Around 1 in 200 children born each year have mutations in the mitochondrial DNA (mtDNA)—the genome of the cellular powerhouse. In most cases this causes only mild disease, often without any symptoms. However, for about 1 in 6,500 individuals, mitochondrial disease causes serious and often fatal conditions, which include blindness, muscular weakness, and heart failure. Currently available therapies for individuals carrying mitochondrial disease are not curative, but at best supportive in nature. Mitochondrial replacement (MR), via three-way in vitro fertilization (IVF) [1–3], may be a tractable intervention to circumvent what are clearly devastating diseases. The approach differs from traditional IVF in that embryos effectively have three parents who each make a genetic contribution, with the embryo possessing nuclear DNA from the mother and father and mtDNA from a donor female that has healthy mtDNA [1–3]. The key merit of the approach is that it limits the carryover of disease carrying mtDNA, while enabling both the mother and the father to have a genetic connection to their offspring, which would otherwise be lost if standard egg donation was used. These therapies promise a way to reduce the risk of having offspring with potentially devastating mitochondrial disease and affected families eagerly await such treatments. However, MR approaches are technically challenging, have low rates of success [1–3], and are not risk free even if healthy offspring can be produced [4, 5]. Despite growing optimism based on successful mitochondrial replacement experiments [1–3], we remain ignorant of the broad effects that the small mitochondrial genome (mtDNA) might have on phenotype, both directly through its role in cellular energetics and indirectly as a manipulator or moderator of nuclear genetic events [6]. A growing literature shows that mtDNA can have wide effects on phenotypes, and that these effects are particularly prominent in males. Experiments in mice and fruit flies have shown that simply varying the mtDNA, while keeping the nuclear genetics constant, can result in major alterations to longevity, fertility, and behavior [7]. In addition, the long-held view that mtDNA was a fairly minor player on phenotype is currently being drastically overhauled: new work shows that mtDNA haplotype has direct and indirect effects on nuclear gene regulation [6]. Faced with this evidence, it seems prudent to move towards clinical application of mitochondria replacement with caution. One of the key considerations will be to determine whether outcomes from such procedures are affected by the similarity between the donor and recipient mtDNAs. A strong prediction is that as mtDNA donor and recipient mtDNAs diverge, offspring survival and health may be reduced as a consequence of the breakdown in co-evolved mtDNA– nuclear interactions [8]. The literature is already rich with examples of how mitochondrial genome variation maintained within the human population alters cellular physiology and disease susceptibility of carriers depending on the nuclear background alongside which this variation is expressed. Reinforcing these concerns is a recent study showing that mismatched mitochondria in nuclear-transfer-derived cells—a technique akin to pronuclear transfer (PNT) suggested for MR—possess alloantigenecity, triggering an immune response in a murine model [5]. Likely, establishing the principle of replacing mutant mtDNA with the most similar non-disease carrying mtDNA is a good one to aspire to, and given advances in sequencing technologies such matching would add a trivial cost to what is a significant manipulation, while likely improving outcomes for both the offspring and their parents. The other issue that requires further thought and experimentation is that a large number of mtDNA disorders appear late in life, and we simply do not yet know what the longer term effects of such replacements might be. While flies, mice, and non-human primates may give us some insights into the utility of such therapies, they are unlikely to help us know what transpires in the later life of humans. One potentially complicating issue here is that mtDNA replacement is not 100%, thus there will be potential for carryover of disease-causing mutant mtDNAs, and these may increase in frequency within an individual over time. Pre-clinical trials applying MR therapies to mammalian model systems led to embryos containing varying amounts of carryover mtDNAs [1, 2]. Perhaps the most comprehensive estimate comes from a recent comparative study applying PNT, spindle-chromosome transfer (ST), and polar body transfer (PBT1, PBT2) in parallel to a mouse model [2]. Probing for heteroplasmy (the presence of more than one mtDNA type in a cell) in F1 infants (tail tips), this study found an average carryover of over 5% for ST individuals, over 20% for PNT individuals, and 0 (i.e. below detection limit of 1%) to 2% for DOI 10.1002/bies.201500008