Home | About | Journals | Submit | Contact Us | Français |

**|**Proc Biol Sci**|**v.279(1745); 2012 October 22**|**PMC3441082

Formats

Article sections

Authors

Related links

Proc Biol Sci. 2012 October 22; 279(1745): 4156–4164.

Published online 2012 August 15. doi: 10.1098/rspb.2012.1449

PMCID: PMC3441082

Received 2012 June 24; Accepted 2012 July 20.

Copyright This journal is © 2012 The Royal Society

This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

This article has been cited by other articles in PMC.

Epistasis between mutations in two genes is thought to reflect an interdependence of their functions. While sometimes epistasis is predictable using mechanistic models, its roots seem, in general, hidden in the complex architecture of biological networks. Here, we ask how epistasis can be quantified based on the mathematical dependence of a system-level trait (e.g. fitness) on lower-level traits (e.g. molecular or cellular properties). We first focus on a model in which fitness is the difference between a benefit and a cost trait, both pleiotropically affected by mutations. We show that despite its simplicity, this model can be used to analytically predict certain properties of the ensuing distribution of epistasis, such as a global negative bias, resulting in antagonism between beneficial mutations, and synergism between deleterious ones. We next extend these ideas to derive a general expression for epistasis given an arbitrary functional dependence of fitness on other traits. This expression demonstrates how epistasis relative to fitness can emerge despite the absence of epistasis relative to lower level traits, leading to a formalization of the concept of independence between biological processes. Our results suggest that epistasis may be largely shaped by the pervasiveness of pleiotropic effects and modular organization in biological networks.

Epistasis describes a fundamental nonlinearity in biological systems, capturing the fact that the phenotypic effect of a genetic mutation or allele variant could depend on another mutation. Epistasis is suggested to play an important role in evolutionary dynamics, e.g. by shaping the fitness landscape [1,2], maintaining sexual reproduction [3,4] and affecting the speed of adaptation [5–8]. Large-scale systematic studies of single and double gene deletions have also revealed that epistasis is useful in understanding the organization of living systems into modules, as genes belonging to the same biological process tend to share similar profiles of epistatic interactions with other genes [9,10]. Given the influence of epistasis on evolutionary processes, and given the modular structure of epistatic interaction networks, one may wonder whether deeper insights can be obtained on how the modular organization of biological networks affects (and is affected by) epistasis and adaptation. The connection between modularity and evolution has been addressed from multiple independent standpoints [11–16]. Yet few concepts seem to be more appropriate than epistasis for trying to relate the architecture of biological networks with the evolutionary forces that gave rise to such networks. This is ultimately due to the fact that epistasis captures in an elementary way the complexity of the genotype–phenotype map.

For quantitative traits, epistasis can be analysed mathematically. Upon a double mutation, a given quantitative trait could turn out to be higher or lower than expected based on individual mutations, leading to a quantifiable positive or a negative epistasis, respectively. Special caution should be used in comparing the definitions of epistasis across different research areas, as multiple, sometime conflicting, terminologies have been adopted. A commonly used classification is the one between synergistic and antagonistic epistasis. Synergistic (antagonistic) epistasis occurs when the joint effect of two alleles is more (less) severe than expected. Note that, based on the terminology adopted here, negative epistasis between deleterious mutations corresponds to synergism, while negative epistasis between beneficial mutations is classified as antagonism (table 1). In addition, the very definition of the baseline expectation relative to which epistasis is quantified is still the subject of active debate [17–20]. In this work, we assume that mutational effects combine multiplicatively in the absence of epistasis (see also §2).

Definitions and conventions for epistasis in the current work. We quantify the degree of epistasis () as the deviation of the joint effect of mutations from the expectation in multiplicative scale (see equation (2.1)). Synergistic epistasis occurs **...**

The role of epistasis in evolution constitutes a particularly important and debated question [21], which involves understanding the sources and the consequences of the average and variance of the distribution of epistatic effects [22]. Therefore, considerable effort has been put into elucidating the distribution of epistasis, both from a theoretical and from an experimental perspective [23,24]. Some early experimental studies, partially motivated by the possible relevance of synergistic effects between deleterious mutations in the evolution of sex [3,4], had found nearly symmetric distributions of epistasis [25–28]. Subsequent large-scale studies under different proportions of beneficial versus deleterious mutations have reported both negative [10] and positive [29] epistasis trends, leaving the problem fundamentally unsettled. From a theoretical perspective, models of fitness landscapes have been used for providing potential explanations for observed epistasis trends. Antagonism between beneficial mutations, for example, can be predicted by a fitness landscape model that assumes ubiquity of stabilizing selection [30]. In addition, stoichiometric models of metabolic networks have been useful in exploring the distribution and network organization of gene–gene interactions in metabolism [9,31], and in providing mechanistic explanations for measured distributions [29,32]. However, none of these prior theoretical works seems to have explicitly addressed the question of how epistasis quantitatively depends on the modular organization of the genotype–phenotype mapping, and on the degree of pleiotropy.

