Strains and plasmids used and created in this study are listed in Tables 1, 2, respectively. Plasmids p414-TEF1p-Cas9-CYC1t and p426-SNR52p-gRNA.CAN1.Y-SUP4t were gifts from George Church (Addgene plasmids #43802 and #43803, respectively). The plasmid pROS13 was a gift from A. J. A. van Maris (Euroscarf plasmid #P30790). Plasmid p426-ATF1 and p426-EAT1 were constructed by introducing the appropriate gRNA sequence into the plasmid through PCR. A 5′ phosphoryl group was introduced on the 5′ ends of the PCR product through 5′ phosphorylated primers. The PCR product was then purified using the DNA Clean & Concentrator^TM-5 (Zymo research) and ligated with T4 ligase (NEB) according to the manufacturer’s protocol. Plasmid p426-ATF2 was constructed from a PCR amplified p426 backbone (Q5, NEB) and a synthetic gBlock (IDT), containing the ATF2 gRNA sequence and 50 bp homologous regions overlapping the linear plasmid backbone. The two fragments were assembled using Gibson assembly^® (NEB), following the manufacturer’s instructions. pROS13-EHT1 and pROS13-EEB1 were constructed by introducing the appropriate gRNAs as described previously (Mans et al., 2015). gRNAs were designed with the help of ChopChop (Labun et al., 2016). Plasmids, ligations and Gibson assemblies were routinely transformed to NEB^® 5-alpha chemically competent cells (NEB) according to the supplier protocol. Correct plasmid construction was confirmed by sequencing (GATC, Macrogen). S. cerevisiae plasmid transformations were performed as described previously (Gietz and Woods, 2002). Gene disruptions in S. cerevisiae CEN.PK2-1D (p414-TEF1p-Cas9-CYC1t) were performed by co-transformation with the appropriate p426 or pROS13 plasmid, together with a synthetic dsDNA repair fragment (IDT), as described previously (DiCarlo et al., 2013). The 140 bp repair fragment consisted of 70 bp homology directly upstream and downstream of the start and stop codon. Successful disruptions were confirmed by PCR, using genomic DNA as template (Lõoke et al., 2011). Plasmids derived from p426-SNR52p-gRNA.CAN1.Y-SUP4t were cured by growing the strains in the presence of 1 mg/mL 5-fluoroorotic acid (Boeke et al., 1984). Plasmids derived from pROS13 were cured by overnight growth without antibiotics and subsequent streaking on selective and non-selective plates. Colonies unable to grow on selective plates were deemed cured and used for the next round of gene disruption. Plasmid p414-TEF1p-Cas9-CYC1t was cured by counter-selection with 0.5 mg/ml 5-fluoroanthranilic acid (Toyn et al., 2000). Cured strains were used for physiological characterization.

Escherichia coli and S. cerevisiae cultures were routinely cultivated in LB (10 g/L tryptone, 5 g/L yeast extract, 10 g/L NaCl) and YPD medium (20 g/L glucose, 10 g/L yeast extract, 20 g/L peptone), respectively. 15 g/L bacteriological agar was added to make plates. Media were supplemented with 50 μg/mL ampicillin and 200 μg/mL geneticin when appropriate. E. coli and S. cerevisiae were cultivated at 37 and 30°C, respectively, unless stated otherwise. S. cerevisiae gene disruptions were performed in YNB-HL medium [5 g/L glucose, 6.7 g/L Yeast Nitrogen Base, vitamins and trace elements (Verduyn et al., 1992), 125 mg/L histidine-HCl and 500 mg/L leucine]. 75 mg/L tryptophan and 150 mg/L uracil were added to YNB-HL as needed.

Saccharomyces cerevisiae strains harboring pCUP1 derived plasmids were characterized in YSg medium (20 g/L glucose, 6.7 g/L Yeast Nitrogen Base without amino acids, 1.92 g/L medium supplements without uracil). Single colonies were picked into 10 mL YSg and incubated overnight while shaking at 150 rpm. The following day, 100 μL of the preculture was transferred to 10 mL fresh YSg in a 50 mL Greiner tube while shaking at 150 rpm. 1 mM CuSO₄ was added for gene induction. Cultures were sampled after 24 h. Samples were frozen at -20°C until analysis. Experiments were performed as biological duplicates.

