Abstract

   Eukaryotic cells can survive the loss of their mitochondrial genome,
   but consequently suffer from severe growth defects. Since the discovery
   of this phenotype in the 1940s, ‘petite yeast’ remained instrumental
   for studying mitochondrial function and physiology. However, the
   molecular cause of the slow-growth defect remained an open question.
   Here, we used adaptive laboratory evolution, and identified recurrent
   mutations within the ATP synthase gene ATP3 that restore the growth
   rate of petite Saccharomyces cerevisiae cells. Then, we use a
   combination of molecular genetics, and -omics approaches to pinpoint
   the metabolic limitations overcome by fast growing petites. Changes in
   ATP metabolism, mitochondrial morphology or retrograde signalling were
   detected in petites, but did not explain the growth recovery in evolved
   petites. Instead, we found that the synthesis of glutamate, glutamine,
   leucine and arginine is growth limiting in petites. The evolution of
   fast growth did overcome these amino acid deficiencies, by alleviating
   a defect in mitochondrial iron metabolism and by restoring a previously
   overlooked defect in the mitochondrial TCA cycle, caused by aconitase
   inhibition. Our results hence explain the slow growth of mitochondrial
   genome-deficient cells with a partial auxotrophy in four amino acids
   that results from distorted iron metabolism and an inhibited TCA cycle.

Introduction

   Mitochondria are key organelles in eukaryotic cells, and function in a
   range of crucial physiological processes, from metabolism, calcium and
   iron homeostasis, signalling, to the execution of apoptosis
   ^[62]1–[63]6 . In most eukaryotes, mitochondria maintain their own
   genome, a remnant of their likely α-proteobacterial origin ^[64]7–[65]9
   . The majority of genes encoding mitochondrial proteins have been
   transferred to the nucleus, but there remain important proteins encoded
   by mitochondrial DNA (mtDNA)^ [66]7,[67]8 . These are translated
   directly in the mitochondrion and are required for mitochondrial genome
   function, oxidative phosphorylation, and in turn, the TCA cycle
   ^[68]1,[69]7,[70]8 . Further, the respiratory chain establishes a
   chemiosmotic proton gradient at the inner mitochondrial membrane, which
   drives mitochondrial transport processes, and oxidative phosphorylation
   ^[71]10,[72]11 .

   Some organisms with a mostly parasitic lifestyle have lost the
   mitochondrial genome and evolved without it ^[73]12–[74]14 . Indeed,
   mtDNA is not under all circumstances essential for basic eukaryotic
   cell functionality. Spontaneous mitochondrial genome loss can be
   observed across multiple cell types and species, including human cell
   lines and yeasts ^[75]15,[76]16 . However, cells that tolerate the
   spontaneous loss of the mtDNA grow slowly. In yeast, this phenotype has
   been described more than 70 years ago, and named the “petite”
   phenotype, as these slow growing yeast cells form small colonies
   ^[77]17 . Petites are commonly referred to as ρ^0 (opposed to wild type
   ρ^+ cells), and to our knowledge, constitute the earliest report of a
   defect directly associated with mitochondria. Accordingly, petites
   became a frequently used tool to study mitochondrial function. Some
   discoveries made with petite cells were of historical importance. For
   example, the petite phenotype led to the discovery of the mitochondrial
   genome itself ^[78]9 . As the research on mitochondrial function is
   still hindered by difficulties in the manipulation of the sequence of
   mtDNA ^[79]18 , petites remain an important model to map the associated
   spectrum of metabolic and physiological roles of mitochondria, such as
   ATP production through oxidative phosphorylation
   ^[80]9,[81]17,[82]19–[83]32 .

   The growth defect of petites can be compensated by secondary suppressor
   mutations. For instance, petite suppressors are common in the
   laboratory S. cerevisiae strain S288C ^[84]33 which spontaneously loses
   mtDNA, a property known as a high petite frequency ^[85]34 . Petite
   suppressor mutations frequently map to the F1 subunit of the ATPase
   (complex V) of the respiratory chain ^[86]21,[87]32,[88]35–[89]39 .
   Other mutations that affect the F1 subunit of the ATPase have been
   associated with thermotolerance ^[90]40 , or enable species such as K.
   lactis and T. brucei where mtDNA is otherwise essential, to survive in
   its absence ^[91]35,[92]41,[93]42 .

   Despite knowing that the petite growth establishes as a direct
   consequence of a lack of the mitochondrial genome, and that the defect
   can be overcome by mutations in the nuclear encoded subunits of the ATP
   synthase, it remained unclear which metabolic constraints are causing
   the slow growth phenotype. We speculated that elucidating the molecular
   changes observed in evolved petites would ultimately guide us to the
   underlying problem that restricts growth in the mitochondrially
   deficient cells. We first conducted an adaptive evolution experiment to
   select for S. cerevisiae petite mutants that effectively suppress their
   slow-growth phenotype. Petites adapted to near wild type growth rates
   despite the absence of mtDNA, by obtaining a set of rapidly and
   independently occurring, dominant mutations in ATP3 gene, that are
   close proximity to previously described alleles ^[94]21,[95]35–[96]39 .
   We then used the newly isolated suppressors as a tool to identify the
   metabolic bottlenecks that slow growth. Many cellular processes
   associated with the function of mitochondria, including redox
   metabolism, nutrient- or retrograde-signalling, were found to be
   altered in petites.

   However, they were not causative to their slow-growth phenotype.
   Instead, we find that petites suffer from an insufficient capacity in
   the biosynthesis of the four amino acids, glutamate, glutamine,
   leucine, and arginine. We show that petite cells overcome this
   multiple, partial auxotrophy by restoring defects in mitochondrial iron
   metabolism and a so far overlooked block in the TCA cycle at the step
   of aconitase.

Results

Adaptive evolution selects for a specific hotspot of mutations in ATP3 that
overcome the petite phenotype

   To test which mutations best suppress the petite phenotype, we designed
   an adaptive evolution experiment. Using the method by Goldring et al
   ^[97]43 , we generated mtDNA-deficient (ρ^0) petite progenitors from a
   diploid, prototrophic Saccharomyces cerevisiae strain (YSBN1) ^[98]44 .
   We selected clones that formed characteristic petite colonies, and
   confirmed loss of mtDNA ([99]Fig. 1a, b). Petite and wild type control
   cultures were then grown in a chemostat under glucose-limiting
   conditions in a highly nutrient restricted minimal (F1) medium ^[100]44
   . The cultures were kept under static nutritional conditions for 65
   days, during which approximately 352 biomass doublings occurred
   ([101]Suppl. Fig. 1). Samples were extracted at regular intervals over
   time, and were genotyped and phenotyped. Large colonies indicative of
   improved growth in petite cultures were readily apparent at the second
   sampling point (65 biomass doublings) and became more abundant with
   time. After 162 doublings, large colonies dominated all three parallel
   chemostats ([102]Fig. 1b). Growth of wild type cells cultivated as
   controls in parallel chemostats remained unaffected ([103]Suppl. Fig.
   2).

Figure 1. Adaptive laboratory evolution selects for ATP3 variants that
overcome the slow-growth phenotype of petites.

   [104]Figure 1
   [105]Open in a new tab

   A. Petites lost their mitochondrial genome. Wild type (ρ^+) yeasts
   (YSBN1) and respective progenitor strains that lost their mitochondrial
   genome following ethidium bromide treatment were analyzed using DAPI
   staining (3 replicates each, c1 - c3). While wild type yeast contains
   dot-like structures indicative of mtDNA, petite (ρ^0) yeast only
   display nuclear DNA staining. Representative images are shown. Scale
   bar = 1 μm.

   B. Evolved petites with improved growth emerge during adaptive
   laboratory evolution. Petite YSBN11 strains (ρ^0) were grown in
   F1-media in a chemostat under glucose-limiting conditions. Samples were
   taken from the inoculum, after 65, 162, as well as 260 generations, and
   spotted on F1 agar plates. After 260 generations, almost all cultures
   had regained the capacity to form larger colonies. Shown is one
   representative of n = 3 biological replicates.

   C. ATP3 substitutions emerge and their prevalence increases during
   adaptive laboratory evolution. Relative frequency of the three most
   abundant heterozygous ATP3mutations (ATP3-6, ATP3-7 and ATP3-8) in
   chemostat cultures was estimated from whole-population sequencing.
   Prevalence of these SNPs increased during the evolution experiment, and
   were recovered from 79.2 % of the isolates from the selected evolved
   population.

   D. Molecular context of isolated suppressor mutations in the yeast
   F1c10 ATPase complex. Detailed view of F1c10 ATPase complex structure
   with α (green), β (turquoise) and γ (grey) subunit interface.
   Identified and confirmed mutations (shown in [106]Suppl. Fig. 3) in the
   γ subunit are highlighted in red. The small inlet structure shows the
   entire yeast F1c10 ATPase complex (PDB ID 2XOK ^[107]118 ) with
   highlighted γ subunit (dark grey) and mutations in C-terminal region
   (red).

   E. Petite isolates from evolved populations that carry ATP3 mutations
   display elevated growth rates. Fast-growing colonies from evolved
   populations were genotyped for ATP3, and growth rates were recorded in
   F1 medium with 0.5 % glucose. The petite ancestor strain exhibits a
   significantly slower growth rate, compared to colonies carrying
   variants of ATP3-6, ATP3-7 or ATP3-8. Statistics based on unpaired
   t-tests comparing naive petite (ρ^0) with other indicated genotypes. *p
   < 0.05; Shown are mean values ± SD of n > 4 biological replicates per
   genotype.

   F, G. Heterologous expression of identified ATP3 substitutions
   increases growth in naive petites. YSBN11 strains were transformed with
   pRS313 plasmids carrying indicated ATP3 variants, and mtDNA was
   depleted. Growth in F1 medium containing 0.5 % glucose was monitored.
   (G) Expression of ATP3 variants significantly improves growth compared
   to vector controls, or additional expression of the wild type allele.
   (F) Petite (ρ^0) cells expressing ATP3 variants display growth
   advantages. Statistics based on two-sided, unpaired t-tests compared
   with other genotypes. *p < 0.05. Mean values ± SD of n = 3 biological
   replicates per genotype are shown.

   To identify the genetic alterations that had occurred, we performed
   whole-genome resequencing at the population level ([108]Suppl. Fig. 1,
   sequenced at generations 0, 260, 340 for all replicates, and
   additionally generation 65 and 162 for replicate 3). In order to rule
   out secondary (nuclear) mutations caused by mtDNA depletion, we
   additionally sequenced respective progenitor strains ([109]Suppl. Table
   1). Minority allele detection analysis revealed a surprisingly specific
   set of 21 non-synonymous and 30 synonymous mutations which were present
   in the evolved petites, but not in the ancestral, naive petites, nor in
   the wild type parent ([110]Suppl. Table 1). Mutations only within one
   gene (ATP3) enriched over time. Moreover ATP3 mutations were detected
   in each of the independent evolution experiments ([111]Fig. 1c,
   [112]Suppl. Table 1).

