Concerted Evolution of Life Stage Performances Signals Recent Selection on Yeast Nitrogen Use

Abstract

Exposing natural selection driving phenotypic and genotypic adaptive differentiation is an extraordinary challenge. Given that an organism’s life stages are exposed to the same environmental variations, we reasoned that fitness components, such as the lag, rate, and efficiency of growth, directly reflecting performance in these life stages, should often be selected in concert. We therefore conjectured that correlations between fitness components over natural isolates, in a particular environmental context, would constitute a robust signal of recent selection. Critically, this test for selection requires fitness components to be determined by different genetic loci. To explore our conjecture, we exhaustively evaluated the lag, rate, and efficiency of asexual population growth of natural isolates of the model yeast Saccharomyces cerevisiae in a large variety of nitrogen-limited environments. Overall, fitness components were well correlated under nitrogen restriction. Yeast isolates were further crossed in all pairwise combinations and coinheritance of each fitness component and genetic markers were traced. Trait variations tended to map to quantitative trait loci (QTL) that were private to a single fitness component. We further traced QTLs down to single-nucleotide resolution and uncovered loss-of-function mutations in RIM15, PUT4, DAL1, and DAL4 as the genetic basis for nitrogen source use variations. Effects of SNPs were unique for a single fitness component, strongly arguing against pleiotropy between lag, rate, and efficiency of reproduction under nitrogen restriction. The strong correlations between life stage performances that cannot be explained by pleiotropy compellingly support adaptive differentiation of yeast nitrogen source use and suggest a generic approach for detecting selection.

selection, adaptation, life history, nitrogen, yeast, metabolism

Introduction

Exposing natural selection and environmental factors driving adaptive differentiation of phenotypes and genotypes is an extraordinary challenge. The classical approach aims to unveil covariations in phenotype, genotype, or environmental factors with a reasonable proxy for fitness. Unfortunately, spurious associations emerge due to genetic drift and population structure. Genetic hitchhiking, whereby nonadaptive alleles and their phenotypes piggyback with adaptive variants at nearby loci, compounds the problem (Barton 2000), as do pleiotropy, whereby nonadaptive phenotypes hitchhike with adaptive by association to the same gene variant (Stearns 2010). Consequently, only a fraction of environmental factors, genotypes, and phenotypes that covary with a fitness proxy is directly linked to selection. An elegant approach toward exposing selection on a phenotype was taken by Orr (1998). If alleles (quantitative trait loci [QTLs]) enhancing a trait value are consistently found in one as compared with another lineage of a species, then positive selection is likely to have acted on that trait. Nevertheless, for most traits in most species, too few QTLs are known for the test to constitute a powerful alternative. Grouping traits into aggregates circumvents this lack of power (Fraser et al. 2010). However, it is rarely clear how a functionally relevant trait grouping is to be achieved.

To be exposed to selection in a particular environment, alleles have to alter net population growth, i.e., birth or death, in that environment. All organisms pass through life cycles composed of distinct life stages (Stearns 1992). Success in any of these life stages, either in the form of increased birth relative death or faster progression to a subsequent life stage in which reproduction occurs, directly affects net population growth (Roff 1992). Performance in these life stages is therefore the most immediate components of fitness and lower level phenotypes affect fitness via them. In many organisms, life stages are intimately linked such that they consistently are subject to the same environmental variations and the same selective pressures. We therefore reasoned that if an environment exposes a population to a long-term selection, optimization of performance in multiple life stages would often emerge. In absence of selection however, such fitness components would tend to fluctuate independently of each other, as dictated by the specifics of genetic drift, genetic hitchhiking, and pleiotropy. This conjecture directly suggested a test for adaptive differentiation in the form of robust correlations between fitness components. To be valid, this test requires fitness components to be genetically independent, as confounding correlations due pleiotropy and genetic hitchhiking otherwise could emerge.

