of glucose showed by far the highest brain accumulation after IV injection (6.9 ± 1.9% injected dose (ID)/g of brain, 17-fold increase compared to the naked NPs) with a wide distribution within the brain, and the highest amount observed in the cerebral cortex (26 ± 7% of accumulation amount within the brain) (Figure 7). These results underscore the importance of both binding avidity and affinity and correlate well with recently reported models [164]. The optimised 52 units glucosebearing PIC micelles were employed to deliver a model ASO and showed efficient and consistent knockdown in various brain regions. Additionally, pH-and reduction-sensitive GLUT-1 targeting PIC micelles constructed from PEG-P(Lys) modified with 3-(2-pyridyldithio)propionate were also recently employed by the Kataoka group to deliver the amyloid-beta (Aβ) aggregation inhibitor 3D6-fragment antigen-binding (Fab) [13]. These PIC micelles achieved 41-fold increased brain accumulation compared to the free Fab and improved Aβ aggregation inhibition when multiple doses were applied to an Alzheimer's disease (AD) mouse model. employed to co-deliver an encoding plasmid open reading frame of human tumour necrosis factorrelated apoptosis-inducing ligand (pORF-hTRAIL pDNA) and DOX for the treatment of glioma [166].
NPs showed increased brain tumour accumulation, enhanced apoptosis in vivo, and superior antitumor efficacy when tested in glioma-bearing mice with prolonged median survival time (almost double compared to saline group and at least 1.2 times better than single therapy groups).
Another ligand to trigger CMT is the rabies virus glycoprotein peptide (RVG29), a 29 aminoacid peptide originated from the rabies virus glycoprotein that specifically binds to the nAchR. By using DGL-PEG-RVG29 NPs carrying a caspase-3 short hairpin RNA (shRNA) encoding plasmid DNA, the level of activated caspase-3 was reduced in a Parkinson's disease (PD) animal model with an increase in the rescue of dopaminergic neurons [167]. Consequently, PD rats treated with these nAchR-targeting NPs showed an enhancement of locomotor activity compared to rats treated with pristine NPs. Similar DGL-PEG-RVG29 NPs were employed to deliver a combination therapy consisting of a therapeutic gene and peptide for AD [168]. Using a gene therapy towards ß-secretase 1 (BACE1) and a specific D-amino acid based-peptide to target Aß accumulation and phosphorylated Tau, respectively, DGL-PEG-RVG29 NPs successfully reduced Aß plaques as well as neurofibrillary tangles in the cortex and hippocampus of an AD mouse model, which resulted in an improvement of memory.

Receptor-mediated Transport
Large peptides and proteins, such as transferrin (Tf), low density lipoproteins, and insulin, rely on receptor-mediated transport (RMT) through specific receptors at the surface of BECs for transcytosis across the brain endothelium [169,170] ( Table 2). Transcytosis via RMT includes four steps: 1) a circulating ligand binds to a cognate receptor expressed on the apical membrane of BECs (e.g., Tf binds to transferrin receptor, TfR); 2) endocytosis takes place through membrane invaginations and eventually the formation of an intracellular vesicle containing receptor-ligand complexes; 3) intracellular trafficking occurs through vesicular and/or vesicular-tubular structures; and then 4) exocytosis occurs with the release of the vesicular content into the brain parenchyma [171]. In BECs, the initial step of transcytosis -endocytosis -is mainly facilitated by clathrin-mediated endocytosis (CME). Conceptually, CME is a fairly simple process that consists of a few sequential and partially overlapping steps [172]. CME is initiated by the clustering of endocytic coat proteins on the inner leaflet of the apical membrane (initiation), which is further continued by the recruitment of other coat proteins (recruitment). Consequently, the assembly of these coat proteins prompts membrane bending, which transforms a flat membrane into a "clathrin-coated pit" (membrane bending). Scission constricts and cuts the neck of the invagination to separate the clathrin-coated J o u r n a l P r e -p r o o f vesicle from the apical membrane. In mammals, this scission step is mediated by dynamin and Bin/Amphiphysin/Rvs (BAR)-containing proteins, including amphiphysin and endophilin. Finally, uncoating disassembles the endocytic protein machinery, releasing the receptor-ligand complexfilled vesicle ( 100 nm in diameter) for further intracellular trafficking [172]. So far, multiple receptors have been shown to undergo CME at the BBB, such as TfR and insulin receptors (IR), and also low density lipoprotein receptor-related protein 1 (LRP1) [173,174].
Notwithstanding the prevalence of CME, other clathrin-independent endocytic mechanisms regulate transcytosis at the BECs, such as caveolin-mediated endocytosis. Caveolae are small flaskshape organelles (60 to 80 nm) containing oligomeric caveolin-1 [175]. In BECs, caveolae assembly at the membrane is negatively regulated by the major facilitator superfamily domain-containing 2a (Mfsd2a), which maintains the integrity of the BBB [150,151].

