A spontaneous mitonuclear epistasis converging on Rieske Fe-S protein exacerbates complex III deficiency in mice

We previously observed an unexpected fivefold (35 vs. 200 days) difference in the survival of respiratory chain complex III (CIII) deficient Bcs1lp.S78G mice between two congenic backgrounds. Here, we identify a spontaneous homoplasmic mtDNA variant (m.G14904A, mt-Cybp.D254N), affecting the CIII subunit cytochrome b (MT-CYB), in the background with short survival. We utilize maternal inheritance of mtDNA to confirm this as the causative variant and show that it further decreases the low CIII activity in Bcs1lp.S78G tissues to below survival threshold by 35 days of age. Molecular dynamics simulations predict D254N to restrict the flexibility of MT-CYB ef loop, potentially affecting RISP dynamics. In Rhodobacter cytochrome bc1 complex the equivalent substitution causes a kinetics defect with longer occupancy of RISP head domain towards the quinol oxidation site. These findings represent a unique case of spontaneous mitonuclear epistasis and highlight the role of mtDNA variation as modifier of mitochondrial disease phenotypes.


Presentation of activities and bioenergetics data in
: The way the authors normalize the data could be potentially artefactual and it would be helpful to see the data in a more straightforward way. In Figure 2 (a), CIII activity is normalized by CIV, why? There is no rational for this. In addition, CIV is measured in two different ways (spectrophotometry and polagrography) depending on the tissues, so the data are not really comparable. It would be more correct and consistent if the authors normalized CIII activity by protein (like they do for CI and CII in Supplemental Figure 2) or by citrate synthase (CS) activity, which is the standard method.
It is unclear how the authors normalized the respiration in figures 2b and c. "Relative O2 flux" seems unconventional? Could the values be expressed as pmol O2/s/mg of protein. This would also make it easier to compare the values of the old and new backgrounds with the already published data by the same group e.g. in: Leveen et al. 2011. The GRACILE mutation introduced into Bcs1l causes postnatal complex III deficiency: A viable mouse model for mitochondrial hepatopathy. Hepatology 53, 437-447 or Davoudi et al. 2014. Complex I Function and Supercomplex Formation Are Preserved in Liver Mitochondria Despite Progressive Complex III Deficiency. PLoS One 9, e86767. Here again, there does not seem to be striking differences in the respiratory parameters between the Mt-Cyb genotypes when comparing them within the same Bcs1l genotype.
The functional assays in Rhodobacter also indicate that the Mt-Cyb mutation does not have striking consequences in CIII activity. I am sure about the usefulness of the +2Ala mutant shown in figure 4 (h) and (i), this is a drastic mutation that is not found in the mice. Therefore, the biochemical and functional data shown in the paper indicate a _mild_ additional effect of the Mt-Cyb mutation on top of the Bcs1l mutation. It is therefore a bit unexpected that Mt-Cyb p.D254N accounts for the striking survival phenotype. The manuscript could do much better fairly dealing with this issue.
Lastly, in the previous publications (cited above) the authors demonstrate that there is a CIII activity decrease with age in the Bcs1l mutant mouse in the old background (Mt-Cyb p.D254N). I wonder if they have checked for the progression of the CIII activity decline with age in the new background (Mt-Cyb WT) to see if it is slower in the critical points and could ameliorate the mouse development and explain the differences in the survival.
Minor point: Nomenclature is not correct Mt-Cyb and MT-CYB (instead of Cytb and CYTB), CYC1 (not CYTC1), RISP is acceptable, but UQCRFS1 is more correct.
Reviewer #2 (Remarks to the Author): In this work Purhonen et al. have performed in-depth genetic and phenotypic analyses of two different in-house bred mouse strains, both of which harbour the disease-related homozygous p.S78G mutation in the mitochondrial complex III (CIII) assembly factor Bcs1l (involved in the right insertion of the catalytic Rieske FeS protein subunit in CIII), which display either short or long (4-fold) survival rates depending on the C57Bl/6J nuclear genetic (nDNA) background surrounding the p.S78G mutation. WGS analyses identified only eight genetic variants within coding regions to differ between the strains, one of which corresponded to the novel homoplasmic variant m.14904G>A in the mitochondrial DNA (mtDNA) Cytochrome b (Cytb) gene that is exclusively present in the colony with short survival and it affects a negatively charged amino acid (p.D254N) conserved across eukaryotes, making it a likely genetic modifier of the survival of Bcs1lp.S78G mice. To assess the effect of maternally inherited mtDNA to the survival and mitochondrial function of Bcs1l mutant mice, the authors crossbred Bcs1lp.S78G heterozygotes from the two colonies, and compared F1 progenies carrying either wild-type or variant mitochondria with equivalent nDNA background. Using spectrophotometric and polarographic assays they show that the presence of the Cytb variant by itself seems sufficient to induce mitochondrial dysfunction and a metabolic disease phenotype, and that this mutation further exacerbates the CIII activity and mitochondrial respiration defects present in Bcs1l p.S78G mice. In silico simulations predicted that the Cytb variant would compromise the conformational flexibility of the Rieske FeS protein (RISP) head domain necessary for full CIII activity, and the authors show that the equivalent substitution in Rhodobacter Cytb may induce a mild shift in the occupancy of the Rieske head domain towards the quinol oxidation site that, however, neither affects the integrity and activity of Rhodobacter CIII nor bacterial growth. This is an interesting manuscript where the authors clearly show the epistatic contribution of the Cytb variant m.14904G>A to the mitochondrial CIII defects promoted by the p.S78G mutation in Bcs1l. However, the potential pathogenic role of the Cytb variant alone is unconvincing under the experimental conditions used in this work. First, Bcs1l wild-type mice carrying the Cytb p.D254N variant show similar weight and life span to the wild type mice and histologically, they do not show a clear phenotypic defect. Second, the main phenotypic effect observed in these mice is a significant decrease in CIII activity only in liver tissue (but not in kidney, heart or muscle) which, however, is not reflected in a significant decrease in mitochondrial oxygen consumption in liver (neither CI-nor CI+CII-linked respiration differ significantly, so the use of ratios to demonstrate functional differences here is misleading). The decrease in CIII activity neither correlates with the normal levels/assembly of hepatic complex III observed in these mice. Therefore, how do the authors explain these apparent contradictions? Moreover, as stated by the authors in the Discussion, the Cytb p.D254N variant appears to be present in the human population as a rare polymorphism, suggesting that this situation could occur in mice. To make a clear statement about the action of the maternally-inherited Cytb variant it is advisable to use alternative experimental models easily subject to stress conditions, for instance, cellular cybrids.
The proposed epistatic mechanistic model that converges in the action of the catalytic RISP subunit is very interesting, but it is solely based in computational simulations that lack a convincing experimental demonstration that needs to be provided. As mentioned before, the equivalent Cytb p.D254N variant in Rhodobacter seems to induce a mild defect in the conformational dynamics of the Rieske head domain that does not alter the integrity and activity of Rhodobacter CIII and therefore, this model is not valid to demonstrate the authors´ hypothesis and an alternative experimental model should be presented.
Reviewer #3 (Remarks to the Author): This paper provides evidence of a double-trouble effect on the biochemical performance of complex III which worsens dramatically the outcome of the Bcs1l mutation associated with the murine equivalent of the GRACILE syndrome, due to the concomitant presence of a missense homoplasmic mutation in the cytB gene. The genetic results are convincingly demonstrating that in the presence of the cytB mutation the lifespan of the Bcs1l mutant animals falls from about 200 days to 4-5 weeks. This cumulative effect id not caused by obvious structural or assembly defect of complex III in addition to the already severe reduction of complex III amount caused by the Bcs1l mutation. The biochemical characterization in different tissues, mainly liver, kidney and skeletal muscle, demonstrates a cumulative effect in cIII activity in the presence of both mutations ( Figure 2a)in liver, kidney and skeletal muscle but not in the heart.The activity of cIII was also slightly but significantly decreased in the liver of Bcs1l-WT, cytB-mutant animals, compared to the Bcs1l-WT, cytB-WT samples, whereas in other tissues no significant differences were detected.These results suggest a very modest, if any, biochemical effect of the cytB mutation when Bcs1l is WT. Other biochemical results require a more detailed explanation, since for instance the normalization procedures are not very clear and justified. For instance, it is unclear what the authors mean by cI-linked and cI-cII linked in figure 2b,c, and what is the meaning of the uncoupling measurement (perhaps to show the maximal respiration rate associated with the different genotypes? This should be explained and the results commented). The same applies to figure 2d (cI&cII/cI), and ETS coupling efficiency (figure 2e). Which is the biological significance of these results in association with the different genotypes considered in the paper. The results shown in figure 2f demonstrate that no increase in H2O2 production is associated with the presence of the mutations, alone or in combination with each other. However, the authors mention in the text that the value of H2O2 normalized to oxygen consumption is increased in the Bcs1l mutant liver and kidney: does this mean that oxygen is used to produce ROS instead of H2O because of the block in the ETC? In any case this effect is restricted to Bcs1l mutants whereas no effect is mentioned attributable to the mutation in cytB. To test (and confirm) that the two mutations are epistatically linked, the authors made in silico prediction and experiments based on spectroscopic measurements on the phase relaxation of RISP Fe-S cluster by oxidized heme bL. I am not familiar with this technology and I must rely on the conclusions of the authors who conclude that the cytB mutation stiffens cytB and interferes with the interaction between cytB and the Rieske protein, worsening the electron flux in a cIII mutant characterized by severe reduction of RISP incorporation, die to the Bcs1l mutation. The conclusions of the authors are that this is an example of synergistic effects of two mutations, one in the nuclear genome, the other in mtDNA, in determining a substantial worsening of the phenotype linked to cIII deficiency. The existence of causative correlations between mtDNA mutations/polymorphisms with each other has been demonstrated in humans (e.g. for LHON mutations). The possibility of a cumulative effect between a mtDNA mutation and the impairment of a nuclear gene has been also hypothesized, for instance in the ND6 mouse mutant associated with a deletion in the NNT gene. However, the present paper indicates an etiological link of these effect based on structural interactions. The explanation in the present case is based mainly on the experiments on Rhodobacter capsulatus (It is unclear what the 2Ala mutant is...), whereas the mutation in cytB has otherwise no biochemical or clinical phenotype. I am not sure that this experimental setup provides a persuasive explanation about the worsening of the phenotype. The in silico predictions suggest an increased "stiffness" of cytB, which can slightly alter the motion of cytB and, possibly, of the RISP. This effect has no consequences in the Bcs1l-WT, whereas it causes a very dramatic worsening of the clinical phenotype, which is disproportionately severe compared to the differences in the cIII activities, O2 consumption, and other biochemical measurements displayed in Fig. 2. This paper reports a Cytochrome b variant (Cytb<sup>p.D254N</sup>) that further decreases CIII activity and capacity of mitochondrial respiration in Bcs1l<sup>p.S78G</sup> mice. Molecular dynamics simulations of <i>S. cerevisae</i> cyt bc1 and variant Cytb<sup>p.D254N</sup> indicate that the observed <i>in vivo</i> differences can be attributed to the compromised mobility of the EF loop and to the Iron-Sulfur Protein Subunit Extrinsic Domain -ISP-ED (or RISP head domain). The mutation was also experimentally tested in <i>R. capsulatus</i> (E278N) where no significant differences to the enzyme activity were observed in relation to the WT. A subtle effect was detected on the motion of ISP-ED towards the Qo site. Rajagukguk, S. <i>et al.</i> previously described this mutation in Rhodobacter sphaeroides (D278N) (Biochemistry. 2007;46 (7): 1791-1798). The novelty of the present work thus resides in the combination of Cytb<sup>p.D254N</sup> with Bcs1lp.S78G . The effect of D254N alone is very subtle, as showed in the present work and previously by Rajagukguk, S. <i>et al.</i> , where only a small change in the rate constant for electron transfer from the iron-sulfur center [2Fe2S] to cyt c1 was found ( 35,000 s 1 for D278N and 60,000 s 1 for the WT enzyme). The MD simulations are state of the art and extensive enough (1 s) to provide adequate sampling. However, the authors should describe in more detail the modelled complex. The model was built from the crystal structure 3CX5, which also includes cytochrome c and the inhibitor stigmatellin, which arrests ISP-ED movement. The inhibitor was obviously deleted, but it seems that the Qoand Qi -sites were left empty since no information is given about the substrates quinol (QH<sub>2</sub>) or quinone (Q). If that is the case the simulations do not describe any intermediate of the catalytic cycle (Q-cycle). This represents a missing opportunity to study the effect of mutations in the actual catalytic cycle. Due to the proximity of the Qo-site to the EF loop it would be interesting to analyse the effect of the substrate on the EF loop dynamics. The authors should clarify this point.
Finally, taking in account that the effect of the mutation is subtle and since the enzyme is a homodimer, the authors should compare the RMSFs of each monomer.

