INTRODUCTION

Top Previous Next References

Mitochondrial DNA (mtDNA) is essential for most eucaryotic (obligate aerobe) species. Evidently, maintenance of the integrity of mtDNA during somatic divisions and sexual reproduction is of extreme importance for their survival. In particular, deletions, duplications, and point mutations in mtDNA have been shown to cause human diseases, either hereditary or sporadic (see references 115, 138, 174, 318, and 337 for reviews). Cases where mutant and wild-type mtDNA molecules are found in the same tissue define a situation termed heteroplasmy. However, the ratio of the two mitochondrial haplotypes often changes over the life of an individual and can be very different between cell types. Despite their fundamental and clinical importance, the factors which act upon the relative distribution of normal and mutant mtDNAs over individual development remain poorly understood. Recently, experimental studies have attempted to address this question (reviewed in reference 22): one of these stresses the roles of the nuclear background (77). In this context, it is noteworthy that several nuclear genes have been implicated in the concomitant accumulation of several classes of deleted mtDNAs in the same individual. Of these, only one has been recently cloned and characterized (219). It encodes thymidine phosphorylase, which is involved in thymidine catabolism and may be essential for mtDNA maintenance.

While very little information is available on nuclear genes which might (directly or indirectly) control mtDNA maintenance in higher eucaryotes, there is an overabundance of data concerning the budding yeast Saccharomyces cerevisiae. In this species, mitochondrial genetics and studies of nuclear control of integrity and transmission of mtDNA are simplified because of two features: (i) this yeast is a facultative aerobe, and (ii) it is very accessible in terms of classical and molecular genetics. However, in spite of its status as a model system, S. cerevisiae exhibits intrinsic weaknesses with respect to these questions: it is a facultative aerobe, it is unicellular and does not stably maintain a heteroplasmic state, and the structure of its mtDNA differs significantly from that of higher eucaryotes. Nonetheless, one can certainly postulate that at least some of the nuclear factors which control mtDNA integrity and transmission have been conserved through evolution. Thus, an understanding of this nucleomitochondrial problem in yeast may shed light on the relevant nuclear genes in higher eucaryotes. Alternate models are rare and have received very little study (see "Conclusions and prospects" below).

This review examines the field of yeast genes involved in mtDNA maintenance. Due to the large number of articles in this area, we may have missed some significant data and apologize for possible omissions. The bibliography was compiled in May 1999.

The symbols used in this review are those currently used by the yeast community: wild-type genes are in capitals and italics, mutant alleles are in lowercase italics, and proteins are in nonitalics and capitalized only on the first letter (e.g., Mip1 or Mip1p).

The cytoplasmic petite mutants (petite colony, vegetative petites, or simply petites) were discovered 50 years ago (84, 85, 278; see reference 87 for a review). Numerous studies then showed that petite mutants lacked a functional mitochondrial (rho⁺) genome but instead showed extensive deletion of mtDNA (rho) or no mtDNA at all (rho⁰) (reviewed in reference 74; see also reference 238). The petite mutations are very pleiotropic because the mutants cannot perform mitochondrial protein synthesis. These mutations differ in this from mit mutations, which are point mutations in mtDNA causing respiratory deficiency through impairment of only one (or a few) specific function (see reference 281 for a definition of mit mutations and the operational differentiation between mit and rho/rho⁰ mutations). The petite mutations are also characterized by their high rate of spontaneous occurrence: in the majority of yeast strains, one finds a small percentage of spontaneously occurring petite clones. However, it should be stressed that the proportion of petite clones with respect to the wild-type clones (often referred to as frequency, percentage, or ratio of petites) is a classical problem of population genetics. The proportion one establishes experimentally results from both the frequency of mutation of rho⁺ cells to rho/rho⁰ cells (measured per cell per generation or per cell per hour) and the ratio of the growth rates of rho⁺ cells to rho/rho⁰ ones. L'Héritier and coworkers (86) have shown that under stationary conditions, when the proportion of petite clones is of the order of 1%, the frequency of mutation is 10 times lower, on the order of 0.1%. Conversely, Ogur and John (223) have shown that the proportion of petite clones, in a population that has multiplied on nonfermentable medium (i.e., ethanol or glycerol) where the petite cells are unable to multiply (but do not die), is equal to the frequency of mutation multiplied by 2. This frequency can be greatly increased (to complete transformation of the cell population into petites) by drugs such as ethidium bromide (280).

In rho mutants, the remaining fragment of rho⁺ mtDNA is regularly repeated along the mtDNA molecules. There are two major types of repeat arrangements: either the sequence is directly repeated with head-to-tail junctions or the arrangement is of the inverted or palindromic type and the repeat unit is an inverted duplication of the retained sequence (91; reviewed in reference 74).

Numerous studies have shown, by crosses between strains with different mitochondrial genotypes, that the heteroplasmic state of the zygotes is very transient. In fact, their vegetative multiplication is coupled with a rapid segregation of mtDNA molecules, leading to homoplasmic clones. After a cross to the wild type, the petite mutants can be classified according to their suppressiveness, i.e., by the relative proportion of petite zygotic clones. Low or moderate suppressiveness probably results from random segregation of mtDNA molecules over cell divisions, while hypersuppressivity implies that rho molecules are preferentially transmitted to the progeny at the expense of the rho⁺ genome. Numerous data have led to the proposal of a replicative advantage for these rho molecules, since they contain several copies of short sequences called ori (rep), which may represent replication origins of mtDNA (reviewed in reference 74). However, more recent data tend to favor the idea of a segregation (transmission) advantage in hypersuppressiveness (see "Nuclear control of biased transmission" below).

This review focuses preferentially on mutations which increase the level of petite production compared to the relevant wild-type background. However, it also concerns the few available mutations which either modify the biased transmission of some rho genomes (see "Nuclear control of biased transmission" below and Table 2) or slow (even abolish) petite production (see "No compromise" below and Tables 10 and 13). It would have been incomplete without a short summary (see "Petite positivity versus petite negativity" below) of recent data concerning the problem of petite positivity versus petite negativity, i.e., the ability or inability of yeast species to produce petite cells. The genes involved in mtDNA maintenance have been roughly classified by their known functions or effects on the mitochondrial genome (see the tables). However, at the beginning of the review, we have focused on the scientific projects which have helped to identify some genes.

Throughout this review, as far as possible, we have attempted to identify the type of petites (rho⁰ and/or rho) produced in a given genetic background. We are in fact convinced that this is an important point for the understanding of two main problems: (i) why some genetic backgrounds lead to a complete loss of mtDNA (rho⁰ petites) if the relevant genes are not (directly or indirectly) required for mtDNA replication, and (ii) why some genes are essential for the maintenance of a complete (rho⁺) genome but dispensable for the maintenance of a truncated version (rho) of this genome. Several criteria are available: (i) ensuring that the petite colonies which appear at high levels in a given strain are respiration deficient; (ii) ascertaining a defect in mtDNA; and (iii) distinguishing between rho⁰ and rho petites (see Table 1 footnotes). "Physiological" methods can address only the first goal. Genetic and molecular methods clarify the second. The third problem, however, can be solved by 4',6-diamidino-2-phenylindole (DAPI) staining or extensive genetic tests and/or clear-cut molecular procedures. Unfortunately, in some cases mtDNA stability was clearly not a focus of the authors and in other cases the methods which distinguish between rho⁰ and rho petites were not rigorously applied. These facts have prompted us to be (sometimes) more cautious about conclusions than were the authors. For example, in our opinion, lack of DNA hybridization with a probe containing a limited mtDNA fragment does not justify the conclusion that the relevant clone is of the rho⁰ type: it may very well contain a rho genome with this fragment deleted. The same is true for genetic tests using too few mtDNA markers. It is indeed easier to show that a petite is rho than to demonstrate a rho⁰ state. In addition to these methodological problems, other parameters require that caution be used in drawing definitive conclusions: the overall genetic background and the physiological state of the cells can strikingly modify the amount and nature of petites. As emphasized in some cases, these parameters may explain discrepancies between results obtained by different groups. To underline the influence of the genetic context, it seems that a mitochondrial genome lacking introns is more stable than an intron-containing genome (see below). To place the physiological parameters in a more concrete form, three observations can be reported. First, environmental parameters can strikingly influence petite production in mutant strains. A remarkable example concerns the effects of both temperature and culture medium on the abf2 mutants (see "The ABF2 paradox" below). Second, most rho mutants are not very stable and may evolve to the rho⁰ state (see, as an example, the discussion of overexpression of ADR1 in "Gene overexpression can increase mtDNA instability" below). Thus, a conclusion about the nature of the petites produced by a given mutant may depend on the age of the culture (see, for instance, the discussion of MGM101 in the following section). Third, petite production due to disruption of a nuclear gene can be significantly different depending on the procedures used. Disruption can be achieved in a haploid vegetative cell, and its effect upon petite production can be immediately observed; disruption can be achieved in a diploid cell, and its effect is observed in haploid cells resulting from sporulation of the heterozygous diploids. These two procedures can lead to different observations (as, for instance, in rpo41 mutants; see "Does transcription play a direct role in mtDNA maintenance?" below and Table 3).

Nuclear genes involved in the maintenance of mtDNA can be identified by a systematic search for mutations increasing rho⁰ production at elevated temperature. Screening for heat-sensitive mutations helps to recover mutations in genes essential for cell survival. This procedure was used by Genga et al. (103): among 360 clones able to grow on glycerol at 25°C but not at 35°C, 31 clones were heat sensitive due to massive production of petites, of which 12 completely lacked mtDNA. These 12 mutants belonged to 11 complementation groups. Only one mutant displayed a deficiency in mtDNA synthesis. This identified the MIP1 gene (Table 1), encoding the catalytic subunit of mtDNA polymerase (94). In addition to the initial mutations and deletion, other mip1 mutations impairing replication fidelity have been obtained. Some yield a high level of petites either at 36°C or at both 36 and 23°C. Interestingly, these mutations show a gene dosage effect, since heterozygous diploids are strong producers of petites (139). This effect was also observed (at 36°C) for diploids containing a single copy of the MIP1 gene (177). The data suggest that Mip1p is produced in limited amounts and that the rho⁺ genome cannot be maintained below a particular threshold of that protein.

In 1992, Jones and Fangman (149) used a similar procedure to identify genes required for mitochondrial genome maintenance (MGM genes). This led to the identification of MGM1, encoding a protein whose GTP-binding domain is related to dynamin (117, 149) (see Table 11). Careful examination of DAPI-stained cells of the mgm1-1 mutant (which lost mtDNA after transfer at high temperature) strongly suggests that the defect of the mutant is not due to an asymmetric distribution of mtDNA during cell divisions. Furthermore, deletion of MGM1 causes not only loss of the rho⁺ genome but also that of rho genomes, even those containing a rep (ori) sequence (149). In addition, Guan et al. (117) showed that disruption of MGM1 led to a structural alteration of mitochondria and a defect in mitochondrial Hsp58 protein import. Recently, Shepard and Yaffe (272) have demonstrated that Mgm1 is an integral protein of the outer mitochondrial membrane, which plays a key role in mitochondrial morphology and inheritance (see "Mitochondrial morphology, the cytoskeleton, and mtDNA" below for a further discussion of MGM1).

More recently, Guan (118) described a similar program which led to the recovery of 120 mutants exhibiting temperature-induced loss of mtDNA (rho⁰). These mutants belong to 36 complementation groups. Thirteen mutants contained mutations that could be assigned to six previously described genes: MIP1 (see above and Table 1), CDC8 and CDC21 (see "mtDNA and the cell cycle" below) (Table 1), RPO41 and MTF1 (see "Does transcription play a direct role in mtDNA maintenance" below and Table 3), and MGM1 (see above, "Mitochondrial morphology, the cytoskeleton, and mtDNA" below, and Table 11). This genetic analysis thus offered 30 new genes involved in mtDNA maintenance, of which 2 have been analyzed to date: MGM101 (Table 1) and MGM104 (see Table 9).

The MGM101 gene was sequenced and disrupted. The predicted amino acid sequence showed no similarity to sequences of known proteins. The disrupted gene led to loss of mtDNA (38). Recently, Meeusen et al. (205) found Mgm101 among proteins associated with mitochondrial nucleoids. These authors have also isolated a new heat-sensitive allele of MGM101 (mgm101-2) and attempted, by extensive biochemical and genetic approaches, to determine the precise function of Mgm101p. The data have led to the following conclusions. The protein binds to DNA, which probably explains its specific association with the nucleoid. However, it is not involved in mtDNA packaging, segregation, or partitioning. It is not required for mtDNA replication. Its loss does not cause defects in morphology or transmission of the organelle. In fact, the only clear-cut phenotypic defect of mgm101-2 is the sensitivity of mtDNA to damage induced by UV irradiation and the hypersensitivity of mtDNA to damage induced by gamma rays and H₂O₂. Therefore, the authors propose that MGM101 performs an essential function in the repair of oxidatively damaged mtDNA required for mtDNA maintenance. Interestingly, using the heat sensitivity of mgm101-2, they observed that after 12 h at 37°C, when the cell population is 100% respiration deficient, only 50% of cells are rho⁰, while after a longer exposure to nonpermissive temperature (16 h), the entire population consisted of rho⁰ cells. Thus, at least in this case, this rho⁰ producer exhibits a transition from the rho⁺ to the rho state before undergoing a complete loss of mtDNA (205).

The mgm104-1 mutation causes a greatly increased production of petites at 25 and 37°C, with a stronger effect at high temperature. Attempts to clone the MGM104 gene led to the identification of the TTS1 gene, which encodes the cytosolic tyrosyl-tRNA synthetase (44). TTS1 is able to rescue the mgm104-1 phenotype, even when present on a low-copy-number vector, but is not thought to be the cognate MGM104 gene. This conclusion was drawn since no specific mutations were found in the coding or promotor regions of the TTS1 gene in the mgm104-1 mutant (118). The author suggests two major hypotheses: either Tts1p has a dual function in cytosolic translation and mtDNA maintenance, or it may interact with the mgm104-1-encoded protein and enhance its activity, thus rescuing the phenotypic defect of the mutant. However, one cannot exclude a third hypothesis: a slight overexpression of the tyrosyl-tRNA synthetase results in mischarging of noncognate tRNAs and thus in informational (missense) suppression of the mgm104-1 mutation. The author suggested that additional plasmids able to complement this mutant were required to identify the cognate gene. However, since the TTS1 mRNA is about 60% less abundant in the mgm104-1 strain than in wild type, one cannot exclude the possibility that the mutation lies in the 3' noncoding sequence of TTS1: this would lead to instability of the messenger and cause the mgm104-1 phenotype. In fact, alteration of cytosolic translation may affect mtDNA stability, as shown in the filamentous fungus Podospora anserina (see "Conclusions and prospects" below).

The main advantage of such a wide screening procedure for nuclear mutations leading to complete loss of mtDNA is that it does not assume any mechanism. This is stressed by the few examples cited above: this type of program has permitted the recovery of genes involved in mtDNA replication, genes involved in mtDNA transcription, and genes whose functions are still unknown. Guan's assessment (118) also shows that the screening procedure is far from saturated. This explains why several genes whose mutations lead to rho⁰ production were not identified in his experiments. Among those involved in mtDNA metabolism, one finds, for instance, PIF1, RIM1, ABF2 (discussed below), and PPA2, which encodes a mitochondrial pyrophosphatase (Table 1). Inorganic pyrophosphate (PP_i) is generated in a number of biosynthetic reactions, notably replication and transcription. Hydrolysis of PP_i is thus an essential process which ensures the forward direction of these reactions. Lundin et al. (197) cloned the gene (PPA1) encoding the cytosolic enzyme and used it as a probe for the gene (PPA2) encoding the mitochondrial enzyme. Cells with PPA2 deleted are unable to grow on respiratory carbon sources. DAPI staining showed that these cells have lost their mtDNA (Table 1), which is in accordance with the evident requirement for this enzyme in DNA replication (197).

In addition to genes specifically involved in mtDNA replication discussed in this section, four genes, required for replication of the mitochondrial and the nuclear genomes (CDC8, CDC21, RNR2, and RNR4) were shown to control mtDNA stability (Table 1). Because of their dual cellular function, it seemed to us more convenient to discuss these genes below in "mtDNA and the cell cycle."

As stressed in the Introduction and in "The petite syndrome" (above), mtDNA maintenance is dependent on two basic parameters: the frequency of mtDNA deletions and the management of heteroplasmic states, i.e., the relative transmission of wild-type and altered molecules during cell divisions. In 1991, Zweifel and Fangman (339) found a mutation which disrupted the biased inheritance of hypersuppressive rho genomes (Table 2), thus identifying the first MGT (for "mitochondrial genome transmission") gene. Strikingly, the MGT1 gene was rediscovered by a totally different procedure. Originally, Kleff et al. (158) were interested in a cruciform cutting endonuclease (Cce), which is required during recombination to resolve Holliday junctions. They developed an assay to detect this activity in crude yeast extracts and sought mutants lacking it. Approximately 150 heat-sensitive mutants were screened before a relevant (cce1) mutant was found. Cloning of the CCE1 gene and assays for Cce1p activity yielded evidence that CCE1 was the structural gene for the cruciform cutting endonuclease. It was shown that cce1 mutants did not affect nuclear recombination but that CCE1 and MGT1 were allelic: the mgt1 mutants showed no Cce1p activity and, in turn, the cce1 mutant was resistant to hypersuppressive rho. It was then demonstrated that Cce1p was localized to the mitochondria and associated with the inner membrane (89). Later, Lockshon et al. (194) demonstrated that the lack of Cce1p led to an increased ratio of mtDNA molecules linked by recombination junctions, resulting in aggregation of mtDNA molecules in a smaller number of cytological structures than in wild-type cells. This effect was more pronounced for rho than for rho⁺ genomes. In accordance with these results, overexpression of MGT1/CCE1 reduced the frequency of branched structures and enhanced the transmission of nonsuppressive rho genomes (Table 2). In a recent report (239), it was shown that the mgt1/cce1 background also influences the transmission of respiration-competent mtDNAs lacking intergenic sequences involved in the biased transmission of hypersuppressive rho mutants. Thus, the resolution of recombination junctions plays a key role in the segregation and transmission of yeast mtDNA. However, deletion of MGT1/CCE1 causes only a slight increase in rho production (158). It is noteworthy that this increase is more marked in the presence of a deletion of the NUC1 gene, encoding the major mitochondrial exoendonuclease. This gene was cloned by reverse genetics, using an antibody against the protein. Unexpectedly, neither deletion nor overexpression of the gene resulted in phenotypic effects (334). However, at that time, the authors did not perform extensive analyses of rho production or recombination rates in these two contexts. This should now be undertaken in the light of the interaction between nuc1 and mgt1/cce1 deletions.

