Incompatibilities Involving Yeast Mismatch Repair Genes: A Role for Genetic Modifiers and Implications for Disease Penetrance and Variation in Genomic Mutation Rates

Genetic background effects underlie the penetrance of most genetically determined phenotypes, including human diseases. To explore how such effects can modify a mutant phenotype in a genetically tractable system, we examined an incompatibility involving the MLH1 and PMS1 mismatch repair genes using a large population sample of geographically and ecologically diverse Saccharomyces cerevisiae strains. The mismatch repair incompatibility segregates into naturally occurring yeast strains, with no strain bearing the deleterious combination. In assays measuring the mutator phenotype conferred by different combinations of MLH1 and PMS1 from these strains, we observed a mutator phenotype only in combinations predicted to be incompatible. Surprisingly, intragenic modifiers could be mapped that specifically altered the strength of the incompatibility over a 20-fold range. Together, these observations provide a powerful model in which to understand the basis of disease penetrance and how such genetic variation, created through mating, could result in new mutations that could be the raw material of adaptive evolution in yeast populations.


Introduction
The scientific literature contains numerous examples in which alleles confer more severe phenotypes in one genetic background relative to another (e.g. [1,2]). Such effects are thought to be due to DNA sequence differences at multiple genetic loci that lead to molecular incompatibilities between gene products or between gene products and cis-acting regulatory sequences that function in specific pathways. Identifying such incompatibilities in a genetically tractable system is of great interest because they provide testable models to study disease penetrance, defined as the proportion of individuals that carry disease-linked alleles that also display the disease phenotype (e.g. [3,4,5]).
We have been studying molecular incompatibilities that relate to DNA mismatch repair (MMR). MMR improves the fidelity of DNA replication ,1000 fold by repairing mismatches that arise during DNA replication. In addition, factors that act in MMR play an important role in preventing interspecific gene transfer by rejecting recombination intermediates containing multiple DNA mismatches [6,7,8]. In E. coli, DNA mispairs are recognized by a homodimer of the MutS protein, which then recruits the MutL homodimer. MutL coordinates recognition of a mispair with downstream excision repair activities. In yeast and other eukaryotes, multiple MutS and MutL homologues (MSH and MLH) have evolved to function as heterodimers with specialized functions in DNA repair and recombination. The primary MLH heterodimer in yeast MMR is MLH1-PMS1 (reviewed in [8]). In humans, mutations in MSH and MLH genes have been implicated in hereditary non-polyposis colorectal cancer (HNPCC), a dominantly inherited syndrome. Recent studies in humans and yeast suggest that genetic background can play an important role in the level of pathogenicity conferred by MMR alleles [9,10].
Previously our laboratories (Heck et al. [11]) examined S. cerevisiae progeny created by mating SK1 and S288c strains that are ,0.7% sequence divergent. We identified single amino acid polymorphisms in MLH1 and PMS1 that, in combination, lead to an elevation in mutation rate and a generalized reduction in longterm fitness. These reductions in fitness are analogous to those seen for interactions thought to be the basis for hybrid incompatibility between incipient or established species [12][13][14][15][16]. Heck et al. [11] proposed a negative-epistasis model in which the S288c and SK1 strains diverged from an ancestral population. When the strains are crossed, an incompatible combination is generated at a 25% frequency.
