Joint effects of genes underlying a temperature specialization tradeoff in yeast

A central goal of evolutionary genetics is to understand, at the molecular level, how organisms adapt to their environments. For a given trait, the answer often involves the acquisition of variants at unlinked sites across the genome. Genomic methods have achieved landmark successes in pinpointing these adaptive loci. To figure out how a suite of adaptive alleles work together, and to what extent they can reconstitute the phenotype of interest, requires their transfer into an exogenous background. We studied the joint effect of adaptive, gain-of-function thermotolerance alleles at eight unlinked genes from Saccharomyces cerevisiae, when introduced into a thermosensitive sister species, S. paradoxus. Although the loci damped each other’s beneficial impact (that is, they were subject to negative epistasis), most boosted high-temperature growth alone and in combination, and none was deleterious. The complete set of eight genes was sufficient to confer ~15% of the S. cerevisiae thermotolerance phenotype in the S. paradoxus background. The same loci also contributed to a heretofore unknown advantage in cold growth by S. paradoxus. Together, our data establish temperature resistance in yeasts as a model case of a genetically complex evolutionary tradeoff, which can be partly reconstituted from the sequential assembly of unlinked underlying loci.


Introduction
Understanding how organisms acquire new traits is a driving question in evolutionary biology. Many traits of interest are adaptations, meaning they provide a fitness benefit to the organism which has driven their rise to high frequency in the population. Genomic methods often find a slew of unlinked changes at the DNA level that associate with a given adaptive trait [1][2][3][4]. For any one candidate locus, gold-standard validation experiments will then swap alleles between taxa and test for an impact on phenotype, at genic [5][6][7][8][9][10] and sub-genic [11][12][13][14][15][16][17][18][19][20] levels of resolution. Though elegant and rigorous, this focused approach on a single gene at a time will by necessity leave polygenic mechanisms of the trait less well characterized.
For a more complete picture of a complex adaptation, we would establish how multiple underlying genes work together, including their interdependence and their joint ability to recapitulate the phenotype. Such questions have come within reach in laboratory evolution, with particular emphasis on genomic methods to infer evidence for epistasis between unlinked adaptive loci [7,[21][22][23][24][25][26][27][28][29]. In a handful of cases, adaptive multi-gene interactions from a lab evolution have been verified experimentally by allelic replacement [30,31]. To date, validating these principles in the context of evolution from the wild has posed a key challenge (although see [32][33][34]).
To study complex genetic mechanisms in adaptation, we set out to use natural variation in Saccharomyces yeasts as a model. S. cerevisiae strains, from the wild and the lab, grow at temperatures up to 41˚C [35][36][37]. All other species in the Saccharomyces clade, which diverged from a common ancestor~20 million years ago, grow poorly at high temperatures, though many outperform S. cerevisiae in the cold [38]. In previous work [39] we developed a genomic version of the reciprocal hemizygosity test to dissect thermotolerance, using S. paradoxus, the closest sister species to S. cerevisiae, as a representative of the inferred ancestral state. Derived alleles of the mapped genes in S. cerevisiae, when tested individually for their marginal effects, were partially necessary or sufficient for thermotolerance, or both [39], and their sequences exhibit evidence for positive selection in S. cerevisiae [39,40]. But how these genes work together has remained unknown. We thus aimed to investigate the extent to which unlinked thermotolerance loci assembled in the same background would explain the trait, and whether and how these genes would depend on one another for their effects. We expected that any answers could also help elucidate other facets of the mechanism and the evolutionary history of thermotolerance.

Combining thermotolerance loci to reconstitute a complex trait
We previously mapped eight genes with pro-thermotolerance alleles in S. cerevisiae (Table 1; [39]). To explore the joint function of these unlinked loci, we introduced the alleles of all eight from DBVPG1373, a soil isolate of S. cerevisiae from the Netherlands, into Z1, a strain of the sister species S. paradoxus, isolated from an English oak tree [41]. Our approach used a stepwise set of gene replacements. With CRISPR/Cas9 we introduced the S. cerevisiae allele of the promoter and coding region of a given gene into the endogenous location in wild-type S. paradoxus; we used the resulting strain as a background for the replacement of the S. paradoxus allele of the next gene by that of S. cerevisiae; and so on until all eight genes were swapped into one genome (Fig 1, bottom).
