Authors: Marie‐Thérèse Henke, Alessandro Prigione, Markus Schuelke
Categories: Euromit, Leigh syndrome, animal models, disease modeling, mitochondrial diseases, organoids, phenotyping, pluripotent stem cells, Review
Source: Journal of Inherited Metabolic Disease
Doi: 10.1002/jimd.12804
Leigh syndrome (LS) is a severe mitochondrial disease that results from mutations in the nuclear or mitochondrial DNA that impairs cellular respiration and ATP production. Mutations in more than 100 genes have been demonstrated to cause LS. The disease most commonly affects brain development and function, resulting in cognitive and motor impairment. The underlying pathogenesis is challenging to ascertain due to the diverse range of symptoms exhibited by affected individuals and the variability in prognosis. To understand the disease mechanisms of different LS‐causing mutations and to find a suitable treatment, several different model systems have been developed over the last 30 years. This review summarizes the established disease models of LS and their key findings. Smaller organisms such as yeast have been used to study the biochemical properties of causative mutations. Drosophila melanogaster, Danio rerio, and Caenorhabditis elegans have been used to dissect the pathophysiology of the neurological and motor symptoms of LS. Mammalian models, including the widely used Ndufs4 knockout mouse model of complex I deficiency, have been used to study the developmental, cognitive, and motor functions associated with the disease. Finally, cellular models of LS range from immortalized cell lines and trans‐mitochondrial cybrids to more recent model systems such as patient‐derived induced pluripotent stem cells (iPSCs). In particular, iPSCs now allow studying the effects of LS mutations in specialized human cells, including neurons, cardiomyocytes, and even three‐dimensional organoids. These latter models open the possibility of developing high‐throughput drug screens and personalized treatments based on defined disease characteristics captured in the context of a defined cell type. By analyzing all these different model systems, this review aims to provide an overview of past and present means to elucidate the complex pathology of LS. We conclude that each approach is valid for answering specific research questions regarding LS, and that their complementary use could be instrumental in finding treatment solutions for this severe and currently untreatable disease.
Keywords: animal models, disease modeling, Leigh syndrome, mitochondrial diseases, organoids, phenotyping, pluripotent stem cells
Leigh syndrome (LS) was first described in 1951 by the British physician Denis Leigh as a subacute necrotizing encephalomyelopathy in a 7‐month‐old male infant. ^1^ After 6 months of normal development, the infant showed symptoms of prolonged sleep, sweating, limb spasticity, and refusal to feed. Three days after admission to the hospital, the boy died. The autopsy revealed one of the hallmarks of LS: bilateral lesions in the basal ganglia of the central nervous system.
More than 70 years after the first clinical description of LS, the underlying disease mechanisms remain largely unknown. In fact, despite increasing knowledge about this rare disease with an incidence of ~1 case per 40 000 new births, the number of undiagnosed patients may be much higher, and no disease‐modifying treatment has been identified and clinically approved. ^2^ ^, ^ ^3^
Here, we present a literature review of the systems and disease models that have been employed to study LS and the potential treatments that have been suggested using these models.
LS is a fatal neurological disorder that predominantly presents with first symptoms around 7 months of age and results in death at a median age of 2.4 years. ^4^ Survival appears to depend on age of onset, genetic cause, and severity of symptoms. ^3^ ^, ^ ^4^ ^, ^ ^5^ ^, ^ ^6^ ^, ^ ^7^ ^, ^ ^8^ ^, ^ ^9^ Cases of late‐onset LS have also been observed. ^4^ ^, ^ ^5^ ^, ^ ^6^ ^, ^ ^9^ LS is considered a severe mitochondrial disease because it affects mitochondrial oxidative phosphorylation (OXPHOS). Mitochondrial diseases are a group of genetic disorders that affect the activity of the mitochondrial respiratory system to supply the cells of the body with sufficient amounts of ATP. ^10^ ^, ^ ^11^
Due to the lack of specific diagnostic criteria, Chang et al. conducted a meta‐analysis and analyzed published symptoms associated with LS in the context of specific OXPHOS deficiencies and mutations. ^12^ They found that developmental delay/retardation was the most common symptom, with other features including epilepsy, poor feeding, weakness, hypotonia, and ataxia. A recent review of LS divided the clinical presentation into neurological and non‐neurological symptoms. ^13^ Indeed, neurological symptoms appear to be the most prominent in LS. Focal and bilateral brain lesions are characteristic of LS. These lesions occur primarily in the brainstem, basal ganglia, and cerebellum ^12^ and affect basic life functions such as breathing, swallowing, and movement control. Another commonly observed diagnostic criterion is lactic acidosis, which refers to the increased concentrations of lactic acid in blood and cerebrospinal fluid (CSF). In fact, due to reduced mitochondrial activity, a compensatory increase in glycolytic rate may occur, leading in turn to further increased production and accumulation of lactate. ^14^
Since the development of magnetic resonance imaging (MRI), brain lesions can be detected in living patients, allowing the diagnosis of LS in combination with genetic and biochemical analyses. ^14^ Nevertheless, the diagnosis of LS still seems to be challenging because the symptoms can be non‐specific and variable between different patients, even among individuals carrying the same mutation or coming from the same family background. ^15^ In fact, because of the heterogeneity of symptoms and prognosis, the scientific community has now introduced the term Leigh Syndrome Spectrum disorder. ^13^
To date, mutations in 16 mitochondrial DNA (mtDNA) encoded and approximately 100 nuclear genes have been described to cause LS. MtDNA mutations are inherited via the maternal line and typically must affect a high proportion of mtDNA molecules to cause LS. ^13^ ^, ^ ^16^ ^, ^ ^17^ This phenomenon of coexistence of mutant and wild‐type mtDNA molecules is known as heteroplasmy, whereas the presence of mutant mtDNA alone is defined as homoplasmy.
