Correspondence: Antonius J.A. van Maris, Department of Biotechnology, Delft University of Technology, Julianalaan 67, Delft 2628 BC, The Netherlands. Tel.: +31 15 278 1616; fax: +31 15 278 2355; e-mail: A.J.A.vanMaris@TUDelft.nl
Laboratory evolution is a powerful approach in applied and fundamental yeast research, but complete elucidation of the molecular basis of evolved phenotypes remains a challenge. In this study, DNA microarray-based transcriptome analysis and whole-genome resequencing were used to investigate evolution of novel lactate transporters in Saccharomyces cerevisiae that can replace Jen1p, the only documented S. cerevisiae lactate transporter. To this end, a jen1Δ mutant was evolved for growth on lactate in serial batch cultures. Two independent evolution experiments yielded growth on lactate as sole carbon source (0.14 and 0.18 h−1, respectively). Transcriptome analysis did not provide leads, but whole-genome resequencing showed different single-nucleotide changes (C755G/Leu219Val and C655G/Ala252Gly) in the acetate transporter gene ADY2. Introduction of these ADY2 alleles in a jen1Δ ady2Δ strain enabled growth on lactate (0.14 h−1 for Ady2pLeu219Val and 0.12 h−1 for Ady2pAla252Gly), demonstrating that these alleles of ADY2 encode efficient lactate transporters. Depth of coverage of DNA sequencing, combined with karyotyping, gene deletions and diagnostic PCR, showed that an isochromosome III (c. 475 kb) with two additional copies of ADY2C755G had been formed via crossover between retrotransposons YCLWΔ15 and YCRCΔ6. The isochromosome formation shows how even short periods of selective pressure can cause substantial karyotype changes.
Laboratory evolution (referred to as ‘evolutionary engineering’ in applied contexts) is a powerful strategy to obtain microorganisms with improved characteristics that are difficult to engineer through rational approaches, especially when the genetic determinants for a phenotype are unknown or complex (Sauer, 2001). Subsequent discovery of the underlying mutation(s) is of key importance, both for functional gene analysis and to enable transfer of the acquired traits to nonevolved industrially relevant strain backgrounds (‘reverse metabolic engineering’; Bailey et al., 1996). In addition to the mutation(s) of interest, random mutations can accumulate during evolution. Combined with our incomplete understanding of metabolic and regulatory networks, this makes elucidation of the genetic basis of acquired phenotypes the main challenge of reverse metabolic engineering. During evolution, different types of mutations can occur: (1) small local changes (substitutions, insertions, deletions and duplications) that result in either differential gene expression or changes in the amino acid sequence of the encoded protein and/or (2) rearrangements of larger DNA fragments, potentially resulting in a different gene dosage or expression (Sauer, 2001).
Several genome-wide techniques have been described in literature to analyse mutations that occur during laboratory evolution (for reviews, see Gill, 2003; Bro & Nielsen, 2004; Gresham et al., 2008a). Classical techniques include construction and analysis of plasmid-based genomic libraries and quantitative trait locus mapping via crossing. More recent developments have enabled whole-genome DNA sequencing, DNA hybridization to DNA microarrays, transcriptome analysis using DNA microarrays, proteome, metabolome and fluxome profiling (Gill, 2003; Bro & Nielsen, 2004; Gresham et al., 2008b). Analysis with DNA microarrays was, until recently, the only affordable genome-wide method. Microarray-based transcriptome analysis only indirectly indicates changes on the DNA level. Several factors, such as complex regulatory networks, transcription factor binding, translation efficiency and stability of mRNA and proteins, complicate the interpretation of the exact genetic basis of changes observed at mRNA and protein level. In contrast, comparative genome hybridization using DNA microarrays can identify genetic changes directly (Winzeler et al., 1999; Daran Lapujade et al., 2003; Schacherer et al., 2007). However, limitations in the maximum probe density imply that either (1) the resolution of these DNA microarrays is limited to several base pairs, necessitating sequencing of selected regions or (2) multiple DNA microarrays are required to cover larger (eukaryotic) genomes. Furthermore, DNA microarrays can only be used for a comprehensive analysis when the strains under investigation are congenic to the strain on whose genome sequence the microarray design was based. In contrast, whole-genome DNA sequencing, which because of recent price developments has become an interesting technique for analysis of evolved strains (Nowrousian, 2010; Sboner et al., 2011), has the potential to directly reveal all mutations accumulated in an evolved strain. However, discriminating between random and relevant mutations remains a challenging task. Combining DNA sequencing with other genome-wide techniques that provide insight in the underlying physiology can help to elucidate the genetic basis of a phenotype acquired via evolution.
