BACE2: A Promising Neuroprotective Candidate for Alzheimer’s Disease

Abstract

Alzheimer’s disease (AD) is the most common cause of dementia that affects millions of predominantly elderly individuals worldwide. Despite intensive research over several decades, controversies still surround the etiology of AD and the disease remains incurable. Meanwhile, new molecular players of the central amyloid cascade hypothesis have emerged and among these is a protease known as β-site APP cleavage enzyme 2 (BACE2). Unlike BACE1, BACE2 cleaves the amyloid-β protein precursor within the Aβ domain that accordingly prevents the generation of Aβ₄₂ peptides, the aggregation of which is commonly regarded as the toxic entity that drives neurodegeneration in AD. Given this non-amyloidogenic role of BACE2, it is attractive to position BACE2 as a therapeutic target for AD. Indeed, several groups including ours have demonstrated a neuroprotective role for BACE2 in AD. In this review, we discuss emerging evidence supporting the ability of BACE2 in mitigating AD-associated pathology in various experimental systems including human pluripotent stem cell-derived cerebral organoid disease models. Alongside this, we also provide an update on the identification of single nucleotide polymorphisms occurring in the BACE2 gene that are linked to increased risk and earlier disease onset in the general population. In particular, we highlight a recently identified point mutation on BACE2 that apparently leads to sporadic early-onset AD. We believe that a better understanding of the role of BACE2 in AD would provide new insights for the development of viable therapeutic strategies for individuals with dementia.

INTRODUCTION

With the passing of every three seconds, someone in the world develops dementia [1]. This is a worrying trend of global concern as dementia remains incurable and is permanently incapacitating that consequently exacts high disease and socio-economic burden throughout the patient’s remaining lifetime. There are currently about 55 million people in the world afflicted with dementia [2], of which Alzheimer’s disease (AD) is the most common form, accounting for about 60% to 80% of dementia cases worldwide [3]. Like other dementia subtypes, AD is a progressively debilitating neurological disorder that is characterized by cognitive decline, memory deficits, and behavioral alterations. The pathological hallmarks of AD include abnormal accumulation of extracellular amyloid-β (Aβ) peptide aggregations (plaques), intracellular neurofibrillary tangles (NFTs) of hyperphosphorylated tau protein, and extensive neuronal loss in the brain parenchyma [4]. The link between these hallmarks to the etiopathogenesis of AD may be explained by the amyloid cascade hypothesis, which remains the most prominent hypothesis to date. The hypothesis states that Aβ aggregation triggers the cascade of events that ultimately lead to AD pathology and symptoms. It posits that the Aβ₄₂ peptide produced via the successive cleavage of the amyloid-β protein precursor (AβPP) by β- and γ-secretases is more aggregate-prone as compared to other Aβ isoforms (Fig. 1). Consequently, these insoluble aggregates form amyloid plaques that trigger a cascade of events promoting the formation of NFTs as well as inflammation and oxidative stress that culminates to neuronal dysfunction and death [5–7]. Supporting this, mutations enhancing β-cleavage of AβPP or altering γ-cleavage of AβPP-C99 (C-terminal fragment generated by β-cleavage of AβPP) are known to be causative of early-onset AD. Furthermore, a study has revealed that a missense mutation in the AβPP suppresses Aβ production and reduces the risk of AD [8]. Notwithstanding the persuasiveness of the amyloid cascade hypothesis, it is tau pathology, rather than Aβ, that seems to correlate with progressive gray matter loss and cognitive impairment [9], which fuels the alternative hypothesis that tau pathology is the primary cause of neurodegeneration in AD. Regardless of the ensuing debate between the amyloid and tau hypotheses, it is generally accepted that both Aβ₄₂ and hyperphosphorylated tau are the two key pathogenic players contributing to the pathogenesis of AD. Accordingly, a targeting strategy focusing on the clearance of both pathogenic entities would be useful, especially given the limited success of anti-amyloid treatments in sporadic AD.

Fig. 1.