   We then randomly selected fast-growing colonies of one evolved
   population (replicate 3), and re-sequenced the ATP3 locus. This
   analysis confirmed the three major ATP3 variants as detected by the
   population sequencing, and revealed four additional ATP3 mutations with
   lower prevalence ([113]Suppl. Fig. 3). Each clone contained exclusively
   one of the identified ATP3 mutations ([114]Suppl. Fig. 3), indicating
   that even within one chemostat culture, multiple ATP3 mutations emerged
   independently. At the final time-point the majority of cells contained
   ATP3 mutations ([115]Fig. 1c, [116]Suppl. Fig. 3, [117]4).

   Most uncovered mutations were located at the C-terminal domain of ATP3,
   and were closely located to previously isolated petite suppressors
   ^[118]32,[119]36,[120]42,[121]45–[122]47 . This included two mutations
   that did match previously identified residues associated with
   slow-growth suppression in a petite S. cerevisiaeyme1 mutant
   ([123]Suppl. Fig. 4a,[124]b) ^[125]36 . All variants affected conserved
   residues. Mapping the identified mutations into the structure of
   complex V highlights a hotspot in a narrow region close to the
   C-terminus of the ɣ subunit ([126]Fig. 1d, [127]Suppl. Fig. 4).

   Next, we tested whether the emergence of the ATP3 alleles was causal
   for the improved growth. We measured growth rate in isolates harbouring
   ATP3 mutations. Their growth rate was significantly increased compared
   to the ancestral petite cultures ([128]Fig. 1e). Then, we cloned the
   two most frequent alleles, ATP3-6 and ATP3-7 ([129]Suppl. Fig. 4a) and
   reintroduced them on a plasmid into the parental wild type background
   and naïve (non-evolved) petite cells. In parallel, we constructed an
   artificial variant which contained the two most frequent mutations in a
   single gene (named ATP3-6,7). The expression of mutant alleles ATP3-6,
   ATP3-7, as well as the artificial allele carrying both mutations
   (ATP3-6,7) restored growth of petites, suggesting a dominant mode of
   inheritance in petites ([130]Fig. 1f, g). In contrast, the plasmid
   control and an additional copy of ATP3 did not compensate for the
   growth defect petites ([131]Fig. 1f, g) nor did their expression affect
   wild type cells ([132]Suppl. Fig. 5). Taken together, a panel of
   mutations within a hotspot region of ATP3 convey a strong fitness
   advantage and suppress the slow-growth phenotype for cells without
   mtDNA. Notably, we isolated exclusively ATP3 alleles as suppressors of
   petite growth during the adaptive evolution experiment that was run in
   replicates, with a strong selection pressure to optimize growth in
   minimal F1 medium, over hundreds of generations, and sequenced with
   high coverage. We hence conclude that ATP3 mutations are the most
   efficient solution to overcome petite growth, at least in the given
   genetic background and condition.

Improved growth of evolved petites correlates with an increase in the
mitochondrial membrane potential ΔΨ[mito]

   Strains suppressing the petite phenotype have an increased
   mitochondrial membrane potential (ΔΨ[mito]) compared to naive petites
   ^[133]29,[134]36,[135]48,[136]49 . Consequently, we examined changes in
   ΔΨ[mito] in strains carrying newly identified ATP3 alleles, and tested,
   to which extent these relate to the changes in growth rate. Using the
   MitoLoc reporter assay, we monitored differences in ΔΨ[mito], as well
   as mitochondrial morphology between naive and evolved petites, as well
   as in “reconstituted” evolved petites heterologously expressing the
   ATP3 alleles ^[137]50 . Wild type cells showed colocalization of the
   MitoLoc markers consistent with a functional ΔΨ[mito] ^[138]50 .
   Colocalization was lost in petites, indicating depolarisation
   ([139]Fig. 2a, left panels). Expression of the petite-suppressing ATP3
   alleles restored mitochondrial colocalization of the MitoLoc markers,
   indicating an improvement of ΔΨ[mito] ([140]Fig. 2a, right panels).
   These results were corroborated by a quantitative colocalization
   analysis ^[141]50 ([142]Fig. 2b), and through staining with the
   ΔΨ[mito]-dependent dyeDiOC[6] ([143]Fig. 2c, [144]Suppl. Fig. 6).
   Similarly, ATP3-mutants isolated from the evolution experiments
   revealed an increase in ΔΨ[mito], compared to the petite controls
   ([145]Suppl. Fig. 7a). Overall, we found that the increase in ΔΨ[mito]
   ([146]Fig. 2b, [147]Suppl. Fig 7a) correlated with the gain in growth
   rate (R^2 = 0.93, [148]Suppl. Fig. 7b), suggesting a link to the
   restored growth phenotype of evolved petites ([149]Fig. 1e, f).

Figure 2. Suppressor mutations increase the mitochondrial membrane potential,
but do not restore mitochondrial morphology or metabolic capacity in petites.

   [150]Figure 2
   [151]Open in a new tab

   A - C. The mitochondrial membrane potential is compromised in petites,
   but improves upon expression of ATP3-6 and ATP3-7. (A) Indicated YSBN11
   strains were transformed with pMitoLoc^ [152]50 , a reporter construct
   containing two fluorescent proteins with ΔΨ^mito-dependent
   (preCOX4-mCherry) or ΔΨ^mito-independent (mtGFP) mitochondrial import.
   Naive petites accumulate preCOX4-mCherry in the cytosol and nucleus,
   indicative of reduced ΔΨ^mito. In contrast, in wild type (ρ^+) and
   petites (ρ^0) expressing ATP3 variants both markers colocalize,
   indicating ΔΨ^mito polarization. Scale bar = 1 μm. (B) Quantitative
   analysis of ΔΨ^mito estimated from Pearson Correlation Coefficients
   (PCC) of pixel-by-pixel protein colocalization as shown in (A). Loss of
   mtDNA leads to a significant decrease of ΔΨ^mito (61 %) compared to the
   wild type, which is significantly rescued by presence of
   plasmid-encoded ATP3 variants (up to 77 %). *p < 0.05, n > 46 cells,
   mean values +/- SEM. (C) Analysis of ΔΨ^mitoin the strains described in
   (A) as estimated by DiOC[6] staining followed by quantification with
   flow cytometry of n = 19956 (ρ^+), n = 16104 (ρ^0), n = 19112 (ρ^0
   ATP3-6), n = 18893 (ρ^0 ATP3-7) events.

   D. Mitochondrial ATPase activity is reduced in strains without mtDNA.
   Mitochondrial proteins of YSBN 11 strains with indicated genotypes were
   isolated and ATPase activity was determined. Petites suffer from a
   significant reduction of the oligomycin-insensitive ATPase activity
   compared to wild type controls. Expression of variant ATP3 alleles did
   not rescue ATPase activity. Mean values of n = 5 biological replicates
   +/- SEM. Statistics is based on two-sided, unpaired t-tests comparing
   wild type against petite and evolved petite strains. *p < 0.05.

   E.-H. The metabolic capacity to metabolise glucose is unaffected in
   evolved petites. Indicated YSBN11 strains were cultured in batch
   fermenters for 45 hours in F1 medium containing 2 % glucose, and
   biomass as well as glucose, glycerol and ethanol concentrations were
   monitored. Wild type (ρ^+) cultures reach stationary phase after 13.9
   h, petites (ρ^0) after 20 h, and petites expressing ATP3-6 and ATP3-7
   variants after 15.9 h (dashed lines). (E) Biomass as estimated from
   OD[600] measurements over time. Wild type cultures accumulate 10.2
   OD[600] units of biomass, while petite (6.1 OD[600] units) and evolved
   petites (6.1 and 6.4 OD[600] units) accumulate less biomass. After the
   diauxic shift, only wild type cultures continue to accumulate biomass.
   (F) Glucose levels in culture media. Note that glucose is entirely
   consumed during exponential growth. (G) Ethanol and levels in culture
   media. All strains produce similar levels of ethanol (up to a
   concentration of 7.7 - 8.0 g/L), while petites are unable to consume
   ethanol subsequent to glucose exhaustion. (H) Glycerol levels in
   culture media. Wild-type produces 0.78 g/L glycerol during exponential
   growth, which is subsequently consumed. Instead, petites produce higher
   levels of glycerol (1.7-1.9 g/L) during exponential phase, but are
   unable to consume glycerol after glucose exhaustion.

   While mitochondria in unstressed wild-type yeast cells are highly
   connected, petites exhibit a fragmented mitochondrial network,
   indicative of mitochondrial fission as it is induced by depolarisation
   of the mitochondrial membrane potential ^[153]50,[154]51 . Monitoring
   mitochondrial morphology using super resolution microscopy, we tested
   whether mitochondrial fission in petites is explained by their
   compromised ΔΨ[mito]. However, the mitochondrial network of petites
   carrying the ATP3 mutations displayed a similar degree of fragmentation
   compared to naive petite cells ([155]Fig. 2a, [156]Suppl. Fig. 8),
   indicating that depolarization of the mitochondrial membrane potential
   is not the only cause of mitochondrial fission in petites.

Evolved petites grow faster than naive petites but form a similar amount of
biomass

   A primary function of the respiratory chain, oxidative ATP
   biosynthesis, is interrupted in petites. Moreover, it has been
   hypothesized that in petites, the remaining and assembled F1 subunit of
   the ATP synthase ^[157]52 could be running in “reverse” to hydrolyse
   mitochondrially imported ATP, and in this way restore ΔΨ[mito] via the
   electrogenic exchange of mitochondrial ADP^3- with cytosolic ATP^
   [158]4–[159]38,[160]41,[161]53–[162]57 . In both scenarios, a decreased
   ATP production could become growth limiting ^[163]31 . However, we
   found that the adenylate energy charge ^[164]58 was unaffected in
   petites and evolved petites compared to wild type cells ([165]Suppl.
   Fig. 9). ATP levels in petites only collapsed after glucose exhaustion
   ([166]Suppl. Fig. 10), when due to the absence of the respiratory
   chain, the petites can not consume the remaining non-fermentable carbon
   substrates, i.e. ethanol and glycerol ^[167]59 . Further,
   quantification of the Complex V ATPase activity revealed that it was
   strongly reduced rather than increased in petites and evolved petites
   compared to wild type ([168]Fig. 2d). Hence, we concluded that the
   evolution of fast growth in petites was not achieved through altering
   ATP metabolism, nor through a futile metabolic cycle in which increased
   ATP hydrolysis would restore the ΔΨ[mito.].

