Current progress with mammalian models of mitochondrial DNA disease

Mitochondrial disorders make up a large class of heritable diseases that cause a broad array of different human pathologies. They can affect many different organ systems, or display very specific tissue presentation, and can lead to illness either in childhood or later in life. While the over 1200 genes encoded in the nuclear DNA play an important role in human mitochondrial disease, it has been known for over 30 years that mutations of the mitochondria's own small, multicopy DNA chromosome (mtDNA) can lead to heritable human diseases. Unfortunately, animal mtDNA has resisted transgenic and directed genome editing technologies until quite recently. As such, animal models to aid in our understanding of these diseases, and to explore preclinical therapeutic research have been quite rare. This review will discuss the unusual properties of animal mitochondria that have hindered the generation of animal models. It will also discuss the existing mammalian models of human mtDNA disease, describe the methods employed in their generation, and will discuss recent advances in the targeting of DNA‐manipulating enzymes to the mitochondria and how these may be employed to generate new models.


| INTRODUCTION
Mitochondrial diseases constitute a range of heritable disorders that, through various mechanisms, end up leading to mitochondrial dysfunction. Despite all of these diseases leading to a similar cellular disorder, a staggering array of variable disease syndromes, tissue specificities, age of onset, and prognostic outcomes for the patient are known. 1,2 Further complicating the diagnosis and understanding of these disorders, the causative mutations can be due to mutations of both the maternally inherited mitochondrial DNA (mtDNA) itself, or by genes whose products function within the mitochondria but are encoded by the nuclear DNA. Current estimates are that 1100 to 1500 nuclear encoded genes 3 are translated by cytosolic ribosomes, and the protein products are then targeted to the various subcompartments of the mitochondria.
To express these genes, the mtDNA also encodes two of the rRNA subunits of their own ribosomes 5 and a minimized array of 22 tRNAs to carry out the translation 6 within the mitochondrial matrix. The mitochondrial genome shows evidence of the loss of these mtDNA encoded RNAs that are present in more ancestral-like mtDNA genomes. 7 For instance, animal mitochondrial ribosomes have evolved to utilize a tRNA in place of the 5S RNA, 8 they have also evolved a protein-only, multisubunit RNaseP enzyme 9 (EC 3.1.26.5), and they generally maintain only the minimal set of tRNAs to decode the modified mitochondrial genetic code. 4 For over 30 years, we have known that mutations of the small, 16.5 kb mitochondrial genome are also important causes of diseases. 10,11 Mutations by base changes, small indels or by large deletions or partial duplication of the mtDNA sequence are also known to lead to mitochondrial disorders. 2 The high copy number and lack of a coordinated mtDNA replication cycle leads to another unusual feature of mtDNA, where two or more variants of the mtDNA can coexist within the same organism, cell, or even mitochondrion in a condition known as heteroplasmy. 12 In most of the mtDNA derived disorders, a high relative proportion of the mutation is necessary to lead to the disease state (referred to as the mitochondrial threshold effect 13 ). In some cases, the disease alleles can be the only allele in the cell, tissue, or organ in a state referred to as homoplasmy, such as seen in some mitochondrial disease alleles leading to Leber Optic Atrophy (LHON, OMIM #535000).
While each specific disease is quite rare in the population, when summed together, frequency estimates for childhood mitochondrial diseases can be as high as 1:6700, 2 and up to 1:4300 in adults. 14 As with all rare diseases, the infrequency of patients leads to many difficulties in uncovering the molecular basis of disease and hampers recruitment in clinical trials. 15 The rarity of these disorders unfortunately also provides economic disincentives for corporate investment in preclinical and clinical trials. In addition, the striking tissue specificity in mitochondrial diseases leads to many instances where cell lines, through genetic manipulation or when derived from patient fibroblasts, can be aphenotypic, or not reveal relevant biochemistry that underlies the pathology of specific organ systems. Thus, the field has had a long interest in animal models to move forward our understanding of these disorders, and for preclinical research into interventions and cures.
Despite a long history of nuclear transgenesis, and recent advances in genome editing, animal mtDNA remains resistant to transgenic manipulation, and as such, there are only a handful of models with relevant disease mutations available. Fortunately, the mitochondrial research field has been quite resourceful, and a number of nondirect methods have been used to generate relevant animal models for research. This review will summarize these advances, and highlight the small but growing number of animal models with pathological mtDNA mutations.

| ANIMAL MODELS IN MITOCHONDRIAL DISEASE RESEARCH
The selection of the animal model always remains a topic of much discussion. Each model organism must be weighed by the research question at hand. There has been a long history of biomedical research using mouse models, and for the nuclear-encoded mitochondrial genes, these have been the models of choice in many previous studies (for reviews of models with nuclear genes involved in mitochondrial disease, see , a trend that will continue with the ease of CRISPR-based nuclear genome editing techniques. 21 As a mammalian model, mice are short lived, breed prolifically, have a long history of transgenic manipulation of the nuclear DNA and, from a purely phylogenetic perspective, share with rats the closest genetic relationship of a model organism to humans and other primates ( Figure 1). Yet this genetic closeness does not always result in the best model of a human disease.
For instance, a very clinically relevant mouse model of Leigh's syndrome generated by disrupting the Complex I NADH dehydrogenase-ubiquinone-FeS 4 (NDUFS4) gene leads to a mouse model displaying ataxia, encephalomyopathy, and other features that model the human disorder. 23 This model is now heavily used in preclinical research. [24][25][26][27] In contrast, mutations in the Surfeit Locus Protein 1 (SURF1) gene, a cytochrome c oxidase assembly factor, in patients also leads to a Leigh syndrome presentation with associated complex IV deficiency. 28,29 Yet a SURF1 knockout mouse line utilizing the loxP-cre recombination system showed little pathological phenotype and in one study, with knockout mice that actually out-lived their wild-type littermates. 30 This

