MITOCHONDRIAL GENOME ENGINEERING COMING-OF-AGE

Abstract

The mitochondrial genome has been difficult to manipulate because it is shielded by the organelle double membranes, preventing efficient nucleic acid entry. Moreover, mtDNA recombination is not a robust system in most species. This limitation has forced investigators to rely on naturally occurring alterations to study both mitochondrial function and pathobiology. Because most pathogenic mtDNA mutations are heteroplasmic, the development of specific nucleases has allowed us to selectively eliminate mutant species. Several “protein only” gene editing platforms have been successfully used for this purpose. More recently, a DNA double strand cytidine deaminase has been identified and adapted to edit mtDNA. This enzyme was also used as a component to adapt a DNA single strand deoxyadenosine deaminase to mtDNA editing. These are major advances in our ability to precisely alter the mtDNA in animal cells.

INTRODUCTION

Human mtDNA is an 16,569 bp double-stranded, circular DNA present in most cells. Large-scale rearrangements/deletions and point mutations are found throughout the mtDNA in any of the 13 protein-coding, 22 tRNA, or 2 rRNA genes [1]. Since 1988, mutations have been associated with a myriad of clinical syndromes. These disorders are extremely heterogeneous, ranging from isolated blindness or ocular myopathy to severe multisystemic disorders. The prevalence of mtDNA-related disease diagnosis is estimated to be 1:5,000 [2]. The mtDNA copy number is maintained at approximately 1,000 to 10,000 copies per diploid cell [3]. Given that there are multiple copies of mtDNA and that mutant mtDNA is commonly heteroplasmic (Figure 1A, see Glossary), several studies have shown that even a small percentage of wild-type mtDNA is protective against the biochemical defects associated with mtDNA mutations [4]. Often, heteroplasmy levels must surpass 80–90% mutant mtDNA for a biochemical phenotype to be observed [4].

FIGURE 1. The use of mitochondrial genome editors for the treatment of mitochondrial genetic disorders.

(A) Within the same cell, the sequence of some mtDNA molecules, denoted as mutant (MT), diverge from the wild type (WT) ones, usually by a single nucleotide polymorphism (SNP). This concept is called heteroplasmy and can be quantified by determining the percentage of WT versus MT alleles. (B) One approach to increase the amount of WT mtDNA molecules consists of introducing a MT-specific nuclease capable of linearizing MT mtDNA molecules. These linearized fragments will be subsequently degraded and removed from the pool of mtDNA molecules available for replication. (C) An alternative approach directly reverts the mutation of interest to generate WT mtDNA molecules from MT mtDNA molecules with the use of an appropriate mitochondrial-targeted DNA base editor.

Currently, there are no effective therapies for mitochondrial DNA disorders. There are palliative treatments, but none that address their cause, namely, the high levels of mutated mtDNA. This concern kickstarted an interest in mitochondrial genetic engineering with the development of the first bacterial restriction enzyme engineered to get imported into the mitochondrial matrix [5]. This proof-of-concept was then followed up by the development of novel technologies that tried to address the main roadblocks that limit the overall efficiency of a mitochondrial specific nuclease: they must be as specific as possible so that their activity does not induce an undesired mtDNA copy number depletion [6]. Furthermore, to be applied to patients, the safest strategies require the use of adeno-associated virus (AAV)-based vector delivery approaches [7–9]. Despite these challenges, the field has advanced by leaps and bounds in the last decade. These breakthroughs have rekindled the field of mitochondrial genetic engineering, taking it closer to clinical use.

