Reciprocal hemizygosity analysis reveals that the Saccharomyces cerevisiae CGI121 gene affects lag time duration in synthetic grape must

Abstract It is standard practice to ferment white wines at low temperatures (10–18°C). However, low temperatures increase fermentation duration and risk of problem ferments, leading to significant costs. The lag duration at fermentation initiation is heavily impacted by temperature; therefore, identification of Saccharomyces cerevisiae genes influencing fermentation kinetics is of interest for winemaking. We selected 28 S. cerevisiae BY4743 single deletants, from a prior list of open reading frames (ORFs) mapped to quantitative trait loci (QTLs) on Chr. VII and XIII, influencing the duration of fermentative lag time. Five BY4743 deletants, Δapt1, Δcgi121, Δclb6, Δrps17a, and Δvma21, differed significantly in their fermentative lag duration compared to BY4743 in synthetic grape must (SGM) at 15 °C, over 72 h. Fermentation at 12.5°C for 528 h confirmed the longer lag times of BY4743 Δcgi121, Δrps17a, and Δvma21. These three candidates ORFs were deleted in S. cerevisiae RM11-1a and S288C to perform single reciprocal hemizygosity analysis (RHA). RHA hybrids and single deletants of RM11-1a and S288C were fermented at 12.5°C in SGM and lag time measurements confirmed that the S288C allele of CGI121 on Chr. XIII, encoding a component of the EKC/KEOPS complex, increased fermentative lag phase duration. Nucleotide sequences of RM11-1a and S288C CGI121 alleles differed by only one synonymous nucleotide, suggesting that intron splicing, codon bias, or positional effects might be responsible for the impact on lag phase duration. This research demonstrates a new role of CGI121 and highlights the applicability of QTL analysis for investigating complex phenotypic traits in yeast.


Introduction
Alcoholic fermentation for most white wines is performed at low temperatures (10-18 C), as this range generally results in greater production and retention of desirable volatiles, leading to highquality wines (Llauradó et al. 2002;Molina et al. 2007;García-Ríos et al. 2017). However, low temperatures also dramatically lengthen the time taken until fermentation completion and increase the risk of ferments becoming stuck or sluggish, which is potentially costly in terms of reduced winery space, product loss, and decreased profits (Bisson 1999;Colombie et al. 2005;Llauradó et al. 2005;Beltran et al. 2007;Ló pez-Malo et al. 2013). Low temperatures encountered during fermentation are particularly stressful to yeast and cause changes in cell membrane fluidity, nutrient uptake and utilization, production of protective compounds, and a decrease in enzymatic reaction rates (Beltran et al. 2007;Redó n et al. 2011;García-Ríos et al. 2016;Ganucci et al. 2018). A greater understanding of the genetics behind the ability of the wine yeast, Saccharomyces cerevisiae, to acclimate to low temperatures and perform fermentation more efficiently in general, is therefore useful for the wine industry.
The duration of the lag period at the start of fermentation, defined as the time between inoculation and the start of CO 2 release, and representing the time necessary for a yeast strain to acclimate to a new environment (Marullo et al. 2006), is greatly impacted by fermentation temperature, along with other variables encountered by yeast during fermentation. The high osmolarity of grape musts, along with the low pH, low-oxygen availability, oxidative stress, and potentially high levels of sulfur dioxide (SO 2 ), low levels of nutrients such as nitrogen, and to a lesser and strain-specific extent, phytosterols and thiamine, all contribute to the duration of the fermentative lag (Treu et al. 2014;Ferreira et al. 2017). Different S. cerevisiae strains also exhibit large variation in their fermentative lag duration ranging from a few hours up to a few days (Marullo et al. 2006;Camarasa et al. 2011). The genetic regulation controlling phenotypic variation in the fermentative lag time of different yeast strains is as complex as the variables involved and largely polygenic (Marullo et al. 2006(Marullo et al. , 2007. During the first few hours after inoculation in enological conditions, yeast must respond to the new environment with a dramatic metabolic reorganization, resulting in an increase in the synthesis of transcripts and proteins involved in carbon and nitrogen metabolism, cellular stress response, ribosomal biogenesis, protein synthesis and oxidative stress (Rossignol et al. 2003;Salvadó et al. 2008). Within this response, there are likely to be numerous genes and quantitative trait loci (QTLs) that influence the duration of the lag phase before the release of CO 2 . This response is more pronounced when the temperature of the must is low, lengthening the duration of the lag further (Salvadó et al. 2008;Albertin et al. 2017).
