Selection on metabolic pathway function in the presence of mutation-selection-drift balance leads to rate-limiting steps that are not evolutionarily stable
© The Author(s). 2016
Received: 21 March 2016
Accepted: 2 July 2016
Published: 8 July 2016
While commonly assumed in the biochemistry community that the control of metabolic pathways is thought to be critical to cellular function, it is unclear if metabolic pathways generally have evolutionarily stable rate limiting (flux controlling) steps.
A set of evolutionary simulations using a kinetic model of a metabolic pathway was performed under different conditions to evaluate the evolutionary stability of rate limiting steps. Simulations used combinations of selection for steady state flux, selection against the cost of molecular biosynthesis, and selection against the accumulation of high concentrations of a deleterious intermediate. Two mutational regimes were used, one with mutations that on average were neutral to molecular phenotype and a second with a preponderance of activity-destroying mutations. The evolutionary stability of rate limiting steps was low in all simulations with non-neutral mutational processes. Clustering of parameter co-evolution showed divergent inter-molecular evolutionary patterns under different evolutionary regimes.
This study provides a null model for pathway evolution when compensatory processes dominate with potential applications to predicting pathway functional change. This result also suggests a possible mechanism in which studies in statistical genetics that aim to associate a genotype to a phenotype assuming independent action of variants may be mis-specified through a mis-characterization of the link between individual gene function and pathway function. A better understanding of the genotype-phenotype map has potential applications in differentiating between compensatory changes and directional selection on pathways as well as detecting SNPs and fixed differences that might have phenotypic effects.
This article was reviewed by Arne Elofsson, David Ardell, and Shamil Sunyaev.
A long standing goal in molecular evolution and comparative genomics is to understand how genes and their functions evolve. Molecular evolutionary and statistical genetics analyses have commonly treated protein function independently of the functions of other proteins and without consideration of genotype-phenotype maps. However, mutation works at the level of the gene, while selection works at the level of the organism in a population in an ecosystem. One critical component of the interplay between molecular biology and organismal biology is the metabolic pathway that combines the actions of multiple proteins (enzymes) in the generation of energy and molecular building blocks. Systems of differential equations based upon Michaelis-Menten kinetics have become a common modeling tool for describing the function of metabolic pathways .
But how do pathways evolve and how do their constituent members co-evolve? Within a given pathway, various enzymes catalyze reactions at different efficiencies and rates. Rate-limiting steps are the bottlenecks in biochemical pathways and can serve as important points of regulation. Kacser and Burns  established the concept of flux control enzymes in a pathway, with a given set of enzyme rate constants. The distribution of rate-limiting steps varies across different biochemical networks, although current biochemical thought from network control theory is that selection has a strong role in maintaining efficient pathway function and regulation through the most controllable points [3–15]. Specific examples in glycolysis are given in . A correlate of this is an expectation of evolutionary stability of rate limiting steps when there is negative (stabilizing) selection on pathway function (for example, steady state flux that is not selected to change) . This issue, however, has not been seriously addressed in the biochemical literature. Related biochemical expectations suggest that the observed distribution of rate-limiting steps is driven by pathway architecture [4, 12, 14]. An examination of the BioModel Database  showed glycolysis as the only pathway with data from multiple species and it shows no evidence for conserved rate limiting steps controlling steady state flux .
Further, it is unclear from a population genetic perspective that stabilizing selection on steady state flux should give rise to evolutionarily stable rate limiting steps in the presence of mutation-selection balance (see [18, 19] for other studies on the role of mutation-selection balance on molecular systems). The effect of new mutations on fitness has been characterized  and has large fractions of both strongly deleterious (lethal) mutations and slightly deleterious mutations. The frequency and magnitude of slightly deleterious mutations depends upon the mutational space surrounding the protein sequence and is linked to its stability and activity [21, 22]. At the two ends of the spectrum, the globally most active sequence will have only degenerative changes possible, while the globally least active sequence will have only activating changes possible. In between, the proportion of changes that increase or decrease activity will depend upon the current activity. With such a mutational process acting on molecular phenotypes (rather than fitnesses directly), it is expected that enzymes with excess activity will accumulate changes that reduce their activity until they affect pathway flux and are acted upon by selection. At the level of an individual enzyme within a pathway, these dynamics have been described .