A hint to how modularity and epistasis relate to each other was recently offered by an experimental evolution study that identified a diminishing returns trend among pairs of beneficial alleles [7] (see also recent studies [8,33]). This study reported antagonistic epistasis between beneficial alleles that arose during 600 generations of evolution of an engineered strain of *Methylobacterium extorquens*. It was found that the observed beneficial alleles improve fitness either by enhancing metabolic capacity or by alleviating protein expression-related costs. This observation suggested that microbial fitness could very coarsely be treated as a modularly structured function of two separate phenotypic traits, a metabolic benefit and a protein expression cost. In fact, a simple benefit–cost model was sufficient to quantitatively predict fitness values for multi-allele strains, and to quantitatively recapitulate the antagonistic trend for epistasis between genes in different loci of the genome [7]. Benefit–cost models had been used to describe fitness in previous studies of evolutionary adaptation, ranging from rapid adaptation of protein expression in new environments [14], to optimal regulatory design upon biochemical noise [34] and optimal transcriptional regulation of metabolism [35].

Here, inspired by the previously demonstrated relevance of benefit–cost models in evolutionary research, we use a benefit–cost model as a starting point for quantifying epistasis based on the dependence of fitness on multiple traits. In particular, we first extend the benefit–cost model from a way to explain an observed case of antagonism between beneficial mutations [7], to a general statistical analysis of expected epistasis distributions among mutations with a given chance of being beneficial or deleterious. In the second part of our work, we generalize these results to show that the degree of epistasis can be estimated analytically for an arbitrary dependence of fitness on simpler quantitative traits, providing a new mathematical link between epistasis, pleiotropy and modularity. Note that our analysis is mostly focused on analysing the interactions between two mutations, without delving into the problem of how multiple perturbations accumulate. In addition, we restrict our calculations almost entirely to a fitness function that depends on two traits (also previously called two-dimensional epistasis [36]), leaving possible extensions to multidimensional epistasis [36] as a topic for potential follow-up research.

The main premise of the current work is that a high-level trait, or phenotype, such as fitness (*f*) can be phenomenologically expressed as a function *F* of two basic observable traits, *X* and *Y*: *f* = *F*(*X*,*Y*). For example, the growth rate of a bacterium may be expressed as a function of its respiratory and fermenting capacities [37]. In addition to a wild-type organism (e.g. a bacterial strain), we consider two mutant strains with genetic modifications at loci *i* and *j*, respectively, and a double mutant strain which has both *i* and *j* alleles modified. For each of these four strains, we hypothesize that it is possible to independently measure the overall fitness (*f*_{0} for the wild-type, *f _{i}* and

Schematic depiction of how we quantify epistasis relative to a fitness function that depends on two quantitative traits, or phenotypes. (*a*) Two alleles or genetic perturbations *i* and *j* are assumed to potentially affect multiple traits, here *X* and *Y* (‘low-level **...**

In the absence of epistasis, it is often assumed that mutational effects should combine multiplicatively, i.e. *f _{ij}*/

2.1

This equation has been used previously [7–10,25,31,38] for assessing the degree of epistasis, given experimentally measured or computationally predicted fitness values for the wild-type and mutant strains. It is through this equation that one can determine whether epistasis is positive (* > 0*) or negative (* < 0*). In this work, we will use this equation to determine how an analytical expression for fitness as a function of other measurable traits (and of the mutations that affect such traits), translates into a prediction of epistasis. We will restrict our analysis to the simplest case in which alleles exhibit no direct epistasis with regard to either of the two underlying traits *X* and *Y* (i.e. *x _{ij}*/

The benefit–cost model used to explain an antagonistic epistasis pattern that emerged during adaptation of an *M. extorquens* strain [7] constitutes a special case of the *F* function introduced earlier. In this case, in analogy to Dekel & Alon [14], *X* = *b* is the growth advantage (benefit), and *Y* = *c* is the burden (cost) associated with the operation of the biological network, and *F = X* − *Y = b* – *c*. In this model, fitness of the wild-type strain can be written as the difference between a basal benefit and a basal cost term:

2.2

The quantities *f*_{0} and *c*_{0} can be experimentally measured to operationally define the basal benefit *b*_{0} as demonstrated in earlier studies [7,14]. A mutant allele *i* is assumed to pleiotropically alter *b*_{0} or *c*_{0} with factors *λ _{i}* and

2.3

To avoid confusion, it is important to stress that *λ _{i}* represents the effect of a single mutation (

2.4

Note that the changes in each trait are not simply given by the *λ* and *θ* factors, but rather also include a dependence on the unperturbed trait, e.g. *Δ**b = λ_{i}b*