View this table: [in this window] [in a new window]   TABLE 2.   Genes which influence biased transmission of rho genomes

Unfortunately, despite its evident interest, very few analyses have been devoted to the problem of biased transmission. However, in addition to the MGT1/CCE1 gene, data concerning the PIF1 gene are available. This gene (described in "The recombination/repair track" below) encodes an mtDNA helicase involved in recombination between rho⁺ and tandemly organized rho genomes. With respect to rho transmission (in crosses with a rho⁺ partner), a high level of suppressiveness was observed when diploids were homozygous for the pif1 mutations. This occurred for all types of rho, regardless of structure (tandem or inverted organization), and was not dependent on the sequence or the size of the repeat (96) (Table 2). This general effect of the pif1 mutants remains unclear. However, when secondary rho clones were obtained from a tandem rho insensitive to the recombinogenic effect of PIF1, those which showed PIF1 sensitivity became hypersuppressive in a pif1 context (98).

Thus, the MGT1/CCE1 and PIF1 data suggest a strong link between mtDNA recombination and transmission at least for rho genomes. With respect to this remarkable point, a systematic analysis of other genes involved in recombination (see "The ABF2 paradox" and "The recombination/repair track" below) should be undertaken.

The presence of a histone-like protein inside mitochondria (HM) was first reported 20 years ago (32). This 20-kDa protein was then rediscovered by Certa et al. (33) in the search for a histone H1 in yeast. It was lastly and serendipitously identified as an ARS binding factor, i.e., a protein binding an autonomously replicating sequence of the nucleus (68). A few years later, it was shown that Abf2 was a mitochondrial protein closely related to DNA-binding proteins of the HMG (high-mobility-group) family (69). These authors showed that abf2 null mutants lose their mtDNA when grown on rich glucose medium (Table 1). However, when tetrads issued from diploids heterozygous for the abf2 null mutation were dissected directly on a nonfermentable substrate, the mutant strains could grow and be maintained indefinitely.

Evidence that HM and Abf2 were the same protein was provided by Megraw and Chae (206). In addition to confirming the data of Diffley and Stillman (69), they added two interesting observations. First, disruption of ABF2 results in a heat-sensitive phenotype on respiratory substrates due to mtDNA instability, as shown by the fact that the abf2 mutant is lethal in a op1 context (see "No compromise" below and Table 13 for the op1 mutation and its lethality in a petite background). Second, overexpression of ABF2 leads to a high frequency of rho⁰ petites (Table 1). The same year, three papers dealt with the issue of the evolutionary relationship of Abf2 and DNA-binding proteins involved in DNA maintenance and/or expression. Abf2p is the functional homologue of the bacterial histone-like protein HU: despite the lack of sequence similarity between the two proteins, Abf2 and HU can substitute for each other (206). When directed toward the mitochondria, a yeast nuclear HMG protein is able to complement an abf2 mutant (153). Functional similarities were discovered between a human mitochondrial transcriptional activator (mtTFA or mtTF1) and Abf2p: the human protein partly rescues the phenotypic defects of a abf2 strain (227). However, the two proteins display structural differences. In particular, the tail region of mtTFA, which plays a critical role in its transcriptional activity, is absent in Abf2p (59a). The data as a whole (see also reference 309a) indicate that, unlike its human homologue, Abf2p plays only a small, if any, role in mtDNA transcription, a notion supported by the fact that abf2 mutants are able to maintain and express mtDNA under specific growth conditions (69, 336).

In a recent report, Zelenaya-Troitskaya et al. (335) exemplified the pleiotropic role that ABF2 may play with respect to mtDNA maintenance. These authors show (i) that the distribution and mixing of mtDNA from two abf2 parental cells after zygote formation differ slightly from the patterns observed in wild-type zygotes; (ii) that recombination between mitochondrial markers is reduced five- to sevenfold in the mutant context; (iii) that rho genomes (introduced through cytoduction experiments) are stable in abf2 mutants, in contrast to rho⁺ genomes; (iv) that (in accordance with the data of Megraw and Chae [206]) hyperexpression of ABF2 leads to a complete loss of mtDNA (Table 1); and (v) that a moderate (two- to threefold) increase in the level of the protein raises the content of mtDNA without affecting mtDNA stability. These data have been completed by Okamoto et al. (224a) and by MacAlpine et al. (198). In the first case, the authors showed that in a rho⁺ × rho⁰ zygote, mtDNA (stained by DAPI) is preferentially transmitted to the diploid bud while (GFP-tagged) proteins have equilibrated throughout the zygote and the bud. In homozygous abf2 crosses, mtDNA sorting is delayed and preferential sorting is reduced (224a). In the second case, the authors analyzed the effect of either a null allele or overexpression of ABF2 upon levels of mtDNA recombination intermediates. In rho⁺ genomes this level is reduced when Abf2p is absent and increased when Abf2p is overproduced. In contrast, the absence of Abf2p does not influence the level of recombination intermediates of rho genomes (198).

Three multicopy suppressors of abf2 phenotypes have been isolated to date. Two (SHM1/YMH1 and YMH2) were identified by their ability to restore the growth of abf2 null mutants at 37°C on a respiratory substrate. They could encode mitochondrial carrier proteins. Deletion of SHM1/YMH1 has no effect on mtDNA in a ABF2 context (152). Interestingly, YMH2, whose deletion leads to respiration deficiency, appears to be the first membrane-bound protein associated with the mitochondrial nucleoid in vivo (43). We discuss this important property of YHM2 in "Is there a centromere-like structure for the mitochondrial genome" (below). The third suppressor is ILV5, isolated by its ability to restore mtDNA stability in an abf2 null mutant grown on rich glucose medium. The gene encodes an enzyme of the biosynthesis of branched-chain amino acids (see "The secret life of well-known proteins", below, and Table 8) (336). This discovery and subsequent analyses revealed that mtDNA is highly stable in abf2 mutants grown on glucose medium lacking leucine, isoleucine, and valine, a condition which enhances the expression of ILV5. The inevitable conclusion is therefore that ABF2 is dispensable for mtDNA stability when ILV5 is sufficiently expressed! The effect of ILV5 on petite production by abf2 mutants at high temperature has not yet been tested.

In any case, the data as a whole suggest that Abf2p plays a role in the recombination process. However, they raise more questions than they solve. Why (as observed in other cases) are rho genomes stable in the abf2 mutants while rho⁺ genomes are not? Why do abf2 mutants completely lose their mtDNA even though they are able to maintain rho genomes? Why does hyperexpression of ABF2 lead to loss of rho⁺ and rho genomes? Why are abf2 null mutants unable to maintain a functional mtDNA on respiratory substrates at high temperature? Zelenaya-Troitskaya et al. (335) favor the idea that the main effects of ABF2 result from its role in recombination. This hypothesis is consistent with many data concerning both the properties of HMG proteins and the role of recombination intermediates in mtDNA transmission (reference 335 and references therein) (see also the discussion of MGT1/CCE1, above). However, as pointed out by the authors (335), this hypothesis is difficult to reconcile with the data obtained from the MGT1/CCE1 analysis: in this case, lack of the protein has a very slight effect on mtDNA stability (194). Another interpretation (198) suggests an indirect effect of Abf2p on replication through recombination structures. However, the naive point of view would be that ABF2 is highly pleiotropic: in fact, as a DNA-packaging protein, which plays a role in mtDNA and nucleoid organization (218), Abf2p may be (directly or indirectly) involved in mtDNA replication, recombination, and distribution. The data would therefore be explained by the sum of discrete effects, more or less pronounced depending on the growth conditions and the structure (rho⁺ or rho) of the mitochondrial genome.

Yeast mitochondria are very active in recombination (76), and it has been suggested that mtDNA deletions which give rise to the rho genomes can occur through homeologous recombination between imperfect repeats (102, 279). Although this mechanism does not apply to all petite deletions, mutations affecting recombination may aid in identifying genes involved in mtDNA maintenance.

A direct screen for recessive mutations altering mtDNA recombination is impossible since the experimental test requires the fusion of two cells carrying different mtDNAs: in the resultant diploid, the nuclear mutation cannot be expressed. However, Foury and Kolodynski (96) succeeded by using a genetic ruse, i.e., a mutation which delays nuclear fusion; thus, the daughter cells of the zygote contain a haploid nucleus along with a mixed mtDNA population. By using this procedure, they obtained three allelic mutants which identified the PIF1 (for "petite integration frequency") gene. The pif1 mutations affect the frequency of recombination between rho⁺ and certain rho genomes, which are tandemly organized. The rho genomes with an inverted organization are not sensitive to these mutations. Similarly, recombination between rho⁺ genomes is identical in the wild-type and mutant backgrounds. The pif1 mutations are pleiotropic since, in addition to their specific effects on recombination, they increase the level of rho induced by UV irradiation, cause a total loss of mtDNA at high temperature (Table 1), and increase the suppressiveness of all types of rho genomes (Table 2). PIF1 encodes an mtDNA helicase which is thus involved in recombination, repair, and mtDNA maintenance (169). Amazingly, it was later shown that the protein also acts in the nucleus by inhibiting telomere elongation and de novo telomere formation (268). In fact, PIF1 encodes two forms of the protein through the alternative use of two AUG codons: the longer form goes to the mitochondria, and the shorter form goes to the nucleus.

Although its role in mtDNA repair and recombination remains to be elucidated, we must mention here the RIM1 (for "replication in mitochondria") gene. RIM1 was discovered as a partial suppressor of the heat-sensitive phenotype of the pif1 null mutants (310). Interestingly, the suppressor effect was observed with a few additional copies of the RIM1 gene (i.e., when carried on a centromeric vector). Unfortunately, it has not been reported if strong RIM1 overexpression could lead to a complete suppression of pif1 mutants. RIM1 encodes a mitochondrial single-stranded DNA-binding (SSB) protein required for mtDNA maintenance; deletion of this gene causes complete mtDNA loss (310) (Table 1). Since the heat sensitivity of the pif1 disruptants remains unclear, it is difficult to speculate on the precise role of Rim1p in mtDNA stability. However, SSB proteins and DNA helicases often perform complementary functions in DNA metabolism, involving unwinding of duplex DNA. Thus, Rim1p might be involved not only in mtDNA replication but also in mtDNA recombination, repair, and transcription (310). As suggested by the authors (310), it would be interesting to test the ability of RIM1 to suppress the defects of the pif1 mutants in mtDNA repair and recombination. It could also be informative to know the effect of strong overexpression of RIM1 in a PIF1 (wild-type) background.

A genetic procedure linking mtDNA repair and recombination was used by Ling et al. (183). Of two mutants displaying an elevated UV induction of rho mutants, one also exhibited a deficiency in gene conversion. The MHR1 (for "mitochondrial homologous recombination") gene is also involved in mtDNA maintenance at high temperature (Table 1). Genetic localization showed that MHR1 was a previously unknown gene.

A direct search for genes involved in DNA repair was undertaken by Reenan and Kolodner (245) to identify the yeast homologue of the mutS gene, which encodes a component of the bacterial MutHLS system. This was performed by reverse genetics, using degenerate oligonucleotide primers and PCR. Two genes were identified, one of which encoded a protein targeted to the mitochondria. Disruption of the MSH1 (for "MutS homologue") gene leads to a strikingly rapid accumulation of rho mutations, mostly hypersuppressive (246) (Table 1). DAPI staining of the mutant cells revealed large fluorescent patches instead of the small dispersed patches observed in wild-type cells. This phenotype might be the result of either altered mtDNA distribution or abnormal morphology and distribution of mitochondria (246). In addition, Chi and Kolodner (42) noted that strains heterozygous for a MSH1 deletion accumulated mtDNA point mutations (mit) faster than wild-type strains did. The authors propose a role for Msh1p in the suppression of homeologous recombination, as supported by the in vitro binding specificity of the protein for DNA with multiple mismatches and loops of unpaired nucleotides. They also suggest that Msh1 could interact with other mitochondrial proteins, especially Pif1 and Rim1 (see above) (Table 1), which are, respectively, a DNA helicase and an SSB protein: they have similarities to components of the Escherichia coli MutHLS system, and both are required for mtDNA maintenance. This assumption could be tested by using the two-hybrid system in yeast or, even more clearly, by performing coimmunoprecipitation experiments, since the three proteins, Pif1, Rim1, and Msh1, have been purified.

Thus, the recombination/repair approach (which has been difficult to perform) appears very promising with respect to mtDNA maintenance, even though most of the genes required for these processes in mitochondria remain to be identified. Furthermore, as described in "Nuclear control of biased transmission" (above), this specific screen also identified the CCE1 gene (allelic to MGT1 [Table 2]), encoding a mitochondrial cruciform cutting endonuclease (158). Although discovered by other methods, ABF2 (see "The ABF2 paradox" above) must be added to the list of the genes involved in recombination. In fact, among the genes required for mtDNA maintenance, four are clearly involved in recombination processes: MGT1/CCE1, PIF1, MHR1, and ABF2. However, in the first three cases, recombination between markers in rho⁺ × rho⁺ crosses is unaffected or only slightly affected in the relevant mutant backgrounds (96, 183, 339). Thus, ABF2 is the only gene among these four to be implicated in intergenic recombination between rho⁺ genomes (335). There is another striking point concerning ABF2: it stands apart from MGT1/CCE1 with respect to recombination intermediates. In fact, their level is reduced in the absence of ABF2 and increased in the absence of MGT1/CCE1; moreover, these effects are seen only on rho⁺ genomes in the first case and preferentially on rho genomes in the second case (194, 198). Clearly, the field of recombination has been underinvestigated.

It has long been known that mitochondrial protein synthesis is required for maintenance of rho⁺ genomes (reference 213 and references therein). This conclusion has been confirmed over time: all nuclear or mitochondrial mutations which completely block mitochondrial translation inescapably lead to rho genome production (see below). In accordance with this constraint, mutations which impair mtDNA transcription should lead to rho genome production. The transcriptional apparatus has also been thought to be directly involved in mtDNA replication. This assumption is based on indirect (mostly in vitro) data, suggesting that (at least in vertebrates) mtRNA polymerases are primases for mtDNA replication (reviewed in reference 49). Thus, transcription might play a dual role in mtDNA maintenance: directly through initiation of mtDNA synthesis and indirectly through translation. However, evidence for a direct role is still questionable.

The RPO41 gene encoding the catalytic subunit of mtRNA polymerase in yeast was cloned by reverse genetics (155). Disruption experiments led to the conclusion that this enzyme was required for mtDNA maintenance (112) (Table 3). Since the authors provided evidence for a complete loss of mtDNA in a rpo41 mutant background, it seemed of utmost importance to determine if Rpo41p was also required for maintenance of rho genomes. This experiment was performed by Fangman et al. (90). They used two types of rho strains. The first was hypersuppressive, containing an (amplified) mtDNA fragment with an ori/rep sequence, which is assumed to offer a replicative advantage in matings with rho⁺ strains. The second was neutral, lacking such sequences and consisting solely of AT base pairs. Such rho genomes replicate by an unknown process and are most often lost after mating with a rho⁺ strain. Both rho genomes were maintained in haploid strains carrying a disrupted rpo41 gene. Thus, Rpo41p did not seem to be required for rho mtDNA replication. This observation prompted the authors to reexamine the effects of RPO41 disruption on rho⁺ genomes. In contrast to the results of Greenleaf et al. (112), they observed that the rpo41 mutation did not cause a complete loss of mtDNA but led mainly to the accumulation of rho genomes (90) (Table 3). As emphasized by the latter authors, this discrepancy is probably due to the experimental procedures used in the two laboratories. In one case (112), the observations were performed on cultures obtained from haploid spores which inherited the rpo41 mutation through meiosis, while in the other case (90), the observations were made on vegetative cells after inactivation of the RPO41 gene. When these authors repeated the experiment by sporulating a diploid strain heterozygous for the rpo41 mutation, they observed a lower recovery of mtDNA than in their previous experiment with vegetative cells. These data as a whole lead to three clear-cut conclusions. First, inactivation of RPO41 does not cause a complete loss of mtDNA. Second, even if the mutant strain produces a large amount of rho⁰ petites, cytoduction experiments provide evidence that it is able to maintain rho genomes. This interesting feature has also been noticed for the abf2 mutants (see "The ABF2 paradox" above). Third, conclusions about the effect of a nuclear mutation upon the production of rho⁰ versus rho petites must be drawn cautiously, since they may depend on the experimental procedures used in different laboratories. In particular, this could explain why rpo41 mutations were recovered by Guan (118) among rho⁰ producers (see "Wide-scale screening for rho⁰ production" above).

Another very interesting way to test the putative role of transcription for replication was to analyze the effects of a rpo41 deletion on biased transmission of hypersuppressive rho, assumed to result from their replicative advantage. This experiment was also performed in Fangman's laboratory (196). Since rho⁺ genomes are unstable in a rpo41 background while rho genomes are stable, the authors examined diploid clones issued from matings between hypersuppressive and neutral petites. The hypersuppressive rho genomes were always preferentially transmitted regardless of the nuclear background, RPO41 or rpo41 (deletion). The authors proposed two hypotheses based on all available data. First, hypersuppressive petites indeed have a replication advantage, but transcription is not required for replication initiation (at least for rho genomes). Second, the preferential transmission of rho genomes containing a ori/rep sequence is not at the level of replication but at the level of segregation. This hypothesis is in agreement with the deletion and overexpression of MGT1/CCE1 (see "Nuclear control of biased transmission" above and Table 2), which suggest that the resolution of recombination junctions could play a key role in the segregation and transmission of mtDNA. A similar hypothesis has also been proposed for ABF2 (see "The ABF2 paradox" above).