Several critical questions emerged from this analysis with respect to the negative epistatic MLH1-PMS1 interactions resulting from hybrid incompatibility. First, is there evidence of genetic divergence at the PMS1 and MLH1 loci, as predicted if the incompatible alleles arose in isolation? Second, do the amino acid polymorphisms isolated by Heck et al. [11] confer incompatibility regardless of allelic background? Third, do intragenic modifiers of the incompatibility segregate in natural populations? Because each of the above questions is best answered using a large sample of PMS1 and MLH1 alleles, we obtained the DNA sequence of the MLH1 and PMS1 genes from 68 geographically and ecologically diverse strains of S. cerevisiae. These strains were genotyped for the single amino acid polymorphisms in MLH1 and PMS1 that conferred a mutator phenotype in S288c-SK1 hybrids. We found that none of the strains sampled contained the incompatible combination. Furthermore, a mutator phenotype was only observed in MLH1-PMS1 combinations that contained the single polymorphisms underlying the S288c-SK1 incompatibility. To our surprise the mutator phenotype observed in this set of combinations varied over a 20-fold range. In contrast, all possible compatible combinations showed mutation rates similar to those seen in natural populations. We mapped the variation/genetic modifiers seen in the incompatible combinations to specific polymorphisms within MLH1 and PMS1. This work demonstrates a mechanism by which yeast can rapidly and reversibly elevate the genomic mutation rate in natural populations. It also illustrates a model for disease penetrance that may explain the basis of some HNPCC-like cancers displaying atypical inheritance [9].

Results
The Model for MLH1-PMS1 Hybrid Incompatibility Holds True Regardless of Allelic Origin Heck et al. [11] identified one site each in MLH1 (aa 761, Asp in S288c, Gly in SK1) and PMS1 (aa 818/822, Arg in S288c, Lys in SK1) that accounted for the MMR defects observed in progeny derived from crossing S288c and SK1 strains. This analysis indicated that only the MLH1-Asp, PMS1-Lys combination conferred the mutator phenotype. Using a limited sampling of Saccharomyces cerevisiae strains and closely-related Saccharomyces species, Heck et al. [11] inferred that MLH1-Gly, PMS1-Arg was the predicted ancestral combination. None of the 14 strains in this small sample contained the MLH1-Asp, PMS1-Lys combination predicted to be a mutator, despite DNA sequence analysis suggesting that recombination had occurred between strains that could generate this combination. An explanation for this observation, which was supported by spore viability after mutation accumulation in bottleneck growth studies, is that the MLH1-Asp, PMS1-Lys combination carries a long-term fitness cost.
Here we examined 68 yeast strains to measure the frequency of the incompatibility alleles ( Figure 1, Table 1). The selected strains included natural, clinical, and domesticated stocks from a variety of collections, including the strains compiled by Justin Fay [17], strains currently being sequenced by the Saccharomyces Genome Resequencing Project at the Sanger Institute (http://www.sanger. ac.uk/Teams/Team71/durbin/sgrp/), the recently sequenced vineyard isolate RM11-1a (http://www.broad.mit.edu/annotation/ fungi/comp_yeasts/), and the clinical strain YJM789 [18]. For each of the strains listed in Table 1, MLH1 and PMS1 were amplified and the sequence surrounding the S288c-SK1 incompatibility defined above was determined (except YJM789, sequence solely from public database). Additional sequencing or publicly available sequence was used to determine the additional SNPs located in the MLH1 and PMS1 open reading frames as well as downstream sequence (Table S1 and  Table S2). Neighbor-joining trees were created for MLH1 and PMS1 using the compiled sequence data for Saccharomyces cerevisiae and closely related Saccharomyces species ( Figure 1A and B). A total of 72 polymorphic sites were found in and around MLH1, with an average pairwise nucleotide diversity (p) of 0.002. 134 polymorphic sites were identified in and around PMS1 (p = 0.0039). Sequencing of MLH1 and PMS1 revealed a pattern similar to that seen by Fay and Benavides [19]; the domesticated strains tended to group, consistent with the idea that they are derived from a subset of wild-type yeast.
