Cellular mechanisms of mtDNA heteroplasmy dynamics

Abstract

Heteroplasmy refers to the coexistence of more than one variant of the mitochondrial genome (mtDNA). Mutated or partially deleted mtDNAs can induce chronic metabolic impairment and cause mitochondrial diseases when their heteroplasmy levels exceed a critical threshold. These mutant mtDNAs can be maternally inherited or can arise de novo. Compelling evidence has emerged showing that mutant mtDNA levels can vary and change in a nonrandom fashion across generations and amongst tissues of an individual. However, our lack of understanding of the basic cellular and molecular mechanisms of mtDNA heteroplasmy dynamics has made it difficult to predict who will inherit or develop mtDNA-associated diseases. More recently, with the advances in technology and the establishment of tractable model systems, insights into the mechanisms underlying the selection forces that modulate heteroplasmy dynamics are beginning to emerge. In this review, we summarize evidence from different organisms, showing that mutant mtDNA can experience both positive and negative selection. We also review the recently identified mechanisms that modulate heteroplasmy dynamics. Taken together, this is an opportune time to survey the literature and to identify key cellular pathways that can be targeted to develop therapies for diseases caused by heteroplasmic mtDNA mutations.

Introduction

Mitochondrial diseases are unique because they can be caused by mutations in either the nuclear or mitochondrial genomes (Gorman et al. 2015, 2016; Frazier et al. 2019). In contrast to the ~1500 mitochondrial proteins encoded by the nuclear genome, human mtDNA encodes only 13 essential protein subunits of the electron transport chain and ATP synthase, which carry out oxidative phosphorylation (Calvo and Mootha 2010; Rath et al. 2021). In addition, the human mtDNA also encodes 22 tRNAs and two rRNAs (16S and 12S), comprising the RNA components of the machinery required for mitochondrial mRNA translation (Anderson et al. 1981; Bibb et al. 1981). Despite this modest genetic contribution, mutations in mtDNA play an outsized role in causing mitochondrial diseases. The prevalence rate of diseases due to mtDNA mutations is estimated to be approximately one in 5000 (Gorman et al. 2015). In contrast, nuclear mutations are estimated to cause mitochondrial diseases in 2.9 per 100,000 individuals (Gorman et al. 2015). Taken together, mutations in mtDNA appear to impose a greater mitochondrial disease burden than nuclear mutations.

The mtDNA represents a unique paradigm for genetic diseases. Human mtDNA is in general, maternally inherited and present at high copy number, with a typical eukaryotic cell containing hundreds to thousands of individual mtDNA molecules (Shokolenko and Alexeyev 2015). New mutations can occur frequently, contributing to a state of heteroplasmy, in which two or more sequence variants of mtDNA (i.e. wild-type and one or more mutants) coexist (Wallace and Chalkia 2013). Due to the high copy number of mtDNA, mutations are not scored as either homozygous or heterozygous, but rather as a percentage of overall mtDNA copies carrying a given mutation, often termed the mutant or heteroplasmy frequency. At low levels, mtDNA mutations can have negligible phenotypic effects. However, rise of the mutant frequency beyond a critical threshold (typically 60–90%) can lead to pathogenicity and the phenotypic manifestation of mitochondrial disease (DiMauro and Moraes 1993; Rossignol et al. 2003) (Figure 1 shows how heteroplasmy increase correlates with pathogenicity). The threshold is usually mutation-specific and can vary from cell to cell or between tissue types (Rossignol et al. 2003; Zhang et al. 2018). Additionally, the replication of additional mtDNA copies is not tightly coupled to the cell cycle (Falkenberg 2018). Consequently, mutant mtDNA frequency can vary over time within an individual, accounting for a wide range of phenotypic effects and disease severities (Calloway et al. 2000; Wallace 2005; Tuppen et al. 2010).

Figure 1.

Schematic of mitochondrial heteroplasmy and consequences for organismal fitness. When present at low intracellular levels, even deleterious mtDNA mutations tend to have negligible phenotypic effects. However, due to organelle turnover and relaxed mtDNA replication, heteroplasmy levels can vary over time and between cells. Consequently, a previously negligible mutation can become pathogenic at higher levels, particularly upon crossing a critical threshold (dashed line).

It has long been recognized that the heteroplasmy frequency of mtDNA mutations can undergo dramatic shifts within one generation from mother to offspring, and amongst somatic tissues of an individual (Howell et al. 1992; Jenuth et al. 1996; Greaves et al. 2014; Li et al. 2015). One explanation for this phenomenon is provided by the concept of the genetic bottleneck that is thought to occur during maternal germline transmission of mtDNA and during somatic cell development (Zhang et al. 2018). The phenomenon of genetic bottleneck posits that only a restricted number of randomly sampled mtDNA molecules seed new cells, in either oocytes or somatic tissues (Brown et al. 2001; Zhang et al. 2018; Barrett et al. 2020). This inheritance of a sub-set of mtDNA molecules could lead to rapid and random segregation of mtDNA variants leading to differential levels of heteroplasmy frequency between siblings and amongst tissues (Jenuth et al. 1996; Chinnery et al. 2000; Brown et al. 2001). In addition to the random segregation, biased mitochondrial biogenesis or degradation in favor of one mtDNA variant over another can cause nonrandom shifts in mtDNA heteroplasmy (an overview of the mechanisms that favor selection and propagation of mutant versus wild-type mtDNA molecules is shown in Figure 2) (Yoneda et al. 1992; Dunbar et al. 1995; Jenuth et al. 1997; Stewart, Freyer, Elson, Larsson 2008; Stewart, Freyer, Elson, Wredenberg, et al. 2008; Malena et al. 2009; Carelli et al. 2015; Stewart and Chinnery 2015; Knorre 2020). These nonrandom shifts are highly dependent on the cellular context and environment (Chinnery et al. 1999; Moreno-Loshuertos et al. 2006; Latorre-Pellicer et al. 2019). The focus of this review is on summarizing our current understanding of the selection mechanisms that modulate heteroplasmy within cells and individuals. We will first review findings from model systems Caenorhabditis and Drosophila, which have led the charge of providing perhaps most insights into these mechanisms. We will also discuss what can only be described as heroic efforts carried out using mouse models that provided early evidence for the hypothesis that heteroplasmy frequency is subject to dynamic modulation. We will review recent studies in humans, made possible due to the advent of whole genome sequencing, which provide exciting evidence for the active modulation of heteroplasmy dynamics. Finally, we end by discussing the development of strategies that aim to reduce mutant mtDNA burden.