Processing of amyloid-β protein precursor (AβPP) via the amyloidogenic or non-amyloidogenic pathway. In the amyloidogenic pathway, AβPP undergoes sequential cleavage first by BACE1 or BACE2 to generate a C99 fragment, which is then cleaved by γ-secretase to release Aβ₄₂. This Aβ₄₂ is aggregate prone and can form neurotoxic plaques that are implicated in the amyloid cascade hypothesis to cause Alzheimer’s disease. In the non-amyloidogenic pathway, Aβ₄₂ can undergo further degradation by BACE2, which can also function as an Aβ-degrading protease (AβDP). The degradation products include non-toxic Aβ species Aβ₁₉, Aβ₂₀, and Aβ₃₄. In addition, AβPP can be either cleaved by BACE2 at the θ-site, or by α-secretase to generate a C80 or C83 fragment, respectively. These C-terminal fragments undergo further cleavage by γ-secretase to release the AβPP intracellular domain (AICD) and a short fragment P3. In this case, no Aβ₄₂ is generated. [Parts of the figure were drawn by using pictures from Servier Medical Art. Servier Medical Art by Servier is licensed under a Creative Commons Attribution 3.0 Unported License (https://creativecommons.org/licenses/by/3.0/)].

INSIGHTS FROM DOWN SYNDROME AND THE ENTRY OF BACE2 AS A POTENTIAL NEUROPROTECTIVE MODIFIER OF AD

The APP gene is found on human chromosome 21, and individuals born with Down syndrome (DS) attributed to Trisomy 21 are born with an extra copy of this gene, which enhances their propensity for developing AD. Consistent with this, non-DS individuals with a rare triplication of the APP gene due to the duplication of a single APP locus (DupAPP) develop early-onset AD with a 100% penetrance by the age of 60 [10, 11]. Technically, the same should be observed for DS individuals with APP gene triplication. However, the intriguing paradox is that only about 70% of individuals with DS develop AD by age 60, i.e., 30% of individuals with DS are spared of clinical signs of dementia. This suggests the attractive possibility that chromosome 21 might harbor potentially neuroprotective genetic modifiers of dementia [12]. One such modifier is BACE2, which like the APP gene, is also located on chromosome 21, i.e., an extra copy of the BACE2 gene is similarly found in individuals with DS [13].

The BACE2 gene is located within the DS Critical Region in 21q22.3 and encodes a type 1 transmembrane aspartyl protease spanning 518 amino acids long [14, 15]. BACE2 is a close homolog of BACE1 which shares about 64% amino acid sequence similarity [16]. While BACE1 is recognized to be the main β-secretase in the brain that participates in the amyloidogenic pathway by partnering with γ-secretase to release the Aβ peptide from AβPP [17], BACE2 can process AβPP via the amyloidogenic or non-amyloidogenic route (Fig. 1). In the latter route, BACE2 cleaves AβPP within the Aβ domain particularly at the θ-site (Phe20) to prevent Aβ₄₂ generation [18, 19] (Fig. 1). Given this amyloidosis-protective role of BACE2, it is not surprising to note that several studies have reported a neuroprotective role for BACE2 in dementia [13, 18–22].

EXPRESSION OF BACE2 IN THE BRAIN

Although BACE2 is intimately involved in AβPP processing, its expression is not confined to the brain. Rather, BACE2 is ubiquitously expressed and found in many peripheral tissues such as the kidneys, pancreas, colon, placenta, prostate, trachea, and stomach. Intriguingly, reports regarding its expression in the brain have been conflicting. BACE2 expression was initially deemed to be low in the brain [14, 23, 24]. This being the case, the lower expression of BACE2 in the brain might render neurons more susceptible to Aβ accumulation as compared to other organs with higher BACE2 activity [25]. This would be consistent with its proposed neuroprotective role of BACE2. However, another study demonstrated considerable levels of BACE2 in the human brain [26], especially in the cortical layer and areas near blood vessels. Notably, BACE2 expression in subregions of the brain such as the ventromedial hypothalamus and mammillary body has also been documented [23]. At the cellular level, BACE2 is found to be expressed in both neurons and astrocytes [27], with higher BACE2 levels residing in the latter [25]. Other studies have shown that BACE2 is also expressed in spinal cord neurons [23], hippocampal neurons, and temporal cortical pyramidal neurons [28]. Thus, BACE2 expression is found in many regions of the central nervous system.