So far, one QTL with strong linkage to lag phase has been mapped to the SSU1 gene, encoding the SO 2 efflux pump (Peltier et al. 2018). Removal of SO 2 from the yeast cell is carried out via Ssu1p, in which there are several allelic variants and translocation events in different strains that alter Ssu1p efficiency (Perez-Ortin et al. 2002;Ferreira et al. 2017). Beneficial genetic variants allow yeast to pump out SO 2 more efficiently, significantly reducing lag time. Previous work in our laboratory investigated QTLs linked to fermentation kinetics and found two regions, one of Chr. VII and one on Chr. XIII, that were significantly linked to fermentative lag (Deed et al. 2017). Linkage analysis was performed on a set of 119/121 completely mapped (>99% of the genome) F 1 progeny from a cross between haploid strains BY4716 and RM11-1a constructed by Brem et al. (2002). Due to the difficulty in phenotyping lag phase in experiments with grape juice, and the large number of candidate genes within the confidence intervals surrounding the high logarithm of the odds (LOD) score peaks on Chr. VII [10 open reading frames (ORFs)] and Chr. VIII (34 ORFs), these 44 candidate genes were not investigated further. This previous identification of chromosomal regions linked to lag phase duration provides an excellent opportunity to investigate the causative genes using a controlled and reproducible fermentation medium, such as synthetic grape must (SGM). Because single reciprocal hemizygosity analysis (RHA) was not feasible for 44 different genes, we first aimed to test the lag duration of BY4743 single deletants of each candidate ORF identified in Deed et al. (2017). Those demonstrating variation in lag time compared to the BY4743 reference strain were deleted in haploids RM11-1a and S288C, followed by the construction of RHA hybrids. Phenotyping of deletants and RHA hybrids confirmed any relationships between the candidate ORFs with lag time phenotypes during fermentation.

Saccharomyces cerevisiae strains
We utilized laboratory strain BY4743 (MATa/a his3D1/his3D1 leu2D0/leu2D0 LYS2/lys2D0 met15D0/MET15 ura3D0/ura3D0) and 28 BY4743 homozygous diploid deletants derived from EUROSCARF containing a Kanamycin resistance construct (KanMX) in place of each ORF of interest (Table 1). The deletants were selected based on an original list of 44 candidates linked to lag phase in Deed et al. (2017) after linkage analysis of 119/121 BY4716 Â RM11-1a F 1 progeny using 2957 mapped loci (Brem et al. 2002). Of the 44 original candidates, 28 were available from EUROSCARF. Single gene deletions in three of the 28 candidates of interest were constructed in S288C (MATa), standing in for the BY4716 parent, and RM11-1a (MATa HO::HphMX) ( Table 2). Combinations of wild-type and deletant versions of S288C and RM11-1a were then used to make hybrids for RHA.

Growth and fermentation conditions
S. cerevisiae cultures were propagated using yeast peptone dextrose (YPD) medium and incubated overnight at 28 C, with orbital shaking at 150 revolutions per minute (rpm). Pre-cultures were washed in sterile water before further use via centrifugation for 5 minutes at 3,000 g. Growth curves were obtained using the Bioscreen C TM MBR Automated Growth Curve Analysis System, operated via the BioScreener TM software (Oy Growth Curves Ab Ltd.). Pre-cultures were used to inoculate YPD at 1 Â 10 6 cells ml À1 in quintuplicate wells of a 100-well honeycomb plate. Cells were grown at 25 C for 72 h following the protocol in Deed et al. (2019). BY4743 and BY4743 deletion mutants were fermented in 250-ml flasks with airlock at 12.5 C and 15 C in 100 ml SGM modeled on the chemical composition of Sauvignon blanc grape juice (Henschke and Jiranek 1993;Kinzurik et al. 2015). For fermentations using the BY4743 strains, SGM was supplemented with additional amounts of the following amino acids: 10 Â histidine (300 mg L À1 ), 10 Â leucine (300 mg L À1 ), and 10 Â uracil (100 mg L À1 ) (Harsch et al. 2010). RM11-1a and S288C wild types, deletants, and RHA hybrids were fermented at 12.5 C in 13-ml tubes with 8 ml SGM. A < 0.5 mm 2 pin-hole was punctured into each tube lid to allow for CO 2 escape (Deed et al. 2017). All fermentations were inoculated at density of 1 Â 10 6 cells ml À1 and were monitored either 8-hourly or daily by measuring cumulative weight loss (g) (Bely et al. 1990).

