The biogenesis and steady-state function of mitochondria require a highly regulated system of protein import, sorting, assembly, and quality control (Figure 2; Supplementary Table S1). The human mitochondrial proteome consists of approximately 1200–1500 individual proteins (Morgenstern et al., 2021; Rath et al., 2021). Being semi-autonomous organelles, mitochondria have the genome (mitochondrial DNA) and the biosynthetic machinery (e.g., mitochondrial ribosomes, RNA/DNA polymerases and tRNAs) to synthesize a handful of their resident proteins, which in humans includes 13 subunits of OXPHOS complexes I, III, IV and V. All other mitochondrial proteins are encoded in nuclear DNA, synthesized on cytosolic ribosomes, and subsequently imported into mitochondria. Multiple pathways exist for the targeting and sorting of nuclear-encoded proteins to the proper mitochondrial membrane or aqueous subcompartment (Busch et al., 2023) (Figure 2A). These proteins are synthesized with mitochondria targeting information encoded in the polypeptide sequence itself, which must be recognized by the dedicated import complexes that direct them to their correct destinations. The majority (about two-thirds) of mitochondria-targeted proteins are imported via the translocase of the mitochondrial inner membrane 23 (TIM23) pathway, which mediates the translocation of soluble proteins into the matrix as well as the integration of membrane proteins into the MIM (Sinha et al., 2014). In this section, we summarize our current understanding of TIM23-based protein biogenesis and the associated mitochondrial proteostasis machinery.

Cellular proteostasis involves the regulation of all stages of the protein life cycle (Labbadia and Morimoto, 2015; Klaips et al., 2018): molecular chaperones guide cotranslational folding of nascent chains during ribosomal synthesis and promote protein folding and assembly of oligomeric complexes, degradation mechanisms such as the ubiquitin-proteasome system (UPS) and the autophagy-lysosome pathway (ALP) remove misfolded and aggregated proteins, and stress response pathways respond to protein folding stress, such as the ER-based unfolded protein response (UPR^ER) (Figure 4, black arrows). All three of these protein quality control processes are implicated in neurodegenerative diseases (Scheper and Hoozemans, 2015; Park et al., 2020; Schmidt et al., 2021).

AD, PD, and HD are proteostatic diseases associated with the misfolding of Aβ, α-syn, and mHtt, respectively, in a process that leads to the formation of amyloid fibrils (Figure 4, grey arrows). This process can begin with precursors that include intrinsically disordered peptides such as Aβ and α-syn, or with partially and misfolded globular proteins. The kinetics of amyloid fibrillization have a marked dependence on protein concentration, as well as factors that affect the tendency for polypeptides to self-associate, including mutations, post-translational modifications (PTMs), cofactors, and oxidative stress (Alexandrescu, 2005; Rezaei-Ghaleh et al., 2016; Hu et al., 2019; Lontay et al., 2020; Chiki et al., 2021; Gottlieb et al., 2021; He et al., 2021; McGlinchey et al., 2021). Fibrillization follows nucleation kinetics, accounting for the “seeding” properties of amyloids (Chiti and Dobson, 2017; Almeida and Brito, 2020). Nucleation can occur through primary or de novo processes, or secondary mechanisms whereby mature fibril surfaces act as templates for new fibril growth (Ferrone et al., 1985; Padrick and Miranker, 2002; Patil et al., 2011; Tornquist et al., 2018). Amyloid fibrils share a common structural property of protein association through a cross-β spine motif, stabilized by intermolecular hydrogen-bonded parallel β-sheet layers arranged perpendicular to the long axis of the fibril (Nelson et al., 2005; Chiti and Dobson, 2017; Almeida and Brito, 2020). They also share similar morphologies, often being microns long with widths of ∼10 nm and a twisting repeat of ∼100 nm (Dobson, 2003; Qiang et al., 2017).

Amyloidogenic aggregates have historically provided the primary histopathological markers of AD, PD, and HD. These aggregates are generally termed “amyloid fibrils” when formed extracellularly and ‘inclusions’ when formed intracellularly. The hallmarks of AD are extracellular plaques composed primarily of Aβ peptides and intracellular neurofibrillary tangles enriched in hyperphosphorylated variants of the protein tau (Zheng and Koo, 2011). PD is characterized by cytoplasmic aggregates called Lewy bodies and inclusions called Lewy neurites, of which α-syn is the main component (Braak et al., 1999). Intracellular accumulation of mHtt into amyloid-like inclusion bodies is a primary feature of HD (Yamamoto et al., 2000; Arrasate et al., 2004).

The concept that insoluble fibrils, whether extra- or intra-cellular, are the primary cytotoxic factors that initiate amyloidogenic disease is the core tenet of the “amyloid cascade hypothesis” (Figure 5, left) first proposed by Hardy and Higgins to describe the role of Aβ amyloids in AD pathogenesis (Hardy and Higgins, 1992). Subsequently it was proposed that α-syn fibrils are the primary cytotoxic factors in PD (Stefanis, 2012) and that amyloid fibril-like inclusions of mHtt are the primary cytotoxic factors in HD (Scherzinger et al., 1999). There is considerable controversy, however, on whether amyloid fibrils are the only, or even the principal culprits in pathology. A confounding factor is the heterogeneity of species associated with amyloid formation. Amyloidogenesis involves a hierarchy of structures that starts from the functional, soluble form of a protein or peptide and proceeds to oligomers, nuclei, β-sheets, protofilaments, protofilament bundles, and finally mature fibrils (Serpell, 2000) (Figure 4). Within the fibrils themselves, there are structural polymorphs that can differ depending on whether they are grown in vitro or isolated from patients. In fact, increasing evidence suggests distinct structural polymorphs are associated with specific disease subtypes. Therefore, considerable uncertainty surrounds which species represent the culprits in pathogenicity (Carulla et al., 2005; Lansbury and Lashuel, 2006; Haass and Selkoe, 2007; Zraika et al., 2010), with some recent proposals advancing that soluble oligomers (Larson and Lesné, 2012; Arbor et al., 2016; Cline et al., 2018) or even individual proteins/peptides (Hoffner and Djian, 2014; Hillen, 2019) could initiate the phenotypic cascade of amyloidogenic diseases. Because mature fibrils are insoluble and extremely stable, many view soluble oligomeric precursors as the most likely candidates for cytotoxicity. However, studies on amyloid oligomers have been hampered by the low concentrations and transitory nature of these intermediates (Engel, 2009), thereby contributing to a lack of evidence for this model. Finally, it unclear whether cytotoxicity is associated with intracellular or extracellular forms of amyloidogenic proteins (Cuello, 2005; Gurlo et al., 2010; Aston-Mourney et al., 2011).

Much of the current research on the mechanisms of amyloidogenic diseases has shifted focus towards processes preceding mature fibril formation. As a result, it has now been widely demonstrated that mitochondrial dysfunction occurs early in the pathogenesis of AD (Wang et al., 2020b), PD (Malpartida et al., 2021), and HD (Carmo et al., 2018). The “mitochondrial cascade hypothesis”, originally proposed by Swerdlow and colleagues, proposes that it is in fact the progressive decline in mitochondrial function that promotes AD pathology and influences the progression of the disease (Swerdlow et al., 2014) (Figure 5, right). Therefore, exactly how amyloidogenic proteins interact with mitochondria, and specifically how they may engage the mitochondrial protein import machinery, is a key mechanistic question in understanding the pathology of neurodegenerative disorders.

