The Underlying Structure of Adaptation under Strong Selection in 12 Experimental Yeast Populations

Abstract

The aims of this study were to determine (i) whether adaptation under strong selection occurred through mutations in a narrow target of one or a few nucleotide sites or a broad target of numerous sites and (ii) whether the programs of adaptation previously observed from three experimental populations were unique or shared among populations that underwent parallel evolution. We used archived population samples from a previous study, representing 500 generations of experimental evolution in 12 populations under strong selection, 6 populations in a high-salt environment and 6 populations in a low-glucose environment. Each set of six populations included four with sexual reproduction and two with exclusively asexual reproduction. Populations were sampled as resequenced genomes of 115 individuals and as bulk samples from which frequencies of mutant alleles were estimated. In a high-salt environment, a broad target of 11 mutations within the proton exporter, PMA1, was observed among the six populations, in addition to expansions of the ENA gene cluster. This pattern was shared among populations that underwent parallel evolution. In a low-glucose environment, two programs of adaptation were observed. The originally observed pattern of mutation in MDS3/MKT1 in population M8 was a narrow target of a single nucleotide, unique to this population. Among the other five populations, the three mutations were shared in a broad target, sensing/signaling genes RAS1 and RAS2. RAS1/RAS2 mutations were not observed in the high-salt populations; PMA1 mutations were observed only in a high-salt environment.

INTRODUCTION

Experimental evolution of microorganisms provides the setting for capturing the underlying architecture of adaptation as it assembles over time. The process of adaptation to defined environments is now being observed in several laboratories (1) at high resolution in experimental microbial populations, with the yeast Saccharomyces cerevisiae as the best representative of eukaryotes (2, 3). In these experiments, populations are sampled and archived over time as they evolve. From the samples, representative genomes are resequenced and compared to that of the ancestor in order to identify the mutant alleles that have accrued.

In large microbial populations, we know that enormous numbers of mutant alleles are present at extremely low frequencies. In some populations, the number of mutational events that have occurred since inception is potentially greater than the number of nucleotide sites in the genome. For example, assuming (conservatively) a mutation rate of 4 × 10⁻¹⁰ (4) in a population with an effective size of 10⁷ propagated for 1,000 generations, about four mutations per site are expected to have occurred. This estimate must be taken as a rough approximation because mutation rates are not uniform across the genome. Nonetheless, the diversity in mutant alleles is a rich source of variation on which natural selection may act.

In large populations, genetic drift is negligible, and without positive selection, any mutant genotype has a vanishingly low probability of ever rising to a high frequency. Even for sites under selection, the majority of mutant genotypes quickly become extinct because of the regular dilution of the population that is imposed on batch cultures or continuous-flow chemostats to maintain constant population size (5, 6).

Of the emergent mutant genotypes that escape dilution and become established in a population, those that show upward trajectories in frequency are especially interesting. These genotypes match the expectation that fitness-enhancing, mutant alleles will emerge. For the candidate mutant alleles, we are now able to accurately measure alterations in molecular function, contribution to phenotype and fitness, and interaction with other alleles (7, 8).

Although experimental populations are artificial constructs, their population genetic structure is by no means simple (3, 9). When multiple genotypes of enhanced fitness are present, the capacity of any one of them to increase in frequency in the population will be less than it would have been in a background of only ancestral genotypes. This effect is known as clonal interference. Under clonal interference, another effect is that the frequency trajectory of any single genotype is likely to be less predictable.

In experimental evolution, two questions remain central. First, are the targets for selection broad, including numerous nucleotide sites in one or more genes, or narrow, including only one or a limited number of nucleotide sites (10)? Second, are the programs of adaptation repeated or unique among replicate populations (10)? The emerging pattern is that the mutations that drive adaptation tend to occur repeatedly within limited sets of genes. For example, repeated programs of adaptation have been observed in the evolution of antifungal drug resistance. In Candida albicans, two repeated patterns of adaptation to the drug fluconazole characterized by global changes in gene expression were seen across both experimental populations and populations residing in humans (11, 12). In Saccharomyces cerevisiae, two repeated patterns of adaptation were reported, and the two patterns were characterized by altered gene expression patterns and underlying mutations that were specific to the concentration and timing of drug supplementation (13). In the recent study on evolution in rich medium (3), selection was of a more general kind than that imposed by an antifungal drug. Here, a multitude of repeatedly mutated genes drove adaptation and were accompanied by a diversity of other mutant alleles interpreted as hitchhikers.