Transport of carboxylic acids plays a key role in weak organic acid stress (Palmqvist & Hahn-Hägerdal, 2000) and in metabolic engineering strategies for organic acid production with Saccharomyces cerevisiae (van Maris et al., 2004a, b; Jol et al., 2010). However, the responsible membrane transporters are often poorly studied and encoded by multiple redundant genes (Casal et al., 2008). The goal of this study was to investigate whether laboratory evolution and subsequent elucidation of the underlying mutations can lead to the identification of alternative lactate transporters in S. cerevisiae. In S. cerevisiae, Jen1p was previously identified as the only efficient lactate importer by generation of a UV mutant unable to grow on lactate and subsequent functional complementation with a genomic library (Casal et al., 1999). The lactate uptake rate of the resulting jen1 null mutant was close to the detection limit (Casal et al., 1999). Interestingly, export of lactate in engineered lactate producing S. cerevisiae is unaffected by deletion of JEN1 (our unpublished results). Even though mechanisms for import or export can differ, these combined observations suggest the presence of at least one alternative lactate transporter. To explore the presence of other lactate transporters in S. cerevisiae, a jen1Δ strain was constructed and evolved for growth on lactate as the sole carbon and energy source. The evolved strains were subjected to a combination of transcriptome analysis and whole-genome DNA (re)sequencing to identify the relevant mutations. The resulting lead genes were tested for their lactate transport activity via knockout studies in the evolved strains and introduction into nonevolved strains.
Materials and methods
Strains and maintenance
The S. cerevisiae strains used and constructed in this study (Table 1) are all derived from CEN.PK113-7D (van Dijken et al., 2000; Entian & Kotter, 2007). Stock cultures were grown at 30 °C in shake flasks containing either 100 mL synthetic medium (Verduyn et al., 1992) with 20 g L−1 glucose as carbon source or 100 mL complex medium containing 20 g L−1 glucose, 10 g L−1 Bacto yeast extract and 20 g L−1 Bacto peptone. After overnight growth, 20% (v/v) glycerol was added and 1-mL aliquots were stored at −80 °C.
Table 1. Saccharomyces cerevisiae strains used in this study
The plasmids and primers used in this study are listed in Tables 2 and 3, respectively. Transformations of S. cerevisiae were carried out according to the LiAc/ssDNA method (Gietz & Woods, 2002). Gene deletions were performed using the loxP-marker-loxP/Cre recombinase system, using pUG6, pUG66, pUG-natNT2 and pUG-hphNT1 as templates for PCR amplification of the knockout cassettes (Gueldener et al., 2002). Selection of knockout mutants was performed on agar plates containing 20 g L−1 glucose, 10 g L−1 Bacto yeast extract, 20 g l−1 Bacto peptone and either 200 mg L−1 G418, 10 mg L−1 phleomycin, 100 mg L−1 nourseothricin or 200 mg L−1 hygromycin B. The natNT2 marker was cloned from pFA6a-natNT2 (Janke et al., 2004) to pUG6 via the SacI and BglII restriction sites, thereby replacing the KanMX4 marker, resulting in pUG-natNT2. The JEN1 knockout cassette was amplified from pUG6 using primers JEN1 KO Fw and JEN1 KO Rv and transformed to CEN.PK113-7D, resulting in strain IMK302. Correct integration of the knockout cassette was confirmed using primer combinations JEN1 Ctrl Fw/KanMX4 Ctrl Rv and KanMX4 Ctrl FW/JEN1 Ctrl Rv. The URA3 knockout cassette was amplified from pUG-hphNT using primers URA3 KO Fw and URA3 KO Rv and transformed to IMK302, resulting in IMK322. Correct integration of the knockout cassette was confirmed using primer combinations URA3 Ctrl Fw/hphNT1 Ctrl Rv and hphNT1 Ctrl FW/URA3 Ctrl Rv. The KanMX4 and hphNT1 markers in IMK322 were removed using the Cre/loxP system with pSH47 (Gueldener et al., 2002), resulting in IMK338. Correct marker removal was confirmed using primer combinations JEN1 Ctrl Fw/JEN1 Ctrl Rv and URA3 Ctrl Fw/URA3 Ctrl Rv.
The first ADY2 knockout cassette was amplified from pUG-hphNT1 using primers ADY2 KO Fw A and ADY2 KO Rv B and transformed to IMK338, IMW004 and IMW005, resulting in IMK341, IMW032 and IMW033, respectively. Correct integration of the knockout cassette was confirmed using primer combinations ADY2 Ctrl Fw/hphNT1 Ctrl Rv and hphNT1 Ctrl FW/ADY2 Ctrl Rv. The second ADY2 knockout cassette was amplified from pUG-natNT2 using primers ADY2 KO Fw B and ADY2 KO Rv B and transformed to IMW032, resulting in IMW040. Correct integration of the knockout cassette was confirmed using primer combinations ADY2 Ctrl Fw/natNT2 Ctrl Rv and natNT2 Ctrl FW/ADY2 Ctrl Rv. The third ADY2 knockout cassette was amplified from pUG66 using primers ADY2 KO Fw C and ADY2 KO Rv C and transformed to IMW040, resulting in IMW041. Correct integration of the knockout cassette was confirmed using primer combinations ADY2 Ctrl Fw/ble Ctrl Rv and ble Ctrl FW/ADY2 Ctrl Rv. Presence or absence of the ADY2 gene was confirmed using primers ADY2 inside Fw and ADY2 inside Rv. The ADY2 gene including promoter and terminator was amplified from genomic DNA of CEN.PK113-7D, IMW004 and IMW005 using primers ADY2p Fw and ADY2t Rv, and the resulting PCR products were cloned into pCR-BLUNT II-TOPO using Gateway Technology (Invitrogen, Carlsbad, NM), resulting in pUD151, pUD153 and pUD152, respectively. The ADY2 alleles including promoter and terminator were subsequently cloned into pRS416 via the BamHI and XbaI sites, resulting in pUDC011, pUDC015 and pUDC012, respectively. pRS416, pUDC011, pUDC012 and pUDC015 were transformed into IMK341 and selected on a synthetic medium without uracil, resulting in IMZ271, IMZ272, IMZ273 and IMZ276.
