Amyloid precursor protein‐mediated mitochondrial regulation and Alzheimer's disease

Abstract

Despite clear evidence of a neuroprotective physiological role of amyloid precursor protein (APP) and its non‐amyloidogenic processing products, APP has been investigated mainly in animal and cellular models of amyloid pathology in the context of Alzheimer's disease. The rare familial mutations in APP and presenilin‐1/2, which sometimes drive increased amyloid β (Aβ) production, may have unduly influenced Alzheimer's disease research. APP and its cleavage products play important roles in cellular and mitochondrial metabolism, but many studies focus solely on Aβ. Mitochondrial bioenergetic metabolism is essential for neuronal function, maintenance and survival, and multiple reports indicate mitochondrial abnormalities in patients with Alzheimer's disease. In this review, we focus on mitochondrial abnormalities reported in sporadic Alzheimer's disease patients and the role of full‐length APP and its non‐amyloidogenic fragments, particularly soluble APPα, on mitochondrial bioenergetic metabolism. We do not review the plethora of animal and in vitro studies using mutant APP/presenilin constructs or experiments using exogenous Aβ. In doing so, we aim to invigorate research and discussion around non‐amyloidogenic APP processing products and the mechanisms linking mitochondria and complex neurodegenerative disorders such as sporadic Alzheimer's disease.

Linked Articles

This article is part of a themed section on Therapeutics for Dementia and Alzheimer's Disease: New Directions for Precision Medicine. To view the other articles in this section visit http://onlinelibrary.wiley.com/doi/10.1111/bph.v176.18/issuetoc

Abbreviations

ABAD

amyloid‐binding alcohol dehydrogenase

AICD

amyloid precursor protein intracellular domain

APP

amyloid precursor protein

Aβ

amyloid β

BACE1

β‐site amyloid precursor protein cleaving enzyme 1

CTF

carboxyl‐terminal fragment

Cu‐ATSM

copper‐labelled diacetyl‐bis(N4‐methylthiosemicarbazone)

Drp1

dynamin‐related protein 1

ER

endoplasmic reticulum

FAD

flavin adenine dinucleotide

FADH₂

reduced flavin adenine dinucleotide

Fis1

mitochondrial fission 1

MAM

mitochondria‐associated membrane

MFN1/2

mitofusin 1 and 2

MPTP

1‐methyl‐4‐phenyl‐1,2,3,6‐tetrahydropyridine

NAD

nicotinamide adenine dinucleotide

NADH

reduced nicotinamide adenine dinucleotide

OPA1

optic atrophy 1

OXPHOS

oxidative phosphorylation

PS‐1

presenilin‐1

sAPPα

soluble APPα

sAPPβ

soluble APPβ

Introduction

Amyloid precursor protein (APP) is a transmembrane protein that consists of an extracellular domain, a single transmembrane domain and a short cytoplasmic fragment. APP is encoded by APP (chromosome 21q11.2‐q21) (Goldgaber et al., 1987), a member of a gene family that includes APP‐like proteins 1 and 2 (APLP1 and APLP2) (Sprecher et al., 1993). Alternative splicing of APP generates eight isoforms designated according to their number of amino acids, of which three are the most common: APP₆₉₅ is expressed predominantly in the Central nervous system (CNS), while APP₇₅₁ and APP₇₇₀ are more ubiquitously expressed (Sandbrink et al., 1996).

APP is the parent protein of a complex family of peptides. APP is processed from its membrane‐bound holoform via two main pathways (Figure 1). The dominant non‐amyloidogenic pathway involves cleavage of APP by α‐secretase (predominantly ADAM10) (Kuhn et al., 2010) and produces a large N‐terminal ectodomain, soluble APPα (sAPPα), which is secreted into the extracellular medium, and a carboxyl‐terminal fragment (CTF) called CTFα/CTF83 (Kojro and Fahrenholz, 2005). CTFα is endocytosed and undergoes further cleavage by the γ‐secretase complex, yielding a small peptide called p3 and APP intracellular domain (AICD) fragments (Kummer and Heneka, 2014). In contrast, amyloidogenic processing of APP involves cleavage by β‐secretase [β‐site amyloid precursor protein cleaving enzyme 1 (BACE1)] and subsequent release of a shorter ectodomain, soluble APPβ (sAPPβ), into the extracellular medium and a 99 amino acid fragment (CTFβ/CTF99) in the membrane (Vassar et al., 1999). Cleavage of CTFβ by the γ‐secretase complex generates amyloid β (Aβ) peptides (mainly Aβ₄₀ and Aβ₄₂) and AICD fragments (Kummer and Heneka, 2014). Aβ peptides are a major component of the cerebral plaques found in patients with Alzheimer's disease (Jarrett et al., 1993; Mann et al., 1996). Cleavage by α‐secretase occurs at a position within the sequence of Aβ and therefore precludes its formation. Further cleavage of AICD fragments by caspases or caspase‐like proteases results in two additional fragments, Jcasp and C31 (Lu et al., 2000; Bertrand et al., 2001).

