Multi-locus Genotypes Underlying Temperature Sensitivity in a Mutationally Induced Trait

Determining how genetic variation alters the expression of heritable phenotypes across conditions is important for agriculture, evolution, and medicine. Central to this problem is the concept of genotype-by-environment interaction (or ‘GxE’), which occurs when segregating genetic variation causes individuals to show different phenotypic responses to the environment. While many studies have sought to identify individual loci that contribute to GxE, obtaining a deeper understanding of this phenomenon may require defining how sets of loci collectively alter the relationship between genotype, environment, and phenotype. Here, we identify combinations of alleles at seven loci that control how a mutationally induced colony phenotype is expressed across a range of temperatures (21, 30, and 37°C) in a panel of yeast recombinants. We show that five predominant multi-locus genotypes involving the detected loci result in trait expression with varying degrees of temperature sensitivity. By comparing these genotypes and their patterns of trait expression across temperatures, we demonstrate that the involved alleles contribute to temperature sensitivity in different ways. While alleles of the transcription factor MSS11 specify the potential temperatures at which the trait can occur, alleles at the other loci modify temperature sensitivity within the range established by MSS11 in a genetic background- and/or temperature-dependent manner. Our results not only represent one of the first characterizations of GxE at the resolution of multi-locus genotypes, but also provide an example of the different roles that genetic variants can play in altering trait expression across conditions.


Introduction
Genotype-by-environment interaction (or 'GxE') occurs when genetically distinct individuals exhibit different phenotypic responses to the environment [1,2]. Work to date suggests that GxE is an important contributor to heritable variation in many agriculturally, evolutionarily, and medically relevant phenotypes (as described in [3][4][5][6] and elsewhere). However, although 'GxE' has been extensively studied, there are few, if any, traits for which the underlying genetic basis of GxE is fully understood. This lack of detailed case studies may have a technical basis, as causal loci involved in GxE can act in an environment-and genetic background-dependent manner [7,8], making them difficult to detect. Improving understanding of GxE could therefore require characterizing how combinations of alleles, rather than individual loci, influence phenotype across environments.
The aforementioned results stem from work performed exclusively at 30°C, the standard temperature used to culture S. cerevisiae in the lab. Here, we extend our research on the rough phenotype to two additional temperatures: 21°C and 37°C. In doing so, we find that many BYx3S ira2Δ2933 segregants express the rough phenotype in a temperature sensitive manner (Fig 1). To determine the genetic basis of this GxE, we perform genetic mapping of several temperature sensitivity classes using a combination of bulk segregant analysis [23][24][25] and selective genotyping of individual cross progeny [10,26]. These efforts lead to the identification of seven environmentally responsive loci, and five specific multi-locus genotypes involving these loci, that influence the expression of the rough phenotype across temperatures. As we describe below, comparison of these multi-locus genotypes provides detailed insights into the genetic architecture of temperature-dependent GxE in our system, and also sheds light on the distinct roles that the causal alleles play in modifying the rough phenotype's expression at different temperatures.

Results and Discussion
Screen for temperature sensitivity among rough BYx3S ira2Δ2933 segregants We screened for rough BYx3S ira2Δ2933 segregants at three temperatures-21, 30, and 37°Cin a backcross of the rough ira2Δ2933 segregant described in [10] to BY (S1 and S2 Tables; Materials and Methods).~3,000 segregants (30 random spore plates of~100 colonies per plate) Fig 1. Segregants that carry ira2Δ2933 show different temperature sensitivities. BY4716 (BY), 322134S (3S), and their recombinant offspring typically exhibit 'smooth' colony morphology. However, in the presence of the ira2Δ2933 frameshift mutation, some BYx3S segregants are capable of expressing an alternative, 'rough' colony phenotype. Examination of rough BYx3S ira2Δ2933 segregants at 21, 30, and 37°C revealed that these individuals largely express the phenotype in a temperature sensitive manner. For the most part, individuals exhibit the phenotype only at 21°C, only at 21°C and 30°C, or at all examined temperatures. We refer to these three classes as highly sensitive (HS), moderately sensitive (MS), and non-sensitive (NS) to temperature, respectively. were examined at each temperature and~9,000 segregants were screened in total across the three temperatures (Materials and Methods). Among the 252 rough individuals obtained from this screen, 173, 107, and 72 were recovered from 21, 30, and 37°C, respectively (S1 and S2 Tables). The majority of these rough segregants were capable of expressing the phenotype at temperatures other than the one from which they were collected. When these rough individuals were individually examined at each of the three initially employed temperatures, they largely fell into three classes: rough at 21°C only, rough at 21 and 30°C only, or rough at all three temperatures (Fig 1; S1 and S3 Tables; Materials and Methods). This implies that the major form of temperature-dependent GxE in our system is temperature sensitivity. For the remainder of the paper, we refer to the three aforementioned classes as highly sensitive ('HS'), moderately sensitive ('MS'), and non-sensitive ('NS') to temperature, respectively (Fig 1).