In this study, our objective was to observe the underlying structure of adaptation in the 12 experimental yeast populations with which we previously established causality between divergent adaptation in different environments and reproductive isolation (14)—the focal principle of ecological speciation. The 12 experimental populations (six labeled S1 to S6 and six labeled M7 to M12) were descended from a single diploid yeast cell, homozygous save for the mating-type locus (14). The 12 populations were allowed to evolve for 500 generations, 6 populations in a high-salt environment and 6 populations in a low-glucose environment. Over the course of experimental evolution, fitness increased substantially in all populations, and without exception, hybrids of populations evolved in different environments were of lower fitness than the original populations in their experimental environment. The reduced fitness of hybrids occurred by two means, ecological isolation as the “dosage” of adaptive alleles was divided in half relative to the evolved populations and epistasis via Dobzhansky-Muller incompatibilities (negative fitness interactions between certain pairs of mutant alleles that had arisen in different environments).

In a subsequent study (7), the underlying genetic determinants of adaptation in representative populations were identified by whole-genome sequencing of a single haploid endpoint representative of 3 populations selected from the original 12 populations: S2 and S6 from a high-salt environment and M8 from a low-glucose environment. Single-nucleotide polymorphisms (SNPs) were then deployed in crossing experiments in which the aggregate effects of each SNP on fitness, as well as interactions between mutations, were measured. In a high-salt environment, contributions to higher fitness came from different mutations between the two populations in PMA1, the main proton efflux pump acting with increased expression of the sodium efflux pumps ENA1, ENA2, and ENA5. The increased expression of the ENA genes in the S2 population was due to expansion in gene copy number in the tandem cluster and presumably by derepression by a mutation in the global transcriptional repressor CYC8. In the low-glucose population, two major determinants were found. The first was a mutation in MDS3 with global effects on transcription and increasing fitness primarily during the fermentative phase of growth. The second was a mutation in MKT1 that increases transcript level of genes important in respiration. The mutant alleles of MDS3 and MKT1 worked in synergy, in positive epistasis. The mutant allele of PMA1 in population S2 showed a negative epistatic interaction with the mutant allele of MKT1. We proposed that this was due to the abnormally low intracellular pH caused by the mutant allele of PMA1 and impaired expression of glucose transporters caused by the mutant allele of MKT1 (8). Were these patterns of adaptation general among our experimental populations?

The objectives of this study required sequencing enough representative genomes to determine (i) whether the mutational target on which selection acted was broad or narrow in the experimental populations and (ii) whether the programs of adaptation observed in populations S2, S6, and M8 are unique or general in the 12 populations. We sampled and resequenced genomes of 115 individuals from populations at generation 200 or 300 and at generation 500. For populations S2 and M8, we also sampled the bulk population at multiple time points to resolve the change in the frequencies of mutant alleles throughout the 500 generations. We found a repeatable pattern of evolution in the high-salt populations with a broad mutational target in the PMA1 gene. In all, we found 11 unique mutations of PMA1 in the six high-salt populations. In a low-glucose environment, mutations in the RAS1 and RAS2 sensing/signaling genes were common, one mutation in each of four low-glucose populations. The MDS3/MKT1 program of adaptation in population M8 was apparently unique to population M8 presumably due to the exceptionally small mutational target in MKT1—just one nucleotide site.

MATERIALS AND METHODS

Populations were sampled in two different ways for: (i) genotyping of individuals and (ii) estimation of the frequencies of mutant alleles in populations. For genotyping individuals, samples from generation 500 and an intermediate time point were removed from the −80°C archive and streaked on yeast extract-peptone-dextrose (YPD) agar medium. Single colonies were then used separately to inoculate 10 ml of liquid YPD. For estimating allele frequencies in populations, 100 μl was removed from the archive and used to inoculate 10 ml of liquid YPD. These bulk samples were taken at 100-generation increments throughout the 500-generation experiment (except for generation 300 in population S2, because the archive was not available).