In this review, we summarize the current state-of-the-art of mitochondrial genome editing approaches (Figure 1B–C) and highlight their current applications, advantages, and disadvantages (Table 1). They are grouped into different categories, according to the conceptual approach they are based upon:

TABLE 1:

Comparing the different features of mitochondrial genome editors

	mitoRE	mitoARCUS	mtZFN	mitoTALEN	mitoTev-TALE	mitoCRISPR	mitoDdCBE	TALED
Mechanism	Nuclease	Nuclease	Nuclease	Nuclease	Nuclease	Nuclease	Base editor	Base editor
Edit type	Depletion	Depletion	Depletion	Depletion	Depletion	Depletion	C>T	A>G
DNA binding domain	RE dependent	I-CreI based	ZF motif	TALE repeat	TALE repeat	PID [76]	TALE repeat	TALE repeat
Targetable sequences	−/+	+++	++	+++	++	++++	+	+
Heterodimeric	No	No	Yesb	Yes	No	No	Yes	Yesb
Multiplex editing	No	No	No	No	No	Yes	No	No
“Protein-only”	Yes	Yes	Yes	Yes	Yes	No	Yes	Yes
Evidence that works in mitochondria	Yes	Yes	Yes	Yes	Yes	Needs Confirmation	Yes	Yes
Minimum construct size a	∼1.2kb	1.2kb	∼1.3kb (x2)	∼2.8kb (x2)	∼2.7kb	∼4.2kb	∼2.5kb (x2)	<2.6kbc (x2)

^(a)Only accounting for the elements represented in Figure 2 and six ZF motifs or 13.5 TALE repeats, respectively. In this case-scenario, the resulting DNA binding domain would recognize either 18bp or 13–14bp per array, respectively.

^(b)As depicted in Figure 2, a slightly larger but monomeric alternative was also reported.

^(c)The longest architecture, mTALED, is depicted here.

NUCLEASE-BASED APPROACHES (mito-nucleases)

This approach is based on recognition and cleavage of specific mtDNA regions, but not actual sequence editing (Figure 1B). These tools rely on the fact that when there is a mtDNA double-strand break, the cleaved molecules are not repaired, but rather rapidly degraded [5, 10, 11]. Accordingly, if mutant mtDNA is specifically eliminated, the residual mtDNA (mainly wild-type) will replicate and repopulate the organelle to restore normal mtDNA levels in a cell-type dependent way [12, 13]. The overall performance of these systems is limited by the amount of off-target double-strand breaks that they produce in wild-type mtDNA, even though heteroplasmy shifts can still be accomplished, if the on-target double strand breaks are produced at a higher rate [6]. Nonetheless, the off-target activity on mtDNA should still be minimized, as much as possible, since it correlates with higher levels of mtDNA depletion after treatment [6], something with a deleterious cellular [14] and and multisystemic phenotype [15].

A few laboratories have focused on developing tools based on the nuclease approach to reduce the mutant mtDNA load. These tools include:

Restriction Endonucleases (mitoRE)

Infrequently, mtDNA mutations generate restriction sites that are not present in the wild-type mtDNA. This concept led investigators to reengineer restriction endonucleases to carry a Mitochondrial Targeting Signal (MTS) (Figure 2A). This approach has been already successfully tested in vitro, ex vivo, and in vivo mouse models of mtDNA heteroplasmy [16–20].

KEY FIGURE 2: The architecture of the different mitochondrial genome editors.

All the mtDNA editing platforms have added to N-terminus of their constructs a Mitochondrial Targeting Signal (MTS). Once in the mitochondrial matrix, their DNA binding domains (depicted in different shades of blue or red) will bind to their corresponding target DNA region (shaded in the same color). When required, this phenomenon will catalyze the dimerization and thus activation of the catalytic domains exclusively at the target site. (A) homodimeric Restriction endonucleases. (B) Heterodimeric mtZFN. A Nuclear Export Signal (NES) was also introduced to suppress the nuclear mis-localization of the constructs. (C) TALE-based nucleases. mitoTALEN is heterodimeric whereas I-TevI is monomeric and only digest target sites that follow a 5′-CNNNG-3′ motif. (D) mitoARCUS in monomeric as the original homodimer was fused into one polypeptide chain. (E) The applicability of CRISPR to mtDNA editing is still controversial. Even though most of the CRISPR binding affinity is determined by the sequence of the gRNA spacer (colored in red), the PAM-Interacting Domain of the Cas protein (PID, also in red) will further expand the targeted sequence to include an extra motif known as Protospacer Adjacent Motif (PAM). This PAM is species-specific. (F) Heterodimeric DddE can only deaminate with high efficiency cytosines that are preceded by a thymine. The UGI is an inhibitor of the mitochondrial Base-Excision repair pathway (mtBER). (G) DddA_tox is required for the system to work with the highest efficiency possible. Any given A can be targeted within the central region of the spacer sequence.