Minor points:
It is stated that : "The atomic partial charges of metal cofactors were taken from an earlier study 49" I suppose the authors wanted to mention that the atomic partial charges and remaining parameters were taken from ref. 49. If the equilibrium bonds, angle and van der Walls parameters were taken from somewhere else that should be stated in the manuscript Below, please find our detailed response (in purple font) and a list of references supporting the response. In the response, the main changes and additions to the manuscript text and figure panels are referred to by figure and line numbers, respectively, and highlighted in red font.

Reviewer #1 (Remarks to the Author):
The manuscript by Purhonen et al. describes interesting genetic interaction of nuclear and mitochondrial DNA mutations in mitochondrial complex III coding genes (Bcs1lp.S78G and m.14904G>A/Cytbp.D254N, respectively) with the phenotypic consequences being more evident when the two mutations are combined. Given that Cytb p.D254N is rather phenotypically mild (see further), this serendipitous finding highlights the possibility that some OXPHOS defects can arise from converging of two variants that individually are benign (or close to benign). The latter is important for some aspects of diagnosis of respiratory chain defects in humans, therefore, the findings are of interest to others in the community and the wider field. The paper is well-written in a brief format and the data clearly presented. However, there are several issues concerning methodology and conclusions the authors should address before publication. The authors claim that the Mt-Cyb p.D254N mutation in the Bcs1l WT background has phenotypic consequences, but they only show a slight reduction in the weight at P33. Similarly, they also state that the Mt-Cyb p.D254N mutation already has consequences in the Bcs1l WT background in the activity of complex III in liver, but it is very mild (with a lot of inter-individual variability) and it is not the case in any other of the tested tissues. This aspect could be investigated and discussed a bit more thoroughly.
The major revelation of our study is the serendipitous and infinitesimally unlikely identification of a novel spontaneous mtDNA variant that has a drastic effect when combined with a known diseasecausing Bcs1l mutation. As for the effect of the D254N alone, we have to stress the fact that the mice carrying this spontaneous variant are healthy "wild-type" mice that have been used by us (see references in manuscript) and several other research groups for about a decade without any suspicion. Therefore, we did not expect D254N to have any measurable effect on its own, and we consider the significantly lower weight and decreased liver CIII activity in the juvenile F1 mice a striking finding. For this revision, we have measured the hepatic gene expression of Fgf21 and Gdf15, two important metabolic regulators and secreted markers of mitochondrial dysfunction. Interestingly, mt-Cyb p.D254N alone induced the expression of Fgf21 to a similar level as the pathogenic Bcs1l mutation, indicating that it is sufficient to cause a clear metabolic stress signal. Importantly, the Fgf21 upregulation was maintained also in mice backcrossed several times to C57BL/6JCrl ( Fig. 1f and Results, lines 83-88).
The fact that D254N alone does not seem to decrease CIII activity in all tissues is not unexpected and is in line with the extensive literature showing tissue specific effects (pleiotropy) of both mtDNA mutations and nuclear mutations affecting mitochondria (Frazier et al., 2017). This is, in fact, the case for human BCS1L and MT-CYB mutations as well. For example, different BCS1L mutations cause very different clinical pictures, sometimes with mainly visceral manifestations (as in the GRACILE syndrome caused by the S78G mutation, which primarily affect the liver and the kidney) and sometimes no visceral manifestations at all Fernández-Vizarra and Zeviani, 2015) (Discussion, lines 260-265).
Concerning the inter-individual variability in liver CIII activity data (Fig. 2a, n=5/genotype), we already replicated the liver CIII activity measurement from the F1 Bcs1l WT mice with and without mt-Cyb p.D254N (n=8/genotype) and these data was shown in the Supplementary Fig 2d. We have now moved this replication data to main Fig. 2b. The heart CIII activity data did indeed show high interindividual variability that likely originated from the unrepresentatively small (5 mg) frozen tissue samples we originally used. We have now optimized the sample preparation (sample amount, buffer and homogenization; Materials and Methods, lines 352-358) and repeated the CIII activity measurement for the heart. This new data clearly shows the synergistic effect of Bcs1l p.S78G and mt-Cyb p.D254N on CIII activity also in this tissue (Fig. 2a). Interestingly, the new data with less variation also suggests that mt-Cyb p.D254N has an independent effect on CIII activity in the heart.
To rule out type 1 error regarding the independent effect of mt-Cyb p.D254N on liver and heart CIII activity, we have now bred an additional mouse panel that is genetically somewhat different from the original C57BL/6JBomTac:C57BL/6JCrl F1 hybrid mice, as these mice have been backcrossed to C57BL/6JCrl males several times as part of the normal maintenance of the colony. We collected a sample panel from these mice at approximately P30, which is a few days earlier that in the original F1 panel and repeated the liver and heart CIII activity measurements. Importantly, the measurements confirmed the synergistic decrease in CIII activity in the Bcs1l p.S78G ;mt-Cyb p.D254N mice after the further backcrossing to C57BL/6JCrl ( Supplementary Fig. 2). However, the nuclear background does seem to have a slight effect as the hepatic CIII activity was not significantly decreased by mt-Cyb p.D254N alone in this panel. In the heart, however, mt-Cyb p.D254N decreased CIII activity also in the Bcs1l WT mice. We have added these data in the supplement ( Supplementary  Fig. 2).
We have also revised the text throughout, better pointing out the matters presented above.
Presentation of activities and bioenergetics data in Figure 2: The way the authors normalize the data could be potentially artefactual and it would be helpful to see the data in a more straightforward way. In Figure 2 (a), CIII activity is normalized by CIV, why? There is no rational for this. In addition, CIV is measured in two different ways (spectrophotometry and polagrography) depending on the tissues, so the data are not really comparable. It would be more correct and consistent if the authors normalized CIII activity by protein (like they do for CI and CII in Supplemental Figure 2) or by citrate synthase (CS) activity, which is the standard method.