Molecular biology techniques
PCR amplification was performed using Phusion Hot Start High Fidelity Polymerase (Finnzymes, Espoo, Finland) according to manufacturer's instructions in a Biometra TGradient Thermocycler (Biometra, Gottingen, Germany). DNA fragments were separated on a 1% (w/v) agarose (Sigma, St. Louis, MO) gel in 1xTAE (40 mM Tris–acetate pH 8.0 and 1 mM EDTA). Isolation of fragments from gel was performed with the Zymoclean Gel DNA Recovery kit (Zymo Research, Orange, CA). Restriction endonucleases (New England Biolabs, Beverly, MA) and DNA ligases (Roche, Basel, Switzerland) were used according to manufacturer's instructions. Transformation and amplification of plasmids was performed in Escherichia coli One Shot TOP10 competent cells (Invitrogen) according to manufacturer's instructions. Plasmids were isolated from E. coli with the Sigma GenElute Plasmid Miniprep Kit (Sigma). DNA constructs were routinely sequenced by Baseclear BV (Baseclear, Leiden, The Netherlands).
Laboratory evolution of strains IMW004 and IMW005
Glycerol stocks of IMK302 were used to inoculate two 500-mL shake flasks containing 100 mL of synthetic medium (Verduyn et al., 1992) with 5 g L−1l-lactic acid (Fluka 09578) and 25 mM MES buffer. The pH of the medium was set to 5.0 with 2 M KOH and 2 M H2SO4 prior to autoclaving (121 °C, 20 min). When growth was observed, a small aliquot (0.1–2 mL) was transferred to a new flask. After 10 transfers, the culture was plated on agar plates containing 20 g L−1 glucose, 10 g L−1 Bacto yeast extract and 20 g L−1 Bacto peptone to obtain single colonies. Per evolution line, four colonies were tested for growth on lactate. One colony per evolution line was stocked and named as IMW004 and IMW005.
Cultivation and media
Shake-flask precultures for characterization experiments were inoculated with glycerol stocks and grown in synthetic medium (Verduyn et al., 1992) with 20 g L−1 ethanol as carbon source.
Characterization of growth on lactate in shake flasks with an initial pH of 5.0 was performed in synthetic medium (Verduyn et al., 1992) with 5 g L−1 lactate and 25 mM MES buffer, as described earlier. Cells growing exponentially on ethanol were used to inoculate 500-mL shake flasks containing 100 mL medium and incubated in an Innova incubator shaker (New Brunswick Scientific, Edison, NJ) at 200 rpm and 30 °C.
Aerobic batch cultures for transcriptome analysis were carried out at 30 °C in 2-L laboratory fermentors (Applikon, Schiedam, The Netherlands) with a working volume of 1 L. Synthetic medium (Verduyn et al., 1992) was supplemented with 20 g L−1 lactic acid (Fluka 09578) as sole carbon source. The pH of this medium was set to 5.0 before autoclaving (120 °C). Antifoam Emulsion C (Sigma) was autoclaved separately (120 °C) as a 20% (w/v) solution and added to a final concentration of 0.2 g L−1. The culture pH was maintained at 5.0 by automatic addition of 2 M KOH and 2 M H2SO4. Cultures were stirred at 800 rpm and sparged with 500 mL air per min. The precultures were grown on ethanol as described earlier. Six hours before the inoculation of the batch cultures, 25 mM lactate was added to the preculture to induce lactate metabolism. Samples for RNA extraction were taken during the exponential growth phase at a cell dry weight concentration of 3–4 g L−1 and a lactate concentration of 5–7 g L−1. Sequential batch cultures for the characterization of growth rate and biomass yield were run in the same setup as described earlier, but with 10 g L−1 lactate to prevent possible nutrient limitations above 5 g L−1 cell dry weight. The first batch was inoculated with a preculture growing exponentially on ethanol, as described earlier. To increase reproducibility of the results, a sequential batch reactor (SBR) was used and the third cycle was sampled (Abbott et al., 2009).
Growth assays on agar plates
Growth on ethanol, lactate and pyruvate was tested by spotting 10 μL of serial dilutions of a culture growing exponentially on ethanol, on synthetic medium agarose (1% w/v) plates supplemented with 25 mM MES and either 5 g L−1 ethanol, lactate or pyruvate. Before spotting, the cultures were first washed 2 times and diluted to the appropriate cell concentration with synthetic medium (Verduyn et al., 1992). Pictures were taken after 6 days incubation at 30 °C.
