Disease models of mitochondrial aminoacyl‐tRNA synthetase defects

Mitochondrial aminoacyl‐tRNA synthetases (mtARS) are enzymes critical for the first step of mitochondrial protein synthesis by charging mitochondrial tRNAs with their cognate amino acids. Pathogenic variants in all 19 nuclear mtARS genes are now recognized as causing recessive mitochondrial diseases. Most mtARS disorders affect the nervous system, but the phenotypes range from multisystem diseases to tissue‐specific manifestations. However, the mechanisms behind the tissue specificities are poorly understood, and challenges remain in obtaining accurate disease models for developing and testing treatments. Here, some of the currently existing disease models that have increased our understanding of mtARS defects are discussed.


| RECESSIVE MUTATIONS IN MITOCHONDRIAL AMINOACYL-tRNA SYNTHETASES CAUSE VARIABLE MITOCHONDRIAL DISEASE PHENOTYPES
Mitochondria are key metabolic regulators in our cells, contributing to numerous cellular processes that can be either part of the maintenance of normal cell function or highly specified depending on the cell type or state. One of the important functions of mitochondria is the production of ATP by oxidative phosphorylation (OXPHOS). It requires a functional respiratory chain (complex I-IV) and the ATP synthase (complex V) on the mitochondrial inner membrane. Four of the five OXPHOS complexes (I, III-V) contain crucial polypeptides that are encoded in the mitochondrial DNA (mtDNA) and translated by a dedicated protein synthesis machinery within mitochondria. A part of this organellar translation system is the mitochondrial aminoacyl-tRNA synthetases (mtARS), a set of ubiquitously expressed enzymes that enable the accuracy of protein synthesis in mitochondria by charging mitochondrial tRNAs with their cognate amino acids ( Figure 1). The 22 mitochondrial tRNAs are encoded in mtDNA, whereas the mtARS enzymes are encoded by 19 nuclear genes, which all have been identified as human disease genes since the year 2007, 1 extensively reviewed by us and others. [2][3][4][5][6][7][8] The set of mtARS genes is separate from those encoding cytosolic aminoacyl-tRNA synthetases, with the exception of genes for glycyl-tRNA synthetase (GARS1) and lysyl-tRNA synthetase (KARS1), which each encode both the cytosolic and mitochondrial enzymes. The mtARS are essential for human development, and thus all of the mtARS diseases are recessively inherited, and caused by mutation combinations that allow for some residual tRNA charging activity.
What has been surprising is the wide variety of tissuespecific manifestations of mtARS diseases, despite each synthetase having its function in the same specific step of mitochondrial protein synthesis. The mtARS diseases typically affect the nervous system, but may also cause phenotypes ranging from early-onset lethal multiorgan diseases to combinations of tissue-specific manifestations. Such variety in disease outcomes is typical in mitochondrial diseases, 9 which counteracts the hypothesis that mtARS disease phenotypes might be caused by defects in non-canonical functions of the mtARS enzymes. Yet the possible "moonlighting" roles of these housekeeping mitochondrial tRNA-synthetases are currently poorly known and may turn out to contribute to some of the identified phenotypes. For example, mitochondrial threonyl-tRNA synthetase (TARS2) was recently reported to be required for threonine-sensitive mTORC1 activation, with a fraction of TARS2 (10% to 15%) localizing in the cytosol, 10 and mitochondrial trypthophanyl-tRNA synthetase (WARS2) has been found to contribute to angiogenesis. 11 Some mitoARS disease phenotypes are not gene but mutation specific. For example, certain mutations in the mitochondrial alanyl-tRNA synthetase (AARS2) cause an infantile-onset lethal cardiomyopathy whereas other AARS2 mutations lead to a progressive leukodystrophy with onset typically after teenage years. [12][13][14][15] Missense mutations in SARS2 can cause HUPRA syndrome (hyperuricemia, pulmonary hypertension, renal failure, and alkalosis), whereas a splicing variant in the same gene causes early-onset spastic paresis without renal involvement. 16,17 Here approaches to model mtARS diseases, as well as some challenges in their modeling, are discussed.

