All organisms can evolve in the face of environmental perturbations. Yet even closely-related individuals may differ in how quickly and effectively they adapt (Kirschner and Gerhart, 1998; Masel and Trotter, 2010; Lauring et al., 2013). Genetic modifiers that create these differences in evolvability are subject to natural selection, and the potential consequences of secondary selection on evolvability has been the subject of extensive theoretical work (van Nimwegen et al., 1999; Wilke et al., 2001; Masel and Bergman, 2003; Ciliberti et al., 2007; Cowperthwaite et al., 2008; Draghi et al., 2010; Soyer and Pfeiffer, 2010). Only recently, however, have experiments begun to identify these genetic modifiers and characterize their effects (Montville et al., 2005; Jarosz and Lindquist, 2010; Woods et al., 2011; McDonald et al., 2012; Stern et al., 2014; McDonald et al., 2016; Xiao et al., 2016).

Modifiers of evolvability act by changing parameters of the evolutionary process. For example, mutator or antimutator alleles often spread through adapting laboratory populations (Sniegowski et al., 1997; McDonald et al., 2012; Wielgoss et al., 2013). Other empirical work has shown that alleles that modify recombination rates can speed the rate of adaptation or help purge deleterious load (Zeyl and Bell, 1997; Colegrave, 2002; Goddard et al., 2005; Cooper, 2007; Becks and Agrawal, 2012; Gray and Goddard, 2012; McDonald et al., 2016; Xiao et al., 2016). Mutations can also affect evolvability in more subtle ways, through epistatic interactions that shape the future directions that evolution can take. These epistatic effects create historical contingency: the presence of an initial mutation changes the fitness effects of other potential mutations, opening up or closing off future adaptive trajectories (Blount et al., 2008; Woods et al., 2011; Salverda et al., 2011; Blount et al., 2012).

The identification and characterization of epistasis among specific sets of mutations has been a subject of intensive recent research. For example, several groups have analyzed combinations of mutations within the same protein that fixed during evolution (Weinreich et al., 2006; Bridgham et al., 2009; Gong et al., 2013; Weinreich et al., 2013; Palmer et al., 2015), or combinations of multiple mutations in a region of a protein (Hietpas et al., 2011; Araya et al., 2012; Jacquier et al., 2013; Harms and Thornton, 2014). These studies often find interesting intra-protein patterns of epistasis, including sign epistasis (a mutation shifts from beneficial to deleterious or vice versa depending on the background genotype) or higher-order epistasis (the combined effect of a set of mutations is not fully explained by the effects of any subset of these mutations). Other studies have measured pairwise epistasis among large collections of genome-wide mutations (e.g. all single gene knockouts) (Jasnos and Korona, 2007; van Opijnen et al., 2009; Costanzo et al., 2010; Szappanos et al., 2011; Babu et al., 2014; Costanzo et al., 2016). Still others have analyzed epistasis among all combinations of a few mutations that accumulated along the line of descent in a microbial evolution experiment (Chou et al., 2011; Khan et al., 2011; Wünsche et al., 2017), measured the fitness effects of individual mutations in divergent genetic backgrounds (Wang et al., 2016), or identified examples where a single mutation opens up an opportunity to colonize a new metabolic niche (Blount et al., 2008; Blount et al., 2012; Quandt et al., 2014; Quandt et al., 2015).

This body of work has extensively analyzed genetic interactions between individual mutations, a phenomenon we call microscopic epistasis (Good and Desai, 2015). These interactions can lead to historical contingency in the order and identities of mutations that spread in an adapting population (Weinreich et al., 2005). However, to affect evolvability a mutation must change the entire distribution of fitness effects of future mutations. That is, it must lead to a region of the fitness landscape which has statistically different properties, a phenomenon we refer to as macroscopic epistasis (Good and Desai, 2015).

Microscopic and macroscopic epistasis are distinct but not mutually exclusive properties of the fitness landscape. For example, a particular pattern of microscopic epistasis between individual mutations might (or might not) change the overall shape of the distribution of fitness effects, and therefore may (or may not) lead to macroscopic epistasis. Conversely, macroscopic epistasis can reflect microscopic epistasis among specific loci, but could also arise in the absence of microscopic epistasis if the pool of adaptive mutations is small and acquiring any individual mutation depletes future opportunities for further adaptation.

Although it is clear that microscopic epistasis is widespread, much less is known about patterns of macroscopic epistasis. Thus it is unclear how often and which mutations lead to differences in evolvability. Several recent studies have addressed this question by analyzing variation in the rate of adaptation among closely related microbial strains (Couce and Tenaillon, 2015). These studies have found that a few initial mutations can affect how quickly a strain adapts in the future (Barrick et al., 2010; Perfeito et al., 2014; Kryazhimskiy et al., 2014; Szamecz et al., 2014). Much of this variation in adaptability can be explained by the fitness of the founder, with fitter founders adapting more slowly (i.e. the rule of declining adaptability).

This work suggests that the rate at which a strain adapts is a heritable trait, influenced not just by stochastic evolutionary factors, such as the random occurrence and fixation of mutations, but also by genotype. However, little is known about variation in evolvability across larger genetic distances. In other words, how different is the evolutionary process starting from very different points in genotype space? Does a systematic pattern of declining adaptability with increasing fitness still apply among more distantly related strains? Or are differences in evolvability caused primarily by specific interactions between any of the initial genetic differences and the pool of potential future mutations? Additionally, previous studies have analyzed evolution in a single environmental condition, and thus could not test how initial genotype might affect the pleiotropic consequences of evolution (i.e. how adaptation in one environment affects fitness in other conditions). Note that here we use the term ‘pleiotropy’ in a sense that is common in experimental evolution but not in the genetics literature: it reflects how mutations that affect one trait (fitness in one environment) also influence other traits (fitness in different environments).

In this study, we address these gaps by measuring genetic variation in adaptability among 230 haploid S. cerevisiae genotypes (‘founders’) derived from a cross between two substantially diverged parental strains (Bloom et al., 2013). The parental strains (a laboratory strain, BY, and a wine strain, RM) differ at ~50,000 loci. Thus our 230 founders differ by on average ~25,000 mutations that are to some degree representative of natural variation in yeast. We evolve four replicate lines descended from each founder for 500 generations in each of two different laboratory environments. As in earlier studies, we can then measure variation in evolutionary outcomes that can be attributed to founder identity, above and beyond inherent evolutionary stochasticity. However, because our founders are chosen from a panel of strains designed for mapping the genetic basis of phenotypes of interest — each founder contains a random subset of the genetic variation between RM and BY strains — we can also use standard methods from quantitative genetics to identify specific loci that affect adaptation.

We focus on quantifying two specific aspects of evolvability: the rate of adaptation of each founder in each environment, and the pleiotropic consequences of adaptation in one environment for fitness in the other. We also address the influence of founder genotype on the genetic basis of future adaptation. Our results demonstrate that variation in both adaptability and pleiotropy is widespread. Despite the fact that our founder genotypes span evolutionary distances that are several orders of magnitude larger than tested in earlier work, the variation in their rates of adaptation still follows an overall rule of declining adaptability. The variation in pleiotropic consequences of adaptation between founders is also partly governed by founder fitness. Above and beyond these effects of founder fitness, we also identified several specific quantitative trait loci (QTLs) that affect adaptability and pleiotropy, including one which has a large influence on the genetic basis of future evolution. Thus both macroscopic and microscopic epistasis are important in our system, and lead to systematic variation in the rate, genetic basis, and pleiotropic consequences of adaptation.

To measure genetic variation in adaptability in budding yeast, we selected 230 segregants (‘founders’) from a cross between a wine strain, RM, and a laboratory strain, BY (Materials and methods). The parental strains RM and BY differ by about 0.5% at the sequence level, so each founder differs from its siblings at about 25,000 loci across the genome. We chose to analyze evolvability in two environments: rich media at an optimal temperature of 30°C (the OT environment), and synthetic media at a high temperature of 37°C (the HT environment). We measured the initial fitness of each founder in each environment and found that the genetic differences between founders lead to substantial variation in starting fitness in both environments (Figure 1). This variation is strongly correlated, implying that at least on average, alleles that are more beneficial in our OT environment also tend to be more beneficial in our HT environment.

Figure 1

Download asset Open asset

We established eight replicate lines from each founder (a total of 1840 populations), and evolved half of these replicate populations in our OT environment and half in our HT environment. After 500 generations of evolution, we measured the fitness of each descendant population in both environments (Figure 1). The difference between the initial fitness of the founder, $X$, and the fitness of a descendant population, $X^{\prime }$, is a measure of the rate at which that population adapted.