Analysis of kinetic parameters
The length of fermentative lag phase (h) of BY4743 and the 28 BY4743 deletants at 15 C was determined using the cumulative weight loss data to calculate the time elapsed between inoculation and the x-axis intercept where the steepest part of the slope transects y 0 , as per Marullo et al. (2006). Lag phase duration for all fermentations performed at 12.5 C was measured using a Gompertz model with curve fitting based on Tronchoni et al. (2009) and executed using the R package nlstools (Baty et al. 2015).

Gene deletions and reciprocal hemizygosity analysis
Deletion of three candidate genes, CGI121, RPS17a, and VMA21, within either the Chr. VII or XIII QTLs linked to lag phase were constructed in RM11-1a Hgm R and S288C using a modification of the Schiestl and Gietz (1989) lithium acetate yeast transformation protocol. Transformation of haploid RM11-1a and S288C was performed independently to generate mutants with KanMX insertions in CGI121, RPS17a, and VMA21 by amplifying the corresponding constructs, CGI121::KanMX, RPS17a::KanMX, and VMA21::KanMX, from BY4743 EUROSCARF deletion library strains. Transformation with a Nat R pFLR-A plasmid was used as a positive control. Successful deletions were confirmed via PCR (list of oligonucleotide primers in Table 3) and gel electrophoresis. Crosses were made between RM11-1a and S288C wild types, and combinations of nondeleted RM11-1a with each S288C deletion mutant and vice versa, in order to construct diploid hemizygous F 1 hybrids for RHA (Steinmetz et al. 2002b) (crosses in Table 2).
Since there were no markers in the S288C parent, this strain had to be present in 100 Â excess of the RM11-1a deletion strain parent for mating (1 Â 10 8 cells ml À1 S288C wild type with 1 Â 10 6 cells ml À1 RM11-1a Hgm R Kan R deletion strain). Hybrids were selected on YPD plates containing 300 lg L À1 hygromycin B and 200 lg L À1 G-418. A multiplex PCR to amplify 10 variable microsatellite markers and two mating-type loci, MATa and MATa, was used to ensure that the hybridization was successful and to finalize strain selection since there would be some RM11-1a Table 3 Oligonucleotide primers used for gene deletions and RHA Primer name Sequence (5 0 to 3 0 ) Purpose Flanking test primer to confirm integration of KanMX in the VMA21 locus of transformants The genotypes are given for each of the RM11-1a and S288C parents. The S288C parent strain in bold was required to be present in 100 Â excess of the RM11-1a parent, due to the lack of selectable markers to differentiate it from RM11-1a. The F 1 hybrid selections marked with * could result in the presence of the RM11-1a parent and the F 1 hybrid. The RM11-1a Â S288c cross was included as a control.