| INABILITY TO CHARGE tRNAs WITH THE COGNATE AMINO ACID MAY ALTER AMINO ACID POOLS
The mtARS are collectively required for functional OXPHOS, but a defect in each of them may affect the pool of the particular amino acid that they charge the mitochondrial tRNA with. The resulting alteration in amino acid metabolism could have variable effects depending on the tissue. Curiously, mutations in both the cytosolic and mitochondrial aspartyl-tRNA synthetase, DARS1 and DARS2, respectively, lead to similar white matter diseases with a unique MRI signature, 1,18,19 which suggests that aspartate could be the common culprit. These conditions are named Hypomyelination with Brainstem and Spinal Cord Involvement and Leg Spasticity (HBSL) and Leukoencephalopathy with Brainstem and Spinal Cord Involvement and Lactate Elevation (LBSL), respectively. HBSL and LBSL patients may present with relapsing-remitting clinical course, showing regression after a viral infection or vaccination, thus they resemble acquired neuroinflammatory disorders such as multiple sclerosis. 20 Aspartate was recently shown to be important for autoimmune T cells, in which a lack of mitochondrial aspartate production prevented the regeneration of cytosolic oxidized nicotinamide adenine dinucleotide (NAD+), disrupting the regulation of the size of endoplasmic reticulum (ER). 21 Whether altered aspartate metabolism is a contributor to the brainstem and spinal cord tract phenotypes, or if neuroinflammation plays a role, are open questions. Possibilities are many, but it could also be speculated that changes in the aspartate pool, resulting from either DARS1 or DARS2 defects could have similar effects through ER dysregulation. F I G U R E 1 Mitochondrial aminoacyl-tRNA synthetases are encoded by 19 nuclear genes, translated by cytosolic ribosomes, and imported into the mitochondrial matrix where their charge mitochondrial DNAencoded tRNAs with their cognate amino acids as the first step in mitochondrial protein synthesis, which produces 13 critical polypeptides of the oxidative phosphorylation system.

| MITOCHONDRIAL tRNA CHARGING IS NOT CRITICAL IN CULTURED GLYCOLYTIC CELLS
Several aspects have complicated the studies of mtARS disease mechanisms. One is that proliferating cell types such as cultured skin fibroblasts from patients can be highly resistant to defects in mitochondrial tRNA charging, as they rely more on glycolysis than OXPHOS for energy production. For example, a homozygous splicing variant in SARS2 that we identified in a patient who had early-onset spastic paresis, led to a 95% reduction in the amount of SARS2 protein but had no apparent effect on mitochondrial protein synthesis in cultured skin fibroblasts. 17 SARS2 charges two different mitochondrial seryl-tRNAs, and SARS2 mutations have been shown to selectively affect the stability of tRNA Ser AGY , which is an unusually small mitochondrial tRNA lacking the entire D domain. 16,17,22 The degree of tRNA instability varies between mutation and cell type. Another example of how the mutation type may cause tissue-specific effects is the most common mutation in DARS2, which is an intronic variant affecting the splicing of DARS2. 1 It was shown that this "leaky" splicing variant results in a lower amount of functional DARS2 protein specifically in neuronal cell types. 23 Thus in studies using cultured proliferating cells, the mutation effect may not be the same as in differentiated cells, or the cultured cells may not manifest an OXPHOS phenotype even when the mitoARS function is severely disrupted.

| CHALLENGES IN GENERATION OF MOUSE MODELS FOR MITOCHONDRIAL AMINOACYL-tRNA SYNTHETASE DISEASES
Genetically modified mice have in general been quite useful in modeling mitochondrial diseases. 24 For the nuclear-encoded mtARS genes, standard genome modification methods of mouse ES cells have been available for generating mouse models. Homozygous full-body mtARS knockout mice are embryonic lethal, showing that these enzymes are not functionally redundant, whereas heterozygous knockouts are phenotypically normal. 25,26 Mice generated to carry variants equivalent to patient mutations could be expected to provide the most accurate disease models, however, for example, the common DARS2 intronic variant is not conserved in the mouse genome, preventing its direct modeling.
The same is true for the European founder mutation R592W in the proofreading domain of AARS2 underlying infantile-onset cardiomyopathy. 13 Besides the aminoacylation domain, two mtARS enzymes, AARS2 and TARS2, possess a proofreading function for removing mischarged tRNAs. 14, 27 We initially hypothesized that the AARS2 cardiomyopathy phenotype was caused by a defective proofreading activity, leading to mistranslation, since the R592W mutation located to this domain of the synthetase, 14 whereas the AARS2-leukodystrophy mutations affected the aminoacylation. As it was not possible to model the AARS2 R592W mutation in mice, we generated two different knock-in mouse models by altering amino acids required for the proofreading, causing a mild and severe defect. However, both homozygous mice were early embryonic lethal, showing that proofreading activity is essential in mammals. 28 Furthermore, in vitro studies later indicated that the disease-associated AARS2 mutations in the proofreading domain do not affect the editing function. 15,29 Even if the mutation site was conserved in mice, obtaining a useful disease model can be challenging. It was recently reported that compound heterozygous SARS2 variants identified in a patient with a multisystem phenotype were both embryonic lethal as homozygotes in mice, as well as in compound heterozygosity, 22 suggesting species-specific differences to tolerance of mitoARS defects.