We measured CIII activities for liver and kidney from isolated mitochondria and normalized the activities to protein concentration as stated in the text, and now also more clearly in the figure legend. We only collected frozen heart and skeletal muscle tissue because it is technically possible to measure CIII activities from homogenates of these tissues (unlike e.g. liver) and because we did not want to isolate mitochondria from these tissues due to poor yield from small amounts of tissues (small P30-35 mice). With the homogenates (in contrast to isolated mitochondria), however, normalization to protein concentration gave unacceptable within-group variation. Therefore, we normalized the CIII activity data to the unaffected mitochondrial inner-membrane enzyme, CIV, for these two tissues. Our validation procedure for any normalization method is to first confirm that the normalization is not biased by the experimental set up (e.g. by genotype). Second, we check that the normalization decreases the within-group variation and correlates (within-group) with the target to be normalized. We observed similar CIV activity in all genotypes, and normalization to CIV activity did not alter relative mean values between the groups. It, however, efficiently decreased the within-group variation and correlated with CIII activity (average within-group R 2 = 0.87 for CIII vs CIV correlation). We agree that the CS activity is a widely used marker and it can be an excellent marker for mitochondrial mass (Larsen et al. J Physiol 2012). We, however, found that CS activity poorly correlated with CIV (R 2 =0.24) or CIII activity (R 2 =0.23) in our heart samples. Normalization of heart CIII activity to CS activity actually increased the within-group coefficient of variation from 31% to 42% whereas normalization to CIV activity decreased the coefficient of variation to 13.6%. Moreover, we have previously shown that in Bcs1l mutant mice skeletal muscle CS activity can be affected by a sole dietary modification without obvious increase in mitochondrial mass (Purhonen et al., 2017). Larsen et al. (J Physiol 2012) found that in human muscle biopsies, among 13 markers of mitochondrial mass, CIV activity correlated third best, after cardiolipin content and CS activity, with mitochondrial mass. In sum, we think that normalization to CIV activity for these two tissues is a valid method. We have now clarified our normalization method (Materials and Methods, lines 365-369). Here again, there does not seem to be striking differences in the respiratory parameters between the Mt-Cyb genotypes when comparing them within the same Bcs1l genotype.
Analysis of mitochondrial function by respirometry requires fresh samples and due to the lowthroughput nature of respiromtery we had to spread the experiment over a few weeks. To minimize interday variability, we chose to normalize the data to CIV activity that can be measured inside the respirometry chamber using TMPD and ascorbate as substrates immediately after all other steps. In other words, the maximal oxygen consumption capacity, by the terminal oxidase CIV, was set as a reference state. This strategy provides a robust in-assay normalization as the measurement is performed immediately after the actual respirometry experiment using the same equipment and exactly identical sample amount (Larsen et al., 2012). For the initial respirometry experiments, we used a simple two-step centrifugation that provides a crude but minimally processed, intact mitochondrial fraction. As the respirometry data is central in this study, we have now repeated the whole respirometry experiment with a new independent mouse panel. We also used more purified mitochondrial fraction that allowed us to measure the mitochondrial protein amount more accurately. We have now presented this data both as percentage of maximal CIV capacity (Supplementary Fig. 4) and relative to mitochondrial protein ( Supplementary Fig. 5). These new data faithfully replicate our previous measurements. We have now clarified this normalization method to the reader (Materials and Methods, lines 386-395).
Standardization of respirometry data is an important issue in the field. Unfortunately, currently, comparisons of absolute respiration values between different laboratories is difficult. Respirometry for the Leveen et al. 2011 article was performed a long time ago in a different laboratory, by different personnel, a slightly different protocol and using mice of 129Sv:C57BL/6J mixed background. Thus, we do not find comparing absolute respiration values between these two studies meaningful.
The functional assays in Rhodobacter also indicate that the Mt-Cyb mutation does not have striking consequences in CIII activity. I am sure about the usefulness of the +2Ala mutant shown in figure 4 (h) and (i), this is a drastic mutation that is not found in the mice.
As the reviewer correctly points out, D278N indeed does not have an effect on CIII activity in the purified Rhodobacter bc 1 complex. This is in line with the subtle effect it has alone in mice (it is not a disease-causing mutation). As we mention in the Discussion, only a slight kinetics defect was previously shown in another bacterial model, Rhodobacter sphaeroides, when this mutation was introduced as an artificial experimental mutation (Rajagukguk et al., 2007). We utilized the +2Ala mutant as a positive reference and to illustrate a more prominent effect on the RISP mobility. We find that showing the +2Ala mutant data in Fig. 4 for comparison is useful to estimate the magnitude of the effect of the D254N mutation. +2Ala is an artificial mutation and the rationale of its use is now explained better in the text (Results,.
Therefore, the biochemical and functional data shown in the paper indicate a _mild_ additional effect of the Mt-Cyb mutation on top of the Bcs1l mutation. It is therefore a bit unexpected that Mt-Cyb p.D254N accounts for the striking survival phenotype. The manuscript could do much better fairly dealing with this issue.
Like explained above, the effect of D254N is necessarily subtle because the mice carrying it are basically healthy wild-type mice. In the Bcs1l mutant mice, however, even a slight further decrease in CIII activity can be detrimental. Of the 20 or so known BCS1L mutations, the S78G mutation causes the most severe phenotype in human patients, likely leading to the severest CIII deficiency possible without embryonic lethality (knock-out alleles of respiratory chain subunits, including RISP, are almost always embryonic lethal). The mice carrying the analogous mutation are not quite as sick, surviving to P200, probably because they avoid an early lethal metabolic crisis, but they are still very sick and growth-retarded from weaning on. We have now elaborated the discussion, better bringing forward this point.
Lastly, in the previous publications (cited above) the authors demonstrate that there is a CIII activity decrease with age in the Bcs1l mutant mouse in the old background (Mt-Cyb p.D254N). I wonder if they have checked for the progression of the CIII activity decline with age in the new background (Mt-Cyb WT) to see if it is slower in the critical points and could ameliorate the mouse development and explain the differences in the survival.
Yes, we think the reviewer's point is highly relevant. The Bcs1l mutant phenotype first manifests soon after weaning (approximately P24). In the mt-cyb p.D254N background this coincides with 50% loss of CIII activity, and the activity in liver drops linearly to as low as 10% just prior to death (Levéen et al., 2011 and the current manuscript). Although we lack a similar time series from this age window in the mt-Cyb wt background, our current data and previously measured CIII activities at P95 (Purhonen et al. 2017) and at end-stage P200 (Rajendran et al. under review) are approximately as follows: P30: 40%, P33: 25%, P95: 30% and P200: 35% of wild-type littermate values. In sum, these data show that in the mt-cyb p.D254N background CIII activity rapidly declines to 10% by P35, resulting in lethal metabolic crisis (blood glucose <2), whereas in the mt-Cyb wt background the activity has a nadir (~25%) at approximately the same age but stabilizes or even slightly recovers thereafter. The plain Bcs1l p.S78G homozygotes never develop lethal metabolic crisis but die of lateonset (>P150) dilating cardiomyopathy (Rajendran et al. under review). Again, our crossbreeding experiment rules out the effect of nuclear background on the difference in the course of CIII activity decline, indicating that it is determined by the mt-cyb p.D254N variant alone.
Minor point: Nomenclature is not correct Mt-Cyb and MT-CYB (instead of Cytb and CYTB), CYC1 (not CYTC1), RISP is acceptable, but UQCRFS1 is more correct.