Optical density at 660 nm was measured with a Libra S11 spectrophotometer (Biochrom, Cambridge, UK). Culture dry weights were determined via filtration of appropriate sample volumes (10–20 mL) over dry preweighed nitrocellulose filters (Gelman Laboratory, Ann Arbor, MI) with a pore size of 0.45 μm. After removal of the medium, the filters were washed twice with demineralized water, dried in a microwave oven for 20 min at 350 W and weighed. Culture supernatants were obtained after the centrifugation of the broth. Supernatants and media were analysed via HPLC using an Aminex HPX-87H ion exchange column operated at 60 °C with 5 mM H2SO4 as mobile phase at a flow rate of 0.6 mL min−1. Off-gas was first cooled in a condenser (2 °C) and dried with a Perma Pure Dryer (Permapure, Toms River, NJ). CO2 concentrations in the off-gas were measured with a NGA 2000 Rosemount gas analyzer (Rosemount Analytical, Orrville, OH).
Microarrays and analysis
Sampling of cells from the batch cultures and total RNA extraction was performed as described previously (Abbott et al., 2007). Probe preparation and hybridization to Affymetrix Genechip® microarrays were performed according to Affymetrix's instructions. The one-cycle eukaryotic target labelling assay was used, starting with 15 μg of total RNA. The quality of total RNA, cRNA and fragmented cRNA was checked using the Agilent BioAnalyzer 2100 (Agilent Technologies, Amstelveen, the Netherlands). Results for each strain were obtained from two independent culture replicates. The Significance Analysis of Microarrays (SAM version 1.12; Tusher et al., 2001) add-in to Microsoft Excel was used for comparison of replicate array experiments using a fold-change threshold of two and an expected false discovery rate of 5%. Transcript data have been deposited in the Genome Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE31634.
Whole-genome sequencing and analysis
DNA of the reference strain CEN.PK113-7D was prepared as described previously (Burke et al., 2000). A library of 200-bp fragments was created and sequenced paired-end using the Illumina Solexa system, generating c. 56 million 36-bp paired reads. These reads were mapped to the genome of S. cerevisiae strain S288C (SGD project) using the Burrows–Wheeler Alignment tool (BWA; Li & Durbin, 2009). Subsequently, a consensus sequence was generated using SAMtools’ pileup2fq (Li et al., 2009) with default parameters, except for the minimum depth and root-mean-square mapping quality, which were set to 10 and 25, respectively. Consensus bases not passing the filtering thresholds, mostly in repetitive regions, were replaced by ‘N’. This consensus was used as CEN.PK113-7D reference genome.
Genomic DNA from IMW004 and IMW005 was isolated using the Qiagen 100/G kit (Qiagen, Hilden, Germany). A library of 200-bp genomic fragments was created and sequenced paired-end (50-bp reads) using an Illumina HiSeq 2000 sequencer by Baseclear BV (Baseclear). The individual reads were mapped onto the genome of CEN.PK113-7D, using BWA (Li & Durbin, 2009) and further processed using SAMtools (Li et al., 2009). Single-nucleotide variations and small insertions and deletions were extracted from the mapping using SAMtools’ varFilter. Default settings were used, except that the maximum read depth was set to 300× (−D300) and the minimum small insertions and deletions score for single-nucleotide variations filtering was set to 60 (−G60). To minimize false positive mutation calls, custom Perl scripts and Microsoft Excel were used for further mutation filtering. First, mutation calls containing ambiguous bases in either reference or mapping consensus were filtered out. Second, only the single-nucleotide variations with a quality of at least 20 and small insertions and deletions with a quality of at least 50 were kept. Variant quality is defined as the Phred-scaled probability that the mutation call is incorrect (Ewing & Green, 1998). Third, mutations with a depth of coverage < 10× were discarded. Fourth, insertion and deletion mutation calls were only kept when at least 85% of the reads spanning the location confirmed the insertion or deletion. Fifth, small insertions and deletions that were close to an ‘N’ in the reference were removed, because this would complicate correct alignments and introduce false positive mutation calls. For small insertions and deletions, a window size of 40 bp was used, and for single-nucleotide variations, a window size of 5 bp. Copy number variation was analysed using CNV-seq (Xie & Tammi, 2009).
Pulsed-field gel electrophoresis and southern blotting
Chromosomes were separated using the CHEF yeast genomic DNA Plug Kit (170-3593; Bio-Rad, Richmond, CA) according to manufacturer's instructions and subsequently transferred onto Hybond-N+ nylon membranes (RPN303; Amersham Biosciences, Piscataway, NJ). Southern blotting, signal generation and signal detection were performed using the Gene Images AlkPhos Kit, CPD Star detection reagent and Hyperfilm ECL (RPN 3680, RPN3682 and 28-9068; Amersham Biosciences). The DNA probe used for southern blotting was amplified from CEN.PK113-7D genomic DNA with primers ADY2 inside Fw and ADY2 inside Rv (Table 3).