Figure 2.

Diverse cellular mechanisms alter mtDNA genotype in heteroplasmic animals. Multiple mechanisms aim to ensure mitochondrial quality control and to respond to mitochondrial stress. Unfortunately, when mutations within the mitochondrial genome serve as the source of mitochondrial stress, some of these mechanisms can result in a vicious cycle that favors the propagation of the mutant genome. For example, the cell initiates continued rounds of mtDNA replication in order to establish a suitable wild-type mtDNA population to sustain respiratory function; since the replication machinery may be blind to mtDNA genotype, deleterious mutations can hitchhike to higher frequency and escape the regulation of mtDNA copy number. Moreover, mechanisms that function to ameliorate physiological stress, such as mitochondrial UPR and FoxO, can promote the proliferation of deleterious mtDNA mutants, either through a vicious cycle of compensatory (stress-induced) mitochondrial biogenesis or by protecting mutant genomes from the targeted degradation of underperforming organelles. Additional mechanisms for maintaining quality can function to bias the propagation of mtDNA in favor of wild-type, functionally competent genomes. For example, mitochondrial fission enables the sequestration of dysfunctional regions of the mitochondrial network, which can expose deleterious mtDNA to degradation machinery. Moreover, in contrast to stress-related mitochondrial biogenesis that can favor mutant mtDNA by replicating the genomes of underperforming organelles, mitochondrial biogenesis can also occur in a manner that preferentially localizes to respiratory-competent organelles, favoring the biased proliferation of wild-type genomes.

Caenorhabditis

C. elegans has many features that make it an ideal model system to study germline transmission of heteroplasmic mtDNA mutations. First, the mtDNA is highly conserved between humans and C. elegans—12 out of 13 protein-coding genes, along with 22 tRNAs and two rRNAs found in human mtDNA are also present in the C. elegans mtDNA (Okimoto et al. 1992). Second, its short generation time of three days along with a large brood size of ~300 progeny makes it possible to track shifts in heteroplasmy frequency across multiple generations and individuals. Third, approximately 90% of the total mtDNA content of an adult hermaphrodite is contained within the germline syncytium, the common cytoplasmic pool shared by the multinucleated germline (Bratic et al. 2009). This highlights the fact that the germline is a major site of mtDNA replication in C. elegans. Indeed, the germline undergoes more than 30-fold expansion in mtDNA copy number during development (see Figure 3 for an illustration of the C. elegans germline) (Tsang and Lemire 2002a, 2002b).

Figure 3.

Differences in maternal germline development have implications for mitochondrial genome dynamics. Cells of the mammalian germline (left) proliferate during development of the female embryo before entering a period of dormancy. These cells will serve as the supply of oocytes later in life. Following birth and weaning, oocytes produced during embryogenesis develop and mature within ovarian follicles. Mitochondria produced in conjunction with germline proliferation in developing embryos enter a quiescent state, thought to preserve mitochondrial quality and minimize mtDNA mutations in oocytes (Hayashi et al. 2020). In contrast to mammals, the female germlines in invertebrates Drosophila (upper right) and Caenorhabditis (lower right) continue to proliferate during sexual maturation and adulthood. In Drosophila, oocytes arise from germline stem cells that proliferate within the germarium, before being segregated into individual egg chambers along with a group of nurse cells. In hermaphroditic Caenorhabditis, germline nuclei proliferate near the distal end of the germline and develop into mature oocytes as they migrate toward the spermatheca for fertilization. Many germline nuclei will export their organelles into the syncytium—a contiguous pool of cytoplasm shared by all germline nuclei—before undergoing apoptosis, thereby functionally serving as nurse cells for the nuclei destined to become mature oocytes. In both Drosophila and Caenorhabditis, mitochondrial biogenesis continues to occur in conjunction with the continued proliferation of oocytes, enabling age-dependent shifts in the mitochondrial genotype of the female germline in a manner that differs from mammals.