The neighbor joining tree analysis allowed us to examine the phylogenetic relationship between strains in each of the three categories. The eleven strains containing MLH1-Asp 761 (S288c class) are highlighted in Figure 1A. Seven of these fall into the same clade, suggesting a single, recent origin of this allele, although the presence of strains carrying MLH1-Gly 761 in this clade, and of MLH1-Asp 761 in other haplotypes, are consistent with the explanation that some recombination has occurred. The strains in this class display world-wide geographic origins, including Europe, Africa, Asia and North America (Table 1). In addition, these strains include clinical, wild, and fermenting yeasts, and thus do not share an obvious functional origin. Thus while we observed molecular evidence for a common origin of the MLH1-Asp 761 site, we were unable to conclude definitively that this allele arose in geographic isolation from the PMS1 Lys 818/822 mutation with which it is incompatible. Recent studies suggest that evidence for geographic isolation will be difficult to obtain in S. cerevisiae strains due to their wide usage and the ease of transfer in a globalized society [17,20]. It is difficult to draw conclusions concerning the origin of the PMS1-Lys 818 allele, since only four such alleles were identified ( Figure 1B). Finally, it is worth noting that there is a lack of concordance between the MLH1 and PMS1 trees with respect to individual isolates. This likely reflects the presence of outcrossing between strains in natural populations of yeast [11].
Heck et al. [11] predicted that no strains would be found that contain the MLH1-Asp, PMS1-Lys combination due to the fitness cost resulting from an elevated mutation rate. They also predicted that MLH1-Asp, PMS1-Lys combinations created in the labora-

Author Summary
For many common afflictions, it is difficult to map diseaseassociated loci because multiple loci are involved, with some loci playing greater roles than others. To explore how complex interactions can contribute to disease, we examined an incompatibility involving the MLH1 and PMS1 DNA mismatch repair proteins in baker's yeast. In our system, an incompatibility is defined as a defect occurring when specific combinations of MLH1 and PMS1 proteins obtained from different baker's yeast strains are tested for function. We identified amino acid differences at only one site in each protein that contributed to this incompatibility. We also showed that amino acid differences that could cause such an incompatibility are found in strains collected from across the globe. No strain contained the incompatible MLH1-PMS1 combination, indicating that it was likely to be deleterious. When such a combination was created in the laboratory, we could detect a wide range of defects that were under the control of genetic modifiers. These observations provide a powerful model in which to understand the basis of disease penetrance and how segregation of defects in mismatch repair may allow for rapid yet reversible changes in genomic mutation rates that can help yeast adapt to changing or novel environments. tory would show a mutator phenotype regardless of allele origin. We tested the latter prediction by cloning MLH1 and PMS1 from a total of eleven strains into plasmids expressing MLH1 and PMS1 through S288c versions of their respective promoters. Four of these strains were from the S288c class (S288c, SB, UCD820, and Y6), three were from SK1 (SK1, YJM320, Y1), and four from the ancestral class (Y4, YPS163, YPS1000, YPS1009). These alleles were introduced as combinations predicted to be compatible and incompatible into the S288c strain EAY1365. The mutator phenotype of the resulting strains was measured in the highly sensitive lys2-A14 mutator assay, which displays a roughly 10,000fold range in mutation rate (mutations per cell per generation) between wild-type and MMR defective strains ( [21]; Table 2, Figure 2B-E; Materials and Methods). All of the natural isolate combinations, in which both the MLH1 and PMS1 alleles originated from the same strain background, displayed mutation rates similar to those reported previously for the wild-type SK1 and S288c strains (,1-to 3-fold relative to S288c). The one exception involved a combination involving one of the two Y1 MLH1 alleles; this strain displayed a partial defect in MMR (Table 2). This partial defect in MMR is likely recessive and may not have an effect because Y1 appears to be an obligate diploid; we were unable to sporulate this strain under a variety of conditions.
The median relative mutation rates for the intra-strain, intragroup, and inter-group compatible combinations were indistinguishable (1.1, 1.2, 1.1 fold above the S288c mutation rate, respectively; Figure 2). Only strains bearing the MLH1-Asp, PMS1-Lys incompatible combination displayed a mutator phenotype that was significantly different from the natural isolate class ( Figure 2E compared to 2B, C, and D). The median mutation rate for the predicted incompatible class was 88, with a range from 24to 409-fold above the S288c mutation rate. A histogram plot of the incompatible combinations suggests a continuum of mutator phenotypes ( Figure 3).