AD is the leading cause of senile dementia, affecting several regions of the brain involved in memory and cognition, including the hippocampus, the neocortex, and the basal forebrain (Auld et al., 2002). While the exact etiology of AD is unknown, it is biologically characterized by the aggregation of two misfolded proteins: Aβ, the primary component of extracellular plaques, and hyperphosphorylated variants of the microtubule-associated protein tau, which form intracellular inclusions known as neurofibrillary tangles (NFTs) (Knopman et al., 2021). Aβ is produced through sequential cleavage of APP, a membrane glycoprotein with isoforms ranging from 100 to 140 kDa that predominantly reside at neuronal synapses (Zhou et al., 2011; Hoe et al., 2012) (Figure 6A). APP plays a vital role in neural development and synaptic plasticity, as it is implicated as a receptor involved in kinesin 1 cargo recognition (Lazarov et al., 2005), the Wnt signaling pathway (Liu T. et al., 2021) and other functions including cell adhesion, synaptogenesis (Baumkötter et al., 2014), and iron export (Duce et al., 2010). As a type I integral membrane protein, APP has a large ectodomain at its N-terminus, a single α-helical transmembrane segment, and a small C-terminal intracellular domain. APP belongs to a highly conserved superfamily of genes (Coulson et al., 2000; Jacobsen and Iverfeldt, 2009). In mammals, alternative splicing generates eight APP isoforms, the three most common being APP₆₉₅, APP₇₅₁, and APP₇₇₀, among which APP₆₉₅ is most highly expressed in neurons (Sandbrink et al., 1996; Belyaev et al., 2010). Like other plasma membrane proteins, the life cycle of APP following synthesis consists of membrane trafficking via the secretory pathway and degradation and recycling by the endocytic system (Lin et al., 2021). The Aβ peptides that form amyloid plaques in AD are derived from the processing of APP through sequential actions of β- and γ-secretases (Selkoe and Hardy, 2016; Zhao et al., 2020) (Figure 6B). The generated Aβ peptides range in size, with the 42-residue Aβ₄₂ being much more aggregation-prone than the more abundant 40-residue Aβ₄₀ (Seubert et al., 1992; Haass and Selkoe, 2007; Chow et al., 2010). Monomeric Aβ₄₀ and Aβ₄₂ are intrinsically disordered in solution (Roche et al., 2016), with Aβ₄₂ having a transient population of β-sheet structure (Kakeshpour et al., 2021).

The neuropathology of AD has historically been described by the amyloid cascade hypothesis, whereby Aβ deposition is the causative event (Hardy and Higgins, 1992). In support of this hypothesis, early-onset familial AD (FAD) is caused by autosomal dominant inheritance mutations in the genes encoding APP or APP processing enzymes (Bekris et al., 2010; Ricciarelli and Fedele, 2017). However, the vast majority of cases are sporadic, late onset AD (LOAD) occurring without mutations in the genes encoding APP or APP processing enzymes (Bekris et al., 2010). While the correlation between Aβ and AD remains strongly supported (Herrup, 2015), intracellular Aβ in the form of soluble monomers and oligomers are now considered to be the primary toxic species rather than insoluble, fibrillar assemblies of Aβ that form extracellular plaques (Goure et al., 2014; Gallego Villarejo et al., 2022). Despite this, soluble Aβ species precede the formation of plaques, which are found in patient brains prior to the onset of clinical symptoms, and it has been shown that Aβ alone is not sufficient to induce disease pathogenesis (Herrup, 2015). This suggests a more complex pathological mechanism at play that led to the development of a variety of alternative hypotheses surrounding AD etiology.

One prevailing proposal is the ApoE cascade hypothesis (Martens et al., 2022), based on the genetic association of LOAD with the apolipoprotein ε4 allele (APOE4) (Corder et al., 1993). APOE4 carriers have a vastly increased risk of developing LOAD and a reduced age of onset compared to carriers of the APOE3 and APOE2 isoforms (Corder et al., 1993; Sando et al., 2008). ApoE is a 34 kDa lipid-binding protein that primarily functions in lipid transport and metabolism (Mahley, 2016), with differences among the ApoE2, E3, and E4 isoforms confined to two sites (residues 112 and 158) (Najm et al., 2019; Martens et al., 2022). The primary functional difference in ApoE4 is its binding preference for very-low density lipoprotein (VLDL) over high-density lipoprotein (HDL), whereas the ApoE2 and ApoE3 isoforms preferentially bind HDL (Raffai et al., 2001; Belloy et al., 2019). While ApoE contributes to Aβ synthesis, accumulation, and clearance in an isoform-dependent manner, there is conflicting evidence on the isoform-specific roles of ApoE in Aβ pathology (Koistinaho et al., 2004; Castellano et al., 2011; Huang and Mahley, 2014; Najm et al., 2019). The ApoE cascade hypothesis does not include Aβ as a contributing factor in disease pathogenesis, stating that the biophysical and structural properties dependent of the ApoE isoform initiate a cascade of events driving AD and aging-related pathogenic condition (Martens et al., 2022). ApoE4 has been shown to alter lipid homeostasis and metabolism, leading to changes in lipid droplet formation, cholesterol turnover, and the PC/PE ratio (Farmer et al., 2019; Lazar et al., 2022; Yang et al., 2023). Consistent with MAM alterations, several lines of work support the involvement of ApoE in mitochondrial dysfunction as a contributing component to disease pathogenesis, including recent findings suggesting mitochondrial dysfunction influences ApoE expression and secretion (Dose et al., 2016; Martens et al., 2022; Wynne et al., 2023).

AD is associated with pronounced defects in mitochondrial function (Cenini and Voos, 2019; Wang et al., 2020c). These include alterations in mitochondrial biogenesis and morphology, and decreases in mitochondrial number (Hirai et al., 2001; Qin et al., 2009; Oka et al., 2016; Brustovetsky et al., 2023); compromised energy metabolism, including glucose hypometabolism (Kumar et al., 2022) and deterioration of the TCA and OXPHOS systems (particularly CIV) (Bubber et al., 2005; Liang et al., 2008; Zhang et al., 2015; Mastroeni et al., 2017; Sorrentino et al., 2017; Adav et al., 2019; Ryu et al., 2021); oxidative stress (Misrani et al., 2021); effects on mitochondrial dynamics, most notably a general increase in fragmentation (Wang et al., 2008a; Wang et al., 2009); impairment of mitochondrial trafficking (Calkins et al., 2011); and altered ER-mitochondria apposition with associated cellular Ca²⁺ dyshomeostasis (Leal et al., 2020; Fernandes et al., 2021; Li et al., 2023). While these effects could be associated with the interaction of APP/Aβ with mitochondria (see below), other AD-related factors could also be involved.

PD involves the selective degeneration of nigrostriatal dopaminergic neurons, manifesting as progressive effects on movement that include rigidity, resting tremor, and bradykinesia (Zhai et al., 2018). The main histological features of PD are cytoplasmic inclusions called Lewy bodies (LBs) and neuritic inclusions called Lewy neurites (LNs) whose core component is α-syn (Baba et al., 1998; Spillantini et al., 1998) but also contain vesicular structures, membranes and fragmented organelles (Shahmoradian et al., 2019; Ericsson et al., 2021). These α-syn aggregates are characteristic of a general group of neurodegenerative disorders termed synucleinopathies. α-syn is a 140 residue protein encoded by the synuclein alpha (SNCA) gene that contains three domains: (i) an N-terminal amphipathic region, (ii) a highly hydrophobic region called the non-amyloid-β component (NAβC), and (iii) a C-terminal acidic region (Bisi et al., 2021) (Figure 7A). Being enriched in the axon terminals of presynaptic neurons, α-syn is a membrane-interactive protein that plays a role in synaptic function, including vesicle trafficking and exocytosis by assembly of SNARE complexes, synaptic membrane remodeling, and maintenance of neurotransmitter vesicle pools (Bendor et al., 2013; Burre, 2015). α-syn also localizes to the nucleus, where it regulates gene expression (Somayaji et al., 2021).