The measured rate of adaptation includes contributions from experimental error (inaccuracies in our fitness measurements lead to apparent differences in the rate of adaptation), from evolutionary stochasticity (some populations are more lucky than others in accumulating beneficial mutations), and from the inherent adaptability of the founder (some founders may tend to adapt faster than others). We are primarily interested in understanding the third component: the effect of founder genotype on adaptability. To disentangle this from evolutionary stochasticity, we can compare the variation in the rate of adaptation between lines descended from the same founder with the variation between lines descended from different founders. Similarly, to disentangle the effects of measurement error, we can compare the variation between lines descended from the same founder with the variation between replicate fitness measurements of the same populations. We could in principle compare these different sources of variation using an analysis of variance (ANOVA) framework, following the approach we used in previous work (Kryazhimskiy et al., 2014). Here, we instead describe this analysis in the language of quantitative genetics. While some of the underlying methods are equivalent, this quantitative genetics framework provides a useful perspective, and makes it possible to address additional questions about the genetic basis of variation in adaptability.

We begin by calculating the broad and narrow sense heritability of founder fitness and adaptability. Here the broad-sense heritability of a phenotype, $H^2$, is the fraction of the variance in that phenotype that can be explained by genotype. The narrow-sense heritability, $h^2$, is the fraction of the total variance in the phenotype which can be explained by an additive model of the individual loci. We calculate both broad and narrow-sense heritability using standard methods (Yang et al., 2010; Zuk et al., 2012; Visscher et al., 2006), which exploit the fact that in an additive model the covariance between founder phenotypes is a linear function of their relatedness (Materials and methods). Since our populations are all haploid, the gap between broad-sense and narrow-sense heritability arises primarily from epistatic interactions between loci and any experimental error in our estimation of these quantities, although it may also carry contributions from untagged rare variants, mitochondrial copy number variation, and any uncontrolled $\text{G}\times \text{E}$ effects (Bloom et al., 2015).

We first consider the heritability of the founder fitness. We find that this has a broad-sense heritability near 100% (Figure 2, Table 1), indicating that measurement errors are small compared to genuine differences in fitness between founders. Consistent with previous work (Bloom et al., 2013), we find a substantial gap between $h^2$ and $H^2$ (Figure 2, Table 1), indicating the importance of epistasis in determining founder fitness.

Figure 2

Download asset Open asset

We next consider the heritability of adaptability. The broad-sense heritability of adaptability, $H_{\mathrm{\Delta }X}^2$, is the fraction of variance in ${\displaystyle \mathrm{\Delta }X=X^{\prime }-X}$ that can be attributed to founder genotype, rather than to inherent evolutionary stochasticity or to measurement errors. In contrast to founder fitness, which we expect to be a property of the genotype, there is no reason that the rate of adaptation must be heritable. However, multiple previous studies have found that genotype does affect the rate of adaptation in microbial populations (Burch and Chao, 2000; Barrick et al., 2010; Woods et al., 2011; Perfeito et al., 2014; Kryazhimskiy et al., 2014). Consistent with this earlier work, we find that adaptability is heritable, with $H_{\mathrm{\Delta }X}^2\approx 0.61$ in the OT environment and $0.65$ in the HT environment (Figure 2, Table 1). The fact that the values of $H_{\mathrm{\Delta }X}^2$ are lower than those of $H_X^2$ is expected because, in addition to measurement errors, our measure of adaptability is affected by the inherent randomness of the evolutionary process.

As before, we fit a linear model to estimate the narrow-sense heritability of adaptability $h_{\mathrm{\Delta }X}^2$. We find $h_{\mathrm{\Delta }X}^2=0.46$ in the OT environment and $h_{\mathrm{\Delta }X}^2=0.44$ at HT (Figure 2, Table 1). The difference between $H_{\mathrm{\Delta }X}^2$ and $h_{\mathrm{\Delta }X}^2$ indicates that about two thirds of the heritable variation in adaptability can be explained by an additive model of the underlying genotype, showing that epistasis is important to this trait (Figure 2).

Predicting adaptability from genotype and founder fitness

We next sought to map the genetic basis of the observed variation in founder fitness and adaptability. To do so, we applied a standard iterative procedure to identify QTLs that affect each of these traits in each environment (see Materials and methods). We identified 6 QTLs that affect founder fitness at OT, and 3 QTLs that affect founder fitness at HT. Consistent with the strong correlation between these two phenotypes, 2/3 of the HT QTLs also appear as OT QTLs (Supplementary file 5). In both cases, these QTLs explain 76% of the trait variance. We also found 4 QTLs that significantly affect adaptability, $\mathrm{\Delta }X$, at OT and 2 QTLs that affect adaptability at HT. Note that these 2 loci are a subset of the OT adaptability loci (Supplementary file 5). Together these adaptability QTLs explain 30% of the total variance in this trait in the OT environment, and 31% in the HT environment, also consistent with our estimates of narrow-sense heritability (Figure 2 and Figure 3 and Table 1). Due to the limited size of our study we may be missing some QTLs that affect fitness or adaptability. However, the fact that the variance in these traits explained by detected QTLs is similar to our estimates of narrow-sense heritability (Figure 2, Table 1) indicates that any missing QTLs are unlikely to have large additive effects on the traits.

Figure 3

Download asset Open asset

Figure 3—source data 1

Parameters for combined fitness and QTL models for adaptability.

https://doi.org/10.7554/eLife.27167.006

Download elife-27167-fig3-data1-v2.xlsx

Interestingly, three of the four adaptability loci were also independently identified as QTLs for founder fitness (Supplementary file 5), suggesting that loci that affect founder fitness also affect adaptability. This is consistent with the conclusions of several recent studies, which show that much of the variation in adaptability between closely related microbial strains can be explained by differences in their initial fitness, according to the rule of declining adaptability (Perfeito et al., 2014; Kryazhimskiy et al., 2014; Couce and Tenaillon, 2015). Motivated by these results, we asked whether a similar effect holds in our system as well, even though our founders are several orders of magnitude more diverged than strains analyzed in earlier studies. We found that indeed founder fitness explains much of the variation in adaptability, and follows the rule of declining adaptability (Figure 3). Specifically, a model where adaptability declines linearly with founder fitness explains 42% of the total variance in adaptability in the OT environment, and 52% of the adaptability at HT, or about 68% and 80% of the heritable variance, respectively (Figure 2). This represents more of the variance explained than the linear QTL models described above, despite the fact that the QTL models have more free parameters. Interestingly, the variance explained by founder fitness is comparable to the narrow-sense heritability, suggesting that founder fitness may be as good of a predictor of adaptability as any linear genetic model.

Since founder fitness explains more variation in adaptability than additive QTL models, and since three of the identified QTLs affect both fitness and adaptability, it is natural to ask whether there are any genetic loci that explain some of the variation in adaptability above and beyond that explained by fitness. Indeed, about 30% and 20% of heritable variation in adaptability remain unexplained by founder fitness in the OT and in the HT environment, respectively (Figure 2, Table 1). Can we then identify QTLs that explain any of this residual variation?

To address this question, we first asked whether any of the previously identified QTLs (either for founder fitness or for adaptability) can help explain variation in adaptability once we have taken fitness into account by including it as a covariate in the QTL models (Materials and methods). We found that indeed several of these QTLs had a modest effect on adaptability above and beyond fitness (Supplementary file 5). The QTL with the largest effect was centered at position 376,315 on chromosome XIV, and we refer to it as the ‘KRE33 locus.’ This QTL explains 2% percent of residual residual variation in adaptability in the HT environment, and 0.4% percent in the OT environment (Figure 3). We previously identified this locus as having the largest effect on founder fitness: segregants with the BY allele at this locus are on average 9% more fit at OT and 19% more fit at HT compared to segregants with the RM allele. After adding pairwise interaction terms between the KRE33 locus and other detected QTLs, the combined model explains 82% of the heritable variation in adaptability in the HT environment and 88% in the OT environment (Table 1). Thus most of the heritable variation in adaptability in our system can be explained by founder fitness, with some additional contributions from genotype at a few adaptability QTL loci. Nonetheless, there remains some residual variation due to further unidentified genotype effects.

To check for strong effect adaptability QTLs not previously identified, we repeated the QTL discovery procedure described above, taking the trait to be the residual variation in adaptability about the regression line in Figure 3. We recovered a subset of the already-identified adaptability QTLs, without identifying any further loci.

