Adaptive Laboratory Evolution and Reverse Engineering of Single-Vitamin Prototrophies in Saccharomyces cerevisiae

Many strains of Saccharomyces cerevisiae, a popular platform organism in industrial biotechnology, carry the genetic information required for synthesis of biotin, thiamine, pyridoxine, para-aminobenzoic acid, pantothenic acid, nicotinic acid, and inositol. However, omission of these B vitamins typically leads to suboptimal growth. This study demonstrates that, for each individual B vitamin, it is possible to achieve fast vitamin-independent growth by adaptive laboratory evolution (ALE). Identification of mutations responsible for these fast-growing phenotypes by whole-genome sequencing and reverse engineering showed that, for each compound, a small number of mutations sufficed to achieve fast growth in its absence. These results form an important first step toward development of S. cerevisiae strains that exhibit fast growth on inexpensive, fully supplemented mineral media that only require complementation with a carbon source, thereby reducing costs, complexity, and contamination risks in industrial yeast fermentation processes.

generation of highly reproducible data, independent variation of the concentrations of individual nutrients, and, in applied settings, design of balanced media for highbiomass-density cultivation and application of defined nutrient limitation regimes (1,2). CDMY such as yeast nitrogen base (YNB) and Verduyn medium are widely used in research on Saccharomyces yeasts (2,3). In addition to carbon, nitrogen, phosphorous, and sulfur sources and metal salts, these media contain a set of seven growth factors: biotin (B 7 ), nicotinic acid (B 3 ), inositol (B 8 ), pantothenic acid (B 5 ), para-aminobenzoic acid (pABA) (formerly known as B 10 ), pyridoxine (B 6 ), and thiamine (B 1 ). Based on their water solubility and roles in the human diet, these compounds are all referred to as B vitamins, but their chemical structures and cellular functions are very different (4). Taking into account their roles in metabolism, they can be divided into three groups: (i) enzyme cofactors (biotin, pyridoxine, and thiamine), (ii) precursors for cofactor biosynthesis (nicotinic acid, pABA, and pantothenic acid), and (iii) inositol, which is a precursor for phosphatidylinositol and glycosylphosphatidylinositol anchor proteins (5).
Previous studies demonstrated that Saccharomyces species are bradytroph for some B vitamins; growth does not strictly depend on addition of all of these compounds, but their omission from CDMY typically results in reduced specific growth rates (6)(7)(8). These observations imply that the term "vitamin," which implies a strict nutritional requirement, is in many cases formally incorrect when referring to the role of these compounds in Saccharomyces cerevisiae metabolism (5). In view of its widespread use in yeast physiology, we will nevertheless use it in this paper.
The observation that Saccharomyces yeasts can de novo synthesize some or all of the "B vitamins" included in CDMY is consistent with the presence of structural genes encoding the enzymes required for their biosynthesis ( Fig. 1 [5]). However, as illustrated by recent studies on biotin requirements of S. cerevisiae CEN.PK113-7D (5,9), a full complement of biosynthetic genes is not necessarily sufficient for fast growth in the absence of an individual vitamin. In the absence of biotin, this strain grew extremely slowly ( Ͻ 0.01 h Ϫ1 ), but fast biotin-independent growth was obtained through prolonged adaptive laboratory evolution (ALE) in a biotin-free CDMY. Reverse engineering of mutations acquired by evolved strains showed that, along with mutations in the plasma membrane transporter genes TPO1 and PDR12, a massive amplification of BIO1 was crucial for fast biotin-independent growth of evolved strains (10). These results illustrated the power of ALE in optimizing microbial strain performance without a priori knowledge of critical genes or proteins and in unravelling the genetic basis for  (5). Vitamins that are usually added to the chemical defined media for cultivation of yeasts are shown in light blue.
Elimination of vitamin requirements could enable cost reduction in the preparation of defined industrial media, and fully prototrophic strains could provide advantages in processes based on feedstocks whose preparation requires heating and/or acid treatment steps (e.g., lignocellulosic hydrolysates [13,14]) that inactivate specific vitamins. In addition, processes based on vitamin-independent yeast strains may be less susceptible to contamination by vitamin-auxotrophic microorganisms such as lactic acid bacteria (15). Thus, chassis strains able to grow fast in the absence of single or multiple vitamins would therefore be of interest for industrial application. Moreover, engineering strategies aimed at enabling fast growth and product formation in the absence of single or multiple vitamins may be relevant for large-scale industrial application of Saccharomyces yeasts.
The goals of the present study were to investigate whether single-vitamin prototrophy of S. cerevisiae for inositol, nicotinic acid, pantothenic acid, pABA, pyridoxine or thiamine could be achieved by ALE and to identify mutations that support fast growth in the absence of each of these vitamins. To this end, the laboratory strain S. cerevisiae CEN.PK113-7D was subjected to parallel aerobic ALE experiments that encompassed serial transfer in different synthetic media, which each lacked a single B vitamin. Independently evolved strains from each medium type were then characterized by whole-genome resequencing and the relevance of selected identified mutations was assessed by their reverse engineering in the parental non-evolved strain.
(This article was submitted to an online preprint archive [16].)