Figure 1.

APP processing and cleavage products. Non‐amyloidogenic processing of APP (right) involves cleavage by α‐secretase, resulting in the generation of sAPPα into the extracellular space and intracellular CTFα/CTF83. CTFα undergoes further cleavage by the γ‐secretase complex, yielding p3 and AICD fragments, which are further processed by caspases, producing Jcasp and C31. The amyloidogenic pathway (left) involves cleavage by β‐secretase, with a subsequent release of sAPPβ into the extracellular medium, and CTFβ/CTF99 in the membrane. Cleavage of CTFβ by the γ‐secretase complex generates Aβ peptides, AICD fragments, Jcasp and C31. Cleavage by α‐secretase occurs at a position within the sequence of Aβ, and therefore precludes its formation.

Together with the intraneuronal tangles of aggregated hyperphosphorylated tau protein, the extracellular accumulation of Aβ into amyloid plaques is a hallmark of Alzheimer's disease. In several pedigrees of early‐onset familial Alzheimer's disease, point mutations in APP and PSEN1, the gene encoding the γ‐secretase component presenilin‐1 (PS‐1), are linked to altered APP processing and abnormal accumulation of Aβ (Mullan and Crawford, 1993; Hunter and Brayne, 2018). Although Alzheimer's disease caused by Mendelian variants in these and other genes constitutes a tiny proportion of disease burden compared with sporadic cases, genes associated with familial Alzheimer's disease have influenced research disproportionately. Consequently, the majority of cellular and animal models in the field involve various combinations of mutant versions of these proteins. A large number of studies have also used synthetic Aβ to show a wide range of toxic effects in vitro, which include mitochondrial toxicity. As a result, research has focused largely on the amyloidogenic properties and toxicity of Aβ, and the physiological function of APP and its non‐amyloidogenic fragments has received less attention. Extensive reviews focused on animal studies of APP mutants and the role of Aβ in mitochondria are available (Pagani and Eckert, 2011; Kaminsky et al., 2015; Kawamata and Manfredi, 2017). Here we focus instead on mitochondrial studies in human Alzheimer's disease and on potential research gaps looking at non‐amyloidogenic APP processing and mitochondrial function.

Mitochondrial bioenergetic metabolism is essential for neuronal function, maintenance and survival, and mitochondrial dysfunction has become an established hallmark of neurodegeneration. Neurons rely on mitochondria for a variety of essential processes, including energy production, through the generation of ATP via the oxidative phosphorylation (OXPHOS) system, which is the main cellular source of ROS (Lopez Sanchez et al., 2016). Beyond their bioenergetic role, mitochondria are involved in other important cellular functions, including intracellular calcium homeostasis, iron metabolism and apoptosis signalling amongst others. Disruption of mitochondrial function and dynamics has been associated with a variety of neurodegenerative processes, including those involved in the development of Alzheimer's disease (Lionaki et al., 2015; Zorzano and Claret, 2015; Pei and Wallace, 2018), and APP and APP‐derived peptides have been reported to affect mitochondrial function. In this review, we focus on the role of full‐length APP and its non‐amyloidogenic fragments, particularly sAPPα, on mitochondrial bioenergetic metabolism. In doing so, we aim to invigorate research and discussion around non‐amyloidogenic APP processing products and the mechanisms linking mitochondria and complex neurodegenerative disorders such as Alzheimer's disease.

