Beneficial mutations are required for adaptation to novel environments, yet the range of mutational pathways that are available to a population has been poorly characterized, particularly in eukaryotes. We assessed the genetic changes of the first mutations acquired during adaptation to a novel environment (exposure to the fungicide, nystatin) in 35 haploid lines of Saccharomyces cerevisiae. Through whole-genome resequencing we found that the genomic scope for adaptation was narrow; all adapted lines acquired a mutation in one of four late-acting genes in the ergosterol biosynthesis pathway, with very few other mutations found. Lines that acquired different ergosterol mutations in the same gene exhibited very similar tolerance to nystatin. All lines were found to have a cost relative to wild type in an unstressful environment; the level of this cost was also strongly correlated with the ergosterol gene bearing the mutation. Interestingly, we uncovered both positive and negative effects on tolerance to other harsh environments for mutations in the different ergosterol genes, indicating that these beneficial mutations have effects that differ in sign among environmental challenges. These results demonstrate that although the genomic target was narrow, different adaptive mutations can lead populations down different evolutionary pathways, with respect to their ability to tolerate (or succumb to) other environmental challenges.
POPULATIONS adapt to stressful environments through the fixation of beneficial alleles. The number of advantageous mutations accessible to a population within one or a few mutational steps [“mutational neighborhood” (Burch and Chao 2000)] remains particularly poorly characterized, especially in eukaryotes. This is an important factor, however, as the number of mutations in concert with their pleiotropic effects will directly influence the range of evolutionary pathways available to different populations. The first beneficial mutations to fix are of particular interest, as genetic and gene–environment (G × E) interactions may dictate the fixation of subsequent mutations. Knowledge of the number of available pathways may help us predict whether two populations subjected to similar selective pressures in allopatry might accumulate and fix different mutations. If this frequently occurs, reproductive isolation could evolve purely by chance fixation of different mutations [the mutation-order hypothesis (Schluter 2009)]. We thus sought to determine the mutational neighborhood of adaptive mutations in one environment and to characterize the pleiotropic effects of these mutations to different environmental challenges.
A fruitful approach to characterize the genotypic basis of adaptation has involved experimental microbial studies, where multiple replicate populations are initiated with the same ancestral culture and evolved under the same conditions for several generations (Conrad et al. 2011). Targeted resequencing of specific genes in replicate populations evolved for hundreds or thousands of generations at large population size (where selection should overwhelm drift) has demonstrated that in many cases the same genes repeatedly acquire mutations (Cooper et al. 2003; Pelosi et al. 2006; Woods et al. 2006; Ostrowski et al. 2008; Barrick et al. 2009). An examination of diverse clinical isolates of Pseudomonas aeruginosa has also repeatedly implicated the same genes during the acquisition of resistance to quinolone (Wong and Kassen 2011). Parallel genotypic evolution is not restricted to the utilization of single genes, as parallel transposition mutations (Chou et al. 2009) and large-scale aneuploid events (Selmecki et al. 2009) have also been documented in replicate lineages evolved under the same conditions. Furthermore, the magnitude of genetic parallelism has been shown to be influenced by the selective environment (Anderson et al. 2003; Gresham et al. 2008), depending on both the size of the genomic target for beneficial mutations and the probability of establishment of different mutations.
Targeted resequencing studies may, however, paint a skewed picture of the extent to which parallel mutations underlie evolutionary change, as only a few genes are typically examined and such candidate genes may be more likely to be repeated targets of beneficial mutations. Only through whole-genome resequencing (WGS) can the full array of beneficial mutations and their chance of appearing repeatedly be assessed. Over the last few years, a broader picture of the types of mutations acquired in long-term experimental evolution lines has been painted by WGS. The results suggest that while the number of different genes available to adaptation and the types of mutations depend on both the species and the environment, some generalities from targeted resequencing studies seem to hold. In all cases a small number of genes have been the target of beneficial mutations in independently evolved lines, with nonsynonymous single-nucleotide changes (SNPs) being the most common type of mutation (Herring et al. 2006; Araya et al. 2010; Kishimoto et al. 2010; Minty et al. 2011; Toprak et al. 2011; Tenaillon et al. 2012). Considerable variation among experiments is found in the total number of different genes targeted, however, and variation is also present for the absolute frequencies of different classes of mutations (e.g., copy number variants, insertions, deletions, and regulatory changes).