Transferrin Receptor
TfR is abundantly expressed by BECs to mediate transport of iron into the brain [182]. A cognate ligand for the TfR is the iron binding protein Tf, which is a transmembrane glycoprotein consisting of two 90 kDa subunits, linked by intermolecular disulfide bonds and with each subunit binding to one molecule of Tf [183]. Although Tf is a specific ligand to TfR, Tf is a questionable targeting moiety for drug delivery, mainly due to the fact that only a small amount of Tf is transcytosed across the BECs [184,185]. Thus, monoclonal antibodies against the TfR, which bind to epitopes on the extracellular domain of TfR distal to the Tf binding side circumventing competition with endogenous Tf, have been widely developed for RMT-based delivery [186,187]. OX26 (murine monoclonal antibody against rat TfR), 8D3 and R17-217 (murine monoclonal antibodies against mouse TfR) exhibited brain uptake in mice [188]. Both 8D3 and R17-27 antibodies exhibited high transport, with brain uptake of 3.1 and 1.6% of the ID/g, respectively, while OX26 uptake was 25 to 50 times lower, with only 0.06% ID/g [188]. Based on these findings, TfR antibodies have been explored for brain delivery by conjugation to a therapeutic cargo, such as brain-derived neurotrophic factor (BDNF) [189], basic fibroblast growth factor (bFGF) [190], and nerve growth factor (NGF) [191], or by genetically engineering chimeric proteins for tumour necrosis factor receptor (cTfRMAb-TNFR), anti-Aß antibodies (cTfRMAb-ScFv), glial cell line-derived neurotrophic factor (cTfRMAb-GDNF), and J o u r n a l P r e -p r o o f erythropoietin (HIRMab-EPO) [192]. However, few studies investigating the transport of anti-TfR monoclonal antibodies have demonstrated that, despite the substantial binding to BECs, the transcytosis into the brain parenchyma is limited with TfR antibodies accumulating in the BECs upon endocytosis [193][194][195]. This accumulation occurs possibly either due to a lack of antibody dissociation from the TfR upon endocytosis or because the intracellular trafficking of the TfR is affected by the binding interaction with the antibody.
A recent series of studies suggests that engineering Tf antibody binding properties (affinity and avidity) affects intracellular trafficking and, consequently, transcytosis [181,[196][197][198]. By engineering two constructs with a single chain (sc) Fab fragment of an anti-TfR monoclonal antibody fused either to one (sFab) or both (dFab) C-terminal ends of the heavy chain of an anti-A monoclonal antibody, it was shown that the monovalent binding (sFab) increases Aß target by 55fold compared to the bivalent (dFab) antibody [197]. In vitro and in vivo evidence demonstrated that monovalent binding mode (sFab) facilitates transcellular transport, whereas bivalent mode (dFab) leads to lysosomal degradation. In a later study, it was further shown that a monovalent TfR-based construct (sFab) is sorted for transcytosis through intracellular tubules while the bivalent construct (dFab) transport is impaired by cellular sorting into lysosomes [181]. Apart from these monoclonal antibodies, peptides interacting with TfR were found by phage display biopanning, including B6 peptide [199], 9-mer CRT peptide (CRTIGPSVC) [200], and 7-mer HAI or T7 (HAIYPRH) and 12-mer THR (THRPPMWSPV) peptide [201,202]. Due to practical reasons, the use of small peptides as targeting moieties in NPs to promote BBB entry, is preferred over conjugation of large antibodies.
T7, was for instance, employed by Liu et al. to co-deliver to the brain pORF-hTRAIL pDNA and DOX using DGL-PEG [203]. DOX was conjugated through a pH-labile hydrazone linker for pH-dependent release. Combination-bearing T7-NPs of 170 nm diameter in size and 3 mV -potential showed synergistic effects (in vitro and in vivo) and increased in almost twice the survival time of braintumour bearing mice when compared with any of the tested controls. DGL-PEG-T7 NPs efficiently delivered RNA interference (RNAi) to several brain areas (cortical layer, caudate putamen, hippocampus, and substantia nigra) for glioma treatment [204]. Gene silencing experiments revealed a more than 2-fold increase in in vivo gene silencing of T7-NPs compared to pristine NPs when tested in an orthotopic human glioma mice model using U-87 MG-luciferase cells. Apart from these studies using polypeptide nanocarriers, in a recent study, Wu et al. engineered an amphiphilic derivative from the GYR peptide (GYRPVHNIRGHWAPG), identified by phage display biopanning, that shows specific binding to BECs [205]. By adding a cysteine to induce cross-linking (via disulfide bridge formation) between two adjacent peptides and a fluorophore to the amino group of the cysteine to initiate π-π stacking among conjugates to the GYR peptide, the resulting amphiphile (CGY) self-J o u r n a l P r e -p r o o f assembled into fiber-like structures [206]. In vitro studies showed that CGY PA interacts with BECs through TfR and RAGE, while intravenous injection of Cy5.5-CGY in mice resulted in targeting of the brain with a peak of the levels at 4 hours (5.7% of ID) [206]. Furthermore, CGY-BACE1-siRNA complexes showed an effective BACE1 down-regulation in the brain without toxicity and inflammation.