RESULTS
Assessment of CEN.PK113-7D specific B vitamin requirements. S. cerevisiae strains belonging to the CEN.PK lineage, which was developed in an interdisciplinary project supported by the German Volkswagen Stiftung between 1993 and 1994 (17), exhibit properties that make them good laboratory models for yeast biotechnology (18). To provide a baseline for ALE experiments, specific growth rates of the haploid strain CEN.PK113-7D were analyzed in aerobic batch cultures on complete synthetic medium with glucose (SMD) and on seven "SMDΔ" media from which either biotin, inositol, nicotinic acid, pantothenic acid, pABA, pyridoxine, or thiamine was omitted. To limit interference by carryover of vitamins from precultures, specific growth rates were measured after a third consecutive transfer on each medium ( Fig. 2A).
Consistent with the presence in its genome of genes predicted to encode all enzymes involved in biosynthetic pathways for all seven vitamins ( Fig. 1 [5]), the strain CEN.PK113-7D grew on all SMDΔ versions. On complete SMD, a specific growth rate of 0.38 Ϯ 0.02 h Ϫ1 was observed, while specific growth rates on SMDΔ lacking biotin, pantothenate, pyridoxine, thiamine, or inositol were 95%, 57%, 32%, 22%, or 19% lower, respectively. After three transfers, specific growth rates on SMDΔ lacking pABA or nicotinic acid did not differ significantly from the specific growth rate on complete SMD ( Fig. 2A). However, in SMDΔ lacking pABA, growth in the first transfer was slower than in the first transfer on complete SMD (Fig. 2B). Extending the number of transfers to five, which corresponded to approximately 33 generations of selective growth, led to higher specific growth rates on several SMDΔ versions ( Fig. 2A), suggesting that serial transfer selected for spontaneous faster-growing mutants.
Adaptive laboratory evolution of CEN.PK113-7D for fast growth in the absence of single vitamins. Serial transfer in independent triplicate aerobic shake-flask cultures on each SMDΔ version was used to select mutants that grew fast in the absence of individual vitamins. Specific growth rates of evolving populations were measured after 5, 10, 23, 38, and 50 transfers and compared to the specific growth rate of strain CEN.PK113-7D grown in complete SMD.
ALE experiments were stopped once the population reached a specific growth rate equal to or higher than 0.35 h Ϫ1 , which represents 90% to 95% of the specific growth rate of strain CEN.PK113-7D on complete SMD ( Fig. 2A) (19)(20)(21)(22). As already indicated by the specific growth rates observed after 3 and 5 transfers in SMDΔ ( Fig. 2A), few transfers were required for reaching this target in SMDΔ lacking inositol, nicotinic acid, or pABA. Conversely, more than 330 generations of selective growth were required to reach a specific growth rate of 0.35 h Ϫ1 on SMDΔ lacking either pantothenic acid, pyridoxine, or thiamine ( Fig. 3A). At least two single-cell lines were isolated from each of the three independent ALE experiments on each SMDΔ version and the fastest growing single-cell line from each experiment was selected (strains IMS0724 to FIG 2 Specific growth rates of S. cerevisiae CEN.PK113-7D in aerobic batch cultures on complete SMD and on SMD lacking single vitamins. (A) Growth rate measurements were performed after 3 (light gray) and 5 (dark gray) consecutive transfers in the same medium. (B) Growth curves of CEN.PK113-7D in complete SMD () and at transfer 1 (o) and 3 (p) in SMD lacking para-aminobenzoic acid (pABA). In the latter medium, a lower specific growth rate was observed at transfer 1, but upon the third transfer, the growth rate was the same as in complete SMD. Error bars represent the standard deviations (n ϭ 9 for complete SMD; n ϭ 3 for all other media).

FIG 3
Heat map showing specific growth rates during ALE of S. cerevisiae CEN.PK113-7D on SMD lacking single vitamins. (A) Aerobic serial transfer experiments on each medium composition were performed in triplicates (rows). The specific growth rate of each evolving population was measured after a specific number of sequential transfers (columns). Yellow color indicates slow growth, while cyan indicates a specific growth rate statistically undistinguishable from that of the positive control (strain CEN.PK113-7D grown on SMD medium with all vitamins). (B) Specific growth rates of single-colony isolates from each independent biological replicate evolution line. The fastest-growing isolates, whose genomes were resequenced, are indicated in blue.

Perli et al.
Applied and Environmental Microbiology IMS0726 from SMDΔ lacking nicotinic acid, IMS0727 to IMS0729 from SMDΔ lacking pABA, IMS0730 to IMS0732 from SMDΔ lacking inositol, IMS0733 to IMS0735 from SMDΔ lacking pantothenate, IMS0736 to IMS0738 from SMDΔ lacking pyridoxine, and IMS0747 to IMS0749 from SMDΔ lacking thiamine) ( Table 1). The specific growth rates of isolates that had been independently evolved in each SMDΔ version did not differ by more than 6%. The largest difference (5.3%) was observed for isolates IMS0733 to IMS0735 evolved on SMDΔ lacking pantothenate (Fig. 3B). Whole-genome sequencing of evolved strains and target identification. To identify mutations contributing to vitamin independence, the genomes of the sets of three independently evolved isolates for each SMDΔ version were sequenced with Illumina short-read technology. After aligning reads to the reference CEN.PK113-7D genome sequence (23), mapped data were analyzed for the presence of copy number variations (CNVs) and single nucleotide variations (SNVs) that occurred in annotated coding sequences.
