ATP Synthase Diseases of Mitochondrial Genetic Origin

Devastating human neuromuscular disorders have been associated to defects in the ATP synthase. This enzyme is found in the inner mitochondrial membrane and catalyzes the last step in oxidative phosphorylation, which provides aerobic eukaryotes with ATP. With the advent of structures of complete ATP synthases, and the availability of genetically approachable systems such as the yeast Saccharomyces cerevisiae, we can begin to understand these molecular machines and their associated defects at the molecular level. In this review, we describe what is known about the clinical syndromes induced by 58 different mutations found in the mitochondrial genes encoding membrane subunits 8 and a of ATP synthase, and evaluate their functional consequences with respect to recently described cryo-EM structures.


INTRODUCTION
Mitochondria support aerobic respiration and produce the bulk of cellular ATP by oxidative phosphorylation (OXPHOS) (Saraste, 1999). Electrons provided by the oxidation of fatty acids and carbohydrates are shuttled to oxygen along four respiratory chain (RC) complexes (I-IV) embedded in the inner mitochondrial membrane (IMM), producing water and releasing the energy necessary to pump protons from the mitochondrial matrix to the intermembrane space (IMS). This results in the formation of transmembrane electrochemical ion gradient across the IMM, also called the proton-motive force (pmf ). The outer side of the IMM is positively charged (the p-side) while the inner side is negatively charged (the n-side). The pmf enables the F 1 F o ATP synthase to produce ATP from ADP and inorganic phosphate (Boyer, 1997). The OXPHOS complexes contain ∼90 structural proteins of which 13 are encoded by the mtDNA in humans.

Disease/Syndrome Phenotypes
Apical hypertrophic cardiomyopathy (AHCM) and neuropathy primary disease of the myocardium (the muscle of the heart) in which a portion of the myocardium is hypertrophied (thickened) without any obvious cause, creating functional impairment of the cardiac muscle; neuropathy is damage to or disease affecting nerves, which may impair sensation, movement, gland or organ function, or other aspects of health, depending on the type of nerve affected Ataxia genetic disorders characterized by slowly progressive incoordination of gait and is often associated with poor coordination of hands, speech, and eye movements, with full mental capacity Autism neurodevelopmental disorder, characterized by impaired social interaction, verbal and non-verbal communication, and restricted and repetitive behavior; noticeably affected by mitochondrial dysfunction which impairs energy metabolism Charcot-Marie-Tooth syndrome (CMT) hereditary disorders that damage the nerves in arms and legs (peripheral nerves); symptoms usually begin in feet and legs, but they may eventually affect hands and arms Encephalopathy abnormal brain function or brain structure, symptoms may be mental or physical dysfunctions, depending on what part of the brain is being affected Epilepsy with Brain Pseudoatrophy brain disorder manifesting by seizures, dementia, convulsions, loss of control on muscles, difficulties with talking This review focuses on mutations in the MT-ATP8 and MT-ATP6 genes (further named ATP8 and ATP6) encoding subunits 8 and a of ATP synthase, respectively, that were identified in patients with various disorders. We summarize what is known about their clinical and functional consequences. Based on recent high-resolution structures (Morales-Rios et al., 2015a;Zhou et al., 2015;Hahn et al., 2016;Guo et al., 2017), we define the topological locations of these mutations, which helps understand their impact on ATP synthase structure, function and assembly.

