Requirements for Carnitine Shuttle-Mediated Translocation of Mitochondrial Acetyl Moieties to the Yeast Cytosol

ABSTRACT In many eukaryotes, the carnitine shuttle plays a key role in intracellular transport of acyl moieties. Fatty acid-grown Saccharomyces cerevisiae cells employ this shuttle to translocate acetyl units into their mitochondria. Mechanistically, the carnitine shuttle should be reversible, but previous studies indicate that carnitine shuttle-mediated export of mitochondrial acetyl units to the yeast cytosol does not occur in vivo. This apparent unidirectionality was investigated by constitutively expressing genes encoding carnitine shuttle-related proteins in an engineered S. cerevisiae strain, in which cytosolic acetyl coenzyme A (acetyl-CoA) synthesis could be switched off by omitting lipoic acid from growth media. Laboratory evolution of this strain yielded mutants whose growth on glucose, in the absence of lipoic acid, was l-carnitine dependent, indicating that in vivo export of mitochondrial acetyl units to the cytosol occurred via the carnitine shuttle. The mitochondrial pyruvate dehydrogenase complex was identified as the predominant source of acetyl-CoA in the evolved strains. Whole-genome sequencing revealed mutations in genes involved in mitochondrial fatty acid synthesis (MCT1), nuclear-mitochondrial communication (RTG2), and encoding a carnitine acetyltransferase (YAT2). Introduction of these mutations into the nonevolved parental strain enabled l-carnitine-dependent growth on glucose. This study indicates intramitochondrial acetyl-CoA concentration and constitutive expression of carnitine shuttle genes as key factors in enabling in vivo export of mitochondrial acetyl units via the carnitine shuttle.

Catabolism of the acetyl-CoA generated during growth of S. cerevisiae on fatty acids involves the mitochondrial tricarboxylic acid (TCA) cycle. Conversely, during growth on glucose, the mitochondria act as an important source of acetyl-CoA, with the pyruvate dehydrogenase (PDH) complex catalyzing the predominant acetyl-CoA generating reaction (8,26). The carnitine acetyltransferase reaction is, in principle, mechanistically and thermodynamically reversible (⌬G R°= ϭ Ϫ1.1 kJ · mol Ϫ1 in the direction of acetyl-L-carnitine formation [27]). This observation suggests that the carnitine shuttle should not only be able to import acetyl units into the mitochondria but also be able to export them from the mitochondrial matrix to the cytosol. Therefore, based on in vitro experiments, it was initially hypothesized that the carnitine shuttle was responsible for export of acetyl moieties from yeast mitochondria (22). Further studies, however, indicated that the PDH bypass, which encompasses the concerted action of pyruvate decarboxylase, acetaldehyde dehydrogenase, and acetyl-CoA synthetase (28), was responsible for cytosolic acetyl-CoA provision in glucose-grown S. cerevisiae cultures (26) (Fig. 1A). Several additional observations argue against an in vivo role of the carnitine shuttle in export of acetyl moieties from mitochondria to cytosol in glucose-grown cultures. In wild-type S. cerevisiae, transcription of genes involved in the carnitine shuttle is strongly glucose repressed (18,19,29), which precludes a significant contribution to cytosolic acetyl-CoA provision in glucose-grown batch cultures. Moreover, even in derepressed, glucose-limited chemostat cultures, supplementation of growth media with L-carnitine cannot complement the growth defect of strains lacking a functional PDH bypass, which is caused by an inability to synthesize cytosolic acetyl-CoA (30). Hence, based on currently available data, the carnitine shuttle of S. cerevisiae appears to operate unidirectionally (i.e., transporting acetyl moieties into the mitochondria) during growth on glucose.
The goal of the present study is to investigate the molecular basis for the apparent unidirectionality of the yeast carnitine shuttle. To this end, we studied growth on glucose of an S. cerevisiae strain in which the carnitine shuttle is constitutively expressed. We recently demonstrated that constitutive expression of the components of the carnitine shuttle enables efficient transport of acetyl moieties from cytosol to mitochondria in glucose-grown, L-carnitine-supplemented batch cultures (8). In the present study, overexpression of the carnitine shuttle proteins was combined with replacement of the native S. cerevisiae pathway for cytosolic acetyl-CoA synthesis by a cytosolically expressed bacterial PDH complex (31). In the resulting strain, cytosolic acetyl-CoA synthesis could be switched off at will by omitting lipoic acid from growth media. After evolving in the laboratory, mutations required for L-carnitine-dependent growth in the absence of lipoic acid were identified by whole-genome sequencing and functionally analyzed by their introduction in the nonevolved parental strain.

RESULTS
Constitutive expression of carnitine shuttle genes does not rescue growth on glucose of S. cerevisiae acs1⌬ acs2⌬ strain. Interpretation of previous studies on the role of the carnitine shuttle in glucose-grown cultures of S. cerevisiae is complicated by the strong glucose repression of the structural genes encoding carni- In the evolved strains IMS0482 and IMS0483, extracellular L-carnitine is imported into the mitochondria via the Hnm1 transporter at the plasma membrane and the Crc1 transporter at the inner mitochondrial membrane. Pyruvate is imported into the mitochondria, following its oxidative decarboxylation by the native mitochondrial PDH complex. The acetyl moiety is then transferred to L-carnitine, followed by export of acetyl-L-carnitine to the cytosol. There, carnitine acetyltransferases move the acetyl moiety back to CoA, yielding cytosolic acetyl-CoA. Abbreviations: Ach1, CoA transferase; Acs, Acs1, and Acs2, acetyl-CoA synthetase; Agp2, regulator; ALD, acetaldehyde dehydrogenase; CAT, carnitine acetyltransferase; Crc1, acetyl-carnitine translocase; Hnm1, carnitine transporter; LplA and LplA2, lipoylation proteins; Mpc1, Mpc2, and Mpc3, mitochondrial pyruvate carrier; OAA, oxaloacetate; PDC, pyruvate decarboxylase; PDH, pyruvate dehydrogenase complex.