A segmental amplification of 34 kb (from nucleotide 802500 to 837000) on chromosome VII, which harbors THI4, was observed in strain IMS0749 (Fig. 4A), which had been evolved in SMDΔ lacking thiamine. THI4 encodes a thiazole synthase, a suicide enzyme that can only perform a single catalytic turnover (24). Segmental amplifications on chromosomes III and VIII were observed in strain IMS0725, which had been evolved in SMDΔ lacking nicotinic acid (Fig. 4B). Since these regions are known to be prone to recombination in the parental strain CEN.PK113-7D (23,25), their amplification is not necessarily related to nicotinic acid independence.
SNV analysis was systematically performed, and data from the three sequenced isolates were compared. To eliminate false positives caused by mapping artifacts, reads of the CEN.PK113-7D strains were mapped back on its own reference assembly. Identified SNVs found were systematically subtracted. SNV analysis was restricted to non-synonymous mutations in coding sequences (Table 2).
In three of the six isolates from ALE experiments in SMDΔ lacking nicotinic acid or inositol, no non-synonymous SNVs were detected (Fig. 5). One strain (IMS0724) from a serial transfer experiment on SMDΔ lacking nicotinic acid showed SNVs in RPG1 and PMR1, while a second strain (IMS0725) showed SNVs in MTO1 and VTH2. A mutation in No pABA 3 0.414 4 a ID, identifier. b Percentage improvement over the specific growth rate of the parental strain after three transfers in the same medium is also shown (n ϭ 1 for each strain).
YFR054W was identified in a single strain (IMS0730) evolved for inositol-independent growth. The absence of mutations in several strains subjected to serial transfer in SMDΔ lacking nicotinic acid or inositol is consistent with the fast growth of the parental strain CEN.PK113-7D in these media ( Fig. 2A). Sequencing of the three isolates evolved in SMDΔ lacking pABA revealed only five SNVs, of which two were in ABZ1 (strains IMS0727 and IMS0729) and one in ARO7 (IMS0728), while SNVs in NUP57 and HTS2 were found in strains IMS0728 and IMS0729, respectively (Fig. 5). NUP57 and HTS2 could not be directly linked to pABA metabolism. Conversely, Abz1 is an aminodeoxychorismate synthase that directs chorismate toward pABA synthesis, and Aro7 is a chorismate mutase that catalyzes the first committed reaction toward phenylalanine and tyrosine and thereby diverts chorismate from pABA synthesis ( Fig. 1) (26,27). These two SNVs therefore represented clear targets for reverse engineering ( Table 2).
In line with the much longer ALE experiments (approximately 332 generations), strains evolved in SMDΔ lacking thiamine, pantothenate, or pyridoxine showed larger numbers of SNVs, with a maximum number of 30 SNVs in the isolate IMS0738 evolved in SMDΔ lacking pyridoxine ( Table 2 and Fig. 5).
Isolates IMS0736 and IMS0737, which had been evolved in SMDΔ lacking pyridoxine, harbored only two and three mutations, respectively, while strain IMS0738 harbored 30 mutations. All three strains carried different mutated alleles of BAS1, which encodes a transcription factor involved in regulation of histidine and purine biosynthesis (33,34). IMS0736 harbored a non-synonymous mutation causing an amino acid change position   II (36). One of the mutations in strain IMS0733 affected ISW2, which encodes a subunit of the chromatin remodeling complex (37). These three genes involved in regulatory processes were selected for reverse engineering, along with SPE2 and FMS1. The latter two genes, encoding S-adenosylmethionine decarboxylase (38) and polyamine oxidase (39), are directly involved in pantothenate biosynthesis and were found to be mutated in isolates IMS0734 and IMS0735, respectively. In summary, based on mutations in the same gene in independently evolved isolates and/or existing information on involvement of affected genes in vitamin biosynthesis, mutations in 12 genes were selected for reconstruction in the parental strain CEN.PK113-7D. These were mutated alleles of ISW2, GAL11, TUP1, FMS1, and SPE2 for pantothenate, in BAS1 for pyridoxine, mutations in CNB1, PMR1, and FRE2 as well as overexpression of THI4 for thiamine, and mutations in ABZ1 and ARO7 for pABA. Since serial transfer on SMDΔ lacking nicotinic acid or inositol did not consistently yield mutations and the parental strain CEN.PK113-7D already grew fast on these media, no reverse engineering of mutations observed in isolates from those experiments was carried out.