ATP SYNTHASE STRUCTURE AND FUNCTION
Mitochondrial ATP synthase is a unique macromolecular rotary machine of ∼625 kDa. It is composed of typically 17 different protein subunits (Figure 1) and organizes into a membraneextrinsic F 1 catalytic and membrane-embedded F o domains, which are connected by a peripheral and central stalk (Allegretti et al., 2015;Morales-Rios et al., 2015a;Zhou et al., 2015). The matrix-oriented F 1 is composed of a prominent (αβ) 3 catalytic head into which the γ δε central stalk rotor penetrates. The F o c-ring typically consists of identical c subunits (subunit 9 in yeast). Together with subunit a, the c-ring shuttles protons across the membrane. The F o domain further consists of subunits 8 (alias A6L), b (4 in yeast), f, d, F6 (h in yeast), and OSCP that together form the peripheral stalk connecting the catalytic head with the membrane stator. Three mitochondria-specific subunits, e, g, k, induce either directly or indirectly the formation of ATP synthase dimers (Hahn et al., 2016) that self-assemble in longer ribbons important for cristae formation (Parsons, 1963;Strauss et al., 2008).
The ATP synthase harbors a unique rotary mechanism driven by the pmf to translocate ions through F o , to generate FIGURE 1 | Cartoon representation of the yeast F 1 F o ATP synthase. The view is horizontally to the membrane shown in grayscale. The structure was drawn according to (Hahn et al., 2016, PDB code 5FL7). For simplicity the structure is shown without subunits e, f, g, I, and a truncated subunit b. The figure was made in Pymol (The PyMOL Molecular Graphics System, Version 0.99, Schrödinger, LLC), using the following color code: α, forest; β, split pea; γ , density; δ, cyan; ε, white; OSCP, red; b (= 4 in yeast), dirty violet; d, orange; h, salmon; 8, green; a, blue; c-ring, yellow. The arrows indicate the path of protons (see also Figure 4) and nucleotide conversion. For details see text.
rotation of its rotor and transmit torque into the F 1 catalytic head where finally ATP is synthesized and released. Subunit a provides a pathway that involves a number of hydrophilic amino acids, which allows protons to enter from the IMS (Figure 1). Approximately in the middle of the membrane the proton can bind to a highly conserved acidic residue of subunit c helix 2 (cH2) (cE59 in H. sapiens) located at the outer surface of the c-ring. It has been suggested that the binding of a proton on this carboxylate residue disrupts a previously established electrostatic interaction of cE59 with a highly conserved, positively charged arginine residue in subunit a membrane helix 5 (aH5) (aR159 in H. sapiens; Vik and Antonio, 1994;Junge et al., 1997;Pogoryelov et al., 2010;Guo et al., 2017). This arginine acts as an electrostatic separator between the proton pathway from the IMS to the middle of the membrane and a second, spatially separated pathway that allows incoming protons still bound on the c-ring glutamate to be released into the matrix (Mitome et al., 2010). The operation direction of this process is primarily driven by the ion gradient that causes a ratchet type mechanism of the neutralized c-ring glutamate in the hydrophobic membrane, which is energetically unfavorable and does not allow the back stepping without externally applied force (Vik and Antonio, 1994;Junge et al., 1997). After an almost complete revolution of the c-ring, the glutamate is deprotonated in the aqueous exit channel (Pogoryelov et al., 2010). This channel is formed by hydrophilic residues of the c-ring/subunit a interface at the matrix side through which the protons can reach the matrix (n-Side) (Allegretti et al., 2015;Morales-Rios et al., 2015a;Zhou et al., 2015;Hahn et al., 2016;Guo et al., 2017). The c-ring is tightly bound to the central stalk subunits γ δε, of which subunit γ protrudes into the F 1 catalytic head, which induces cyclic conformational changes when rotating (Abrahams et al., 1994). Consequently, ADP and P i are sequentially converted at the catalytic sites of the three subunits β, according to the binding change mechanism (Boyer, 1997). Cryo-EM structures of the bovine Bos taurus and yeasts Yarrowia lipolytica and Saccharomyces cerevisiae F 1 F o ATP synthases, that are basically of the same subunit composition and structural construction as the human enzyme, have been described recently [15][16][17]. These structures show a very similar overall architecture and differ only with respect to the subunit c (9) stoichiometry (8 in mammals, 10 in yeasts), the loss of the dimerization domain subunits (e/g/k) during purification (yeast) and the non-essential and yeast specific subunits i and k (Zhou et al., 2015;Hahn et al., 2016;Guo et al., 2017). It therefore has become feasible to build reliable structural models of the membrane domain (F o ) of the eukaryotic, mitochondrial, ATP synthase (Figure 1), and to map the human disease-causing mutations at the molecular level within the ATP synthase structure and to pin-point their potentially adverse effects on the above-described mechanism.