Synopsis
To date, animal mitochondrial DNA has remained resistant to genome engineering technologies for the generation of animal models of these rare diseases; this review will discuss the currently available mammalian models for mtDNA disease, how they were generated, and future prospects for enhancing these resources.
was despite a 30% to 50% reduction in complex IV enzyme activity and decreased complex IV activity as assessed by histochemistry. 30,31 More recent work with these SURF1 knockout mice has shown that the mild complex IV deficiency activates various mitochondrial stress response pathways which appears to fully compensate for the genetic deficiency under normal laboratory conditions. 32 A recent experiment in pigs, where TALENs and CRISPR/Cas9 nucleases were used to generate SURF1 −/− cells then were used to generate pigs by Somatic Cell Nuclear Transfer technologies. In contrast to the mice, the pigs suffered from failure to thrive, muscle weakness and highly reduced life span, with elevated perinatal mortality, and neuropathological disruptions consistent with a delay in central nervous system development. 33 In this instance, the genetically more distantly related, yet larger and more anatomically similar pig better modeled the human pathology.
While this review will focus on mammals, and mostly mice, I wish to emphasize the important contributions of the classic invertebrate models, Drosophila melanogaster 34,35 and Caenorhabditis elegans and Caenorhabditis briggsae. 36 These powerful genetic models have begun to be commonly used to check for mitochondrial pathology of potential new disease alleles, and have strong potential in early high throughput screens for pathways of molecules in models of mitochondrial disease. [37][38][39] As for mtDNA mutants, C. elegans and C. briggsae have a long history in the study of mtDNA deletions mutants (reviewed in 36 ), and recent work with targeted restriction endonucleases has allowed the generation of Drosophila models with mtDNA mutations, including very versatile temperature sensitive mutations. 40,41 Another potential vertebrate model of mtDNA diseases is the zebrafish, as a number of studies of nuclear mitochondrial disease genes are underway. [42][43][44] Recent work on mtDNA mutations and the germline, and the manipulation of the mitochondrial replication machinery 42,43,45 portents the appearance of zebrafish mtDNA mutant models.