To explore the power of this test, we studied natural isolates of the model yeast Saccharomyces cerevisiae, which has a well-understood life history with fitness components that are straight forward to define, measure and dissect genetically. Sex in natural S. cerevisiae is rare and heavily oriented toward self-fertilization (Ruderfer et al. 2006; Tsai et al. 2008). Selection is therefore expected to act primarily on the asexual life cycle (Warringer and Blomberg 2014). Accordingly, the time it takes for an asexual population to pass through the lag phase to the reproductive stage (lag time), its net rate of growth in the reproductive stage (doubling time), and the total population density achieved before the reproductive stage ends (efficiency) are key fitness components. Population growth of chemoheterotrophic microbes, such as yeasts, is limited by access to nutrients, primarily energy, carbon, reducing equivalents, and nitrogen. Saccharomyces cerevisiae obtains energy, carbon, and reducing equivalents from sugar, while using a wide diversity of nitrogen sources to satisfy its nitrogen needs (Cooper 1982). Decomposing or damaged fruit, flowering plant nectar, and tree saps (exudates) are primary microhabitats occupied by yeast (Landry et al. 2006; Hittinger 2013). These are rich in sugar but poor in nitrogen, with one or a few nitrogen sources dominating each microhabitat (supplementary fig. S1A and Supplementary Data, Supplementary Material online) (Gardener and Gillman 2001). Selection for optimal use of single nitrogen sources can therefore be expected to have been strong in natural yeasts and to have driven recent adaptive differentiation. To evaluate the validity of the proposed test for selection, we therefore measured asexual population growth of natural yeasts in a vast array of nitrogen-restricted, single nitrogen-source environments.

Results

Natural Yeasts Are Highly Differentiated with Regard to Nitrogen Source Use

To evaluate fitness components of natural yeasts we considered four S. cerevisiae strains, each representing one of the nonreproductively isolated yeast populations discovered outside China: the West African DBVPG6044 (WA), the North American YPS128 (NA), the European DBVPG6765 (E), and the Sake Y12 (S). Lineages differed genetically by 0.3–0.7%, corresponding to several million generations of evolution, and encompassed >50% of the known SNP and phenotypic variation outside China (Liti et al. 2009; Warringer et al. 2011; Bergstrom et al. 2014) (supplementary fig. S1C, Supplementary Material online). Strains were clonally propagated in 28 nitrogen environments corresponding to all low-complexity nitrogen sources used by yeast. These were present at equal and yield-limiting nitrogen concentrations (supplementary fig. S1D and Supplementary Data, Supplementary Material online). From high-resolution growth curves, it was abundantly clear that natural strains vary greatly in their capacity to use different nitrogen sources (fig. 1A). Quantifying these variations, we resolved the growth supported by the various nitrogen sources into its components and extracted the rate (population doubling time), lag (time to initiate growth), and efficiency of population expansion (total change in population density) (fig. 1B). Variations among natural lineages were pervasive with significant strain differences (false discovery rate [FDR], q ≤ 5%) in 45% of the 84 fitness component measures (fig. 1C and supplementary fig. S2, Supplementary Material online). The lower the mean capacity to convert nitrogen into growth, the larger was the variation between strains, with the West African standing out as a generally poor performer (fig. 1D and E). Thus, natural S. cerevisiae strains are highly differentiated with regards to nitrogen source use and the differentiation is strongest for poor nitrogen sources. The latter is in agreement with recent wine yeast observations (Gutierrez et al. 2013). Given that entry into stationary phase could occur for reasons other than nitrogen depletion, we investigated whether total population densities achieved truly reflected exhaustion of nitrogen. Surveying the nitrogen remaining in the medium as a function of time in strains WA and NA, we found nitrogen depletion to neatly coincide with exit from the exponential growth phase (fig. 1F). In fact, 10–35% of the total population expansion occurred after this time point, presumably by mobilization of stored intracellular nitrogen. Thus, stationary phase levels reflected depletion of not only external but also internally stored nitrogen.

