Identification of multiple interacting alleles conferring low glycerol and high ethanol yield in Saccharomyces cerevisiae ethanolic fermentation

Selection of a rare segregant displaying the trait of low glycerol/high ethanol yield and lacking the major causative allele ssk1^{E330N…K356N}

Previous work has identified the S. cerevisiae strain CBS6412 as a strain with an unusually low ratio of glycerol/ethanol yield and genetic analysis identified the ssk1 ^{E330N…K356N} allele as a major causative gene[19] (Figure 1a). In order to identify the minor QTLs and their causative genes responsible for determining this complex trait, we have first screened all superior segregants with a glycerol/ethanol ratio as low as the superior parent strain, for a segregant that lacked the ssk1 ^{E330N…K356N} allele. Among the 44 superior segregants available, only a single such segregant, 26B, was present. Its glycerol yield was equally low and its ethanol yield equally high as the superior parent strain CBS4C, both in minimal medium with 5% glucose and in rich YP medium with 10% glucose (Figure 1b). Hence, 26B showed the same phenotypic difference with the inferior parent strain ER7A as CBS4C (Figure 1b).

Backcross of the unique superior segregant 26B with the inferior parent ER7A and screening for superior segregants

We next switched the mating type of 26B from Matα to Mata (see Materials and methods) and crossed the Mata 26B strain with the Matα inferior parent strain, ER7A, which is a derivative of the industrial strain Ethanol Red, currently used worldwide in bioethanol production. The hybrid diploid ER7A/26B showed a glycerol/ethanol yield phenotype, which was intermediate between that of ER7A and 26B (Figure 1b). The hybrid was sporulated and 260 meiotic segregants were screened for low glycerol yield (and corresponding higher ethanol production) in 100 ml fermentations with YP 10% glucose. The parent strains 26B and ER7A, and the hybrid diploid, were used as controls in each batch of fermentations.

Glycerol and ethanol yield of the segregants in each batch were normalized to those of 26B, which were set to 100%. ER7A and the diploid 26B/ER7A showed an average glycerol yield of 146% and 124% and a decreased ethanol yield of 98.1% and 99.4% (Figure 2a). The glycerol and ethanol yield of the segregants showed a Gaussian distribution, which extended over the range of the two parental strains. In the case of the lowest glycerol yield, this extension was only marginal. The population means of the glycerol yield (123%) and ethanol yield (98.8%) were close to those of the diploid 26B/ER7A. In general, glycerol and ethanol yield of the segregant population correlated inversely (as determined with a Pearson test), meaning that low glycerol yield was usually accompanied by high ethanol yield. Nearly all exceptions to this correlation were segregants with an unusually low ethanol yield that failed to show a correspondingly higher glycerol yield. To compose the pool of selected superior segregants, two cut-off criteria were defined, a glycerol yield lower than 120% of 26B and an ethanol yield higher than 99% of 26B. These cut-off criteria resulted in the selection of a set of 34 superior segregants. These were all retested in 100 ml fermentations with YP 10% glucose and 22 segregants showed again a low glycerol yield combined with a correspondingly higher ethanol yield using the same cut-off criteria (Figure 2b). These 22 segregants were selected for QTL mapping with pooled-segregant whole-genome sequence analysis. A second pool with 22 randomly selected segregants was also subjected to pooled-segregant whole-genome sequence analysis and referred to as the unselected control pool (Figure 2b).

Pooled-segregant whole-genome sequence analysis and QTL mapping

The genomic DNA of the selected and unselected pools, as well as the parent strain 26B, was extracted and submitted to custom sequence analysis using Illumina HiSeq 2000 technology (BGI, Hong Kong, China). The genome sequence of the parent strain ER7A has been determined in our previous study (data accession number SRA054394)[19]. Read mapping and single nucleotide polymorphism (SNP) filtering were carried out as described previously[20, 29]. The SNP variant frequency was plotted against the SNP chromosomal position (Figure 3). Of the total number of 21,818 SNPs between CBS4C and ER7A, 5,596 SNPs of CBS4C were found back in 26B. These SNPs were used for mapping minor QTLs in the genomic areas that were not identical between 26B and ER7A. The other genomic areas were completely devoid of SNPs because they were identical between the 26B and ER7A parents (white gaps in Figure 3). The scattered raw SNP variant frequencies were smoothened and a confidence interval was calculated, as previously described[20, 29]. The Hidden Markow Model, EXPloRA (see Materials and methods) was used to evaluate whether candidate regions showed significant linkage to the low glycerol phenotype. EXPloRA indicated six significant QTLs: on chr. I (3859–11045), chr. II (584232–619637), chr. IV (316389–375978 and 696486–748140), and chr. XIII (600902–610995 and 634582–640415) for the selected segregants pool.