Genes encoding proteins involved in mitochondrial respiration via the electron transport chain (ETC) at the inner mitochondrial membrane carry most of the mutations associated with LS. Common genetic defects typically cause complex I deficiency, ^12^ ^, ^ ^18^ ^, ^ ^19^ for example, mutations in nuclear genes encoding structural subunits of complex I (e.g., NDUFS4) or mutations in the mtDNA‐encoded genes MT‐ND1 to MT‐ND6. Other genetic defects result in complex IV deficiency, such as mutations in the nuclear complex IV assembly gene SURF1 or in the mtDNA‐encoded structural genes of complex IV, for example, MT‐CO1, or in mitochondrial ATPase deficiency caused by mutations in the mtDNA‐encoded genes MT‐ATP6 or MT‐ATP8
^12^ ^, ^ ^13^ ; for a recent review of the genetic causes of LS, see Ref. [13].
The severity of LS may also be related to the underlying mutation. For example, while both m.8993T>C and m.8993T>G mutations in the MT‐ATP6 gene can lead to LS, the m.8993T>G mutation has been found to result in a more severe phenotype. ^20^
Reduced ATP production due to impaired OXPHOS is compensated by cells through increased glycolysis, which may influence cell fate during development and may be one of the reasons why developmental delay is the most common feature of LS. ^21^ ^, ^ ^22^
A variety of models of LS disease have been created to study the underlying disease mechanisms or to test potential treatments. In the case of mutations in the nuclear DNA (nDNA), the generation of model organisms and isogenic controls is comparatively easy, especially since the development of various CRISPR/Cas9‐based genome editing approaches. ^23^ However, a recurring problem is that the models do not show the phenotypes observed in the patients. Furthermore, in the case of mtDNA mutations, generating a disease model is not as straightforward given the difficulties associated with mtDNA editing. ^24^ ^, ^ ^25^ Despite these challenges, a variety of LS disease models have been developed, leading to important insights into LS mechanisms and potential treatment strategies.
Here, we provide a summary of different disease models for LS with their respective advantages and pitfalls, and we highlight important insights into LS that have been gained through the use of these model systems. Figure 1 provides a graphical overview of these findings.
FIGURE 1 Models of Leigh syndrome. Leigh syndrome disease models (first column) with modeled respiratory chain (RC) defects in cyan (second column) and not modeled RC defects in white. The pictograms in the second column show the mitochondrial oxidative phosphorylation machinery. The third column shows the correlation of the model phenotype with the observed phenotype in patients based on the Human Phenotype Orthology (HPO; ORPHA:506). Phenotypes were separated in symptoms of the cardiovascular and respiratory system (anatomical heart symbol), nervous system (brain symbol), musculature (muscle symbol), metabolism (mitochondrion symbol), immunology (antibody symbol), growth (separating cells symbol), eye, ear, skin, hair, nails (face symbol), and cellular phenotype (apoptotic cell symbol). Cyan symbols indicate overall matching phenotypes, light gray symbols indicate not matching or not evaluated phenotypes. Created with BioRender.com.
Yeast has many advantages as a model organism, not only for LS. Most cellular functions, such as DNA replication or enzymatic activities, are indeed conserved between yeast and humans. ^26^ In addition, their rapid reproduction and resistance to different growth conditions make them convenient to work with in the laboratory and ensure the reproducibility of experiments.
In 1996, the complete genome of the Saccharomyces cerevisiae yeast stain was published. It contains more than 6000 genes. ^27^ ^, ^ ^28^ This knowledge led to the identification of several yeast homologs of human disease genes ^29^ and further promoted the role of yeast as an organism to study human disease at the molecular and pharmacological level. The main advantage over disease models is that homoplasmic mtDNA variants can be easily introduced into yeast because mtDNA replication in yeast does not depend on protein synthesis or require an origin of replication sequence. In addition, yeast can survive in the absence of mtDNA and can be transformed with bacterial plasmid DNA carrying a mitochondrial gene. ^30^ ^, ^ ^31^ Several mitochondrial deficient yeast models have been established over the years to study mitochondrial diseases; reviewed in Ref. [32]. Table 1 shows published yeast strains with LS‐causing mutations. As shown, most of the presented strains harbor mutations in OXPHOS complexes I, IV, and V and are restricted to only a few genes, mainly human SURF1 and MT‐ATP6.
Extensive studies in S. cerevisiae by Kucharczyk et al. showed that the m.9176T>G MT‐ATP6 mutation leads to complex V assembly defects, ^38^ while MT‐ATP6 mutations at position 8993 ^39^ ^, ^ ^46^ result in complex IV deficiency, suggesting a regulatory mechanism of complex V functionality on Cox1p expression. ^39^ ^, ^ ^40^ ^, ^ ^47^ However, this feature was not observed in human cells. ^48^ This lack of conserved effects highlights the importance of integrating different tools to study human pathologies. Recently, Baranowska et al. introduced two novel LS‐associated MT‐ATP6 mutations (m.8950G>A, m.9025A>G) into yeast and showed a significant reduction in ATP production and growth. ^41^
Four years after the discovery of mutations in the assembly factor SURF1 as the underlying cause of complex IV (COX) deficiency, ^49^ ^, ^ ^50^ Barrientos et al. established a S. cerevisiae model of deficiency by editing the SURF1 homolog Shy1
^34^ and confirmed its importance for COX assembly. Based on this, further studies in yeast were instrumental in providing insight into the assembly pathway of complex IV. ^35^ ^, ^ ^51^ ^, ^ ^52^ ^, ^ ^53^
Saccharomyces cerevisiae has limitations as a tool to study mitochondrial diseases because it does not express OXPHOS complex I. In contrast, Yarrowia lipolytica, a yeast of the Dipodascaceae family, has been used to study complex I‐associated LS mutations. However, only a few mutations have been introduced so far (Table 1). Site‐directed mutagenesis of the human nuclear‐encoded NDUFS7 and NDUFS8 homologs was able to generate an LS phenotype in yeast. ^44^ At the same time, Y. lipolytica carrying a mutation in its human NDUFS3 homolog showed no reduction in complex I activity or content. ^45^
Overall, most of the LS yeast tools showed impaired growth on non‐fermentable carbon (Table 1). Taken together, yeast appears to be a very useful tool for modeling genetic defects of LS, and, in particular, for studying the biochemical consequences of mitochondrial defects. Over the past 30 years, studies in S. cerevisiae have provided important insights into the function and assembly of proteins whose mutations are known to cause LS. In addition, yeast can also be used for drug screening. ^54^ ^, ^ ^55^ However, newer approaches for LS or NARP have been lacking since 2011.
