Limits...
Iterative reconstruction of a global metabolic model of Acinetobacter baylyi ADP1 using high-throughput growth phenotype and gene essentiality data.

Durot M, Le Fèvre F, de Berardinis V, Kreimeyer A, Vallenet D, Combe C, Smidtas S, Salanoubat M, Weissenbach J, Schachter V - BMC Syst Biol (2008)

Bottom Line: The predictions of the final version of the model, which included three rounds of refinements, are consistent with the experimental results for (1) 91% of the wild-type growth phenotypes, (2) 94% of the gene essentiality results, and (3) 94% of the mutant growth phenotypes.To facilitate the exploitation of the metabolic model, we provide a web interface allowing online predictions and visualization of results on metabolic maps.The iterative reconstruction procedure led to significant model improvements, showing that genome-wide mutant phenotypes on several media can significantly facilitate the transition from genome annotation to a high-quality model.

View Article: PubMed Central - HTML - PubMed

Affiliation: Genoscope (Commissariat à l'Energie Atomique) and UMR 8030 CNRS-Genoscope-Université d'Evry, 2 rue Gaston Crémieux, CP5706, 91057 Evry, Cedex, France. mdurot@genoscope.cns.fr

ABSTRACT

Background: Genome-scale metabolic models are powerful tools to study global properties of metabolic networks. They provide a way to integrate various types of biological information in a single framework, providing a structured representation of available knowledge on the metabolism of the respective species.

Results: We reconstructed a constraint-based metabolic model of Acinetobacter baylyi ADP1, a soil bacterium of interest for environmental and biotechnological applications with large-spectrum biodegradation capabilities. Following initial reconstruction from genome annotation and the literature, we iteratively refined the model by comparing its predictions with the results of large-scale experiments: (1) high-throughput growth phenotypes of the wild-type strain on 190 distinct environments, (2) genome-wide gene essentialities from a knockout mutant library, and (3) large-scale growth phenotypes of all mutant strains on 8 minimal media. Out of 1412 predictions, 1262 were initially consistent with our experimental observations. Inconsistencies were systematically examined, leading in 65 cases to model corrections. The predictions of the final version of the model, which included three rounds of refinements, are consistent with the experimental results for (1) 91% of the wild-type growth phenotypes, (2) 94% of the gene essentiality results, and (3) 94% of the mutant growth phenotypes. To facilitate the exploitation of the metabolic model, we provide a web interface allowing online predictions and visualization of results on metabolic maps.

Conclusion: The iterative reconstruction procedure led to significant model improvements, showing that genome-wide mutant phenotypes on several media can significantly facilitate the transition from genome annotation to a high-quality model.

Show MeSH

Related in: MedlinePlus

Model correction examples. Examples of model corrections performed between iAbaylyiv2 (left) and iAbaylyiv3 (right) models. Metabolites are depicted by blue circles and triangles, triangles indicating essential biomass precursors. Reactions are represented by arrows colored in red if they are predicted essential and in green if they are predicted dispensable. Gene names are indicated next to reaction arrows; they are written in red if they are experimentally essential and in green if they are dispensable. Genes with inconsistent predictions are written in bold italic. Dashed boxes indicate components that have been modified. Further evidence for model corrections are shown in main text and Additional file 3. (A) First steps of histidine biosynthesis. Unpredicted essentiality of ACIAD2907 encoding for ribose-phosphate diphosphokinase activity was corrected by removing the alternate gene ACIAD0964 from the reaction GPR. Unpredicted essentialities of ACIAD0661 and ACIAD1257, catalyzing the ATP phosphoribosyltransferase reaction, were corrected by assigning them as complex subunits instead of isozymes in the reaction GPR. (B) Isoprenoids biosynthesis. Unpredicted dispensability of ACIAD2968, catalyzing farnesyl-diphosphate and geranyl-diphosphate synthases activities, was corrected by adding ACIAD1374 (undecaprenyl-diphosphate synthase) and ACIAD2940 (octaprenyl-diphosphate synthase) as isozymes. Unpredicted essentiality of ACIAD1374 was resolved by adding undecaprenyl-PP to the set of essential biomass precursors. (C) Synthesis of charged glutamine-tRNA(gln) and asparagine-tRNA(asn). Unpredicted essentiality of ACIAD1920, encoding for glutaminyl-tRNA synthetase activity, was corrected by removing from the model the alternate pathway using aspartyl/glutamyl-tRNA amidotransferase enzyme (ACIAD0822-0824). (D) Biosynthesis of polysaccharides. Unpredicted dispensabilities of all genes involved in GDP-mannose, UDP-glucose, and dTDP-rhamnose synthesis were corrected by removing these three metabolites from the list of essential biomass precursors.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC2606687&req=5