Mutation-selection-drift balance and rate limiting steps
Pathway evolution was simulated according to several sets of mutational processes and selective regimes to evaluate the evolutionary dynamics with simultaneous mutational and selective pressures acting on pathway function. After 20,000 generations, all experiments except the mutation-only negative control (where selection was absent) showed that the fitness equilibrium had been reached (Additional file 1: Figures S20-S21). However, when the mutational process was designed to mimic biological mechanisms and adaptive mutation was limiting, there was still co-evolutionary directional movement in some parameters without fitness effects, most notably KM (Additional file 1: Figures S5-S19).
The overall proportion of generations that each reaction spent as rate limiting is shown, pooled across each replication, for each experiment
Selection on flux only
Selection on flux and against protein expression cost
Selection on flux and against a deleterious intermediate
It is important to note that within each experiment, there was variability among the replicates. That is, within each replicate, a particular reaction may have had higher numbers of consecutive generations in which it was rate limiting, and this was not necessarily constant among all replicates. In particular, for the experiment shown in Fig. 2a, panel B, the variability among replicates is shown in Fig. 2b.
Allele segregation within population
In order to assess the consistency of the observations made in this simulation with population genetic expectations given the mutational profile, the amount of segregating variation in the population was characterized. As estimated from Kimura and Crow  as described in the Methods section, assuming neutrality, the expected number of alleles for each parameter in the population is 1.6 alleles per parameter segregating at any time. The observed values that were calculated from the selection on flux only experiment are greater than the expectation when neutral, but of the same order of magnitude. Over 2000 generations, the mean for all of the parameters was found to be 2.75 with a standard deviation of 0.33. The minimum and the maximum of the range were found to be 1.85 and 3.77 correspondingly. For the reaction parameters involved only in the forward direction, the mean for 2000 generations was found to be 2.48 (standard deviation 0.37), with the range minimum of 1.53 and maximum of 3.8. The difference in the parameters reflects the action of selection, particularly with regard to the reverse parameters.
Patterns of co-evolution
Is mutation-selection-drift balance just drift in a small population in disguise?
Haldane’s relationship and mutational processes
Haldane’s relationship describes the relationship between the various kinetic parameters in establishing an equilibrium that is consistent with thermodynamic observations of energy differences between reactants and products. In the simulations shown thus far, the mutational parameters are independently free to vary. Haldane’s relationship constrains the values of the parameters as acted upon by mutation by the thermodynamics of the reaction, although the joint effects of mutations are not currently modelable (see  for some discussion of this in the context of glycolysis as well as ). In the simplest case, Haldane’s relationship reduces 4 parameters to 3° of freedom, although more degrees of freedom are added with additional products and substrates, regulation, cofactors, and more complex equilibria. It is well known that modulation of KM is used biologically to regulate the direction of the reaction . However, the precise scheme that was simulated under had one too many degrees of freedom, so the effect of this was tested.
In the experiment constrained by Haldane’s relationship, the evolutionary stability of the rate-limiting steps has a similar average generation length of each reaction as rate-limiting, which is approximately 200 generations in both cases (Fig. 5a, b). Simulating under Haldane’s relationship generates a noticeably more flat distribution across reactions steps as well as less variance.
Low evolutionary stability of rate-limiting steps caused by mutation-selection balance appears to be an expectation for control in metabolic pathways when pathway flux is under negative (stabilizing selection), even when there is a preference for a particular reaction to be flux controlling (as in the reaction leading to a deleterious intermediate). It is clear that in nature, not all pathways are under selection for a constant flux, but may be temporally regulated. More complex regulatory schemes are expected to result in a more complex landscape and longer times to reach a fitness equilibrium, but mutation-selection-drift balance should still play an important role. Relatedly, processes that constantly shift the fitness equilibrium, such as shifting selection (see for example ) or shifting population sizes might be expected to show interesting evolutionary dynamics. Although the fitness equilibrium is never reached, this would provide a very different mechanism of generating flux controlling steps than control theory suggests.
In this study, only a linear pathway was examined. There is no reason to expect mutation-selection balance to not apply to branched pathways or cycles, although the dynamics of equilibration may in fact be different. This remains an interesting topic for future study.