Mitochondrial bioenergetics in Alzheimer's disease

Early studies in Alzheimer's disease patients showed decreased brain glucose metabolism using fluorodeoxyglucose PET (Benson et al., 1983; Chase et al., 1984). This finding has been validated and is an established correlate of the disease, although the extent to which this is a cause or effect continues to be debated (reviewed by Chen and Zhong, 2013). One possible cause of decreased glucose utilization is mitochondrial dysfunction, which has been associated with a variety of neurodegenerative disorders (Johri and Beal, 2012).

The OXPHOS system is responsible for most of the energy production in cells. It consists of five multimeric protein complexes located in the inner mitochondrial membrane (complexes I–IV and the ATP synthase or complex V), encoded by both nuclear and mitochondrial DNA (mtDNA). Briefly, glucose is oxidized to pyruvate in the cytosol through glycolysis. Pyruvate enters mitochondria where it is converted into acetyl coenzyme A, which is further oxidized within the tricarboxylic acid cycle to yield reduced nicotinamide adenine dinucleotide (NADH) and reduced flavin adenine dinucleotide (FADH₂). These two molecules serve as electron donors for OXPHOS complexes I and II. ATP is produced by the shuttling of electrons by carrier molecules along the respiratory complexes, which is accompanied by the pumping of protons from the mitochondrial matrix into the intermembrane space. This electrochemical gradient across the inner mitochondrial membrane is used by the ATP synthase to catalyse the conversion of ADP and inorganic phosphate to ATP (Smeitink et al., 2001).

Mitochondrial OXPHOS metabolism was first associated with Alzheimer's disease through the discovery that cytochrome c oxidase (complex IV) enzymatic activity is decreased in some patients with sporadic Alzheimer's disease (Parker et al., 1990; Kish et al., 1992; Mutisya et al., 1994), which was linked to increased Aβ levels in vitro (Parks et al., 2001), and contributed largely to the development of the Alzheimer's disease mitochondrial cascade hypothesis (Swerdlow and Khan, 2004). The specificity of complex IV decreases in Alzheimer's disease has been questioned, as cytochrome c oxidase activity is also reduced in patients affected by spinocerebellar ataxia type I and Friedreich's ataxia (Kish et al., 1999) and other neurodegenerative disorders (Cavelier et al., 1995; Arnold, 2012). However, the balance of evidence is that cytochrome c oxidase is decreased in the brain tissue of patients with Alzheimer's disease.

The early demonstration of semi‐quantitative histochemical staining for cytochrome c oxidase in post‐mortem brain samples (Kageyama and Wong‐Riley, 1982) led to several studies showing higher numbers of cytochrome c oxidase‐negative neurons in various regions of Alzheimer's disease brain (Simonian and Hyman, 1993; Gonzalez‐Lima et al., 1997; Wong‐Riley et al., 1997). The veracity of this defect has been reinforced by biochemical studies in peripheral cells (Parker et al., 1990) and the brain (Parker et al., 1994; Maurer et al., 2000). It has been hypothesized that mitochondrial dysfunction due to complex IV deficiency drives an overproduction of Aβ (Crouch et al., 2007), as the presence of intra‐mitochondrial Aβ has been demonstrated in Alzheimer's disease brain (Lustbader et al., 2004). An elegant study addressed this question by using neuron‐specific complex IV‐deficient mice carrying mutant APP and PS‐1, which surprisingly resulted in lower amyloid plaque burden and Aβ levels (Fukui et al., 2007). Therefore, this suggests that the cytochrome c oxidase deficiency in Alzheimer's disease does not drive amyloid production.

Some evidence for other OXPHOS defects has been reported in samples from Alzheimer's disease patients (Johri and Beal, 2012). Complexes I and IV transcripts were found to be decreased in one study (Aksenov et al., 1999), while another found evidence for complex I subunit mRNAs to be specifically decreased in Alzheimer's disease brain (Manczak et al., 2004). Furthermore, complex I subunit protein levels were found to be decreased in different areas of Alzheimer's disease and Down syndrome brain (Kim et al., 2001), but biochemical studies of enzymatic activity corroborating these findings are lacking in Alzheimer's disease. A meta‐analysis of complex I and complex IV findings found that only complex IV was consistently reported to be decreased in Alzheimer's disease (Holper et al., 2018). This analysis also showed that decreases of both complexes I and IV were consistently found in studies of human brain ageing (Holper et al., 2018). It is now well established that ageing is associated with lower OXPHOS capacity in human muscle (Trounce et al., 1989) and, in human, non‐human primate and rodent brain (Bowling et al., 1993; Ferrandiz et al., 1994; Ojaimi et al., 1999). Since ageing is the strongest risk factor for sporadic Alzheimer's disease, it is clear that an age‐related decline in mitochondrial OXPHOS can be a key contributor to Alzheimer's disease pathogenesis, where additional, disease‐specific mitochondrial perturbations exist (Crouch et al., 2007).