YEAST AND HUMAN CELLULAR MODELS OF mtDNA DISEASES
Human cells contain up to thousands copies of mtDNA (Miller et al., 2003). Mutations in this DNA are highly recessive and usually co-exist with wild type mtDNA molecules, a situation referred to as heteroplasmy. A mutational load above 60% is usually required to induce a clinical phenotype (Stewart and Chinnery, 2015). Given the high mutational rate of the mitochondrial genome and the presence of numerous family or population-specific polymorphisms, it can be difficult to distinguish between a neutral mtDNA variant and a diseasecausing mutation. Additionally, the effects of deleterious mtDNA mutations might be exacerbated by mtDNA nucleotide changes that are not pathogenic per se and by unknown factors in nuclear genetic background, i.e., the so-called modifier genes (Cai et al., 2008;Swalwell et al., 2008). These features make it difficult from patient's cells and tissues to precisely know how specific mtDNA mutations influence oxidative phosphorylation.
To better characterize the effects of mtDNA mutations, homoplasmic cell lines, i.e. with a 100% mutational load, in a defined nuclear genetic background are required. To this end King and Attardi (King and Attardi, 1989) developed an approach that used cybrid (cytoplasmic hybrid) cell lines obtained by fusing enucleated cytoplasts from patient's cells with cells lacking mtDNA (ρ 0 ). This approach was used to evaluate the bioenergetics consequences of 11 ATP6 mutations (Trounce et al., 1994;Majander et al., 1997;Nijtmans et al., 2001;Carrozzo et al., 2004;Mattiazzi et al., 2004;Pallotti et al., 2004;Jonckheere et al., 2008;Sikorska et al., 2009;Aure et al., 2013;Blanco-Grau et al., 2013;Lopez-Gallardo et al., 2014;Hejzlarova et al., 2015;Wen et al., 2016). Another approach exploits unique features of S. cerevisiae. Mitochondria from this single-celled fungus and humans show many similarities (Steinmetz et al., 2002;Prokisch et al., 2004;Reinders et al., 2006;Pagliarini et al., 2008;Rhee et al., 2013), and mitochondrial genetic transformation can be achieved in this yeast in a highly controlled fashion, by the biolistic delivery into mitochondria of in-vitromade mutated mtDNA fragments, followed by their integration into wild type mtDNA by homologous DNA recombination (Bonnefoy and Fox, 2001). Being unable to stably maintain heteroplasmy (Okamoto et al., 1998), it is easy to obtain yeast homoplasmic populations where all mtDNA molecules carry a mutation of interest. Owing to its good fermenting capacity, yeast models of human mitochondrial diseases can be kept alive when provided with sugars like glucose even when oxidative phosphorylation is completely inactivated (Baile and Claypool, 2013;Lasserre et al., 2015). This approach was used to investigate the impact on ATP synthase of nine ATP6 mutations identified in patients (Rak et al., 2007;Kucharczyk et al., 2009aKucharczyk et al., ,b,c, 2010Kucharczyk et al., , 2013Kabala et al., 2014;Lasserre et al., 2015;Niedzwiecka et al., 2016;Wen et al., 2016).

