Limits...
Epistatic interaction maps relative to multiple metabolic phenotypes.

Snitkin ES, Segrè D - PLoS Genet. (2011)

Bottom Line: Finally, we found that genes involved in many interactions across multiple phenotypes are more highly expressed, evolve slower, and tend to be associated with diseases, indicating that the importance of genes is hidden in their total phenotypic impact.Our predictions indicate a pervasiveness of nonlinear effects in how genetic perturbations affect multiple metabolic phenotypes.The approaches and results reported could influence future efforts in understanding metabolic diseases and the role of biochemical regulation in the cell.

View Article: PubMed Central - PubMed

Affiliation: Program in Bioinformatics, Boston University, Boston, Massachusetts, United States of America.

ABSTRACT
An epistatic interaction between two genes occurs when the phenotypic impact of one gene depends on another gene, often exposing a functional association between them. Due to experimental scalability and to evolutionary significance, abundant work has been focused on studying how epistasis affects cellular growth rate, most notably in yeast. However, epistasis likely influences many different phenotypes, affecting our capacity to understand cellular functions, biochemical networks adaptation, and genetic diseases. Despite its broad significance, the extent and nature of epistasis relative to different phenotypes remain fundamentally unexplored. Here we use genome-scale metabolic network modeling to investigate the extent and properties of epistatic interactions relative to multiple phenotypes. Specifically, using an experimentally refined stoichiometric model for Saccharomyces cerevisiae, we computed a three-dimensional matrix of epistatic interactions between any two enzyme gene deletions, with respect to all metabolic flux phenotypes. We found that the total number of epistatic interactions between enzymes increases rapidly as phenotypes are added, plateauing at approximately 80 phenotypes, to an overall connectivity that is roughly 8-fold larger than the one observed relative to growth alone. Looking at interactions across all phenotypes, we found that gene pairs interact incoherently relative to different phenotypes, i.e. antagonistically relative to some phenotypes and synergistically relative to others. Specific deletion-deletion-phenotype triplets can be explained metabolically, suggesting a highly informative role of multi-phenotype epistasis in mapping cellular functions. Finally, we found that genes involved in many interactions across multiple phenotypes are more highly expressed, evolve slower, and tend to be associated with diseases, indicating that the importance of genes is hidden in their total phenotypic impact. Our predictions indicate a pervasiveness of nonlinear effects in how genetic perturbations affect multiple metabolic phenotypes. The approaches and results reported could influence future efforts in understanding metabolic diseases and the role of biochemical regulation in the cell.

Show MeSH

Related in: MedlinePlus

Schematic depiction of the main steps we used to generate predictions of metabolic fluxes for yeast single and double deletion mutants.(A) As in any flux balance model, we implement a steady state approximation, yielding for each metabolite in the network a linear constraint on reaction rates (fluxes). (B) We then generate the best possible flux state for the wild type, using flux and growth rate data from the literature [30] as additional constraints, and (C) minimizing the sum of the absolute values of all fluxes. This last step prevents the generation of arbitrarily large loops of fluxes associated with alternative optima. (D) Lastly, to generate flux predictions for the deletion mutants, we impose that the fluxes associated with the deleted genes be set to zero, and identify the flux state for the mutant that is as close as possible to the wild type state, identified in step C. Note that this approach does not employ growth rate maximization, as often done in flux balance analysis. Instead, using the concept of minimization of metabolic adjustment [29], it searches for mutant fluxes that undergo minimal rerouting relative to the (experimentally tuned) wild type flux solution. This approach was proven to be the most accurate way of predicting fluxes in yeast knockout strains.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC3037399&req=5

pgen-1001294-g002: Schematic depiction of the main steps we used to generate predictions of metabolic fluxes for yeast single and double deletion mutants.(A) As in any flux balance model, we implement a steady state approximation, yielding for each metabolite in the network a linear constraint on reaction rates (fluxes). (B) We then generate the best possible flux state for the wild type, using flux and growth rate data from the literature [30] as additional constraints, and (C) minimizing the sum of the absolute values of all fluxes. This last step prevents the generation of arbitrarily large loops of fluxes associated with alternative optima. (D) Lastly, to generate flux predictions for the deletion mutants, we impose that the fluxes associated with the deleted genes be set to zero, and identify the flux state for the mutant that is as close as possible to the wild type state, identified in step C. Note that this approach does not employ growth rate maximization, as often done in flux balance analysis. Instead, using the concept of minimization of metabolic adjustment [29], it searches for mutant fluxes that undergo minimal rerouting relative to the (experimentally tuned) wild type flux solution. This approach was proven to be the most accurate way of predicting fluxes in yeast knockout strains.