Van Rossum et al.
tine acetyltransferases and acetyl-carnitine translocase (18,19,29,32). To reexamine whether the carnitine shuttle can translocate acetyl units from mitochondria to cytosol, a strain was constructed in which provision of cytosolic acetyl-CoA could be made strictly dependent on a constitutively expressed carnitine shuttle. Its construction ( Fig. 2A) started with a strain in which cytosolic acetyl-CoA metabolism had been modified by replacing the acetyl-CoA synthetase genes ACS1 and ACS2 by the six-gene {PDHL} cluster (we use the curly brackets to indicate a chromosomally integrated cluster of PDH complex {PDHL} genes as discussed in "Strain construction" below in Materials and Methods) (33) ( Table 1), which enables functional expression in the yeast cytosol of the Enterococcus faecalis PDH complex (Fig. 1B). This strain provided an experimental model in which cytosolic acetyl-CoA synthesis could be switched off at will by omitting lipoic acid from growth media. The functionality of alternative (introduced) routes to cytosolic acetyl-CoA could thus be tested by omitting lipoic acid and checking for growth. Expression cassettes were constructed in which the yeast carnitine shuttle genes (AGP2, CAT2, CRC1, HNM1, YAT1, and YAT2) were controlled by strong, constitutive promoters. The resulting six DNA fragments were assembled and integrated as a single cluster of carnitine genes ({CARN}; Fig. 2B; Table 1) into the genome of the strain carrying the {PDHL} cluster. Consistent with an earlier study on cytosolic expression of the E. faecalis PDH complex in S. cerevisiae (31), growth of the resulting strain IMX745 (acs1⌬ acs2⌬::{PDHL} sga1⌬::{CARN}) on synthetic medium containing glucose depended on the addition of lipoic acid to the growth medium.
Enzyme activities in cell extracts of strain IMX745 showed a carnitine acetyltransferase (CAT) activity of 3.2 Ϯ 0.1 mol · mg protein Ϫ1 · min Ϫ1 , while activities in extracts of the parental strain IMX719 (acs1⌬ acs2⌬::{PDHL}) and of the reference strain IMX585 (ACS1 ACS2) were below the detection limit of the assay (Ͻ0.01 mol · mg protein Ϫ1 · min Ϫ1 ). Growth of strain IMX745 was not observed when lipoic acid was replaced by L-carnitine or when both growth factors were omitted from the glucosecontaining synthetic medium (Fig. 3). This result demonstrated that, even when constitutively expressed, the S. cerevisiae carnitine shuttle cannot export acetyl units from mitochondria at a rate that is sufficient to meet cytosolic acetyl-CoA requirements in an acs1⌬ acs2⌬ strain background.
Laboratory evolution yields mutants in which the carnitine shuttle provides cytosolic acetyl-CoA. To investigate whether laboratory evolution can enable the carnitine shuttle to support export of acetyl units from the mitochondrial matrix, a laboratory evolution experiment was started with strain IMX745 (Acs Ϫ {PDHL} {CARN}) by starting two independent shake flask cultures on synthetic medium with 20 g · liter Ϫ1 glucose and 400 mg · liter Ϫ1 L-carnitine (Fig. 2C). Following 2 weeks of incubation, growth was observed in both shake flasks, and after six or seven subsequent transfers (corresponding to ca. 70 generations), single-cell lines were isolated from each experiment, resulting in strains IMS0482 and IMS0483. These two evolved strains readily grew on glucose-containing synthetic medium supplemented with either lipoic acid or L-carnitine, but they did not grow when both compounds were omitted from the medium (Fig. 3). In

FIG 2
Construction of a lipoic acid-dependent, carnitine shuttle-constitutive S. cerevisiae strain and its laboratory evolution for lipoic acid-independent, carnitine-dependent growth. (A) In a previous study (33), the {PDHL} cluster, consisting of six cassettes required for cytosolic expression of a functional Enterococcus faecalis pyruvate dehydrogenase complex and flanked by 60-bp sequences, was assembled in vivo via homologous recombination (indicated with black crosses) and introduced in ACS2 after introduction of a Cas9-induced double-strand break. ACS1 was removed using a 120-bp DNA repair fragment (figure adapted from reference 33). (B) In this strain, the {CARN} cluster, consisting of six cassettes for constitutive expression of carnitine shuttle genes, was similarly in vivo assembled and introduced into the SGA1 locus, resulting in strain IMX745 (acs1⌬ acs2⌬::{PDHL} sga1⌬::{CARN}). Activity of the E. faecalis PDH in the yeast cytosol is lipoic acid dependent (31). (C) As strain IMX745 did not show L-carnitine-dependent growth when lipoic acid was omitted from growth media, an evolution experiment was initiated using synthetic medium with 20 g · liter Ϫ1 glucose (dextrose) (SMD) and 400 mg · liter Ϫ1 L-carnitine. Abbreviations: chrI, chromosome I; chrIX, chromosome IX; chrXII, chromosome XII.
shake flask cultures on glucose-containing synthetic medium, addition of L-carnitine supported specific growth rates of 0.14 h Ϫ1 (IMS0482) and 0.10 h Ϫ1 (IMS0483) ( Table 2). When the synthetic gene cluster encoding the E. faecalis PDH complex {PDHL} was removed from the evolved strains, growth of the resulting strains on glucose could no longer be supported by the addition of lipoic acid and, instead, became uniquely dependent on L-carnitine (Fig. 4). Conversely, deletion of the six carnitine shuttle expression cassettes {CARN} from the evolved strains abolished their L-carnitine-dependent growth, leaving the strains uniquely dependent on lipoic acid (Fig. 4). Together, these results unequivocally show that, in the evolved strains, export of the acetyl moiety of mitochondrially produced acetyl-CoA via the constitutively expressed carnitine shuttle supported cytosolic acetyl-CoA provision (Fig. 1C).