Our first goal is to determine analytically the magnitude and sign of epistasis relative to fitness under the assumptions of the benefit–cost model. As shown in detail in electronic supplementary material, §B, this can be achieved by substituting the expressions for the fitness of single and double mutants (equations (2.3) and (2.4)) into the definition of epistasis (equation (2.1)), yielding:

3.1

This is a remarkably simple expression, in which epistasis turns out to be computable as the product of a term that depends only on the unperturbed state parameters (*b*_{0}, *c*_{0}), and on a term that depends only on the phenotypic effects of the mutations (the *λ _{i}, θ_{i}, λ_{j}* and

Equation (3.1) predicts the degree of epistasis for specific combinations of two mutations. In this section, we show how this same equation can be also used to draw general conclusions about the expected distribution of epistasis between a pair of perturbations in the benefit–cost model. The fact that the sign of epistasis is dictated by the product of (*λ*–*θ*) terms suggests that a statistical analysis of how these terms are distributed could provide information about the distribution of epistasis itself.

To reason about this problem, one can visualize possible choices of perturbations on the (*λ*,*θ*) plane (figure 2). Each perturbation in the benefit–cost model can be represented as a point in this plane. The point of coordinates (1,1) corresponds to the wild-type, i.e. the unperturbed system. For simplicity, we assume here that *λ* and *θ* cannot exceed a given value *W*. Two fundamental lines can be drawn on this plane. One line, which we call the *isochange line*, is defined by the equation *λ* = *θ*, and corresponds to all individual perturbations that change both the benefit and the cost by the same multiplicative factor. Note that the isochange line does not have in itself an immediate interpretation in terms of epistasis (e.g. it does not represent a boundary between positive and negative epistasis); rather, as reasoned later, it is an abstract geometrical construct that will help us to determine the chance of observing a given sign of epistasis for two mutations, through equation (3.1). If we think of two mutations as two points with coordinates (*λ _{i}*,

Estimating epistasis through a geometrical representation of perturbations in phenotype space. (*a*) The (*λ*,*θ*) plane, a geometrical representation of possible mutant alleles in a benefit–cost model of fitness. Any allele (e.g. *i* **...**

A second fundamental line in the (*λ*,*θ*) plane is the line that partitions beneficial from deleterious mutations. This line, which we call *neutrality line* (*λ _{i}b*

Contingency table for the phenotypic values of strictly beneficial alleles. The categories of classified by the four conditions are analogous to the possible outcomes of tossing a coin twice, and allow us to compute the overall probability of **...**

While the geometrical arguments illustrated earlier provide estimates of the bias in the distribution of as a function of the fraction, *ρ*, of beneficial mutations, they do not allow us to predict the shape of distribution itself. We used computer simulations (see the electronic supplementary material, §A) to explore the full spectrum of epistasis distribution as a function of *ρ*. Again, here *ρ* is a phenotypic measure; we make no assumption on the connection between random mutations and specific values of *ρ*. The simulations confirm that negative is more likely to occur over the whole range of *ρ* values (figure 3*a*). The bias (prevailing negative epistasis) reaches a maximum in both extreme cases (strictly beneficial or strictly deleterious; figure 2*c*,*d*) while it becomes less and less pronounced as *ρ* approaches 0.5 (figure 3*a*,*b*). Note that for distributions derived from combinations of beneficial and deleterious mutations, a negative cannot be easily associated with synergistic or antagonistic trends, as different pairs contributing to the distribution will have different effects (including sign epistasis, an interesting case where an allele is beneficial on some genetic backgrounds but deleterious on others; see electronic supplementary material, figure S7 for more details). Computer simulations also indicate that these epistasis trends are robust over a broad range of values for *c*_{0} (see the electronic supplementary material, figure S2), and that they would equally ensue in a more complex model involving multiple cost components (e.g. *f*_{0} = *b*_{0} − *c*_{0} − *d*_{0} − *e*_{0}, electronic supplementary material, figure S1).

Most of our study so far has been focused on a specific dependence of fitness on two traits, i.e. the difference between a benefit and a cost trait. We next generalize our analysis to ask whether it is possible to estimate epistasis when fitness depends in an arbitrary way on two quantitative traits *X* and *Y*, e.g. it is the sum, the product or any arbitrary function *F* of such traits. As performed for the derivation of equation (3.1), we need to substitute the expressions for fitness of single and double mutant strains into equation (2.1) to compute * = F*(*x _{ij},y_{ij}*)