Reviewer #2 (Remarks to the Author):
In this work Purhonen et al. have performed in-depth genetic and phenotypic analyses of two different in-house bred mouse strains, both of which harbour the disease-related homozygous p.S78G mutation in the mitochondrial complex III (CIII) assembly factor Bcs1l (involved in the right insertion of the catalytic Rieske FeS protein subunit in CIII), which display either short or long (4-fold) survival rates depending on the C57Bl/6J nuclear genetic (nDNA) background surrounding the p.S78G mutation. WGS analyses identified only eight genetic variants within coding regions to differ between the strains, one of which corresponded to the novel homoplasmic variant m.14904G>A in the mitochondrial DNA (mtDNA) Cytochrome b (Cytb) gene that is exclusively present in the colony with short survival and it affects a negatively charged amino acid (p.D254N) conserved across eukaryotes, making it a likely genetic modifier of the survival of Bcs1lp.S78G mice. To assess the effect of maternally inherited mtDNA to the survival and mitochondrial function of Bcs1l mutant mice, the authors crossbred Bcs1lp.S78G heterozygotes from the two colonies, and compared F1 progenies carrying either wild-type or variant mitochondria with equivalent nDNA background. Using spectrophotometric and polarographic assays they show that the presence of the Cytb variant by itself seems sufficient to induce mitochondrial dysfunction and a metabolic disease phenotype, and that this mutation further exacerbates the CIII activity and mitochondrial respiration defects present in Bcs1l p.S78G mice. In silico simulations predicted that the Cytb variant would compromise the conformational flexibility of the Rieske FeS protein (RISP) head domain necessary for full CIII activity, and the authors show that the equivalent substitution in Rhodobacter Cytb may induce a mild shift in the occupancy of the Rieske head domain towards the quinol oxidation site that, however, neither affects the integrity and activity of Rhodobacter CIII nor bacterial growth.
"This is an interesting manuscript where the authors clearly show the epistatic contribution of the Cytb variant m.14904G>A to the mitochondrial CIII defects promoted by the p.S78G mutation in Bcs1l. However, the potential pathogenic role of the Cytb variant alone is unconvincing under the experimental conditions used in this work. First, Bcs1l wild-type mice carrying the Cytb p.D254N variant show similar weight and life span to the wild type mice and histologically, they do not show a clear phenotypic defect. Second, the main phenotypic effect observed in these mice is a significant decrease in CIII activity only in liver tissue (but not in kidney, heart or muscle) which, however, is not reflected in a significant decrease in mitochondrial oxygen consumption in liver (neither CI-nor CI+CII-linked respiration differ significantly, so the use of ratios to demonstrate functional differences here is misleading)." We actually do not claim that mt-Cyb p.D254N is pathogenic, and therefore we refer to it as a variant, not as a mutation. Indeed, the mice carrying it have been used as healthy wild-types by us and several other research groups for years. Therefore, we were very surprised that our results showed subtle changes in growth and respiratory chain function in these mice as compared to mice carrying wild-type C57BL mitochondria. Mitochondrial defects commonly produce highly tissue-specific phenotypes and even many severely pathogenic mutations do not usually manifest in every tissue.
In many tissues CIII activity has to decline at least by 50% for a pathological phenotype to arise. A subtle CIII defect may not limit CI&CII-linked respiration in all tissues and cell types, depending on their metabolic status and rate. Ratios of respiratory states have been used since the very first respirometry experiments (Chance and Williams, 1955). They provide an internal normalization and, thus are more sensitive than absolute respiration values to pick subtle differences (Pesta and Gnaiger, 2012). We disagree that the use of ratios is misleading. They provide a functional fingerprint of the electron transport system and, in our case, show that the mt-Cyb p.D254N variant does modify the mitochondrial function. We have now elaborated the text and explained the parameters more thoroughly. We have also added data on oligomycin-induced leak respiration which was significantly decreased by the D254N variant (Fig. 3a, b and Supplementary Fig. 5 and  6).
"The decrease in CIII activity neither correlates with the normal levels/assembly of hepatic complex III observed in these mice. Therefore, how do the authors explain these apparent contradictions?" Indeed, our conclusion from BNGE analysis of CIII composition/assembly was that D254N did not affect CIII assembly, i.e. the amount of RISP assembled into CIII, so as to explain the decrease in CIII activity. We reasoned that its effect has to be more subtle, outside the sensitivity of any measurements we could do using isolated mitochondria. Therefore, we decided to perform molecular dynamics simulations and functional studies in an organism (Rhodobacter capsulatus), in which the cytochrome bc 1 complex can be mutagenized and purified. These two additional approaches both showed that the defect lies in RISP dynamics. Therefore, we do not see any contradiction.
"Moreover, as stated by the authors in the Discussion, the Cytb p.D254N variant appears to be present in the human population as a rare polymorphism, suggesting that this situation could occur in mice. To make a clear statement about the action of the maternally-inherited Cytb variant it is advisable to use alternative experimental models easily subject to stress conditions, for instance, cellular cybrids." The mt-Cyb p.D254N variant may theoretically exist in mice as a rare polymorphisms that was amplified by random drift in the colony we used, but in any case, as mentioned in the manuscript, it is not present in any Mus musculus mtDNA sequence available in GenBank. The robust in vivo stress condition that revealed the presence of this variant in the first place was the Bcs1l mutation compromising CIII function. In other words, the mt-Cyb p.D254N carriers were much more sensitive to the effect of the Bcs1l p.S78G mutation than mice with wild-type mitochondria. It is fully possible that the mice are more sensitive to other genetic, environmental or chemical stress factors affecting or dependent on mitochondrial function, e.g. other mutations or chemicals inhibiting the respiratory complexes, but we feel that such further in vivo studies are outside the scope of this manuscript. The reviewer is right in that cybrid cell lines are an established model to assess the effect of mtDNA mutations. However, in our case, a long history of studies of patient fibroblasts shows that CIII deficiency tends to manifest very mildly if at all in cell lines, and this is the case for the BCS1L p.S78G mutation as well, even though the patients are otherwise extremely sick already at birth Fellman et al., 1998;Kotarsky et al., 2010;Visapää et al., 2002). Recently, a MT-CYB missense mutation (m.15557G>A, E271K), causing a drastic change of a negatively charged amino acid into a positive one in an extremely conserved sequence motif was modelled in cybrid cells. This mutation, predicted to dramatically compromise CIII function, induced only a mild mitochondrial dysfunction in human transmitochondrial cybrids (Iommarini et al., 2018). In sum, we think that the probability of mt-Cyb p.D254N cybrid cells producing meaningful additional data is too small to justify their time-consuming generation for this study. We agree that cybrids may be of interest in further studies if it is feasible to generate such from cell types relevant for the disease pathology, such as a hepatocyte cell line.
The proposed epistatic mechanistic model that converges in the action of the catalytic RISP subunit is very interesting, but it is solely based in computational simulations that lack a convincing experimental demonstration that needs to be provided. As mentioned before, the equivalent Cytb p.D254N variant in Rhodobacter seems to induce a mild defect in the conformational dynamics of the Rieske head domain that does not alter the integrity and activity of Rhodobacter CIII and therefore, this model is not valid to demonstrate the authors´ hypothesis and an alternative experimental model should be presented.
In fact, the epistatic mechanism we propose is not based solely on computational simulations, but on them combined with assessment of CIII function in intact isolated mitochondria (respirometry), mitochondrial membranes (CIII enzymatic activity) and extensive analysis of purified Rhodobacter bc 1 complex. The mouse crossbreeding experiment showed unequivocally that the maternally inherited Cytb p.D254N variant dictated the survival difference between the two colonies. Our further unpublished work on these mice has shown that the survival difference is stable and remains the same after further backcrossing of the Cytb p.D254N carriers to C57BL/6JCrl males. We show that the variant decreased CIII activity in key affected tissues (liver and kidney) and in skeletal muscle and heart of Bcs1l mutant mice. The results from structural modelling, molecular dynamics simulations and analyses of the purified Rhodobacter bc 1 complex, showing a subtle effect on ef loop and RISP motility, are of the magnitude we expected because D254N is not a disease-causing mutation (it is present in a "wild-type" mouse colony). Compared with the various disease-causing patient and artificial mutations previously simulated and assayed in the Rhodobacter model (Borek et al., 2015;Ekiert et al., 2016;Lee et al., 2011), we did not expect the D254N variant to cause a severe defect. Nevertheless, to corroborate the data from the F1 mice, we have now bred another experimental mouse panel using parental mice that have been backcrossed to C57BL/6JCrl for several times (as opposed to the original C57BL/6JBomTac:C57BL/6JCrl F1 data presented in the manuscript) and repeated the respirometry and CIII activity assays from this independent panel, with almost identical results ( Supplementary Figs. 2, 4 and 5). The Rhodobacter model has been widely used and accepted as a model system to study effects of mitochondrial mutations in complex III due to high homology of the core subunits between bacterial and eukaryotic cytochromes bc 1 . See examples in (Borek et al., 2015;Lanciano et al., 2013). The reviewer does not elaborate which alternative model he/she would advise us to use, but the only eukaryotic model organism into which mtDNA mutations can be introduced artificially is yeast. While we do acknowledge the general importance of corroborating results in several experimental models, we find it highly unlikely that studying the D254N variant in isolated yeast mitochondria would provide further insight as compared to mouse mitochondria and the purified Rhodobacter enzyme that we have used in this study. It is theoretically implausible that any other mechanism than the novel mtDNA variant could account for the maternally inherited phenotypic difference between the two Bcs1l mutant strains. The only X chromosomal variant (in Rhox6 gene, see Supplementary Table 1) we identified by WGS is synonymous (Leu51Leu) and the crossbreeding experiment rules out also a sex-specific effect. In sum, we feel that the data we present from three different experimental models is coherent and sound, and that an additional experimental model is not required.