Adaptability depends on initial fitness in the ‘home’ environment

We have seen that the rule of declining adaptability applies in both of our evolution conditions. However, because the initial fitnesses of our founders in the two environments are correlated (Figure 1), it is possible that the two patterns of decline in adaptability are not independent. Instead, they both could be driven by some factor common to both environments, but not by the initial fitnesses in those environments. If this were the case, then we would expect that the rate of adaptation of all the strains would be controlled by this factor, and we would not expect a strain with particularly disparate fitnesses in the two environments (i.e. lying away from the trend in Figure 1B) to adapt faster in the environment where its fitness was lower. Alternatively, the rule of declining adaptability could apply independently in each environment. In this case, we would expect that a strain with low initial fitness at OT and high initial fitness at HT would adapt rapidly at HT and slowly at OT (and vice versa).

To address this question, we first classify each founder according to how disparate its initial fitness is in HT versus OT (Figure 4A). Note that this is the same data as in Figure 1B after an appropriate normalization (see Materials and methods), with color representing distance from the diagonal. In Figure 4B, we show the average fitness gains of descendants of each of these founders after evolution in each environment. Consistent with a rule of declining adaptability that holds independently in each environment, we find that founders with particularly high initial fitness in HT (green) tend to adapt particularly slowly in HT but not in OT, and vice versa for founders with particularly high initial fitness in OT (purple). In Figure 4C, we quantify this trend by plotting the difference in the normalized fitness gains in the two environments as a function of the difference in initial fitnesses in the two environments (see Materials and methods for details). If a common factor drove the patterns of declining adaptability in both environments, then the difference in a founder’s initial fitnesses should be uncorrelated with the difference in its adaptabilities in the two environments. In contrast, we see a negative correlation (${\displaystyle r^2=0.28,p<{10}^{-4}}$), again consistent with a rule of declining adaptability holding independently in both environments. Thus initial fitness in the home environment is a predictor of its adaptability in that environment, above and beyond a possible common factor that might be driving adaptability in both conditions. We note that these results also hold after controlling for the KRE33 allele (Materials and methods, Figure 4—figure supplement 1).

Figure 4 with 1 supplement see all

Download asset Open asset

The genetic basis of adaptation

Our observations of heritable variation in adaptability indicate that macroscopic epistasis is widespread in our system, and can be largely explained by a model involving founder fitness and a small number of QTLs. However, it is not clear whether this is accompanied by microscopic epistasis between alleles in the founders and potential future mutations. In other words, do different founders accumulate different sets of beneficial mutations as they adapt? And if so, can we identify QTLs or other factors (e.g. founder fitness) that modify the spectrum of mutations that accumulate during adaptation?

To address this question, we sequenced 273 whole-population samples from the final timepoint of all independently evolved populations descended from each of 35 founders (Supplementary file 3). We discarded nineteen populations for technical reasons before calling mutations (see Materials and methods), leaving us with a total of 254 sequenced lines (mean 23x depth). We identified SNPs and small indels in this mixed-population sequence data (Materials and methods). We focus here only on mutations that were fixed or estimated to be at high frequency (above 50%) within the population at this final timepoint. We identified a total of 603 mutations (on average 1.3 per OT population and 3.6 per HT population) meeting these criteria. Of these, 449 are nonsynonymous mutations or indels in genes (293 from HT populations, 156 from OT populations).

More fit founders can adapt more slowly because they have access to fewer beneficial mutations, or to beneficial mutations of weaker effect, or both. In any of these cases, we expect that mutations accumulate more slowly over time in populations descended from fitter founders. Consistent with this expectation, we found that the number of mutations identified in a population declines weakly but significantly with the initial founder fitness (Figure 5).

Figure 5

Download asset Open asset

Beyond variation in the number of mutations, did founder genotype affect the spectrum of mutations available to future adaptation? To address this question we focused on genes that were mutated in multiple populations (genes mutated in a single population are uninformative for this purpose). We found 27 genes in which missense or putative loss of function mutations were called in two or more independent populations (Figure 6); these ‘multi-hit’ genes are expected to be enriched for targets of selection.

Figure 6

Download asset Open asset

Figure 6—source data 1

Mutual information between de novo mutations and the KRE33 allele, the evolution environment, and the genotype.

https://doi.org/10.7554/eLife.27167.011

Download elife-27167-fig6-data1-v2.xlsx

The gene with the second-largest number of identified de novo mutations was KRE33: 31 independent sequenced populations acquired a mutation in this gene (a total of 12% of our 254 sequenced lines; Figure 6). All 31 of these populations are descended from founders that had the less-fit RM allele at the QTL containing the KRE33 gene (i.e. the same QTL that we identified above as affecting adaptability; Figure 6). This observation suggests that genetic variation in the KRE33 gene itself might have caused the observed variation in adaptability. Since the Kre33 protein is a member of the 90 s preribosomal pathway, we reasoned that the deleterious effect of the RM allele could potentially be compensated not only by mutations in KRE33 itself, but also by mutations in genes that code for other members of the ribosome biogenesis pathway. Consistent with this hypothesis, we found 108 putatively functional mutations in our sequenced lines in 10 genes (including KRE33) that were categorized as belonging to this pathway, all of which arose in descendants of founders that carry the RM allele at the KRE33 locus (Figure 6). Thus, compensation for the deleterious effect of the RM allele in the KRE33 gene is likely one of the strongest targets of selection in our experiment, and it likely contributes to the increased adaptability of the founders which carry the RM version of KRE33.

Of the remaining 101 mutations in multi-hit genes, 14 out of 15 mutations in GPB1, GPB2 and PDE2 — genes involved in cAMP regulation — occurred exclusively in populations descended from founders that carry the more-fit BY allele at the KRE33 locus. This suggests a potential genetic interaction between KRE33 and this pathway, such that mutations in GPB1, GPB2, and PDE2 are more beneficial in the BY KRE33 background. However, it is also possible that these mutations are always beneficial, but are less likely to fix in the KRE33-RM background where they can be lost to clonal interference with more strongly beneficial mutations (e.g. those in the ribosome biogenesis pathway).

The effect of the KRE33 allele on the genetic basis of future adaptation suggests that other loci that contribute to fitness or adaptability differences between RM and BY strains might also be targets for adaptive de novo mutations. To test this hypothesis, we asked whether multi-hit genes are preferentially found within the 8 QTLs that we identified as influencing founder fitness or adaptability. We found two additional examples of multi-hit genes that lie within the confidence interval for one of these 8 QTLs: IRA2 and BFR2 (Supplementary file 5). These 2 genes, along with KRE33, represent an enrichment over the expected number of genes in common between the set of multi-hit genes and those within QTL confidence intervals (Materials and methods). Additionally, RAS2, an essential gene which acts as a master regulator in the cAMP pathway that includes 5 multi-hit genes (including IRA2 Figure 6), is found within one of the QTLs affecting fitness. These results suggest that some loci that underlie variation in fitness among founders are also targets of selection during subsequent evolution. They also suggest that the cAMP pathway, which has been implicated as a target of adaptive de novo mutations in several other experimental evolution studies (Venkataram et al., 2016; Kvitek and Sherlock, 2013; Lang et al., 2013), may also explain some of the fitness differences caused by natural genetic variation in this cross, at least in the specific experimental conditions we analyze here.

With the exception of the cases noted above, which all involve QTLs previously identified as affecting fitness or adaptability, we lack power to identify individual loci in the founder genotype that affect the genetic basis of future adaptation. That is, we cannot find specific QTLs that affect whether mutations in any of the 27 multi-hit genes will be observed in the descendants of a given founder. However, we can still ask whether the mutations that occur during adaptation show a nonrandom association with the founder genotype. We can also ask whether the mutations that occur during adaptation depend on the environment in which evolution was conducted. To do so, we calculated the mutual information $I(G,E;M)$ provided by the founder genotype $G$ and evolution environment $E$ about the set of mutations $M$ that accumulate during adaptation (see Materials and methods for details; note in particular that we control for the identity of the KRE33 allele throughout this analysis). Intuitively, the mutual information $I(G,E;M)$ measures the amount of information (in bits) that we gain about $M$ by knowing $G$ and $E$. Higher values of $I$ indicate that $G$ and $E$ more strongly predict $M$.

We used $I(G,E;M)$ as a test statistic to determine whether there were nonrandom associations between founder genotype, evolution environment, and the mutations that underlie adaptation. Controlling for the number of mutations acquired by each population, we detected a significant association between the evolution condition and the mutations that accumulate during adaptation. For example, we found putatively functional mutations in SIR3 and SIR4 in 34 different populations, 30 of which evolved at HT. However, the overall magnitude of this effect of $E$ on $M$ is small, on the order of 0.1 bits. In addition to this effect of environment, we also find that the genotype $G$ affects $M$. Specifically, controlling for the number of mutations, the KRE33 allele, and the evolution environment, there is a small additional effect of founder on the identities of de novo mutations. This suggests that while founders with the same allele at the KRE33 locus draw many common adaptive mutations from the same pool, there are additional founder-specific epistatic interactions.

