Characterizing criticality of proteins by systems dynamics: Escherichia coli central carbon metabolism as a working example
© Li and Liu; licensee BioMed Central Ltd. 2012
Published: 16 July 2012
Systems biology calls for studying system-level properties of genes and proteins rather than their individual chemical/biological properties, regarding the bio-molecules as system components. By characterizing how critical the components are to the system and classifying them accordingly, we can study the underlying complex mechanisms, facilitating researches in drug target selection, metabolic engineering, complex disease, etc. Up to date, most studies aiming at this goal are confined to the topology-based or flux-analysis approaches. However, proteins have tertiary structures and specific functions, especially in metabolic systems. Thus topological properties such as connectivity, path length, etc., are not good surrogates for protein properties. Also, the manner of individual sensitivity analysis in most flux-analysis approaches cannot reveal the simultaneous impacts on collateral components as well as the overall impact on the system, thus lacking in system-level perspective.
In the present work, we developed a method to directly assess protein system-level properties based on system dynamics and in silico knockouts, regarding to the conceptual term "criticality". Applying the method to E. coli central carbon metabolic system, we found that multiple enzymes including phosphoglycerate kinase, enolase, transketolase-b, etc., had critical roles in the system in terms of both system states and dynamical stability. In contrast, another set of enzymes including glucose-6-phosphate isomerise, pyruvate kinase, phosphoglucomutase, etc., exerted very little influences when deleted. The finding is consistent with experimental characterization of metabolic essentiality and other studies on E. coli gene essentiality and functions. We also found that enzymes could affect distant metabolites or enzymes even greater than a close neighbour and asymmetry in system-level properties of enzymes catalyzing alternative pathways could give rise to local flux compensation.
Our method creates a different angle for evaluating protein criticality to a biological system from the conventional methodologies. Moreover, the method leads to consistent results with experimental references, showing its efficiency in studying protein system-level properties. Besides working on metabolic systems, the application of the method can be extended to other kinds of bio-systems to reveal the constitutive/functional properties of system building blocks.
Systems biology focuses on studying properties of bio-molecules like genes and proteins at the system level, especially their constitutive/functional roles as system components. By exploring their interplay structure in the system, we can evaluate how critical a building block is and how different parts vary in properties [1, 2]. Based on such knowledge, we can understand how a system is formed, how the system-level function is achieved and whether it can be modified according to our needs, enhancing researches in drug target selection, metabolic engineering, complex diseases, etc [3, 4]. E. coli is the best-studied organism, with knowledge accumulated in each of its biological hierarchies, e.g. genetic regulation, genomic information, metabolism, etc [5–7]. The central carbon metabolism contains glycolysis and pentose phosphate pathways as principal parts (Additional file 1). It is the most common and conservative pathway among prokaryotes, with close resemblance in eukaryotes [5, 7, 8].
Up to date, multiple genome-scale networks have been built on the organism with regard to the pathway to reveal essentiality of genes and proteins [9, 10]. However, most of such studies are based on network topology or flux analysis. In topology-based approaches, system-level properties are defined as the connectivity of a molecule or shortest path lengths, etc [11, 12]. Such properties usually have poor consistency with experimental characterizations, especially on the protein level. For example, multiple studies suggest that proteins with large connectivity in protein-protein interaction networks are not essential. Also, many enzymes associated with large number of accompanies exert very little influence on cell growth [6, 13, 14]. We think the possible reason is that mere topology does not encode specific biochemical/biological functions of proteins, whereas topology-based approaches purely regard the bio-molecules as vertices in an abstract graph. While in flux-analysis approaches (e.g. flux balance analysis - FBA; metabolic control analysis - MCA), they calculate the extent of how a perturbation on a system parameter influences a specific, pre-defined system objective. Although such individual sensitivity analyses give a quantitative measure of a component's control on a functional pathway, they cannot reveal the simultaneous impacts on other parts of the system and the overall system. In other words, besides the pre-defined objective in interest, we will not know if a perturbation triggers collateral influences on other parts of the system and what it implicates to the overall system. The lack in system-level perspective possibly gives rise to false positive results, because a simulated perturbation favouring an objective may not be actually carried out as we do not know if it has lethal impacts on collateral but crucial components in the system.