The mitochondrial PDH complex is the predominant source of acetyl-CoA in evolved, L-carnitine-dependent acs1⌬ acs2⌬ strains. In S. cerevisiae, mitochondrial acetyl-CoA can be generated by the native, mitochondrial PDH complex and by the mitochondrial succinyl-CoA:acetate CoA-transferase Ach1 (8,26,34). To study which of these reactions provided mitochondrial acetyl-CoA in the evolved strains IMS0482 and IMS0483, the mitochondrial PDH complex was inactivated by deleting PDA1 (35,36), and Ach1 activity was abolished by disrupting ACH1. In both evolved Van Rossum et al.
strains, deletion of PDA1 abolished L-carnitine-dependent growth on glucose, while ACH1 disruption did not have a detectable impact on growth (Fig. 5). These results demonstrate that, in glucose-grown batch cultures of the evolved strains, the S. cerevisiae PDH complex is the predominant source of mitochondrial acetyl-CoA and, via the constitutively expressed carnitine shuttle, of cytosolic acetyl-CoA. Whole-genome sequencing and reverse engineering of evolved L-carnitine-dependent strains. To identify the mutations that enabled L-carnitine-dependent growth of the evolved carnitine-dependent acs1⌬ acs2⌬ strains, the genomes of strains IMS0482 and IMS0483 (Acs Ϫ {PDHL} {CARN}, isolated from evolution lines 1 and 2, respectively) and of their parental strain IMX745 (Acs Ϫ {PDHL} {CARN}) were sequenced. Analysis of single-nucleotide changes and insertions/deletions (indels) in open reading frames revealed only three mutations in strain IMS0482 (evolution line 1) and four mutations in strain IMS0483 (evolution line 2) relative to the parental strain ( Table 3). Analysis of copy number variations (37) showed that strain IMS0482 carried a duplication of chromosome X (data not shown). Chromosome X did not carry either one of the two synthetic gene clusters or any of three mutated genes. No copy number variations relative to the parental strain were detected in strain IMS0483.
Both evolved strains carried mutations in MCT1, which is predicted to encode the mitochondrial malonyl-CoA:acyl carrier protein (ACP) transferase that catalyzes the second step of mitochondrial fatty acid synthesis (21,38,39). In strain IMS0482, the T-to-G change at position 641 encoded by MCT1 (MCT1 T641G ) caused an amino acid change from leucine to tryptophan at position 214, and in strain IMS0483, an MCT1 C292T mutation caused a premature stop codon at position 98. Strain IMS0482 carried an additional mutation in RTG2, which resulted in a W168L amino acid change. Rtg2 is involved in communication between mitochondria and the nucleus, and deletion of RTG2 negatively affects activity of citrate synthase (oxaloacetate ϩ acetyl-CoA ϩ H 2 O ¡ citrate ϩ CoA; 40,41). A third mutation in strain IMS0482 was found in the introduced expression cassette for YAT2, which has been reported to encode a cytosolic carnitine acetyltransferase (15) and caused a P58R amino acid change in the evolved strain. In strain IMS0483, the abovementioned MCT1 C292T mutation was accompanied by single-nucleotide changes in the coding regions of RPO21 and STB2 and a deletion of either HXT6 or HXT7. Since the protein products of these three genes did not show an obvious relation with mitochondrial metabolism (Table 3), further analysis was focused on the mutations found in strain IMS0482 which, moreover, exhibited the highest specific growth rate on glucose of the two evolved strains ( Table 2).
Mutations in MCT1, RTG1, and YAT2 together enable in vivo reversal of the mitochondrial carnitine shuttle. To investigate their biological relevance, the three mutations found in evolved strain IMS0482 were introduced individually and in different combinations into the nonevolved parental strain IMX745 (Acs Ϫ {PDHL} {CARN}). As expected, all resulting strains grew on synthetic medium with glucose and lipoic acid. However, on solid medium, only strains IMX909 (Mct1 L214W Rtg2 Yat2 P58R ) and IMX913 (Mct1 L214W Rtg2 W168L Yat2 P58R ) showed L-carnitine-  dependent growth (Fig. 6), suggesting that both Mct1 L214W and Yat2 P58R were essential for the acquired phenotype. On spot plates, no clear impact of the mutation in RTG2 was observed after 100 h of incubation (Fig. 6). For a quantitative analysis of the impact of the Rtg2 W168L mutation on specific growth rates, strains IMX909 (Mct1 L214W Rtg2 Yat2 P58R ) and IMX913 (Mct1 L214W Rtg2 W168L Yat2 P58R ) were grown in shake flask cultures on synthetic medium with glucose and L-carnitine (Table 2 and Fig. 7). Strain IMX909 showed decelerating exponential growth rates of 0.10 h Ϫ1 to 0.06 h Ϫ1 , while strain IMX913 exhibited monophasic exponential growth at a specific growth rate of 0.14 h Ϫ1 , which resembled the specific growth rate of evolved strain IMS0482 ( Fig. 7). This result showed that all three mutations in the laboratory-evolved strain IMS0482 contributed to its acquired phenotype. Exponentially growing cultures of the reverse engineered strain IMX913 on synthetic medium with glucose and L-carnitine exhibited a high viability (Ͼ99%), resembling that of the reference strain IMX585.
To investigate whether the mutations in MCT1, RTG2, and YAT2, acquired by strain IMS0482 during laboratory evolution, might have caused a complete loss of function, three Acs Ϫ {PDHL} {CARN} strains were constructed in which deletion of one of the three genes was combined with the acquired point mutations of the remaining two genes. The three resulting strains, synthetic medium with 20 g · liter Ϫ1 glucose, supplemented with lipoic acid (strains IMW074 and IMW076) or L-carnitine (strains IMW075 and IMW077) and spotted on plates containing synthetic medium with glucose (dextrose) without lipoic acid or L-carnitine (SMD), with lipoic acid (SMD lipoate) and with L-carnitine (SMD carnitine). The plates were incubated for 100 h at 30°C. Relevant strain descriptions are given in the figure. Photographs of the entire spot plates are shown in Data Set S1 in the supplemental material. IMX932, IMX933, and IMX934, all showed growth after 100-h incubation on solid medium with glucose and lipoic acid (Fig. 6). (Fig. 6). This result indicated that the amino acid changes in the Mct1 L214W and Yat2 P58R variants did not result in complete loss of function. Interestingly, the genetic context of the other evolved strain IMS0483, in which MCT1 contained a premature stop codon, did appear to enable carnitine-dependent growth in the absence of a functional Mct1 protein. The slightly lower L-carnitine-dependent growth of strain IMX933 (Acs Ϫ {PDHL} {CARN,Yat2 P58R } Mct1 L214W rtg2⌬) compared to a congenic strain expressing the mutant Rtg2 W168L variant, suggests that this amino acid change does not lead to a completely nonfunctional protein.