Generated via chemical mutagenesis, a strain harboring a mutant mtDNA variant uaDf5 with a 3.1 Kb deletion has provided most insights into the modulation of heteroplasmy dynamics in C. elegans (Tsang and Lemire 2002a, 2002b). We and others showed that uaDf5 is subject to negative or purifying germline selection—uaDf5 levels are lower in oocytes and embryos compared to the levels in the maternal germline (Ahier et al. 2018; Gitschlag et al. 2020). As approximately half of the germline cells undergo apoptosis (Gumienny et al. 1999), a question naturally arises as to whether apoptosis provides the means to select against mutant mtDNA uaDf5. However, this possibility seems unlikely given that germline cells export their mitochondria into the syncytium before apoptosis occurs (Raiders et al. 2018). Alternatively, there may be selection against organelles with uaDf5. Consistent with the possibility that PINK1/parkin-dependent mitophagy selects against mutant mtDNA, deletion of the parkin homolog pdr-1 results in elevated levels of uaDf5 (a description of the genes implicated in the cellular mechanisms reported to modulate heteroplasmy dynamics in C. elegans, Drosophila, and mouse are described in Table 1) (Valenci et al. 2015).

Table 1.

Genes involved in cellular mechanisms of heteroplasmy dynamics across organisms.

Organism	Gene name	Acronym(s)	General cellular function	Type of selection
Caenorhabditis elegans (nematode)	PTEN-induced kinase	pink-1	Mitophagy	Favors wild-type mtDNAa
	Parkinson’s disease related	pdr-1
	Activating transcription factor associated with stress	atfs-1	Mitochondrial unfolded protein response (UPRmt)	Favors mutant mtDNAb,c
	Abnormal dauer formation	daf-16	Stress sensitive fork-head domain-containing transcription factor	Favors wild-type or mutant mtDNAd
	Abnormal dauer formation	daf-2	Insulin-like growth factor 1 receptor	Favors mutant mtDNAd
Drosophila melanogaster (fruit fly)	PTEN-induced putative kinase 1	Pink1	Inhibits local protein synthesis	Favors wild-type mtDNAe,f,g
	Mitochondrial assembly regulatory factor	Marf, mfn, mitofusin	Mitochondrial fusion	Favors mutant mtDNAg,h
	DNA polymerase gamma subunit 1	PolG1,tamas,tam	mtDNA replication	Favors mutant mtDNAi
	Dynamin related protein 1	Drp1, dynamin-2	Mitochondrial fission	Favors wild-type mtDNAh
	BCL2 interacting protein 3	BNIP3	Mitophagy	Favors wild-type mtDNAh
Mus musculus (mouse)	Supercomplex assembly factor 1	Scaf1,Cox7a2l	Superassembly of complexes III and IV	Favors one mtDNA variantj
	Nicotinamide nucleotide transhydrogenase	Nnt	Production of NADPH in mitochondria	Favors one mtDNA variantj
	OMA1 zinc metallopeptidase	Oma1	Mitochondrial fragmentation and clearance	Favors one mtDNA variantj
	GTPase, IMAP family member 3	Gimap3	ER-localized GTPase	Favors one mtDNA variantk
	GTPase, IMAP family member 5	Gimap5	Lysosome-localized GTPase	Favors one mtDNA variantl

^aValenci et al. (2015).

^bGitschlag et al. (2016).

^cLin et al. (2016).

^dGitschlag et al. (2020).

^eZhang et al. (2019).

^fZhang et al. (2016).

^gKandul et al. (2016).

^hLieber et al. (2019).

ⁱChiang et al. (2019).

^jLatorre-Pellicer et al. (2019).

^kJokinen et al. (2010).

^lJokinen et al. (2015).

Remarkably, uaDf5 undergoes sufficient proliferation during larval development and adult germline maturation to compensate for the initial purifying selection from parents to embryos, to the degree that adult progenies carry uaDf5 at even higher frequency on average than their parents (Gitschlag et al. 2020). These results are intriguing given that high uaDf5 levels have deleterious consequences for the host including delayed development and reduced oxygen consumption (Gitschlag et al. 2016; Lin et al. 2016; Gitschlag et al. 2020). These studies suggest that purifying selection against deleterious mtDNA variants from parents to embryos is accompanied by “selfish” proliferation—favoring deleterious mtDNA variants—during development. This also provides an explanation for the stable persistence of the mutant genome within a heteroplasmic population for hundreds of generations (Tsang and Lemire 2002a, 2002b). Another mutant variant Δctb-1 was also recently reported to persist for several generations in spite of deleterious host fitness effects (Dubie et al. 2020), suggesting that selfish proliferation may be a generalizable theme of deleterious mtDNA mutants in the C. elegans germline. Perhaps most impressively, in a closely related species C. briggsae, the mutant variant nad5Δ is observed in multiple geographically diverse populations (Clark et al. 2012), suggesting that it persists across evolutionary timescales (Estes et al. 2011).

The proliferation of uaDf5 mutant mtDNA during development in C. elegans is highly analogous to the process of clonal expansion in human muscle cells (Coller et al. 2002; Trifunov et al. 2018). Muscle biopsies from humans with mtDNA associated disease often reveal extreme heterogeneity in mutant mtDNA levels between muscle fibers (Sciacco et al. 1994; Bua et al. 2006; Rygiel et al. 2016). These data are consistent with preferential proliferation or clonal expansion of mutant mtDNA, which results in cells with high mutant mtDNA levels. If the process of mutant mtDNA proliferation in the C. elegans germline and human muscle cells is conceptually similar, then mechanistic insights gained from C. elegans may be directly applicable to humans (Picard et al. 2016).

