High-resolution mutation mapping reveals parallel experimental evolution in yeast

Abstract

Understanding the genetic basis of evolutionary adaptation is limited by our ability to efficiently identify the genomic locations of adaptive mutations. Here we describe a method that can quickly and precisely map the genetic basis of naturally and experimentally evolved complex traits using linkage analysis. A yeast strain that expresses the evolved trait is crossed to a distinct strain background and DNA from a large pool of progeny that express the trait of interest is hybridized to oligonucleotide microarrays that detect thousands of polymorphisms between the two strains. Adaptive mutations are detected by linkage to the polymorphisms from the evolved parent. We successfully tested our method by mapping five known genes to a precision of 0.2-24 kb (0.1-10 cM), and developed computer simulations to test the effect of different factors on mapping precision. We then applied this method to four yeast strains that had independently adapted to a fluctuating glucose-galactose environment. All four strains had acquired one or more missense mutations in GAL80, the repressor of the galactose utilization pathway. When transferred into the ancestral strain, the gal80 mutations conferred the fitness advantage that the evolved strains show in the transition from glucose to galactose. Our results show an example of parallel adaptation caused by mutations in the same gene.

Figure 1. Schematic Description of the Pooled Mapping Method Applied to a Single Pool of Segregants in Budding Yeast

(A) A strain expressing the trait of interest (target strain) is crossed to a highly polymorphic strain that lacks the trait (reference strain). The DNA polymorphisms between the strains, represented by yellow squares for the target strain's genotype and blue squares for the reference's genotype, are the genetic markers used for the linkage analysis. After mating, the hybrid diploid is sporulated, yielding a pool of haploids (segregants) that are genetically diverse due to random recombination along the chromosomes. A large pool of segregants that express the trait of interest (cells with red background) is selected, enriching for segregants that carry the alleles that give rise to the trait of interest (target locus; represented by red circle). As a control, a pool of segregants of comparable size is randomly collected. For simplicity, only a single chromosome is shown. (B) The location of the target locus is inferred by a genome-wide comparison of the fraction of the target strain's genotype (yellow squares) within the selected pool to that within the control pool. The genomic DNA of the selected and control pools are extracted and their patterns of polymorphisms along the genome are analyzed. At polymorphic sites that are unlinked to the target locus, half of the segregants in the selected pool are expected to display the target strain's genotype and half the reference strain's genotype. However, at linked polymorphic sites, the fraction of selected segregants that carry the target strain's genotype (black line) should be higher than 50% and inversely proportional to the distance between the polymorphic site (squares) and the target locus (red circle). For the control pool, the genotype of both parental strains should be equally represented throughout the entire genome (orange line). The target loci lie in chromosome regions where the target strain's genotype is significantly overrepresented in the selected pool relative to the control pool.

Figure 2. Successful Mapping of Known Genes

Two separate pools of ˜10 ⁷ W303/SK1 segregants, one resistant to canavanine that grew without lysine (can1, LYS5) and one resistant to geneticin, hygromycin, and nourseothricin (KAN ^R, NAT ^R, HYG ^R), were selected for and mapped. The five genes mapped are: (A) can1, centered on Chromosome 5, position 32.6 kb; (B) LYS5, on Chromosome 7, position 215.7 kb; (C) KAN ^R, on Chromosome 7, position 413.4 kb; (D) HYG ^R, on Chromosome 15, position 619.1 kb; and NAT ^R, on Chromosome 15, position 960.6 kb. The LMS was calculated across the whole genome for each pool (shown in Figure S3) and is plotted here along the chromosomes that carry the five target genes, as a function of chromosome position in 100-kb units. The five peaks that correspond to the five target genes all fell above the significant peak thresholds estimated at 99% confidence for each selected pool (horizontal dashed lines, which are so close to the x-axis as to be invisible in [C] and [D]). The arrows mark the actual center of the target genes, the solid lines within the drawn chromosomes mark the predicted center of the genes, and the grey boxes within the chromosomes mark the 95% confidence intervals estimated with simulations (see Materials and Methods). The peak for LYS5 is low relative to the significant peak cutoff, because of the low local SFP density (see Figure S3G– S3I for discussion). Note that the scale of the y-axis is different in the four panels ( Protocol S1). (E) The mapping deviations of the genes' predicted centers from their actual centers and their 95% confidence intervals. Their corresponding average genetic distance in cM is written in parentheses. All five genes were found within their 95% confidence intervals.

Figure 3. Computer Simulations Evaluate the Effect of Different Factors on Mapping Precision

The absolute distance of the predicted position of a simulated target locus from its real position (mapping deviation) is plotted as a function of (A) number of array replicates, (B) coefficient of variation (standard deviation divided by the mean) of hybridization intensities, (C) SFP FDR, (D) SFP density, and (E) recombination rate. Aside from the varying factor, the parameters were set to the values observed in our test case ( Figure 2) and are marked on each panel with a red arrow. With the exception of (A), four replicate arrays were used for the selected and control pools, and eight replicates for the target strain. The mean coefficient of variation values of all our test case hybridizations at SFPs varied between 8%–15%. A smoothing window of 50 SFPs was used for all simulations except for SFP density, where a window of constant chromosome size (50 kb) was used. The mean and standard errors were computed from 5 repetitions of 1,000 simulation runs each for every datapoint. Error bars that are not visible are smaller than the dot. Note the logarithmic scale of the x-axis of SFP density.