| MOUSE AND OTHER ANIMAL MODELS OF REDUCED mtARS FUNCTION
The so far most comprehensively studied mtARS in mouse models is DARS2, with inducible knockouts produced in several different tissues or cell types by the Trifunovic lab ( Figure 2). First, a muscle creatine kinase promoter for Cre recombinase was used to induce the loss of DARS2 starting on embryonic day 15.5, resulting in the complete absence of DARS2 by 6 weeks of age in skeletal muscle and heart. 26 As expected, the progressive reduction of DARS2 function led to deregulation of mitochondrial protein synthesis. The mice developed cardiac hypertrophy, with compensatory induction of mitochondrial biogenesis, and survived only 6-7 weeks. Interestingly, the authors identified a strong activation of proteotoxic stress responses in mouse heart, with markers for mitochondrial unfolded protein response (UPRmt) and mitochondrial integrated stress response (mtISR) highly induced. 26 The UPRmt refers to the induction of mitochondrial proteases and chaperones that respond to accumulation of abnormal proteins inside mitochondria, whereas mtISR integrates the mitochondrial proteotoxicity with cellular stress that activates the ATF4/ATF5 transcriptional response and results in the rewiring of one-carbon metabolism as well as induction of secreted systemic regulators of metabolism such as FGF21 and GDF15. These responses have been described in a number of models of mitochondrial disease or dysfunction, [30][31][32][33][34][35] and are not specific to DARS2, although their induction was particularly pronounced in the hearts of DARS2 mice.
Next, DARS2 was depleted separately in mouse forebrain-hippocampal neurons and in myelin-producing cells, and the outcomes were compared. 36 In both models, OXPHOS function was severely disturbed, however, the resulting phenotypes were completely different. DARS2 loss in neurons led to apoptotic cell death and brain atrophy, reducing the survival of the mice up to 32 weeks of age. Although OXPHOS deficiency was present already at 15 weeks of age, massive cell death took place at 20 weeks, followed by the development of severe brain atrophy in the next weeks, showing that neurons were able to compensate for the OXPHOS defect for a limited time. Interestingly, the authors identified activated neuroinflammatory processes such as microgliosis and astrogliosis already before neuronal death, leading to the hypothesis that disrupted mitochondrial protein synthesis signaled to trigger neuroinflammation. On the contrary, loss of DARS2 in oligodendrocytes did not cause neurodegeneration or neuroinflammation. 36 Importantly, these results argue that neurons and not glial cells are the main targets of DARS2 deficiency. Another study of the neuronal DARS2 model demonstrated induction of mtISR markers, although at a much lower level than in DARS2 heart. 37 It should be also noted that the bulk RNAseq analysis of brain samples does not reveal the cell type, which produced the signal. Finally, DARS2 was also knocked out in mouse cerebellar Purkinje cells and shown to be critical for the survival of that neuronal type, and protecting from cerebellar ataxia. 38 Interestingly, the removal of CLPP, the proteolytic subunit of the caseinolytic protease, which is a mitochondrial matrix protease, delayed the progression of OXPHOS deficiency and neurodegeneration in the neuronal DARS2 mouse models, 39 and partially rescued the DARS2 cardiomyopathy model. 40 In the neuronal models, the rescue mechanism was postulated to owe to the role of CLPP in the stabilization of OXPHOS complex I and the supercomplexes, which improved NAD+ metabolism. As aspartate has an important role in the regulation of NAD+/NADH through the malateaspartate shuttle, CLPP removal may indirectly correct a defect in the redox state caused by dysfunctional DARS2, at least partially.
To model hearing loss in Perrault syndrome, Xu et al. recently generated conditional HARS2 knockout mice targeted to cochlear hair cells. 41 Full-body knockout of HARS2 was embryonic lethal as expected. The hair cellspecific knockout mice showed progressive hearing loss, becoming deaf by the age of 60 days. Hair cells were lost by apoptosis, increased levels of reactive oxygen species were identified, and inner hair cell synaptic transmission was disrupted due to reduced calcium influx. The reduction of HARS2 protein level was demonstrated but its effect on mitochondrial protein synthesis was not addressed in this mouse model. Perrault syndrome is a combination of premature ovarian failure and hearing loss, which can be caused by mutations in HARS2 or LARS2, as well as in several other mitochondrial disease genes. 42 Mouse model carrying a HARS2 or LARS2 disease mutation or compound heterozygous variants would be of interest to investigate the mechanisms and cell type-specific alterations leading to Perrault syndrome. The closest mouse model resembling global reduced mtARS activity is the N-ethyl-N-nitrosourea (ENU)induced mouse mutant harboring a recessive WARS2 (mitochondrial tryptophanyl-tRNA synthetase) allele F I G U R E 2 Reported viable mouse models of mtARS defects. DARS2 and HARS2 mice are conditional knockouts targeted to specific tissues or cell types. WARS2 mice are hypomorphs from a mutagenesis screen, which have a variant affecting WARS2 gene splicing. V117L, which caused in-frame skipping of exon 3, still allowing the production of some full-length transcript. 43 As a result, WARS2 protein was nearly undetectable in all tested tissues. Remarkably, despite severe reduction in WARS2 amount, different tissues showed very different effects on mitochondrial protein synthesis as measured by OXPHOS subunit levels on western blot or by respiratory complex activities. For example, heart and liver showed severe defects in OXPHOS, whereas kidney and skeletal muscle were nearly unaffected. The authors concluded that the tissue-specificity of the respiratory chain dysfunction was caused by compensatory mitochondrial biogenesis in the spared tissues. The WARS2 mice had complex pathology with hearing loss, adipose tissue dysfunction, and hypertrophic cardiomyopathy. The authors concluded that a robust increase of p-eIF2a, indicative of ISR activation, was identified only in the heart, whereas FGF21 expression was found to increase in heart, skeletal muscle, and iWAT but not in any other tissue. UPRmt markers were not changed in any tissue. Although the WARS2 mutant mouse is not a direct model of human disease, because the WARS2 diseasecausing mutations are different and found in neurological phenotypes, it is an interesting model showing the complexity of mtARS regulation in different tissues.
Knockdowns of mtARSs have also been studied in other animal models, particularly in zebrafish and flies. The zebrafish models include FARS2, 25,44 RARS2,45 VARS2, 46 YARS2, 47 and WARS2 11 knockdowns, with phenotypes ranging from retarded embryonic development to brain hypoplasia, heart failure, and CNS and skeletal muscle involvement. Findings resembling those in mtARS mouse models have been reported in zebrafish models, including respiratory chain deficiency, ISR activation, and disrupted fatty acid oxidation. 46 Knockdowns and mutants in Drosophila melanogaster have been studied for FARS2, 48,49 MARS2, 50 SARS2 51 and WARS2. 50 The fly models have shown variably compromised developmental viability, motility, and tissue development and have had alterations in mitochondrial respiration, lactic acidosis and reactive oxygen species accumulation. 51 The zebrafish and fly models could be useful in assessing therapeutic approaches but have not so far been largely utilized. Finally, with some limitations, yeast mutant strains can be useful models in determining whether a human mtARS mutation is pathogenic. 52