We computed the criticality for all enzymes in the system, and observed that they can be categorized into two classes: those with critical properties and those with uncritical properties. An enzyme is characterized as critical if its deletion caused large influences on system states and qualitative changes to system dynamical stability.
System state fluctuation
We first simulated the system to obtain metabolite kinetics and flux distributions under normal conditions. Next, we carried out in silico enzyme knockouts by modifying the corresponding parameters and re-simulating the system. Following the definition in previous studies, we regarded concentrations as the primary system state [6, 16]. We calculated state deviations of the modified system and computed the fluctuation amplitude of each metabolite upon the enzyme's removal. Here we encoded them with a vector f. Second, we assessed the impact area by calculating the distances of metabolites from the removed enzyme and encoded them with a vector d. This allowed us to see whether the influence was within a limited radius or propagated to distant parts of the system. In short, we used a vector pair U = (d,f) to represent system state fluctuation, and we could quantify the overall impact with a measurement formula (see section "Methods" for details). All results here were summarized in Additional file 2.
Meanwhile, we also found that there were another group of enzymes, in contrast to those mentioned above, having very little influences on system states. For example, multiple enzymes such as glucose-6-phosphate isomerase (PGI), pyruvate kinase (PK), phosphoglucomutase (PGM), etc., only resulted in very small state fluctuations when deleted. The amplitudes were slight, and the influences were mostly within limited areas as the amplitudes were negligible at the distance ≥ 2 (Figure 2D, Additional file 2). Hence, for system-level properties so far as system state fluctuation was considered, the former enzymes were much greater to those in the latter group.
More interestingly, we found that the most severely influenced metabolites did not always concentrate in the close neighborhood of the perturbed enzyme. For example, the largest impacts of TKb deletion were at the distances of 2 and 3 but not at the distance of 1 (Figure 2A). Likewise, the largest impacts of PGK deletion occurred at the distances of 3 and 4 also not at the distance of 1 (Figure 2B). Similar patterns were also seen from the results of other enzymes like ENO, R5PI, ALDO, GAPDH, TA and PGI (Figure 2, Additional file 2). This suggested that in contrast to the intuition that perturbation would cause largest changes to its neighborhood, distant effects could occur due to the leverage of system dynamics.
We also examined the impacts of enzyme knockouts from the enzyme-centric view with our method. With each enzyme representing a reaction and using fluxes as system states, we computed flux change amplitudes and impact radiuses on the enzyme-centric network in the same way stated previously. The results showed a similar pattern with the results presented here (Additional file 3).
System dynamical stability
Deleting TIS or ALDO caused the system to re-establish equilibrium over an extreme distance beyond the ordinary range (Figure 4D). This indicated that after such a deletion, if the residual system was running on its own, it would approach an extreme position beyond the regular state space due to its special dynamics. In other words, the residual system could not maintain its own regular operating and functionality, thus the deletions of ALDO and R5PI were both regarded as having large qualitative influence in system dynamics.
Moreover, the system had no equilibrium at all after deleting enzymes such as PGK, ENO or GAPDH (Figure 4E). This meant that the original equilibrium was destroyed and the residual system could not re-establish another one. This was because that the residual system upon the removal of anyone of the three enzymes was so ill-suited that its trajectories did not exhibit the normality of well-imposed biological kinetic systems, in which all trajectories tended to stabilize near some regions in the state space. This also indicated that the residual system, if operating on its own, could not effectively maintain its functionality. Hence, the deletions of PGK, ENO and GAPDH were regarded as having even larger qualitative influences in system dynamics compared with the previously mentioned enzymes.