Reverse engineering of target gene mutations and overexpression. To investigate whether the selected targets contributed to the phenotypes of the evolved strains, single point mutations or single-gene overexpression cassettes were introduced in a non-evolved reference strain, followed by analysis of specific growth rate in the relevant SMDΔ variant. For most target genes, a two-step strategy was adopted, so that a single-gene knockout mutant was constructed in the process ( Fig. 6A and B). For the SPE2 mutant strains IMX2308 and IMX2289, point mutations were introduced in a single step (Fig. 6C). The THI4-overexpressing strains IMX2290 and IMX2291 were constructed The Venn diagrams show genes that acquired non-synonymous mutations in multiple independent evolution experiments for a specific medium as well as genes that were affected in a single replicate. Apparent mutations also found in the genome of the parent strain CEN.PK113-7D were subtracted and not shown. Target genes that were selected for reverse engineering are shown in red.

Single Vitamin Prototrophies in S. cerevisiae
Applied and Environmental Microbiology by integrating the overexpression cassette at the YPRcTau3 locus (40) (Fig. 6D). Subsequently, multiple mutations that were found in strains evolved in the same SMDΔ version were combined into single engineered strains to test for additive or synergistic effects.
Thiamine. The specific growth rate of S. cerevisiae CEN.PK113-7D was only 27% lower in SMDΔ lacking thiamine than in complete SMD ( Fig. 2A). Nevertheless, it took more than 300 generations of selective growth to obtain evolved isolates that compensated for this difference ( Fig. 3 and Table 1). The role of mutations in CNB1, FRE2, and PMR1 in this evolved phenotype was first investigated in the single knockout strains IMX1721, IMX1722, and IMX1723, respectively. While deletion of PMR1 negatively affected the specific growth rate on SMDΔ lacking thiamine, deletion of either CNB1 or FRE2 resulted in a 17% increase of the specific growth rate on this medium relative to that of CEN.PK113-7D. However, strains IMX1721 (cnb1Δ) and IMX1722 (fre2Δ) still grew significantly slower than the evolved isolates (Fig. 7). Subsequently, the mutated alleles found in the evolved isolates were introduced at the native chromosomal locus, resulting in strains IMX1985 (CNB1 L82F ), IMX1986 (PMR1 S104Y ), and IMX1987 (FRE2 T110S ). In addition, THI4 was overexpressed (strain IMX2290) to simulate the copy number increase observed in IMS0749. Strains IMX1987 (FRE2 T110S ) and IMX2290 (THI41) grew as fast as the evolved isolates on SMDΔ lacking thiamine (0.35 to   (Fig. 7). Combination of these mutated alleles of PMR1 and FRE2, which occurred together in isolate IMS0748, as well as of the two mutations resulting in growth improvement (FRE2 T110S and THI41) was also tested. None of these combinations yielded a higher specific growth rate than observed in the evolved strains and in the reverse-engineered FRE2 T110S and THI41 strains. para-Aminobenzoic acid. In SMDΔ lacking pABA, strain CEN.PK113-7D grew 50% slower than in complete SMD. However, it took only a few transfers to achieve fast pABA-independent growth. The independently evolved isolates IMX2057 and IMX1989 harbored mutations affecting genes that encode chorismate-utilizing enzymes, the precursor of pABA (ABZ1 N593H and ARO7 L205S , respectively) (Fig. 1). As these strains were able to grow in SMD without amino acid supplementation, these mutations did not cause a complete loss of function. However, they might well have affected the distribution of chorismate over pABA and aromatic amino acid biosynthesis (26,27). Introduction of either ABZ1 N593H or ARO7 L205S , while replacing the corresponding wild-type allele, eliminated the slower growth observed in strain CEN.PK113-7D during the first transfer on SMDΔ lacking pABA. Specific growth rates of these reverseengineered strains IMX2057 (ABZ1 R593H ) and IMX1989 (ARO7 L205S ) were not statistically different from those of the corresponding evolved isolates (Fig. 7).
Pantothenic acid. Omission of pantothenic acid from SMD led to a 57% lower specific growth rate of strain CEN.PK113-7D than observed in complete SMD ( Fig. 2 and Table 1). Of a total number of 29 mutations found in three independently evolved isolates that showed fast growth in SMDΔ lacking pantothenate, SNVs in ISW2, GAL11, TUP1, SPE2, and FMS1 were analyzed by reverse engineering. Single deletions of SPE2, FMS1, and GAL11 resulted in an inability to grow on SMDΔ lacking pantothenate. This result was anticipated for the spe2Δ and fms1Δ mutants, in view of the roles of these genes in pantothenate biosynthesis. However, GAL11 was not previously implicated in pantothenate biosynthesis. The gal11Δ strain was conditional, as the mutant did grow on complex yeast extract-peptone-dextrose (YPD) and SMD media. Of the remaining two deletion mutants, the tup1Δ strain IMX1817 showed a 68% higher specific growth rate on SMDΔ than strain CEN.PK113-7D, while deletion of ISW2 did not result in faster growth on this medium (Fig. 7). Of seven SNVs that were individually expressed in the non-evolved strain background, only the GAL11 Q383stop mutation found in IMS0735 supported a specific growth rate of 0.33 h Ϫ1 on SMDΔ lacking pantothenate that was only 8% lower than that of the evolved isolates.
Combination of the GAL11 Q383stop mutation with TUP1 Q99stop , and TUP1 with FMS1, did not lead to additional improvement, indicating that the GAL11 Q383stop mutation was predominantly responsible for the improved growth of evolved strain IMS0735 in the absence of pantothenate.