The catalytic (Rpo41) subunit of mtRNA polymerase requires an additional factor for promoter recognition. This factor (Mtf1 or sc-mtTfb) was initially characterized biochemically (264), while the gene was first identified as a high-copy suppressor of an rpo41 heat-sensitive mutant (188) (Table 3). The relationship between the protein and its structural gene was then established by Jang and Jaehning (144) using reverse genetics. The geneticists (188) claimed that disruption of MTF1 led to a complete loss of mtDNA. However, since their experimental evidence was limited to a lack of hybridization with a single probe (21S rDNA), we must be more cautious (Table 3). Directed mutageneses of the MTF1/mtTFB gene were performed by Shadel and Clayton (271). The mutants showing respiration deficiency were not analyzed at the mtDNA level. This would have been especially interesting with respect to a small deletion (of 5 amino acids) resulting in a mutant phenotype, while the protein showed significant levels of transcription in vitro. This was the only mutant of the series exhibiting a contradictory (in vitro/in vivo) phenotype. This led the authors to propose that the Mtf1/mtTfb protein could have a mitochondrial function other than transcription.

Thus, to date, the data neither prove nor disprove a role of transcription in replication initiation of the rho⁺ genome. However, they lead to two conclusions concerning petite mutants: (i) rho genomes (either hypersuppressive or neutral) do not require transcription for their replication, and (ii) preferential transmission of hypersuppressive rho genomes is maintained in the absence of transcription.

THE TRANSLATION-ATP SYNTHASE CONNECTION

Top Previous Next References

Background

The idea that mitochondrial translation is required for mtDNA stability was first suggested by experiments demonstrating that growth of yeast in the presence of inhibitors of mitochondrial protein synthesis caused a high frequency of rho genomes (references in reference 213). This idea has been reinforced by these authors, who showed that the inactivation of four genes involved in mitochondrial translation led to loss of wild-type mtDNA (213). They also showed that induction of rho was observed only if the mutations were stringent and not if the relevant nuclear genes carried leaky mutations. Finally, they provided evidence that rho genomes could be maintained in these contexts, which thus appeared deleterious only for the wild-type mitochondrial genome. In these experiments (213) (Table 3), the genes studied encoded two aminoacyl-tRNA synthetases, one elongation factor, and a putative ribosomal protein later shown to be a bona fide r-protein (60). To date, there has been no exception to the rule: rho genomes are produced at high frequency as long as a stringent mutation occurs in a mitochondrial or nuclear gene encoding an essential component of the mitochondrial translational apparatus. It would be redundant to quote all reports on the subject published over the last 14 years.

One must keep in mind that the effect of a mutation may be indirect, if the product of the relevant gene acts upstream of translation. Among many examples, two illustrate this fact. The first case concerns the RPM2 gene, encoding the protein component of a ribonucleoprotein required for 5' maturation of mitochondrial tRNAs (154, 210). A less obvious example concerns the NAM1 gene, first identified as a multicopy suppressor of mitochondrial splicing deficiencies caused by mitochondrial mutations (73). The effects of NAM1 inactivation were carefully analyzed in two mitochondrial genetic backgrounds, i.e., in strains carrying either an intron-containing or an intronless mitochondrial genome. The data led to the conclusion that Nam1p was required for the processing and/or stability of cytb and cox1 intron-containing pre-mRNAs and of the atp6 transcript (which does not contain introns) (116). Interestingly, NAM1 inactivation leads to 20 to 50% petite production in a strain with an intron-containing mitochondrial genome (16) while there is no increase in petite production (with respect to wild type) when the mitochondrial genome does not contain introns (116). According to the authors, the high level of petites observed in the first context may be explained by a defect in mitochondrial translation due to a starvation for tRNA^Glu, which is cotranscribed with the cytb gene, the intron-containing cytb transcript being unstable in the absence of Nam1p. Although this attractive hypothesis has yet to be experimentally tested, it strikingly illustrates the chain of events which can indirectly link a mutation with mitochondrial translation and hence with petite production.

In 1985, Myers et al. (213) proposed two hypotheses concerning the role of mitochondrial translation in maintenance of the rho⁺ genome. First, mtDNA would encode a protein required for replication or repair. Second, perturbation of the inner membrane (due to the lack of a component encoded by mtDNA) might specifically prevent the import of proteins required for replication or repair of the rho⁺ genome. Fourteen years later, the question remains open. However, many data have meanwhile focused on the mitochondrial ATP synthase.

The yeast mitochondrial ATP synthase (ATPase) contains at least 14 subunits, of which 5 (, , , , and ) are found in the catalytic F₁ sector (responsible for both ATP synthesis and hydrolysis) while the rest participate in the proton membrane channel F₀ sector (see reference 232 for a general review on ATPases and Table 4 for yeast mtATPase). Three of the F₀ subunits (Atp6, Atp8, and Atp9) are encoded by the mitochondrial genome, and the remainder of the complex is specified by nuclear genes (Table 4).

Two of the three mitochondrial genes, those encoding Atp9 (54) and Atp6 (254), were initially identified through mutations conferring resistance to oligomycin, an inhibitor of mitochondrial ATPase (OLI1 and OLI2, respectively) (10). Later, mit mutations leading to the loss of ATPase activity were obtained, defining two genes, PHO2 and PHO1, which mapped near the OLI1 and OLI2 loci, respectively (54, 97). These loss-of-function mutations produced high levels of petites (Table 4). To simplify the nomenclature, the genes are now called ATP9 and ATP6. A final group of mit mutations helped to identify the third gene, AAP1 (ATP8), encoding Atp8 (199). Again, the mutants were high producers of petites (214) (Table 4). At the time, it seemed clear that mutations in the ATPase mitochondrial genes so far identified were inherently unstable with respect to the mitochondrial genome. This conclusion was strengthened by the observation that very few mit mutations were found in these three genes compared to the mitochondrial genes encoding cytochrome oxidase subunits and cytochrome b. For instance, it was reported that mutants with deficiencies in cytochrome b and in cytochrome oxidase were, respectively, 50 and 300 times more frequent than mutations in the ATP9 gene (54, 163, 307). As discussed by these authors, the rare occurrence of the ATP9 mutants could have two explanations. First, one must take into account the relative sizes of the targets: ATP9 is a small gene (this is also the case for ATP6 and even more so for ATP8, which encodes a 48-amino-acid protein). Second, all loss-of-function mutations in the ATP9 gene (and in ATP6 and ATP8 as well) are unstable. In fact, one can presume that all ATP9 (and ATP6 and ATP8) mutations able to produce nearly 100% petites could not have been recovered as mit mutations. This hypothesis implies that most mit mutations in the three relevant genes are leaky. Although frameshift and nonsense mutations have been obtained in these genes, this does not imply stringent mutations: phenotypic suppression occurs in mitochondria, due to its error-prone translational apparatus (see reference 284 for an example of a leaky frameshift mutation in the ATP6 gene).

This high instability of the rho⁺ genome, observed when the F₀ sector of ATPase is deficient, is not limited to the mitochondrial genes. In fact, mutations in nuclear genes encoding the other F₀ subunits also display this property (Table 4). Three F₀ subunits (e, g, and k) have been omitted in Table 4 since they are not required for the formation of enzymatically active ATP synthase (6).

In 1973, Ebner and Schatz (79) showed that a nuclear mutant (pet936) lacking F₁ produced a large number of petites. However, this observation was forgotten, and before 1992 it was believed that the F₁ ATPase sector was not involved in mtDNA stability (see, for instance, reference 2). In fact, at that time two pet mutations, later shown to lie in the structural genes for  and  subunits (260, 297), appeared genetically stable with respect to mtDNA (Table 4). In any case, the situation was strikingly modified when it appeared that deletions of the genes encoding the three other F₁ subunits (, , and ) led to a more or less pronounced production of petites (Table 4). Very recently, Lai-Zhang et al. (170) showed that strains carrying a deletion of either ATP1 (subunit ) or ATP2 (subunit ) produce less than 1% petites. The authors did not discuss the fact that deletions of either one of these two genes give a petite level 10-fold below what they observed in the wild-type strain from which the deletions were obtained (Table 4). Our hypothesis is that these mutants cannot survive in a petite context (see "No compromise," below, for a more extensive discussion of nuclear mutants killed by mtDNA instability).

In addition to these structural genes, several nuclear genes whose products are involved either in ATPase subunit synthesis or in assembly of the complex have been identified. Three concern the expression of the mitochondrial ATP9 gene: AEP1 (for "ATPase expression protein"), AEP2 (ATP13), and NCA1 (for "nuclear control of ATPase"). The first is necessary for ATP9 translation (231). Disruption of the AEP1 gene leads to a high frequency of petites (230). The second gene, isolated by two research groups, is required for ATP9 transcript maturation or stabilization. The original pet mutant, which helped identify the ATP13 gene (1), showed a low level (8%) of petite production (306). This can be explained by some leakiness of the mutant, since disruption of AEP2 (ATP13) clearly shows an instability of mtDNA (93). The third gene (NCA1) involved in ATP9 expression is required for transcript stability, but there are no data concerning the effects of either the original mutation or the NCA1 disruption upon petite production (338).

Three other genes are required for stabilization (NAM1) or normal maturation (NCA2 and NCA3) of the ATP8-ATP6 cotranscripts. As mentioned above, inactivation of the NAM1 gene leads to a high production of petites in a strain bearing an intron-containing mitochondrial genome (probably due to an indirect effect on translation) whereas there is only a basal level of petites when the mitochondrial genome contains no introns. In fact, in the latter condition, the only defect of the nam1 mutant is a diminution of the ATP6 transcripts. However, this is a leaky phenotype, as shown by the fact that the nam1 strain is able to grow (although only slowly) on glycerol-containing medium at 28°C: this probably explains the mtDNA stability. It would have been interesting to observe petite production by the mutant at 36°C, a temperature at which it displays a complete respiratory growth defect (116). The two other genes so far identified, which affect the synthesis of ATP6 and ATP8 subunits, are part of a complex procedure that is not yet understood. It was demonstrated that the simultaneous presence of mutations in two genes resulted in a cold-sensitive phenotype (on a nonfermentable carbon source), associated with a small amount of one of the ATP8-ATP6 cotranscripts (234). Isolation of the NCA2 (29) and NCA3 (233) genes shed no light on their functions. The petite levels were not reported, either for the original double mutant or for the double-disruption strain.

Three genes (ATP10, ATP11, and ATP12) specifically involved in ATPase assembly have so far been described. The Atp11 and Atp12 proteins are required for normal assembly of the F₁ sector. The atp11 and atp12 mutants display 10% of wild-type ATPase activity and seem to be defective in a late step of F₁ assembly, the incorporation of  and  subunits into an active oligomer (3). The two mutants display a basal (1 to 5%) level of petite production (2). The genes were cloned (2, 23), but the effects of null alleles on the fate of mtDNA have been reported only for ATP11 by Lai-Zhang et al. (170): a strain carrying a deletion of this gene produces 1% petites. With respect to ATP10, biochemical data suggest that the mutants have an abnormal F₀ structure, which would impair the coupling of F₁ to F₀. Biochemical criteria, in accordance with growth phenotypes, also suggest that the mutants (the original pet as well as the disruptant) are leaky. One of the pet mutants has been analyzed with respect to mtDNA stability. According to the authors, it is genetically stable since it accumulates only 10 to 15% petites after long-term culture (4).

The final class of nuclear genes reported here are those whose products are required to facilitate the assembly of ATPase and other complexes in the mitochondrial inner membrane as well, namely, cytochrome oxidase. They are the AFG3 (YTA10), RCA1 (YTA12), and OXA1 genes (reference 248 and references therein). Again, all the mutants (even the null mutants) display a leaky phenotype with respect to the ATPase complex. For instance, as reported by Rep et al. (249), afg3 rca1 double mutants contain 56% of fully assembled ATPase relative to wild-type. The single- and double-mutant strains are genetically stable with respect to mtDNA (5). The same situation is encountered with the oxa1 mutants (G. Dujardin, personal communication).

These data as a whole lead to the following conclusions and comments. First, null alleles of all genes encoding essential subunits of the F₀ sector cause a strong increase in petite production. Second, deletion (disruption) of the genes encoding subunits , , and  of the F₁ sector also increase mtDNA instability. Third, the status of the  and  subunits (which are the catalytic subunits of ATP synthase) remains ambiguous. Ebner and Schatz (79) showed that absence of ATP synthase activity (due to an uncharacterized pet mutation) increases petite production. However, the petite colonies could be detected only after 6 to 10 days (instead of 2 to 3 days in the usual procedures). This raises a fundamental question concerning the conclusions of Lai-Zhang et al. (170). In fact, these authors observed that deletion of ATP1 ( subunit) or ATP2 ( subunit) resulted in less than 1% petites (compared to 9% in their reference strain). However, they did not state whether they found any petites and whether they examined the plates after long-term culturing. The last conclusion concerns the regulatory proteins. Clearly, some of them facilitate only ATPase assembly, since, in their absence, significant amounts of ATPase are assembled. In these cases, the mutant strains do not accumulate petites at high rates. In contrast, deletions of AEP1 or AEP2, required for ATP9 expression, increase mtDNA instability.

We must now reformulate the problem of mitochondrial translation. If the lack of one (or several) protein encoded by mtDNA is responsible for the high rates of petites observed when translation is blocked, one can easily exclude cytochrome b and subunits I, II, and III of cytochrome oxidase: stringent mutations in the four genes do not increase the levels of petites. However, in direct contrast, each time a mutation abolishes the synthesis (or drastically alters the structure) of one of the three ATPase subunits encoded by mtDNA, an extreme level of petite production is observed (see above and Table 4). The question must be posed: what is the link between Atp6, Atp8, Atp9, and mtDNA stability?

To answer this question, one must keep in mind that mutations which abrogate the synthesis of any of the ATPase subunits, even those which are nuclearly encoded (with the possible exception of  and ), cause the same phenotype, i.e., high production of petites. A careful analysis of the data leads one to favor a special role for Atp6 (see references in Table 4 for most of the relevant data). First, Atp6 assembly depends on Atp9 and Atp8 (reviewed in reference 214). Second, the lack of either of the nuclearly encoded F₀ subunits is always associated with the absence of Atp6. Third, in a strain lacking the F₁ subunit , the concentrations of the F₀ subunits Atp6, Atp8, and Atp9 are reduced by more than 70% (although petite production is only 20% in the culture analyzed) (228). Fourth, when the F₁ subunit  is absent, Atp6, Atp8, and Atp9 are present but there is a proton leak, suggesting a conformational change of F₀ (119). The relevant observations cannot be performed with subunit , since deletion of the gene leads to 100% petites. Finally, disruption of ATP2 (subunit ) blocks the functional assembly of F₀ (298). Thus, the data as a whole show (i) that the F₀ sector is altered when one of the F₀ or F₁ subunits is absent and (ii) that Atp6 is the ultimate and most sensitive target of F₀ assembly.

These data do not argue against the proposals of Myers et al. (213) but suggest the following considerations: in the absence of mitochondrial translation, either the import of proteins required for replication/repair of the rho⁺ genome would be specifically prevented by alteration of the F₀ sector or the putative mtDNA-encoded protein required for replication/repair of the rho⁺ genome would be Atp6. In our opinion, these hypotheses now appear quite unlikely and we favor a structural and more direct role of Atp6 in the maintenance of the rho⁺ genome: for instance, Atp6 might be involved in physical attachment of the rho⁺ molecule to a centromere-like structure required for its faithful transmission (see "Is there a centromere-like structure for the mitochondrial genome?" below).

In any case, a critical test with respect to the translation-ATPase connection would be to supply the mitochondria with Atp6, Atp8, and Atp9 encoded by artificial nuclear genes specifying imported forms of the proteins (already performed with Atp8 [214]) and then examine the effect of a mitochondrial translational block on petite production. If in this context the level of petites remains stable, one may conclude that at least one of these subunits is indeed the target of mitochondrial translation with respect to rho⁺ maintenance. However, if a high production of petites is observed, one must assume that either another (uncharacterized) protein encoded by the mitochondrial genome or translation per se of the relevant subunit is involved in the rho⁺ maintenance.

The problem of ATPase does not end here: we will find it again as a possible key in the old problem of petite positivity versus petite negativity (see "No compromise" and "Petite positivity versus petite negativity" below).

POSTTRANSLATIONAL EVENTS

Top Previous Next References

Expected and Surprising Roles for Proteases (Peptidases) in mtDNA Stability

As reviewed in reference 247, mitochondrial proteases can be divided into two classes: processing of precursor proteins and involvement in protein degradation. Most mitochondrial proteins encoded by nuclear genes are translated as precursors whose N-terminal presequences are required for mitochondrial targeting. These presequences are eliminated by proteolytic cleavage during their transport across the mitochondrial membranes. This is achieved by the mitochondrial processing peptidase (Mpp). One may assume that Mpp is indirectly required for mtDNA stability through processing of numerous proteins directly (or indirectly) involved in this process. However, this prediction cannot be tested, because the genes encoding the two subunits of Mpp are essential for cell viability (147, 329). This is not surprising if Mpp is a global mitochondrial processing enzyme: mitochondria are indeed required for many functions besides respiration (reviewed in reference 305). A few proteins, encoded by nuclear genes and delivered to the intermembrane space or encoded by the mitochondrial genome and devoted to the inner membrane, are processed by Imp, a inner membrane peptidase (reference 221 and references therein). This protein is beyond the scope of this review, since mutations of the two genes encoding its catalytic subunits (though leading to respiration deficiency) do not cause mtDNA instability.

A subset of nuclearly encoded proteins are processed to the mature form in two steps. The precursors are first cleaved by Mpp and then cleaved by the mitochondrial intermediate peptidase "Mip" (we put this acronym in quotation marks to avoid confusion with the acronym used above for the mitochondrial DNA polymerase [Table 1]). The yeast "MIP1" gene (Table 5) was cloned by hybridization with the cDNA encoding this enzyme in the rat. Disruption of the gene leads to respiration deficiency (142). Among the precursors which require "Mip" in order to be processed in the mature form, one r-protein, one translational elongation factor, and the Rim1 precursor were found (24). Because mitochondrial translation (see the previous section) (Table 3) and Rim1p (see "The recombination/repair track" above) (Table 1) are involved in mtDNA stability, the petite production observed when "MIP1" is inactivated is an indirect but expected effect (24).