Monomeric α-syn is intrinsically disordered in solution, dynamically interconverting among many different conformations that differ with respect to short- and long-range electrostatic interactions and transient secondary structures (Weinreb et al., 1996). The N-terminus of α-syn in solution has helical propensity with a low fractional population assuming α-helix conformation; however, when bound to membranes, it adopts a stable helical structure that is either a continuous helix or two separate helices (Davidson et al., 1998; Bussell and Eliezer, 2003; Chandra et al., 2003; Ulmer et al., 2005; Croke et al., 2008; Georgieva et al., 2008; Jao et al., 2008; Stockl et al., 2008; Cho et al., 2009; Bodner et al., 2010; Rao et al., 2010; Croke et al., 2011; Ullman et al., 2011; Adao et al., 2020; Musteikyte et al., 2021). Membrane binding is facilitated by seven imperfect repeats of the 11-residue pattern xKTKEGVxxxx that extend from the amphipathic domain into the NAβC domain, a motif also found in apolipoproteins A2 (Bussell and Eliezer, 2003). This membrane-interactive region imparts classic amphitropic behavior to α-syn, which consequently promotes its interaction with curved bilayers, particularly with those enriched in negatively charged phospholipids (Jo et al., 2002) and cholesterol (Fantini et al., 2011). This membrane-binding activity is critical for its function in synaptic vesicle interaction (Fortin et al., 2005; Sarchione et al., 2021). The hydrophobic NAβC domain nucleates α-syn aggregation, forming the canonical β-structure of amyloid fibrils along with parts of the N-terminal domain (Rodriguez et al., 2015). The fraction of α-syn that natively exists in neurons in a membrane-bound state, as an unfolded monomer (Fauvet et al., 2012; Theillet et al., 2016), or as aggregation-resistant folded tetramers (Bartels et al., 2011; Wang et al., 2011) remains an open question.

Familial autosomal dominant PD is linked to defects in several genes, most of which have a connection to mitochondrial physiology (Cieri et al., 2017; Billingsley et al., 2018). Among them, the six known heritable mutations in the SNCA gene (A30P, E46K, H50Q, G51D, and A53E/T) all localize to the amphipathic N-terminal region of α-syn and promote aggregation (Flagmeier et al., 2016). α-syn aggregation can also be potentiated by increases in its copy number (Srinivasan et al., 2021). However, PD is primarily an idiopathic disease in which aging-related increases in α-syn aggregation and/or certain post-translational modifications of α-syn are commonly involved (Anderson et al., 2006; Manzanza et al., 2021; Roshanbin et al., 2021). The balance between α-syn function and pathogenicity is depicted in Figure 7B.

Mitochondrial dysfunction plays a key role in the pathogenesis of PD (Vicario et al., 2018). The enhanced expression of α-syn or particular mutants of α-syn known to promote PD cause the disruption of mitochondrial dynamics (Xie and Chung, 2012; Thorne and Tumbarello, 2022) and promote fragmentation (Kamp et al., 2010; Nakamura et al., 2011; Gui et al., 2012; Pozo Devoto et al., 2017), impair mitochondrial trafficking (Pozo Devoto et al., 2017) and autophagic clearance (Gao et al., 2017), impact mitochondrial energetics (Banerjee et al., 2010), and alter Ca²⁺ flux by the enhancement (Cali et al., 2012) or disruption (Guardia-Laguarta et al., 2014; Guardia-Laguarta et al., 2015; Paillusson et al., 2017) of specific ER-mitochondria tethering sites. One of the most prominent effects of PD is the impairment of respiratory Complex I associated with reduced energetic output and an overproduction of ROS (Devi et al., 2008; Marella et al., 2009). This multitude of effects is in line with the many mitochondrial subcompartments α-syn has been found to target, including the MOM (Li et al., 2007; Cole et al., 2008; McFarland et al., 2008; Nakamura et al., 2008; Zhang et al., 2008; Kamp et al., 2010; Nakamura et al., 2011; Di Maio et al., 2016a; Pozo Devoto et al., 2017), the MIM/IMS (McFarland et al., 2008; Nakamura et al., 2011; Zhu et al., 2011; Lu et al., 2013; Robotta et al., 2014; Amorim et al., 2017), the matrix (McFarland et al., 2008; Ludtmann et al., 2016; Ludtmann et al., 2018) and the MAM (Cali et al., 2012; Guardia-Laguarta et al., 2014; Paillusson et al., 2017). Importantly, this extensive interaction of α-syn with mitochondria is mediated in part through lipid bilayer interactions, particularly with regions enriched with the mitochondrial lipid cardiolipin (Ramakrishnan et al., 2003; Cole et al., 2008; Nakamura et al., 2008; Grey et al., 2011; Zigoneanu et al., 2012; Robotta et al., 2014; Ryan et al., 2018) that provides the negative surface, high curvature, and acyl packing defects necessary to promote α-syn binding (Sharon et al., 2003; Nuscher et al., 2004; Middleton and Rhoades, 2010; Pfefferkorn et al., 2012; Ouberai et al., 2013; Gilmozzi et al., 2020).

HD is associated with neurodegeneration of the basal ganglion, with preferential deterioration of striatal medium spiny GABAergic neurons, that clinically presents as progressive loss of motor control and cognition (Morigaki and Goto, 2017). This monogenic disease is caused by heritable alterations in Htt (Jurcau and Jurcau, 2022). Structurally, Htt is a 348 kDa protein that contains two domains of HEAT tandem repeats that form α-solenoid structures connected by a bridge domain, and an N-terminal region with a tripartite organization that contains a highly conserved sequence of 17 N-terminal amino acids (Nt17), a stretch of glutamine residues (poly-Q), and a proline-rich domain (PRD) (Saudou and Humbert, 2016; Guo et al., 2018) (Figure 8A). Htt is conformationally dynamic, with an extensive list of interaction partners likely owing to the protein-interaction functions of its HEAT repeats (Shirasaki et al., 2012). As such, it serves as a multivalent molecular scaffold, a function that likely supports its many interactions in synapses including axonal transport, vesicle recycling, autophagy, transcriptional regulation, and endocytosis (Barron et al., 2021).

HD is an autosomal dominant disorder caused by an extension of the poly-Q tracts resulting from polymorphic CAG trinucleotide repeat expansion in exon 1 of the HTT gene (MacDonald et al., 1993). HD is part of a larger family of neurodegenerative diseases associated with polyglutamine expansion of proteins (Lieberman et al., 2019). Whereas Htt in healthy individuals contains fewer than 36 CAG repeats, the pathogenically expanded poly-Q region can contain between 42 and 250 CAG repeats, with disease severity and earlier age of onset directly related to the extent of expansion (Rubinsztein et al., 1996; Penney et al., 1997). Because Htt is ubiquitously expressed, HD affects not only the brain but also peripheral tissues with high metabolic activity (Chuang and Demontis, 2021). Poly-Q expansion causes the resulting mHtt protein to have altered folding (Vijayvargia et al., 2016) and interactions (Ratovitski et al., 2012; Greco et al., 2022). This makes mHtt prone to aggregation, forming inclusion bodies in the cytoplasm and nucleus (Davies et al., 1997) with diverse aggregate structures (Tanaka et al., 2001; Poirier et al., 2002) and a complex composition that includes lipids, proteins, and membrane-bound organelles (Kegel-Gleason, 2013; Riguet et al., 2021). Notably, the primary cytotoxic, aggregation-prone species in HD may not be full-length mHtt, but rather heterogeneous populations of N-terminal fragments of mHtt exon 1 (Httex1) (Mangiarini et al., 1996; Cooper et al., 1998; Hackam et al., 1998; Martindale et al., 1998; Hoffner et al., 2005; Barbaro et al., 2015) that can arise by cleavage of the mature protein by caspase, calpain, or other proteases (Goldberg et al., 1996; Wellington et al., 1998; Kim et al., 2001; Gafni and Ellerby, 2002; Lunkes et al., 2002; Graham et al., 2006; Schilling et al., 2007; Landles et al., 2010; Tebbenkamp et al., 2012; Martin et al., 2019), or by aberrant splicing and premature polyadenylation of pathogenic HTT exon 1 with expanded CAG repeats (HTTex1 transcripts) (Sathasivam et al., 2013; Neueder et al., 2017; Neueder et al., 2018) (Figure 8B).