| TRANSGENIC MANIPULATION OF ANIMAL mtDNA-PROGRESS AND PROBLEMS
Unlike the mitochondria of plants, 46,47 yeast, and Chlamydomonas, 48 the mtDNA of animals has remained resistant to transfection. While we have long been able to move intact mitochondria between animal cells, 49 or into the germ cells of Drosophila 50 and mice, 51,52 transfection of animal mitochondria remains elusive. In plants, yeast and Chlamydomonas, the methods to introduce DNA into the mitochondria are membrane disrupting, such as the induction of temporary pores through electroporation, or the use of bio-ballistic "gene guns" to punch DNA coated projectiles into the mitochondria to deliver the transgene. In animals, these treatments are likely to release cytochrome c from the mitochondria, which F I G U R E 1 Representative cladogram of the genetic divergence in animals models mentioned in this review, which are used in mitochondrial disease research. At the nodes are estimated genetic divergence times (mya = million years ago), with confidence intervals in brackets extracted from http://www.timetree.org/ (June 28, 2020). 22 Branch lengths are only representative, and not to scale. Images from http://phylopic.org would be expected to induce the mitochondria-dependent apoptosis pathways 53 which ultimately lead to the destruction of the cell bearing the treated mitochondria. In cell culture, the reported successful microinjection techniques of King and Attardi, 54,55 viral-mediated membrane fusion of isolated mitocytoplasts, 56 and even bacterial conjugation methods 57 may have been able to limit cytochrome c release. Investigations into novel techniques, such as the photothermal nanoblade 58 or liposome-based methods such as the MITO-porter system 59 give hope that high-efficiency ways to introduce exogenous DNA into mitochondria are forthcoming.
However, these methods would still require the introduction of an entire replacement mtDNA, as a second obstacle remains to the integration of smaller transgenic oligonucleotides into the mtDNA itself. Plant, Chlamydomonas, and yeast systems have clearly defined mechanisms of mtDNA recombination which allow for the integration of the transgene. 48 However, recombination appears to be a rare phenomenon in animal mtDNA. 40,[60][61][62] The use of nuclease-driven genomeediting technologies currently in use to modify nuclear DNA would face specific problems. It is clear we can target functional nucleases into animal mitochondria, such as restriction enzymes 41,62-64 zinc finger nucleases (ZFN), 65 or transcription activator-like effector nucleases (TALEs). 66,67 One recent article also reveals that the CRISPR nucleases CAS9, but more efficiently CAS12a, can be imported into mammalian mitochondria in cells, and that the gRNAs colocalize with the mitochondria. 68 However, these authors concede that this gRNA localization may not represent mitochondrial matrix import, but only association with the mitochondrial membranes or the mitochondrial intermembrane space. History speaks to a number of cases where true mitochondrial matrix import was not actually achieved in attempts to import nucleic acids. 69 An earlier report of mitochondrial CRISPR manipulation does not define the mitochondrial subcompartment localization of the gRNA. 70 This leaves the possibility that the negative cell phenotypes observed may not have been due to the specific cutting of mtDNA but were overexpression artifacts, or may even represent reported RNA-independent DNA cleavage by the matrixtargeted CAS9. 71 Indeed, Antón et al report mtDNA damage in their CAS12a import experiments and dramatic mitochondrial dysfunction when importing CAS9 to the mitochondria. 68 Regardless of the efficacy of importing protein-only or protein-RNA nucleases, mammalian mitochondria appear to lack the nonhomologous end joining, microhomologymediated end joining, or homology-directed repair DNA repair pathways that lead to genome editing in the nucleus. 72 Double strand breaks in mammalian mtDNA have been shown to be rapidly degraded, with only rare recircularization of the mtDNA to form deletion-bearing mtDNAs. 62,73 These results were in agreement with earlier work on the mouse, which found that germline transmitted heteroplasmic mice rarely, if ever, recombine, implying a similar lack of recombination in the mouse germline, limiting the utility of these pathways in transgenesis. 60,61 In contrast, transient expression of the mitochondrially targeted PstI to adult mouse neurons did demonstrate that these double-strand breaks were rejoined to form canonical mtDNA deletions. 74 Also in Drosophila, the targeting of restriction enzymes has been successfully used to induce recombination in heteroplasmic flies, 41 albeit at a very low rate and under a strong selective regime to recover the target recombinant mtDNA. Also, human cell lines heteroplasmic for the pathogenic mt-TL1 A3243G and mt-TI A4269G mutation, did display recombinant mtDNAs 61 after 9 months of culture. 61 These tissue-specific differences in these results warrant further study and characterization, as the ability to induce neuronal-like recombination within the germline may help in the generation of models bearing mtDNA deletions, and would be valuable for inducing yeast-like recombination for targeted manipulation of the mtDNA.
A major conceptual breakthrough has recently emerged, inspired by the CRISPR "base editor" strategies. 21 In a classic case of unrelated research leading to a breakthrough for another field, researchers characterized a bacterial toxin deaminase (DddA) from the Gramnegative bacterium Burkholderia cenocepacia, which deaminated cytidines to uracil, leading to a C:G > T:A mutations at TC or TCC sites. Unlike other cytidine deaminases such as apolipoprotein B mRNA editing catalytic polypeptide-like (APOBEC) or activation-induced cytidine deaminase proteins, dddA has specificity to double-stranded DNA and undetectable activity on RNA. 75 Fortunately, mtDNA continues to replicate in postmitotic tissues outside of the cell cycle, 12 allowing replication-associated mutations from C > U > T before the U is repaired in the mtDNA. This was demonstrated earlier when APOBEC1 had been targeted to the mitochondria of living flies, leading to robust C:G > T:A mutagenesis throughout the mtDNA without the need to induce strand breaks or limit base-excision repair systems. 76 To lead to specific editing, the team split the DddA protein into two parts, combining them with two TALE sequences, which would bring the domains together and limit the deamination to the space between the two TALE recognition sequences (dubbing the constructs DdCBEs). A similar strategy had been employed to induce ZFN-directed methylation of mtDNA, 77 and to direct the paired FokI domains in mitochondrially targeted TALE or ZFN nucleases 65,66 to limit nuclease activity to specific mtDNA target sequences. Uracil glycosylase inhibitor sequences can also be included to inhibit repair of the uracil bases after targeting, and was shown to increase the editing activity with some constructs. When expressed in human cell lines, the more optimal DdCBE constructs induced up to 40% to 49% editing efficiency for the targeted base. Though all TC or TCC sequences between the TALE recognition sequences could be edited, outside of this region, off-target mutation screening found that other mutations were undetectable above background heteroplasmy levels for most of the TALE recognition sequences, demonstrating that good TALE specificity is the critical aspect of the specific editing in this system. 75 This impressive work finally allows for targeting mutations to the 13 to 14% of the sites represented by the modifiable TC or TCC motifs with the mouse and human mtDNAs, respectively. This is particularly exciting for those using another developing model for the study of mitochondrial disease, the induced pluripotent stem cells (iPSCs) or organoid systems. 78,79 The ability to engineer mtDNA mutations into the cells would remove the necessity of obtaining the cells from patients bearing these rare diseases, and would allow for mutant and control cells or organoids to be studied under identical nuclear genetic backgrounds. For iPSCs from nucleusencoded disease mutations, such an editing strategy may also supply a means to remove unwanted mtDNA mutations that can arise during the development of iPSCs. 80,81 This technology will undoubtedly open the doors to the generation of targeted animal models, or in the reversion of pathogenic mutations back to wild-type alleles. However, of the 93 pathogenic human mtDNA mutations which are classified as "confirmed" disease causing in the MitoMap server 82 (accessed July 2020) only 11 involve TC, TCC, GA, or GGA sites. There may be further limitations possible in the ability to generate appropriate TALE (or ZFN) nucleases for the construct, or the presence of other mutable bases in the vicinity. Due to variations in the mtDNAs, 13 of these human mtDNA mutations could be generated in analogous positions in the mouse mtDNA (Table S1). An expanded mutagenic toolkit to target different sequence contexts and different bases will be required to generate analogues of the common MELAS (OMIM #540000) and MERRF (OMIM #54500) alleles (mt-L1 3243A > G and mt-TK 8344A > G) or the Maternally Inherited Leigh Syndrome (#256000)/NARP (#551500) allele (mt-ATP6 8993 T > G). I would also caution against limiting the animal models to only direct sequence analogues of human mtDNA disease mutations, as species-specific differences are to be expected as to whether a specific substitution is pathogenic, or the relative severity of a mutation in another species. A striking example is the lack of sequence conservation in the TψC loop of the mt-TK which harbors the 8344A > G MERRF allele, making the design of the direct analog of the MERRF mutation impractical (see alignment at Reference 83).
As introduced previously, the overwhelming majority of mitochondrial proteins are encoded by the nuclear DNA, cytosolicly translated, and imported into the various mitochondrial subcompartments. Thus, a longstanding question has remained as to why the mitochondria maintain an independent chromosome, 84,85 and why could we not just express recoded variants of the mtDNA-encoded proteins targeted to the mitochondria to produce animal models or to use as therapies? Such allotopic expression experiments have been reported, and remain quite controversial. There is one report of nuclear expression of mouse wild-type mt-ATP6 and a variant with an analogous Leigh syndrome/NARP 8993 T > G (L156R) mutation. 86 Biochemically, the mice were indistinguishable from nontransgenic controls, but the mice expressing the allotopic ATP6 with the 8993 T > G mutation displayed poorer performance on a number of behavioral and physiological tests. More unusually, two independent rat models were reported where the human proteins derived from the LHON mt-ND4 G11778A (R340H) mutation were transfected into the eyes, where they were reported to lead to symptoms consistent with LHON pathology. [87][88][89] The authors also reported being able to reverse the damage by a second round of transfection with the wild-type, human allotropic ND4. 88,89 These results have led to clinical trials in human patients. 90 However, the claims have met with much skepticism. In the case of the rat work, is seems difficult to reconcile the apparent uptake of the human ND4 protein into the rat complex I, as cross-species mito-nuclear incompatibilities have often been reported, [91][92][93][94] and the within species rapid coevolution of the nuclear and mitochondrial components of the respiratory complexes should provide a significant obstacle to cross-species complementation. 95 Other work in cell culture has repeatedly failed to demonstrate that the allotopically expressed proteins are properly imported into the mitochondria, or that they are assembled into the respiratory chain complexes. [96][97][98] The mitochondrially encoded proteins are very hydrophobic, and the mitochondrial ribosome has been demonstrated to interact directly with the inner membrane to cotranslationally insert the nascent polypeptides. 99,100 It is unclear how the matrix import machinery would import these hydrophobic polypeptides and insert them in the correct orientation in the inner membrane. These concerns are supported by reports of polypeptide aggregates and failed incorporation of allotopically expressed proteins. 96-98