The problem of the Pim1/Lon protease is much more puzzling. It belongs to the second group of mitochondrial proteases involved in protein degradation (247). The gene encoding this ATP-dependent protease was cloned independently by two groups. In one case (313), it was discovered during the systematic sequencing project of chromosome II and identified due to similarities of the deduced protein to bacterial Lon proteases. In the other case (295), the gene was obtained through PCR experiments, using the sequence conservation between these bacterial proteases and their human homologues. Evidence for proteolytic activity of the protein in vivo was obtained. The data also showed its role both in selective protein turnover (295) and in degradation of misfolded proteins (316). The PIM1/LON gene was rediscovered as a multicopy suppressor of mutations in two nuclear genes encoding proteins of the mitochondrial inner membrane, AGF3/YTA10 and RCA1/YTA12 (248). Mutations in these two genes lead to defects in degradation of mitochondrially synthesized proteins and in assembly of inner-membrane complexes. Several arguments led to the conclusion that these proteins may play a dual role, with both degradation and chaperone activities (reviewed in reference 247). Interestingly, overproduction of the Pim1/Lon protein suppressed respiration-dependent growth defects and protein assembly defects of the agf3/yta10 and rca1/yta12 mutants but not defects in protein degradation. Furthermore, suppression was maintained and even enhanced when the proteolytic activity of the Pim1/Lon protein was destroyed (249). This suggests that this protein (at least when overproduced) also displays a chaperone-like function in the assembly of mitochondrial complexes.

Disruption of the PIM1/LON gene is not lethal but leads to respiration deficiency and rho genome production (295, 313) (Table 5). The stability of mtDNA depends on the proteolytic activity of the Pim1/Lon protein. Two types of data lead to this conclusion. First, a missense mutation which impairs this activity results in mtDNA aberrations. Second, a strain carrying a deletion of the PIM1 gene is complemented (at 30 but not 37°C) by the E. coli LON gene with respect to mtDNA stability, provided that the bacterial protein was proteolytically active (299). These observations prompted the authors to propose that the Pim1/Lon protein could be involved either in the proteolytic activation of a protein required for mtDNA integrity or in the turnover of a protein which would negatively affect mtDNA stability. Notably, van Dyck et al. (312) succeeded in dissociating the role of Pim1/Lon on mtDNA stability from another function of its protease activity in biogenesis of the respiratory chain. In this fascinating problem, elucidation of the specific target(s) of the mitochondrial Pim1/Lon protease with respect to maintenance of the rho⁺ genome is now necessary. New insights into the functions of its bacterial homologues will undoubtably help in this research. In fact, it has been recently found that the Caulobacter Lon protease plays a crucial role in cell cycle control of DNA methylation (330) and that the E. coli Lon protease is a site-specific DNA-binding protein (101). This property has also been reported for the mitochondrial human Lon protease (100).

In addition to proteins bifunctional as proteases and chaperones (see above) (Table 5), mitochondria contain more conventional chaperones. Most of the nuclear genes encoding these chaperones are necessary for cell survival. This might be expected if proteins which perform essential functions in mitochondria (see reference 305 for a review) require these chaperones for translocation across the membranes and/or proper folding within the organelle. Since these genes are essential, it has often been impossible to address the specific question of their role with respect to mtDNA stability. However, a few points can be made.

The yeast mtHsp70 is encoded by the nuclear gene SSC1 (for "stress seventy subfamily C"). This gene was isolated by hybridization with the SSA2 gene, which encodes another member of the Hsp70 family in yeast. SSC1 is essential (55) and encodes a mitochondrial protein (56). Further studies have shown that mtHsp70 is required for the translocation of proteins into mitochondria, folding of these proteins in the matrix, and folding of proteins synthesized in the organelle (reviewed in reference 215). Although two heat-sensitive mutants have been used in these studies, no effects were reported on mtDNA stability at subrestrictive temperatures. However, Moczko et al. (209) found that double-mutant strains bearing either of the two heat-sensitive ssc1 mutations and a deletion of the gene encoding mtHsp78 were unable to grow on nonfermentable carbon sources at 30°C, conditions which allow growth of the two single mutants. The double mutant is in fact in a rho (or rho⁰) state (Table 5). Thus, the absence of mtHsp78 leads to mtDNA instability when the activity of mtHsp70 is limiting. The exact role of mtHsp78 (179) remains unclear, although some data suggest that it can partially substitute for mtHsp70 (265).

In E. coli, Hsp70 (DnaK) is assisted by two proteins, DnaJ and GrpE. DnaJ is a chaperone in its own right; however, it also interacts with DnaK and stimulates its ATPase activity. GrpE acts as a nucleotide exchange factor for DnaK (see reference 104 for a review). Although the mitochondrial Hsp70 does not seem to be the functional homologue of DnaK (62), the functional homologues of DnaJ and GrpE were found in yeast mitochondria. The nuclear gene encoding the mitochondrial GrpE was independently discovered by three groups during the same year. Named YGE1 by Ikeda et al. (140), it was a serendipitous discovery, since these authors identified the gene as responsible for resistance to staurosporine (an inhibitor of protein kinases) when present on a multicopy plasmid! The article also reports that the gene is essential and able to complement a grpE mutant of E. coli and that a strain with YGE1 disrupted has mitochondria stacked in the mother cell and thus unable to enter the buds. Bolliger et al. (21) purified the protein by using its interaction with mtHsp70 and then cloned the gene, which was shown to be essential for cell survival. Laloraya et al. (172) discovered the gene (named MGE1 in this article) by DNA sequence analysis. They showed that MGE1 is essential and encodes a mitochondrial protein that is probably required for protein translocation into mitochondria. This assumption was fully confirmed by later studies using heat-sensitive mutants of YGE1/MGE1, and it was shown that Mge1p acts in concert with mtHsp70 (see reference 215 for a review). Laloraya et al. (171) reported an increased frequency of petites in a heat-sensitive mge1 mutant (Table 5).

In bacteria, DnaJ is (like GrpE) an essential partner of Hsp70 (see above). Its yeast mitochondrial homologue was identified during a DNA sequencing project (258). Deletion of MDJ1 (for "mitochondrial DnaJ") causes death at 37°C and an inability to grow on nonfermentable carbon sources at 24°C. This second phenotype cannot be rescued by the wild-type gene: evidence was obtained that these petites displayed a rho⁰ state (Table 5). Mdj1p is not required for protein import but is required in later steps of protein folding. However, Mdj1p cooperates with mtHsp70 to prevent aggregation of misfolded proteins, thus allowing their degradation by Pim1/Lon (see above). Finally, again in association with mtHsp70, Mdj1 can act as a chaperone for nascent proteins synthesized in mitochondria (258, 316, 323). Thus, the role of Mdj1 with respect to mtDNA stability is probably indirect, through its effect upon proteins that are directly or indirectly involved in mtDNA metabolism.

Another canonical chaperonin system, consisting of Hsp60 and Hsp10, is also required in the mitochondrial matrix. The genes encoding these two proteins are essential for viability (reviewed in reference 215). There is, to our knowledge, no report concerning their effect on mtDNA stability under subrestrictive conditions.

This section terminates the first part of this review, which tentatively concerned genes clearly involved in mtDNA metabolism and in gene expression including posttranslational events. The genes discussed in the following sections have been grouped (more or less artificially) according to a few governing ideas.

Mitochondrial carriers ensure the traffic of a number of small molecules (adenine nucleotides, inorganic ions, amino acids, fatty acids) across the inner membrane (see reference 168 for a review). It is thus not surprising that some of them are indirectly involved in mtDNA metabolism and consequently mtDNA stability.

The MIR1 (for "mitochondrial import receptor") gene was initially identified as encoding an integral membrane protein (p32) which was proposed to act as a receptor for protein import into mitochondria (211, 225). Indirect (in vitro) data suggested that p32 contained a binding site specific for the signal sequence of preproteins directed toward mitochondria (212). However, it appeared that the MIR1 gene encodes a protein displaying 40% identity to mammalian mitochondrial membrane proteins which could act as phosphate carriers (211). Furthermore, a protein of the mitochondrial inner membrane which could function as a phosphate translocator was purified and sequenced (120): it is 100% identical to p32. Sequencing of the relevant gene led to the same conclusion (237). Data from Zara et al. (333) gave new insights with regard to the controversy about the function(s) of p32. In fact, these authors provided evidence that lack of p32 causes both a block in the import of phosphate and a pronounced reduction in the import of proteins devoted to the mitochondrial inner membrane and matrix. The second defect would be indirect, due to a strong reduction of the membrane potential in the mutant mitochondria. It was shown that (in vitro) restoration of a membrane potential restored protein import without relieving the phosphate transport deficiency. The viability of the mutant implies that, in vivo, protein import is sufficient for mitochondria biogenesis. However, this rate would be limiting with respect to mtDNA stability because the mutant accumulates petites at a higher frequency than the wild type does (333) (Table 6). Another explanation would be that a lower potential directly affects mtDNA stability. Nonetheless, Murakami et al. (211) demonstrated that the inability of a mir1 mutant (partial deletion) to grow on glycerol was not due to the loss of the rho⁺ genome. This contradictory result can be explained either by the genetic background or by the fact that the MIR1 gene was only partially deleted. In addition, the high production of petites observed in a mir1 disruptant whose mtDNA contains introns is abolished when the mitochondrial genome lacks introns (P. Hamel and G. Dujardin, personal communication). This is one of several examples, as reported in this review, of the strain influence in yeast due to variations in the nuclear and/or mitochondrial genomes.

The RIM2/MRS12 gene was identified as a multicopy suppressor of respiration deficiency both of a pif1 null strain (at 36°C) and of a mrs2 disruptant. As mentioned above, PIF1 encodes a DNA helicase essential for mtDNA stability at high temperature (Table 1). Mrs2p (for "mitochondrial RNA splicing") is bifunctional: it is required for group II intron splicing and for assembly of the respiratory chain (325). Multicopy suppressors of the respiratory defect of either pif1 or mrs2 mutants were independently sought. This allowed, in the first case, identification of the RIM1 gene required for mtDNA replication (see "The recombination/repair track" above) (Table 1). In the second, a multitude of suppressors was obtained (317). It appeared that one (MRS12) was also discovered as a suppressor of pif1 and was named RIM2 (311). Deletion of RIM2/MRS12 causes a slow-growth phenotype on glucose medium and an inability to grow on glycerol. This respiration deficiency is in fact due to a massive loss of mtDNA (Table 6). This rho⁰ phenotype is seen as early as spore germination (issued from a diploid heterozygous for the deletion). The RIM2/MRS12 gene encodes a member of the mitochondrial carrier family (311) whose substrates are as yet unknown. However, the authors suggest that it could be involved either in the transport of mtDNA precursors (to explain the suppression of pif1) or, more likely, in the transport of Mg²⁺, which could explain its pleiotropic suppression effect on both pif1 and mrs2 mutations. It is noteworthy that, in the latter case, suppression acts on both defects: splicing and respiratory chain assembly (311).

ATP-binding cassette (ABC) transporters also assure transport of substrates across membranes. However, their substrates are extremely diverse, ranging from small molecules to large proteins. Each transporter is substrate-specific (reviewed in reference 132; see also reference 296 for a review of yeast ABC proteins). ATM1 (ABC transporter of mitochondria) encodes a protein located in the mitochondrial inner membrane (178). Disruption of ATM1 leads to a dramatic growth defect in rich glucose medium and an inability to grow at all on minimal medium. Although it grows (very slowly) on rich medium containing nonfermentable carbon sources, the disruptant loses this ability because it becomes rho (Table 6). The mutant lacks all major cytochromes, including cytochromes c and c₁, which are encoded by nuclear genes. However, the authors provided evidence that this defect is not due to a failure to export heme from the mitochondria. Thus, the true function of ATM1 remains to be characterized, as do the reasons for which this gene is essential not only for mtDNA stability but also for optimal growth. These two properties appear related, since some atm1 disruptants show a moderate growth defect and do not become rho. This variation in the phenotype strength of the mutants is probably due to the spontaneous appearance of nuclear suppressors. Analysis of these suppressors will certainly aid in the elucidation of the (dual?) function of ATM1 (178). Interestingly, ATM1 was rediscovered as a gene able to complement a mutant defective in mitochondrial cytochromes (156). The data reported suggest that Atm1p could play a key role in mitochondrial iron homeostatis; this aspect is discussed in the following section.

With respect to the purpose of this review, it is interesting that mitochondrial carriers and transporters comprise three classes. In the first class, absence of the protein has no effect on mtDNA stability, even though in some cases the relevant genes are multicopy suppressors of abf2 (see "The ABF2 paradox" above) (43, 152). In the second class, one finds the AAC2 gene, encoding the major mitochondrial ADP/ATP translocator, required for viability of cytoplasmic petites (see "No compromise" below). The third class regroups the RIM2/MRS12 and ATM1 genes (Table 6), in which mutations lead to a high production of petites. Depending on the genetic background, MIR1, which encodes a phosphate carrier, can enter in either the first or the third class.

Oxygen, although essential for life, is deleterious when present in one of its reactive species (ROS). While some cytosolic oxidative processes produce ROS, it seems that mitochondrial respiration is the main source of production of superoxide (O₂·) and other ROS (H₂O₂ and the hydroxyl radical ·OH). Most superoxides are probably produced by complex III (reference 195 and references therein). This radical is normally converted into H₂O₂ by superoxide dismutase, while H₂O₂ is destroyed by catalases and peroxidases. The hydroxyl radical, assumed to be the most deleterious, is formed by iron-catalyzed reactions (Fenton reactions) of O₂· and H₂O₂. Generally, these reactions are minimized by compartmentalization of free iron and superoxide (reviewed in reference 128).

Superoxide (O₂·) is removed by superoxide dismutase, which catalyzes the reaction 2O₂· + 2H⁺H₂O₂ + O₂. Like most eucaryotic cells, S. cerevisiae contains a cytosolic and a mitochondrial SOD, respectively the products of the SOD1 and SOD2 genes (reviewed in reference 109).

Disruption of the gene encoding the mitochondrial Sod (Sod2p) was first performed by van Loon et al. (314), who showed that the mutant was oxygen sensitive, especially when grown in 100% oxygen. However, it was sensitive even to normal air when grown in liquid medium with nonfermentable carbon sources. These authors also noted that there was no increase in the production of petites in this mutant context. This search was performed in air (G. Schatz, personal communication). Seven years later, Guidot et al. (121) showed that growth in hyperoxia of a strain with SOD2 disrupted is fully restored in a petite (rho⁰) background. This is explained by the fact that petites produce less O₂· and thus overcome the absence of Sod2p. In any case, the status of sod2 mutants with respect to petite production still requires careful examination.

In contrast to the possible innocuity of Sod2p with respect to mtDNA stability, very low levels of glutathione (GSH) lead to a loss of mtDNA (157) (Table 6). GSH performs many functions, among them protection against ROS, especially H₂O₂ (see reference 207 for a review). The GSH1 gene, encoding the first enzyme of the two-step biosynthesis of GSH, was cloned by complementation of a gsh1 mutant (224) and rediscovered as a multicopy suppressor of a pet mutation impairing mitochondrial translation (185). Strains carrying a deletion of GSH1 are unable to grow in the absence of exogenous GSH (110, 331). It was also demonstrated that these strains are hypersensitive to H₂O₂, and that reducing agents (such as dithiothreitol or -mercaptoethanol) can restore the growth of the mutants, at least under nonstress conditions (110). Thus, as emphasized by the authors, GSH is essential only as a reductant. However, the effect of low levels of glutathione on mtDNA remains to be clarified. An effect of oxidative stress cannot be excluded. However, one might imagine an indirect effect which could involve ribonucleotide reductase. This enzyme, which catalyzes the conversion of ribonucleotides into deoxyribonucleotides, requires GSH for its recycling. As discussed below (see "mtDNA and the cell cycle"), mtDNA replication is highly sensitive to limiting amounts of DNA precursors. In contrast to gsh1 mutants, strains lacking the second enzyme of GSH biosynthesis appear unaffected in mitochondrial function (111). As shown by the authors, gsh2 mutants accumulate -glutamylcysteine, the intermediate in GSH synthesis, which would partly substitute for GSH.

Friedreich's ataxia is an autosomal recessive degenerative disease which primarily affects the nervous system and the heart. The FDRA gene was identified and shown to encode a protein (frataxin) whose function could not be inferred from its amino acid sequence (31). The yeast homologue was independently found through protein sequence similarity (31) and as a multicopy suppressor of a mutant unable to grow on iron-limited media (11). Disruption of YFH1 (for "yeast frataxin homologue") leads to the inability to grow on respiratory substrates, due to instability of mtDNA (11). The petites are in fact rho⁰ (328) (Table 6). However, this production of petites seems highly strain dependent (95). Similarly, growth defects of the yfh1 mutants on fermentable carbon sources can be either severe or not evident, depending on the strain. Disruption of YFH1 results in selective accumulation of iron in mitochondria, at the expense of the cytosol: the amount of iron is approximately 10-fold that in wild-type mitochondria in mutants (11, 95). Yfh1p localizes to the mitochondria (11, 328). This was also demonstrated for the human protein (164). Notably, Rötig et al. (256) did not detect any quantitatively significant rearrangements of the mitochondrial genome in Friedreich's ataxia patients. However, they observed a deficiency in the activity of the iron-sulfur (Fe-S) proteins of mitochondrial respiratory complexes I, II, and III. This could be explained as being due to oxidative stress (Fenton reactions) caused by the increased mitochondrial iron uptake, if one takes into account the fact that Fe-S proteins are critical targets for oxygen free radicals (reference 256 and references therein). Yeast strains carrying a YFH1 disruption appear hypersensitive to oxidative stress (11). These authors, as well as Koutnikova et al. (164), suggested that yeast mtDNA instability might be a secondary effect of frataxin absence: inhibition of oxidative phosphorylation would reduce the production of ROS and thus buffer the mitochondrial iron accumulation. This implies that the yfh1 mutant context would cause a selection of the petite state and not a real increase in the petite frequency. While very attractive, this hypothesis neglects the point that the petites are rho⁰ (328). One would in fact expect that partial deletions of mtDNA (rho petites) could equally counter the lack of frataxin.

Knight et al. (159) sought mutants with abnormal iron metabolism and analyzed one which showed a respiratory defect. It appeared that this strain carried a nonsense mutation in the SSC2 gene encoding a minor mitochondrial Hsp70, previously described and named SSH1 (263). Knight et al. (159) showed that their ssc2 mutant exhibited an increased total mitochondrial iron content, twofold higher than that observed in yfh1 mutants. These authors also demonstrated that the ssc2 mutant displayed a partial defect in Yfh1p maturation, and they consequently favored the idea that Ssc2p might regulate mitochondrial iron accumulation through its effect upon Yfh1 and other proteins. The nonsense ssc2 mutant exhibits a high production of petites (Table 6) and slow growth at 30°C, which is exacerbated at both 23 and 37°C (159). These data contradict those of Schilke et al. (263), who showed that a strain with SSH1 (SSC2) deleted grew quite normally at 37°C but very slowly at 30°C and below and that this mutant background does not induce petite production at the permissive temperature (see "No compromise" below and Table 13).