Mitochondrial dysfunction plays a central role in HD pathogenesis (Bossy-Wetzel et al., 2008; Carmo et al., 2018; Jurcau and Jurcau, 2023). HD is associated with reduced mitochondrial biogenesis and quality control (Steffan et al., 2000; Cui et al., 2006; Weydt et al., 2006; Jin and Johnson, 2010; Wong and Holzbaur, 2014; Khalil et al., 2015; Guo et al., 2016; Dubois et al., 2021; Sonsky et al., 2021), mtDNA heteroplasmy (Wang et al., 2021), defective energy metabolism and OXPHOS activity (Sawa et al., 1999; Klivenyi et al., 2004; Seong et al., 2005; Benchoua et al., 2006; Pandey et al., 2008; Mochel et al., 2012; Silva et al., 2013; Naia et al., 2015; Burtscher et al., 2020), altered mitochondrial Ca²⁺ handling and sensitivity to the mitochondrial permeability transition pore (mPTP) (Panov et al., 2002; Choo et al., 2004; Milakovic et al., 2006; Giacomello et al., 2013), increased mitochondrial oxidative stress (Sorolla et al., 2008; Chae et al., 2012; Ribeiro et al., 2012; Ribeiro et al., 2014; Chen et al., 2017; Moretti et al., 2021; Lopes et al., 2022; Egidio et al., 2023), defective mitochondrial dynamics and hyper-fission (Costa et al., 2010; Song et al., 2011; Shirendeb et al., 2012; Manczak and Reddy, 2015; Cherubini et al., 2020), defective mitochondrial trafficking (Trushina et al., 2004; Chang et al., 2006; Orr et al., 2008; Shirendeb et al., 2012; Berth and Lloyd, 2023), and altered cytochrome c release with apoptosis (Chen et al., 2000; Kiechle et al., 2002; Zhang et al., 2006; Wang et al., 2008b). Although many of these effects are related to spurious interactions of mHtt with cytosolic proteins, there are several reports that full-length and N-terminal fragments of Htt/mHtt interact directly with mitochondria (Panov et al., 2002; Choo et al., 2004; Petrasch-Parwez et al., 2007; Orr et al., 2008; Song et al., 2011; Guo et al., 2016). Some of these studies specifically implicate interactions with the MOM (Hamilton et al., 2020) and some show interactions with internal mitochondrial compartments (Yano et al., 2014; Yablonska et al., 2019). In general, accumulation of N-terminal mHtt fragments with mitochondria increases with age (Orr et al., 2008). Potential interactions between Htt/mHtt and the mitochondrial import machinery are discussed below.

Most proteins that are targeted to cellular compartments other than their site of ribosomal synthesis contain unambiguous targeting sequences that faithfully direct them to a specific location. For example, as described above, the MTSs of TIM23 substrates form an N-terminal amphipathic α-helix that serves as the primary recognition element of the TIM23 import machinery. Three examples of MTSs shown in Figure 9A illustrate their amphipathic character: they have an appreciable number of nonpolar residues quantified as the mean hydrophobicity (<H>) (Fauchere and Pliska, 1983), a net charge (z) that reflects a preponderance of basic residues, and an asymmetric distribution of nonpolar and basic residues quantified by the hydrophobic moment (<μH>) (Eisenberg et al., 1982). Yet it is also notable that MTSs vary significantly with respect to sequence, length, and extent of amphipathicity. Thus, while canonical MTSs encompass a range of physicochemical properties, they are sufficiently well-defined enough to be recognized by a single import pathway.

For some proteins, however, the relationship between the targeting sequence and the organellar import pathway is not straightforward and some targeting sequences can direct passenger proteins to multiple cellular compartments by non-mutually exclusive mechanisms (Karniely and Pines, 2005). Indeed, targeting fidelity requires multiple levels of regulation and precise recognition because mitochondrial MTSs, ER signal peptides, chloroplast transit peptides, and even some peroxisomal targeting signals share similarity in that they all form N-terminal amphipathic α-helices with a hydrophobic face and a basic (or polar) face (Kunze and Berger, 2015).

Some proteins contain an ambiguous (or cryptic) targeting sequence that can be recognized by multiple import machineries (Lu et al., 2004; Ventura et al., 2004; Chatre et al., 2009). Other proteins contain well-defined targeting signals whose accessibility or targeting efficiency can be modified by folding, protein binding, or post-translational modifications (Karniely and Pines, 2005). Still other proteins contain multiple targeting signals that direct them to different compartments (Petrova et al., 2004; Pino et al., 2007). For example, some cytochromes P450 (CYPs) expressed in hepatocytes contain chimeric signals with amino-terminal ER-targeting sequences in tandem with a mitochondrial-targeting sequence (Addya et al., 1997; Anandatheerthavarada et al., 1999; Robin et al., 2001; Robin et al., 2002). The bipartite targeting sequence of these cytochromes contain a cryptic mitochondria localization signal that remains idle until activation. In the case of CYP2E1, the mitochondrial targeting signal is activated by cyclic-AMP-dependent phosphorylation of a Ser residue in the cryptic sequence (Robin et al., 2002). Moreover, an inducible endoprotease has been identified that activates mitochondrial import by cleaving bimodal targeting signals to expose the cryptic mitochondrial targeting signal (Boopathi et al., 2008). It is unclear whether such cleavage of chimeric signals is a general requirement for mitochondrial targeting.

By allowing greater diversity of potential cellular locations, noncanonical targeting imparts flexibility to the functional range of individual proteins that can be subject to regulation by the cell in response to different physiological demands. However, cryptic targeting sequences can also cause the pathogenic accumulation of proteins in organelles. In the following sections, we review the cryptic mitochondria targeting sequences present in Aβ/APP, α-syn, and Htt/mHtt that may facilitate their interactions with the mitochondrial import machinery.

Several lines of evidence indicate that full-length APP engages the TIM23 import pathway. The canonical biogenesis route of APP entails targeting to the ER via the SRP/Sec61 pathway, translocation across the ER membrane attendant with glycosylation of the ectodomain, integration as a type I membrane protein, and trafficking by the secretory, endocytic and recycling routes (Muresan and Ladescu Muresan, 2015). APP contains a classic N-terminal Sec-targeting signal with a predicted Signal Peptidase cleavage site (AxA) between positions 17 and 18 (Auclair et al., 2012; Almagro Armenteros et al., 2019). Following this ER-targeting signal is a potential cryptic MTS spanning residues 37 to 67, segments of which could form an amphipathic helix (Figure 9Bi). Thus, together they form a chimeric ER-mitochondria targeting signal (Devi and Anandatheerthavarada, 2010). Indeed, it has been shown using human cortical neuronal cells that a non-glycosylated form of APP₆₉₅ can stall in the mitochondrial import pathway with a predicted N^mito/C^cyto topology. This stalled intermediate made crosslinking-detected interactions with Tom40, Tim23, and Tim44, in a manner that required the Δψ_m and the positive residues Arg40, His44, and Lys51, all three of which reside on a common face of the putative MTS (Anandatheerthavarada et al., 2003) (Figure 9Bi). Similar translocation intermediates of APP were observed in mitochondria of transgenic AD mouse models (Anandatheerthavarada et al., 2003) and in postmortem samples of AD brains (Devi et al., 2006), the latter showing stable association of APP with ∼480 kDa complexes containing the TOM machinery and ∼620 kDa complexes containing the TIM23 machinery. The stalling of APP in the import machinery appeared to be due to the tight folding of the APP acidic domain (residues 220–290) blocking transport along the TOM complex, as deletion of this domain facilitated complete translocation (Anandatheerthavarada et al., 2003). This observation is consistent with the formation of stalled translocation intermediates observed with native mitochondria-targeted substrates with tightly folded (cofactor-bound) C-terminal domains and fusion constructs with MTSs fused to cofactor-stabilized domains like DHFR (Schneider, 2018). In agreement with the known effects of AD on mitochondria function, APP translocation intermediates were shown to interfere with the import of native mitochondrial proteins (Devi et al., 2006). The extent to which mitochondrial targeting of APP may be facilitated by endoprotease cleavage of the ER targeting signal to expose the cryptic MTS (Boopathi et al., 2008) is an open question. Figure 10A (left) summarizes the interactions of APP with mitochondria.