Interestingly, individuals diagnosed with mild cognitive impairment, preclinical AD, and AD exhibit an increase in the expression and enzymatic activities of BACE2 [27]. Importantly, in the AD brain, BACE2 protein levels inversely correlate with higher Braak stages [29], suggesting that low BACE2 levels portends increasing severity of AD. Although why BACE2 expression describes such a trend in the AD brain remains unclear, it is noteworthy that a recent report has identified RCAN1 (regulator of calcineurin 1) to be a regulator of BACE2 expression [30]. In their report, Qiu and colleagues demonstrated that RCAN1 promotes BACE2 expression by inhibiting its turnover by the proteasome [30]. Notably, RCAN1 is highly expressed in the human brain and its expression is further increased in aged and AD brains, which BACE2 expression mirrors. RCAN1 levels are reportedly upregulated by stress (cortisol), APOE4, and inflammation (NF-κB), all of which are associated with AD. Thus, the initial increased levels of BACE2 observed in AD brains may be at least in part explained by the positive effects of RCAN1 on its expression. However, whether there is a corresponding decrease in RCAN1 levels with higher Braak stages of AD remains to be characterized. Moreover, RCAN1 also promotes BACE1 expression [31], which makes it a challenging candidate to target to enhance BACE2 expression as a strategy to afford neuroprotection in AD.

AMYLOIDOGENIC AND NON-AMYLOIDOGENIC ACTIVITIES OF BACE2

Although this review is focused on the non-amyloidogenic role of BACE2, it is important for us to also highlight some studies that implicate an amyloidogenic role for BACE2. For example, Wang and colleagues have recently reported that artificial and AD-associated mutations that perturb the juxtamembrane helix of AβPP may—at a slight extent—trigger BACE2 to function as a conditional β-secretase to cleave at the same β-site as BACE1 [32]. BACE2 can also cleave at the β-site when clusterin, also known as the third most predominant genetic risk factor for late-onset AD [33, 34], binds to the juxtamembrane helix of AβPP [32]. Other studies have also hinted at the prospect of BACE2’s involvement in Aβ generation and AD by being a β-secretase. In one of these studies, cells co-transfected with AβPP bearing the Flemish mutation (which typically causes early-onset AD with cerebral amyloid angiopathy and large senile plaques [35]) and BACE2 resulted in an increase in C99 fragments that can possibly be cleaved by γ-secretase to yield the eponymous Aβ peptide [36]. Another study suggested that the triplication of BACE2 as a result of Trisomy 21 could contribute to AD in individuals with DS, as immunoreactivity of BACE2 was found to co-localize with NFTs in the brains of individuals with DS-AD but not in their counterparts without AD pathology [37]. Furthermore, the cleavage activity of BACE2 at the β-site was confirmed in mass spectrometric analyses that revealed identical cleavage fragments when BACE1 or BACE2 was incubated with a shorter AβPP peptide containing the β-site [36]. In addition, BACE2 overexpression increased the presence of soluble peptide AβPPb (sAβPPb) which is a BACE1 cleavage product [28]. It was assumed that an increase in sAβPPb would lead to a concomitant increase in Aβ levels. Unexpectedly, this increase in sAβPPb concentration is associated with diminished (rather than increased) Aβ levels [23] which is of a significant seven-fold decrease in magnitude [28]. This is probably due to the alternative non-amyloidogenic pathways by which BACE2 processes AβPP and Aβ, which will be elaborated below.