As a proof-of-concept, mitoPstI was the first mitoRE to be successfully introduced into mammalian mitochondria [5], since in the human and mouse wild-type mtDNA there are two restriction sites for this enzyme. Following this report, SmaI was then repurposed so as to target the pathological m.8399G single-point-polymorphism (SNP) that generates the corresponding restriction site in the mutant mtDNA, which is associated with the NARP and Leigh syndromes [21]. Similarly, to validate and define the kinetics of this approach in vivo, MitoApaLI was also generated. This system successfully produced heteroplasmy shifts in the NZB/BALB mouse model that carries two different haplotypes (NZB and BALB), which can be distinguished by the presence (BALB) or absence (NZB) of an ApaLI restriction site [19]. Using the same model, both a systemic AAV gene delivery approach [16] and oocyte mRNA microinjection [20] have successfully induced an heteroplasmy shift towards the NZB haplotype. In a similar fashion, mito-XhoI [22] and mito-EcoRI [23] have also been used to cleave mtDNA.

Unfortunately, this strategy depends on spontaneous mutations creating unique restriction sites, something which hardly ever happens spontaneously. An exception is the aforementioned m.8399T>G mutation, which creates an unique SmaI site in the human mtDNA [24]. As successful as they may have proven to be, their stringent sequence requirements make them the least versatile gene editing approach in our current mitochondrial gene editing toolbox.

Zinc-Finger nucleases (mtZFN)

The restriction enzyme FokI has independent DNA-binding and cleavage domains. Therefore, it is possible to change the binding domain to modify the targeted DNA sequence without negatively impacting the performance of its unspecific catalytic site. Libraries of Zinc fingers (ZFs) with predetermined DNA-binding specificities have been thus generated using computational design and phage display, which can be tested against custom target DNAs. Each engineered ZF can recognize a specific trinucleotide, and the targeted region can be enlarged by linking 3 – 6 consecutive ZFs. Furthermore, as the FokI catalytic domain was modified to be an obligatory heterodimer, two sets of dimers are designed and directed to neighboring DNA sequences in inverted strand orientation. This way, when both sets bind to their respective target sequence, they subsequently dimerize and cleave the region in between both targeted sequences, which is usually designed to be 5–7bp long [25]. Nevertheless, a monomeric version has also been developed by connecting two consecutive FokI domains with a flexible linker (Figure 2B) [26].

However, the affinity and specificity of each individual ZF are highly context dependent. Because of this, a significant proportion of the predicted in silico dimer candidates are, in practice, non-functional. Therefore, they must be empirically screened in a time-consuming process to select the best performer for each individual DNA region to target [27, 28]. Despite these technical difficulties, The Minczuk lab in Cambridge has taken advantage of this platform to target mtDNA [6] and successfully developed the so-called mtZFNs. Besides an MTS, they also added a nuclear export signal (NES) to a custom ZF array targeting the m.8993T>G [29], the common deletion [30] in culture and the m.5024C>T [6] in vivo (Figure 2B).

This system has been tested in vivo in the m.5024C>T tRNA^Ala mouse model. For this, the constructs were packaged using the cardio-tropic AAV9.45 serotype [31] and delivered by intravenous (IV) tail injection into 2- to 8-month-old male mice. The mice were sacrificed 65 days post-injection and showed a successful induction of heteroplasmy shift in the heart. As a direct consequence of this phenomenon, the levels of tRNA^Ala, lactate and pyruvate were restored and, with it, the physiological function of heart mitochondria was effectively re-established.