Specific Amino Acid Polymorphisms in MLH1 and PMS1 Act as Intragenic Modifiers of the Incompatible Phenotype
We observed a roughly 20-fold range in mutation rate for the predicted incompatible combinations ( Table 2). Because all of the MLH1 and PMS1 genes were expressed through S288c promoters in the same strain background, we suspected that SNPs within the open reading frames of MLH1 and PMS1 were responsible for modifying the incompatible phenotype. As shown previously, the S288c MLH1, SK1 PMS1 combination conferred a rate that was ,100-fold (116-fold in this study) higher than the S288c MLH1, S288c PMS1 rate [11]. In contrast, S288c MLH1, Y1 PMS1-allele A and SB MLH1, SK1 PMS1 combinations displayed 409-and 39fold higher mutation rates, respectively. DNA sequence analysis allowed us to map the amino acid polymorphisms responsible for the different rates ( Figure 4). SK1 PMS1 and Y1 PMS1-allele A differ by only one SNP, a non-synonymous mutation causing the F165C change. Therefore this change is responsible for the difference in mutation rate. This change is located near the conserved ATP binding domain motifs within the N-terminus of MLH1 [22]. In contrast, four SNPs were found between S288c MLH1 and SB MLH1, one synonymous and three non-synonymous. We tested each non-synonymous SNP in S288c MLH1 paired with SK1 PMS1. In only one case, S288c MLH1-L271P, SK1 PMS1, did we see a decrease in the mutation rate (27fold) that was similar to that seen in the SB MLH1, SK1 PMS1 combination. This polymorphism is located near the putative DNA binding region of MLH1 [23]. It is important to note that the SB MLH1, SB PMS1 combination showed a mutation rate that was lower than the S288c MLH1, S288c PMS1 combination. However the fact that the SB modifier only affected the incompatible combination ( Figure 2, Table 2) indicates that it specifically affected the penetrance of the mutator phenotype. This work provides a unique example in which both an incompatibility and modifiers of the incompatibility have been directly mapped.

Discussion
In Heck et al. [11] we developed a model to explain the incompatibility observed for MMR genes in crosses between the S. cerevisiae S288c and SK1 strains. In this model, S288c and SK1 S. cerevisiae strains diverged from an ancestral population, with an incompatibility occurring in 25% of the progeny obtained in crosses between the strains (Figure 2). To rigorously test this model we genotyped 68 Saccharomyces cerevisiae strains and found that while the vast majority belonged to the ancestral class (79.4% allele frequency), significant numbers of strains belonged to the S288c and SK1 classes (16.2 and 4.4% allele frequency, respectively). No strains were found to contain the incompatible MLH1-D761, PMS1-K818 combination that was shown previously to confer a long-term fitness cost. Consistent with the model, all MLH1-PMS1 combinations predicted to be compatible showed mutation rates similar to those seen in natural isolates. In all cases the predicted incompatible combinations displayed mutator phenotypes that were significantly higher than compatible combinations. The surprising result was that the incompatible combinations displayed mutator phenotypes that varied over a 20-fold range and could be mapped to intragenic modifiers. As described below, these results have implications for disease penetrance and the modulation of genomic mutation rates and the potential for adaptive evolution.

Implications for Disease Penetrance
Over 500 mutations and polymorphic variations in the MMR pathway have been linked to HNPCC [24,25]. These mutations are found in all of the major components of the pathway including MSH2, MSH6, MLH3, MLH1, and PMS2 (PMS2 is the human homolog of yeast PMS1) with the majority of the mutations identified in MSH2 and MLH1. Many of the identified missense mutations do not show a clear link to cancer predisposition, and for roughly half of the kindreds displaying the Amsterdam criteria for HNPCC, mutation in a MMR gene has not been identified [24,25]. In addition, many cases of colorectal cancer resembling HNPCC have been identified that display a non-Mendelian pattern of inheritance [9,26].