What accounts for the proliferation of deleterious mtDNA variants during development? A number of important underlying mechanisms have been described in recent research in C. elegans, which generally center around the themes of homeostasis and stress resistance. For example, we found that wild-type mtDNA copy number is maintained within a narrower range than the copy number of deleterious mutant mtDNA variant uaDf5 (Gitschlag et al. 2016), suggesting that uaDf5 may proliferate by escaping from homeostatic mechanisms controlling copy number, in agreement with prior computational modeling (Capps et al. 2003) and evaluation of heteroplasmic mtDNA variants in humans (Durham et al. 2007). In addition to escaping from copy number homeostasis, we and others found that heteroplasmic mutant uaDf5 proliferates, at least in part, by activating the mitochondrial unfolded protein response (UPR^mt) and subsequently exploiting UPR^mt activation (Gitschlag et al. 2016; Lin et al. 2016). UPR^mt entails communication of mitochondrial dysfunction to the nucleus via “retrograde signaling,” in which the stress-activated transcription factor ATFS-1 plays a critical part by regulating the expression of nuclear genes that function to alleviate mitochondrial stress (Nargund et al. 2012). This mechanism guarantees the maintenance of mitochondrial quality control, allowing functional integrity of the mitochondrial proteome (Shpilka and Haynes 2018). The presence of uaDf5 undermines mitochondrial function in a manner that induces the activation of UPR^mt (Gitschlag et al. 2016; Lin et al. 2016). We hypothesized that by attempting to alleviate mitochondrial stress, UPR^mt activation shields uaDf5-harboring organelles from degradation via mitophagy, thereby allowing the mutant mtDNA to persist. Consistent with this idea, when UPR^mt and PINK1/parkin-dependent mitophagy are disabled simultaneously, the biased proliferation of uaDf5 is at least partially restored (Gitschlag et al. 2016). In addition to protecting uaDf5 from mitophagy, UPR^mt activation also involves the upregulation of mitochondrial biogenesis, leading to the preferential amplification of uaDf5 (Lin et al. 2016). Thus, acting in more than one way, UPR^mt has emerged as a prominent pathway that can modulate heteroplasmy dynamics in C. elegans. As all the work with UPR^mt has used only a single mutant mtDNA variant, it will be interesting to determine how generalizable are these principles across a diverse set of mutant mtDNA variants.

A recent study from our laboratory, which explored the role of nutrient status in heteroplasmy dynamics, found that mtDNA heteroplasmy dynamics are susceptible to metabolic perturbations (Gitschlag et al. 2020). We find that acting via the insulin receptor homolog DAF-2, nutrient abundance promotes mtDNA biogenesis in the germline during development, which is necessary for the proliferation of mutant mtDNA uaDf5. Interestingly however, mtDNA biogenesis is not sufficient for the preferential proliferation of the mutant genome, which requires activity of the DAF-2 target FoxO/DAF-16, a transcription factor involved in stress tolerance (Gitschlag et al. 2020). Although the mechanistic basis is unclear, a likely possibility involves the role of FoxO/DAF-16 in stress tolerance. FoxO/DAF-16 regulates numerous genes important for energy metabolism and antioxidant defense (Tepper et al. 2013; Depuydt et al. 2014; Webb et al. 2016), and is known to promote organismal survival during nutrient deprivation (Greer et al. 2007; Kramer et al. 2008; Hibshman et al. 2017). Consistent with these findings, FoxO/DAf-16 partially mitigates the cost that uaDf5 imposes on host fitness, particularly during conditions of nutrient deprivation (Gitschlag et al. 2020). Hence, by alleviating physiological stress, FoxO/DAF-16 may weaken the selective pressure against mutant mtDNA within the developing germline, thus promoting its proliferation. Together with the studies exploring the role of UPR^mt, these findings suggest that stress responses may represent a general category of mechanisms that allow mutant mtDNA to overcome the effects of purifying or negative selection. Whether such stress response mechanisms can modulate heteroplasmy dynamics in mammals remains to be fully elucidated. However, there is one intriguing study on this front (Khan et al. 2017). The study used a mouse model carrying a patient-derived dominant mutation in the mtDNA replicative helicase Twinkle. Like patients with this mutation, these mice develop mitochondrial myopathy, characterized by accumulation of mtDNA deletions in their postmitotic tissues, which causes deficiency in cytochrome c oxidase (COX) activity. The study identified mTORC1 as a master regulator of the stress responses in this mouse model (Khan et al. 2017). Inhibition of mTORC1 by rapamycin slowed disease propagation in these mice and, importantly, decreased the number of COX-deficient muscle fibers, which is usually indicative of a decrease in mtDNA deletion load (Khan et al. 2017). The data suggest the possibility that inhibiting mTORC1-dependent stress responses may decrease mutant mtDNA levels, by modulating heteroplasmy dynamics. However, we suggest performing further experiments to test other equally plausible explanations such as decreased generation of mtDNA deletions or rescue of COX activity via mtDNA-independent mechanisms.

Drosophila

Drosophila provides another powerful invertebrate model system to study heteroplasmy dynamics. Drosophila embryos are particularly suitable for microinjections, providing an effective way to create heteroplasmies via germplasm transplantation (de Stordeur et al. 1989; Matsuura et al. 1989; Ma et al. 2014). Using this technique, heteroplasmic flies with intra- and interspecific mtDNA combinations were generated in which heteroplasmy frequency changes occurred across generations in a temperature and nuclear background dependent manner (Matsuura et al. 1991, 1993, 1997; Farge et al. 2002; Le Goff et al. 2002; Ma et al. 2014). These findings suggest the existence of functional differences between mtDNA variants, which are acted upon by the different environmental and genetic contexts.