| Indirect mutagenesis of mtDNA
To date, mutagenesis of the mtDNA in animals has been accomplished through indirect methods. The earliest examples were derived from the spontaneous appearance of deleterious mtDNA mutations in cell culture 101,102 or from aged mouse tissues. 103 Early manipulation of mtDNA in cell culture employed a combination of 4,5 0 ,8-trimethylpsoralen and UV light exposure. [104][105][106] Such cells have been used to transfer the mutants to embryonic stem (ES) cells for injection into oocytes to generate mouse models (see Section 2.2).
However, mammalian mtDNA has been shown to be remarkably resistant to chemical mutagenesis in the living mouse. One recent study treating live mice with the chemical mutagens benzo[a]pyrene and N-ethyl-Nnitrosourea showed the extensive mutagenesis of the nuclear DNA, and the presence of the expected chemical adducts on the mtDNA. Despite this, the mtDNA mutation rate in these mice remained indistinguishable from the untreated mtDNA mutation rate. 107 In vivo mtDNA mutagenesis can be achieved through genetic means, through the use of manipulated components of the mitochondrial replication machinery. The first such mouse was the mtDNA mutator mice, which bears an exonuclease activity-disrupting D257A mutation in the endogenous locus of the catalytic subunit of the mitochondrial DNA polymerase, polymerase gamma (PolG). Reports in the literature show this mouse having been independently generated in at least three labs. [108][109][110] Phenotypically, all three display very similar strong progeria phenotypes and early death, despite being maintained in a number of different lab environments, and being maintained on different nuclear genetic background strains. 111 From the outset, the mice were observed to accumulate high levels of mtDNA point mutations and small indels, 108,109 which are transmissible down the female germline, and thus a source of indirect mtDNA mutagenesis. 112 A number of other structural rearrangement mutations are known in the mtDNA mutator mice. The mice also harbor an unusual truncated, linear, double stranded mtDNA molecule, which encodes the major arc sequences between the two origins of replication. 108,113 To date this linear fragment accumulation has only been seen in the presence of the exonuclease deficient POLG 114,115 , in homozygous larvae of the analogous PolG mutation in Drosphila 116 or through disrupted 5 0 -3 0 exonuclease MGME1 in human patients 114,117 or the Mgme1-knockout mouse model. 118 The persistence of these linear molecules can be explained by the compromised mtDNA degradation activity in the absence of MGME1 or functional PolG exonuclease activity. 114,115,119 Excess replication stalling by the polymerase is also reported in mtDNA mutator mice. 113 In addition, brain and heart 120 were found to contain duplicated control region sequences, dubbed control region multimers. 120 Similar to exo-deficient polymerases used in cycle sequencing reactions, POLG D257A acquires a strand-displacement activity, suggesting an alternative, replication-mediated model for the formation of the linear mtDNA fragments and control region multimers. 121 There are also reports of classic mtDNA deletions in the mtDNA mutator mice, 122 and an increase in deletions when a mitochondrially targeted restriction endonuclease and the POLG D257A are coexpressed. 115 mtDNA deletions in human patients can often be caused by other mutations in human PolG, or by mutations in the mitochondria's replication-associated helicase, TWINKLE. 2 Yet the levels remain much lower than the mtDNA deletion inducing models, such as the TWINKLE mutant "deletor" mice 123,124 or even the MGME1 knockout mouse. 118 These rearranged mtDNAs, including the canonical mtDNA deletions, have not been demonstrated to be transmitted through the mouse germline, limiting the available mouse models bearing mtDNA deletions.
These replication machinery mutants may be employed in cell culture to isolate mutations of interest. One group has been able to manipulate the expression levels of the exonuclease deficient POLG in order to limit mutagenesis to only 1one to two mutations per mtDNA to isolate pathogenic mtDNA point mutations within cell culture, resulting in 812 mtDNA mutations being identified in the study. 125 Similar manipulations with TWIN-KLE or MGME1 cell lines may produce cell lines bearing mtDNA deletions that are of interest. A similar concept used the platelets of mtDNA mutator mice to create cybrids to capture and concentrate mtDNA mutations. 126 Using heterozygous mtDNA mutator mice would reduce the mutation rate, and may generate cybrids with only a few mtDNA mutations, which would ease the correlation of phenotype to specific mutations in the downstream studies. These mtDNA mutation-bearing cell lines could be transferred into ES cells for implantation in mouse embryos (see Section 2.2).