Fig. 1.

Natural yeasts are highly differentiated for nitrogen source use. (A) Population density as a function of time for Y12 (S), YPS128 (NA), and the S288C derivative YSBN in sample single nitrogen-source environments (n = 2). (B) Extraction of lag (time to initiate proliferation), rate (population doubling time), and efficiency (total change in population density) of population growth from high-density growth curves. (C) Natural yeasts vary in capacity to convert different nitrogen sources into population growth. Nitrogen sources significantly (FDR, q < 5%) better (red) or worse (green) for one natural yeast lineage (n = 6) than others (n = 18) are shown. (D) Variation in nitrogen source utilization between natural yeast lineages is largest for poor nitrogen sources. Variance over the four natural strains is plotted as a function of mean. Population growth efficiency is displayed. Line = linear regression. (E) The North American is superior and the West African inferior at utilizing single-nitrogen sources. Fraction of fitness component measures in which each strain (n = 6) is significantly (FDR, q < 5%) better (blue) or worse (red) than other natural yeasts (n = 18) is shown. (F) Yeast variation in total population density reached reflects variations in nitrogen use efficiency. Population size (left y-axis, continuous bold lines) and nitrogen remaining in the medium (right y-axis, broken bold lines) was quantified as a function of time in West African and North American strains. Broken lines = time of external nitrogen depletion.

Concerted Selection on Lag, Rate, and Efficiency of Yeast Nitrogen Use

To test whether the differentiation of nitrogen source use was caused by adaptation and selection, we compared measures of lag, rate, and efficiency of population growth over all nitrogen-restricted environments and strains. Correlations between all pairs of fitness components were excellent (fig. 2A). Thus, the efficiency of population growth covaried extensively (Pearson, r = 0.85 and 0.61) with both the rate and lag. This correlation between fitness components has two components: covariation that is due to genetic variants that are shared between strains and covariation that is due to genetic variants that differ between strains. The former corresponds to concerted evolution of fitness components in the ancestral lineage. The latter corresponds to concerted evolution of fitness components in lineage(s) after their separation. As >98% of nucleotides are shared between strains, we expect much of the covariation between fitness components to arise from these invariant positions. We can estimate this ancestral coevolution by considering trait averages, over the four strains, as a proxy for ancestral traits. Comparing trait averages for different fitness components over the different environments, we find these to be strongly correlated (supplementary fig. S3A, Supplementary Material online). Thus, fitness components have largely coevolved with regards to nitrogen source use in the ancestor of current yeasts. Fitness component coevolution after the separation of lineages would be reflected in a correlation over the four strains. Given the low sample size (four), which adds substantial randomness that hides any existing covariation, we cannot provide an accurate estimate of recent fitness component coevolution. The average correlation (r = 0.30, considering each environment and pair of fitness component independently) is likely to be a substantial underestimate and should be regarded as lower bound for coevolution. Under an assumption of variations in fitness components being due to different alleles, these correlations are challenging to explain without invoking selection. However, if underlying alleles are pleiotropic and consistently affect multiple fitness components in the same direction, selection need not necessarily be involved. Rejecting pleiotropy is not straight forward for alleles that are shared between strains. However, it can be done for alleles that differ between strains because following mating between two strains, we expect variations in two fitness components to cosegregate in offspring recombinants only if they are explained by the same, or genetically linked, alleles. We therefore mated the four natural lineages in all six pairwise combinations. Resulting hybrids were sporulated to obtain 92 F1 meiotic progenies from each cross. F1 segregants were genotyped at nucleotide sites for which parents were known from genome sequencing to be polymorphic (Cubillos et al. 2011). We then precisely followed the net population growth of these 552 recombinants in all nitrogen environments, extracting >42,000 fitness component measures. Finally, we evaluated the cosegregation of each SNP marker and each life history trait by QTL linkage mapping, for each cross and nitrogen environment separately. Overall, we detected 230 robust (permutation test, α = 0.05) QTL. 87.4% of all QTL were unique for a single fitness component (fig. 2B), rejecting the null hypothesis of genetic variants with pleiotropic effects on fitness components. For example, QTLs private to lag, rate, or efficiency determined population growth in the West African × Wine/European cross in citrulline (fig. 2C), leucine, and isoleucine (supplementary fig. S3B, Supplementary Material online). In agreement with this fitness component specificity of QTLs, only weak correlation (r = 0.15) between fitness components remained in the average environment and cross. Together with the strong correlations between natural isolates, this supports that natural variation in yeast nitrogen source use indeed reflects adaptive differentiation.