A further layer of complexity is that this work has proceeded with a fixed network whereas network structure evolves in natural systems. Duplication  and the existence and evolution of promiscuous functions [29, 30] are known to give rise to specific processes of network growth . The dynamics of this type of differential equation system evolution have been studied in a community ecology setting  and the co-evolutionary landscapes that emerge may be different from those with a static structure.
With an understanding of the co-evolution of parameters under negative selection, it will be interesting to observe if this pattern changes when positive directional selection is applied to a pathway flux. This would give a probabilistic basis to examining patterns of co-evolution in a pathway to differentiate between compensatory processes and directional selection. These models could also potentially be used to differentiate between negative and positive directional selection in an Approximate Bayesian Computing framework, where constraint on pathway flux gives rise to lineage-specific patterns of enzyme evolution that can be compared to data from gene family analysis.
Lastly, one debate that has consistently arisen in the molecular evolution community is that of the relative importance of changes in gene expression and changes in coding sequence evolution . Mechanistic frameworks like this with roots in either a Boltzmann Distribution or Michaelis-Menten Kinetics, when coupled to a protein level mutational model (see for example ), have the potential to describe the mutational opportunity to affect phenotypes through changes in either protein concentrations or protein coding sequence function parameters (like KM or kcat although predictions on enzymatic reactions are more complex than binding). Deviations from this mutational opportunity (for example, from additional levels of constraint) would be informative about the molecular nature of both compensatory and adaptive evolution.
Relatedly, the field of statistical genetics has commonly made an assumption that the action of a variant is constant against all genetic backgrounds. In the simulations here, the effect of a variant that reduces enzyme activity will have a flux and fitness effect in some parameter (genetic) backgrounds and not in others. Statistically, this averaging would result in low power to detect causal variants. An understanding of the dynamics associated with processes like mutation-selection balance could be used to generally improve models used for understanding the genotype-phenotype map in various biological systems, including in human genetic disease.
Many studies in comparative genomics study each gene in isolation and thereby miss the equilibrium that mutation, selection, and drift generate, including inter-molecular compensatory changes. Under several population genetic and selective regimes, the dynamics of enzyme co-evolution with ultimate negative selection on pathway flux were characterized, resulting in a general lack of evolutionarily stable rate-limiting steps. From this, expected patterns of enzyme co-evolution with negative selection were generated using a clustering approach. This ultimately provides a null model for pathway evolution under stabilizing selection.
Simulated evolution of metabolic pathways
To evaluate the role of mutation-selection-drift balance in biochemical pathway evolution, a population of cells with a key metabolic pathway was evolved under different selective schemes. The simplified kinetic model designed to capture features of glycolysis  and methylglyoxal metabolism  is shown in Fig. 1. The glycolysis-like aspects of the pathway include the feedback loop (as an approximation to glycolysis regulation) and the synthesis of final metabolite F as analogous to pyruvate in a linear pathway. The methylglyxoxal-like pathway elements include the toxic intermediate (B) as analogous to methylglyoxal (a highly toxic intermediate) and again, the synthesis of the final metabolite (F) is analogous to pyruvate.
This model is expressed in terms of a system of ordinary differential equations where reactions are described by reversible Michaelis-Menten kinetics. Each enzyme has parameters for enzyme concentration [Enzyme] (mmol/l), the catalytic constant (kcat) (mmol/l/s), the Michaelis constant for the substrate (KM) (mmol/l), the reversible catalytic constant (kcatr) (mmol/l/s), and the Michaelis constant for the product (KMr) (mmol/l). The kinetic model has a single inhibitory reaction that is described in the system by the inhibition constant KI (mmol/l). The COPASI  modeling environment is used to solve this system of equations. The steady state solution is used, with a constantly replenishing concentration of A and mass action to utilize F, as described in Additional file 1: Table S1.
This mutational scheme allows for scaling across orders of magnitude in kinetic parameters and generates a distribution of mutational effects with a bias towards slightly degrading change that is dependent upon the activity and expression level of the protein. The mutational scheme is consistent with current thought in molecular evolution, where the range and distribution of mutational effects are influenced by the current state . Most of the mutations are slightly deleterious or neutral, while advantageous mutations are rare, although slightly less so as the activity of the molecule decreases. Intuitively, as a sequence decreases in fitness contribution, the number of sequences with higher potential fitness contribution increases and as it increases in fitness contribution, the number of sequences with a higher potential fitness contribution decreases, as expected by Fisher’s geometric model.