We first assayed growth of the eight-gene transgenic and the wild-type parental species at high temperature, with biomass accumulation (growth efficiency; the increase in optical density after 24 hours relative to that at the start) as a readout of strain performance. The results revealed an advantage of 2.07-fold attributable to the eight S. cerevisiae alleles in the S. paradoxus background at 39˚C (Fig 1, inset and S1A Table), and no such effect in 28˚C control conditions (S1 Fig). This joint phenotype recapitulated 15% of the trait divergence between wild-type S. cerevisiae and S. paradoxus at 39˚C (Fig 1, inset), with predictive power tailing off near 37˚C (S2 Fig). These data make clear that even at 39˚C, the full genetic architecture of S. cerevisiae thermotolerance must involve more loci than the eight we have manipulated herehighlighting the potential for high genetic complexity of this trait divergence between species.

Negative epistasis among thermotolerance genes
We hypothesized that S. cerevisiae thermotolerance determinants might depend on one another to confer their effects. We sought to test this at the whole-gene level, treating the allele of each thermotolerance locus (including the promoter and coding region) as a module, and investigating the interactions between them. The S. cerevisiae allele of each module, when introduced on its own into S. paradoxus, was sufficient for a <1.4-fold benefit in biomass accumulation at 39˚C, as expected (S3 Fig and [39]). We summed these measurements to yield an expected phenotype under the assumption of independent gene function, which we compared to the true measurement from the S. paradoxus strain harboring all eight thermotolerance loci from S. cerevisiae. The latter came in significantly below the estimate from the model assuming independence (Fig 2A). Thus, the combination of all eight genes was subject to negative epistasis, improving growth at high temperature to an extent less than the sum of its parts. We next turned to the strains we had made in the service of the eight-gene transgenic. These harbored S. cerevisiae modules of one thermotolerance gene in the S. paradoxus background, two genes, three genes, and so on (Fig 1, bottom). We considered this strain panel as an arbitrary trajectory through a gene-wise genetic landscape from the S. paradoxus wild-type to the eight-locus swap. Though it represented just one of tens of thousands of possible paths to the eight-fold transgenic genotype, we anticipated that it could help inform our understanding of the genetic architecture of thermotolerance.
In assays of biomass accumulation at 39˚C in each strain of the panel, we saw a generally monotonic relationship between thermotolerance and the number of S. cerevisiae gene modules introduced into S. paradoxus (Fig 1 and S1A Table). No such pattern emerged from a 28˚C control (S1 Fig and S1B Table). For a quantitative analysis, we converted the phenotypic profiles in the strain set to effect sizes, each reporting the impact of the S. cerevisiae allele of a gene on thermotolerance, in the chimeric background into which it was introduced ( Fig 2B). Inspection of this metric confirmed that each S. cerevisiae gene addition boosted the phenotype along the path to the eight-fold transgenic, in most cases significantly so (S1A Table), with one exception. The pro-thermotolerance function of S. cerevisiae SCC2 (encoding a cohesin loading factor), discernable when it was introduced on its own into S. paradoxus (S3 Fig),   Fig 1. S. cerevisiae alleles of thermotolerance loci jointly improve growth at high temperature. In each plot, the yaxis reports growth efficiency at 39˚C, the cell density after a 24-hour incubation as a difference from the starting density. In the main plot, each column reports data from a transgenic S. paradoxus strain harboring S. cerevisiae alleles of the indicated thermotolerance loci, or the wild-type S. paradoxus progenitor. At bottom, each cell reports the genotype at the indicated locus in the indicated strain. The inset shows purebred wild-types and the S. paradoxus strain harboring all eight thermotolerance loci from S. cerevisiae (8x swap); the y-axis is log-scaled. Black points report individual biological replicates and white dots report means. Boxes span the interquartile range. Whiskers are 1.5 times the interquartile range. Statistical analyses are reported in S1A Table. https://doi.org/10.1371/journal.pgen.1009793.g001 was either below our detection limit or absent altogether in combination with S. cerevisiae DYN1 and MYO1 (encoding motor proteins; Figs 1 and 2B). Despite this potential case of masking epistasis, the remaining trend for beneficial effects by S. cerevisiae alleles as we built up the eight-locus strain was highly non-random (binomial p = 0.03). In no case did we observe a defect from swapping in an S. cerevisiae allele, meaning we had no evidence for sign epistasis in this system.
We also used our multi-genic strain panel, alongside single-gene transgenics in S. paradoxus, to compare S. cerevisiae allele-replacement effects across backgrounds. In most cases, a given thermotolerance gene from S. cerevisiae had less impact in the presence of other S. cerevisiae loci than when tested on its own in S. paradoxus (Fig 2B). This finding confirmed the negative (magnitude) epistasis between the S. cerevisiae alleles of thermotolerance genes that we had inferred with the eight-gene transgenic (Fig 2A). We noted that the damping of allelic effect by genetic background was most apparent at thermotolerance genes annotated in chromosome segregation/mitosis (DYN1, MYO1, SCC2, APC1, and ESP1; Fig 2B). By contrast, introducing the S. cerevisiae allele of TAF2 or AFG2, involved in transcription and translation respectively, drove nearly the same benefit on its own in S. paradoxus and in the respective multi-swap chimera ( Fig 2B).
