Schrödinger’s yeast: the challenge of using transformation to compare fitness among Saccharomyces cerevisiae that differ in ploidy or zygosity

Motivation for study: ploidy specific effects of mutations

Among the many distinguishing characteristics of life is the number of copies of an organism’s genome (“ploidy”). While many organisms spend the majority of their life with one copy of their genome (most bacteria but also multicellular haplont organisms such as dictyostelid slime moulds and green alga of the genus Chara), others have two (most animals), or many (polyploid plants). Yet others exist as multicellular forms in both ploidy stages, called haploid-diploid or haplodiplontic (many green, brown, and red macroalgae). Even in multicellular diplont organisms such as humans, the difference in selection in the haploid and diploid cell are of great importance. For example, mutations that render fitness advantages in gametes may have unpredicted consequences for a developing zygote (Immler, 2019).

Studies measuring differences in the reproductive fitness of haploid and diploid stages of the life cycle have discovered a strong interactive effect of the environment (Gerstein & Otto, 2009; Mable, 2001; Mable & Otto, 1998; Thornber, 2006; Zörgö et al., 2013). Many of these studies are conducted in budding yeast, Saccharomyces cerevisiae, that can be maintained in both its haploid and diploid form. The researchers have reported broad differences in growth rate or morphology of the two ploidy levels. Nevertheless, few studies have been conducted to measure mutational effects on fitness in the two different states.

Gerstein (2012) compared the fitness and phenotypic effect of 40 different nystatin-tolerant mutations in S. cerevisiae and found that haploids acquired stronger resistance to the toxin and acquired greater increases in growth rate compared to their homozygous diploid counterparts. Szafraniec et al. (2003) measured differences in growth rate of mutant haploids created by mutagenesis and heterozygous diploids but did not report a comparison of the homozygous diploids to haploids. Here, we ask if fitness effects of spontaneous mutations from a mutation-accumulation (MA) experiment (Sharp et al., 2018) differ between ploidy levels.

Dominance

The genetic dominance of a mutation describes the difference in phenotype with or without the presence of a functional version. There are mutations that are imperceptible in the heterozygous state but fully expressed in the homozygous state and as such have a dominance of 0. The dominance coefficient, h, determines the likelihood that a beneficial allele spreads and fixes in a population (known as Haldane’s sieve, Haldane (1927)). Conversely, the frequency of deleterious mutations at mutation-selection balance will be inversely proportional to the dominance coefficient (Haldane, 1937). Whether through population structure (Whitlock, 2002) or inbreeding (Morton, 1971), the dominance coefficient will determine the mutation load in populations with non-random mating. Understanding the average as well as the variance in dominance of spontaneous mutations is of interest in both conservation biology and medicine.

There is also reason to believe that dominance effects may correlate with mutational effect and type of protein. The shape of this relationship determines the possible mutational load and inbreeding depression of a species. While outside the scope of this study, the underlying mechanism for differences in dominance is one of the oldest and most debated topics in evolutionary biology (Billiard & Castric, 2011).

In addition to determining fixation probabilities, mutation frequencies, and the impacts of inbreeding, the dominance coefficient can explain the prevalence of haploid-diploid life cycles. We should expect a diploid-dominant life cycle if deleterious mutations are sufficiently recessive in diploids (Perrot, Richerd & Valéro, 1991), while a haploid-dominant life cycle is expected when deleterious mutations are dominant (Jenkins & Kirpatrick, 1994, 1995). While it is commonly assumed that the mutational effect on fitness in a haploid is equal to that of a diploid homozygous for the mutation, recent theoretical work has demonstrated that intrinsic differences in how mutations affect the haploid and diploid stage of a life cycle can explain the stable co-existence of haploid-diploid organisms (Scott & Rescan, 2017). By assigning separate fitness effects of mutations in the homozygous diploid ( $s_d$) and haploid ( $s_h$) phase, the authors identified conditions under which predominantly haploid (“haplont”), predominantly diploid (“diplont”), or biphasic life cycles (“haploid-diploid”) might evolve.