| Introducing mtDNA into embryos
The first attempts at generating heteroplasmic mice utilized mitochondria bearing mtDNA mutations leading to chloramphenicol resistance (CAP R ) in teratocarcinoma and melanoma cell lines. Cybrids were derived and injected into blastocysts, where they generated mosaic animals. 127 However, transmission of these mtDNAs to further generations was not documented and subsequent attempts to generate heteroplasmic CAP R mice also did not result in a stable mouse line for analysis. [128][129][130] The first models of heteroplasmic mtDNA transmission in transmitochondrial mice were generated by isolating cytoplasts from one mtDNA haplogroup, injecting and electrofusing these to the cells of an embryo carrying another mtDNA haplogroup, and then screening the progeny for heteroplasmic pups to produce heteroplasmic lines stably transmitting the mt-NZB and mt-BalbC mtDNA haplogroups. 51 Shortly thereafter, karyoplast transfers from one mouse strain's oocyte into another led to the generation of C57Bl/6 and NZB heteroplasmic mice 52,131 (Figure 2A). Direct injection of mouse mitochondria isolated from nonsperm sources, into the oocyte also appears to be a potential way to introduce intact mitochondria bearing mtDNA into mice, but has not yet resulted in a disease model. [132][133][134] 2.1.1 | ΔmtDNA 4696 "Mito-mice" This cytoplast electrofusion method was employed by the Hayashi group to generate the ΔmtDNA 4696 "mito-mice," the first heteroplasmic mouse model with a pathogenic F I G U R E 2 Three methods to generate mouse models with mtDNA mutations. A, Cytoplasts containing the desired mtDNA are generated to transfer mitochondria, but not the nucleus, to the embryo. Electrofusion merges some of the cytoplasm with cells of an early embryo, introducing the mtDNA, which can give rise to a chimeric founder (see Section 2.1). B, Cybrid embryonic stem (ES) cells are generated in cell culture to introduce the desired mtDNA through cytoplasmic fusion. The resulting ES cells can be homoplasmic or heteroplasmic, depending on whether ES cells are pretreated to remove their own mtDNA. ES cells are introduced into an early embryo to create chimeras (see Section 2.2). A + B, Female offspring are screened to determine if they are transmitting the desired mtDNA variants, and the correct founders are bred to create maternal lines of the mice. C, Breeding/screening methodology using female PolG D257A/WT females to transmit mtDNA to their offspring. After breeding N2 females to secure the female line, the mothers are sacrificed and their colons are screened for the presence of mitochondrial deficiency and the cosegregating mtDNAs are identified by sequencing of lasercaptured colonic crypts (cartoon sequence traces showing a heteroplasmic C > T mtDNA mutation are illustrated below the histology section containing representative COX-deficient colonic crypts). Selected females are then bred to establish the heteroplasmic lines transmitting the desired mtDNA mutations mtDNA mutation. 103 Synaptosomal fragments from mouse brains bearing mitochondria and mtDNA were isolated and then fused into a ρ 0 mouse cell line. One cell line bearing a 4696 bp deletion breakpoint (positions mt-TK 7759 to mt-ND5 12 454) was selected to generate cytoplasts to fuse into mouse embryos (Figure 2A). While the first three generations of mice showed evidence of the mtDNA deletion, a 27.9 kb partially duplicated species in addition to the WT mtDNA sequence, the 27.9 kb molecule was reported to be lost during the early generations. Mice displayed decreased complex IV activity, and had a pronounced pathology of the kidneys. Surprisingly, the mice faithfully transmit this mtDNA deletion, despite this being an exceedingly rare phenomenon in humans. 135

| Manipulated via embryonic stem cells
In a refinement of the aforementioned method, mtDNA from a target cell line has been transferred into chromosomally normal ES cells, which were then injected into mouse oocytes. These experiments lead to chimeric founder animals, and the pups are screened in order to identify germline transmission of the ES cell, and the desired mtDNA variants in a manner similar to established mouse nuclear transgenesis techniques. 136 This technique has been used to generate a number of relevant mouse models with mtDNA mutations ( Figure 2B).

| mt-COI T6589C (V421A) mice
The first mouse line was homoplasmic for a T6589C mutation in mt-COI, leading to a V421A amino acid substitution. 137 This mtDNA mutation was identified in a B82COI M cell culture, where the cells were found to exhibit half of the expected mitochondrial complex IV activity. ES cells depleted of their own mtDNA through rhodamine-6G treatment were fused to enucleated B82COI M cells to introduce the mtDNA mutation of interest. These ES cells were microinjected into eightcell-stage mouse embryos to generate the chimeric founder mice. The pups of the female chimeras were bred and screened for pups who had inherited the T6589C-bearing B82COI M mitochondria, to found this mouse line ( Figure 2B). The authors reported that the mouse hearts showed reduced complex IV activity through COX histochemistry, increased blood lactate levels after glucose loading, and a slightly reduced body mass over 18 weeks. In a later study, the mice showed no lifespan differences compared to wild-type mice. 138 Overall, the phenotype of the mouse was quite mild when compared to the ΔmtDNA 4696 mito-mice.
An mtDNA chromosome from LA9 cells, homoplasmic for both a 13885inC frame-shift mutation in mt-ND6 and the same mt-COI T6589C mutation described above was utilized to independently generate the same animal model in a second laboratory. One ES cell line in which 4% of the mtDNA contained a spontaneous deletion (13885insCdelT) which repaired the mt-ND6 frameshift mutation led to the successful generation of ES cell cybrids and eventually a single female chimeric mouse after introduction into blastocysts. 101 Intriguingly, the mice showed a profound selection against the 13885insC + T6589C mtDNA, in favor of the 13885insCdelT + T6589C. In the homosplamic state, the T6589C mutation (in concert with 0-14% of the 13885insC) also led to decreases in complex IV activity, but showed evidence of ragged red fibers in the skeletal muscle, cardiomyopathy, and abnormal mitochondrial morphology in the heart and muscle at 1 year of age despite outwardly appearing no different from age-matched control mice. 101 The apparent differences between the two lines of mice bearing the same mutation could be simply a matter of the differences in ages used, or by the exact housing condition of the mice used by the two groups in the respective papers. However, differences in the exact mouse strain used, 111 the precise mtDNA haplogroup on which the mutation is carried, 139 or the presence of the low levels of the 13885insC mt-ND6 frameshift mutation 101 may all play a role in the divergent phenotypes reported.

