Mitochondrially-targeted APOBEC1 is a potent mtDNA mutator affecting mitochondrial function and organismal fitness in Drosophila

Somatic mutations in the mitochondrial genome (mtDNA) have been linked to multiple disease conditions and to ageing itself. In Drosophila, knock-in of a proofreading deficient mtDNA polymerase (POLG) generates high levels of somatic point mutations and also small indels, but surprisingly limited impact on organismal longevity or fitness. Here we describe a new mtDNA mutator model based on a mitochondrially-targeted cytidine deaminase, APOBEC1. mito-APOBEC1 acts as a potent mutagen which exclusively induces C:G>T:A transitions with no indels or mtDNA depletion. In these flies, the presence of multiple non-synonymous substitutions, even at modest heteroplasmy, disrupts mitochondrial function and dramatically impacts organismal fitness. A detailed analysis of the mutation profile in the POLG and mito-APOBEC1 models reveals that mutation type (quality) rather than quantity is a critical factor in impacting organismal fitness. The specificity for transition mutations and the severe phenotypes make mito-APOBEC1 an excellent mtDNA mutator model for ageing research.

The dat a in t his paper appear t o be sound, and proper cont rols have been im plem ent ed. However, I feel that the authors' interpretations are a bit too narrow and should be backed up with addit ional dat a and m odified if needed. Aft er t hese issues are fixed ( see below) , I have no doubt t hat t his will be a significant cont ribut ion t o t he field t hat deserves t o be published in NC.
Maj or issues 1. The m aj or phenot ypic difference bet ween t he t am as m ut at or and t he m it o-APOBEC1 expressing flies t o m e suggest s t he possibilit y t hat t he lat t er has addit ional effect s on m it ochondrial funct ions dist inct from t hose caused by t he m ut at ion burden per se. One obvious t hing t hat should be t est ed is effect s on m it ochondrial RNA levels, since t he presence of a considerable num ber of abasic sit es and/ or U residues over t he genom e could lim it t he overall rat e of t ranscript ion ( m ism at ches should be t olerat ed by t he t ranscript ion m achinery but abasic sit es in t he t em plat e st rand or even j ust t he presence of t he BER m achinery m ight not ) . The aut hors should t herefore m easure t he levels of at least a few represent at ive t ranscript s from each st rand of t he m aj or gene clust ers using quant it at ive RT-PCR, versus cont rols.
2. Alt hough t he aut hors have checked effect s on m t DNA copy num ber by PCR, t his does not dist inguish bona fide m t DNA from incom plet ely or abort ively replicat ed m olecules. I t would t hus be helpful t o include a Sout hern blot analysis alongside t o t est whet her m t DNA is st ill int act in t he m it o-APOBEC1 st rain. There m ay also be effect s on m t DNA replicat ion which could part ially underlie t he phenot ype. St alled, regressed or degraded forks should be easily det ect ed on 2D agarose gels ( t he ent ire genom e need not be int errogat ed, j ust one or t wo rest rict ion fragm ent s covering e.g. t he replicat ion origin and a region dist ant from it ) .
3. I n t he light of t hese doubt s, and even if t he addit ional experim ent s fully support t he aut hors' original int erpret at ions, I t hink it is not helpful t o t he field t o use a st rident ly worded int erpret at ion as t he t it le of t he paper. I nst ead, som et hing like 'APOBEC1 expression in Drosophila m it ochondria act s as a pot ent m t DNA m ut at or and produces a phenot ype of m it ochondrial dysfunct ion' would be a t it le t hat m ore accurat ely fit s t he m ain findings and does not m ake definit ive claim s about t he m echanism of this effect, which rem ain t o be fully verified.

