Mitochondria are essential cytoplasmic organelles that generate energy (ATP) by oxidative phosphorylation and mediate key cellular processes such as apoptosis. They are maternally inherited and in humans contain a 16,569-base-pair circular genome (mtDNA) encoding 37 genes required for oxidative phosphorylation. Mutations in mtDNA cause a range of pathologies, commonly affecting energy-demanding tissues such as muscle and brain. Because mitochondrial diseases are incurable, attention has focused on limiting the inheritance of pathogenic mtDNA by mitochondrial replacement therapy (MRT). MRT aims to avoid pathogenic mtDNA transmission between generations by maternal spindle transfer, pronuclear transfer or polar body transfer: all involve the transfer of nuclear DNA from an egg or zygote containing defective mitochondria to a corresponding egg or zygote with normal mitochondria. Here we review recent developments in animal and human models of MRT and the underlying biology. These have led to potential clinical applications; we identify challenges to their technical refinement.
Mitochondria are energy-producing organelles in the cytoplasm of most eukaryotic cells. Mitochondria generate approximately 90% of the cell's energy requirements, in the form of ATP, through oxidative phosphorylation (OXPHOS) reactions involving the electron transport chain. Mitochondrial function and replication are controlled by mitochondrial and nuclear genomes. In humans, the mitochondrial genome is a double-stranded circular DNA molecule (mtDNA) of 16,569 base pairs encoding 37 genes, of which 13 encode proteins, 22 encode tRNAs and 2 encode rRNAs; the RNAs are required for translation of mitochondrial proteins. These products are essential for oxidative phosphorylation, and other components of the electron transport chain are encoded by 79 nuclear genes1.
Mitochondria have other important roles in cellular physiology, notably in programmed cell death (apoptosis) and steroid synthesis, although these depend on genes encoded by nuclear DNA (nDNA). Nuclear genes are also responsible for encoding structural components of mitochondria as well as factors controlling their replication, mtDNA gene expression, fusion and fission. Adult (somatic) cells typically contain a few hundred to several thousand mitochondria, each with two to ten copies of mtDNA. The number of mitochondria in each cell type reflects its energy demand; tissues such as muscle and brain have large numbers of mitochondria due to their high energy requirements. The number of mitochondria, estimated by mtDNA copy number, can also vary with the developmental stage2,3: human and mouse fertilizable oocytes have around 250,000 copies of mtDNA, each cell of a blastocyst (late-stage preimplantation embryo) has about 1,000, and there may be many fewer, from 10–200 per cell, during early development of the embryo after implantation, at the time the primordial germ cells are specified. Interestingly, direct measurements of mtDNA copy numbers in isolated mouse primordial germ cells suggest that their number may be higher (>200) than those of somatic cells at the same developmental stage4. These very low numbers may be the basis of the mitochondrial bottleneck, in which the number of mtDNA molecules transmitted at each generation is restricted2,5; only after this do mtDNA copy numbers increase in the developing fetus, in a cell-type specific manner. However, selective amplification of specific mtDNAs during oocyte development may play an alternative or additional role6.
Here we review assisted reproductive technologies that aim to reduce or prevent the transmission of mtDNA-dependent mitochondrial diseases. We first discuss the contributions that mutations in mtDNA make to a range of often severe and life-limiting diseases and the various factors that determine severity. We consider current methods to avoid or reduce the transmission of pathogenic mtDNA, before focusing on MRT–techniques to reconstruct oocytes or zygotes using nDNA from the prospective mother and normal mitochondria from a female egg donor. We review pre-clinical data concerning the safety and efficacy of these techniques, including those from model organisms and human research embryos. Finally, we consider potential ways to refine these techniques and the challenges involved. Throughout, we note the ethical and legal issues that these techniques have generated and emphasize the need for a robust regulatory environment and the importance of cautious clinical implementation.