| mt-ND6 G13997A (P25L) mice
The next model was generated from a complex I deficient metastatic lung carcinoma cells bearing a G13997A mutation in mt-ND6. 140 This substitution mirrored the human disease allele G14600A (P25L) which leads to Leigh syndrome with optic atrophy 141 or Leigh Syndrome with sensory neural deafness. 142 By 3 months of age, the mice showed evidence of impaired complex I + III activity, but normal complex II + III and IV activities, in line with an isolated complex I deficiency. Blood lactate was again elevated after glucose loading, and there were elevated H 2 O 2 levels detected specifically in the bone marrow. Histology was carried out on the optic nerves and retina and no evidence of optic atrophy was observed. 140 A later study on these mice at more advanced age found no significant differences in lifespans and no stark advancement of mitochondrial pathology, but the mice developed a glucose intolerance that was reported to be mitigated with treatment by the antioxidant N-acetylcysteine. 138 The mice were seen to develop hematopoietic tumors, especially B-cell lymphomas in the spleen, liver, and lungs at frequencies ×7.5 higher than the age matched controls, and these tumors appeared to have high propensity for malignancy. 143 A second strain was generated by the Wallace lab, by psoralen/UV treatment of LMTK − cells. In their studies, they were able to detect swelling and then degeneration of the smaller retinal ganglion cells over the window of 14 to 24 months. 106 More mitochondria, and those with abnormal morphology, were observed in the axons of the optic nerves and the mice showed signs of vision problems. The lab was also able to confirm an isolated complex I deficiency, and elevated ROS levels in the mice, but did not report the tumorigenesis phenotype in their mice.
The nuclear-encoded mitochondrial Adenine Nucleotide Translocator 1 gene (ANT1 or Slc25a4) is important to mitochondrial function by exchanging matrix ATP derived from mitochondrial complex V with ADP in the cytosol. A study of two ANT1 ko/ko strains, bearing either the mt-ND6(P25L), or the mt-COI (V421A) mtDNA showed that the ANT1 ko/ko ; mt-ND6(P25L) combined mutations lead to an enhanced cardiomyopathy phenotype and shortened the lifespans of the mice. In contrast, the ANT KO/KO mt-COI(V421A) did not lead to a similar additive effect on the phenotype of the mice. 144 Intriguingly, a similar observation was made in a line of Bcs1l S78G knock-in mice for the mitochondrial inner membrane AAA-family translocase BCS1L, a factor important for the Rieske iron-sulfur protein, UQCRFS1 subunit's assembly into complex III. A homoplasmic, normally neutral haplogroup variant in mt-Cytb G14904A (D245N), when introduced into the Bcs1l S78G colony, led to a strong increase in the severity of the phenotype in this model. 145 These studies again point to the importance of mito-nuclear epistasis effects, and the importance of understanding the exact strain of mice used, 111 and additionally, knowledge of the precise mitochondrial haplogroup involved in your mouse models of mitochondrial disorders.

| mt-TK G7731A mice
A mt-TK G7731A mutation was found and concentrated from P29 cell lines before introduction into ES cell cybrids. This mutation would disrupt base pairing in the anticodon stem of the tRNA Lys, and aligned to the known mt-TK G8328A mutation in humans. 146,147 This early study reported that 4-month-old mice with >76% relative levels of the mutation were slightly smaller than their low-level counterparts or wild-type controls, that there was a general reduction of mitochondrial respiratory chain activity in the kidney and skeletal muscle, and the grip strength of the mice was reduced. 102 Mice aged to 26 months retained these phenotypes, but began to show markers of dysfunction in the blood, including reduced hematocrit, hyperglycemia, and elevated blood urea nitrogen levels. 143 Muscle atrophy, but the absence of ragged-red fibers, and renal failure were also discovered in postmortem analyses.

| Breeding and selecting for mutations in POLG D257A/WT lineages
Our method differs from other successful methods in producing mice with mtDNA mutations in that there is no embryo manipulation. Earlier, we had attempted to use mtDNA mutator mouse females to transmit mtDNA point mutations and small indel mutations through female lineages. 112 We found that these mice contained far too many mtDNA mutations to lead to a clear genotype-phenotype association between any specific mutations. For instance, a mouse bearing a heteroplasmic mt-TM 3875delC and homoplasmic mt-tC T5425C mutations also bore 7 rRNA mutations, 3 additional tRNA mutations, a mutation in the OriL region, and 23 protein coding gene mutations (10 synonymous and 13 nonsynonymous). 148 Despite the deletion-bearing tRNA Met mutation, reduced tRNA Cys levels, and a germline selection limiting mice to <87%, no evidence of respiratory chain dysfunction was observed.
Further work on the mtDNA mutator mice revealed that the heterozygous females were also mutagenic, and that successive breeding of heterozygous females led to mutation accumulation, the eventual loss of fecundity, impairments in brain development 149 and decreased lifespan of POLG WT animals that had accumulated these mtDNA mutations. 150 Based on the observed mutation rate in WT siblings of mtDNA mutator mice, we estimated that heterozygous mtDNA females with no prior maternal exposure to the POLG D257A allele would induce on average two to three mtDNA mutations per molecule that transmitted through the female germline. A later collaboration on heterozygous mtDNA mutator mice and the accumulation of different clonally expanded point mutations within each of their colonic crypts lead us to an in vivo method to prescreen mouse lineages of interest. 151 Using COX/SDH histochemistry to screen for individual colonic crypts displaying mitochondrial dysfunction, and using laser-capture microdissection to sequence the individual crypts, we were able to identify female lineages that were transmitting pathogenic mtDNA mutations, and identify the specific mutations cosegregating with the dysfunction while the mice still harbored subthreshold levels of the mtDNA mutation.
Thus, the mtDNA mutations from the POLG D257A/WT female are transmitted via the oocytes to the firstgeneration backcross females (N1 animals), who are selected if they lack the POLG D257A allele. These females are again backcrossed to produce an N2 generation of females bearing and segregating the mtDNA mutations from their mother. Once the N3 litters have produced females to stabilize the female line, the N2 mothers are sacrificed, and the colonic crypts screened for the presence of crypts bearing complex IV deficiency, as indicated by COX/SDH dual staining histochemisty, indicating the lines of interest for further breeding an characterization. 152 COX deficient crypts are then laser-capture micro-dissected and the mtDNA amplified and sequenced to determine which mtDNA mutations are cosegregating with the mitochondrial deficiency ( Figure 2C).