In contrast to the above, enzymes like PGI, PK and PGM again showed a different property. After deleting anyone of them, the residual system still had an equilibrium locating very near to X eq . Moreover, this equilibrium was also asymptotically stable, with all dimensions converging to it (Figure 4F). Therefore, PGI, PK or PGM knockout did not qualitatively change the system dynamics. Hence, for system-level properties so far as dynamical stability was considered, enzymes like PGK, ALDO, TKb, etc. were more critical than enzymes like PGI, PK, and PGM. Based on all above, we could see that one class of enzymes exemplified by PGK, ENO, TKb, ALDO, TIS, R5PI, GAPDH, and TA have critical properties in terms of both impact on system states and dynamical stability. And the other class of enzymes exemplified by PGI, PK, and PGM had opposite properties. Therefore, the former class was characterized as "critical" and the latter was "uncritical".
Comparison with experimental characterizations
Comparison with multiple E. coli gene essentiality studies
Studying system-level properties of bio-molecules is essential to systems biology [1, 2]. But most studies are based on either network topology that is not working very well at the protein level, or flux analysis that lacks in system level perspective [13, 14, 21]. To overcome such drawbacks, we propose a method of criticality characterization on the basis of kinetic modeling. In a kinetic system, every interaction is expressed by a kinetic rate equation. How a component influences the system is determined by both its position and the kinetic parameters. Position is equivalent to topological property, while kinetic parameters encode specific biochemical/biological functions. Both kinds of information are integrated in modeling and revealed by dynamical simulation [15, 22]. According to the typical formulism of biochemical systems, the kinetic rate equations constitute the deterministic part of the complex system dynamics and they can be viewed as the "driving force" of the system . Thus theoretically, the criticality characterization proposed in our method is the study of structural factors built into the "driving force" of a system.
Differing from topology-based methods, our method characterizes system-level properties on the quantitative basis. But unlike the conventional sensitivity analysis, we employ the network structure information by calculating the distances from the deleted spot to the affected entities besides computing the fluctuations. Moreover, unlike conventional flux-analysis approaches, we explore the system stability and retrieve system dynamics structure. Incorporating the network/dynamics structure information allows us to reveal the simultaneous/collateral influences and the overall impact on the system. Another major difference from the sensitivity analysis is that we use in silico deletions instead of mild perturbations (e.g. 5% or 10%, as most flux-analysis approaches do). Because a well-casted biological network usually has parametric properties favouring the robustness in dynamics, critical components may well tolerate mild perturbations (i.e. parameters exhibiting the Lyapunov stability). Therefore, individual sensitivity analysis often fails to identify such critical spots, and its inability to reveal simultaneous influences worsens the situation. That is why we develop the "criticality characterization". In silico deletion is equivalent to investigating how the system would be if the component is forcefully assumed to be absent, eliminating the parametric properties stated earlier. Furthermore, our method's capacity of revealing simultaneous/overall impacts at the system level enables it to distinguish real critical spots from uncritical ones more effectively. In addition, utilizing kinetic model as the analytical basis is a superiority over the stoichiometric flux-balance modeling in traditional flux-analysis methods, enabling us to appropriately explore system behaviours in the real-time scale . For example, both traditional topology-based and flux-analysis approaches regard TIS as peripheral as it is not highly connected and it is not on any uni-directional or rate limiting reactions. However, there were experimental studies showing that knocking out tpiA (i.e. the gene encoding TIS) attenuated the cell growth by about 70%. And our method appropriately revealed that TIS could exert large impacts on the system if deleted, because of the designs we made (mentioned above). Hence methodologically, our method creates a different angle from topology-based methods and can be viewed as an improvement of conventional flux-analysis approaches.
After in silico deleting a protein, the residual system is actually a virtual structure. We assume that this structure encodes important information about whether the mutant can maintain its functionality and how it would dynamically behave/evolve provided that it stills operates on its own. The residual system fails to maintain functionality when its kinetics goes beyond regular ranges (e.g. occurring negative values, or soaring to extreme values exceeding regular intracellular molecular concentrations), or its dynamics is trapped in a mode where the stable equilibrium is sabotaged, as stable equilibrium gives rise to robustness and is an essential prerequisite for valid mathematical formulations of living cell dynamics [17, 24–26]. Either case indicates that deleting the protein makes the system so ill-suited that it cannot run on its own.