Mitochondrial function and disease Defective mitochondria are responsible for a range of devastating diseases (Table 1). Tissues with high energy requirements, and therefore large numbers of mitochondria, are most susceptible to disease from defective mitochondria. Defects can result from mutations in nDNA required for mitochondrial function, and these nDNA mutations are inherited in the same way as other nuclear genetic traits; transmission can be avoided by the use of prenatal or preimplantation genetic diagnosis. In approximately 1 in 5,000 births (higher in fetuses), an mtDNA mutation is responsible for disease (Fig. 1). Each cell can have a mixture of mitochondria; some with a mutation in the mtDNA and some without. This mixture is known as heteroplasmy. A cell with only one type of mtDNA (either normal or abnormal) is said to be homoplasmic.
There is substantial mtDNA sequence diversity between individuals and across human populations. Individuals can be classified by their mtDNA haplotypes, corresponding to the set of variant sequences, which in turn fall into particular haplogroups, based on characteristic sets of these polymorphisms. These haplogroups arose during human evolution and, in association with migrations of populations across the globe, became enriched in certain geographic regions (Fig. 2). Large population movements can therefore be mapped according to their common haplogroup. Polymorphisms are distributed throughout mtDNA, but two hypervariable regions, HVR1 and HVR2, are located within the control region (CR), the only sizeable non-coding segment in the mitochondrial genome, that contains the D-loop and is associated with transcription and the initiation of DNA replication5 (Fig. 1).
Inheritance of mtDNA mutations The inheritance of mtDNA differs from that of nDNA: mtDNA is transmitted solely from the mother via the oocyte. In a conserved process involving concerted autophagy (mitophagy) and proteasomal degradation pathways7, sperm-derived mitochondria are eliminated after fertilization before the two-cell stage8,9,10 although alternative mechanisms, including elimination of sperm mtDNA before fertilization, may operate in the mouse11.
Thus, a woman who has a pathogenic germline mtDNA mutation, whether or not she expresses the phenotype, may pass on the mutation to most of her children. However, not all children inheriting the mutation will necessarily become symptomatic, as disease severity will depend on the particular mutation, the proportion of normal to abnormal mtDNA in each cell, and the energy requirement of each tissue; the higher the mutation load the more likely that disease will manifest. Many mitochondrial diseases become evident late or increase in severity with age. Heteroplasmy may also vary between tissues, sometimes resulting in specific manifestations of the disease according to the most affected organ or system. Homoplasmy for abnormal mtDNA is generally associated with the most severe disease, although sometimes it is compatible with life and associated with less severe symptoms. For example, Leber hereditary optic neuropathy (Table 1) is typically caused by homoplasmy for mutated mtDNA and causes blindness in early adulthood with relatively low penetrance12,13,14.
Avoidance of mitochondrial disease Currently, there are no cures for mitochondrial diseases; treatment relies on palliative support. To reduce the chance of a woman who carries abnormal mitochondria having an affected child, strategies aim to prevent or reduce transmission of the abnormal mitochondria. For those at risk, egg donation avoids transmission but is unacceptable for some because the resulting child is not genetically related to the woman undergoing treatment. Some couples may decide to remain childless rather than pursue this option, and others prefer to adopt. For some, prenatal diagnosis may be an option, but the degree of heteroplasmy in tissue samples taken by amniocentesis or chorion villus biopsy may not predict the degree of heteroplasmy in all tissues, or disease severity.
For those with nDNA mutations, preimplantation genetic diagnosis (PGD) is a reliable method to avoid the birth of a child with mitochondrial disease, because only embryos predicted to be unaffected are selected for implantation. However, in the case of mtDNA mutations, the embryo biopsy reveals only the proportion of heteroplasmy present in the biopsy, which may be high or low, but is rarely zero (Fig. 3). Thus, rather than being a means to definitively prevent transmission, PGD for mtDNA mutations is useful only as a risk-reduction strategy, where the risk can be further defined by taking into account the specific mutation and its expression in other family members. The first successful PGD for an mtDNA mutation was reported in 2006 (ref. 15), with only ten more cases reported by the end of 2016, and with varying levels of heteroplasmy deemed acceptable for embryo replacement (Table 2). Decisions about safe levels of heteroplasmy vary widely between clinics practicing PGD for mitochondrial disease since the small number of cases means that there is insufficient information on individual mutations for clear guidelines to be drawn up. Even when the level of heteroplasmy low in the biopsy sample, there is some evidence that the mutation level can become higher in the embryo's tissues in later development5,16. Clearly, PGD is not suitable for women who carry very high levels of abnormal mtDNA in their oocytes.