The locus on chr. I was present in both the selected and unselected pool and was thus likely linked to an inadvertently selected trait, such as sporulation capacity or spore viability. It was excluded from further analysis. EXPloRA also reported two significantly linked loci on chr. VI (169586–170209) and chr. VII (472620–493523) for the unselected pool. Both loci were linked to the inferior parent, ER7A. For the region on chr. VII, the linked locus with the inferior parent genome was also present in the selected pool. Both loci likely represent linkage to inadvertently selected traits, such as sporulation capacity or spore viability. It is unclear why the locus on chr. VI was only present in the unselected pool. Since both loci were not linked to the low glycerol phenotype they were not investigated further.

The locus on chr. II was interesting since it also appeared in the previous mapping with the two original parents, CBS4C and ER7A, but in that case it was not pronounced enough to be significant[19]. The mapping with the backcross has now confirmed the relevance of this locus. On chr. IV and XIII, two new QTLs with a significant linkage to the low glycerol/high ethanol yield phenotype were detected. These QTLs were not present in our previous mapping with the original parent strains CBS4C and ER7A.

All QTLs with a significant link to the phenotype under study, i.e. those on chr. II, IV and XIII, were further investigated in detail. Selected SNPs within the respective QTLs were scored in the 22 individual superior segregants to determine precisely the SNP variant frequency and the statistical significance of the linkage. Using the binomial test previously described[20, 29] none of the three loci was found to be significantly linked to the genome of the superior parent strain 26B with the low number of superior segregants available. Therefore, we isolated 400 additional F1 segregants of the diploid 26B/ER7A and screened them for low glycerol/high ethanol production. In addition, we performed four rounds of random inbreeding (mating and sporulation) with all F1 segregants from the diploid 26B/ER7A to increase the recombination frequency[23] and subsequently also evaluated 400 F5 segregants in small-scale fermentations for glycerol/ethanol yield. The results for the 400 F1 and 400 F5 segregants are shown in Figure 4a. The glycerol and ethanol yields are again expressed as percentage of that of the superior parent strain 26B. There was again a clear inverse correlation between glycerol and ethanol yield. From the 800 segregants, we selected in total 48 superior segregants, i.e. 22 F1 segregants and 26 F5 segregants (Figure 4b).

We next scored selected SNPs in the putative QTLs on chr. II, IV and XIII in the 22 additionally selected F1 segregants and the 26 selected F5 segregants. Next, we determined the SNP variant frequency and the corresponding P-value, as described previously[20, 29], for the following groups of segregants: the 22 initially selected segregants of the sequenced pool, the 22 additionally selected F1 segregants, the total of 44 selected F1 segregants, the 26 selected F5 segregants and the total of 70 selected F1 and F5 segregants. They are shown in Figure 4c. By increasing the number of superior segregants, we were now able to demonstrate significant linkage (P-value < 0.05) to the genome of the superior parent strain 26B for the three QTLs under study. For the QTLs on chr. II and IV the linkage was very strong, while for the QTL on chr. XIII it was still weak, but significant. In contrast, the second region on chr. IV did not show any significant linkage with none of the pools.

Identification of causative genes in the QTLs on chr. II, IV and XIII

For further analysis, we first selected three potential candidate genes within the three QTLs, based on their known function in glycerol metabolism. SMP1, which is located in the QTL on chr. II (594,864 to 593,506 bp), encodes a putative transcription factor involved in regulating glycerol production during the response to osmostress[30]. The gene is located in the chromosomal region from 584,232 to 619,637 bp, which was predicted as most significant by the EXPloRA model. The 26B SMP1 allele has two point mutations within its coding sequence, which are changing the primary protein sequence at position 110 from arginine to glutamine and at position 269 from proline to glutamine. Hence, we have named this allele smp1 ^R110Q,P269Q .

The SNP with the highest linkage within the QTL found on chr. IV, was located at position 411,831 bp (Figure 4c), which is within the open reading frame of GPD1 (411,825 to 413,000 bp). This is the structural gene for the NAD⁺-dependent cytosolic GPDH[15, 16]. This enzyme catalyzes the conversion of DHAP to glycerol 3-phosphate through the oxidation of NADH and has been shown to be the rate-controlling step in glycerol formation[1, 16]. The GPD1 allele of 26B harbors a point mutation, changing leucine at position 164 into proline. This mutation was found before (DDBJ database data, accession number AY598965). The GPD1 allele of 26B was named gpd1 ^L164P .

The SNP with the highest linkage within the QTL found on chr. XIII was located at position 606,166 bp (Figure 4c), which is within the open reading frame of HOT1 (605,981 to 608,140 bp). HOT1 encodes a transcription factor required for the response to osmotic stress of glycerol biosynthetic genes, including GPD1, and other HOG-pathway regulated genes[31, 32]. The 26B HOT1 allele contains two non-synonymous point mutations, changing proline to serine at position 107 and histidine to tyrosine at position 274. We have named the HOT1 allele of 26B, hot1 ^P107S,H274Y .