Saccharomyces cerevisiae gene disruption strains were assayed in YPD-80 (YPD medium with 80 g/L glucose) and YSg-ura (YSg medium supplemented with 150 mg/L uracil). Single colonies were picked into 10 mL medium and incubated overnight while shaking at 150 rpm. The following day, 100 μL of the preculture was transferred to 10 mL fresh medium in a 50 mL Greiner tube while shaking at 150 rpm for 24 h. 2 mL sample was transferred to a 10 mL glass vial, sealed, and frozen at -20°C until analysis. Experiments were performed as biological duplicates. Samples were analyzed as technical duplicates.

White grape juice was prepared by diluting the Arsegan^® white grape juice concentrate five times. 1 g/L yeast extract was added to the medium which was filter sterilized. Subsequent HPLC analysis revealed that 58.8 g/L glucose and 53.3 g/L fructose were initially present in the medium. Precultures were grown in 10 mL medium in a 50 mL Greiner tube while shaking overnight at 150 rpm. The next day, 2 mL of the preculture was transferred to 50 mL fresh medium in a 100 mL Schott bottle. The bottles were closed with a rubber stopper and a water lock. Cultures were cultivated statically at 20°C for 7 days. 2 mL sample was transferred to a 10 mL glass vial, sealed, and frozen at -20°C until analysis. Experiments were performed as biological duplicates. Samples were analyzed as technical duplicates.

OD₆₀₀ was measured in flat-bottom 96-well plates (Greiner) in 200 μL sample using a Synergy MX microplate reader (BioTek). The light path was calculated as 0.632 cm and used to adjust the OD₆₀₀ values to the standard 1 cm light path. CO₂ production in white grape juice fermentations was estimated by measuring the weight of the bottles with the closed waterlock before and after the fermentation. The difference was assumed to be due to loss of CO₂ (van Rijswijck et al., 2017).

Sugars were measured by High Pressure Liquid Chromatography (HPLC) on an ICS5000 HPLC system (Thermo Scientific) equipped with a DP pump, AS-AP autosampler and a VWD UV detector (Dionex), at 210 nm and an RI detector (Shodex) at 35°C. An HPX-87H cation-exchange column (Aminex) was used with a mobile phase of 0.016 N H₂SO₄. The HPLC was operated at 0.8 mL/min and 60°C. A final concentration of 2 mM DMSO in 0.04 N H₂SO₄ was used as internal standard.

Ethanol was measured by GC-FID as described before (Kruis et al., 2017) on a Shimadzu 2010 gas chromatograph equipped with a temperature controlled 20i-s autosampler. 0.5 μL of liquid sample was injected on a Stabilwax column (30 m × 0.25 mm, 0.5 μm coating, Restek). The column temperature was held at 60°C for 1 min and increased to 120°C at a rate of 20°C/min. The split ratio was 20. 2 mM 1-butanol was used as internal standard.

The ester production profile of the yeast strains was assessed using Headspace-Solid Phase Microextraction Gas Chromatography-Mass Spectrometry (HS-SPME GC-MS) as described previously, with minor adaptations (van Rijswijck et al., 2017). A Trace 1300 Gas Chromatograph (Thermo Fisher) with a TriPlus RSH autosampler (Thermo Fisher) and an ISQ QD mass spectrometer (Thermo Fisher) was used for analysis of volatiles. Frozen samples were incubated at 60°C for 10 min. Volatile compounds were extracted for 20 min at 60°C using an SPME fiber (Car/DVB/PDMS, Supelco). The compounds were desorbed from the fiber for 2 min onto a Stabilwax^®–DA column (30 m length, 0.25 mm ID, 0.5 μm d_f, Restek). The PTV was heated to 250°C and operated in split mode at a ratio of 1:25. The GC oven temperature was kept at 40°C for 2 min, raised to 240°C with a slope of 10°C/min and kept at 240°C for 5 min. Helium was used as carrier gas at a constant flow rate of 1.2 ml/min. Mass spectral data were collected over a range of m/z 33–250 in full-scan mode with 3.0030 scans/s. Data was analyzed using Chromeleon^® 7.2. The ICIS algorithm was used for peak integration and the NIST main library to match the mass spectral profiles with the profiles of NIST. Peak areas were calculated using the MS quantification peak (highest m/z peak per compound).