There have been several attempts to measure the distribution of dominance coefficients (Table 1 in Agrawal & Whitlock, 2011; and Table 1 in Manna et al., 2012). Data from the Yeast Knockout Collection (YKO) (Shoemaker et al., 1996) were used to estimate the distribution of dominance coefficients (Phadnis & Fry, 2005; Agrawal & Whitlock, 2011; Manna et al., 2012), revealing a negative correlation between selection strength and dominance. One limit of knockout data is that they are, by definition, large effect mutations with a complete loss of function in the homozygous diploid and do not represent the multitudes of small effect mutations that are likely to arise spontaneously. As Manna et al. (2012) pointedly note, the YKO was not created to answer questions of dominance. In addition, these studies have not included organisms of different ploidy levels. There is a need for datasets of mutational effects measured with high replication and precision in haploids, heterozygous, and homozygous diploids to help parameterize models of life cycle evolution.

Our contribution

In this study, we use previously described MA lines (Sharp et al., 2018) to produce lines of different ploidy and copies of the MA genome: from haploid lines to heterozygous and homozygous diploid. Our goal was to measure how selective effects in haploids correlate with the selective effects in homozygous diploids, while at the same time measuring the average of and variance among dominance effects in heterozygous yeast.

Justification for method

A commonly used method to produce homozygous diploids from haploids is by mating type switching. Haploid yeast have two different mating types, MATa and MAT $\alpha$. All yeast have both mating type loci present in their genome, but one is inactive. Wild yeast are “homothallic” and can, via gene conversion using an endonuclease called HO, switch which mating type is active. Homothallism makes it possible for haploid cells in the absence of mating partners to switch the mating type they express and mate with a neighboring cell. In contrast, laboratory strains of yeast are typically heterothallic and cannot switch mating type, due to a deletion mutation of the HO locus. To produce homozygous yeast, researchers can insert the HO locus on a plasmid into the haploid yeast, which will readily switch mating type and mate. The problem with using this approach to compare mutational effects across ploidy levels is that the diploid that is formed will have a different mating type, namely MATa/MAT $\alpha$, and will therefor differ in more than genomic copy number (Birdsell & Wills, 1996).

An alternative approach, used by Gerstein (2012), is to temporarily transform MATa haploids with a plasmid containing the opposite mating type. With sufficient copies of the MAT $\alpha$ plasmid the transformed haploid yeast will act like a MAT $\alpha$ haploid and mate readily with untransformed MATa cells. Through propagation post mating, the plasmid with the added mating type is lost, and the researcher has acquired a diploid homozygote of the MATa/MATa genotype. This method is not without risk, however, as it can allow for further mating of the MATa/MATa diploids to haploids still possessing the MAT $\alpha$-containing plasmid, leading to lines with higher ploidy levels.

In this experiment, we introduce a new method. By deleting both the MAT locus and the STE4 locus we rendered haploid yeast sterile. We then temporarily transform the yeast with two different kinds of plasmids that contain the lost genes and restore mating ability as MATa or MAT $\alpha$. Once mating has occurred, the diploid is selected to lose the plasmid, leading to a mat $\mathrm{\Delta }$0/mat $\mathrm{\Delta }$0 mating type.

Unfortunately given our goals, in the present study transformations induced large changes in fitness that obscure the smaller mutational effects we wanted to measure. In particular, we found high rates of loss in respiratory function, also known as petiteness (respiratory deficiency known to decrease growth rate) in yeast. Even when considering sets of yeast lines with the same petite status, we found aberrant growth patterns across the haploid, heterozygous, and homozygous states. This indicates that our experimental lines changed in more ways (genetically, epigenetically, or otherwise) than those easily scored by petiteness and that these changes have greater impacts on growth rate than the mutations in the MA lines. We conclude that the haploid, heterozygous, and homozygous genotypes of the MA lines have acquired new variation, preventing us from extracting any information on the dominance of the small-effect spontaneous mutations acquired during MA. This conclusion highlights the challenge of estimating the effects of spontaneous mutations across ploidy levels: methods designed to hold “all else equal” instead induce differences.

Strain creation

We used 100 haploid lines of both mating types that had gone through mutation accumulation (Sharp et al., 2018), as well as 33 replicates of their SEY6211 ancestor as controls (MATa/ $\alpha$, ho, leu2-3 112, ura3-52, his3- $\mathrm{\Delta }$200, trp1- $\mathrm{\Delta }$901, ade2-101, suc2- $\mathrm{\Delta }$9). Half of the lines were rdh54 $\mathrm{\Delta }$::KANMX. The RDH54 gene is involved in recombinational repair (Klein, 1997), and its effect on mutation accumulation in diploids was studied in the original MA study (Sharp et al., 2018). The lines were streaked out from frozen on yeast peptone glycerol (YPG) agar plates to verify that the lines had functional respiratory pathways. Single colonies were then streaked out in patches on yeast peptone dextrose plates supplemented with additional adenine (YPAD) to inhibit reversion of the ade2 mutation (Achilli et al., 2004).