Fig. 2.

Concerted selection on lag, rate, and efficiency of yeast nitrogen use. (A) The efficiency, rate, and lag of population growth are strongly correlated in natural yeast strains. The log₂ of population growth efficiency, rate, and lag were pairwise compared over all natural yeasts and nitrogen environments. Means (n = 6), Pearson correlations (r), and linear regressions are shown. (B) Efficiency, rate, and lag of nitrogen source use are genetically independent. QTL (permutation test, α = 0.05) private to a fitness component or shared between components were summed over six yeast crosses and 28 nitrogen environments. (C) QTL private to a single fitness component control variation in the efficiency, rate, and lag of citrulline use in the WE × WA cross. Logarithm of the odds (LOD) scores represent cosegregation of growth and genetic markers in 92 F1 recombinants. Bands represent chromosome boundaries. Dotted lines show threshold for significance (permutation test, α = 0.05). (D) Nitrogen use QTL is private to a single nitrogen source. Fraction of total QTL detected for one or more than one nitrogen source is displayed. Each cross and fitness component was considered separately. There is a weak support for more environment pleiotropy than expected by chance (Mann–Whitney–Wilcoxon, P = 0.016).

QTLs were largely unique to a single nitrogen source (fig. 2D and supplementary fig. S4A, Supplementary Material online), meaning that different alleles tended to control variation in the use of different nitrogen sources and are acted upon independently by selection. Nevertheless, a few extremely pleiotropic QTL, affecting the same fitness component in a large number of nitrogen-restricted environments, were observed. For example, a single QTL on chromosome VI accounted for poor growth efficiency of the Wine/European strain in almost all nitrogen-restricted niches, suggesting a general relaxation of selection for efficient nitrogen use in this strain. Predominantly, QTL also emerged only in a single of the three crosses in which a particular genetic background was represented (supplementary fig. S4B, Supplementary Material online). Therefore, the penetrance of the underlying alleles was highly dependent on genetic context, suggesting widespread epistasis. Although the results reported above provide strong indications, the limitations of QTL data should be acknowledged. Due to lack of power, detected QTLs do not explain all of the heritable variation in traits (Bloom et al. 2013). Furthermore, the breakup of parental allele structures during meiosis and the emergence of novel allele combinations can both disrupt and promote epistasis, affecting trait values. Finally, QTL represents the combined effect of all alleles in a region. These effects call for some caution and means that stringent conclusions can only be reached by identifying and measuring the effects of individual mutations.

Alleles Private to a Single Fitness Component Control Natural Variation in Yeast Nitrogen Source Use