Statistical analysis was performed using R (R Foundation for Statistical Computing, 2016). A one tailed Student’s t-test was performed. The p(critical) values were adjusted according to the Bonferroni method (Dunn, 1961) to account for multiple comparisons.

The protein-coding nucleotide sequences of ATF1, ATF2, EAT1, EEB1, EHT1, and IMO32 were retrieved from the 157 S. cerevisiae strains sequenced in Gallone et al. (2016) (de novo assemblies downloaded from NCBI BioProject accession: PRJNA323691). Local BLAST databases were set up for all the genomes and BLASTN (vBLAST+ 2.5.0) searches were performed (1E-04 E-value cut-off). The protein coding sequences of each gene from S. cerevisiae CEN.PK2-1D were used as queries. Multiple sequence alignments (MSAs) were obtained for each gene using MAFFT (v7.187), with default settings and 1,000 refinement iterations (Katoh and Standley, 2013). Maximum likelihood (ML) trees were constructed for each MSA using RAxML (v8.2.8), under the GTRGAMMA model (Stamatakis, 2014). Rapid bootstrapping (2000 bootstrap replicates) and search for the best-scoring ML tree were conducted in one single run (Pattengale et al., 2009). Trees were visualised and rooted in FigTree (v1.4.2) using Saccharomyces paradoxus (Scannell et al., 2011) as outgroup¹. Shannon entropy (H) was calculated for every position within the protein translated alignment of each gene (Shannon, 1948). Gaps were excluded from the calculation. Estimated ploidy and copy number variation data were obtained from Gallone et al. (2016).

Ester production profiles of diverse S. cerevisiae strains used in beer brewing, wine making and other industrial processes vary significantly. Gallone et al. (2016) previously reported the detection of ethyl acetate, isoamyl acetate, ethyl hexanoate and ethyl octanoate production by 157 industrial yeast strains with known genome sequences. The total amount of esters produced by the strains varied from 7 to 57 mg/L. Ethyl acetate was the most abundant ester (88.0–98.5% of the esters measured), followed by isoamyl acetate (0.9–10.0% of the esters measured). Ethyl hexanoate and ethyl octanoate constituted between 0.1 and 3.0% of the total ester produced by the strain collection. We investigated whether this large variability can be explained by differences at the genomic level. We compared the sequence diversity and copy number variation (CNV) of ATF1, ATF2, EHT1, EEB1, IMO32, and EAT1. All six genes were present in all strains of the yeast collection. The conservation of the translated nucleotide sequences was measured using Shannon entropy (Shannon, 1948). This method calculates the variability of each amino acid position in the sequence alignment. The conserved DUG3 and the variable ZRT1 genes were included in the analysis as a control. The average Shannon entropy of the protein sequences showed that Atf2p was the most variable while Atf1p was the most conserved AAT (Figure 1). However, all AATs were generally highly conserved, relative to DUP3 and ZRT1. We sorted the ester production profiles of the yeasts according to the phylogenetic distribution of each individual AAT. Even upon visual inspection it became clear that there was little to no correlation between the primary protein-sequence variation of the AATs and the ester production profiles (Supplementary Figure S1).