We sought to determine the underlying genetic basis of the temperature-dependent GxE by selectively genotyping individuals in each temperature sensitivity class. To generate the necessary populations for this genetic mapping strategy, we screened an additional 60 plates of random spores from the BY backcross mentioned above, and also mated the rough ira2Δ2933 segregant described in [10] to its 3S parent and screened 90 random spore plates from this 3S backcross (Materials and Methods). Given that spores were plated at a density of~100 colonies per plate, we estimate that we examined an additional~15,000 individuals as a part of this second screen. This second screen was exclusively conducted at 21°C, as all three temperature sensitivity classes can be recovered from this condition (S2 Table). Collected individuals were then stringently phenotyped at 21, 30, and 37°C (Materials and Methods). In combination with our preliminary screen, we recovered 544 and 466 rough backcross progeny from the BY and 3S backcrosses, respectively, with 78.4% of these individuals classified as HS, MS, or NS (S1 and S4 Tables; Materials and Methods).

Bulk segregant mapping of the three temperature sensitivity classes
We first attempted to determine the genetic bases of the three temperature sensitivity classes using bulk segregant mapping by sequencing [23][24][25] (Fig 2A). Between 51 and 126 individuals were pooled per backcross and temperature sensitivity class, and each pool was sequenced to at least 114X coverage (S4 Table; Materials and Methods). Across the six pools, eight distinct loci were detected using MULTIPOOL [27] (Figs 2B and S1; Materials and Methods). Seven of these loci overlapped ira2Δ2933 or causal alleles that we previously identified as contributors to the phenotype at 30°C: END3 BY , FLO8 3S , MGA1 BY , MSS11 BY , SFL1 BY , and TRR1 3S [10,11] ( Fig 2B). We used genetic engineering to show that the final locus, which was detected on Chromosome IX in the NS class, corresponds to the 3S allele of FLO11, which encodes a cell surface glycoprotein that is required for the rough phenotype [11] (Materials and Methods). Specifically, we replaced the 3S version of the FLO11 coding region with the BY allele in an NS individual from the BY backcross, and found the resulting allele swap strain only exhibited rough morphology at 21 and 30°C (Fig 3).
Loci that contributed to temperature sensitivity could be distinguished from those that did not based on the bulk segregant mapping results. FLO8 3S and ira2Δ2933 were detected in every temperature sensitivity class (Fig 2B; S1 Note), suggesting they are involved in the general expression of the rough phenotype but do not influence its temperature sensitivity. In contrast, the other six involved alleles-END3 BY , FLO11 3S , MGA1 BY , MSS11 BY , SFL1 BY , and TRR1 3Swere each detected in just one or two of the classes, indicating they contribute to the observed differences in temperature sensitivity ( Fig 2B). Of these six alleles, zero, three, and six were detected in the HS, MS, and NS classes, respectively, and the three that were detected among MS individuals-END3 BY , MSS11 BY , and TRR1 3S -were also identified among NS individuals. These findings indicate that the temperature sensitivity of the rough phenotype is largely  by the same loci that were originally determined to underlie the trait's expression at 30°C [10,11], and that the rough phenotype's temperature sensitivity can be reduced or even eliminated by combining particular alleles at these loci. In particular, we note that individuals with the genotype END3 BY FLO8 3S FLO11 3S ira2Δ2933 MGA1 BY MSS11 BY SFL1 BY TRR1 3S did not exhibit any temperature sensitivity in our experiments.