Pyridoxine. Strain CEN.PK113-7D grew 35% slower on SMDΔ lacking pyridoxine than on complete SMD (Fig. 2). Three different mutated alleles of BAS1 were identified in strains that had been independently evolved for fast growth on the former medium ( Fig. 3 and Table 2). Deletion, in a non-evolved reference strain, of BAS1 (IMX2128) did not result in faster pyridoxine-independent growth (Fig. 7). Individual expression of the evolved BAS1 alleles in strain IMX2128 yielded strains IMX2135 (BAS1 Q152R ), IMX2136 (BAS1 D101N ), and IMX2137 (BAS1 S41P ). All three BAS1 mutant strains grew faster on SMDΔ lacking pyridoxine than strain CEN.PK113-7D, reaching specific growth rates on this medium that were not significantly different from the average of those of evolved The specific growth rate of strain CEN.PK113-7D in SMD medium lacking the relevant vitamin is shown and highlighted by a vertical line to help to visualize improved performance of engineered strains. Error bars represent the standard deviations (n ϭ 9 for complete SMD, n ϭ 6 for strain IMX1721; otherwise n ϭ 3). A Student's t test was performed to compare the wild-type and evolved CEN.PK113-7D growth rate to the engineered strain growth rates, and non-significant differences are indicated with white letters (P value Ͼ 0.05). strains IMS736, IMS737, and IMS738 (Fig. 7). These results suggest that BAS1, which was previously shown to be involved in regulation of purine and histidine biosynthesis (33,34), may also be involved in regulation of pyridoxine biosynthesis in S. cerevisiae.

DISCUSSION
Vitamin requirements of S. cerevisiae. Most S. cerevisiae genomes harbor the full complement of genes required for synthesis of the seven B vitamins that are commonly included in chemically defined media for yeast cultivation (CDMY; for a recent review see references 5 and 10). Previous studies indicated that the presence of a complete set of biotin biosynthesis genes supported only slow growth on CDMY. The present study shows that, similarly, none of the other six B vitamins included in CDMY (inositol, nicotinic acid, pantothenic acid, pABA, pyridoxine, and thiamine) are strictly required for growth. Remarkably, the impact of individually eliminating these six vitamins from glucose-containing CDMY differently affected specific growth rates in aerobic glucosegrown cultures, with growth rate reductions varying from 0% to 57%. It should, however, be noted that requirements for these growth factors, which for aerobic yeast cultivation cannot be formally defined as vitamins, and their absolute and relative requirements may well be condition and strain dependent. For example, it is well documented that synthesis of nicotinic acid and pantothenic acid by S. cerevisiae is strictly oxygen dependent (41). The data set compiled in the present study will, hopefully, serve as reference for investigating vitamin requirements of diverse natural isolates and laboratory and industrial strains and thereby help to obtain a deeper understanding of the genetics and ecology of vitamin prototrophy and vitamin biosynthesis in S. cerevisiae.
ALE and reverse engineering for identifying genes involved in fast B vitaminindependent growth. A serial transfer strategy was applied to select for spontaneous mutants that grew as fast in aerobic batch cultures on CDMY lacking either inositol, nicotinic acid, pyridoxine, thiamine, pantothenic acid, or para-aminobenzoic acid as in CDMY containing all these six vitamins as well as biotin. In the ALE experiments on media lacking nicotinic acid or inositol, fast growth was observed within a few cycles of batch cultivation, and not all fast-growing strains were found to contain mutations. These observations indicated that, under the experimental conditions, the native metabolic and regulatory network of S. cerevisiae was able to meet cellular requirements for fast growth in the absence of these "vitamins." As demonstrated in other ALE studies, performing independent replicate evolution experiments helped in identifying biologically relevant mutations upon subsequent whole-genome sequencing (11,12). The power of this approach is illustrated by the ALE experiments that selected for pyridoxine-independent growth, in which the independently evolved mutants IMS0736 and IMS0738 harbored 2 and 30 mutated genes, respectively, of which only BAS1 also carried a mutation in a third independently sequenced isolate (Fig. 5 and Table 2).
In total, the role of 12 genes that were found to be mutated in the ALE experiments were selected for further analysis by reverse engineering of the evolved alleles and/or deletion mutations in the parental non-evolved genetic background ( Fig. 5 and Table  2). These genes comprised three groups: (i) genes encoding enzymes known or inferred to be involved in the relevant vitamin synthesis pathway (SPE2 and FMS1 for pantothenate, THI4 for thiamine, and ABZ1 and ARO7 for pABA), (ii) genes encoding transcriptional regulator proteins (TUP1 and GAL11 for pantothenate and BAS1 for pyridoxine), and (iii) non-transcriptional-regulator proteins whose functions were not previously associated with vitamin biosynthesis (ISW2 for pantothenate and CNB1, PMR1, and FRE2 for thiamine).
Of the first group of mutations defined above, only those in SPE2 and FMS1 were not found to contribute to faster growth in the absence of the relevant vitamin. The second group yielded interesting information on the regulation of vitamin biosynthesis in S. cerevisiae. In particular, the key role of mutations in BAS1 in enabling fast pyridoxineindependent growth and the role of GAL11 and TUP1 mutations in fast pantothenate-independent growth dependency provided interesting insights and leads for further research.