The mitochondrial ABC transporter Atm1 previously identified by Leighton and Schatz (178) (see "The world of carriers and transporters" above) (Table 6) was further analyzed by Kispal et al. (156). They demonstrated that a disrupted atm1 mutant exhibits an oxidative stress phenotype and that the mutant mitochondria accumulate free iron at levels 30-fold higher than do wild-type organelles and thus threefold more than that observed in yfh1 mitochondria. This observation may explain the oxidative damage of heme-containing proteins in the atm1 mutant (156). These authors gave no information about petite production in their atm1 background.

It is well known that yeast reference strains used in different laboratories are not isogenic. Thus, it is not surprising to observe that mutant phenotypes can differ more or less depending on the genetic background. Furthermore, transformation procedures (used to perform disruption experiments) are highly mutagenic and can induce suppressor mutations able to alleviate the mutant phenotype. Such effects are particularly striking in the case of atm1 and yfh1 mutants. The atm1 mutant shows a strong or slight growth defect depending on the disruptant, and this effect is correlated with a high or low level of petite production (178). Tetrad analysis aided these authors in the identification of a suppressor of the atm1 disruptant. A similar correlation between growth defect and petite production was observed, depending on the strain in which the YFH1 disruption was achieved. However, in this case, the two disruptants display a similar mitochondrial iron content, about 10-fold more than in the two YFH1 reference strains (95). This latter observation is really provocative and leads one to question the exact linkage between mitochondrial iron content, oxidative stress, and mtDNA instability. This observation is strengthened by the fact that disruption of YFH1, in the most permissive background, causes hypersensitivity to oxidant reagents (also reported by Babcock et al. [11]) and shows 100% petite production in the presence of 1 mM FeSO₄ (95).

In this complex problem, it is important to note that two homologous genes encoding mitochondrial iron (Fe) transporters (MTF1 and MTF2) have been characterized (182). A double-deletion strain grows well on both respiratory and nonrespiratory substrates. This observation was valid when either of the two genes is overexpressed. In that case, total mitochondrial iron levels are increased about fivefold, which is less than in yfh1 mitochondria. The fact that strains overexpressing MTF1 or MTF2 do not exhibit respiratory defects led the authors to suggest two hypotheses: (i) iron is not toxic for mitochondria below a precise threshold, and (ii) iron accumulates in different mitochondrial subcompartments in the two mutant backgrounds involving lack of YFH1 and overexpression of MTF1/MTF2 (182).

In any case, a high mitochondrial iron content is associated with mtDNA instability. It remains to be shown if there is a direct relationship between these two parameters, how this effect is manifested, and why the atm1 mutant accumulates rho genomes while the yfh1 mutants accumulate rho⁰ petites (Table 6). The data reported above clearly show that further experiments are required to understand the ways that mitochondria use to combat the ROS they produce and to understand the relationship between iron homeostasis, oxidative stress, and mtDNA instability.

Most genes involved in the synthesis of either long-chain fatty acids, which form the water-insoluble part of these lipids, or their polar head segments have been cloned and characterized (see references 61 and 129 for recent reviews). In four cases, mutations have been reported to increase mtDNA instability. These are single mutations in the OLE1, CHO1, and PSD1 genes and the mutations in the CHO2 CPT1 EPT1 triple-mutant strain (Table 7) (see below).

Although the data are incomplete, they focus on the fact that mtDNA instability can be increased by changes in the lipid composition of the mitochondrial membranes. This idea is strengthened by the recent discovery of a mitochondrial fatty acid synthase (FAS) whose defects can lead to a high production of petites. However, with the exception of lipoic acid, the compounds synthesized by this complex remain to be identified.

The KD115 (ole1) strain was obtained as a mutant auxotrophic for an unsaturated fatty acid, oleic or linoleic acid (250). An increased frequency of petites was observed when the mutant was grown in limited amounts of Tween 80 (a source of oleic acid), with the level of petites (from 0.4 to 80%) being determined by the degree of fatty acid depletion (201). In subsequent studies, Graff et al. (108) showed that mtDNA stability depends not only on an adequate supply of cognate fatty acids but also on the quality (i.e., the structural features) of the relevant fatty acids, even though these "inadequate" compounds are incorporated into cellular phospholipids. The OLE1 gene was cloned by complementation of the KD115 (ole1) mutant (289) and shown to encode the 9 fatty acid desaturase (290). Amazingly, the OLE1 gene was rediscovered as the gene complementing the heat-sensitive mdm2 mutant, which exhibits an altered mitochondrial morphology and a general defect in mitochondrial movement at high temperatures (287). The authors show that addition of unsaturated fatty acid (oleic acid) to the medium restores mdm2 growth at 37°C and corrects the mutant's defects in mitochondrial movement and morphology.

As proposed by Graff et al. (108), the data suggest that mitochondrial lipids containing unsaturated fatty acids may exhibit specific quantitative and qualitative requirements with respect to these compounds in order to ensure the correct insertion of proteins into the membrane. As an extension of this idea, one could imagine that mtDNA stability depends on its own attachment to one of these proteins (see the following sections and "Is there a centromere-like structure for the mitochondrial genome," below, for further discussion).

PS synthase is encoded by the CHO1 gene, cloned by complementation of a mutant auxotrophic for choline or ethanolamine (180). Two groups provided contradictory data on mtDNA stability in the mutant strain. In the first case (165), the proportion of petites in the mutant cultures was in the range of 5%. In the second (9), a high proportion of petites was found, markedly affected by temperature: at 25°C, almost 100% of the cells became respiration deficient after a few days of growth, while at 30°C, the petite level did not exceed 50%. These properties cosegregated with the cho1 mutation (9). It would be interesting to examine the effects of temperature in a null cho1 mutant (this strain is now available [12]). Atkinson et al. (9) suggest that the observed defect could be due to mitochondrial membrane alterations, caused either by the lack of PS or by more subtle changes, since the mutants can synthesize PC only through the salvage pathway: in particular, the balance between PC and PE appears significantly disturbed in the mutants.

Transformation of PS into PE requires PS decarboxylases, encoded by two genes (PSD1 and PSD2). The major form of the enzyme is encoded by PSD1 and is located at the inner mitochondrial membrane, which provides PE directly to the mitochondria. This gene was cloned and disrupted by two laboratories (47, 303). The latter authors provided evidence that the disrupted strains (although able to grow on nonfermentable carbon sources) show an increased production of petites on glucose medium. The Psd2 enzyme is not localized to the mitochondria (304). The observation that the psd1 mutants have significantly less PE in the mitochondria than the wild-type strain does prompted the authors to propose that PE formed by Psd2p is not efficiently transported into the organelle: this depletion would cause the phenotype observed in the psd1 mutants.

The three final steps required to synthesize PC from PE involve two phospholipid methyltransferases, encoded by the CHO2/PEM1 gene for conversion of PE into PMME (phosphatidylmonomethylethanolamine) and by the OPI3/PEM2 gene for conversion of PMME into PDME (phosphatidyldimethylethanolamine) and PDME into PC. The two genes were cloned by complementation of the relevant mutants (160, 293). The fact that mutations in either gene cause (160) or do not cause (204, 293) choline auxotrophy is explained by different strain backgrounds (61). Prototrophy for choline of the cho2 mutants is probably due to partial substitution by the Opi3 enzyme. With respect to the opi3 mutants, it is postulated that accumulation of PMME can partially replace PC. The sole report on petite production concerns a cho2 mutant which does not accumulate more petites than the reference strain (113). With respect to mtDNA, we would like to mention two of the five genes required for the salvage pathways towards either PE or PC (61). CPT1 encodes sn-1,2-diacylglycerol cholinephosphotransferase (133), while EPT1 encodes sn-1,2-diacylglycerol ethanolaminephosphotransferase (134). Griac and Henry (113) reported that a cpt1 ept1 double mutant, in which the salvage pathways through both free choline and free ethanolamine are blocked, does not produce more petites than the reference strain but that a cho2 cpt1 ept1 triple mutant exhibits a high level of petites (113). In this case, cell survival would be assured by the presence of PE (from the de novo pathway) and small amount of PC (due to the Opi3 activity and its ability to partly substitute for the Cho2 enzyme).

In all cases reported above (Table 7), the authors suppose that mitochondria are more sensitive to phospholipid starvation (or to desequilibrium in phospholipid composition of their membrane) than is the cell per se. Here again, the reason why these changes in membrane structure cause mtDNA instability relies on a putative link between mtDNA and the inner membrane (see "Is there a centromere-like structure for the mitochondrial genome" below). However, the recent discoveries reported below (and given in Table 7) open new windows on this major problem.

Until recently, it was assumed that mitochondrial needs for fatty acids were satisfied by import of molecules synthesized in the cytosol. However, several genes encoding subunits of a mitochondrial fatty acid synthase (FAS) complex similar to the bacterial (and chloroplast) complex (FAS type II) have now been characterized. Inactivation of these genes leads to a pet (respiration-deficient) phenotype and in some cases, to an increased production of petites (Table 7). The first gene discovered was CEM1 (for "condensing enzyme with mitochondrial function"), encoding a protein with strong similarities to -keto-acyl synthases (also called condensing enzymes). This gene was cloned by complementation of a pet mutant. Inactivation of CEM1 does not lead to a high production of petites (126). Two years later, Schneider et al. (267) disrupted the ACP1 gene encoding the mitochondrial acyl carrier protein (ACP), previously identified by systematic sequencing (40). In this case, inactivation of the gene causes loss of the wild-type mtDNA (Table 7). Schneider et al. (266) then characterized two other genes of the mitochondrial FAS on the basis of similarities between two open reading frames of the yeast genome and bacterial FAS subunits. These genes were named OAR1 (for "3-oxo-acyl-ACP reductase") and MCT1 (for "malonyl coenzyme A:ACP transferase"). Disrupted oar1 and mct1 strains exhibit an increased level of petites (266) (Table 7).

The role of mitochondrial FAS (excluding that of the CEM1-encoded subunit) in mtDNA stability has been recently supported by data from Stuible et al. (288). These authors identified (by sequence similarity) the PPT2 gene, encoding a mitochondrial phosphopantheine:protein transferase which specifically activates the mitochondrial ACP protein. Inactivation of PPT2 causes loss of the wild-type mtDNA, as previously shown for the acp1 disruptants (25, 267) (Table 7).

Harington et al. (126, 127), on one hand, and Schneider et al. (266, 267), on the other, propose two (nonexclusive) hypotheses to explain the pet phenotype of the four mitochondrial FAS mutants presently available. In the first case, the authors favor the hypothesis that the defect is situated at the level of the membrane insertion of proteins, a process which can require the covalent addition of fatty acid-like molecules to these proteins. This assumption is based on the fact that in the cem1 strain, mitochondrial apo-cytochromes are synthesized while holo-cytochromes are absent. Since neither the global synthesis of fatty acids nor their overall content and composition in the mitochondrial membranes appear altered in the cem1 disruptant, the authors suppose that Cem1p is involved in the synthesis of a specialized fatty acid required for proper anchorage of some proteins in the membrane (126, 127). Because the mutant does not produce a high level of petites, one must assume that this putative fatty acid-like component is dispensable for correct positioning into the membrane of all proteins required for mtDNA stability. The second hypothesis, proposed by Schneider et al. (266, 267), proposes that the mitochondrial FAS provides fatty acids required for remodeling and/or repair of phospholipids damaged by oxidation. This hypothesis may explain why the lack of any one of the four FAS subunits causes a respiration-deficient phenotype. There was no discussion of the increase in petite production which the authors observed in their mutants (266, 267). Invoking oxidative damage appears unsatisfactory, since the role of such processes in the maintenance of mtDNA is far from clear, as shown in the above section.

We would prefer to explore and broaden the ideas of Harington et al. (127). Specialized fatty acid-like molecules (as yet uncharacterized) may be required for the insertion of proteins into the mitochondrial membrane. Some might involve cytochromes: their absence would cause a pet phenotype, and their synthesis would depend on the entire FAS complex (with the four subunits so far identified). Others would concern the membrane insertion of (at least) one protein involved in mtDNA stability: their absence would lead to an increased production of petites, and their synthesis would not require Cem1p activity. The main virtue of this hypothesis is to offer a unifying view, including phospholipids (see above), of the relationship between fatty acid depletion and instability of mtDNA.

The identification of the molecules synthesized by the mitochondrial FAS should shed light on the problem. With respect to the pet phenotype of the cem1 mutants, Harington et al. (127) discussed the possible involvement of lipoic acid (which is a cofactor of the pyruvate dehydrogenase and -ketoglutarate dehydrogenase complexes) but rejected this candidate for two reasons. First, although lipoic acid-deficient mutants display a pet phenotype, they differ from the cem1 mutants because they have spectrally detectable cytochromes (308). Second, the addition of lipoic acid to the medium does not restore respiratory growth to the cem1 strains. In contrast, Brody et al. (25) showed that strains carrying a disrupted acp1 gene have only 5 to 10% of the wild-type lipoic acid content. The mutants are obviously respiration defective since they are massively converted into petites, as previously shown by Schneider et al. (267). The role of Acp1p in lipoic acid synthesis is well supported in that a similar deficiency in this compound is observed when the PPT2 gene is disrupted, a situation in which Acp1p is not activated (288).

The fact that mtAcp1p is clearly involved in lipoic acid synthesis does not disprove the idea that the mitochondrial FAS complex (with the exception of its Cem1 subunit) could be involved in the synthesis of another fatty acid-like molecule required for mtDNA stability. That acp1 mutants are not supplementable by addition of lipoic acid to growth medium (25) is not surprising, because the massive production of petites occurs as early as spore germination. This experiment should be performed on strains carrying a disrupted acp1 gene complemented by the wild-type gene present on a plasmid, such as those constructed by Schneider et al. (266) and Stuible et al. (288): loss of the plasmid in the presence of lipoic acid would help to clarify whether the massive production of petites characteristic of the mutants defective for Acp1p is due to the lack of lipoic acid or to another function of Acp1p. For the moment, a possible role of lipoic acid in mtDNA stability cannot be excluded. However, data from Sulo and Martin (292) concerning the LIP5 gene (see below) suggest that the absence of lipoic acid per se is not responsible for a high production of petites. Thus, the reason why three of the four subunits of the mitochondrial FAS are required for mtDNA stability remains unclear.

In fact, the idea that other mitochondrial fatty acids in addition to lipoic acid are required for respiratory competence and/or mtDNA stability is strongly supported by data obtained by Hoja et al. (137) on acc1 and bpl1/acc2 mutants. BPL1 encodes a biotin-protein ligase which is required for acetyl coenzyme A (acetyl-CoA) carboxylase holoenzyme formation. This enzyme (whose apoform is encoded by ACC1/FAS3) catalyzes key steps of fatty acid synthesis. Viable mutants of the two genes require long-chain fatty acids for growth and are respiration deficient, even though lipoic acid is reduced only to about half its normal level. Furthermore, the acc1 mutant studied was a petite, because its respiration defect could not be cured by transformation with the wild-type ACC1 gene (137) (Table 7).

In addition to the involvement of lipids, the role of the mitochondrial membrane for mtDNA maintenance will be further discussed (see "Mitochondrial morphology, the cytoskeleton, and mtDNA" and "Is there a centromere-like structure for the mitochondrial genome" below).

A growing number of enzymes whose functions in metabolism have been well documented appear to play an unexpected role in mtDNA stability. To our knowledge, the first case reported was the E1 subunit of the pyruvate dehydrogenase complex (322). This complex links glycolysis to the citric acid (Krebs) cycle by conversion of pyruvate to acetyl-CoA. Deletion of the PDA1 gene encoding the E1 subunit leads to two unexpected phenotypes: a partial auxotrophy for leucine and an increased production of rho⁰ petites (Table 8). The first defect could be explained by a high concentration of CoA in the mitochondria, inhibiting a reaction specific for the leucine pathway, while the effect of PDA1 deletion on rho⁰ production remains unexplained (322). It would be interesting to know if petite production is also increased in mutants with mutations of the genes encoding the other subunits of the pyruvate dehydrogenase complex (reference 243 and references therein).

Sulo and Martin (292) were interested in mitochondrial tRNA processing. For this reason, they cloned a gene able to complement a respiratory mutant which showed an accumulation of mitochondrial tRNA precursors. The gene in question was involved in lipoic acid synthesis. Disruption of this gene (LIP5) leads to the production of rho mutants (Table 8). Addition of lipoic acid to the culture medium permits the recovery of protein-bound lipoate but does not prevent the destabilization of mtDNA. In contrast to acp1 mutants, which massively lose their rho⁺ genome (25) (see the previous section), the kinetics of rho accumulation in lip5 disruptants are slow (292). Thus, in this case, the attempt to prevent petite production with lipoic acid makes sense. Because other genes involved in lipoate metabolism do not show this phenotype, the authors propose that Lip5p may have dual functions, with the second function being required for mitochondrial translation. However, this second role might be dispensable, because disruption of LIP5 in a op1 mutant background is viable (292). Notably, in Kluyveromyces lactis, a mutation in the LIPB gene (which encodes a lipoyl-protein ligase) causes a destabilization of the mitochondrial genome (36).

NAD⁺-dependent isocitrate dehydrogenase catalyzes an essential step in the Krebs cycle, the conversion of isocitrate to -ketoglutarate. The genes encoding the two subunits of this enzyme (IDH1 and IDH2) have been cloned (58, 59). The complex was rediscovered through its RNA-binding property. In fact, it binds specifically to the 5' untranslated leaders of mitochondrial mRNAs (reference 82 and references therein). The authors proposed that the protein plays a role in mitochondrial translation. This hypothesis was strengthened by the observation that strains with either of the IDH1 and IDH2 genes disrupted displayed increased rates of petite production (Table 8). However, as reported to us by L. Grivell, the problem is more complicated than it appears. First, the increase in petite frequencies is not as high as that observed with mutations in genes encoding components of the mitochondrial translational apparatus. Second, a wild-type level of petites can be observed in a idh1 mutant complemented by a mutant form, nonbinding to RNA, of Idh1p. Third, there is no significant difference (observed by several tests) between the wild type and the nonbinding mutant in mitochondrial translation. The slight but significant effect of the idh disruptions on mtDNA stability might be explained in two ways, as suggested by Grivell. First, RNA binding may not be important to this process. This hypothesis suggests that the Idh subunits may have three functions: one in the Krebs cycle (which is probably not related to mtDNA stability), one in mtRNA binding (with a nonessential function with respect to translation), and an unknown role distantly related to mtDNA stability. In the second hypothesis, the RNA-binding property of Idh subunits would be related to mtDNA stability but the low level of RNA binding activity remaining in the idh (non binding) mutants may be sufficient to fulfill the needs of mitochondria to maintain their genome (L. Grivell, personal communication).

