A potential source of mitochondrial genotypic heterogeneity is simply the quantity of mtDNA per cell. A proliferating cell is expected to increase its complement of mtDNAs by a factor of 2 over the cell cycle, and noisily partition those molecules upon division (Johnston and Jones, 2015) (Figure 1Ai). However, a number of studies suggest that cytoplasmic mtDNA density is tightly controlled within a variety of replicating cell types (although mtDNA copy number, and density, varies radically between different cell types; Wilson et al., 2001). Work in HeLa cells (Posakony et al., 1977) and budding yeast (Rafelski et al., 2012) have indicated that mitochondrial volume increases approximately in proportion to cytoplasmic volume, although we note that a recent in vivo study involving hypertrophic mouse hepatocytes suggested that mtDNA density may reduce with cell size (Miettinen et al., 2014). Further single-cell studies are required to validate this observation. Studies in proliferative human cell lines (Iborra et al., 2004; Tauber et al., 2013), budding yeast (Osman et al., 2015) and fission yeast (Jajoo et al., 2016) have shown that the distribution of inter-nucleoid spacings is significantly perturbed from random, suggesting that mtDNA density is controlled in proliferating cells. Indeed, mathematical modeling using a constant mitochondrial density (Johnston et al., 2012) was able to explain a range of single-cell data for replicating cells (Das Neves et al., 2010).

The conservation of mitochondrial density is somewhat surprising, given that mitochondrial density is a potential axis for cells to control power production in response to differing demands, especially in the context of differing cell volume. Smaller cells have a larger surface area to volume ratio, therefore power demand is not expected to scale linearly with cell volume. Mathematical modeling has suggested that cells may instead modulate their mitochondrial membrane potential, rather than their mtDNA density, to satisfy cellular demands in mammalian cells (Miettinen and Björklund, 2016; Aryaman et al., 2017a), perhaps affording the cell more control since membrane potential may change on a faster timescale than mtDNA biogenesis. The extent to which mtDNA density homeostasis holds in the absence of cell volume variation driven by the cell cycle, i.e., quiescent cells, has yet to be carefully explored (Figure 1Ai) despite its relevance for mosaic dysfunction in aging post-mitotic tissues (Kauppila et al., 2017).

In humans, a variety of nuclear mutations which induce defects in mtDNA maintenance cause mitochondrial depletion syndromes; these are severe disorders and clinically diverse in their physiological impact (El-Hattab and Scaglia, 2013). Conversely, it has been shown that increasing mtDNA copy number can rescue male infertility in mice engineered to accumulate mtDNA mutations, despite unaltered heteroplasmy (Trifunovic et al., 2004; Jiang et al., 2017). It has been hypothesized that failure to maintain homeostasis in the density of functional mtDNAs may underlie the pathology of one of the most common mtDNA mutations associated with mitochondrial disease (3243A>G tRNA mutation) (Aryaman et al., 2017b). A mathematical model of human cybrid cells with the 3243A>G mutation was consistent with a range of omics data (Picard et al., 2014), made by assuming that cells attempt to maintain mtDNA density homeostasis through cytoplasmic volume reduction, until a minimum cell volume is reached where cells undergo a switch in their metabolic response (Aryaman et al., 2017b). Indeed, assuming constant mitochondrial functionality, the study of Johnston et al. (2012) predicts that a reduction in mtDNA density results in lowered ATP concentrations, which results in lowered transcription rate (Das Neves et al., 2010). These studies highlight the potential pathophysiological relevance of maintaining mtDNA density homeostasis.

Mitochondrial DNA is replicated and degraded, even in non-proliferating tissues, which generates opportunities for mtDNA mutations to arise and proliferate. Studies of mtDNA mutation spectra in humans have suggested that point mutations predominantly arise from replication errors (Kennedy et al., 2013; Williams et al., 2013; Stewart and Larsson, 2014) as opposed to oxidative damage (Kauppila and Stewart, 2015; Kauppila J.H. et al., 2018), as is also the case for the “common” 4997 bp deletion (Phillips et al., 2017).