PATHOGENIC MUTATIONS IN ATP8 AND ATP6
Subunits 8 and a are synthesized from a bi-cistronic mRNA unit (Figure 2). The two genes show a 46 nucleotide overlap. Thus, mutations in this unit can affect either subunit a or 8, or both. Currently, 36 different ATP8 and ATP6 mutations with a confirmed or suspected pathogenic character are recorded in MITOMAP database (Figure 3). We here review 22 additional mutations found in literature. The nucleotide and amino acid changes induced by these mutations, and what is known about their functional and clinical consequences is summarized in Table 2

Most Frequent Mutations
• Two mutations at the same amino acid position of subunit a, m.8993T>G (aL156R) and m.8993T>C (aL156P), were identified in numerous patients presenting with the NARP or MILS syndrome depending on the mutation load (Uziel et al., 1997;Jonckheere et al., 2012). The first one was consistently found to severely compromise mitochondrial ATP production, with deficits of up to 90%. While most studies concluded this was due to a block in proton translocation, some suggested a less efficient coupling or defects in the assembly/stability of ATP synthase (Tatuch et al., 1992;Trounce et al., 1994;Houstek et al., 1995;Vazquez-Memije et al., 1998;Garcia et al., 2000;Nijtmans et al., 2001;Carrozzo et al., 2004;Mattiazzi et al., 2004;Pallotti et al., 2004;Morava et al., 2006;Sgarbi et al., 2006;Baracca et al., 2007;Cortes-Hernandez et al., 2007). Although its pathogenic character is firmly established, the second mutation has less severe consequences on ATP synthase with a 70% drop in ATP production mainly because of a less efficient assembly or diminished stability of subunit a (Vilarinho et al., 2001;Morava et al., 2006;Craig et al., 2007;Kucharczyk et al., 2009a;Aure et al., 2013). In addition to bioenergetic deficits, the two mutations lead to an enhanced production of ROS and aberrant mitochondrial morphologies that may contribute to the disease process as well. • Similar diseases were associated with two mutations at amino acid position 217, m.9176T>G (aL217R), and m.9176T>C  Table 2.
(aL217P). The first one is extremely detrimental with a block in subunit a assembly that leads to extreme clinical phenotypes when highly abundant in cells and tissues. The second one does not obviously compromise assembly of subunit a indicating that it affects the functioning of ATP synthase. • The m.9185T>C (aL220P) mutation was identified in individuals presenting with MILS, MNS, periodic paralyzes, spinocerebellar ataxia syndromes (SCA) or CMT (Castagna et al., 2007;Childs et al., 2007). Biochemical analyses revealed a substantial drop (50-90%) in ATP production, and study in yeast indicated that this mutation compromises the functioning of ATP synthase (Kabala et al., 2014).  De Meirleir et al., 1995;Honzik et al., 2013;Ye et al., 2013). Studies with yeast revealed that m.8851T>C leads to major drops (95%) in mitochondrial ATP synthesis owing to a block in F o -mediated proton transfer (Kucharczyk et al., 2013). The m.9134A>G (aE203G) was identified in patient suffering from MS accompanied with cardiomyopathy (Honzik et al., 2012). • The m.9035T>C (aL170P) was identified at high (>90%) mutation load in 21 ataxia patients (Sikorska et al., 2009;Pfeffer et al., 2012). Cybrids carrying this mutation had reduced ATP levels (40-50% vs. controls) and produced 5-7 times more ROS than control cells. Another mutation m.8611insC (aL29PfsX36) was found in patient presenting ataxia with encephalomyopathy (Jackson et al., 2017). identified in children with neuromuscular disorders (Felhi et al., 2016). The first one was also identified in prostatic cancer cells (Petros et al., 2005). In a yeast model of this mutation, ATP synthase assembly/stability was found substantially affected . • The m.9205delTA (aSTOP elimination) mutation was identified in patients with a severe encephalopathy leading to premature death. Since the stop codon of ATP6 overlaps with the start codon of COX3, the expression of both genes is compromised, which results in a lower content in complex IV and ATP synthase (Jesina et al., 2004;Hejzlarova et al., 2015). It is puzzling that mutations that cluster in specific regions of ATP6 or ATP8 give rise to such a wide variety of clinical symptoms (Table 1). These phenotypic differences may be due to other unknown genetic variations in patients within mitochondrial or nuclear DNA that could exacerbate  (Table 2) and the mutations modeled in S.c., respectively. The arrows are colored according to Figure 4. At the bottom, the secondary structural elements are drawn according to PSIPRED prediction (2D) and a cryo-EM structure (3D) (Hahn et al., 2016). or attenuate the consequences on health of defects in ATP synthase subunits. Another important source of variability in the clinical outcome likely resides in the levels of heteroplasmy and different distributions of mtDNA mutations in cells and tissues. Furthermore, in addition to a lack of ATP, defects in ATP synthase may have multiple secondary effects, such as increased production of ROS and changes in upstream metabolic processes (Korshunov et al., 1997) that together will influence the disease process unpredictably.