Two in vivo studies utilizing transgenic mice revealed that BACE2 might not play a role in the amyloidogenic pathway. Firstly, mice that overexpressed BACE2 revealed no alterations in the concentration of Aβ₄₀ and Aβ₄₂ levels in the hippocampus and cerebral cortex—areas that are implicated in AD. Moreover, there were no observable differences in cognitive dysfunction and cholinergic degeneration, which are typical hallmarks synonymous with people with AD [38]. Subsequently, the same group observed similar findings in double transgenic mice that overexpressed BACE2 and APP. This suggests that BACE2 overexpression is not involved in the amyloidogenic pathway and does not trigger cognitive dysfunction or cholinergic degeneration associated with AD [39]. Interestingly, the authors also hinted that BACE2 may have some protective roles in the double mutant mice, as its expression was increased in response to the transgenic co-expression of AβPP [39]. Besides the transgenic approach, BACE2 knockout mice have also been generated. Like its transgenic counterparts, BACE2 exon 6 knockout mice lacked any observable physical and behavioral changes, even though exon 6 comprises one of two active sites of BACE2. However, fibroblasts derived from these mice secreted more Aβ although this was not the case for BACE2 knockout-derived primary neurons [40]. Notwithstanding these findings, the potential protective role of BACE2 was further corroborated in several overexpression-based experiments involving transfected cell lines. Firstly, conditioned medium collected from SK-N-SH cells overexpressing BACE2, and not cells overexpressing BACE1, demonstrated a sevenfold decrease in Aβ₄₀ levels [28]. Next, in HEK293T cells transfected with BACE2-expressing plasmids, BACE2 was more efficient at cleaving the product of β-secretase (AβPP β-CTF) than cleaving at the β-site. The cleavage of AβPP β-CTF occurs within the Aβ region after Phe-19 and Phe-20 to generate a non-neurotoxic Aβ₁₉ fragment [36]. Subsequent studies reported similar observations regarding the strong affinity of BACE2 for the Phe-19 and Phe-20 θ-cleavage site [19, 20, 41, 42]. For this reason, BACE2 is widely regarded as a θ-secretase. In alignment with these studies, antisense oligomer-mediated knockdown of BACE1, not BACE2, decreased Aβ levels whereas the overexpression of BACE2 reduced Aβ peptide levels [41]. Consistent with this, HEK293 cells expressing BACE2 alone or double transfected with BACE2 and BACE1 generated less Aβ_40/42 as compared to cells expressing BACE1 alone [20]. Furthermore, targeted RNAi-mediated inactivation of BACE2 in HEK293 cells transfected with wildtype APP695 or APP695sw double mutant increased the amount of secreted AβPP and Aβ [22]. Similarly, Sun and colleagues demonstrated that a cell line expressing BACE2 and Swedish mutant APP695 also considerably reduced Aβ production [43], with the same group recapitulating these findings in transgenic AD mice and identifying a novel BACE2 θ-cleavage site on AβPP to generate a non-amyloidogenic C80 fragment that is further processed by γ-secretase to yield a truncated Aβ₄₂ peptide [19]. Notably, individuals harboring the Swedish mutation produce about 6–8 times more Aβ than normal individuals [44]. More recently, Huentelman and colleagues demonstrated that overexpressing BACE2 in BE(2)-M17 human neuroblastoma cells that normally express BACE2 at a low level led to decreased Aβ₄₀ and Aβ₄₂ concentrations in cell culture media [45]. Taken together, these various studies strongly support a non-amyloidogenic role of BACE2 that is protective against Aβ₄₂-induced pathology.

Interestingly, BACE2 can also degrade Aβ itself. This was discovered in a functional screen of 352 proteases where BACE2 emerged as the top Aβ-lowering protease [18]. In this study, recombinant BACE2 degraded synthetic Aβ at an optimal pH of 3.5 at a rate of about 150 times faster than BACE1. Furthermore, BACE2 was found to cleave Aβ at three different sites: Phe19-Phe20, Phe20-Ala21, and Leu34-Met35, where the latter was identified to be the initial and main cleavage site. Cells overexpressing APP alone generated aggregation-prone Aβ species Aβ₄₂, Aβ₄₀, Aβ₃₇, Aβ₃₈ and Aβ₃₉, whilst cells that co-expressed APP and BACE2 decreased the levels of the aforementioned Aβ species and produced non-AD related species Aβ₁₉, Aβ₂₀, and Aβ₃₄ instead [18]. As an acidic pH is needed for BACE2 to exhibit its Aβ-degradation activity, it is likely to involve the lysosome. Indeed, Alic et al. found that the degradation of Aβ by BACE2 occurs in LAMP2-positive compartments of neurons as exemplified by the high degree of colocalization between BACE2, its substrate Aβ₄₀, and degradation product Aβ₃₄ with the components of the chaperone-mediated autophagy pathway [13]. BACE2 may also be degraded via the macroautophagy-lysosome pathway [46] and the ubiquitin-proteasome pathway [30]. Consistent with this, the inhibition of BACE2 degradation via lysosomal inhibitor treatment with chloroquine or NH₄Cl in 4EB2 cells, which express both BACE2 and the human APP Swedish mutant, led to an increase in BACE2 levels and non-amyloidogenic C80 levels [46].

INSIGHTS FROM HUMAN BRAIN ORGANOID MODELS OF AD

Although the promising results above collectively suggest that BACE2 cleavage activity precludes the generation of aggregation prone Aβ species synonymous with AD, they still beg the question whether this phenomenon is mirrored in the human brain because these studies involve the artificial manipulation of the gene dose of BACE2 and were carried out in cultured cells or in rodents. Hence, the use of human cerebral organoids has started to gain traction as the new frontier of neurodevelopmental and disease modelling in a bid to address the inadequate translational potential of current preclinical models to better recapitulate the physiological context of the human brain.