The pleiotropic consequences of adaptation

Adaptation to one environment can lead to concomitant fitness gains or losses in other environments, an effect we refer to as pleiotropy for fitness. To characterize these pleiotropic effects in our system, we analyzed how evolution in our OT environment affects fitness in the HT environment, and vice versa. We can describe the founder fitnesses in both environments as $\overrightarrow{W}=(X,Y)$, where $X$ is the initial fitness at OT and $Y$ is the initial fitness at HT. As noted above, the founder fitnesses in the two environments were strongly correlated, especially within each of the two KRE33 allele groups (Figure 7A), suggesting that the alleles that distinguish RM and BY have correlated effects in our environments.

Figure 7

Download asset Open asset

We next calculated the fitness gains for populations adapted to each environment. After evolution in a given environment, a founder with fitness $\overrightarrow{W}=(X,Y)$ will have adapted, yielding a descendant line with fitness ${\overrightarrow{W}}^{\prime }=(X^{\prime },Y^{\prime })$. We denote the fitness increments of this population as $\mathrm{\Delta }\overrightarrow{W}\equiv (\mathrm{\Delta }X,\mathrm{\Delta }Y)=(X^{\prime }-X,Y^{\prime }-Y)$; this vector represents the increase or decrease in fitness of the descendant population in both environments, after evolution in one of the two environments. For clarity, we will use the notation ${\mathrm{\Delta }}_{OT}$ to refer to the fitness increments of descendant populations after evolution in the OT environment, and ${\mathrm{\Delta }}_{HT}$ to refer to the fitness increments after evolution in the HT environment. We refer to the environment in which a given population evolves as its ‘home’ environment, and the other environment as its ‘away’ environment. Thus ${\mathrm{\Delta }}_{OT}X$ and ${\mathrm{\Delta }}_{HT}Y$ are fitness increments in the home environments, and ${\mathrm{\Delta }}_{OT}Y$ and ${\mathrm{\Delta }}_{HT}X$ are fitness increments in the away environments.

We find that, on average, populations adapting to either of our two environments also gained fitness in the other environment (Table 2), consistent with our observation that the sets of multi-hit genes in the two environments were highly overlapping. Thus as populations adapt, their fitnesses at OT and HT remain positively correlated (Figure 7B,C). Despite this overall pattern of correlated adaptation, we find that the average fitness gain at HT was significantly higher after evolution at HT than after evolution at OT (Table 2), as we might expect. However, the converse was not true: populations increased in fitness at OT by about the same amount after evolving at HT as they did after evolving at OT. Because populations evolved at HT and those evolved at OT made almost identical fitness gains at OT, but different gains at HT, evolution in these two conditions must involve different sets of adaptive mutations, despite the overlap in the targets of selection between the two environments. This is consistent with our mutual information analysis above.

Table 2

	Kre33 -RM		Kre33 -BY
	Evolved at OT	Evolved at HT	Evolved at OT	Evolved at HT
Fitnessat OT, %	8.33(8.27, 8.38)	8.86(8.81, 8.92)	5.01(4.94, 5.07)	4.834.76, 4.90
Fitnessat HT, %	9.57(9.42, 9.77)	20.33(20.18, 20.53)	6.25(6.12, 6.38)	11.51(11.38, 11.64)

Predicting pleiotropic effects from genotype and founder fitness

We next used the framework of quantitative genetics to analyze the effects of founder genotype on pleiotropy. We find that the pleiotropic consequences of adaptation to our HT environment are highly heritable, with broad-sense heritability $H^2$ of the fitness increases in the away environment, ${\mathrm{\Delta }}_{HT}X\approx 0.62$. This is similar to the heritability of adaptability we found above. In contrast, the pleiotropic consequences of adaptation to our OT environment are much less heritable, with broad-sense heritability of ${\mathrm{\Delta }}_{OT}Y\approx 0.29$, indicating that in this case stochastic evolutionary forces are more important to the outcome than genotype.

As with the adaptability traits analyzed above, we find that the narrow-sense heritability of pleiotropy is substantially lower than the broad-sense heritability (Figure 8). Specifically, we find $h^2\approx 0.48$ for ${\mathrm{\Delta }}_{HT}X$ and $h^2\approx 0.09$ for ${\mathrm{\Delta }}_{OT}Y$. Thus most of the heritable variation in this trait cannot be explained using an additive QTL model of the underlying founder genotype.

Figure 8

Download asset Open asset

Figure 8—source data 1

Parameters for combined fitness and QTL models for pleiotropy.

https://doi.org/10.7554/eLife.27167.015

Download elife-27167-fig8-data1-v2.xlsx

We next sought to understand which factors best explain the heritable variation in pleiotropy. Analogous to our analysis of adaptability, we consider the potential effects of specific QTL loci as well as of founder fitness (now in both home and away environments). We begin by fitting a model in which the fitness increments in the away environment decline linearly with the founder fitness in both the home and the away environments. We find that founder fitness does explain much of the heritable variation in pleiotropy (68% of the heritable variation in ${\mathrm{\Delta }}_{HT}X$ and 43% for ${\mathrm{\Delta }}_{OT}Y$). However, we find that it is only the initial fitness in the away environment that is important: once founder fitness in the away environment is included in the model, founder fitness in the home environment adds no further explanatory power, while the converse is not true.

To assess the dependence of pleiotropic outcomes on specific parental alleles, we next looked for QTLs that influence fitness increments in the away environment, over and above founder fitness in that environment. Specifically, we asked whether any of the eight previously-identified QTLs could help predict pleiotropic outcomes over and above founder fitness in the away condition. We find that 6 of these loci, including KRE33, do have a significant effect on pleiotropic outcomes (Supplementary file 5). Together with founder fitness in the away condition, they explain 75% and 58% of the heritable variation in pleiotropy at OT and HT, respectively (Figure 8). To check for strong effect pleiotropy QTLs not previously identified, we also repeated the QTL discovery procedure described above, taking the trait to be the residual variation in fitness increments in the away environment after regression against founder fitness in that environment. We recovered a subset of the previously identified QTLs, without identifying any further loci.

In order for evolution to increase or decrease adaptability, there must be genetic variation in this trait that natural selection can act on. Here, we analyzed the extent and genetic basis of variation in adaptability among 230 segregants from a cross between two divergent yeast strains (Bloom et al., 2013). By exploiting the structure of the cross, we mapped QTLs that explain observed differences in adaptability. We then compared the predictive power of these additive genetic models with an alternative model where adaptability depends on founder fitness.

Consistent with earlier work, we find that the rate of adaptation is a heritable trait: different founders consistently adapt at different rates. This means that some of the ~50,000 alleles that distinguish BY from RM alter future adaptive trajectories. In principle, these effects could involve just a few founder alleles whose effects on adaptability are independent of each other (e.g., if particular alleles that distinguish RM and BY create opportunities for improvement of independent biological processes). If this were the case, an additive model involving these variants would explain the observed heritable variation in adaptability between founders. However, we find that the best additive QTL model explains at most about half of the observed heritable genetic variation in adaptability.

Instead, we find that a model based on founder fitness explains more of the heritable variation in adaptability, despite having fewer free parameters. This is broadly consistent with earlier work showing that initial fitness explains variation in adaptability among founders that differ only by a few mutations (Barrick et al., 2010; Perfeito et al., 2014; Kryazhimskiy et al., 2014; Wang et al., 2016). However, over the larger genetic distances we study here, a substantial fraction of heritable variation in adaptability is not explained by founder fitness. About half of the remaining variation can be explained by including KRE33 and a small number of other loci as predictors in addition to founder fitness, leaving about 20 percent of the heritable variation unexplained.

We next investigated the pleiotropic consequences of adaptation in one (‘home’) environment on fitness in another (‘away’) environment. We found that in the two conditions we considered, adaptation to the home environment leads on average to a fitness gain in the away environment. As with adaptability, these fitness increments in the away environment were heritable. To some extent this is expected: since founder genotype affects adaptability, we expect the founder genotype to affect pleiotropy as well, simply because fitness gains in the home environment are correlated with fitness gains in the away environment. Consistent with this, we find that founder fitness does explain much of the heritable variation in pleiotropy. Surprisingly, however, only initial fitness in the away environment is important. These results are consistent with a simple model in which the fitness effects of mutations in any environment are modulated by the fitness of the genetic background in that environment, regardless of how these mutations were acquired. Thus the effects in environment B of mutations accumulated through evolution in environment A are determined primarily by the fitness of the genetic background in environment B.