Minor issues
4. An im portant corollary of t he findings is t hat t hey cast som e doubt on t he assum pt ion t hat t he developm ent al let halit y of t he t am as m ut at or when hom ozygous is really due t o m ut agenesis, and not t o som e ot her effect of t he t am asexo-allele on m t DNA replicat ion. I feel t here is no need for t he aut hors t o delve int o this point m echanistically, but it should nevert heless appear in t he Discussion.
5. The aut hors already point out t hat the t ype of m ut at ion produced by APOBEC1 is likely to be dam aging because of t he high A+ T cont ent of t he Drosophila m itochondrial genom e, wit h C/ G pairs confined t o crit ical sit es. This point could be bet t er fleshed out by explaining t hat alm ost all t hird-base posit ions in t he coding sequence are eit her A or T, whilst C and G are found predom inant ly in first or second base posit ions, giving rise t o a high frequency of nonsynonym ous changes brought about by m it o-APOBEC1. Can t he expect ed pat t erns of subst it ut ions be m odelled com put at ionally, t o t est how far t he specific m ut at ional pat t erns of t he t wo m ut at ors fit t hese predict ions? One of t he 'unknowns' m ay be t he frequency of different t ypes of error int roduced by Pol gam m a, w hich such an analysis m ight r eveal.
6. The parent -of-origin difference in t he m ut at ion levels conferred by t he het erozygous t am as m utator is consistent with the literature on the developm ental effects of severe or even null alleles of t am as, which give let halit y only at L3 st age. This support s t he idea t hat t he st ore of m at ernally supplied Pol gam m a is sufficient t o carry t he zygot e far int o developm ent . I suggest t o add t his point and cit e t he relevant lit erat ure. 7. Alt hough t he A+ T region cannot be reliably sequenced by NGS m et hods, it can be sequenced using convent ional Sanger sequencing, t hough I do not see any specific advant age in carrying out such a laborious analysis at t his point . However, t he relevant figure legends should not e t hat t he non-coding A+ T region is excluded from t he analysis, or at least specify it, as 'm t DNA codingregion m ut at ion load' et c. Even t hough t his is m entioned in the Materials and Methods, a casual reader m ight not not ice. . Andreazza et al., subsequent ly produce a novel m itochondrial m utator m odel by expressing the cytidine deam inase APOBEC1 specifically in the m itochondrial. APOBEC1 expression produces sim ilar levels of m utation as the pol gam m a exonuclease m ut ant , however, APOBEC1 expression in t he m it ochondria induces m it ochondrial dysfunct ion and an aging phenot ype. The aut hors analyze t he m ut at ion spect rum induced by APOBEC1 com pared t o t he pol gam m a m ut ant and det erm ine t hat APOBEC1 expression induces exclusively C t o T subst it ut ions. They also find t hat APOBEC1-induced m ut at ions are m ore likely t o produce non-synonym ous changes, which t he aut hors argue is t he likely cause of t he m itochondrial dysfunction and aging. Overall, t he work is well cont rolled and int erest ing. The lack of an aging phenot ype in flies expressing act ive APOBEC1 t hat is not t arget ed t o t he m it ochondria or cat alyt ically inact ive m it o-APOBEC1 clearly indicat e t hat t he aging phenot ype in t hese flies is due t o act ive deam inat ion of t he m it ochondrial genom e. However, I believe addressing t he following com m ent s will st rengt hen t he m anuscript . 