   Next, we performed fermentation experiments to determine the efficiency
   of petite and evolved petite cells in converting the glucose into
   biomass. We determined carbon balances of wild type, petite, and
   evolved petites by measuring glucose uptake, ethanol and glycerol
   production, CO[2] and O[2] levels in the bioreactor, as well as the
   obtained biomass during growth on glucose ([169]Fig 2 e-h, [170]Suppl.
   Fig. 11). Further, we used the obtained data on glucose consumption,
   ethanol and glycerol production as well as obtained biomass in a model
   to estimate physiological parameters such as yield coefficients, growth
   rate, and uptake/excretion rates during the exponential phase of all
   strains (first 10.5 hours, [171]Suppl. Table 2). Wild type yeast
   consumed glucose and entered diauxic shift after 13.9 hours, having
   accumulated 168 % of the biomass generated by petites ([172]Fig. 2e).
   Petite cells reached glucose exhaustion after 20.0 hours ([173]Fig.
   2f), reflecting their slower growth rate. Strainscarrying ATP3-6 or
   ATP3-7 reached stationary phase and glucose exhaustion considerably
   faster than petites (15.9 hours; [174]Fig. 2e, f). Despite accelerating
   growth, the ATP3 adaptationsdid not largely affect the biomass yield,
   i.e. the biomass produced per glucose consumed during exponential phase
   ([175]Suppl. Table 2), or the biomass obtained until the diauxic shift
   ([176]Fig. 2e), nor the ability of the petite cells to consume
   non-fermentable carbon sources ([177]Fig. 2g,h, [178]Suppl. Table 2).
   Petite suppression does hence not affect the general metabolic
   properties of petites. Instead, these results indicated that growth of
   petites is compromised because biosynthetic pathways do not meet the
   required rates to support faster growth.

Insufficient supply of four amino acids limits growth of cells without mtDNA

   In order to pinpoint the biochemical processes affected in petite
   cells, we performed untargeted proteomics using microLC-SWATH-MS, a
   proteomic method designed for obtaining precise enzyme quantities in
   large sample series ^[179]60 . The proteomic data indicated that
   petites suffer from a broad range of metabolic perturbations that
   revolve around central carbon metabolism, the tricarboxylic acid (TCA)
   cycle in particular, and amino acid metabolism ([180]Fig. 3a and
   [181]Fig. 3b, squares). Although a contribution of the TCA cycle was
   expected due to the loss of crucial components of the respiratory
   chain, the degree of perturbed amino acid biosynthesis pathways was
   somewhat surprising. Curiously, recent findings from another yeast,
   Schizosaccharomyces pombe, which is normally petite negative, indicated
   that that supplementation with amino acids derived from α-ketoglutarate
   supported growth upon pharmacological inhibition of the respiratory
   chain ^[182]61 , supporting the hypothesis that in respiratory
   deficient cells, metabolism of α-ketoglutarate and downstream amino
   acids are challenged ^[183]62 .

Figure 3. Amino acid metabolism is perturbed in petites, and their growth
rate defect is rescued by the addition of selected amino acids.

   [184]Figure 3
   [185]Open in a new tab

   A. Pathway enrichment analysis of proteomics data indicates metabolic
   defects associated with petites. Differentially expressed proteins
   between wild type (ρ^+) and naive petite (ρ^0) YSBN1 yeast were
   determined using SWATH-MS and subjected to pathway enrichment analysis
   using String-db. 11 Reactome pathways were significantly enriched (FDR
   < 5 %), with most significantly enriched pathways being biological
   functions related to TCA and amino acid synthesis. Enrichment is shown
   as bar size (-log10 enrichment p-value) and bar colors indicating
   enrichment over background frequency (number of changed proteins
   compared to all proteins of that pathway), and 9 TCA pathway proteins
   (enrichment = 0.56) were changed in naive petites. Protein abundance
   data from n = 4 biological replicates per genotype. Note that the
   p-values stated in the corresponding barchart represent non-log
   transformed values.

   B. Differential expression of proteins and metabolites of naive and
   evolved petites compared to wild type. A simplified pathway
   representation of central carbon metabolism in S. cerevisiae. Mean
   fold-changes of enzyme (squares) and metabolite levels (circles) of
   petite (middle column, empty plasmid control) and evolved petites
   (right column, expressing ATP3-6) against wild type (left column, empty
   plasmid control) are shown. Data shown is mean fold change from n = 4
   biological replicates per genotype.

   C-E. The slow growth of petite cells is rescued by a supplementation
   with glutamate, glutamine, arginine and leucine. (C) Growth rate of
   YSBN11 wild type (containing empty vector control), naive petites
   (containing empty vector control) and evolved petites (containing
   ATP3-6) grown in minimal media with 2 % glucose in 96-well liquid
   culture, without (-AA) or with addition with a complement mixture of
   proteinogenic amino acids (+AA), or combinations of 2 mM glutamate (E),
   2 mM glutamine (Q), 2 mM arginine (R) and 2 mM leucine (L). The
   slow-growth phenotype (0.075 OD min^-1) of petites increases to 0.137
   OD min^-1 upon supplementation with QERL. Mean values +/- SD of n = 5
   biological replicates. (D) The relative growth defect of petite cells
   (normalized to percentage of wild type growth rate) is compensated from
   43.3 % to 13.5 % upon QERL supplementation, as calculated from the data
   as shown in (A). Mean values +/- SD of n = 5 biological replicates. (E)
   Indicated strains were pre-grown in minimal media with (+QERL) or
   without supplements (-AA). A serial dilution series (1:5) was spotted
   with an initial OD[600] of 0.5 on agar plates with the indicated media
   composition. Shown is one of n = 2 biological replicates.

   We hence continued with a quantification of amino acids and respective
   precursors, which confirmed perturbation of intermediary metabolism in
   petites ([186]Fig. 3b. circles). Petites were characterized by strongly
   reduced concentrations of leucine, arginine (as well as its precursors
   citrulline, ornithine and α-ketoglutarate). In parallel, enzymes of the
   associated synthesis pathways were differentially expressed in petites.
   Upon expression of the mutant ATP3 alleles, these expression changes
   were partially alleviated ([187]Fig. 3b squares, [188]Suppl. Fig. 12).
   Curiously, the amino acid levels did not follow the expression changes,
   and remained low in evolved petites compared to wild-type cells
   ([189]Fig. 3b, circles). We speculated that this result could indicate
   that the demand for these metabolites was indeed limiting.

   We therefore conducted supplementation experiments with combinations of
   leucine, arginine, as well as glutamate and glutamine, which both are
   directly associated with α-ketoglutarate metabolism. Supplementation of
   these amino acids, in separation, and specifically in their
   combination, lead to a disproportionate increase in the growth rate of
   petites compared to how wild type or evolved petites responded to
   supplementation ([190]Fig. 3c, [191]Suppl. Fig. 13). Consistent with
   previous observations, leucine shortages are particularly growth
   limiting due to the high biosynthetic requirements for this amino acid
   ^[192]63–[193]65 , compensating with leucine had a particularly strong
   impact on the petite growth rate ([194]Fig. 3c). The supplementation of
   arginine, which has also been described as beneficial to petites in
   another study ^[195]66 , but also the combination of glutamine and
   glutamate benefited petite growth clearly more than it benefited wild
   type or evolved petites. When we supplemented all four amino acids
   (QERL) at once, the growth rate of petite cells became similar to the
   growth rate of wild-type cells without amino acids (AA) supplementation
   ([196]Fig. 3c, d, e). Supplementation with all proteinogenic AA also
   accelerated growth, but the complex treatment was less effective than
   the addition of just QERL alone ([197]Fig. 3c). The lower efficacy of
   the broad rather than the specific supplementation is potentially
   caused by multiple amino acids competing for the same transporters
   ^[198]67 , or by substrate-induced down-regulation of amino acid
   transport in more complex media ^[199]68–[200]71 . Further, we tested
   if improved growth by QERL supplementation coincides with an increase
   in ΔΨ[mito], that is known to depend on growth rate. Indeed, the
   membrane potential was generally higher in the fast growing cells.
   However, differences between wild type, petite, and evolved petite
   overall remained ([201]Suppl. Fig. 14). Taken together, these data
   suggest that insufficient rates in the synthesis of specific amino
   acids, especially of leucine, arginine, glutamine and glutamate, is
   limiting the growth rate of petites.

Perturbations of NADPH metabolism, TOR signalling, and the retrograde
response are not causal for the impaired amino acid biosynthesis in petites

   The observation that multiple amino acid biosynthesis pathways were
   affected in parallel indicated that regulatory processes, the
   cellular-chemical environment, or common metabolic precursors, or
   combinations of these, were limiting the growth in petite cells. First,
   we tested if an amino acid deficiency is associated with a lack of
   reducing equivalents in the form of NADPH, which is required for amino
   acid anabolism ^[202]72,[203]73 . We exploited the phenotype that yeast
   cells are sensitive to the synthetic thiol oxidant diamide when NADPH
   supply is limited, as NADPH is also providing reducing power for the
   antioxidant machinery ^[204]73–[205]76 . We did not detect strong
   differences in petites in their tolerance to diamide ([206]Suppl. Fig.
   15a). Further, we conducted measurements of the glutathione (GSH)
   pools, that depend on NADPH availability. GSH pools were comparable
   between wild type and petite strains ([207]Suppl. Fig. 15b–[208]d).
   Finally, we measured superoxide levels. These were marginally changed,
   and indeed, we detected lower superoxide levels in petites than in
   wild-type cells ([209]Suppl. Fig. 15e). Taken together, these results
   argue against a significant NADPH shortage or strong oxidative
   imbalances as a cause of the amino acid deficiency of petite cells.

   Next, we focussed on amino acid sensing and signalling. We studied the
   response of petites and evolved petites to TOR inhibition by rapamycin
   ^[210]77 . A dose response curve revealed that the chemical sensitivity
   (IC50) to rapamycin does not differ between cells with or without mtDNA
   ([211]Suppl. Fig 16).

   We then investigated the role of mitochondrial retrograde (RTG)
   signalling, which has been attributed to mitochondrial dysfunction
   ^[212]1 . In addition to regulating the expression of mitochondrial
   genes such as the TCA cycle, the RTG response has been implicated in
   regulation of peroxisomal processes ^[213]2 . As peroxisomal
   proliferation has hence been associated with mitochondrial deficiency
   ^[214]78 , we used a peroxisomal fluorescent marker (Pts1p-GFP) to
   quantify these organelles. The number of peroxisomes was significantly
   increased in petites compared to the wild type (18.5 ± 1.0 vs. 13.0 ±
   0.6 in WT; p < 0.0001). Moreover, the number of peroxisomes returned
   close to wild type levels in evolved petites (11.2 ± 0.6; p = 0.05)
   ([215]Suppl. Fig. 17a–[216]b). Similarly, differential expression of
   proteins associated with the RTG regulon, Aco1p, Dld3p, Idh1p, Idh2p
   and Pyc1p ^[217]1,[218]79 was observed in petites, and this response
   was alleviated in evolved petites([219]Suppl. Fig. 17c).These results
   indicated that the RTG response pathway is activated in petites, but
   not in evolved petites. To test whether suppression of retrograde
   signalling was causally associated with the amino acid deficiency, we
   deleted RTG2, the central mediator of the retrograde response, and
   CIT2, a target of a downstream regulated process, the glyoxylate cycle
   ^[220]1 . Deletion of RTG2 or CIT2 did not restore petite growth,
   instead, the growth defect in petites increased. Moreover, rtg2 and
   cit2 mutants did not affect the ability of ATP3 mutants to suppress the
   petite phenotype ([221]Suppl. Fig. 17d). These results suggest that
   activation of the retrograde response pathway is a protective response
   for cells without mtDNA, but not causing the petite phenotype.