Enzyme assays do not confirm carnitine acetyltransferase activity of Yat2. The prior classification of Yat2 as a cytosolic carnitine acetyltransferase (20,21,24) was based on its homology with other carnitine acetyltransferase genes and on a reported 50% decrease of carnitine acetyltransferase activity (not normalized for protein content) in cell extracts of ethanol-grown cultures of a yat2⌬ strain (15). To compare carnitine acetyltransferase activities of Yat2 and Yat2 P58R , YAT2 and YAT2 C173G genes under control of the constitutive ADH1 promoter were introduced in reference genetic backgrounds. Since the native YAT1, YAT2, and CAT2 carnitine acetyltransferases are repressed by glucose, enzyme assays on cell extracts of glucose-grown batch cultures should reflect activity of only these constitutively expressed YAT2 genes. Surprisingly, no detectable (Ͻ0.01 mol · mg protein Ϫ1 · min Ϫ1 ) carnitine acetyltransferase activity was found in such experiments with strains expressing the wild-type YAT2 or evolved alleles of YAT2 from single-copy or multicopy, pADH1-controlled expression cassettes ( Table 4). The same negative results were obtained with the carnitine acetyltransferase assay procedure described by Swiegers et al. (15). In contrast, strains IMX868 (sga1⌬::{CARN}) and IME233 (multicopy plasmid with constitutively expressed CAT2) showed high activities (Table 4). To exclude the theoretical possibility that Yat2 is subject to glucose catabolite inactivation, a yat1⌬ cat2⌬ YAT2 strain (CEN.PK215-4A) was constructed and subsequently tested under glucosederepressed, respiratory growth conditions. However, in ethanolgrown cultures of this strain, the Yat2-dependent carnitine acetyltransferase activity remained below the detection limit. Under the same conditions, the reference strain CEN.PK113-7D showed a carnitine acetyltransferase activity of 1.75 mol · mg protein Ϫ1 · min Ϫ1 (Table 4).
Possible explanations for our inability to detect Yat2dependent carnitine acetyltransferase activity include the following. (i) Yat2 is active within a heteromeric complex only when another carnitine acetyltransferase is present. (ii) Yat2 is a catalytically inactive regulator of other carnitine acetyltransferases. (iii) Assay conditions and/or Yat2 protein instability preclude accurate measurement of in vitro Yat2 carnitine acetyltransferase activity. In the first two scenarios, the mutated form of Yat2 might still show a detectable impact on total carnitine acetyltransferase activity.  (Table 4).

DISCUSSION
Requirements for reversal of the mitochondrial carnitine shuttle. To our knowledge, this study is the first to demonstrate that the carnitine shuttle can connect the mitochondrial acetyl-CoA pool to cytosolic, acetyl-CoA-consuming pathways in a eukaryote. Three requirements had to be met to enable export of acetyl units from mitochondria of glucose-grown S. cerevisiae. L-Carnitine, which cannot be synthesized by S. cerevisiae (9,15), needed to be added to growth media. Furthermore, glucose repression of key genes encoding carnitine shuttle proteins had to be circumvented, which in this study was done by expression from constitutive promoters. While these first two criteria also have to be met to enable the carnitine shuttle to effectively import acetyl units into mitochondria (8,9,11,15), its operation in the reverse direction additionally required mutations in the yeast genome. Single-amino-acid changes in three proteins (Mct1 L214W , Rtg2 W168L , and Yat2 P58R ) together enabled export of acetyl units from mitochondria via a constitutively expressed carnitine shuttle. Mct1 is predicted to encode mitochondrial malonyl-CoA:ACP transferase (38), which is required for mitochondrial fatty acid synthesis. This process uses mitochondrial acetyl-CoA as a precursor and might therefore compete for this substrate with the carnitine shuttle. Mct1 uses malonyl-CoA, formed by the mitochondrial acetyl-CoA carboxylase Hfa1 (42), rather than acetyl-CoA, as a substrate. Inhibition of Hfa1 by malonyl-CoA, a property shared by several acetyl-CoA carboxylases (43,44), could decrease its ability to compete for acetyl-CoA when Mct1 functions suboptimally. Rtg2, a sensor protein involved in the retrograde regulation pathway for nuclear-mitochondrial communication (40), was previously shown to affect levels of mitochondrial citrate synthase (41), which also uses mitochondrial acetyl-CoA as a substrate. We therefore propose that, in the evolved strains, mutations in MCT1 and RTG2 improved the driving force and/or kinetics of the export of acetyl units via the mitochondrial carnitine shuttle by negatively affecting pathways that compete for its substrate, intramitochondrial acetyl-CoA.