We first investigated the effect of smp1 ^R110Q,P269Q , gpd1 ^L164P and hot1 ^P107S,H274Y on the low glycerol/high ethanol phenotype using reciprocal hemizygosity analysis (RHA)[25]. For that purpose, we constructed for each gene a pair of hemizygous diploid 26B/ER7A hybrid strains, in which each pair contained a single copy of the superior or the inferior allele of SMP1, GPD1 or HOT1, respectively, while the other copy of the gene was deleted. The three pairs of hemizygous diploids were tested in the same 100 ml YP 10% glucose fermentations as previously used for the screening. The parent strains 26B and ER7A and the hybrid diploid 26B/ER7A were added as controls. The glycerol and ethanol yields were again expressed as percentage of those of 26B, which were set at 100%. The significance of any differences between the strains was evaluated using a two-tailed unpaired t-test with a P-value < 0.05 considered to indicate a significant difference. The results of the RHA are shown in Figure 5. They indicate that both smp1 ^R110Q,P269Q and hot1 ^P107S,H274Y , but not gpd1 ^L164P , derived from the superior parent 26B cause a significant drop in the glycerol/ethanol ratio compared to the alleles of the inferior parent strain ER7A. For smp1 ^R110Q,P269Q only the reduction in glycerol, and not the increase in ethanol, was significant with the P-value < 0.05 used. These results indicate that smp1 ^R110Q,P269Q is probably a causative gene in the QTL on chr. II. They do not exclude that the QTL may contain a second causative gene, especially since smp1 ^R110Q,P269Q is not located in the region with the strongest linkage (lowest P-value).

The RHA with the GPD1 alleles failed to show any difference both for glycerol and ethanol production (Figure 5). Hence, the superior character of the gpd1 ^L164P allele could not be confirmed with RHA. This is remarkable because the SNP with the strongest linkage (lowest P-value) in the QTL on chr. IV was located in the open reading frame of GPD1 and showed very strong linkage to the low glycerol/high ethanol phenotype. The hot1 ^P107S,H274Y allele of the superior strain 26B, in contrast, caused a reduction in glycerol and an increase in ethanol production, and both changes were significant (P-value < 0.05) (Figure 5). Hence, these results indicate that hot1 ^P107S,H274Y is a causative allele in the QTL on chr. XIII and because it contains the SNP with the strongest linkage (lowest P-value), it is likely the main causative allele in this QTL.

The glycerol yield for the inferior parent ER7A and the diploid 26B/ER7A were on average 143% and 126% of the 26B yield (Figure 5). Ethanol yield of both strains was correspondingly reduced to 98% of the 26B yield. Clearly, the smp1 ^R110Q,P269Q and hot1 ^P107S,H274Y alleles can only be responsible for part of the difference in the glycerol/ethanol ratio between the parent strains. The same was found previously for the ssk1 ^{E330N…K356N} allele[19]. This confirms that the glycerol/ethanol ratio in yeast fermentation is a true polygenic, complex trait, determined by an interplay of multiple mutant genes.

Expression of the gpd1 ^L164P allele from 26B in haploid gpd1∆ strains reveals its superior character

Several explanations could account for the failure to confirm the superior character of the gpd1 ^L164P allele from 26B in the RHA test. A closely located gene may be the real causative gene, the gpd1 ^L164P allele may be effective only in a haploid genetic background or the effect of the gpd1 ^L164P allele may be suppressed through epistasis by one or both of the other two superior alleles, smp1 ^R110Q,P269Q and hot1 ^P107S,H274Y . To distinguish between these possibilities, we amplified the gpd1 ^L164P allele from strain CBS4C and the GPD1 allele from strain ER7A by PCR (410,523 to 413,479 bp, including promotor, ORF and terminator). The PCR fragment was ligated in the centromeric plasmid YCplac33, resulting in plasmids YCplac33/gpd1 ^L164P -CBS4C and YCplac33/GPD1-ER7A. Both plasmids were transformed into gpd1∆ strains of the two parents 26B and ER7A, the hybrid diploid 26B/ER7A and the lab strain BY4742[33, 34]. All strains were tested in 100ml fermentations with YP 10% glucose. Glycerol and ethanol yields were determined after 120 h of fermentation. The results are shown in Figure 6.

