Amyloid Precursor Protein and Mitochondria

Abstract

Amyloid Precursor Protein (APP) processing to amyloid beta (Aβ) is a major hallmark of Alzheimer’s disease (AD). The amyloid cascade hypothesis postulates that Aβ accumulation and aggregation causes AD, however many therapeutics targeting Aβ have failed recently. Decades of research describe metabolic deficits in AD. Mitochondrial dysfunction is observed in AD subjects within the brain and systemically. APP and γ-secretase are localized to mitochondria. APP can be processed within mitochondria and its localization to mitochondria affects function. Here we discuss the evidence showing APP and γ-secretase localize to mitochondria. We also discuss the implications for the function of APP and its cleavage products in regulating mitochondrial function.

Introduction

Diagnosis of AD is currently limited to postmortem neuropathological confirmation of amyloid plaques and tau tangles [1]. Plaques are composed of aggregating amyloid beta (Aβ), a cleavage product derived from amyloid precursor protein (APP). Most therapeutic efforts have targeted the reduction in amyloid plaque formation or the increased clearance of amyloid plaques [2].

APP is processed by secretase enzymes within lipid bilayers [3]. Two canonical processing pathways are known for APP, the non-amyloidogenic and the amyloidogenic pathways (Figure 1). Non-amyloidogenic APP processing begins with APP cleavage by α-secretase (ADAM10) generating a C-terminal fragment (C83) and releasing soluble APP α (sAPPα) into the extracellular space. Next, C83 is cleaved by the γ-secretase enzyme complex generating the APP intracellular domain (AICD) and p3. Amyloidogenic processing begins with APP cleavage by β-secretase (BACE) generating a C-terminal fragment (C99) and releasing soluble APP β (sAPPβ) into the extracellular space. C99 is then cleaved by the γ-secretase enzyme complex generating AICD and Aβ.

Figure 1. APP Processing Pathways.

A. Non-amyloidogenic APP processing begins with APP cleavage by α-secretase (ADAM10) generating a C-terminal fragment (C83) and releasing soluble APP α (sAPPα) into the extracellular space. Next, C83 is cleaved by the γ-secretase enzyme complex generating the APP intracellular domain (AICD) and p3. B. Amyloidogenic processing begins with APP cleavage by β-secretase (BACE) generating a C-terminal fragment (C99) and releasing soluble APP β (sAPPβ) into the extracellular space. C99 is then cleaved by the γ-secretase enzyme complex generating AICD and Aβ.

Autosomal dominant mutations in APP, PS1, and PS2 cause early-onset familial AD (FAD) [4]. Approximately 32 mutations in APP have been identified in FAD and most mutations are near the γ-secretase cleavage site. Approximately 176 mutations have been identified in PS1 but only 14 mutations in PS2 are linked to AD. Mutations in both PS1 and PS2 increase Aβ₄₂ to Aβ₄₀ ratios. APP is located on chromosome 21 and those with Down Syndrome have an extra copy of APP. The extra copy of APP leads to an increase in Aβ production and aggregation, which correlates with 40–80% of Down Syndrome patients developing early onset AD [5]. Other forms of FAD are caused by duplication of the APP gene, independent of Down Syndrome. These findings are the basis for the amyloid cascade hypothesis, which postulates that Aβ aggregation is what causes AD [2]. However, doubt has been raised concerning the validity of the amyloid cascade hypothesis due to the failure of therapeutics targeting Aβ [2].

Alzheimer’s disease (AD) appears to be a metabolic disease [6, 7]. Increasing evidence supports a role for metabolic deficits in AD onset and progression as reviewed elsewhere [8, 9]. Furthermore, many studies show localization of both APP and the γ-secretase complex at mitochondria. Here we will discuss the evidence of mitochondrial localization of APP, γ-secretase, and APP fragments with special consideration on the implications to mitochondrial function.

Amyloid Precursor Protein Localizes to Mitochondria

The first study describing localization of APP at mitochondria was published in 2003[10]. The N terminus of APP contains an ER targeting domain from amino acids 1–36. APP also contains three positive amino acid residues at +40, +44, and +51 which direct its mitochondrial localization. Human cortical neuron 1A (HCN1A) cells and isolated mitochondria were used to show import of APP into membranes. In 2004, a separate group reported mitochondrial localization of APP in rat pheochromocytoma derived cells (PC12) [11]. Based on these two independent studies it was concluded that APP localizes to mitochondria in vitro.