Transformation protocol

We followed the LiAc/SS carrier DNA/PEG method described by Gietz & Schiestl (2007). We grew overnight cultures from single colonies that were used to grow competent cells (cultures in exponential phase of growth). The culture of competent cells was washed with sterile water and a lithium acetate solution before we added transformation mix (containing the PCR product of interest, ssDNA, 50 $\mathrm{\%}$ PEG, water and 1 M lithium acetate). The cells were heat shocked with the transformation mix for 1 h at 42 °C. The cells were then spun down and resuspended in water before being plated on selective media corresponding to the transformation at hand.

Step 1: deletion of the STE4 locus

To avoid mating during transformation of our MAT $\alpha$ lines into $\mathrm{\Delta }$mat transformants (that act as MATa and therefore could mate with their non-transformant MAT $\alpha$ siblings (Strathern, Hicks & Herskowitz, 1981)) we chose to make our lines sterile by deleting the STE4 locus and replacing it with the TRP1 gene (for which our original lines were deletion mutants). We amplified the TRP1 locus from the pFA6a-TRP1 plasmid (a gift from John Pringle: Addgene plasmid #41595; http://n2t.net/addgene:41595; RRID:Addgene_41595, Longtine et al., 1998), using primers constructed to contain homologous sequences upstream and downstream of STE4 and bind to regions flanking TRP1 on the plasmid (see Table S1 for primer sequences). The resulting PCR product was used to replace STE4 with TRP1, see Fig. 1A.

Transformation was conducted using 1-h heat shock at 42 °C. Yeast that were successfully transformed and had their STE4 locus replaced by TRP1 were selected through growth on agar plates with synthetic complete medium lacking tryptophan (SC-Trp). A single colony from the SC-Trp plate was then streaked to single colonies again on SC-Trp. These colonies were used to verify the knockout through PCR (see laboratory notebooks on GitHub for gel pictures) and to proceed with strain creation.

Knockout of the MAT and STE4 knockout lead to increased fitness, and new variation in growth rates

Four MA lines fell 2.5 standard deviations below the average knockout effect, with large decreases in their growth rates post knockout (line IDs 2, 106, 122 and 127). We exclude these lines in the comparison of lines before and after knockout. The knockout of MAT and STE4 led to increases in growth rates of both control and MA lines (t-value 15.87, ${\chi }^2$ of model with compared to without effect of KO = 241.4, P < 10^-5), see Fig. 2A.

We found a significant difference in growth rate among MA lines, but not control lines, before knockout ( ${\chi }^2$ = 42.3, P < 10^-5 for MA lines, ${\chi }^2$ = 3.2, P = 0.073 for control lines). After knockout of the STE4 and MAT loci, we verified this result ( ${\chi }^2$ = 195, P < 10^-5 for MA lines, ${\chi }^2$ = 1.94, P = 0.16 for control lines).

Despite the promising variance in growth rates among MA lines we found no correlation between the growth rate of lines before and after knockout (Kendall’s rank correlation $\tau$ = 0.087, P = 0.22, see Fig. 2B). This is problematic because it indicates that whatever signal existed before about the impact of mutations on growth had been swamped by variation induced by the knock-outs.

High frequency of respiratory incompetent yeast (“petites”) following plasmid transformation

In an initial experiment (DS1), we generated diploids by combining two plasmid-transformed haploids, followed immediately by selection for plasmid loss. We found a large reduction in the average growth rate of these diploid lines, compared to the corresponding haploid lines, but the latter had not undergone transformation with OLP003 or OLP004. To determine if this plasmid transformation was the source of the large decreased fitness of the diploids, we re-transformed our lines, saving the haploid lines after transformation to compare the growth of these to the diploids created from them (DS2). On average, we found that the haploids had also decreased in growth rate compared to DS1. However, we also found very strong bimodality in our data, see Fig. 3.

To determine whether variation among lines might have been caused by loss of mitochondrial function we spotted our lines on YPG to test for respiratory competence. While the number of petites was low among haploids in DS1 (seven out of the 131 lines without plasmid transformation: four of these were the four lines we detected as outliers in our comparison of growth rates before and after knockout), we found a high prevalence in the diploids (169 out of 260 lines). In DS2, where haploids were isolated post plasmid transformation, we found about an equal proportion of petites in the haploid and diploid lines (about half of our lines were petite in both groups). After further analysis, we found that diploid lines that had the RDH54 gene intact were more likely to become petite ( ${\chi }^2$ for DS1 = 36, P < 10^-4, ${\chi }^2$ for DS2 = 61, P < 10^-4). We also found that the frequency of petites was significantly higher in the haploids that were transformed with OLP004 (carrying STE4 and MAT $\alpha$ and URA3) compared to the haploids that were transformed with OLP003 (carrying STE4 and LEU2) ( ${\chi }^2$ = 65, df = 1, P < 10^-5).