Figure 4. Four Evolved Strains That Independently Adapted to Glucose–Galactose Transition Acquired One or Two Missense Mutations in GAL80

(A) A schematic description of the evolution experiment. In each cycle of evolution, haploid cells (light blue) were grown in glucose-containing media for 4 d, mated on YPD plates, and then transferred to galactose-containing media for 2 d. The diploid cells (dark blue) were then put through a sporulation cycle (meiosis) with potassium acetate as their carbon source. Each evolved population was put through 36 such cycles. (B) All four evolved strains were mapped to a single locus on Chromosome 13. The LMS is plotted as a function of chromosome position in 100-kb units for Chromosome 13, where a significant peak was detected in all four strains. The predicted peak centers and estimated 95% confidence intervals are: Ev2, 185.8 ± 27 kb (orange); Ev14, 165.8 ± 39 kb (red); Ev42, 169.8 ± 40 kb (green); and Ev43, 171.8 ± 36 kb (black). The solid cyan line within the schematic chromosome marks the mean of the four predicted peak centers (173.3 kb), and the grey box within the chromosome marks the mean estimated 95% confidence interval (±35.5 kb [±14.2 cM]). The genes that fall within a 30-kb interval around the mean peak center (cyan dashed line) are depicted below (black or gray boxes) with the GAL80 gene colored in red (centered at 172.2 kb; gene coordinates taken from the Saccharomyces Genome Database). The peak centers of the four evolved strains are marked with lines color-coded according to their LMS plot. Figure S6 shows the LMS across the entire genome for this experiment. (C) GAL80 was sequenced in the four evolved strains (Ev2, Ev14, Ev42, and Ev43) and in the ancestor, and the mutations found are presented at the nucleotide and amino acid levels. The mutated nucleotides are underlined. In addition, a deletion of a single T was found 96 nucleotides upstream to the translation start site of GAL80 in Ev42. This mutation is unlikely to have a significant effect on the activity of Gal80, as the mutation in amino acid 222 alone was sufficient to recapitulate the adaptive phenotype (see Figure 5D). The GAL80 sequence of the reference strain, SK1, is identical to that of the ancestor, W303, at the nucleotide level.

Figure 5. Transformation of gal80 Mutations into the Ancestral GAL80 Reconstructs the Adaptive Phenotype

(A) A schematic depiction of the regulation of the galactose utilization pathway. In the absence of galactose, Gal80 inhibits the transcriptional activator, Gal4, by binding to Gal4 in the nucleus. When galactose is present it enters the cell through Gal2 transporters and binds Gal3, a coinducer of the pathway, which in turn binds Gal80 in the cytoplasm, sequestering Gal80 away from the nucleus. This relieves the repression of Gal4 allowing it to induce the transcription of genes required for galactose uptake and catabolism ( GAL genes), including GAL2, GAL3, GAL80, and the genes encoding the enzymes of galactose catabolism [ 38]. A similar phenotype is obtained through loss-of-function of the repressor, GAL80. (B–E) The gal80 mutations confer a fitness advantage in transfers of exponentially growing haploid cells from glucose- to galactose-containing medium, but not in transfers in which the carbon source does not change ( Figure S7). Three different sets of gal80 mutations in the coding region were transformed into the ancestral GAL80 gene in an ancestral haploid strain (A0) (Ev2 indicates the mutation is from evolved culture 2, etc.). Ev42 and Ev43 have the same mutation in the coding region. Cell density (OD) was measured for each of these strains and for the ancestor and a GAL80 knockout strain following transfer from glucose- to galactose-containing medium. (F–I) The four evolved strains are more fit than their ancestor (A0) when transferred from glucose- to galactose-containing medium. Cell density (OD) was measured for the haploid evolved strains Ev2, Ev14, Ev43, and Ev42, and their ancestor, following transfer from medium containing only glucose to medium containing galactose. (J–M) The gal80 mutations have a semidominant effect when present in one copy in the ancestral diploid strain following transfer from glucose- to galactose-containing medium. The cell density (OD) of ancestral diploids carrying one copy of the gal80 mutations from either Ev2, Ev14, or Ev43 were compared to that of an ancestral diploid (A0) and a diploid lacking both copies of GAL80 (gal80Δ/gal80Δ), following transfer from glucose- to galactose-containing medium. As a control, an ancestral diploid was made hemizygous for GAL80 (M; GAL80/gal80Δ). Mean cell density and a standard deviation from at least three independent cultures were plotted for each datapoint for (B–M). Error bars that are not visible are smaller than the datapoint.