1) I n Fig. 1, t heir appears t o be a 4 fold higher load of C: G t ransit ions ( y-axis scale is 10^-3) in panel D com pared t o t he point m ut at ion load ( y-axis scale is 10^-4) in panel A. This needs t o be explained because t he t ransit ions should be a subset of t he point m ut at ion so t he dat a in panel D should be lower t han in panel A. 2) Please include t he unit s for m ut at ion load in t he fig. 1. Mut at ion load could m ean t ot al m ut at ions or m ut at ion density ( m ut ations per nucleot ide) , which is used here. 3) I n lines 275-277, t he aut hors st at e t hat t he high rat io of NS t o S m ut at ions is due t o m it o-APOBEC1 causing exclusively C: G t ransit ions and t hen suggest t his underlies t he aging phenot ype. This effect would presum ably be due t o inact ivation of m it ochondrial prot eins t hrough t he NS subst it ut ions. However, in Fig 1 t he m aj orit y of point m ut at ions in t he exo-pol gam m a flies are also C: G t ransit ions. There is less t han a 2-fold difference in C: G t ransit ion load bet ween t he exopol gam m a flies and t he m it o-APOBEC1 flies. That difference does not seem enough t o account for t he difference in t he NS: S subst it ut ion rat io. 4) I n relat ion t o point 3, APOBEC1 deam inat es DNA in a sequence specific m anner, t arget ing TC dinucleot ides. Could t his sequence specificit y be cont ribut ing t o t he elevat ed NS: S rat io? A com parison of t he sequence cont ext of t he m ut at ions in t he exo-pol gam m a and m it o-APOBEC1 would address t his. 5) Mit o-APOBEC1 m ut at ions also appear t o be st rand biased. Based on an analysis of t he m utations in supplem ental table 1, 3-4 fold m ore G m utations occur than C m utations in the m ito-APOBEC1 flies. Exo-pol gam m a flies have m ore C m ut at ions t han G m ut at ions. This indicat es t hat APOBEC1 is deam inat ing one of t he DNA st rands of t he m it ochondrial DNA m ore t han t he ot her st rand. Since APOBEC1 is ssDNA specific, is t here any idea what t he ssDNA t arget of t he enzym e is in t his syst em ? I n lines 327-328, t he aut hors speculat e m it o-APOBEC1 deam inat es t ranscript ion int erm ediat es. The analysis of st rand bias, could answer t his quest ion. Could it also cont ribut e t o t he NS: S rat io? 6) I n t he discussion, I believe it would be useful for t he aut hors t o discuss why no aging phenot ype is observed in t he exo-pol gam m a flies when it induces a significant num ber of insert ion/ delet ion m ut at ions t hat would be as dam aging ( if not m ore) as t he C: G t ransit ion m ut at ions. 7) While t he aut hors nicely show no difference occurs in t he num ber of m olecules of m it ochondrial DNA upon expression of m it o-APOBEC1 or exo-pol gam m a, could t he t wo m ut at ors different ially im pact t he rem oval of defect ive m it ochondria by m it ophagy? Select ive rem oval of m it ochondrial genom es cont aining delet ions has been suggest ed in at least one ot her publicat ion ( Kandul, et al., Nat ure Com m unicat ions, 2016) . 8) Presum ably ot her m ut at ors wit h less specificit y t han t he m it o-APOBEC1, could also induce an aging phenot ype if t hey produced enough m ut at ions t o accum ulat e t he sam e levels of C: G t ransit ions. Would t hat level of m ut agenesis be physiologically relevant t o m ake t he m it o-APOBEC1 an effect ive m odel for m ore relevant m ut agens t hat would be expect ed t o act in a non-engineered m it ochondria?
We thank the reviewers for their positive support for this study and their constructive critiques. In line with their suggestions we have amended the test to clarify our methodology, meaning or intention, and we have added several new analyses which together we feel further strengthen the study.