Multiple genotypes underlie the HS and MS classes
Although bulk segregant mapping is known to be a statistically powerful technique when large numbers of cross progeny are examined [24], it can fail to detect causal loci if distinct combinations of alleles that interact with each other or the environment exhibit indistinguishable phenotypes [10,26]. This phenomenon, which we refer to here as 'genotypic heterogeneity', clearly occurred in the present data for the MS class. As described in the introduction, we previously showed that two specific multi-locus genotypes-END3 BY FLO8 3S ira2Δ2933 MSS11 BY TRR1 3S and END3 3S FLO8 3S ira2Δ2933 MGA1 BY MSS11 BY SFL1 BY -express the rough phenotype at 30°C [11] (Fig 4A). Phenotyping of previously described segregants [10,11] revealed that both of these allele combinations lead to moderate temperature sensitivity-i.e., expression of the trait at 21 and 30°C, but not 37°C (S2 Fig). In our current data for the MS class, alleles required for both genotypes were fixed, while alleles involved in just one of the two genotypes were merely enriched or not even detected (Fig 2B; S2 Note). To directly show that our sample of MS individuals was comprised of both previously identified multi-locus genotypes, we genotyped 19 random MS segregants using restriction markers for END3 and MGA1. This effort revealed that 14 and 5 of these individuals possessed the END3 BY FLO8 3S ira2Δ2933 MSS11 BY TRR1 3S and END3 3S FLO8 3S ira2Δ2933 MGA1 BY MSS11 BY SFL1 BY genotypes, respectively (S3 Note; Materials and Methods).
which we had previously resolved to specific genes and one of which we cloned in the current paper. The identities of these genes are stated above their corresponding loci. Alleles detected from BY and 3S are shown in blue and orange, respectively, while the ira2Δ2933 mutation is denoted in red. The color intensity of a locus corresponds to its allele frequency in the bulk segregant mapping data. A legend with the correspondence between allele frequency and color intensity is provided at the bottom of the figure. To aid in visualization, loci are depicted in the main text figures as having the same widths.
. This allele replacement resulted in a conversion from rough to smooth colony morphology specifically at 37°C. In this picture, the FLO11 3S individual is a segregant that has not been genetically modified, while the FLO11 BY individual is the same strain with its FLO11 allele swapped. The phenotypes of both of these strains at 21, 30, and 37°C are shown. After recognizing the genotypic heterogeneity underlying the MS class (Fig 4A), we investigated whether the HS class, for which only FLO8 3S and ira2Δ2933 were detected by bulk segregant mapping (Fig 2A), might also be genotypically heterogeneous. To examine this possibility, we individually genotyped each HS segregant by performing low coverage whole genome sequencing (Materials and Methods). We then used χ 2 tests to scan the genomes of the ira2Δ2933 segregants in the HS class for pairs of loci that exhibited correlated allele states (Materials and Methods). Such associations might be expected if alleles at two loci participate We previously showed that two distinct multi-locus genotypes underlie trait expression at 30°C [10,11], and in this paper we determined that these same genotypes underlie the MS class. In addition to ira2Δ2933, these genotypes both involve FLO8 3S and MSS11 BY . However, individuals carrying END3 BY also require TRR1 3S ('END3 BY -dependent genotype'), while individuals carrying END3 3S instead require MGA1 BY and SFL1 BY ('END3 3S -dependent genotype'). We refer to the fact that both of these multi-locus genotypes specify the same trait as 'genotypic heterogeneity'. When such genotypic heterogeneity is present, alleles involved in only one of the multi-locus genotypes can be masked. For example, in the bulk segregant mapping data for the MS class, only the alleles involved in the genotype that involves END3 BY were detected. (B) In the current data, we found evidence for genotypic heterogeneity in the HS class. By partitioning individuals in the HS class from the 3S backcross population and calculating their allele frequencies across the genome, we determined that two predominant multi-locus genotypes underlie this temperature sensitivity class. These genotypes are FLO8 3S ira2Δ2933 MSS11 BY ('MSS11 BY -dependent genotype') and XII 3S END3 BY FLO8 3S ira2Δ2933 MGA1 BY MSS11 3S SFL1 BY ('MSS11 3S -dependent genotype'). As described in the main text, XII 3S refers to a locus that was detected specifically among HS individuals carrying MSS11 3S . The same coloring scheme is used in this figure as in  in the same multi-locus genotype. At a 1% false discovery rate threshold, we detected no significant pairs of loci in the BY backcross and two significant pairs of loci in the 3S backcross: Chromosome XII-Chromosome XIII and Chromosome XIII-Chromosome XV (S3 Fig; Materials and Methods). Both of these significant locus pairs included a region of Chromosome XIII that overlapped MSS11. As for the other two detected loci, the Chromosome XV region overlapped SFL1 and the Chromosome XII region was novel relative to our past work [10,11]. As we have yet to determine the causal gene at the Chromosomes XII locus (S4 Note), we hereafter refer to it by its chromosome number: 'XII'.