DNA was extracted from overnight yeast cultures with a Gentra Puregene Yeast/Bact. kit (Qiagen), and paired-end libraries were prepared and subjected to Illumina sequencing. Alignment of the raw 100-base reads was to the yeast genome reference (R64-1-1_20110203) with Geneious software (v6.1.5). SNPs were called by setting an approximate P value of 10⁻³⁴, a minimum read frequency of 0.24, and coverage of a candidate SNP site by 12 or more reads. The probability of the value was set empirically by recovery of 10 SNPs that were already known to exist. The maximum P value for these SNPs was 10⁻³⁴.

This protocol generated several hundred SNP calls per sample, exceeding plausible expectations for an individual in a population evolving for such a relatively short time (10). Candidate lists of SNPs require filtering that can be based on replicated discovery (3) within a population or correction of misalignments (15). Several filtering rules were imposed here. First, any mutation that appeared in 3 to 12 populations was considered to be ancestral to the 12 populations and therefore not of direct interest in this study. Mutations of interest in this study were only those that occurred during the 500 generations of evolution. In all, there were approximately 500 SNPs that distinguished the ancestor of this experiment from the sequenced S288C reference, and the vast majority of these were found in all or most of the 12 populations. Second, we used a replication filter, reasoning that the SNPs of most interest in explaining adaptation would appear multiple times among the individuals sampled from any given population. Therefore, singleton and doubleton observations have been removed (Table 1; a corresponding table including the singletons and doubleton observations is included in Table S1 in the supplemental material). Third, genes such as those in the FLO and DAN gene families routinely returned large numbers of potential SNPs. These genes were not considered because they exist in duplicated copies that have diverged significantly from one another and because the reads frequently misaligned with the reference. Finally, in each population, the open reading frame of URA3 was replaced with one of four unique sets of 20-base bar codes flanked by either a G418 or nourseothricin (NAT) resistance cassette (eight unique combinations were used to mark the 12 populations). SNPs called at the beginning and end of the URA3 gene were attributable to alignment artifacts and were removed. We opted for one modification to these rules. Once a mutation in a gene passed the filters, we decided to list all singleton and doubleton mutations in the same gene in Table 1. This modification applied to only one gene, PMA1. For each of the 115 individual genomes and the 11 bulk genome sequences for the time course experiment, the presence of the correct bar code and resistance marker were confirmed by aligning raw sequence reads with the sequence of the KANMX5 resistance cassette (13).

TABLE 1.