3.2

where , , , and are the partial derivatives of *F* computed at *x*_{0}, *y*_{0}. The first two terms in (3.2) quantify the contribution to epistasis through individual traits *X* and *Y*, respectively, whereas the third term is associated with their functional coupling. As a generalized form of equation (3.1), equation (3.2) provides a direct quantitative link between the magnitude and direction of epistasis and the magnitude of individual mutations. For a function *F = X − Y*, upon expressing the *Δ**x* and *Δ**y* parameters in terms of *λ* and *θ*, it is easy to show that equation (3.2) yields the benefit–cost result of equation (3.1) (see electronic supplementary material, §D). One can further use equation (3.2) to explore expected epistasis under other possible functional dependencies. For example, if *F* is a linear combination of the two phenotypes, *F*(*X*,*Y*) = *aX* + *bY*, then the degree of epistasis converges to a generalized form of equation (3.1), where . In this case, the relative signs of *a* and *b* determine whether the expected distribution of epistasis will display a positive or negative bias. In addition, as expected, one can verify that, if *F*(*X*,*Y*) = *XY*, then epistasis is always zero, compatible with the idea that multiplicative effects on individual traits will combine to provide an overall multiplicative effect on fitness. Interestingly, this is also true for functions of the form *F*(*X,Y*) *= X ^{n}Y^{m}*, hinting to a broader view of the relationship between epistasis and independence, as explored in detail later.

An important question one can ask using the expression of equation (3.2) is what degree of epistasis should be expected between two mutations each affecting only one of the two traits (i.e. non-pleiotropic mutations; figure 1*b*). Does our model support the general intuition that mutations affecting independent modules in a biological system should have no epistasis? Equation (3.2) allows us to ask this question in a formal way. In our framework, lack of pleiotropy is expressed by assuming, for example, that allele *i* affects only trait *X*, and allele *j* affects only *Y* (i.e. *Δ**x _{j} =*

3.3

Equation (3.3) implies that, even if each mutation affects only one of the two phenotypes (figure 1*b*), epistasis will be zero only for the set of functions *F* that satisfy the condition ^{2}log*F*/*x**y =* 0. One can verify (see the electronic supplementary material, §D) that this condition is satisfied by any function decomposable as

3.4

Note that, as opposed to functions such as *F*(*X,Y*) *= X ^{n}Y^{m}*, which would always give zero epistasis (irrespective of whether mutations are pleiotropic or not), one can think of several other

The consequences of equations (3.2)–(3.4) can be summarized as follows: (i) in the general case (arbitrary *F*), irrespective of whether mutations affect individual or multiple traits, one cannot necessarily expect zero epistasis relative to fitness, even if no epistasis is assumed relative to each individual trait; (ii) when *F* can be decomposed as in equation (3.4), epistasis can exist only in the presence of pleiotropy, i.e. if each mutation affects more than one trait. Hence, equation (3.4) can be viewed as an operational definition of independence between two traits; and (iii) in particular cases (e.g. if *F* is simply the product of two traits), epistasis will be always zero, irrespective of pleiotropy (table 3).

Genome-wide epistatic profiles of fitness have been used to study the functional organization of biological systems [10], suggesting the existence of functionally coherent modules, characterized by specific epistatic interaction network properties such as monochromaticity [9,40]. Yet the relationship between the modular organization of the cell and epistasis remains poorly understood. Inspired by the successful attempt to explain epistasis data with a simple benefit–cost model of microbial fitness [7], we sought to explore the broader implications of expressing fitness as a function of two quantitative traits. We found that epistasis, and specific biases in its distribution, could be a natural outcome of the dependence of fitness on multiple phenotypes. If we interpret different phenotypes as metrics associated with different subsystems, or modules, we can determine whether epistasis exists at the system-level, and what might be the bias in its distribution based on how two modules interact with each other to produce fitness. Thus, we suggest that our approach establishes a novel link between biological modularity, pleiotropy and epistasis.

The benefit–cost model, originally employed to explain a small number of interactions in a single evolutionary experiment, has been shown here to lead to a global bias in the distribution of epistasis, under a broad range of beneficial versus deleterious mutation frequencies. In particular, upon deviating from symmetry in the amount of beneficial and deleterious mutations, we predict an overall pattern of negative . A similar pattern was previously suggested to be informative in identifying physically interacting partners or gene pairs belonging to redundant, but parallel functional pathways [10]. Besides its relevance to functional genomics, a trend towards negative also plays an important role in evolutionary theories seeking to interpret origin and maintenance of sex and recombination [3,4]. On the other hand, for adaptation, negative indicates that beneficial alleles combine antagonistically and suggests a diminishing returns trend in the fitness improvements as more beneficial alleles are acquired, consistent with some previous experimental results and theoretical analyses [7,8,41–43]. Thus, based on simple assumptions about the functional dependence of fitness on multiple phenotypes and genetic perturbations, a single ‘symmetry breaking’ mechanism could provide a potential explanation for both the deceleration of adaptation upon accumulation of beneficial mutations, and the prevalence of synergistic interactions between deleterious mutations. In pondering the general relevance of the above results to biology, one cannot avoid asking whether and why a simple benefit–cost model should truly underlie trends of epistasis, potentially across different organisms and biological scales. Indeed, we do not expect that a benefit–cost model should be the common mechanism behind all observed epistasis trends (see counter-example in Chou *et al*. [44]). However, it is not inconceivable that under some circumstances, fitness could be effectively represented as the overall difference of two independently measurable terms, for example, in cases where most mutations are estimated to affect the abundance of different proteins in the cell. While direct experimental testing is beyond the scope of the current paper, it is worth mentioning that the quantitative prediction of negative epistasis bias in our benefit–cost model does not deviate much from the corresponding value observed for epistatic interactions between yeast deletion mutants [10] upon matching our model's deleterious/beneficial allele ratio to the 3 : 1 ratio present in the yeast data (see the electronic supplementary material, figure S8). Given that our model is only remotely related to the yeast deletion data, this result may be pure coincidence. However, it exemplifies how the result we are presenting could be tested against experimental data in the future.