Detection of correlated loci among individuals from the 3S backcross suggested that genotypic heterogeneity might have led to only FLO8 3S and ira2Δ2933 being identified by bulk segregant mapping focused on the HS class ( Fig 2B). Because MSS11 was present in both significant locus pairs, we split the 3S backcross progeny by individuals' genotypes at MSS11 and separately examined genome-wide allele frequencies in the two resulting groups (Materials and Methods). This analysis revealed that individuals with MSS11 BY only need the specific alleles ira2Δ2933 and FLO8 3S to express the phenotype at 21°C, whereas individuals with MSS11 3S require a number of additional alleles to show rough morphology under the same condition ( Fig 4B; Materials

Roles of detected alleles in modulating temperature sensitivity
Based on our genetic mapping efforts described in this paper, we have identified seven environmentally responsive loci that influence the expression of the rough phenotype across temperatures: XII, END3, FLO11, MGA1, MSS11, SFL1, and TRR1. Furthermore, through our current and past efforts [10,11], we have characterized five predominant multi-locus genotypes involving these loci that exhibit different levels of temperature sensitivity. These include two HS genotypes, two MS genotypes, and a single NS genotype (Fig 5).
Comparison of the alleles involved in these five multi-locus genotypes suggests that the seven environmentally responsive loci play different roles in modifying the rough phenotype's temperature sensitivity. In particular, MSS11 appears to determine the range of temperatures at which the phenotype can be expressed (Fig 5). Indeed, while both MSS11 alleles can facilitate expression of rough morphology at 21°C, only individuals that carry MSS11 BY have the genetic potential to express the trait at 30 or 37°C (Fig 5). In contrast, the other identified alleles appear to collectively increase or decrease temperature sensitivity within the range established by MSS11 (Fig 5). These modifier alleles show phenotypic effects that can depend on both genetic background and temperature, although the degree of such dependencies varies. For example, END3 BY , MGA1 BY , and SFL1 BY were detected in one multi-locus genotype in each temperature sensitivity class (Fig 5), and appear to influence temperature sensitivity in a genetic background-dependent manner (see S4 and S5 Figs, as well as [10], for genetic engineering results supporting this point for MGA1 BY ). The remaining alleles appear to have effects on temperature sensitivity that depend on both genetic background and temperature. Specifically, XII 3S and FLO11 3S were each detected in a single multi-locus genotype and temperature sensitivity class, while TRR1 3S was only identified in the END3 BY MSS11 BY genetic background among MS and NS individuals (Fig 5).

Conclusion
Across the current manuscript and our previous work [10,11], we have now described eight loci and five predominant multi-locus genotypes that influence whether BYx3S segregants carrying ira2Δ2933 can express the rough phenotype in at least one temperature. Of the eight identified loci, only one-the transcription factor FLO8-does not contribute to temperature sensitivity. This likely reflects the fact that FLO8 encodes a transcriptional activator that is required for expression of the rough phenotype [10], and BY harbors a nonfunctional version of this gene [26,28]. For the remaining seven loci, it is difficult to distinguish between their effects on trait expression in general and their effects on temperature sensitivity. Indeed, our results indicate that genetic background and temperature together determine which alleles are required for expression of the rough phenotype.