No correlation in growth rates across data sets

We found no significant correlation between the growth rate of lines in DS1 and DS2 with the same genotype and petite status (Table 1). Hence, we analyzed the two datasets separately.

Inferences of mutational effects across haploid, heterozygote, and homozygote yeast

The linear mixed effect models we fit to DS1 and DS2 have the same structure: maximum slope is fitted as a function of petite status, RDH54 status, MA status (whether MA or control), genotype, and initial OD as fixed effects. We also include an interaction between MA status and genotype. We control for day, machine (in DS1 only, as all DS2 replicates were measured on one machine), and haplotype effect (an identifier shared by the haploid, heterozygote, and homozygote from the same haploid genome) by including them as random effects in our model. Neither MA nor the interaction of MA and genotype is significant in DS1 (t-value = −1.45, P = 0.15 for MA). Haploid lines had a significantly higher growth rate than diploids in DS1 (t-value = −14.163 and −13.950 for heterozygotes and homozygotes respectively, P = 10^-5 for both). In DS2 MA is also not significant, but the interaction between MA and homozygosity is (t-value = 5.78, P = 10^-5), with homozygous MA lines having a higher average growth rate than homozygous control lines.

Because we found no correlation between how diploid lines of the same petite status grew in DS1 and DS2 (Table 1), we considered the possibility that unique fitness-affecting mutations occurred in each line during transformation. In effect, this would mean that the aim of our study, to compare the growth rate of identical mutated genomes in three different genotypic conditions, would be compromised. If this were the case, we would not expect a correlation between the fitnesses of the haploid and diploid homozygous state of each haplotype, because the fitness effects of pre-existing mutations would be swamped by new and different mutations in the two states. We compared our original model that used MA identity as a random effect on intercept (with the same identity for the haploid and homozygous lines derived from the same MA line) to a model using line identity as a random effect (that treats each genotype as its own group). In this latter model, there is no information on the original haplotype used to establish the line, but the fixed effects remain the same (petite status, RDH54 status, MA status (whether MA or control), genotype, and initial OD). The model using line identity rather than haplotype is preferred in both DS1 ( $\mathrm{\Delta }$AIC = −1,021) and DS2 ( $\mathrm{\Delta }$AIC = −2,579), suggesting that our MA mutations are overwhelmed by unique fitness-affecting mutations that occurred in each line during transformation. These results remained unchanged when excluding the four lines that were outliers in the KO to MA comparison. Our findings are evident also in plotting the correlation of heterozygote and homozygote fitness (Fig. 4): we find a significant correlation only in the grande MA lines in DS2. The fact that the correlation was not found in DS1 further emphasizes the lack of repeatability in our experiment.

Deletion of RDH54 protects against petite-formation

We found an unexpected pattern where the rdh54 $\mathrm{\Delta }$ deletion mutants were less likely than the wildtype to become respiratory deficient following transformations. Petite yeast are generally more heat tolerant than wildtype yeast (Van Uden, 1985). rdh54 $\mathrm{\Delta }$ deletion mutants at stationary phase have also been shown to be tolerant to heat shock (Jarolim et al., 2013). Even though our strains were subjected to heat shock for transformation in the exponential phase, there may be interactions between temperature, the Rdh54 protein, and mitochondrial function that accounts for the pattern.

A plethora of genes are involved in mitochondrial genome stability (see Wide-scale screening for rho0 production in Contamine & Picard (2000)). Although there is no evidence as far as we know that the Rdh54 protein is involved in mitochondrial repair or recombination (Chen, 2013), the suggestion has been made that petite yeast could arise from homologous recombination between imperfect duplicates (Faye et al., 1973; Gaillard, Strauss & Bernardi, 1980; Slonimski & Lazowska, 1977; Weiller et al., 1991); see also the “Recombination/repair track” in Contamine & Picard (2000)), and Rdh54 is a vital protein for homologous recombination in the nucleus. Thus another explanation is that the deletion of RDH54 could reduce mitochondrial recombination rates.