Beyond in vitro studies, APP localization at mitochondria has been observed in postmortem human brain samples. In the first study it was reported that APP was found within mitochondria but only in those with AD [12]. Aβ was also found within mitochondria and electron microscopy (EM) data suggested an association of APP with the translocase of the outer mitochondrial membrane (TOMM40). High-molecular weight APP complexes were observed within AD postmortem brain tissues and similar weight complexes containing TOMM40 and translocase of the inner mitochondrial membrane (TIM23) were observed. Further data showed an impairment of import of cytochrome oxidase subunits into the mitochondria in AD subject postmortem brain. Based on these findings it was concluded that APP “clogs” the mitochondrial protein import machinery, however, to date no data directly show this [12].

More recent studies reported APP within mitochondria of postmortem brain tissue, however these studies reported APP was found in both non-demented and AD samples. In 2012, APP was found within mitochondria, and it associated with heat shock protein 60 (HSP60) [13]. In a 2021 study, APP and APP C-terminal fragments were found in enriched mitochondrial fractions from postmortem brain tissue. APP and C-terminal fragments were observed at mitochondria in both non-demented and AD cases but where elevated in those with AD[14].

Studies depicting mitochondrial localization of APP span decades, several research groups, and model organisms. Despite this, the role of APP at mitochondria is not understood. Research interest has mostly focused on potential APP processing at or within mitochondria. As a result, additional studies showed γ-secretase components also localize to mitochondria, as discussed below.

Amyloid Precursor Protein Processing at Mitochondria

The γ-secretase enzyme complex consists of presenilin 1 and/or presenilin 2 (PS1, PS2), anterior pharynx-defective 1 (APH1), Nicastrin, and presenilin enhancer 2 (PEN2). PS1 and/or PS2 serve as the catalytic subunit of the γ-secretase complex [15]. In 2011, Pavlov et al. described localization of Nicastrin and PS1 to mitochondria in mouse embryonic fibroblast (MEF) cells [16]. It was further shown that full-length APP as well as N-terminal and C-terminal fragments of APP localized to mitochondria in MEFs and SH-SY5Y neuroblastoma cells. The localization of APP and its fragments within mitochondria was shown to be at the outer mitochondrial membrane and within mitoplasts (the inner mitochondrial membrane/matrix compartments). An in vivo study showed accumulation of C99 APP fragments within mitochondria of the 5X FAD mouse model [17]. Data also suggested that C-terminal fragments of APP within the mitochondria were cleaved into AICD by γ-secretase localized to mitochondrial compartments [16].

γ-secretase components are also found within mitochondria of non-transgenic rats. In rat brain, Nicastrin, APH1, PS1, and PEN2 were shown to localize to mitochondria using EM [18]. It was further shown that these proteins form high molecular weight complexes using blue-native polyacrylamide gel electrophoresis (BN-PAGE). APP C83 and AICD fragments were generated in isolated mitochondria from rat brain, and this could be inhibited with γ-secretase inhibitors [18].

A further study showed other mitochondrial proteases might be required for γ-secretase function within mitochondria in vitro. The serine protease, Omi/HtrA2 was shown to form complexes with PS1 in mitochondria and induce its cleavage [19]. Using MEF cells lacking Omi/HtrA2, it was shown that APP processing did not occur within mitochondria. These data suggest that to form an active γ-secretase complex the serine protease Omi/HtrA2 is required to cleave PS1 [19]. Overall, several studies leveraging both in vitro and in vivo models have shown mitochondrial localization of γ-secretase and APP fragments derived from γ-secretase cleavage.

Other work has implicated doubt in the localization of γ-secretase and APP at or within mitochondria. Mamada et al. used SH-SY5Y cell fractionations to show APP, BACE, ADAM10, and γ-secretase components are present in crude mitochondria and microsomes [20]. However, in more purified mitochondrial fractions lower levels of these proteins were observed. It must be noted though, that they were still observed, and more “pure” fractions of mitochondria still contained MAM and lysosome markers. This is of interest because mitochondria which contain APP and APP fragments might be undergoing mitophagy and localize to lysosomes. Increased MAM contact sites are also a hallmark of autophagy and mitophagy [21].