Multiple reports indicate additional mitochondrial abnormalities in Alzheimer's disease, including increased mtDNA oxidative damage (reviewed by Santos et al., 2013), pathogenic mtDNA deletions (Corral‐Debrinski et al., 1994; Hamblet and Castora, 1997), altered mtDNA methylation (Blanch et al., 2016) and mtDNA gene expression perturbations (Lunnon et al., 2017), which may result in increased oxidative stress. Indeed, imbalanced ROS, oxidative stress and damage are detected in Alzheimer's disease, including increased levels of oxidative markers in mtDNA (Mecocci et al., 1994) and free radicals (Reddy, 2006). This supports the hypothesis that chronic oxidative stress and changes in mtDNA may play a role in neurodegenerative processes. However, the specificity of these mitochondrial changes in Alzheimer's disease is uncertain, given that mtDNA abnormalities and oxidative damage are common features of several neurodegenerative disorders (reviewed in Cha et al., 2015). Similarly, studies have shown decreased protein and mRNA levels of the key transcriptional regulator of mitochondrial biogenesis PPARG co‐activator 1α in Alzheimer's disease brain, suggesting that mitochondrial biogenesis may also be impaired (Qin et al., 2009; Sheng et al., 2012).

Perturbations in mitochondrial dynamics and in proteins responsible for mitochondrial fusion [optic atrophy 1 (OPA1) and mitofusin 1 and 2 (Mfn1/Mfn2)] and fission [dynamin‐related protein 1 (Drp1, also referred to as DLP1) and mitochondrial fission 1 (Fis1)] have also been implicated in Alzheimer's disease. One study showed co‐localization of Drp‐1 and Aβ, increased levels of Drp1 and decreased levels of mitochondrial fusion proteins in brain samples (Manczak et al., 2011). However, another study reported decreased protein levels of Drp1, OPA1, Mfn1 and Mfn2 and significantly increased Fis1 levels in Alzheimer's disease brain tissue (Wang et al., 2009). Despite discrepancies in the changes in levels of Drp1, these studies suggest that enhanced mitochondrial network fission may be involved in neuronal dysfunction in Alzheimer's disease. This subject has been reviewed by others (Zhu et al., 2013). Further work is necessary to clarify whether mitochondrial fission/fusion abnormalities in the brain are causal or a product of other upstream perturbations in Alzheimer's disease.

APP and mitochondrial metabolism

The identification of a chimeric mitochondrial targeting sequence suggested that full‐length APP could be targeted to mitochondria in neuronal models or transgenic mice overexpressing APP (Anandatheerthavarada et al., 2003). It has been proposed that APP associates with two components of the mitochondrial protein translocation machinery, TOM40 and TIM23, and its accumulation prevents the import of nuclear‐encoded mitochondrial proteins, including OXPHOS subunits, therefore impairing mitochondrial function (Anandatheerthavarada et al., 2003; Devi et al., 2006) (Figure 2).

Figure 2.

Mechanisms of APP regulation of mitochondrial OXPHOS metabolism. APP accumulates in the protein import channels of mitochondria (TIM/TOM), preventing the import of nuclear‐encoded mitochondrial proteins, including subunits of OXPHOS complexes (I‐IV, ATP synthase) (1). Binding of sAPPα to an unknown cell membrane receptor, potentially the InsR, may activate the PI3K/Akt pathway to decrease mtDNA transcription of OXPHOS subunits (2). APP and APP‐derived peptides can be produced and processed at MAMs (3). An accumulation of unprocessed CTF99 and AICDs in mitochondria has been linked to altered lipid composition and disruption of mitochondrial membranes that interfere with OXPHOS function (4). Pathological binding of ABAD to Aβ in mitochondria causes leakage of ROS and mitochondrial dysfunction (5), but the effects of this binding on mtDNA processing have not been investigated yet. TIM, translocase of the inner membrane; TOM, translocase of the mitochondrial outer membrane.