The Aβ peptide has also been widely reported to accumulate in mitochondria through the import machinery. The presence of Aβ within mitochondria is well documented by its direct interactions with many mitochondrial proteins, including MOM proteins VDAC (Manczak and Reddy, 2012) and Drp1 (Manczak et al., 2011), matrix proteins Hsp60 (Caspersen et al., 2005), Aβ-binding alcohol dehydrogenase (ABAD) (Lustbader et al., 2004) and cyclophilin D (Du et al., 2008), and OXPHOS complexes including Complex IV (Hernandez-Zimbron et al., 2012) and ATP synthase (Beck SJ. et al., 2016). Furthermore, a pronounced feature of AD pathogenesis is the mitochondrial accumulation of Aβ (Caspersen et al., 2005; Manczak et al., 2006; Cha et al., 2012). Aβ₄₂ has been shown to enter mitochondria through the TOM complex, associating with the MIM fraction in a Δψ_m-independent manner and localizing to cristae in both in vitro import systems and in mitochondria from human brain biopsies (Hansson Petersen et al., 2008). The specific interactions of Aβ with the TOM complex were explored using a yeast mitochondria model system, indicating that Aβ binds directly to the Tom22 receptor, but not Tom20 or Tom70, and that Aβ residues 25–42 were indispensable for this interaction (Hu et al., 2018). The accumulation of Aβ within the import machinery is functionally relevant because it inhibits the import of native nuclear-encoded proteins (Sirk et al., 2007). However, it should also be noted that cytosolic Aβ may inhibit the import of mitochondrial proteins by coaggregation in the cytosol that does not involve blocking the mitochondrial import machinery (Cenini et al., 2016). A common theme in these studies is that the more aggregation-prone Aβ₄₂ has a stronger interaction with the import machinery than the Aβ₄₀ variant (Sirk et al., 2007; Cenini et al., 2016; Hu et al., 2018). Figure 10A (right) summarizes the interactions of Aβ with mitochondria.

As mentioned, ApoE4, the primary genetic risk factor for AD, induces increased connectivity between MAMs and mitochondria in different AD models (Area-Gomez et al., 2012; Tambini et al., 2016). MAMs have lipid raft-like membranes, enriched in sphingolipids and cholesterol, which promote the interaction between APP and its processing enzymes (β-secretase and γ-secretase) that have been found to localize at the MAM. The biophysical properties of these microdomains may enhance amyloidogenic processing of APP (Diaz et al., 2015; Del Prete et al., 2017; Area-Gomez et al., 2018). Additionally, the protein interactome of MAMs in cells with mutant APP is enriched in mitochondrial import components Tom22, Tim17b, and Sam50 (Del Prete et al., 2017). Enhanced proximity of APP to the mitochondrial import machinery could contribute to spurious import of APP.

There are some potentially confounding factors regarding the import of the Aβ peptide through the mitochondrial import machinery. The first is that Aβ produced from APP proteolysis is released into exocytoplasmic compartments (ER lumen or the extracellular space), whereas engagement with the mitochondrial import machinery occurs from the cytosol. However, there is evidence that proteolytically produced or externally added Aβ can be taken up by cells via clathrin-dependent or -independent endocytosis, followed by endosomal escape, which could allow it to localize to mitochondria (Saavedra et al., 2007; Hansson Petersen et al., 2008; Berridge, 2010; Friedrich et al., 2010). Furthermore, Aβ produced in the vicinity of mitochondria by MAM-localized secretases could directly access mitochondria (Area-Gomez et al., 2009). The second is that the presence of Aβ in the matrix would necessitate TIM23-mediated translocation; however, to date there is no evidence of a direct interaction between Aβ and the TIM23 complex. One explanation could be that the association between the short Aβ peptide and the TIM23 receptors and channel is too transient to be captured by techniques like crosslinking or immunoprecipitation. Lastly, the Aβ sequence does not contain any segments with strong amphipathicity (opposing basic and nonpolar faces) that are present in canonical MTSs (Figure 9Bii). This feature may hinder transport along the “acid chain” of the TIM23 pathway, possibly explaining how Aβ could stall nonproductively at the TOM complex.

It is noteworthy that AD is associated with a general disruption of the mitochondrial import machinery. Analysis of RNA-seq datasets from brains of AD patients versus age-matched healthy individuals showed that expression of mitochondrial import genes was decreased with AD (Sorrentino et al., 2017), consistent with the observed decrease in expression of Tom20 and Tom70 in postmortem neocortex samples of AD patients that correlate with higher Aβ₄₂/Aβ₄₀ ratios (Chai et al., 2018). It has also been proposed that the length of a poly-T polymorphism in the TOMM40 gene correlates with LOAD, supporting the involvement of the TOM complex channel in AD pathogenesis (Roses et al., 2010); however, this finding has been questioned (Chiba-Falek et al., 2018). It is possible that altered expression of TOM complex subunits could modulate mitochondrial interactions of APP/Aβ; for instance, decreased Tom20/70 expression relative to Tom22 could promote Aβ binding to the TOM complex (Hu et al., 2018).

The most compelling evidence that the mitochondrial import and processing machinery serves as a mechanism for amyloid clearance comes from the interaction between Aβ and PreP, the matrix-localized protease that digests MTSs (Pinho et al., 2014). This degradation step is critical because MTSs, being amphipathic and membrane-interactive, can disrupt MIM integrity and their buildup can lead to Δψ_m collapse and OXPHOS uncoupling (Teixeira and Glaser, 2013). Importantly, PreP was shown to degrade Aβ variants in vitro (Falkevall et al., 2006) and PreP activity is lower in the temporal lobes of AD patients and transgenic AD mice compared with controls, which may be further attributed to disease-related oxidative modifications (Alikhani et al., 2011b). Given the functional coupling between precursor processing and MTS turnover (Kucukkose et al., 2021), the accumulation of Aβ could overcome the MTS-clearing capacity of PreP and cause feedback inhibition of MIP and MPP, thereby inhibiting import by the toxic buildup of precursors and MTSs in the matrix (Mossmann et al., 2014). Conversely, overexpression of PreP in AD mouse models decreases mitochondrial Aβ concentrations and improves organellar function (Fang et al., 2015; Brunetti et al., 2016; Du et al., 2021). Given the key role of PreP in proteostasis (Brunetti et al., 2021), mitochondrial import of Aβ and its subsequent degradation may be essential for balancing cellular Aβ concentrations.

Several studies support the entry of α-syn into mitochondria through the import complexes. α-syn was first shown to accumulate in mitochondria in a manner that required the Δψ_m across the MIM and an accessible Tom40 channel (Devi et al., 2008). Furthermore, the possibility that the N-terminus of α-syn targets mitochondria is supported by the indispensable role of residues 1–32 in mitochondrial uptake (Devi et al., 2008) and of residues 1–11 in association of α-syn with isolated mitochondria (Robotta et al., 2014). When folded as an α-helix (stabilized by membrane interactions), the N-terminal 32 residues of α-syn form an amphipathic structure (Figure 9Biii), made up in part by the first two imperfect KTKEGV repeats that produce a Lys-rich basic face and a Val-rich nonpolar face. The N-terminus of α-syn therefore has strong potential to act as a cryptic MTS (Devi and Anandatheerthavarada, 2010).