Reviewer #3 (Remarks to the Author):
"This paper provides evidence of a double-trouble effect on the biochemical performance of complex III which worsens dramatically the outcome of the Bcs1l mutation associated with the murine equivalent of the GRACILE syndrome, due to the concomitant presence of a missense homoplasmic mutation in the cytB gene. The genetic results are convincingly demonstrating that in the presence of the cytB mutation the lifespan of the Bcs1l mutant animals falls from about 200 days to 4-5 weeks. This cumulative effect id not caused by obvious structural or assembly defect of complex III in addition to the already severe reduction of complex III amount caused by the Bcs1l mutation. The biochemical characterization in different tissues, mainly liver, kidney and skeletal muscle, demonstrates a cumulative effect in CIII activity in the presence of both mutations ( Figure  2a)in liver, kidney and skeletal muscle but not in the heart.The activity of CIII was also slightly but significantly decreased in the liver of Bcs1l-WT, cytB-mutant animals, compared to the Bcs1l-WT, cytB-WT samples, whereas in other tissues no significant differences were detected.These results suggest a very modest, if any, biochemical effect of the cytB mutation when Bcs1l is WT. Other biochemical results require a more detailed explanation, since for instance the normalization procedures are not very clear and justified. For instance, it is unclear what the authors mean by cIlinked and cI-cII linked in figure 2b,c, and what is the meaning of the uncoupling measurement (perhaps to show the maximal respiration rate associated with the different genotypes? This should be explained and the results commented). The same applies to figure 2d (CI&CII/cI), and ETS coupling efficiency (figure 2e). Which is the biological significance of these results in association with the different genotypes considered in the paper." We have now thoroughly revised the Results text concerning respirometry to be more readerfriendly (Results, lines 110-145). By CI-linked respiration we mean respiration driven by substrate combination providing NADH for CI. Similarly, CI&CII-linked respiration is respiration driven by CI-linked substrates and CII substrate, succinate. In some tissues (e.g. mouse liver) the phosphorylation system (adenine nucleotide translocase, phosphate transporters and the ATP synthase) limits respiration, thus we also measured uncoupled respiration. As mentioned in the response to Reviewer #1, technical replication and the new data obtained with a new mice cohort shows that mt-Cyb p.D254N has an independent effect on CIII activity also in the heart.
"The results shown in figure 2f demonstrate that no increase in H2O2 production is associated with the presence of the mutations, alone or in combination with each other. However, the authors mention in the text that the value of H2O2 normalized to oxygen consumption is increased in the Bcs1l mutant liver and kidney: does this mean that oxygen is used to produce ROS instead of H2O because of the block in the ETC? In any case this effect is restricted to Bcs1l mutants whereas no effect is mentioned attributable to the mutation in cytB." We interpret that higher H 2 O 2 production per O 2 consumption means higher relative leakage of electrons to oxygen before the terminal oxidase, CIV that reduces O 2 to produce H 2 O.
"To test (and confirm) that the two mutations are epistatically linked, the authors made in silico prediction and experiments based on spectroscopic measurements on the phase relaxation of RISP Fe-S cluster by oxidized heme bL. I am not familiar with this technology and I must rely on the conclusions of the authors who conclude that the cytB mutation stiffens cytB and interferes with the interaction between cytB and the Rieske protein, worsening the electron flux in a CIII mutant characterized by severe reduction of RISP incorporation, die to the Bcs1l mutation. The conclusions of the authors are that this is an example of synergistic effects of two mutations, one in the nuclear genome, the other in mtDNA, in determining a substantial worsening of the phenotype linked to CIII deficiency. The existence of causative correlations between mtDNA mutations/polymorphisms with each other has been demonstrated in humans (e.g. for LHON mutations). The possibility of a cumulative effect between a mtDNA mutation and the impairment of a nuclear gene has been also hypothesized, for instance in the ND6 mouse mutant associated with a deletion in the NNT gene. However, the present paper indicates an etiological link of these effect based on structural interactions. The explanation in the present case is based mainly on the experiments on Rhodobacter capsulatus (It is unclear what the 2Ala mutant is...), whereas the mutation in cytB has otherwise no biochemical or clinical phenotype. I am not sure that this experimental setup provides a persuasive explanation about the worsening of the phenotype. The in silico predictions suggest an increased "stiffness" of cytB, which can slightly alter the motion of cytB and, possibly, of the RISP. This effect has no consequences in the Bcs1l-WT, whereas it causes a very dramatic worsening of the clinical phenotype, which is disproportionately severe compared to the differences in the CIII activities, O2 consumption, and other biochemical measurements displayed in Fig. 2." The Bcs1l mutant mice have severe CIII deficiency and are sick from early life, so it is not at all unexpected that even a subtle further decrease in CIII function may dramatically worsen the disease progression via a threshold effect. In humans, different BCS1L mutations are extremely pleiotropic, i.e. cause an unexpectedly wide spectrum of phenotypes even with apparently similar CIII deficiency. Thus, in the light of the literature on mitochondrial disorders and our own work, we do not find the biochemical measurements disproportional in relation to the phenotype observed in the mice. The +2Ala mutant in Rhodobacter work was used as a reference to illustrate the more prominent effect of arresting RISP movement on the temperature-dependent phase relaxation to compare its magnitude with the effect of D278N mutation. +2Ala is an artificial mutation and the rationale of its use is now explained better in the text. We do, in fact, clearly show in the current work that the mt-Cyb variant alone has biochemical and metabolic consequences: significantly lower weight, decreased liver and heart CIII activity, lower leak respiration and altered ratios of different respiratory states. Additionally, for the revision, we have measured the hepatic gene expression of Fgf21 and Gdf15, two important metabolic regulators and secreted markers of mitochondrial dysfunction. Interestingly, mt-Cyb p.D254N alone induced the expression of Fgf21 to a similar level as the pathogenic Bcs1l mutation, indicating that it is sufficient to cause a clear metabolic stress signal ( Fig. 1f and Results, lines 83-88). The mt-Cyb variant dramatically affects survival of the Bcs1l mutant mice. We do not claim that it dramatically worsens the overall clinical phenotype. It is impossible to determine for sure their immediate cause of death, but the most likely cause is lethal metabolic crisis due to extreme hypoglycemia (blood glucose commonly below detection limit, Fig. 1e), which the plain Bcs1l mutant mice escape. In other words, the worsening is directly related to energy deficiency due to the CIII defect. Otherwise, the organ manifestations (liver and kidney disease), which progress fairly slowly, are similar between the colonies, with only proteinuria clearly more pronounced in the kidneys of the Bcs1l p.S78G ;mt-Cyb p.D254N mice ( Supplementary Fig. 1).