Mitochondria form dynamic physical interactions with the endoplasmic reticulum (ER), known as mitochondria‐associated membranes (MAMs). This association constitutes a key signalling hub involved in fundamental cellular processes, including mitochondrial tethering and dynamics, lipid biosynthesis and intracellular calcium signalling (Gomez‐Suaga et al., 2018). There is strong evidence that MAMs are an intracellular site of APP processing (Schreiner et al., 2015; Del Prete et al., 2017). The physical and biochemical apposition of MAMs with mitochondria (Hayashi et al., 2009) supports the possibility that various APP‐derived peptides are able to interact with mitochondria (Figure 2). This may also account for previous reports that indicated processing of full‐length APP occurs in mitochondria (Ankarcrona and Hultenby, 2002; Hansson et al., 2006; Pavlov et al., 2011), which has been revisited and questioned recently, given that mitochondria do not possess the enzymes necessary to produce Aβ peptides from APP (Mamada et al., 2017). Instead, it is thought that Aβ is produced at mitochondria‐ER contact sites (Schreiner et al., 2015). The presence and processing of APP in MAMs has led to the hypothesis that alterations in ER‐mitochondria connectivity may result in perturbations in lipid homeostasis (Grimm et al., 2012), which may contribute to Alzheimer's disease pathology (Area‐Gomez et al., 2012; Del Prete et al., 2017).

In addition to full‐length APP, APP‐derived peptides have been detected in mitochondria, including Aβ (Crouch et al., 2005; Pagani and Eckert, 2011), AICD (Pavlov et al., 2011) and CTF99 (Devi and Ohno, 2012), or in MAMs (Pera et al., 2017), where they may alter mitochondrial function. The presence of AICD in the inner mitochondrial membrane has been linked to disruptions in mitochondrial distribution, morphology and bioenergetics (Ward et al., 2010), while an accumulation of unprocessed CTF99 has been associated with altered lipid composition of MAMs and mitochondrial membranes that interferes with the assembly of OXPHOS complexes and mitochondrial respiration (Pera et al., 2017) (Figure 2).

Amyloid‐binding alcohol dehydrogenase (ABAD), known under a variety of names including 17β‐hydroxysteroid dehydrogenase type 10 (HSD17B10) or mitochondrial RNase P protein 2 (MRPP2), has also been linked to mitochondrial perturbations in the context of amyloid pathology. It was shown that pathological binding of ABAD to Aβ in mitochondria increases leakage of ROS and results in mitochondrial dysfunction (Lustbader et al., 2004). As a key component of the RNase P complex responsible for the 5′ processing of mitochondrial tRNAs, ABAD has a direct role in mitochondrial gene expression and respiratory function (Sanchez et al., 2011) and is also involved in the modulation of mitochondrial function by oestrogens (Sanchez et al., 2015). Inhibition of the ABAD–Aβ interaction significantly reduced mitochondrial Aβ accumulation, leading to improvements in mitochondrial function and attenuation of mitochondrial ROS production (Yao et al., 2011). Surprisingly, despite its direct role in mitochondrial RNA processing, the consequences of ABAD's interaction with Aβ in the regulation of the mitochondrial genome have not been investigated.

Our in vitro studies have sought to elucidate the effects of wild‐type APP and its non‐amyloidogenic fragments, particularly sAPPα, on mitochondrial function, seeking to distinguish the physiological role of APP and its non‐amyloidogenic fragments from the known toxic effects of excessive Aβ resulting from overexpression of mutant versions of APP. Overexpression of wild‐type APP reduces mitochondrial respiration in vitro, while higher levels of Aβ, either by overexpression of mutant APP harbouring the Swedish mutation (Lopez Sanchez et al., 2017), or via overexpression of BACE1 (Schaefer et al., 2016), do not alter mitochondrial respiration. We found that this reduction in respiratory capacity in neuronal cells overexpressing APP correlated with reduced transcription of mtDNA and a metabolic switch to glycolytic metabolism, suggesting a programmed down‐regulation of OXPHOS (Lopez Sanchez et al., 2017). Furthermore, these results suggest that non‐amyloidogenic APP processing can impact mitochondrial regulation.