TIM23-mediated uptake of α-syn is also supported by its physical and genetic interactions with subunits of the import machinery. Based on proteomics analysis of the interactome of the α-syn C-terminal peptide and its variants phosphorylated at Tyr125 and Ser129 that are known to promote aggregation (Samuel et al., 2016; Fayyad et al., 2020), the unmodified and phosphorylated peptides were found to preferentially interact with mitochondrial and cytosolic proteins, respectively (McFarland et al., 2008). In this study, the unmodified α-syn C-terminal peptide bound Tom40, Sam50, and Tom22, with the interaction to MOM proteins Tom40 and Sam50 strongly reduced by phosphorylation, suggesting the preferential import of unmodified α-syn (McFarland et al., 2008). This connection between α-syn aggregation propensity and mitochondrial import is further supported by the fact that all known PD-related missense mutations reside in the N-terminal region (Flagmeier et al., 2016) near the cryptic targeting sequence (Figure 7A). Perhaps more compelling are the specific ways in which N-terminal mutants of α-syn feature altered cellular interactions; namely, the A53T mutant shows increased affinity for mitochondria and the MAM, whereas the A30P mutant shows weaker affinity (Cole et al., 2008; Devi et al., 2008; Guardia-Laguarta et al., 2014; Pozo Devoto et al., 2017). This may be related to the decreased binding affinity of A30P mutant α-syn for lipid membranes (Jo et al., 2002; Fortin et al., 2004). Therefore, these PTMs and site mutations likely alter the aggregation propensity and membrane interactions of α-syn that subsequently affect its interaction with the mitochondrial import machinery in complex ways.

The involvement of the TOM complex in α-syn import was further investigated in two other studies. In one report, brain tissue taken from postmortem PD patients and transgenic mice overexpressing α-syn showed decreased expression of Tom40 concurrent with mtDNA damage, oxidative stress, and reduced bioenergetic efficiency (Bender et al., 2013). Another report used a rotenone-induced Complex I dysfunction model of PD, as well as postmortem brain tissue of PD patients, to show that α-syn (specifically, the S129 phosphomimetic and soluble oligomers) demonstrated strong but reversible binding to Tom20 and reduced Tom20 expression. This binding thereby prevented interactions between the Tom20/Tom22 receptors, inhibited the import of mitochondrial proteins, and inhibited respiration (Di Maio et al., 2016b). In these studies, the overexpression of either Tom40 (Bender et al., 2013) or Tom20 (Di Maio et al., 2016b; De Miranda et al., 2020) reversed α-syn aggregation and its associated mitochondrial defects. Taken together, these results show a direct interaction between α-syn and the TOM complex, although exactly how the oligomeric state of α-syn or its PTMs may modify such interactions remain open questions. It should be noted that to date, there is no strong evidence for the specific interaction between α-syn and the TIM23 import complex of the MIM. However, transport and sorting of α-syn via the TIM23 machinery is highly likely given the submitochondrial distribution of α-syn, which includes the MIM and matrix. Figure 10B summarizes the interactions of α-syn with mitochondria.

Another feature supporting the mitochondria-targeting capacity of the α-syn N-terminus is its membrane-interactive nature. In its functional role of fusing synaptic vesicles with the presynaptic membrane, α-syn binds anionic phospholipid bilayers through its N-terminus, which then adopts a stable α-helical structure and displaces the unstructured C-terminal end resulting in an elongated conformation (Bartels et al., 2010). Similar membrane interactions have long been known for bona fide mitochondrial MTSs, which can undergo a coil-to-helix transition upon anchoring to lipid bilayers without perturbing membrane integrity (Skerjanc et al., 1987; Hoyt et al., 1991; Wieprecht et al., 2000). Functionally, this could allow the mitochondrial precursor to undergo a two-dimensional random walk on the membrane surface to seek out its cognate import receptor more efficiently. It is possible that α-syn adopts a similar mechanism to accumulate on membrane surfaces to increase its local concentration in the vicinity of import complexes.

The interaction of α-syn with the mitochondrial import machinery has important mechanistic implications for mitochondrial stress responses and quality control related to PD (Thorne and Tumbarello, 2022). One of the main pathways for mitophagy-based removal of damaged mitochondria is the PINK1/Parkin system. Under non-stressed conditions, the serine/threonine kinase PINK1 engages mitochondrial import complexes and becomes proteolyzed and efficiently degraded; however, under stress conditions that lower the Δψ_m of the MIM, PINK1 accumulates at the MOM and recruits the E3 ubiquitin ligase Parkin, which ubiquitinylates mitochondrial proteins. This signals for autophagic degradation of dysfunctional organelles and consequently blunts cellular expansion of the damage (Ge et al., 2020). Importantly, defects in PINK1 and Parkin (encoded by PINK1 and PRKN genes, respectively) cause the deregulation of mitochondrial quality control and together they represent the preeminent monogenic forms of heritable PD (Klein and Westenberger, 2012).

Moreover, the specific affinity of the α-syn N-terminal region for cardiolipin (Ramakrishnan et al., 2003; Cole et al., 2008; Nakamura et al., 2008; Grey et al., 2011; Robotta et al., 2014) may explain its interaction with mitochondria during stress. The translocation of cardiolipin from the MIM to the MOM is an early signaling event in mitophagy and apoptosis (Li et al., 2015), and it has been shown using α-syn mutant models of PD that cardiolipin becomes externalized to the cytosol, where it then binds α-syn and promotes the refolding of α-syn monomers from aggregated fibrils (Ryan et al., 2018). This may represent a feed-forward process whereby the stress of α-syn burden causes cardiolipin externalization, thereby resulting in the recruitment of more α-syn to the mitochondrial surface. Mitophagy then results when α-syn burden outmatches the refolding ability of cardiolipin.

It should be emphasized that α-syn may play a physiological role in mitochondria (Faustini et al., 2017). For example, consistent with the general role that α-syn appears to play in cellular lipid metabolism and signaling (Jenco et al., 1998; Golovko et al., 2005; Narayanan et al., 2005; Golovko et al., 2006), the lack of α-syn in SNCA ^-/- mice causes mitochondria to have reduced cardiolipin with altered acyl compositions (Ellis et al., 2005; Barcelo-Coblijn et al., 2007). This effect on cardiolipin in turn alters the physical properties of the MIM coupled with reduced Complex I and III activity (Ellis et al., 2005) and is accompanied by increases in neutral lipids including cholesterol and cholesterol esters (Barcelo-Coblijn et al., 2007). Other studies indicate that α-syn may play critical roles in mitochondrial dynamics, quality control and transport (reviewed in (Pozo Devoto and Falzone, 2017)). While these roles of α-syn could in principle be exerted within or outside the mitochondrion, some reports indicate a role of α-syn within the organelle. For example, α-syn appears to interact directly with the matrix-facing catalytic domain of mitochondrial ATP synthase, with monomers positively regulating its catalytic activity (Ludtmann et al., 2016) and oligomers promoting the mitochondria permeability transition involved in cell death (Ludtmann et al., 2018). This functional duality indicates that, as with cytoplasmic α-syn, it may be the abundance and/or aggregation of α-syn within the mitochondrion and not the mere presence of α-syn itself, that dictates pathogenicity inside the organelle. As such, the TIM23 machinery must then play a key role in regulating this balance of mitochondrial α-syn.

