MtDNA mutations are one of the hallmarks of ageing and age-related diseases. It is well established that somatic point mutations accumulate in mtDNA of multiple organs and tissues with increasing age and heteroplasmy is universal in mammals. However, the origin of these mutations remains controversial. The long-lasting hypothesis stating that mtDNA mutations emanate from oxidative damage via a self-perpetuating mechanism has been extensively challenged in recent years. Contrary to this initial ascertainment, mtDNA appears to be well protected from action of reactive oxygen species (ROS) through robust protein coating and endomitochondrial microcompartmentalization. Extensive development of scrupulous high-throughput DNA sequencing methods suggests that an imperfect replication process, rather than oxidative lesions are the main sources of mtDNA point mutations, indicating that mtDNA polymerase γ (POLG) might be responsible for the majority of mtDNA mutagenic events. Here, we summarize the recent knowledge in prevention and defence of mtDNA oxidative lesions and discuss the plausible mechanisms of mtDNA point mutation generation and fixation.
Old theories, new twists
In recent years, gerontological studies have revealed different molecular pathways involved in the ageing process and pointed out mitochondrial dysfunction as one of the hallmarks of ageing. The recognition of the role of mitochondria in ageing can be traced back to Harman’s Free Radical Theory of Ageing that highlighted oxidative damage as the most important contributor to this process . Soon after, following another suggestion by Harman , Miquel postulated mitochondria as the main source of intracellular free radicals and mtDNA as their most critical target [3,4]. Those theories were further amended by the mitochondrial free radical theory of ageing (MFRTA) or the ‘vicious cycle theory’ that assumes a self-perpetuating mechanism of oxidative mtDNA lesions, leading to global catastrophe in the cell .
Undeniably, mitochondria are the main, although not unique, source of reactive oxygen species (ROS) in a cell . Both, under normal metabolic conditions and upon respiratory chain dysfunction, mitochondrial respiratory complexes I and III generate an electron leak that can lead to ROS formation, although the amount of ROS has been largely overestimated in the past [6,7]. Nevertheless, traces of oxidative damage arising in tissues of aged animals and senescent cells have been observed and well documented in many different systems and organisms [8,9]. This includes oxidative lesions in proteins, lipids and DNA . However, the actual link in the interplay between ROS and ageing remains surprisingly correlative [7,9,11,12]. Implications of ROS as critical signalling molecules, indispensable for normal cellular physiology and environmental fitness in an organism, amended their ‘bad boys’ reputation [12–14]. Actually, through its signalling function, ROS also exert a beneficial effect on longevity, suggesting that oxidative damage might be just a hallmark, rather than a driving force of organismal decline with age [15,16].
The other important part of the MFRTA was based on the observation that mtDNA mutations and deletions accumulate with increasing age. Therefore, it was intellectually very appealing to assume that the close proximity of the major ROS-producing sites and a lack of histone protection makes mtDNA a perfect candidate for the ROS damage and resulting mutations a driving force in ageing . However, recent findings showed that mitochondria’s precious genetic material, mtDNA, is protein coated and packaged into nucleoids by mitochondrial transcription factor A – TFAM  and spatially separated from the sites of extensive ROS production via tight microcompartmentalization of the mitochondrial matrix .
Since the initial measurements, attempts to repeat the observation of increased oxidation of mtDNA have resulted in a range of measurements that span over four orders of magnitude, including values that are as low as published values for nuclear DNA . Over the years, it became apparent that minute amounts of purified DNA, as the ones usually obtained for mtDNA and used in these studies, are particularly prone to artifactual oxidation in the course of quantification . When repeated with more stringent controls, the analysis of oxidative damage showed comparable levels of oxidative lesions in mtDNA and nuclear DNA . Remarkably, the age of the tested animals did not have an effect on the amounts of DNA lesions in rat liver, suggesting that the damage does not accumulate, but was turned over or repaired . Some oxidative lesions such as 8-oxo-7,8-dihydroguanine (8-oxo-dG), 2,6-diamino-4-hydroxy-5-N-methylformamidopyrimidine (FAPy) guanosine and 5-hydroxyl cytosine were even found at a higher frequency in nuclear DNA than mtDNA in rat liver .