Laboratory evolution of Jen1p-independent growth on lactate
To test whether laboratory evolution and subsequent analysis of the underlying mutations is a powerful strategy to identify mutants that can transport lactate in S. cerevisiae, a jen1Δ knockout strain was constructed and evolved for growth on lactate. Hitherto, Jen1p is the only efficient importer of lactate in S. cerevisiae described in literature (Casal et al., 1999). Consistent with earlier reports, S. cerevisiae IMK302 (jen1∆) did not show growth on plates with lactate after 6 days, whereas the reference strain CEN.PK113-7D (JEN1) grew normally (Fig. 1). In shake-flask cultures with 5 g L−1 lactate at pH 5, the same results were obtained [μ < 0.001 h−1 for IMK302 (jen1∆)], but after prolonged incubation (5–15 days), growth was observed in two independent experiments. To select for faster growth, small aliquots of the shake-flask cultures were transferred to fresh medium for nine consecutive times. After c. 100 generations in 10 shake-flask cultures, single-colony isolates were obtained by plating on nonselective agar plates containing glucose, yeast extract and peptone. Per evolution line, four single-colony isolates were tested for growth on lactate and yielded essentially the same growth rates as the mixed population from which they were isolated. One single-colony isolate from each evolution experiment was chosen for further characterization. These resulting evolved strains were named IMW004 and IMW005 and grew at maximum specific growth rates of 0.14 and 0.18 h−1, respectively, on 20 g L−1 lactate in aerobic bioreactor batch cultures (pH 5.0).
Mutation analysis through whole-genome DNA sequencing and transcriptome analysis
Two techniques were used to study the molecular basis of Jen1p-independent growth on lactate in the evolved strains IMW004 and IMW005: transcriptome analysis with DNA microarrays and whole-genome DNA resequencing. In view of highly reproducible growth conditions and the ability to control the specific growth rate, chemostat cultures are normally the preferred cultivation type for comparative transcriptome analysis (Daran-Lapujade et al., 2008). However, chemostat cultures growing on lactate demonstrated persistent oscillations that precluded steady-state analysis (data not shown). Therefore, cell samples for transcriptome analysis were obtained from aerobic bioreactor batch cultures grown on 20 g L−1 lactate and at pH 5.0. To minimize differences in nutrient concentration and growth conditions, all cultures were sampled at a biomass dry weight concentration between 3 and 4 g L−1, corresponding to 5–7 g L−1 residual lactate. The average deviation of the mean of all genes with mRNA expression levels above background (in this case 12) was between 16% and 21%, which was only slightly higher than chemostat-based transcriptome analysis (10–15%; Basso et al., 2011). When comparing the transcriptome data of CEN.PK113-7D (JEN1) with IMW004 and IMW005 (both jen1Δ evolved), only a very small number of genes (11 for IMW004 and 18 for IMW005) were expressed at different levels based on the statistical criteria applied in this study (absolute fold difference ≥ 2; false discovery rate 5%, see 'Materials and methods'; Fig. 2). In addition to JEN1 (deleted in the evolved strain) and the introduced KANMX4 dominant marker gene, the general amino acid permease GAP1 was the only gene that was differentially expressed in both evolution lines IMW004 and IMW005. However, the transcript levels of GAP1 were only 2.6- and 2.0-fold higher in IMW004 and IMW005, respectively, than in the reference strain CEN.PK113-7D.
In parallel, genomic DNA of IMW004 and IMW005 (both jen1Δ evolved) was sequenced using Illumina technology, yielding c. 12.8 million 50-bp reads for IMW004 and c. 13.7 million reads for IMW005. The average depth of coverage was 51.5-fold for IMW004 and 56.6-fold for IMW005. The sequencing reads were mapped against the genome sequence of the reference strain CEN.PK113-7D, which was constructed by mapping CEN.PK113-7D sequencing reads to the S288C sequence. This resulted in 20 and 16 single-nucleotide variations for IMW004 and IMW005, respectively. No insertions or deletions were found. The set of relevant mutations was reduced to 7 for IMW004 and 10 for IMW005 (Table 4) by eliminating single-nucleotide variations outside coding regions, because no links were found with mRNA expression levels of adjacent genes. Nucleotide changes that only result in different codon usage for the same amino acid might affect translation efficiency and were therefore not discarded. Interestingly, several of the remaining single-nucleotide variations were either identical in both IMW004 and IMW005 or different mutations had occurred in the same genes (Table 4). Such mutations might (1) present genes that are relevant for the Jen1p-independent growth on lactate, (2) have accumulated during the construction of the jen1Δ strain and have been present before the start of the evolution experiment or (3) represent errors in the reference sequence.