The S. cerevisiae transcriptional activator Bas1 is involved in the regulation of purine and histidine (33,34). Interestingly, Bas1 is also involved in the repression of genes involved in C 1 metabolism and of SNZ1 (42). Snz1 is a subunit of a two-component pyridoxal-5=-phosphate synthase, which catalyzes the first step of the synthesis of pyridoxal-5-phosphate, the active form of pyridoxine in S. cerevisiae (43). Interrogation of the Yeastract database (44) for occurrence of transcription binding sites in promoter regions of pyridoxine biosynthesis genes confirmed the link already established between BAS1 and SNZ1 (42,45). Moreover, this analysis revealed that all pyridoxine biosynthesis genes in S. cerevisiae contain a consensus Bas1 cis-regulatory binding motif (Fig. 8). Consistent with the regulatory role of Bas1 on SNZ1 expression, Bas1 has been experimentally shown to repress transcription of genes involved in pyridoxine biosynthesis (46). The mutations found in BAS1 may, therefore, have attenuated Bas1mediated repression of pyridoxine biosynthetic genes and, thereby, enabled increased pyridoxine biosynthesis.
ALE experiments in pantothenate-free medium yielded different mutations in TUP1 and GAL11, two major components of the yeast regulatory machinery. TUP1 encodes a general transcriptional repressor that, in a complex with Cyc8, modifies chromatin structure such that genes are repressed (47)(48)(49). GAL11 (also known as MED15) encodes a subunit of the mediator complex required for initiation by RNA polymerase II and consequently plays a critical role in transcription of a large number of RNA polymerase II-dependent genes (50,51). Despite its involvement in general cellular transcriptional regulation, GAL11 is not an essential gene for growth in complete medium (52). The inability of a gal11Δ strain to grow on glucose synthetic medium without pantothenate represents the first indication for a possible involvement of Gal11 in regulation of pantothenate metabolism. Gal11 interacts with transcriptional activators through various peptidic segments, including an N-terminal KIX domain. This region shows homology with the B-box motif found in the mammalian activating protein SRC-1 and is essential for recruitment of the mediator complex by other regulatory proteins (e.g., Gcn4) (53). Of two mutations found in GAL11, the most potent was a nonsense mutation at nucleotide 383. In contrast to a gal11Δ strain, a reverse-engineered strain carrying this premature stop codon grew on SMDΔ pantothenate, which indicates that the GAL11 Q383stop allele encodes a functional peptide. Such a functional truncated Gal11 version was not previously described and is sufficiently long to include a complete KIX domain (AA 9 to AA 86 ) for recruitment of the RNA polymerase II machinery by an as yet unidentified transcription factor involved in regulation of pantothenate biosynthesis. Further research is required to resolve and understand the role of the wild-type and evolved alleles of GAL11 in regulation of pantothenate metabolism.
A third group of non-transcription factor genes had not yet been associated with the biosynthesis of vitamins. Reverse engineering of a mutation in ISW2, which encodes an ATP-dependent DNA translocase involved in chromatin remodeling (54) identified in the pantothenate evolution, did not yield a growth improvement. We cannot exclude that this mutation in association with ERG3, AMN1, DAN4, and ERR3 identified in IMS0733 ( Fig. 5 and Table 2) might have a significant impact, but systematic combinatorial analysis of the mutations was not performed.
Mutations in CNB1, PMR1, and FRE2 identified in evolved isolates all improved growth of S. cerevisiae in the absence of thiamine (Fig. 7, top). These three genes all encode proteins involved in metal homeostasis: Fre2 is a ferric or cupric reductase (55), and Cnb1 is the regulatory B subunit of calcineurin, a Ca 2ϩ /calmodulin-regulated type 2B protein phosphatase which regulates the nuclear localization of Crz1. This transcription factor influences the expression of a large number of genes. Its targets include PMR1, which encodes a high-affinity Ca 2ϩ /Mn 2ϩ P-type ATPase involved in Ca 2ϩ and Mn 2ϩ transport into the Golgi apparaus (30,56). Neither of these three genes has hitherto been directly associated with thiamine. However, thiamine pyrophosphokinase (Thi80), thiamine phosphate synthase (Thi6), and hydroxymethylpyrimidine phosphate (Thi21 and Thi20) all require Mg 2ϩ or Mn 2ϩ as cofactors (57,58). At low concentration, Mn 2ϩ was shown to be a stronger activator of Thi80 than Mg 2ϩ (59). In an ALE experiment with engineered xylose-fermenting assimilating S. cerevisiae, a nonsense mutation or deletion of PMR1 caused selectively and strongly increased intracellular concentrations of Mn 2ϩ , which was the preferred metal ion for the heterologously expressed Piromyces xylose isomerase (60). Although intracellular metal ion concentrations were not measured in the present study, the different phenotypes of a pmr1Δ deletion strain (IMX1722) and a PMR1 S104Y strain (IMX1986) (Fig. 7, top) indicate that the latter mutation does not act through a massive increase of the intracellular Mn 2ϩ concentration.