| CONCLUSIONS AND FUTURE PERSPECTIVES
Although the basic function of mtARS enzymes is well known, the underlying disease mechanisms have been difficult to study. We still lack information on the biology of mtARS, which could help explain some of the associated phenotypes. Additional non-canonical functions are likely to exist, and those may have a role in some of the diseases. Questions such as how the mtARS function together and how they localize within mitochondria have only recently been addressed. 53,54 To enable the development and testing of treatment options, accurate genetic mouse models mimicking human disease mutations and phenotypes need to be generated and fully characterized. There is still much to learn about cell type-specific metabolic regulation and cell and tissue communication in these diseases.
Opportunities to investigate mtARS disease mutations in human models are now also available by using patient-specific induced pluripotent stem cells (iPSC) and their differentiation, which can particularly aid in studies of neuronal cell types that are otherwise difficult to reach. Brain organoids derived from iPSC may also provide useful models. One study investigating mitochondrial involvement in GARS1 disease, which affects both the cytosolic and mitochondrial glycyl-tRNA synthetases, compared induced neuronal progenitor cells from patients to those of carrier parents and unrelated controls. 55 Genome editing techniques now allow correcting the patient mutations in iPSC, which is important for producing isogenic controls for comparisons. Such comprehensive studies with edited controls are not yet published on mtARS diseases but are certainly on their way in several laboratories including ours.
Excitingly, treatment trials with specific amino acids in single patients who have cytosolic ARS mutations have been tolerated well and have shown beneficial effects. 56 Recently one child with FARS2 deficiency was treated with daily oral L-phenylalanine supplementation, and showed improvement in gross motor skills, movement abilities, and postural stability. 57 Reports of whether a similar strategy could be beneficial in other mtARS diseases are to be expected.

ACKNOWLEDGMENTS
Tyynismaa group, CureARS network, and MetaStem Centre of Excellence are acknowledged for support.

FUNDING INFORMATION
Financial support has been received from the Academy of Finland, University of Helsinki, and Sigrid Jusélius Foundation.

CONFLICT OF INTEREST STATEMENT
Henna Tyynismaa declares that she has no conflict of interest.

DATA AVAILABILITY STATEMENT
Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.

ETHICS STATEMENT
This article does not contain any studies with human or animal subjects performed by any of the authors.