Mitochondria also benefit from effective ROS scavengers that are particularly prominent in pathological conditions [7,22] and possess a complex base excision repair (BER) machinery that is used to correct hydrolytic, alkylated and oxidative lesions as well as ROS-induced DNA breaks [23,24]. Still, some studies showed that in cell culture, mtDNA may be more vulnerable (than nuclear DNA) to lesions resulting from treatment with an excess of H2O2 . It seems that H2O2 treatment results primarily in strand breaks or abasic sites that lead to strand breaks , suggesting that instead of base modifications and therefore mtDNA point mutations, single-stand breaks are the main products of oxidative damage. Thus, the relationship between free radicals and mtDNA mutations is not as straightforward as often portrayed.
Still, a large body of evidence suggests a consistent accumulation of different mtDNA mutations with increasing age that comes in two different flavours: (i) large deletions that result in smaller, circular mtDNA molecules [27,28] or (ii) single base changes that can be nucleotide substitutions, base insertions or deletions , of not yet fully understood origin.
Following mtDNA mutation signatures
Historically, the most frequently detected somatic mtDNA mutations were large deletions that resulted in smaller, circular or linear mtDNA molecules, whose abundance is very low in the whole tissue, but could reach high levels in individual cells [27,28,30–32]. These deletions almost always include loss of the major arc, between the two origins of mtDNA replication and are very similar to deletions found in mitochondrial disease patients . Eighty percent of large mtDNA deletions are type I, which are flanked by two repeats, while type II deletions are not flanked by repeats, but colocalize with 2D and 3D mtDNA structures as hairpins, G- quadruplexes and cruciforms [33–36]. Current understanding is that mtDNA deletions of both the types are formed spontaneously during replication via erroneous slippage of the replication fork on structurally challenging parts of the DNA molecule [35,36] or replication stalling that leads to formation of double-strand breaks (DSBs), whose subsequent repair contributes to formation of mtDNA deletions [33,37]. Remarkably, oxidative damage was also proposed to generate mtDNA deletions through induction and repair of DSB, but this has been demonstrated only in severe, non-physiological conditions, without a direct role for the DSB repair pathway and therefore is currently considered to be largely correlative [37–39].
Another puzzling question is the mechanism of the clonal expansion of mtDNA deletions, that over the years, was explained by either replicative advantage of the smaller, deleted mtDNA molecules  or by random genetic drift . Recently, the positive selection hypothesis was largely disproved as it was shown that the rate of clonal expansion does not correlate with the size of mtDNA molecules in human muscle . Instead, mathematical modelling studies have shown that random genetic drift is sufficient for the clonal expansion of mtDNA deletions in humans . However, this can still not explain how post-mitotic tissues with a very low number of replication events, e.g. neurons in substantia nigra, accumulate large numbers of cells with a high mtDNA mutation/deletion load during the course of ageing [43–45]. Another study questioned the random drift explanation, as it showed that this hypothesis works for lifespans of approximately 100 years, but for the vast majority of animals, that are short-lived, the resulting degree of heteroplasmy is incompatible with those experimental observations . Therefore, currently no concept exists that collectively explains the observed clonal expansion of mtDNA deletion mutants in short-lived as well as long-lived animals . Recently, the positive selection hypothesis, dependent on the tight connection between transcription and priming of replication, not on mtDNA deletion size, has been put forward , and might be a missing link in the explanation of mtDNA deletion accumulation in somatic tissues.
mtDNA point mutations
It is increasingly recognized that the most common types of mutations observed in the mtDNA are single nucleotide substitutions, single base insertions or single base deletions , and that most of us carry appreciable numbers of low-level mtDNA variants in our cells . The frequency of somatic mtDNA mutations exceeds the mutation frequency of the nuclear genome by several orders of magnitude [49,50]. Accelerated development of high-throughput, next-generation sequencing techniques, allowed the elegant demonstration that age-dependent accumulation of somatic mtDNA mutations is quite universal in all the animals and different tissues [51–58]. MtDNA mutations also show a trend of accumulation in tissues that are particularly affected by age-associated pathologies . More recently, mtDNA point mutations have also been linked to stem cell fitness during ageing [60,61]. Finally, the generation of mtDNA mutator mice has provided the first direct evidence that accelerating the mtDNA mutation rate can result in premature ageing .