To address the shortcoming of the QTL analysis, we traced the genetic basis of three QTLs down to single alleles. All nonessential genes within these three QTL regions were identified. We then crossed reference strain (BY4741) gene knockouts for each of these genes to both well and poor-performing natural strains, resulting in pairs of diploid hemizygotes. A trait difference within such hemizygotic pairs was taken to imply that the two alleles at the hemizygotic locus affect the trait differently. To exclude confounding effects from haploinsufficiency, we next mated a poor- and well-performing natural strain. In two of the resulting diploid hybrids, we reciprocally deleted either of the two alleles. A trait difference between these was considered final confirmation of the causative loci. We first focused on the chromosome VI QTL controlling variation exclusively in population growth efficiency in all crosses involving the Wine/European strain, and in almost all nitrogen environments (fig. 3A and supplementary fig. S5A, Supplementary Material online). Hemizygotic diploids only containing the WE allele of RIM15 mimicked the poor growth efficiency of the WE parent (fig. 3B). Furthermore, the WE RIM15 consistently imposed poor efficiency on hemizygotes from crosses with other natural yeasts (fig. 3C and supplementary fig. S5B, Supplementary Material online). In all cases, RIM15 fully explained the defects of the WE parent. The WE RIM15 contains an early two base pair insertion, rim15c.459_460insCA, shifting the reading frame to cause an early stop codon (supplementary fig. S5C, Supplementary Material online). Thus, WE RIM15 is a null allele. No rate or lag QTL were found in the RIM15 region in the relevant crosses and nitrogen sources and diploids only containing the WE RIM15 allele showed no impairments of the rate or lag (supplementary fig. S5D and Supplementary Data, Supplementary Material online). RIM15 encodes a poorly understood protein kinase known to control stationary phase entry (Wanke et al. 2005) and sporulation efficiency (Bergstrom et al. 2014). As sporulation is induced by nitrogen depletion, it is tempting to speculate that rim15 sporulation defects arise as a consequence of nitrogen use impairment.

Fig. 3.

Alleles private to single fitness components control variation in yeast nitrogen source use. (A–C) RIM15 accounts for poor population growth efficiency of the WE lineage during nitrogen limitation. (A) LOD score plots of the cosegregation of growth efficiency and genetic markers among 92 F1 recombinants from the WE × S cross in proline. Bands represent chromosome boundaries and dotted line shows threshold for significance (permutation test, α = 0.05). (B) Proline growth efficiency of diploid hemizygotes retaining only the WE (blue) or S (red) allele in crosses between WE or S and single BY4741 gene deletions. Gene deletions correspond to nonessential genes in the chromosome VI QTL region. Mean (n = 2) log₂ values were normalized to those of the YSBN control (n = 4). Error bars = SEM. (C) Proline growth efficiency of reciprocal diploid hemizygotes retaining only the WE or only the alternate RIM15 allele in crosses between WE and other natural lineages. Mean (n = 8) log₂ values were normalized to those of the YSBN control (n = 5). Error bars = SEM, P-values = Student’s t-test. (D–F) PUT4 allele accounts for poor proline growth rate in the WA lineage. (D) LOD score plot of cosegregation of proline growth rate and genetic markers among 92 F1 recombinants from the WA × WE cross. Bands represent chromosome boundaries. Dotted line indicates threshold for significance (permutation test, α = 0.05). (E) Proline growth rates of diploid hemizygotes retaining only the WE (blue) or WA (red) allele in crosses between WE or WA and single BY4741 gene deletions. Gene deletions correspond to nonessential genes in the chromosome XV QTL region. Mean (n = 2) log₂ values were normalized to those of the YSBN control (n = 4). Error bars = SEM, P-values = Student’s t-test. (F) Proline growth rates of reciprocal diploid hemizygotes retaining only the WA or only the alternate PUT4 allele in crosses between WE and other natural lineages. Mean (n = 8) log₂ values were normalized to those of the YSBN control (n = 5). Error bars = SEM (n = 2), P-values = Student’s t-test. PUT4 explained 97 ± 6% of the WA–NA, 67 ± 4% of the WA–WE, and 54 ± 3% of the WA–S variation. (G–I) A frameshift mutation in DAL4 and a nonsynonymous mutation in DAL1 both impair WA growth rate on allantoin. (G) LOD score plot of cosegregation of allantoin growth rate and each genetic marker among 92 F1 recombinants from the WA × WE cross. Bands represent chromosome boundaries. Dotted line shows threshold for significance (permutation test, α = 0.05). (H) Allantoin growth rates of diploid hemizygotes retaining only the WE (blue) or WA (red) allele in crosses between WE or WA and single BY4741 gene deletions. Gene deletions correspond to nonessential genes in the chromosome IX QTL region. Mean (n = 8) log₂ values were normalized to those of the YSBN control (n = 5). Error bars = SEM. (K) Addition of the S288C DAL1 or DAL4 to a haploid WA via a centromeric plasmid and replacement of the candidate SNP in the other DAL gene with the S288C nucleotide variant identified dal4c.1201delA and dal1c.415C>T as independently impairing the allantoin growth rate of WA. Mean (n = 6–12) log₂ population growth rates are displayed. Error bars = SEM.