Because of the ability to study intercellular communication of different cell types in a physiological context, non‐mammalian animals have been widely used to model several rare genetic diseases; reviewed in Ref. [56]. For LS, most non‐mammalian models have been established for complex I deficiencies, and only a few for complex II (e.g., for the SDH gene), complex IV (e.g., the SURF1 gene), and complex V (e.g., the MT‐ATP6 gene) (Table 2).
Burman et al. generated a Drosophila melanogaster model carrying a short in‐frame deletion in the mt*‐nd2* gene. ^57^ The generated flies showed a milder phenotype than mammalian MT‐Nd2 knockout models. Nevertheless, overall complex I activity and energy production were reduced, and behavioral and histological analyses revealed features of mitochondrial disease. Using this model the authors found further evidence for the role of the mt‐nd2 subunit in proton pumping. In 2016, the mt‐nd2 ^del1^ model was used to test rapamycin as a potential treatment. ^58^ Wang et al. found that rapamycin rescued the fat storage defect of mt‐nd2 ^del1^ flies and observed an autophagy‐independent increase in lifespan. However, the behavioral phenotype was not altered by the treatment.
One of the earliest non‐mammalian LS models was two Drosophila melanogaster models with whole body (Actin5C‐GAL4) or brain‐specific (elav‐GAL4) post‐transcriptional knockdown of surf1. ^63^ The authors found a phenotype similar to human LSSURF1−/− patients. Predominantly, these fruit flies showed impaired development and light‐induced locomotion. However, severe motor defects were only observed in the Actin5C‐GAL4 surf1 knockdown flies. In contrast, elav‐GAL4 surf1 knockdown flies had a longer lifespan than controls, although they showed impaired visual responses, probably due to alterations in several visual circuits. ^63^ More recently, using the CRISPR/Cas9 gene editing technology, Haroon et al. generated two surf1 ^ −/− ^ zebrafish (Danio rerio) deletion strains that expressed many LS phenotypic hallmarks. ^70^ They also found altered oxidative stress induction and glutathione metabolism after treatment with sodium azide, a complex IV inhibitor. Strikingly, this response was less profound after prophylactic treatment with cysteamine bitartrate (Cyst‐BIT) or N‐acetylcysteine (NAC), both thiol‐containing reagents that may represent promising drug candidates for SURF1‐associated LS and COX deficiency. ^70^
In 2006, Celotto et al. found an mt‐atp6 mutation in their Drosophila sesB1 (ant1) mutant strain. Their analysis showed that these flies recapitulated many symptoms of LS. Furthermore, the authors demonstrated that normal respiration can occur despite defects in ATP synthase activity. ^65^ Later in 2019, this model was also used to study sleep and circadian function in mitochondrial encephalomyopathies. ^66^
In summary, non‐mammalian LS models have been used to study development and lifespan, as well as neurological deficits and locomotor effects associated with specific mutations, and have been instrumental in suggesting potential treatment strategies (Table 2).
In general, mammalian animal models have several advantages over non‐mammalian animals (described in Section 2.1.2) for disease modeling. Their closer relationship with humans allows for a more detailed analysis of cognition, movement, and development. However, as recently reviewed by Mukherjee et al., ^71^ the choice of an appropriate animal model that can reproduce the pathology in the same way as in humans is crucial. Furthermore, despite the ever‐evolving methods for generating transgenic animal models, scientists face only a few successes and many hurdles when it comes to modeling LS or mitochondrial diseases in general; reviewed by Ref. [72].
The most prominent mammalian model of LS is the Ndufs4 knockout mouse (Table 3). Ndufs4 is a nuclear gene encoding the accessory NADH‐dehydrogenase subunit S4 of OXPHOS complex I. To date, several mutations of this gene can cause LS or mitochondrial complex I deficiency in humans (OMIM no. 252010), reviewed in Refs. [99, 100]. In 2008, Kruse et al. ^74^ published the first homo‐ and heterozygous Ndufs4 knockout mice. These mice exhibited symptoms similar to those seen in LS patients with complex I (CI) deficiency. Later, additional Ndufs4 knockout mice were generated with tissue‐specific knockouts, mainly in brain, heart, and skeletal muscle (Table 3). Lagrue et al. ^85^ tried a different approach to mimic CI deficiency by inhibiting CI function in young mice using 1‐methyl‐4‐phenyl‐1,2,3,6‐tetrahydropyridine (MPTP). While the mice showed elevated lactate levels and striato‐nigral network degeneration as seen in LS, they did not exhibit the characteristic symptoms of LS.
A Surf1 knockout mouse was published in 2003. ^87^ However, the authors could only partially mimic the human LS phenotype with their model. In fact, a hallmark of LS, the lesions in the basal ganglia and brainstem, were absent in the Surf1 knockout mice. In 2018, a new approach was taken by Quadalti et al. ^98^ to model COX deficiency by generating Surf1 knockout pigs using the new TALEN and CRISPR/Cas9 gene editing tools. Although the pigs did not show COX deficiency, they suffered from severe growth retardation and died shortly after birth. Therefore, this model did not fully display the human SURF1‐associated LS phenotype but indicated the presence of severe developmental features that may be related to the overall pathogenesis of LS.