Strategies to replace defective mitochondria Where PGD is not appropriate, mtDNA disease transmission can be avoided in principle by maternal spindle transfer (MST), pronuclear transfer (PNT) or polar body transfer (PBT). These approaches, collectively called mitochondrial donation or mitochondrial replacement therapy (MRT), isolate from an oocyte either the maternal metaphase II spindle (in MST) or the first polar body (in PB1T), or isolate from a zygote the pronuclei (in PNT) or the second polar body (in PB2T). Thus, MST and PB1T are performed on unfertilized oocytes, while PNT and PB2T are performed on one-cell embryos (zygotes). The genome is encapsulated either in a karyoplast (chromosomes and/or nucleus, with a small amount of cytoplasm, surrounded by plasma membrane) or in a polar body and is transferred to an enucleated oocyte or zygote provided by a woman with normal mtDNA. In effect, this replaces abnormal mitochondria with the normal mitochondria of a healthy donor oocyte or zygote but allows the transmission of both parents' nDNA (Fig. 4).
These methods have been explored in animals and in vitro with human oocytes and zygotes. Very recently, they have also been applied clinically, either for the avoidance of mitochondrial disease or as a speculative treatment for fertility problems, in a small number of cases. We discuss these cases, and the risks and benefits of MRT, its potential hazards and possible ways to circumvent them.
Studies on mitochondrial donation This section considers preclinical studies of the safety and efficacy of MRT. MRT techniques are depicted in Figure 4.
Mammalian model organisms. MRT protocols have largely been made possible by developments in mouse micromanipulation, most of which have passed the important test of producing healthy offspring. They include PNT17, intracytoplasmic sperm injection (ICSI)18, MST19, PB1T20, PB2T21 and somatic cell nuclear transfer22. Mammalian models have revealed how mitochondria are inherited and maintained and how they interact with nuclear genomes.
The fate of mitochondria carrying a pathogenic mutation was first studied by fusing mouse zygotes with respiration-deficient cytoplasts (to create “cybrids”) carrying a 4,696 base-pair mtDNA deletion23. Mutant mtDNA was transmitted maternally over several generations in the resulting 'mito-mice', and its accumulation induced mitochondrial dysfunction in different tissues23,24. A second trans-mitochondrial mouse model, the 'mutator mouse', expresses a proofreading-deficient variant of the mtDNA polymerase, Polg25. The 'mutator' model suggests a role for mtDNA in aging26,27, whereas the 'mito-mouse' suggests that mtDNA mutations regulate mitochondrial diseases and metastasis but not aging28.
The divergence between human Eurasian and African mtDNAs has been mimicked in a mouse model (generated by backcrossing) in which a conplastic strain, B6-NZB, has a C57BL/6 (B6) nuclear genome and NZB/OlaHsd (NZB) cytoplasm29. B6-NZB mice unexpectedly exhibited a median lifespan extended by 16%, manifesting no decline in respiration and fewer signs of aging compared with B6 controls. The study reports that mtDNA haplotype influences mitochondrial proteostasis, metabolic syndrome and reactive oxygen species generation29. Mouse models suggest that metabolic syndrome is transmitted via mitochondria transgenerationally owing to defective mitophagy (destruction of mitochondria by the cell)30,31.
PNT has been used for reciprocal mitochondrial exchange between B6 and PWD subspecies of mice that diverged from a common ancestor over 0.5 Myr ago32 and whose mtDNAs differ by 391 single-nucleotide polymorphisms (SNPs; 2.4% of the mtDNA genome). Post-implantation development of embryos with a B6 nuclear genome and PWD mtDNA (B6-PWD) was similar to controls. Male, but not female, fertility was reduced. In contrast, embryos with PWD nDNA and B6 mtDNA (PWD-B6) were far less viable, suggesting that mtDNA genotype divergence plays a major role in reproductive isolation32.