Results
First screening of 28 BY4743 deletion mutants fermented in SGM at 15 C identified five candidate ORFs that may influence lag time Of the 44 S. cerevisiae genes identified within the 95% confidence intervals of the high LOD score peaks for QTLs on Chr. VII and XIII linked to fermentation lag duration in Deed et al. (2017), 28 single-gene deletion mutants were available from EUROSCARF (listed in Table 1). Of the 16 ORFs that were unavailable, seven were classified as essential genes and hence inviable in a null mutant according to the Saccharomyces Genome Database. The remaining nine either encoded transposable elements (six ORFs) or were classified as dubious and unlikely to encode a protein (three ORFs). Cumulative weight loss (g) of the 28 BY4743 deletants fermented in 100 ml SGM at 15 C was measured at 8-h intervals for 72 h as a quick initial screen to identify whether any of the ORFs have an impact on the duration of the fermentative lag compared to the BY4743 reference (Figure 1, A-D). Because it was not feasible to perform RHA on 28 different candidate genes, this initial step was conducted to narrow down the number of candidates. Due to the large number of fermentations in triplicate, the deletants were fermented in four separate batches, each with the BY4743 reference for standardization, and an uninoculated control as a measure of evaporation and to ensure there was no contamination. Figure 1, A-D shows that the 28 deletants demonstrated a range of fermentation abilities at 15 C in SGM, with strong visual indications of variation in lag phase time compared to the BY4743 reference. The lag duration of BY4743 and the 28 deletants was calculated from the weight loss curves and presented in Figure 2, A-D. The lag time for BY4743 across the four batches ranged from 40 to 52.8 h, with a mean of 45.7 h (n ¼ 12). This degree of variation demonstrates the difficulty of measuring lag time due to the high level of noise at the start of fermentation. There were no significant differences between the BY4743 deletants in batch 1 compared to BY4743 (Figure 2A). In batches 2 and 3, the lag phase times of BY4743 Drps17a (56.6 h) ( Figure 2B) and BY4743 Dvma21 (48.7 h) ( Figure 2C) were significantly longer than BY4743 (43.6 h), while BY4743 Dclb6 (37.3 h) had a significantly shorter lag phase ( Figure 2C). In batch 4, two deletants, BY4743 Dapt1 and BY4743 Dcgi121, had two replicates each that had not yet left lag phase ( Figure 2D). For a useful comparison to be made against BY4743 (44.9 h), the lag times for these replicates were set at 70 h, giving an average duration of 63.5 h for BY4743 Dapt1 and 63.6 h for BY4743 Dcgi121, although the actual measure of lag time is likely to be longer for these deletants.
Further screening at 12.5 C confirms that BY4743 single deletions of Dcgi121, Drps17a, and Dvma21 significantly alter fermentative lag time Because the five candidate genes identified above were selected across three different fermentation batches with a degree of noise, and with some strains still in fermentative lag or unable to ferment, a repeat single-batch 100-ml fermentation was performed for the five deletants and BY4743 to confirm that the lag phase differences observed were repeatable. The fermentations were also performed over a longer timeframe (528 h) than was used previously to determine whether the mutants that did not initiate fermentation were still in lag phase or were unable to ferment. A temperature of 12.5 C was selected to provide a greater resolution in lag phase duration compared to 15 C, whilst maintaining an enologically relevant temperature. Prior to the Numbers are band sizes in bp. The 12 loci detected correspond to 10 variable microsatellite loci and two mating-type loci, MATa and MATa, as described in Richards et al. (2009). fermentation experiment, the growth of the six strains in YPD at 25 C was also measured for 72 h to ensure that the number of viable cells in each YPD pre-culture was equivalent. This check was performed to ensure that potential variation in the starting number of live cells was not a factor, since differences would confound the length of fermentative lag. By $24 h, all strains had reached stationary phase ensuring that the fresh inoculum added to each ferment, after adjusting based on cell counts, was the same (Supplementary Figure S1). Figure 3 shows the weight loss curves at 12.5 C for the five deletants and BY4743. The results from the first screening at 15 C were conserved at 12.5 C, with BY4743 and BY4743 Dclb6 demonstrating an earlier exit from fermentative lag compared to BY4743 Dcgi121, BY4743 Drps17a, and BY4743 Dvma21. Surprisingly, with the extension of the fermentation timeframe, it was revealed that the performance of the Dapt1 deletant was equivalent to the uninoculated control, with no initiation of fermentation. The Dapt1 deletant was capable of growth in YPD in  Batches from 1 to 4 and each batch included BY4743 for standardization. Error bars represent 95% confidence intervals. Student's t-test was used to generate P-values between BY4743 and every single deletant (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001).