By utilizing the power of kinetic model for approaching real-time events, we simulated fluxes after enzyme deletions and compared the results with a previously validated study of metabolic essentiality . The comparison shows that our characterization of criticality is well supported by functional essentiality. Interestingly, we discovered that the asymmetry in criticalities of building blocks might give rise to local flux compensation. For instance, multiple metabolites (e.g. ribulose-5-phosphate, sedoheptulose-7- phosphate, etc.) in the pentose phosphate pathway have increased flux-sums after PGI knockout (Figure 5D and 6C). The cutoff of PGI induces the two alternative pathways for generating the essential metabolite fructose-6-phosphate, TKb and TA, to operate at a greater volume. Thus fluxes through relative reactions are compensated, resulting in local amplified fluxes. This is a likely result in accordance to the MOMA mechanism . Although MOMA can compensate system fluxes/states to some degrees, our results show that the effects caused by deletions of critical components such as TKb, TA, PGK, etc. cannot be smoothed by such compensations (Figure 5A-5B, Additional file 5). This is because such compensations are mainly mediated by alternative pathways . When a critical component is deleted, leaving inferior components as backup to rely on, the system cannot work efficiently. On the contrary, deleting PGI leaves its two alternative pathways that are of superior properties at the "ON" state and the system still works, thus fluxes/states can be efficiently compensated. This gives a hint on how criticality characterization can help in bio-system modifications such as in metabolic engineering. We can delete some system components with inferior properties, leaving alternative pathways with superior properties to work. And phenotypes in the local areas relating to such alternative pathways might be compensated due to the leverage of system structure and the MOMA mechanism. Therefore, comprehensive methods of exploring system-level properties can help us make use of bio-complexity in engineering, as well as in knowledge discovery.
It is noteworthy that functionally important components are not necessarily critical, as studies suggest that the more important a reaction is in function, the more likely that it has a backup pathway [6, 13]. For example, PK connects very fundamental chemical compounds but it is regarded as uncritical at the system level, because there are alternative paths (e.g. the phosphotransferase system - PTS, in bacteria glycolysis; Additional file 1) that can prevent large impacts on system kinetics/dynamics. This exemplifies that bio-system components have dichotomy. They have "importance" as biochemical molecules, and they also have "criticality" to the system as constitutive building blocks. Actually, our method does not aim to find the "functionally important" molecules, but those "critical" to the system, i.e. components that cannot be absent, or the system will be severely aberrant. Since the criticality of an enzyme depends on many factors (e.g. kinetic parameters, substrates inhibiting/activating other reactions, degree of the effects, etc.), the assignments of system boundaries in modeling might affect prediction results. As the enzymes located on the boundary might have incomplete interplay structure, the above factors may not occur properly in the kinetic equations. Therefore, accurate criticality characterization is facilitated by appropriate system inclusiveness in modeling. For example, glucose-1-phosphate adenyltransferase (G1PAT) only connects the external polysaccharide synthesis pathway, with few interactions with large-capacity reactions both in the system and outside pathways. Thus as the boundary is assigned up to it, the validity of the results are enhanced (Table 1). Furthermore, fundamental, common and conserved pathways must be chosen for comparison with genome-scale gene essentiality studies that regard to global cellular functionality. For instance, the bacteria central carbon metabolism here is an appropriate example [5, 7, 8], thus various predictions of protein criticality are well consistent with global gene essentiality characterizations [13, 14, 19].
Although we used a metabolic system as the working model, the application of our method is not confined to metabolic systems. For instance, we can model gene transcription dynamics by deriving gene transcription rate with the power-law formulism, the Hill equation, or equations of chemical kinetic actions [27–29]. Or we can describe ligand-receptor and protein-protein binding actions with the mass action law and build models for signaling networks . We even do not have to obtain exact parameters fitting the modeled solutions to assay measurements when analyzing the generic behavioral potential of the system, e.g. in what parameter ranges the system exhibits certain dynamics and how they change with parameters. Such qualitative predictions are also useful in revealing general principles governing complex bio-systems. Naturally, complicated bifurcation dynamics will be harder to analyze; but the idea of our method can be well applied once the coexisting dynamical characteristics in bifurcation are associated with biological implications . By integrating knowledge and using theoretical generic forms of models [15, 30], kinetic modeling will be eventually feasible for more organisms. Hence, instead of the traditional approaches, we propose that complex systems be studied by casting the network into kinetic equations and computing the system-level properties with respect to system kinetics/dynamics (criticality). Overall, our method may provide a new viewpoint in revealing constitutive/functional properties of building blocks in a biological system.