Finite mutation rates during replication of mtDNA are expected to give rise to a set of closely-related sequences which do not all necessarily maximize fitness (Eigen and Schuster, 1977; Nowak, 2006). Therefore, at the intra-cellular level, we expect to observe mtDNA sequence diversity (see e.g., Jayaprakash et al., 2015). Recent experimental work in primary cultures of mouse neurons and astrocytes has shown this to be the case (Morris et al., 2017). The authors found 3.9 ± 5.7 single nucleotide variations (SNVs) (± standard deviation) per mitochondrion (Morris et al., 2017), with a mitochondrion expected to contain around 5 molecules of mtDNA (Satoh and Kuroiwa, 1991). The authors of this study found that the distribution of allele frequencies was skewed toward 0% heteroplasmy (Morris et al., 2017), suggesting the existence of negative selection acting at the intra-cellular level (Birky et al., 1983). However, several moderate/high impact mutations with >90% heteroplasmy were also discovered (Morris et al., 2017), suggesting that certain mutations are able to evade intra-cellular selection and reach high levels of heteroplasmy. Further studies are required to determine whether such high-heteroplasmy variants become established through neutral drift, or whether a positive selection mechanism exists; for example, it is possible that heterogeneity in the intra-cellular environment establishes intra-cellular niches which favor different mtDNA sequences. We use the term “microheteroplasmy” to denote abundant ultra-low heteroplasmic mutations within single cells (Figure 1Aii).

Whilst the pathological implications of microheteroplasmy remain unclear, comparisons could be drawn to experiments where a naturally occurring, but foreign, mtDNA haplotype is introduced into a cell with which it has not co-evolved. Increasing genetic distance amongst haplotypes has been shown to induce tissue-specific selective pressures in mice (Burgstaller et al., 2014), and heteroplasmy between otherwise healthy haplotypes have been shown to induce fitness disadvantages (Acton et al., 2007; Sharpley et al., 2012), as corroborated by bottom-up mathematical modeling (Hoitzing et al., 2017a). Recent work has shown that such fitness disadvantages may be mediated through oxidative damage (Bagwan et al., 2018) (Figure 1Aiii).

Another consequence of the finite mutation rate of mtDNA, and its stochastic turnover, is that mutations will occasionally drift to very high level heteroplasmies simply by genetic drift (Birky et al., 1983; Ewens, 2004), i.e., “macroheteroplasmy” (Figure 1Aiii). Cell-to-cell heterogeneity in heteroplasmy (heteroplasmy variance) therefore widens through time. Certain deleterious mutations may therefore drift to high levels of heteroplasmy and cause pathology (Rossignol et al., 2003) (see Figure 2). Theoretical work predicts that heteroplasmy variance increases approximately linearly, and increases with the rate of mitophagic turnover of mtDNAs (Poovathingal et al., 2009; Johnston and Jones, 2016), but decreases with total mtDNA copy number with a non-linear dependence upon mean heteroplasmy (Johnston and Jones, 2016). A linear increase in heteroplasmy variance with time has recently been observed in mouse oocytes, and over pup lifetimes (Burgstaller et al., 2018). Mathematical modeling has suggested that increasing heteroplasmy variance might increase the energetic cost of maintaining a tissue (Hoitzing et al., 2017a).

Increases in heteroplasmy variance can also purge mitochondrial mutations when combined with selection, as observed between generations in the human mitochondrial bottleneck (Johnston et al., 2015; Burr et al., 2018; Floros et al., 2018). During development, through a reduction of cellular mtDNA copy number, heteroplasmy variance is increased. When combined with a selective apoptotic threshold, whereby cells above a heteroplasmic threshold undergo cell death, mean heteroplasmy can be reduced. The mechanisms and timings of the mitochondrial bottleneck are debated (Jenuth et al., 1996; Cao et al., 2007; Cree et al., 2008; Wai et al., 2008); however, stochastic modeling has shown that several proposals are compatible with the induction of heteroplasmy variance through a combination of random mtDNA partitioning at division and passive turnover of mtDNA (Johnston et al., 2015; Johnston and Jones, 2016; Hoitzing et al., 2017b).