TAL-Effector nucleases (mitoTALEN, mitoTev-TALE)

A different platform for the customization of DNA binding domains was readapted from the transcription activator-like effectors (TALEs) expressed by plant pathogenic Xantomonas proteobacteria species [32]. Their DNA binding domain is a modular and highly repetitive region in which a series of 33–35 consecutive amino acids, also referred as TALE repeats, are concatenated with modifications on the 12^th and 13^th amino acids. The amino acids at these positions, also known as the Repeat Variable Di-residue (RVD), determine the nucleotide affinity for the whole repeat. By choosing the appropriate RVDs, it is thus possible to synthetize a DNA binding domain for any given DNA sequence.

Despite the cloning challenges associated with the synthesis of repetitive DNA sequences and the fact that TALE-derived DNA binding domains are bigger than their ZFN-based counterparts, multiple mitochondria-targeted TALE nucleases (mitoTALEN) have been developed. Each monomer contains an MTS, the custom TALE DNA binding domain, and the unspecific (but obligatory heterodimer) FokI cleavage domain (Figure 2C). This approach was successful in changing heteroplasmy in cultured cells for m.14459G>A, the “common deletion” [11], m.8344A>G, m.13513G>A [33], m.9176T>C [20] and the m.13513G>A in iPSC [34].

However, the dimeric architecture of mitoTALENs makes the two corresponding coding sequences long. In addition, genes coding for mitoTALENs have highly repetitive sequences even after degenerating their codon usage, something which can lead to sequence instability [35]. Moreover, in vivo experiments require the injection of two viral preps, each with one mitoTALEN monomer, diluting the viral titer that is critical for in vivo transduction. Nevertheless, in vivo testing has been successfully performed [10]. Using the same m.5024C>T tRNA^Ala mouse model, a pair of mitoTALEN heterodimers were packaged on two separate AAV9 viral vectors and delivered by intramuscular or retro-orbital injection (i.v.) into young mice. As soon as 4 weeks post-injection, shifts in mtDNA heteroplasmy were observed in the skeletal muscle and the heart and the levels of tRNA^Ala were consequently restored.

Successful attempts to target mitochondrial mutations with this system have been performed on mouse embryos [20]. In the heteroplasmic NZB/BALB mouse model and by using the appropriate set of mitoTALENs, the levels of NZB mitochondrial genomes were specifically degraded in M-II oocytes so that they were not transmitted to the next generation anymore. Similar attempts were made after successfully introducing the human m.14459G>A and m.T9176T>C pathological mutations into mouse oocytes and then targeting them with the corresponding set of mitoTALENs. Of special note, following the mRNA microinjection and mitoTALEN expression, there was no subsequent recovery of the total levels of mtDNA as oocytes and pre-implantation embryos are not capable of replicating mtDNA [36]. The production of non-viable embryos after a mito-nuclease treatment is thus a concern that must be considered for oocytes with high mutant loads. Despite this, in comparison to the mitochondrial replacement techniques currently available [37], they are not as challenging to perform from a technical perspective, and they also avoid the ethical concerns that the presence of donor mtDNA generates [37].

MitoTev-TALE was designed as a monomeric alternative of the mitoTALEN strategy [38] that can be delivered with just one AAV-derived viral preparation [39]. In contrast with mitoTALENs, mitoTev-TALEs do not have a FokI domain as the DNA cleaving catalytic domain. Instead, they have the monomeric I-TevI homing nuclease which can generate DSBs within the mtDNA at any close-by 5’-CNNNG-3’ sequence (Figure 2C). This alternative successfully reduced the mutation load present in patient-derived cybrid cells carrying the m.8344G>A mutation and improved their overall oxidative phosphorylation function [38].