Figure 8: Model correction examples. Examples of model corrections performed between iAbaylyiv2 (left) and iAbaylyiv3 (right) models. Metabolites are depicted by blue circles and triangles, triangles indicating essential biomass precursors. Reactions are represented by arrows colored in red if they are predicted essential and in green if they are predicted dispensable. Gene names are indicated next to reaction arrows; they are written in red if they are experimentally essential and in green if they are dispensable. Genes with inconsistent predictions are written in bold italic. Dashed boxes indicate components that have been modified. Further evidence for model corrections are shown in main text and Additional file 3. (A) First steps of histidine biosynthesis. Unpredicted essentiality of ACIAD2907 encoding for ribose-phosphate diphosphokinase activity was corrected by removing the alternate gene ACIAD0964 from the reaction GPR. Unpredicted essentialities of ACIAD0661 and ACIAD1257, catalyzing the ATP phosphoribosyltransferase reaction, were corrected by assigning them as complex subunits instead of isozymes in the reaction GPR. (B) Isoprenoids biosynthesis. Unpredicted dispensability of ACIAD2968, catalyzing farnesyl-diphosphate and geranyl-diphosphate synthases activities, was corrected by adding ACIAD1374 (undecaprenyl-diphosphate synthase) and ACIAD2940 (octaprenyl-diphosphate synthase) as isozymes. Unpredicted essentiality of ACIAD1374 was resolved by adding undecaprenyl-PP to the set of essential biomass precursors. (C) Synthesis of charged glutamine-tRNA(gln) and asparagine-tRNA(asn). Unpredicted essentiality of ACIAD1920, encoding for glutaminyl-tRNA synthetase activity, was corrected by removing from the model the alternate pathway using aspartyl/glutamyl-tRNA amidotransferase enzyme (ACIAD0822-0824). (D) Biosynthesis of polysaccharides. Unpredicted dispensabilities of all genes involved in GDP-mannose, UDP-glucose, and dTDP-rhamnose synthesis were corrected by removing these three metabolites from the list of essential biomass precursors.

Mentions: A majority of the model improvements (34/56) were applied to the GPR component, with a clear bias towards false dispensable inconsistencies: 26 GPR corrections pertained to experimentally essential genes against only 8 to experimentally dispensable genes (see Table 3). This large set of false dispensable predictions includes two main inconsistency types. In 22 cases, isofunctional genes with annotations of medium confidence were in fact unable to replace the activity of their deleted isozymes. For instance, ACIAD0964 and ACIAD2907 (prs) were identified in the initial reconstruction as isozymes for the catalysis of the ribose-phosphate diphosphokinase activity, which is required for the biosynthesis of 5-phosphoribosylpyrophosphate (PRPP) (see Figure 8A). The association of both genes to the activity relied on homologies with previously annotated genes in other organisms. The expected and predicted dispensability of ACIAD2907 was yet contradicted by its experimental essentiality. Looking further into the annotation evidence, ACIAD0964 function was supported by only limited homologies to previously known genes (second best hit after ACIAD2907 with E. coli gene prsA, with 25% identity). Conversely, ACIAD2907 function was supported by a stronger homology with E. coli gene prsA (68% identity) whose ribose-phosphate diphosphokinase has been experimentally confirmed [32]. The combination of the observed gene essentialities with the limited homology supporting the annotation of ACIAD0964 led us to correct the model by removing ACIAD0964 from ribose-phosphate diphosphokinase GPR. On the other hand, the functions of some isozymes with medium confidence level were corroborated by the gene essentialities. For instance, two isozymes were indirectly confirmed to have a dihydroxy-acid dehydratase activity, which is essential for the synthesis of valine, leucine and isoleucine. Two duplicate genes were associated with this activity: ACIAD1266 (ilvD) and ACIAD3636. While the annotation of ACIAD1266 is supported by a strong homology with E. coli gene ilvD (74% identity) whose activity has been experimentally shown [33], ACIAD3636's function was supported only by weaker homologies with the reference genes (37% identity with E. coli gene ilvD). Gene knock-outs revealed that both genes were dispensable while the essentiality of other genes in the pathway strongly suggested that the dihydroxy-acid dehydratase activity was required. This result strongly suggests that both genes could back up each other and therefore indirectly corroborates the functional assignment to ACIAD3636.