The ILV5 gene encodes the mitochondrial protein acetohydroxy acid reductoisomerase, catalyzing a step in the biosynthesis of the amino acids isoleucine, leucine, and valine (235). This gene was rediscovered as a multicopy suppressor of the mtDNA instability phenotype of abf2 mutants (see "ABF2 paradox" above) (Table 1). Interestingly, suppression is also seen when ILV5 is carried by a low-copy-number plasmid. Furthermore, it was shown that deletion of ILV5 leads to a high frequency of petites (336) (Table 8). This mtDNA instability is clearly not the consequence of a block in biosynthesis of branched-chain amino acids. The authors propose a bifunctional role for acetohydroxy acid reductoisomerase: in addition to its catalytic activity, the protein could also function as do both Abf2 (a DNA-packaging protein) and Idh (an mRNA-binding protein). Interestingly, while abf2 petites are rho⁰, those produced in the ilv5 context are rho. Furthermore, there is a synergistic effect of the two deletions, with a more extreme mtDNA instability than that seen in either of the two single deletions: in contrast to the single mutants, the double mutant loses its rho⁺ genome as early as spore germination. This suggests a synergistic effect of ABF2 and ILV5 in mtDNA maintenance, which may explain the suppression effect of ILV5 on abf2 (336). It would be interesting to see the phenotype caused by overexpression of ABF2 in a ilv5 context.

Throughout the history of genetics, different screening procedures have sometimes led to the discovery of different genes, even though all are involved in the same function. Also, different screens can lead to the discovery of the same gene: the reader will find several examples in this review. It is therefore not surprising that some genes required for mtDNA stability have been fortuitously identified. In most cases covered in this review, the actual mechanisms of the relevant proteins responsible for their involvement in mtDNA stability seem plausible. However, there are (at least) seven serendipitous discoveries which remain unexplained. They concern the four enzymes discussed in the above section and the proteins encoded by the ADR1, SLS1, and SMP1 genes. The story of ADR1 is presented below in "Gene overexpression can increase mtDNA instability" (see Table 12).

The SLS1 gene (for "synthetic lethal of ssm4") was identified by a mutation colethal with a deletion of the SSM4 gene. The latter, when inactivated, is able to suppress some (but not all) mutations of RNA14, which encodes a protein localized in the nucleus and required for mRNA polyadenylation. Ssm4 is an integral nuclear membrane protein (reference 257 and references therein). To better understand the function of Ssm4p, these authors sought mutations colethal with an SSM4 deletion (257). The sls1-1 mutation was one of these. It was shown to display a pet phenotype, as did a complete deletion of the gene. Furthermore, the data strongly suggest that Sls1 is a mitochondrial integral membrane protein. The protein does not seem to be involved in mtDNA replication or transcription or in mtRNA splicing, maturation, or stability. However, the mutants show a defect in the assembly of complexes III and IV (lack of cytochromes c₁, b, and aa₃). The original mutant, as well as the deletion, leads to a high production of petites (Table 9) when the strains harbor a mitochondrial genome containing introns. This production is reduced when the cells contain an intronless mitochondrial genome (257), but, this value was not compared to the level of petites produced in the isogenic wild-type context. Although the increased stability of intronless mitochondrial genomes has been found in other nuclear mutant contexts (for instance, the nam1 mutants [see "The translation-ATP synthase connection" above]), the problem remains unsolved. Concerning the SLS1 gene, its requirement for the assembly of two respiratory complexes does not shed light on its role in mtDNA stability. However, the authors did not report on its effects on mitochondrial translation or in the assembly of ATP synthase, two processes clearly linked to mtDNA stability (see "The translation-ATP synthase connection" above). Even more mysterious is the reason why mutations of the SLS1 gene, encoding a mitochondrial membrane protein, can be colethal with deletion of the SSM4 gene, encoding a nuclear membrane protein (257).

The smp1 (for "stable maintenance of pSR1") mutation was identified among mutations increasing the stability of pSR1, a Zygosaccharomyces rouxii plasmid which is maintained in S. cerevisiae less stably than in Z. rouxii. The pSR1 molecule shows three open reading frames: one encodes a protein essential for intramolecular recombination of the plasmid, while the two others encode proteins involved in its maintenance, which would also require a specific host factor (reference 141 and references therein). The smp1 mutation enhances the stability not only of this foreign plasmid but also of S. cerevisiae plasmids, if they contain an ARS sequence but no centromere. For an as yet unexplained reason, the smp1 cells are rho⁰ petites (Table 9). Amazingly, the stability of the pSR1 plasmid is also increased (by a factor of 20, as in the smp1 context) in SMP1 cells which are rho or rho⁰ (141). This effect is not due to respiration deficiency per se, because pet mutants (which maintain their rho⁺ genome) do not show this phenotype. These data pose two (probably related) questions. First, how can mtDNA loss enhance the stability of a nonmitochondrial plasmid? Second, how can a nuclear mutation lead to both increased stability of a plasmid and loss of mtDNA? With respect to the first question, "incompatibility" between mtDNA and a foreign plasmid was previously reported. For instance, Gunge and Yamane (122) reported two linear DNA killer plasmids from K. lactis which were stably transmitted in S. cerevisiae if the recipient strain was a neutral petite. In contrast, mit mutations exerted an "incompatibility" effect on plasmid transmission, as observed in a rho⁺ context. An ad hoc hypothesis to explain this effect of a petite background relies on established observations that the status of the mitochondrial genome can modify the expression of some nuclear genes (151, 226). This phenomenon, called retrograde regulation, has been extensively studied in yeast (reviewed in references 35 and 240). New technologies, which can describe the transcriptional status of the entire yeast nuclear genome in a given genetic background, will aid in obtaining a catalogue of genes whose expression is altered in a petite context. However, we believe that a simple titration idea could account for the observed situations. If a protein X is titrated both by mtDNA and by some specific plasmid DNAs, loss of one of the partners could increase the availability of X for the other partners. With respect to the second question, one should note that nuclear mutations stabilizing a plasmid do not systematically cause mtDNA instability. For example, Thrash-Bingham and Fangman (302) reported that the amm1 (for "altered minichromosome maintenance") mutation, although leading to a reduced growth in glycerol medium at 34°C, did not increase petite production. Clearly, further experiments (especially cloning of SMP1) are required to understand why the smp1 mutation leads to loss of mtDNA.

In contrast to nuclear DNA replication, which occurs at a specific step of the cell cycle (the S phase), mtDNA replication occurs throughout the cell cycle (270, 326). However, the localization of mtDNA molecules within the organelle and the organization of the mitochondrial network inside the cell must be tightly controlled so that mitochondria and the mitochondrial genomes are precisely distributed between mother and daughter cells, both during the mitotic cycle and among the spores after meiosis. In both cases, careful cytological analyses have been performed. With respect to mitotic cell divisions, mitochondria can be seen at the mother cell/bud neck as early as bud emergence. The growing bud is then colonized by mitochondria (reviewed in reference 286). In meiosis, mitochondria are arranged around the meiotic nucleus in a very specific pattern (208). Other studies have carefully described the fate of mitochondria and their genomes during and after mating (reference 222 and references therein). However, the precise mechanisms which orchestrate mitochondrial shape, fusion, fission, and movement, as well as the mechanisms by which the pool of mtDNA molecules is distributed in mitotic and meiotic progeny, remain unclear. These problems are discussed below (see "Mitochondrial morphology, the cytoskeleton, and mtDNA" and "Is there a centromere-like structure for the mitochondrial genome?"). To date, studies of the cell cycle have provided very little information on that topic.

In fact, very few CDC (for "cell division cycle") genes have been reported to influence mtDNA stability. One can distinguish two groups, whose effects on mtDNA are either quite trivial or still very puzzling. The first includes the CDC8 and CDC21 genes (Table 1), which encode thymidylate kinase (150) and thymidylate synthase (19), respectively. It was initially reported that the cdc8 and cdc21 mutants not only stop nuclear replication at high temperature but also arrest mtDNA replication, which is not a general feature of cdc mutants impaired in the initiation or achievement of the S phase (216). Later, a high frequency of petites was observed in these mutants at permissive temperature (217). It is not surprising that blocks in the pathway of deoxyribonucleotides lead to an arrest of both nuclear and mitochondrial DNA replication. The fact that the fate of mtDNA is altered at temperatures still permissive for nuclear replication is in accordance with other data and suggests that mtDNA synthesis is highly sensitive to a decrease in the supply of DNA precursors. This is exemplified by the effects of low-activity mutants or overexpression of the RNR genes encoding subunits of ribonucleotide reductase, which catalyzes the conversion of ribonucleotides into deoxyribonucleotides. A partial deletion of RNR2 causes a 10-fold increase in petite formation (Table 1) (80), while a strain carrying a deletion of the RNR4 gene displays a petite phenotype due to the loss of functional mtDNA (Table 1) (319). Reciprocally, the RNR1 gene (81) acts as a multicopy suppressor of mutations in the MIP1 gene, which encodes the mtDNA polymerase (see "Wide-scale screening for rho⁰ production" above) (Table 1) (177). The suppressor effect was observed not only in several heat-sensitive mip1 mutants but also in diploid strains harboring a single wild-type copy of the MIP1 gene which accumulate petites at high temperature: overexpression of RNR1 decreased the level of petite production. The data as a whole strongly suggest that the deoxyribonucleotide concentration is rate limiting for mtDNA replication. Although it is unclear if the relevant petites are rho and/or rho⁰, it should be noted that Guan (118) found mutations in the CDC8 and CDC21 genes during his systematic search for rho⁰ producers (see "Wide-scale screening for rho⁰ production" above).

The second class of mutants impaired in both the cell cycle and mtDNA stability can be illustrated by three examples: the cdc28-srm allele, the gcs1 sed1 double-mutant strain, and the erv1 mutant. Devin and Koltovaya (66) directly sought nuclear mutations decreasing petite production. They began with a strain producing 50% petites and selected four mutations giving rise to a lower level of petites (Table 10) (see "No compromise" below). In further studies, Devin et al. (67) found a fifth mutation, shown to be allelic to CDC28. This gene encodes the protein kinase called p34^cdc28 (p34^cdc2 in Schizosaccharomyces pombe and other eucaryotes) which, in yeasts, controls both the G₁/S and G₂/M transitions of the cell cycle by interactions with specific cyclins (reviewed in reference 181). Mutations in CDC28 which impair either one of these key steps lead to heat-sensitive mutants. The cdc28-srm allele described by Devin et al. (67) is not conditional. However, the mutation is pleiotropic: it leads to alterations in cell morphology, loss of extra chromosomes in disomic strains, and loss of circular centromeric plasmids and decreases the spontaneous and ethidium bromide-induced rates of petites (Table 10). This effect on petite frequency was also found for one of two conditional cdc28 mutants at permissive or semipermissive temperatures (162) (Table 10). The reason why some CDC28 alleles have this striking effect on mtDNA stability remains unknown.

Drebot et al. (70) described a strain (carrying two mutations) defective for reentry into the cell cycle from the stationary phase at low temperature. One of the mutations causes a general cold sensitivity (gcs1), while the other (sed1) suppresses this defect in proliferating cells. The double mutant (but not the single mutants) accumulates petites of the rho⁰ type at high rates (92) (Table 9). Here again, the observation remains unexplained.

The ERV1 gene (for "essential for respiration and viability") was identified by a pet mutation. The mutant was respiration deficient at high temperature, showing strongly reduced levels of mitochondrial transcripts. Disruption of the complementing gene demonstrated that it was essential for viability (184). Complementation of the erv1 disruption by the wild-type gene under the control of an inducible promoter permitted the author to monitor the fate of cells transferred to culture conditions which no longer allow ERV1 expression (184). It appeared (by DAPI staining) that the cells first lose their mtDNA (Table 9) and then arrest, after a few divisions, with a cdc phenotype in a budded state, after migration of the nucleus into the neck between the mother cell and the bud (186). This phenotypic trait does not clearly determine if the cells are arrested in the G₁ or G₂ state. The protein was later shown to be associated with the mitochondria (187). The identification of a human homologue of the ERV1 gene (105, 189) has not shed light on the function of the protein.

With respect to cell cycle topics, it is interesting that some mutations which impair mitochondrial morphology (see "Mitochondrial morphology, the cytoskeleton, and mtDNA" below) also lead to a cdc phenotype. For instance, the mmm1 (for "maintenance of mitochondrial morphology") mutants not only display a defect in organelle morphology and inheritance but also are defective in the completion of cytokinesis at restrictive temperature when grown on a nonfermentable carbon source (28). At the time, the authors could propose the attractive hypothesis of a cell cycle checkpoint associated with mitochondria: when the bud does not receive mitochondria, cell division is forbidden. However, the situation is no longer so clear. Among the mdm (for "mitochondrial distribution and morphology") mutants selected by Yaffe's laboratory (202), some, but not all, display specific cell cycle defects. For instance, the mdm1 (203) and mdm2 (287) mutants are unable to achieve cytokinesis. In contrast, mutations in the MDM10 and MDM12 genes (see the next section and Table 11) do not exhibit this feature. Furthermore, there is no correlation between the cell cycle defect and the ability of the mutants to maintain their mitochondrial genome: the mmm1 mutants do not accumulate petites (28), while at least some mdm2/ole1 mutants exhibit this property (see "Fatty acids and phospholipids," above, and Table 7). Clearly, elucidation of the relationship between the cell cycle and mtDNA transmission remains a formidable challenge.

In growing yeast cells, mitochondria form reticular structures localized at the cell cortex (reviewed in reference 286). The continuity of this network depends on a balance between the fusion and fission events required, along with mitochondrial movements, for correct transmission of the organelles at cell division, mating, and meiosis (references 222 and 282 and references therein). Genes involved in mitochondrial movements and morphology have been identified either by a specific search for mutants altered for these processes or by a posteriori inspection of mutants screened for other purposes. Some encode components of either the cytoskeleton or the outer membrane of mitochondria, while others could play regulatory roles in these processes (reviewed in reference 18). However, the link between the fate of the organelle and its genome remains elusive, because only a subset of these genes also control mtDNA stability.

McConnell et al. (202) selected, by visual inspection, temperature-sensitive mutants exhibiting defects in mitochondrial distribution and morphology (mdm mutants). The MDM10 (283) and MDM12 (17) genes encode proteins of the mitochondrial outer membrane. In either case, loss of the proteins results in giant mitochondria defective for transfer into daughter buds. The authors suggest a role for these proteins in the mitochondrion-cytoskeleton linkage. This attachment would be required for maintenance of the normal (tubular) mitochondrial morphology, with the defects in organelle inheritance being a secondary consequence of abnormal morphology. The mdm10 and mdm12 mutants produce petites at rates much higher than the wild type does (17) (Table 11). Two other proteins of the outer mitochondrial membrane are also involved in morphology and mtDNA stability: Fzo1 and Mgm1. They share a GTP-binding property, as discussed below. Interestingly, there is (at least) a fifth outer membrane protein of the mitochondria essential for maintenance of organelle morphology; it is encoded by the MMM1 gene, whose mutations also lead to giant mitochondria unable to enter buds. It differs from the four other genes in that mmm1 mutants contain functional mtDNA (28). This suggests that abnormal mitochondrial morphology does not inescapably lead to petite production.

Hermann et al. (130) also screened a collection of mutants for strains unable to transport mitochondria into the buds at high temperature. Isolation of the mdm28 mutant has contributed to the comprehension of the timing of mitochondrial movement at this stage of the cell cycle (255). In fact, the mutant shows a delay (but not a block) in mitochondrial inheritance: while in wild-type cells mitochondria can be seen in the buds as early as their emergence, in the mutant most small buds are devoid of mitochondria while most large buds receive a mitochondrial network. Cloning and sequencing of the MDM28 gene revealed that it corresponds to the previously identified gene PTC1, encoding a type 2C serine/threonine phosphatase (200) and initially identified in a screen for mutants deficient in tRNA biosynthesis (TPD1 [315]). In their study of MDM28/PTC1, Roeder et al. (255) showed that ptc1 mutants generate petites at a higher frequency than the wild type does (24.4 and 1.8%, respectively, at 30°C) (Table 11). Although Ptc1p is assumed to negatively regulate mitogen-activated protein kinases in the high-osmolarity glycerol response (HOG) pathway (which plays a critical role in the osmostress response in yeast), the authors provide evidence that Ptc1p does not act by this pathway with respect to mitochondrial transport. Ptc1p may act directly on proteins required to move mitochondria into buds, or it may affect mitochondrial inheritance indirectly, by a pathway different from HOG. Since the actin cytoskeleton is known to play important roles in mitochondrial distribution, morphology, and movement (71, 176, 277), Roeder et al. (255) compared actin structures in ptc1 and PTC1 cells but observed no differences. Thus, the reasons why ptc1 mutants exhibit this delay in mitochondrial transmission and the possible existence of a link between this defect and mtDNA instability remain unexplained.

The fuzzy onion gene (fzo) was first identified in Drosophila melanogaster as essential for a developmentally regulated step in mitochondrial fusion (124). Mitochondrial aggregation, fusion, and elongation alongside the growing flagellar axoneme are successive events during Drosophila spermatogenesis and are necessary for sperm motility. The mitochondrial structure then resembles a slice of onion, as viewed in cross-section by electron microscopy. Mutations in fzo cause a specific defect in the fusion step, leading to aberrant structures ("fuzzy onions") and male sterility. The gene encodes a transmembrane GTPase targeted to the mitochondria. Related proteins have been found in several organisms, including S. cerevisiae (124). The yeast gene has been studied by two groups (131, 244). In both cases, the data lead to the conclusions that Fzo1 is an integral protein of the mitochondrial outer membrane and fzo1 mutants display fragmented mitochondria and mtDNA instability (Table 11). However, one of the two groups supplied several important observations (131). First, in conditional mutants transferred to restrictive conditions, the mitochondrial network initially fragments into very small mitochondria which then aggregate into a few clusters, a profile also observed in the fzo1 mutants. Second, the absence of Fzo1p does not affect mitochondrial inheritance but blocks mitochondrial fusion during mating. Third, the fzo1 petites are rho⁰ (Table 11). Fourth, loss of mtDNA is observed many hours after changes in mitochondrial morphology (131). Although further experiments are required to understand why fzo1 mutants lose their mtDNA, one cannot exclude a trivial possibility: some of the small mitochondria might not contain DNA. Because the mutant mitochondria never fuse, the level of organelles devoid of DNA would continuously increase in the population. If this hypothesis was confirmed, the implication would be that loss of mtDNA in the fzo1 mutants is a passive consequence of mitochondrial fragmentation which, in this case, cannot be balanced by fusion events. However, a fragmented mitochondrial pattern does not seem systematically linked to mtDNA instability (see reference 252, for example).