I-CreI Homing endonuclease (mitoARCUS)

MitoARCUS was recently developed to address the barriers encountered in the platforms discussed above. The ARCUS gene-editing platform, developed by Precision Biosciences, is based on a heavily reengineered and monomerized I-CreI homing endonuclease that can be rewired to recognize and cut almost any DNA sequence through in silico and directed evolution techniques. A mitoARCUS specific to the m.5024C>T tRNA^Ala has been delivered to a mouse model that carries the mutation to assess whether it could induce mtDNA heteroplasmy shift in vivo [8]. The AAV9-mitoARCUS construct (Figure 2D) was injected retro-orbitally in either 2.5- or 6-weeks old mice and sacrificed at 6 to 24 weeks post-injection. Heteroplasmy levels were then assessed in the heart, tibialis anterior, quadriceps, gastrocnemius, kidney, and spleen. The greatest shift was observed in the liver, and this may be due to the strong hepatic tropism shown by AAV9 [40]. Furthermore, tRNA^Ala levels, analyzed in this same organ, were successfully restored.

CRISPR-based nucleases (mitoCRISPR)

The CRISPR system, a widespread bacterial viral resistance pathway [41], was first repurposed for nuclear gene editing in 2013 [42, 43] due to its capability to recognize and cleave DNA in a sequence-specific manner. Two key components make this system: the Cas protein and the guide RNA (gRNA). Cas is an unspecific endonuclease that remains inactive in the absence of a small RNA known as gRNA. In its wild-type version, the gRNA is made of two pieces of ssRNA that form a complex with a T-shape comprised of one tetraloop and three stem loops. Later on, both RNAs were fused together with a linker to create a synthetic alternative that will adapt the same secondary structure as the wild-type complex [43]. The gRNA 5′ end is designed to be complimentary to the target DNA sequence. The gRNA binds to the Cas protein inducing a conformational change that converts the inactive protein into its active form. The introduction of these two components into the cell nucleus promotes a site-specific double-strand break. Furthermore, the Cas protein can also be genetically engineered to be inactive and fused with any other non-specific enzyme to restrict their activity to a gRNA-determined site-specificity [44]. In a similar way, the gRNA can be further reengineered to include extra sequences with multiple applications [44].

Unfortunately, this system is not readily usable to modify the mammalian mitochondrial genome since it is not widely accepted whether RNA can be imported into human mitochondria [45, 46]. Still, multiple reports have claimed that mitoCRISPR could work in cultured human cells [47–51], isolated yeast mitochondria [47], yeast [52] and in zebrafish [48]. For this, the Cas9 is always linked to an MTS instead of a Nuclear Localization Sequence (NLS), whereas the gRNA is either reengineered to include a putative RNA import motif sequence or simply left unmodified (Figure 2E). Strikingly, none of them has yet shown heteroplasmy shift, they instead focus on demonstrating a mitoCRISPR-specific mtDNA depletion. The lack of this critical piece of information, in conjunction with the inconsistencies in the methodology, used to introduce the gRNA in the mitochondrial matrix and the lack of any follow-ups makes this approach not yet widely accepted by the research community.

If successful, the mitoCRISPR system would be substantially simpler to administer as a gene therapy agent, compared to the existing “protein only” platforms. Furthermore, it is the only system that can be directed at more than one target site at a time, something which may prove useful in basic and translational research.

BASE-EDITOR APPROACHES

Although mito-nucleases can specially target and eliminate mtDNA copies carrying deleterious mutations, they cannot correct mutant genomes. Because of this, such strategies will not work for the the rescue of pathological, homoplasmic mutations. A different approach is thus necessary for modifying the mitochondrial genome in this scenario (Figure 1C).

In stark contrast, there are a number of well-developed strategies currently available for nuclear genome editing capable of inducing either C>T or A>G transitions at any given custom DNA region, without inducing double strand breaks in the DNA [53]. These constructs rely on either a cytidine or adenosine deaminase domain, respectively, to induce a deamination that, if left unrepaired, will eventually lead to the aforementioned mutation after one cycle of DNA replication. Unfortunately, as they currently stand, such domains can only work on single stranded regions of DNA [54]. This poses a serious hurdle for mitochondrial gene editing as our currently available protein-only, customizable DNA binding domains (ZFN, TALE) do not unwind the double-stranded mtDNA. For the generation of a functional mito-base editor, it is thus necessary to either identify new DNA binding domains that have this ability (e.g mitoCRISPR) or new catalytic domains that can work on dsDNA.