Our findings also suggest molecular mechanisms that might underlie the rough phenotype's temperature sensitivity. For instance, some of the alleles involved in temperature sensitivity may have reduced biochemical activity and/or diminished structural stability at 30 or 37°C. The most striking example of this is MSS11 3S , which can only facilitate trait expression at 21°C. Given that Mss11 acts as a heterodimer with Flo8 [29], it might be that Flo8-Mss11 3S does not dimerize well, poorly binds DNA, or is unable to stimulate RNA polymerase II activity at 30 and 37°C. Supporting such possibilities, we determined that the causal variant in MSS11 is an isoleucine to serine amino acid change that occurs in the LisH motif required for dimerization with Flo8 [29] (S6 Fig). 3S and roughly half of the other available sequenced S. cerevisiae isolates carry the derived version of this site, which is the serine allele that results in temperature sensitivity (S6 Fig). The data also suggest FLO11, which plays a crucial role in cell-cell and cellsurface adhesion [30,31], harbors a temperature sensitive polymorphism in its coding region. Instability of Flo11 BY at 37°C could result in temperature-dependent suppression of the rough phenotype among FLO11 BY individuals and would explain why FLO11 3S is required by the NS class. Comparison of multi-locus genotypes underlying differences in temperature sensitivity among BYx3S FLO8 3S ira2Δ2933 segregants. Across our current study and past work, we have identified five predominant multi-locus genotypes that underlie the three temperature sensitivity classes. Excluding FLO8 3S , which is generally required for expression of the trait in the presence of ira2Δ2933, these genotypes involve specific alleles at seven environmentally responsive loci. Respectively, blue and orange indicate the BY or 3S allele of a given gene participates in a specific multi-locus genotype, while grey indicates that neither allele is required. Additionally, brown lines emphasize how adding particular combinations of alleles to a FLO8 3S ira2Δ2933 MSS11 BY genetic background can lead to reduction or elimination of temperature sensitivity. Temperature sensitivity at the phenotypic level may also be caused by the combined effects of genetic background and temperature on Ras signaling and Ras-dependent gene regulation. We note that Flo8-Mss11, Mga1, and Sfl1 are all Ras-regulated transcription factors [18]. Furthermore, Flo8-Mss11 and Sfl1 are known to play particularly important roles in the expression of multicellular phenotypes in yeast, as they compete to bind DNA and are regulated in an antagonistic manner at the posttranslational level by Protein Kinase A, the effector kinase of the Ras cascade [30,32,33]. In fact, we previously showed that ira2Δ2933 reveals the rough phenotype in certain genetic backgrounds by reducing Sfl1-mediated transcriptional repression [11]. Viewing our current work in light of our past findings suggests that the temperature sensitivity described in this paper results in part from genotype-temperature combinations that conditionally reduce Ras signaling and/or Ras-regulated gene expression. Consistent with this possibility, others have noted functional relationships between Ras signaling and both End3 [34] and the oxidative stress response, of which Trr1 is a component, in yeast [34,35]. Additionally, temperature has been reported to affect levels of Ras signaling in human fibroblast cultures [36].
Our results also shed light on the genetic basis of phenotypic capacitance-i.e., the uncovering of cryptic genetic variation by environmental or mutational perturbation [37][38][39][40][41][42]. As we previously described, all of the polymorphisms that influence rough morphology in the BYx3S ira2Δ2933 cross can be considered cryptic genetic variants, as they do not cause the rough phenotype under standard conditions in the absence of the IRA2 mutation [11]. Here, we have shown the multi-locus genotypes that provide the genetic potential for ira2Δ2933 to uncover the rough phenotype differ across temperatures. Moreover, we have demonstrated that these multi-locus genotypes that facilitate phenotypic capacitance vary not only in their initial temperature sensitivities, but also in their potential to reduce their temperature sensitivities through segregating genetic variation (Fig 5). This latter finding has potential relevance for our understanding of genetic assimilation, the process by which environmentally induced traits are converted into constitutively expressed phenotypes by natural selection [42][43][44][45][46][47][48][49][50][51][52][53][54][55][56], as our results provide an example of the genetic architecture that might underlie this phenomenon.
Lastly, our study provides technical insights into research aimed at determining the genetic basis of GxE. First, we have shown that selective genotyping of individuals that exhibit particular levels of environmental sensitivity can identify multi-locus genotypes that cause GxE, rather than just individual contributing loci. Second, we have demonstrated that genotypic heterogeneity can complicate efforts to genetically dissect GxE and have described a strategy to overcome this challenge. Third, we have illustrated how characterizing the genetic basis of GxE at the resolution of multi-locus genotypes can clarify the different roles that contributing loci play in altering trait expression across conditions. These technical insights that have emerged from our work will likely be relevant to future studies of GxE in other species and traits.