Mitochondrial mutations which survive the bottleneck can result in severe congenital diseases (Schon et al., 2012), and accelerate aging in the case of severe mutant loads in mice (Ross et al., 2013). Under physiological conditions, mtDNA mutations are also observed to accumulate with age across tissues in humans (Khrapko et al., 1999; Taylor et al., 2003; Bender et al., 2006; Kraytsberg et al., 2006; Li et al., 2015; Kauppila et al., 2017). Certain cases of mutant accumulation (or “clonal expansion”) may be explained by neutral drift, for instance partitioning noise in highly proliferative colonic crypt cells can result in highly mutated crypts (Taylor et al., 2003). However, positive selection through replicative advantage (Samuels et al., 2013) and tissue-specific niches (Avital et al., 2012; Li et al., 2015; Ahier et al., 2018) can also influence heteroplasmy dynamics. The extent to which mtDNA mutations present at birth, as opposed to somatic mutation post-development, contribute to healthy aging remains open. Bulk heteroplasmy measurements in monozygotic twins have shown that inherited mutations can contribute mitochondrial mutations, including heteroplasmic mutations with a low mutant load (Avital et al., 2012). The vast amount of DNA replication during development itself is also likely to seed and expand a number of potentially pathological mtDNA mutations (Hahn and Zuryn, 2018), as is thought to be the case in the nuclear genome (Keogh et al., 2018). Mitochondrial mutations which arise and focally expand post-development in skeletal muscle contribute to pathological age-related loss in muscle mass (Khrapko et al., 1999; Aiken et al., 2002; Bua et al., 2006; Vincent et al., 2018). In the case of human muscle fibers of healthy individuals, the number of fibers exhibiting electron transport chain (ETC) abnormalities increases from approximately 6% at age 49–31% at age 92 (Bua et al., 2006). Serial sections through laser-captured single skeletal muscle fibers showed that heteroplasmy in mtDNA deletions may exceed 90% (Bua et al., 2006). Deletion mutations have been associated with local fiber atrophy and breakage in mice (Wanagat et al., 2001; Khrapko and Vijg, 2009) and rhesus monkeys (Aiken et al., 2002) (although the causative role of mtDNA mutations in aging for shorter-lived animals is contested; Vermulst et al., 2007; Kauppila T.E. et al., 2018; Lakshmanan et al., 2018). It is noteworthy that mitochondrial dysfunction in a particular tissue is able to induce stress responses in distal tissues through the mitochondrial unfolded protein response (Zhao et al., 2002; Durieux et al., 2011; Zhang et al., 2018) and other hormonal signaling pathways (Tyynismaa et al., 2010; Khan et al., 2017), suggesting potential organism-wide consequences of focal mitochondrial mutations. The development of novel, quantitative, dyes for electron transport chain (ETC) deficiency may allow more refined measurements of the causes of deletion proliferation in somatic tissues in the future (Simard et al., 2018; Vincent and Picard, 2018).

Mitochondria have been observed to be transferred between cells (Spees et al., 2006; Torralba et al., 2016), and the intercellular transfer of other organelles has also been observed (Rustom et al., 2004). Several mechanisms by which mitochondrial transfer is achieved are known, including tunneling nanotubes (Koyanagi et al., 2005) and microvesicles (Phinney et al., 2015), amongst others (Torralba et al., 2016). Intercellular mitochondrial exchange has largely been understood through in vitro experiments (Spees et al., 2006; Jayaprakash et al., 2015; Berridge et al., 2016a); however, recent developments have shown the effect to also be relevant in vivo (e.g., Islam et al., 2012; Ahmad et al., 2014; Berridge et al., 2016a). For instance, Islam et al. (2012) have shown that mouse mesenchymal stem cells (MSCs) are able to release microvesicles containing mitochondria, and protect against sepsis-induced acute lung injury in mice. Recent evidence from xenograft mouse models have also shown that mtDNA can be transferred between cells via exosomes, mediating an escape from dormancy in therapy-resistant breast cancer cells (Sansone et al., 2017). Furthermore, astrocytes have been found to donate mitochondria to neurons after stroke (Hayakawa et al., 2016), although interpretations of these data exist which do not require mitochondrial exchange (Berridge et al., 2016b). In terms of mitochondrial genetic dynamics, intercellular exchange of mtDNA appears to slow down heteroplasmic drift (Jayaprakash et al., 2015) and thereby reduce cell-to-cell mitochondrial genetic heterogeneity. It is possible that inter-cellular exchange of mitochondria, and mitochondrial DNA, is an evolved mechanism to ameliorate heteroplasmy variance and therefore the build-up of cells with pathological levels of heteroplasmy. Although the existence of inter-cellular mitochondrial exchange has now been established, the extent and dynamics of mtDNA transfer remain incompletely understood.