Reviewer #4 (Remarks to the Author):
This paper reports a Cytochrome b variant (Cytb p.D254N ) that further decreases CIII activity and capacity of mitochondrial respiration in Bcs1l p.S78G mice. Molecular dynamics simulations of S. cerevisae cyt bc1 and variant Cytb p.D254N indicate that the observed in vivo differences can be attributed to the compromised mobility of the EF loop and to the Iron-Sulfur Protein Subunit Extrinsic Domain -ISP-ED (or RISP head domain). The mutation was also experimentally tested in R. capsulatus (E278N) where no significant differences to the enzyme activity were observed in relation to the WT. A subtle effect was detected on the motion of ISP-ED towards the Qo site. Rajagukguk, S. et al. previously (7): 1791-1798). The novelty of the present work thus resides in the combination of Cytb p.D254N with Bcs1lp.S78G . The effect of D254N alone is very subtle, as showed in the present work and previously by Rajagukguk, S. et al. , where only a small change in the rate constant for electron transfer from the iron-sulfur center [2Fe2S] to cyt c1 was found ( 35,000 s 1 for D278N and 60,000 s 1 for the WT enzyme).
We thank the reviewer for pointing out this important detail. Indeed, in agreement with this earlier data as well as the data presented in this work, we find the effect of the D254N mutant to be subtle also in the simulations. We have now performed additional simulations as suggested by the reviewer and amended the manuscript text accordingly (Supplementary Figs. 6, 8-9).
The MD simulations are state of the art and extensive enough (1 s) to provide adequate sampling. However, the authors should describe in more detail the modelled complex. The model was built from the crystal structure 3CX5, which also includes cytochrome c and the inhibitor stigmatellin, which arrests ISP-ED movement. The inhibitor was obviously deleted, but it seems that the Qo-and Qi -sites were left empty since no information is given about the substrates quinol (QH 2 ) or quinone (Q). If that is the case the simulations do not describe any intermediate of the catalytic cycle (Qcycle). This represents a missing opportunity to study the effect of mutations in the actual catalytic cycle. Due to the proximity of the Qo-site to the EF loop it would be interesting to analyse the effect of the substrate on the EF loop dynamics. The authors should clarify this point.
We have now added a more detailed description of the modeling setup, including the modeling of quinone molecules at the Q o and Q i sites. We performed additional ~1 s simulations of wild-type and mutant systems each by modeling doubly reduced, doubly protonated QH 2 and an oxidized Q at the Q o and Q i sites, respectively ( Supplementary Fig. 6). As the reviewer pointed out, we indeed find that the modeling of QH 2 molecule partly arrests the flexibility of the domains around the site D254, which is only 11-13 Å from the Q o site, and somewhat more in the case of D254N mutant, in agreement with the current and earlier biochemical data.
Finally, taking in account that the effect of the mutation is subtle and since the enzyme is a homodimer, the authors should compare the RMSFs of each monomer.
We have now added new simulations (Supplementary Figs. 8,9) showing data from the individual monomers and also discussed the differences observed in between the monomers in more details.
Minor points: It is stated that: "The atomic partial charges of metal cofactors were taken from an earlier study 49". I suppose the authors wanted to mention that the atomic partial charges and remaining parameters were taken from ref. 49. If the equilibrium bonds, angle and van der Walls parameters were taken from somewhere else that should be stated in the manuscript We have now clarified this in the manuscript. Given the new evidences provided, the conclusion that the subtle Cytb variant m.14904G>A epistatically contributes to the mitochondrial CIII defects promoted by the p.S78G mutation in the assembly factor Bcs1l remains unconvincing, as other different genetic causes (both homozygous and heterozygous, the latter not being contemplated at all by the authors, as well as potential mutations outside coding regions) cannot be excluded as causative of the defect (as also indicated by the authors in their rebuttal letter). Moreover, the bacterial model presented to support the functional influence of such mild mutation is inadequate. The detrimental effect of the cytb mutation by itself needs to be convincingly demonstrated, as the effects shown are very mild (in the best case) or non-existent and the calculations of the statistical significance are based on a very limited number of measurements per experiment. Since the authors have obtained strains that are equalized in nDNA and differ in mtDNA, they could effectively demonstrate the functional influence of the cytb mutation by stressing MEFs or fibroblasts by metabolic switch, among other different experimental approaches. The authors need to convincingly demonstrate the detrimental effect of the mutation in cytb by itself (as it is satisfactorily shown in the paper by Iomarinni et al mentioned by the authors in their rebuttal letter, among others). They could alternatively make cybrids to ensure that the detrimental effect comes exclusively from the mutation detected in the mtDNA since without a clear proof, any other genetic variant exclusively present in the variantcarrying strain can be equally responsible for the phenotypic effects described in this work. In addition, there is no clear correlation between complex III activities, O2 consumption assays and complex III assembly levels. The respiratory chain activity of complex III cannot be normalized by complex IV activity as this complex does not reflect mitochondrial mass. Even if the authors claim that there are no differences in complex IV activities between strains, complex IV is often affected by defects of complex III and it cannot be used for normalization at the authors´ will. If the normalization by standard procedures such as citrate synthase or by miligrams of protein results in increased inter individual variability, this probably reflects the lack of significant differences among sample measurements. The same goes for the use of ratios and the non-standard procedures for the normalization of O2 consumption experiments. In this regard, I am particularly worried by the method used to reach such great statistical significance given the small number of measurements provided and the clear overlapping error bars in most of the experiments. In summary, my major concerns remain unsatisfactorily addressed.
Reviewer #3 (Remarks to the Author): The revised paper has resolved some of the questions I raised in my comments. The work, especially the genetic results, is convincing in showing a synergistic double-trouble effect of a virtually benign polymorphism in cyt b vs. a severe mutation in the UQCRFS1 assembly factor Bcs1l. As I mentioned in my previous comments, this is an interesting finding although the presence of these effects is not an absolutely novelty in the specification of mitochondrial phenotypes. The mechanistic explanation offered by the Authors is interesting and supported by some experimental data obtained in a mtDNA-editable organism such as R. capsulatus.