Studies in D. melanogaster have arguably provided most significant insights into the mechanisms that regulate germline purifying selection. Using microinjection technique, two groups generated heteroplasmic strains consisting of wild-type and temperature sensitive mutant mtDNA (Hill et al. 2014; Ma et al. 2014). The authors observed strong selection against mutant mtDNA at the restrictive temperature. As the decrease in mutant mtDNA levels is observed in eggs, they have concluded that selection occurs within the female germline. Unlike in the C. elegans germline where mitochondria exist in a common cytoplasmic pool until the very end stages of oogenesis, mitochondria in Drosophila segregate early during germline development (as depicted in Figure 3). Specifically, asymmetric division of germline stem cells gives rise to cells, which go on to divide without cytokinesis to form 16-cell egg chambers containing a single oocyte and 15 nurse cells (Hinnant et al. 2020). This early segregation of mitochondria into separate egg chambers may provide opportunity for cellular mechanisms to effectively assess mitochondrial quality, and then to preferentially replicate functional mitochondria. Consistent with this hypothesis, disrupting mitochondrial function inhibits mtDNA replication in the developing egg chambers (Hill et al. 2014).

How do cells recognize and inhibit mtDNA replication of dysfunctional mitochondria harboring deleterious mutant mtDNA? Xu and colleagues showed that protein synthesis, including that of the mtDNA replication machinery, on the outer mitochondrial membrane is locally regulated and important for mtDNA replication (Zhang et al. 2016). Importantly, the same group showed more recently that PINK1—known for its role in promoting mitophagy—accumulates on dysfunctional mitochondria harboring mutant mtDNA (Zhang et al. 2019). However, rather than promoting mitophagy, PINK1 inhibits local protein synthesis, including that of the mitochondrial polymerase (POLG), thereby inhibiting the replication of mutant mtDNA. Consistent with these results, rescuing local protein synthesis prevents selection against mutant mtDNA.

In addition to the biased replication of wild-type mtDNA, recent data show that the Drosophila germline is also equipped to target dysfunctional mitochondria harboring mutant mtDNA for degradation (Lieber et al. 2019). Specifically, mitochondria fragment in the developing egg chambers, setting the stage for organelles with wild-type genomes to be distinguished from those harboring mutant mtDNA (Liu et al. 2020). Consistent with this idea, decreasing fragmentation by overexpression of mitofusin or loss of fission factor Drp1 produces embryos carrying mutant mtDNA at higher frequency, whereas the knockdown of mitofusin or overexpression of Drp1 enhances selection against mutant mtDNA (Lieber et al. 2019). Further experiments showed that fragmented mitochondria harboring mutant mtDNA are likely detected because of their lower ATP levels. These fragmented mitochondria are then targeted for degradation via a BNIP3-dependent form of mitophagy, which is known to promote elimination of mitochondria in red blood cells (Lieber et al. 2019) Together, these data show that inhibition of mutant mtDNA replication and its degradation constitute alternate but complementary mechanisms underlying the purifying germline selection against deleterious mtDNA.

Mitophagy is thought to play an especially important role in somatic tissues such as the brain and muscle where deleterious mtDNA variants are otherwise expected to accumulate and lead to age-related decline (Kim et al. 2007; Carelli et al. 2015; Chen et al. 2020). Blocking autophagy slightly but significantly increases deleterious mtDNA levels in Drosophila muscles (Kandul et al. 2016). In contrast, overexpressing PINK1 or Parkin strongly selects against mutant mtDNA in Drosophila muscle cells (Kandul et al. 2016). Combined, these data suggest that while mitophagy can select against deleterious mtDNA in the somatic tissue of Drosophila, it likely plays a minor role to curb their levels, potentially owing to their short lifespan. Nevertheless, these data highlight the usefulness of Drosophila as a model to study the modulation of heteroplasmy dynamics in somatic cells via mitophagy.

Despite the presence of purifying or negative selection mechanisms in the Drosophila germline, deleterious mtDNA can overcome these selection barriers to successfully propagate, suggesting selfish proliferation like that observed in C. elegans. In a dramatic example of such selfish propagation, a deleterious mtDNA variant outcompeted a functional genome so successfully that several heteroplasmic fly stocks went extinct in 4–5 generations (Ma and O’Farrell 2016). What mechanism underlies this selfish drive? To address this question, Ma and colleagues performed an elegant deficiency screen from which they identified POLG as playing a key role (Chiang et al. 2019). Lowering the availability of POLG shifts the replication bias in favor of functional mtDNA in heteroplasmic flies (Chiang et al. 2019). These data suggest that normal POLG levels promote the maintenance of the mutant mtDNA at high frequency (Chiang et al. 2019). It is not yet clear how the mutant genome takes advantage of POLG for its preferential propagation.

Mice

Studies in mice provide some of the clearest evidence of germline purifying selection against deleterious mtDNA. The “mito-mice” was the first animal model generated carrying a 4696-bp mtDNA deletion which removes several tRNA and protein-coding genes (Inoue et al. 2000; Nakada et al. 2001). The majority of the mutant mice with high proportions of the mutant mtDNA die at ~6-months of age due to renal failure. The pups from the first litter have the same mutant mtDNA levels as their mothers, which initially suggested that the transmission of the deleterious genome is not subject to selection (Sato et al. 2007). However, there is a reduction in the deletion levels in the pups from subsequent litters of all five mothers tested, with many of the pups from the third and fourth litters showing complete disappearance of the mtDNA deletion. Oocytes from all four females examined show a similar decline in mutant mtDNA levels that is dependent on maternal age, demonstrating a role for germline selection (Sato et al. 2007). In a tour de force effort, another group also succeeded in generating a single heteroplasmic female mouse with a severe mutation in a complex I subunit (Fan et al. 2008). In this female, the deleterious mutation was present at approximately 50% heteroplasmic frequency but was transmitted at increasingly lower levels so that the mutant was completely eliminated from pups in the fourth litter. These data suggest the existence of a germline filter that rather strikingly acts in a cumulative fashion throughout the lifetime of the female. A question arises as to the nature of such filter. Two alternatives are possible given the biology of oocyte production in mice (see Figure 3). Unlike in C. elegans and D. melanogaster where oocyte production occurs throughout the reproductive life of adult females, in mice, all oocytes are generated during early development (Kerr et al. 2006; Zhang et al. 2012). Thus, one possibility is constant atresia of oocytes with the highest mutant mtDNA levels. Alternatively, intracellular selection biased against mutant mtDNA may be constantly operating in oocytes, leading to a gradual reduction in mutant mtDNA levels over time.