Environmental and background effects have been shown to affect the penetrance of many cancers (e.g. [27,28]). For MMR, Tanyi et al. [29] reported that the association of a non-pathogenic hMSH2 polymorphism with an HNPCC-associated hMSH2 mutation lowered the age of onset for the disease. Kariola et al. [9] identified individuals with colon cancer that carried two inherited polymorphisms, one in msh2 and one in msh6 that were strains. The nucleotide site responsible for the PMS1 polymorphism at amino acid 822 was excluded from the phylogenetic reconstruction. The strains that are shaded in green all contain the PMS1-K822 polymorphism originally identified in the SK1 strain. The low frequency of this allele makes inferences regarding its origin difficult. See Materials and Methods for details. The blue/green color scheme is maintained in Figure 2, Table S1, and  Table S2 inherited from separate parents. Each appeared non-pathogenic, but the inheritance of both increased cancer susceptibility. Finally, Wanat et al. [10] showed in yeast that strain background effects can alter the penetrance of HNPCC mlh1 alleles that contain missense mutations. In this study we examined the effects of intragenic modifiers on the mutation rate of the MLH1-PMS1 hybrid incompatibility. We saw an ,20-fold range in mutation rate among the different combinations of MLH1-PMS1 that contained the hybrid D-K incompatibility ( Table 2 Figures 2E, 3). This variation can be explained by specific polymorphisms within MLH1 and PMS1 that modify the hybrid incompatibility. We found that MLH1-L271P decreased the mutation rate of the S288c MLH1, SK1 PMS1 combination by four-fold and PMS1-F165C increased the mutation rate by a similar level. However, it is clear from the continuous variation observed for the mutator phenotype ( Figure 3) that additional intra-and intergenic polymorphisms are likely to modify the MLH1-PMS1 incompatibility. While our work illustrates a relatively simple example of how background modifiers can affect disease penetrance, we believe that this will be of great interest to human geneticists who are interested in mapping modifiers that affect human disease. To our knowledge, this is the first example in which both an incompatibility and modifiers of the incompatibility have been mapped to specific amino acid variants.

MMR and the Rapid Modulation of Genomic Mutation Rates
We were intrigued by the continuum of mutator phenotypes observed in the incompatible combination set (Figure 3). Studies in a variety of organisms and fitness studies indicated that genomic mutation rates are somewhat similar and that deviation from wildtype rates would be selected against [11,30,31]. In our system we observed evidence of recombination between yeast strains that would yield progeny containing the incompatible MMR combination [11]. Thus the ability to rapidly create a wide range of mutator phenotypes upon mating might be comparable to that seen in natural and laboratory populations of bacteria.
Approximately 10% of natural isolates of E. coli show a mutator phenotype, with about 1-3% of these strains displaying such phenotypes due to defects in the MMR pathway [32,33]. This frequency is much higher than the rate predicted for mutators arising in a population without selection [34]. The fact that an increased mutation rate could be deleterious to the cell, but observable at high frequency within natural populations, suggests that mutators may contribute to the adaptive evolution of a population [35][36][37][38][39][40]. When adapting to a new environment, mutators may have an advantage due to their increased probability of acquiring the first adaptive/beneficial mutation within a population. Once adapted to an environment, however, the accumulation of deleterious mutations will outweigh the advantages of the beneficial mutation. In this model, loss or reduction of MMR functions can result in a burst of divergence, and reacquisition of fully functional MMR genes may restore fitness and sustain that divergence, owing to their role in suppressing recombination between diverged sequences. Phylogenetic studies of E. coli strains suggest that mutators can reacquire MMR function through horizontal gene transfer and that loss and reacquisition are selected alternately, presumably for adaptation to new environments [35].