The order of mutational steps can have a tremendous impact on the fitness effect of subsequent mutations (Weinreich et al. 2005), both in magnitude and in sign, implying that the first adaptive step taken can alter the path of evolution. While WGS has allowed us to leap forward in our understanding of the genetic basis of adaptation, fewer WGS studies have focused on the first step of adaptation, and none have yet characterized the first steps in a eukaryote. The data that exist in viruses and prokaryotes suggest that the first mutations to be selected also tend to be clustered in relatively few genes. Rokyta et al. (2005) identified 10 unique nonsynonymous single-step mutations through WGS within two different viral genes of ΦX174. Similar results were found through targeted resequencing in both P. aeruginosa [where 15 unique mutations were identified in rpoB (MacLean and Buckling 2009)] and P. fluorescens [five nonsynonymous SNPs were found in gyrA and four mutations were found within three efflux pump regulatory sites (Bataillon et al. 2011)]. As with all targeted resequencing studies, it remains unknown how many additional mutations were present in the lines acquired in Pseudomonas. It has also not yet been determined whether eukaryotes accumulate mutations in a similar (i.e., largely parallel) manner. A number of fundamental characteristics differ between prokaryotic and eukaryotic genomes [e.g., chromosome structure, the number of replication origins, the amount of noncoding DNA, and the degree of transcript processing, just to name a few (Poole et al. 2003)]. These or other factors could affect the nature of mutations acquired under stressful conditions in eukaryotic genomes.
To assess the mutational neighborhood allowing adaptation to a novel stressful environment in a eukaryote we developed an assay to isolate multiple adapted lines of haploid Saccharomyces cerevisiae. We exposed 240 replicate lines initiated with ∼100,000 progenitor cells to a level of stressor that inhibits growth of the ancestral strain. By immediately isolating cells that were able to grow in this environment, we limited the number of mutational hits in the genome and reduced the potential influence of clonal interference. Through WGS we pinpointed the genetic basis of adaptation for each lineage. We chose a polyene antibiotic, nystatin, as the stressful environment in which to acquire mutations. Nystatin binds to ergosterol (the primary sterol in the fungal membrane) to form porin channels that increase membrane permeability and allow cellular components (including potassium ions, sugar, and metabolites) to leak out of the cell (Carrillo-Munoz et al. 2006; Kanafani and Perfect 2008). The resulting change in potassium concentration leads to an osmotic imbalance between the vacuole and cytoplasm and an enlarged vacuole (Bhiyan et al. 1999). Transcriptional profiling has identified membrane transporters and the cell stress response as the major cellular components affected by exposure to nystatin (Hapala et al. 2005). Previous work has identified resistant mutants in Candida albicans and S. cerevisiae that show defects in genes involved in the ergosterol biosynthesis pathway, with mutants exhibiting an altered sterol content in the cell membrane (Bhiyan et al. 1999; Ghannoum and Rice 1999; Kanafani and Perfect 2008). Whether the ergosterol pathway would always be involved in resistance evolution, or whether beneficial mutations could be recruited from other membrane components or from altered ion pumps, remains unknown. The first goal of our research was thus to document the genetic basis of adaptation to nystatin in many replicate lineages and to measure the fitness benefit gained by each adapted lineage.
We then determined whether the mutations we identified exhibited different responses to other stressful environments (G × E). Given that populations often face multiple environmental challenges simultaneously, the scope for adaptation would be greatly reduced if adaptive mutations in the presence of a single environmental change always exhibit reduced tolerance to other environmental challenges (strict trade-offs). The idea that strict trade-offs should exist during adaptation to novel environments is long-standing and multiple hypotheses have been put forward to explain the physiological basis of trade-offs (Pörtner et al. 2006), yet trade-offs are not universally found (Ostrowski et al. 2005; Bennett and Lenski 2007; Hereford 2009). Another possibility is that mutations vary in the subset of environments in which they are beneficial [i.e., there is “sign G × E,” by analogy with sign epistasis (Weinreich et al. 2005)]. In this case, some mutations may be simultaneously beneficial to multiple types of change in the environment, allowing the organism to adapt more readily to complex environmental challenges. Furthermore, such gene–environment interactions imply that lineages carrying different first-step mutations would find themselves at different locations on the adaptive fitness surface after further changes in the environment. To explore the nature of gene–environment interactions among single adaptive mutations, we conducted a set of experiments to measure the fitness effects under different stressful conditions of mutations whose genetic basis is known, allowing us to compare mutations in different genes and at different sites within a gene. To explore a variety of environmental challenges, we varied levels of copper, ethanol, and salt, measuring growth of each line in each environment.
When exposed to nystatin, we found strong parallelism in the adaptive mutations that appeared within our S. cerevisiae lines at both the gene and the pathway levels, with only a limited number of genes being involved in the first step of adaptation. Mutations in different genes had significantly different fitness effects across environments, with some lines showing increased tolerance and others decreased tolerance to other stressors (sign G × E). Our results thus provide support for the mutational-order hypothesis that adaptation to one environmental challenge may well drive isolated populations down different evolutionary pathways, with significant differences in starting fitness when faced with further environmental challenges.
Materials and Methods
Strain background and mutation acquisition
Mutations were acquired in haploids of genotype BY4741 (MATa his3△1 leu2△0 met15△0 ura3△0) derived from S288C. Stocks of the ancestral culture and relevant gene deletion lines (erg6Δ, erg3Δ, and erg5Δ, see below) were ordered from Open Biosystems (Thermo Fisher Scientific) and streaked on a YPD plate, and a single colony was isolated and frozen. This single colony from BY4741 served as the progenitor culture for all experiments, and we refer to this as the ancestral strain. Mutation accumulation was carried out in 1-ml culture in 96-deep–well plates (2-ml polypropylene plates with a conical bottom), with shaking at 200 rpm at 30°. We acquired mutations in two screens, separated by 2 weeks. To initiate each screen, the ancestral strain was streaked onto a YPD plate from the frozen isolate and grown for 48 hr. A single random colony was then picked and grown for 24 hr in 10 ml YPD at 30°, with shaking at 200 rpm.