Together, these genetic data characterize an example path toward thermotolerance of incremental advances from S. cerevisiae gene modules-most limiting each other's effects to some extent, but without frank deleterious consequences.

Thermotolerance loci improve viability only during active growth
We next aimed to investigate cellular mechanisms of thermotolerance, using as a tool the strain with all eight of our focal genes from S. cerevisiae replaced into S. paradoxus. We focused on cell viability, as assayed by counts of colony-forming units (CFUs) from aliquots of liquid culture at 39˚C. In a first characterization of strain performance in growing cultures under this setup, S. paradoxus cells were much less viable than those of S. cerevisiae, across a range of warm temperatures (S4 Fig), as expected. We anticipated that S. cerevisiae alleles of our thermotolerance loci would rescue this phenotype, at least in part. This prediction bore out in our CFU assays from actively growing cultures: at 39˚C, the eight-gene transgenic survived 7-fold better than did its S. paradoxus progenitor ( Fig 3A). An analogous test of wild-type S. cerevisiae revealed three logs higher viability than that of S. paradoxus during active growth at 39˚C (Fig 3A).
The poor performance of strains of S. paradoxus origin in the CFU assay ( Fig 3A) was much bigger in magnitude than that seen in our measurements of biomass accumulation (Fig  1), in growing cultures. Such a discrepancy suggests that some cells with S. paradoxus genotypes managed to divide early in the heat treatment and then ultimately die, contributing at the end of the incubation to measured biomass but not viability. Qualitatively, however, both experiments led to the same conclusion: S. cerevisiae alleles of thermotolerance loci recapitulate part but not all of the advantage by S. cerevisiae relative to S. paradoxus during active growth at 39˚C. We detected no viability differences between strains during active growth at 28˚C (S5A Fig).
To gain insight into why S. paradoxus cells die at high temperature, we took account of the role in mitosis for most of our thermotolerance genes (Table 1). We hypothesized that failure of the S. paradoxus cell growth machinery was the proximal cause of death for this species at 39˚C. If so, we expected that the underlying alleles would not be a liability if cells did not enter the cell cycle in the first place. As a test of this notion, we retooled our viability assay to start by incubating a liquid culture at a permissive temperature until it reached stationary phase (when nutrients are exhausted and cell division arrests). We then switched these non-growing cultures to 39˚C, and finally took aliquots for assays of CFUs. The results revealed that, when exposed to heat as a stationary-phase culture, S. paradoxus survived nearly as well as did S. cerevisiae (Fig 3B), in contrast to the many logs of difference between the species during active growth ( Fig 3A). Likewise, transgenesis of our eight thermotolerance genes had no impact on high temperature survival in stationary phase ( Fig 3B). Viability experiments on stationaryphase cultures at 28˚C also found no difference between strains (S5B Fig).
These viability data show that the thermotolerance defect of S. paradoxus alleles from our eight-gene transgenic strain only manifests in actively growing cells, consistent with the hallmarks of cell cycle breakdown seen in microscopy assays of S. paradoxus at 39˚C [39]. Together, our results support a model in which passage through mitosis itself is lethal at high temperature for S. paradoxus, whereas cells in an arrested state are protected from damage and death. S. cerevisiae, meanwhile, grows and divides successfully at 39˚C, owing in part to its thermotolerant mitotic genes.

An evolutionary tradeoff in tolerance of extreme temperatures
Given that S. cerevisiae is unique within its clade for its ability to grow at high temperatures, we anticipated that this trait could have evolved as part of a tradeoff, and that cold tolerance would be a logical potential opposing character. Consistent with this picture, we observed generally better cold resistance across a panel of wild S. paradoxus relative to environmental isolates of S. cerevisiae (S6 Fig). No such difference is detectable at 28˚C [39]. We reasoned that thermotolerance alleles at our eight focal genes could contribute to the poor growth by S. cerevisiae in the cold. Indeed, our eight-fold transgenic strain grew significantly worse than did wild-type S. paradoxus at 4˚C (Fig 4). With respect to biomass accumulation, this strain recapitulated 15% of the divergence between the wild-type species at 4˚C-paralleling the analogous quantity at 39˚C (Fig 1), and establishing antagonistic pleiotropy by S. cerevisiae alleles at our genes of interest.
