Alleviating the unwanted effects of oxidative stress on Aβ clearance: a review of related concepts and strategies for the development of computational modelling

Enzymatic processing of amyloid precursor protein (APP) results in production of several derivatives with biological functions. Toxic products of APP are known as the factors involved in the pathology of AD. APP proteins, produced by the endoplasmic reticulum, are transported to the Golgi complex and ultimately to the plasma membrane. APP cleavage by β-secretase between positions 16 and 17 produces the β-C-terminal fragment (C99) and the large ectodomain (sAPPβ), through the amyloidogenic pathway. This process ultimately enhances Aβ aggregation. APP cleavage by α-secretase between positions 10 and 11 produces the α-C-terminal fragment (C83) and the ectodomain (sAPPα) through the non-amyloidogenic pathway, which prevents Aβ aggregation [30, 31]. C99 and C83 are then cleaved by γ-secretase to produce Aβ and p3 peptides through amyloidogenic and non-amyloidogenic pathways, respectively (Fig. 1). In addition to the amyloidogenic pathway, C99 can be cleaved by α-secretase to produce other Aβ species [32]. Previous studies have identified that many cleavages of APP, including short Aβ isoforms (Aβ1-17/18/19/20), are produced through the amyloidogenic pathway [33‐35]. Although the mechanisms and functions of APP are not completely understood, studies have shown that the production of APP is related to transcriptional control, axonal transport, and apoptosis [36‐38].

Aβ monomer, known as Aβ protein, has two dominant forms of Aβ proteins, Aβ40 and Aβ42. Evidence indicates that the Aβ monomer may be non-toxic [39, 40]. Aβ monomers may aggregate into an aggregation state, resulting in Aβ oligomers (AβOs) and protofibrils, which have low and high molecular weights, respectively. These protofibrils ultimately form fibrils and amyloid plaques [41]. Researchers have reviewed the relationship between the products of the amyloidogenic pathway using in vitro, in vivo, and in silico (computer simulation) experiments [42, 43]. Pharmacological experiments have also shown the relationship between the products of the amyloidogenic pathway needed for the development of pharmacological interventions [42, 43].

Although researchers once thought that amyloid plaque was associated with Aβ aggregation or a pathogenic form of Aβ, they have now identified AβOs as the pathogenic form of Aβ, as indicated by the AβOs present in animal and human models [44‐46]. For instance, low levels of Aβ but not AβOs are present in severe cognitive impairments like AD [47, 48].

Studies have identified AβOs as contributors to poor memory [49]. Experimental studies in cell biology have proven this: non-transgenic mice injected with small quantities of AβOs show poor memory performance [50, 51]. Likewise, in vivo studies have shown that AβOs disrupt long-term potentiation and contribute to long-term depression [52]. AβOs may be more cytotoxic to neuronal synapses than protofibrils and fibrils based on experimental results of in vivo studies [53‐55]. To understand oligomer production and toxicity, researchers have investigated the monomer-dependent secondary nucleation (MDSN) of Aβ in an in vitro study [56]. They found that the rate of the MDSN process plays a key role in amyloid-forming peptides; knowing the rate of the MDSN process, modulated by hydrophobic and electrostatic interaction of surrounding proteins, will ultimately help in the development of inhibitors which suppress MDSN.

Researchers have investigated other mechanistic aspects related to Aβ toxicity. For example, researchers have found that mutation of insulin-degrading enzyme (IDE) at the allosteric site, resulting in the cysteine-free IDE mutant which is catalytically inactive against insulin, impairs Aβ degradation [57]. This is a known cause of Aβ deposition and toxicity. While the deposition of functional toxic forms of Aβ in the central nervous system can cause AD, non-functional toxic forms of Aβ deposited in the tissue lead to other diseases such as amyotrophic lateral sclerosis (ALS) [58]. Researchers have explored Aβ toxicity in AD using in vitro experiments and by disrupting membranes [59]. The latter has demonstrated that cell membrane disruption is comprised of two steps: soluble AβOs binding to the membrane, and Aβ fibrils causing membrane fragmentation. In the first step, the soluble AβOs bind to the membrane to form calcium-permeable pores, and they are known as primary pathologic species of AD. In the second step, elongated Aβ fibrils, as the detergent molecules, interact with cell membranes, causing membrane fragmentation by detergent effect [59].