Not all deleterious mtDNA mutations are equally susceptible to elimination via the maternal germline (Stewart, Freyer, Elson, Larsson 2008; Stewart, Freyer, Elson, Wredenberg, et al. 2008; Freyer et al. 2012). One study showed that deleterious tRNA mutations are better tolerated and undergo purifying selection during embryonic development when compared to protein-coding mtDNA mutations, which directly impact the electron transport chain or the ATP synthase (Freyer et al. 2012). Another study used mice expressing a proof-reading defective POLG to generate random mutations (Stewart, Freyer, Elson, Larsson 2008; Stewart, Freyer, Elson, Wredenberg, et al. 2008). The authors then restored wild-type POLG and followed the fate of the accumulated mtDNA mutations over five generations, specifically analyzing a total of 1069 unique mutations from 190 animals. Consistent with the existence of germline selection, the authors observed rapid elimination across generations of non-synonymous changes in protein-coding genes (Stewart, Freyer, Elson, Larsson 2008; Stewart, Freyer, Elson, Wredenberg, et al. 2008). However, a greater number of mutations persisted in rRNAs and tRNAs. These data suggest that these are either not as deleterious as those affecting protein-coding genes, or the germline filter is not as effective in eliminating them. Regardless, data from these and other studies (Kauppila et al. 2016) are consistent with the observations from humans, which show that more than half of the known pathogenic mutations affect tRNA genes, even though tRNA genes occupy only about 10% of the mitochondrial genome (Florentz et al. 2003; Moreno-Loshuertos et al. 2011).

Recent efforts have begun to elucidate mechanisms underlying germline transmission dynamics of mtDNA heteroplasmies in mice (Latorre-Pellicer et al. 2019). In an artificially created heteroplasmy with two non-pathological mtDNA variants from C57 and NZB mouse strains, a massive analysis of 819 offspring from 43 females revealed a progressive shift in favor of C57 mtDNA as a function of maternal age (Latorre-Pellicer et al. 2019). This shift, which occurs in animals with C57 homozygous nuclear background, is markedly reduced in animals with a mixed C57/NZB heterozygous nuclear background. These data suggest that differences in the nuclear background can influence the selection for either one of the two mtDNA variants. To identify the nuclear-encoded factors involved in this process, the authors focused on genes that differ between the NZB and C57 mouse strains. They have identified supercomplex assembly factor 1 (SCAF-1), which is required for the super assembly of complexes III and IV, as being mutated in the C57 strain (Latorre-Pellicer et al. 2019). Remarkably, reintroducing a functional copy of this gene was sufficient to switch the maternal-age-dependent preference in oocytes away from C57 mtDNA and toward the NZB mtDNA. Nicotinamide nucleotide transhydrogenase (NNT), an integral mitochondrial inner membrane protein that produces NADPH in the mitochondrial matrix, is another gene that is variable between the different nuclear backgrounds. Knocking out this gene from the original C57 nuclear background results in complete ablation of the age-dependent segregation preference for C57 mtDNA in oocytes (Latorre-Pellicer et al. 2019). Having an intuitive understanding of why these genes are important for heteroplasmy modulation will require further experimentation. Presumably, however, biased differences in the rate of synthesis and degradation of the two mtDNA variants will ultimately account for intracellular shifts in heteroplasmy levels. Consistent with this idea, loss of metalloendopeptidase OMA1, involved in mitochondrial fragmentation and clearance, resulting in the loss of preference for the C57 mtDNA over the NZB variant in oocytes (Latorre-Pellicer et al. 2019). Given the role of mitochondrial fragmentation and mitophagy in modulating heteroplasmy dynamics in D. melanogaster (Lieber et al. 2019), these data speak to the broad evolutionary conservation of these mechanisms, as well as the utility of invertebrate model systems.