Table 4. Mutation calls inside genes in IMW004 and IMW005 identified via whole-genome sequencing
Amino acid change
Genes mutated in both IMW004 and IMW005
Mitochondrial carnitine acetyltransferase
DNA damage checkpoint protein
Genes mutated only in IMW004
Part of transcriptional regulatory complex
Genes mutated only in IMW005
Cell wall mannoprotein
Member of seripauperin multigene family
Rearrangement of chromosome III
Rearrangement of larger DNA fragments, including deletions and duplications, is a well-known phenomenon during (yeast) evolution (Wolfe & Shields, 1997; Dunham et al., 2002; Wisselink et al., 2010; Dhar et al., 2011). To analyse these larger structural variations, the genome-wide mRNA expression levels were plotted according to their chromosomal location. Interestingly, both the mRNA expression levels and depth of coverage of DNA sequencing showed 2log ratios between 1 and 2 for a group of genes on chromosome III, when comparing IMW004 and CEN.PK113-7D (Fig. 3). Detailed investigation of the depth of coverage of DNA sequencing indicated a duplication of chromosome III between 83 and 126 kb, including the centromere, and a triplication between 126 kb to the end of the right arm (Fig. 4a). These observations led to the hypothesis that, in addition to chromosome III, an additional pseudopalindromic isochromosome III was formed containing the centromere and two copies of the right arm of chromosome III. Karyotyping confirmed the formation of an additional chromosome with a size of c. 475 kb (Fig. 5), which corresponds well to the proposed rearrangement (Fig. 4).
To exactly determine the chromosomal rearrangement, the DNA sequences 5 kb upstream and downstream of the observed breakpoints (i.e. 78–88 and 121–131 kb) were aligned with each other. The highest homology was observed between the Ty1-type long terminal repeat retrotransposons YCLWΔ15 and YCRCΔ6, with 96% homology over a region of 176 bp. To verify whether crossover took place between YCLWΔ15 and YCRCΔ6 as proposed in Fig. 4, primer combinations were designed (P5 and P6) that should give a 3041-bp PCR product if crossover had indeed taken place between YCLWΔ15 and YCRCΔ6. In line with the proposed mechanism of rearrangement, the expected 3-kb PCR product was obtained from IMW004, but not from CEN.PK113-7D genomic DNA. Sequencing of this DNA fragment showed that crossover took place between the last 98 homologous base pairs of YCLWΔ15 and YCRCΔ6.
Verification of the identified mutations in the acetate transporter ADY2
The ADY2 gene, which encodes an acetate transporter (Paiva et al., 2004), contained mutations that resulted in amino acid changes in both independently evolved strains (Table 4) and is located on the triplicated region of chromosome III of IMW004. To analyse whether the different copies of ADY2 in IMW004 all contained the same mutation, the individual sequencing reads mapping to ADY2 were analysed. At all positions, a constant signal was observed, showing that IMW004 carried three identical copies of ADY2C755G. ADY2 was not amongst the genes whose mRNA levels were significantly different from those in the reference strain (1.4-fold in IMW004 and 0.6-fold IMW005).
To investigate the physiological relevance of the mutation (C755G; Ala252Gly) and the apparent triplication of ADY2 in IMW004, knockout strains were constructed. Diagnostic PCR on genomic DNA of the single, double and triple knockout strains (IMW032, IMW040 and IMW041) only showed absence of ADY2 in the triple knockout strain (IMW041), thereby confirming that IMW004 contained three copies of ADY2C755G. Only the triple knockout strain was unable to grow on lactate (Fig. 1), confirming that Ady2pAla252Gly was essential for lactate import in the evolved strain IMW004.
To analyse the effect of ADY2C755G gene dosage on the growth rate on lactate, aerobic shake-flask experiments were performed with 5 g L−1 lactate at pH 5. IMW004 (3 copies of ADY2C755G) grew at a growth rate of 0.10 ± 0.01 h−1; IMW032 (2 copies) at 0.09 ± 0.00 h−1, IMW040 (1 copy) at 0.07 ± 0.00 h−1 and IMW041 (0 copies) did not grow (μ < 0.001 h−1).
To analyse whether the identified mutation (C655G; Leu219Val) in ADY2 could also fully explain the acquired phenotype in IMW005, a knockout of ADY2C655G was constructed. IMW033 (IMW005 ady2Δ) was not able to grow on lactate, thereby confirming the crucial role of Ady2pLeu219Val in lactate import in IMW005 (Fig. 1).
Reverse metabolic engineering of mutated ADY2 alleles
To verify whether mutation of ADY2 alone is sufficient to allow for growth on lactate, the mutated alleles were introduced into a nonevolved background (jen1Δ ady2Δ). IMZ271 (jen1Δ ady2Δ) and IMZ272 (jen1Δ ady2Δ + ADY2) did not show growth on agar plates with 5 g L−1 lactate after 6 days of incubation, whereas IMZ273 (jen1Δ ady2Δ + ADY2C755G) and IMZ276 (jen1Δ ady2Δ + ADY2C655G) did grow on these plates (Fig. 1). This confirmed that these alleles can confer the ability to transport lactate. IMZ271 (jen1Δ ady2Δ) could still grow on acetate (data not shown), which prevented kinetic analysis of acetate transport mediated by the mutated Ady2p alleles. Interestingly, IMZ271 (jen1Δ ady2Δ) could not grow on pyruvate, whereas IMZ272 (jen1Δ ady2Δ + ADY2), IMZ273 (jen1Δ ady2Δ + ADY2C755G) and IMZ276 (jen1Δ ady2Δ + ADY2C655G) could, showing that both the nonmutated and the mutated Ady2p alleles can function as efficient pyruvate importers in this strain background (Fig. 1), in contrast to previous reports on other S. cerevisiae strains (Akita et al., 2000).