In the lag phase or early stages of AD, the conformation of AβOs may be dominated or changed by numerous toxic pathways. These pathways include various agents such as metal, other amyloid proteins (such as synuclein and tau), and lipids. For instance, aberrant metal homeostasis may dominate the conformation of AβOs. Zinc, copper, and iron may mediate the aggregation of AβOs [60]. These metals may react with the products of free radicals, leading to cellular toxicity [61]. Studies have identified that pre-formed AβOs, which interact with lipids as well as Aβ monomers, can affect the conformation of AβOs [62, 63]. Regarding the toxicity of AβOs, the interaction between AβOs and the lipid membrane has been identified as a prominent factor in neuronal cell damage [63]. Interactions between other amyloid proteins and AβOs have been explored to understand different amyloid diseases [64]. For example, prion protein, which can cause fatal diseases, may be a high-affinity receptor for AβOs leading to conformational change and aggregation of AβOs [65]. This phenomenon, or the links between other amyloid proteins (misfolded proteins) and AβOs (which are based on molecular and pathogenic mechanisms), are known as cross-seeding. Cross-seeding interrupts the conformation of AβOs and Aβ aggregation. Recent studies have investigated the cross-seeded polymerisation of Aβ to understand the underlying mechanism of amyloid-forming proteins [66]. Examinations of cross-seeding have provided crucial information for those developing therapeutic interventions for AD, using the mechanisms of related diseases [67, 68].

Immunotherapies using antibodies against AβO toxicity are known as one of the most promising approaches for pharmacological interventions of AD. Injection of an AβO-specific antibody has been shown to rescue memory performance (spatial learning) in transgenic mice (5×FAD mice) [69]. Kinetic analyses have shown that antibodies protect cell membranes against AβO toxicity [70, 71]. Using high-throughput screening technology, studies have shown that 5 small molecules can inhibit or prevent AβO toxicity [72]. An antibody-based immunotherapeutic approach has been applied to inhibit activities of soluble oligomers and insoluble fibrils [73]. The antibody-based immunotherapeutic approach reported by Sevigny et al. selects human B-cell clones triggered by the unique antigens (neo-epitopes) present in pathological Aβ aggregates during the process of Aβ aggregation [73].

Scientists have also developed antibodies to reduce or counteract the effects of certain metals. Various metals contribute to AβO toxicity and senile plaques, which can increase concentrations of transition metals in AD-affected brains [74, 75]. However, the development of drugs and nonpharmacological procedures to slow down or halt AD has been hindered by the fact that the complex heterogeneous properties of AβOs are not completely understood, particularly from a mechanistical perspective [76, 77].

There are approximately 20 Aβ-degrading proteases (both intracellular and extracellular) involved in Aβ clearance. These proteases include the IDE, neprilysin (NEP), endothelin-converting enzyme 1 and endothelin-converting enzyme 2 [80]. Aβ-degrading proteases, produced by glial cells, cleave Aβ peptides into smaller fragments at different sites. An in silico study has shown that the Aβ-degrading proteases possess many cleavage sites in the Aβ peptide [81]. Up-regulation of these proteases helps to control the aggregation of Aβ peptides and thus presents a possible avenue for therapeutic intervention [81]. Scientists have recently investigated the characteristics of peptide fragments degraded by the prominent proteases –NEP and IDE [79].