Mutations in mitochondrial lipid synthesis were previously shown to affect carnitine shuttle activity in human cells. When mitochondrial ␤-oxidation of fatty acids in human cells is compromised, acyl-carnitines are exported from the mitochondria to the cytosol and can even be found in blood plasma (45,46). Especially when yeast carnitine shuttle genes can be functionally replaced by their human orthologs (47), the L-carnitine-dependent strains described in this study provide interesting platforms for studying the role of the carnitine shuttle in healthy and diseased human cells. Many eukaryotes use a citrate-oxaloacetate shuttle, consisting of mitochondrial citrate synthase, a mitochondrial citrate transporter, and cytosolic ATP-dependent citrate lyase, for export of acetyl units from their mitochondria (48)(49)(50). Conversion of mitochondrial acetyl-CoA to acetate, followed by its export and cytosolic ATP-dependent activation to acetyl-CoA, occurs in Trypanosoma brucei (51). The latter mechanism also supports slow growth of pyruvate decarboxylase-negative S. cerevisiae mutants, which cannot use the PDH bypass for cytosolic acetyl-CoA synthesis (52). The ATP requirement of these naturally occurring acetyl-CoA shuttles is consistent with our hypothesis that in vivo concentrations of acetyl-CoA in cytosol and mitochondria of wild-type yeast cells do not allow outward translocation of acetyl units via the energy-independent carnitine shuttle. Quantification of trade-offs between ATP efficiency and in vivo kinetics of cyto-  solic acetyl-CoA provision via different pathways requires analysis of mitochondrial and cytosolic acetyl-CoA pools in wild-type and engineered strains. Such studies will, however, have to await development of techniques for accurate measurement of acetyl-CoA concentrations in different cellular compartments. YAT2, the third gene in which a point mutation stimulated carnitine-dependent growth of acs1⌬ acs2⌬ strains, was reported to encode a carnitine acetyltransferase (15). Yat2 shows substantial sequence identity with the two other yeast carnitine acetyltransferases (28% and 22% amino acid sequence identity with Yat1 and Cat2, respectively [53]). However, Yat2 is substantially longer than Yat1 and Cat2, by 236 and 253 amino acids, respectively, and its 169-amino-acid C-terminal sequence is conserved only in some closely related orthologs within the Saccharomycetaceae (54). The mutation in YAT2 is intriguing because Cat2 (active in the mitochondrial and peroxisomal matrices) and Yat1 (active in the cytosol) should in theory suffice to form a functional mitochondrial carnitine shuttle. Prompted by its essential role in reversal of the mitochondrial carnitine shuttle in evolved strain IMS0482, we sought to compare enzyme kinetics of wild-type Yat2 and Yat2 P58R . Our inability to detect activity of either Yat2 isoform in cell extracts does not rule out the possibility that these proteins are carnitine acetyltransferases. Combined with the impact of a mutation in YAT2 on in vivo carnitine shuttle activity, this result underlines the need for further biochemical characterization of Yat2.
(Energetic) implications of the carnitine shuttle in cytosolic acetyl-CoA provision for biotechnological applications. In the native S. cerevisiae pathway for cytosolic acetyl-CoA synthesis, cytosolic acetate is activated by the Acs1 and/or Acs2 acetyl-CoA synthetases (2,26,55,56). This activation involves hydrolysis of ATP to AMP and pyrophosphate which, when pyrophosphate is subsequently hydrolyzed to inorganic phosphate, is equivalent to the hydrolysis of 2 mol of ATP to ADP and inorganic phosphate. Cytosolic acetyl-CoA is an important precursor for many industrially relevant compounds, and much effort has been invested in metabolic engineering of alternative, more-ATP-efficient pathways for cytosolic acetyl-CoA supply into S. cerevisiae. Examples of such strategies include cytosolic expression of heterologous phosphoketolase and phosphotransacetylase, acetylating acetaldehyde dehydrogenase, pyruvate-formate lyase, and a heterologous pyruvate dehydrogenase complex (31,57,58). The present study demonstrates that reversal of the mitochondrial carnitine shuttle can directly link acetyl-CoA synthesis via the mitochondrial PDH complex, the predominant source of acetyl-CoA in aerobic, glucose-grown S. cerevisiae cultures (36), to provision of cytosolic acetyl-CoA. The low specific growth rates of the evolved and reverse engineered L-carnitine-dependent strains indicate that this novel strategy for engineering cytosolic acetyl-CoA provision in S. cerevisiae requires optimization before industrial implementation can be considered. Progress in this direction would provide a strong incentive to engineer a complete L-carnitine biosynthesis pathway in S. cerevisiae. Despite recent advances (59), synthesis of the key precursor trimethyl-lysine in S. cerevisiae remains an important metabolic engineering challenge.
Export of acetyl units from mitochondria via the carnitine shuttle may also be relevant for eukaryotic cell factories other than S. cerevisiae. Oleaginous eukaryotes such as the yeast Yarrowia lipolytica employ the mitochondrial PDH complex and a citrateoxaloacetate shuttle to provide cytosolic acetyl-CoA for lipid syn-thesis (49,60). The citrate-oxaloacetate shuttle requires 1 ATP for each molecule of mitochondrial pyruvate converted into cytosolic acetyl-CoA. Eliminating this ATP requirement could further improve the ATP efficiency of lipid synthesis and, consequently, the lipid yield in oleaginous eukaryotes.
Outlook. By demonstrating in vivo reversibility of the mitochondrial carnitine shuttle, a ubiquitous mechanism in eukaryotes, this study provides new leads for investigating and understanding the role of this shuttle in yeast and other eukaryotes. The "switchable" L-carnitine-dependent yeast strains described here provide valuable experimental platforms for functional analysis of the native yeast carnitine shuttle, for heterologous complementation studies on carnitine shuttle components from other eukaryotes, and for engineering of a complete L-carnitine biosynthesis pathway into S. cerevisiae (59). After further optimization of the kinetics, the "reverse" mitochondrial carnitine shuttle offers a potential new strategy for energetically efficient synthesis of cytosolic acetyl-CoA as a precursor for a wide range of biotechnologically relevant compounds by eukaryotic cell factories.

MATERIALS AND METHODS
Growth media. Yeast extract-peptone (YP) medium contained 10 g · liter Ϫ1 Bacto yeast extract (BD, Franklin Lakes, NJ, USA) and 20 g · liter Ϫ1 Bacto peptone (BD) in demineralized water. Synthetic medium with ammonium as the nitrogen source (SM-ammonium) was prepared by the method of Verduyn et al. (61). Synthetic medium with urea as the nitrogen source (SM-urea) contained 38 mM urea and 38 mM K 2 SO 4 instead of (NH 4 ) 2 SO 4 . SM-ammonium was autoclaved at 121°C for 20 min, and SM-urea was sterilized using 0.2-m bottle-top filters (Thermo Fisher Scientific, Waltham, MA, USA). Solid media were prepared by the addition of 20 g · liter Ϫ1 agar (BD), prior to autoclaving at 121°C for 20 min. Where indicated, urea was added after heat sterilization of the solid media from a filter-sterilized 100-fold-concentrated stock solution.