Two enzymatic steps of the TCA cycle are inhibited in petite cells

   The retrograde signalling is typically associated with the synthesis of
   α-ketoglutarate and downstream amino acids ^[222]62 . We therefore
   focused on the TCA cycle, of which α-ketoglutarate is a key
   intermediate. Quantification of both proteins and metabolites involved
   in TCA cycle reactions revealed that the upper TCA cycle is severely
   affected in petites compared to wild type cells ([223]Fig. 4a, b,
   [224]Suppl. Fig. 18, 19). Pathway intermediates including citrate,
   aconitate and succinate accumulated, whereas the concentration of
   α-ketoglutarate was reduced ([225]Fig. 4b).

Figure 4. Aconitase is inhibited inpetites and blocks the TCA cycle, which is
alleviated in evolved petites.

   [226]Figure 4
   [227]Open in a new tab

   A-B. Differentially expressed proteins and metabolites of the TCA
   cycle. A simplified pathway representation of central carbon metabolism
   in S. cerevisiae. Shaded areas indicate respective cellular
   localization (mitochondrial, peroxisomal, cytosolic) labelled according
   to the small inlet (top right). (A) Enzymes (squares) are indicated
   next to associated reactions (arrows). Mean fold-changes of protein
   levels (squares) of wild type YSBN11 (left column, plasmid control)
   compared to petites (middle column, plasmid control) and an evolved
   petite strain (right column, expressing plasmid based ATP3-6) are
   shown. Note that electron transport chain and succinate dehydrogenase
   are impaired in petites ^[228]1,[229]7,[230]8 . Associated enzymes are
   accordingly dysregulated, and not alleviated in evolved petites.
   Instead, the reactions from citrate to α-ketoglutarate are
   differentially regulated in petites, but partially restored in evolved
   petites. Shown are mean values of n = 4 biological replicates per
   genotype. (B) Mean fold-changes of metabolite steady-state levels
   (circles) indicated next to the associated reaction in the pathway.
   Genotypes and data normalization like in (A). Metabolite levels
   indicate two blocks in the TCA cycle at the steps of aconitase and
   succinate dehydrogenase. Shown are mean values of n = 3 biological
   replicates per genotype.

   C. Aconitase and isocitrate dehydrogenase protein levels are increased
   in petites. Protein expression illustrated relative to the wild type
   (ρ^+) strain highlight upregulation of aconitase and
   isocitrate-dehydrogenase in petites. Levels of the major NAD- and
   NADP-dependent isocitrate dehydrogenase (Idh2p/Idp1p respectively), as
   well as the main mitochondrial aconitase (Aco1p) are upregulated in
   petites. Impairments are partially alleviated in an evolved petite
   strain. Mean values +/- SD of n = 3 biological replicates. Statistics
   based on two-sided, unpaired t-tests comparing wild type to other
   genotypes. Note that the data is from the same experiment as shown in
   (A).

   D. Aconitase is inhibited, while (NAD dependent)
   isocitrate-dehydrogenase activity is increased in petites. Activity of
   the indicated enzymes was quantified using enzyme assays with indicated
   yeast extracts and normalized to the wild type (ρ^+) strain. Despite
   the increased expression level (C), Aconitase activity was reduced
   below the detection limit in petites, and regained activity in evolved
   petites. In contrast, NAD-dependent isocitrate dehydrogenase activity
   mirrors the determined protein levels (shown in C) and increases in
   activity in petites. Mean values +/- SD of n = 3 biological replicates
   per genotype. Statistics based on two-sided, unpaired t-tests.

   E. Major carbon fluxes in wild-type, petites, and evolved petites
   during exponential growth on glucose. Wild-type (top row), petite
   (middle row), and evolved petite (bottom row) were grown in bioreactors
   (corresponding to [231]Fig. 1) and fermentation products were
   monitored. Shown are the estimated carbon balanced - glucose uptake
   rate, ethanol and glycerol production rate, as well as calculated
   remaining carbon flux allocated to CO2 that is indicated as the TCA
   cycle during exponential growth phases. Top row corresponds to
   wild-type, middle row to petite, and bottom row to evolved petite.
   Indicated rates are given as c-mol / (gDW*h) with six carbon
   equivalents for glucose, three for glycerol, and three for ethanol
   (assuming stoichiometric production of one CO2 per mole of ethanol).
   Shown uncertainties were calculated from the model based estimations.
   For corresponding data, see [232]Suppl Table 2.

   F-G. Stable isotope labelling incorporation into and levels of
   pyruvate, TCA cycle, and glutamate, glutamine, GABA. YSBN11 wild type,
   petite, and evolved petites carrying plasmid-encoded ATP3-6 as
   indicated were grown exponentially, harvested, and resuspended in media
   with [U-^13C]-glucose. After 1, 5, 15 and 30 min cells were quenched,
   extracted, and analyzed using GC-MS. (F) Fractional labelling
   (estimated as percent of the metabolite pool containing ≥ one ^13C
   carbon atoms) is similar between genotypes. (G) Pool sizes (estimated
   from the sum of all isotopologues) of indicated metabolites differ
   between genotypes. While wild type and evolved petites display largely
   similar levels of indicated metabolites, petites accumulate
   citrate/cis-aconitate, whereas pool sizes of glutamate, glutamine and
   GABA are reduced. Shown are mean values +/- SD of n = 3 (wild type,
   evolved petite), n = 2 (petite) biological replicates.

   A partial explanation for the perturbation of the TCA cycle in petites
   is its connection with the respiratory chain. The loss of the
   mitochondrial genome deletes Cox1p, Cox2pandCox3p that are implicated
   in electron transport from succinate dehydrogenase (SDH, complex II) to
   cytochrome c reductase (complex III) ^[233]1,[234]7,[235]8 . Without a
   functional succinate dehydrogenase, and potentially fostered by
   activation of the glyoxylate shunt, succinate accumulates in petites
   ([236]Fig. 4b).

   While our data is consistent with a loss of SDH activity, it did reveal
   a second and so far overlooked perturbation within the TCA cycle
   ([237]Fig. 4b). Levels of citrate and cis-aconitate were increased in
   petites, while the α-ketoglutarate pool was reduced ([238]Fig. 3b). On
   the protein level, α-ketoglutarate dehydrogenase (Kgd1p, Kgd2p) and
   succinyl-CoA ligase (Lsc1p, Lsc2p) were down-regulated, while aconitase
   (Aco1p, Aco2p) and especially isocitrate dehydrogenase (Idh1p, Idh2p,
   Idp1p, Idp3p) were up-regulated in petites ([239]Fig. 4a, c).
   Upregulation of aconitase and isocitrate-dehydrogenase isomers in
   petites contrasted to the accumulation of their respective substrates
   (citrate and cis-aconitate). Moreover, enzymes involved in catalysis of
   citrate to α-ketoglutarate returned to wild type levels in evolved
   petites.

   These results implied that one or more of the upper TCA cycle enzymes
   could be less active specifically in petites, and therefore be
   upregulated as a compensatory response.We quantified enzyme activities
   from wild-type, petite, and evolved petite extracts using enzyme assays
   to monitor substrate conversion ([240]Fig. 4d). NAD-dependent
   isocitrate dehydrogenase (IDH) activity was increased in petites
   according to the increased expression levels of the enzyme (Idh1p and
   Idh2p, [241]Fig. 4a, c, d). The NADP dependent isocitrate dehydrogenase
   was unchanged ([242]Fig. 4d). In sharp contrast, despite the increase
   in aconitase protein expression in petites ([243]Fig. 4c), the activity
   of this enzyme was reduced below the detection limits of the assay
   ([244]Fig. 4d). Remarkably, aconitase activity was again detected in
   the evolved petites ([245]Fig. 4d).

   We asked whether the inhibition of the TCA cycle enzymes is reflected
   in lower activity of this cycle. Subtracting the carbon balanced
   production rate of the major fermentation products ethanol and glycerol
   from the glucose uptake rate allowed us to estimate the major carbon
   fluxes between wild-type, petite, and evolved petites. Indeed,
   considering their biomass yield and growth rate, petites channel more
   carbon from glucose to ethanol/glycerol compared to wild-type
   ([246]Suppl. Table 2, [247]Fig. 4 e). This also suggests less carbon is
   allocated to the CO[2] producing TCA cycle. Then, we performed ^13C
   tracing experiments to estimate the relative activity of the TCA cycle.
   We followed labelling incorporation of [U-^13C]-glucose to pyruvate,
   and through the upper TCA cycle to succinate ([248]Fig. 4 f, g,
   [249]Suppl. Fig. 20). Of note, as both metabolomics and a metabolite
   labelling based strategy can not distinguish shared pathway
   intermediates of the mitochondrial TCA cycle with the
   cytosolic/peroxisomal glyoxylate cycle, these data have to be seen as a
   mix of both signals. Moreover, as a surrogate for α-ketoglutarate,
   which could not be quantified in petites during tracer experiments, as
   by addition of labelling, the signals were too diluted, we analysed
   label incorporation into and pool sizes of glutamate, glutamine, and
   GABA as downstream metabolites. While labelling incorporation after 30
   min was largely similar, the substrate levels of aconitase, citrate and
   cis-aconitate were elevated. Instead, the downstream metabolites
   glutamate/glutamine/GABA were reduced in petites, but not in evolved
   petites ([250]Fig. 4f, g). Especially considering that petites have
   less carbon flux ([251]Fig. 4e), these data are consistent with a
   reduced activity of reactions towards glutamate/glutamine/GABA
   specifically in petites. In contrast, succinate levels were elevated in
   petites but also in evolved petites ([252]Fig. 4b, Fig. 4g), which is
   in line with the loss of SDH activity in all strains lacking mtDNA. In
   summary, aconitase, isocitrate dehydrogenases and the glyoxylate shunt,
   targets of the retrograde response, are upregulated in petites. Petites
   are further characterized by strong reduction in aconitase and SDH
   activity, while evolved petites overcome the inhibition of aconitase.