In Htt, the site of pathogenic poly-Q expansion is flanked by the first 17 residues of the N-terminus (Nt17) and the proline-rich domain (PRD) (Figure 8A). The Nt17 sequence has the hallmarks of a moderately amphipathic MTS (Figure 9Biv). Many structural and computational studies have addressed the conformational dynamics of the Nt17 sequence in the context of the tripartite structure of Htt. Nt17 itself does not adopt a stable secondary structure but has the characteristics of a compact coil (Thakur et al., 2009), which is similar to other IDPs (Uversky, 2002), and has the propensity to form α-helical structures, particularly in the presence of membranes (Davies et al., 1997; Tam et al., 2009). Indeed, Nt17 forms α-helices in the context of oligomers or fibrils of Httex1 fragments (Kim et al., 2009; Sivanandam et al., 2011; Michalek et al., 2013). The α-helical Nt17 structure promotes aggregation of Httex1 by promoting helical structure within the poly-Q tract (Thakur et al., 2009). By contrast, the PRD forms a polyproline II structure (PPII), which reduces aggregation propensity (Bhattacharyya et al., 2006). Importantly, Htt constructs lacking the Nt17 fail to localize to mitochondria, supporting a role of this segment in mitochondrial targeting (Rockabrand et al., 2007).

Using a combination of biochemical and microscopy-based analyses with striatum-derived cell lines and murine models of HD, Friedlander and colleagues found that Htt and mHtt (and N-terminal fragments thereof) interact directly with the mitochondrial TIM23 complex and are localized to the IMS (Yano et al., 2014; Yablonska et al., 2019) (Figure 10C). In these studies, immunoprecipitation/mass spectrometry and surface plasmon resonance spectroscopy were used to show interactions with TIM23 complex subunits Tim23, Tim50 and Tim17a. They revealed that mHtt had higher affinity interactions with TIM23 subunits and inhibited the TIM23-mediated import of native mitochondrial proteins to a greater extent than wild type Htt counterparts. Importantly, these studies showed that both the Nt17 segment and the expanded poly-Q tract are crucial for interactions with the TIM23 complex. The authors concluded that this inhibition of mitochondrial import by mHtt is an early event in HD pathology (detected pre-symptomatically) and that this effect alone can result in neuronal cell death. Figure 10C summarizes the interactions of mHtt with the mitochondrial import machinery.

These observations provide potential clues as to how Htt, specifically N-terminal segments of mHtt, may engage and disrupt the TIM23 machinery during HD-related proteostatic stress. First, the requirement of Nt17 for TIM23 interactions suggests that this sequence acts like a MTS, assuming an amphipathic α-helical structure to engage the TOM/TIM23 receptors and accumulate Htt/mHtt at mitochondria during HD progression. Second, because TIM23 substrates must be unfolded to traverse the import pathway, N-terminal fragments of Htt/mHtt may be more likely to engage TIM23 than full-length Htt/mHtt, as this would require significant unfolding for import. Finally, the selective interaction of mHtt fragments with TIM23 suggests that the poly-Q expansion provides a sufficiently long, unstructured tether for the Nt17 to access the TIM23 binding sites. Indeed, it has been shown that a sufficiently long presequence is required for TIM23 precursors to span both mitochondrial membranes and engage matrix mtHsp70, which translocates substrates by an active power stroke or Brownian ratchet mechanism (Matouschek et al., 1997; Okamoto et al., 2002). The localization of Htt/mHtt to the IMS suggests incomplete translocation of these polypeptides into the matrix, creating stalled translocation intermediates that could explain the inhibition of TIM23 import. It should be noted that the findings of the Friedlander group were subsequently questioned in a study that found mitochondria-localized mHtt to reside only on the cytosolic side of the MOM and to have no measurable effect on the import of TIM23 substrates (Hamilton et al., 2020). Future work will be required to reconcile these contradictory findings.

Mitochondrial proteostasis involves many stress response pathways that either directly or indirectly involve the protein import machinery (Figure 11), and any combination of them could be marshalled in response to amyloid burden at the TIM23 complex. First, some responses involve the degradation of toxic peptides or proteins in the matrix subcompartment. As noted above, the MTS-degrading enzyme PreP (Figure 11A) has an established role in the breakdown of Aβ (Falkevall et al., 2006) and may assume a more general role in the clearance of other amyloidogenic peptides. Other pathways have been described whereby mitochondria import unfolded or aggregated proteins into the matrix for degradation, providing a proteostatic mechanism for the clearance of cytotoxic proteins. These mechanisms, termed Mitochondria as Guardian in Cytosol (MAGIC) (Ruan et al., 2017) and the FUNDC1/HSC70 pathway (Li Y. et al., 2019), both entail the uptake of aggregation-prone proteins into mitochondria with subsequent degradation by matrix proteases. Thus, these pathways may explain why some amyloidogenic proteins are ectopically imported into mitochondria (Figure 11B). Alternative quality control pathways involve feedback mechanisms between the mitochondria and the nucleus to signal defects in mitochondrial protein biogenesis. Among them, the best characterized is the mitochondrial unfolded protein response (UPR^mt), which senses the accumulation of unfolded proteins and ROS in mitochondrial compartments and activates transcription factors to express nuclear genes encoding mitochondrial chaperones, proteases, and antioxidant systems (Wodrich et al., 2022) (Figure 11C). Indeed, UPR^mt activation is implicated in neurodegeneration (Zhu et al., 2021), and particularly in AD (Beck JS. et al., 2016; Sorrentino et al., 2017; Shen Y. et al., 2019), PD (Cooper et al., 2017), and HD (Berendzen et al., 2016; Fu et al., 2019). Other stress responses serve to clear the cytosol of mitochondrial precursor proteins when import is compromised. Namely, the UPR activated by the mistargeting of proteins (UPR^am) (Wrobel et al., 2015) and the associated mitochondrial precursor overaccumulation stress (mPOS) response (Wang and Chen, 2015) (Figure 11D) respond to increased cytosolic concentrations of precursors, which are normally very low, by activating the proteasome and downregulating cytosolic protein synthesis, respectively. Other similar responses are designed to extract stalled translocation intermediates from the TOM complex and direct them to the proteasome for degradation. These include the mitochondrial compromised import response (MitoCPR) system (Weidberg and Amon, 2018) and the mitochondrial protein translocation-associated degradation (MitoTAD) system (Martensson et al., 2019) (Figure 11E). Although many of these stress responses have been resolved in lower eukaryotes, there is evidence that they exist in mammalian systems as well, although the mechanistic details have yet to be elucidated.

One confounding factor in these mitochondria-based proteostasis mechanisms is the question of how aggregated proteins in the cytosol could access the TOM/TIM23 machinery, given that the import requires an unfolded precursor. A potential solution may come from the observation that a cytosolic HSP70/co-chaperone system is capable of disaggregating amyloid structures (Wentink et al., 2020). Such a system could in principle dislodge amyloid proteins and deliver them as monomers in a post-translational, chaperone-bound fashion to the mitochondrial import machinery. Indeed, several amyloids are found as distinct disease-associated polymorph structures (Qiang et al., 2017; Yang et al., 2022) that could differ in their stabilities and proclivities to dissociate soluble species. A second possibility is that since amyloid fibril structures feature rigid cores surrounded by disordered “fuzzy coats”, the solvent exposure of these flexible regions could promote PTMs and/or dissociation of monomers (Lin et al., 2017). Thus, shed monomers or proteolytic fragments of amyloidogenic proteins from fibrillar aggregates could then be free to interact with the mitochondrial protein machinery.

In summary, the degree to which PreP, MAGIC, and FUNDC1/HSC70 may assist as general clearance systems for amyloidogenic proteins like Aβ, α-syn, or mHtt requires further investigation, as do the pathways by which these amyloids may trigger mitochondrial import-related stress responses.