Leptin Receptor
Leptin receptor (LepR), or OB-R, is a single transmembrane glycoprotein belonging to the family of cytokine receptors. There are two isoforms of LepR: the short (OB-Ra), which is present in the brain endothelium, and the long one (OB-Rb) [207]. A cognate ligand for LepR is leptin, a protein (16 kDa) that regulates adipose tissue activity and appetite by acting in the CNS (i.e., hypothalamus), after secretion into the blood circulation by adipose tissues. At the BECs, leptin is transcytosed via the LepR (OB-Ra) [208,209], and consequently LepR targeting has been explored through the use of

Low-density Lipoprotein Receptors
The low-density lipoprotein receptor (LDLR), a single transmembrane glycoprotein that recognises low-density lipoprotein (LDL) particles, and LDLR-related proteins (LRP1 and LRP2) are abundantly expressed at BECs to mediate the transcytosis of lipoproteins and other ligands [212]. Specifically, LRP1 is a multifunctional receptor that interacts with a range of ligands, such as apolipoprotein E (ApoE), α 2 -macroglobulin, aprotinin, amyloid precursor protein, and Aß [213]. One of the first examples of targeted gene delivery to the brain by means of P(Lys) nanocarriers was reported by Mousazadeh and co-workers using this receptor to shuttle them across the BBB [214]. An LDL receptor-binding peptide dimer (LRKLRKRLLR-LRKLRKRLLR) derived from its endogenous ligand ApoE was conjugated to P(Lys) 16 . This P(Lys) 16 NP system was used to complex a reported DNA plasmid encoding for β-galactosidase (β-Gal). In this study, although a proof of concept was obtained with an J o u r n a l P r e -p r o o f Journal Pre-proof increased expression of β-Gal in the brain (increased by 600 times by using 30 ng/mg of NPs), brain targeting was lost due to a high level of expression of β-Gal in other organs. A variation of P(Lys) 16 non-viral vector functionalised with a different ApoE-derived sequence (LRVRLASHLRKLRKRLLRDA) has been successfully used for the delivery of an array of both small drugs and proteins to the brain via intravenous (IV) injection [215][216][217][218]. However, the development of these P(Lys) 16 NPs has been hampered by a peptide dose-dependent toxicity associated to an unexpected inhibition of acetylcholinesterase, which still remains to be fully understood [219]. , for the treatment of brain and brainrelated disorders. In particular, the ANG1005 is currently in Phase II clinical trial for the treatment of recurrent brain metastases [222,223]. Therefore, angiopep-2 is, possibly, the most exploited LRP1ligand for the design of nanocarriers targeting the brain. In one example concerning peptidic nanocarriers, angiopep-2 was attached to DGL-PEG [224]. Spherical DGL-PEG NPs ( 120 nm in diameter) were formed from condensing the neuroprotective gene encoding human GDNF. Interestingly, our recent study demonstrated that avidity of angiopep-2-functionalised nanocarriers affects intracellular trafficking and ultimately transcytosis across BECs [164]. A high number of angiopep-2 peptides at the surface of the nanocarriers (polymersomes) directs LRP1 towards an endosomal sorting and lysosomal degradation, while a mid-avidity favours the transport across BECs through syndapin-2-stabilised tubular structures.

J o u r n a l P r e -p r o o f Journal Pre-proof
Apart from angiopep-2, L57 peptide (TWPKHFDKHTFYSILKLGKH-OH) was identified by phage display biopanning for the targeting of LRP1, showing significant brain uptake in mice [225]. The receptor-associated protein (RAP) is a chaperone protein that, tightly binds to LRP1, and participates in folding and trafficking of LRP1. Pan et al. demonstrated that RAP is relatively stable in blood for 30 minutes, and permeates intact across the BECs (0.5-1% ID/g in the mouse brain, 30 minutes after IV injection) [226]. Recently, using computer-aided design, a short RAP12 (EAKIEKHNHYQK, amino acids 251-262 of RAP) peptide was identified for LRP1-mediated transport across the BBB [227].