Loss of mtDNA increases intracellular iron levels and is associated to
aconitase inhibition in petites

   Next, we addressed potential mechanisms that could inhibit aconitase
   activity in petites. Aconitase is a prominent iron-sulfur-cluster (ISC)
   dependent enzyme ^[253]80 . Defects of mtDNA and ΔΨ[mito] have been
   implicated with perturbed iron- and ISC-homeostasis
   ^[254]30,[255]81–[256]83 . We hence speculated that an impaired iron
   /and ISC metabolism could be responsible for the loss of aconitase
   activity. First, we tested for the activation of the iron regulome
   ^[257]84,[258]85 by recording proteomes of cells depleted and replete
   with iron ([259]Fig. 5a, b). In order to distinguish confounding
   effects caused by the differences in growth rate we applied principal
   component analysis (PCA, [260]Suppl. Fig. 21). Growth-rate (PC1) and
   iron depletion (PC2) response were the two dominant drivers of proteome
   changes. Growth-rate sensitive proteins (PC1) changed substantially in
   petites, but not in wild type and evolved petites ([261]Fig. 5a).
   Instead, the top 50 iron-response proteins (PC2) changed according to
   iron depletion in all wild type, petite, and evolved petite cells
   ([262]Fig. 5b). This includes proteins expected to change in response
   to iron depletion, such as ISC proteins that are down-regulated upon
   iron starvation^ [263]86 ([264]Fig. 5b red dots, [265]Suppl. Fig. 22).
   Thus, petites mount a functional response to iron depletion.

Figure 5. Iron overload inhibits aconitase, which is required for petite
suppression.

   [266]Figure 5
   [267]Open in a new tab

   A-B. The iron starvation-response is activated in petites. YSBN11 wild
   type (plasmid control), petite (plasmid control), and an evolved petite
   (expressing plasmid-based ATP3-6) were cultured with (+) or without (-)
   iron supplementation as indicated. Protein abundance was determined by
   SWATH-MS, and fold-changes were calculated in comparison to
   iron-replete (+iron) wild type and analyzed by principal component (PC)
   analysis ([268]Suppl. Fig. 21). (A) Heatmap displaying the 50 proteins
   with highest loadings in PC1 (associated with growth rate).
   Irrespective of iron addition, levels are consistently up regulated in
   petites, compared to both wild type or evolved petitesexpressing
   ATP3-6. (B.) Iron response mounts irrespective of the genotype. Heatmap
   illustrating the 50 proteins with highest loadings in PC2 (associated
   with iron-depletion). ISC proteins (red dots) are severely down
   regulated; levels of an iron transporter (blue dot) increases. Data
   from n = 4 biological replicates.

   C. Elevated cellular iron levels in petites. Iron levels in a YSBN11
   wild type (plasmid control), a petite (plasmid control), and an evolved
   petite strain (expressing plasmid-based ATP3-6) were analyzed by ICP-MS
   and normalized to protein content. Cellular iron is increased in
   petites (0.17 nmol Fe/μg protein) compared to wild type (0.81 nmol
   Fe/μg) and evolved petites (0.80 nmol Fe/μg). Mean values +/- SD of n =
   3 biological replicates. Statistics based on unpaired, two-sided
   t-tests. *p < 0.05.

   D. Petites are more resistant to iron depletion compared to wild type.
   (D) Dose response curve of YSBN11 wild type (black), petite (red), and
   evolved petites expressing ATP3-6 (blue). Shown is the normalized
   growth rate against increasing concentration of the iron chelator DIP.
   Growth rate was normalized to the maximum growth rate obtained with
   physiological iron concentration of the respective genotypes, as
   indicated by the dashed line. Mean values are indicated by lines, and
   standard deviation by area. Data from n = 6 biological replicates.

   E. Aco1 binds excessive iron in petites. BY4741 wild type (expressing
   Aco1-GFP, black), a petite (expressing Aco1-GFP, red), and an evolved
   petite (expressing Aco1-GFP and ATP3-6, blue) were exposed to ^55Fe.
   Aco1-GFP was enriched from cell extracts, and bound ^55Fe was
   quantified by scintillation counting. While GFP alone bound neglectable
   amounts of iron (18 cpm ^55Fe/mg protein), Aco1-GFP incorporated 636
   cpm (wild type), 6734 cpm (petite), and 1296 cpm (evolved petite) of
   ^55Fe/mg protein respectively. Thus, Aco1p binds excessive iron
   specifically in petites. Statistics based on unpaired, two-sided
   t-tests of n = 3 biological replicates. *p < 0.05.

   F. Mitochondrial aconitase activity is reduced in petites. Wild type
   and petite BY4741 yeast were transformed with plasmid controls, or
   plasmids harbouring native aconitase (ACO1) or additionally fused to a
   mitochondrial signaling sequence (mtACO1) respectively. In empty vector
   controls, aconitase activity was detected in wild type (6.9 mU/mg), but
   not in petites (0.5 mU/mg). Heterologous expression of ACO1 and mtACO1
   increased aconitase activity in wild type (16.7 mU/mg and 13.1 mU/mg).
   In petites, only ACO1 expression increased aconitase activity (5.4
   mU/mg), but not when targeted as mtACO1 (0.6 mU/mg). *p < 0.05, n=3.

   G. Aconitase is sensitive to increasing iron levels in vitro . Extracts
   of YSBN11 wild type (ρ^+, plasmid control) were incubated with
   increasing concentrations (0 - 500 μM) of Fe[II]-chloride and aconitase
   activity was determined. Exposure to iron concentrations ≥ 50 μM
   significantly reduced aconitase activity. Mean values are indicated by
   lines, and standard deviation by the grey area. Statistics based on
   unpaired, two-sided t-tests of n = 3 biological replicates per
   time-point. *p < 0.05.

   H. Iron-dependent inhibition aconitase is alleviated by addition of the
   iron chelator deferoxamine. Genotypes and experimental setup as
   described in (E). Cell lysates were incubated with or without 100 μM
   iron[II] chloride and 20 mM deferoxamine (DFO-B) and assayed for
   aconitase activity. While addition of 100 μM iron led to a significant
   reduction in aconitase activity, addition of 20 mM DFO-B alleviated
   iron-induced enzyme inhibition. Statistics based on unpaired, two-sided
   t-tests of n = 3 biological replicates per time-point. *p < 0.05.

   I. Petites no longer benefit from expression of ATP3-6 upon deletion of
   aconitase. Growth of indicated strains (genotypes and color code like
   in (G)) was monitored and growth rates were determined. The slow-growth
   phenotype of petites was rescued by expression of ATP3-6, but not in a
   Δaco1 background. Statistics based on unpaired, two-sided t-tests from
   n = 3 biological replicates. *p < 0.05.

   J. Proteome profile of Δaco1 does not respond to expression of ATP3-6 .
   Levels of 1929 proteins were determined in BY4741 wild type (ρ^+)
   expressing plasmid control (black) or ATP3-6 (grey), petites (ρ^0)
   expressing plasmid control (red) or ATP3-6 (blue), and petite Δaco1
   expressing plasmid control (yellow) or ATP3-6 (orange). A PCA was
   performed with normalized intensities of all proteins. Proteome
   profiles of wild type (black, grey) cluster closely together, while
   slow- (red) and fast-growing (blue) ρ^0 strains are well separated
   along PC2. In contrast, ρ^0 Δaco1 (yellow) is not separated from Δaco1
   expressing ATP3-6 (orange). Data from n = 3 biological replicates.

   Next, we tested whether excessive ion concentrations could be a
   problem. We conducted a total elemental composition analysis to
   quantify a range of elements including iron in wild type, as well as
   naive and evolved petites by inductively coupled plasma-mass
   spectrometry (ICP-MS, [269]Suppl. Table 3). We found iron levels were
   significantly elevated (> 2-fold) in petites compared to the wild type.
   Other element concentrations were not affected. ([270]Fig. 5c,
   [271]Suppl. Table 3). Strikingly, iron levels returned to wild type
   upon expression of the petite suppressor ATP3-6 ([272]Fig. 5c).

   In order to test whether the highly elevated iron levels could be
   associated with petite growth, we first manipulated growth media using
   iron chelators. The relative growth rate of wild type cells dropped
   with increasing iron chelator concentrations. Relative to their slower
   growth rate, petites did better tolerate iron depletion ([273]Fig. 5d,
   [274]Suppl. Fig. 23). To test if aconitase might be directly affected
   by elevated iron levels in petites, we supplemented the cells with the
   radioactive iron isotope ^55Fe, and affinity-purified mitochondrial
   Aco1p-GFP. Protein-bound iron from immuno-purified Aco1p-GFP
   wasquantified by scintillation counting. Aconitase enriched from wild
   type yeast showed weak radioactivity. Instead, aconitase purified from
   petites bound a high degree of radioactivity ([275]Fig. 5e). We then
   heterologously overexpressed ACO1, which is localized in both cytoplasm
   and mitochondria^ [276]87 . Furthermore, we overexpressed ACO1 fused
   with an additional mitochondrial targeting signal ^[277]50 . In a wild
   type background, aconitase activity was detected, and overexpression of
   both ACO1 constructs increased aconitase activity consistent with its
   overexpression ([278]Fig. 5f). Instead, in petite extracts, aconitase
   activity was reduced to the detection limit of the assay. Moreover, in
   the petite, no increase in total aconitase activity was detected when
   the overexpressed protein was specifically targeted to mitochondria
   ([279]Fig. 5f).

   Aconitase activity could be low because it is not correctly
   folded/assembled (e.g. not correctly loaded with it`s ISC).
   Furthermore, a principally functional enzyme could be sensitive to high
   iron levels. To test this, we used wild-type extracts, and tested
   aconitase activity in vitro, with an increasing iron concentration. As
   expected, in wild-type extracts, aconitase was active. However, the
   activity decreased with increasing iron concentrations ([280]Fig. 5g).
   Adding the iron chelating agent deferoxamine (DFO-B) prevented this
   effect ([281]Fig. 5h). Nonetheless, wild-type aconitase retained a
   higher activity in the presence of iron, compared to aconitase purified
   from petites ([282]Fig. 5g, f). These results imply that two scenarios
   apply: aconitase purified from petites is dysfunctional and has bound
   excessive iron ([283]Fig. 4 d, [284]Fig 5e, 5f), but also, a per se
   functional aconitate is sensitive to increased iron levels ([285]Fig
   5g, h).