Mentions: The first issue is the reliability of flux predictions for deletion mutants. The availability of experimentally determined growth phenotypes for all gene deletion mutants in S. cerevisiae has allowed for extensive evaluation of the yeast model's capacity to predict mutant growth. These previous studies [25], [26], including a comparison of model predictions against experimental growth measurements for 465 gene deletion mutants under 16 metabolically diverse conditions [26], have demonstrated that the yeast model can predict deletion mutant viability with high accuracy. Furthermore, observed discordances between model predictions and experimentally determined mutant growth phenotypes have been used in refinements of the existing yeast model, further bolstering the ability of the model to accurately mimic the effect of different gene deletions [25], [27]. In addition to effectively predicting single mutant growth, flux balance models have also been shown to predict viabilities of double deletion mutants with high accuracy [28]. However, while model predictions of mutant growth have been evaluated extensively, comparisons between measured and predicted fluxes through the underlying metabolic reactions in different mutants are less readily available [29]. To address this need we recently evaluated the ability of the yeast model to predict the fluxes through central carbon metabolism in single gene deletion mutants by comparing model predictions to a previously released compendium of experimentally measured mutant fluxes [30]. An assessment of different approaches for mutant flux prediction revealed that an experimentally driven variant of the minimization of metabolic adjustment [31] gives the best correlation with measured fluxes (Spearman rank correlation greater than 0.90, Figure S1), and hence chose it for our calculations (See Materials and Methods). In essence, this method implements the hypothesis that the metabolic response to genetic perturbation will be a minimal rerouting of flux around the insult. A conceptual illustration of the methodology for predicting mutant fluxes is shown in Figure 2, with a detailed quantitative description provided in the Materials and Methods and Text S1.


Epistatic interaction maps relative to multiple metabolic phenotypes.

Snitkin ES, Segrè D - PLoS Genet. (2011)

Schematic depiction of the main steps we used to generate predictions of metabolic fluxes for yeast single and double deletion mutants.(A) As in any flux balance model, we implement a steady state approximation, yielding for each metabolite in the network a linear constraint on reaction rates (fluxes). (B) We then generate the best possible flux state for the wild type, using flux and growth rate data from the literature [30] as additional constraints, and (C) minimizing the sum of the absolute values of all fluxes. This last step prevents the generation of arbitrarily large loops of fluxes associated with alternative optima. (D) Lastly, to generate flux predictions for the deletion mutants, we impose that the fluxes associated with the deleted genes be set to zero, and identify the flux state for the mutant that is as close as possible to the wild type state, identified in step C. Note that this approach does not employ growth rate maximization, as often done in flux balance analysis. Instead, using the concept of minimization of metabolic adjustment [29], it searches for mutant fluxes that undergo minimal rerouting relative to the (experimentally tuned) wild type flux solution. This approach was proven to be the most accurate way of predicting fluxes in yeast knockout strains.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC3037399&req=5