We next addressed the genetic architecture of cold sensitivity, again focused on our eight thermotolerance loci. In tests of marginal effects by each of these genes individually, the S. cerevisiae allele of DYN1 (encoding the dynein heavy chain) reduced biomass accumulation markedly at 4˚C when introduced into S. paradoxus on its own; a number of other genes had smaller effects (S7 Fig). Likewise, of our combinatorial gene replacements, most grew only slightly worse than the single-gene DYN1 swap at 4˚C (S8 Fig). This contrasted with the stairstep-like trend of increasing thermotolerance as S. cerevisiae alleles were stacked in the S. paradoxus background (Fig 1). We conclude that, with respect to our focal genes, the genetic mechanisms of heat and cold tolerance do not strictly parallel each other, and that antagonistic pleiotropy between the temperatures is most salient for a single locus, DYN1.
In viability assays during growth at 39˚C, we had found robust divergence in survival between S. cerevisiae and S. paradoxus, as well as survival effects of variation at our thermotolerance loci (Fig 3A). It remained an open question whether the genetics of cold tolerance would have effects on cell death. In viability assays we saw very little difference between S. cerevisiae, S. paradoxus, and our eight-gene transgenic in the S. paradoxus background, in terms of CFUs after incubation at 4˚C (S9 Fig). This result underscores the difference in mechanism between cryotolerance and thermotolerance in this system, and establishes that S. cerevisiae alleles (at thermotolerance loci and others) slow growth in the cold rather than killing cells outright.

Discussion
Evolution often uses variants at genes scattered throughout the genome to build an adaptive trait. The complete set of these loci, once in hand, can reconstitute the trait in an exogenous background, which is the ultimate goal for many biomedically and industrially relevant characters. Along the way, the underlying genes can shed light on the process by which a trait arose, which may involve events from millions of years ago. In this work, we have used  (8x swap). For a given strain, points on a given day report biological replicates; lines report the average fit from a logistic regression across replicates (S2 Table). (B) The y-axis reports growth rate, in units of cell density (optical density, OD) per day, from the average logistic fit of the timecourse in (A) for the indicated strain. (C) The y-axis reports, for day 10 of the timecourse in (A) for the indicated strain, growth efficiency, the cell density after a 10-day incubation at 4˚C as a difference from the starting density. In (B) and (C), points report individual biological replicates. Box and whisker format is as in thermotolerance, a putatively adaptive, fitness-relevant character [39,40], to explore genetic, biological, and evolutionary mechanisms of adaptation.
Introducing eight unlinked genes from S. cerevisiae into S. paradoxus, we reconstituted 15% of the difference in thermotolerance between the respective purebred species. Much of the architecture for this trait thus remains unmapped, likely due in part to limitations of coverage and power in our original reciprocal hemizygosity scan [39], and to its restriction to the nuclear genome. Incisive experiments have established the role of mitotype in temperature tolerance as it differs between S. cerevisiae and either S. paradoxus or farther-diverged species of the complex, S. uvarum and S. eubayanus [42][43][44][45]. These mitochondrial variants presumably make up much of the heritability not accounted for by our current gene set. And S. cerevisiae alleles of the nuclear-encoded genes that we study here could well depend on the S. cerevisiae mitochondrial genome for their full effect.
In our focus on nuclear-encoded thermotolerance factors, we found that most S. cerevisiae alleles improved high-temperature growth, when tested on their own in S. paradoxus or in a chimeric multi-swap background. In principle, these effects could be limited by partial incompatibilities in S. paradoxus. That is, a given gene could have made an even bigger contribution to thermotolerance in the presence of other functional partners from S. cerevisiae. Even in such a case, the qualitative conclusion from our own data would remain: S. cerevisiae alleles mostly help and do not hurt at high temperature, alone or combined. This profile provides an intriguing contrast to the sign epistasis often detected between adaptive amino acid changes within any one protein, as they boost fitness in some combinations and compromise it in others [46,47]. Since we have not surveyed all possible subsets of our mapped thermotolerance genes, and more remain unidentified, we cannot rule out the possibility of toxic interactions at high temperature between some S. cerevisiae versions of unlinked loci when they come together. That said, our thermotolerance data as they stand conform to the idea that combining adaptive variants across unlinked sites will be less constrained than, say, repacking a protein [19]. This model is especially compelling in the search to understand evolutionary dynamics in sexual systems. If a gene module can recombine into most backgrounds and confer a benefit, it would speed the advance toward fitness of the population as a whole.