We initiated mutation acquisition in the first screen by transferring 10 μl of the ancestral overnight culture into 1 ml of YPD + 4 μM nystatin into the 60 inner wells of a 96-deep–well plate. The second screen was identical except we initiated 180 replicates into the inner wells of three 96-deep–well plates. The level of nystatin was determined in preliminary experiments as the level that showed only sporadic growth of the ancestral strain within 2–7 days of incubation, suggesting that growth required a mutational event (results not shown). Growth in nystatin for each screen was checked and recorded daily by visual examination of the bottom of the 96-well plates. A small amount of growth would typically be observed one day, with full growth on the second or third day (where full growth is approximately equivalent to the amount of turbidity and precipitate seen in the ancestral strain after 24 hr of growth in YPD). Occasionally full growth took up to 4 days. Each well that showed growth (even slow growth) was marked as a “putative mutation” line.
On the first day that full growth was recorded for each putative mutation line, the well was thoroughly mixed by pipetting and culture was streaked onto a YPD plate. Putative mutation lines were obtained in this way from 64 of the 240 inoculated wells, with no growth observed in the remaining wells over 7 days. After 48 hr of growth on the plate, we visually assessed each line for petite mutations (mutations that affect mitochondrial function and prevent respiration; these colonies present as much smaller than normal on a YPD plate). About half of the putative mutation lines showed evidence of petite mutations. All suspected petites were confirmed by lack of growth on a YPG plate, a medium that requires respiration for growth, and these lines were discarded to focus on nuclear mutations. For each of the remaining 35 lines, eight colonies were haphazardly picked off the YPD plate, placed back into 8 wells containing 1 ml of YPD + 4 μM nystatin, and assayed for growth. Often all eight colonies picked would exhibit similar growth patterns, but sometimes colonies varied with respect to the number of days to full growth or even whether any growth was observed in nystatin within 48 hr. To avoid analyzing nonmutant cells that might have been segregating, we randomly picked a single well that showed any level of growth in nystatin for each mutation line. The 1-ml culture from this well was mixed with 1 ml 30% glycerol and frozen; this freezer culture constitutes the material for all future experiments. These 35 lines were labeled BMN1–35 (BMN: “beneficial mutation nystatin”), as described in the Table 1 legend.
Freezer culture from each BMN line was streaked onto YPD plates and grown for 48 hr. A single colony was then haphazardly picked for each line and grown for 24 hr in YPD. DNA was extracted (Sambrook and Russell 2001) and sequenced in 100-bp single-end fragments, using Illumina’s (San Diego) HiSeq 2000. Library preps followed standard Illumina protocols (2011 Illumina, Inc., all rights reserved), with 12 uniquely barcoded strains run together on a lane. The resulting genomic sequence data were processed using Illumina’s CASAVA-1.8.0. Specifically, configureBclToFastq.pl was used to convert to fastq and separate the sequences by barcode (allowing one mismatched base pair). ConfigureAlignment.pl (based on the alignment program ELAND) was then used to align each sequence to the yeast reference genome (scergenome.fasta downloaded from the Saccharomyces Genome Database, http://downloads.yeastgenome.org/genome_release/r64/).
SNPs and small insertions and deletions (indels) were then called using configureBuild.pl. Average coverage per mapped site across the strains (excluding mitochondrial genes) was 44.0 (with a minimum average coverage per site of 4.7 for BMN13). Data on relative coverage per chromosome are illustrated in Supporting Information, Figure S1, which indicate that one line (BMN27) had an additional copy of chromosome 2 (denoted as a black circle in Figure S1). Custom UNIX and perl scripts were then used to parse the output files. Illumina data from an independent project using the same ancestral strain were used to identify mutations that were common to the ancestor, and all such mutations were ignored. Given that our initial lines were haploid, variants identified by configureBuild.pl as heterozygous were also discarded as likely alignment or sequencing errors. Similarly, variants involving repeat elements were discarded. All remaining variants were checked in the alignments, using tview in samtools-0.1.7a (Li et al. 2009); variants that were not supported by multiple fragments starting from different positions were also discarded (typically near deletions or gaps in the alignment). Finally, the same procedure was repeated, but using the bwa software package to perform the alignment (Li et al. 2009) along with samtools-0.1.7a to identify SNPs (Li et al. 2009), using the -bq 1 option to limit data to reliable alignments. All SNPs (Table 1 and Table S1) were identified using both methods.