SNPs in 12 yeast populations

Population (no. of recombination events detected)a	Chromosome	Base position	Base change	Amino acid change	Gene	No. of single colonies at generation 200d:						No. of single colonies at generation 500d:
						1	2	3	4	5	6	1	2	3	4	5	6
S1 (1)	VII	481076	G→T	P→T	PMA1			HE
	II	789822	T→A		SUL1	HO	HE		HE	HE		HE	HO		HE	HE
	V	108806	C→A	V→L			HE			HE		HE	HO	HE	HE	HE
S2 (2)	XIV	543273	A→G	D→G	LAP2b		HO					HE	HE	HE		HE
	X	456765	C→A	P→H	MET3b							HE
	VII	481966	G→C	S→C	PMA1b							HE
	VII	481077	C→A	M→I	PMA1									HE		HE
	III	261470	G→T	A→E	TUP1b				HE			HE			HE	HE
S3 (1)	VII	482288	C→A	V→F	PMA1							HE	HE			HE
	VII	481658	C→T	G→S	PMA1									HO
	VII	481786	C→A	G→V	PMA1										HE
	I	139656	C→T		SSA1	HO			HE							HE
S4	XII	1004775	C→A		CNA1	HO		HO	HO	HO	HO	HO	HO	HO	HO	HO	HO
	VII	482198	C→A	V→F	PMA1b							HE	HE	HE		HO	HE
	VII	189308	C→T		PMR1b							HE		HE			HE
S5	VII	480417	C→G	W→C	PMA1b							HE	HE	HE	HE	HE
	XIV	300340	G→T	Q→K	RHO5b							HE	HE	HE	HE	HE
	XII	323590	A→T	D→V	SUL2b							HE	HE	HE	HE	HE
	VII	1039818	G→A		ncdgc							HE	HE	HE	HE
	VII	984362	C→T		ncdg							HE	HE	HE	HE	HE
S6	XI	83451	C→A	N→K	CNB1	HE		HO							HE
	XV	174630	C→A		IRA2		HE					HE	HE
	IV	74835	C→A	D→Y	PRR2		HE					HE	HE	HE		HE
	IV	742173	G→A	R→C	SAN1		HE					HE	HE
	XVI	167249	T→G	L→V	TPK2	HE		HE							HE
	X	118286	A→T	N→K		HE		HE							HE
M7 (12)	VII	116963	C→G	S→C	ARO8							HE	HO		HE
	XI	272789	C→A		CUE2	HO	HE	HE				HE				HE
	V	195985	G→C	R→T	GPA2							HE		HO	HE
	IX	123071	T→C	L→P	KGD1							HE	HO		HE
	X	205579	T→C	F→S	MDV1	HE	HE	HE				HE		HO	HE
	X	495598	A→G	L→P	RAV1	HE						HE			HE
M8	VII	126867	T→G	F→V	MDS3b							HO	HO	HO	HO
	XIV	467219	A→G	D→G	MKT1b							HO	HO	HE	HO
	XV	432850	G→A	R→K	SGT1b							HO		HE	HO
M9	XV	794703	A→G		DGA1	HO	HO	HO	HO	HO		HO	HO	HO	HO
	XV	515374	A→C	E→A	RAS1b	HE				HE		HE	HO	HO	HO
M10 (11)	V	113456	A→T	K→I	GEA2	HO	HO		HO	HO		HO	HO	HO	HO	HO
	XV	509639	A→T	C→T	NUP1	HE	HE	HE	HE	HE			HO	HO	HE
	XIV	440441	C→G	E→Q	RAS2b							HE		HE	HE
	XVI	490702	T→G	R→S	SVL3	HE	HE		HO	HE		HE	HO	HO
	VII	1030562	T→C		YTA7	HO	HO	HO	HO	HO			HO	HO	HO	HO
	IV	1165639	C→A		ncdg				HE	HE		HE	HO		HE
M11	IV	1314162	C→A	A→E	ARO80								HE	HE	HE	HE
	XIII	386688	G→A		BUB2								HE	HE		HE
	XI	272789	C→A		CUE2	HE	HE	HE	HE			HE	HE	HE	HE	HE
	III	28065	C→A	T→K	KAR4								HE	HE		HE
	XIV	440516	C→T	G→S	RAS2b								HE	HE	HE	HE
	V	395644	G→T	Q→K	SLX8								HE	HE	HE	HE
	XI	532009	A→G	Y→C								HE	HE	HE	HE	HE
M12	XV	794703	A→G	Y→C	DGA1	HO	HO	HO	HO			HO	HO	HO	HO
	XVI	609562	C→G	S→T	EAF3				HE			HE	HE	HE	HE
	XV	515374	A→C	E→A	RAS1b	HE			HE			HE	HE	HE	HE

^aNumber of recombination events detected. Examples: S1, chromosome II base position 789822 (II 789822) and chromosome V base position 108806 (V 108806); S2, XIV 543273 and III 261470; S2, VII 481077 and III 261470; S3, VII 482288 and I 139656; M7, VII 116963 and XI 272789; M7, VII 116963 and V 195985; M7, VII 116963 and X 205579; M7, VII 116963 and X 495598; M7, XI 272789 and V 195985; M7, XI 272789 and IX 123071; M7, XI 272789 and X 205579; M7, XI 272789 and X 495598; M7, V 195985 and IX 123071; M7, V 195985 and X 495598; M7, IX 123071 and X 205579; M7, IX 123071 and X 495598; M10, V 113456 and XV 509639; M10, V 113456 and VII 1030562; M10, XV 509639 and XIV 440441; M10, XV 509639 and XVI 490702; M10, XV 509639 and IV 1165639; M10, XIV 440441 and XVI 490702; M10, XIV 440441 and VII 1030562; M10, XIV 440441 and IV 1165639; M10, XVI 490702 and VII 1030562; M10, XVI 490702 and IV 1165639; M10, VII 1030562 and IV 1165639.