The solute carrier (SLC) genes comprise a superfamily of 55 gene families with a total of >360 protein‐coding genes. ^101^ Mutations in genes belonging to the SLC19 and SLC25 families ^102^ and the SLC39 family are associated with LS. ^13^ In 2017, Terzenidou et al. established a mouse model carrying a nonsense mutation in the highly conserved Slc25a46 gene, resulting in loss of protein function. The mice exhibited some typical symptoms of LS, such as ataxia, seizures, and early lethality. At the molecular level, the authors found that most of the mitochondrial interactome in the cerebellum was negatively affected by the mutation. ^94^
Overall, many different mammalian models have been generated over the past decades to study LS. However, the simple introduction of LS‐associated mutations has not always led to the phenotype observed in patients. This complicates the study of LS, as there appear to be multiple underlying causes for the onset of LS symptoms demonstrating the complexity of the disease. At the same time, some models, such as the Ndufs4 knockout mice, appear to be a robust model system and are therefore widely used in the community, as reviewed by Ref. [100].
In summary, mammalian LS models provide a better insight into how specific mutations might affect body weight and feeding, the interplay between different organs (e.g., cardiac and neural functions), specific brain regions, and the regulation of metabolites. In addition, mouse models can be used to study the toxicity and physiology of potential drug treatments in vivo, providing an important platform for translational studies (Table 3).
Obtaining primary cells, mostly fibroblasts, from skin biopsy specimens is a common procedure in the diagnostic process for LS. ^14^ We have focused here on publications that examine primary cell cultures in more detail to investigate disease mechanisms and treatments, rather than just for diagnostic purposes (Table 4).
In 2007, Koopman et al. ^103^ published a study using 16 patient‐derived fibroblasts carrying 16 different nuclear‐encoded complex I mutations and 6 control fibroblast lines. Their aim was to study human NADH:ubiquinone oxidoreductase deficiency in more physiological and molecular detail. The main experiments were performed to compare mitochondrial morphology and reactive oxygen species (ROS) levels. According to the authors, the included CI‐deficient patient‐derived fibroblasts could be divided into two classes, with class I showing more fragmented mitochondria after rhodamine 123 staining in living cells and higher ROS levels as compared with class II. In a follow‐up study with the same fibroblast lines, mitochondrial membrane potential (Δψ), calcium storage, and ATP production were comprehensively investigated. ^104^ The authors found a reduced calcium content in the ER and reduced ATP production in the mitochondrial matrix of CI‐deficient fibroblasts, as well as a depolarized Δψ. Ten years later, Iannetti et al. ^106^ used three of the previously studied patient fibroblasts carrying mutations in NDUFS7‐p.V112M, NDUFS8‐p.R94C, and NDUFV1‐p.R59X/p.T423M ^103^ in pyruvate‐ and glucose‐free, galactose‐substituted cell culture conditions. While control lines survived, patient fibroblasts died in the glucose‐free culture medium and could only be rescued by the addition of pyruvate and eNAD in a dose‐dependent manner. The authors conclude that eNAD restores Δψ, while pyruvate reduces ROS levels.
Further studies of complex I deficiency have been performed in a variety of cell lines, with fibroblasts remaining the cell type of choice. NDUFS4 knockouts have been generated in primary mouse embryonic fibroblasts (MEFs), isocortical astrocytes and neurons, ^111^ and HEK293 cells. ^112^ Despite a reduced CI function and ATP production in all cell types, Bird et al. were able to show an impaired Δψ only in the MEFs, but not in the neuronal cells, arguing that this may be due to the limitations of the study. ^111^
Wojtala et al. ^107^ studied ROS production in patient‐derived fibroblasts carrying mtDNA mutations (degrees of heteroplasmy) in MT‐ND1 (100%), MT‐ND3 (75%), and MT‐ND5 (85%). In their cellular models, they showed that phosphorylation of the p66Shc protein (which is absent in mice) can activate a pro‐oxidative pathway that can be inhibited by the natural fungal compound hispidin.
In addition to CI‐deficiency, other cellular models of LS have been studied. COX deficiency was studied in 12 patient‐derived immortalized B‐lymphocytes (with 15 different SURF1 mutations) by Li et al. ^114^ The advantage of this study was the less invasive method of obtaining patient samples. By comparing the different cell lines, the authors argued that the missense mutations studied were associated with a better disease prognosis compared with the SURF1 null mutations.
Furthermore, French‐Canadian LS was also studied in fibroblasts carrying the c.1119C>T mutation in the LRPPRC gene. ^110^ The authors found no increased ROS production, but a decrease in Δψ, a fragmented mitochondrial network, impaired OXPHOS‐capacity, and increased sensitivity to Ca^2+^‐induced permeability transition.
Overall, patient‐derived fibroblasts are the human cell type of choice to study the pathophysiology and underlying molecular mechanisms of LS. The advantage of patient‐derived cells is that they carry all the genetic information (both nuclear and mitochondrial) of the patient and can be conveniently cultured. However, fibroblasts from patients with LS are not as severely affected as muscle or brain cells. However, these other cell types are more difficult to maintain in culture and can only be obtained from patients using more invasive methods or for post‐mortem studies. Nevertheless, two studies included LS patient‐derived myoblasts carrying SURF1
^115^ and ECHS1
^116^ mutations. The ECHS1 mutation caused respiratory chain defects and the immortalized patient‐derived myoblasts have been comprehensively studied. ^116^ The patient‐derived SURF1 mutant myoblasts showed a bioenergetic imbalance toward glycolysis, disturbed cell cycle, increased proliferation, and upregulation of the proteasome. ^115^ In conclusion, LS patient‐derived cellular models are relatively convenient to derive and have been analyzed in terms of proliferation in different growth media compositions and correlation with their mtDNA heteroplasmy levels. Their use provides important insight into the pathogenicity of mutations to highlight affected intracellular pathways.