Reviewer #4 (Remarks to the Author):
In my opinion the authors significantly improved the manuscript by adding to the revised version two additional 1 s simulations of the protein with the substrates ubiquinone and ubiquinol modeled at the Qo and Q1 sites and by analyzing the differences between the monomers. There are still some minor issues with the paper. The labels in the new figures (S8 and S9) are misleading. It seems that the authors are referring to different simulations, when in fact the graphs are for the two monomers. So instead of wild-type 1 and wild-type 2, I would suggest something like WT monomer 1, WT monomer 2.
Furthermore, and since the effects reported in the simulations are subtle, I suggest changing "Accordantly, we observed a simultaneous reduction" to " a simultaneous small reduction" (Page 8, line 170) Finally, the distance between the dihydroxyquinone in the modeled structures and D254/N254 is too large for any direct interaction. So the sentence "this is in part due to the vicinity of the QH2 molecule at the Qo site to the D254/N254 residue (average distance from the simulations between the Q head group and the CG atom of D254/N254 is 13.2/11.6 Å)" should be deleted and instead the authors should analyze the distance(s) between other ef loop residue(s) or nearby residues that are closer to QH2. (Page 8, line 177).

Reviewer #5 (Remarks to the Author):
The ms thoroughly analysis the effect of a mutant of the bc1 complex in mice, and propose it is due to a restricted mention of the Rieske protein, as a result of a D254N mutation. In spite of the impressive amount of work performed, all data regarding the activity of the bc1 mutant, including the work performed with a similar mutant in the complex from the bacterium R. capsulatus, show only, and this are the authors´ words, subtle effects, if any at all. Therefore, this raises a key question -is this mutant really important in causing diseases, when the activity of the complex is barely affected? Also, the molecular dynamics simulations reveal only a very minor effect on the dynamics of the Rieske protein, when the calculations are performed upon docking the quinone substrates. I am not an expert on Pulsed EPR, but at least the EPR spectra of the WT and mutant complexes are also identical. Other specific points are: -Lines 117-118 -If the activity of the bc1 were affected, how the absence of effect on CI-linked respiration increased? It should have happened exactly the opposite, if what the authors claim is really important -Lines121-122 -Again, CI-CII linked phosphorylation is not affected by the mutation, neither maximal electron transfer capacity, which should have been observed if the D254N mutation would be relevant. (lines 141-142).
-Lines 147-152 -ROS production is also not affected (also in the R. capsulatus mutant).
-In fact, the MD simulations contradict the results -while it is claimed, on its basis, that the mobility of the Rieske protein is affected, this does not translate into an effect on its activity! In summary, all biochemical and respirometry data clearly show that the D254N mutation has indeed no effect on the mitochondrial respiration of the mice mutants. Therefore, some other reasons must exist to explain the phenotypes observed (and those, only in some tissues). To all reviewers: 4 We initiated this multi-approach study after observing a consistent, extraordinary 5-fold difference 5 in the survival of CIII deficient Bcs1l mutant mice between two colonies of different strains (in 6 Lund, Sweden and in Helsinki, Finland). Therefore, we performed whole genome sequencing of 7 both strains and were stunned to discover a previously unknown mtDNA variant in the Lund 8 colony, in which the homozygotes had a short survival. Amazingly, of the astronomically high 9 number possible nuclear and mitochondrial genetic variants that could theoretically modify 10 mitochondrial function and energy metabolism (and subsequently the complex multiorgan disease 11 phenotype), the variant we identified alters a subunit of CIII, the same complex compromised due 12 to the Bcs1l mutation. Even more stunningly, the variant is located in the binding site for the 13 electron-transferring subunit, RISP, the assembly of which is compromised by the Bcs1l mutation.

Purhonen et al. A spontaneous cytochrome b variant exacerbates complex III
14 As opposed to nuclear DNA, mtDNA is inherited only maternally. Therefore, it only took a single 15 F1 crossbreeding experiment from the two inbred mouse lines (mice homozygous for every allele) 16 to produce F1 hybrid mice (all nDNA differences in heterozygous state) and prove that the short 17 survival trait is maternally inherited and, thus, must necessarily be caused by the only difference in 18 mtDNA, mt-Cyb p. D254N . Affecting the mtDNA-encoded CIII subunit MT-CYB, around which the 19 whole mammalian 11-subunit CIII is assembled, and which functions as the core structural and 20 catalytic subunit, it was also obvious and theoretically inevitable that the only way this variant 21 could aggravate the mouse phenotype is by further compromising CIII function.

35
The main novelty of this study is that we identify the genetic cause for the drastic survival  (Purhonen et al. 2017, Tegelberg et al. 2017, Purhonen et al. 2018, Rajendran et al. 2018. These mice never become as hypoglycemic as the Lund mice and live to P200 because they escape the 46 early metabolic crisis. We have now clarified these premises of the current study in the manuscript.
We agree with the reviewer that the Fgf21 and Gdf15 mRNA expression data does not explain the 48 phenotypic difference, but that was not our hypothesis either. The purpose of this data was to 49 strengthen the assessment of the possible metabolic effect of the mt-Cyb D254N variant alone, as 50 suggested by the reviewers previously. There was no particular reason to presume that in our CIII The comment about the presentation of RCC activities was virtually ignored. It is still not clear why 59 CIII activity could not be measured in low spin homogenate across the tissues -the method which 60 is commonly used in the field.