Whether or not the negative bias we observe in our model is directly relevant for the discussion on the evolution of recombination remains to be seen. Experimental data have provided conflicting results [23,29,32,45], depending on the system used (yeast, bacteria, viruses), the methods involved (classical versus high throughput), the different criteria for selecting genes to be analysed (highly deleterious only, YPD essential genes, etc.), and the definition of epistasis employed. In addition, while robust with respect to several parameters, the bias observed in the benefit–cost model may still in principle change in magnitude or sign, under different assumptions on the underlying distribution of individual mutations (see also electronic supplementary material, figures S2–S6). In any case, similar to prior computational models [9], we believe that the framework we are proposing will be useful in explaining, and potentially motivating experimental measurements relevant for this question.

Beyond the benefit–cost model, we showed in equation (3.2) that it is possible to estimate how epistasis depends on the magnitude of individual mutations and on the functional dependence of fitness on the two traits *X* and *Y*. For simple functional dependencies, this equation leads to direct insight into the type of epistasis to be expected. For example, linear combinations of traits lead to epistatic effects formally similar to the ones obtained with the benefit–cost model, except that the sign (and therefore the distribution biases) of epistasis is heavily influenced by the signs of the coefficients of such linear combinations. Future experiments (e.g. measuring epistasis upon perturbations of metabolic pathways that combine additively to produce fitness) could directly test this prediction. Our analytical expression is derived upon analysing the partial derivative of fitness with respect to traits it depends on, similar to the concept of ‘phenotype landscape’ proposed before to address the evolution of canalization, phenotypic plasticity and integration [46]. In our approach, however, we explicitly take into account both the effects of different mutations on distinct traits, and the functional dependence of fitness on such traits. One of the most interesting consequences of our general expression for epistasis as a function of two quantitative traits is the possibility to infer a general class of functional dependencies that guarantee lack of epistasis in the absence of pleiotropy. This finding echoes the viewpoint of a recent review that pleiotropy is an important prerequisite for epistasis [24]. Our result establishes a formal link between epistasis and pleiotropy, and suggests a new way to think about independence in biological networks.

Our method is simple and analytically solvable. Future variants of our framework could address more complex or alternative scenarios: first, given the multiplicity of traits that may be thought of as contributing to fitness, one could extend the current approach to fitness functions that depend upon more than two phenotypes (as we preliminarily explored here through computer simulations for the benefit–cost model). Second, while we have assumed here that perturbations correspond to genetic mutations, one could explore the consequences of a similar model for environmental perturbations. Third, it may be interesting to generalize our expression for epistasis to the case in which the basic traits themselves do have some degree of epistasis. In such case, one could seek an ‘epistasis propagation law’, showing how epistasis at a low-level phenotype affects epistasis at higher levels. Fourth, it has been suggested that the purging of deleterious mutations depends on the magnitude of mutational effects, in addition to the bias in the distribution of epistasis [23,47]. Future extension of our equation (3.2) in regard to linkage disequilibrium could further address this view point from another perspective to further elucidate this critical point about the evolution of sexual reproduction. Fifth, as we show in electronic supplementary material, figure S7, our model can account for instances of reciprocal sign epistasis between two alleles, providing potential new avenues for studying the ruggedness of evolutionary landscapes.

Finally, while in the current work we have focused on fitness and on its dependence on other traits, our formulation is quite general, and should equally apply to the functional dependence of any trait on any other set of traits. Even if most genetic interaction data are obtained relative to fitness, epistasis relative to multiple phenotypes will probably become increasingly available [31,48], offering opportunities to study how epistasis propagates between different traits to ultimately shape the genotype–phenotype mapping. We hope also that our equation will help understand epistasis relative to traits associated with genetic diseases, and provide insight on the interplay between evolution and modular organization of biological systems.

We are grateful to Evan Snitkin, Niels Klitgord, Ed Reznik, Nigel Delaney and Deepa Agashe and other members of the Segrè and Marx Laboratories for helpful suggestions and feedback on the manuscript, and to Bjorn Poonen for suggesting a criterion for functional decomposability. This work was supported by NIH grants (R01 GM078209 and R01 GM089978), and by the NASA Astrobiology Institute.