Another crucial process for Aβ clearance involves the active transport of these proteins from the brain. Many proteins play a crucial role in Aβ clearance, including apolipoprotein E (APOE) and α2-macroglobulin (α2-m) [82]. APOE and α2-m interact with various receptors, including lipoprotein receptors. As a result of these interactions, small fragments or Aβ sequences cross the blood–brain barrier (BBB) [83]. In short, the level of Aβ clearance may not only be determined by proteolytic degradation, but also by whether it is actively transported out of the brain across the BBB.

Both genetic and non-genetic factors can cause an increase in the aggregation of Aβ peptides and lead to Aβ catabolism [22, 81]. Researchers have identified genetic mutations and non-genetic factors as causes of familial AD and sporadic AD (SAD), respectively. Non-genetic SAD account for 90% of all AD cases [84]. In addition, non-genetic factors play a greater role than genetic factors in the impairment of Aβ clearance [79].

In amnestic mild cognitive impairment (aMCI), or early-stage AD, several non-genetic factors may affect the metabolism of glucose, leading to mild cognitive impairment [28]. Intervention is crucial at this stage to ensure aMCI individuals can continue to function well [85]. In aMCI, oxidative stress is one non-genetic factor associated with impaired Aβ clearance and enhanced Aβ aggregation via HNE modification [11, 86]. Also, oxidative stress resulting in impaired Aβ clearance and enhanced Aβ aggregation is considered as an initial lag phase [86] (Fig. 1).

Oxidative stress is an imbalance between free radicals and antioxidants, where the number of free radicals outweighs the number of antioxidants. Free radicals are oxygen-containing molecules with an uneven number of electrons, which enables them to interact with other molecules known as oxidants or reductants [87]. Free radicals like hydroxyl radicals, hydrogen peroxide, superoxide anion radicals, hypochlorite, nitric oxide radicals, and peroxynitrite radicals are believed to cause neurological diseases such as Parkinson’s disease and AD [88]. In the context of chronic oxidative stress, these free radicals negatively impact certain processes, leading to oxidative protein modification, DNA oxidation, and lipid peroxidation of the cell membrane [89, 90]. These processes eventually result in homeostatic disruption and cell damage [87]. In normal conditions, antioxidants neutralise free radicals by donating electrons to them. As a result, free radicals become less reactive and more stable [91]. The balance between the production of free radicals and the antioxidant activity is indicated by redox signalling [14, 92]. The redox status has been explored in neurodegenerative diseases such as AD and Parkinson’s disease [93, 94].

The covalent modification of aldehydes by lipid peroxidation, known as oxidative carbonylation, will ultimately lead to oxidative protein modification. This plays a key role in metabolic diseases [20]. Oxidative carbonylation means that free radicals directly attack specific amino acids which are vulnerable to oxidation (e.g., proline, arginine, lysine, and threonine), leading to protein hydrophobicity (protein unfolding) and the risk of protein aggregation [95, 96]. Likewise, the free radicals oxidise DNA bases (adenine, guanine, cytosine, and thymine), leading to DNA damage. For example, guanine, which has high oxidation potential [97, 98], is attacked by free radicals at its imidazole ring. As a result, guanine is transformed into 8-hydroxyguanine, causing DNA lesions [99]. Lipid peroxidation is known as a prominent source of cell membrane damage [90]. Lipid peroxidation of the cell membrane occurs when free radicals, under oxidative stress conditions, attack the cell membrane at the carbon–carbon double bond(s), causing hydrogen removal from carbons and oxygen insertion. A lipid peroxyl radical is formed and an abstract hydrogen atom, in fatty acyl chain in a lipid bilayer, forms lipid hydroperoxide (LOOH) [100]. In the propagation phase, the free radicals can react with the lipid peroxyl radicals, by removing hydrogens from the lipid molecule, resulting in the production of new free radicals and lipids [101, 102]. Many studies have found that lipid peroxidation contributes to the development of pathological states and accelerates aging [20, 103, 104]. The brain is vulnerable to attacks by free radicals because the phospholipid form, which is the backbone of neuron membranes, contains high levels of polyunsaturated fatty acids (PUFAs). PUFAs, like glycolipids, phospholipids, and cholesterol, are a family of lipids with carbon–carbon double bonds [90, 105]. In addition to LOOH, lipid peroxidation can produce many toxic secondary products, of which HNE is the most toxic one [90, 106]. HNE, when bound to the key neuronal membrane, results in dysfunction of key neuronal proteins, leading to neuronal death. HNE is produced by lipid peroxidation through two pathways: enzymatic and nonenzymatic pathways [90, 107].