In the face of such a strong relationship between mtDNA point mutations and ageing, dissection of the mutation source is very important to further understand the mechanisms behind it and explore the possible treatments. The overall mutation load of the cell depends on numerous factors including, performance and precision of DNA replication, adjustment of the DNA repair machinery to mutagenic properties of the local environment, rates of hazardous insults in a cellular milieu and sufficient DNA protection against mutagens. Evidence from a recent study suggests that early to midlife mtDNA mutations are likely to be clonally expanded and therefore are much more important, than late life de novo mtDNA mutational events that make up only negligible contribution to the ageing phenotype .
In dividing cells, changes in heteroplasmy mainly occur through vegetative segregation, e.g. the unequal partitioning of different mitochondrial genotypes during cell division. In some cases, when a mutation has a severe effect on cell function, the negative selection will occur at the cellular level, as shown for the m.3243A>G mutation in the mtDNA-encoded tRNA leucine 1 gene (MTTL1), whose levels decrease exponentially in blood during lifespan .
In post-mitotic tissues, mtDNA is copied through a process of relaxed replication that ensures renewal of the mtDNA pool in non-dividing cell. In a heteroplasmic cell, a mutated molecule might be replicated more frequently than the wild-type molecule, leading over time to a change in the overall heteroplasmy level within the cell. In most cases, this neutral drift model explains pretty/quite well the accumulation of somatic point mutations in post-mitotic tissues [41,64]. The existence of recurring variants in the mitochondrial control region, some of which appear to be preferentially clonally expanded in different tissues of individuals with increasing age, have also been reported [65–67]. The discovery of tissue-specific selection of somatic mtDNA variants, lead to the hypothesis that deleterious mutations that occur on the same molecule with such a variant may be drawn to high levels via a ‘hitch-hiking’ effect with these otherwise harmless alleles . Other highly divergent mtDNA haplotypes have also been found to preferentially accumulate within tissues over time . However, their existence has to be confirmed by additional studies, including rigorous controls for the variation in the corresponding mitochondrial pseudogenes found within the nuclear DNA that may hamper the analysis .
Another hypothesis that explains mtDNA mutation accumulation suggests an increased mitochondrial proliferation activated as a compensatory mechanism to maintain an optimal amount of wild-type mtDNA . If one of the molecules is mutated, the cell responds by replicating the entire mtDNA content to restore the balance and hence proliferates both wild-type and mutated molecules, that over time can exceed the threshold levels . As the proliferation will preferentially occur in cells containing mutant mtDNA, the net effect for the tissue will be to increase the overall level of the mutation over time.
Precise assessment of each mutation source can be challenging, however, most of the mutagenic interventions leave more or less straightforward signatures that can be used to trace their true origin [71,72]. We can generally distinguish between transition and transversion mutations in DNA. Transitions assume changes of one type of purine into another (A←→G) or one type of pyrimidine into another (C←→T). Thus on a double strand template this results in (i) C:G→T:A or (ii) T:A→C:G changes. In contrast, transversions refer to substitutions of purine with pyrimidine and vice versa, therefore, four different forms of transversion mispairing on a DNA template can be distinguished: (i) C:G→A:T, (ii) C:G→G:C, (iii) T:A→A:T and (iv) T:A→G:C.
Next-generation sequencing allowed deeper insight into mtDNA mutational patterns in brain biopsies of young and aged human individuals. Predominant mutations detected in these studies are transition mutations with the most common C:G→T:A change [51,52]. In contrast, G:C→T:A transversions, that are predominant changes generated through oxidative lesions , were neither present in significant levels in mtDNA nor accumulated visibly with progressing age in studied samples [51,52]. This suggests that transitions rather than transversions should be considered as leading mutational signature of the mammalian mitochondrial genome during ageing.