the fermentation pre-cultures (Supplementary Figure S1), suggesting that this strain may either be deficient in a specific factor required for fermentation, the enological environment was not permissible for the growth of this strain, and/or a lack of nitrogen in the SGM limited nucleotide biosynthesis given that Apt1p is involved in the purine salvage pathway (Alfonzo et al. 1995). The lag phase duration was calculated for the remaining strains using a modified Gompertz curve-fitting model to obtain greater accuracy compared to the intercept method used in the quick screen (Tronchoni et al. 2009). Overall, lag times at 12.5 C compared to 15 C were approximately twofold longer, as expected when decreasing fermentation temperature (Charoenchai et al. 1998;Torija et al. 2003; Figure 4). The lag times confirm the prior observations from the weight loss curves in Figure 3, but with no significant difference between the lag times of the two fastest strains, BY4743 (64.9 h) and BY4743 Dclb1 (59.1 h) (Figure 4). The lag times of BY4743 Dcgi121 (149.6 h), BY4743 Drps17a (130.7 h), and BY4743 Dvma21 (119.9 h) were not significantly different from one another based on the 95% confidence intervals, but were significantly longer than the lag times of BY4743 and BY4743 Dclb6.
To summarize, fermentation screening successfully identified three genes resulting in a longer lag phase when deleted (Dcgi121, Drps17a, and Dvma21). These were further investigated using single RHA.
Construction of RM11-1a and S288C single gene deletions and RHA hybrids reveals that the CGI121 gene impacts on lag phase duration To determine whether any of the three candidates, CGI121 (Chr. XIII), RPS17a (Chr. XIII), or VMA21(Chr. VII), were responsible for the high LOD scores and genetic linkage to fermentative lag phase in the original 119 BY4716 Â RM11-1a mapped progeny, single deletions of these three ORFs were constructed in two haploid S. cerevisiae strain backgrounds, RM11-1a (Hgm R ) and S288C. S288C was used as a substitute for BY4716, as in Deed et al. (2017). For the three candidate genes, all combinations of RM11-1a and S288C single deletants with the corresponding wild-type were hybridized for RHA (Table 2). Successful hybridization was confirmed using microsatellite typing (Table 4).
Fermentation in SGM at 12.5 C was performed for 192 h, with 8-hourly monitoring, using the RM11-1a and S288C parent strains, the haploid Dcgi121, Drps17a, and Dvma21 single deletants in RM11-1a and S288C, the RM11-1a Â S288C F 1 hybrid and the RHA F 1 hybrids constructed by crossing combinations of RM11-1a and S288C. The RHA hybrids were hemizygous for a null allele and either the RM11-1a copy or the S288C copy of CGI121, RPS17a, or VMA21. Cumulative weight loss curves show that the diploid RM11-1a Â S288C F 1 hybrid had a superior fermentation performance compared to the haploid parents, RM11-1a and S288C, based on the emergence from fermentative lag and rate of fermentation ( Figure 5, A-C). RM11-1a and S288C performed similarly, and in all cases exhibited a much shorter lag time compared to all RM11-1a and S288C single deletion mutants in Dcgi121, Drps17a, and Dvma21, in agreement with the results observed for BY4743. This result confirms that the presence of CGI121, RPS17a, and VMA21 results in faster lag times. The RM11-1a Â S288C Dcgi121 hybrid appeared to exit fermentative lag at the same time as RM11-1a Â S288C, while the lag phase of RM11-1a Dcgi121 Â S288C was longer ( Figure 5A). There did not appear to be any difference between RM11-1a Â S288C Drps17a or RM11-1a Drps17a Â S288C in terms of fermentation performance, and potentially only a minor difference in lag time compared to RM11-1a Â S288C ( Figure 5B). The same trend was observed for RM11-1a Â S288C Dvma21 and RM11-1a Dvma21 Â S288C; however, both hemizygotes showed a noticeably longer lag time than RM11-1a Â S288C ( Figure 5C). Figure 6A confirms that the lag times for RM11-1a and S288C Dcgi121, Drps17a, and Dvma21 single deletants were significantly longer than nondeleted RM11-1a and S288C (average of 390 h compared to 126 h), as suggested from the weight loss curves in Figure 5, A-C. The long lag times of the deletion mutants corroborate the results shown by the BY4743 Dcgi121, Drps17a, and Dvma21 deletants, but with even greater lag duration in RM11-1a and S288C due to the generally poor fermentation performance of haploid strains (Li et al. 2010). There were no significant differences between the nondeleted RM11-1a and S288C strains or