Sequence alignments were also manually checked using tview in samtools-0.1.7a for each of the four genes harboring beneficial mutations (ERG3, ERG5, ERG6, and ERG7) to look for larger rearrangements or other changes not identified by the above procedure. Two additional large-scale mutations were identified from gaps in the alignments. To determine the nature of these rearrangements, the fastq files containing the unaligned short-read sequences were directly searched for sequences on either side of the alignment gap, confirming a 60-bp deletion in ERG5 within line BMN35 and a 29-bp duplication in ERG3 within line BMN28 (Table 1). For confirmation, we Sanger sequenced the appropriate gene from BMN lines representing 10 of the 20 unique mutations found (Table 1). In all cases the Sanger sequence data matched our analysis of the genomic sequence data. It is worth noting repetitive sequences were often not mapped and were ignored when mapped, so that any mutations in such sites or any larger-scale rearrangements not involving these four genes would likely have been missed. Nevertheless, 95.3% (SE = 0.15%) of the known sites in the S. cerevisiae reference genome had coverage in our alignments, so that we are likely to have detected most nucleotide mutations and small indels that occurred.
We compared the sterol profile of the ancestral strain (BY4741) and BMN lines, using a spectrophotometry-based assay. When more than one BMN line shared the same ergosterol mutation, we randomly chose one line to represent that group. Sterols were extracted using the alcoholic potassium hydroxide method as previously described (Arthington-Skaggs et al. 1999). BMN culture streaked to a single colony on a YPD plate was inoculated into 50 ml of YPD and grown at 30° for 48 hr at which point the optical density (at 630 nm) was measured to record cell concentration. Cells were then harvested by centrifugation at 2700 rpm for 5 min and washed twice with sterile distilled water. Three milliliters of 25% alcoholic KOH was added to each pellet and vortexed for 1 min. The sample was then incubated in an 80° water bath for 1 hr and then cooled to room temperature. To extract the sterols, 1 ml of sterile distilled water and 3 ml of heptane were added and vortexed for 3 min. A 200 μl aliquot of the heptane layer was added to 800 μl of 95% ethanol, and the absorbance was immediately read every 3 nm between 200 and 300 nm with a Thermo BioMate 3 spectrophotometer.
To determine the breadth of nystatin tolerance conferred by each mutation, a growth assay was performed to measure the half-maximal inhibitory concentration (IC50) of nystatin. Freezer stock from each BMN line was inoculated into 1 ml of YPD in one well of a 96-deep–well plate and grown for 48 hr. To standardize the starting density of cells, the optical density (OD) of 200 μl from each well was measured using the BioTek microplate reader (BioTek Instruments, Winooski, VT) and diluted to the sample with the lowest OD (usually between 0.7 and 0.9). Two hundred microliters of the standardized culture was then added to 400 μl of YPD to obtain the final volume necessary for the assay inoculations. For each BMN line, 12 μl of the dilute culture was then inoculated into a well containing 1 ml YPD plus 1 of 10 levels of nystatin (0 μM, 2 μM, 4 μM, 8 μM, 12 μM, 16 μM, 46 μM, 96 μM, 116 μM, and 200 μM), each replicated in 4 different nonadjacent wells. Plates were sampled at 72 hr to measure OD. Wells were manually mixed and 150 μl aliquots were taken from each well and read on the BioTek reader.
A maximum-likelihood model was fit to the data to determine IC50. The logistic function(1)was used, where x represents the tested concentration of nystatin, y represents the observed OD following 72 hr of growth, and N(0, σ) represents a normal deviate with mean zero and standard deviation, σ. The fitted parameters were ymax (the maximal OD under full growth), IC50 (the nystatin concentration at which OD is half maximal), a (the slope of the logistic curve at x = IC50 divided by ymax/4), and σ. Prior to fitting the data using this likelihood procedure, all nystatin concentrations were ln transformed (so that percentage changes, not absolute differences, in nystatin matter), although we report all values of nystatin concentration and IC50 on the original scale. The maximum-likelihood point was found in R, using the subplex method of optim, as implemented in the find.mle routine of the diversitree package (FitzJohn et al. 2009). The find.mle routine allows lower and upper limits to the parameters to be specified in the search routine (we used lower, ymax = 0.8, IC50 = 0.0000001, a = −50, and σ = 0; and upper, ymax = 1.2, IC50 = 116, a = 0, and σ = 10; lower and upper ymax and IC50 were based on observations).
To determine whether the IC50 of a mutant line was significantly different from that of the ancestral strain, a likelihood model was fitted to the data from the mutant line and the ancestral line, allowing each of these two lines to have its own values of ymax, IC50, a, and σ. This “full” model was then compared to a constrained model where IC50,mutant = IC50,ancestral, using a likelihood-ratio test. If the drop in log-likelihood between the full and the constrained model was > = 1.92, we rejected the hypothesis that IC50 was the same for the two lines.