Functional γ-secretase enzyme is also localized to endoplasmic reticulum (ER) membranes associated with mitochondria, or mitochondrial associated membranes (MAMs) [22]. APP can be processed within MAMs to generate Aβ in vitro [23]. A more recent study showed that palmitoylated APP localizes to MAMs in neural progenitor cells (NPC) derived from familial AD subjects [24]. MAM content and activity is upregulated in AD [25]. Silencing of MAMs using both genetic knockdown and pharmacological inhibition reduces Aβ production [24]. Upregulation of MAMs appeared to drive APP to the cell surface which facilitated BACE cleavage and Aβ production. The majority of Aβ production driven by MAM activity was localized to axons [24].

Numerous studies spanning decades have described localization of APP and γ-secretase at mitochondria. There are inconsistencies between studies, and this might stem from variance in models used including reliance upon overexpression of proteins in vitro. As discussed below, these same issues are creating difficulty in understanding the function of APP within or at mitochondria.

Amyloid Precursor Protein and Mitochondrial Function

Studies examining the effects of APP on mitochondrial function have leveraged a wide variety of models including cell culture and in vivo work. Overall, there is not a current consensus on the function of APP in general nor within mitochondria. This is likely due to inconsistent results across studies and research groups as discussed below.

In HCN1A cells WT APP overexpression induced bioenergetic stress through reduced ATP production and cytochrome oxidase activity [10]. Differences in the effects of WT APP overexpression were examined between PC12 and human embryonic kidney (HEK) cells in a separate study. ATP levels were reduced in both cell types, but mitochondrial membrane potential, apoptosis, and nitric oxide changes were diverse [11]. This highlights the inconsistent findings between studies and cell types examined.

Overexpression of WT APP or APP harboring Swedish familial mutations (K670N/M671L) had no effect on mitochondrial function in N2a cells [26]. However, when cells were serum starved APP expression caused decreased ATP levels and increased apoptosis, ROS, and calcium levels. It was found that APP interacted with B-cell lymphoma protein 2 (Bcl-2) and heat shock proteins (HSPs) during apoptosis induced by starvation [26].

In SH-SY5Y cells stably transfected with WT APP 695, mitochondrial oxygen consumption was lower, mitochondrial ETC protein expression was lower, and mtDNA levels were higher [27]. There was also an increase in glycolytic flux with a reduction in COX activity. These effects were not blocked by inhibiting BACE and were not observed in SH-SY5Y cells expressing familial Swedish APP mutations (K670N/M671L) [27]. These data overall suggest a role for full-length APP at modulating mitochondrial function in the absence of Aβ production. Furthermore, mutated forms of APP associated with familial AD (and high Aβ production) did not exert the same effects on mitochondrial function.

Other studies have presented data supporting the opposite. In HEK293 cells expressing WT APP, mitochondrial oxygen consumption was reduced but this was not observed when BACE was overexpressed [28]. Inhibition of γ-secretase rescued mitochondrial oxygen consumption in HEK cells expressing WT APP [28]. Similar effects were observed in N2a cells.

Beyond full-length APP, accumulation of APP fragments within mitochondria have been associated with mitochondrial dysfunction. In the 5X FAD mouse model of AD, increased C99 APP fragments within mitochondria resulted in increased cytochrome c release, loss of complex II activity, and increased DNA damage [17]. These effects could be modulated with partial knockout of BACE, which reduced localization of C99 APP fragments at mitochondria and improved memory performance. These benefits of BACE partial knockout were only observed at 6 months of age [17]. Aβ can also be imported directly into mitochondria through the translocase of the outer mitochondrial membrane 40 (TOMM40) where it localizes to cristae and further studies show that mitochondrial localized Aβ induces mitochondrial dysfunction [11, 29–31].

The C99 APP fragment also accumulates in MAMs. Accumulation of C99 in MAMs and mitochondria was associated with decreased mitochondrial respiration and altered lipid content at both the total cellular level and mitochondrial level [22]. Specifically, sphingolipids were decreased, and ceramide was increased. The overall model used within this work were PS knockout MEFs. Reduction of PS function led to accumulation of C99 in this model. Accumulation of C99 was associated with an increase in MAM content, a phenomenon also observed in other AD models [25]. Inhibition of BACE function protected the MEFs from changes in mitochondrial function. These data overall show that APP processing occurs at contact sites between mitochondria and ER leading to altered bioenergetic function.