Our method creates a new angle from traditional topology-based methodologies for evaluating system-level properties of bio-molecules. Moreover, the proposed method can be viewed as an improvement of the conventional flux-analysis approaches such as FBA and MCA. In addition, the method leads to results that are consistent with experimental references, showing that it is efficient in characterizing protein criticality and studying biological systems. Furthermore, the method's application can be extended to other types of bio-systems (e.g. transcriptional networks and signaling networks) to reveal the constitutive/functional properties of system building blocks.
We utilized existing kinetic data in E. coli central carbon metabolism and adopted a previous modeling framework as our working platform . The kinetic model consists of 30 metabolites (including external metabolites and biosynthesis products) and 30 biochemical reactions (24 enzymes and 6 lumped reactions standing for transport/biosynthetic processes relating to external pathways; Additional file 1). The model can also be re-casted into an enzyme-centric network, by adding a directed connection from enzyme A to B if any of A's products was B's substrate. We could explicitly see the interactions among enzymes from the enzyme-centric view (Additional file 1).
Here vector X denoted system state and P denoted kinetic parameters. R was a function vector collocated by all rate equations, and A was the stoichiometric matrix. B was the term standing for extra reactions (e.g. transport, metabolite utilization for cellular growth, etc). Most parameters were found in published studies and the rest could be estimated using the experimental conditions, steady-state reaction rates and concentrations reported in previous studies [5, 31, 32]. For complete descriptions of metabolites, reactions, forms of kinetic rate equations and ODEs, see Additional file 6.
Dynamical simulation and state fluctuation
Normal bio-systems are subjected to robustness as they structurally consist of abundant alternative pathways and feedback loops [6, 17, 24]. Thus valid formulations of bio-systems usually have stable equilibrium, attracting neighborhood trajectories and allowing slight changes to be tolerated without disturbing normality [5, 25, 26]. The trajectories tend to some area over adequately large ranges of time and parameter spaces if the system has equilibrium. And if it did not, trajectories spread out along some dimensions traversing several orders of magnitudes. To locate the equilibrium, we utilized the state at the end time point of simulation as an initial guess and used the trust-region method to solve the problem [26, 34]. By carefully refining the numerical approach, the equilibrium could be computed and distances from the original X eq were calculated by the Euclid norm.
If the bifurcation caused by an in silico deletion (parameter modification) yields multiple equilibriums, the impact on dynamical stability is regarded as large if anyone of the equilibriums exhibit qualitative difference from X eq in dynamical properties.
MOMA and flux-sum
Here P μ was the parameter set with the relevant enzymatic parameters deleted, X 0 was the original state and C 0 was the initial value. Some states that were close to X 0 in the feasible space could be solved with the genetic algorithm, a heuristic numerical approach that can alleviate computation difficulty in large variable space to some extent.
where Ω k was the index set of reactions producing metabolite k.
After MOMA computation, we obtained one (or more) set of parameters and system states. Using rate equations, we simulated the fluxes and calculate flux-sums according to Eqn (7).
We thank Yi-Xue Li for valuable advices about system dynamics analysis.
This article has been published as part of BMC Systems Biology Volume 6 Supplement 1, 2012: Selected articles from The 5th IEEE International Conference on Systems Biology (ISB 2011). The full contents of the supplement are available online at http://www.biomedcentral.com/bmcsystbiol/supplements/6/S1.