We next considered the strong chromosome XV QTL affecting population growth rate on proline in all crosses involving the West African DBVPG6044 (fig. 3D). Hemizygotic diploids from crosses with reference strain gene knockouts identified the West African PUT4, encoding a high-affinity proline permease (Andreasson et al. 2004), as the only allele in the region contributing to the trait variation (fig. 3E). Reciprocal PUT4 hemizygotes from crosses between the West African and other natural strains confirmed that the West African PUT4 impairs proline growth rate, accounting for 54–96%, depending on cross, of the West African defect (fig. 3F). No efficiency or lag proline QTL was found in the PUT4 region and no diploids hemizygotic for the WA PUT4 allele showed any impairments of population growth efficiency or lag (supplementary fig. S6A–C, Supplementary Material online). Thus, the PUT4 defect is private to growth rate. WA PUT4 allele harbors no nonsynonymous but several synonymous and promoter mutations, implying that expression differences cause the proline growth variation (supplementary fig. S7A, Supplementary Material online). WA PUT4 is inherited by the SK1 and Y55 lab strains, but the proline defect is completely buffered in Y55, emphasizing that its exposure to selection depends on genetic context (supplementary fig. S7B and Supplementary Data, Supplementary Material online).

Finally, we considered the exceedingly slow growth of the West African when utilizing allantoin (fig. 3G), the nitrogen secretion product of mammals other than apes (Young et al. 1944). Allantoin lag time is unaffected, but the enormous growth rate effect precluded growth efficiency estimation. The defect affected close to 50% of WA offspring, regardless of cross, suggesting a monogenic effect (supplementary fig. S8A, Supplementary Material online). Arguing against monogenicity however, results from the lab strain cross implied that neither DAL4 nor DAL1 WA alleles were able to support allantoin growth (fig. 3H). DAL4 and DAL1 encode enzymes catalyzing the first two steps in allantoin uptake and degradation and are arranged back-to-back in an allantoin use gene cluster (supplementary fig. S8B and Supplementary Data, Supplementary Material online). This arrangement fuelled concern that deletion of one DAL gene impairs expression of its neighbor, as suggested (Naseeb and Delneri 2012). Inspecting WA DAL1 and DAL4, we uncovered a single nucleotide frameshifting insertion in DAL4 and an early proline to serine nonsynonymous mutation, predicted to be strongly detrimental, in DAL1 (supplementary fig. S8D, Supplementary Material online). Repairing either of these mutations with S288C variants failed to restore allantoin growth (fig. 3I), implying that they independently disrupt allantoin utilization. To confirm this, we repaired either of the two mutations while also introducing a centromeric plasmid with the functional S288C version of the other allele. Perfect allantoin growth was restored in both constructs. Thus, the DAL1 and DAL4 are null alleles arising from dal4c.1201delA and dal1c.415C>T mutations, respectively. Each mutation disrupts growth to a degree that is comparable to the effect of both mutations together, reflecting positive epistasis between loss-of-function mutations in components of a linear pathway (Lehner 2011). Curiously the lab strain Y55, which has inherited both WA null alleles, achieved reasonably fast allantoin growth, suggesting the existence of unknown routes for allantoin use (supplementary fig. S8E and Supplementary Data, Supplementary Material online).