1. Weinreich D. M., Delaney N. F., DePristo M. A., Hartl D. L.
2006.
Darwinian evolution can follow only very few mutational paths to fitter proteins. Science
312, 111–114 (doi:10.1126/science.1123539)10.1126/science.1123539 [PubMed] [Cross Ref]

2. Hayden E. J., Ferrada E., Wagner A.
2011.
Cryptic genetic variation promotes rapid evolutionary adaptation in an RNA enzyme. Nature
474, 92–95 (doi:10.1038/nature10083)10.1038/nature10083 [PubMed] [Cross Ref]

3. Kondrashov A. S.
1988.
Deleterious mutations and the evolution of sexual reproduction. Nature
336, 435–440 (doi:10.1038/336435a0)10.1038/336435a0 [PubMed] [Cross Ref]

4. Otto S. P.
2009.
The evolutionary enigma of sex. Am. Nat.
174, S1–S14 (doi:10.1086/599084)10.1086/599084 [PubMed] [Cross Ref]

5. Sanjuan R., Cuevas J. M., Moya A., Elena S. F.
2005.
Epistasis and the adaptability of an RNA virus. Proc. Natl Acad. Sci. USA
170, 1001–1008 (doi:10.1534/genetics.105.040741)10.1534/genetics.105.040741 [PubMed] [Cross Ref]

6. Kryazhimskiy S., Tkačik G., Plotkin J. B.
2009.
The dynamics of adaptation on correlated fitness landscapes. Proc. Natl. Acad. Sci. USA
106, 18 638–18 643 (doi:10.1073/pnas.0905497106)10.1073/pnas.0905497106 [PubMed] [Cross Ref]

7. Chou H.-H., Chiu H.-C., Delaney N. F., Segrè D., Marx C. J.
2011.
Diminishing returns epistasis among beneficial mutations decelerates adaptation. Science
332, 1190–1192 (doi:10.1126/science.1203799)10.1126/science.1203799 [PMC free article] [PubMed] [Cross Ref]

8. Kahn A. I., Dinh D. M., Schneider D., Lenski R. E., Cooper T. F.
2011.
Negative epistasis between beneficial mutations slows the rate of population adaptation. Science
332, 1193–1196 (doi:10.1126/science.1203801)10.1126/science.1203801 [PubMed] [Cross Ref]

9. Segrè D., DeLuna A., Church G. M., Kishony R.
2005.
Modular epistasis in yeast metabolism. Nat. Genet.
37, 77–83 (doi:10.1038/ng1489)10.1038/ng1489 [PubMed] [Cross Ref]

10. Costanzo M., et al.
2010.
The genetic landscape of a cell. Science
327, 425–431 (doi:10.1126/science.1180823)10.1126/science.1180823 [PubMed] [Cross Ref]

11. Hartwell L. H., Hopfield J. J., Leibler S., Murray A. W.
1999.
From molecular to modular cell biology. Nature
402, C47–C52 (doi:10.1038/35011540)10.1038/35011540 [PubMed] [Cross Ref]

12. Lawrence J. G., Roth J. R.
1996.
Selfish operons: horizontal transfer may drive the evolution of gene clusters. Genetics
143, 1843–1860 [PubMed]

13. Kreimer A., Borenstein E., Gophna U., Ruppin E.
2008.
The evolution of modularity in bacterial metabolic networks. Proc. Natl Acad. Sci. USA
105, 6976–6981 (doi:10.1073/pnas.0712149105)10.1073/pnas.0712149105 [PubMed] [Cross Ref]

14. Dekel E., Alon U.
2005.
Optimality and evolutionary tuning of the expression level of a protein. Nature
436, 588–592 (doi:10.1038/nature03842)10.1038/nature03842 [PubMed] [Cross Ref]

15. Kashtan N., Alon U.
2005.
Spontaneous evolution of modularity and network motifs. Proc. Natl Acad. Sci. USA
102, 13 773–13 778 (doi:10.1073/pnas.0503610102)10.1073/pnas.0503610102 [PubMed] [Cross Ref]

16. Wagner G. P., Pavlicev M., Cheverud J. M.
2007.
The road to modularity. Nat. Rev. Genet.
8, 921–931 (doi:10.1038/nrg2267)10.1038/nrg2267 [PubMed] [Cross Ref]

17. Wade M. J., Winther R. G., Agrawal A. F., Goodnight C. J.
2001.
Alternative definitions of epistasis: dependence and interaction. Trends Ecol. Evol.
16, 498–504 (doi:10.1016/S0169-5347(01)02213-3)10.1016/S0169-5347(01)02213-3 [Cross Ref]

18. Cordell H. J.
2002.
Epistasis: what it means, what it doesn't mean, and statistical methods to detect it in humans. Hum. Mol. Genet.
11, 2463–2468 (doi:10.1093/hmg/11.20.2463)10.1093/hmg/11.20.2463 [PubMed] [Cross Ref]