Iterative reconstruction of a global metabolic model of Acinetobacter baylyi ADP1 using high-throughput growth phenotype and gene essentiality data.

Durot M, Le Fèvre F, de Berardinis V, Kreimeyer A, Vallenet D, Combe C, Smidtas S, Salanoubat M, Weissenbach J, Schachter V - BMC Syst Biol (2008)

Model correction examples. Examples of model corrections performed between iAbaylyiv2 (left) and iAbaylyiv3 (right) models. Metabolites are depicted by blue circles and triangles, triangles indicating essential biomass precursors. Reactions are represented by arrows colored in red if they are predicted essential and in green if they are predicted dispensable. Gene names are indicated next to reaction arrows; they are written in red if they are experimentally essential and in green if they are dispensable. Genes with inconsistent predictions are written in bold italic. Dashed boxes indicate components that have been modified. Further evidence for model corrections are shown in main text and Additional file 3. (A) First steps of histidine biosynthesis. Unpredicted essentiality of ACIAD2907 encoding for ribose-phosphate diphosphokinase activity was corrected by removing the alternate gene ACIAD0964 from the reaction GPR. Unpredicted essentialities of ACIAD0661 and ACIAD1257, catalyzing the ATP phosphoribosyltransferase reaction, were corrected by assigning them as complex subunits instead of isozymes in the reaction GPR. (B) Isoprenoids biosynthesis. Unpredicted dispensability of ACIAD2968, catalyzing farnesyl-diphosphate and geranyl-diphosphate synthases activities, was corrected by adding ACIAD1374 (undecaprenyl-diphosphate synthase) and ACIAD2940 (octaprenyl-diphosphate synthase) as isozymes. Unpredicted essentiality of ACIAD1374 was resolved by adding undecaprenyl-PP to the set of essential biomass precursors. (C) Synthesis of charged glutamine-tRNA(gln) and asparagine-tRNA(asn). Unpredicted essentiality of ACIAD1920, encoding for glutaminyl-tRNA synthetase activity, was corrected by removing from the model the alternate pathway using aspartyl/glutamyl-tRNA amidotransferase enzyme (ACIAD0822-0824). (D) Biosynthesis of polysaccharides. Unpredicted dispensabilities of all genes involved in GDP-mannose, UDP-glucose, and dTDP-rhamnose synthesis were corrected by removing these three metabolites from the list of essential biomass precursors.
© Copyright Policy - open-access
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC2606687&req=5