We also found that the broad-sense heritability of fitness changes in the away environment is asymmetric. For populations evolved at HT, fitness gains in the OT environment were highly heritable, and predicted well by founder fitness in the OT (away) environment. This suggests that the HT populations acquired a set of mutations whose effects at OT are predictable based on the fitness of the genetic background in this condition. Conversely, the populations evolved at OT saw much more variable gains at HT. This implies that the mutations accumulated at OT have effects at HT that are less predictable from founder fitness or genotype.

In addition to founder fitness, we find that the KRE33 allele and a few other smaller-effect QTLs significantly affect evolutionary outcomes. To some extent, these QTLs affect evolution through their impact on founder fitness. However, five QTLs, including KRE33, affect adaptability and in some cases pleiotropy above and beyond their effects on founder fitness. The founder allele at the KRE33 locus also strongly affects the spectrum of mutations that accumulate during adaptation: founders that start with the less-fit RM allele at KRE33 accumulate mutations in KRE33 and in other genes involved in the 90 s preribosomal pathway. These genes are all essential, so these mutations likely involve gain or attenuation rather than loss of function. Surprisingly, despite this effect, KRE33 still has only a modest impact on adaptability and pleiotropy above and beyond the founder fitness. For example, populations descended from founders with the BY and RM alleles of KRE33 had similar patterns of fitness gains in the two environments, despite acquiring different sets of mutations (Figure 7D). This suggests that the pleiotropic effects of these two classes of mutations are similar.

Together, these observations suggest the following simple phenomenological picture of adaptation in this system. The average fitness of an adapting yeast population in its home environment is statistically predictable based on the fitness of its founding genotype, as well as the allele at the KRE33 locus and a few other loci, even for highly divergent strains. Mutations accumulated during adaptation tend to lead to pleiotropic fitness gains in other, related, environments. Our ability to predict these pleiotropic fitness gains depends on the environment: it is strong in one case but weak in the other. This points to the possibility that fitness trade-offs between environments, and other complicating factors, disrupt our ability to predict pleiotropic fitness gains.

We note that our experimental approach has several important limitations. First, we define adaptability as the expected increase in fitness of a line descended from a given founder after 500 generations of evolution in one of two specific and artificial laboratory environments. The effects of founder fitness and genotype on adaptability and pleiotropy might change over different timescales or in a different set of environmental conditions. In particular, the heritability of pleiotropic effects of adaptation may be a consequence of the similarity of the environments we chose, and might be lower across other conditions. Our results also depend on our choice of this specific yeast cross as our source of founder genotypes. In addition, due to the relatively small number of segregants we are able to analyze and the intrinsic noisiness of measurements of adaptability and pleiotropy, our power to identify the genetic basis of these traits is limited. As a result, we have likely missed some QTLs affecting adaptability and pleiotropy (though our estimates of narrow-sense heritability do suggest that any potentially missing QTLs have minor additive effects on these traits). Further work across other environments and with other strains and organisms will be needed to test the generality of our conclusions. However, despite these important caveats, our results provide additional support to a growing body of evidence that a rule of declining adaptability may be general feature of microbial evolution.

Our results are consistent with earlier studies showing that microscopic epistasis is common among mutations that accumulate along the line of descent in laboratory microbial evolution experiments. However, it has been unclear whether this microscopic epistasis leads to macroscopic epistasis, in which related genotypes have different distributions of fitness effects of new mutations. In principle, many patterns of microscopic epistasis could be consistent with the same pattern of macroscopic epistasis, or with no macroscopic epistasis at all. In turn, many types of macroscopic epistasis could be consistent with the same pattern of variation in adaptability, since the expected rate of adaptation depends in a complex way on the entire distribution of fitness effects of new mutations and on population parameters. For example, it is possible to have extensive macroscopic epistasis without any variation in adaptability.

One way to study differences in adaptability would be to take a bottom-up approach, measuring patterns of microscopic epistasis and using these to characterize macroscopic epistasis and then adaptability. We have instead taken the opposite approach, by directly measuring genetic variation in adaptability. We have shown that even across large genetic distances, less-fit initial genotypes adapt more rapidly than more-fit initial genotypes. Our results cannot explain what type of microscopic epistasis, if any, underlies this pattern of declining adaptability. Previous work indicates that some adaptive mutations exhibit negative (‘diminishing returns’) epistasis, which is implicated in driving the pattern of declining adaptability (Chou et al., 2011; Khan et al., 2011; Kryazhimskiy et al., 2014; Wünsche et al., 2017), but we do not have direct evidence for or against this type of microscopic epistasis in the present experiment. However, we do find at least one significant microscopic epistatic interaction that affects adaptability at both fitness and genotype levels in our HT environment, which involves the allele at the KRE33 locus. This microscopic epistasis also exists in our OT environment, as evidenced by its effect on the genetic basis of adaptation, and on the pleiotropic consequences in the HT condition. Surprisingly, however, this microscopic epistasis has a very weak effect on the rate of adaptation in the OT environment, above the effect we expect from lower founder fitness. This is an example of a case where idiosyncratic microscopic epistasis does not lead to differences in adaptability, either because the microscopic epistasis does not lead to macroscopic epistasis, or because it leads to macroscopic epistasis which does not significantly affect adaptability.

1. 1. Jerison ER
   2. Kryazhimskiy S
   3. Mitchell J
   4. Bloom JS
   5. Kruglyak L
   6. Desai MM

   (2017) Data from: Genetic variation in adaptability and pleiotropy in budding yeast

   Publicly available at the NCBI Sequence Read Archive (accession no. SRP102877).

   https://www.ncbi.nlm.nih.gov/sra/?term=SRP102877

1. Book

   1. Adams A
   2. Gottschling DE
   3. Kaiser CA
   4. Stearns T

   (1998)

   Methods in Yeast Genetics: A Cold Spring Harbor Laboratory Course Manual, Vol. 12

   Harbor: Cold Spring Harbor Laboratory Press Cold Spring.

   • Google Scholar
2. 1. Araya CL
   2. Fowler DM
   3. Chen W
   4. Muniez I
   5. Kelly JW
   6. Fields S

   (2012) A fundamental protein property, thermodynamic stability, revealed solely from large-scale measurements of protein function

   PNAS 109:16858–16863.

   https://doi.org/10.1073/pnas.1209751109

   • PubMed
   • Google Scholar
3. 1. Babu M
   2. Arnold R
   3. Bundalovic-Torma C
   4. Gagarinova A
   5. Wong KS
   6. Kumar A
   7. Stewart G
   8. Samanfar B
   9. Aoki H
   10. Wagih O
   11. Vlasblom J
   12. Phanse S
   13. Lad K
   14. Yeou Hsiung Yu A
   15. Graham C
   16. Jin K
   17. Brown E
   18. Golshani A
   19. Kim P
   20. Moreno-Hagelsieb G
   21. Greenblatt J
   22. Houry WA
   23. Parkinson J
   24. Emili A

   (2014) Quantitative genome-wide genetic interaction screens reveal global epistatic relationships of protein complexes in Escherichia coli

   PLoS Genetics 10:e1004120.

   https://doi.org/10.1371/journal.pgen.1004120

   • PubMed
   • Google Scholar
4. 1. Barrick JE
   2. Kauth MR
   3. Strelioff CC
   4. Lenski RE

   (2010) Escherichia coli rpoB mutants have increased evolvability in proportion to their fitness defects

   Molecular Biology and Evolution 27:1338–1347.

   https://doi.org/10.1093/molbev/msq024

   • PubMed
   • Google Scholar
5. 1. Baym M
   2. Kryazhimskiy S
   3. Lieberman TD
   4. Chung H
   5. Desai MM
   6. Kishony R

   (2015) Inexpensive multiplexed library preparation for megabase-sized genomes

   PLoS One 10:e0128036.

   https://doi.org/10.1371/journal.pone.0128036

   • PubMed
   • Google Scholar
6. 1. Becks L
   2. Agrawal AF

   (2012) The evolution of sex is favoured during adaptation to new environments

   PLoS Biology 10:e1001317.

   https://doi.org/10.1371/journal.pbio.1001317

   • PubMed
   • Google Scholar
7. 1. Bloom JS
   2. Ehrenreich IM
   3. Loo WT
   4. Lite TL
   5. Kruglyak L

   (2013) Finding the sources of missing heritability in a yeast cross

   Nature 494:234–237.

   https://doi.org/10.1038/nature11867

   • PubMed
   • Google Scholar
8. 1. Bloom JS
   2. Kotenko I
   3. Sadhu MJ
   4. Treusch S
   5. Albert FW
   6. Kruglyak L