RAGE
RAGE, a transmembrane protein of the immunoglobulin superfamily, is a multiligand receptor, which binds to advanced glycation end products (AGEs) (i.e., adducts from non-enzymatic glycation of proteins and lipids), S100/calgranulins, HMGB and Aß protein [228]. At the BBB, RAGE mediates the transport of Aß from blood-to-brain contributing to the formation of Aß plaques [229]. In the Aß peptide, the major binding site to RAGE is localised to an eight amino acid stretch of residues at position 16-23 (KLVFFAED) -a series of hydrophobic residues flanked by two negatively charged amino acid residues at the C-terminal [230]. In addition to the 16-23 peptide, the RP1 peptide (APDTKTQ), identified by phage display biopanning with homology to 16-23 regions of A shows a high-affinity to RAGE [231]. Due to its function in Aß transport, this receptor has been widely explored to shuttle NPs towards AD treatments. In one example, KLVFFAED-tagged PEG-P(Lys)-based micelles with reactive oxygen species (ROS)-responsiveness and scavenging ability provided by the phenylboronic groups attached to the polymer backbone, were used to encapsulate the AD multitarget drug curcumin [232]. Curcumin targets several pathological hallmarks of AD by inhibiting/disrupting A plaques formation [233,234], clearing phosphorylated Tau [234], inhibiting acetylcholinesterase (AChE) [235], acting as antioxidant activity for oxidative stress [236] and modulating several steps of the inflammatory cascade [237]. The peptide KLVFFAED [238] was

Targeting Alternative Receptors at the BBB
Although not-BBB specific, other vascular receptors have been used to deliver peptide NPs to the brain. The widely explored RGD peptides, which target αVβ3 and αvβ5 integrins, have been used in conjunction with polypeptide nanocarriers for brain delivery. RGD peptides have a selective affinity for the integrins overexpressed in the endothelial cells of tumour angiogenic vessels [14].

J o u r n a l P r e -p r o o f
Thus, conjugation of RGD to nanocarriers appears as a way to cross the brain-blood tumour barrier for both imaging and treatment of GBM. The Kataoka group have employed cRGD-installed PEG-P(Glu) micelles to entrap a platinum (II) analogue, DACHPt, via complexation with carboxyl groups for treatment of GBM [14]. Different ligand densities were examined with the 20% cRGD containing micelles showing a higher level of brain accumulation ( 6 % ID/ g in the tumour after 10 hours of IV administration) and a significant decrease in tumour volume compared to saline (10-fold decrease in tumour volume). Recently, these cRGD-PEG-P(Glu) were also used for delivery of epirubicin for treatment of GBM [47]. Epirubicin was attached by post-polymerisation of modified PEG-P(Asp) polymer backbone through pH-sensitive hydrazine-linker. cRGD-epirubicin micelles showed a 12-fold increase in anti-tumour activity in an orthotopic GBM model compared to pristine micelles.
Conjugation of cRGD was further employed to image neovasculature in an orthotopic model of GBM [239]. PIC vesicles (PICsomes) formed by PEG-P(Asp) and the homocatiomer n-butyl-poly( studies are yet lacking. The BBB plays a critical role in maintaining suitable brain folic acid (FA) concentrations (1.5 to 3-fold increase compared to the serum) through its active transport via folate receptor (FR).
Although widely expressed, FR is upregulated in numerous solid tumours, including GBM, and neovasculature. Based on this, folate has been used as a ligand to functionalise nanocarriers for the transport of cargo to the brain. Du Chen and co-workers used FA to co-deliver DOX and BCL-2 (an anti-apoptotic protein) siRNA as a way to overcome drug-resistance in the treatment of glioma [240]. FA-PEG-P(Glu) was used to coat self-assembled NPs from linear (PEI)-poly(e-caprolactone) (PCL) PEI-PCL pre-loaded with siRNA and DOX [241]. FA-functionalised NPs effectively delivered BCL-2 siRNA and sensitized C6 rat glioma cells to chemotherapy both in vitro and in vivo, showing greater efficacy almost suppressing tumour growth and showing extended median survival times [241]. A dually targeting polymersome formulation of DOX was designed by Chen and co-workers [242]. Polygamma(Glu)-based NPs with backbones decorated with FA and des-octanoyl ghrelin, a 28 amino acid J o u r n a l P r e -p r o o f endogenous ligand known to transport from blood-to-brain direction via des-octanoyl ghrelin binding sites RMT [243,244], demonstrated an enhanced BBB crossing in vitro and enhanced antitumour effects compared to controls.
Several infectious pathogens such as prions, bacteria or some neurotropic viruses are able to reach the brain via laminin receptor binding [245][246][247]. Laminin is the major structural component of the basal membrane at the extracellular matrix, and its receptor abundantly expressed in adult neurons and glial cells has demonstrated essential roles in tumour invasion and metastasis [247,248]. To target laminin receptor, DGL-PEG nanocarriers were decorated with a Streptococcus pneumoniae-derived peptide EPRNEEK [119]. EPRNEEK-DGL-PEG NPs containing a plasmid DNA encoding for luciferase exhibited increased BBB translocation and accumulation within a glioma tumour compared to pristine NPs.
Yi et al. designed an amphiphilic R 3 V 6 peptide, which self-assembled into small NPs, for the delivery of vascular endothelial growth factor (VEGF)-siRNA and carmustine to C6 glioma cells [249].
It was shown that carmustine is efficiently entrapped into the hydrophobic core of the micelles, and that the positively charged R 3 V 6 peptide forms a stable complex with the VEGF siRNA. However, this study offered no details regarding in vivo efficacy of the VEGF-siRNA/carmustine-PA complexes and targeting to the BECs or tumour. To target the tumour vasculature, CREKA peptide was employed for an active binding to fibrin in tumour vasculature. Fluorescently labelled Cy7-PEG-CREKA micelles, when administered intravenously in an intracranial GL261 glioma mouse model, exhibited accumulation at the tumour site within 1 hour [250].