Cerebral organoids are three-dimensional aggregates of cells that develop organized and distinct brain regions composed of various neuronal and non-neuronal subtypes of cells. This can be achieved by using the seminal method first described by Lancaster and colleagues [47, 48]. In this approach, pluripotent stem cells (PSCs) are seeded onto low attachment 96-well plates to generate embryoid bodies (EB). Over the next few days, each well will yield an EB that will undergo germ layer differentiation followed by neural ectodermal induction, which is an important step in embryogenesis as brain tissue is derived from the ectoderm. After which, EBs are embedded in Matrigel to promote the development of neuroepithelial rosettes composed of radially organized neuroprogenitors. These EBs will mature into cerebral organoids that can be maintained in culture on an orbital shaker or bioreactor for extended periods of time. Subsequently, these cerebral organoids should stain positively for hindbrain, midbrain, and forebrain markers. Over time, cerebral organoids will express a rich diversity of cell types including those from all the six layers of the cerebral cortex, as well as hippocampal neurons, ventral forebrain neurons, dopaminergic neurons, glutamatergic neurons, GABAergic neurons, astrocytes, and oligodendrocytes. Sometimes, other regions such as the choroid plexus and retinal pigmented epithelium can also be observed [47, 49, 50]. Emerging evidence has established similarities between cerebral organoids and human fetal brains in terms of epigenomics [51] and proteomics [52]. Remarkably, brain organoids can produce coordinated electrical oscillations in groups of neurons resembling brain waves observed in newborns [53]. Therefore, the cytoarchitectural and functional resemblance shared with some degree of fidelity between brain organoids and the human brain render the use of organoids as an attractive option to model neurological disorders, especially neurodevelopmental disorders. As these organoids may be cultured for prolonged periods, several groups including ours have also used the system to model age-related neurodegenerative diseases.

The first organoids modelling AD utilized ReNcell VM human neural stem cells that overexpressed APP and presenilin-1 to mimic familial AD (FAD) mutations. These FAD organoids exhibit enhanced levels of Aβ deposits and phosphorylated tau (p-tau), where the latter was significantly reduced by the treatment of these disease-associated organoids with β- or γ- secretase inhibitors [54]. Not only did this study demonstrate that organoids can recapitulate disease-specific hallmarks, it also illustrated that organoids are amenable to pharmacological intervention and have utility for drug development. In another study, brain organoids were generated from individuals with dupAPP or PSEN1 FAD mutations, both of which are associated with AD. These disease-associated organoids exhibited a time-dependent increase in the size and quantity of Aβ aggregates, as well as the concentration of Aβ oligomers. A significant difference in p-tau levels between disease-associated and control organoids was only observed at a later timepoint as compared to Aβ, which was observed earlier, suggesting that these organoids recapitulate the trajectory of AD pathologies as outlined in the amyloid cascade hypothesis. Moreover, these AD-associated organoids also contained enlarged endosomes with abnormalities typical of those observed in mouse AD models and AD patients [55]. Other cerebral organoids generated from cell lines with a missense mutation (A243E) in presenilin 1 and individuals with DS similarly exhibit elevated level of Aβ deposits, p-tau pathology, and cell death compared to their normal controls [56]. Not surprisingly, the same trend is observed in APOE4-bearing organoids derived from CRISPR/Cas9-edited APOE3 induced pluripotent stem cells (iPSCs) relative to APOE3-containing organoids. This is expected as the APOE4 allele is the strongest risk factor for sporadic AD. Conversely, in the same study, APOE3 organoids generated from the APOE4 iPSCs attenuated the AD pathologies observed in the APOE4 organoids [57]. Thereafter,

Zhao et al. used ELISA, immunostaining, and transcriptomic profiling with AD organoids to demonstrate that AD status has an effect on boosting Aβ₄₀ and Aβ₄₂ levels independently of APOE4, as well as increasing cell death, synaptic loss, and stress granule formation. APOE4 status was related to p-Tau generation, which suggests that APOE4 may drive p-Tau generation without Aβ. Consistent with this, isogenic conversion of APOE4 to APOE3 decreased the AD phenotypes observed earlier in APOE4 organoids, which corroborates with the fact that APOE4 is a significant AD risk factor [58]. Another recent paper also performed transcriptomic analyses utilizing brain organoids exposed to human serum to simulate the serum components associated with the malfunctioning of the blood brain barrier in AD. In this study, serum-exposed organoids showed a decline in synaptic function in neurons and an increase in insoluble Aβ [59].