   (2015) Genetic interactions contribute less than additive effects to quantitative trait variation in yeast

   Nature Communications 6:8712.

   https://doi.org/10.1038/ncomms9712

   • PubMed
   • Google Scholar
9. 1. Blount ZD
   2. Barrick JE
   3. Davidson CJ
   4. Lenski RE

   (2012) Genomic analysis of a key innovation in an experimental Escherichia coli population

   Nature 489:513–518.

   https://doi.org/10.1038/nature11514

   • PubMed
   • Google Scholar
10. 1. Blount ZD
    2. Borland CZ
    3. Lenski RE

    (2008) Historical contingency and the evolution of a key innovation in an experimental population of Escherichia coli

    PNAS 105:7899–7906.

    https://doi.org/10.1073/pnas.0803151105

    • PubMed
    • Google Scholar
11. 1. Bolger AM
    2. Lohse M
    3. Usadel B

    (2014) Trimmomatic: a flexible trimmer for Illumina sequence data

    Bioinformatics 30:2114–2120.

    https://doi.org/10.1093/bioinformatics/btu170

    • PubMed
    • Google Scholar
12. 1. Bridgham JT
    2. Ortlund EA
    3. Thornton JW

    (2009) An epistatic ratchet constrains the direction of glucocorticoid receptor evolution

    Nature 461:515–519.

    https://doi.org/10.1038/nature08249

    • PubMed
    • Google Scholar
13. 1. Burch CL
    2. Chao L

    (2000) Evolvability of an RNA virus is determined by its mutational neighbourhood

    Nature 406:625–628.

    https://doi.org/10.1038/35020564

    • PubMed
    • Google Scholar
14. 1. Chou HH
    2. Chiu HC
    3. Delaney NF
    4. Segrè D
    5. Marx CJ

    (2011) Diminishing returns epistasis among beneficial mutations decelerates adaptation

    Science 332:1190–1192.

    https://doi.org/10.1126/science.1203799

    • PubMed
    • Google Scholar
15. 1. Ciliberti S
    2. Martin OC
    3. Wagner A

    (2007) Robustness can evolve gradually in complex regulatory gene networks with varying topology

    PLoS Computational Biology 3:e15.

    https://doi.org/10.1371/journal.pcbi.0030015

    • PubMed
    • Google Scholar
16. 1. Colegrave N

    (2002) Sex releases the speed limit on evolution

    Nature 420:664–666.

    https://doi.org/10.1038/nature01191

    • PubMed
    • Google Scholar
17. 1. Cooper TF

    (2007) Recombination speeds adaptation by reducing competition between beneficial mutations in populations of Escherichia coli

    PLoS Biology 5:e225.

    https://doi.org/10.1371/journal.pbio.0050225

    • PubMed
    • Google Scholar
18. 1. Costanzo M
    2. Baryshnikova A
    3. Bellay J
    4. Kim Y
    5. Spear ED
    6. Sevier CS
    7. Ding H
    8. Koh JL
    9. Toufighi K
    10. Mostafavi S
    11. Prinz J
    12. St Onge RP
    13. VanderSluis B
    14. Makhnevych T
    15. Vizeacoumar FJ
    16. Alizadeh S
    17. Bahr S
    18. Brost RL
    19. Chen Y
    20. Cokol M
    21. Deshpande R
    22. Li Z
    23. Lin ZY
    24. Liang W
    25. Marback M
    26. Paw J
    27. San Luis BJ
    28. Shuteriqi E
    29. Tong AH
    30. van Dyk N
    31. Wallace IM
    32. Whitney JA
    33. Weirauch MT
    34. Zhong G
    35. Zhu H
    36. Houry WA
    37. Brudno M
    38. Ragibizadeh S
    39. Papp B
    40. Pál C
    41. Roth FP
    42. Giaever G
    43. Nislow C
    44. Troyanskaya OG
    45. Bussey H
    46. Bader GD
    47. Gingras AC
    48. Morris QD
    49. Kim PM
    50. Kaiser CA
    51. Myers CL
    52. Andrews BJ
    53. Boone C

    (2010) The genetic landscape of a cell

    Science 327:425–431.

    https://doi.org/10.1126/science.1180823

    • PubMed
    • Google Scholar
19. 1. Costanzo M
    2. VanderSluis B
    3. Koch EN
    4. Baryshnikova A
    5. Pons C
    6. Tan G
    7. Wang W
    8. Usaj M
    9. Hanchard J
    10. Lee SD
    11. Pelechano V
    12. Styles EB
    13. Billmann M
    14. van Leeuwen J
    15. van Dyk N
    16. Lin ZY
    17. Kuzmin E
    18. Nelson J
    19. Piotrowski JS
    20. Srikumar T
    21. Bahr S
    22. Chen Y
    23. Deshpande R
    24. Kurat CF
    25. Li SC
    26. Li Z
    27. Usaj MM
    28. Okada H
    29. Pascoe N
    30. San Luis BJ
    31. Sharifpoor S
    32. Shuteriqi E
    33. Simpkins SW
    34. Snider J
    35. Suresh HG
    36. Tan Y
    37. Zhu H
    38. Malod-Dognin N
    39. Janjic V
    40. Przulj N
    41. Troyanskaya OG
    42. Stagljar I
    43. Xia T
    44. Ohya Y
    45. Gingras AC
    46. Raught B
    47. Boutros M
    48. Steinmetz LM
    49. Moore CL
    50. Rosebrock AP
    51. Caudy AA
    52. Myers CL
    53. Andrews B
    54. Boone C

    (2016) A global genetic interaction network maps a wiring diagram of cellular function

    Science 353:aaf1420.

    https://doi.org/10.1126/science.aaf1420

    • PubMed
    • Google Scholar
20. 1. Couce A
    2. Tenaillon OA

    (2015) The rule of declining adaptability in microbial evolution experiments

    Frontiers in Genetics 6:99.

    https://doi.org/10.3389/fgene.2015.00099

    • PubMed
    • Google Scholar
21. 1. Cowperthwaite MC
    2. Economo EP
    3. Harcombe WR
    4. Miller EL
    5. Meyers LA

    (2008) The ascent of the abundant: how mutational networks constrain evolution

    PLoS Computational Biology 4:e1000110.

    https://doi.org/10.1371/journal.pcbi.1000110

    • PubMed
    • Google Scholar
22. 1. Deatherage DE
    2. Barrick JE

    (2014) Identification of mutations in laboratory-evolved microbes from next-generation sequencing data using breseq

    Methods in molecular biology 1151:165–188.

    https://doi.org/10.1007/978-1-4939-0554-6_12

    • PubMed
    • Google Scholar
23. 1. Draghi JA
    2. Parsons TL
    3. Wagner GP
    4. Plotkin JB

    (2010) Mutational robustness can facilitate adaptation

    Nature 463:353–355.

    https://doi.org/10.1038/nature08694

    • PubMed
    • Google Scholar
24. 1. Engel SR
    2. Dietrich FS
    3. Fisk DG
    4. Binkley G
    5. Balakrishnan R
    6. Costanzo MC
    7. Dwight SS
    8. Hitz BC
    9. Karra K
    10. Nash RS
    11. Weng S
    12. Wong ED
    13. Lloyd P
    14. Skrzypek MS
    15. Miyasato SR
    16. Simison M
    17. Cherry JM

    (2014) The reference genome sequence of Saccharomyces cerevisiae: then and now

    G3 4:389–398.

    https://doi.org/10.1534/g3.113.008995

    • PubMed
    • Google Scholar
25. 1. Goddard MR
    2. Godfray HC
    3. Burt A

    (2005) Sex increases the efficacy of natural selection in experimental yeast populations

    Nature 434:636–640.

    https://doi.org/10.1038/nature03405

    • PubMed
    • Google Scholar
26. 1. Gong LI
    2. Suchard MA
    3. Bloom JD

    (2013) Stability-mediated epistasis constrains the evolution of an influenza protein

    eLife 2:e00631.

    https://doi.org/10.7554/eLife.00631

    • PubMed
    • Google Scholar
27. 1. Good BH
    2. Desai MM

    (2015) The impact of macroscopic epistasis on long-term evolutionary dynamics

    Genetics 199:177–190.

    https://doi.org/10.1534/genetics.114.172460

    • PubMed
    • Google Scholar
28. 1. Gray JC
    2. Goddard MR

    (2012) Sex enhances adaptation by unlinking beneficial from detrimental mutations in experimental yeast populations

    BMC Evolutionary Biology 12:43.

    https://doi.org/10.1186/1471-2148-12-43

    • PubMed
    • Google Scholar
29. 1. Harms MJ
    2. Thornton JW

    (2014) Historical contingency and its biophysical basis in glucocorticoid receptor evolution