Based on the above model and evidence for recombination between yeasts that would generate an incompatible combination,  Figure 2) with respect to MLH1 761 (G or D) and PMS1 818 (R or K). The S288c-like genotype (MLH1-D, PMS1-R) is indicated as C, the SK1-like genotype (MLH1-G, PMS1-K) as K, and the predicted ancestral genotype (MLH1-G, PMS1-R) as A. doi:10.1371/journal.pgen.1000103.t001  we entertain the possibility that the mating and segregation of various MLH1 and PMS1 alleles in yeast could result in the loss and reacquisition of MMR functions. Such events could provide a balance between fitness-loss and potential for adaptive evolution.
As shown above, we were able to create a wide range of mutator phenotypes through the segregation of only two genes, indicating that a small sampling of amino acid polymorphisms could have a major effect on fitness and potentially adaptive evolution. Could the incompatibility found between MLH1 and PMS1 in S. cerevisiae contribute to incipient speciation? Support for such an idea is based on the fact that reproductive barriers have already been shown to exist between some of the yeast strains reported in this study; for example, inter-strain crosses of SK1 and S288c show reduced spore viability of 73% [11]. We took advantage of the large databases of genome sequences obtained from yeasts isolated throughout the world to directly test MLH1-PMS1 combinations predicted to have been sampled in nature in highly sensitive mutator assays. This analysis provided molecular evidence for the S288c class of MLH1 alleles having a single origin. However, the strains containing this allele were scattered over four continents and from many different sources including lab strains, a clinical isolate, and several domesticated strains. This is not surprising due to the ubiquity of yeast in nature, the ease of mobility of the organism, and the extensive colonization of the species for commercial use [19,41]. It is clear from previous studies that it is difficult to assign geographic locations for even undomesticated populations of yeast. For example, in a study of wild S. paradoxus strains, evidence was obtained by sequence comparison for geographical divergence between North American and European strains. However, secondary contact with no admixture was recorded between these two groups from isolates found on the same tree in Pennsylvania [42]. In addition, defining species among yeasts has become difficult due to lack of pre-zygotic isolation [36,43], lateral gene transfer [44], and introgression [20] between Saccharomyces species. A lack of detailed population sample data from many localities throughout the species range and a lack of understanding of the relative frequencies of clonal vs. sexual reproduction in natural populations of S. cerevisiae prevent a critical evaluation of the role that the MLH1-PMS1 incompatibility played in past reproductive isolation. Nonetheless, the ability to use genetic approaches in yeast to identify incompatibilities and their effects on fitness make it an attractive model to study molecular evolution.

Plasmids
The plasmids used in this study are shown in Table S3. Plasmids pEAA213 (S288c MLH1, ARSH4 CEN6, LEU2) and pEAA214 (SK1 MLH1, ARSH4 CEN6, LEU2) were described previously [2,11]. In both vectors MLH1 expression is driven by the S288c MLH1 promoter. MLH1 from different strains of yeast were cloned into pEAA213 by amplifying MLH1 from genomic DNA [46] using Pfu turbo polymerase (Stratagene) and primers AO324 (59ATAGTG-TAGGAGGCGCTG) and AO821 (59AACTTTGCGGCCGCG-GATCCAGCCAAAACGTTTTAAAGTTA). The PCR amplified product containing the entire MLH1 open reading frame was digested with BamH1-and NheI and inserted into the corresponding sites of pEAA213. The entire PCR fragment was DNA sequenced. All of the resulting constructs expressed MLH1 via the S288c MLH1 promoter.