Transmitochondrial cellular hybrids (cybrids) are obtained by fusion of a cell line without mtDNA (called p ^0^‐cells) with cytoplasts from another cell line. This method was first described by King and Attardi in 1989 ^117^ and has been widely used in the following 30 years. The advantage of this method was that it allowed the study of LS and other mitochondriopathies in a more robust cellular model than primary cells, using cell lines that could be easily expanded for multiple passages. It has also been used to define and validate whether the mitochondriopathy is due to a genetic defect in the mtDNA or the nDNA. For a comprehensive overview of cybrids as a disease model for mitochondriopathies, including the advantages and limitations of the method; see the review by Ref. [118].
For LS, many cybrid lines have been established for complex I, IV, and V deficiencies. The earliest cybrids were made in the early 1990s for COX deficiency. Tiranti et al. ^119^ used cybrids to discover that mutations in the nDNA were responsible for causing COX deficiency. They created a cybrid line carrying the patient's mtDNA and healthy nDNA (143BSCy.TK^+^) and another cell line carrying the patient's nDNA and healthy mtDNA (SACy‐Neo^r^). Only the SACy‐Neo^r^ line showed COX defects, leading to the conclusion that the patient's specific COX deficiency was caused by nDNA mutations. In a follow‐up study, the authors were able to propose that the COX deficiency in eight of their patients was caused by mutations in the same gene, by the fusion of the seven COX‐negative patient lines with the previous COX‐negative line. ^120^ One year later, this hypothesis was confirmed by mutational analysis, which identified pathogenic variants in the gene encoding the complex IV assembly factor SURF1. ^49^
To study the biochemical effects of the m.8993T>G MT‐ATP6 mutation, Trounce et al. created a cybrid model in 1994 ^121^ and measured reduced maximal respiration as well as a reduced ADP/O ratio. In addition, Manfredi et al. ^122^ produced three cybrids carrying m.8993T>G mutations with different levels of heteroplasmy and used galactose‐oligomycin supplementation of the cell culture media to select for cells with different levels of heteroplasmy. They found a reduction in mutant load after 5 days of selection and 1 day of recovery. Other MT‐ATP6 mutations have been studied in cybrids. In 2009, D'Aurelio published five cybrid lines carrying the m.8993T>G, m.8993T>C, and m.9176T>G mutations. ^17^ The authors found that not only heteroplasmy levels contribute to the severity of the LS phenotype, but also the overall mtDNA background, as several variations in structural genes of other respiratory chain complexes were detected in these cybrids. Thus, the importance of MT‐ATP6 sequencing when mitochondrial disease is suspected has been reinforced by studies by Auré et al., ^123^ Blanco‐Grau et al., ^124^ and López‐Gallardo et al., ^125^ who all used cybrid models of novel MT‐ATP6 mutations (Table 5).
For complex I deficiency, LS cybrids were generated with mutations in MT‐ND1, ^130^ ^, ^ ^131^
MT‐ND2, ^129^
MT‐ND3, ^141^
MT‐ND4, ^126^
MT‐ND5, ^130^ ^, ^ ^133^ and MT‐ND6. ^127^ ^, ^ ^128^ ^, ^ ^130^ ^, ^ ^132^ ^, ^ ^142^ However, the results of these studies were mixed, while some mutations showed reduced CI activity ^127^ ^, ^ ^128^ ^, ^ ^130^ ^, ^ ^131^ ^, ^ ^133^ or even assembly defects, ^127^ ^, ^ ^129^ ^, ^ ^130^ a recent study by Chen et al. did not observe any severe metabolic defects in the cells, despite reduced ATP levels ^132^ (Table 5). In addition, an interesting approach was taken by D'Aurelio et al. by fusing two cybrids with mutations in complex IV and III to study mitochondrial functional complementation. ^137^
Transmitochondrial cybrids have been generated not only from patient‐derived cells, but also from cells derived from mouse models. ^143^ ^, ^ ^146^ ^, ^ ^147^ Kirby et al. ^144^ generated mouse transmitochondrial embryonic stem cells (ESCs) carrying different mutations in complex I and IV and differentiated them into neurons to model the neurodegenerative effects of mtDNA mutations. They showed that mutations leading to severe electrochemical defects also showed impaired differentiation. In a follow‐up study, Abramov et al. ^145^ focused on the mitochondrial metabolism of these affected neuronal lines and found an increased Δψ and increased ROS production in neuronal cybrids with a mutation in complex I, leading to cell death.
In conclusion, transmitochondrial cybrids are important models for the study of LS and other mitochondrial diseases. In particular, it has become a state‐of‐the‐art approach to generate cybrids to validate the pathogenicity of novel mtDNA mutations identified in patients, as these cells can be used to analyze biochemical properties and validate the genetic location of the defect.
However, with the development of patient‐derived induced pluripotent stem cell (iPSC) technologies (see Section 2.3.1), the cybrid approach is being used less and may fade into the background of the toolkit for studying mitochondria‐related diseases.
In 2006, Takahashi and Yamanaka published a novel method for reprogramming fibroblast cultures into induced pluripotent stem cells (iPSCs). ^148^ Since then, a new era of disease modeling and drug screening has begun, with important implications for LS.
It is important to note that during the stressful process of reprogramming, cells can undergo genomic changes that may also affect the mtDNA. ^149^ This is particularly important for modeling mitochondrial diseases, where genetic background and heteroplasmy levels are critical parameters in determining the disease phenotype and severity. Several studies have shown that mtDNA variants can change during the reprogramming process, as reviewed in Ref. [25]. Therefore, in addition to a common quality control analysis of all generated iPSC clones (including karyotype, morphology, pluripotency, and DNA fingerprinting analysis), it is becoming clear that mtDNA sequence and heteroplasmy testing should also be included, as reviewed in Ref. [150, 151, 152, 153].