With regards to BACE2, several groups have generated cerebral organoids from individuals with DS whose BACE2 and APP are naturally triplicated due to their position on Chromosome 21. In one of these studies, Murray and colleagues generated an isogenic DS iPSC model [60] that was used in a follow-up study that analyzed conditioned media from organoids grown from these iPSCs. The study revealed that the amount of non-amyloidogenic BACE2 cleavage products Aβ₁₉, Aβ₂₀, and Aβ₃₄ increased in these DS organoids. This phenomenon was corroborated in cerebrospinal fluid samples obtained from individuals with DS. Subsequently, a copy of the triplicated BACE2 allele in these DS organoids was eliminated through CRISPR-Cas9 technology. As anticipated, DS organoids harboring BACE2 in a diploid state exhibited early amyloid and tau pathology as compared to unedited trisomic organoids. The subsequent treatment of these CRISPR-edited DS organoids with β-secretase and γ-secretase inhibitors prevented the formation of amyloid and tau pathology observed previously [13]. This supports a neuroprotective role for BACE2 in AD pathogenesis. In a similar vein, recent research findings from a group in China revealed rare BACE2 loss-of-function variants in patients with Hirschsprung disease and found that brain organoids generated from these individuals exhibit increased amounts of Aβ oligomers and cell death. In contrast, they found that BACE2 overexpression in brain organoids carrying the familial AD APP Swedish/Indiana mutation promoted a dual reduction of Aβ and cell death [21].

SINGLE NUCLEOTIDE POLYMORPHISMS IN BACE2 CORRELATE WITH THE AGE OF DEMENTIA ONSET

Although the evidence presented above collectively supports a neuroprotective role for BACE2 in AD pathogenesis, it remains intriguing as to why not all individuals with DS harboring an extra copy of BACE2 are protected from AD. Single nucleotide polymorphisms (SNPs) for BACE2 exist in the population of people with DS, suggesting that genetic background may determine the predisposition of DS individuals to AD. The premise that BACE2 genetic variations may modify the risk of individuals for AD has certainly been examined previously and has gained considerable traction recently. In 2005, Myllykangas and colleagues have reported the association of BACE2 haplotype with AD [61]. In particular, the rs2252576 (T allele) from the Myllykangas study is also associated with an earlier age of dementia onset in individuals with DS as reported in a separate study that also identified other significant SNPs in the population of DS individuals [62]. More recently, Huentelman and colleagues have also identified SNPs within the BACE2 locus that are associated with altered AD risk preferentially in APOE4 non-carriers, of which a subset with mild cognitive impairment or AD contained the rs2012050 SNP that is correlated with decreased BACE2 gene expression and increased Aβ₄₂ cerebrospinal fluid levels [45]. Further, as mentioned above, Luo and colleagues have recently also reported rare variants of BACE2 that occur in individuals with Hirschsprung disease [21]. Perhaps the most direct evidence supporting a role of BACE2 in AD is the identification of a patient with early-onset AD (EOAD) who harbours a mutation in BACE2. In a screen for de novo variants that might participate in the genetic determinism of sporadic EOAD (typically with disease onset before 65 years), Rovelet-Lecrux and colleagues have identified two de novo copy number variations in two EOAD patients, with one (Patient EXT 804) harboring a 12 kb deletion within intron 1 of BACE2 (while the other harboring an APP duplication) [63]. The BACE2 intronic deletion overlaps with enriched H3K4Me3 and H3K27Ac histone marks, which are epigenetic modifications that correlate with promoters and enhancers, respectively, suggesting the likelihood of reduced BACE2 expression in Patient EXT 804. Supporting this, reverse transcription quantitative multiplex PCR performed by the group on cultured fibroblasts of the patient and 10 control individuals revealed that the abundance of BACE2 transcripts in the patient are among the lowest of the samples [63]. Although the large inter-individual variability among controls precluded a firm conclusion of the putative pro-amyloidogenic effect of the BACE2 mutation, the premise is consistent with several reported studies supporting a neuroprotective role of BACE2 as discussed above [13, 18–22]. Nonetheless, the disease causality and mechanism related to this index BACE2 intronic mutation remain to be clarified. In our attempt to address this, our preliminary (unpublished) results have revealed that cerebral organoids generated from this patient produced more amyloid plaques and phosphorylated tau than those generated from his asymptomatic parental control. Importantly, we documented that the appearance of amyloid plaques preceded that of phosphorylated tau, which aligns with the order of pathology proposed by the amyloid cascade hypothesis (Yeap et al., unpublished observations).