In the enzymatic pathway, PUFAs are cleaved by phospholipase A2 (PLA2) at the sn-2 position, a process which frees PUFAs from neuron membranes [108]. There are two major classes of PUFAs: omega-3 PUFAs (n-3 PUFAs) and omega-6 PUFAs (n-6 PUFAs). Both classes are metabolised by the same esterification reaction [109] (Fig. 2), and both require release of PUFAs from cell membranes [109]. The n-3 PUFA family comprises α-linolenic acid, docosahexaenoic acid (DHA), and eicosapentaenoic acid (EPA). The n-6 PUFA family includes linoleic acid, arachidonic acid (AA), and dihomo-γ-linolenic acid. While n-3 PUFAs exert anti-inflammatory effects and vasodilation, n-6 PUFAs cause inflammation and platelet aggregation. The anti-inflammatory activities of n-3 PUFAs are related to the fact that n-3 PUFAs—EPA and DHA—can be enzymatically converted to generate bioactive and anti-inflammatory products. They also exert anti-inflammatory effects through modulating nuclear factor-κB signaling, NLRP3 (NOD-like receptor family pyrin domain containing 3) inflammasome, G protein-coupled receptors, and transforming growth factor β signalling [110]. For n-6 PUFA oxidative metabolism, AA can be converted to prostaglandin H2 (PGH2) by cyclooxygenases 1 and 2 [111]. AA can also be converted to leukotriene B4 (LTB4) by 5-lipoxygenase (5-LOX) and leukotriene A4 (LTA4) hydrolase. Both PGH2 and LTA4 can cause a variety of illnesses [112]. In addition to n-6 PUFA oxidative metabolism, AA can also be transformed into HNE through the metabolism of 15-lipoxygenase (15-LOX) [113]. Of the two, the non-enzymatic lipid peroxidation pathway has received more scientific attention as the non-enzymatic lipid peroxidation of PUFAs leads to the formation of intensively reactive electrophilic aldehydes—HNE, malondialdehyde, and acrolein [114].

In the non-enzymatic pathway, production of HNE is involved in the free radical lipid peroxidation [86]. PUFA first undergoes abstraction of allylic hydrogen atom from the methylene group to produce a carbon-centred alkyl radical. Then, the alkyl radical produces peroxyl radical. As illustrated in Fig. 2 and explained below, HNE is produced via five mechanisms. First, the peroxyl radical forms hydroperoxyl radicals, which in turn, are involved in hydrogen abstraction [115]. Then, the hydroperoxyl radical and the alkoxyl radical produce HNE through transition metal ions—such as Fe²⁺—and beta-scission regulated by a hydroxy alkoxy radical. Second, cyclisation of the peroxyl radical forms dioxetane, followed by oxygenation of dioxetane, which leads to peroxyl dioxetane. This fragmentation of peroxy-dioxetane leads to 4-hydroperoxy-2E-nonenal (4-HPNE); eventually, the hydrogen abstraction of 4-HPNE results in HNE. Third, oxygenation of the hydroperoxyl radical results in hydroperoxyl dioxetane, which is further fragmented to 4-HPNE, the immediate precursor of HNE. Fourth, alkoxyl radicals, produced by a reaction between reduced forms of transition metals and bicyclic endoperoxides, are oxygenated and fragmented, generating HNE. Fifth, the alkoxyl radical is cyclised and oxygenated. It undergoes a Hock rearrangement. This process generates 15-hydroperoxyeicosatetraenoic acid (15-HPETE) or 13-hydroperoxy-linoleic acid (13-HPODE), both of which are known as immediate precursors of HNE.