We found that the magnitude of the phenotypic effect of our focal genes depended on genetic background. For our analyses of epistasis, we focused on thermotolerance, under the assumption that this trait or its correlates have mattered for the fitness of S. cerevisiae in the environment, and that genetic interactions could have been at play during its evolution. We found ample evidence for negative epistasis, with the S. cerevisiae alleles of some of our unlinked loci obscuring the thermotolerance effects of others, in replacements into S. paradoxus. Negative epistasis has been a linchpin of classical genetic studies placing unlinked lossof-function mutants in pathways [48,49], and their genomic equivalent, reverse-genetic double-mutant screens [50][51][52][53]. Negative epistasis also features in the polygenic architectures of putatively neutral and disease traits that vary in populations [54][55][56]. Our thermotolerance system complements these scenarios, in that we focus on alleles that act as gains of function, at least at high temperature [39]. Under the latter condition, we detected the most marked negative epistasis among thermotolerance genes with annotations in chromosome segregation and cell division. We can speculate that the S. cerevisiae allele at any such locus, as it rescues one aspect of mitosis at 39˚C, also pulls up the function of other parts of the mitotic machinery to some extent, e.g. by stabilizing heteromeric protein complexes. S. cerevisiae alleles of additional genes, introduced into such a background, would make less dramatic improvements to the phenotype than would be expected from their respective single-gene allele-swap strains.
Importantly, a sizeable literature has described negative epistasis between unlinked adaptive loci in laboratory evolution experiments, especially in microbes [7,[21][22][23][24][25][26][27][28][29]57], where in some studies, suites of unlinked variants have been validated in terms of effects on fitness [30,31]. A chief emphasis here has been on evolutionary dynamics, in that negative epistatic interactions limit the availability of mutational steps with big fitness effects that would otherwise speed adaptation [46,47]. This principle plays out in part through diminishing-returns epistasis [7,58], the difficulty of improving the fitness of somewhat-fit backgrounds, which can be explained by heightened susceptibility of fast-growing strains to pleiotropic effects of new mutations [59]. Though the latter has been a landmark of the recent literature, we do not consider it an immediately relevant model for our own data, because we work with backgrounds that perform very poorly at 39˚C. In more advanced stages of an allele-swap progression, which would reconstitute more of the yeast thermotolerance phenotype, we would expect genes to interact under the diminishing-returns mechanism.
Along with our insights into the genetics of thermotolerance, we discovered a cold resistance defect in S. cerevisiae relative to S. paradoxus, complementing previous temperature profiling using other assays across the genus [36,38]. We found that pro-thermotolerance alleles at our focal genes contribute to this character, most notably DYN1, which elicited the most dramatic drop in cold resistance when tested on its own in S. paradoxus, and overshadowed the effects of other loci in multi-gene swap backgrounds. This echoes the known breakdown of S. cerevisiae DYN1 protein function below 8˚C, as distinguished from the mammalian ortholog [60]; additional deleterious effects could also manifest in conditions other than those we study here. The emerging picture from our work and that of others is that the evolution of thermotolerance in S. cerevisiae required, at least in part, the sacrifice of the ability to deal with cold. The tradeoff would have involved DYN1 and its modifiers as well as mitochondrial genes, whose S. cerevisiae alleles are known to boost heat resistance and compromise cold growth [42][43][44][45]. And yet such a tradeoff need not be the whole evolutionary story. Some of the architecture of S. cerevisiae cold sensitivity could have arisen under other selective forces altogether. We can envision, for example, some degree of relaxed selection on cold tolerance in the S. cerevisiae ancestor, later in its evolution after commitment to a niche as a high-temperature specialist. Under one compelling model [61], supported by ethanol growth data [62,63], S. cerevisiae's more particular niche would have been as an avid glucose fermenter, with the resulting heat and ethanol both helping to kill off microbial competitors. Whether this or some other scenario proves to be the right description of S. cerevisiae ecology, our results attest to the complexity of the challenge faced by evolution, as it built the traits that define this species.

Strain construction
Strains used in this work are listed in S3 Table. To study combinations of S. cerevisiae alleles in an S. paradoxus background, we used as the latter the wild-type homozygous diploid S. paradoxus Z1, originally isolated from tree bark in England [41]. We used our CRISPR/Cas9 method [39], essentially as described but with slight modifications detailed below, to replace both copies of a given thermotolerance locus, at the endogenous location, with the allele from S. cerevisiae DBVPG1373, a soil isolate from the Netherlands [41]. To build the eight-gene transgenic, we started with the single-gene transgenic in the Z1 background harboring the S. cerevisiae allele of DYN1 from [39]. We introduced S. cerevisiae alleles at additional genes in an iterative series of transformations; after each, we cultured an isogenic stock from a single colony for Sanger sequence confirmation and storage, and, where appropriate, we used this stock as input into the next transformation.