A typical characteristic of industrial S. cerevisiae strains is their high variability in chromosome content (ploidy). These strains often exhibit abnormal numbers of chromosomes relative to their expected ploidy (aneuploidy), or several small and local changes in CNV, such as duplications and deletions (Gallone et al., 2016). Based on the estimated ploidy levels and copy number profiles reported by Gallone et al. (2016), we examined the copy number levels of each AAT gene. Generally, copy number changes of the AAT genes were associated with duplications and deletions involving full chromosomes or large chromosomal fragments, relative to the rest of the genome. The comparison of the copy number changes of each AAT with their ester production profiles again showed no correlation with either the total ester amount produced, or with the ratios between the four measured esters (data not shown). The results of the in silico analysis suggest that the dramatic differences in ester production in the yeast collection cannot be explained by the variation of the coding regions of the AATs on a genomic level.

We determined the effect of ATF1, ATF2, EHT1, EEB1, IMO32, and EAT1 overexpression on ester production in S. cerevisiae CEN.PK2-1D. This haploid yeast strain was previously sequenced (Nijkamp et al., 2012) and contains only one copy of each AAT gene. The parental strain produced five different acetate esters (ethyl-, propyl-, isobutyl-, isoamyl-, and phenylethyl acetate), and two propanoate esters (ethyl- and isoamyl propanoate). These esters were produced from acetyl-CoA or propionyl-CoA, and a range of (fusel) alcohols formed innately by S. cerevisiae CEN.PK2-1D. The strain also produced four ethyl-MCFA esters (ethyl -hexanoate, -octanoate, -decenoate, and -decanoate). Overexpression of ATF1 resulted in a broad increase in acetate ester production, while overexpression of ATF2 caused increased levels of phenylethyl acetate and to a lesser extent isoamyl acetate production. An increase in phenylethyl acetate levels was also observed in strains overexpressing EHT1 and EEB1, although this increase was much lower compared to ATF1 and ATF2 (Figure 2A). The strain overexpressing EAT1 produced more acetate esters and ethyl propanoate but had no effect on MCFA production. The overexpression of IMO32 had no effect on ester production (Figure 2A).

Next, we assessed whether acetate and propanoate ester production were increased by other EAT genes. We overexpressed 15 members of the EAT family in S. cerevisiae CEN.PK2-1D, originating from nine yeast species (Figure 2B). It was found that 10 out of 15 overexpression strains showed increased production of acetate and propanoate esters, while medium chain fatty acid (MCFA) ester production was not affected. This was comparable to the overexpression of the S. cerevisiae EAT1. Ordering the ester production profiles of the overexpression strains according to a phylogenetic tree of the Eat protein sequences revealed that three groups seem to exist within the Eat family (Figure 2B). Two groups consist of EAT1 genes that are able to increase ester production. The third group consists of genes which had only a negligible effect on the ester profile compared to the other two EAT groups. It is possible that these enzymes are not involved in ester synthesis in yeast, although it cannot be excluded that these genes are not expressed well in S. cerevisiae CEN.PK2-1D. Curiously, W. anomalus and W. ciferrii EAT1 are more closely related to the non-producing EAT genes, but still increased ester production in S. cerevisiae CEN.PK2-1D. S. cerevisiae CEN.PK2-1D (pCUP1:Huv2) showed a different ester production profile compared to the other EAT1 expressing strains. The strain showed a 1291- and 590-fold increase in isoamyl acetate and phenylethyl acetate production, respectively (Supplementary Figure S2). These were some of the highest increases in ester production observed in this experiment. Huv EAT2 was also the only gene that significantly (>5-fold) increased the production of isoamyl propanoate. The remaining EAT1 homologs evoked comparable ester production profiles to S. cerevisiae CEN.PK2-1D (pCUP1: Sce Eat1).