Aβ-degrading proteases are an important feature of the immune system found in the brain and play a key role in Aβ clearance [79]. Understanding how these Aβ-degrading proteases break down the Aβ peptide via proteolysis may help develop targeted AD treatments [116]. One study has found that up-regulation of these proteases helps control the accumulation of Aβ peptides [81]. IDE and NEP are the most significant enzymes involved in Aβ degradation. They are released from microglial cells [79, 117]. Both of these enzymes are known as intracellular and extracellular Aβ-degrading enzymes [22]. Intracellularly, interaction between the NEP and APP intracellular domain (AICD) enables amyloid clearance through the regulation of NEP [118]. Subsequently, ACID may be released into the cytosol to be degraded by IDE [106]. Both NEP and IDE are well-known extracellular degrading enzymes [117, 119, 120].

Aging and high cholesterol levels may cause low level and low activity of NEP and IDE [121, 122]. Furthermore, several studies have shown that oxidative stress, through post-translational modification and lipid peroxidation, can impair the expression of NEP and IDE in AD [123‐125]. Protein misfolding, caused by interactions between lipid peroxidation products and proteins, impairs protein activities of NEP and IDE, and ultimately leads to Aβ aggregation [126, 127]. For instance, HNE-NEP adduction can reduce Aβ cleavage, a key factor in Aβ accumulation [127]. Likewise, HNE-IDE adduction may lower the enzymatic activity of IDE [79, 127].

The binding of HNE to amino acids may be explained by two principles: Schiff’s base formation and Michael’s addition (Fig. 3). HNE often reacts with lysine (Lys) and histidine (His), amino acids of NEP and IDE; it also reacts with cysteine (Cys) at the same velocity as Lys and His [20, 128, 129]. The interaction between HNE and these amino acids may modify chemical structure of NEP and IDE. For example, the HNE-induced modification of Cys, His and Lys can impair the enzymatic activity of NEP [130]. Similarly, an abundance of Cys and His, modified by HNE, can lower the activity of IDE [131, 132]. One study found that the reactions of HNE with Lys, His, and Cys have the same velocity. Cys has the highest reactivity, followed by His, Lys and Arg [20].

Degradation of the HNE-modified proteins occurs via two pathways: the lysosomal pathway and the UPP [137, 138]. The UPP and lysosomal pathways control the degradation of intracellular modified proteins, while extracellular modified proteins are only degraded by the lysosomal pathway [139, 140]. UPP is responsible for the degradation of abnormal cytosolic and nuclear proteins in eukaryotic cells: the 26S proteasome plays a key role in the degradation of proteins [141‐143]. The 26S proteasome comprises the 20S proteasome and 19S regulatory particles [144, 145]. The 20S proteasome has been identified as the catalytic core [146, 147]. Studies have shown that the 20S proteasome degrades oxidised proteins without ATP hydrolysis and conjugation of ubiquitin [148‐150]. For the ubiquitination of proteins, the 26S proteasome requires ATP hydrolysis and the conjugation of ubiquitin to modify the protein target (Fig. 4) [151, 152]. The aggregation of HNE-modified proteins may lead to cellular dysfunction and cellular aging; in short, the clearance of HNE-modified proteins is crucial to proper cell functioning [133, 153, 154].