Values in this logistic function control the asymptotic fitness and the gradient of the flux to fitness relationship. As enzymes reach limits of adaptation because of the ability to utilize products, so do pathways, where the end products are also subjected to the rules of binding and catalysis [23, 38, 39]. The asymptote of 650 and slope of 0.07 are arbitrary, but are chosen to reflect the ultimate utilizable flux. Changing them would be expected to alter the distribution of fitness effects (fraction of deleterious changes at equilibrium), but not the overall evolutionary dynamics of the system. A second (negative control) scheme was implemented to examine mutational opportunity and mutational pressure. In this experiment individuals were sampled at random from the population and only the mutational process acted.
Here, [B] is the concentration of the deleterious metabolite and s (9.4 × 10−4) is a scalar chosen to control the flux and the intersection point of the two curves. As indicated, the mean of the mutational distribution is set at 0, and the distribution is parameter-independent.
Each of these simulations was run for 22,000 generations and the point of mutation-selection balance was reached by generation 20,000 under each of these selective schemes (the scheme with no selection did not reach equilibrium because there is no mutation-selection balance without selection). The point of mutation-selection balance was determined by the stability of the fitness of the median individual across generational time as assessed by observation of approximately equal rates of positive and negative changes (Additional file 1: Figures S20–S21). The point of balance was confirmed for the experiment with selection on flux alone by replicate experiments approaching the same point from lower fitnesses that were reached from higher fitnesses (Additional file 1: Figure S1).
Identification of rate limiting steps
The sensitivity of each of the reactions across the last 2000 generations was determined by reducing each reaction rate of the median individual by 10 % while fixing the rest of the reaction rates. The difference in flux between the perturbed and unperturbed systems was used a measure of sensitivity, and the most sensitive step was determined by the reaction for which this value was the largest.
Examination of evolution and coevolution
Examination of parameter evolution and co-evolution was based upon the values in the median individual at each generation for the 2000 generations after equilibrium was reached. Since the reversible and inhibitory reaction constants have minimal impact on the system, they were removed from the analysis. Five replicates of the same experiment were analyzed together and the rate of change of each parameter was calculated for every generation. In order to control for directional change within enzyme concentrations, catalytic, and binding constants, the average amount of change is calculated for each group and removed from each parameter within the group. 10,000 replicates were bootstrapped from this dataset by random re-sampling within each replicate and complete linkage clustering was performed using absolute correlations as a measure of relatedness between the rates of change (Additional file 1: Figures S22–S26) . The largest clusters significant at the 0.05 level are used to identify co-evolving parameters.
Simulations with variable population sizes
representing the fixation probability, where Ne is the population size, c is the ploidy (haploid, c = 1), s is the selective coefficient (f’/f0-1, where f’ is the fitness after mutation and f0 before) and p is the initial frequency of the allele in a population. The initial frequency p was set to ½ rather than 1/N for computational efficiency, giving the property that a neutral mutation has a 50 % chance of fixation, which scales the selective coefficient. The effects of population size played out in rising from a 0.5 frequency to fixation and the introduction of new mutations was independent of population size.
The population scheme was run for 200,000 generations per experimental replicate and the rate-limiting step length was calculated as was previously described. Both population sizes were run for 30 replicates.
Simulations with more thermodynamic realism
Here, Keq is the equilibrium constant driven by the thermodynamics of the reaction. For this experiment, kinetic parameter initial values were set according to Haldane’s relationship (Additional file 1: Table S3). To maintain the ratio, the mutational scheme was modified from that for other experiments described above. Mutations for KMr and KM are drawn from a normal distribution with a mean at −1 %. The mutational effect for kcatis also drawn from a normal distribution with a mean at −1 % and has a modifier that is dependent on the original ratio of kcat and kcatr and the ratio of mutated KM and KMr. Kcatr is calculated from Haldane’s relationship with the mutated kcat, KM, KMr. This experiment was replicated for 30 times. Rate-limiting step lengths were evaluated as was previously described.