Mitochondrial replacement has also been investigated in Macaca mulatta (Rhesus macaque). When Lee et al.33 generated macaque oocytes that were heteroplasmic (50:50) for two subspecies mtDNA variants, the resulting embryos underwent marked partitioning of the mtDNA between different blastomeres and to some extent between trophectoderm and inner cell mass (ICM) in the resulting blastocysts. Some fetuses and some stem cell lines derived from the embryos exhibited skewed mtDNA ratios, although there was no evidence of preferential mtDNA selection. This suggested that both variants functioned equally, even though the two mtDNA those of sequences were as different from each other as they were from some other primate species33.
In these MST experiments, isolated macaque oocyte karyoplasts carried over approximately 0.6% of the total number of mitochondria33. Comparisons between different MRT protocols in the mouse34 reported that the number of mtDNA molecules carried over per transfer was 359 for PB1T; 1,092 for PB2T; 34,392 for PNT and 2,318 in MST, with development proceeding near control rates in all groups34. These findings are discussed below in the context of human mitochondrial carryover elimination.
MRT in humans: research and clinical applications. Recent years have seen a number of studies of MRT using human eggs and zygotes. We discuss the assessment of the quality of the embryos generated, the derivation of stem cells, the phenomenon of 'reversion', and, finally, clinical applications.
Two of the first human MRT studies describe the use of MST in human oocytes35,36. Both offered insight into the challenges of optimizing MRT in humans, such as avoidance of premature oocyte activation (in the case of MST), optimization of karyoplast fusion, timing of procedures and visualization of key developmental landmarks. Tachibana et al.35 reported a 53% incidence of abnormal fertilization; however, the remaining embryos developed to the blastocyst stage and yielded embryonic stem (ES) cells at similar rates to controls. These ES cell lines were euploid, and donor mtDNA was undetectable. Paull et al.36 used parthenogenetic activation following MST in order to focus on the issue of mtDNA carryover of abnormal mtDNA (leading to heteroplasmy), which was detected at proportions below 1% in morula or blastocysts, and to simplify an assessment of potential mitochondrial–nuclear DNA incompatibility by avoiding a paternal genome contribution (see below). Heteroplasmy levels remained low or undetectable in stem cell lines derived from these (but see section on reversion). These two studies offered an early indication that MST was compatible with efficient preimplantation development of the human embryo and low levels of mtDNA carryover.
Early experiments using PNT in abnormal (unipronuclear or tripronuclear) human zygotes also revealed the importance of optimization37. Average mtDNA carryover was less than 2% when less cytoplasm was transferred. Despite the difficulties of working with abnormal zygotes, these experiments indicated that, like MST, PNT has the potential to prevent the transmission of mtDNA disease in humans.
Further optimization of MST38,39 and PNT40 was reported more recently. Kang et al.38 carried out MST using oocytes from individuals carrying pathogenic mtDNA mutations (causing Leigh syndrome or MELAS), in order to explore the logistics of clinical applications of MST. The fertilization rates of oocytes generated by spindle transfer were comparable to those of controls. Modified protocols reduced the rate of abnormal oocyte activation and around 75% of zygotes developed to the blastocyst stage; quality and aneuploidy rates of MST-derived blastocysts and controls were similar. Carryover in the MST embryos was consistently less than 1%. Similar observations were made when analyzing embryos generated using MST and artificial oocyte activation39. Hyslop et al.40 reported a new PNT protocol, involving the transplantation of pronuclei soon after the completion of meiosis, which was important to improve the survival rates of reconstituted embryos when using normally fertilized zygotes. Carryover in PNT blastocysts was <2% in 79% of blastocysts and none had >5% carryover. Again, the quality of PNT blastocysts was comparable with controls: quality was assessed by a number of parameters, including single-cell expression profiles.
Ma et al.41 describe the transfer of polar body 1 (PB1) from metaphase II oocytes into donor metaphase II ooplasm from which the maternal spindle had been removed (PB1T) of PB1T. Only two out of five reconstructed oocytes formed a bipolar spindle, but despite a fall in the rate of blastocyst formation following fertilization, genetic, epigenetic and transcriptomic analyses of ES cells derived from PB1T blastocysts indicated a marked similarity to controls. The feasibility of using polar body transfer as a clinical tool has already been established in mice34. However, it is worth noting that the use of PBT is currently not permitted in the UK by the regulations that were introduced into the Human Fertilisation and Embryology Act in order to permit MST and PNT (http://www.legislation.gov.uk/uksi/2015/572/contents/made).