Gene-editing technologies targeted at the mitochondrial genome are under development (Bacman et al., 2013, 2018; Gammage et al., 2014, 2018; Reddy et al., 2015; Pereira et al., 2018). Such technologies sequence-specifically bind and cleave mitochondrial DNA, which is subsequently degraded rapidly (Peeva et al., 2018). Mathematical modeling of mitochondrially-targeted gene therapies predicts that tissues with high mean heteroplasmy and large heteroplasmy variance are generally more difficult to treat (Hoitzing et al., 2017a); there is therefore a close link between these promising therapeutic technologies and inter-cellular heterogeneity in heteroplasmy.

The inner mitochondrial membrane (IMM) is heterogeneous in both its composition and its topology. The connection between the IMM and the mitochondrial genotype is particularly relevant since mitochondrial DNA is situated in the mitochondrial matrix and in close proximity to the IMM (Brown et al., 2011). Consequently, one might expect mtDNAs to affect their local respiratory units more than distal mtDNAs (Busch et al., 2014), thus allowing greater control of respiration at a local level (Allen, 1993, 2003; Lane, 2011). Below we discuss aspects of how IMM physiology may generate mitochondrial heterogeneity, and how cardiolipin may act as a sensor of the mitochondrial genotype. We discuss how cardiolipin reacts to differences in reactive oxygen species (ROS) production and pH, and may amplify the mitochondrial phenotype through supercomplex formation and cristal structure.

The stoichiometry of respiratory units within the IMM can vary, since respiratory units organize into supramolecular structures termed supercomplexes (Schägger and Pfeiffer, 2000; Enŕıquez, 2016) (Figure 1Biii). A variety of supercomplex stoichiometries exist in mammalian cells (Schägger and Pfeiffer, 2000; Schäfer et al., 2006) as well as free resipriatory subunits (Schägger and Pfeiffer, 2000). The “plasticity model” has therefore been suggested (Aćın-Pérez et al., 2008), whereby a landscape of combinations between respiratory complexes coexist. Dysfunction in the assembly of mitochondrial supercomplexes in mice results in decreased muscle activity and heat production in the cold due to reduced CIV activity (Ikeda et al., 2013), showing that supercomplex assembly is required for fully-functional respiration. Consequently, dysfunction in supercomplex assembly could affect mitochondrial quality control pathways and influence the mitochondrial genotype.

The topologies of mitochondrial membranes are heterogeneous (Mannella, 2006; Enŕıquez, 2016), and this heterogeneity may potentially be driven by mitochondrial genetic heterogeneity (see above), as we will discuss at the end of this subsection. ATP-synthase forms dimers in the IMM which often arrange into micron-scale rows, which are associated with high local curvature to form mitochondrial cristae (Strauss et al., 2008; Davies et al., 2012). Recent work has suggested that the mitochondrial fusion protein OPA1 stabilizes ATP synthase oligomers by modulating cristal shape (Quintana-Cabrera et al., 2018). Crista membranes show an enrichment of respiratory complexes relative to the inner boundary membrane (Gilkerson et al., 2003; Vogel et al., 2006), and it has been suggested that cristae exist to increase the packing density of respiratory units (Rieger et al., 2014). Individual cristae are morphologically heterogeneous in mammalian cells (Frey and Mannella, 2000), and can be modulated in response to altered metabolic demands (Eisner et al., 2018): cristae become narrower in mammalian cells in response to starvation (Patten et al., 2014), suggesting that cristal shape influences respiratory efficiency, perhaps by modulating local substrate concentrations (Mannella et al., 2001) (Figure 1Bii). Remodeling of mitochondrial morphology occurs during cell death (Scorrano et al., 2002; Yamaguchi et al., 2008), whereby supercomplexes and dimers of ATP synthase disassemble and cristal structure becomes disorganized (Cogliati et al., 2016), allowing the release of cytochrome c (an electron carrier of the ETC) to trigger the intrinsic cell death pathway (Taylor et al., 2008) (although the importance of crista remodeling for cytochrome c release has been questioned; Tam et al., 2010). The intimate connection between mitochondrial physiology and cell death provides insight into the recent observation that mitochondrial heterogeneity can partially explain variability in chemotherapeutic resistance in HeLa cells (Márquez-Jurado et al., 2018).