The iPSCs in the undifferentiated state do not represent the disease model of choice. In fact, the metabolism of pluripotent stem cells is mainly geared toward glycolysis rather than OXPHOS. ^23^ ^, ^ ^154^ Therefore, to develop model systems for mitochondrial diseases, iPSCs need to be differentiated into specific cell types. The possibility to differentiate iPSCs into cell lines that are highly affected in mitochondrial diseases (e.g., neurons or cardiomyocytes) offers the greatest advantage for studies of the underlying disease mechanisms and for drug screening, as reviewed in Ref. [25, 152, 155].
In the context of LS, the first iPSC models were generated in 2015 using patient‐derived fibroblasts carrying the heteroplasmic m.13513G>A MT‐ND5 and the heteroplasmic and homoplasmic m.8993T>G MT‐ATP6 mutations. ^156^ The aim of this study was to eliminate the presence of the mtDNA mutation to obtain cell lines with corrected functionality. This was successfully achieved using somatic cell nuclear transfer (SCNT), demonstrating the potential for cell therapy applications in LS.
In 2017, Lorenz et al. ^22^ generated iPSCs lines for MT‐ATP6 associated LS carrying the m.9185T>C mutation and differentiated them into neural progenitor cells (NPCs) and dopaminergic neurons. In the mutant NPCs, they found an abnormally elevated Δψ, which they used as a readout for high‐throughput drug screening. They found that the PDE5 inhibitor avanafil was able to reduce the abnormally elevated mitochondrial membrane potential (MMP) in mutant NPCs and ameliorate the calcium signaling defects in both mutant NPCs and neurons. In the following year, the group published several additional lines carrying different MT‐ATP6 mutations. ^157^ ^, ^ ^158^ ^, ^ ^159^
Between 2016 and 2020, Galera et al. published several LS iPSC lines derived from patients carrying mutations in the MT‐ND5 gene (m.13513G>A) ^160^ ^, ^ ^161^ ^, ^ ^162^ and the MT‐ATP6 gene (m.8993T>G). ^163^ For the MT‐ND5 mutation, they found that the ability to differentiate into cardiomyocytes was strongly dependent on the mutation load, which is consistent with the prominent symptoms of cardiomyopathy in patients carrying this mutation. ^162^ In addition, they suggested that iPSCs with higher mutation load might favor differentiation into the neuroectoderm lineage. In another study, they differentiated the same iPSCs into neural stem cells and neurons and observed reduced OXPHOS function and calcium buffering capacity. ^161^
The complex I gene NDUFS4, which has been extensively studied in mouse models (see Section 2.1.3), has also been studied in iPSCs. Two NDUFS4 knockout iPSC lines were generated by CRISPR/Cas9 gene editing. ^112^ ^, ^ ^164^ Yoon et al. further differentiated the iPSCs into cardiomyocytes and were able to generate a cellular phenotype with a lower mitochondrial membrane potential, higher ROS levels, and slower SERCA Ca^2+^ reuptake. ^112^ Consistent with these findings, Galera‐Monge et al. differentiated the *NDUFS4‐*mutant iPSC line toward the neuronal lineage. Their NDUFS4‐ko neurons were functional but exhibited increased cell death and altered calcium regulation. ^161^
The authors compared the effects of a Ndufs4‐deletion in mice, HEK293 cells, and iPSC‐derived cardiomyocytes. They observed complementary results as the cardiac defects of Ndufs4
^−/−^ mice were recapitulated in iPSC‐derived cardiomyocytes as documented by decreased sodium‐(Nav1.5) and Ca^2+^‐(SERCA2a) transients.
^112^
They also suggested nicotinamide riboside (NR) as a potential treatment strategy for LS.
^165^
Daneshgar et al. differentiated NDUFS4 knockout iPSCs into neurons and brain organoids and found increased glutamatergic excitotoxicity under pro‐inflammatory conditions.
^164^
NPCs carrying mutations in the NDUFS4 gene were found to have defective neuronal growth capacity and impaired Δψ.
^166^
Overall, as highlighted by the studies reviewed here (Table 6), patient‐derived iPSCs provide an important platform to model LS. At the same time, comprehensive and time‐consuming quality control analyses are required during reprogramming and after differentiation processes to ensure that the model faithfully recapitulates the genetic characteristics of the patient. iPSCs may allow the development of a disease model that is genetically close to the patient, with many advantages for advancing our understanding of the disease and discovering innovative and potentially personalized treatment options.
Organoids are self‐organizing three‐dimensional (3D) cell clusters that reproduce specific aspects of the structure and function of a human organ. They can be derived either from somatic stem cells isolated directly from patients or through the differentiation of pluripotent stem cells. The latter is the most widely used for all those organs that are not easily accessible, such as the brain. Differentiation is based on the modulation of signaling pathways that are important during the development of the target tissue; for a review, see Ref. [180, 181].
In the case of LS, since specific defects mainly affect the central nervous system, organoid models have mainly focused on brain models (Table 6). Among these, both unguided cerebral brain organoids, ^166^ ^, ^ ^178^ or regionalized guided cortical organoids ^164^ have been used to model LS.