When the gpd1 ^L164P -CBS4C allele or the GPD1-ER7A allele were expressed in the gpd1∆ strains of the superior parent 26B or the hybrid diploid 26B/ER7A, the increase in glycerol production and the decrease in ethanol production was the same for the two alleles. On the other hand, expression of the gpd1 ^L164P -CBS4C allele in the gpd1∆ strains of the inferior parent ER7A or the lab strain BY4742, enhanced glycerol production and reduced ethanol production significantly more than expression of the GPD1-ER7A allele. The latter shows that the gpd1 ^L164P -CBS4C allele is superior compared to the GPD1-ER7A allele. The difference between the two alleles is apparently not dependent on the haploid or diploid background of the strain but seems to be related with the presence of the two other superior alleles, smp1 ^R110Q,P269Q and hot1 ^P107S,H274Y . They are both present in the two strains, 26B and 26B/ER7A, in which gpd1 ^L164P -CBS4C has no differential effect and absent in the two strains, ER7A and BY4742, in which gpd1 ^L164P -CBS4C has a differential effect. Hence, the superior potency of gpd1 ^L164P -CBS4C may be suppressed through epistasis by smp1 ^R110Q,P269Q and/or hot1 ^P107S,H274Y . On the other hand, we cannot exclude that the effect of gpd1 ^L164P -CBS4C is suppressed by one or more other mutant genes present in the superior parent 26B or the hybrid diploid 26B/ER7A.

We have scored the final 70 superior segregants with a glycerol yield < 120% and an ethanol yield > 99% of that of the superior parent 26B, for the presence of the three causative alleles, smp1 ^R110Q,P269Q , gpd1 ^L164P and hot1 ^P107S,H274Y . The results are shown in Figure 7a. The largest group of superior segregants contained all three mutant alleles, followed by smaller groups with only two of the three mutant alleles and finally the three smallest groups with only one mutant allele. Hence, there was a clear correlation between the number of mutant alleles and low glycerol/high ethanol yield in this group of selected segregants. On the other hand, although there was a tendency for a lower mean glycerol/ethanol yield ratio with an increasing number of mutant alleles, the differences between the means of the different groups were small and the variation remained large and with the same range for the three largest categories.

We have also investigated a possible correlation between the different mutant alleles and the strength of the low glycerol/high ethanol phenotype. For that purpose, we determined the percentage of segregants with a specific mutant allele in sets of strains with a decreasing glycerol yield or an increasing ethanol yield. The results show that there is no preference between the different alleles in the strains with a higher glycerol yield, but in the strains with the lowest glycerol yield, the gpd1 ^L164P allele is preferentially present, followed by the hot1 ^P107S,H274Y allele, although this only holds for the category with the lowest glycerol yield (Figure 7b). Hence, the order of potency of the three alleles appears to be: gpd1 ^L164P  > hot1 ^P107S,H274Y  ≥ smp1 ^R110Q,P269Q . There was no correlation between the variant frequency of the three alleles for high ethanol yield, indicating that other minor QTLs may affect ethanol yield independently from glycerol yield and act together with the currently identified alleles.

Identification of superior alleles as gene tools for reduction of the glycerol/ethanol yield ratio

The goal of the present work was to investigate whether natural yeast strains may harbor specific alleles, e.g. in structural and/or regulatory genes of glycerol metabolism, that would allow reducing the glycerol yield and increasing the ethanol yield in yeast fermentation without causing negative side-effects on other essential functions. We successfully identified three mutant alleles, which separately and together reduce the glycerol yield in a subtle way without affecting, at least not in a conspicuous way, the overall rate and characteristics of the fermentation process. Combined with the previous discovery of the ssk1 ^{E330N…K356N} allele in strain CBS4C[19], this indicates that the original diploid parent strain CBS6412 contains at least four specific alleles causing reduced glycerol production and concomitantly higher ethanol production. This suggests that a low glycerol/ethanol yield ratio may have been advantageous for the survival of the S. cerevisiae strain CBS6412 in nature. Hence, we have now identified four mutant alleles acting at different levels, either transcriptional, regulatory or structural, in the biosynthesis of glycerol and its regulation, and that have been pre-filtered by natural selection and evolution for compatibility with survival in the natural environment. This spreads the reduction of glycerol yield not only over different target genes, each causing a subtle reduction, but also over different nodes in the cellular network and thus likely minimizes further the risk of negative side-effects.

For the phenotypic screening, we primarily selected on low glycerol yield. Although there was in general an inverse relationship between glycerol and ethanol yield, this was not true in all strains. Moreover, the variant frequency of the three mutant alleles for low glycerol yield and high ethanol yield (Figure 7b) did not match. In the segregants with the lowest glycerol yield, as opposed to those with higher glycerol yield, the gpd1 ^L164P allele was preferentially present, but this difference was not observed between segregants with higher and lower ethanol yield. This suggests that there are other factors that determine high ethanol yield independent of glycerol yield or that it is the combination of the mutant alleles that is required to enhance ethanol yield rather than the presence of specific alleles. The previously identified major ssk1 ^{E330N…K356N} allele, which was present in nearly all F1 segregants with low glycerol/ethanol ratio[19], might also have played a more important role in establishing high ethanol yield than the three currently identified minor causative alleles.