^bSNP verified by PCR and Sanger sequencing.

^cncdg, noncoding.

^dNumbers designate individual colonies. Up to six colonies were selected for each population at each sampling time. HO, SNP homozygous; HE, SNP heterozygous.

RESULTS AND DISCUSSION

The prevailing pattern of adaptation in the 12 populations examined here was parallel among replicates reared in the same environment, but not among those reared in different environments. The mutations that drove adaptation occurred at a diversity of sites within a limited set of genes, of which the most repeatedly hit were PMA1 in a high-salt environment and RAS1 and RAS2 in a low-glucose environment. Another program in the M8 population, with mutations in MDS3 and MKT1, had a positive and large effect on fitness but occurred only once, a tiny target for mutations conferring a fitness benefit.

Among the 53 mutations in the filtered list (Table 1), all but two pairs were unique to a population. The exceptions were an A-to-C base change in the same site in RAS1 in populations M9 and M12 and an A-to-G change in DGA1 also in populations M9 and M12. Of the 53 mutations, 3 were noncoding, 11 were coding and synonymous, and 39 were coding and nonsynonymous. Of the 50 mutations in coding regions, only 3 (all nonsynonymous) were in unnamed genes.

In the original experiments, eight populations were propagated with regular rounds of mating and sporulation (S1 to S4 and M7 to M10) and four populations (S5, S6, M11, and M12) were strictly asexual. In the mating populations, we broke up the tetrads and mixed the spores to promote outbreeding; in nature, most mating is intratetrad, and outbreeding is rare. We therefore checked the SNP data for consistency with the expected signatures of sexual versus asexual reproduction with the four-gamete test of Hudson and Kaplan (16). Genetic exchange and recombination are expected to produce a subset of genotypes with all four genotypic combinations of the two alleles. Such examples can also arise from recurrent mutation, but this is a low probability in these populations and can be discounted as a plausible explanation here. The four-gamete test was originally designed for haploids, but nonetheless, it can be applied to the diploids here because genetic exchange between individuals and recombination would still be required to produce all four combinations of alleles.

In applying the four-gamete test, we know that each population carries the ancestral-ancestral combination for each pair of sites from the outset; we need to check only for the remaining three combinations in the data. Consistent with the experimental design of sexual versus asexual reproduction, there were 27 examples of recombination between pairwise combinations of bases at SNP sites in the eight sexual populations and none in the four asexual populations (Table 1).

High-salt populations.

The most prominent feature of adaptation to high-salt levels was recurrent mutation in PMA1, which encodes the essential, ATP-driven, proton efflux pump. Table 1 lists eight unique mutations in PMA1 found in the single-colony isolates of populations S1 to S6, and Table 2 includes these eight mutations plus three additional PMA1 mutations. One mutation in the S6 population was detected by Anderson et al. (7) but was not recovered in the present sample, and two mutations were detected only in the time course experiment in population S2, but not in the single-colony isolates listed in Table 1. The mutations occurred in a diversity of domains of PMA1, including transmembrane, cytoplasmic, and extracellular domains. A total of four amino acid sites listed in Table 2 were also altered in the extensive mutagenic study of Morsomme et al. (17), but to a different amino acid. One amino acid change in Table 2 matched one reported by Morsomme et al. (17); in that study, this amino acid mutation had no effect on ATPase function or on proton pumping.

TABLE 2.

SNPs in PMA1

Population	Base position on the chromosome	Base position on the gene	Base change	Amino acid position	Amino acid change
S3	482288	381	C→A	127	V→Fa
S4	482198	471	C→A	157	V→Fc
S2	481966	702	G→C	234	S→Ca,b
S3	481786	881	C→A	294	G→V
S3	481658	1009	C→T	337	G→Sa
S6	481579	1088	A→C	363	L→Wa,d
S2	481077	1590	C→A	530	M→I
S1	481076	1605	G→T	535	P→T
S2	480976	1691	A→G	564	I→Te
S2	480472	2195	G→A	732	A→Ve
S5	480417	2250	C→G	750	W→C

^aAmino acid position mutated in Morsomme et al. (17) but to a different amino acid than the one predicted here.