Discussion

From a genetics perspective, it is not surprising that distinct alleles control variation in lag, rate, and efficiency of population expansion. Effects on any single one of these direct components of fitness, both by environmental factors (Warringer et al. 2008) and gene deletion (Warringer et al. 2003), are common. Biochemically, a positive coupling between the rate and efficiency of population expansion is controversial because the rate of metabolic reactions is negatively correlated with the energy remaining after reactions (Westerhoff et al. 1983; Heinrich et al. 1997; Pfeiffer et al. 2001). Under energy restriction, this is expected to force a trade-off between population growth rate and efficiency and there is ample experimental support for this (Postma et al. 1989; Spor et al. 2008, 2009). However, under nitrogen restriction, accelerated burning of energy by fast metabolic reactions does not necessarily reduce biomass yields and increase in biomass yield does not require slow metabolism. Energy is available in excess. Consequently, the observed correlation between rate and efficiency does not violate thermodynamics. Assimilated nitrogen can either be stored or used inside cells or channeled into reproduction. Reflecting this distinction, the nitrogen content per cell can vary over orders of magnitude. Enhanced population growth efficiency under nitrogen restriction represents the prioritized channeling of nitrogen to the next generation. This is important, because it suggests an evolutionary mechanism whereby efficiency enhancing mutations may be selected. In absence of internal storages, efficiency enhancing mutations cannot increase frequency in well-mixed populations because nonmutants enjoy equal access to the resource and benefit equally from efficiency enhancements (Hardin 1968; MacLean 2008). Privatization of nitrogen by internal accumulation means that only genomes carrying efficiency enhancing mutations benefit from them, enabling selection on population growth efficiency.

The fundamental assumption of this article is that a strong correlation between fitness components that reflect performance in different life stages, but are nonpleiotropic, only can emerge through concerted selection. Concerted selection follows from the intrinsic link between life stages underlying these fitness components, which are exposed to the same environmental variations and selective pressures. However, the conjecture does not necessarily predict such a strong correlation as observed here. It is tempting to speculate that diminishing return of consecutive mutations on a trait (Chou et al. 2011; Khan et al. 2011; Kryazhimskiy et al. 2014) may constitute part of the explanation for the strength of detected correlations. First, it provides a powerful incentive for parallel accumulation of mutations enhancing different fitness components. Two mutations enhancing different fitness components will simply tend to have a larger aggregate fitness contribution than two mutations enhancing the same fitness component. Second, the diminishing return of consecutive mutations means that even if different numbers of beneficial mutations have been accumulated in different fitness components, their aggregate trait contributions will tend to be rather similar.

Although there are few intrinsic limitations to the applicability of the proposed test for selection, caveats need to be recognized. First, it is prudent to point out that our evidence for coevolution of fitness components is strongest with regard to coevolution in the ancestral lineage. In contrast, our evidence for nonpleiotropy between fitness components relates to variants that differ between strains, i.e., more recent evolutionary events. It is hard to envision why fixed variants would be largely pleiotropic, whereas polymorphisms would be largely nonpleiotropic with regard to effects on fitness components. Nevertheless, we can currently not completely rule out this possibility. Second, the test is straight forward to apply only to traits that directly reflect performance in different life stages. Great care should be taken before extending the test to organisms with hard to define, hard to measure, or hard to genetically dissect fitness components. Third, absence of correlation between fitness components is not a sufficient ground for rejecting selection. Over shorter time periods, optimization of a single-fitness component may occur, even if multiple fitness components are under selection, as few mutations are accumulated in each genome. This has been made abundantly clear in artificial laboratory selections (MacLean and Gudelj 2006; Novak et al. 2006; Bachmann et al. 2013). Optimization of single fitness components may also occur in organisms where life stages are less intimately linked and occur in different environmental contexts, as selection pressures then may be radically different. Fourth, the test is unable to distinguish between positive and recently relaxed selection. Here, the RIM15, PUT4, DAL1, and DAL4 alleles were lineage-specific loss-of-function mutations, implying recent relaxation of selection in that lineage (Zorgo et al. 2012). Nevertheless, extension-of-function mutations in RIM15, DAL4, and PUT4 emerge rapidly in nitrogen-restricted artificial laboratory selections (Hong and Gresham 2014). Hence, it cannot be excluded that loss-of-function mutations are ancestral and that independent repair recently have occurred in well-functioning lineages. Fifth, all fitness components should ideally be accounted for. We could not quantify effects on sexual recombination, which although exceedingly rare may be under selection or on diploid growth traits, which often differs from haploid growth traits (Zorgo et al. 2013). Furthermore, we could not measure performance in the yeast spore state, which may be under selection in nature (Neiman 2011), or in the net death phase, when growth has ceased. From a general perspective, not accounting for all possible phenotypic and genetic relationships between fitness components leaves a margin for error. Assigning selection may still be valid, but the interpretation of which fitness components that are under selection may be confounded due to undetected pleiotropy between them.