Our results therefore confirm that natural yeast strains harbor mutant alleles of the well known structural and regulatory genes identified in laboratory yeast strains that have been filtered by natural selection and evolution for compatibility with survival in the natural environment. The chance that these alleles exert significant negative effects on other essential functions of the yeast cells is probably not completely absent but at least minimized compared to drastic genetic modifications like gene deletion or overexpression. Screening of biodiversity for such specific alleles therefore appears to be a fruitful strategy to identify mutant alleles that can be used as specific gene tools for strain improvement by targeted genetic modification. A related example is the development of S. cerevisiae wine strains with a higher glycerol/lower ethanol production ratio[12–14]. Higher glycerol production is a preferred characteristic because it improves the mouthfeel of wine and lower ethanol production is a preferred characteristic as well because of the restrictions imposed on ethanol consumption for driving. Our results with the many segregants tested for glycerol/ethanol yield in fermentation has revealed many strains with a much higher glycerol and much lower ethanol yield than the parent strains (Figures 2,4). This suggests that also for this beneficial trait specific alleles could be identified that would allow to improve this trait in a more subtle way. Possibly, this would allow to enhance glycerol production and reduce ethanol production without the dramatic increase in undesirable acetate production that was the result of classical genetic engineering of structural genes for GPDH[1].

Identification of minor QTLs and causative genes

While identification of major QTLs has become straightforward with pooled-segregant whole-genome sequence analysis[19, 20, 22, 23], identification of minor QTLs remains a major challenge. This is especially true for phenotypes that require a high workload for scoring and for which as a result only low numbers of selected segregants can be obtained to assemble the pool for whole-genome sequence analysis. In the present paper we have successfully established a novel approach for minor QTL identification. After mapping major QTLs and identification of the causative genes, the F1 segregants displaying the phenotype-of-interest are screened for absence of one or more superior alleles. As a result, these segregants should have all or most minor QTLs able to confer the phenotype-of-interest. This is not only because they are needed to confer the phenotype in the absence of the superior allele, but also because their effect is often suppressed by a major superior allele through epistasis. Hence, the use of a rare segregant displaying the trait-of-interest, might be advantageous in case of gene interference. To display the trait-of-interest, the segregant must have inherited a set of compatible mutant alleles, with any interfering mutations being absent. When such a rare F1 segregant is backcrossed with the inferior parent, the segregants from this cross displaying the phenotype-of-interest, should again contain all or most of these minor QTLs facilitating their mapping and identification of their causative allele. As shown in Figure 7a, several of these segregants contained only one of the three causative alleles and still displayed a low glycerol yield under the cut-off of 120%. This suggests that there are additional alleles present in these strains able to confer low glycerol yield and that a new backcross of such a segregant with the inferior parent may allow identification of additional alleles conferring low glycerol yield. In principle this approach could thus be repeated with each new generation of segregants. Previously, similar approaches have been used in which either F1 segregants were backcrossed to eliminate a major QTL[27] or major QTLs were fixed in one of the parents and the crossing repeated[26]. In these cases the parents displayed a reduced phenotypic difference, which may make the phenotyping and the stringency of selection in the next cross more cumbersome.

Random coincidence versus linkage in small segregant populations

In small populations of segregants, random coincidence can easily cause falsely predicted QTLs, which are difficult to distinguish from QTLs with significant, but weak linkage[35]. In this case, the unselected pool is of little use because it usually does not contain at the same position the same false QTL caused by random coincidence. Higher stringency in QTL selection can eliminate false QTLs but also weakly linked true QTLs. An essential difference between a false QTL caused by random coincidence and a true QTL with weak linkage, is that the latter should be reproducible. Therefore, we screened three different pools of segregants resulting in three independent pools of small-size with segregants displaying low glycerol yield. This allowed us to distinguish the false QTLs on chr. IV (696486–748140) and chr. XIII (634582–640415) from the true, weakly linked QTLs on the same chromosomes, chr. IV (316389–375978) and chr. XIII (600902–610995). It is well known that a higher number of segregants increases the reliability of minor QTL detection. In previous work, many minor QTLs could be identified by using millions of segregants and a selectable phenotype[23]. However, most complex traits are not selectable and phenotypic screening of millions of segregants is not feasible for many traits. In these cases, reliable minor QTL identification remains a major challenge. In addition, the number of segregants that can be pooled for pooled-segregant whole genome sequence analysis is in principle unlimited, but in practice the useful number of segregants is limited by the average coverage in whole-genome sequencing. When the number of segregants exceeds the average coverage of sequencing, the surplus no longer enhances the reliability of mapping and is thus useless.