- Hood L: Systems biology: Integrating technology, biology and computation. Mech Ageing Dev. 2003, 124: 9-16. 10.1016/S0047-6374(02)00164-1.View ArticlePubMedGoogle Scholar
- Ideker T, Galitski T, Hood L: A new approach to decoding life: Systems biology. Annu Rev Genomics Hum Genet. 2001, 2: 343-372. 10.1146/annurev.genom.2.1.343.View ArticlePubMedGoogle Scholar
- Bailey JE: Toward a science of metabolic engineering. Science. 1991, 252: 1668-1675. 10.1126/science.2047876.View ArticlePubMedGoogle Scholar
- Li H, Zhan M: Systematic intervention of transcription for identifying network response to disease and cellular phenotypes. Bioinformatics. 2006, 22: 96-102. 10.1093/bioinformatics/bti752.View ArticlePubMedGoogle Scholar
- Chassagnole C, Noisommit-Rizzi N, Schmid JW, Mauch K, Reuss M: Dynamic modeling of the central carbon metabolism of Escherichia coli. Biotechnol Bioeng. 2002, 79: 53-73. 10.1002/bit.10288.View ArticlePubMedGoogle Scholar
- Kim PJ, Lee DY, Kim TY, Lee KH, Jeong H, Lee SY, Park S: Metabolite essentiality elucidates robustness of Escherichia coli metabolism. Proc Natl Acad Sci USA. 2007, 104: 13638-13642. 10.1073/pnas.0703262104.PubMed CentralView ArticlePubMedGoogle Scholar
- Neidhardt FC, Curtiss R, Ingraham JL, Lin ECC, Low KB, Magasanik B, Reznikoff W, Riley M, Umbarger HE: Escherichia coli and Salmonella: Cellular and molecular biology. 1996, Washington, DC: ASM PressGoogle Scholar
- Milo R, Itzkovitz S, Kashtan N, Levitt R, Shen-Orr S, Ayzenshtat I, Sheffer M, Alon U: Superfamilies of Evolved and Designed Networks. Science. 2004, 303: 1538-1542. 10.1126/science.1089167.View ArticlePubMedGoogle Scholar
- Ma H, Zeng AP: Reconstruction of metabolic networks from genome data and analysis of their global structure for various organisms. Bioinformatics. 2003, 19: 270-277. 10.1093/bioinformatics/19.2.270.View ArticlePubMedGoogle Scholar
- Covert MW, Palsson BO: Transcriptional Regulation in Constraints-based Metabolic Models of Escherichia coli. J Biol Chem. 2002, 277: 28058-28064. 10.1074/jbc.M201691200.View ArticlePubMedGoogle Scholar
- Jeong H, Tombor B, Albert R, Oltvai ZN, Barabasi AL: The large-scale organization of metabolic networks. Nature. 2000, 407: 651-654. 10.1038/35036627.View ArticlePubMedGoogle Scholar
- Vitkup D, Kharchenko P, Wagner A: Influence of metabolic network structure and function on enzyme evolution. Genome Biol. 2006, 7: R39-10.1186/gb-2006-7-5-r39.PubMed CentralView ArticlePubMedGoogle Scholar
- Ghim CM, Goh KI, Kahng B: Lethality and synthetic lethality in the genome-wide metabolic network of Escherichia coli. J Theor Biol. 2005, 237: 401-411. 10.1016/j.jtbi.2005.04.025.View ArticlePubMedGoogle Scholar
- Gerdes SY, Scholle MD, Campbell JW, Balazsi G, Ravasz E, Daugherty MD, Anderson I, Gelfand MS, Bhattacharya A, Kapatral V, D'Souza M, Baev MV, Grechkin Y, Mseeh F, Fonstein MY, Overbeek R, Barabasi AL, Oltvai ZN, Osterman AL: Experimental Determination and System Level Analysis of Essential Genes in Escherichia coli MG1655. J Bacteriol. 2003, 185: 5673-5684. 10.1128/JB.185.19.5673-5684.2003.PubMed CentralView ArticlePubMedGoogle Scholar
- Li RD, Li YY, Lu LY, Ren C, Li YX, Liu L: An improved kinetic model for the acetone-butanol-ethanol pathway of Clostridium acetobutylicum and model-based perturbation analysis. BMC Syst Biol. 2011, 5: S12-PubMed CentralView ArticlePubMedGoogle Scholar