between the corresponding pairs of RM11-1a and S288c single deletion mutants in Dcgi121, Drps17a, or Dvma21. In addition, there were no significant differences in lag time between RM11-1a Dcgi121, Drps17a, and Dvma21 single deletants. The same result was observed for the S288C single deletants. For the RHA hybrids ( Figure 6B), the lag time of the RM11-1a Â S288C Dcgi121 hybrid was not significantly different from the RM11-1a Â S288C wild type (average of 122 and 121 h, respectively). However, the RM11-1a Dcgi121 Â S288C hybrid had a significantly longer lag time (149 h), suggesting that the presence of the RM11-1a CGI121 allele results in a lag time equivalent to wild type, but the S288C version results in increased lag time. This result is strong evidence validating the role of CGI121 on impacting the duration fermentative lag and corresponds to mapping data indicating that the longer lag time is consistent with the presence of the S288C CGI121 allele and not the RM11-1a copy in the homozygous F 1 progeny from the original cross (Deed et al. 2017). We aligned the RM11-1a and S288C nucleotide sequences of CGI121 to determine whether there were any allelic differences (Supplementary File S1). However, nucleotide alignment showed that the sequences were 99% identical and the single base difference observed at 282 bp (G in RM11-1a and A in S288C) was synonymous, with both codons corresponding to a phenylalanine (AAG vs. AAA). Further alignment of 1 kb in front of the coding sequence of the RM11-1a and S288C CGI121 sequences did not uncover any nucleotide differences in the promoter region. Because Cgi121p is one of five members of the endopeptidase-like and kinase associated to transcribed chromatin (EKC)/kinase, endopeptidase and other proteins of small size (KEOPS) protein complex (Srinivasan et al. 2011), we also aligned the RM11-1a and S288C nucleotide sequences for the four other genes, BUD32, GON7, KAE1, and PCC1 (Supplementary File S2). Nucleotide alignment for the RM11-1a and S288C BUD32 and PCC1 alleles were 100% identical, while the alignment for GON7 was 99.2% (3 bp) and was 99.7% for KAE3 (4 bp). All GON7 and KAE3 substitutions were synonymous.
For RPS17a, as suggested by the weight loss curves, there was no significant difference in lag time between RM11-1a Â S288C Drps17a or RM11-1a Drps17a Â S288C, suggesting that neither allele impacts on lag time, even though RM11-1a Drps17a Â S288C did have a slightly longer lag than RM11-1a Â S288C (138 h vs. 121 h). RM11-1a Â S288C Dvma21 and RM11-1a Dvma21 Â S288C were also not significantly different from one another, with no allele-specific impacts on lag duration for VMA21. The lag times for both hemizygotes were significantly longer than RM11-1a Â S288C (144 and 149 h vs. 121 h) suggesting an additive effect with two copies of the VMA21 gene being beneficial for a shorter lag time.
Overall, these results have demonstrated a clear role of CGI121 on Chr. XIII for altering fermentative lag time, and although RPS17a and VMA21 did not show allelic differences in terms of their impact on lag time, both genes have a clear effect on lag duration when deleted.

Discussion
Through genetic linkage analysis from a set of completely mapped 119 BY4716 Â RM11-1a F 1 progeny, fermentation screening of single BY4743 deletants in candidate genes to narrow down the field, and RHA using RM11-1a and S288C, we have identified the relationship between the CGI121 gene on Chr. XIII with fermentative lag time duration, which likely corresponds to the high LOD score on Chr. XIII (Deed et al. 2017). Deletion of Dcgi121 in homozygous diploid BY4743, and haploids RM11-1a and S288C, resulted in a significant increase in fermentative lag in SGM at 12.5 C, compared to the corresponding wild types. The effect of the CGI121 gene in fermentative lag phase was different in the hemizygous single RHA F 1 hybrids, depending on whether they harbored the RM11-1a or the S288C allele, i.e., the RM11-1a Dcgi121 Â S288C F 1 hybrid had a significantly longer fermentative lag duration than RM11-1a Â S288C and RM11-1a Â S288C Dcgi121. Mapping data from Deed et al. (2017) determined that the difference in CGI121 in the F 1 progeny was derived from the S288C allele. Transcriptomics data also demonstrated that CGI121 transcripts are upregulated by at least twofold in an M2 Â S288C F 1 hybrid versus the M2 parent during the early stages of fermentation (at 2% weight loss) at 12.5 C, suggesting a key difference in the regulation of the S288C CGI121 allele.