Strains, growth conditions, and storage. All S. cerevisiae strains used in this study (Table 1) share the CEN.PK genetic background (62,63). Shake flask cultures in 500-ml flasks with 100 ml SM-urea and 20 g · liter Ϫ1 glucose were grown at 30°C in an Innova incubator shaker (New Brunswick Scientific, Edison, NJ, USA) set at 200 rpm. Stock cultures were grown in YP medium with 20 g · liter Ϫ1 glucose. Where indicated, lipoic acid was added to sterile media to a concentration of 50 ng · liter Ϫ1 . A 50-mg · liter Ϫ1 stock solution of lipoic acid was prepared by dissolving 5 g · liter Ϫ1 (Ϯ)-␣-lipoic acid (Sigma-Aldrich, St. Louis, MO, USA) in ethanol and diluting the resulting solution 100-fold in sterile demineralized water. L-Carnitine (Sigma-Aldrich) was added to sterile media from a 40-g · liter Ϫ1 filter-sterilized stock solution at the concentration indicated. Frozen stock cultures of yeast strains were prepared by adding glycerol (30%, vol/vol) to exponentially growing shake flask cultures and freezing 1-ml aliquots at Ϫ80°C.
Plasmid construction. Guide RNA (gRNA) plasmids for clustered regularly interspaced short palindromic repeat (CRISPR)/Cas9-based genome editing (see Table S1 in the supplemental material) were constructed as described previously (33). In short, double-gRNA cassettes were PCR amplified using the primer(s) indicated in Tables S1 and S2. Plasmid backbones containing the desired marker gene were obtained by PCR with primer 6005, using the appropriate pROS plasmid (Table S1) as a template. The two fragments were then assembled into a plasmid with the Gibson Assembly kit (New England Biolabs, Ipswich, MA, USA) or NEBuilder HiFi DNA assembly cloning kit (New England Biolabs). Multicopy plasmids carrying wild-type YAT2 and mutated YAT2 variants were based on the pRS426 expression vector (64). pADH1-YAT2-tYAT2 and pADH1-YAT2 C173G -tYAT2 fragments were PCR amplified from strains IMX745 and IMS0482, respectively, using primers 8902 and 8903 (sequences of these cassettes are presented in Table S3) and then inserted into the EcoRI-XhoI-linearized pRS426 backbone with the NEBuilder Van Rossum et al.
HiFi DNA assembly cloning kit. After transforming the resulting plasmids to Escherichia coli and confirmation of their DNA sequences by Illumina sequencing, this yielded pUDE390 (2m ori URA3 pADH1-YAT2-tYAT2) and pUDE391 (2m ori URA3 pADH1-YAT2 C173G -tYAT2). A multicopy plasmid carrying the CAT2 gene under control of the TDH3 promoter was similarly obtained by assembling a pRS426 backbone with a CAT2 PCR fragment using the Gibson Assembly kit. The TDH3 promoter and CYC1 terminator sequences were synthesized and assembled into the pRS426 vector by GenScript (Piscataway, NJ, USA). The resulting plasmid was linearized by PCR amplification using primers 3627 and 3921. The CAT2 open reading frame (ORF) was amplified via PCR from S. cerevisiae CEN.PK113-7D genomic DNA using primers 5948 and 5949. Gibson Assembly of the two fragments yielded pUDE336 (2m ori URA3 pTDH3-CAT2-His 6 -tCYC1). The DNA sequence of the pTDH3-CAT2-His 6 -tCYC1 cassette is presented in Table S3.
Strain construction. S. cerevisiae strains were transformed by the method of Gietz and Woods (65), and transformants were selected on solid YP medium with 20 g · liter Ϫ1 glucose. Appropriate antibiotics were added at the following concentrations: G418 (InvivoGen, San Diego, CA, USA), 200 mg · liter Ϫ1 ; hygromycin B (InvivoGen), 200 mg · liter Ϫ1 ; nourseothricin (Jena Bioscience, Jena, Germany), 100 mg · liter Ϫ1 . Lipoic acid was added as indicated above. Throughout the text we refer to chromosomally integrated gene clusters with four-capital acronyms surrounded by curly brackets (based on the common practice in set theory for indicating a collection of elements). A mutation in a gene that is part of the cluster is indicated within the curly brackets. For example, {CARN,YAT2 C173G } refers to the {CARN} set in which the YAT2 gene carries a C173G nucleotide change.