Subsequently, two strains expressing the native ADY2 allele were compared. IMK302 (jen1Δ ADY2), which contains the native ADY2 gene on chromosome III, did not grow on lactate (μ < 0.001 h−1) in shake-flask cultures with 5 g L−1 lactate at pH 5. The fact that IMZ272 (jen1Δ ady2Δ + ADY2), containing the native ADY2 under control of its own promoter on a centromeric vector, grew very slowly on lactate (μmax around 0.02 h−1) demonstrates that even nonmutated Ady2p can transport lactate, although at a very low rate. The observed differences in growth may be due to small differences in expression level between chromosomal (IMK302) and centromeric (IMZ272) expression. This might raise the lactate consumption rate in IMZ272 marginally above the flux required for maintenance metabolism and thus allow for very slow growth (Pirt, 1975; Boender et al., 2009).
To further characterize the mutated alleles of ADY2, the reverse engineered strains IMZ273 (jen1Δ ady2Δ + ADY2C755G) and IMZ276 (jen1Δ ady2Δ + ADY2C655G) were tested in aerobic sequential batch fermentations with 10 g L−1 lactate at pH 5.0. The reference strain CEN.PK113-7D (JEN1 ADY2) was used for comparison, because in this strain, the large majority of lactate transport proceeds via Jen1p and not via Ady2p (Fig. 1). Both IMZ273 (jen1Δ ady2Δ + ADY2C655G) and IMZ276 (jen1Δ ady2Δ + ADY2C755G) grew rapidly on lactate at maximum specific growth rates of 0.14 ± 0.00 and 0.12 ± 0.00 h−1, respectively. In cultures of strain IMZ276, the Ady2pAla252Gly-dependent lactate consumption rate decreased towards the end of the fermentation, indicating a low affinity for lactate (Fig. 6). The biomass yield on lactate of IMZ273 (0.37 ± 0.00 g g−1; jen1Δ ady2Δ + ADY2C655G) was considerably higher than those of IMZ276 (0.30 ± 0.01 g g−1; jen1Δ ady2Δ + ADY2C755G) and CEN.PK113-7D (0.26 ± 0.01 g g−1; JEN1 ADY2).
Ady2p as lactate transporter
No protein structure of Ady2p or close homologues is available, and therefore, the effect of the mutations on the structure and function of Ady2p remains uncertain. However, the observation that single-nucleotide mutations turned the acetate transporter Ady2p into an efficient lactate transporter provides valuable information for future structure-function analysis. In both evolution lines, the identified mutations were located in putative transmembrane domains and resulted in amino acids that are one carbon atom smaller than the original amino acid (alanine to glycine in IMW004 and leucine to valine in IMW005). As lactate (C3H6O3) is structurally similar, but one carbon atom larger than acetate (C2H4O2), this suggests that extra space is created in the translocation site of Ady2p.
According to the literature, Jen1p catalyses an electroneutral lactate-proton symport, which is energetically equivalent to diffusion of the undissociated acid (Cassio et al., 1987; Casal et al., 1999). Complete aerobic dissimilation of 1 lactate then yields 7 ATP: 6 from oxidative phosphorylation at an in vivo P/O ratio of 1.0 (Van Gulik & Heijnen, 1995; Bakker et al., 2001) and 1 from substrate level phosphorylation in the tricarboxylic-acid cycle. To generate and maintain a proton-motive force, the plasma membrane H+-ATPase in yeast exports (only) a single proton for each ATP that it hydrolyses (van Leeuwen et al., 1992; Weusthuis et al., 1993). Lactate transport mechanisms that result in outward translocation of a proton and/or inward translocation of a negative charge can save ATP equivalents that would otherwise be used for the build-up or maintenance of the proton-motive force. These savings can therefore not exceed the in vivo activity of the plasma membrane H+-ATPase. In this way, lactate import via a lactate anion uniport or via a lactate-proton antiport mechanism could increase the apparent ATP yield from lactate dissimilation to 8 or 9 mol mol−1, corresponding to increases in the biomass yield on lactate of 14% and 29%, respectively. The reverse engineered strain IMZ273 (Ady2pLeu219Val) displayed a 24 ± 5% higher biomass yield on lactate than IMZ276 (Ady2pAla252Gly) and a 41 ± 3% higher biomass yield than the reference strain CEN.PK113-7D (Fig. 6), in which lactate import primarily proceeds via Jen1p (Fig. 1). Taking into account that an additional c. 5% of the biomass yield increase can be explained owing to higher specific growth rates of the evolved strains, which causes a lower impact of maintenance-energy requirements (Pirt, 1975; Boender et al., 2009), these biomass yield values might be consistent with a different mode of energy coupling of lactate transport. Further analysis of the energy coupling of lactate transport by these Ady2p isoforms, involving in vitro studies, will provide valuable insights into the energetics of organic acid transport in yeast and its impact on growth energetics.