DddA-derived cytosine base editor (mitoDdCBE)

In 2020, work led by the Labs of David Liu at the Broad Institute and Joseph D. Mougous at the University of Washington developed a new class of base editor that is not dependent on the CRISPR architecture. Instead, it is based on the identification of DddA, an unusual deaminase from Burkholderia cenocepacia, a Gram-negative bacterium. The catalytic domain (DddA_tox) deaminates cytosine residues in double-strand DNA, an essentially unheard property for a nucleic acid deaminase up until now [54]. DddA_tox was split into two halves that were not active, but if physically close to each other, could assemble and deaminate cytosines in double-stranded DNA. They fused each half of the catalytic domain to a mitochondrial-targeted TALE monomer, to direct the binding of DddA_tox to specific mtDNA sites. This way, the DddA_tox is reconstituted and shows deaminase activity within the spacer region in between both TALE DNA binding domains, which is usually 14–18bp long. Any cytosine within this region, regardless of which DNA strand they stand at, could be de-aminated by this base editor. However, the catalytic domain works with the highest efficiency only for cytosines preceded by a thymine and that are approximately 4–7 nucleotides upstream of both 3′ ends of the spacing region [55].

The final synthetic proteins, named DddA-derived cytosine base editor (DdCBE), consisted of an MTS, a custom TALE DNA binding domain, one complementary half of the split deaminase and an uracil glycosylase inhibitor (UGI) domain that increases the overall performance of the system by inhibiting the mitochondrial Base-Excision Repair (BER) pathway [56] (Figure 2F). In the HEK293T cell line, these two sets of heterodimers did not localize to the nucleus and successfully induced the desired C>T transition in all the five different mtDNA cytosines that were tested for [55]. More recently, the same group evolved different forms of the enzyme with more relaxed requirement for a “T” preceding the target “C” [57].

This approach was also shown to edit mtDNA in mouse embryos [58]. In these experiments, mouse oocytes were microinjected with two different sets of mito-DdCBEs that targeted the wild-type version of the ND5 mitochondrial gene. By doing so, different mtDNA mutations were created at this gene, among them the pathological m.1291G>A and m.12336C>T. They then implanted the resulting embryos into surrogate mothers to generate new mitochondrial mouse animal models that could potentially reproduce human disorders like the Leigh syndrome, MELAS or LHON. Unfortunately, these new mice lines did not show any obvious phenotype, which may need higher levels of heteroplasmy.

This approach has also been applied to rats to induce the ortholog mutations corresponding to the human pathological m.8363G>A and m.14710G>A [59]. In this case, no editing was observed unless the construct was transfected as a plasmid instead of a mRNA. Whereas the m.8363G>A model was not fully characterized, the m.14710G>A mutation was passed on to the off-spring and, compared to wild-type rats, showed mild molecular, locomotive, and cardiac phenotypes.

In zebrafish, three different loci were targeted instead [60]. The corresponding edits were orthologs to the human pathological m.3733G>A, m.13513G>A, and the same m.8363G>A generated in rats. By injecting mRNA encoding the construct in one-day-old zygotes, they successfully generated zebrafish carrying the desired edits, capable of transmitting them to their off-spring. Furthermore, the animal models for m.3733G>A and m.13513G>A showed defective motility and abnormal mitochondrial morphology. Surprisingly, the animal model of m.8363G>A did not show any obvious phenotype even in individuals that carried mutation loads higher than 80%. Using the same technical approach, a non-sense heritable mutation (m.10215C>T) at the mt-co3 zebrafish gene and the m.3739G>A at a conserved site of the mt-tl1 zebrafish gene were successfully induced with pleiotropic molecular consequences in 7 days-post-injected larvae [61]. Other uncharacterized, non-sense mutations in the human MT-ND2 and MT-ND4 genes were also introduced in HEK293T cells [61].