Characterizing allele segregation with explicit populations
Here n is the number of alleles for particular parameter, μ is mutation rate, and Ne the effective population size.
Statistical tests and bootstrap confidence intervals
For the simulation experiments exploring the evolutionary stability of the rate limiting step, we performed permutation tests under the null hypothesis of no stability. Stability was measured by the number of consecutive generations that a reaction remained rate limiting, once it became the rate limiting step, and under the null hypothesis, each reaction should have the same average number of consecutive generations. For each of the replicates, the correspondence between each reaction and its average time spent as rate limiting was permuted, and the average absolute deviation of each reaction from the overall mean was calculated. In this manner, a null distribution was generated through 100,000 permutation replicates, and an empirical p-value was found by comparing the average absolute deviation in the actual data as compared to this null distribution.
Confidence intervals for each average number of consecutive generations that a reaction was rate limiting were constructed by first bootstrapping the replicates, and then bootstrapping the consecutive runs within each selected replicate. In this manner, a Monte Carlo sampling distribution for each average was generated, and 95 % confidence intervals were generated by taking the 2.5th and 97.5th percentiles from each bootstrap sampling distribution. Error bars in the corresponding figures reflect these confidence intervals.
To test the question of whether selection has an effect on expression cost, we examine enzyme concentration, kcat and KM against the length of the enzyme. Using the 2000 generations at equilibrium across the five replications and comparing the selection against flux experiment to the selection against flux and protein expression, we ran a linear mixed-effects model with random effects for the replicates, and an interaction term for enzyme length and experiment. The null hypothesis is that the interaction term is equal to 0, or in other words, that the different selection regimes have the same effect on expression cost. Due to computational burden within the mixed-effects model when attempting to account for the correlation structure induced by the Markovian nature of the simulations, only the unique values of each corresponding outcome variable were selected to be fit in the model, and standard linear mixed-effects models were run with a random effect for each replicate. While this loss of information is suboptimal, this should result in conservative inference. Assumptions of homoskedasticity and linearity of the relationship between enzyme length and parameter observations appear to be satisfied. The small effective sample size may be a concern, but should also lead to conservative inference. Statistical significance of the interaction term was assessed via likelihood ratio tests comparing the interaction model to the null model, which did not contain the interaction term.
We thank the reviewers for their reviews of our manuscript. Reviewer 2 included minor comments that have not been included for publication, but have improved the readability of the manuscript.
Reviewers’ report 1: Arne Elofsson, Stockholm University, Sweden
The authors describe a simulations of enzymatic reactions and identifies rate limiting steps.
Reviewer recommendations to author
One problem with this paper is that I am not convinced that the main reason to do this study is correct. The authors claim: “While commonly assumed in the biochemistry community that the control of metabolic pathways is thought to be critical to cellular function.”. However, they do not provide a single reference to that this really is commonly assumed, or if this is really true. Certainly it varies from pathway to pathway. It is certainly very important to have a very close control over glucose levels in the blood (see the problem for people with diabetes), while lactase efficiency can vary by many orders of magnitude without a large impact on fitness. I would interpret the results in a different way than the authors do. I do not agree with the claim “The evolutionary stability of rate limiting steps was low in all simulations with non-neutral mutational processes. ” Instead I think: If you select for positive control or against deleterious intermediates the first step is almost always rate limiting step if not it can be any step.
Author Response: The main objective of the study is the ultimate differentiation between compensatory processes and directional selection in pathways after the characterization of the expected evolutionary dynamics of pathways with negative selection on pathway function. This is currently not well understood. There is indeed an implicit assumption of pathway stasis, from biochemistry textbooks that describe “glycolysis” and other pathways, to model organism studies that seek to transfer functions from one organism to another to GWAS and QTL studies that assume that pathway regulation and sensitivity are constant.
The selective regime we have applied in this study is fairly simple. It is definitely the case that the nature of the evolutionary dynamics will be altered with more complex regulatory schemes. However, these do not provide selective pressures for extra activity and evolutionarily stable control. With selection against the deleterious intermediate, there is a preponderance of cases where this step re-emerges as rate limiting more frequently than others. However, it is not stable in an evolutionary sense as rate limiting, given its short half-life in remaining flux controlling when it emerges as such. Further, the first step is only the rate limiting step when that is the step leading to the deleterious intermediate; it is not generally the first step that is rate limiting when there is a deleterious intermediate. With lactase, it is unclear to us that the excess activity in the enzyme is stable across distantly related species and we would predict otherwise.