TOPOLOGY WITHIN THE F O OF MUTATIONS IN SUBUNITS a AND 8
To define the topology of the ATP6 and ATP8 mutations identified in patients, we used the recently published structures of Y. lipolytica and S. cerevisiae ATP synthases (Hahn et al., 2016;Guo et al., 2017); the complete model of subunits a, 8, and the c-ring is shown in Figure 4. The amino acid alignments in Figure 3 establish the correspondences with human subunits a and 8 amino acids.

Subunit a Mutations
Helix 5 of subunit a (aH5) is kinked due to the presence of proline at position 153, a residue well-known for its propensity to bend or break alpha helices owing to its inability to participate fully in protein backbone hydrogen bonding. aP153 enables aH5 to follow the curvature of the c-ring and seal the two hydrophilic pockets that connect the a/c-ring interface to the intermembrane and matrix spaces. Five substitutions from hydrophobic alanine or leucine residues into proline are located on aH5 (aH5) in proximity to the essential aR159 residue (aA105P, aA155P and aL156P) that faces the proton binding glutamate of the c-ring, or close to aH168/aE203 in the proton entry channel (aL169P, aL170P). These mutations may distort or break aH5. Those at positions 155 and 156 at one helix turn from aR159 may compromise the ion translocation mechanism, for example by ion short circuiting (Mitome et al., 2010) or by preventing aR159 to interact with the c-ring glutamate due to its structurally shifted position. On aH6, the aL217P, aL220P, and aL222P mutations are on the edge of the exit channel close to the matrix. Their severe functional consequences possibly result from a change in the topology of the nearby aD224 residue that was suggested to be of critical importance for the exit of protons toward the mitochondrial matrix (Guo et al., 2017). Similarly, the aP136S change in the loop connecting aH4 and aH5 possibly alters the accessibility of protons in this region of subunit a.
The extremely detrimental consequences of aL217R on ATP synthase assembly/stability (Kucharczyk et al., 2009b) possibly results from the inability of subunit a to pack tightly owing to replacement of a hydrophobic residue with a bulkier and positively charged one within the membrane. The absence of proton conduction induced by the aL156R mutation at the a/c-ring interface (Rak et al., 2007), without any defect in ATP synthase assembly, may be caused by the inability of protons to exit from the ring or by electrostatic or steric hindrance that prevent rotation of the ring. Being located near the N-terminal side of aH5, the block in proton translocation induced by the aS148N mutation (Wen et al., 2016) possibly results from obstruction of the proton exit pore. The detrimental consequences of aH168R (Lopez-Gallardo et al., 2014) are not surprising considering its location in the p-side cleft in the proximity of the c-ring. This positive charge cuts off the connection of the p-side to the c-ring. A similar effect on the n-side of the membrane may result from the aW109R, where aH5 and aH6 diverge. A mutation at this position has an impact in proton translocation without impacting ATP synthase assembly/stability (Kucharczyk et al., 2013), indicating that this location of subunit a is close to the pathway along which protons are evacuated into the matrix. The clinical consequences of aK122E probably also result from a less efficient proton conduction toward the matrix.
While aM140T, aG167S, aA177T, aI192T, aA205T, and aY212H decrease the hydrophobicity of the a/c-ring interface, it is less obvious from our structural model to predict the consequences of other mutations that replace hydrophobic to non-charged hydrophilic residues (aI24T, aI106T, aA126T) or vice-versa (aT53I, aT59A, aT178A, aS182L) and those that lead to small hydrophobicity changes (aM57I, aM104V, aV142I, aA162V, aI164V, aL190F). However, meaningfully, most are within helices aH4-6 that are important for the movement of protons through F o .
half of subunit a (Hahn et al., 2016). Four of the mutations identified in patients (8T6A, 8P10S, 8I13T, 8M16V) cluster at the N-terminal region of helix 8H1 in proximity to aH4, suggesting that they may affect the stability of subunits a and 8 (Hahn et al., 2016). The six other mutations identified in patients (8Y33C, 8P36L, 8P39L, 8N46I, 8K49E, 8E52K) are in the matrix-exposed helix 8H2. These mutations might affect the flexibility of the outer stalk, and thereby compromise the stability of F o and/or, indirectly, the ion translocation mechanism or ATP synthase assembly process. This hypothesis is supported by the reduced functionality and stability of ATP synthase in these patients and by various studies on subunit b in the bacterial enzyme Altendorf, 1984, 1985;Wehrle et al., 2002;Greie et al., 2004).