Although HNE-modified proteins can be cleared via UPP, this is insignificant for studies of biological reactions because, in some cases, membrane receptors facilitate their degradation via the lysosomal pathway [137, 155]. This finding indicates that both the UPP and the lysosomal pathway are involved in the degradation of HNE-modified proteins (Fig. 4) [23]. In the lysosomal pathway, the HNE-modified proteins are transferred as protein substrates to the receptor of the lysosomal membrane [156, 157]. Consequently, the protein substrates are transformed into the lysosomal lumen and degraded. One study has shown that inhibiting the ubiquitin–proteasome system or the lysosomal proteolytic system alone, leads to a partial decrease in the degradation of the methylglyoxal (MGO)-modified proteins. In contrast, inhibition of both shows a significant aggregation of MGO-modified proteins [158].

AβOs, but not the monomers, impair the 26S proteasome activity (proteasomal activity) [163]. Interactions between the AβOs and the 20S proteasome may lead to the impairment of the proteasomal activity [164]. Since AβO binding to the 20S proteasome may occur in neurodegenerative diseases [165], regulating the proteasome gate may provide a strategy to protect the proteasomal activity against AβO conjugation [166].

Although proteasomal activity contributes to Aβ clearance (through degradation of the HNE-modified proteins), it may also be impaired by conjugation of the AβOs (Fig. 5). Therefore, the proteasomal activity and the AβOs can interact with each other.

Along with increased levels of oxidative damage, decreased antioxidant enzyme activity may cause AD; this process is thought to be due to the resulting oxidative damage to related residues [167]. Thus, enhancement of the antioxidant activity may provide a form of intervention for AD. GSH has been identified as the most prevalent antioxidant in cells of the brain [24, 25]. Studies have shown two effective precursors to GSH: N-acetyl-L-cysteine (NAC) and γ-glutamyl cysteine ethyl ester (GCEE) [135, 168]. NAC crosses the BBB and provides cysteine for GSH synthesis, thereby increasing GSH levels [25]. Researchers have developed mitochondria-targeted nanocarriers to deliver NAC, and reported that they may halt or slow the degenerative process, which ultimately leads to decreased oxidative damage [169]. GCEE, another precursor which increases GSH synthesis through catalysation, could interact with GSH to easily cross the cell membrane and BBB [25]. This mechanism may provide protection against myocardial dysfunction and mitochondria damage [170].

Although antioxidants play a major role in defending against oxidative stress, the levels of oxidative damage are still greater than the antioxidant defence. Therefore, the antioxidant defence strategies could be integrated into the antioxidant defence network, which could include maintaining conserved mechanisms involving kinases and transcription factors, restricting food intake, increasing exercise, using chemical compounds to increase antioxidant levels by inhibiting reactions catalysed by iron and copper, and promoting lifestyles that reduce oxidative stress [171].

Changing gene expression levels using gene-based techniques to resist oxidative stress is challenging. Mimicking enhanced antioxidant levels has been proposed as a defensive strategy that could be used to protect cells against oxidative stress. This aim could be achieved by elevating levels of glucose-regulated proteins (e.g., GRP78 and GRP94), and the heat chock protein [56, 172]. To identify genes involved in resistance to oxidative stress, one study performed gene microarray experiments in mammalian cells showing resistance to oxidative stress [173]. Altering certain stress-responsive genes such as HO-1 (heme oxygenase-1), C-JUN, and GADD15 results in elevations of the antioxidant level and GSH level, enabling greater resistance to oxidative stress [174]. An error-prone PCR technique to mutate cAMP receptor protein genes has been used to generate three E. coli mutants with improved oxidative stress resistance [175].

Protein design is a useful technique with potential therapeutic applications, and has been used to reduce oxidative stress via a few techniques, including the development of nanozymes and the creation of protein inhibitors and activators.