In 2014, the data available to the expert panel convened by the UK Human Fertilisation and Embryology Authority (HFEA) showed that levels of carried-over mtDNA appeared consistently low in human blastocysts generated either by MST or PNT. However, because mitochondria are thought to be quiescent during pre-implantation stages of development, the panel recommended carefully examining the behavior of carried-over mtDNA in a context where mtDNA replication occurs, namely in ES cells derived from the embryos and in differentiated cell types obtained from these (http://hfeaarchive.uksouth.cloudapp.azure.com/www.hfea.gov.uk/docs/Third_Mitochondrial_replacement_scientific_review.pdf). All three studies summarized above38,39,40 derived ES cells from blastocysts generated by MST or PNT and, in the case of Kang et al.38 and Yamada et al.42, also by somatic cell nuclear transfer. In most cases, these lines were passaged extensively in order to determine whether the initial low levels of carried-over mtDNA were maintained after multiple rounds of cell division in vitro. Each of these studies reported that, in a minority of stem cell lines (around 15–25%), the proportion of carried-over mtDNA increased with passaging, typically reaching 100% (homoplasmy). We refer to this phenomenon as 'reversion', because in these cases the mtDNA associated with the karyoplast came to predominate. The behavior of the karyoplast-associated mtDNA was occasionally unstable, and could return to much lower levels after continued passaging.
The clinical significance of these in vitro experiments is difficult to estimate, not least because ES cells are unusual, with properties that might predispose them to the reversion phenomenon (see Box 1 for discussion of the validity of the ES cell model). Nevertheless, because they raise the possibility of increased levels of karyoplast-associated, and therefore pathogenic, mtDNA in post-implantation embryos and fetuses in clinical applications, it is important to understand the molecular basis of reversion. In these studies, there was no clear relationship between the degree of reversion and the starting level of heteroplasmy, which was consistently low, nor with the extent of diversity between the karyoplast-associated and donor mtDNA haplogroup.
Finally, while in some jurisdictions the clinical use of MRT is unlawful or its regulation non-existent or unclear43,44, there have been two reports of the modification of embryos through MRT and their subsequent transfer to women45,46. Zhang et al.45 described the use of PNT to establish a pregnancy for a 30-year-old nulligravid woman with unexplained infertility. After the transfer of five embryos, a triplet pregnancy resulted that was surgically reduced to twins. The two remaining fetuses survived only to mid-gestation, probably owing to the obstetric complications of the multiple pregnancy, rather than the PNT itself. Neither fetus had detectable levels of the maternal (karyoplast-derived) mtDNA haplotype. In addition, the use of MST to treat a case of Leigh syndrome has been reported46. Transfer of an XY euploid blastocyst following MST resulted in a live birth. The neonate had a mtDNA mutation load of 2.36–9.23% in its tissues and is reported to be healthy at seven months of age. In addition to the ethical and legal controversies elicited by this report of a live birth following MRT43, including a challenge from the US Food and Drug Administration (https://www.fda.gov/downloads/BiologicsBloodVaccines/GuidanceComplianceRegulatoryInformation/ComplianceActivities/Enforcement/UntitledLetters/UCM570225.pdf), an editorial accompanying its publication drew attention to several unanswered scientific questions, including why the quality of the patient's oocytes was low and why levels of heteroplasmy in the transferred blastocyst (5.7%) were higher than in other reported uses of MST, and rose to higher levels in some tissues of the newborn47. It is likely that more clinical applications will be reported in due course, not least due to the decision by the UK HFEA to grant a license to perform PNT to Newcastle Fertility at Life (https://www.hfea.gov.uk/about-us/news-and-press-releases/2017-news-and-press-releases/hfea-statement-on-mitochondrial-donation/). It is worth noting here, however, that UK law permits the use of MRT only to reduce the risk of mtDNA disease transmission; use of MRT to treat infertility is unlawful. The UK expert panel supported this restriction in its reports to the HFEA, given the absence of a substantive evidence base demonstrating the role of mitochondria in fertility, or the utility of MRT in treating infertility.