pEAA238 (S288c PMS1, ARSH4, CEN6, HIS3) and pEAA239 (SK1 PMS1, ARSH4, CEN6, HIS3) were described previously [2,11]. In both vectors PMS1 expression is driven by the S288c PMS1 promoter. PMS1 from different strains of yeast were cloned indicating the origin of MLH1 used in assay and head of arrow indicating origin of PMS1 used in assay. Panels B-D show intra-strain combinations (B, MLH1 and PMS1 are from the same strains), intra-group combinations (C, MLH1 and PMS1 are from the same group) and inter-group combinations (D, MLH1 and PMS1 from predicted compatible combinations). Panel E shows inter-group combinations predicted to display an incompatible phenotype. doi:10.1371/journal.pgen.1000103.g002  Table 2. Mutation rates (mutations per cell per generation) are presented relative to the S288c strain ( into pEAA238 by amplifying PMS1 from genomic DNA using Pfu turbo polymerase and primers AO548 (59CGATTCTAATACA-GATTTTAATGACC) and AO481 (59CCACGTTCATATTCT-TAATGGCTAAGC). The PCR amplified product containing the entire PMS1 open reading frame was digested with AatII-and MluI and inserted into the corresponding sites of pEAA238. The entire PCR fragment was DNA sequenced. All of the resulting constructs expressed PMS1 via the S288c PMS1 promoter.
All point mutations were made in pEAA213 using the QuickChange XL Site-Directed Mutagenesis protocol (Stratagene, USA). A fragment that contained the point mutation was then subcloned into unmutagenized pEAA213 and sequenced to determine that only the desired mutations were created.

Determination of Mutation Rates
pEAA213 (MLH1) and pEAA238 (PMS1) plasmids and derivatives were transformed into EAY1365 using standard methods [47] and were maintained on minimal histidine, leucine dropout plates. All transformants were assayed for their ability to grow on lactate as a carbon source. Because EAY1365 is a mutator strain, we were concerned that variation in genetic background between different strain isolates could affect mutation rate. For each MLH1, PMS1 combination, 15-30 independent cultures, from three to six independent transformants derived from the same EAY1365 frozen stock, were assayed to determine the mutation rate. We found that independent transformants containing the same plasmid combination displayed mutation rates that did not deviate significantly from each other; this can be seen in the 95% confidence intervals presented in Table 2. Reversion of the lys2::insE-A14 allele to Lys + was measured as described previously [10,11]. The reversion rate m (mutations per cell per generation) was calculated using the equation m = f/ln(Nm), where f is the frequency of revertants and N is the total number of revertants in The S288c strain EAY1365 was transformed with ARS CEN plasmids containing MLH1 and PMS1 from the indicated strains (Table 1). Mutation rates (mutations per cell per generation) were determined using the lys2::insE-A 14 reversion assay as described in the Materials and Methods. For each strain, 15-30 independent cultures, from three to six independent transformants, were assayed. See Figure 2 for class groupings. 95% C. I. = 95% confidence interval. doi:10.1371/journal.pgen.1000103.t002 Table 2. cont.
1 ml of overnight culture ( [21,48], Dmitry Gordenin, personal communication). This equation, derived by Drake [50], is transcendental and can be solved only by iteration; we use a computer method as suggested by Dmitry Gordenin. This equation is used to calculate the mutation rate for each culture and the median of the independent cultures is defined as the mutation rate. The 95% confidence intervals were determined as described [49].

Sequencing, Alignment, and Phylogenetic Reconstruction
The sequence of MLH1 and PMS1 from each of the strains in Table 1 was determined in house or taken from publicly available sequence (http://www.yeastgenome.org/; http://www.sanger.ac. uk/Teams/Team71/durbin/sgrp/; http://www.broad.mit.edu/ annotation/fungi/comp_yeasts/). For the in house sequencing, MLH1 and PMS1 coding regions and approximately 200 bp of downstream sequence were amplified by PCR, purified and then sequenced using Big Dye chemistry and an ABI 3730xl DNA Analyzer at the Cornell Life Sciences Core Laboratories Center.
Sequences were aligned using the CLUSTALW algorithm [50], using the resources of the Computational Biology Service Unit from Cornell University, which is partially funded by Microsoft Corporation. Neighbor-Joining trees were constructed based on nucleotide sequences, using PAUP*4.0b10 [51]. For MLH1, the nucleotide site responsible for the D761G polymorphism was excluded from phylogenetic reconstruction; similarly, the site responsible for R818K in PMS1 was excluded.