Adsorptive-mediated Transport of peptides
Adsorptive-mediated transport (AMT) is a mechanism of transport triggered by the interaction of polycationic peptides to anionic components, including carboxylic acid groups of sialoglycoproteins and sialoglycolipids and sulphate groups of heparan sulphate proteoglycans, at the surface of BECs [251]. Although AMT is described in drug delivery for the brain, the mechanism of intracellular trafficking and, consequently, transcytosis is not well-understood. In particular, it is established that polycationic species interact strongly with plasma proteins such as albumin, fibrogen, and immuneglobulins, most of which are negatively charged, forming the so-called protein corona [252,253]. As such, rather than being due to electrostatic interactions, many of the transport processes attributed in the literature to this adsorptive mechanism may rather be driven by the protein corona via an undisclosed RMT pathway.
Regardless of the precise details of the underlying transport mechanism, cationic albumin, basic polypeptides (including, ebiratide, 001-C8, E-2078, histone, avidin, wheat germ agglutinin and J o u r n a l P r e -p r o o f protamine), and cell penetrating peptides (CPPs) have been reported in the literature as undergoing AMT-based delivery into the brain [251]. In particular, CPPs are short peptides (< 30 amino acids in length), which are classified according to their physicochemical properties; cationic, amphipathic or hydrophobic [254]. The cationic class consists of peptides abundant in positively charged amino acids (Arg and Lys), such as peptides derived from human immunodeficiency virus (HIV) transactivator of transcription (TAT) protein (TAT 48-60 , TAT 49-57 ), penetratin, polyarginines and DPV1047. The majority of reported studies exploiting CPPs as delivery vectors for transport across the BBB include TAT. However, additional CPPs have shown potential for brain delivery, such as dNP2, that was used in the delivery of the cytotoxic T-lymphocyte antigen for autoimmune encephalomyelitis [255].
The CPP platform has been combined with a singular class of polyamino acids used in the delivery of therapeutics, the ELPs discussed in Section 3.1 [256,257]. Their relevance to the field of drug delivery arises from their convenient LCST (40-41ºC), which allows for the application of mild hyperthermia, a mature clinical modality currently used in clinics [258]. In one example, Hearst et al.
used ELPs fused to Synb1 CPP to demonstrate that external thermal cycling increased localisation of fluorescent-labelled Synb1-ELP into the brain, particularly, in the cerebellum after IV injection [259].
The group used the tandem Synb1-ELP to deliver a Notch inhibitory peptide, dominant negative MAML (dnMAML) [260] or DOX through a pH sensitive hydrazine linker [261] for the treatment of

5-FAM-CGYRPVHNIRGHWAPG
TfR and RAGE BACE1 siRNA Alzheimer's Disease [206] J o u r n a l P r e -p r o o f

Localised Delivery
Although considered an invasive and aggressive method, FDA approval of Gliadel wafers in 1997 demonstrated the efficacy and safety of local delivery of therapeutics to the brain [266]. In one example, Abid Sheikh and co-workers [267] used PEI-P(Lys) to complex and deliver VEGF pDNA to PD holding rats through direct administration to the brain by stereotactic brain surgery. These polyplex NPs exerted great transfection efficiencies both in vitro and in vivo and the delivery of VEGF decreased neuronal loss, reduced apoptosis events and microglia activation, and therefore prevented motor deficits in 6-hydroxydopamine (6-OHDA) induced rat PD model. The Kataoka group has also explored the ICV route for the delivery of therapeutics to the brain using polyplex nanomicelles. In one study, the polycation PEG-P(AspDET), a PEG-P(Asp) polymer carrying diethylenetriamine (DET) at the side chains, was used to produced pH-sensitive micelles of 50 nm in diameter and core-shell structure [268]. DET moieties with a two-step protonation behaviour at a s of 9.5 and 6.0 were introduced to provide pH-triggered destabilization of the cell membrane and endosomal escape [269]. Delivery of messenger RNA (mRNA) by ICV infusion into mice brains using these polyplex micelles resulted in sustained protein expression in cerebrospinal fluid (CSF) and overcome the two main limitations of mRNA, its stability and immunogenicity. mRNA encoding the mouse neprilysin protein (NEP), a protease that degrades Aβ [270] was efficiently introduced in mice brains by ICV using these polyplex micelles. Mice pre-treatment with NEP mRNA micelles increased NEP concentration and almost doubled its proteolytic activity in brain and diminished almost to half the amount of exogenous Aβ 1-40 injected thereafter as compared to other groups. The authors recently reported an upgrade of these polypeptide nanocarriers by fine-tuning the hydrophobicity (logP) of the P(Asp) residues in the side chains [271]. A direct correlation among logP and polyplex stability and cell internalisation efficiency was found.
J o u r n a l P r e -p r o o f