Fitness proxies in a permissive environment and the evolutionary environment
We assessed growth in both the evolutionary environment (YPD + 4 μM nystatin) and an unstressful environment (standard laboratory YPD), using two fitness proxies. We measured the maximal growth rate to measure how quickly yeast cells are taking in nutrients and growing during the exponential phase of growth and optical density at 48 hr to capture the ability to turn resources into biomass. Both fitness proxies were determined using previously described methods (Gerstein and Otto 2011) that utilize the Bioscreen C Microbiological Workstation (Thermo Labsystems), which measures OD in 100-well honeycomb plates. In brief, plates were streaked from frozen stock onto YPD plates for all lines and the ancestral strain and allowed to grow for 72 hr. Inoculations containing one colony (haphazardly chosen) for each BMN line and five separate inoculations of the ancestral strain (each from a different single colony) were then allowed to grow for 48 hr in 10 ml YPD. One hundred microliters was transferred into 10 ml of fresh YPD and mixed thoroughly, and four 150 μl aliquots for each line were placed into nonadjacent bioscreen wells. The bioscreen plates were grown at 30° for 48 hr with constant shaking; OD readings were automatically taken every 30 min. We determine the maximal growth rate for each well as the spline with the highest slope, from a loess fit through ln-transformed optical density data, using an analysis program written by Richard FitzJohn in R (R Development Core Team 2011). OD at 48 hr was used as a second fitness measure (OD48). As can be seen from the raw growth curves (Figure S2), the lines have stopped growing by 48 hr in YPD, and this measure thus reflects efficiency (i.e., ability to turn nutrients into cellular material). In nystatin, by contrast, some lines may still be growing, and this assay thus represents a combined measure of growth rate and efficiency.
Assessing gene–environment interactions
The ecological tolerance (measured as IC50) was determined for each line in copper (CuS04), ethanol, and salt (NaCl). The tolerance assays in these environments were conducted as previously described for nystatin; we measured growth after 72 hr in eight levels of copper (0 mM, 1 mM, 2 mM, 4 mM, 6 mM, 8 mM, 10 mM, and 12 mM), seven levels of ethanol (0%, 2%, 4%, 6%, 8%, 12%, and 14%), and eight levels of salt (0.2 M, 0.4 M, 0.6 M, 0.8 M, 1.2 M, 1.4 M, 1.6 M, and 2 M). These levels were chosen based on preliminary data that indicated the approximate position of IC50. Significance was determined as with tolerance to nystatin, using a maximum-likelihood test that compared a model fit with two IC50 parameters (one for the mutation line and one for the ancestor) to a model with only one IC50 value.
We acquired 64 haploid lines of S. cerevisiae that were resistant to nystatin in two independent screens. We eliminated all lines that could not respire, which left us with 35 lines that we term BMN lines. Through WGS with the Illumina platform, we found that each line carried a single mutation in one of four late-acting genes in the ergosterol biosynthesis pathway (Figure 1). Within these lines, we found 1 line each with mutations in ERG7 (“BMN-erg7”) and ERG5 (“BMN-erg5”), 7 unique mutations in ERG6 within 19 lines (collectively referred to as “BMN-erg6 lines”), and 11 unique mutations within 14 lines in ERG3 (“BMN-erg3 lines”). We recovered multiple classes of mutations including nonsynonymous SNPs, premature stop codons, small indels (<3 bp), one 60-bp deletion, and one 29-bp duplication (Figure 1); the precise nucleotide and amino acid changes as well the numbering scheme for BMN lines are presented in Table 1.
For five mutations, the same sequence change was observed in multiple lines (Table 1). There are three potential explanations for this finding. The most likely is that mutations were initially segregating in the source population before the stressor was applied (see File S1 and File S2). A second explanation, that contamination occurred among wells in the 96-well plates, is possible, but fails to explain the similar timing of appearance of identical hits (Table S1). The independent appearance of the same sequence changes is also possible, but in no case did we observe the same mutation in the two screens (Table S1). While we treat each line as independent for statistical purposes, we note that combining lines with the same ergosterol mutation leads to the same conclusions (see File S1).
We did not expect to see many mutations other than those conferring a fitness benefit given the relatively small initial population size (∼105 cells), small genome-wide mutation rate (Lynch et al. 2008), and short time frame of the experiment (we stopped once growth could be observed, thus minimizing the number of generations; File S2). Nevertheless, we identified a small number of additional point mutations (five synonymous, eight nonsynonymous, and one nonsense changes in genes that are not part of the ergosterol pathway and five mutations in intergenic regions; Table S2). The majority of these mutations were unique to a single BMN line, but two mutations were found in multiple lines. A nonsynonymous change from glutamic acid to lysine in FCY2 was found in four lines (BMN24–27), and a synonymous mutation was found in GDA1 in five lines (BMN11–15); in both cases, these two sets of lines also shared a primary ergosterol mutation (Table 1), strongly suggesting that these two sets may be derived from the same mutations that arose in the precursor population. The genome sizes of all lines were measured using flow cytometry, and no deviations from haploidy were found. Examining the depth of coverage from Illumina data (see Materials and Methods) uncovered one case of chromosomal aneuploidy (BMN27 had a duplicated chromosome II, Figure S1). We did not find strong evidence that any of the nonergosterol mutations influence fitness in the environments measured (see File S1), and thus we focus our discussion on the ergosterol mutations.