Origin of mtDNA transitions
Transitions as a result of insult and repair
Lesion specificity of DNA glycosylases involved in mitochondrial BER
UNG1 (uracil-DNA glycosylase) removes uracil bases misincorporated into DNA strands and is able to remove 5-formyluracil and oxidation products of cytosine adducts from DNA, effectively preventing GC to AT transitions . The in vivo role of UNG1 in mitochondria is still not well understood, as contradicting results for  and against [79,80] its role in the maintenance of mtDNA integrity have been reported.
NTH1 (Nth-like DNA glycosylase 1) seems to be able to recognize and eliminate damaged pyrimidines like thymine glycol, cytosine glycol and cytosine hydrate, formed as a result of aerobic ionizing radiations, oxidative stress and UV irradiation, generating a free apurininc/apyrimidinic (AP) site susceptible to further BER . Nevertheless, alike UNG-deficient animals, Nth1 knockout mice are healthy, fertile and overall phenotypically indistinguishable from their wild-type littermates .
NEIL1/NEIL2 (endonucleases VIII like 1/2) can repair damaged bases, including formamidopyrimidine and 5-hydroxyuracil (Figure 1) on replicated and transcribed DNA as they exhibit activity towards DNA bubbles formed by mismatched dsDNA . In contrast with UNG1- or NTH1-deficient animals, neil1 knockout mice develop metabolic syndrome characterized by severe obesity, dyslipidaemia and liver failure . They also display elevated levels of mtDNA deletions, although the qPCR method used to assess this might be prone to artefacts . Correspondingly, depletion of NEIL2 in HEK293 cells increases the rate of single-strand breaks . These findings suggest the role of NEIL1/2 in mtDNA maintenance and integrity, independent of the removal of oxidatively damaged DNA bases.
Overall, the experimental evidence on different DNA glycosylases/endonucleases that might be involved in mitochondrial BER currently do not support their prominent role in maintenance of mtDNA integrity, although this could be due to their redundancy (Figure 1).
MtDNA transitions as endogenous errors mediated by POLG
The majority of experimental evidence points that mitochondrial mutations in human tissues are mainly generated by replication errors likely to be mediated by DNA polymerase γ (POLG) [74,85]. This preferential origin of transitions recapitulates the in vivo mtDNA mutation spectra produced by wild-type or defective POLG [52,57,86] and also overlaps with the most common changes in mtDNA (G→A and A→G) that contribute to inherited polymorphisms and disease-causing somatic mutations (see http://www.mitomap.org/). Subsequent studies revealed that a similar trait in the mtDNA mutational spectrum is present in aged human  and murine tissues, as well as in mtDNA mutator mice, a model for premature ageing .
Fidelity of DNA replication constitutes a major determinant of the genetic information stability in the cell. Proper balance between correct and incorrect DNA synthesis must be maintained by organisms to avoid mutations that can lead to development of diseases and even death. On the other hand, low DNA synthesis fidelity offers diversity of genetic information and thus allows flexibility in an organism to react with variable environmental conditions. Indeed, mtDNA fragments obtained from somatic tissues of different primates manifested a ten-fold higher rate of evolution than their counterpart sequences in the nuclear genome . The association between mtDNA polymorphic variants and a response to climate changes, susceptibility to diseases and general mitochondrial fitness has been reviewed recently .
DNA polymerases, as key enzymes involved in DNA synthesis, are one of the major determinants of faithful DNA replication. Error rates for single base substitutions obtained by proofreading-deficient polymerases range from 10–3 to >10–6, while the major proofreading-proficient polymerases from Pol A, B and C families display in vitro base-substitution error rates that vary from 10–6 to 10–8 [88–90]. Mitochondrial POLG belongs to the Pol A family of faithful polymerases with low error rates in the range of <3.8 × 10–6 (from chicken embryo; ), <2.0 × 10–6 (from pig liver; ) and 5.6 × 10–7 (human POLG; ). The high accuracy of DNA polymerases that possess intrinsic proofreading activity comes as a result of three main factors: selection of the correct nucleotide during DNA synthesis, selective extension of a well-paired primed 3′-terminus and preferential excision of incorrect nucleotides during a 3′→5′ exonuclease reaction (Figure 2) [92,93]. Remarkably, POLG proofreading activity contributes at least 20-fold to the fidelity of base selection  and its importance in vivo was highlighted by studies on mtDNA mutator mice .