Saccharomyces cerevisiae EAT1 clearly has the potential to produce acetate and propanoate esters in vivo, while the effect of IMO32 is unconfirmed. To determine the impact of EAT1 and IMO32 on the total ester production, we disrupted the two genes in combination with the other four known S. cerevisiae AAT genes (ATF1, ATF2, EHT1, and EEB1) in S. cerevisiae CEN.PK2-1D by CRISPR-Cas9 (DiCarlo et al., 2013; Mans et al., 2015). A total of 14 disruption strains were generated; six single knockouts (Δatf1,Δatf2,Δeht1,Δeeb1,Δeat1,Δimo32), two double knockouts (Δatf1Δatf2,Δeht1Δeeb1), two triple knockouts (Δatf1Δatf2Δeat1,Δeht1Δeeb1Δeat1), one quadruple knockout (Δatf1Δatf2Δeht1Δeeb1), two quintuple knockouts (Δatf1Δatf2Δeht1Δeeb1Δeat1, Δatf1Δatf2Δeht1Δeeb1Δimo32), and finally a sextuple disruption strain in which all six genes were disrupted (Δatf1Δatf2Δeht1Δeeb1Δeat1Δimo32). We determined their ester production profiles under three cultivation conditions. Two were based on routine laboratory cultivations in minimal YSg medium and rich YPD-80 medium while shaking. The latter medium was used previously to assess the effects of AAT disruptions on ester production (Verstrepen et al., 2003b). The third cultivation condition simulated industrial white wine fermentations and was performed statically. We first assessed the fermentation performance of the strains and found that there were only small differences in growth, glucose consumption or ethanol production in either YSg medium or YPD-80 medium (Supplementary Figure S3). However, the strains showed substantial differences in their ester production profiles. Three acetate esters (ethyl-, isoamyl-, and phenylethyl acetate) and three ethyl MCFA esters (ethyl- octanoate, 9-decenoate, and decanoate) were measured under these cultivation conditions.

Single deletions of ATF1 and ATF2 resulted in a slightly lower acetate ester production in both YSg (Figure 3A) and YPD-80 medium (Figure 3B). In contrast, CEN.PK2-1D Δeht1 produced less MFCA esters, but showed no differences in acetate ester levels in either media (Figure 3). The effects of these single disruptions were observed previously (Verstrepen et al., 2003b; Saerens et al., 2006), although the effect of the ATF1 and ATF2 deletions on acetate ester production in these studies was stronger. This may be due to the repression of ATF1 under aerobic conditions (Mason and Dufour, 2000). Disruption of EEB1 did not show any detectable effect on ester production in YSg or YPD-80 medium (Figure 3). CEN.PK2-1D Δeat1 showed significantly reduced levels of ethyl acetate in YPD-80 medium (Figure 3B). In some cases, the roles of the AAT genes in ester production became more apparent when multiple genes were disrupted simultaneously. For example, the ethyl acetate production of CEN.PK2-1D Δeat1 grown on YSg medium was similar to the CEN.PK2-1D parental strain. However, ethyl acetate production was reduced in CEN.PK2-1D Δatf1Δatf2Δeat1, CEN.PK2-1D Δeht1Δeeb1Δeat1 and CEN.PK2-1D Δatf1Δatf2Δeht1Δeeb1Δeat1 compared to their respective strains without the eat1 deletion (p < 0.0007). The gene disruptions mostly showed a cumulative effect. The quintuple disruption strain CEN.PK2-1D Δatf1Δatf2Δeht1Δeeb1Δeat1 produced the least esters. However, ester production was not completely abolished, indicating that other ester forming reactions exist in S. cerevisiae CEN.PK2-1D. This was the most apparent in the case of ethyl acetate and isoamyl acetate, where more than 50% ethyl acetate production remained in some cases (Supplementary Figure S4). Imo32p was previously associated with ester production in the absence of Atf1p (Holt et al., 2018). The single deletion strain CEN.PK2-1D Δimo32 produced 10% less ethyl acetate in both YSg and YPD-80 media. This effect disappeared when IMO32 was disrupted in CEN.PK2-1D Δatf1Δatf2Δeht1Δeeb1 and CEN.PK2-1D Δatf1Δatf2Δeht1Δeeb1Δeat1 (Figure 3). The strain where all six genes were disrupted still produced similar amounts of esters compared to CEN.PK2-1D Δatf1Δatf2Δeht1Δeeb1Δeat1. This indicates that IMO32 does not play a direct role in ester production in S. cerevisiae CEN.PK2-1D grown on YSg and YPD-80 media. The production of other esters, such as phenylethyl acetate and all MCFA esters was almost completely eliminated in CEN.PK2-1D Δatf1Δatf2Δeht1Δeeb1Δeat1 and CEN.PK2-1D Δatf1Δatf2Δeht1Δeeb1Δeat1Δimo32 grown on YSg or YPD-80 medium (Supplementary Figure S4). The effect of the AAT disruptions was then investigated under more industrially relevant conditions. The strains were cultivated statically in white grape juice, which contained 112.1 g/L sugars (58.8 g/L glucose and 53.3 g/L fructose). The gene disruptions affected the fermentation performance of the strains. This was most apparent in the strain where all AAT genes were disrupted. The growth, sugar consumption, ethanol formation and CO₂ formation were approximately 10% lower (Supplementary Figure S5). Five additional esters (isobutyl acetate, ethyl hex-4-enoate, ethyl hexanoate, isoamyl octanoate, and ethyl dodecanoate) were detected in white grape juice compared to YSg and YPD-80 medium (Figure 3).