Chromosomal rearrangement is a well-known phenomenon during (yeast) evolution (Wolfe & Shields, 1997; Dhar et al., 2011) and occurs often by recombination of Ty1-type retrotransposons (reviewed by Garfinkel, 2005). The occurrence of a c. 475-kb isocytochrome III in strain IMW004 shows how even a short (c. 100 generations) period of selective growth can select for a massive chromosomal rearrangement, in this case resulting in duplication of around 25 genes and triplication of about 100 genes. In addition to enhancing performance under the conditions that led to their selective advantage, such rearrangements may be ‘hopeful monsters’ that open up novel evolutionary paths via specialization of duplicated genes (Theißen, 2009).
In-depth analysis of the sequencing reads showed that all three copies of ADY2 in IMW004 contained the same nucleotide change (C755G), which suggests that the mutation in ADY2 occurred before the triplication event. This indicates that first a gain of function mutation in ADY2 provided the cells with a selective advantage over the parental strain. The subsequent retrotransposon-mediated triplication of ADY2C755G apparently gave sufficient evolutionary advantage to compensate for the burden of an additional chromosome. In line with this hypothesis, a strain derived from IMW004 containing only one copy of ADY2C755G (IMW040) showed a 30% lower growth rate on lactate. The observation that the second and third copy of ADY2C755G did not lead to a two- and threefold higher growth rates can be explained by the saturation of transcription, possibly due to a limited availability of transcription factors that regulate ADY2 expression, such as Cat8p, Adr1p and Snf1p (Casal et al., 2008). Indeed, the ADY2 mRNA expression level of IMW004 (three ADY2C755G copies) was only 40% higher than that of CEN.PK113-7D (one ADY2 copy). Simulation of the evolution experiment, based on the observed cell densities and the relatively small difference in growth rate between strains expressing one (0.07 h−1) or three (0.10 h−1) ADY2C755G copies, indicated that the formation of the isochromosome III occurred early in the evolution experiment before the first transfer. Accordingly, no further increase in the growth rate on lactate was observed in the remaining nine shake flasks of the evolution experiment.
Functional analysis through laboratory evolution and reverse engineering
Laboratory evolution combined with molecular analysis of the evolved strains enabled the successful identification in S. cerevisiae of genes encoding transporter mutants that can import lactate. Although this strategy can only be used in cases where the gene function can be linked to a selective advantage, it can be very useful in situations where only a low residual enzyme activity is observed for which the genetic determinants are unknown, such as for glucose transport or disaccharide hydrolysis in mutant strains of S. cerevisiae in which all known genes for these activities have been deleted (Wieczorke et al., 1999; De Kok et al., 2011). In addition, this approach can contribute genes and proteins with new, interesting properties for metabolic engineering. Elucidation of the genetic basis of the acquired phenotype is a crucial step in laboratory evolution-based functional analysis, which in this study was approached by a combination of transcriptome analysis and whole-genome resequencing. Subsequent verification via knockout studies and reverse metabolic engineering showed that the relevant changes could be identified via whole-genome resequencing, but not via transcriptome analysis. In yeast, only few studies have been performed to identify the relevant genetic changes and reverse engineer them into nonevolved strain background. When studies on both prokaryotic and eukaryotic laboratory evolved strains are considered, a clear picture emerges that, in many studies, transcriptome analysis did not allow rapid identification of relevant genetic changes that had been selected for in laboratory evolution experiments (Ferea et al., 1999; Riehle et al., 2003; van Maris et al., 2004c; Fong et al., 2005, 2006; Hu & Wood, 2010), even when combined with metabolome analysis (Vijayendran et al., 2008) or proteome analysis (Lewis et al., 2010). In contrast, studies in which the genetic basis for the acquired phenotype was elucidated and reconstructed via reverse metabolic engineering analysed the DNA level via whole-genome resequencing or hybridization of genomic DNA to microarrays (Herring et al., 2006; Gresham et al., 2008b; Conrad et al., 2009; Atsumi et al., 2010; Lee & Palsson, 2010; Tremblay et al., 2010; Hong et al., 2011). These observations support the view that analysis of the genome sequence rather than the transcriptome is the most promising first-line analytical approach for elucidating relevant changes that occurred during evolution. Transcriptome analysis can then be used as a follow-up in cases where whole-genome resequencing does not generate obvious and productive leads. In this respect, mRNA sequencing is an interesting development that allows simultaneous measurement of mRNA levels and identification of mutations inside coding regions (Wang et al., 2009; Oshlack et al., 2010). Independent of the technique used, analysing multiple independently evolved strains strongly facilitates dissection of relevant and random genetic changes.
We thank Marinka Almering, Mark Bisschops, Lizanne Bosman, Daniel Gonzalez Ramos, Marit Hebly and Duygu Yilmaz for experimental assistance. This work was financially supported by Tate & Lyle Ingredients Americas Inc. The Kluyver Centre for Genomics of Industrial Fermentations is supported by the Netherlands Genomics Initiative.