Cardiolipin is a phospholipid found in the IMM, and stabilizes both supercomplexes and cristal structure. When exposed to a pH gradient, cardiolipin-containing lipid vesicles spontaneously form crista-like membrane invaginations (Khalifat et al., 2008, 2011), thus providing a potential connection between respiratory activity, which is influenced by mtDNA genotype, and cristal shape. Furthermore, flies with deficient cardiolipin levels show reduced ATP synthase abundance in high-curvature regions of cristae, resulting in disorganized cristae, cardiac insufficiency, motor weakness and early death (Acehan et al., 2011). In yeast, cardiolipin has been shown to be necessary for supercomplex stabilization (Zhang et al., 2002; Pfeiffer et al., 2003). Indeed, patients with Barth syndrome who are deficient in cardiolipin due to a mutation in the tafazzin gene show both aberrant crista formation (Acehan et al., 2011) and reduced supercomplex formation (McKenzie et al., 2006).

Cardiolipin is particularly susceptible to damage by ROS. In isolated bovine mitochondria, ROS exposure resulted in loss of CI activity, but exogenously added cardiolipin could restore CI activity (Paradies et al., 2002). It has recently been shown that a mitochondrially-targeted antioxidant (MitoQ) is able to increase cardiolipin expression and content in liver mitochondria of rats fed on a high-fat diet, resulting in increased mitochondrial functionality and ATP synthase activity (Fouret et al., 2015). Together, this suggests that mitochondrial dysfunction, which alters pH gradients across the IMM and ROS production, can cause cristae to become disorganized and affect supercomplex assembly via cardiolipin, potentially resulting in further loss of mitochondrial efficiency and ROS production. It is possible that heterogeneity in pH and ROS production, for instance through mtDNA mutation heterogeneity (Lane, 2011), could be amplified through such mechanisms and thus strengthen the genotype-phenotype link between mtDNA and their local respiratory complexes.

The inner membrane potential (ΔΨ) is an indicator of mitochondrial functionality, generated by the ETC, which drives the synthesis of ATP by ATP-synthase. Mitochondrial output is highly sensitive to ΔΨ: a 14 mV change in ΔΨ corresponds to a 10-fold change in the maximum ATP/ADP ratio (Nicholls, 2004), where ΔΨ typically ranges between 150 and 180 mV (Perry et al., 2011). Quantification of the absolute value of ΔΨ in millivolts at a single-cell level through fluorescence probes (Perry et al., 2011) is possible but technically challenging, requiring deconvolution from other confounding factors such as fluctuations in plasma membrane potential, the matrix: cell volume ratio, dye activity and binding affinity in the matrix/cytosol, and spectral changes resulting from binding (Gerencser et al., 2012, 2016). Many studies involving ΔΨ measurements through Nernstian dyes neglect these possible confounding variables and assume differences in fluorescence are always directly attributable to differences in ΔΨ, so some caution is required. With these caveats in mind, measurements of mitochondrial mass scaled by ΔΨ has been shown to explain much of the variation in transcript elongation rate (Das Neves et al., 2010; Johnston et al., 2012) and protein noise (Guantes et al., 2015), as well as predicting phenomena such as cell cycle duration (Johnston et al., 2012) and chemotherapeutic resistance (Márquez-Jurado et al., 2018) in mammalian cells (Figure 1Bvi).

In addition to inter-cellular ΔΨ heterogeneity, individual mitochondria have been shown to undergo transient depolarization/repolarization cycles, termed “flickers” in animals (Duchen et al., 1998; O'Reilly et al., 2003) or “pulses” in plants (Schwarzländer et al., 2012) (Figure 1Biv). In freshly dissociated smooth muscle cells, flickers range from < 10 mV to >100 mV, typically lasting on the order of seconds (O'Reilly et al., 2003). It has been proposed that mitochondrial flickers are regulated by various mechanisms, including mitochondrial inner membrane fusion (Santo-Domingo et al., 2013), Ca²⁺ influx (Duchen et al., 1998; Jacobson and Duchen, 2002), and transient opening of the mitochondrial permeability transition pore (mPTP) (Hüser and Blatter, 1999; Jacobson and Duchen, 2002). The mPTP is thought to be a non-selective mitochondrial channel that induces cell death when open for prolonged periods (Bernardi et al., 2006). In rat myocytes, opening of the mPTP through pharmacological intervention has been found to correlate with the frequency of transients in circularly permuted yellow fluorescent protein (cpYFP) fluorescence (Wang et al., 2008). cpYFP has been shown to be sensitive to pH (Schwarzländer et al., 2011, 2014; Santo-Domingo et al., 2013), therefore transient cpYFP fluorescence corresponds to transient alkalinization of the mitochondrial matrix (i.e., an increase in pH gradient across the IMM). Since one would expect a reduction in pH gradient if the mPTP were behaving in a non-selective mode, transient alkalinization potentially implies an ion-selective mode of the mPTP. The coincidence of an increase in pH gradient, and loss of membrane potential, indicates a redistribution of the proton motive force, which has pH and electrical contributions (Nicholls, 2004). This redistribution may be mediated by selective ion movement. Flickers have been shown to exist in vivo in mouse astrocytes, and are enhanced by both neuronal activity and oxidative shifts (Agarwal et al., 2017), which is consistent with previous in vivo observations in mice (Breckwoldt et al., 2014, 2016). Although no clear consensus exists on the mechanisms of transient mitochondrial depolarization/alkalinization cycles, such cycles appear to be a likely means of regulating metabolic rate, and perhaps ROS production, at the single mitochondrion level.