   These findings suggested that aconitase itself would be required to
   suppress the petite phenotype. To test this, we generated an aco1
   deletion strain. Δaco1 loses respiratory function and mtDNA
   ([286]Suppl. Fig. 24a) ^[287]88 , and is auxotrophic for glutamate
   ([288]Suppl. Fig. 24b), which is in line with perturbation of the TCA
   cycle reactions towards α-ketoglutarate ^[289]89 . Likewise, growth of
   petites further deteriorates upon deletion of aco1 ([290]Fig. 5i).
   Based on classification with PCA, petites with intact ACO1 have a
   strong shift along PC2 upon expression of ATP3-6 ([291]Fig. 5j,
   [292]Suppl. Fig. 25). This effect was abolished upon aco1 deletion from
   petites, indicating that the suppressor mutation ATP3-6 lost its
   function when aconitase was disabled ([293]Fig. 5j). Moreover,
   petiteΔaco1 cells expressing mutant ATP3 could no longer suppress the
   slow-growth phenotype ([294]Fig. 5i). Thus, a functional aconitase is
   required for petite-suppression, and the inhibition of the Krebs cycle
   is directly associated with the slow-growth phenotype of petites.

Discussion

   The loss of mitochondrial DNA in yeast causes a severe growth defect
   described in 1949 as the petite phenotype, whose molecular basis was
   never fully clarified ^[295]17 . Mitochondria are a major organelle
   implicated in many cellular processes ^[296]1–[297]6 , and
   consequently, the loss of the mitochondrial genome has broad
   consequences on yeast physiology. In order to pinpoint the processes
   whose interruption is causal for the slow growth phenotype, we
   harnessed the power of adaptive evolution. Remarkably, accelerated
   growth of petites was readily evolved, and was explained by a set of
   mutations located in the C-terminal region of Atp3p, the ɣ subunit of
   the Complex V. Mutations in this hotspot region are known as one of the
   most common substitutions in the yeast genome
   ^[298]32,[299]36,[300]42,[301]45,[302]46 . We recently provided
   evidence that the frequent occurrence of such mutations are coinciding
   with an increase in the spontaneous mutation frequency, once the
   mitochondrial genome is deficient ^[303]32 . Our adaptive evolution
   experiment adds to these studies, by explaining that the selective
   advantage of ATP3 alleles is provided by maximizing growth rate, and by
   demonstrating that ATP3 mutations are a main solution to restore growth
   in petite yeast. This is because the design and long-term nature of our
   experiment would have allowed a number of possible adaptations to be
   selected, but a series of and independently emerging alleles all
   affected the ATP3 locus. In petites, the ATP3 mutations are dominant
   and similar to previously isolated mutant alleles that restore the
   mitochondrial membrane potential. Our data is effectively ruling out
   futile metabolic cycles to underlie the restoration of the ΔΨ[mito]
   because biomass formation does not change with petite suppression. It
   is thus possible that ATP3 suppressor mutations are gain of function
   mutations that generate a so far unknown biochemical activity. Although
   additional work will be required to elucidate the the mechanistic
   details, it is sensible to assume that elevated ΔΨ[mito] benefits
   mitochondrial functionality and thus various processes involving
   mitochondria, such as mitochondrial protein import and homeostasis
   ^[304]10,[305]90 .

   In this study, we focused on the physiology of these evolved petites,
   and used them as a model for identifying growth limiting factors of
   cells that have lost the mtDNA. We employed the basic assumption that
   perturbing the mechanism “repaired” in evolved mutants would leave the
   otherwise dominant allele dysfunctional. The effectiveness of this
   strategy is illustrated by ruling out retrograde response as a cause of
   the slow growth phenotype. We find that the RTG response is activated
   upon loss of the mitochondrial genome, and mitigated in evolved
   petites. However, the deletion of crucial components of the retrograde
   response impaired rather than restored petite growth. Further, the
   selected ATP3 alleles could still improve growth of petites in which
   the RTG was disrupted. Activation of the RTG is hence beneficial to
   petites as a downstream protective response, but cannot explain the
   growth benefit of isolated suppressor mutations.

   Eventually, our data converged in a model in which petites grow slowly
   because they are unable to synthesize amino acids, specifically
   glutamate, glutamine, arginine and leucine at a sufficient rate. We
   then continued to identify the key underlying molecular mechanisms. We
   discover that the TCA cycle in petites, which supplies biosynthetic
   precursors, or shares metabolites or iron sulfur cluster enzymes with
   the four metabolic pathways, is interrupted at two enzymatic steps,
   succinate dehydrogenase and aconitase. SDH is interrupted due to the
   loss of the mitochondrial genome ^[306]1,[307]7,[308]8 and remains
   deficient in the evolved petites. The inhibition of aconitase is
   overcome by evolved petites. Furthermore, once aconitase is deleted,
   the ATP3 mutations lose their function, and do no longer benefit
   petites.

   Our data indicates that two mechanisms contribute to the sensitivity of
   aconitase upon loss of mtDNA. First, we found that aconitase binds
   excessive amounts of iron in petites, and increased iron levels inhibit
   aconitase, even if it is functionally loaded with an ISC. Second, our
   results together with previous findings argue that in petites ISC
   assembly and export is perturbed due to a reduction of ΔΨ[mito]
   ^[309]30,[310]81–[311]83 . Reduced export of ISC can in turn explain
   elevated mitochondrial iron levels, and can contribute to incorrect
   loading of ISCs ^[312]81 . Of note, these results seem relevant for the
   pathology of Friedreich’s ataxia. This rare metabolic disease is
   associated with a mutation in the mitochondrial iron chaperone
   Frataxin, leading to accumulation of iron in mitochondria and aconitase
   deficiency ^[313]3,[314]91 .

   All four amino acids (QERL) that we uncovered to rescue the growth
   defect in petites rely on at least partial mitochondrial biosynthesis
   ^[315]92,[316]93 . Glutamate, which is in turn required for synthesis
   of glutamine and arginine, is directly affected by the perturbation of
   aconitase as its biosynthesis depends on α-ketoglutarate. Leucine ties
   into this perturbation as its synthesis requires Acetyl-CoA, and is
   inhibited by CoA ^[317]92,[318]94 . Another aforementioned common
   denominator in the biosynthesis of these amino acids is the requirement
   for intact iron sulfur clusters, specifically in the case of leucine
   (ILV3, LEU1) and glutamate/glutamine (GLT1) ^[319]81,[320]82,[321]92 .
   Nevertheless, as petites are viable, there must be residual ISC
   assembly, and in turn loading of ISC-dependent enzymes. Of note, yeast
   is particularly sensitive to perturbations in leucine metabolism.
   Leucine is one of the most frequently encoded amino acids, needs to be
   synthesized at high rates, and the strong growth defect of leucine
   auxotrophs cannot fully be compensated by supplementation
   ^[322]63–[323]65 . It hence seems plausible that a shortage in ISC
   metabolism would manifest in the leucine pathway, not only because it
   is in high demand, intertwined with the Krebs cycle and mitochondrial
   metabolism, but additionally because it depends on mitochondrial ISC
   export ^[324]95 .

   Taken together, while numerous cellular processes depend on
   mitochondria, the main growth rate restriction caused by mitochondrial
   genome loss is explained by a well-defined, biosynthetic defect in
   amino acid metabolism-most of it that can be overcome with a simple
   nutritional supplementation. These findings suggest that even complex
   mitochondrial dysfunctions, as they are seen in numerous genetic and
   acquired human diseases, could be overcome by mapping and compensating
   for their precise biochemical defects.

Material & Methods

Chemicals and media

   All chemicals were obtained from Sigma-Aldrich unless stated otherwise.
   Prototrophic yeast strains were cultured in F1 ^[325]44 or SM (6.8 g/L
   yeast nitrogen base (Sigma) medium, and growth of auxotrophic strains
   in SM was enabled by supplementing with 20 mg/L histidine, 60 mg/L
   leucine, 20 mg/L uracil, 20 mg/L methionine and/or 50 mg/L lysine as
   indicated. Synthetic complete (SC) medium was prepared by supplementing
   SM medium with CSM-HIS-LEU-MET-TRP-URA (MP Bio), 10 mg/L adenine, 20
   mg/L uracil, 40 mg/L tryptophan, and amino acids as above. YP (20 g/L
   peptone (Bacto), 10 g/L yeast extract (Bacto) medium was used where
   indicated. Iron-free culture of yeast was performed using SM medium
   prepared with yeast nitrogen base without amino acids and without iron
   (Formedium), and analytical grade EMSURE water (Merck). Where
   indicated, iron-free medium was supplemented with 200 μg/L FeCl[3] 6
   H2O. Media were supplemented with 2 % glucose unless stated otherwise.
   Δaco1 yeast strain was grown by additional supplementation with 5 mM
   glutamate.

Yeast strains

   Adaptive evolution experiment was carried out using wild type YSBN1, a
   prototrophic diploid variant of S. cerevisiae S288c ^[326]44 . For
   transformation of yeast with plasmid constructs, derivatives of YSBN1
   with single deletions of ura3 or his3 were used ^[327]44 . Where
   indicated, BY4741 strain was used. Single gene deletion mutants BY4741
   Δcit2, Δzwf1 and Δrtg2 were obtained from the yeast deletion collection
   ^[328]96 and identity verified by sequencing. BY4741 Δaco1 was
   generated by inserting a natMX targeting cassette ^[329]97 into the
   ACO1 locus and verification by sequencing.

   Strains depleted of mtDNA (ρ^0) were generated by plating yeast on YPD
   agar containing 0.1 g/L ethidium bromide and incubation at 30°C for 2
   d. Absence of mtDNA was confirmed in isolated surviving clones by DAPI
   staining and growth assays using the non-fermentable carbon sources
   ethanol and glycerol. For ρ^0 strains carrying a plasmid, yeast was
   transformed according to standard procedures, grown on selective agar
   and positive clones isolated. Single clones were then subjected to
   mtDNA depletion procedure as described above.

Plasmid constructs

   Plasmids used in this study are listed in [330]Suppl. Table 4 and are
   available via Addgene. For construction of various pRS313_ATP3
   plasmids, respective ATP3 variants (T911A, G919C, T911A+G919C) were
   first obtained by PCR site directed mutagenesis. Resulting alleles, and
   the respective native promoter region (the ~600 bp upstream region)
   were sub-cloned using the pGEM-T System (Promega). Subsequently,
   respective ATP3 alleles (including wild type) were cloned into the
   multiple cloning site of pRS313 ^[331]98 by digests with SalI/BamHI.
   Successful insertion was evaluated using blue/white selection. Finally,
   Sanger sequencing confirmed the respective ATP3 alleles of pRS313-ATP3
   constructs and excluded additional mutations. Plasmid expressing
   Aco1p-GFP has been described elsewhere ^[332]99 , and plasmid
   expressing GFP-Pts1 was a gift from E. Hettema ^[333]100 . Construct
   for expression mitochondrially targeted aconitase was generated by
   sub-cloning of ACO1 and addition of preSu9 sequence using homologous
   recombination, and insertion into p416GPD using BamHI/SalI sites.