Reviewer #1 (Remarks to the Author):
This is a nicely-written paper that convincingly introduces a new mitochondrial mutator system in Drosophila and compares it to the classic POLG model in terms of lifespan, mtDNA mutations and mitochondrial function. The new model is likely to be of broad interest and utility in the field.
Minor point: the choice of substrates for the mitochondrial experiments is odd, and more appropriate for mammalian mitochondria. Mitochondria from flies use proline and glycerol phosphate much better than the glutamate+malate and succinate used here, so these substrates would have been better able to show up functional deficits than the ones chosen.
We agree that the precise details of the protocol are more typical for mammalian than insect mitochondrial physiology but while fly mitochondria do indeed utilise proline more efficiently than glutamate+malate (likely due to free permeability of fly mitochondria to proline, although this seems to be unresolved), we consider that any potential difference is probably offset somewhat in our assay by conducting the analysis at 30 o C. Nevertheless, we appreciate the reviewer's point and will adapt our protocols in future to maximise the potential to detect small differences. In the context of the current study, we are confident that we have detected robust differences where they are present and the lack of observable phenotypes is a true reflection of the condition.

Reviewer #2 (Remarks to the Author):
This is an extremely interesting paper which potentially opens up the field of mitochondrial mutagenesis. The three most important findings are that the originally created tamas (Polg) mutator, which in Drosophila is recessive lethal, has a minimal phenotypic effect as a heterozygote, whereas the implementation of the human APOBEC1 gene, which encodes an RNA editing enzyme that also acts as a C to T mutator in DNA, produces a severe and typically 'mitochondrial' phenotype of respiratory chain impairment, disturbed locomotor function and decreased lifespan. The third finding concerns the parent-of-origin effect on the mutation load conferred by the tamas (but not the mito-APOBEC1) mutator, being much stronger in the female line.
The data in this paper appear to be sound, and proper controls have been implemented.
However, I feel that the authors' interpretations are a bit too narrow and should be backed up with additional data and modified if needed. After these issues are fixed (see below), I have no doubt that this will be a significant contribution to the field that deserves to be published in NC.
Major issues 1. The major phenotypic difference between the tamas mutator and the mito-APOBEC1 expressing flies to me suggests the possibility that the latter has additional effects on mitochondrial functions distinct from those caused by the mutation burden per se. One obvious thing that should be tested is effects on mitochondrial RNA levels, since the presence of a considerable number of abasic sites and/or U residues over the genome could limit the overall rate of transcription (mismatches should be tolerated by the transcription machinery but abasic sites in the template strand or even just the presence of the BER machinery might not). The authors should therefore measure the levels of at least a few representative transcripts from each strand of the major gene clusters using quantitative RT-PCR, versus controls. This is a good suggestion. We have now analysed by qRT-PCR the expression levels of several mitochondrial transcripts across the genome. Specifically, we have analysed 4 transcripts; Cyt-b, ND3 and CoI from the major strand and ND4 from the minor strand. Interestingly, these results show a consistent decrease in transcript levels for the non-mutating mito-APOBEC1[E63A] control while only Cyt-b showed a significant change compared to control in mito-APOBEC1 flies, which was in fact increased. These data are shown in new Fig. 3f. We interpret these results to indicate that mito-APOBEC1 is not causing a generalised or gross disruption of mitochondrial transcription, and any small changes observed are unlikely to significantly contribute to the organismal phenotypes.
2. Although the authors have checked effects on mtDNA copy number by PCR, this does not distinguish bona fide mtDNA from incompletely or abortively replicated molecules. It would thus be helpful to include a Southern blot analysis alongside to test whether mtDNA is still intact in the mito-APOBEC1 strain. There may also be effects on mtDNA replication which could partially underlie the phenotype. Stalled, regressed or degraded forks should be easily detected on 2D agarose gels (the entire genome need not be interrogated, just one or two restriction fragments covering e.g. the replication origin and a region distant from it).
We appreciate the value of this analysis as we are keen to determine as much as possible whether the phenotypes we see are principally due to point mutations. As suggested, we have analysed the mitochondrial genome integrity in the mito-APOBEC1 model (and controls) by Southern blot analysis, and further complemented this by long-range PCR. Encouragingly, neither of these approaches revealed any evidence of mtDNA with deletions. However, quantification of the Southern blots revealed a small but significant decrease in mtDNA levels. These data are shown in new Fig. 3c-e. These results led us to consider our previous qPCR analysis of mtDNA levels which had shown a consistent trend towards lower levels, although it appeared that the technical variability rendered these differences non-significant. Reassessing these data, we realised that we had used a more conservative statistical analysis than was necessary (no matched comparisons).
Applying a more appropriate and lenient analysis (matched comparisons), the mtDNA levels in mito-APOBEC1 and mito-APOBEC1[E63A] 20-day-old flies are significantly decreased, in line with the Southern blot data. These results have been amended in the respective figures and text.
Although some of these differences are statistically meaningful, changes are consistently modest in size. Thus, considering the weight of evidence from the range of techniques used (qPCR, Southern blot, long-range PCR), in conjunction with the sequence and transcript analysis, we believe that little if any contribution to the organismal phenotypes comes from disruption of other aspects of mitochondrial genome than the mtDNA mutations directly.
3. In the light of these doubts, and even if the additional experiments fully support the authors' original interpretations, I think it is not helpful to the field to use a stridently worded interpretation as the title of the paper. Instead, something like 'APOBEC1 expression in Drosophila mitochondria acts as a potent mtDNA mutator and produces a phenotype of mitochondrial dysfunction' would be a title that more accurately fits the main findings and does not make definitive claims about the mechanism of this effect, which remain to be fully verified.
We understand the reviewer's point of view but there were a number of key elements that we wanted to convey in our title. One was that this was a novel type of mtDNA mutator model, since, as discussed, the field has been heavily reliant on the classic POLG mutator. Second, we wanted to make a clear counter-point to the recent PNAS paper from Kauppila et al. entitled "Mutations of mitochondrial DNA are not major contributors to aging of fruit flies" which is in stark contrast to our study. It is clear from our study that mtDNA mutations are in fact capable of influencing Drosophila lifespan, but it depends on the nature of the mutations (and the model). While the reviewer's proposed title is certainly adequate, this outcome would not be reflected in that title.
Moreover, we also wanted to convey what we consider is a very important outcome of our analysis; that is, a much more profound and accurate insight into the nature of mtDNA mutations is gained from a much more detailed analysis of the exact nature of the observed mutations. It is only through this deep analysis of the mutations that we have been able to rationalise why two models with an apparently similar mutational burden can have such dramatically different organismal consequences. This clearly indicates that it isn't the quantity of mtDNA mutations that specifically affects organismal fitness but what those mutations are (i.e. their quality). We feel that this takehome message needs the prominent visibility afforded in the title.