DNA polymerase γ mediated point mutations on sound mtDNA
The single point mutation (D257A) introduced into the mtDNA mutator genome fully abolished the 3′→5′ POLG exonuclease activity, leading to a progressive, random accumulation of somatic mtDNA mutations during the course of mitochondrial biogenesis . As the proofreading in these mice was efficiently prevented, they developed an mtDNA mutator phenotype with a three–five fold increase in the levels of random somatic point mutations . Remarkably, these mutations accumulate at a higher rate during the time of development from oocytes to early embryonic life of mtDNA mutator mice, than during the rest of their lives when mutations accumulate in rather linear fashion .
Studies over the last decade have identified over 200 mutations in POLG that are associated mainly with neurodegenerative disease (http://tools.niehs.nih.gov/polg/) . Intriguingly, POLG mutations present in the close proximity of the exonuclease domain hardly ever associate with affected proofreading activity or accumulation of point mutations . In contrast, Y955C, the most common POLG mutation that affects the conformation of the hydrophobic dNTP-binding pocket at the polymerase active site, alters the dNTP selectivity leading to diminished fidelity of DNA synthesis . Therefore, this mutation has been linked to the occurrence of ageing-like phenotypes including Parkinsonism and premature menopause .
In the nucleus and yeast mitochondria, base mismatches generated during DNA synthesis are actively identified and removed by mismatch repair (MMR) system components . In contrast, evidence for an existing MMR in mammalian mitochondria is very subtle (Figure 2) . It has been proposed that POLG proofreading activity allows for sufficiently high fidelity of replication and that this circumvents the need for mismatch DNA repair in mammalian mitochondria . However, that does not explain why budding yeast mitochondria have canonical MMR, despite having similar POLG proofreading activity. Interestingly, although POLG has been considered as a rather high-fidelity enzyme , a number of antimutator variants of this enzyme have been identified suggesting the POLG fidelity is not optimal and might be further enhanced . Additionally, single nucleotide polymorphism variants of human POLG were shown to modulate its enzymatic performance, disease susceptibility and drug sensitivity, allowing the possibility that in vivo accuracy of mtDNA replication can vary in the human population [103–105].
Therefore, in light of all these recent findings, it becomes increasingly clear that the majority of mtDNA mutations are likely to arise as a consequence of errors during mtDNA replication with POLG as major mediator of mtDNA point mutations.
What happens with mtDNA transversions?
As already mentioned, more than 90% of all mtDNA mutations that accumulate during the course of ageing or in pathological conditions are transitions. However, the question remains: Are the oxidative lesions that contribute to transversions not generated at all due to the robust protein coating and spatial DNA separation? Are they efficiently removed from mtDNA via a specialized BER machinery? Or may be they are not visible to the mitochondrial replicase – polymerase γ (POLG) that is able to pair the DNA bases correctly despite the damage? ← do not understand this? The answer to these questions is not trivial. Most likely, all these events contribute partially to diminish transversion input in the total mtDNA mutation pool (Figure 3).
The likely fate of oxidative lesions in mtDNA
The most common oxidative lesions that lead to transversion formation are 8-hydroxypurines (including the most studied 8-hydroxyguanine or 8-OH-G) and formamidopyrimidines (FapyG and FapyA) base modifications. These base modifications may contribute to formation of DNA point mutations, while many other oxidative lesions will mostly lead to deletions or replication stalling . Purines with some OH– adducts may undergo subsequent oxidation or reduction with ring opening that drives the formation of the most common oxidative DNA lesions, 8-hydroxypurines or formamidopyrimidines respectively .
Strikingly, these two lesions are hallmarks of the nuclear genome rather than that of mtDNA . In the nucleus, the major signature of oxidative damage is 8-oxo-dG and FapyG lesion mispair with adenine leading to G→T transversion mutations [107–109]. Remarkably, inside mitochondria, 8-oxo-dG lesions seem to preferentially stall mtDNA synthesis and POLG tends to insert the adequate cytosine instead of mispairing adenine, while operating in front of an 8-oxo-dG-modified base (Figure 3) [110–112].