In addition to FZO1, MGM1 encodes a mitochondrial GTP-binding protein, as mentioned above (see "Wide-scale screening for rho⁰ production"). In this case, we know not only that the mutant produces rho⁰ petites (Table 11) but also that it is unable to maintain rho genomes, even those which are otherwise very stable (149). The MGM1 gene encodes a protein related to dynamin, whose protein family may have different functions. In several cases, a role in intracellular vesicle movement has been suggested. Interestingly, Guan et al. (117) observed that mitochondria of an mgm1-deleted strain are present in condensed structures rather than in the dispersed pattern seen in MGM1 strains (either rho⁺ or rho⁰). Although reminiscent of the giant mitochondria found in mdm10, mdm12, and mmm1 mutants, the aberrant structures in mgm1 partition at cell divisions, at least in the strain background used by these authors. The data also suggest a defect in protein import. Mgm1p has been recently rediscovered by Shepard and Yaffe (272) by using the heat-sensitive mutation mdm17. Studies of mdm17 and a null allele of MGM1 confirm previous observations concerning mitochondrial aggregation and loss of mtDNA (Table 11) and add important information. First, Mgm1p is located to the mitochondrial outer membrane. Second, the GTP-binding site is required for Mgm1p functions. Third, in contrast to the observations of Guan et al. (117), Shepard and Yaffe (272) observed altered mitochondrial distribution to the buds (suggesting that this defect may be strain dependent). Fourth, after a shift to the nonpermissive temperature, loss of mtDNA occurs after alterations in mitochondrial morphology and distribution. This delay in mtDNA loss was also reported for the fzo1 mutants (131). Another nucleotide-binding protein must be mentioned here. It is encoded by the YME1 (for "yeast mitochondrial DNA escape") gene, which is discussed below (see "No compromise"). Yme1p is an ATP- and zinc-dependent protease, localized to the inner mitochondrial membrane, whose substrates are not yet clearly defined (320). In any case, yme1 mutants display abnormal mitochondrial morphology: rather than being reticular, mutant mitochondria form small globular compartments (30). In contrast to the above-mentioned situations, yme1 mutations do not cause a high production of petites but are nearly lethal in a petite background (see "No compromise" below and Table 13).

The putative link between mtDNA stability and the cytoskeleton is emphasized by the observation that deletion of the TPM1 gene encoding tropomyosin (192), an actin-binding protein, leads to increased petite production (193) (Table 9). This phenotypic property of the tpm1 mutant cannot be generalized to all components of the cytoskeleton. For instance, some actin mutants display aberrant mitochondrial organization while maintaining normal rates of petite formation (71). It is noteworthy that sporulation conditions (which require respiration) enhance defects in mitochondrial organization: under these conditions, some actin mutants exhibit giant mitochondria similar to those observed in the mdm and mmm mutants (282).

Where is Ariadne's thread in this complex network of probably indirect effects in the cytoskeleton, mitochondrial morphology, and mtDNA stability? One can follow two (nonexclusive) trails. The first relies on a link between the frameworks required for mitochondrial morphology and mtDNA maintenance, and the second relies on a link between organelle morphology and the cytoskeleton, and/or an intact import process required for mtDNA stability.

The first proposal is evoked by Burgess et al. (reference 28 and references therein). These authors posit an internal framework (mitoskeleton) and an external framework (attachment of mitochondria to the cytoskeleton) which could be involved in mitochondrial morphology. The external framework was also proposed for the MDM10 and MDM12 genes (17, 283). It has recently been shown that Mmm1p and Mdm10p play a key role in linking mitochondria to the actin cytoskeleton (20). However, the three mutants clearly differ with respect to mtDNA stability, which is not affected in the mmm1 mutant. These are cellular nuances we have not yet grasped, although the idea that mtDNA transmission could be influenced by organelle morphology appears sensible. There is in fact an enormous gap in our understanding of the possible interactions between mtDNA and the inner membrane, which certainly reacts to modifications of the outer membrane and its attachment to the cytoskeleton (see "Is there a centromere-like structure for the mitochondrial genome?" below). The TPM1 problem may also be explained in the context of a putative external framework. Similarly, Mgm1p and Fzo1p might be involved in one of these frameworks. Here again, the suppressor method may help to distinguish between this hypothesis and that proposed by Jones and Fangman (149), who favor the idea of a signal transmission (by GTP binding and hydrolysis) required for mtDNA replication.

The second proposal is based on the still unproven but attractive hypothesis that efficient import of proteins into the mitochondria involves a selective transport of the relevant mRNAs to the vicinity of the organelle (reference 190 and references therein). Particular defects in the cytoskeleton (TPM1) or in the attachment of mitochondria to the cytoskeleton (MMM1, MDM10, and MDM12) may disturb the delivery of proteins by causing a defect in the routing of mRNAs. On the other hand, and without a cytoskeletal implication, the abnormal organelle morphology might alter the import process, as shown for the mitochondrial Hsp58 protein in mgm1 mutants (117), even though the relevant proteins (in their wild-type form) are not directly involved in the process. It is noteworthy that a null allele of YGE1/MGE1 (encoding a protein required for protein translocation into mitochondria [see "Conventional mitochondrial chaperones" above]) results in mitochondria stacked in the mother cell and unable to enter the bud while a heat-sensitive allele results in a high production of petites (Table 5). Depending on these more or less specific defects, mtDNA stability might be either impaired (MGM1, MDM10, MDM12, or FZO1) or unaffected (MMM1). Careful analysis of the import process in the relevant mutants would be very informative.

Although gene overexpression can give insight into gene functions, this approach has been little used on studies of mtDNA stability. We have previously noted that overexpression of MGT1/CCE1 enhances the transmission of nonsuppressive rho genomes (see "Nuclear control of biased transmission" above and Table 2) and that overexpression of ABF2 leads to rho⁰ petites while deletion of the gene increases the production of rho genomes (see "The ABF2 paradox" above and Table 1). We would like to discuss two genes at this point, PET127 and ADR1 (Table 12), whose overexpression increases mtDNA instability while their absence does not.

PET127 was identified by two different screening procedures (123, 324). In the first, the pet127 mutation acted as a suppressor of mutations in PET122, a gene encoding a translational activator of the mitochondrially encoded COX3 mRNA. In the second, pet127 suppressed cox3 mutations with decreased mRNA stability. Deletion of PET127 causes only a slight respiratory defect at 37°C. However, the data suggest that Pet127, which is a membrane-associated protein, plays important roles in mtRNA turnover and 5'-processing cleavages. In contrast, overexpression of PET127 shifts cells to the petite phenotype (324). The authors suggest that excess of Pet127p results in the destruction of an mtRNA required for mtDNA replication or translation.

Adr1p is the primary transcriptional activator of the ADH2 gene, encoding the glucose-repressible alcohol dehydrogenase in S. cerevisiae (45, 63). Since Adr1p has no obvious mitochondrial function, it was quite surprising to find that overexpression of the ADR1 gene led to a high production of petites (41) (Table 12). The authors provided evidence that this phenomenon was not due to increased expression of ADH2 but to the Adr1 activity per se. Furthermore, they performed a careful analysis of petites which extended far beyond that of most of the studies discussed in this review. Using genetic criteria, they made several observations. First, 25% of these petites were rho⁰ (or contained severely deleted mtDNA) while the others were rho with nonspecific deletions and a bias in marker loss similar to what is observed in wild-type strains. Second, the rho petites were unstable and progressed to the rho⁰ state. Third, only daughter cells became petites whereas a mother cell was still able to give rise to normal progeny after producing a petite bud. Fourth, resistance to Adr1 hyperactivity (in a certain genetic context) correlated with resistance to euflavine and high temperature effects for petite production; these two conditions also cause only buds to become petites (83, 273). Suppressors of the overproduction of petites were sought in order to clarify the Adr1p effect. Unfortunately, all (about 100) revertants displayed a reduced activity of Adr1p. The authors suggest that, when overproduced or hyperactive, Adr1p may bind (specifically or nonspecifically) to regulatory sequences of nuclear genes required for mtDNA stability and decrease their expression (41). In fact, more recent data have shown that the regulatory functions of Adr1p are much broader than previously assumed (references 53 and 276 and references therein). However, the function of Adr1p with respect to mtDNA stability remains unclear.

S. cerevisiae is a petite-positive yeast, which means that it can survive without mtDNA (27). However, mutations in several nuclear genes transform S. cerevisiae into a petite-negative yeast which no longer (or barely) tolerates the cytoplasmic petite state (Table 13). In one case (adc1 mutants), the explanation relies on energy production. In the others, there is a common problem which concerns the membrane potential () required for mitochondrial biogenesis and thus for cell survival.

S. cerevisiae contains two cytosolic alcohol dehydrogenases: AdhI is constitutively expressed, while AdhII is glucose repressed and can be induced only under respiratory conditions. They are responsible for alcoholic fermentation, by driving the glycolytic pathway to ethanol, which leads to the production of ATP. Under anaerobic conditions the entire energy supply of the cell depends on AdhI, because AdhII cannot be induced. AdhII activity is also absent in respiration-deficient mutants. Therefore, under nonrespiratory conditions, energy production is absolutely dependent on AdhI. This easily explains why cells lacking AdhI (adc1 mutants) cannot survive as petites: both their energy sources, fermentation by constitutive AdhI and respiration, are blocked (references 45, 46, and 327 and references therein).

To fully understand the other situations, one must keep in mind the following proposal: it is currently assumed that the membrane potential of respiration-deficient mutants relies on the entry of ATP (via the ADP-ATP carriers) in exchange for ADP produced by the F₁ sector of ATPase, which, in this context, acts as a ATP hydrolase (reference 107 and references therein).

The first case reported was that of the op1 mutation, which prevents growth on nonfermentable media (166, 167) due to a missense mutation in the AAC2 gene (175), the most important of three genes encoding the mitochondrial ADP-ATP carriers (reference 161 and references therein). The colethality of the op1 and rho/rho⁰ mutations was demonstrated as follows. Mutants bearing the op1 mutation lose the ability to multiply even on fermentable media after growth in the presence of acriflavine, which converts wild-type cells into petites. However, rho/rho⁰ mutations indeed occur in the op1 context, since the acriflavine-treated cells can give rise to viable respiration-deficient diploids when crossed with OP1 petite cells. Furthermore, petite production is restored in op1 revertants (167). The reason why the op1 mutation is lethal in a petite context is still unclear. However, ATP certainly is at the heart of the problem: in petites, ATP is provided to mitochondria via the AAC translocators (this import is probably reduced when the major [Aac2p] translocator is nonfunctional [op1 mutant]). However, the fact that op1 is unable to withstand the petite state cannot be explained simply by the limited amount of ATP provided to the mitochondria by the minor translocators Aac1 and Aac3: op1 is viable in presence of mit mutations. This mutant has even been used as a powerful tool in isolating mit mutations, without the background noise of rho/rho⁰ petites (163). Interestingly, not a single stringent mit mutation has been found, in the op1 context, in the genes encoding Atp6, Atp8, and Atp9 subunits of ATPase, among hundreds of mit mutations located in COX1, COX2, COX3, and cytb genes (see references 163 and 241 for examples). This observation can be explained by the fact that atp6, atp8, and atp9 stringent mutants would be quickly converted into petites. In fact, one major difference between mit mutants (mutated in genes encoding a component of the electron transport chain) and petites is that the F₀ sector of ATPase is present in the first case and absent in the second. The F₁ complex might be less stable (or less active) when it is not anchored (petite mutants) than when it is anchored to the F₀ sector (mit mutants): differences between these two forms of F₁ were suggested by Schatz (262). Thus, in op1, limited entry of ATP (via the minor Aac1/Aac3 translocators) in addition to insufficient ATP hydrolysis could cause a dramatic decrease in the ATP-ADP exchange and thus in the potential , below the threshold required for viability. A search for extragenic suppressors of the aac2 petite lethality could be very useful in clarifying this problem. The above hypothesis predicts that mutations increasing the stability (activity) of the "soluble" petite F₁ would be found among these suppressors.

The second mutation demonstrated as intolerant of the petite state lies in the PEL1 (for "petite lethal") gene (146, 291). It has been recently demonstrated that PEL1 (renamed PGS1) encodes phosphatidylglycerophosphate synthase, the first enzyme in the cardiolipin biosynthesis pathway (34). The Pel1/Pgs1 protein is localized to mitochondria (78). Cardiolipin is a lipid found predominantly in the inner mitochondrial membrane (reviewed by Daum et al. [61]). The biosynthesis of this dimeric phospholipid is a three-step pathway, with the final step being catalyzed by cardiolipin synthase. The gene (CLS1) encoding this enzyme was identified in the yeast genome by similarities to eucaryotic cardiolipin synthases (148). It appeared that CLS1 inactivation was not lethal and did not cause a pet phenotype (below 37°C), although growth on respiratory substrates was lowered. To explain this surprising result, the authors suggested that PG, the precursor of cardiolipin which accumulates in cls1 mutants, might adequately substitute for cardiolipin, at least below 37°C. This hypothesis was confirmed a year later, when the gene (PEL1/PGS1) encoding the first enzyme of the cardiolipin biosynthesis pathway was clearly identified (34). Mutants of this gene lack not only cardiolipin (145) but also PG (34). Disruption of PEL1/PGS1 leads to the same phenotypic defects reported for the original pel1 mutants: respiration deficiency, heat sensitivity for growth on fermentable conditions, and lethality in a petite background (34). Many data have focused on the pleiotropic role that cardiolipin (and its precursor) can play in mitochondria (reviewed in reference 135). This situation makes it difficult to speculate on the petite-negative phenotype of the pel1/pgs1 mutants. However, an interesting approach was suggested by Hoffmann et al. (136). These authors showed that in vitro activity of a reconstituted aac2 mutant carrier specifically and absolutely requires cardiolipin: they proposed that this compound would be required in vivo for activity of the wild-type ADP-ATP carrier. If this idea was confirmed, it would provide a single and unified explanation for both the op1 and pel1 mysteries: a decrease in potential below the threshold required for cell viability in a petite background, due to either the absence or the inactivity of the major mitochondrial ADP-ATP translocator.

Besides ADH1, AAC2, and PEL1/PGS1, which appear totally essential in a petite background, other genes are almost essential when the rho⁺ genome is absent: mutations in these genes cause severely impaired growth in a petite context (Table 13). The most heavily documented situation concerns YME1 (for "yeast mitochondrial DNA escape"), but there are also a few reports concerning the F₁ sector of ATPase.

It was first reported by Ebner and Schatz (79) that a mutant strain lacking F₁ barely grew in a petite background. This observation has been strengthened by the following reports. A strain carrying a deletion of ATP3 (subunit ) grows very slowly when mtDNA is lacking (321). Strains with deletions of either ATP16 or ATP15, encoding subunits  and , respectively, also exhibit slow growth (107). The growth defect of atp3 and atp16 mutants was also noticed by Lai-Zang et al. (170). As discussed above (see "The translation-ATP synthase connection"), the data reported by these authors with respect to ATP1 (subunit ) and ATP2 (subunit ) deletions might be explained if these mutations were (almost) lethal in a petite background: the petite levels of atp1 and atp2 strains are below 1%, while the reference strain produces 9% petites (170). As a whole, the data strongly suggest that viability of the petites can be ensured only if enough ATP is hydrolyzed in mitochondria to provide the threshold of ATP import needed to generate the minimum membrane potential, which is in turn required for mitochondrial biogenesis and thus for cell viability. This idea links the tales of the op1 and pel1 mutants and the mutants impaired in F₁ ATPase.

Thorsness and Fox (300) were interested in the transfer of DNA from mitochondria to the nucleus, a phenomenon which most probably occurred during eucaryotic evolution. They therefore developed a genetic screen for mutations increasing the frequency of such events and obtained mutations in six genes. Mutations in YME1 are pleiotropic: they cause heat sensitivity under respiratory conditions, cold sensitivity on rich glucose medium (300, 301), altered morphology of the mitochondrial compartment (30) (see "Mitochondrial morphology, the cytoskeleton, and mtDNA" above), and severely impaired growth in a petite background (321). With respect to the latter phenotype, it was initially reported that yme1 cells were inviable in the absence of a rho⁺ genome (301). This was ascertained by several methods, including careful genetic tests. Spontaneous petite mutants (which appear at a frequency of 1% in the wild-type strain) were not found in the yme1 strains. These strains did not grow on medium containing ethidium bromide (which induces rho very efficiently). The yme1 mutations were colethal with a nuclear mutation which causes destabilization of mtDNA and viable in combination with another pet mutation which prevents respiration but does not destabilize mtDNA (301). It was later shown that yme1 rho (rho⁰) cells were able to form colonies if they were incubated at 30°C for 7 to 10 days instead of 3 to 4 days as in the original experiments (321). YME1 encodes an ATP- and zinc-dependent protease tightly associated with the mitochondrial inner membrane (320). Unfortunately, the precise functions of Yme1 are not yet well established, which does not aid us in understanding why yme1 petite cells are so sick. However, the search for specific suppressors of the slow-growth phenotype of these strains should shed light on this problem. This has been undertaken by Weber et al. (321), who identified six dominant suppressor mutations located in two genes. It is noteworthy that the major gene (five of six mutations) is ATP3, which encodes the  subunit of ATP synthase (see "The translation-ATP synthase connection" above) (Table 4) and that the suppressor mutations are missense mutations. This is a major discovery with respect to the general problem of petite negativity and its connection to ATPase, as emphasized in the next section.