Challenges and solutions There are a number of technical and biological challenges to optimizing existing MRT protocols.
The reversion phenomenon. Mitochondrial haplotypes may differ in their functionality. These differences could affect the predominance of one haplotype over another in the presence of heteroplasmy following MRT, and could be a determinant of the reversion seen in ES cell models (Box 1). Both phenomena may depend on how efficiently the new haplotype interacts with co-resident nDNA48 and/or whether it confers a selective disadvantage, for example due to altered rates of replication or OXPHOS. But such effects may be difficult to predict precisely.
It has also been suggested that mtDNA haplotypes with specific D-loop polymorphisms might be amplified preferentially compared to others following MST, leading to reversion38. This could be addressed in MRT by selecting oocyte recipients with D-loop sequences similar to those of spindle-associated mtDNA, or with a known replicative advantage. However, D-loop functionality also depends on factors encoded by the nucleus, including RNASEH1 (ref. 49), which might affect the risk of reversion. Another suggestion is that perinuclear mitochondria are somehow preferentially replicated50. At present, there is no clear consensus on the relevance or mechanism of reversion. If it did turn out to be a risk factor in MRT, then methods to avoid carryover of any mutant mtDNA, which could conceivably involve its targeted destruction using mitophagy or genome editing methods51, might provide solutions (see below).
Mitochondrial–nuclear DNA incompatibility. There have been some expressions of concern that MRT may disrupt mitochondrial–nuclear (mito-nuclear) DNA interactions by generating novel combinations of mtDNA and nDNA52. The arguments behind these concerns have been contested53. There are few reports of experiments involving prescriptive interference with mitochondrial haplotypes to study human mito-nuclear DNA compatibility. In one report, transfer of fibroblast nuclei from a patient with Leigh syndrome into recipient oocytes yielded pluripotent stem cells containing wild-type (oocyte-derived) mtDNA that differed from the patient haplotype at 47 nucleotide positions. The stem cells were transcriptionally and metabolically similar to controls (in contrast to impaired oxygen consumption and ATP production in nuclear donors), indicative of normal mito-nuclear DNA compatibility, notwithstanding the haplotype difference32.
A study of human MST followed by parthenogenetic activation examined the resultant pluripotent stem cells that contained identical nDNA but were homoplasmic for different (L0 or K1) haplotypes39. This suggested that there were no major differences, implying that the maternal nuclear genome is compatible with different mtDNA haplotypes.
Some PNT protocols40 yielded blastocysts in which OXPHOS transcript levels varied but were indistinguishable (by unsupervised hierarchical clustering) from controls, regardless of whether donor karyoplast and recipient cytoplasts contained the same or different mtDNA haplotypes40.
Likewise, fertilization and blastulation rates were similar among human MST embryos with differing degrees of divergence between the karyoplast and cytoplast mtDNA haplotypes, ranging from close (six SNPs), to medium (33 SNPs), to distant (57 SNPs), supporting the conclusion that MST does not adversely affect mito-nuclear DNA interactions during human embryonic development in vitro38. These and other findings lead to the conclusion that switching nuclear genomes does not detectably affect mitochondrial gene expression38,39,40.
Human population studies The evaluation of human cohorts and populations should also yield valuable insights into mito-nuclear DNA compatibility. Analysis of 12 tissues obtained at autopsy from each of 152 individuals (who died aged three days to 96 years from a wide range of causes) revealed high frequencies of age-related heteroplasmy in different tissues, with apparent positive selection for reduced mitochondrial function in the liver54. The selective accumulation of mtDNA mutations outside the liver is indicative of tolerance in mito-nuclear DNA compatibility. Indeed, population genetics meta-analyses predict that deleterious mito-nuclear DNA interactions are unlikely to be more prevalent in individuals born after MRT than they are in normal reproduction55,56.