It will be important to disentangle epiphenomena arising from overexpression of APP in cellular and mouse models from physiological effects of APP processing. In this regard, there are remarkably few studies that have attempted to investigate holo‐APP or sAPPα levels in Alzheimer's disease brain. Davidsson et al. (2001) found lower APP levels in sporadic Alzheimer's disease brain samples compared to age‐matched controls. Two other studies support this finding, also showing that both sAPPα and sAPPβ levels were lower in sporadic Alzheimer's disease brain (Wu et al., 2011) and sAPPα levels were lower in the cerebrospinal fluid of patients with sporadic Alzheimer's disease (Sennvik et al., 2000). Furthermore, these authors stressed the need for measurement of APP‐derived peptides besides Aβ to gain further understanding of Alzheimer's disease pathogenesis. Thus, while Aβ levels are the strongest link between sporadic and Mendelian Alzheimer's disease, it remains to be determined what factors drive the increased β‐cleavage of APP in sporadic Alzheimer's disease cases. The possibility that decreased levels of the neurotrophic sAPPα is the key pathogenic driver in Alzheimer's disease remains under‐explored (Hunter and Brayne, 2018).

In summary, there is strong evidence that APP and its processing products are present in mitochondria and MAMs. In mitochondria, they interact with the protein import machinery, mtDNA and OXPHOS components, with detrimental consequences on mitochondrial dynamics and bioenergetic function. However, the mechanisms leading to mitochondrial dysfunction are not yet clear. For instance, mitochondria possess systems to prevent damage caused by misfolded proteins, including their own unfolded protein response mechanism, which is activated when misfolded proteins accumulate in the matrix, and mitophagy, which eliminates damaged mitochondria (Pellegrino and Haynes, 2015). The participation of these protective mechanisms in the context of amyloid pathology warrants further investigation. Furthermore, it is important that APP, sAPPα, Aβ and other APP fragments are measured and reported in studies that employ APP overexpressing models (Hunter and Brayne, 2018). Aβ levels are increased in models of amyloid pathology, and therefore, it has become accepted that Aβ accumulation is directly responsible for mitochondrial changes. However, this may reflect the fact that full‐length APP and other non‐Aβ fragments are not systematically measured, and these may be responsible for the cellular and phenotypic changes reported.

Neuroprotective role of APP and sAPPα – metabolic connections

APP has been investigated largely in animal and cellular models that express higher levels of APP, either in its wild‐type form or as mutant variants associated with familial Alzheimer's disease (reviewed by Drummond and Wisniewski, 2017; Sasaguri et al., 2017). These include first‐generation transgenic mice overexpressing mutant APP, such as APP V717F (also known as Indiana mutation) (Murrell et al., 1991) and the commonly used Swedish mutation (K670N/M671L) (Citron et al., 1992), including Tg2576 (Hsiao et al., 1996) and APP23 mice (Sturchler‐Pierrat et al., 1997). However, as discussed above, APP overexpression in these models also results in increased levels of full‐length APP and other amyloidogenic and non‐amyloidogenic fragments (Sasaguri et al., 2017) that is often overlooked.

Loss of function models have provided important insights into the roles that APP plays in neuronal function. APP knockout mice are viable and fertile; however, they display CNS anatomical and behavioural abnormalities (reviewed in Drummond and Wisniewski, 2017; Sasaguri et al., 2017) and age‐related neurodegenerative symptoms (Dawson et al., 1999; Seabrook et al., 1999) that have been associated with impaired copper homeostasis (White et al., 1999; Crouch et al., 2005), lipid metabolism (Grimm et al., 2005) and iron homeostasis (Ayton et al., 2015), indicating that APP plays a critical role not only during CNS mammalian development (reviewed in van der Kant and Goldstein, 2015) but also in the maintenance of cognitive functions with ageing. Furthermore, APP knockout mice show greater cortical neuronal loss and worsened motor and cognitive defects with mild diffuse brain impact injury compared to wild‐type mice (Corrigan et al., 2012a; Ayton et al., 2014). The protective fragment has been identified as sAPPα (Corrigan et al., 2012b). Interestingly, knock‐in of the gene fragment encoding the sAPPα domain is also sufficient to prevent the anatomical and behavioural abnormalities seen in APP knockout mice (Ring et al., 2007).