pgen-1001294-g002: Schematic depiction of the main steps we used to generate predictions of metabolic fluxes for yeast single and double deletion mutants.(A) As in any flux balance model, we implement a steady state approximation, yielding for each metabolite in the network a linear constraint on reaction rates (fluxes). (B) We then generate the best possible flux state for the wild type, using flux and growth rate data from the literature [30] as additional constraints, and (C) minimizing the sum of the absolute values of all fluxes. This last step prevents the generation of arbitrarily large loops of fluxes associated with alternative optima. (D) Lastly, to generate flux predictions for the deletion mutants, we impose that the fluxes associated with the deleted genes be set to zero, and identify the flux state for the mutant that is as close as possible to the wild type state, identified in step C. Note that this approach does not employ growth rate maximization, as often done in flux balance analysis. Instead, using the concept of minimization of metabolic adjustment [29], it searches for mutant fluxes that undergo minimal rerouting relative to the (experimentally tuned) wild type flux solution. This approach was proven to be the most accurate way of predicting fluxes in yeast knockout strains.
Mentions: The first issue is the reliability of flux predictions for deletion mutants. The availability of experimentally determined growth phenotypes for all gene deletion mutants in S. cerevisiae has allowed for extensive evaluation of the yeast model's capacity to predict mutant growth. These previous studies [25], [26], including a comparison of model predictions against experimental growth measurements for 465 gene deletion mutants under 16 metabolically diverse conditions [26], have demonstrated that the yeast model can predict deletion mutant viability with high accuracy. Furthermore, observed discordances between model predictions and experimentally determined mutant growth phenotypes have been used in refinements of the existing yeast model, further bolstering the ability of the model to accurately mimic the effect of different gene deletions [25], [27]. In addition to effectively predicting single mutant growth, flux balance models have also been shown to predict viabilities of double deletion mutants with high accuracy [28]. However, while model predictions of mutant growth have been evaluated extensively, comparisons between measured and predicted fluxes through the underlying metabolic reactions in different mutants are less readily available [29]. To address this need we recently evaluated the ability of the yeast model to predict the fluxes through central carbon metabolism in single gene deletion mutants by comparing model predictions to a previously released compendium of experimentally measured mutant fluxes [30]. An assessment of different approaches for mutant flux prediction revealed that an experimentally driven variant of the minimization of metabolic adjustment [31] gives the best correlation with measured fluxes (Spearman rank correlation greater than 0.90, Figure S1), and hence chose it for our calculations (See Materials and Methods). In essence, this method implements the hypothesis that the metabolic response to genetic perturbation will be a minimal rerouting of flux around the insult. A conceptual illustration of the methodology for predicting mutant fluxes is shown in Figure 2, with a detailed quantitative description provided in the Materials and Methods and Text S1.

Bottom Line: Finally, we found that genes involved in many interactions across multiple phenotypes are more highly expressed, evolve slower, and tend to be associated with diseases, indicating that the importance of genes is hidden in their total phenotypic impact.Our predictions indicate a pervasiveness of nonlinear effects in how genetic perturbations affect multiple metabolic phenotypes.The approaches and results reported could influence future efforts in understanding metabolic diseases and the role of biochemical regulation in the cell.

View Article: PubMed Central - PubMed

Affiliation: Program in Bioinformatics, Boston University, Boston, Massachusetts, United States of America.

ABSTRACT
An epistatic interaction between two genes occurs when the phenotypic impact of one gene depends on another gene, often exposing a functional association between them. Due to experimental scalability and to evolutionary significance, abundant work has been focused on studying how epistasis affects cellular growth rate, most notably in yeast. However, epistasis likely influences many different phenotypes, affecting our capacity to understand cellular functions, biochemical networks adaptation, and genetic diseases. Despite its broad significance, the extent and nature of epistasis relative to different phenotypes remain fundamentally unexplored. Here we use genome-scale metabolic network modeling to investigate the extent and properties of epistatic interactions relative to multiple phenotypes. Specifically, using an experimentally refined stoichiometric model for Saccharomyces cerevisiae, we computed a three-dimensional matrix of epistatic interactions between any two enzyme gene deletions, with respect to all metabolic flux phenotypes. We found that the total number of epistatic interactions between enzymes increases rapidly as phenotypes are added, plateauing at approximately 80 phenotypes, to an overall connectivity that is roughly 8-fold larger than the one observed relative to growth alone. Looking at interactions across all phenotypes, we found that gene pairs interact incoherently relative to different phenotypes, i.e. antagonistically relative to some phenotypes and synergistically relative to others. Specific deletion-deletion-phenotype triplets can be explained metabolically, suggesting a highly informative role of multi-phenotype epistasis in mapping cellular functions. Finally, we found that genes involved in many interactions across multiple phenotypes are more highly expressed, evolve slower, and tend to be associated with diseases, indicating that the importance of genes is hidden in their total phenotypic impact. Our predictions indicate a pervasiveness of nonlinear effects in how genetic perturbations affect multiple metabolic phenotypes. The approaches and results reported could influence future efforts in understanding metabolic diseases and the role of biochemical regulation in the cell.

Show MeSH
Related in: MedlinePlus