Generation of backcross segregants for the preliminary screen and genetic mapping
The rough BYx3S MATa F 2 segregant used for genetic mapping in [10] was mated to MATα versions of both BY and 3S. Diploid zygotes were obtained from each backcross mating using microdissection, and then sporulated at 21°C using standard yeast sporulation methods [57]. Spore cultures were digested with β-glucoronidase and random MATa spores were selected on yeast nitrogen base (YNB) plates containing canavanine using the Synthetic Genetic Array (SGA) marker system [58], as described previously [10,24]. Implementation of the SGA system was possible because the MATa F 2 segregant used in backcrossing possessed the markers can1Δ::STE2pr-SpHIS5 and his3Δ [58], and matings were performed to BY and 3S MATα his3Δ strains. Spores were plated at low density (~100 spores per plate) so that individual colonies could be easily distinguished. After five days of growth on YNB + canavanine plates, colonies were replicated onto yeast extract-peptone-ethanol (YPE) plates. These YPE plates were incubated at the specified temperature (21, 30, or 37°C) for five days and then screened by eye for colonies exhibiting the rough phenotype. Strains identified as rough were picked from the plates, inoculated in liquid yeast extract-peptone-dextrose (YPD) media, and grown overnight at 30°C. Freezer stocks of the rough backcross segregants were generated by mixing equal volumes of 40% glycerol solution with a portion of the liquid YPD cultures and then storing these cultures at -80°C.

Phenotyping of rough segregants at multiple temperatures
Cells from the freezer stocks described above were inoculated into 800 μl of liquid YPD media. These YPD cultures were grown for two days at 30°C and then manually pinned onto three YPE plates. Of these three plates, one was incubated at 21°C, one was incubated at 30°C, and one was incubated at 37°C. Individuals were scored manually on a scale of 0 to 5 based on their degree of expression of the rough phenotype. The low and high ends of this scale correspond to individuals that were smooth and rough, respectively. Individuals with intermediate scores were bumpy to varying degrees, which may reflect weaker expression of the phenotype. Three biological replicate cultures were screened for each segregant. For the purposes of the paper, we viewed an individual's phenotype at a given temperature as their median score obtained for that temperature. The results described in the paper are based on individuals that clearly and consistently showed smooth or rough phenotypes at a given temperature. We considered median scores between 0 and 1 or 4 and 5 as clearly indicating the smooth and rough phenotypes, respectively. Phenotype data are reported in S1 Table. Bulk segregant mapping of temperature sensitivity Cultures were inoculated from the freezer stocks described above into 800 μl of liquid YPD. After two days of growth at 30°C, 100 μl of each culture from the same backcross and temperature sensitivity class were mixed together, and DNA was extracted from these pools using the Qiagen DNeasy Blood and Tissue kit. DNA sequencing libraries were then constructed using the Illumina Nextera kit. Each library was barcoded with a distinct sequence tag to facilitate multiplex sequencing. Libraries were mixed in equimolar fractions and sequenced on an Illumina NextSeq machine using 75x75 or 150x150 base pair reads. Sequencing reads were mapped to both the reference genome for S288c, which is the progenitor of BY, and the 3S draft genome (available through http://www.yeastgenome.org) using the Burrows-Wheeler Aligner (BWA) version 7 with options mem -t 20 [59]. We then used SAMtools to obtain mpileup files for each sample [60]. Based on these mpileups, we determined that the pools were sequenced to an average per site coverage of at least~114X (S4 Table). Genome-wide allele frequencies were determined at 36,756 high confidence SNPs that were previously identified by mapping Illumina sequencing reads for 3S to the S288c reference genome [10,11]. Because roughly half of the genome is fixed in each backcross, we subsetted out the data for segregating regions in each backcross and analyzed each of these regions individually. To identify significantly enriched loci, we used MULTIPOOL [27] with the settings: replicates mode, 3,300 bp centimorgans, 100 bp bins. Significant loci were defined as genomic regions that had a maximum LOD score of at least 5 for a span of at least 20 kb. Confidence intervals were estimated as the bounds of a locus that correspond to a 2 LOD drop from the point of maximal significance. To generate allele frequency plots, the data was smoothed by averaging allele frequency over sliding windows of 25 SNPs.