    Nature 512:203–207.

    https://doi.org/10.1038/nature13410

    • PubMed
    • Google Scholar
30. 1. Hietpas RT
    2. Jensen JD
    3. Bolon DN

    (2011) Experimental illumination of a fitness landscape

    PNAS 108:7896–7901.

    https://doi.org/10.1073/pnas.1016024108

    • PubMed
    • Google Scholar
31. 1. Jacquier H
    2. Birgy A
    3. Le Nagard H
    4. Mechulam Y
    5. Schmitt E
    6. Glodt J
    7. Bercot B
    8. Petit E
    9. Poulain J
    10. Barnaud G
    11. Gros PA
    12. Tenaillon O

    (2013) Capturing the mutational landscape of the beta-lactamase TEM-1

    PNAS 110:13067–13072.

    https://doi.org/10.1073/pnas.1215206110

    • PubMed
    • Google Scholar
32. 1. Jarosz DF
    2. Lindquist S

    (2010) Hsp90 and environmental stress transform the adaptive value of natural genetic variation

    Science 330:1820–1824.

    https://doi.org/10.1126/science.1195487

    • PubMed
    • Google Scholar
33. 1. Jasnos L
    2. Korona R

    (2007) Epistatic buffering of fitness loss in yeast double deletion strains

    Nature Genetics 39:550–554.

    https://doi.org/10.1038/ng1986

    • PubMed
    • Google Scholar
34. Software

    1. Jerison E
    2. Kryazhimskiy S

    (2017) Adaptability, version b71d3b9

    github.

    https://github.com/erjerison/adaptability
35. 1. Khan AI
    2. Dinh DM
    3. Schneider D
    4. Lenski RE
    5. Cooper TF

    (2011) Negative epistasis between beneficial mutations in an evolving bacterial population

    Science 332:1193–1196.

    https://doi.org/10.1126/science.1203801

    • PubMed
    • Google Scholar
36. 1. Kirschner M
    2. Gerhart J

    (1998) Evolvability

    PNAS 95:8420–8427.

    https://doi.org/10.1073/pnas.95.15.8420

    • PubMed
    • Google Scholar
37. 1. Kryazhimskiy S
    2. Rice DP
    3. Jerison ER
    4. Desai MM

    (2014) Microbial evolution. Global epistasis makes adaptation predictable despite sequence-level stochasticity

    Science 344:1519–1522.

    https://doi.org/10.1126/science.1250939

    • PubMed
    • Google Scholar
38. 1. Kvitek DJ
    2. Sherlock G

    (2013) Whole Genome, Whole Population Sequencing Reveals That Loss of Signaling Networks Is the Major Adaptive Strategy in a Constant Environment

    PLoS Genetics 9:e1003972–11.

    https://doi.org/10.1371/journal.pgen.1003972

    • Google Scholar
39. 1. Lang GI
    2. Botstein D
    3. Desai MM

    (2011) Genetic variation and the fate of beneficial mutations in asexual populations

    Genetics 188:647–661.

    https://doi.org/10.1534/genetics.111.128942

    • PubMed
    • Google Scholar
40. 1. Lang GI
    2. Rice DP
    3. Hickman MJ
    4. Sodergren E
    5. Weinstock GM
    6. Botstein D
    7. Desai MM

    (2013) Pervasive genetic hitchhiking and clonal interference in forty evolving yeast populations

    Nature 500:571–574.

    https://doi.org/10.1038/nature12344

    • PubMed
    • Google Scholar
41. 1. Lauring AS
    2. Frydman J
    3. Andino R

    (2013) The role of mutational robustness in RNA virus evolution

    Nature Reviews Microbiology 11:327–336.

    https://doi.org/10.1038/nrmicro3003

    • PubMed
    • Google Scholar
42. 1. Masel J
    2. Bergman A

    (2003) The evolution of the evolvability properties of the yeast prion [PSI${}^{\mathrm{+}}$+]

    Evolution 57:1498–1512.

    https://doi.org/10.1111/j.0014-3820.2003.tb00358.x

    • PubMed
    • Google Scholar
43. 1. Masel J
    2. Trotter MV

    (2010) Robustness and evolvability

    Trends in Genetics 26:406–414.

    https://doi.org/10.1016/j.tig.2010.06.002

    • PubMed
    • Google Scholar
44. 1. McDonald MJ
    2. Hsieh YY
    3. Yu YH
    4. Chang SL
    5. Leu JY

    (2012) The evolution of low mutation rates in experimental mutator populations of Saccharomyces cerevisiae

    Current Biology 22:1235–1240.

    https://doi.org/10.1016/j.cub.2012.04.056

    • PubMed
    • Google Scholar
45. 1. McDonald MJ
    2. Rice DP
    3. Desai MM

    (2016) Sex speeds adaptation by altering the dynamics of molecular evolution

    Nature 531:233–236.

    https://doi.org/10.1038/nature17143

    • PubMed
    • Google Scholar
46. 1. Montville R
    2. Froissart R
    3. Remold SK
    4. Tenaillon O
    5. Turner PE

    (2005) Evolution of mutational robustness in an RNA virus

    PLoS Biology 3:e381.

    https://doi.org/10.1371/journal.pbio.0030381

    • PubMed
    • Google Scholar
47. 1. Palmer AC
    2. Toprak E
    3. Baym M
    4. Kim S
    5. Veres A
    6. Bershtein S
    7. Kishony R

    (2015) Delayed commitment to evolutionary fate in antibiotic resistance fitness landscapes

    Nature Communications 6:7385.

    https://doi.org/10.1038/ncomms8385

    • PubMed
    • Google Scholar
48. 1. Perfeito L
    2. Sousa A
    3. Bataillon T
    4. Gordo I

    (2014) Rates of fitness decline and rebound suggest pervasive epistasis

    Evolution 68:150–162.

    https://doi.org/10.1111/evo.12234

    • PubMed
    • Google Scholar
49. 1. Quandt EM
    2. Deatherage DE
    3. Ellington AD
    4. Georgiou G
    5. Barrick JE

    (2014) Recursive genomewide recombination and sequencing reveals a key refinement step in the evolution of a metabolic innovation in Escherichia coli

    PNAS 111:2217–2222.

    https://doi.org/10.1073/pnas.1314561111

    • PubMed
    • Google Scholar
50. 1. Quandt EM
    2. Gollihar J
    3. Blount ZD
    4. Ellington AD
    5. Georgiou G
    6. Barrick JE

    (2015) Fine-tuning citrate synthase flux potentiates and refines metabolic innovation in the Lenski evolution experiment

    eLife 4:e09696.

    https://doi.org/10.7554/eLife.09696

    • PubMed
    • Google Scholar
51. 1. Salverda ML
    2. Dellus E
    3. Gorter FA
    4. Debets AJ
    5. van der Oost J
    6. Hoekstra RF
    7. Tawfik DS
    8. de Visser JA

    (2011) Initial mutations direct alternative pathways of protein evolution

    PLoS Genetics 7:e1001321.

    https://doi.org/10.1371/journal.pgen.1001321

    • PubMed
    • Google Scholar
52. 1. Sniegowski PD
    2. Gerrish PJ
    3. Lenski RE

    (1997) Evolution of high mutation rates in experimental populations of E. coli

    Nature 387:703–705.

    https://doi.org/10.1038/42701

    • PubMed
    • Google Scholar
53. 1. Soyer OS
    2. Pfeiffer T

    (2010) Evolution under fluctuating environments explains observed robustness in metabolic networks

    PLoS Computational Biology 6:e1000907.

    https://doi.org/10.1371/journal.pcbi.1000907

    • PubMed
    • Google Scholar
54. 1. Stern A
    2. Bianco S
    3. Yeh MT
    4. Wright C
    5. Butcher K
    6. Tang C
    7. Nielsen R
    8. Andino R

    (2014) Costs and benefits of mutational robustness in RNA viruses

    Cell Reports 8:1026–1036.

    https://doi.org/10.1016/j.celrep.2014.07.011

    • PubMed
    • Google Scholar
55. 1. Szamecz B
    2. Boross G
    3. Kalapis D
    4. Kovács K
    5. Fekete G
    6. Farkas Z
    7. Lázár V
    8. Hrtyan M
    9. Kemmeren P
    10. Groot Koerkamp MJ
    11. Rutkai E
    12. Holstege FC
    13. Papp B
    14. Pál C
    15. Koerkamp MJG

    (2014) The genomic landscape of compensatory evolution

    PLoS Biology 12:e1001935.