In S. cerevisiae, synthesis of the thiazole moiety of thiamine biosynthesis involves sulfide transfer from an active-site cysteine (Cys205) residue of the thiazole synthase Thi4. This sulfur transfer reaction is iron dependent and generates inactive enzyme by formation of a dehydroalanine. Fe(II) plays an essential role in this sulfide transfer, which remains poorly understood (24). Further research is needed to investigate if the FRE2 mutation in strain IMS0749 in some way increases the efficiency of the reaction catalyzed by the energetic single-turnover enzyme Thi4 and to resolve the role of metal homeostasis in vitamin biosynthesis.
Towards mineral media for cultivation of S. cerevisiae. With the exception of the carbon and energy sources for growth, B vitamins are the sole organic ingredients in standard CDMY recipes for aerobic cultivation of wild-type and industrial S. cerevisiae strains. In view of the chemical instability of some of these compounds, vitamin solutions cannot be autoclaved along with other medium components but are usually filter sterilized. In research laboratories and, in particular, in industrial processes, the costs, complexity, and contamination risks associated with the use of vitamins is significant. Complete elimination of vitamins from CDMY, without compromising specific growth rate, yield, or productivity, could therefore result in considerable cost and time savings as well as in improved standardization and robustness of cultivation procedures.
The present study demonstrates that, by ALE as well as introduction of small sets of defined mutations into S. cerevisiae, it is possible to achieve specific growth rates in single-vitamin-depleted CDMY that are close or identical to those found in CDMY supplemented with a complete vitamin mixture. While these results represent a first step toward the construction of prototrophic growth of S. cerevisiae and related yeasts, further research is required which trade-offs are incurred upon simultaneous introduction of the genetic interventions identified in this study and how they can be mitigated. This issue may be particularly relevant for mutations that affect genes involved in global regulation processes (51,61), which may interfere with other cellular processes. In addition, simultaneous high-level expression of multiple enzymes with low-catalytic turnover numbers, with the suicide enzyme Thi4 (24, 62, 63) as an extreme example, may affect cell physiology due to the required resource allocation (64,65).
In such cases, it may be necessary to expand metabolic engineering strategies beyond the native metabolic and regulatory capabilities of S. cerevisiae by expression of heterologous proteins and/or pathways with more favorable characteristics (66).

MATERIALS AND METHODS
Strains, media, and maintenance. The S. cerevisiae strains used and constructed in this study are shown in Table 3 and they all derive from the CEN.PK lineage (17,67). Yeast strains were grown on synthetic medium with ammonium sulfate as a nitrogen source (SM) or YP medium (10 g/liter Bacto yeast extract, 20 g/liter Bacto peptone) as previously described (2). SM and YP media were autoclaved at 121°C for 20 min. Then, SM medium was supplemented with 1 ml/liter of filter-sterilized vitamin solution [0.05 g/liter D-(ϩ)-biotin, 1.0 g/liter D-calcium pantothenate, 1.0 g/liter nicotinic acid, 25 g/liter myoinositol, 1.0 g/liter thiamine hydrochloride, 1.0 g/liter pyridoxol hydrochloride, 0.20 g/liter 4-aminobenzoic acid]. Vitamin dropout media were prepared using vitamin solutions lacking either thiamine, pyridoxine, pantothenic acid, inositol, nicotinic acid, or para-aminobenzoic acid, yielding SMΔthiamine, SMΔpyridoxine, SMΔpantothenic acid, SMΔinositol, SMΔnicotinic acid, and SMΔpABA, respectively. A concentrated glucose solution was autoclaved at 110°C for 20 min and then added to the SM and YP media at a final concentration of 20 g/liter, yielding SMD and YPD, respectively. Five hundred-milliliter shake flasks containing 100 ml medium and 100-ml shake flasks containing 20 ml medium were incubated at 30°C and at 200 rpm in an Innova incubator (Brunswick Scientific, Edison, NJ). Solid media were prepared by adding 1.5% Bacto agar and, when indicated, 200 mg/liter G418 or 200 mg/liter hygromycin. Escherichia coli strains were grown in LB (10 g/liter Bacto tryptone, 5 g/liter Bacto yeast extract, 5 g/liter NaCl) supplemented with 100 mg/liter ampicillin or kanamycin. S. cerevisiae and E. coli cultures were stored at Ϫ80°C after the addition of 30% (vol/vol) glycerol.
Molecular biology techniques. PCR amplification of DNA fragments with Phusion Hot Start II high-fidelity polymerase (Thermo Scientific, Waltham, MA) and desalted or PAGE-purified oligonucleotide primers (Sigma-Aldrich, St. Louis, MO) was performed according to the manufacturers' instructions. DreamTaq polymerase (Thermo Scientific) was used for diagnostic PCR. Primers used in this study are shown in Table 4. PCR products were separated by gel electrophoresis using 1% (wt/vol) agarose gels (Thermo Scientific) in Tris-acetate-EDTA (TAE) buffer (Thermo Scientific) at 100 V for 25 min and purified either with a GenElutePCR Clean-Up kit (Sigma-Aldrich) or with a Zymoclean Gel DNA Recovery kit (Zymo Research, Irvine, CA). Plasmids were purified from E. coli using a Sigma GenElute Plasmid kit (Sigma-Aldrich). Plasmids used in this study are shown in Table 5. Yeast genomic DNA was isolated with the SDS-lithium acetate (LiAc) protocol (68). Yeast strains were transformed with the lithium acetate method (69). Four to eight single colonies were restreaked three consecutive times on selective media, and diagnostic PCRs were performed in order to verify their genotype. E. coli XL1-Blue was used for chemical transformation (70). Plasmids were then isolated and verified by either restriction analysis or by diagnostic PCR.