A given transformation involved donor DNA and constructs encoding guide RNAs for Cas9 to target replacement of the S. paradoxus allele with that from S. cerevisiae at either one or two loci, which Sanger sequencing then verified to be successful at one or both. For each locus we used designs of guide RNAs from [39] that targeted, for double-strand breaks, the endogenous Z1 allele by Cas9; one guide targeted a site~1000bp upstream of the coding start of the gene of interest (or the 3' end of the closest upstream gene) in the Z1 genome and the other guide targeted a site near the coding stop. Cas9 editing proceeded as described [39]. Briefly, for each transformation step of strain construction, one or two guide RNA pairs, targeting one or two loci respectively, were cloned into plasmid pBC712, which also encodes Streptococcus pyogenes Cas9. Next, DNA to serve as a repair template was generated for each relevant locus, via PCR from S. cerevisiae DBVPG1373, with 90bp primers that contained 70bp of sequence homologous to S. paradoxus Z1 on each side of the amplified DNA product. Finally, plasmid and repair template were transformed into the Z1 descendant as described [39] at a ratio of 0.3 to 3.5 based on the length of the donor DNA, equivalent to 10 12 dsDNA molecules for 10 μg of plasmid. Putative transformants were purified and sequence-verified.

Growth assays
Measurements of biomass accumulation (growth efficiency) at 39˚C in Fig 1 were done essentially as described [39] with modifications as follows. For a given day's worth of experiments, wild-type S. paradoxus Z1 and one or more other strains of interest were streaked onto a yeast peptone dextrose (YPD) plate from a -80˚C freezer stock and incubated at 28˚C for 2 days. 2-8 colonies of a given strain were each inoculated separately in 5mL of liquid YPD and grown for 24 hours at 28˚C with shaking at 200rpm to saturation; we refer to the cultures at this stage as pre-cultures. Each such replicate pre-culture was back-diluted into 10mL of YPD to achieve an OD 600 of 0.05; incubated at 28˚C until it reached an OD 600 of 0.4-0.8; back-diluted again to achieve an OD 600 of 0.1; and then incubated for 24 hours at 39˚C. We tabulated the difference in OD 600 between the final and initial timepoints across this 24-hour incubation for each culture. This procedure, from streaking on solid medium through inoculation, heat treatment, and biomass measurement, was repeated at least four times for each strain in the analysis of Fig 1. The resulting vector of biomass measurements across all replicates from all days for each strain was compared to that for each other strain with a one-tailed Wilcoxon test in S1 Table. Multiple testing was corrected for using the Benjamini-Hochberg method.
Measurements of biomass accumulation at 28˚C in S1 Fig were done as described [39]. Strains were streaked on solid plates and one colony per strain was pre-cultured in liquid at 28˚C as above. Each such saturated pre-culture was back-diluted to achieve an OD 600 of 0.05 and grown for an additional 5.5 hours at 28˚C until it reached logarithmic phase. We transferred cells from each such pre-culture, and YPD, to five replicate wells of a 96-well plate, with volumes sufficient to yield a total volume of 150 μL per well at an OD 600 of 0.02. The plate was covered with a gas-permeable membrane (Sigma) and incubated with orbital shaking in an M200 plate reader (Tecan, Inc.) at 28˚C for 24 hours. We tabulated the difference in OD 600 between the final and initial timepoints across this 24-hour incubation for each replicate culture. The vector of these replicate measurements for each strain was compared to that from S. paradoxus with a two-tailed Wilcoxon test. Multiple testing was corrected for using the Benjamini-Hochberg method.
For measurements of biomass accumulation at 37˚C-39˚C in S2 Fig, strains were streaked, and three colonies of a given strain were pre-cultured in liquid as above. Each such replicate liquid culture was back-diluted into 10mL of YPD to achieve an OD 600 of 0.05; incubated at 28˚C until it reached an OD 600 of 0.4-0.8; back-diluted again to OD 600 of to achieve an OD 600 of 0.1; and then incubated for 24 hours at the temperature of interest. We tabulated the difference in OD 600 between the final and initial timepoints across this 24-hour incubation for each replicate culture. The vector of these replicate measurements for each strain at a given temperature was compared to that from S. paradoxus with a one-tailed Wilcoxon test. In S2 Fig, lines are the result of a polynomial regression on the points, created using Seaborn's regplot in Python 3.7.
For measurements of growth at 4˚C in Figs 4 and S6-S8, for a given day's worth of experiments, strains were streaked on solid plates and three to eight colonies per genotype, respectively, were pre-cultured in liquid, each as an independent biological replicate, as above. After the second back-dilution, each liquid culture was incubated for 11 days at 4˚C in a rotating shaker at maximum speed in a cold room. The OD 600 was measured on days 0, 4, 5, 7, 8, 9, 10, and 11, and also on day 6 for S6-S8 Figs. To measure biomass accumulation (growth efficiency) we tabulated the difference in OD 600 between the final and initial timepoints across this 11-day incubation for each replicate culture. Two such days' worth of experiments were carried out for each strain in S7 and S8 Figs, and one in Figs 4 and S6. We collated the measurements from all replicate culture measurements across all days for a given strain and compared each transgenic against S. paradoxus with a one-tailed Wilcoxon test. Separately, we fit a logistic curve to the timecourse measurements for each replicate using Scipy's curve_fit function as a part of Scipy's optimize package (Python 3.7). Bounds for the parameters of the logistic equation (the carrying capacity, growth rate, and time to half-maximal growth) were constrained to the range -1.0 to 10.0, and the Trust Region Reflective algorithm was used to find the best fit. We collated the growth rate estimates from all replicate culture measurements across all days for a given strain and compared each transgenic against S. paradoxus with a one-tailed Wilcoxon test. Multiple testing was corrected for using the Benjamini-Hochberg method.