19. Mani R., St Onge R. P., Hartman J. L., IV, Giaever G., Roth F. P.
2008.
Defining genetic interaction. Proc. Natl Acad. Sci. USA
105, 3461–3466 (doi:10.1073/pnas.0712255105)10.1073/pnas.0712255105 [PubMed] [Cross Ref]

20. Gao H., Granka J. M., Feldman M. W.
2010.
On the classification of epistatic interactions. Genetics
184, 827–837 (doi:10.1534/genetics.109.111120)10.1534/genetics.109.111120 [PubMed] [Cross Ref]

21. Loewe L., Hill W. G.
2010.
The population genetics of mutations: good, bad and indifferent. Phil. Trans. R. Soc. B
365, 1153–1167 (doi:10.1098/rstb.2009.0317)10.1098/rstb.2009.0317 [PMC free article] [PubMed] [Cross Ref]

22. Phillips P. C., Otto S. P., Whitlock M. C.
2000.
Beyond the average: the evolutionary importance of gene interactions and variability of epistatic effects. In Epistasis and the evolutionary process (eds Wolf J. B., Brodie E. D. III, Wade M. J., editors. ), pp. 20–38
New York, NY: Oxford University Press

23. Kouyos R. D., Silander O. K., Bonhoeffer S.
2007.
Epistasis between deleterious mutations and the evolution of recombination. Trends Ecol. Evol.
22, 308–315 (doi:10.1016/j.tree.2007.02.014)10.1016/j.tree.2007.02.014 [PubMed] [Cross Ref]

24. de Visser J. A. G. M., Cooper T. F., Elena S. F.
2011.
The causes of epistasis. Proc. R. Soc. B
278, 3617–3624 (doi:10.1098/rspb.2011.1537)10.1098/rspb.2011.1537 [PMC free article] [PubMed] [Cross Ref]

25. Elena S. F., Lenski R. E.
1997.
Test of synergistic interactions among deleterious mutations in bacteria. Nature
390, 395–398 (doi:10.1038/37108)10.1038/37108 [PubMed] [Cross Ref]

26. de Visser J. A. G. M., Hoekstra R. F., van den Ende H.
1997.
Test of interaction between genetic markers that affect fitness in *Aspergillus niger*. Evolution
51, 1499–1505 (doi:10.2307/2411202)10.2307/2411202 [Cross Ref]

27. Wloch D. M., Borts R. H., Korona R.
2001.
Epistatic interactions of spontaneous mutations in haploid strains of the yeast *Saccharomyces cerevisiae*. J. Evol. Biol.
14, 310–316 (doi:10.1046/j.1420-9101.2001.00267.x)10.1046/j.1420-9101.2001.00267.x [Cross Ref]

28. Szafraniec K., Wloch D. M., Sliwa P., Borts R. H., Korona R.
2003.
Small fitness effects and weak genetic interactions between deleterious mutations in heterozygous loci of the yeast *Saccharomyces cerevisiae*. Genet. Res.
82, 19–31 (doi:10.1017/S001667230300630X)10.1017/S001667230300630X [PubMed] [Cross Ref]

29. Jasnos L., Korona R.
2007.
Epistatic buffering of fitness loss in yeast double deletion strains. Nat. Genet.
39, 550–554 (doi:10.1038/ng1986)10.1038/ng1986 [PubMed] [Cross Ref]

30. Martin G., Elena S. F., Lenormand T.
2007.
Distributions of epistasis in microbes fit predictions from a fitness landscape model. Nat. Genet.
39, 555–560 (doi:10.1038/ng1998)10.1038/ng1998 [PubMed] [Cross Ref]

31. Snitkin E. S., Segrè D.
2011.
Epistatic interaction maps relative to multiple metabolic phenotypes. PLoS Genet.
7, e1001294. (doi:10.1371/journal.pgen.1001294)10.1371/journal.pgen.1001294 [PMC free article] [PubMed] [Cross Ref]

32. He X., Qian W., Wang Z., Li Y., Zhang J.
2010.
Prevalent positive epistasis in *Escherichia coli* and *Saccharomyces cerevisiae* metabolic networks. Nat. Genet.
42, 272–276 (doi:10.1038/ng.524)10.1038/ng.524 [PMC free article] [PubMed] [Cross Ref]

33. Kryazhimskiy S., Draghi J. A., Plotkin J. B.
2011.
In evolution, the sum is less than its parts. Science
332, 1160–1161 (doi:10.1126/science.1208072)10.1126/science.1208072 [PubMed] [Cross Ref]

34. Tănase-Nicola S., ten Wolde P. R.
2008.
Regulatory control and the costs and benefits of biochemical noise. PLoS Comput. Biol.
4, e1000125. (doi:10.1371/journal.pcbi.1000125)10.1371/journal.pcbi.1000125 [PMC free article] [PubMed] [Cross Ref]