| mt-TA C5024T mice
The first line derived by this method was found to contain two linked mtDNA mutations; mt-TA C5024T mutation, and mt-ND6 C13715T (G119D). While no strong complex I phenotype was observed, and the G119D mutation sits in a poorly conserved region of mt-ND6, it is impossible to rule out a contributing role of this mutation to the phenotype of the mice. 152 The C5024T mutation disrupts the same aminoacyl-stem basepairing as the human mt-TA G5650A mutation found in a patient suffering from a strict mitochondrial myopathy phenotype, 153 and in a complex case of CADASIL (OMIM #125310) where the patient had both a NOTCH3 mutation commonly associated with CADASIL and also the mt-TA G5650A mutation. 154 In the mice, steady-state levels of the tRNA Ala were reduced in animals bearing higher relative levels of the mutation, eventually leading to decreased capacity in in organello translation assays performed on mitochondria isolated from animal tissues. 152 Mice retained less body fat than age-matched controls, and the heart mass was elevated, and showed elevated levels of the cardiomyopathy markers ANF and ATF5. 155 Beyond the colonic crypts, mosaic complex IV deficiency was originally observed in the heart and colonic smooth muscle and very rarely in the skeletal muscle. 152 However, our continued work has revealed that in mice above 75% relative levels of the mutation, the liver, kidneys, small intestine, choroid plexus of the brain ventricles, and the ocular muscles show signs of complex IV defects using NBTx histochemistry. 31 The phenotype differs substantially from the strict myopathy in the young mt-TA G5650A patient, but is surprisingly reminiscent of the mother in that study, who had high levels of the mutation and strikingly high levels of muscle complex IV deficiency, but had not succumbed to a disease phenotype at the time of the report. 153 The scheme has also produced another, currently unpublished mouse line that we are currently characterizing.

| Naturally occurring mtDNA mutations
While this method has not led to the isolation of models of pathogenic mtDNA mutations in mice, spontaneous mutations have been studied in rats, 156,157 and veterinary practice has identified dogs with mitochondrial disease phenotypes. [158][159][160] A molecular diagnosis of a mt-CYTB G14474A (V98M) mutation was made in Australian Cattle dogs and Shetland Sheepdogs, which lead to a spongiform leukoencephalomyelopathy in the more severely afflicted dogs. 161 More recently, Golden Retrievers from a maternal lineage were identified with sensory ataxic neuropathy. 162,163 The animals were found to bear the small deletion mt-TY 5304delA, which would disrupt the first basepairing after the variable loop, TψC stem of the tRNA, leading to an unstable tRNA with reduced steady-state levels. All afflicted dogs had >88% of this mutation, while animals with lower relative levels remained aphenotypic. ATP production rates and respiratory chain enzyme activities were reduced in the animals, and COX/SDH histochemisty in the skeletal muscle showed a generalized reduction of COX staining, and bluish color due to the complex II counterstaining. 162 Generally, these mutations have led to severe disability in animals bearing high relative levels of the mutation. Ongoing contact with veterinary medicine practitioners that observe classic mitochondrial phenotypes in pets and domestic animals may allow for unique collaborations, and expedite preclinical research. Such collaborations could be of great interest to pet owners while simultaneously being of help to human patients in expediting research and safety assessments into experimental treatments.

| PRECLINICAL WORK WITH MTDNA ANIMAL MODELS
These mtDNA mouse models generated to date (Table 1) have raised considerable interest, not only in allowing deeper investigations into the molecular mechanisms underlying the pathophysiology of mtDNA disease, but they have also allowed for preclinical investigations into potential therapies to treat human mtDNA disease patients. In 2005, the Hayashi group used the ΔmtDNA4696 "Mito-mice" to demonstrate that nuclear transplantation from these mice to an enucleated oocyte lacking the mtDNA deletion was possible, and protected the resulting pups from the pathology of these "Mito-mice." 166 These early results appear to have inspired the human mitochondrial replacement therapies, which are now being refined for families transmitting mtDNA disease. [167][168][169][170] One potential method to help patients with heteroplasmic mtDNA mutations would be to employ methods that would favor an increase in the relative amount of wild-type mtDNA while decreasing that of mutant mtDNAs within the cells. As the amount of mutant mtDNA is reduced below the biochemical threshold for the given mutation, one would expect the mitochondrial network within the cell to regain proper function and rescue the cellular phenotype. Early work on mtDNA NZB/BalbC heteroplasmic mice (see Section 2.1) showed that the targeted degradation of one of the mtDNA haplogroups with restriction endonucleases allowed for the selective removal of one mtDNA type. The cells were able to re-populate their mtDNA with the untargeted mtDNA haplotype, thereby protecting the cell from mtDNA depletion. 63,171,172 Further studies using mitochondrially targeted TALE nucleases also successfully reduced the amount of the targeted mtNZB in mouse oocytes or early embryos. 173 In back-to-back studies, the heteroplasmic mt-TA C5024T mice (Section 2.3.1) were used to further investigate these methods. 65,66 Using the muscle-specific adenoassociated virus 9 (AAV9), the Moraes group were able to deliver the mitoTALEN constructs to the heart and skeletal muscle. This strategy utilized a pair of TALE recognition sites, where one mitoTALE would bind to both WT and mutant mtDNA molecules, but the other would only bind to the C5024T mutation. This strategy ensures the cleavage of only the mutation-bearing mtDNA. The treatment led to a relative reduction of the C5024T allele without resulting in a pathogenic mtDNA depletion. Importantly, we were able to confirm a restoration of the tRNA Ala levels within the mitochondria of the treated tissues, confirming an improvement in mitochondrial expression as a result of the treatment. 66 The Minczuk group used AAV9.45 with their mitoZFN constructs to also reduce the levels of the mtDNA mutation, rescue the tRNA Ala steady-state levels, and found metabolomics signatures consistent with increased oxidative respiration with decreased dependence on glycolysis. 65 At present, these methods rely on AAV-mediated gene therapy and studies into peptide delivery are still lacking. While gene therapy approaches are often complicated to extend to human treatments, work on delivering the mitoZNF to oocytes is underway 174 which may potentially decrease the amount of the pathogenic mtDNA transmitted. Such a protocol could be coupled with in vitro fertilization techniques, or be combined with mitochondrial replacement therapies to limit the carryover of mutant mtDNA during these treatments. Aside from nucleases, other methods to selectively shift heteroplasmy are being explored. For instance, the use of small molecules that bind to mutations that produce G-quadraplex sequences have been shown in cell culture to selectively reduce the amount of the mutant mtDNA by inhibiting its replication. 175 Unfortunately, no animal models with appropriate mutations are available to test this method in an in vivo study.
Instead of shifting mtDNA heteroplasmy, another strategy has been to alter mtDNA copy number within the cells. In an illustrative example, a pathogenic mitochondrial threshold of 81% in a cell with 100 mtDNA would predict that the 20 wild-type mtDNAs would be sufficient to maintain normal cellular function. By doubling the mtDNA content of the cell, levels as high as 90% would still provide the required 20 wild-type copies and maintain healthy mitochondrial function within the cell. 176 It has been long known that manipulations of the TFAM gene, which serves a dual role as a component of the mitochondrial transcription initiation complex, 177 and acts as a histone-like protein which binds to and compacts mtDNA, 178,179 alters the cellular mtDNA copy number. 180 TFAM heterozygous knockout mice are viable, with approximately half the normal mtDNA copy number, 181 and overexpression of TFAM in the mouse leads to a 150% -250% increase of the mtDNA copy number. 182,183 Comparing the phenotypes of the standard mt-TA C5024T mice and the mt-TA C5024T mice with both TFAM overexpression, and TFAM heterozygous knockout nuclear backgrounds supports this hypothesis. Mice with similar relative levels of the C5024T mutation showed improvements under TFAM overexpression, such as a recovery of body mass, suppression of the cardiac enlargement and cardiomyopathy markers, reduction of COX-deficient fibers, and a rescue of enzyme activity defects. 155 While genetic overexpression of TFAM is most likely not feasible in patients, the authors point out that mitochondrial biogenesis is a process that has responded to small molecules, and may offer an avenue for the development of drug therapies for heteroplasmic mitochondrial diseases. Indeed, a few candidates such as bezafibrate, resveratrol and retinoic acid are already in use in experimental settings, but with contradictory results (reviewed in Reference 184).