61
We apologize for not answering directly to the reviewer's initial proposal of presenting the data 62 relative to protein amount. These methods have not been standardized across different laboratories,

63
especially not among non-clinical laboratories such as ours. Many technical details of the assay 64 such as choice of protein standard and assay chemistry may already introduce two-fold difference in "absolute" values. Thus, we found it more informative to present the CIII activity data relative to 66 healthy control group as this allows inspections of threshold effects (e.g. 50% and 25% of wild-type 67 values) across different tissues and time points.

68
We agree that it is possible to measure activity of several mitochondrial enzymes, such as succinate

75
We have now clarified throughout the manuscript that the mice are dramatically sicker because their 76 CIII activity decreases below survival threshold (25%) in liver and below 50% of wild-type values 77 (threshold for pathology) in other assessed tissues due to mt-Cyb p.D254N , as we show in Fig. 2. This 78 further aggravation of CIII deficiency in multiple tissues could easily collapse the whole-body 79 metabolism, as indeed our new indirect calorimetry data suggests (Fig. 6).

80
The purpose of the respirometry analyses was to assess how the mt-Cyb D254N  in hepatocyte apoptosis (Fig. 1g). In contrast, the Bcs1l p.S78G mice with wild-type mt-Cyb showed milder ATP depletion and even some recovery by P35-36, and never developed massive hepatocyte Davoudi et al. 2016, Rajendran et al. 2016, Purhonen et al. 2017, Purhonen et al. 2018 In addition, there is no clear correlation between complex III activities, O2 consumption assays and 169 complex III assembly levels. 170 We regret that our figure panels did not highlight the most important pieces of the data well enough.

171
We previously provided additional respirometry data from the backcrossed mice as a supplement, 172 which we have now moved to a main figure for clarity. We have now also plotted the correlation 173 between the CIII activity and maximal phosphorylating respiration, and it is actually almost perfect.
The fact that the respiration driven by sole NADH-producing substrates (CI-linked respiration) is 177 not compromised by the CIII deficiency is in line with our previous published work and with 178 studies on threshold effects related to respiratory enzymes. For instance, rat liver CIII has to be 179 inhibited by more than 85% before CI-linked respiration is affected (Rossignol et al. 1999).

180
Therefore, to reveal CIII deficiency in respirometry one has to use convergent electron flow to the 181 coenzyme Q pool via at least two quinol oxidoreductases such as CI and CII (CI&CII-linked 182 respiration).

183
The second fact is that the uncoupled respiration did not separate the genotypes as significantly as 184 the phosphorylating respiration. Interpretation of this respiratory state is, however, quite 185 complicated as CIII is a major proton translocase in the respiratory electron transfer and the 186 uncoupling may shift the rate-limiting steps in the system. It has been shown that the activity of 187 uncoupled CIII activity is approximately 3 times higher than when it is working against membrane 188 potential (Rich andClark 1982, Rottenberg et al. 2009). Moreover, the accurate estimation of 189 maximal respiration by uncoupling is somewhat technically challenging due to different tolerances 190 of healthy and compromised mitochondria to protonophores and off-target effects of oligomycin 191 preparations (uncoupling after oligomycin) (Ruas et al. 2016). 192 In summary, the most important piece of respirometry data is the quasi-maximal phosphorylation The respiratory chain activity of complex III cannot be normalized by complex IV activity as this 207 complex does not reflect mitochondrial mass. 208 Larsen et al (2012)  We apologize for not highlighting this clearly, but we actually did provide the respirometry data 221 relative to mitochondrial proteins as a supplement in the first revision. The conclusions were 222 unaffected. Moreover, liver and kidney CIII activity data are normalized to mitochondrial protein. 223 We find that it is better to normalize the data to the CIII-CIV segment of the respiratory chain by show the magnitude of the effect.

234
In this regard, I am particularly worried by the method used to reach such great statistical  which is the only theoretically feasible way how it can modify disease progression in the CIII 255 deficient mice.

256
As it seems that we initially failed to present the true scope of CIII activity and respirometry 257 measurements in a reader-friendly way, we have now thoroughly revised the manuscript text and 258 the figure panels.

259
References not included in manuscript:  We have now revised this section as referee suggested. The ms thoroughly analysis the effect of a mutant of the bc1 complex in mice, and propose it is due 303 to a restricted mention of the Rieske protein, as a result of a D254N mutation.

304
In spite of the impressive amount of work performed, all data regarding the activity of the bc1 305 mutant, including the work performed with a similar mutant in the complex from the bacterium R.

306
capsulatus, show only, and this are the authors´ words, subtle effects, if any at all. Therefore, this 307 raises a key question -is this mutant really important in causing diseases, when the activity of the 308 complex is barely affected?

309
It is unfortunate that we failed to present and explain clearly in the manuscript and in the first 310 response the crucial genetic (crossbreeding) experiment in Fig. 1b  Yes, the reviewer is correct in that simulations performed by modeling ubiquinones at Q o and Q i 316 sites reveal lower and subtle differences ( Supplementary Fig. 11). However, it is highly likely that 317 the binding of ubiquinol at Q o site is also somewhat affected in the first place, as much larger 318 differences are seen in our simulations without the ubiquinones modeled. Moreover, we have now 319 discussed additional interactions in the revised text that explains the reason behind reduced mobility in the case of Qo/Qi sites occupied. Furthermore, and as noted above, the mt-Cyb p.D254N is a non-321 pathogenic variant present in a wild-type mouse colony. Therefore, it cannot have a drastic or 322 detrimental effect on any parameter on its own.

323
I am not an expert on Pulsed EPR, but at least the EPR spectra of the WT and mutant complexes are 324 also identical.

325
It is actually quite important that the EPR spectra of the Rieske cluster of the WT and mutant do not Please, see our response to reviewer 2. In brief, CI is a relatively slow-rate enzyme and insufficient 336 to reduce the coenzyme Q pool enough to reveal the CIII deficiency inside respirometry chamber.

337
Moreover, NADH cannot be provided directly for CI, as it is impermeable to mitochondrial inner 338 membrane, and thus NADH has to be generated indirectly via substrates for TCA cycle enzymes 339 which may introduce additional rate-limiting steps into the system. Therefore, a convergent electron 340 flow via CI and CII is needed to reveal an acutely sublethal CIII deficiency inside respirometry 341 chamber.

342
-Lines121-122 -Again, CI-CII linked phosphorylation is not affected by the mutation, neither CI&CII-linked phosphorylation respiration as expected from CIII activity measurements.

349
-Lines 147-152 -ROS production is also not affected (also in the R. capsulatus mutant). 350 We agree that our data do not suggest increased ROS production by CIII. However, increased ROS 351 production from other sources may still play a role.

352
-In fact, the MD simulations contradict the results -while it is claimed, on its basis, that the  In summary, all biochemical and respirometry data clearly show that the D254N mutation has 370 indeed no effect on the mitochondrial respiration of the mice mutants.

371
We have now made major revisions to the figure panels, as they seem to have been misinterpreted.

372
The CIII activity measurements showed consistent further decrease in activity in four different 373 tissues by mt-Cyb p.D254N in Bcs1l mutant mice. In line with this, respirometry also showed further 374 decrease in CI&CII-linked phosphorylating respiration. Moreover, the results are consistent across 375 two independent sample sets. Please, see also response to reviewer 2.

376
Therefore, some other reasons must exist to explain the phenotypes observed (and those, only in 377 some tissues).

378
It is unfortunate that we failed to present and explain clearly in the manuscript and in the first 379 response the crucial genetic (crossbreeding) experiment in Fig. 1b and c, which shows 380 unequivocally that the short survival was inherited maternally and therefore must necessarily be 381 caused by the mt-Cyb p.D254N variant. Please, see also response to reviewer 2. We hope that in the 382 current revised manuscript we present and discuss these crucial data more clearly and thoroughly.