In contrast to these mutants, others might survive only in a petite context. We describe here a mutant cured by a complete loss of mtDNA. The SSH1 (for "stress seventy subfamily H") gene encodes a protein of the Hsp70 family which is localized in the mitochondria (263). A strain bearing a deletion of SSH1 grows quite normally at 37°C but very slowly at 30°C and below. Overexpression of the SSC1 gene (encoding the major mitochondrial Hsp70) (see "Conventional mitochondrial chaperones" above) (Table 5) leads to partial suppression of ssh1 cold sensitivity. In contrast, overexpression of SSH1 cannot rescue the heat sensitive phenotype of a ssc1 mutant. Furthermore, and unlike Ssc1p, Ssh1p does not seem to be required for translocation of proteins into mitochondria. Spontaneous revertants of the ssh1 mutant, growth competent below 30°C, were obtained at high frequency (in the range of 1%). Convincing criteria demonstrated the presence of rho⁰ mutants. Among 15 revertants analyzed, only 1 showed mtDNA, in an amount fourfold below the wild-type level. These data were satisfyingly completed by a screening of the ssh1 mutant at 37°C (the permissive temperature) for clones which had become respiration deficient. Among 4,500 colonies, 6 showed this phenotype but only 1 was a suppressor of ssh1 at low temperature, and it was a petite of the rho⁰ type; the others, which could be siblings (not independent isolates), were nuclear pet mutants. Thus, it was impossible to obtain rho mutations in the ssh1 context. Does this mean that these petites are lethal in an ssh1 context at high (permissive) temperature? In this case, why does this nuclear background perform a selection of the rho⁰ state, especially at restrictive temperatures (263)? The authors propose two hypotheses. First, SSH1 might be involved in mtDNA replication, so that its absence (at low temperatures) would lead to a replication mode that is deleterious for the organelle and thus for the cell. Second, lack of mtDNA might affect the expression of nuclear genes able to suppress the cold sensitivity of the mutant. This hypothesis is reminiscent of the retrograde regulation already evoked above (see "Serependipitous discoveries") for the smp1 mutants (see reference 240 for a review). However, in the retrograde regulation system, both the rho and rho⁰ states lead to modifications in nuclear gene expression. With respect to the first hypothesis, it remains to be seen if the ssh1 context cannot tolerate a rho state at low temperature. This test could be performed through cytoduction experiments (at high temperature) between a ssh1 rho⁰ strain and a strain carrying a very stable rho genome, as in the procedure used by Jones and Fangman (149) in their analysis of the MGM1 gene (see "Mitochondrial morphology, the cytoskeleton, and mtDNA" above). There are in fact two questions. Is the ssh1 mutant able to maintain a rho genome at high temperature? Does the transfer at low temperature of the ssh1 cytoductants carrying the rho genome lead to the loss of mtDNA? In any case, the mutation causes a strong decrease in the petite level. In fact, in the ssh1 background, only 1 petite (rho⁰) was found among 4,500 colonies at 37°C. This is far below the spontaneous rate of petites normally observed in a wild-type context.

As to this decrease in petite production, there is to our knowledge only one report of a direct search for nuclear mutations producing this effect (Table 10). Devin and Koltovaya (66) began with a strain which produced 50% petite mutants and selected four nuclear mutations giving rise to a low level of petites (1% in three cases and 9% in the fourth). The four mutations are recessive and map to different complementation groups. As reported above (see "mtDNA and the cell cycle") (Table 10), a fifth mutation was shown to be allelic to CDC28 (67), while Koltovaya et al. (162) observed the same phenomenon with one of the "classical" cdc28 alleles. It is quite surprising that these mutations have not yet been introduced into a more "classical" background (with 1% spontaneous petites) to determine whether they are able to reduce the petite production and identify the relevant genes.

In contrast to S. cerevisiae, which spontaneously produces petites at a high rate, most yeast species are petite negatives: they do not give rise to petite cells even after treatment with ethidium bromide (which massively converts S. cerevisiae rho⁺ cells into rho⁰). This inability to lose mtDNA could easily be explained if these yeasts were strict aerobes. This is not the case, at least for S. pombe (reference 123 and references therein) and K. lactis (38), in which nonrespiratory mutants have been obtained. In fact, recent data on these petite-negative yeasts have provided a novel viewpoint on this problem. We will focus on K. lactis, for which the most documented information is so far available. It has been shown that large deletions (and even total loss) of mtDNA do occur but are lethal in the wild-type nuclear background (48). However, in strains carrying mutations in either one of several nuclear genes, named MGI (for "mitochondrial genome integrity"), petite production is observed spontaneously (in the range of 1 to 5%) and at high frequencies after treatment with ethidium bromide (38). Thus, mgi mutations allow petites to survive. It has been shown that MGI2, MGI1, and MGI5 encode, respectively, the , , and  subunits of mitochondrial ATPase. The mgi mutations are missense mutations in highly conserved positions. Null mutations in these genes abolish the petite-forming capacity, strongly suggesting that integrity of the F₁ complex is required for expression of the mgi phenotype. These authors propose the following model: the absence of F₀ subunits (encoded by the mitochondrial genome) in rho (rho⁰) cells in the presence of a wild-type F₁ would create a proton-permeable channel abrogating the membrane potential, thus killing the mitochondria and consequently the cell. Subtle and specific changes in , , or  subunits would cause the formation of an altered F₁ complex that was able to prevent the leakage of protons, allowing the survival of petites (37). This hypothesis has been rather complicated by recent data which show that cells lacking the mitochondrial ATP9 gene or disrupted for three nuclear genes (ATP4, ATP5, ATP7) encoding F₀ subunits are viable (39). The obvious conclusion is that lack of F₀ per se is not responsible for petite death. As emphasized by the authors, it remains to be determined if loss of ATP6 and/or ATP8 leads to cell death (39). If not, the mitochondrial component for the vital function of mtDNA will have to be identified, along with the role played by F₁.

As suggested by Chen and Clark-Walker (37), one can postulate that petite-positive yeasts are natural mgi mutants. For S. cerevisiae, the status of F₁ remains ambiguous: on one hand, loss of F₁ subunits strongly increases petite levels (see "The translation-ATP synthase connection" above and Table 4); on the other, the petites appear very ill when F₁ is lacking (see the previous section and Table 13). The most promising MGI candidate in S. cerevisiae remains YME1. As mentioned in the previous section, yme1 mutations are sublethal in the presence of rho (rho⁰) mutations. Extragenic suppressors of this defect have been obtained. Interestingly, five of the six correspond to mutations occurring in the ATP3 gene, encoding the ATPase  subunit. Two were sequenced and show missense mutations in strictly conserved positions (321). The remarkable fact is that one corresponds to the same substitution event as that observed in the mgi5-1 mutant of K. lactis. Deletion of the ATP3 gene in S. cerevisiae does not restore petite viability in the yme1 context: on the contrary, the double-mutant strains show a very slow growth on any media, probably because inactivation of ATP3 causes a high level of petites. It is too early to speculate on the role Yme1p plays in the viability of S. cerevisiae petites. What we presently know is that it is involved in a complex located in the mitochondrial inner membrane and that it displays a protease activity involved in the degradation of unassembled or abnormal mitochondrially synthesized proteins of the inner membrane (see reference 294 for a review). However, Atp3 is probably not a substrate for Yme1p (320). In any case, the near-essential role of YME1 in rho⁰ (rho) contexts and the fact that its absence is suppressed by a subtle change in ATPase  subunit are provocative. We are really not so far from the situation observed in K. lactis. It will be interesting to obtain more data on the status of the YME1 gene in K. lactis and the effects of the mgi2 ( subunit) and mgi1 ( subunit) missense mutations in S. cerevisiae as putative suppressors of the yme1 mutants.

Although these data have made a significant breach in the wall of an old problem, further interdisciplinary studies are still necessary to find a satisfactory answer. However, a unitary view of the problem in both S. cerevisiae (see the previous section) and K. lactis (this section) relies on the essential role of F₁ as an ATP hydrolase in petite cells. It is noteworthy that, beyond yeasts, functional F₁ is essential in maintaining the mitochondrial membrane potential and the growth of human cells devoid of mtDNA (26).

To our knowledge, there is as yet no satisfactory explanation for the high intrinsic instability of the mitochondrial genome of S. cerevisiae (and other yeasts as well; as reported in the previous section, some petite-negative yeasts can be converted into petite-positive forms which spontaneously produce 1 to 5% petite cells). There are many parameters which can influence the fate of mtDNA. Among the most obvious are the structure of the genome per se (for instance, mtDNA without introns appears more stable than intron-containing molecules), the high rate of recombination observed in mitochondria (and, consequently, a possible high probability of recombination between direct repeats), the selective advantages the defective mtDNA molecules may exhibit (in terms of replication and/or transmission), and the rapid loss of heteroplasmy due to the transmission of a small number of molecules (or a preferential transmission of related molecules) to the buds. One way to clarify this problem may be to study mutations which decrease (without abolishing) petite production (Table 10) (see the end of "No compromise" above). Another method is to screen for genes whose overexpression could decrease petite levels: this might identify genes whose products (normally synthesized in limited amounts) are required for mtDNA stability. One possible candidate is MSH1, which is assumed to protect mtDNA against recombination events between imperfect repeats (see "The recombination/repair track" above). This path will certainly be difficult to follow if several genes are normally expressed under the threshold required for an efficient mtDNA stability. Moreover, it is obviously easier to observe an increase than a decrease in petite production. However, even with respect to an increase in petite levels, a systematic study of gene overexpression is still lacking (Table 12) (see "Gene overexpression can increase mtDNA instability" above).

There are at least two possible answers to the second question. First, rho⁰ production may simply demonstrate that the mitochondrial compartment is unable to replicate any type of DNA. Thus, in the relevant genetic contexts, mtDNA replication is (directly or indirectly) abolished. The second possibility is based on the fact that some genetic backgrounds may select for a complete loss of mtDNA, as if any type of mtDNA molecules were deleterious to the cell in these contexts. The reasons why rho⁰ petites would be the only survivors in these situations remain unclear. This hypothesis has been proposed, for instance, to explain why the ssh1 mutants are able to grow at low temperature only if they are in the rho⁰ state (Table 13) (see "No compromise" above) and why an increased iron concentration might lead to rho⁰ production in the yfh1 mutants (Table 6) (see "Oxidative stress, iron homeostasis, and mtDNA above). However, the situation can be envisaged as more complex and very puzzling. In fact, some mutants which exhibit a complete loss of mtDNA (rho⁰ producers) are nonetheless able to maintain rho genomes. The relevant experimental tests have been performed in only a few cases (see "The ABF2 paradox" and "Does transcription play a direct role in mtDNA maintenance," above, for abf2 and rpo41 mutants, respectively).

The same sort of assumption can also explain the high production of petites (here rho and rho⁰) in a few situations. However, most situations would be more easily accounted for if one assumes that the wild-type (rho⁺) genome requires a specific structure to be consistently transmitted to the bud. Analysis of the sorting of mtDNA in zygotes also led Okamoto et al. (224a) to propose the existence of a segregation apparatus for the mitochondrial genome. Such a centromere-like structure has also been recently proposed for the bacterial chromosome (88). Concerning yeast mitochondria, the data suggest that this putative structure would be dispensable for rho molecules due to their abilities to use other routes of transmission. Presently, one can only emphasize some observations which focus on a possible attachment of the mitochondrial genome to the inner membrane. In fact, very few biochemical and cytological studies have been devoted to this problem. We would like to review two of these.

The first concerns specific structures of the mitochondrial membrane, seen in close association with the mitochondrial nucleoid. These myelin-like structures were clearly observed under nonrespiratory conditions (in the wild type or pet mutants grown on glucose-rich medium) but disappeared under respiratory conditions (332). It was not reported if these structures were present in petite mitochondria. However, the data led the author to suggest that these structures would be similar to those assumed, at that time, to link the bacterial chromosome to the membrane and be responsible for the distribution of daughter molecules at division (258a). Such ultrastructural analyses would be profitable in terms of the techniques available 30 years later. The second study concerns the isolation of mtDNA-protein complexes under different experimental conditions. The most interesting fact is the effect of RNase on the sedimentation pattern and composition of the nucleoids. The authors suggest an attachment of mtDNA to the inner membrane via nascent mRNAs in the course of their translation by membrane-bound ribosomes (251). In the intervening period, it has been shown that proteins encoded by mtDNA and devoted to the inner membrane are indeed synthesized on membrane-bound ribosomes. Specific proteins might ensure the link between mRNAs and the membrane (see reference 99 for a review). Since blocks in mitochondrial translation lead to conversion of rho⁺ cells to rho (Table 3) (see "The translation-ATP synthase connection" above), one could assume that translation per se is required for rho⁺ maintenance. However, the data suggest a more specific relationship between these two parameters, which focus on the Atp6 subunit. We have previously suggested experiments which might clarify whether Atp6 is required per se or must be translated in mitochondria (see "The translation-ATP synthase connection" above). In either case, it is clear that morphological alterations of mitochondria (Table 11) (see "Mitochondrial morphology, the cytoskeleton, and mtRNA" above) may influence the structure of the inner membrane and thus weaken mtDNA attachment. However, one cannot exclude a more indirect effect on mtDNA stability through specific impairments of the import process. Thus, analysis of this process and of the inner membrane structure in mdm/mgm conditional mutants (at subrestrictive temperatures) would be very informative. Notably, mtDNA instability is a postponed effect of morphological alterations, at least in cases where the relationship between the two defects has been examined (see "Mitochondrial morphology, the cytoskeleton, and mtDNA" above).

In the frame of the hypothesis of a centromere-like structure (by attachment of mtDNA to the inner membrane), one must keep in mind that some fatty acids might play a key role in anchoring the relevant proteins to the membrane, although we cannot exclude an indirect role for the lipid composition of the membranes on the import process (see "Fatty acids and phospholipids" above). Last but not least, there is the provocative discovery of Ymh2p. The YHM2 gene was discovered as a multicopy suppressor of abf2 null mutants. As reported above (see "The ABF2 paradox"), Ymh2p is localized to the inner membrane and associated with mtDNA in vivo. This is, in fact, the first case of a membrane-bound protein associated with the nucleoid (43). The confusing point of this problem is that disruption (with a partial deletion) of YMH2 does not increase petite production. However, the authors mention that ymh2 cells lose significant amounts of their mtDNA even when they are grown under respiratory conditions (this property was also reported for abf2 cells [335]). Clearly, further studies performed on strains bearing a complete deletion of the gene are required to fully understand the role played by Ymh2p.

The situation as a whole resembles a maze, whose branches include mitochondrial translation, the Atp6 subunit, fatty acids, mitochondrial morphology, and so on. Multidisciplinary approaches are necessary to know if this labyrinth is to emerge into either a centromere-like structure or another form liable to explain the requirements for rho⁺ maintenance.

Meanwhile, more classical approaches may give some insight into these questions. Unfortunately, there are very few model systems in this field. One must briefly mention the filamentous fungus Podospora anserina, which, although far behind S. cerevisiae, appears (in our opinion) the most advanced system with respect to the nuclear control of mtDNA stability. Its mitochondrial genome is nearly 100 kb and contains many introns (57). Thus, in this aspect, it resembles S. cerevisiae. However, it is an obligate aerobe. Furthermore, as a filamentous fungus, it is on the borderline between unicellular and multicellular eucaryotes. P. anserina displays a spontaneous degenerative process, called senescence (253). Short mtDNA sequences (senDNAs) are found amplified as circular multimeric molecules in senescent cultures (see references 15 and 75 for reviews). This phenomenon is under multigenic control (reference 15 and references therein). With respect to this control, there are two major differences between P. anserina and S. cerevisiae. First, while in P. anserina numerous mutations altering the cytosolic translational apparatus can modify the timing of cellular death and/or the nature of the relevant mtDNA rearrangements (14, 274, 275), there is no report of this type for S. cerevisiae petite production (with the possible exception of the mgm104-1 mutation [see comments in "Wide-scale screening for rho⁰ production" above]). Second, while blocks in mitochondrial translation increase mtDNA instability in yeast (see "The translation-ATP synthase connection" above), the opposite effect is observed in P. anserina: a slowing in mitochondrial translation delays senescence (13).

With respect to mtDNA maintenance, another degenerative process, called premature death, has been more recently discovered in P. anserina. It is linked to the accumulation of mtDNA molecules in which a specific part of the genome is deleted (14, 259, 269). It occurs only in the presence of either one of two (among five) mutations of the AS1 gene encoding a cytosolic ribosomal protein (14, 65). In some ways, premature death exhibits similarities to human diseases characterized by the accumulation of the so-called "common" deletion (174). This situation should also be observed in yeast. However, searches for a mutation leading to the accumulation of a specific rho deletion have so far probably been hampered by the high spontaneous rate of petites. In fact, the approaches used in yeast and in Podospora (with respect to premature death) have been quite different, while complementary. With a few exceptions (see, in particular, the early sections of this review), yeast researchers examine a posteriori the effects of mutations on petite production in terms of levels and petite types (rho and/or rho⁰). Our aim has been to find mutations which can delay and even abolish the accumulation of the deletions characteristic of premature death. To date, this has helped to identify eight genes (51, 52; D. Dequard-Chablat, personal communication), of which four have been cloned. Two encode proteins which show no similarity to known proteins and are thus absent in the yeast genome (V. Contamine, unpublished data; Dequard-Chablat, personal communication). Two others have yeast homologues (143). These are TOM70, encoding a component of the receptor for protein import into mitochondria (see references 191 and 236 for reviews), and MDM10, involved in mitochondrial morphology and distribution (283). Interestingly, mutations in the two P. anserina genes modify the spectrum of mtDNA rearrangements associated with cellular death and mitochondrial morphology. While effects of tom70 mutations on petite production in yeast have not been reported, mdm10 mutations increase petite production (Table 11) (see "Mitochondrial morphology, the cytoskeleton, and mtRNA"). This stresses the point that, although unclear (but see the previous section), the relationship between mitochondrial morphology and mtDNA maintenance has probably been conserved throughout evolution.

It may appear presumptuous and untimely to compare the well-documented situation in S. cerevisiae (which accumulates heterogeneous rho deleted genomes) and the underexploited situation of premature death in P. anserina (which is characterized by the accumulation of mtDNA molecules lacking a specific part of the genome). However, this emphasizes that a clear comprehension of the genetic control of mtDNA maintenance remains an important challenge for the future. We are very concerned about the interest the scientific community will invest in this field in the next century.