Adaptive evolution and genome sequencing

   Wild type and petites were cultured in F1 carbon-limited medium
   ^[334]101 with 2.5 g/L glucose as carbon source in a DASGIP Parallel
   Bioreactor Systems chemostat for 67 days, with pH controlled at 4.5,
   agitation at 200 rpm, and temperature at 30°C. Three chemostat cultures
   were grown for each cell line a dilution rate of 0.1. Aliquots of each
   culture were archived daily with 25 % glycerol at -80°C.

   For genome sequencing, YPD was inoculated with glycerol stocks of
   chemostat samples and harvested in exponential phase. A 5 mL yeast
   culture was grown overnight in YPD and 1.4 mL culture was added to a
   microcentrifuge tube. Genomic DNA was extracted with a Wizard Genomic
   DNA Purification Kit (Promega) according to the manufacturer’s
   protocol.

   Whole-genome sequencing was performed on an Illumina HiSeq 2500
   platform with 50 bp paired end sequencing. The adapter sequence was
   AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCTCTTCCGATCT. Data was
   analyzed essentially as described ^[335]102 . 96% of reads could be
   aligned against the database
   Saccharomyces_cerevisiae.EF4.69.dna_sm.toplevel.fa obtained from
   Ensembl using bwa v0.5.9-r16 ^[336]103 after sequencing adapter
   removal; duplicates were marked using Picard “MarkDuplicates” and
   coverage was between 61x and 452x. SNP calling was done after local
   realignment using “UnifiedGenotyper” from the GATK v2.2 pipeline with
   default (sensitive) parameters. Annotation was done using ANNOVAR
   ^[337]104 with information from Ensembl Saccharomyces cerevisiae v69.
   Positions with coverage below 10x in any of the experiments were
   excluded from the analysis, as well as positions which were mutated in
   any of the wild type timepoints as found in > 1 % of reads with a
   minimum of 2 reads. Suppressor mutations of petites were included if
   they were found in > 1 % of reads with a minimum of 2 reads at the
   final time point of at least one replicate culture.

Growth curves and chemostat experiments

Determination of growth rates

   For determination of growth rates, yeast was inoculated at an optical
   density (OD[600]) ~ 0.1 in the respective liquid medium in 96-well
   plates and cultured in quadruplicates in a FLUOstar OPTIMA plate reader
   (BMG LABTECH) for 40 h at 30°C, with recording of OD[600] every 20 min.
   Maximum specific growth rate (μ[max]) and lag phase were both
   determined from growth curves using a model-richards fit from the R
   ‘grofit’ package v. 1.1.1-1 ^[338]105 .

Chemostat experiments and quantification of fermentation products

   200 mL pre-culture medium (per liter: 10 g yeast extract (Merck), 20 g
   soy peptone (Kerry Bio-Science), 22 g glucose monohydrate (Roth) were
   inoculated with 750 μL cryostock of the respective yeast strains and
   grown on an orbital shaker at 30°C and 180 rpm overnight. This culture
   was used for inoculation of the bioreactors at an OD[600] of 0.4. The
   batch cultivations were performed in 1.4 L bioreactors (DASGIP Parallel
   Bioreactor System) with 1.0 L F1 medium containing 2 % glucose.
   Cultivation temperature was controlled at 30°C, pH was controlled at
   5.0, and the dissolved-oxygen concentration was maintained above 20 %
   saturation by controlling the stirrer speed between 200 to 1200 rpm as
   well as the air flow between 6 sL/h to 40 sL/h. Inlet and outlet gases
   were followed with the off-gas sensor provided by the bioreactor system
   (DASGIP Off-Gas Analyzer GA4, DASGIP AG). Foam formation was prevented
   by the automatic addition of a 1 % solution of Struktol SB2121
   (Schill+Seilacher).

   12 mL of samples were taken at regular intervals throughout the whole
   cultivation duration. Biomass production (OD[600]) was determined using
   a spectrophotometer (Ultrospec 1100pro, Amersham Biosciences). A
   correlation between the OD[600] and the cell dry mass (CDM) was
   established. The concentrations of glucose, glycerol and ethanol in the
   culture broth were determined by HPLC analysis (Shimadzu Corporation)
   with a Phenomenex Rezex ROA column (300 mm × 7.8 mm) and a refractive
   index detector (RID-10A, Shimadzu Corporation). The column was operated
   at 60°C temperature, 1 mL/min flow rate and 4 mM H[2]SO[4] as a mobile
   phase. HPLC samples were prepared by adding 100 μL of 40 mM H[2]SO[4]
   to 900 μL culture supernatant. Subsequently the samples were filtered
   on 0.20 μm RC membrane filters and 10 μL were injected for analysis.

   Rates for glucose uptake, ethanol / glycerol production as well as
   growth rate were estimated by fitting the time courses (including time
   points up to 10.5 hours of growth) to an ordinary differential equation
   model (ODE) assuming exponential growth and constant yield. The model
   was used as described elsewhere ^[339]106 , and gPROMS Model Builder
   (v. 4.0, Process Systems Enterprise Ltd.) was used for parameter
   estimation.

Microscopy

Quantification of mitochondrial membrane potential and morphology by MitoLoc

   Microscopic analysis of ΔΨ^mito using the MitoLoc system was carried
   out as previously described ^[340]50 . Pixel-by-pixel colocalization of
   preSU9-GFP and preCOX4-mCherry was used as a measure for ΔΨ^mito using
   the Pearson Correlation Coefficient (PCC) as described ^[341]50 .
   Mitochondrial fragmentation was determined as described ^[342]50 .

Quantification of mitochondrial membrane potential with DIOC[6] analyzed by
microscopy

   ΔΨ^mito analysis using 3,3′-dihexyloxacarbocyanine iodide (DiOC[6]) was
   performed by growing strains in SM for 4 h to an OD[600] of ~0.5. Where
   indicated, cells were pre-treated with 15 μM CCCP for 5 min to
   depolarize mitochondria, and 5 mio cells stained with 175 nM DiOC[6] in
   HEPES with 5 % glucose for 15 min at 30°C. After washing with PBS,
   cells were resuspended in fresh medium, embedded in agarose pads and
   subjected to microscopy as described ^[343]50 , using
   excitation/emission wavelengths of 488/528 nm.

Quantification of mitochondrial membrane potential with DIOC[6] analyzed by
flow cytometry

   Freshly streaked colonies were resuspended in 30 μL water and 10 μL of
   suspension were used to inoculate 500 μL SM or SM with 2mM glutamate,
   glutamine, arginine, and leucine (+QERL) and incubated overnight with
   agitation at 30°C. The following day, cultures were diluted 1:2 and
   cultured for another 4-8 hours to an OD[600] ~0.5. Cells were then spun
   down at 4000 rpm for 3 min and the media exchanged with SM or SM+QERL
   supplemented with 175 nM DiOC[6], incubated at 30°C with agitation for
   a further 15 min. Cells were then repelleted and washed with 500
   μL/well PBS, before final centrifugation and resuspension in 500 ul
   PBS. 200 μL of each culture was transferred to a U-bottomed 96 well
   plate and taken for high-throughput flow cytometry analysis on a
   Fortessa X20 HTS cytometer. Acquisition parameters were 120 μL sample
   injection at a flow rate of 1 μL/sec. DiOC[6] fluorescence was captured
   using 488 nm blue laser excitation followed by a 530/30 bandpass
   filter.

Quantification of peroxisomes

   The number of peroxisomes was determined in GFP-Pts1p expressing BY4741
   yeast using super-resolution microscopy as described ^[344]50 . In
   short, BY4741 expressing GFP-Pts1p was cultured to exponential phase in
   SC-HU medium containing 2 % glucose, harvested by centrifugation,
   washed twice in PBS and fixed by resuspension in formaldehyde solution
   (4 g/L PFA, 3.6 % sucrose). The number of peroxisomes was counted using
   CellProfiler software 2.0 ^[345]107 .

ROS quantification

   ROS production was assessed by growing strains in SM for 4 h to an
   OD[600] of ~0.5. Where indicated, cells were pre-treated with 2 mM
   tert-Butyl hydroperoxide (tert-BOOH) for 5 min to induce formation of
   ROS, and and 5 mio cells stained with 2.5 μg/mL dihydroethidium (DHE)
   for 5 min in the dark. After washing with PBS, cells were resuspended
   in fresh medium, embedded in agarose pads and subjected to microscopy
   as described ^[346]50 , using excitation/emission wavelengths of
   510/595 nm. Images were analyzed using CellProfiler software 2.0
   ^[347]107 , counting the proportion of cells with detectable
   fluorescence.

Metabolomics and proteomics

Metabolomics

   For quantification quantification of intracellular metabolites, single
   colonies of respective yeast strains were grown to mid-log phase
   (OD[600] ~0.75) in selective medium, and a culture volume equalling 7.5
   OD[600] was harvested by injecting into cold methanol and subsequent
   metabolite extraction as described ^[348]73 . Amino acid quantification
   by LC-MS/MS was performed as previously described ^[349]108 . Sugar
   phosphate quantification by LC-MS/MS was performed as previously
   described ^[350]73 . Quantification of ADP, ADP and AMP was performed
   as described ^[351]109 . For quantification of reduced and oxidized
   glutathione we performed LC-MS/MS experiments on a 6460 Triple
   Quadrupole Mass Spectrometer (Agilent) coupled to UPLC (1290 Infinity,
   Agilent) operated with a binary gradient using 0.1 % formic acid in
   water as buffer A and methanol as buffer B and a Zorbax Eclipse Plus
   C18 column (Agilent, RRHD, 2.1x100 mm, 1.8 μm, column temperature:
   20°C). Chromatographic separation of extracted analytes was achieved by
   isocratic flow at 100 % A at 0.5 mL/min for 0.5 min, which was then
   ramped to 10 % B within 2.5 min. After a 0.5 min washing step at 10 % B
   the column was re-equilibrated to 100 % A resulting in a total cycle
   time of 5 min. GSSG and GSH were quantified in positive ESI mode via
   the MRM transitions 613→484, 613→355, 613→231 (fragmentor: 120,
   collision energy: 18), and 308→233, 308→179, 308→161 (fragmentor: 105,
   collision energy: 9), respectively. Medium glucose was quantified using
   LC-MS/MS as described ^[352]110 modified by including the transition
   (179→89, 70 V Fragmentor, 5 V collision energy, negative mode).

Protein profiling, data analysis, and pathway enrichment

   For protein profiling, single colonies of respective yeast strains were
   grown to mid-log phase (OD[600] ~0.7) in selective medium containing 2
   % glucose, harvested by centrifugation and snap-frozen in aliquots
   equalling 10 OD[600] units. Sample preparation was carried out as
   previously described ^[353]60 . SWATH LC-MS/MS analysis was performed
   essentially as previously described ^[354]60 on a TripleTOF5600
   instrument (SCIEX) online coupled to a nanoACQUITY chromatographic
   system (Waters) operating at 3 μL/min flow rate. Data was analyzed with
   Spectronaut software (version 14, Biognosys AG) and post-processed in
   statistical language R. Principal component analysis was carried out
   using the unfiltered data set and a prcomp function. Protein fold
   change was calculated with reference to the wild type strain, and
   differential abundance was defined as a fold change > 1.5 and
   FDR-corrected p-value < 0.01. Pathway enrichment was performed in
   String-db v11 ^[355]111 , and Reactome pathways ^[356]112 were reported
   as significantly enriched if FDR-corrected enrichment p-value < 0.05.