35. Wessely F., Bartl M., Guthke R., Li P., Schuster S., Kaleta C.
2011.
Optimal regulatory strategies for metabolic pathways in *Escherichia coli* depending on protein costs. Mol. Syst. Biol.
7 (doi:10.1038/msb.2011.46)10.1038/msb.2011.46 [PMC free article] [PubMed] [Cross Ref]

36. Kondrashov F. A., Kondrashov A. S.
2001.
Multidimensional epistasis and the disadvantage of sex. Proc. Natl Acad. Sci. USA
98, 12 089–12 092 (doi:10.1073/pnas.211214298)10.1073/pnas.211214298 [PubMed] [Cross Ref]

37. Pfeiffer T., Bonhoeffer S.
2002.
Evolutionary consequences of tradeoffs between yield and rate of ATP production. Z. Phys. Chem.
216, 51. (doi:10.1524/zpch.2002.216.1.051)10.1524/zpch.2002.216.1.051 [Cross Ref]

38. Sanjuán R., Moya A., Elena S. F.
2004.
The contribution of epistasis to the architecture of fitness in an RNA virus. Proc. Natl Acad. Sci. USA
101, 15 376–15 379 (doi:10.1073/pnas.0404125101)10.1073/pnas.0404125101 [PubMed] [Cross Ref]

39. Phillips P. C.
2008.
Epistasis: the essential role of gene interactions in the structure and evolution of genetic systems. Nat. Rev. Genet.
9, 855–867 (doi:10.1038/nrg2452)10.1038/nrg2452 [PMC free article] [PubMed] [Cross Ref]

40. Michaut M., Baryshnikova A., Costanzo M., Myers C. L., Andrews B. J., Boone C., Bader G. D.
2011.
Protein complexes are central in the yeast genetic landscape. PLoS Comput. Biol.
7, e1001092. (doi:10.1371/journal.pcbi.1001092)10.1371/journal.pcbi.1001092 [PMC free article] [PubMed] [Cross Ref]

41. Gerrish P. J., Lenski R. E.
1998.
The fate of competing beneficial mutations in an asexual population. Genetica
102–103, 127–144 (doi:10.1023/A:1017067816551)10.1023/A:1017067816551 [PubMed] [Cross Ref]

42. de Visser J. A. G. M., Zeyl C. W., Gerrish P. J., Blanchard J. L., Lenski R. E.
1999.
Diminishing returns from mutation supply rate in asexual populations. Science
283, 404–406 (doi:10.1126/science.283.5400.404)10.1126/science.283.5400.404 [PubMed] [Cross Ref]

43. Elena S. F., Lenski R. E.
2003.
Evolution experiments with microorganisms: the dynamics and genetic bases of adaptation. Nat. Rev. Genet.
4, 457–469 (doi:10.1038/nrg1088)10.1038/nrg1088 [PubMed] [Cross Ref]

44. Chou H.-H., Berthet J., Marx C. J.
2009.
Fast growth increases the selective advantage of a mutation arising recurrently during evolution under metal limitation. PLoS Genet.
5, e1000652. (doi:10.1371/journal.pgen.1000652)10.1371/journal.pgen.1000652 [PMC free article] [PubMed] [Cross Ref]

45. Bonhoeffer S., Chappey C., Parkin N. T., Whitcomb J. M., Petropoulos C. J.
2004.
Evidence for positive epistasis in HIV-1. Science
306, 1547–1550 (doi:10.1126/science.1101786)10.1126/science.1101786 [PubMed] [Cross Ref]

46. Rice S. H.
2000.
The evolution of developmental interactions: epistasis, canalization, and integration. In Epistasis and the evolutionary process (eds Wolf J. B., Brodie E. D. III, Wade M. J., editors. ), pp. 82–98
New York, NY: Oxford University Press

47. Kouyos R. D., Otto S. P., Bonhoeffer S.
2006.
Effect of varying epistasis on the evolution of recombination. Genetics
173, 589–597 (doi:10.1534/genetics.105.053108)10.1534/genetics.105.053108 [PubMed] [Cross Ref]

48. Rowe H. C., Hansen B. G., Halkier B. A., Kliebenstein D. J.
2008.
Biochemical networks and epistasis shape the *Arabidopsis thaliana* metabolome. Plant Cell
20, 1199–1216 (doi:10.1105/tpc.108.058131)10.1105/tpc.108.058131 [PubMed] [Cross Ref]

Articles from Proceedings of the Royal Society B: Biological Sciences are provided here courtesy of **The Royal Society**

PubMed Central Canada is a service of the Canadian Institutes of Health Research (CIHR) working in partnership with the National Research Council's national science library in cooperation with the National Center for Biotechnology Information at the U.S. National Library of Medicine(NCBI/NLM). It includes content provided to the PubMed Central International archive by participating publishers. |