The AAT disruption strains produced less esters in white grape juice compared to the parental strain (Figure 4). The lower production could hypothetically be explained by the reduced fermentation performance of the disruption strains. However, the changes in ester production were more dramatic than, e.g., the reduction of growth or sugar consumption. For example, CEN.PK2-1D Δatf1Δatf2Δeht1Δeeb1Δeat1 consumed approximately 10% less sugar than the parental strain, while ester production was decreased by more than 90% (Figure 4). The independence of ester formation from growth in this experiment was also apparent by comparing the relevant disruption strains. For example, CEN.PK2-1D Δatf1Δatf2Δeht1Δeeb1 and CEN.PK2-1D Δatf1Δatf2Δeht1Δeeb1Δeat1 showed no significant differences in fermentation performance (Supplementary Figure S5), whereas ethyl acetate and phenylethyl acetate were almost 50% or more lower in the strain where EAT1 was disrupted (Supplementary Figure S6, p-value < 0.0007). This result supports the observation in YSg and YPD-80 media (Figure 3).

Some CEN.PK2-1D AAT disruption strains behaved differently in white grape juice (Figure 4) compared to the cultivation in YSg and YPD-80 media. The most notable difference was observed in CEN.PK2-1D Δeat1, which showed a decreased production of both acetate and MCFA esters. This observation is not in line with the behavior of CEN.PK2-1D Δeat1 in laboratory media (Figure 3), or the overexpression experiment (Figure 3), where only an effect on acetate and propanoate ester production was observed. The deletion of EAT1 did not show the same effect on MCFA ester production in CEN.PK2-1D Δatf1Δatf2Δeat1, compared to CEN.PK2-1D Δatf1Δatf2. Rigorous genotyping and three independent repeats of the fermentation confirmed that the observed results are reproducible and not caused by technical issues. Moreover, CEN.PK2-1D Δeat1 served as the precursor strain for generating other disruption strains that did not show this behavior.

Disruption of ATF2 and EEB1 also resulted in different production profiles in white grape juice, compared to YSg and YPD-80. CEN.PK2-1D Δatf2 did not show a difference in the ester production profile compared to the parental strain, whereas it reduced acetate ester production in YSg and YPD-80 cultures. In contrast, the disruption of EEB1, which had no significant effect in laboratory media, resulted in a broad decrease of MCFA esters in white grape juice.

Deletion of ATF1 and EHT1 showed reduced production of acetate and MCFA esters, respectively, which is comparable to the profiles observed in laboratory media (Figure 3). The quintuple disruption strain in CEN.PK2-1D Δatf1Δatf2Δeht1Δeeb1Δeat1 again produced the least esters. The disruption of IMO32 either did not have an impact on ester production in wine juice, or these differences were not statistically significant (Figure 4). Residual levels of esters were still detected (Supplementary Figure S6), matching the observations in YSg and YPD-80 media (Figure 3).