Unless indicated otherwise, genetic engineering was done using CRISPR/Cas9 (33). The platform strain with constitutive expression of the genes involved in the carnitine shuttle (HNM1, AGP2, CRC1, YAT1, YAT2, and CAT2) was constructed by modification of the previously constructed strain IMX719 (33), which had ACS1 and ACS2 replaced by the genes required for an active, lipoylated cytosolic Enterococcus faecalis PDH complex {PDHL}. Analogous to a previous description (8), the genes involved in the carnitine shuttle were placed under the control of strong constitutive promoters and integrated into the SGA1 locus of strain IMX719, resulting in strain IMX745 (acs1⌬ acs2⌬::{PDHL} sga1⌬:: {CARN}) ( Table 1). To remove the E. faecalis PDH genes {PDHL} or the set of carnitine shuttle expression cassettes {CARN} from strains IMS0482 and IMS0483, either plasmid pUDR072 (to remove {PDHL}) or pUDR073 (to remove {CARN}) was transformed together with a repair fragment obtained by annealing oligonucleotides 7349 and 7350 or oligonucleotides 8012 and 8013 (see Table S2 in the supplemental material), respectively, resulting in strains IMW074 to IMW077. Deletion of PDA1 and ACH1 in strains IMS0482 and IMS0483 was done by transformation with pUDR047 (with oligonucleotides 6157 and 6158) and pUDR085 (with oligonucleotides 6160 and 6161), resulting in strains IMW078 to IMW082. To introduce the MCT1 T641G mutation, plasmid pUDR080 and a repair fragment obtained by annealing oligonucleotides 8417 and 8418 was transformed into strain IMX745 (Table 1), resulting in strain IMX847. Similarly, the RTG2 G503T mutation was introduced in strain IMX745 by transforming plasmid pUDR078 and oligonucleotides 8430 and 8431, resulting in strain IMX849. The MCT1 T641G RTG2 G503T double mutations were introduced in strain IMX745 using plasmid pUDR079 using oligonucleotides 8417, 8418, 8430, and 8431, resulting in strain IMX852. To selectively introduce the YAT2 C173G mutation in the ADH1 promoter-driven gene, not in the YAT2-promoter driven gene (at chromosome V), the single-nucleotide polymorphism (SNP) was introduced in {CARN} via a two-step strategy. First, a synthetic CRISPR target site was introduced by transformation of strains IMX745, IMX847, IMX849, and IMX852 with plasmid pUDR073 and oligonucleotides 8621 and 8622, thereby removing part of the ADH1 promoter and part of the YAT2 ORF. Next, the fragment containing the YAT2 C173G mutation was PCR amplified from the IMS0482 genome using primers 8618 and 8619 and cotrans-formed with plasmid pUDR105, introducing the YAT2 C173G mutation and resulting in strains IMX907, IMX909, IMX911, and IMX913. In all these cases, after introduction of the desired mutations, the double-gRNA plasmids were removed, followed by confirmation of the SNPs by Sanger sequencing (BaseClear BV, Leiden, The Netherlands) using the primers indicated in Table S2. The ORFs of YAT2 (the copy present in {CARN}), RTG2, and MCT1 were deleted from the genomes of strains IMX852, IMX909, and IMX911, respectively, by transforming the following plasmids and repair fragments: for strain IMX852, plasmid pUDR073 and oligonucleotides 8874 and 8875; for strain IMX909, plasmid pUDR078 and oligonucleotides 8428 and 8429; and for strain IMX911, plasmid pUDR080 and oligonucleotides 8415 and 8416. After gene knockout was confirmed by diagnostic PCR (Table S2), the resulting strains were named IMX932 to IMX934, respectively.
The pADH1-YAT2-tYAT2 variants were integrated in the cas9-bearing reference strain IMX585. pADH1-YAT2-tYAT2 (wild-type) and pADH1-YAT2 C173G -tYAT2 cassettes were amplified with PCR using primers 8647 and 8648 from genomic DNA of strains IMX745 and IMS0482, respectively. The resulting cassettes had overlaps with the promoter and terminator of SGA1, enabling integration into the SGA1 locus. Cas9 was directed to the SGA1 locus using the gRNA plasmid pUDR119 (see Table S1 in the supplemental material), following integration of the cassette by in vivo homologous recombination. After confirmation of correct integration and sequence by PCR and Sanger sequencing, plasmid pUDR119 was removed as described earlier (33), resulting in strains IMX923 and IMX925, respectively. To obtain the multicopy-based YAT2and CAT2-expressing strains, plasmids pUDE336, pUDE390, and pUDE391 were transformed to strain CEN.PK113-5D, resulting in strains IME233, IME320, and IME321, respectively (Table 1).
To obtain strain CEN.PK215-4A (cat2⌬ yat1⌬), CAT2 and YAT1 were deleted by transformation of a kanMX marker cassette, obtained by PCR using pUG6 as the template (66) and primers 9237 and 9238 for the CAT2 deletion cassette and primers 9239 and 9240 for the YAT1 deletion cassette. The amplified kanMX cassettes were used as selectable markers to replace the target genes in the prototrophic diploid strain CEN.PK122. Transformants were verified for correct gene replacement by diagnostic PCR (see Table S2 in the supplemental material). After sporulation and tetrad dissection, the corresponding haploid deletion strains, CEN.PK194-2C (MATa cat2⌬) and CEN.PK196-2C (MAT␣ yat1⌬), were obtained. To obtain a strain with both CAT2 and YAT1 deleted, strains CEN.PK194-2C and CEN.PK196-2C were crossed. After tetrad dissection, spores were subsequently analyzed by diagnostic PCR to confirm correct deletion of both genes, resulting in strain CEN.PK215-4A (cat2⌬ yat1⌬) ( Table 1).
Molecular biology techniques. PCR amplification with the Phusion Hot Start II high-fidelity polymerase (Thermo Fisher Scientific) was performed according to the manufacturer's instructions, using highperformance liquid chromatography (HPLC)-or polyacrylamide gel electrophoresis (PAGE)-purified oligonucleotide primers (Sigma-Aldrich). Diagnostic colony PCR was performed on randomly picked transformed colonies, using DreamTaq (Thermo Fisher Scientific) and desalted primers (Sigma-Aldrich). DNA fragments obtained by PCR were separated by gel electrophoresis on 1% (wt/vol) agarose gels (Thermo Fisher Scientific) in TAE (Tris-acetate-EDTA) buffer (Thermo Fisher Scientific). Alternatively, fragments were purified using the GenElute PCR cleanup kit (Sigma-Aldrich). Plasmids were isolated from E. coli with Sigma GenElute plasmid kit (Sigma-Aldrich) according to the supplier's manual. Yeast genomic DNA was isolated using a YeaStar genomic DNA kit (Zymo Research) or using a sodium dodecyl sulfate/lithium acetate-based lysis protocol (67). E. coli XL1-Blue (GE Healthcare Life Sciences, The Netherlands) was used for chemical transformation or for electroporation. Chemical transformation was conducted by the method of Inoue et al. (68). Electroporation was performed in a 2-mm cuvette (catalog no. 1652086; Bio-Rad, Hercules, CA, USA) using a Gene Pulser Xcell electroporation system (Bio-Rad), following the manufacturer's protocol. Elec-trocompetent E. coli cells were prepared according to the same protocol, with the exception that, during preparation of competent cells, E. coli was grown in LB medium without sodium chloride. Laboratory evolution. Strain IMX745 was inoculated in 500-ml shake flasks containing 100 ml SM-urea with 20 g · liter Ϫ1 glucose and 400 mg · liter Ϫ1 L-carnitine. When stationary phase was reached, 1 to 3 ml of culture was transferred to a new shake flask. After six or seven serial shake flask transfers, eight individual cells were isolated from each evolution experiment using a micromanipulator (Singer Instruments, Watchet, United Kingdom) and placed on SM-urea plates with 20 g · liter Ϫ1 glucose and 400 mg · liter Ϫ1 L-carnitine. For each evolution experiment, one colony was selected and restreaked once, yielding strains IMS0482 (evolution line 1) and IMS0483 (evolution line 2) ( Table 1).