Dynamic labelling experiments

Sample preparation

   For dynamic labelling experiments, yeast cultures were inoculated at
   OD[600] of 0.15 in F1 medium containing 2 % glucose and agitated at
   30°C until reaching mid-log phase. 45 OD[600] units (~6 × 10^8 cells)
   were pelleted, re-suspended in 6 mL F1 medium containing 1 % (w/v)
   [U-^13C[6]]-glucose (Cambridge Isotope Laboratories) and kept at 30°C
   with occasional agitation. 1 mL of cells (~1 × 10^8 cells) was removed
   after 1, 5, 15 and 30 min, shock-frozen by injection into 20 mL of
   -40°C MeOH and pelleted by centrifugation (2 min, 4000 × g, 4°C).
   Methanol was removed, and metabolites extracted by resuspension of the
   pellet in 600 μL chloroform/methanol (600 μL, 2:1 v/v), transfer to a
   1.5 mL tube, brief vortexing, and subsequent periodic sonication in a
   water bath for 1 hr at 4°C. Samples were pelleted by centrifugation (10
   min, 16000 ×g, 4°C), the metabolite extract transferred to a new tube,
   and the pellet re-extracted with methanol/water (600 μL, 2:1 v/v,
   containing 1 nmol scyllo-inositol). Samples were pelleted as above, the
   second extract added to the first and dried in a rotary vacuum
   concentrator. Samples were biphasic partitioned by addition of 100 μL
   chloroform, followed by 600 μL methanol/water (1:1 v/v) and vigorous
   vortexing. After centrifugation (as above), the upper (polar) phase was
   removed, dried, and processed for gas chromatography-mass spectrometry
   (see below).

Gas Chromatography-Mass Spectrometry (GC-MS)

   Polar metabolites were derivatized and analysed by GC-MS (Agilent
   7890B-5977A) and identification and abundance of individual metabolites
   was estimated as previously described ^[357]113 . In brief, dried
   metabolite samples were washed with methanol (twice), and derivatized
   overnight at RT with methoxyamine (20 mg/mL in pyridine, Sigma)
   followed by addition of BSTFA + 1% TMCS (Sigma) for > 1 hr at RT. GC-MS
   was performed using splitless injection (injection temperature 270°C)
   onto a 30 m + 10 m × 0.25 mm DB-5MS+DG column (Agilent J&W), with
   helium carrier gas, in electron impact ionization mode. The oven
   temperature was initially 70°C (2 min), followed by a temperature
   increase to 295°C at 12.5°C/min and subsequently to 320°C at 25°C/min
   (held for 3 min). GAVIN software ^[358]114 was used for metabolite
   identification and quantification by comparison to the retention times,
   mass spectra, and responses of known amounts of authentic standards.

Iron depletion experiments, and ^55Fe incorporation binding assays

Iron depletion for proteomics

   Culture was performed by inoculating a pre-culture (iron-free SM medium
   with or without iron, 2 % glucose) with washed yeast cells and
   incubation overnight at 30°C. Main culture was inoculated to OD[600] of
   0.1, and incubated at 30°C with shaking until cultures reached OD[600]
   of 0.7. 10 OD[600] units were collected by centrifugation, washed with
   iron-free water and stored at -80°C.

Iron depletion in liquid media

   Single colonies of indicated strains were transferred to 10 mL SM
   situated in Erlenmeyer flasks, and incubated overnight. Cultures were
   diluted to OD[600] ~0.2 and incubated for 5 hours. Cells were
   harvested, supernatants were discarded, and cells were washed 3 times
   in sterile MiliQ H[2]O, and OD[600] was adjusted to 0.2. Then, 5 μL of
   each culture was transferred to a microtiter plate with 195 μL of
   minimal media with iron, without iron, and without iron and increasing
   dipyridyl (DIP) concentration. The base formulation of this media
   corresponds to 6.7 g/L yeast nitrogen base (YNB, Sigma Y0626) with 2 %
   glucose. For iron depletion, the ferric chloride was omitted, and where
   indicated DIP was added from a 50 mM stock dissolved in DMSO to a final
   concentration of 50, 100, 150, 175 μM respectively. Growth curves were
   obtained on a Tecan Spark-Stacker using the mean values of five
   multi-well reads at OD[600] obtained every 30 minutes. Growth rates
   were estimated from obtained growth data using the spline fit
   implemented in the R package grofit (v1.1.1-1).

^55Fe incorporation assay

   Quantification of Fe-S cluster formation was performed essentially as
   described ^[359]115 . BY4741 yeast transformed with pUG35_Aco1
   construct were verified for mitochondrial GFP expression using
   fluorescence microscopy, and pre-cultured overnight in iron-free SM+LMW
   medium with 2 % glucose. Main culture was inoculated to OD[600] ~0.1,
   and incubated at 30°C until all strains were in mid-log phase. Cells
   were harvested by centrifugation, washed once with analytical grade
   water, re-suspended in 10 mL fresh medium per 0.5 g wet cell mass, and
   incubated for 10 min (30°C, 200 rpm). 10 μCi ^55FeCl[3] (Perkin Elmer)
   in 100 μL 0.1 M sodium ascorbate was added to each sample, and iron
   incorporation performed for 2 h. After collecting cells by
   centrifugation (3000 xg, 5 min), pellet was washed once with 10 mL
   citrate buffer (50 mM sodium citrate, 1 mM EDTA, pH 7.0) and once with
   1 mL HEPES-KOH pH 7.4. Cells were disintegrated by bead shaking in a
   FastPrep device (MP Biomedicals) with 500 μL TNETG+PI buffer (20 mM
   Tris-HCl pH 7.4, 2.5 mM EDTA, 150 mM NaCl, 10 % (w/v) glycerol, 0.5 %
   (w/v) Triton X-100, 1x cOmplete Protease Inhibitor Cocktail (Roche) and
   0.5 volume acid-washed glass beads for 30 s at 25/s speed. After
   centrifugation (10 min, 17000 xg, 4°C), protein concentration of
   supernatants was determined, adjusted to 15 μg/μL and GFP-Aco1p
   enriched using 25 μL washed GFP-Trap magnetic beads (ChromoTek).
   Samples were inverted (1 h, 4°C), beads collected and washed
   extensively with TNETG+PI buffer. Bound ^55FeCl[3] was quantified by
   scintillation counting.

ICP-MS

   For quantification of elemental composition, yeast cells were
   resuspended in 4.5 mL analytical grade water, and proteins were cleaved
   by addition of 1.4 mL concentrated HNO3 (final concentration 16.4 %)
   and incubation for 2 h at 80°C. Following centrifugation (10 min, 4000
   xg), the supernatants were then analyzed undiluted in triplicate using
   inductively coupled plasma spectrometry (ICP-MS) on a Perkin Elmer Elan
   DRC II instrument. The instrument was calibrated by dilution of Fe
   standard solution (1 ppb to 1 ppm) and the standard was re-analyzed
   every 16 samples to correct for instrumental drift. A 1 ppb solution of
   In, Re and Rh was used as an internal standard. Concentrations of
   elements were normalized to protein concentration of yeast lysate as
   determined by BCA assay.

Biochemical and molecular biology assays

Rapamycin sensitivity testing

   For rapamycin sensitivity assay, strains were inoculated at 0.1 OD[600]
   with 11 rapamycin concentrations from 50 pg/mL to 1 μg/mL (n = 3) in a
   384 multi-well plate (70 μL/well) in SC-H medium and growth was
   monitored for 48 h. The maximum specific growth rate and EC50 values
   for dose response curves were determined with R package grofit
   (v1.1.1-1).

Yeast dilution spot tests

   For oxidant spot tests, a 2 mL overnight culture of yeast was prepared.
   Cells were harvested, washed once in H2O and re-suspended in fresh,
   autoclaved H2O to an OD[600] of 2.5 unless otherwise indicated. 200 μL
   of the resulting suspension was transferred to a 96-well plate and a
   1:5 dilution series was prepared. Oxidant plates were prepared freshly
   by cooling media containing agar to ~50°C, 1.2 mM diamide was added to
   aliquots, mixed, cast into petri dishes and cooled to RT. Cell
   suspensions were vigorously mixed, and 5 μL of each dilution was
   spotted onto agar plates. Spots were air-dried under a flow hood and
   incubated at 30°C for 2-3 days.

   For amino acid supplementation spot tests, overnight cultures of
   indicated strains were prepared in 15 mL SM situated in flasks, and
   supplemented with or without indicated amino acids. The overnight
   culture was diluted 1:20 in respective media, and incubated for 5
   hours. Cells were harvested, set to an OD[600] of 0.5 in the respective
   media, and spotted and incubated as described above.

   For iron depletion spot tests, the indicated strains were pre-grown in
   SM with 2 % glucose. A serial dilution series (1:5) was spotted with an
   initial OD[600] of 0.5 on SM agar plates with or without 1 mM
   ferrozine.

Quantification of enzyme activities

   Isocitrate dehydrogenase activity was quantified using a coupled
   diaphorase assay as described ^[360]116 , using 20 μg/mL yeast lysate
   and 2 mM NAD^+ or 2 mM NADP^+. The reaction was started by addition of
   16 mM isocitrate and followed continuously in 96-well plates at 340 nm
   in a plate reader (Tecan Infinite 200Pro). Aconitase activity was
   quantified as described ^[361]99 , with 1 U/mL IDH1 (Roche), 0.4 mM
   NADP^+, and 20 μg/mL yeast lysate. The reaction was started by addition
   of 2 mM citrate and absorption at 340 nm followed continuously as
   above. ATPase (complex V) activity was quantified as described
   ^[362]117 , and specific activity (μmol ADP/mg protein/min) of the
   oligomycin-insensitive fraction was calculated according to the
   Beer-Lambert law equation.

Aconitase overexpression and localization tagging

   Wild type and petite BY4741 was transformed with a plasmid expression
   of Aco1p (ACO1), Aco1p constitutively targeted to mitochondria
   (mtACO1), or the empty vector. Cells were grown to mid-exponential
   phase in SC-U medium, harvested, broken, and aconitase activities
   determined as described above.

Supplementary Material

   Suppl. Fig. 1
   [363]EMS140507-supplement-Suppl__Fig__1.pdf^ (3.6MB, pdf)
   Suppl. Table 1
   [364]EMS140507-supplement-Suppl__Table_1.xlsx^ (1.4MB, xlsx)

Acknowledgements