We first measured the sterol profile of all lines. This assay takes advantage of the characteristic four-peak curve produced by ergosterol and the late sterol intermediate 24(28)dehydroergosterol (DHE) that are present in wild-type cells (Arthington-Skaggs et al. 1999). All lines that carried mutations in the same ergosterol gene showed nearly identical sterol profiles (Figure 2). Interestingly, only BMN-erg5 (the line with a mutation in the gene closest to the end of the pathway) had a sterol profile similar to that of the ancestral strain. The sterol profiles for BMN-erg6 lines and BMN-erg3 lines have a similar shape to previously published results of erg3Δ and erg6Δ obtained using the same protocol (Jensen-Pergakes et al. 1998; Mukhopadhyay et al. 2002). Surprisingly, however, our own measures of erg6Δ and erg3Δ (and erg5Δ) recovered the ancestral sterol phenotype (not shown).
All mutation lines had a significantly higher tolerance to nystatin than the ancestral strain, and many lines could tolerate nystatin at much higher levels than the 4 μM concentration used to isolate beneficial mutations (Figure 3). We measured the breadth of tolerance as IC50, i.e., the inhibitory concentration of the drug that reduced growth by 50%. The significance of changes in IC50 relative to that in the ancestor was determined by likelihood-ratio tests (Table S3). Replicate lines that carried different mutations in the same ergosterol gene showed similar IC50 values (Figure 3). A two-way ANOVA found that IC50 in nystatin has a very strong association with the ergosterol gene bearing a mutation (F3 = 252.4, P < 0.0001) but was not affected by either the class of mutation within a gene (i.e., nonsynonymous SNP, premature stop codon, or indel) or their interaction (mutation type, F2 = 0.92, P = 0.41; interaction, F2 = 0.66, P = 0.53). We then compared the tolerance of our lines to that of S. cerevisiae strains that carry gene knockouts for ERG6, ERG3, and ERG5 (erg7Δ is inviable and could not be tested). Although all gene knockout lines did show increased nystatin tolerance compared to the ancestor (Figure 3), we found that while BMN-erg5 and erg5Δ had similar IC50 values (Figure 3), BMN-erg6 lines had a significantly higher nystatin tolerance than erg6Δ, and all but two BMN-erg3 lines had a significantly lower nystatin tolerance than erg3Δ.
We also measured two fitness-related proxies for all lines in both the evolutionary environment (YPD + 4 μM nystatin) and an unstressful environment (standard laboratory YPD). When grown in nystatin, all BMN lines reached a higher optical density at 48 hr (OD48, Figure 4A) and had a higher maximal growth rate (Figure 4C) than the ancestral strain (significance determined by a t-test compared to five ancestral colonies, Figure 4 and Table S4, Table S5, Table S6, and Table S7). Conversely, the ancestor performed better than all BMN lines in YPD for both fitness proxies (Figure 4, B and D). Growth rate and OD48 were significantly correlated with each other when BMN lines were grown in nystatin, consistent with the idea that both assays measure an aspect of growth rate (cor = 0.87, t33 = 10.3, P < 0.0001), and both were significantly correlated with nystatin IC50 (OD48, cor = 0.74, t33 = 6.4, P < 0.0001; growth rate, cor = 0.76, t33 = 6.8, P < 0.0001). When lines were grown in YPD, however, growth rate and OD48 were not significantly correlated with each other, consistent with growth having completed by 48 hr in YPD and OD48 measuring solely the efficiency of converting resources into cellular material (cor = 0.26, t33 = 1.5, P = 0.13). Growth rate in YPD was significantly correlated with IC50 in nystatin (cor = 0.56, t33 = 2.8, P = 0.01), while OD48 was not (cor = 0.04, t33 = 0.18, P = 0.20). Interestingly, the correlation between growth rate in YPD and IC50 in nystatin was positive; that is, mutations with the broadest tolerance to nystatin were among the best to grow in YPD. The ergosterol gene that bore a mutation was significantly associated with both fitness proxies in both environments (growth rate in nystatin, F3 = 26.4, P < 0.0001; OD48 in nystatin, F3 = 15.5, P < 0.0001; growth rate in YPD, F3 = 25.8, P < 0.0001; OD48 in YPD, F3 = 4.9, P = 0.007). The type of mutation was found to have a significant effect on growth rate in nystatin (F2 = 5.0, P = 0.014), although we note that of all significant statistical results this is the only one that does not remain significant when we combine multiple lines with the same ergosterol mutation (see File S1). In all other comparisons we found no significant association with the type of mutation (OD48 in nystatin, F2 = 0.86, P = 0.44; growth rate in YPD, F2 = 0.27, P = 0.77; OD48 in YPD, F2 = 0.09, P = 0.91) nor was the interaction between ergosterol gene and type of mutation significant (growth rate in nystatin, F2 = 0.16, P = 0.85; OD48 in nystatin, F2 = 0.68, P = 0.51; growth rate in YPD, F2 = 0.33, P = 0.72; OD48 in YPD, F2 = 0.73, P = 0.50).