Convection-Enhanced Delivery
Convection-enhanced delivery (CED) is an alternative method that bypasses the BBB entirely, potentially greatly increasing delivery of therapeutics [272]. This technique is based on the direct pumping of therapeutics to the brain by means of a pressure gradient at the tip of an infusion catheter. Since small drugs usually fail to distribute within the tissue, the combination of CED and nanomaterial-based delivery systems has demonstrated great potential in dealing with brain tumours [273]. Inoue and co-workers applied CED to delivery PEG-P(Asp)-DOX conjugated micelles in rat brains with 9L (gliosarcoma cells) intracranial tumours [274]. DOX-bearing micelles infused by CED achieved significantly wider distribution both in brain tumour and surroundings and exhibited enhanced antitumor effects as reported by an almost twice increased median survival time compared with free DOX. The same authors used PEG-P(benzyl-Asp) to deliver Am80 [275], a synthetic agonist of the nuclear retinoid acid receptor (RAR) highly expressed in glioma cells, reported to inhibit the growth rate of several cancers [276]. Combination of CED of the micellar The Hanes group studied the influence of NP size and PEG corona density in brain diffusion of NPs [279]. While NPs with low density PEG coating showed adhesive interactions with the brain extracellular matrix regardless of size, surface shielding with a dense PEG corona enhanced diffusion in the brain of 40 and 100 nm NPs, but not 200 nm NPs [280]. These results defy the convectional opinion that only NPs smaller than 64 nm access the brain parenchyma through the extracellular space [281]. Based on this knowledge, biodegradable PEGylated P(Lys) NPs were used by the group as non-viral vectors [282]. Ellipsoidal PEGylated NPs carrying enhanced green fluorescence protein eGFP-expressing DNA showed enhanced brain distribution in vivo following CED with efficient transfection. In a more recent example, the authors used the same concept to deliver anticancer agents. NPs based on P(Asp) to which PEG chains were conjugated to the polymer backbone were used to complex cisplatin yielding 70 nm NPs with a dense PEG corona [283]. CED of NPs in F98 rat glioma tumours resulted in a significantly enhanced antitumor efficacy with 80% of the rats exhibiting long-term survival. This technique has also been employed to deliver MRI macromolecular J o u r n a l P r e -p r o o f probes such as gadolinium diethylenetriaminepentaacetic acid-installed PLys derivatives developed by Hardy et al [284].

Intranasal Administration
Nose-to-brain delivery or intranasal (IN) administration of therapeutics has recently arisen as a promising approach to directly reach the brain and bypass the BBB. The possibility for selfadministration, as well as low systemic accumulation and the rapid and enhanced absorption of therapeutics following this route, provides fast action with reduced off-target effects making IN an interesting alternative for the treatment of CNS disorders [285]. Although IN administration allows for higher brain accumulation (up to 10-fold) when compared to IV, only a few therapeutics hold specific mucoadhesion/mucodiffusion balance to be applied within this route. This fact, together with the need for further navigation once at the olfactory bulb to access other brain areas, highlights the need for its combination with nanosized delivery systems [286]. To this end, Baba and coworkers used the previously mentioned polyplex nanomicelles based on PEG-P(AspDET) block copolymers to deliver mRNA via IN administration to treat sensory nerve diseases [287]. Mice with drug-induced olfactory dysfunction treated daily with brain-derived neurotrophic factor (BDNF)expressing mRNA using these micelles through the IN route allowed for almost complete rescue of olfactory function and the regeneration of the olfactory epithelium structure as observed by behavioural test and histology respectively. Similarly, GDNF pDNA was intranasally delivered to the brain of PD rats complexed to a PEG-P(Lys) 30 carrier [288]. The generated pGDNF DNA NPs were intranasally administered 1 week prior to a 6-OHDA lesion in rats used as a PD model.
Administration of the NPs yielded longer-lasting GDNF brain expression compared to naked GDNF and consequently, exerted greater neuroprotection.
The abovementioned combination of ELP and CPP carriers have also been explored through the IN route. cAMP-dependent protein kinase A (PKA) inhibitor was delivered efficiently to the brain following IN route by the aid of the tandem Synb1-ELP in order to treat spinocerebellar ataxia-1 (SCA1) [289]. However, a follow-up study showed that the addition of CPPs (either Tat or Synb1) to ELP significantly affected the biodistribution of ELP after IN administration by increasing their retention at the nasal epithelium in detriment of the amounts reaching perivascular spaces at the brain [290].