ATP SYNTHASE DIMERS AND MITOCHONDRIAL MORPHOLOGY
The mitochondrial ATP synthase exists as dimers (Schagger and Pfeiffer, 2000;Paumard et al., 2002) that associate into rows that contribute to cristae formation (Davies et al., 2011;Hahn et al., 2016). The mutations in subunits 8 and a often correlate with pathological forms of mitochondria cristae as for example seen in the Leigh syndrome (Kucharczyk et al., 2009b). The defects caused by these mutations therefore not only affect the ATP synthase function but they can also affect the assembly process. The reduced amount, or lack thereof, of native and completely assembled ATP synthase dimers would certainly affect cristae formation, which is crucial for the accommodation of the OXPHOS respiratory chain complexes and the ATP synthase. This explains some of the pathologic forms of mitochondria found in the diseases and syndromes described in this review.

CONCLUSIONS
Diseases associated to mutations in the mitochondrial ATP6 and ATP8 genes are particularly challenging to study due to factors like heteroplasmy, complex inheritance, variable penetrance, and interactions with (e.g., nuclear) modifier genes, which makes it difficult to verify their pathogenicity. The possibility to create and keep alive homoplasmic strains of S. cerevisiae with defined mtDNA mutations in a controlled nuclear genetic background makes it possible to study their functional consequences. With the recently obtained cryo-EM structures of F 1 F o ATP synthase from various mitochondrial origins it has become feasible to map mutations in ATP6 and ATP8 at the molecular level within the F o . These "open eyes" provide the chance for a completely new level of understanding of how the mutations may affect ATP synthase structure, assembly, and mechanism. This knowledge also enables to evaluate the observed pathogenic forms of mitochondrial morphology that are associated with these syndromes on the structural level, from the mutation at the molecular level to its associated consequences at the macroscopic scale of the organelle. The recent technical advances enabling the structural analysis of macromolecular complexes by cryo-EM (Kühlbrandt, 2014), the advent of complete structures of ATP synthases (Allegretti et al., 2015;Morales-Rios et al., 2015b;Zhou et al., 2015;Hahn et al., 2016;Sobti et al., 2016;Guo et al., 2017) and the availability of genetically approachable systems like S. cerevisiae are just the first steps in these new shoes; they will considerably improve the comprehension of human diseases associated to defects in this key mitochondrial enzyme. We are still at the beginning of understanding these complex processes.

AUTHOR CONTRIBUTIONS
AD, AH, TM, DT-T, J-PdR, and RK discussed findings, analyzed literature and wrote the manuscript; AD, AH, and TM designed and made the Figures.