Genotyping of MS individuals using PCR and restriction digestion
DNA was extracted from 19 randomly chosen BY backcross segregants from the MS class using the Qiagen DNeasy Blood and Tissue kit. Small regions of these genes that contained a SNP were amplified by PCR and then digested with restriction enzymes (see S5 Table for a description of specific reagents). The amplified regions were chosen so that one parental allele would be cut a single time while the other parental allele would not be cut at all. Each diagnostic restriction digest was tested on BY and 3S. Digested PCR products were examined on a 1.5% agarose gel containing ethidium bromide to determine each individual's genotype at a given locus.

Detailed genetic characterization of the HS class
Respectively, 68 and 89 HS segregants were independently prepared for sequencing from the backcrosses to BY and 3S. Individual cultures were generated by inoculating 800 μl of liquid YPD with cells from the freezer stocks described above. After two days of growth at 30°C, DNA was extracted from each individual culture using the Qiagen DNeasy 96 Blood and Tissue kit. Libraries were prepared for sequencing using the Illumina Nextera kit, with each individual receiving a unique sequence barcode. Sequencing was performed on an Illumina NextSeq machine using 75x75 base pair reads. Illumina reads were then mapped to the S288c genome, and mpileups were generated from these alignments using BWA [59] and SAMtools [60] in the same manner described earlier for the bulk segregant mapping data. Individuals with an average per site coverage below 1.02X were excluded from subsequent analyses. Respectively, one and five individuals were excluded from the BY and 3S backcrosses due to low coverage. Furthermore, 16 sequenced individuals from the BY backcross were excluded from downstream analysis because they possessed a wild type IRA2 (S1 Note). As previously described [10], a Hidden Markov Model (HMM) was used to determine the haplotype of each segregant from the aforementioned 36,756 SNP differences between the BY and 3S genomes. As with the bulk segregant data, we only performed our statistical tests on regions of the genome that segregated in a given backcross. Furthermore, to reduce our total number of statistical tests, we collapsed linked SNPs that showed the same pattern of inheritance across backcross segregants into unique segregating regions. There were 1,399 and 767 such segregating regions in the BY and 3S backcross populations, respectively. We used the R statistical programming environment to identify pairs of loci on different chromosomes that showed correlated alleles states based on Χ 2 tests. Pairs of loci were considered statistically significant if they exhibited a point-wise false discovery rate (q-value) of 1% or less, as determined when p-values for all segregating regions were converted into q-values by the qvalue() package in R [61,62]. Regions within 20,000 bases of the ends of chromosomes were excluded from this analysis due to the problems in mapping Illumina reads to telomeric regions. Based on the results from the initial scan for correlated loci, HS individuals from the 3S backcross were subsetted by their genotype at MSS11 and genome-wide allele frequency plots were generated from the aforementioned HMM tables. Allele frequencies were averaged over 25 SNP sliding windows, and loci were called as significant in a given MSS11 background if they exhibited allele frequencies below 10% or greater than 90%.