Minor issues
4. An important corollary of the findings is that they cast some doubt on the assumption that the developmental lethality of the tamas mutator when homozygous is really due to mutagenesis, and not to some other effect of the tamasexo-allele on mtDNA replication. I feel there is no need for the authors to delve into this point mechanistically, but it should nevertheless appear in the Discussion.
We have added a comment in the Discussion on this, and cited the recent paper of Samstag et al. (PLoS Genetics, 2018) who have also discussed this matter. 5. The authors already point out that the type of mutation produced by APOBEC1 is likely to be damaging because of the high A+T content of the Drosophila mitochondrial genome, with C/G pairs confined to critical sites. This point could be better fleshed out by explaining that almost all third-base positions in the coding sequence are either A or T, whilst C and G are found predominantly in first or second base positions, giving rise to a high frequency of nonsynonymous changes brought about by mito-APOBEC1. Can the expected patterns of substitutions be modelled computationally, to test how far the specific mutational patterns of the two mutators fit these predictions? One of the 'unknowns' may be the frequency of different types of error introduced by Pol gamma, which such an analysis might reveal.
In our original manuscript we calculated the C:G content in first, second, and third codon positions for each protein within the mitochondrial genome, and demonstrated the depletion of C:G sites in the third codon position (new Supplementary Fig. 6c). We have now expanded upon this analysis in three ways: 1) We characterised the prevalence of mutations in first, second, and third codon positions in mito-APOBEC1 and maternal tam exomutator models, demonstrating that mito-APOBEC1 flies show an enrichment in first and second codon position mutations relative to tam exoflies ( Supplementary Fig. 6d).
2) We analysed the theoretical rate of non-synonymous vs synonymous substitutions for each mutation type throughout the protein-coding sequence to clarify that the predominant mutation type observed in mito-APOBEC1 flies frequently results in non-synonymous substitutions ( Supplementary Fig. 6e).
3) We have analysed the mutation load within each trinucleotide context ( Supplementary Fig. 7).
Together, these data support the model that mito-APOBEC1 flies bear an enrichment in first and second codon position mutations in a highly specific genomic context. Conversely, tam exoflies display a diverse mutation spectrum, including a much greater proportion of mutations in third codon positions and a large number frequently benign T:A>C:G mutations. These mutations are more likely to be synonymous substitutions, influencing the skewed NS:S ratio we previously discussed. We have added these new analyses to the manuscript and discussed their implications in context. 6. The parent-of-origin difference in the mutation levels conferred by the heterozygous tamas mutator is consistent with the literature on the developmental effects of severe or even null alleles of tamas, which give lethality only at L3 stage. This supports the idea that the store of maternally supplied Pol gamma is sufficient to carry the zygote far into development. I suggest to add this point and cite the relevant literature.
We have added a comment and citation in the Discussion to this effect. 7. Although the A+T region cannot be reliably sequenced by NGS methods, it can be sequenced using conventional Sanger sequencing, though I do not see any specific advantage in carrying out such a laborious analysis at this point. However, the relevant figure legends should note that the non-coding A+T region is excluded from the analysis, or at least specify it, as 'mtDNA codingregion mutation load' etc. Even though this is mentioned in the Materials and Methods, a casual reader might not notice.
We have added this clarification in the relevant Methods and reiterated it in the first figure legend.
8. The post-hoc correction to ANOVA is by Tukey, not Turkey. Corrected.
9. The APOBEC1 signal in the Western of Fig. 2B is obviously poor. Whilst I see nothing 'suspicious' about this, it would be good to substitute a better image for publication, especially given the much cleaner signals in Fig. 2C.
We understand the reviewer's point but this level of signal is an accurate reflection of the relative level of expression. The purpose of showing this blot is to verify that a single protein species is produced as expected, and to give a sense of the relative expression levels both between the two mito-APOBEC1 transgenes and to the control mito-HA-GFP transgene. The conditions used for this blot also affords comparisons between mito-and cyto-APOBEC1 transgenes, as the same control, da>mito-HA-GFP, is used. Overexposure of blot in Fig. 2b would make comparison with cyto-APOBEC1 ( Supplementary Fig. 3b) unreasonable. This is an important comparison as it demonstrates that even though the cyto-APOBEC1 expression is benign, protein levels are much higher than mito-APOBEC1. In summary, we feel that this blot provides a true reflection of the relatively low levels of expression from these transgenes, and offers appropriate comparisons where necessary.
10. Although it is already stated in M&M, it's appropriate to mention in the legend to Fig. 3 that the assays were conducted on males.
This has been added.