Viability assays
For the survey of viability phenotypes at high temperatures across environmental isolates in S4 Fig, strains were streaked out and four colonies of each were pre-cultured in liquid as for 39˚C growth above, except that the initial pre-culture to achieve saturation lasted 48 hours. Each pre-culture was back-diluted into 10 mL of YPD to reach an OD 600 of 0.05 and then cultured for 24 hours at the temperature of interest (35˚C-38˚C). The OD 600 at the end of this timecourse was measured for each such replicate culture. Then to measure viability for each, we diluted aliquots from the culture in a 1:10 series and spotted 3μL of each dilution for growth on a solid YPD plate. After incubation at 28˚C for two days, we used the dilution corresponding to the most dense spot that was not a lawn for the final report of viability: we counted the number of colonies in each of the two technical replicate spots, formulated the number of colony-forming units per mL of undiluted culture (CFU/mL), and divided this ratio by the OD 600 we had measured at the end of the liquid timecourse, to account for differences in the number of dead cells that contribute to the latter. At a given temperature, the vector of viability measurements across all replicate liquid cultures for all S. cerevisiae strains was compared to that for all S. paradoxus strains with a one-tailed Wilcoxon test.
For the comparison of viability between strains during log-phase growth at 39˚C in Fig 3A, strains were streaked out and three colonies of each were pre-cultured at 28˚C, back-diluted, and cultured at 39˚C, as for the 39˚C growth assays above. After 24 hours of incubation at 39˚C, for each such replicate culture, spotting of dilutions and colony counting to yield CFU/mL/OD 600 for each replicate liquid culture was done as above, except that we used two technical replicate spotting assay replicates for each culture, taking the average across them as the final report of viability. The vector of these viability measurements across replicates for a given strain was compared to that from S. paradoxus with a one-tailed Wilcoxon test.
For the comparison of viability between strains in stationary phase at 39˚C in Fig 3B, strains were streaked out and three colonies of each were pre-cultured at 28˚C as for the 39˚C growth assays above, except that that the initial pre-culture to achieve saturation lasted 72 hours. Each such replicate culture was then incubated (without back-dilution) at 39˚C for 24 hours, after which spotting, colony counting, and statistical testing were as for Fig 3A. For the comparison of viability between strains during log-phase growth at 28˚C in S5A Fig, strains were streaked out and four colonies of each were pre-cultured at 28˚C and backdiluted as for the 39˚C growth assays above; each back-diluted replicate culture was incubated at 28˚C for 24 hours, after which spotting and colony counting was as above. For the comparison of viability between strains in stationary phase at 28˚C in S5B Fig, strains were streaked out as above, and three colonies of each were inoculated into liquid YPD at 28˚C and incubated for 96 hours, after which spotting, colony counting, and statistical testing were as for Fig 3A, except that two-sided Wilcoxon tests were performed.
For the comparison of viability between environmental strains during log-phase growth at 4˚C (S6 Fig), aliquots from cultures set up for cold growth assays (see above) were taken at day 8 of the cold timecourse for spotting, colony counting, and statistical testing as for Fig 3A.

Epistasis analysis
For Fig 2A, we calculated the growth phenotype at 39˚C of the strain harboring all eight S. cerevisiae loci in the S. paradoxus background as expected under a model of independent locus effects as follows. We first tabulated the mean growth efficiency at 39˚C of each isogenic strain in turn with just one gene swapped in from S. cerevisiae and the analogous mean for S. paradoxus, and took the difference between them, representing the mean effect of the respective swap; we then summed the latter effect values across all eight loci. Error was estimated by bootstrapping as follows. For each locus, we generated a random sample of the replicate measurements of the growth efficiency of the respective swap strain at 39˚C with replacement, and took the mean; we calculated an analogous mean from a random sample of replicates of the S. paradoxus wild-type; and we took the difference between these means, representing one bootstrapped estimate of the effect of the swap. We then took the sum of such effects across all loci, representing one bootstrap's worth of the estimate of the eight-gene transgenic's phenotype under the additive model. We repeated this procedure 10,000 times to set up a distribution of the estimated sums in Fig 2A, showing 100 randomly chosen points to minimize figure complexity.