Reviewers’ report 2: David Ardell, University of California- Merced, USA
I am not an expert on Metabolic Control Analysis (MCA) or Biochemical Systems Theory. My assessment of the present work is that it is a highly original synthesis of ideas and models that appears to have been well-implemented, that the conclusions drawn are well-founded from the results obtained, and that the results challenge apparently accepted wisdom regarding the evolutionary stability of rate-limiting steps in non-branching metabolic pathways. The review of prior literature and contextualization of results in the introduction and discussion are sufficient, yet further efforts here might benefit some readers, to dispel confusions and to better connect the results to prior work. For example, in his textbook “A First Course in Systems Biology,” E.O. Voit writes “MCA was originally conceived to replace the formerly widespread notion that every pathway has one rate-limiting step, which is a slow reaction that by itself is credited with determining the magnitude of flux through the pathway… In linear pathway sections without branches, the rate-limiting step was traditionally thought to be positioned at the first reaction, where it was possibly inhibited through feed-back exerted by the end product. In MCA, this concept of a rate-limiting step was supplanted with the concept of shared control, which posits that every step in a pathway contributes, to some degree, to the control of the steady-state flux.” In light of this, perhaps the author’s results might profitably be considered as extending MCA’s notion of distributed control in metabolic pathways to distributed selection pressures on their component enzymes coevolving on a rugged landscape. A valuable contribution made by this work, in my opinion, is the example it provides of another relatively simple biological system that under the simplest evolutionary scenarios — namely stabilizing selection on its output — explores a large neutral network of solutions (in this case, of kinetic parameters). In summary, I believe the present work is an important contribution to its field.
Reviewer recommendations to author
There are many points where the presentation and narrative could be improved to increase impact, especially for those unfamiliar with some among the many different disciplines and subjects touched on by this work.
Author Response: We thank the reviewer for his summary. We have tried to improve the readability of the manuscript and to better introduce disparate concepts.
Reviewers’ report 3: Shamil Sunyaev, Harvard Medical School, USA
This manuscript challenges the idea of evolutionary stability of rate limiting steps in linear pathways. Extensive computer simulations demonstrate that rate limiting steps exist for only short evolutionary times. This is an interesting result.
Reviewer recommendations to authors
I find the section “Allele Segregation within Population” confusing. I suggest that the authors would clarify this section. Also, how stable are the results with respect to effect sizes and directions of incoming mutations?
Author Response: We thank the reviewer for his summary. We have added a new introduction to the allele segregation section to improve clarity. The trajectories of effect sizes and directions of incoming mutations that are sampled for each parameter are shown in the Supplementary Figures. They derive from the mutational process described in the Methods section.
[E], [Enzyme], the concentration of the enzyme; kcat, catalytic constant; KI, inhibition constant; KM, Michaelis constant; mRNA, messenger ribonucleic acid; Ne, effective population size; μ, mutation rate.
We thank Claudia Weber for a careful reading of this manuscript. This work was funded by NSF grant DBI-0743374.
The research was conceived and supervised by DAL. Simulation studies were run by AO. AIT and PBC performed statistical analysis on the data generated from simulations. All authors were involved in writing the manuscript. All authors read and approved the manuscript.