We observed substantial differences among the nystatin resistance lines in their tolerance to other stressful environments (breadth of tolerance measured as IC50 in all environments, Figure 5). We found significant negative correlations between tolerance to nystatin and tolerance to both copper and ethanol (copper, cor = −0.80, t34 = −7.6, P < 0.00001; ethanol, cor = −0.63, t34 = −4.7, P < 0.00001) and no correlation between nystatin and salt tolerance (cor = −0.07, t34 = −0.39, P = 0.70). The tolerance breadths exhibited by lines with mutations in the same ergosterol gene were fairly consistent, with only a few exceptions. Importantly, although reduced tolerance to all other environments tested was observed for some genes bearing nystatin resistance mutations (especially ERG6 mutations), mutations in other ergosterol genes had no effect or even a positive effect on growth in the face of other environmental challenges (e.g., positive fitness effects were observed for BMN-erg7, BMN-erg3, and BMN-erg5 lines in copper). That is, mutations in different ergosterol genes exhibited significant sign G × E when comparing growth in nystatin and copper. Overall, the majority of lines differed in fitness from the ancestor in most environments (Figure 5, Table S8), but the pattern was heavily dependent on both environment and gene.
Genes that act late in the ergosterol biosysnthesis pathway were found to be the primary (and possibly exclusive) target for the first step of adaptation by S. cerevisiae to low levels of nystatin. The mutational neighborhood was reasonably large, as we uncovered 20 unique mutations within four genes exhibiting increased tolerance to nystatin. At the gene level, however, the genomic scope for beneficial mutations was quite narrow in this environment, as all but two lines carried mutations within ERG6 and ERG3. Lines with different mutations in the same gene tended to exhibit similar tolerance phenotypes in all environments tested, including altered levels of the original stressor, an unstressful environment (YPD), and three different stressful environments (copper, ethanol, and salt). We found nonparallel fitness effects of mutations in different ergosterol genes in the face of different environmental challenges, indicating the unpredictable nature of gene–environment interactions. Although some lines showed a decreased fitness in all other stressful environments tested (i.e., BMN-erg6 lines), other lines showed a mixture of fitness costs and benefits in other environments besides nystatin, with some lines having high tolerance in all environments tested (i.e., BMN-erg5).
Parallel evolution is more likely to occur via loss-of-function mutations than via gain-of-function (Christin et al. 2010), and the different mutations that we observed in ERG6, ERG3, and ERG5 may well have caused loss-of-function alleles. Consistent with this hypothesis, knockout lines for these three genes (erg6Δ, erg3Δ, and erg5Δ) are viable and have been shown to increase fitness in low levels of nystatin in a screen of all deletion collection lines (Hillenmeyer et al. 2008). Our own nystatin tolerance assay of these null mutations in nystatin showed subtle but significant differences between BMN-erg6 and BMN-erg3 lines and the appropriate knockout lines (Figure 3). These results suggest that our mutations are similar, but not identical, to the null mutations and may indicate that the enzymes these genes encode retain some activity. Different amino acid changes in ERG6 have previously been shown to have different kinetic properties (Nes et al. 2004), and so we wish to be cautious in concluding that all of these mutations represent complete loss of function. By contrast to ERG6 and ERG3, we identified only one mutation in ERG5. The lack of parallel mutations in ERG5 is surprising, given that this is a longer gene (1616 bp) than either ERG6 (1151 bp) or ERG3 (1097 bp). Because erg5Δ is respiratory deficient (Merz and Westermann 2009), we checked 25 of the petite lines we isolated from our screens for sequence changes in ERG5, yet found no evidence of additional mutations in ERG5. The genomic DNA from four additional petite lines could not be amplified using two different sets of primers, and thus additional mutations in ERG5 may be present in those lines. The fact that BMN35 was not respiratory deficient and could grow in glycerol (unlike erg5Δ) indicates that this mutant is not equivalent to a whole-gene knockout. We thus conjecture that relatively few adaptive mutations were found in ERG5 because relatively rare changes were required to allow adaptation to nystatin without full loss of function.
BMN1 (with a mutation in ERG7) cannot be a loss-of-function mutation as erg7Δ is inviable. Consistent with this claim, the only mutation that arose in ERG7 was a nonsynonymous change very close to the end of the gene. It is thus plausible that this particular change was a gain-of-function mutation. Furthermore, our sterol profile of BMN1 is unique among the mutations we acquired. Similarly, BMN1 does not seem to share a sterol profile with any of the mutants identified in earlier studies on nystatin resistance, although sterol profiles in these early mutants match the profiles exhibited by our BMN lines that carry mutations in ERG6, ERG3, and ERG5 (Woods 1971; Bard 1972; Grunwald-raij and Margalith 1990). The unique phenotype generated by the mutation in ERG7 deserves future investigation. It may be that other gain-of-function mutations could have been beneficial in nystatin but were not sampled due to rarity or a bias toward large-effect loss-of-function mutations in our screens. Interestingly, S. cerevisiae knockout lines for all other genes that act late in the ergosterol pathway (except hmg1Δ and hmg2Δ, which are isozymes, so that deleting either one alone is not expected to have a strong effect on growth) are inviable [erg10Δ, erg13Δ, erg12Δ, erg8Δ, mvd1Δ, erg20Δ, erg9Δ, erg1Δ, erg11Δ, and erg25Δ (Giaver et al. 2002)], are unable to grow aerobically under our growth conditions [erg24Δ (Lees et al. 1995)], are ergosterol auxotrophs [erg2Δ (Parks and Casey 1995)], or have reduced fitness in nystatin [erg4Δ (Hillenmeyer et al. 2008)], helping to explain the narrow gene target of adaptation to nystatin that we have observed.