Although the single nucleotide difference between the RM11-1a and S288C CGI121 alleles was synonymous, it has been reported that synonymous mutations can result in differences in gene expression, with the use of particular codons significantly increasing transcript numbers (Plotkin and Kudla 2011). In addition, CGI121 contains an intron at 457-562 bp, which is a relatively uncommon feature of yeast protein-coding genes (only 5%). Differences in the regulation of genes from strain to strain can be caused by variation in intron splicing efficiencies, which can be modulated by the stress response (Pleiss et al. 2007). The CGI121 intron is also classified as having an unstable and unstructured branch point (BP) (Gahura et al. 2011) and a predicted novel type of BP (Gould et al. 2016), which may further impact on splicing efficiency. Therefore, it would be interesting to determine whether the splicing efficiencies of the RM11-1a and S288C CGI121 introns are different. Alternatively, there could be cis or trans regulatory effects depending on the CGI121 allele position (Brem et al. 2002;Sinha et al. 2006). Investigation into whether any of the other members of the EKC/KEOPS complex displayed allelic difference in RM11-1a compared to S288C did not provide any important differences, with only synonymous base changes, highlighting how highly conserved this complex is (Srinivasan et al. 2011). Role of CGI121 and evidence for impact on fermentative lag time CGI121 (YML036W) is a 652 bp gene encoding a small polypeptide component EKC/KEOPS protein complex with roles in transcription, telomere uncapping, chromosome segregation, and DNA repair, and is specifically required for threonine carbamoyl adenosine (t 6 A) tRNA modification and telomeric TG 1-3 recombination and length regulation (Kisseleva-Romanova et al. 2006;Srinivasan et al. 2011;Liu et al. 2018). There are five proteins within this complex, encoded by BUD32, CGI121, GON7, KAE1, and PCC1. Of the five genes, only Dkae1 null mutants are inviable, due to the severe growth impairment and chromosomal instability caused by deleting this essential gene, which encodes an ATPase (Downey et al. 2006;Mao et al. 2008). The role of Cgi121p in the EKC/KEOPS complex is to regulate Bud32p kinase activity by interacting with the N-terminal lobe, which in turn regulates the Kae1p ATPase, allowing for downstream function and catalytic activities (Mao et al. 2008;Zhang et al. 2015). Cgi121p does not directly participate in the t 6 A tRNA modification function of the complex, but is important for telomere length regulation and recombination (Downey et al. 2006;Srinivasan et al. 2011;Peng et al. 2015), and may also be involved in creating stable connections between each KEOPS subunit, allowing for correct assembly (Perrochia et al. 2013). In the S. cerevisiae EKC/KEOPS complex, Cgi121p is not required for retaining functionality but is required for maximal activity, with the phenotypes of Dcgi121 mutants being much milder than those displayed by Dbud32, Dkae1 or Dpcc1 mutants (  Lag time duration (h) of RM11-1a, S288C, and respective single deletants in Dcgi121, Drps17a, and Dvma21 (A) and RHA hybrids comparing the impact of RM11-1a and S288C alleles of CGI121, RPS17a, and VMA21 (B) fermented in SGM at 12.5 C for 192 h (n ¼ 3). Error bars represent 95% confidence intervals. Samples sharing the same letter are not significantly different (ANOVA followed by post-hoc Tukey's HSD).