Population-wide whole human genome sequencing projects, such as the 100,000 genomes project (https://www.genomicsengland.co.uk/the-100000-genomes-project) and AstraZeneca's compilation of genome sequences and health records from two million people57, should provide further information about mtDNA haplotype-nDNA compatibility and mitochondrial disease. Studies focused on sequencing whole genomes of patients with mitochondrial disease and their unaffected relatives also promise to reveal disease-predisposing nDNA and mtDNA variants. One study found that potentially pathogenic mtDNA point mutations in 24.6% of patients had arisen de novo and one of the five cases analyzed in detail carried a mutation in the nuclear gene required for mtDNA replication, POLG58. These findings suggest that sequencing will uncover new pathogenic nDNA and mtDNA alleles and the mechanisms underlying de novo mutagenesis in mitochondrial disease.
Minimizing carryover of mtDNA Ensuring that no mitochondria are transferred with the patient's nuclear genome would guarantee that offspring do not inherit pathological mitochondria and hence mitochondrial disease. It would also eliminate any possibility of reversion. However, the complete elimination of carried-over mtDNA is currently challenging, and further refinements to MRT methodologies will be required to change this. These improvements are likely to reflect our growing knowledge of mitochondrial biology. The following methods may help to minimize, rather than eliminate, carryover.
Mitophagy. It may be possible to manipulate mitophagy so that carried-over mitochondria are eliminated in MRT protocols. In mammalian fertilization, mitophagy naturally removes sperm-borne mitochondria via a pathway that ubiquitylates mitochondrial components for proteasomal destruction10. Mitophagy may be restricted to cells of the male germline, as mitochondria from mouse spermatids but not hepatocytes are degraded9, which is why the process would have to be modified so that carried-over mitochondria were ubiquitylated in MRT protocols.
PBT. It has long been known that mouse genomes in first and second polar bodies can contribute to efficient full-term development20,21. Moreover, mitochondria are largely excluded from the first polar body34,59. With this in mind, recent studies in mice have compared donor mitochondrial carryover from four MRT protocols: PB1T, PB2T, MST and PNT34. PB1T offspring (F1) contained the fewest donor-derived mitochondria of the four groups, with low levels being stably transmitted to the next (F2) generation following natural mating34. Importantly, as is the case with mouse polar bodies34,60, human polar bodies contain few mitochondria60. Clearly, the low level of donor mtDNA in human first polar bodies suggests a direct basis for donor mitochondria reduction via PB1T. Moreover, unlike PNT and PB2T, PB1T does not require the generation of embryos as genome donors. However, more research will be required in human oocytes and zygotes to optimize PB1T and PB2T procedures and carefully assess safety and efficacy in both cases.
Genome editing. Several findings point to the possibility of eliminating mtDNA from donor samples in MRT. Transcription activator–like effector nucleases (TALENs) have been targeted to mitochondrial genomes of a specified haplotype in heteroplasmic mouse 1-cell embryos51, and mitochondrially targeted zinc finger nucleases (mtZFNs) can efficiently discriminate between, and selectively eliminate, mutant mitochondrial genomes in a human cell line in vitro61. One report suggests that Cas9 containing a mitochondria-targeting peptide fusion ('mitoCas9') can also effectively edit mtDNA, although it is unclear how the gRNA component enters the mitochondria or to what extent the editing efficiency depends on the targeted mtDNA sequence62. A major caveat with mtDNA destruction is the potential for off-target nDNA cleavage, and it needs to be demonstrated (via whole genome analysis) that high levels of specificity under idealized conditions translate to acceptably high levels across the spectrum of heterogeneity in clinical MRT.
Cytoskeletal disruption. Mammalian mitochondria associate with each other and with other organelles63,64, increasing the likelihood of carryover in MRT. Several reports suggest that mitochondria are interconnected via a network of microtubules63,65. It is therefore possible that microtubule-mediated connections could be undone with inhibitors such as nocodazole before MRT, although residual levels of the inhibitor could potentially disrupt chromosome segregation during the first mitotic cleavage, resulting in aneuploidy; the inhibitory effects of nocodazole are reversible, but any damage to, for instance, chromosome–spindle attachments may not be. In addition, there is a meshwork of polymerized actin filaments (microfilaments) in growing mouse oocytes66, and microfilament networks surround pronuclei in mouse zygotes67, so it is possible or likely that they also entrap mitochondria. If so, microfilament disruption with polymerization inhibitors such as latrunculin or cytochalasin B could reduce carryover in MST and PNT.