Due to their flexible design and economical production costs, nanozymes provide a viable way to study enzymatic and non-enzymatic activities [176]. Nanozymes have been developed to mimic endogenous antioxidants GSH, NAC, and SOD, because these antioxidants scavenge free radicals associated with oxidative damage [177‐179]. GSH is an intracellular protective antioxidant that protects cells against oxidative stress in the endoplasmic reticulum. It reduces the non-native disulphide bond and forms the native disulphide bond, and it regenerates other antioxidants such as ascorbate and tocopherols [180, 181]. NAC is a cysteine precursor which contributes to GSH activities by maintaining the synthesis of GSH. Both GSH and NAC function to convert hydrogen peroxide (H₂O₂) into water (H₂O) and dioxygen (O₂) [182]. This hydrogen peroxide is a result of SOD catalytical conversion of superoxide (O₂°), one of the free radicals, with the help of cofactors like copper, zinc, and manganese [178, 182]. Some materials have been developed to mimic antioxidants, including metal oxides, carbon-based nanomaterials, and noble metals [183, 184].

For instance, Chen’s group have developed a polyvinylpyrrolidone-modified Prussian blue nanoparticle (PPB) which mimics antioxidant enzyme activities [185]. The goal of PPB is to convert O₂° to H₂O₂ and convert the H₂O₂ into H₂O and O₂, mimicking GSH, NAC, and SOD. This process ultimately prevents lipid peroxidation and oxidative damage [186]. However, before embracing such technology, we must first understand the biological effects of nanoparticles (their kinetic binding properties and equilibrium), because the inherent shortcomings of nanozymes such as nanoparticles interacting with lipid and DNA impair the expression of the original enzymes [187, 188]. Furthermore, an insufficiency of nanoparticles, and a low ability to interact with the target proteins, are significant drawbacks associated with the development of nanozymes. Both of these limitations will have a negative effect on the treatment [26].

Scientists have found a link between oxidative stress and insulin resistance, suggesting that type 2 diabetes mellitus (T2DM) contributes to AD. This discovery provides an opportunity for pharmacological intervention through the development of protein inhibitors and activators. T2DM, associated with insulin resistance, may cause oxidative stress which contributes to the progression of AD [28, 189]. Studies investigating the mechanisms of the generation of oxidative species have shown that oxidative stress may be directly linked to insulin resistance [190]. Recent studies have identified that insulin resistance is also present in early stages of AD [28, 191]. These studies indicate that alleviating insulin resistance using pharmacological methods may reduce oxidative stress. In pharmaceutical research, scientists have developed various inhibitors and activators to reduce the activities of the Jun NH2-terminal kinase (JNK) and other enzymes that contribute to insulin resistance, such as protein tyrosine phosphatases 1B (PTP1B), fructose-1–6-bisphosphate (FBPase), and glucokinase [192, 193]. Although these inhibitors and activators have been developed to reduce insulin resistance, which may in turn prevent AD, these drugs have side effects; further pharmacodynamics studies are needed to clarify their mechanisms of action, adverse effects, and drug interactions [192, 194‐196].

Directed evolution is an efficient method for improving the stability and activity of enzymes such as protease subtilisin E [198], cytochrome and GSH transferases [199]. Directed evolution mimics the process of natural selection which includes mutagenesis and selection in vitro [199]. Using this method, researchers have been able to identify selected and mutated residues which improve the catalytic reaction. One experimental study has shown that directed evolution occurs in natural evolution; hence, it can improve the stability and activity of various enzymes [200]. The process of direct evolution begins with the selection of randomly mutated genes from the gene library (> 10,000 clones). Researchers then select and/or screen the gene candidates with the desired function, and isolate the gene candidates, followed by biochemical testing.

Rational design has been developed to improve the thermostability and change the mechanisms of enzymes [201]. Rational design requires information on three-dimensional protein structure and the amino acid sequence as input data and is used for individual mutated gene screening. There are 144,464 protein structures with identified structures available in the Protein Data Bank (PDB) [197]. For enzyme stabilisation, the designed enzymes need to be catalysts for high catalytic activation under mild conditions and require the restricted number of mutations. One study has also developed a strategy to reduce the number of generated variants for rational design, resulting in a reduced screening workload [202]. After protein candidates have been generated by the mutated gene screening process, residue targets may be selected based on substrate selectivity [203]. Unlike directed evolution, in rational design one must understand the interactions between amino acids and protein structures from the beginning of the process.