Classical genetics studies have shown that null mutants of Dcgi121 have increased replicative lifespan and viability, and reduced single-stranded DNA at uncapped telomeres which functions to initiate telomere recombination (Downey et al. 2006;Peng et al. 2015). Deletion of Dcgi121 in BY4742 resulted in cells with a 50% longer lifespan, as the absence of CGI121 inhibits telomere recombination and therefore provides greater genome stability (Peng et al. 2015). Large-scale surveys have implicated the Dcgi121 deletion in causing reduced vegetative and fermentative growth rates; however, data from Srinivasan et al. (2011) suggests that the vegetative growth of a W303-1A Dcgi121 mutant was close to wild type on solid medium after two days growth at 30 C. In the propagation of BY4743 Dcgi121 for fermentation in this research there did not appear to be any difference in vegetative growth in YPD compared to BY4743, with equivalent cell titres (data not shown), but there could be a difference in lag phase earlier on in vegetative growth which was not observable after 24 h of growth at 28 C. In terms of fermentative growth, Dcgi121 was identified by Steinmetz et al. (2002a) as showing reduced growth on YPD with 2% glucose; however, this screen was aerobic and does not adequately represent the fermentation environment. Hoose et al. (2012) identified S288C mutants in Dcgi121 as having an increased duration of cell cycle progression in G 1 phase, with the percentage of S288C Dcgi121 G 1 cells greater than two standard deviations (41.6%) above wild-type S288C at equivalent measurement times. The longer period spent in G 1 phase would mean that Dcgi121 cells do not divide as often as wild type and can explain the longer lag time during fermentation. Cell division and vegetative growth influences the timeframe of the fermentative lag phase and stressful environmental conditions, such as those encountered in the enological environment can significantly prolong G 1 (Hoose et al. 2012), which could be why the impact of the Dcgi121 was more pronounced during fermentation at low temperature. The presence of certain nutrients also influences the timing through G 1 to START, from where the rest of the growth cycle can be completed. When shifting from poor to rich medium, the G 1 phase can be prolonged temporarily until the cells reach a critical size allowing them to commit to a phase of cell division (Hoose et al. 2012). In Dcgi121 mutants, there is a decreased rate of carbon and nitrogen utilization, with abnormal glucose and arginine metabolism (VanderSluis et al. 2014), as well as an upregulation of carbohydrate metabolism genes in Dcgi121 mutants compared to wild type (Chou et al. 2017). Therefore, abnormal usage of glucose, a primary carbon source in grape juice and SGM, as well as decreased nitrogen consumption and accumulation of arginine, could greatly influence the lag duration of Dcgi121 mutants during fermentation.

Impact of RPS17a and VMA21 on fermentative lag duration
We have also shown that along with the single deletion in Dcgi121, single deletions in Drps17a [encoding a ribosomal protein of the small 40S subunit (Abovich et al. 1985)], and Dvma21 [encoding an integral membrane protein required for V-ATPase function (Hill and Stevens 1994)] resulted in an extended lag time duration in BY4743, RM11-1a, and S288C; however, neither RPS17a nor VMA21 provided clear evidence for any allelic differences via RHA analysis. Interestingly, Drps17a mutants demonstrate a prolonged G 1 phase in the cell cycle, in the same way as Dcgi121 (Hoose et al. 2012), which could explain the influence of the null mutant on fermentative lag. Null mutations in Dvma21, result in a multitude of phenotypes in S. cerevisiae, with decreased resistance to oxidative and osmotic stress (Dudley et al. 2005), and decreased thermotolerance (Jarolim et al. 2013), all of which can result in a longer fermentative lag time (Ferreira et al. 2017). Although Dvma21 mutants also had a decreased carbon utilization rate, these were for nonfermentable carbon sources (Dudley et al. 2005;VanderSluis et al. 2014). The QTL responsible for the high LOD score on Chr. VII in Deed et al. (2017) is yet to be identified, but may be derived from the RM11-1a parent, which would mean that the initial BY4743 screen was not so useful for pinpointing the QTL responsible. Future work could investigate whether any of the seven essential genes that were not assessed out of the original 44 candidates play a role in lag duration, by screening for evidence of haploinsufficiency in the fermentative lag phenotypes of diploids that are hemizygous at these loci.

Conclusions
We have shown that single deletions of Dcgi121, Drps17a, and Dvma21 result in increased fermentative lag duration in S. cerevisiae. This research has also demonstrated that the CGI121 gene, encoding a component of the EKC/KEOPS complex, plays a role in modulating the fermentative lag phase in S. cerevisiae. RHA confirmed that the S288C-derived CGI121 allele accounted for a longer lag time. A greater understanding of the role of the CGI121 in stress tolerance will allow easier manipulation and/or selection of S. cerevisiae strains to shorten or lengthen lag time and provide growth advantages during the fermentation of foods and beverages.