| CONCLUSIONS
Progress is being made in our ability to generate mammalian models with pathogenic mtDNA mutations. To date, the mouse models of mtDNA disease have either been generated by introducing pathogenic mtDNA mutations within mitochondria into mouse embryos, introducing them into ES cells which are implanted into embryos, or by the generation and selection of mtDNA mutations produced by mtDNA mutator mice. To date, one striking difference between the mouse models and patients with mtDNA mutations has been a consistent trend towards a less severe phenotype in the mice. As mentioned earlier, some of the nuclear gene mouse models have also shown a rather mild phenotype in comparison to the human patients they were meant to model, such as the SURF1 knockout, 28,29 or the TWINKLE deletor mice. 123 The disease phenotypes observed in the NDUFS4 knockout model of Leigh Syndrome does show that mice can, under the correct conditions, recapitulate human disease to some degree. 23 While some of this is likely due to differences in tissue specific metabolic demands between the species, another source of this variation may be the mechanism of mitochondrial biogenesis. The relaxed replication model of mtDNA, with its clonal expansion and vegetative segregation, is thought to lead to strong cell-by-cell mosaicism with respect to mtDNA mutations. 12 While the human and mouse embryos do not start out with vastly different numbers of mtDNA inherited in the oocyte, humans are regarded to have substantially more cells. For instance, a human brain is estimated to contain 1200 times more neuronal cells than that of a mouse. 185 This drastically expanded amount of cell division and mitochondrial replication in human development could lead to much greater variation in the relative levels of the mtDNA mutation in a largely stochastic system. With more cells sampling these effects, a human will be far more likely to have cells that cross the mitochondrial threshold effect. This in turn is hypothesized to induce compensatory biogenesis mechanisms to maintain the cell's mitochondrial function. However, this corrective function will eventually become self-defeating as the wild-type mtDNA molecules to copy become more rare, and thus eventually fail to compensate for the disrupted mitochondrial function. 12 This phenomenon has been observed in patients with mtDNA disease. 186 Thus, the smaller size and cell content of a mouse may inherently protect the mouse from some of the severe consequences of mtDNA disease.
A similar trend has been noted in aging studies with mouse models of mitochondrial dysfunction, as mice normally show little COX deficiency in their colons at the end of their life, a phenomenon quite common in aged humans. 187 Yet heterozygous mtDNA mutator mice, with their ×10 increase in the mtDNA mutation load, 149 do recapitulate the aged human levels of colonic COX deficiency in advanced age. 151 Perhaps mouse models of human mtDNA diseases will require the use of compounded mutations within a mitochondrial pathway, such as the combined ANT1 ko/ko ; mt-ND6(P25L) 144 or the mt-Cytb (D245N); Bcs1l S78G models, 145 if a strong pathological phenotype is required.
Besides these current methods to generate the mouse models, we hope to see an expansion of other methods to generate these models. The use of mtDNA mutator cell lines, 125 coupled to existing embryo manipulation techniques holds promise. More recently, targeted genome editing, such as that promised by the DdCBEs, 75 will undoubtedly garner interest and hopefully lead to the generation of more models, again by manipulation of cells in culture, movement to ES cells, and then implantation into embryos. Direct delivery into the mouse may be more efficient, but germline transfection by AAVs are still not widely available. Finally, work into other base editor domains with other mutational signatures will greatly help in our ability to specifically mutate the mtDNA to both generate and potentially heal our animal models, with the eventual goal of finding curative therapies to help the human patients suffering from these diseases.

ACKNOWLEDGMENTS
The author would like to acknowledge the past and ongoing collaborations with the Chinnery, Greaves, Larsson, Minczuk, and Moraes groups. The author would like to thank Nina Bonekamp and Laila Singh for their helpful comments on this manuscript. The author has received Research Group support from the Max Planck Society, funding from the United Mitochondrial Disease Foundation (13-053R), and support by a Marie-Sklodowska-Curie ITN European Training Network, "REMIX" (EU 721757). The author confirms independence from the funding sources and from Max Planck Innovation; the content of the article has not been influenced by the sponsors. Open access funding enabled and organized by Projekt DEAL.

CONFLICT OF INTEREST
The author is a named inventor for corporate licensing agreements offered by Max Planck Innovation, for commercial use of the mt-TA C5024T mouse mentioned in the review.

AUTHOR CONTRIBUTION
James Bruce Stewart is the sole author of the manuscript.