Novel mutant alleles and possible epistatic interactions

Smp1 is a transcription factor, belonging to the MEF2 family, that regulates the expression of stress-responsive genes, such as GPD1. Its DNA binding domain is located at the amino acid residues 1–90[36]. Upon osmotic stress, Smp1 is phosphorylated by Hog1, which physically interacts with its C-terminal domain. Four different phosphorylation sites were identified, i.e. Ser348, Ser357, Thr365, and Ser376, all located within a region coincident with the Hog1 binding domain. Phosphorylation of Smp1 is essential for its functioning, since an allele unable to be phosphorylated caused an impaired stress response[30]. The point mutations in the 26B allele, smp1 ^R110Q,P269Q , are not located in the DNA or the Hog1 binding domain. However, the change of a proline to a glycine, close to the phosphorylation sites, might change Smp1 structure, thereby influencing its ability to be bound and/or phosphorylated by Hog1. The smp1 ^R110Q,P269Q allele is dominant since its expression in the hybrid 26B/ER7A diploid decreased glycerol yield and increased ethanol yield.

GPD1 encodes NAD⁺-dependent cytosolic glycerol 3-phosphate dehydrogenase. It catalyzes the conversion of DHAP to glycerol 3-phosphate through the oxidation of NADH. The expression of GPD1 is induced by the HOG pathway and it is essential for growth under high osmolarity[16] Possible domains for binding of NADH, H⁺ and DHAP have been predicted based on similarity with proteins with a comparable function[37]. The single point mutation present in the 26B allele, gpd1 ^L164P , may be located in the putative NADH-binding domain, but the location of this domain is not well predicted. This mutation was found earlier and called a ‘natural variant’ (DDBJ database data, accession number AY598965). No linkage of gpd1 ^L164P with low glycerol yield was observed in RHA but its effect was revealed by expression in a GPD1-deficient mutant. The L164P mutation could reduce the intrinsic activity of the Gpd1 enzyme or its expression by lowering GPD1 mRNA stability. Both possibilities would result in reduced glycerol production and thus explain the low glycerol yield of CBS6412. Both explanations are consistent with the lower GPDH activity that we measured in CBS4C (0.128 ± 0.027 U/mg protein) compared to ER7A (0.225 ± 0.053 U/mg protein).

The gpd1 ^L164P allele was apparently subject to epistatic suppression in the superior strains. In the BY4742 and ER7A backgrounds, which lack the smp1 ^R110Q,P269Q and hot1 ^P107S,H274Y alleles, the gpd1 ^L164P allele had a clear reducing effect on glycerol yield compared to the wild type GPD1 allele (Figure 6). On the other hand, its expression in the strains 26B and 26B/ER7A, which contain the smp1 ^R110Q,P269Q and hot1 ^P107S,H274Y alleles, resulted in a similar glycerol yield as the wild type GPD1 allele. This suggests that the smp1 ^R110Q,P269Q and hot1 ^P107S,H274Y alleles and/or other alleles present in the superior strains suppress the effect of the gpd1 ^L164P allele. The epistatic effect may be explained at the biochemical level by the fact that the reduction in expression of GPD1, caused by the smp1 ^R110Q,P269Q and hot1 ^P107S,H274Y alleles, is so strong that the mutation in GPD1 itself has no significant effect anymore. Hot1 activates transcription of GPD1 and other HOG-dependent genes under osmostress[31, 32, 38]. Alepuz et al.[31] proposed that Hot1 serves as an anchor for Hog1, which directly recruits the RNA polymerase II complex. The position of the Hog1 binding domain in Hot1 is unknown. It is unclear how the two mutations in the 26B allele, hot1 ^P107S,H274Y , could affect the functioning of the protein.

Interestingly, although all mutant genes revealed in this work had already been identified previously using classical molecular genetics approaches, our work clearly indicates the potential of complex trait analysis for identifying new alleles of known components, and possibly also completely new components, in signaling pathways and other cellular functions. In our case, it seems plausible that continuation of backcrossing with new segregants from subsequent generations, displaying low glycerol/high ethanol yield and lacking (most of) the previously identified alleles, might reveal new players in the HOG signaling pathway or in transcriptional control of GPD1.