Mitochondria also possess quite elaborate BER repair system intended to cope with dangerous oxidative lesions, although the significance of this in physiological or pathophysiological conditions is still unclear. 8-oxoguanine DNA glycosylase (OGG1), the 8-oxoguanine DNA glycosylase of the BER system specifically recognizes and removes 8-oxo-dG adducts , while adenines mispaired with 8-oxo-dG are distinguished and eliminated by MUTYH to allow subsequent pairing . The role for both the enzymes in the nuclear BER is clearly defined. For example, MUTYH-deficient animals are more prone to spontaneous tumours due to increased nuclear DNA mutation rates . The role of MUTYH in mitochondria is far less clear; although some indications exist that its mitochondrial isoform might be important for mtDNA stability . Correspondingly, OGG1 mostly acts in the nucleus, and the majority of the seven different OGG1 isoforms present in mitochondria seem to be inactive, as they lack the DNA glycosylase activity . Again, in contrast with a severe phenotype manifested on the nuclear genome , in vivo studies showed that OGG1 deficiency neither affects overall mitochondrial function and oxidative stress levels, nor significantly influences mtDNA mutation accumulation . Likewise, a loss-of-function mutation in Ogg1 did not significantly influence the somatic mtDNA mutation frequency in flies with extremely reduced activity of manganese superoxide dismutase 2 (SOD2), a primary enzyme that detoxifies superoxide anion within mitochondria .
The findings over the last decade indicate that mtDNA seems to be well protected from oxidative damage; likely to be by protein coating that provides tight and efficient bundling. Furthermore, thanks to specific properties of POLG, oxidative lesions are not easily converted into transversion mutations. Although POLG is a high-fidelity enzyme, majority of mtDNA mutations seem to originate from the replication errors probably laid down by POLG. This might be a result of relaxed replication, as mtDNA is much more frequently copied, independent of cell cycle. This does not exclude the role for the mitochondrial BER pathway in removing different adducts from the mtDNA, although currently, this pathway seems not to be the most prominent factor in mtDNA protection against mutagenesis.
Is there a role for mitochondria in ageing and ageing-associated diseases beyond oxidative stress and the MFRTA? A large amount of evidence has accumulated over a couple of decades of intense research which demonstrated that both levels of point mutations and large-scale mtDNA deletions, increase with age in humans and different model organisms. The causative effect of increased mtDNA point mutation rates, excluding increased ROS levels, on the appearance of premature ageing phenotypes, has been proven by the development of mtDNA mutator mice [62,120].However, it is hard to estimate the relative contribution of mtDNA mutations to normal ageing and therefore, we would highly benefit from reciprocal experiments, to test if decreasing the amount of mtDNA mutations would extend the lifespan, in order to clearly demonstrate the link between mtDNA mutations and aging. Finally, a novel mitochondrial theory of ageing is emerging that, going beyond the accumulation of mtDNA mutations, should be extended to include abnormalities in mitochondrial turnover, dynamics and proteostasis, but also pathways that seem to counteract age-related decline as increased mitochondrial biogenesis and the mitochondrial unfolded protein response (UPRmt).
mtDNA point mutations are hallmarks of ageing, but their origin remains controversial.
The enduring standpoint that suggests that mtDNA point mutations emerge as a consequence of oxidative damage has been extensively challenged in recent years.
Recent data suggest that mtDNA is well protected from oxidative damage and the majority of mtDNA point mutations arise from the imperfect replication mediated by mtDNA polymerase γ.
We thank Dr Alexandra Kukat for the critical reading of the manuscript and her input.
This work was supported by the European Research Council [grant number ERC-StG-2012-310700]; and the German Research Council (DFG) [grant number TR 1018/3-1].
The authors declare that there are no competing interests associated with the manuscript.
base excision repair
human embryonic kidney cells 293
mitochondrial free radical theory of ageing
endonuclease VIII like 1/2
Nth-like DNA glycosylase 1
DNA polymerase γ
quantitative polymerase chain reaction
reactive oxygen species
transcription factor A, mitochondrial