^bEffects of this mutation on proton transport and intracellular pH reported by Parrieras et al. (8).

^cAmino acid position mutated in the study of Morsomme et al. (17) to the same amino acid as the one predicted here.

^dFound in the study of Anderson et al. (7), but not in this study.

^eFound in the time course experiment (Fig. 1), but not among the single-colony isolates (Table 1).

Mutations were detected in additional genes with direct or indirect effects on sodium efflux. These include mutations potentially having a major effect on the expression of the sodium efflux genes ENA1, ENA3, and ENA5. A mutation in the TUP1 repressor gene in the S2 population specifies a drastic amino acid substitution possibly resulting in loss of function (i.e., loss of repression). Another mutation in CYC8 had been observed in population S6 by Anderson et al. (7) but was not detected in the present study. The products of TUP1 and CYC8 act as corepressors; presumably loss of function in either one could result in loss of repression and overexpression of the ENA genes.

Two additional mutations were observed in genes that play important roles in sensing and signaling: RHO5, which encodes a GTPase, Ras-like protein involved in PKC1 signal transduction, with effects on cell integrity in population S4, and IRA2, which encodes a GTPase-activating protein controlling reduction of cyclic AMP (cAMP) levels under nutrient-limiting conditions (descriptions from the Saccharomyces Genome Database [http://www.yeastgenome.org/]).

Low-glucose populations.

There were two patterns of adaptation in the low-glucose populations, one present in four different populations and an entirely different pattern of adaptation unique to population M8. Representing the pattern of adaptation in the four populations, there were four mutations in the RAS1 and RAS2 genes, which are duplicates. Both of these genes encode GTPases and function in G-protein signaling with downstream effects on cell proliferation (descriptions from the Saccharomyces Genome Database [http://www.yeastgenome.org/]). One potential anomaly was that identical RAS1 mutations were found in M9 and M12 populations. This would be consistent with recurrent mutation and parallel selection, but for the observation that all representatives of the two populations also share the same mutant allele of DGA1. An alternate possibility is that these RAS1 and DGA1 mutations may have occurred during time of culturing required for insertion of the bar codes and antibiotic resistance tags before the experimental populations were initiated: populations M9 and M12 share the same bar codes and antibiotic resistance tags (i.e., eight unique marker combinations were used for the 12 populations). The possibility of mutations during the culture phase in preparation for experimental evolution has been considered before (2). Nonetheless, the RAS1 mutation is likely to be important in adaptation in M9 and M12 populations, because the mutant allele shows an increase in frequency between generation 200 and 500 (Table 1). Setting aside this possible anomaly still leaves three RAS1 and RAS2 mutations in the low-glucose populations and none in the high-salt populations.

The mutations in MDS3 and MKT1 had already been observed in population M8 (7) and are known to exert strong effects on fitness in the fermentative and respiratory phases of growth, respectively. That these mutations were unique to population M8 may be due to their narrow targets. The MKT1 mutation at position 89 evolves repeatedly in the presence of the MDS3 mutation in recapitulation experiments (8). This change does not appear when the background genotype is ancestral for MDS3. The mutation observed in MKT1 in experimental evolution appears to represent a reversion to what is actually the wild-type allele, which is fixed in natural populations of S. cerevisiae and Saccharomyces paradoxus (18). The status of the other mutations listed in Table 1 can be searched in genome databases from the study of natural populations of S. cerevisiae and S. paradoxus by Liti et al. (18).

Time course of adaptation in populations S2 and M8.