There is no consensus within in vitro and in vivo models of the distinct effects of full-length APP and APP processing products on mitochondrial function. However, APP null models have provided some profound data that APP is important for mitochondrial function and cognition. Whole body knockout of APP leads to behavioral changes in mice. APP knockout mice are viable which suggests related proteins like APPL1 or APPL2 have redundant functions, however phenotypes are still observed. Loss of APP in mice reduces motor performance on the rotarod test and increases gliosis in the cortex and hippocampus [32]. Astrocytes from APP knockout mice have altered calcium dynamics, with reduced calcium spikes. Mitochondrial calcium signaling was also reduced and increased fragmented mitochondrial structure was observed [33]. Replacing APP expression in the astrocytes restored mitochondrial morphology [32].

In HeLa cells, loss of kunitz-type protease inhibitor (KPI) regions in longer APP isoforms reduce its localization at mitochondria. Loss of mitochondrial localized APP resulted in fragmented mitochondrial morphology, increase ROS, reduced ATP, and mitochondrial membrane depolarization [34]. Further evidence suggests that APP may be important for autophagy and mitophagy function. APP interacts with ligases and possibly serves as a recognition receptor for ubiquitination during autophagy. The cytosolic domain of APP appears to facilitate the ubiquitination of proteins important for synaptic transmission [35].

Loss of APP in cell and animal models affects mitochondrial function and overexpression of APP also affects mitochondrial function. Further studies are warranted in understanding the exact mechanisms and to discern the roles of full-length APP versus APP processing products in the regulation of mitochondrial function.

Concluding Remarks

Studies from multiple model types and numerous groups support the localization of APP and γ-secretase at mitochondria. These studies which span decades include in vitro, in vivo, and post-mortem brain tissue samples. APP and γ-secretase components are lipophilic proteins which are notably associated with membranes throughout the cell. Therefore, it is not surprising that they would localize to a hydrophobic mitochondrial membrane. APP is typically thought to be processed at the plasma membrane or in the endosomal system, where secretase enzymes are also bound to lipid bilayers. This type of environment exists within mitochondria and near the outer mitochondrial membrane. APP and γ-secretase localization to mitochondria or to mitochondrial associated ER membranes are not mutually exclusive. Some studies indicate that the APP localized at mitochondria is not glycosylated (based on molecular weight) and glycosylation of proteins can dictate cellular trafficking. Future work should focus on understanding if APP is trafficked from the ER to the mitochondria or through a separate endosomal transport mechanism.

The function of APP and/or γ-secretase at mitochondria is also not understood. A clear role does exist, as knockdown of APP significantly alters mitochondrial function in cells and animal models. However, most data was generated in studies of APP overexpression which itself can cause off target effects on mitochondrial function. There are inconsistencies between prior published work which could stem from overexpression of APP and altered expression levels between studies as well as variance in cell types used. Future studies need to harmonize models to capitalize on expertise surrounding mitochondrial biology and APP function.

Understanding the role of APP at mitochondria is imperative for AD research. This includes understanding the role of full-length APP and APP processing products. There is strong evidence that mitochondrial function can affect APP localization and APP processing [3, 36–39]. As we have reviewed here APP and its processing can also affect mitochondrial function [3, 36]. This relationship likely explains AD biomarkers including low cerebrospinal fluid Aβ levels and increased insoluble brain Aβ deposits. While the cause of sporadic AD is not understood, metabolic deficiencies are obvious. Targeting these deficiencies could affect Aβ burden and cognition in sporadic AD. Furthermore, FAD and Down Syndrome are caused by mutations in APP or PS1/PS2 and APP overexpression. Understanding the role of FAD mutations and effects of APP overexpression on mitochondria could inform new therapeutic targets for these groups.

Highlights.

• Amyloid Precursor protein (APP) localizes to mitochondria and associated membranes

• γ-secretase localizes to mitochondria and participates in APP processing

• Changes to APP expression or mitochondrial localization affect mitochondrial function