Figure 8: Model correction examples. Examples of model corrections performed between iAbaylyiv2 (left) and iAbaylyiv3 (right) models. Metabolites are depicted by blue circles and triangles, triangles indicating essential biomass precursors. Reactions are represented by arrows colored in red if they are predicted essential and in green if they are predicted dispensable. Gene names are indicated next to reaction arrows; they are written in red if they are experimentally essential and in green if they are dispensable. Genes with inconsistent predictions are written in bold italic. Dashed boxes indicate components that have been modified. Further evidence for model corrections are shown in main text and Additional file 3. (A) First steps of histidine biosynthesis. Unpredicted essentiality of ACIAD2907 encoding for ribose-phosphate diphosphokinase activity was corrected by removing the alternate gene ACIAD0964 from the reaction GPR. Unpredicted essentialities of ACIAD0661 and ACIAD1257, catalyzing the ATP phosphoribosyltransferase reaction, were corrected by assigning them as complex subunits instead of isozymes in the reaction GPR. (B) Isoprenoids biosynthesis. Unpredicted dispensability of ACIAD2968, catalyzing farnesyl-diphosphate and geranyl-diphosphate synthases activities, was corrected by adding ACIAD1374 (undecaprenyl-diphosphate synthase) and ACIAD2940 (octaprenyl-diphosphate synthase) as isozymes. Unpredicted essentiality of ACIAD1374 was resolved by adding undecaprenyl-PP to the set of essential biomass precursors. (C) Synthesis of charged glutamine-tRNA(gln) and asparagine-tRNA(asn). Unpredicted essentiality of ACIAD1920, encoding for glutaminyl-tRNA synthetase activity, was corrected by removing from the model the alternate pathway using aspartyl/glutamyl-tRNA amidotransferase enzyme (ACIAD0822-0824). (D) Biosynthesis of polysaccharides. Unpredicted dispensabilities of all genes involved in GDP-mannose, UDP-glucose, and dTDP-rhamnose synthesis were corrected by removing these three metabolites from the list of essential biomass precursors.
Mentions: A majority of the model improvements (34/56) were applied to the GPR component, with a clear bias towards false dispensable inconsistencies: 26 GPR corrections pertained to experimentally essential genes against only 8 to experimentally dispensable genes (see Table 3). This large set of false dispensable predictions includes two main inconsistency types. In 22 cases, isofunctional genes with annotations of medium confidence were in fact unable to replace the activity of their deleted isozymes. For instance, ACIAD0964 and ACIAD2907 (prs) were identified in the initial reconstruction as isozymes for the catalysis of the ribose-phosphate diphosphokinase activity, which is required for the biosynthesis of 5-phosphoribosylpyrophosphate (PRPP) (see Figure 8A). The association of both genes to the activity relied on homologies with previously annotated genes in other organisms. The expected and predicted dispensability of ACIAD2907 was yet contradicted by its experimental essentiality. Looking further into the annotation evidence, ACIAD0964 function was supported by only limited homologies to previously known genes (second best hit after ACIAD2907 with E. coli gene prsA, with 25% identity). Conversely, ACIAD2907 function was supported by a stronger homology with E. coli gene prsA (68% identity) whose ribose-phosphate diphosphokinase has been experimentally confirmed [32]. The combination of the observed gene essentialities with the limited homology supporting the annotation of ACIAD0964 led us to correct the model by removing ACIAD0964 from ribose-phosphate diphosphokinase GPR. On the other hand, the functions of some isozymes with medium confidence level were corroborated by the gene essentialities. For instance, two isozymes were indirectly confirmed to have a dihydroxy-acid dehydratase activity, which is essential for the synthesis of valine, leucine and isoleucine. Two duplicate genes were associated with this activity: ACIAD1266 (ilvD) and ACIAD3636. While the annotation of ACIAD1266 is supported by a strong homology with E. coli gene ilvD (74% identity) whose activity has been experimentally shown [33], ACIAD3636's function was supported only by weaker homologies with the reference genes (37% identity with E. coli gene ilvD). Gene knock-outs revealed that both genes were dispensable while the essentiality of other genes in the pathway strongly suggested that the dihydroxy-acid dehydratase activity was required. This result strongly suggests that both genes could back up each other and therefore indirectly corroborates the functional assignment to ACIAD3636.

Bottom Line: The predictions of the final version of the model, which included three rounds of refinements, are consistent with the experimental results for (1) 91% of the wild-type growth phenotypes, (2) 94% of the gene essentiality results, and (3) 94% of the mutant growth phenotypes.To facilitate the exploitation of the metabolic model, we provide a web interface allowing online predictions and visualization of results on metabolic maps.The iterative reconstruction procedure led to significant model improvements, showing that genome-wide mutant phenotypes on several media can significantly facilitate the transition from genome annotation to a high-quality model.

View Article: PubMed Central - HTML - PubMed

Affiliation: Genoscope (Commissariat à l'Energie Atomique) and UMR 8030 CNRS-Genoscope-Université d'Evry, 2 rue Gaston Crémieux, CP5706, 91057 Evry, Cedex, France. mdurot@genoscope.cns.fr

ABSTRACT

Background: Genome-scale metabolic models are powerful tools to study global properties of metabolic networks. They provide a way to integrate various types of biological information in a single framework, providing a structured representation of available knowledge on the metabolism of the respective species.

Results: We reconstructed a constraint-based metabolic model of Acinetobacter baylyi ADP1, a soil bacterium of interest for environmental and biotechnological applications with large-spectrum biodegradation capabilities. Following initial reconstruction from genome annotation and the literature, we iteratively refined the model by comparing its predictions with the results of large-scale experiments: (1) high-throughput growth phenotypes of the wild-type strain on 190 distinct environments, (2) genome-wide gene essentialities from a knockout mutant library, and (3) large-scale growth phenotypes of all mutant strains on 8 minimal media. Out of 1412 predictions, 1262 were initially consistent with our experimental observations. Inconsistencies were systematically examined, leading in 65 cases to model corrections. The predictions of the final version of the model, which included three rounds of refinements, are consistent with the experimental results for (1) 91% of the wild-type growth phenotypes, (2) 94% of the gene essentiality results, and (3) 94% of the mutant growth phenotypes. To facilitate the exploitation of the metabolic model, we provide a web interface allowing online predictions and visualization of results on metabolic maps.

Conclusion: The iterative reconstruction procedure led to significant model improvements, showing that genome-wide mutant phenotypes on several media can significantly facilitate the transition from genome annotation to a high-quality model.

Show MeSH
Related in: MedlinePlus