BEYOND AD: THE ROLE OF BACE2 IN DIABETES AND ITS IMPLICATIONS FOR THE DEVELOPMENT OF BACE2 THERAPEUTICS

Given the promise of BACE2 in offering neuroprotection, it seems intuitive to develop BACE2-based therapeutics for AD. In reality, the development of BACE2-specific drugs for AD has yet to catch on. The bulk of BACE2 drugs in development are actually BACE2 inhibitors designed mainly for the treatment of Type 2 diabetes (T2D) and not AD [64, 65] (Table 1). An earlier review has documented the patents for these inhibitors from 2010 to 2012 [66] and a recent search on SureChEMBL has yielded a few filed patents for BACE2 inhibitors from 2014 to date [67–73], but it seems that clinical trials have yet to be conducted using these inhibitors. Nonetheless, the use of BACE2 inhibitors for diabetes is important for us to appreciate given the emerging link between diabetes and AD, and the potential consequence that such an approach may pose to the brain. Indeed, there is a growing appreciation that poor sugar control, insulin resistance, and cognitive decline are all intertwined [74, 75], and that insulin resistance in the brain can cause disruptions in glucose metabolism and contribute to the pathogenesis of AD [76, 77]. For this reason, AD is also known as the “diabetes of the brain” or “Type 3 diabetes”.

Table 1.

Patents of BACE2 inhibitors published between 2013-present

Year of Publication	Title & (Patent Number)	Target(s)	Reference(s)
2017	Bace-2 inhibitory compounds and related methods of use (WO2017066742A1)	BACE2	[73]
2016	Compounds for inhibition of memapsin 1 (US9512099B2)	BACE2	[64, 65, 72]
2016	1,4 oxazines as BACE1 and/or BACE2 inhibitors (US9242943B2)	BACE1 or BACE2	[71]
2015	Spiro-[1,3]-oxazines and spiro-[1,4]-oxazepines as BACE1 and/or BACE2 inhibitors (US9079919B2)	BACE1 or BACE2	[70]
2015	Halogen-alkyl-1,3 oxazines as BACE1 and/or BACE2 inhibitors (US8987255B2)	BACE1 or BACE2	[69]
2015	Cyclopropyl-fused-1,3-thiazepines as BACE1 and/or BACE2 inhibitors (US8927535B2)	BACE1 or BACE2	[68]
2013	N-[3-(5-amino-3,3a,7,7a-tetrahydro-1H-2,4-dioxa-6-aza-inden-7-yl)-phenyl]-amides as BACE1 and/or BACE2 inhibitors (US8404680B2)	BACE1 or BACE2	[67]