The SNP detection in single-colony isolates provided multilocus, genotypic information on individuals. To examine the temporal dynamics in individual allele frequencies, we sequenced DNA collected from bulk samples of two populations, S2 and M8 at 100-generation time increments over the 500-generation experiment. In population S2, the pattern was one of competition among four different mutations in PMA1 (Fig. 1). Interestingly, the mutation in the repressor TUP1 became associated with at least two different PMA1 mutations; this was likely the result of genetic exchange and recombination because this population included sporulation and intertetrad mating. The effect of mutation in MET3 was auxotrophy for methionine. The MET3 mutation has been shown (7) to act synergistically with mutations in two other genes, GCD2 and PMA1, to increase fitness in a high-salt environment, with the strong pleiotrophic effect of no growth in minimal medium. In a previous study (7), the mutant LAP2 allele had no significant effect on fitness, and it may merely be a hitchhiker. Alternatively, variation in the LAP2 protein is correlated with salinity levels in mussels and is postulated to play an adaptive role (19).

FIG 1.

SNP frequency and ENA gene copy number over time in high-salt population S2. Mutant alleles are designated with the gene name followed by the letter “e” which stands for “evolved.” The four PMA1e alleles are shown with the amino acid position affected before the final “e” (Table 2). All SNPs are listed in Table 1. The ENA gene copy number reflects the number of Illumina reads aligning to the ENA region, divided by the total number of Illumina reads aligning to the entire chromosome, all normalized as a percentage to the maximum value at generation 500.

The ENA genes have been shown (7) to increase in copy number. This expansion likely arose in a single event because the trait segregated as a single Mendelian determinant. Here we mapped the numbers of Illumina reads of ENA1, ENA2, and ENA5 in the reference genome as a fraction of the total number of reads mapping to chromosome 4. The increase in ENA gene copy number occurred late, from generations 400 to 500.

In contrast to the prevailing pattern of clonal competition in population S2 (Fig. 1), a single program of adaptation characterized population M8 (Fig. 2). Here, as expected from references 1 and 18, the mutant allele in MDS3 rose in frequency first and was followed by the rise of the mutant allele in MKT1. The temporal order of first appearance cannot be inferred from Fig. 2 and Table 1 alone, but this is known from two previous studies (7, 8). Although t this program of adaptation is rare, we conclude that it was powerful, because one genotype rose to near fixation, with no evidence of clonal completion among genotypes. The mutation in SGT1 was observed earlier, but its contribution to fitness increase appears to be nil.

FIG 2.

SNP frequency over time in low-glucose population M8. Mutant alleles are designated with the gene name followed by the letter “e” which stands for “evolved.” These SNPs are also listed in Table 1.

Conclusions.

In both S and M populations, evolution was clearly parallel at the level of the target genes, but not at the level of the nucleotide sites. In a high-salt environment, adaptation repeatedly involves mutation in PMA1 (broad target for mutation) often coupled with increased expression of the ENA genes either through gene cluster expansion and/or derepression. In a low-glucose environment, the most common theme is mutations in genes involved with sensing and control of cell proliferation, while the rare but most effective program of adaptation involved major increases in fermentative and respiratory growth with the underlying changes representing very narrow targets.

The main contribution of this study is in documenting prevailing patterns of adaptation in a series of experimental populations in which the link between divergent adaptation and reproductive isolation, a core principle of speciation, was first established. We now have a compendium of mutant alleles and multilocus genotypes that are either verified or strongly suspected as underlying determinants of adaptation in the two environments. That we have sequenced enough representative genomes to detect examples of parallelism at the level of the gene, especially PMA1 and RAS1/RAS2, is an additional contribution of this study. Of course, additional programs of adaptation, as well as more examples of parallelism in adaptation may well exist in the two environments. At first glance, it might seem best to examine these possibilities in populations of even greater size than examined here. In principle, as population size increases, evolution should become more deterministic, as the best possible genotype ultimately wins. In practice, raising population sizes further is not likely to resolve the issue of undetected programs of adaptation and parallelism because of increased clonal interference as multiple genotypes of enhanced fitness compete with one another. With clonal interference, fitness relationships among genotypes may be complex and not necessarily transitive (e.g., A is better than B, B is better than C, but A may not necessarily be better than C); the best genotype does not always win. We conclude that larger numbers of replicate populations will be needed to discover additional programs of adaptation that may exist and to measure more accurately the extent of parallelism between programs.