The effort involved to perform experiments in mice to detect nonrandom segregation of heteroplasmies in the female germline is not trivial. It requires tracking mutant mtDNA levels across generations from a large number of animals. Consequently, more studies have investigated the modulation of heteroplasmy dynamics in the somatic tissues of mice (Jenuth et al. 1997; Burgstaller et al. 2014; Lechuga-Vieco et al. 2020). In addition, the half-life of mtDNA is calculated to be on the order of days (Korr et al. 1998). As such, mice are better suited to study heteroplasmy shifts in somatic cells, given that months to years are available for such shifts to occur. Using the same “mito-mice” with a large deletion described earlier, statistically significant tissue-specific increase of the mtDNA deletion was observed in heart, skeletal muscles, kidney, liver, testis, and ovary, while little to no change in mutant levels occurs in pancreas, spleen, brain, and blood (Sato et al. 2007). One of the most striking pieces of evidence for somatic modulation comes from a study by Shoubridge group who created a heteroplasmic mouse model carrying two functional mtDNA genotypes—NZB and BALB (Jenuth et al. 1997). The two mtDNA genotypes differ from each other at 106 bases including at 15 sites that result in an amino acid change. They observed that the proportion of NZB mtDNA dramatically increased in the liver and kidney over the course of three months, in all the 37 mice analyzed (Jenuth et al. 1997). In contrast, NZB mtDNA levels were unchanged in other tissues including the brain, lung, and muscles, and even modestly decreased in spleen and blood. These data suggest tissue-specific differences in mtDNA turnover that can distinguish between the two mtDNA genotypes. To determine the genetic basis for this tissue-specific segregation, the same group crossed these heteroplasmic females in the BALB nuclear background with males from the CAST/Ei subspecies (Battersby et al. 2003). They then assessed NZB mtDNA levels in the liver, a tissue in which the NZB mtDNA levels rise from ~3% to more than 50% in animals with BALB nuclear background. Remarkably, the authors observed that one-fourth of the 326 three-month-old F2 progeny had the lowest NZB mtDNA levels ever seen in the parental strain, suggesting that the tissue-specific directional selection of mtDNA segregates in the F2 progeny (Battersby et al. 2003). Quantitative trait locus analysis showed a strong linkage to three nuclear loci accounting for 16–35% of the variance. Impressively, the group succeeded in mapping one of these nuclear loci as encoding for Gimap3, an immune-related GTPase involved in leukocyte development and survival that localizes to the endoplasmic reticulum membrane (Jokinen et al. 2010, 2015). Subsequently, it was also shown that changes in the expression of the paralogue Gimap5 affect mtDNA segregation in hematopoietic tissues (Jokinen et al. 2015). Although it is not yet clear how Gimap3 and Gimap5 function, they are the first genes identified to modulate somatic mtDNA segregation in mammals, and thus represent a major milestone.

Taking a conceptually similar approach, Burgstaller et al. created a panel of four different mouse models in which the C57BL/6N mtDNA was paired with different mtDNA variants (Burgstaller et al. 2014). The heteroplasmy data obtained from dozens of mice across different tissues and at different time-points show a diverse range of segregation patterns in postmitotic tissues such as heart and skeletal muscle. Importantly, the segregation patterns were more pronounced between mtDNA variants with larger genetic differences, suggesting that differences between variants are acted upon by the tissue environment.

One of the most difficult questions to address is whether changes in heteroplasmy in somatic tissues reflect shifts in the proportion of cells with one mtDNA type over the other in the same cytoplasm, or whether these changes are occurring intracellularly. To address this question, a study recently created chimeric animals that contained two cell populations, each homoplasmic for either the C57 or NZB mtDNA (Lechuga-Vieco et al. 2020). Data in this study clearly show no segregation bias over time in favor of homoplasmic cells with one mtDNA variant over homoplasmic cells for the other variant. These data demonstrate that any segregation bias observed in at least these heteroplasmic animals reflects intracellular dynamics. Via massive effort, the same study also showed that metabolic and genetic interventions modulate heteroplasmy dynamics in many tissues. However, there is no clear pattern to these shifts, suggesting that the underlying biology is complex and will require additional careful and detailed characterization.

Human

Investigations into heteroplasmy dynamics are naturally limited in humans, but there are recent advances. Applying next-generation sequencing technology to mtDNA from single cells showed that the proportion of non-synonymous mtDNA variants significantly decreases in the primordial germ cells as embryos progress from 4 weeks to 8 weeks of age (Floros et al. 2018). The number of mutations per base pair in tRNA genes also decreases during this time. Taken together, these data are consistent with the authors’ conclusion that purifying selection acts in the primordial germ cells to eliminate deleterious mtDNA mutations. This conclusion is further supported by a massive study that generated almost 13,000 whole-genome sequences and analyzed mtDNA sequences from over 1500 mother-offspring pairs (Wei et al. 2019). This study found that compared to common variants, rare novel heteroplasmic variants are less likely to be transmitted from mother to offspring, suggesting negative selection. Furthermore, mtDNA with non-synonymous mutations are less likely to be transmitted than those with synonymous mutations, and fewer mutations that occur at conserved sites are transmitted than mutations at nonconserved positions. Finally, the heteroplasmy frequency of mtDNA variants with mutations in rRNA is significantly lower in offspring relative to their mother. Although these data are indicative of a germline filter, it is important to note that this germline filter is not completely effective, as many deleterious disease-causing mtDNA mutations are maternally inherited (Chinnery and Turnbull 2000; McFarland et al. 2007; Tuppen et al. 2010). Why certain mtDNA mutations are effectively detected and eliminated, while others are not, is an important future goal to pursue.

Purifying selection has also been observed in somatic cells in humans. When assessing heteroplasmy levels of A3243G mtDNA point mutation from single peripheral-blood mononuclear cells, heteroplasmy levels were significantly lower in T cells (Walker et al. 2020). These data indicate that purifying selection occurs within the T-cell lineage for the A3243G mtDNA point mutation (Walker et al. 2020). Further studies will have to determine whether this selection occurs interor intra-cellularly. There is also evidence of positive selection for mtDNA variants in the somatic cells of humans, with the clonal expansion of the common deletion in skeletal muscles providing the clearest example (Holt et al. 1989; Moraes et al. 1989, 1995; Campbell et al. 2014; Trifunov et al. 2018). Furthermore, assessment of mtDNA heteroplasmy with next-generation sequencing from multiple tissues of two unrelated individuals revealed evidence for positive selection (Samuels et al. 2013). Specifically, it was found that certain mutations were present in some but not all tissues of both individuals, suggesting that these mtDNA variants have a tissue-selective advantage. Replicative advantage was proposed as the mechanism, given that these mutations occur close to the sites of mtDNA replication.