The authors have no competing interests to declare.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Ingalls B. Mathematical modeling in systems biology: An Introduction. Cambridge: The MIT Press; 2013.Google Scholar
- Kacser H, Burns JA. The control of flux. Symp Soc Exp Biol. 1973;27:65–104.PubMedGoogle Scholar
- Campbell S, Khosravi-Far R, Rossman K, Clark G, Der C. Increasing complexity of Ras signaling. Oncogene. 1998;17:1395–413.View ArticlePubMedGoogle Scholar
- Rausher M, Miller R, Tiffin P. Patterns of evolutionary rate variation among genes of the anthocyanin biosynthetic pathway. Mol Biol Evol. 1999;16:266–74.View ArticlePubMedGoogle Scholar
- Berg JM, Tymoczko JL, Stryer L. Biochemistry. 5th ed. New York: W H Freeman; 2002. Section 16.2.Google Scholar
- Olsen K, Womak A, Garret A, Suddith J, Purugganan M. Contrasting evolutionary forces in the Arabidopsis thaliana floral developmental pathway. Genetics. 2002;160:1641–50.PubMedPubMed CentralGoogle Scholar
- Riley R, Jin W, Gibson G. Contrasting selection pressures on components of the Ras-mediated signal transduction pathway in Drosophila. Mol Ecol. 2003;12:1315–23.View ArticlePubMedGoogle Scholar
- Cork J, Purugganan M. The evolution of molecular genetic pathways and networks. Bioessays. 2004;26:479–84.View ArticlePubMedGoogle Scholar
- Wright K, Rausher M. The evolution of control and distribution of adaptive mutations in a metabolic pathway. Genetics. 2009;184:483–502.View ArticlePubMedGoogle Scholar
- Alvarez-Ponce D, Aguade M, Rozas J. Comparative genomics of the vertebrate insulin/TOR signal transduction pathway: a network-level analysis of selective pressures. Genome Biol Evol. 2010;3:87–101.View ArticlePubMedPubMed CentralGoogle Scholar
- O’Connell M. Selection and the cell cycle: positive Darwinian selection in a well-known DNA damage response pathway. J Mol Evol. 2010;71:444–57.View ArticlePubMedGoogle Scholar
- Alvarez-Ponce D. The relationship between the hierarchical position of proteins in the human signal transduction network and their rate of evolution. BMC Evol Biol. 2012;12:192.View ArticlePubMedPubMed CentralGoogle Scholar
- Olson-Manning C, Lee C, Rausher M, Mitchell-Olds T. Evolution of flux control in the glucosinolate pathway in Arabidopsis thaliana. Mol Biol Evol. 2012;30:14–23.View ArticlePubMedPubMed CentralGoogle Scholar
- Rausher M. The evolution of genes in branched metabolic pathways. Evolution. 2012;67:34–48.View ArticlePubMedGoogle Scholar
- Hermansen RA, Mannakee BK, Knecht W, Liberles DA, Gutenkunst RN. Characterizing selective pressures on the pathway for de novo biosynthesis of pyrimidines in yeast. BMC Evol Biol. 2015;15:232. doi:10.1186/s12862-015-0515-x.View ArticlePubMedPubMed CentralGoogle Scholar
- Orlenko A, Hermansen RA, Liberles DA. Flux control in glycolysis varies across the tree of life. J Mol Evol. 2016;82:146–61.View ArticlePubMedGoogle Scholar
- Li C, Donizelli M, Rodriguez N, Dharuri H, Endler L, Chelliah V, Li L, He E, Henry A, Stefan MI, Snoep JL, Hucka M, Le Novère N, Laibe C. BioModels database: an enhanced, curated and annotated resource for published quantitative kinetic models. BMC Syst Biol. 2010;4:92.View ArticlePubMedPubMed CentralGoogle Scholar
- Taverna D, Goldstein R. Why are proteins marginally stable? Proteins. 2001;46:105–9.View ArticleGoogle Scholar
- Lynch M, Hagner K. Evolutionary meandering of intermolecular interactions along the drift barrier. Proc Natl Acad Sci U S A. 2014;112:30–8.View ArticleGoogle Scholar
- Tamuri AU, dos Reis M, Goldstein RA. Estimating the distribution of selection coefficients from phylogenetic data using sitewise mutation-selection models. Genetics. 2012;190:1101–15.View ArticlePubMedPubMed CentralGoogle Scholar
