The central technical innovation that unlocked this field is the DdCBE system, a twin adenine-associated protein base editor in which two halves of a bacterial toxin-derived deaminase are each fused to a mitochondria-targeting sequence. When the two halves are brought together on opposite strands of the mitochondrial genome, the reconstituted enzyme converts one DNA base into another without making a double-strand break. A companion approach, termed mitoTALED, applies the same split-enzyme logic to adenine base conversion, and subsequent iterations have expanded the repertoire of accessible edits. Because these systems bypass the requirement for guide RNAs entirely, they sidestep the mitochondrial RNA import problem that stymied conventional CRISPR-based strategies for over a decade. The result is a toolset that can, in principle, install or reverse any transition mutation across the circular mitochondrial chromosome, opening the door to precise manipulation of a genome that carries pathogenic variants underlying a wide range of neuromuscular, metabolic, and neurodegenerative disorders.

The review emphasizes that generating a disease model is not as simple as pointing an editor at a mutation. Target selection demands careful attention to several constraints that the authors organize into a coherent decision framework. First, because the editors operate on specific base conversions, only a subset of the hundreds of known pathogenic mitochondrial DNA mutations can be faithfully recapitulated with current chemistry. Second, the editing window, the narrow stretch of DNA over which the deaminase acts once the two editor halves dock, must align with the position of the intended variant; off-target conversions within the window can create confounding mutations. Third, strand bias is a recurring technical reality, with editing efficiencies differing substantially depending on which mitochondrial DNA strand the deaminase ultimately acts upon, a consequence of the asymmetric architecture of the split-enzyme design. Fourth, the sequence context surrounding a target site, including neighboring nucleotides and local secondary structure, can dramatically modulate editing outcomes, making empirical validation at every candidate locus unavoidable.

Cross-species conservation emerges as another decisive consideration. Many mitochondrial disease models are pursued in mice, yet the murine mitochondrial genome differs from the human one at key positions, meaning that a mutation that is pathogenic in patients may not be directly installable at the orthologous site in an animal. The authors discuss how researchers must weigh the fidelity of a candidate model against the biological questions it is meant to answer, sometimes choosing a homologous but non-identical variant, or turning to alternative model organisms whose mitochondrial genomes more closely match the human sequence. This tension between practicality and physiological relevance is a recurring theme throughout the review, and the authors argue that rigorous justification of target selection is essential if edited models are to yield interpretable conclusions about human disease mechanisms.

Once an edited model has been generated, validation becomes the next hurdle, and the review proposes a three-tiered strategy operating at the functional, molecular, and organismal levels. At the molecular level, techniques such as ultra-deep amplicon sequencing quantify the fraction of mitochondrial genomes carrying the intended edit, a critical measurement given that heteroplasmy, the coexistence of mutated and wild-type mitochondrial DNA within a single cell, often determines whether a pathogenic variant crosses the biochemical threshold that triggers cellular dysfunction. At the functional level, the downstream consequences of an edit must be demonstrated: defects in respiratory chain complex activity, altered oxygen consumption rates, changes in membrane potential, shifts in ATP production, and activation of the mitochondrial unfolded protein response all serve as readouts that connect genotype to bioenergetic phenotype. At the organismal level, edited animal models must recapitulate the clinical features of the corresponding human condition, including tissue-specific pathology, age of onset, and inheritance patterns, before they can be trusted as platforms for drug discovery or preclinical testing.

The therapeutic dimension of the field is where momentum has been most striking. Rather than installing mutations, several recent studies have deployed mitochondrial base editors to reverse them, correcting pathogenic variants in vivo and rescuing disease phenotypes in animal models. The review highlights this therapeutic rescue work as proof of principle that the same enzymatic machinery used to build models can, with appropriate delivery vehicles, reach the mitochondria of affected tissues and shift heteroplasmy away from the mutant genome. The implications for mitochondrial medicine are considerable. Current therapeutic options for mitochondrial DNA diseases, which include dietary supplementation, exercise regimens, and in select cases genetic therapies aimed at reducing mutant genome load, address symptoms rather than the underlying genetic lesion. A base editor that permanently converts a pathogenic mutation back to the wild-type sequence offers something closer to a genuine cure, at least in principle.

Safety, however, remains a central preoccupation, and the authors devote sustained attention to the trade-off between editing efficiency and unintended effects. Off-target editing within the mitochondrial genome, where the deaminase acts at sequences resembling the intended target, has been documented for several editor architectures, and more recent work has raised the possibility of off-target effects in the nuclear genome as well, since components of the editor can occasionally be mislocalized to the nucleus. Prolonged expression of the editor increases the cumulative risk of such events, prompting interest in delivery strategies that limit exposure time, including mRNA-based delivery and engineered variants with narrower editing windows. The review frames this as an optimization problem: editors must be efficient enough to shift heteroplasmy to sub-threshold levels in a meaningful fraction of mitochondria, yet restrained enough that collateral edits, whether mitochondrial or nuclear, remain acceptably rare. Balancing these competing demands, the authors argue, will determine whether mitochondrial base editing can transition from laboratory tool to clinical platform.

The broader significance of this synthesis lies in its timing. Mitochondrial diseases collectively affect roughly one in several thousand live births, and no cure exists for the majority of the more than 300 pathogenic point mutations cataloged in human populations. The inability to manipulate mitochondrial DNA precisely has been a structural limitation of the entire field, forcing researchers to rely on patient-derived cells with unpredictable heteroplasmy, cybrid models with uncertain fidelity, and transmitochondrial animal lines generated through cumbersome techniques such as cytoplasmic transfer into embryos. Programmable base editors change this calculus. A researcher can now specify a mutation, install it at will in a cell line or an animal, and study its consequences under controlled conditions, then use the same or a complementary editor to erase it. That symmetry between disease modeling and therapeutic correction, the review argues, is the defining feature of the current era.

The work was conducted by teams based at Seoul National University and Korea University College of Medicine, with Youngho Kim affiliated with Edgene, Inc., a company whose involvement signals growing commercial interest in mitochondrial editing therapeutics. Hyunji Lee, the corresponding author, has filed patent applications related to the technology, including international filings under WO2022060185A1 and PCT/KR2021/012872 and a Korean patent application, underscoring that the intellectual property landscape around mitochondrial base editing is actively forming. The research was supported by grants from the Korea Institute for Advancement of Technology funded by the Ministry of Trade, Industry and Energy, and by the National Research Foundation of Korea through multiple programs funded by the Ministry of Science and ICT. Figures in the article were created with BioRender, and the paper was published open access under a Creative Commons license, with Claudia Feng serving as the primary editor during peer review.

As the field moves forward, the framework laid out in this review, spanning target selection, cross-species validation, window engineering, strand-bias correction, and staged functional testing, is likely to become a reference standard for laboratories entering the mitochondrial editing space. The remaining challenges are formidable: delivering editors efficiently to the nervous system and muscle, the tissues most ravaged by mitochondrial disease; achieving durable and tissue-appropriate heteroplasmy shifts; and excluding off-target activity with the rigor that regulatory authorities will demand. Yet the trajectory is unmistakable. Within a few short years, mitochondrial DNA has gone from an uneditable genomic outpost to a programmable therapeutic target, and the studies reviewed here, culminating in demonstrated disease rescue in animal models, suggest that the first clinical applications of mitochondrial base editing may be approaching faster than many in the field once thought possible.