APP loss has been linked to impaired iron efflux and iron accumulation in neurons (Wong et al., 2014). Lower APP protein levels and higher iron levels were detected in substantia nigra tissue from human patients with Parkinson's disease and in APP knockout mice, while mice overexpressing APP were resistant to 1‐methyl‐4‐phenyl‐1,2,3,6‐tetrahydropyridine (MPTP)‐induced substantia nigra neuronal loss (Ayton et al., 2015). This study used Tg2576 mice which, while exhibiting higher levels of Aβ due to the presence of the Swedish mutation in APP, also have higher levels of sAPPα due to the overexpression of APP. This study provides circumstantial evidence for sAPPα being protective of a mitochondrial toxin‐induced injury, since MPP⁺, the active metabolite of MPTP, is an OXPHOS complex I inhibitor. Similarly, sAPPα has been shown to afford protection post‐hypoxia (Hefter et al., 2016), and since hypoxia/reperfusion injury results in part from mitochondrial impairment, this is also circumstantial evidence for protection from mitochondrial insult.

Mechanisms of sAPPα‐mediated protection – mitochondrial links

Non‐amyloidogenic fragments, particularly sAPPα, have been shown to be neuroprotective in a variety of contexts. sAPPα supports normal neuronal function and survival in vitro and in vivo (Araki et al., 1991; Hayashi et al., 1994; Roch et al., 1994; Meziane et al., 1998) and confers protection from neurotoxicity, including glucose deprivation, glutamate toxicity and Aβ‐induced oxidative injury (Mattson et al., 1993; Schubert and Behl, 1993; Goodman and Mattson, 1994; Barger and Harmon, 1997).

Despite extensive efforts, the receptor for sAPPα has not been defined and insights into the mechanism(s) of protection via sAPPα are limited. The PI3K/Akt survival pathway was identified as one downstream effector of sAPPα in protecting against in vitro excitotoxicity (Cheng et al., 2002). sAPPα was also found to protect from trophic factor deprivation in vitro, via activation of the PI3K/Akt pathway that also depended on holo‐APP, suggesting that full‐length APP acted as a receptor for sAPPα (Jimenez et al., 2011). However, other protective effects of sAPPα have been found to be independent of holo‐APP, such as protection from Bcl‐2‐associated athanogene 3‐linked aggresome formation (Kundu et al., 2016).

The insulin receptor (InsR) is another candidate for sAPPα binding (Figure 2). Wallace et al. (1997) discovered that picomolar levels of sAPPα could potentiate nerve growth factor activity and sustain neurite outgrowth in vitro and that this effect involved both the InsR and PI3K/Akt activation. In primary mouse neuronal cultures, sAPPα has been shown to increase InsR phosphorylation in the absence of insulin and to protect against features of brain pathology in insulin‐depleted diabetes in vivo (Aulston et al., 2018). One common consequence of PI3K/Akt activation is a metabolic switch from OXPHOS to aerobic glycolysis, as classically seen in the activation of T cells (reviewed by Finlay, 2012). We speculate that acute down‐regulation of OXPHOS via sAPPα‐mediated PI3K/Akt activation is a mechanism of protection from OXPHOS‐linked oxidative stress under conditions of neuronal injury. Further work on the interaction of sAPPα with the InsR and downstream effects is warranted.

Mitochondrial function as a biomarker in Alzheimer's disease

Current Alzheimer's disease biomarkers include fluorodeoxyglucose PET (to quantify regional brain glucose utilization), structural magnetic resonance imaging (to assess brain tissue volume), amyloid PET (to detect the presence of Aβ plaque and estimate plaque burden) and CSF Aβ, tau and phosphorylated tau levels, which vary depending on disease progression (Blennow et al., 2010). Recent efforts are now moving towards earlier detection and diagnosis, as it is now accepted that disease progression in Alzheimer's disease takes between 10 and 20 years and preclinical pathology may involve mechanisms that are independent of Aβ (Chetelat, 2013).

In vivo measures of mitochondrial function may be worthy of development so that mitochondrial functional changes can be mapped into the time course of prodromal to clinical Alzheimer's disease, as limited methods are currently available. Radioisotope Cu‐ATSM is one candidate (Okazawa et al., 2014). While known in clinical nuclear medicine as a hypoxic imaging agent, it was found that it accumulates in proportion to altered redox state in cells, where nicotinamide adenine dinucleotide (NAD)⁺ is limited and NADH is in excess, as it occurs when OXPHOS is inhibited (Donnelly et al., 2012). However, it would be desirable to avoid the expense and challenges of PET, and more direct measures of cellular NAD⁺/NADH (Schaefer et al., 2017) or flavin adenine dinucleotide (FAD/FADH₂) may be possible.