- Wylie C, Shakhnovich E. A biophysical protein folding model accounts for most mutational fitness effects in viruses. Proc Natl Acad Sci U S A. 2011;108:9916–21.View ArticlePubMedPubMed CentralGoogle Scholar
- Dasmeh P, Serohijos A, Kepp K, Shakhnovich E. The Influence of Selection for Protein Stability on dN/dS Estimations. Genome Biol Evol. 2014;6:2956–67.View ArticlePubMedPubMed CentralGoogle Scholar
- Hartl DL, Dykhuizen DE, Dean AM. Limits of adaptation: the evolution of selective neutrality. Genetics. 1985;111:655–74.PubMedPubMed CentralGoogle Scholar
- Wagner A. Energy constraints on the evolution of gene expression. Mol Biol Evol. 2005;22:1365–74.View ArticlePubMedGoogle Scholar
- Kimura M, Crow JF. The number of alleles that can be maintained in a finite population. Genetics. 1964;49:725–38.PubMedPubMed CentralGoogle Scholar
- Uhr ML. The influence of an enzyme on the direction of a reaction. Biochem Educ. 1979;7:15–6.View ArticleGoogle Scholar
- Soyer O, Pfeiffer T. Evolution under fluctuating environments explains observed robustness in metabolic networks. PLoS Comput Biol. 2010;6:e1000907.View ArticlePubMedPubMed CentralGoogle Scholar
- Soyer O, Creevey C. Duplicate retention in signalling proteins and constraints from network dynamics. J Evol Biol. 2010;23:2410–21.View ArticlePubMedGoogle Scholar
- Kim J, Kershner J, Novikov Y, Shoemaker R, Copley S. Three serendipitous pathways in E. coli can bypass a block in pyridoxal-5’-phosphate synthesis. Mol Syst Biol. 2010;6:436.View ArticlePubMedPubMed CentralGoogle Scholar
- Kim J, Copley S. Inhibitory cross-talk upon introduction of a new metabolic pathway into an existing metabolic network. Proc Natl Acad Sci U S A. 2012;109:2856–64.View ArticleGoogle Scholar
- Light S, Kraulis P. Network analysis of metabolic enzymes evolution in Escherichia coli. BMC Bioinformatics. 2004;5:15.View ArticlePubMedPubMed CentralGoogle Scholar
- Shoresh N, Hegreness M, Kishony R. Evolution exacerbates the paradox of the plankton. Proc Natl Acad Sci U S A. 2008;105:12365–9.View ArticlePubMedPubMed CentralGoogle Scholar
- Hoekstra H, Coyne J. The locus of evolution: evo devo and the genetics of adaptation. Evolution. 2007;61:995–1016.View ArticlePubMedGoogle Scholar
- Grahnen JA, Nandakumar P, Kubelka J, Liberles DA. Biophysical and structural considerations for protein sequence evolution. BMC Evol Biol. 2011;11:361.View ArticlePubMedPubMed CentralGoogle Scholar
- Teusink B, Passarge J, Reijenga CA, Esgalhado E, van der Weijden CC, Schepper M, Walsh MC, Bakker BM, van Dam K, Westerhoff HV, Snoep JL. Can yeast glycolysis be understood in terms of in vitro kinetics of the constituent enzymes? Testing biochemistry. Eur J Biochem. 2000;267:5313–29.View ArticlePubMedGoogle Scholar
- Weber J, Kayser A, Rinas U. Metabolic flux analysis of Escherichia coli in glucose-limited continuous culture. II. Dynamic response to famine and feast, activation of the methylglyoxal pathway and oscillatory behavior. Microbiology. 2005;151:707–16.View ArticlePubMedGoogle Scholar
- Hoops S, Sahle S, Gauges R, Lee C, Pahle J, Simus N, Singhal M, Xu L, Mendes P, Kummer U. COPASI--a COmplex PAthway SImulator. Bioinformatics. 2006;22:3067–74.View ArticlePubMedGoogle Scholar
- Bershtein S, Mu W, Serohijos AWR, Zhou J, Shakhnovich EI. Protein quality control acts on folding intermediates to shape the effects of mutations on organismal fitness. Mol Cell. 2013;49:133–44.View ArticlePubMedGoogle Scholar
- Jiang L, Mishra P, Hietpas RT, Zeldovich KB, Bolon DNA. Latent effects of Hsp90 mutants revealed at reduced expression levels. PLoS Genet. 2013;9:e1003600.View ArticlePubMedPubMed CentralGoogle Scholar
- Suzuki R, Shimodaira H. Pvclust: an R package for assessing the uncertainty in hierarchical clustering. Bioinformatics. 2006;22:1540–1542.29.View ArticlePubMedGoogle Scholar