Reviewer #3 (Remarks to the Author):
In their manuscript, "A new mitochondrial DNA mutator model shows that quality not quantity of mutations affects organismal fitness in Drosophila, Whitworth and colleagues investigate the role of mutation of the mitochondrial genome in aging of the fly. While the accumulation of mutations in mitrochondrial DNA has been linked to aging in mammalian systems, Andreazza et al. show that an exonuclease deficient DNA polymerase gamma, which replicates the mitochondrial genome, induces mutation, however, unlike mouse model systems, this increase in mutagenesis is not associated with an aging phenotype or mitochondrial dysfunction. This result is consistent with a previous recent publication showing that mutants of the Drosophila DNA pol gamma fail to induce an aging phenotype (Kauppila et al., PNAS, 2018). Andreazza et al., subsequently produce a novel mitochondrial mutator model by expressing the cytidine deaminase APOBEC1 specifically in the mitochondrial. APOBEC1 expression produces similar levels of mutation as the pol gamma exonuclease mutant, however, APOBEC1 expression in the mitochondria induces mitochondrial dysfunction and an aging phenotype. The authors analyze the mutation spectrum induced by APOBEC1 compared to the pol gamma mutant and determine that APOBEC1 expression induces exclusively C to T substitutions. They also find that APOBEC1-induced mutations are more likely to produce non-synonymous changes, which the authors argue is the likely cause of the mitochondrial dysfunction and aging.
Overall, the work is well controlled and interesting. The lack of an aging phenotype in flies expressing active APOBEC1 that is not targeted to the mitochondria or catalytically inactive mito-APOBEC1 clearly indicate that the aging phenotype in these flies is due to active deamination of the mitochondrial genome. However, I believe addressing the following comments will strengthen the manuscript. We calculate mutation frequencies across the whole genome as the total number of mutations identified, including multiple counts for higher heteroplasmies, divided by the total coverage of bases sequenced. For calculations that concerns specific genomic subsets, e.g. by base type ( Fig.   1d and Supplementary Fig. 5a) or genomic feature (Supplementary Fig. 5b and Supplementary   Fig. 7), the denominator is determined by the subset category indicated on the X-axis (e.g. number of sequenced Cs, Gs, etc; or sequenced protein coding bases; or number of sequenced bases in a given codon). Specifically regarding the mutation load in Fig. 1d, as C:G transitions are only possible at C:G sites in the genome, only C:G sites are tabulated for the denominator of this calculation. So, the frequencies shown in Fig. 1a are not intended to be, and indeed should not be, directly compared to those in Fig. 1d. We have further clarified this in the main text.
2) Please include the units for mutation load in the fig. 1. Mutation load could mean total mutations or mutation density (mutations per nucleotide), which is used here.
We agree with the reviewer that the term "mutation load" is ambiguous. We have changed this term to "mutation frequency" throughout the revised manuscript and figures, and the added text describing the calculation of mutation frequency should help to clarify this for the readership.

3)
In lines 275-277, the authors state that the high ratio of NS to S mutations is due to mito-APOBEC1 causing exclusively C:G transitions and then suggest this underlies the aging phenotype. This effect would presumably be due to inactivation of mitochondrial proteins through the NS substitutions. However, in Fig 1 the majority of point mutations in the exo-pol gamma flies are also C:G transitions. There is less than a 2-fold difference in C:G transition load between the exo-pol gamma flies and the mito-APOBEC1 flies. That difference does not seem enough to account for the difference in the NS:S substitution ratio.
We have clarified and expanded upon this hypothesis with a more thorough characterization of the mutational spectra of these flies. As explained in our response to Reviewer 2 above, we have now shown that there is an almost total lack of third codon mutations in mito-APOBEC1 flies (new Supplementary Fig. 6d), whereas tam exostill has a substantial proportion of mutations in the third codon position. This likely has a considerable impact on the occurrence of NS mutations and the reason that the NS:S ratio shown in Fig. 7a, b is much lower for tam exothan for mito-APOBEC1.