It is noteworthy to highlight that BACE2 mRNA expression is highest in pancreatic islets [78, 79] that lends its role in diabetes. Supporting this, mice with an in-frame bi-allelic deletion of exon 6 of Bace2 exhibited lower blood glucose levels, better intraperitoneal glucose tolerance, higher β-cell mass and numbers of Ki67-positive β-cells as compared to their wild-type counterparts [40]. Moreover, islet cells harvested from these mutant mice did not show any signs of apoptosis and released more insulin when stimulated with glucose as compared to those derived from wild-type mice. Similarly, the proliferation of MIN6 cells, a pancreatic β-cell line, increased upon treatment with a Bace2 inhibitor. On the contrary, overexpressing Bace2 led to a decrease in cell proliferation that was restored after treatment with the same Bace2 inhibitor. This phenomenon is attributed to the ability of Bace2 inhibition to prevent the cleavage of Tmem27, a β-cell-transmembrane protein and a Bace2-specific substrate that stimulates both β-cell growth and insulin activity. Alongside this, human islet cells exposed to this inhibitor also exhibit enhanced insulin secretion, although the proliferative capacity of the human islets did not change by much [78]. Interestingly, most individuals with type 2 diabetes (T2D) are presented with islet amyloid deposits [80] that are long considered as a classical pathological feature of T2D [81]. The deposits contain aggregates of amylin, also known as islet amyloid polypeptide (IAPP) [82], that promotes pancreatic islet amyloidosis upon its fibrillization, eventually leading to β-cell death [83]. Under normal circumstances, soluble IAPP functions as a neuro-pancreatic hormone that stabilizes blood glucose levels by decreasing the release of glucagon [84]. It is co-secreted with insulin to augment the latter’s function to trigger glucose uptake [85]. In rat pancreatic β-cell line INS1E overexpressing human IAPP, BACE2 expression was found to be upregulated by 2.2-fold. Subsequent overexpression of BACE2 in these cells decreased β-cell proliferation by 60%, increased reactive oxygen species levels by three-fold, and decreased insulin secretion by 25% as compared to controls. Once BACE2 was silenced, β-cell function was restored [79]. Intriguingly, IAPP can cross the blood-brain barrier and its amyloid deposits can cause brain damage [86]. Indeed, the presence of IAPP amyloid deposits has been documented in AD brains [87], even in the absence of clinical signs of diabetes in these individuals [88]. Furthermore, a recent report revealed that an AD-associated tau species directly interacts with IAPP, whereby the oligomerization of this tau fragment increased in the presence of IAPP [89].

Taking the findings above into account, the inhibition of BACE2 is expected to promote the function of pancreatic β-cells, which fueled the proposition that BACE2 may be a potential therapeutic target for diabetes. However, the complication is that IAPP is apparently a substrate of BACE2 as reported by Rulifson and colleagues [82], who further demonstrated that cleavage of hIAPP at the F15 and F23 residues reduces its propensity to aggregate. They proposed that enhancing, rather than inhibiting, BACE2 may provide translational benefits. Along the same vein, Diaz-Catalan and colleagues showed that Bace2 knockout mice fed with a high-fat diet led to increased weight gain, hyperinsulinemia, and insulin resistance, despite increased β-cell proliferation and higher levels of insulin and leptin. The authors concluded that these mice had damaged leptin and insulin signaling, which raised caution when using BACE2 inhibitors for the treatment of T2D [90]. Notwithstanding the controversies surrounding the outcomes of BACE2 inhibition in diabetic models, an important additional caveat to note is that BACE2 inhibition, when applied systemically, runs the danger of promoting amyloidosis in the brain and may prove to be catastrophic. Clearly, whether BACE2 inhibitors would exacerbate or ameliorate the pathology present in individuals with both AD and diabetes is a question that begs to be resolved. All in all, the link between BACE2, AD, and diabetes remains an enigma.

CONCLUDING REMARKS

To date, AD-related clinical trials have primarily leveraged on the inhibition of Aβ generation via β-secretase inhibitors. Unfortunately, current β-inhibitors are non-selective and inhibit both BACE1 and BACE2 [13, 91, 92], which might in part explain why drug trials have failed throughout the years [93]. This strategy thus seems counterproductive, given the multiple lines of evidence supporting a neuroprotective role for BACE2 in AD that we have presented above. Although the development of selective BACE1 inhibitors would certainly be useful, we propose that targeting BACE2 for therapeutic purposes represents a viable alternative strategy that merits consideration. Besides pharmacological approaches, BACE2 expression may be upregulated by genetic means involving microRNAs such as let-7c, which binds to BACE2 and triggers its expression via RNAa to release C83 or C80 fragments instead of C99 that greatly decrease Aβ generation. Notably, let-7c is downregulated in AD mice and individuals with DS [94]. Using exogenously introduced miRNAs such as let-7c to boost BACE2 gene expression therefore represents an intuitive approach. In addition, a recent study revealed that BACE2 can cleave Kv2.1, the main potassium efflux channel involved in neuronal apoptosis [95], which provides yet another reason why augmenting BACE2 function may be beneficial for AD. However, given the role of BACE2 in diabetes, the systemic effects of AD-targeting BACE2-enhancing drugs particularly in pancreatic β-cells need to be carefully assessed. Taken together, our review has highlighted the important role of BACE2 in the pathogenesis of AD and emphasized the need to focus more attention on BACE2 as a potential therapeutic target for AD.