Genetic engineering
Adaptamer-mediated allele replacement [26,63] was used to alter the allele state at MGA1 and FLO11 in a NS segregant with the genotype END3 BY FLO8 3S FLO11 3S ira2Δ2933 MGA1 BY MSS11 BY SFL1 BY TRR1 3S , as well as END3, MGA1, and SFL1 in a HS segregant with the genotype XII 3S END3 BY FLO8 3S ira2Δ2933 MGA1 BY MSS11 3S SFL1 BY . Using PCR, an amplicon of the gene of interest was tailed at the 3' end with the 5' end of the kanMX cassette, and an amplicon of kanMX was tailed on the 3' end with the region immediately downstream of the gene [26]. The two PCR products were co-transformed into a given strain using the lithium acetate method [65] and plated on YPD agar containing G418 to screen for successful integration. Colonies that showed G418 resistance were then checked by PCR and Sanger sequencing to ensure they harbored the allele replacement. To clone the causal nucleotide in MSS11, adaptamermediated allele replacement was performed multiple times using MSS11 amplicons that spanned only a part of the gene's coding region. These specific engineerings are shown in more detail in S6 Fig. Each of these engineerings to clone the causal variant in MSS11 were checked by Sanger sequencing. The gene deletions described in S4 Note were performed by replacing a gene of interest with the CORE cassette [64]. Regions corresponding to 60 bases upstream and downstream of the target gene were tailed to the CORE cassette using PCR. This product was transformed into cells using the lithium acetate method [65], and selection with G418 was used to screen for integration of the cassette. PCR was then used to verify that deletion strains recovered from the G418 selection lacked the gene of interest. All primers used for genetic engineering are described in S5 Table. Supporting Information    The causal nucleotide in MSS11 results in an isoleucine to serine amino acid substitution in the LisH domain required for Flo8-Mss11 dimerization. 3S carries the derived, serine allele of this amino acid. Inspection of the MSS11 genotypes of other sequenced S. cerevisiae isolates revealed that roughly 56% of strains also harbor the serine allele. Mss11 protein sequence data were obtained from the Saccharomyces Genome Database (http:// www.yeastgenome.org). A scale bar is provided in amino acids. (TIF) S1 Note. The ira2Δ2933 allele was highly enriched (86.1% frequency) but not completely fixed among HS individuals in the BY backcross. Based on whole genome sequencing described later in the paper, we identified 16 IRA2 BY individuals. These individuals were excluded from further consideration, as our goal in this paper was to characterize GxE in an ira2Δ2933 background. We also note that the number of these individuals was too low to enable detection of loci that enable rough morphology in the absence of ira2Δ2933. (DOCX) S2 Note. Because the END3 BY -and END3 3S -dependent genotypes require five and six alleles, respectively, the latter genotype is expected to only occur half as often. In practice, the bias towards the END3 BY -dependent genotype is typically even higher. This is because a locus that confers a selective advantage during random spore isolation in the BYx3S cross is closely linked to END3, with the BY allele of this locus conferring a benefit [10,11]. (DOCX) S3 Note. Nineteen MS individuals from the 3S backcross population were randomly chosen, and genotyped at END3 and MGA1. Fourteen of these individuals carried END3 BY , and the BY and 3S alleles of MGA1 were present in equal frequencies among these END3 BY MS individuals. In contrast, all five segregants carrying END3 3S also harbored MGA1 BY . These are consistent with our past results that two genotypes-END3 BY FLO8 3S ira2Δ2933 MSS11 BY TRR1 3S and END3 3S FLO8 3S ira2Δ2933 MGA1 BY MSS11 BY SFL1 BY -underlie the MS class, as MGA1 BY co-segregates with END3 3S but exhibits no such association with END3 BY . (DOCX) S4 Note. The Chromosome XII interval was delimited to a 34,519 base region. Within this interval, the candidate genes HAP1, HSP60, GSY2, LCB5, PDR8, SYM1, YLR257W, and YPT6 were independently deleted in a 3S backcross segregant expressing the HS phenotype and carrying the XII 3S END3 BY FLO8 3S ira2Δ2933 MGA1 BY MSS11 3S SFL1 BY genotype (Materials and Methods). None of these gene deletions resulted in a loss of rough morphology. This indicates that either the causal allele at XII 3S is a loss-of-function polymorphism or none of the tested genes are the causal gene at this locus. (DOCX) S1 Table. Phenotypes of BYx3S ira2Δ2933 backcross segregants in preliminary and secondary screens. (XLSX) S2 Table. Initial screen for rough morphology among segregants isolated at three different temperatures. Random spore plates from the BY backcross were screened for rough colonies at 21, 30, or 37°C (Materials and Methods). Rough individuals isolated from each of the temperatures were then examined at all three temperatures for the ability to express the phenotype. (DOCX) S3 Table. Classification of BY backcross segregants obtained from the preliminary screen into the three temperature sensitivity classes. These data are based on the same individuals and phenotyping results described in S2 Table. (DOCX) S4 Table. Bulk segregant mapping populations were generated for each temperature sensitivity class. Subsets of segregants from each backcross and sensitivity class were gathered for pooled sequencing. Of the 1,010 collected segregants at 21°C, 78.4% of them exhibited the HS, MS, or NS phenotype. DNA from between 51 and 131 individuals of each backcross and class was combined to form six pools. Each pool was sequenced to a minimum coverage of 114X. (DOCX) S5 Table. Primers used in this study. (XLSX)