Inak et al. generated NPCs, dopaminergic neurons, and cerebral organoids from LS patient‐derived iPSCs carrying the c.530T>G and c.769G>A SURF1 mutations and used the CRISPR/Cas9 technology to correct the mutations in the patient iPSCs and conversely to introduce them into control iPSCs, to obtain different isogenic systems. SURF1 mutant neurons and brain organoids showed reduced neuronal maturation and defective neuronal outgrowth capacity. Further analysis revealed an impaired transition from glycolysis to OXPHOS at the level of NPCs, which may be the pathogenetic basis for the disrupted morphology in mutant organoids. The data supported the beneficial use of bezafibrate to increase mitochondrial biogenesis via PGC1α signaling, ^166^ which in turn improved neuronal branching capacity. Thus, modulation of progenitor cell metabolism could positively impact neurogenesis and potentially ameliorate the neurodevelopmental aspects of LS. ^21^ This disease feature has been used to identify potential repurposable drugs based on their ability to correct morphogenesis defects in iPSC‐derived neurons and midbrain organoids. ^182^
Romero‐Morales et al. generated iPSCs carrying mutations in the genes MT‐ATP6/PDH (m.8993T>G/c.79delC), PDH (c.79delC), and DLD (c.100A>G). They differentiated them into NPCs, neural rosettes, and cerebral organoids. In contrast to 2D cultures of LS neuronal cells, which showed almost no significant differences in viability and metabolism compared with controls, 3D cerebral organoids of LD neuronal cells showed disturbed cortical development, a disrupted mitochondrial network, and altered metabolism. ^178^ This study highlights the potential advantage of organoid models over 2D cultures in recapitulating some of the complex features observed in LS patients.
These two comprehensive LS organoid studies have demonstrated the potential of organoid research as a growing model platform for LS. Additional organoid models are needed to draw more precise conclusions about the impact of LS‐associated mutations during organ and tissue development. In the context of brain organoids, innovative approaches are needed to incorporate blood vessels and immune cells such as microglia, which would be necessary to address the neuroinflammatory aspects associated with LS. ^183^ Overall, several questions remain regarding the use of organoids in modeling LS and mitochondrial diseases, for review, see Ref. [25]. Ongoing and future studies will hopefully shed new light on the pathology of LS and possibly identify new targets for intervention.
In this review, we have highlighted the different types of disease models that are available for LS. Given the complexity of mitochondrial diseases in general, the development of disease models for LS has been challenging. ^13^ Innovations in gene editing techniques, such as CIRSPR/Cas9 ^23^ and mtDNA‐based editing ^184^ play an important role in the continuous development of innovative disease models.
In the early 1990s and 2000s, yeast was the first model organism for LS. Due to its enzymatic similarity to human cells and the ease of mtDNA editing, it has been possible to study fundamental proteomic and functional aspects of LS. However, only Shy1 (SURF1), ^35^ ^, ^ ^51^
MT‐ATP6, ^37^ ^, ^ ^38^ ^, ^ ^41^ ^, ^ ^43^ ^, ^ ^47^ ^, ^ ^185^ and complex I genes ^44^ ^, ^ ^45^ have been studied, which represent only a small percentage of the genetic causes of LS. Although yeast has been suggested as a useful tool for drug screening against LS, this approach has not been strongly pursued in the review. ^54^ ^, ^ ^55^ The next most common organisms for LS research are small non‐mammalian animals such as Drosophila melanogaster, ^57^ ^, ^ ^59^ ^, ^ ^60^ ^, ^ ^63^ ^, ^ ^65^ ^, ^ ^66^ ^, ^ ^67^
Caenorhabditis elegans, ^68^ or Danio rerio (zebrafish). ^70^ All of these organisms are generally useful for studying neurodegenerative diseases. In contrast to yeast models, non‐mammalian models with a variety of edited genes (both nDNA and mtDNA) have been published over the years. In addition, some of these species are more amenable to phenotypic analysis because they often exhibit motor symptoms such as exercise intolerance and muscle weakness that are also seen in patients.
Mice are genetically even closer to humans and have therefore been used extensively to study LS; for review, see Ref. [72]. The most widely used model is the Ndufs4 knockout mouse, ^74^ which shows phenotypic features similar to those observed in patients. However, the more complex an organism is, the more difficult it is to model disease. For example, some engineered gene mutations in mice designed to cause LS, failed to do so, causing aberrant phenotypes, such as increased longevity and often raising more questions than providing answers.
LS primarily affects the brain and neurological symptoms are similar in all patients. ^12^ Therefore, studying LS in neuronal cell lines is another approach to gain insight into this disease. Since neuronal lines cannot be obtained from living patients, the innovative development of reprogramming fibroblasts into iPSCs, ^148^ which can (theoretically) be differentiated into any other cell line, has transformed LS research. Over the past 9 years, many LS iPSC lines have been published, mostly carrying mutations in MT‐ATP6, SURF1, or genes encoding structural subunits or assembly factors of complex I. LS iPSCs have been used to generate differentiated cells such as NPCs, neurons, astrocytes, as well as cardiomyocytes and muscle cells. In addition, the generation of organoids from iPSCs now allows the dissection of the effects of LS mutations on organ and tissue development in the human context. ^25^ Thus, iPSCs may pave the way for personalized and mutation‐specific medicine as well as high‐throughput drug screening.
With this literature review, we have attempted to provide a comprehensive overview of LS disease models from the last 30 years. The review aims to describe the models that have been generated and highlight their key findings. We show how LS disease models have evolved with the continuous innovation of laboratory techniques. Scientists around the world have worked to elucidate the underlying mechanisms of LS and to develop new treatments and cures for LS using different models. This review shows that the combination of different model systems can provide us with a deep understanding of the cellular effects caused by specific mutations and the resulting symptoms.
While nuclear mutations can be easily introduced into an organism using CRISPR/Cas9 technology, introducing mutations into the mtDNA is more challenging. In addition, the heteroplasmy level of mtDNA mutations is a critical factor in disease outcome, symptoms, and severity. Therefore, genetic engineering of mtDNA is another area of research aimed at developing better disease models and treatments; for review, see Ref. [186].