Laboratory evolution. Laboratory evolution of S. cerevisiae CEN.PK113-7D for fast growth in SMD medium lacking a single vitamin was performed by sequential transfer in aerobic shake-flask batch cultures. A frozen aliquot of strain CEN.PK113-7D was inoculated in a preculture shake flask containing SMD medium supplemented with all vitamins. Cells were then spun down, washed twice with sterile water, and used to inoculate a second shake flask containing SMD lacking one of the vitamins. The culture was then grown until stationary phase and transferred in a third shake flask containing the same fresh medium. At each transfer, 0.2 ml culture broth was transferred to 20 ml fresh medium, corresponding to approximately 6.7 generations in each growth cycle. The evolution experiment was performed in SMΔthiamine, SMΔpyridoxine, SMΔpantothenic acid, SMΔinositol, SMΔnicotinic acid, and SMΔpABA media. Each evolution experiment was performed in triplicates. After a defined number of transfers, intermediate strains were stocked and characterized for the growth rate. The experiment was stopped once the target specific growth rate of 0.35 h Ϫ1 was reached. From each evolved population, three single colonies were then isolated and stored. The specific growth rate of these single cell lines was measured to verify that they were representative of the evolved population. The best performing isolate from each evolution line was selected for whole-genome sequencing.
Shake flask growth experiments. For specific growth rate measurements of strains (evolved populations as well as single cell lines), an aliquot was used to inoculate a shake flask containing 100 ml of fresh medium. For specific growth rate measurements of the engineered strains, a frozen aliquot was thawed and used to inoculate a 20-ml starter culture that was then used to inoculate the 100-ml flask. An initial optical density at 660 nm (OD 660 ) of 0.1 or 0.2 was used as a starting point. The flasks were then incubated, and growth was monitored using a 7200 Jenway spectrometer (Jenway, Stone, United Kingdom). Specific growth rates were calculated from at least four time points in the exponential growth phase of each culture.
Plasmid cloning. Plasmids carrying two copies of the same genomic RNA (gRNA) were cloned by in vitro Gibson assembly as previously described (72). In brief, an oligonucleotide carrying the 20-bp target sequence and homology to the backbone plasmid was used to amplify the fragment carrying the 2-m origin of replication sequence by using pROS13 as the template. The backbone linear fragment was amplified by using primer 6005 and either pROS12 or pROS13 as the template (73). The two fragments were then gel purified, combined, and assembled in vitro using the NEBuilder HiFi DNA Assembly master  The plasmid carrying the expression cassette for THI4 was cloned by golden gate assembly using the yeast toolkit (YTK) parts (74). The THI4 coding sequence was amplified using the primer pair 12174/12175 and CEN.PK113-7D genomic DNA as a template in order to add YTK compatible ends to the gene. The PCR product was then purified and combined together with plasmids pYTK009, pYTK056, and pYTK096 in a BsaI golden gate reaction that yielded plasmid pUDI180.
After genotyping of the transformants, correct isolates were grown in 20 ml YPD in a 50-ml vented Greiner tube at 30°C overnight by inoculating a single colony. The next day, 1 l was transferred to a new tube containing the same amount of medium and the sample was grown overnight. The day after, each liquid culture was restreaked to single colony by plating on YPD agar plates. Plates were incubated at 30°C overnight, and the next day, single colonies were patched on both YPD and YPD plus the relative antibiotic (either G418 or hygromycin) to assess which clones had lost the gRNA plasmid. One clone for each strain that had lost the plasmid was then grown in YPD, and 30% (vol/vol) glycerol was added prior to stocking samples at Ϫ80°C.
Data availability. The sequencing data of the evolved and of the BAS1 deletion Saccharomyces cerevisiae strains were deposited at NCBI (https://www.ncbi.nlm.nih.gov/) under BioProject accession number PRJNA603441. All measurement data used to prepare the figures of the manuscript are available at the data.4TU.nl repository (https://doi.org/10.4121/uuid:53c9992f-d004-4d26-a3cd-789c524fe35c).

ACKNOWLEDGMENTS
Experiments were designed by T.P., J.-M.D., and J.T.P. Strain evolution and isolation was performed by T.P. Analysis of next-generation sequencing data was performed by M.V.D.B. and T.P. Reverse engineering of target mutations and phenotypical characterization of the strains was performed by T.P. and D.P.I.M. T.P. and J.-M.D. wrote the first version of manuscript. All authors critically read this version, provided input, and approved the final version.
This work has received funding from the European Union's Horizon 2020 research and innovation program under the Marie Sklodowska-Curie grant agreement no. 722287.
We declare no competing interests.