Therapeutic developments

We are still far from being able to predict with any certainty the possibility of an individual inheriting a heteroplasmic mutation. Consequently, offering genetic counseling for mtDNA-associated diseases remains difficult (Poulton et al. 2017). Gaining a better understanding of the underlying mechanisms that modulate heteroplasmy dynamics is therefore crucial. It is further possible that these mechanistic insights will reveal therapeutic strategies to reduce or eliminate deleterious mtDNA variants. For example, forcing mitophagy induction has been shown to lower mutant mtDNA levels in cell culture (Suen et al. 2010; Gilkerson et al. 2012). Having a better understanding of the mechanisms that modulate heteroplasmy dynamics is also crucial to ensure long-term success of the mitochondrial replacement therapy. This therapy involves the transfer of the nucleus from either a fertilized (pro-nuclear transfer) or unfertilized egg (maternal spindle transfer) carrying an mtDNA mutation into an enucleated egg from a healthy donor (Tachibana et al. 2018). Inevitably, there is some carry over of the original mutant mtDNA into the donor egg, posing a risk of reversion if the mutant mtDNA outcompetes donor’s wild-type mtDNA (Tachibana et al. 2013). We will have to wait to determine whether reversion occurs in children born using mitochondrial replacement therapy, which was approved in the UK in 2015 (Dyer 2015; Craven et al. 2016). However, more than one study has cited complete reversion to homoplasmy of the mutant mtDNA in embryonic stem cell lines derived from human embryos after mitochondrial replacement (Hyslop et al. 2016; Kang et al. 2016; Yamada et al. 2016; Hudson et al. 2019). We think these data warrant caution in widely adapting the mitochondrial replacement therapy until methods are developed to reliably eliminate the original mutant mtDNA.

One way to reduce mutant mtDNA is to use mitochondrially targeted TALE effector nucleases (mitoTALENs) and zinc-finger nucleases (mtZFNs) (Pereira and Moraes 2017; Jackson et al. 2020). The architecture of these technologies consists in sequence-specific DNA-binding domains linked to nonspecific heterodimeric endonuclease Fok1 domains to target specific mtDNA sequences for cleavage. Since mitochondria seem to be devoid of efficient double-strand break repair mechanisms, linearization of mtDNA molecules leads to their degradation (Nissanka et al. 2018; Peeva et al. 2018), followed by re-population of the cells with the residual wild-type mtDNA. These precision nucleases have been used to successfully shift the heteroplasmy in mouse oocytes to prevent germline transmission of specific mtDNA variants (Reddy et al. 2015). Recently, these tools have been successfully applied in vivo in mice to target a point mutation in tRNA^Ala (Bacman et al. 2018; Gammage et al. 2018). Both mitoTALENs and mitoZFNs effectively reduced mutant mtDNA levels in muscle and heart of heteroplasmic mice, which was accompanied by an increase in the tRNA^Ala steady-state levels (Bacman et al. 2018; Gammage et al. 2018). An alternative to eliminating mutant mtDNA is to perform precise base editing, an approach that would be helpful for mtDNA point mutations. Excitingly, precise mitochondrial DNA base editing was recently achieved with the use of a newly discovered bacterial cytosine (Mok et al. 2020; Lee et al. 2021). Taken together, these gene-editing technologies offer ways to prevent mutant mtDNA transmission, as well as to reduce mutant mtDNA levels in the somatic tissues of patients with mtDNA disease.

Concluding remarks

It is crucial to understand the molecular and cellular forces that can modulate mtDNA heteroplasmy dynamics as the pathogenicity and disease severity are associated with the heteroplasmic frequency of deleterious mtDNA mutations. In this review, we have summarized the current literature on these mechanisms from the most commonly used model systems. Amongst the discussed cellular mechanisms, mitophagy is likely evolutionarily conserved playing an important role in modulating heteroplasmy dynamics across metazoans. We will have to await future experiments to determine whether other mechanisms, such as the UPR^mt that modulates heteroplasmy dynamics in C. elegans, are similarly conserved in mammals. Given the importance of model systems in providing mechanistic insights, it would be fruitful to expand the study of heteroplasmy dynamics to additional models. For instance, the bivalve mollusks are particularly well-suited for this purpose because they have a naturally occurring heteroplasmy in the form of doubly uniparental inheritance of mtDNA (Milani and Ghiselli 2020; Iannello et al. 2021). Solid evidence has emerged from multiple systems showing that heteroplasmy dynamics are actively modulated by the cellular context, with underlying mechanisms just beginning to be elucidated. With the development of high-resolution live-imaging techniques, the compartmentalization and segregation of the mtDNA can be better understood (Prole et al. 2020). Together with ultra-deep mtDNA sequencing and transcriptomics analysis, it will become possible to better characterize the mechanisms of selection and mtDNA heteroplasmy dynamics. Looking forward, we anticipate the discovery of additional mechanisms with greater pace. Simultaneously, we also look forward for the application of these emerging mechanisms to prevent germline transmission of deleterious mtDNA mutations and to reduce their levels in somatic tissues of individuals suffering from mtDNA diseases.