Materials and Methods

Strains

Saccharomyces cerevisiae natural isolates were collected and haploidized as described (Cubillos et al. 2009; Liti et al. 2009). Haploids were crossed in all pairwise combinations. Twenty-three tetrads from each cross were sporulated to obtain 92 F1 recombinants. These were genotyped at 164–180 polymorphic positions (Cubillos et al. 2011). Diploid hemizygotes were obtained by mating haploids to single gene deletion BY4741 gene deletions using robotics. RIM15 and PUT4 reciprocal hemizygotes of crosses between natural isolates were obtained by one-step PCR and manual crossing. WA dal4c.1201delA and dal1c.415C>T mutations were replaced by the S288C variants using site-specific in vivo mutagenesis (Stuckey et al. 2011). WT and versions with either the dal4c.1201delA or dal1c.415C>T repaired were transformed with centromeric p5472 MoBY plasmids (Ho et al. 2009). A haploid S288C derivative, YSBN10, was used as control.

Medium and Cultivation

Microcultivation was performed in synthetic defined (SD) medium as described (Warringer and Blomberg 2003; Warringer et al. 2003), using a single nitrogen source present at 30 mg N/l. Experiments were run over 72–144 h until all isolates had entered stationary phase (Warringer and Blomberg 2003). Where a stationary phase had not been reached at the end of the experiment, experiments were discarded to avoid confounding effects. Population growth lag, rate, and efficiency were extracted as described (Warringer et al. 2008). Fitness components were log₂ transformed and normalized to those of at least four controls. Means of replicates (n = 2) were used for linkage analysis. Other experiments were performed with larger numbers of repeats, as indicated.

Nitrogen Uptake

Microcultivation of yeast cells was performed in SD medium as described above. At each sampling time point, 20 biological replicates of each microcultivated sample were pooled. Supernatants were collected and analyzed by diethyl ethoxymethylenemalonate derivatization and high-performance liquid chromatography analysis as described (Gomez-Alonso et al. 2007). Concentrations of the relevant nitrogen source were calculated using internal and external standards.

QTL Analysis

QTL analysis was performed using a nonparametric model in R/QTL at 2-cM density (Broman et al. 2003). Significance was estimated by permutation tests (α = 0.05). QTL positions were counted as identical if positions were within one average marker distance.

Sequence and SNP Analysis

Sequence data were taken from Bergstrom et al. (2014). Negative SNP consequences were predicted by Sorting Intolerant From Tolerant analysis (Kumar et al. 2009).

Statistics

Two group comparisons were performed using a homoscedastic two-tailed Student’s t-test and FDRs at q ≤ 5%. Significance of pleiotropy was tested using randomized permutations and a Mann–Whitney–Wilcoxon test.

Acknowledgments

This work was supported by the Research Council of Norway under the eVITA program (grant numbers 178901/V30 and 222364/F20) and from the Carl Trygger foundation (grant number 08-400).

Author notes

^†These authors contributed equally to this work.

Associate editor: Joshua Akey