    https://doi.org/10.1371/journal.pbio.1001935

    • PubMed
    • Google Scholar
56. 1. Szappanos B
    2. Kovács K
    3. Szamecz B
    4. Honti F
    5. Costanzo M
    6. Baryshnikova A
    7. Gelius-Dietrich G
    8. Lercher MJ
    9. Jelasity M
    10. Myers CL
    11. Andrews BJ
    12. Boone C
    13. Oliver SG
    14. Pál C
    15. Papp B

    (2011) An integrated approach to characterize genetic interaction networks in yeast metabolism

    Nature Genetics 43:656–662.

    https://doi.org/10.1038/ng.846

    • PubMed
    • Google Scholar
57. 1. van Nimwegen E
    2. Crutchfield JP
    3. Huynen M

    (1999) Neutral evolution of mutational robustness

    PNAS 96:9716–9720.

    https://doi.org/10.1073/pnas.96.17.9716

    • PubMed
    • Google Scholar
58. 1. van Opijnen T
    2. Bodi KL
    3. Camilli A

    (2009) Tn-seq: high-throughput parallel sequencing for fitness and genetic interaction studies in microorganisms

    Nature Methods 6:767–772.

    https://doi.org/10.1038/nmeth.1377

    • PubMed
    • Google Scholar
59. 1. Venkataram S
    2. Dunn B
    3. Li Y
    4. Agarwala A
    5. Chang J
    6. Ebel ER
    7. Geiler-Samerotte K
    8. Hérissant L
    9. Blundell JR
    10. Levy SF
    11. Fisher DS
    12. Sherlock G
    13. Petrov DA

    (2016) Development of a comprehensive genotype-to-fitness map of adaptation-driving mutations in yeast

    Cell 166:1585–1596.

    https://doi.org/10.1016/j.cell.2016.08.002

    • PubMed
    • Google Scholar
60. 1. Visscher PM
    2. Medland SE
    3. Ferreira MA
    4. Morley KI
    5. Zhu G
    6. Cornes BK
    7. Montgomery GW
    8. Martin NG

    (2006) Assumption-free estimation of heritability from genome-wide identity-by-descent sharing between full siblings

    PLoS Genetics 2:e41.

    https://doi.org/10.1371/journal.pgen.0020041

    • PubMed
    • Google Scholar
61. 1. Wang Y
    2. Diaz Arenas C
    3. Stoebel DM
    4. Flynn K
    5. Knapp E
    6. Dillon MM
    7. Wünsche A
    8. Hatcher PJ
    9. Moore FB
    10. Cooper VS
    11. Cooper TF

    (2016) Benefit of transferred mutations is better predicted by the fitness of recipients than by their ecological or genetic relatedness

    PNAS 113:5047–5052.

    https://doi.org/10.1073/pnas.1524988113

    • PubMed
    • Google Scholar
62. 1. Weinreich DM
    2. Delaney NF
    3. Depristo MA
    4. Hartl DL

    (2006) Darwinian evolution can follow only very few mutational paths to fitter proteins

    Science 312:111–114.

    https://doi.org/10.1126/science.1123539

    • PubMed
    • Google Scholar
63. 1. Weinreich DM
    2. Lan Y
    3. Wylie CS
    4. Heckendorn RB

    (2013) Should evolutionary geneticists worry about higher-order epistasis?

    Current Opinion in Genetics & Development 23:700–707.

    https://doi.org/10.1016/j.gde.2013.10.007

    • PubMed
    • Google Scholar
64. 1. Weinreich DM
    2. Watson RA
    3. Chao L

    (2005) Perspective: Sign epistasis and genetic constraint on evolutionary trajectories

    Evolution; international journal of organic evolution 59:1165–1174.

    https://doi.org/10.1554/04-272

    • PubMed
    • Google Scholar
65. 1. Wielgoss S
    2. Barrick JE
    3. Tenaillon O
    4. Wiser MJ
    5. Dittmar WJ
    6. Cruveiller S
    7. Chane-Woon-Ming B
    8. Médigue C
    9. Lenski RE
    10. Schneider D

    (2013) Mutation rate dynamics in a bacterial population reflect tension between adaptation and genetic load

    PNAS 110:222–227.

    https://doi.org/10.1073/pnas.1219574110

    • PubMed
    • Google Scholar
66. 1. Wilke CO
    2. Wang JL
    3. Ofria C
    4. Lenski RE
    5. Adami C

    (2001) Evolution of digital organisms at high mutation rates leads to survival of the flattest

    Nature 412:331–333.

    https://doi.org/10.1038/35085569

    • PubMed
    • Google Scholar
67. 1. Woods RJ
    2. Barrick JE
    3. Cooper TF
    4. Shrestha U
    5. Kauth MR
    6. Lenski RE

    (2011) Second-order selection for evolvability in a large Escherichia coli population

    Science 331:1433–1436.

    https://doi.org/10.1126/science.1198914

    • PubMed
    • Google Scholar
68. 1. Wünsche A
    2. Dinh DM
    3. Satterwhite RS
    4. Arenas CD
    5. Stoebel DM
    6. Cooper TF

    (2017) Diminishing-returns epistasis decreases adaptability along an evolutionary trajectory

    Nature Ecology & Evolution 1:0061.

    https://doi.org/10.1038/s41559-016-0061

    • PubMed
    • Google Scholar
69. 1. Xiao Y
    2. Rouzine IM
    3. Bianco S
    4. Acevedo A
    5. Goldstein EF
    6. Farkov M
    7. Brodsky L
    8. Andino R

    (2016) RNA Recombination Enhances Adaptability and Is Required for Virus Spread and Virulence

    Cell Host & Microbe 19:493–503.

    https://doi.org/10.1016/j.chom.2016.03.009

    • PubMed
    • Google Scholar
70. 1. Yang J
    2. Benyamin B
    3. McEvoy BP
    4. Gordon S
    5. Henders AK
    6. Nyholt DR
    7. Madden PA
    8. Heath AC
    9. Martin NG
    10. Montgomery GW
    11. Goddard ME
    12. Visscher PM

    (2010) Common SNPs explain a large proportion of the heritability for human height

    Nature Genetics 42:565–569.

    https://doi.org/10.1038/ng.608

    • PubMed
    • Google Scholar
71. 1. Zeyl C
    2. Bell G

    (1997) The advantage of sex in evolving yeast populations

    Nature 388:465–468.

    https://doi.org/10.1038/41312

    • PubMed
    • Google Scholar
72. 1. Zuk O
    2. Hechter E
    3. Sunyaev SR
    4. Lander ES

    (2012) The mystery of missing heritability: Genetic interactions create phantom heritability

    PNAS 109:1193–1198.

    https://doi.org/10.1073/pnas.1119675109

    • PubMed
    • Google Scholar

Author details

1. Elizabeth R Jerison

   1. Department of Organismic and Evolutionary Biology, Harvard University, Cambridge, United States
   2. Department of Physics, Harvard University, Cambridge, United States
   3. FAS Center for Systems Biology, Harvard University, Cambridge, United States

   Contribution

   Conceptualization, Software, Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review and editing

   Contributed equally with

   Sergey Kryazhimskiy

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-3793-8839

2. Sergey Kryazhimskiy

   Section of Ecology, Behavior and Evolution, Division of Biological Sciences, University of California, San Diego, San Diego, United States

   Contribution

   Conceptualization, Software, Formal analysis, Investigation, Methodology, Writing—original draft, Writing—review and editing

   Contributed equally with

   Elizabeth R Jerison

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-9128-8705

3. James Kameron Mitchell

   Department of Physics, Harvard University, Cambridge, United States

   Contribution

   Investigation, Data acquisition and analysis

   Competing interests

   No competing interests declared

4. Joshua S Bloom

   Department of Human Genetics, University of California, Los Angeles, Los Angeles, United States

   Contribution

   Formal analysis, Methodology, Writing—review and editing

   Competing interests

   No competing interests declared

5. Leonid Kruglyak

   Department of Human Genetics, University of California, Los Angeles, Los Angeles, United States

   Contribution

   Investigation, Methodology, Writing—review and editing

   Competing interests

   Reviewing editor, eLife

   "This ORCID iD identifies the author of this article:" 0000-0002-8065-3057

6. Michael M Desai

   1. Department of Organismic and Evolutionary Biology, Harvard University, Cambridge, United States
   2. Department of Physics, Harvard University, Cambridge, United States
   3. FAS Center for Systems Biology, Harvard University, Cambridge, United States

   Contribution

   Conceptualization, Formal analysis, Funding acquisition, Writing—original draft, Writing—review and editing

   For correspondence

   mdesai@oeb.harvard.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-9581-1150

• 4,030

  views

• 548

  downloads

• 67

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.