Microbial strains, cultivation conditions and plasmids

Mating, mating type switch, sporulation and internal crosses

Mating and sporulation were carried out according to standard procedures[40]. Mating type of segregants was determined by diagnostic PCR for the MAT locus[41]. Mating type switching was performed by induction of the HO-gene expressed from the plasmid pFL39 GAL1 HO KanMX. Meiotic spores of the hybrid 26B/ER7A were isolated by random spore analysis[42]. The hybrid 26B/ER7A was plated on a sporulation plate for a period of about two weeks up to 1 month until asci were observed. The yeast cells were washed off the sporulation plate and suspended in 25 ml of MilliQ water, in a 300 ml Erlenmeyer flask together with sterile 0.45 mm glass beads. 500 μl Zymolyase (10 mg/ml) and 10 μl of β-mercaptoethanol were added to the cell suspension in order to degrade the asci. This cell suspension was incubated overnight at 30°C by shaking at 200 rpm. The cell suspension was transferred to a 50 ml tube together with the glass beads, and shaken vigorously. The cell debris was centrifuged at 20000 rpm for 20 min. The supernatant was discarded and the pellet was suspended in 5 ml of Nonidet P-40 and placed on ice for 15 min. This was followed by 4 rounds of sonication (30 s, 75%). The cell suspension rested 2 min on ice between two rounds. After that, the cell suspension was centrifuged for 10 min at 3000 rpm, the supernatant was discarded, and the cells were re-suspended in 1.5% Nonidet P-40. This procedure was repeated once, followed by 4 more rounds of sonication and incubation on ice. Lastly, the cell suspension was centrifuged for 10 min at 3000 rpm. The supernatant was discarded and cells were re-suspended in 300 ml of MilliQ water. The cell suspension was diluted to obtain single colonies on plates. Plates were incubated at 30°C until single colonies were visible. These single colonies were re-plated and checked for mating type to confirm haploidy. Usually, this procedure yielded 90% haploids. Crossing of 26B and ER7A was carried out as follows. First, diploids were isolated from the cross of ER7A and 26B. For this purpose, the mating type of single colonies resulting from the cross between ER7A and 26B was checked. In the first step of internal crossing, the diploids were incubated on a sporulation plate until sufficient asci were visible. In the second step, spores were isolated using random spore analysis (see further). The isolated spores were all plated on YD and incubated for 2 days at 30°C to ensure that enough diploids had been formed. In the last step, newly formed diploids were transferred to a new sporulation plate to start the next cycle of internal crossing.

Fermentation conditions

The 26B, ER7A and CBS4C strains were tested in two fermentations: minimal medium and YP 10% glucose. The minimal medium was composed of 1.9 g l^-1 yeast nitrogen base (Difco), 5 g l^-1 ammonium sulphate, 250 mg l^-1 leucine, 50 mg l^-1 uracil, 100 mg l^-1 histidine, 30 mg l^-1 lysine, 20 mg l^-1 methionine and 50 g l^-1 glucose. The inoculum culture was grown overnight in minimal medium containing 2% [w/v] glucose and was used for inoculation of the fermentation medium at an initial OD of 1. Fermentations were carried out in Erlenmeyer flasks, which were equipped with air locks, ensuring the exclusion of oxygen but allowing the release of CO₂. The fermentations were performed at 30°C and cultures were continuously stirred at 200 rpm.

The parental strains, ER7A and 26B, as well as CBS4C, were additionally tested in YPD, which contained 0.2% [w/v] yeast extract, 0.6% [w/v] peptone, and 10% [w/v] glucose, to mimick the free amino nitrogen content present in wheat liquefact. The fermentations were carried out in cylindrical glass tubes, which were closed with a rubber stopper containing a glass pipe, sealed off with a cotton plug to release CO₂. 100 ml fermentation medium was added to each tube. Inoculum cultures were grown statically overnight at 30°C in 5 ml of YD medium one day ahead of the fermentation, and this culture was used completely to inoculate the 100 ml fermentation tube. The empty weight, starting weight and weight after 72 h fermentation of the tubes was measured to determine the net weight loss of the medium during the fermentation. Fermentations ran for 72 h at 30°C and were stirred at 200 rpm. After 72 h, the final weight was determined. The fermentation broth was cooled for 1 night at 4°C prior to the analysis of glycerol and ethanol concentrations, in order to minimize evaporation of ethanol during the sample taking. The fermentations of the initial screen of 26B/ER7A segregants, those of the reciprocal hemizygosity analysis, and those of the gpd1 Δ strains were carried out in 100 ml YP 10% glucose.

Screening of additional selected segregants was downscaled to 5 ml fermentation cultures. The pre-culture was started one day ahead of the fermentation in 3 ml of YD medium. Cultures were grown statically overnight at 30°C. The next day, the pre-cultures were used to inoculate the 5 ml fermentation in a proportion of 1/20. Fermentations were kept for 96 h at 30°C and afterwards placed at 4°C overnight prior to the analysis of the fermentation broth in order to prevent ethanol evaporation.

Determination of fermentation parameters

In all fermentations weight loss was used to follow the progress of the fermentation. Glucose, glycerol and ethanol in the medium were determined by HPLC (Waters® isocratic Breeze^TM HPLC, ion exchange column WAT010290). Column temperature was 75°C, 5 mM H₂SO₄ was used as eluent with a flow rate of 1 ml min^-1 and refractive index detection was used (Waters, 2414 RI detector). The product yield was calculated from the final product concentration (g. l^-1) and the difference in glucose concentration at the start and end of the fermentation (consumed glucose in g. l^-1). Yields of strains used for screening, RHA and the gpd 1∆ complementation analysis were related to the yield of 26B in the same experiment in order to decrease variance between different experiments.