4)
In relation to point 3, APOBEC1 deaminates DNA in a sequence specific manner, targeting TC dinucleotides. Could this sequence specificity be contributing to the elevated NS:S ratio? A comparison of the sequence context of the mutations in the exo-pol gamma and mito-APOBEC1 would address this.
We have further characterised our mutator lines by analysing the mutations with respect to the trinucleotide (rather than dinucleotide) spectrum. The distribution of the number of mutations across all possible trinucleotide sequences reveals that while tam exomutates in a wide variety of genomic contexts, mito-APOBEC1 mutates in a highly specific dinucleotide context (TCn), particularly at TCT sites. It is possible that this specificity contributes to the difference in NS:S ratios observed between the two strains but it has not been explored further. These data are presented in new Supplementary Fig. 7.

5)
Mito-APOBEC1 mutations also appear to be strand biased. Based on an analysis of the mutations in supplemental table 1, 3-4 fold more G mutations occur than C mutations in the mito-APOBEC1 flies. Exo-pol gamma flies have more C mutations than G mutations. This indicates that APOBEC1 is deaminating one of the DNA strands of the mitochondrial DNA more than the other strand. Since APOBEC1 is ssDNA specific, is there any idea what the ssDNA target of the enzyme is in this system? In lines 327-328, the authors speculate mito-APOBEC1 deaminates transcription intermediates. The analysis of strand bias, could answer this question. Could it also contribute to the NS:S ratio?
This is a very interesting phenomenon. We have now more deeply analysed the strand bias of the mito-APOBEC1 mutator and found that there is indeed a dramatic bias (new Supplementary Fig.   2a). In fact, there is a nearly 20-fold difference in mutation rate between the major and minor strands; the minor strand being much more mutated. The reason for this striking strand bias is unknown and mapping the mutations provided no obvious clues (new Supplementary Fig. 2b). We speculate the minor strand, whose replication is discontinuous (lagging strand), may be more vulnerable to accumulating mutations. Moreover, current literature also supports some role for APOBEC1 in mutating during transcription (Lada et al., PLoS Genetics, 2015); this has been cited in the discussion. It will be interesting to resolve the mechanistic cause in subsequent studies.

6)
In the discussion, I believe it would be useful for the authors to discuss why no aging phenotype is observed in the exo-pol gamma flies when it induces a significant number of insertion/deletion mutations that would be as damaging (if not more) as the C:G transition mutations.
Our sequencing analysis shows that when indels are present, they rarely exist in heteroplasmies above 10% (Supplementary Table 2). Thus, the relatively limited occurrence of indel mutations is likely readily complemented by wild-type molecules.

7)
While the authors nicely show no difference occurs in the number of molecules of mitochondrial DNA upon expression of mito-APOBEC1 or exo-pol gamma, could the two mutators differentially impact the removal of defective mitochondria by mitophagy? Selective removal of mitochondrial genomes containing deletions has been suggested in at least one other publication (Kandul, et al., Nature Communications, 2016).
This is, at least theoretically, possible but we have no data at this stage to suggest that the two mutator models are inducing mitochondrial quality control processes differently. This is the subject of ongoing work and beyond the scope of the current study.

8)
Presumably other mutators with less specificity than the mito-APOBEC1, could also induce an aging phenotype if they produced enough mutations to accumulate the same levels of C:G transitions. Would that level of mutagenesis be physiologically relevant to make the mito-APOBEC1 an effective model for more relevant mutagens that would be expected to act in a nonengineered mitochondria?
We would agree that other mutagens could indeed induce an ageing phenotype if enough 'damaging' (i.e. NS) mutations arose. However, a less specific mutagen that the reviewer conceives of would also presumably need to contend with a relatively higher amount of mutations in non-coding genomic elements including tRNA and rRNA which may lead to a more generalised and possibly more detrimental mitochondrial disruption.