This base editor system also has been successfully delivered into post-natal mice [62]. Using a viral preparation packed with the cardiotropic AAV9.45 serotype and a systemic injection as the delivery strategy, they transduced heart tissue in 8-week-old and 1-day-old mice with one set of mitoDdCBEs to induce two different deleterious edits, m.9576G>A and m.9577G>A, at the MT-Nd3 locus. Consequently, mitoDdCBE-specific editing was observed twenty-four weeks post-injection in adult mice and three weeks post-injection in neonatal mice. Even though no mtDNA depletion nor nuclear off-target effects were observed by qPCR or Next Generation Sequencing approaches, respectively, at any point after treatment, widespread mitochondrial off-target editing ranging from 3 to 7-fold increase from the untreated controls was indeed detected.

As a cautionary note, a report of mitoDdCBE-derived nuclear off-targeting in mouse embryos has been reported. Genome-wide Off-target analysis by Two-cell embryo Injection (GOTI) were performed on Ai9 mice after the mRNA injection of two different sets of mitoDdCBEs that targeted the MT-Nd5 locus and induced the deleterious m.12918G>A and m.12336G>A edits, respectively [63, 64]. While spurious, sequence-independent mitochondrial off-target effects were detected again. However, the most concerning observation was the 50 to 75-fold increase in nuclear SNP variant calling for the samples treated with mitoDdCBEs versus the negative controls. Furthermore, these variant calls were only enriched of C-to-T/G-to-A conversion and 5’-TC-3’ motif in the presence of mitoDdCBE. The aforementioned SNPs were evenly spread across the nuclear genome and did not overlap across the different technical replicates.

Because of the limiting target requirements, most pathogenic human mtDNA mutations cannot yet be corrected by DdCBE [65]. Furthermore, the system performance is also context dependent. Therefore, the editing efficiency and off-target effects are mainly driven by the orientation of the split deaminase and the amount of off-target TC pairs also found within the spacer region and must be screened empirically in a case-by-case scenario. Bystander modifications are also a critical concern shared with their nuclear base-editor counterparts that must be accounted for [53], as they may prove to have unwanted, pathological consequences. Furthermore, the nuclear mislocalization that this technology shows must be addressed before moving forwards. To reduce the frequency these off-targets, future iterations of this system may include higher fidelity versions of DddA_tox and a NES in their sequence.

TALE-linked adenine deaminases (TALED)

Following the DdCBE discovery, the first ever mitochondrial A-to-G base editor, termed TALED, has been reported [66]. This technology has three different architectures available, ranging from one monomeric set-up and two dimeric alternatives (Figure 2G). Nevertheless, all of them covalently fuse a MTS, a TALE DNA binding domain and the TadA-8e protein [67]. This last component is an adenine deaminase genetically engineered from the E. coli TadA enzyme that can induce A-to-G transitions in ssDNA. For reasons that remain to be elucidated, one set of DddA_tox domain (preferably catalytically inactive), is required for the system to work with high efficiency. Furthermore, this base editor does target any given A within the spacer region irrespective of their context or strand, preferring those placed in the middle section of it rather than the edges.

As a proof-of-concept, this novel system was tested in different human cell lines, where it edited 17 different target sites within the mitochondrial genome [66]. Even though most of these edits did not show any mitochondrial phenotype, the one targeted at the gene MT-RNR2 generated a mutation that caused chloramphenicol resistance. Furthermore, in the presence of the drug, the edited cell line eventually became homoplasmic for the mutation.