Yields were calculated with the following formulas:

Absolute yield:$Y_{S/i=}\frac{c_i^{\mathit{final}}}{\left(c_s^{\mathit{ini}}-c_S^{\mathit{final}}\right)}$

Relative yield:$Y_{\%}=\frac{Y_{S/i}^{\mathit{segr}}}{Y_{S/i}^{26B}}$

DNA methods

Yeast genomic DNA was extracted with Phenol/Chloroform/Isoamyl-alcohol (25:24:1)[43] and further purified with diethyl-ether extraction or ethanol precipitation if required. PCR was performed with high-fidelity polymerases Phusion^TM (Finnzymes) or ExTaq^TM (TaKaRa) for cloning and amplification of deletion or insertion cassettes, and sequencing purposes. Sequencing was carried out using the dideoxy chain-termination method[44] at the VIB Genetic Service Facility (Antwerp). The sequences were analyzed with geneious (Geneious Basic 5.3.4), SeqMan (Lasergene Coresuite 8) or CLC DNA workbench (CLC bio) software.

Pooled-segregant whole-genome sequence analysis

The segregant 26 was isolated from the cross of CBS4C and ER7A. This segregant was backcrossed with its own parent ER7A. From this backcross, the 22 most superior segregants (lowest glycerol production) were assembled in the ‘selected pool’ while 22 random segregants were used to assemble the ‘unselected pool’. The two pools were made by combining equal amounts of cells based on OD₆₀₀. High molecular weight DNA (3 μg, ~ 20kb fragments) was isolated from the pools and parent strains according to Johnston and Aust[45]. The purity of the DNA sample was estimated from UV measurement (260/280 = 1.7-2.0). The DNA samples were provided to BGI (Hong Kong, China) for whole-genome sequence analysis by Illumina technology.

Mapping of short read sequences, variant calling and QTL analysis were carried out as described earlier by Swinnen et al.[20] and by Hubmann et al.[19]. The SNP variant frequencies were calculated by dividing the number of the alternative variant by the total number of aligned reads. A very high or a very low frequency was a sign of a one-sided SNP segregation preferentially coming from one parent, indicating a genetic linkage to the trait of interest. Genetic linkage was statistically confirmed using EXPloRA (Duitama et al. in preparation) or the methods described earlier[20].

Detection of SNP markers

Individual SNPs were scored by PCR. The forward and reverse primer contained the nucleotide of ER7A or CBS4C as the 3’ terminal nucleotide, respectively. The annealing temperature was optimized using DNA of ER7A and CBS4C so as to allow only hybridization with primers containing an exact match.

Reciprocal hemizygosity analysis (RHA)

For RHA analysis[25], two diploid strains were constructed by crossing 26B and ER7A wild type or deletion strains for the candidate gene, so that the resulting diploids only contained a single allele from either 26B or ER7A for the candidate gene being evaluated. Deletion cassettes for SMP1, HOT1 and GPD1 were constructed as described by Gueldner et al.[46] using the phleomycin resistance marker ble ^R . After the transformation, the gene deletions were verified by PCR. RHA was performed with three independent isolates of all tested diploids.

Construction of YCplac GPD1 plasmids

The primers, A-3709 and A-3743, were used to PCR amplify genomic DNA of CBS4C and ER7A chr. IV (410523 – 413479), containing the promoter, GPD1 ORF, and terminator. The resulting PCR fragment was digested with KpnI, purified, and ligated to the plasmid YCplac33, which was digested with KpnI prior to ligation. The constructs YCplac33 GPD1-ER7A and YCplac33 gpd1 ^L164P -CBS4C were verified using Sanger sequencing. URA3 was deleted in the strain 26B gpd1∆ and ER7A gpd1∆. Both strains were mated to obtain the diploid strain ER7A/26B gpd1∆/∆ ura3∆/∆. The deletion cassette was constructed as described by Gueldner et al.[46] with the geneticin resistance marker KanMX6. Transformants were selected simultaneously for the two selectable markers on phleomycin and geneticin, to avoid a cassette switch. URA3 gene deletion was confirmed by PCR and absence of growth on SD-ura plates. YCplac33 GPD1-ER7A, YCplac33 gpd1 ^L164P -CBS4C and the empty plasmid were transferred to the strain BY4742 gpd1∆, 26B gpd1∆ ura3∆, ER7A gpd1∆ ura3∆, and ER7A/26B gpd1∆/∆ ura3∆/∆, using the LiAc/PEG transformation method[47].

Data deposition

Sequencing data have been deposited at the SRA database (NCBI),http://www.ncbi.nlm.nih.gov/sra, with the account number SRA059109.