For a given panel of Fig 2B, the left-hand white point on the plot reports the mean effect of the S. cerevisiae allele of the respective gene when swapped alone into S. paradoxus. For this we tabulated the mean growth efficiency at 39˚C of this single-gene transgenic across all replicates and the analogous mean across all replicates of S. paradoxus, and took the difference between them. For error estimates, we generated a random sample of the replicate measurements of the growth efficiency of the respective swap strain at 39˚C with replacement, and took the mean; we calculated an analogous mean from a random sample of replicates of the S. paradoxus wildtype; and we took the difference between these means, representing one bootstrapped estimate of the effect of the swap. We repeated this procedure 10,000 times to set up a distribution of the estimated effect values in Fig 2B, showing 100 randomly chosen points to minimize figure complexity. The right-hand white point on the plot reports the mean effect of the S. cerevisiae allele of the respective gene when swapped into a multi-genic strain of the S. paradoxus background in the series culminating in the eight-fold transgenic, in the order of Fig 1. Call the strain before and after the replacement of the gene of interest X-1 and X, respectively; we tabulated the mean growth efficiency at 39˚C across all replicates of strain X and the analogous quantity for strain X-1, and took the difference between them. Error was estimated by bootstrapping as above.
Supporting information S1 Table. Table. Strains used in this study. A. Wild-type diploid strains, including those used as parents of allele-replacement transgenesis; SGRP, the Saccharomyces Genome Resequencing Project, version 2. B. Purebred allele replacement strains in S. paradoxus Z1 diploid homozygote backgrounds. In genotype notes, e.g., in an S. paradoxus background, ΔYFG(-X to +Y):: scYFG(-Z to +W) indicates that in S. paradoxus Z1, bases -X to +Y from both homologs of gene YFG have been removed and replaced by bases -Z to +W of the allele of YFG from the indicated S. cerevisiae strain. Positive coordinates count in the 5' to 3' direction from the start codon (+1 corresponds to the A in the ATG), and negative coordinates count in the 3' to 5' direction from the start codon (-1 corresponds to the base directly 5' of the ATG). In cases where the replacement extended into a region of 100% conservation between species, the position of the last divergent nucleotide is shown. (XLSX) S1 Fig. Allelic variation at thermotolerance loci has little growth impact at 28˚C. Data and symbols are as in Fig 1 of the main text except that growth was measured at 28˚C. Statistical analyses are reported in S1B Table. (TIF) S2 Fig. S. cerevisiae alleles of thermotolerance loci jointly improve growth at temperatures from 37˚C-39˚C. In a given panel, each trace reports growth efficiency, the cell density after a 24-hour incubation at the indicated temperature as a difference from the starting density, of the wild-type of the indicated species or the S. paradoxus strain harboring eight thermotolerance loci from S. cerevisiae (8x swap). Lines are the result of a polynomial regression on the points (S2 Table). Each cell reports viability after heat treatment of the indicated strain and species: the number of colonies formed on solid medium from 1 mL of heat-treated liquid culture in logarithmic growth, normalized by the turbidity of the latter. S. paradoxus Z1, N17, and A12 are isolates from UK, Russia, and Quebec, respectively; S. cerevisiae strains DBVPG1373, DBVPG1788, YPS128 are from the Netherlands, Finland, and Pennsylvania, respectively. Viability was different between species at Wilcoxon p < 0.00002 for all temperatures. Each trace reports a timecourse of growth at 4˚C of the wild-type of the indicated strain. For a given strain, points on a given day report biological replicates; lines report the average fit from a logistic regression across replicates (S2 Table). YPS128, a North American S. cerevisiae known to have recently acquired freeze-thaw resistance as a derived character distinct from the ancestral program [64], is shown in faint blue. (B) The y-axis reports growth rate, in units of cell density (optical density, OD) per day, from the average logistic fit of the timecourse in (A) for the indicated strain. (C) The y-axis reports, for day 10 of the timecourse in (A) for the indicated strain, growth efficiency, the cell density after a 10-day incubation at 4˚C as a difference from the starting density. In (B) and (C), Black points report individual biological replicates. White dots report means. Boxes span the interquartile range; whiskers are 1.5 times the interquartile range, and do not report outliers. Comparisons of growth rate and efficiency at 4˚C between S. paradoxus strains and S. cerevisiae strains yielded Wilcoxon p � 0.00003 and p � 0.002, respectively.  Table). � , corrected one-sided Wilcoxon p � 0.05. (TIF)  Table). � and �� , corrected onesided Wilcoxon p � 0.05 and 0.01 respectively. (TIF)