The interactions between amyloidogenic proteins and the mitochondrial import machinery may inform strategies for developing novel therapeutic interventions for neurodegenerative diseases. Clinical studies and drug discovery efforts have led researchers to question the role of amyloid fibrils themselves as the primary causative agents in AD, PD, and HD. For example, the extent of fibril formation does not always correlate with disease severity (Breydo et al., 2012; Ow and Dunstan, 2014), fibrils can be present even in the absence of disease symptoms (Ke et al., 2020), and therapeutic strategies designed to disrupt fibril formation have been met with mixed success (Pardridge, 2016; Barker and Mason, 2019; Paolini Paoletti et al., 2020; Pardridge, 2020; Yiannopoulou and Papageorgiou, 2020). Hence, alternative intervention approaches, including targeting mitochondrial dysfunction, are being developed for neurodegenerative disorders (Xu et al., 2022b). Here we review some promising interventions that directly or indirectly involve the amyloid-TIM23 pathway interaction.

From a broad perspective, given that AD, PD, and HD are all diseases of aging (Blumenthal, 2004; Taylor and Dillin, 2011), targeting general aging-related mitochondrial dysfunction may be an effective strategy. For example, the efficiency of the UPR^mt declines with age (Bennett and Kaeberlein, 2014; Munoz-Carvajal and Sanhueza, 2020) making it a potential target for neurodegenerative diseases (Suarez-Rivero et al., 2022). Indeed, compounds that increase cellular NAD⁺ levels, including nicotinamide riboside and olaparib, improved longevity in C. elegans models of Aβ proteotoxicity by activation of UPR^mt and mitophagy (Sorrentino et al., 2017). Because the UPR^mt involves upregulation of the TIM23 machinery (Xin et al., 2022), mitochondrial import may either be an important target of UPR^mt-modulating compounds, or serve as a marker for pharmacological UPR^mt activation. Another mitochondrial parameter that declines with age is the Δψ_m (Sugrue and Tatton, 2001; Berry and Kaeberlein, 2021; Berry et al., 2023). Caloric restriction (CR) and rapamycin, which extend lifespan in several wild type organisms (Nakagawa et al., 2012; Bitto et al., 2016; Shindyapina et al., 2022), can mediate their effects by regulating Δψ_m (Paglin et al., 2005; Cheema et al., 2021; Berry et al., 2023). These treatments could therefore operate in part by rescuing TIM23-mediated protein import, which functions in a Δψ_m-dependent manner. However, thorough pre-clinical testing of these interventions in AD, PD, and HD models would be warranted as CR and UPR^mt activation can shorten or extend lifespan depending on the context in which the treatments are applied (Liao et al., 2010; Bennett and Kaeberlein, 2014; Angeli et al., 2021; Xin et al., 2022). Notably, a recent study with rapamycin in a mouse model of AD showed increased amyloid plaque formation upon treatment (Shi et al., 2022).

Given the potential role of mitochondria-resident proteases in amyloid degradation, these may represent pharmacological targets. For instance, as mentioned, the mitochondrial PreP enzyme degrades MTSs and has been implicated in degradation of Aβ in the matrix (Falkevall et al., 2006; Alikhani et al., 2011b; Mossmann et al., 2014; Fang et al., 2015; Du et al., 2021); thus, PreP may represent a potential drug target for neurodegeneration (Brunetti et al., 2021). In fact, researchers using the Senescence Accelerated Mouse Prone 8 (SAMP8) AD model found that restoring global mitochondrial function with metabolic modulators (essential and branched-chain amino acids) was accompanied by a reversal of aging-related declines in PreP levels (Brunetti et al., 2020). Similarly, researchers using the APPswe/PS1dE9 AD model, associated with FAD mutations in APP and γ-secretase, found that treatment with the neuroprotective compound Ligustilide reduced disease progression with a concurrent increase in PreP levels (Xu et al., 2018). Very recently, Pioglitazone, an antagonist of the PPAR-γ transcription factor that regulates mitochondria structure and function, was shown to upregulate PreP and the insulin degrading enzyme IDE, thereby restoring the peptide processing machinery (Di Donfrancesco et al., 2023). Original efforts toward developing agonists of PreP were focused on small molecule benzimidiazole derivatives (Vangavaragu et al., 2014), but these results were called into question (Li NS. et al., 2019). Therefore, the potential efficacy of specific effectors of PreP requires further validation.

Finally, there are several therapeutic compounds effective in treating neurodegenerative diseases that improve mitochondrial function as part of their mechanism of action. Three such compounds, described here, could have direct or indirect effects on mitochondrial protein import. First, the synthetic tetrapeptide SS-31 (elamipretide) targets mitochondria by interacting with the cardiolipin-rich MIM (Szeto, 2014). SS-31 is protective against mitochondrial dysfunction associated with a range of diseases, including myopathy, cardiac, retinal, and kidney diseases, and has demonstrated neuroprotective function (Zhu et al., 2018; Zhao et al., 2019; Liu Y. et al., 2021; Nhu et al., 2021), including efficacy in models of AD (Manczak et al., 2010; Calkins et al., 2011; Reddy et al., 2017; Reddy et al., 2018), PD (Yang et al., 2009), and HD (Yin et al., 2016). In these studies, SS-31 has been shown to restore mitochondrial biogenesis, dynamics and energetic output, reduce oxidative stress, and preserve mitochondria structure, with a notable effect on upregulating TOM complex receptor expression (Reddy et al., 2018). SS-31 and its side chain variants may act by modulating membrane electrostatics (Mitchell et al., 2020; Mitchell et al., 2022), mitigating Ca²⁺ stress at the MIM (Mitchell et al., 2020), and/or preserving the Δψ_m (Zhang et al., 2020). Furthermore, SS-31 has an extensive interactome in mitochondria (Chavez et al., 2020) and may therefore attenuate pathogenic interactions of peptides such as Aβ and α-syn with different mitochondrial proteins. Second, the curcumin derivative J147 is a neuroprotective compound that has been explored as an effective treatment in AD models (Prior et al., 2013; Goldberg et al., 2018; Currais et al., 2019; Goldberg et al., 2020; Kepchia et al., 2021; Kepchia et al., 2022). Based on these studies, several molecular mechanisms have been ascribed to J147, including improved energy metabolism, modulation of Ca²⁺ flux and activation of the AMPK/mTOR pathway, reduction of plasma free fatty acid levels, and regulation of acetyl-CoA metabolism. Third, polyphenols constitute a broad class of phytochemicals, many of which show neuroprotective properties (Naoi et al., 2019). Among them, urolithin A has demonstrated efficacy in models of AD (Ballesteros-Alvarez et al., 2023) and PD (Liu et al., 2022). Mechanistically, urolithin A activates mitophagy (Ryu et al., 2016; Amico et al., 2021), which may reverse defects in PINK1/Parkin mitophagy that are implicated in both familial and sporadic PD (Dawson and Dawson, 2010). Because the PINK1/Parkin mechanism is based on interactions with the TIM23 pathway, urolithin A may help reduce proteostatic stress caused by interactions of α-syn with the import machinery. Notably, SS-31, J147, and polyphenolics are all known to interact with mitochondrial F₁F_O ATP synthase (Ahmad and Laughlin, 2010; Goldberg et al., 2018; Chavez et al., 2020), an enzyme whose dysfunction is implicated in neurodegeneration, particularly AD (Ebanks et al., 2020). Increasing evidence supports that dysfunction of ATP synthase coincides with AD progression (Liang et al., 2008; Terni et al., 2010; Beck SJ. et al., 2016), which may mechanistically occur by interactions of Aβ with the OSCP subunit (Beck SJ. et al., 2016). It has also been shown that loss of OSCP drives the opening of the mPTP and activates the UPR^mt (Angeli et al., 2021). Hence, compounds that bind ATP synthase may help maintain its function during amyloidogenic stress and/or inhibit mPTP pore opening induced by the pore initiator cyclophilin D, producing a result similar to the action of cyclosporin A (Connern and Halestrap, 1994; Gauba et al., 2019). Such interventions could increase ATP synthase activity and prevent membrane depolarization, thereby improving available ATP levels to enhance proteostatic clearance and matrix-directed import through the TIM23 pathway.