Minimizing mtDNA carryover mechanically. Refinements in micromanipulation may reduce the cytoplasmic volume (and hence mitochondrial carryover) in PNT and MST karyoplasts, for example, by 'pinching off' excess cytoplasm with the nDNA injection pipette40. Such methods may be limited if there are physical links associating mitochondria with the spindle or pronuclei, and additional micromanipulation increases the risk of damaging nDNA.
Oocytes generated in vitro. Female meiosis that includes oogenesis in vitro has been demonstrated starting from mouse ES and induced pluripotent stem (iPS) cells68. The prospect of something similar in humans would potentially address carryover by enabling the isolation of patient-derived iPS cell lines that contained no affected mitochondria, either by selection or manipulation, and the recapitulation of meiosis from these lines to yield oocytes. Indeed, iPS cell lines containing exclusively wild-type mtDNA have previously been derived through spontaneous segregation from patients with Leigh syndrome and MELAS69. However, extrapolating in vitro oogenesis protocols from mouse to human is unlikely to be trivial; in vitro maturation (IVM) of immature oocytes to produce fertilizable oocytes in the mouse70 took over two decades to translate to IVM in humans71. Indeed, extensive preclinical research will be required to assess the safety and efficacy of using human in vitro-derived oocytes, and in vitro-derived gametes more generally, before their clinical use could be contemplated; numerous ethical issues will also need to be addressed72,73.
Social and regulatory challenges The sensitivity that surrounds clinical use of MRT was exemplified by the media interest in the UK HFEA's decision to license a clinic to perform MRT (https://www.hfea.gov.uk/about-us/news-and-press-releases/2017-news-and-press-releases/hfea-statement-on-mitochondrial-donation/) and the birth of a child following treatment in a clinic in Mexico46. A number of challenges have been made to MRT on legal and ethical grounds43,74,75,76,77. These include the objection that there are safe alternatives for affected women if they relinquish the requirement for genetic relatedness. Others have focused on possible identity-altering aspects of these technologies78, notwithstanding numerous philosophical complications that affect our understanding of identity. Perhaps paramount among such objections is the fact that MRT alters the germ line79. Daughters born following MRT may transmit new combinations of mtDNA molecules to the next generation. This has caused concern because of the perceived potential for negative consequences. For this reason, the US National Academies of Science report on MRT recommended that only XY embryos should be used to establish pregnancies: male offspring generated by MRT would not transmit mtDNA and thus the 'trans-generational justice' point disappears (http://www.nationalacademies.org/hmd/Reports/2016/Mitochondrial-Replacement-Techniques.aspx). It should be noted that the exclusive selection of XY embryos would halve the pool of embryos potentially available for replacement, and therefore substantially reduce the chance of achieving a live birth following MRT.
Other technologies could also be used to prevent transmission of mitochondrial disease, including genome editing. However, it is worth emphasizing that MRT, unlike CRISPR–Cas9 genome editing, does not alter any DNA sequences or employ exogenous nucleic acids. Nevertheless, the use of any such germline therapies remains controversial, and it is likely that the ethical debates will continue as clinical usage of MRT increases. Given this, it is strongly recommended that clinical use of MRT be tightly regulated80 and that detailed data from all cases, including long-term follow-up, be made available (http://hfeaarchive.uksouth.cloudapp.azure.com/www.hfea.gov.uk/docs/Fourth_scientific_review_mitochondria_2016.pdf). It is also imperative that research into improvements in the methodology and understanding of reversion and risks of mito-nuclear DNA incompatibility continue.
Additional information Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
A.C.F.P. is grateful for support from the Medical Research Council, UK (grants MR/N000080/1 and MR/N020294/1). The authors thank Rob Taylor, Charlotte Alston, Emma Watson, Sam Byerley, Jane Stewart and Robert McFarland (Wellcome Centre for Mitochondrial Research Newcastle University and Newcastle upon Tyne Hospitals NHS Foundation Trust) for unpublished data included in Table 2.