Interest has recently turned to retinal imaging as a non‐invasive surrogate measure of mitochondrial function in vivo. The retina, the neural layer of the eye, is a developmental extension of the brain and is the only part of the CNS that can be optically imaged in vivo. The retina manifests features of CNS diseases, and methods to detect retinal biomarkers of Alzheimer's disease are in various stages of development (Hinton et al., 1986; Jentsch et al., 2015; Koronyo et al., 2017; van Wijngaarden et al., 2017; Ko et al., 2018). Several imaging approaches have been postulated to measure NADH and FAD autofluorescence in vivo (Elner et al., 2008; Jentsch et al., 2015). As NADH is fluorescent in its reduced state and FAD is fluorescent in its oxidized form, measurement of the relative intensities of these fluorophores may provide an indication of the metabolic state of the retina (Maleki et al., 2013).

Fluorescence imaging of the eye is confounded by multiple sources of autofluorescence in the cornea, lens, neural retina and retinal pigment epithelium with overlapping excitation and emission spectra (Spaide, 2009). This is further confounded by substantial variation in autofluorescence due to the influence of cataract, variation in macular pigmentation, and age‐ and disease‐related variation in lipofuscin accumulation in the retinal pigment epithelium (Spaide, 2009). The recent development of fluorescence lifetime imaging ophthalmoscopy has the advantage of added temporal resolution due to the measurement of fluorescence decay, allowing greater precision in the attribution of fluorescence to a given source in the retina. Preliminary studies in people with Alzheimer's disease indicate that fluorescence lifetime signals may vary between patients with and without the disease; however, the source of this difference has not been identified (Jentsch et al., 2015). Further developments in metabolic retinal imaging may enable large‐scale prospective studies of Alzheimer's disease due to the ease of obtaining retinal images compared to PET. Such large studies can be expected to differentiate Alzheimer's disease‐specific from age‐related changes in mitochondrial function, allowing identification of prodromal Alzheimer's disease and discriminating age‐related from disease‐specific mitochondrial dysfunction in the pathogenic cascade leading to Alzheimer's disease.

Conclusions

• Our current understanding of Alzheimer's disease is evolving, and the central role of Aβ toxicity as the disease catalyst is increasingly questioned. A greater understanding of the role of loss of protective non‐amyloidogenic APP fragments in the context of Alzheimer's disease, where amyloidogenic APP processing is unquestionably increased, may lead to new insights into effective therapeutic targets.

• We argue for caution in the interpretation of mitochondrial perturbations in models where APP is overexpressed, as it has been shown that not only Aβ but also sAPPα and other non‐amyloidogenic APP fragments are acute modulators of mitochondrial metabolism. APP is subject to complex processing events that generate several peptides, each of which likely performs specific physiological functions. It is critical that studies using models that overexpress APP systematically measure and report the levels of APP, sAPPα, Aβ and other APP‐derived fragments, to elucidate their contribution to cellular functions.

• Although energy deficiency and mitochondrial dysfunction have been recognized as early events in Alzheimer's disease, the link between APP or APP‐derived peptides, mitochondrial metabolism and pathogenesis remains unclear.

• Novel, non‐invasive, surrogate measurements of mitochondrial function in vivo, such as retinal imaging, may be worthy of development so that mitochondrial functional changes can be mapped into the time course of prodromal to clinical Alzheimer's disease.

Nomenclature of targets and ligands

Key protein targets and ligands in this article are hyperlinked to corresponding entries in http://www.guidetopharmacology.org, the common portal for data from the IUPHAR/BPS Guide to PHARMACOLOGY (Harding et al., 2018), and are permanently archived in the Concise Guide to PHARMACOLOGY 2017/18 (Alexander et al., 2017a,b).

Conflict of interest

P.v.W. heads Enlighten Imaging, which aims to develop retinal biomarkers in Alzheimer's disease.

Lopez Sanchez M. I. G., van Wijngaarden P., and Trounce I. A. (2019) Amyloid precursor protein‐mediated mitochondrial regulation and Alzheimer's disease, British Journal of Pharmacology, 176, 3464–3474. doi: 10.1111/bph.14554.