DNA sequencing and sequence analysis. After isolation of genomic DNA (69) from strains IMX745, IMS0482, and IMS0483, 350-bp insert libraries were constructed and paired-end sequenced (100-bp reads) with an Illumina HiSeq 2500 sequencer (Baseclear BV, Leiden, The Netherlands). At least 500 Mb of sequence data, corresponding to a ca. 40-fold coverage, was generated for each strain. Plasmids pUDE390 and pUDE391 were sequenced in-house using the Illumina MiSeq platform (San Diego, CA, USA). After quantification of plasmid DNA with the Qubit 2.0 fluorometer (Thermo Fisher Scientific), DNA libraries were prepared using the Nextera XT DNA kit (Illumina). Paired-end reads (300 bp) of plasmid DNA generated on the MiSeq platform were mapped to an in silico-generated plasmid sequence using the Burrows-Wheeler alignment tool (70) and processed with Pilon (71). Sequence reads of genomic DNA were mapped onto the CEN.PK113-7D genome (63), supplemented with sequences containing the modified SGA1, ACS2, and CAN1 loci, using the Burrows-Wheeler alignment tool (70). Data were further processed with Pilon (71), and sequence variations were extracted from the Pilon output file ".changes." The uniqueness of sequence differences in strains IMS0482 and IMS0483 was manually confirmed by comparison with strain IMX745 using the Integrative Genomics Viewer (72). Copy number variations in strains IMS0482 and IMS0483, relative to strain IMX745, were determined with the Poisson mixture model-based algorithm Magnolya (37).
Growth studies in shake flasks and using spot plate assays. For growth studies in shake flasks and using spot plates, strains were pregrown in shake flasks with SM-urea and 20 g · liter Ϫ1 glucose with lipoic acid or L-carnitine, where appropriate. For growth studies in shake flasks, cells were washed twice with synthetic medium (61) and transferred to new shake flasks with SM-urea containing 20 g · liter Ϫ1 glucose and 40 mg · liter Ϫ1 L-carnitine or 50 ng · liter Ϫ1 lipoic acid, where indicated. Growth rates were based on optical density at 660 nm (OD 660 ) measurements using a Libra S11 spectrophotometer (Biochrom, Cambridge, United Kingdom). Culture viability was estimated with the FungaLight AM-CFDA (acetoxymethyl ester 5-carboxyfluorescein diacetate)/propidium iodide yeast viability kit (Invitrogen, Carlsbad, CA) and a Cell Lab Quanta SC MPL flow cytometer (Beckman Coulter, Woerden, The Netherlands) as described previously (73). For the preparation of spot plates, precultures were washed once with synthetic medium and diluted in synthetic medium to an OD 660 of 0.273 (corresponding to 2 ϫ 10 6 cells · ml Ϫ1 ). Five-microliter samples of a dilution series, containing an estimated 2 ϫ 10 5 , 2 ϫ 10 4 , and 2 ϫ 10 3 cells per ml, were spotted on SM-urea agar plates with 20 g · liter Ϫ1 glucose and L-carnitine (400 mg · liter Ϫ1 ) or lipoic acid (50 ng · liter Ϫ1 ) as indicated.
Enzyme activity assays. Cell extracts were prepared as described before (8) from mid-exponentially growing cultures. The growth medium was SM-ammonium with either 20 g · liter Ϫ1 glucose or 2% (vol/vol) ethanol as the carbon source and, where required, lipoic acid. Activities in cell extracts of carnitine acetyltransferase activity (8) and glucose-6phosphate dehydrogenase (74) (the latter activity was used to verify the quality of cell extracts) were assayed spectrophotometrically as described previously (8). Protein concentrations in cell extracts were determined by the Lowry method (75).
Nucleotide sequence accession number. Raw sequencing data of strains IMX745, IMS0482, and IMS0483 are deposited at the NCBI Sequence Read Archive (http://www.ncbi.nlm.nih.gov/sra) under Bio-Project identifier (ID) or accession number PRJNA313402.

ACKNOWLEDGMENTS
We thank Peter Kötter, Annabel Giezekamp, Marlous van Dijk, Henri Duine, Ioannis Papapetridis, and Xavier Hakkaart for help in strain construction and growth studies. Pilar de la Torre and Melanie Wijsman are gratefully acknowledged for sequencing plasmids pUDE320 and pUDE321. Marcel van den Broek and Thomas Abeel are thanked for their help with sequence analysis. We declare no conflicts of interest related to the results described in this study.

FUNDING INFORMATION
This work, including the efforts of Harmen M. van Rossum, Barbara U. Kozak, Matthijs S. Niemeijer, James C. Dykstra, Marijke A.H. Luttik, Jean-Marc G. Daran, Antonius J.A. van Maris, and Jack T. Pronk, was funded by BE-Basic.
The PhD projects of Harmen van Rossum and Barbara Kozak were financially supported by the BE-Basic R&D Program, which was granted an FES subsidy from the Dutch Ministry of Economic Affairs, Agriculture and Innovation (EL&I) and received additional financial contributions from DSM Biotechnology Center and Amyris Inc. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.