Targeting the BBB in pathological conditions
In most of the examples described above, the design of these peptide NPs is based on the physiology of a healthy BBB and thus overlooks the significant effects of brain disorders on integrity of the barrier and transport across the BECs [144,292]. Indeed, reductions in the cerebral blood flow (CBF), microvascular pathologies, aberrant angiogenesis, breakdown of the BBB due to disruption of the TJs, and altered transport systems in BECs, almost certainly affect the fate of peptide nanocarriers.
In terms of blood flow, reductions in resting CBF or altered responses to brain activation may occur in different regions of the brain in AD, PD, stroke, and other disorders [293,294]. In normal ageing, a modest reduction of 20% in CBF is observed, while more severe regional reductions in CBF (> 50%) occur in chronic neurodegenerative disorders (AD and PD) and stroke [144,295].
Additionally, an increase in blood vessel tortuosity observed in neurodegeneration results in changes in blood rheology that impair the ability of nanocarriers to cross the BECs, thus magnifying the issue of CBF [296]. Similarly, in gliomas, owing to the confined space within the brain, tumour growth can impair CBF and alterations occur in blood vessel geometry, leading to increased tortuosity compared to healthy vasculature [297,298]. Significant microvascular pathologies, including reduced microvascular density, an increased number of fragmented vessels with few intact branches, atrophic string vessels, alterations in vessel diameter, and capillary basement thickening with an accumulation of collagen, have been described in AD and PD [299,300]. Furthermore, findings suggest that degeneration of BECs in AD may reflect an aberrant angiogenesis [301,302], which is also characteristic of brain tumours, such as gliomas [303]. Altogether, these alterations in the architecture of the vasculature along with the characteristic abnormal angiogenesis may again impact flow through the blood vessels as well as BEC-crossing mechanisms. Thus, access to the brain parenchyma may be restricted in brain disorders. On the other hand, breakdown of the BBB has been demonstrated in AD and PD [304][305][306], glioma [307,308], stroke [309] and traumatic brain injury [310], which may facilitate diffusion of nanocarriers from the blood to the brain parenchyma.
Importantly, in addition to the structural changes of the brain blood vessels, molecular alterations in terms of receptors expressed at BECs have been identified, impacting the design of NPs for the crossing across the BBB. In ageing and AD, reduced levels of LRP1 [311] and GLUT1 [312] has been found in rodents and AD individuals, while RAGE levels are increased in AD [313]. Based on these, it is important to understand the physiology of the BECs to precisely target the BBB and achieve J o u r n a l P r e -p r o o f efficient crossing. As example of that, RAGE has been considerably employed for brain delivery in AD, while the peptide RGD sequence targeting the integrins overexpressed in BECs of tumour angiogenic vessels is exploited for delivery in glioma (Table 3).
Although it is essential to understand the biology and pathology of the BBB to precisely design nanocarriers for brain delivery, challenges remain. At the BBB level, a number of carrier-and receptor-mediated transporters remain to be studied in pathological conditions to further understand their role and their potential as a targeting for brain delivery. Developing new genomic and proteomic discovery platforms will allow us to identify transporters that can be potentially used as portals of entry for the peptide nanocarriers.