The canonical means by which mitochondrial membrane potential feeds back into the genetic state is through mitophagy and mitochondrial network dynamics (Twig et al., 2008). Mitochondria are not static organelles but undergo dynamic fusion and fission, the purpose of which is incompletely understood (Hoitzing et al., 2015). In rat pancreas cells, it has been shown that fission often results in a daughter mitochondrion which has a lower ΔΨ than its sister (Twig et al., 2008). These depolarized mitochondria have a lowered propensity for fusion, and are more likely to be degraded (Twig et al., 2008). Selective fusion, when combined with non-selective mitophagy, is sufficient to preferentially degrade depolarized/damaged mitochondria (Hoitzing et al., 2015; Aryaman et al., 2018), although the selective strength of mitophagy itself is not yet fully understood. The extent to which mitochondria which are degraded via selective/non-selective forms of mitophagy possess mitochondrial genomes which are perturbed from the consensus sequence also remains incompletely understood (Lemasters, 2014). However, if depolarized mitochondria are more likely to be perturbed from the consensus mitochondrial sequence, mitochondrial quality control mechanisms would exert a negative selective pressure against variant alleles. Indeed, negative selection pressures have been observed in human tissues (Li et al., 2010, 2015; Avital et al., 2012; Ye et al., 2014) and at the intracellular level in mice (Morris et al., 2017), which may be due to a combination of mitochondrial networks and mitophagy.

The existence of mitochondrial diseases (Schon et al., 2012), the ubiquity of heteroplasmy (Payne et al., 2012; Morris et al., 2017), and the accumulation of heteroplasmy with age (Li et al., 2015), suggest that mitochondrial quality control may be weak for certain sequences. Building on insights from previous work (Mouli et al., 2009; Tam et al., 2013, 2015; Johnston and Jones, 2016), recent mathematical modeling (Aryaman et al., 2018) suggests that, if quality control is weak, heteroplasmy variance 𝕍(h) follows the equation

where t is time, f_s is the fraction of unfused mitochondria, μ is the mitophagy rate, n is copy number and h₀ is the initial heteroplasmy, which is equivalent to mean heteroplasmy under neutral drift (see Figure 2). This equation arises through the assumption that larger mitochondrial fragments are at a reduced susceptibility to degradation, as is observed empirically (Twig et al., 2008). As a consequence, total mitochondrial turnover is modulated by the fraction of unfused mitochondria, independently of the absolute magnitude of fusion-fission rates (Aryaman et al., 2018). Since heteroplasmy variance is proportional to mitochondrial turnover through mitophagy (Johnston and Jones, 2016), mitochondrial fragmentation may therefore modulate the rate of accumulation of pathologically mutated cells in a tissue, independently of selective effects (Aryaman et al., 2018) (Figure 1Bv). Mitochondrial network fragmentation may also slow de novo mutation (Aryaman et al., 2018) through a rescaling of mitochondrial turnover (which is known to modulate the de novo mutation rate; Poovathingal et al., 2009). As a consequence, promoting mitochondrial fusion earlier in life, when mean heteroplasmy is low, may delay the rate of accumulation of cells with pathological levels of mutated mtDNA (Aryaman et al., 2018), which may have implications for healthy aging.