NUCLEIC ACID-BASED APPROACHES

Further groundbreaking gene editing approaches are possible if a reliable way of introducing synthetic nucleic acids into the human mitochondrial matrix is uncovered. In the current literature there are reports of RNA import into mitochondria, including: mRNAs [68], anti-replicative RNAs [69] and ssDNAs [48, 50]. These were reported to occur without the use of any carriers into human mitochondria. Furthermore, any kind of nucleic acid, from DNA [70] to anti-sense RNA [71], mt-tRNA [72], mt-rRNAs [73] or mt-mRNA [74] could, in theory, be imported with nanoparticle carriers or the synthesis of a nucleic acid–protein hybrid [75]. However, these have either not been well documented or yet confirmed by different labs to work ex vivo or in vivo, and therefore, the jury is still out. Even if mitochondrial genome engineering may not be CRISPR-Ized [45] in the near future, mitoCRISPR may be developed as a in vivo monitoring system for mitochondrial RNA import, by examining mtDNA depletion, heteroplasmy shift or acquisition of drug resistance. Until then, “protein-only” platforms will remain as the gold standard.

CONLCUDING REMARKS

For the development of gene therapy approaches for mitochondrial diseases, different mitochondrial genome editing tools and delivery strategies have emerged. In this review, we have discussed all currently available platforms that have successfully modified the mitochondrial genome ex vivo and in vivo. Disregarding some niche mutations at the MT-RNR2 locus which confer specific drug resistances, the lack of proper selectable markers has been a major limitation to mitochondrial manipulations in many systems, hence the assessment of mtDNA depletion, heteroplasmy shift or mitochondrial dysfunction phenotype reversal as endpoints of interest for the characterization of the performance of these technologies. Recent exciting breakthroughs have opened a brand-new approach to modify the mitochondrial genome by base editing. Further optimization of this platform will make it more generally applicable to pathogenic mtDNA mutations. At the same time, improvements in gene delivery aimed at neuromuscular targets would facilitate the translation of these advances to clinical trials. Despite the remaining barriers and limitations (see Outstanding Questions), genetic manipulation of mammalian mtDNA in vivo is now a reality.

OUTSTANDING QUESTIONS.

• How can gene delivery efficiencies be improved in vivo?

• With the use of mito-nucleases, which strategies would reduce their off-target effects without negatively impacting their on-target activity?

• With the use of mitochondrial base editors, which strategies would reduce the amount of off-target, by-stander modifications?

• Do the new animal models carrying mitochondrial mutations reproduce any actual human mitochondrial dysfunction phenotype?

• Can RNA be efficiently imported into human mitochondria?

• Can base editors be modified to expand specificity?

HIGHLIGHTS.

• Approximately 1000 – 10000 copies of the mtDNA can be found per cell. Molecules may carry deleterious mutations that will induce into a mitochondrial disorder if there are not enough wild-type copies to compensate for the defect.

• There are no efficient methods to directly modify the mtDNA in vivo. However, proteins can be delivered to mitochondria by expressing genes coding a mitochondrial target sequence in the nucleus.

• It is possible to introduce specific mitochondrial-targeted nucleases in the mitochondrial matrix to produce double-strand breaks and degrade the pathological molecules. Different strategies have been developed and validated in vivo.

• We are now able to express a mitochondrial-targeted base editor to directly modify the mtDNA in vivo.

GLOSSARY

Bystander modifications

Off-target edits within the spacer region of a base editor. They receive this special name because the base editor architecture does not allow for narrower spacer regions that only contain one potential targetable position, making these off-target effects something expected to happen unless demonstrated otherwise

Heteroplasmy

Cell state in which different copies of mtDNA encoding distinct sequences for the same mitochondrial region co-exist

Homoplasmy

Cell state in which all the copies of mtDNA encode the same sequence for any given mitochondrial region

NARP

Acronym standing for Neuropathy, ataxia, and retinitis pigmentosa syndrome. It is a multisystemic mitochondrial disorder whose clinical presentation negatively impacts the nervous system the most

Pleiotropic

Variable phenotypic consequences

SNP

Acronym standing for Single Nucleotide Polymorphism. It is a minor allele whose only difference with the major allele is one base call. It is the most common genetic variation reported

TALEs

Acronym standing for Transcription Activator-Like Effectors. Xanthomonas bacteria produce these proteins to recognize and activate the expression of specific host genes to assist their infection