The distribution of fitness effects of beneficial mutations is an important factor that dictates how populations might adapt to a novel stressor. In his seminal work on adaptive mutations, Gillespie used the extreme value theory to suggest that one-step beneficial mutations might be expected to exhibit exponentially distributed selective effects (Gillespie 1983, 1984, 1991). The 20 unique one-step nystatin adaptive mutations we have acquired here do not immediately appear to fit this prediction, as we recovered an abundance of large-effect mutations (Figure 3) whose tolerance to nystatin far exceeds the 4 μM exposure concentration at which they were acquired. A number of explanations contribute to this finding. Our assay would not have detected small-effect mutations, because we required mutations of large enough effect to enable growth in 4 μM nystatin. Furthermore, the mutations are not independent, as many are in the same genes. As discussed above, we suspect that the 7 different mutations in ERG6 and 11 unique mutations in ERG3 are largely loss-of-function mutations in the ergosterol pathway. In any environment where large-effect loss-of-function mutations are available, many different nonsynonymous SNPs or indels could be selected, and we might expect these mutations to skew the distribution of beneficial effects toward the maximal fitness effect possible via eliminating the target of selection, here ergosterol. We also expect that the first mutations acquired have a distribution skewed toward large-effect mutations, compared to the distribution of all possible beneficial mutations, because of their selective advantage. Our results are thus more consistent with the extreme value properties in the Weibull domain (where there is a maximal fitness benefit) than in the Gumbel domain used by Gillespie (Joyce et al. 2008).
We found that tolerance across environments frequently exhibited gene–environment interactions, which were typically consistent across different mutations within the same gene. All BMN lines had a decreased growth rate and decreased biomass production (OD48) in the unstressful environment, YPD. Interestingly, we found no evidence that mutations with a larger benefit in nystatin had a greater negative effect in other environments. This is highlighted by a significant positive correlation between IC50 in nystatin and growth rate in YPD. When we examined growth in three other stressful environments (copper, ethanol, and salt), we found that beneficial mutations to nystatin had pleiotropic effects that differed substantially among environments (G × E). For example, while all ergosterol mutations examined were beneficial in nystatin, ERG6 and ERG7 mutations had reduced tolerance to ethanol, while ERG3 and ERG5 mutations were very similar in tolerance to the ancestor. The G × E interactions were so extreme that some mutations exhibited opposite selective effects in some environments. In particular, ERG6 mutations were less tolerant to copper, while ERG3, ERG5, and ERG7 were more tolerant. We call this phenomenon, where two mutations that are beneficial in one environment have selective effects that differ in sign in another environment, sign G × E [by analogy to “sign epistasis” (Weinreich et al. 2005)]. These experiments demonstrate that although adaptive mutations may show parallel phenotypes in a particular environment (here, in nystatin), effects in other environments of interest can be idiosyncratic and must be specifically examined.
Our results demonstrate that even with a narrow target for adaptation at the gene level (all 35 lines isolated in our screens carried mutations in only four different genes), mutations that appear phenotypically similar in one environment may well present variability in others. As a final example from our data set, although BMN35 with a mutation in ERG5 shares a similar IC50 phenotype with BMN-erg3 lines in nystatin, ethanol, and copper, it has a very different phenotype in salt. Consequently, different subsets of adaptive mutations are likely to be favorable under environmental conditions that require adaptation to more than one selective agent. Furthermore, our results demonstrate that if different first mutations are acquired by separated populations during adaptation to nystatin, this may well place different populations at different locations on the adaptive landscape following shifts in other environmental variables, altering the future evolutionary pathways accessible to these populations. The ability to sequence the entire genomes of multiple adapting lines provides an extremely useful way to explore the range of genetic pathways that evolution can take.
The authors thank R. FitzJohn for assistance with R; A. Kuzmin for Illumina library construction; N. Kane for many helpful discussions on the analysis of next generation sequence data; S. Lee, W. Li, A. Van Tol, M. Campbell, and J. Ono for laboratory assistance; and the Otto Laboratory group, F. DéBarre, J. Hill, and two anonymous reviewers for helpful comments on the manuscript. This work was supported by the Canadian National Science and Engineering Research Council (A.C.G. and S.P.O.) and the Killam Trusts (A.C.G.).
Communicating editor: J. J. Bull
- Received February 8, 2012.
- Accepted June 11, 2012.
- Copyright © 2012 by the Genetics Society of America