Mitochondrial “complementation” consists of mitochondria sharing their contents through fusion-fission events, and potentially complementing each others' genetic defects (Hayashi et al., 1994; Yoneda et al., 1994; Enrıquez et al., 2000; Ono et al., 2001). This is supported by experiments involving photoactivatable fluorescent proteins which show that intra-mitochondrial contents mix over time (Twig et al., 2008; Wilkens et al., 2013). Inhibition of fusion and fission have also been shown to induce heterogeneity in the distribution of aged mitochondrial proteins (Ferree et al., 2013). Beyond depending on fission and fusion rates, and the degree of mitochondrial mobility, the strength of complementation is closely linked to the diffusivity of the mitochondrial matrix: if the matrix is a high-diffusivity environment then gene products are expected to be promiscuous within the matrix and not remain local to their parental mtDNA. Therefore, in cells heteroplasmic between wild-type and a pathological mutant, healthy versions of any particular transcript would be found in the matrix. The idea of complementation is also closely linked to the mitochondrial threshold effect (Rossignol et al., 2003; Stewart and Chinnery, 2015; Aryaman et al., 2017b), whereby complementation effects allow cells to withstand high levels of mutant load (60–90%, Chomyn et al., 1992; Miyabayashi et al., 1992; Rossignol et al., 2003) before displaying a respiratory defect.

Much of the progress in understanding mitochondrial complementation derives from cell fusion studies where cells harboring different mtDNA mutations are fused, and mitochondrial functionality recovers through sharing of transcription products (Ono et al., 2001; Gilkerson et al., 2008; Yang et al., 2015). Such experiments highlighted physiological subtleties which remain incompletely understood. For instance, the experiment of Ono et al. (2001) between cells harboring two different mt-tRNA mutations required an adaptive period of 10–14 days before respiratory activity was restored—the reason for this was not fully understood. Furthermore, cell fusion studies between cells harboring partially-functional (3271T>C) and completely dysfunctional (3243A>G) mitochondrial tRNA-Leu mutations resulted in a sigmoidal relationship between heteroplasmy and COX-activity, whereas a linear relationship was observed for simple co-culture between the two cell types (Ono et al., 2004), which again lacks a complete explanation. In contrast to the experiment of Ono et al. (2001), an experiment by Gilkerson et al. (2008) between two non-overlapping mitochondrial deletions showed recovery of MTCO2 protein production on a relatively fast timescale (4 days). Furthermore, Gilkerson et al. (2008) found that fused cells cycled between heteroplasmic and homoplasmic states during long-term culture, despite being on a medium which mildly selected for mitochondrial function (but still allowed ATP production through glycolysis). Yet, a straightforward interpretation of complementation would suggest that heteroplasmic states would show maximal fitness. This effect was originally explained as a tension between a neutral genetic drift effect, deriving from the assumption that nucleoids possess several mtDNAs (Jacobs et al., 2000), and selective pressure for heteroplasmy (Gilkerson et al., 2008). However super-resolution microscopy studies have revealed that nucleoids harbor only 1–2 mtDNAs per nucleoid (Kukat et al., 2011, 2015). It is possible that large drift rates could be explained by the passive partitioning of very large clusters of genetically homogeneous mtDNA at cell division. Furthermore, active asymmetric apportioning of damaged or aged mitochondria has been observed in yeast (McFaline-Figueroa et al., 2011) and stem cells (Katajisto et al., 2015) (which has been shown to influence cellular fate decisions in immune cells; Adams et al., 2016); therefore, active mechanisms of asymmetric apportioning of mitochondrial genomes may also exist. Cycles between heteroplasmy and homoplasmy highlight an area for future experimental and theoretical investigation.

Experiments by Wilkens et al. (2013) fused HeLa cells with mitochondrially-targetted fluorescent proteins of differing colors to investigate the extent of mitochondrial diffusivity. The authors found that, through cycles of mitochondrial fusion and fission, IMM proteins appear to experience slow diffusion relative to the outer mitochondrial membrane, and retain cristal structure (Wilkens et al., 2013). Consequently, complementation may be diminished since gene products may remain local to their parental mtDNA (Busch et al., 2014), allowing local control of respiration (Allen, 1993, 2003; Lane, 2011). This is supported by the observation that mitochondrial transcripts are particularly concentrated around mtDNA (Ozawa et al., 2007). Mathematical modeling of progressive increases in heteroplasmy of the pathological 3243A>G mutation is compatible with the interpretation that tRNAs are enriched in the vicinity of their local mtDNA, and that mutated mtDNAs experience a local depletion of ATP resulting in a transcriptional defect (Picard et al., 2014; Aryaman et al., 2017b). Hence, restricted complementation and local phenotype-genotype links appear to be explanatory.