In 2013, Bacman et al. introduced mitoTALENs (mitochondria‐targeted transcription activator‐like effector nucleases), an engineered system of mitochondria‐targeted nucleases with the ability to recognize specific pathogenic mutations in the mtDNA of living cells (cybrids) and of mouse organs. They were designed to induce a double‐strand break (DSB), thereby reducing the heteroplasmy levels of the mutant mtDNA copies only. ^187^ In 2015, the same laboratory published mitoTALENS specific for two other mutations, one of them m.13513G>A of the MT‐ND5 gene, which is associated with LS. ^188^ However, the group was not able to reduce the mutation load to zero. The research continued and the group of Moraes et al. introduced mitochondrial targeted meganucleases (mitoARCUS) in 2021 that were more specific and efficient. ^189^
In 2020, Mok et al. introduced RNA‐free DddA‐derived cytosine base editors (DdCBEs), which unlike mitoTALENS or mitoARCUS, do not induce double‐strand breaks but have the ability to directly correct a mutation. ^184^ However, only CG‐to‐TA conversions were possible. Efficiencies ranged from 5% to 50%, and off‐target effects were common.
The teams of Minczuk et al. and Liu et al. were able to advance the method by combining it with zinc fingers. ^190^ For example, Silva‐Pinheiro et al. published a library (named MitoKO) of precise mtDNA base editors with the ability to induce premature stop codons. ^191^ In the study, the authors performed in vivo base editing and introduced the MT‐ATP6 m.8096G>A mutation into a mouse model. Later they also used AAVs to deliver DdCBEs to edit mtDNA in the heart tissue of living mice. ^192^ The promising results demonstrate the potential of the method to develop reliable mtDNA‐associated mammalian disease models as well as gene therapies against LS and other mitochondrial diseases. A protocol for the MitoKO library has recently been published, ^193^ making this method more accessible to the scientific community, which will lead to more model organisms and insights into the disease mechanisms and therapeutic solutions in the future.
In terms of the impact of the LS models and tools described here on patient management or treatment, some very useful discoveries have been made. As previously reviewed, LS treatment usually only involved vitamins, supplements, or a ketogenic diet. ^194^ However, the Ndufs4 mouse model in particular has often been used to test potential drugs for LS, for example, the mTOR inhibitor Rapamycin (NCT03747328), ^195^ which was initially discovered using a Drosophila model of LS (see Section 2.1.2 and Table 2). Another mTOR inhibitor, ABI‐009, is presently being investigated in a clinical trial for LS (NCT03747328). As treatment developments for rare diseases are not very profitable, open‐label studies are more common for LS, the EPI743‐12‐002 (NCT02352896) study being an example. More recently the European Medical Agency (EMA) issued an orphan drug designation for Cannabidiol (EU/3/23/2800), which had been studied in the Ndufs4 mouse model ^73^ (see Table 3). Further, Sildenafil was also given an orphan drug designation by the EMA (EU/3/23/2831) after the discovery of PDE5 inhibitors as promising drug candidates in LS patient‐derived neuronal progenitor cells. ^22^ Nicotinamide riboside has been suggested after studies in NDUFS4 iPSCs ^112^ (Table 6) but is still lacking sufficient pre‐clinical data. ^165^ Azole compounds such as Talarazole and Sertaconazole have also been suggested as potential repurposable molecules in the treatment of LS. ^182^
Another field of treatment is gene replacement therapy, which has been studied in the Ndufs4 mouse model. ^196^ Further, gene replacement therapy using the AAV9/hSURF1 vector (TGTX‐102) in LS patients with SURF1 mutations has obtained an orphan drug designation by EMA and FDA (EU/3/21/2531) after successful studies in Surf1 knockout mice. ^197^ The therapy is planned to be tested in a clinical study shortly.
In conclusion, although there is still no adequate treatment or cure for LS patients, the developed LS models and tools are widely used to study potential drugs. In particular, the development of the Ndufs4 knockout mouse and iPSC‐derived models has led to the testing of more drugs in the last decade, even based on older observations. This highlights that only a deep and comprehensive understanding of the disease and the availability of sufficient tools will allow the study and development of treatments, especially for such a complex mitochondrial disease as LS.
Conceptualization: Marie‐Thérèse Henke, Alessandro Prigione, Markus Schuelke. Funding acquisition: Alessandro Prigione, Markus Schuelke. Investigation: Marie‐Thérèse Henke. Methodology: Marie‐Thérèse Henke. Supervision: Alessandro Prigione, Markus Schuelke. Validation: Alessandro Prigione, Markus Schuelke. Writing – original draft: Marie‐Thérèse Henke. Writing – review and editing: Alessandro Prigione, Markus Schuelke. All authors approved the final manuscript as submitted.
Marie‐Thérèse Henke declares that she has no conflict of interest. Alessandro Prigione declares that he has no conflict of interest. Markus Schuelke declares that he has no conflict of interest. Alessandro Prigione and Markus Schuelke have submitted patent applications for the use of Sildenafil in Leigh syndrome. Alessandro Prigione has submitted patent applications for the use of Talarozole and Sertaconazole in Leigh syndrome.
We acknowledge support from DFG (Deutsche Forschungsgemeinschaft) (PR1527/6‐1 to A.P.), the European Joint Programme for Rare Diseases (EJPRD) supported in Germany by the BMBF (Bundesministerium für Bildung und Forschung) (01GM2002A to A.P.), the Foundations MitoHelp, CureMito, and CureATP6, and the Fondation Maladies Rares (to A.P.), and the European Union (SIMPATHIC #101080249 to A.P. and M.S.). M.S. was supported by the DFG (Deutsche Forschungsgemeinschaft) under Germany's Excellence Strategy (EXC‐2049‐390688087) via the NeuroCure consortium at Charité – Universitätsmedizin Berlin.
Henke M‐T, Prigione A, Schuelke M. Disease models of Leigh From yeast to organoids. J Inherit Metab Dis. 2024;47(6):1292‐1321. 10.1002/jimd.12804
Alessandro Prigione, Email: alessandro.prigione@hhu.de.
Markus Schuelke, Email: markus.schuelke@charite.de.