Conclusions and Future Outlook
There is an unquestionable increase in the prevalence of brain disorders, including brain tumours, PD, and AD. This trend will almost certainly continue due to an ageing population, generating a major problem in healthcare systems across the world. While many efforts have been made in the field to improve current therapies and develop new disease-modifying therapies, there is a clear need for further investment. The treatment of brain disorders is particularly challenging due to (i) the presence of the most impenetrable metabolic and structural barrier, the BBB, and (ii) the lack of diffusivity of the majority of the CNS drugs and therapeutics once at the brain parenchyma. This second issue leaves compounds unable to reach all affected areas, leading to poor therapeutic outcomes. Nanotechnology and nanomedicine are promising tools to overcome these limitations.
Despite being in their early development for brain delivery, biodegradable peptide-based NPs in particular show great potential as illustrated here by the many examples in preclinical stages. To reach the brain, choosing the right administration route (IV, IN or local administration) as well as a careful selection of the targeting ligand (according to the route, disease condition and stage) needs consideration. Additionally, as recently demonstrated, optimisation of ligand densities to reach the brain parenchyma within sufficient concentrations to provide enough therapeutic response is key in the design of nanocarriers [164].
However, the challenges in translating peptide NPs to market are manifold. Many of the limitations in this area are generic to the field of medicine development, such as the need for more realistic in vitro and in vivo models to establish structure-activity relationships and predict and test NP outcomes [314,315]. Normalisation of quantitative methods and parameters to report pharmacokinetics and pharmacodynamics, including percentage of brain accumulation, will help in the understanding of the use and concentration of certain targeting ligands and the true potential of nanocarriers. The frequent neglect at the preclinical stages of potential NP-induced neurotoxicity, J o u r n a l P r e -p r o o f off-target and long-term effects, as well as their understudied brain and whole-body clearance and degradation routes, pose a barrier to their evaluation at later stages and translation from bench to bedside. The degradation pathways of peptides must be carefully considered, since several amino acids such as γ-aminobutyric acid (GABA), glycine, and glutamic or aspartic acid function as key primary excitatory transmitters in mammals. As such, the degradation of peptide nanocarriers may disrupt their concentrations at the synaptic feet of neurons and therefore interfere with amino-acid mediated synaptic transmission [316,317].
There remain significant shortcomings in our understanding of how NPs can be designed to cross the BBB. Although it is generally recognised that BECs facilitate the transport of essential molecules by using a CMT or RMT, these mechanisms of transport are yet to be fully understood, particularly in a scenario of pathology. Recent findings on intracellular transport across the BBB raise important questions. How many pathways can actually drive transcytosis? What determines which pathway will sort different receptors and whether a cargo undergoes transcytosis or lysosomal degradation? Intracellular sorting mechanisms (e.g., tubules or vesicular endosomes) that control successful transcytosis in BECs are still to be deciphered. By answering these fundamental questions, we then will be able to fully understand what exactly affects the intracellular transport within BECs to improve the delivery of therapies into the brain. Furthermore, developing genomic and proteomic discovery tools will enable us to identify new transporters in healthy or diseased BECs, which can then be hijacked for CMT-or RMT-based strategies for delivery of drugs into the brain.
There are also significant shortcomings in peptide NP synthesis and characterisation techniques, which have led to major reproducibility issues in the field [318][319][320]. Despite being the subject of study for decades, peptide and polypeptide self-assembly remains remarkably poorly understood. The overwhelming majority of the studies discussed in this work use polyplexes to achieve brain delivery. This is, at least in part, due to their simplicity and the relative ease with which they can be produced and functionalised. There is another reason for their popularity, however, which is that the methods by which more complex, self-assembled peptide structures can be produced at a scale and reproducibility suitable for drug delivery studies are simply not known.
Traditional top-down approaches rooted in statistical mechanics have to-date generally only been successful in describing the formation and self-assembly of the more commonly seen motifs in biology, such as -sheets, helices, and turns. At the same time, simulations-based approaches to study peptide self-assembly are limited by the colossal computational requirements needed to simulate large ensembles of polypeptides in water on microsecond timescales. Experimental studies that apply the evolutionary approaches discussed in Section 3.3 may prove much more productive, while having far lower technical barriers to entry. Popular techniques such as DLS and TEM are rather limited and suffer from selective reporting.
Newer methods, such as tunable resistive pulse sensing and nanoparticle tracking analysis are potentially useful, although older methods such as asymmetric flow field flow fractionation with inline DLS and multi-angle light scattering would represent a significant improvement over the norm.
Surface functional group density remains an under-scrutinised quantity; the development of techniques that can report more than just an average presents a particular challenge. Assays built on the recently developed principles of superselectivity [321][322][323], in which NP binding is a strong function of receptor density over a narrow range of values, may prove promising here. Finally, there is a striking shortage of studies on protein corona formation in peptide NPs [324][325][326][327].
Understanding how the protein corona reduces, or even enhances, the efficacy of engineered NPs is of paramount importance for their rational design. Indeed, the highly controlled interactions between proteins and biomolecules observed in nature suggest that peptide NPs are a particularly promising platform not for removing, but for engineering the protein corona for drug delivery applications [328][329][330].
Despite the challenges listed above, peptide nanocarriers are perfectly positioned to overcome many of the obstacles that prevent successful delivery of therapeutics to the brain. The material diversity of the building blocks of peptides allows for design and tuning of a multitude of supramolecular interactions. Polypeptides can be synthesised with a low polydispersity, complexity, and reproducibility almost entirely unique among polymers. The body of work surrounding protein folding provides an unparalleled toolkit for rational design and characterisation of peptide NPs. On top of this, peptide synthesis and self-assembly techniques make them highly amenable to automated, high-throughput study and discovery that is driven by machine learning methods and affordable lab-made robotics. Improved peptide NP synthesis methods will, in turn, provide researchers with new probes for identifying novel pathways and mechanisms for the delivery of therapeutics to the brain. We believe that this diversity of both material properties and experimental approaches make peptide NPs an extremely promising platform for brain delivery.