As rational design requires in-depth knowledge and high-throughput scanning, and directed evolution, based on the randomisation method, can be performed without in-depth knowledge, researchers developed semi-rational design to improve the screening and selection efficiency. This hybrid model is a combination of directed evolution and rational design [204]. Semi-rational design takes advantage of directed evolution and rational design [204]. The smart library has been used to improve randomisation of the directed evolution method by applying in-depth knowledge of the rational design method (knowledge of stability via mutation of enzyme’s residues) to the whole genome [204, 205]. For example, researchers have employed knowledge of Cre recombinase recognition of DNA to improve the randomisation method for rearrangement of DNA [206]. Compared with rational design and directed evolution, the semi-rational design may lead to a higher probability of synergistic mutations [204]. Semi-rational design has its limitations. To obtain the desired protein activities, researchers must still grapple with a complicated design. Also, this approach can only use a limited number of proteins.

De novo protein design refers to the inverse protein folding, a process which involves creating a protein from scratch instead of using a known protein structure. The topology of the structural protein design, based on primary sequence, is the basis for de novo design. De novo protein design begins by identifying the scaffold protein. It investigates new functions of proteins, a process depending on knowledge of biomedical and synthetic biology [207]. De novo design consists of two steps: generating protein backbone conformation and detecting combinatorial sidechain packing [208]. De novo design generally uses the Monte Carlo procedure to random peptide fragments—based on the backbone conformation—and calculates the lowest energy needed to stabilise the enzyme [208]. Researchers designed this technique using computational design principles [209], via protein backbones retrieved from the PDB [210]. Rosetta, a well-known computational approach and a software package developed by the Rosetta group, was developed to determine an enzyme’s chemical structure [209]. A remarkable study has applied Rosetta to de novo design (as the protein-design strategy), to improve catalytic efficiency [211]. This study focused on improving enzyme catalysis at the transition state of target enzymes. This study concluded that better binding between the modified enzymes (a result of optimisation in the reactant state) and substrates improves enzyme catalysis, by stabilising the catalytic reaction, in the transition state. Since the metalloproteins (metalloenzymes) are involved in biological and chemical processes in nature—in particular stabilising the catalytic reaction, de novo protein design has been explored for designing and redesigning the metalloproteins [212]. Since the transition states of the catalytic reaction at atomic and electronic levels are crucial to be explored for the protein design, quantum mechanics (QM)/molecular mechanics (MM) calculation method has been applied with the de novo protein design [213]. However, despite this advantage, researchers must still discuss the reaction in the transition state and the reactant state, at the atomic level, to improve the accuracy of QM/MM simulations [211].

The combination of MM and QM calculations, known as a QM/MM calculation, has benefited charge-density analysis (based on calculation of electrostatic interaction and charge-density distribution) as it enables researchers to explore enzymatic catalysis at atomic and electronic levels [214]. The QM calculation focuses on the region of treated quantum (the QM region) (e.g., bond formation and bond breaking during the chemical reaction). The MM calculation focuses on the surrounding portion of the ligand-receptor interactions (the MM region) (e.g., the interaction between active site residues and ligand residues, explained in Box 1). Due to the limitations associated with time complexity in QM calculations –the order of ten of atoms [215], QM region selection continues to be an active area of research. While computational efficiency has greatly improved over the past decade, handling QM regions with more than 100 atoms [216] is still complicated; more than hundreds of thousands of energy states need to be computed for energy evaluations [217]. Due to this issue, methods that help researchers to select the optimal residue(s) to include in the QM region are essential to extend the QM/MM calculation for large-scale electronic structure simulations [218], e.g., selection of the Aβ and IDE residues for the calculation of electrostatic interaction and charge-density distribution in the transition state of the catalytic activation [219].