- Segrè D, Vitkup D, Church GM: Analysis of optimality in natural and perturbed metabolic networks. Proc Natl Acad Sci USA. 2002, 99: 15112-15117. 10.1073/pnas.232349399.PubMed CentralView ArticlePubMedGoogle Scholar
- Kitano H: Biological robustness. Nat Rev Genet. 2004, 5: 826-837.View ArticlePubMedGoogle Scholar
- Wagner A: Robustness and evolvability in living systems. 2005, Princeton: Princeton University PressGoogle Scholar
- Baba T, Ara T, Hasegawa M, Takai Y, Okumura Y, Baba M, Datsenko KA, Tomita M, Wanner BL, Mori H: Construction of Escherichia coli K-12 in-frame, single-gene knockout mutants: the Keio collection. Mol Syst Biol. 2006, 2: 2006.0008-PubMed CentralView ArticlePubMedGoogle Scholar
- Hashimoto M, Ichimura T, Mizoguchi H, Tanaka K, Fujimitsu K, Keyamura K, Ote T, Yamakawa T, Yamazaki Y, Mori H, Katayama T, Kato J: Cell size and nucleoid organization of engineered Escherichia coli cells with a reduced genome. Mol Microbiol. 2005, 55: 137-149.View ArticlePubMedGoogle Scholar
- Desai RP, Harris LM, Welker NE, Papoutsakis ET: Metabolic flux analysis elucidates the importance of the acid-formation pathways in regulating solvent production by Clostridium acetobutylicum. Metab Eng. 1999, 1: 206-213. 10.1006/mben.1999.0118.View ArticlePubMedGoogle Scholar
- Lee JM, Gianchandani EP, Eddy JA, Papin JA: Dyanmic analysis of integrated signaling, metabolic, and regulatory networks. PLoS Comput Biol. 2008, 4: e1000086-10.1371/journal.pcbi.1000086.View ArticlePubMedGoogle Scholar
- Ao P: Metabolic network modeling: Including stochastic effects. Comput Chem Eng. 2005, 29: 2297-2303. 10.1016/j.compchemeng.2005.05.007.View ArticleGoogle Scholar
- Kim D, Kwon YK, Cho KH: Coupled positive and negative feedback circuits form an essential building block of cellular signaling pathways. BioEssays. 2007, 29: 85-90. 10.1002/bies.20511.View ArticlePubMedGoogle Scholar
- Kuznetsov YA: Elements of applied bifurcation theory, Springer. 2004, 3View ArticleGoogle Scholar
- Strogatz SH: Nonlinear dynamics and chaos: with applications to physics, biology, chemistry, and engineering. 2001, Westview PressGoogle Scholar
- Savageau MA, Voit EO: Recasting nonlinear differential equations as S-systems: a canonical nonlinear form. Math Biosci. 1987, 87: 83-115. 10.1016/0025-5564(87)90035-6.View ArticleGoogle Scholar
- Huang S, Guo YP, May G, Enver T: Bifurcation dynamics in lineage-commitment in bipotent progenitor cells. Dev Biol. 2007, 305: 695-713. 10.1016/j.ydbio.2007.02.036.View ArticlePubMedGoogle Scholar
- Bintu L, Buchler NE, Garcia HG, Gerland U, Hwa T, Kondev J, Phillips R: Transcriptional regulation by the numbers: models. Curr Opin Genet Dev. 2005, 15: 116-124. 10.1016/j.gde.2005.02.007.PubMed CentralView ArticlePubMedGoogle Scholar
- Lee LW, Yin L, Zhu X, Ao P: Generic enzymatic rate equation under living conditions. J Biol Syst. 2007, 15: 495-514. 10.1142/S0218339007002295.View ArticleGoogle Scholar
- Bhattacharya M, Fuhrman L, Ingram A, Nickerson KW, Conway T: Single-run separation and detection of multiple metabolic intermediates by anion-exchange high-performance liquid chromatography and application to cell pool extracts prepared from Escherichia coli. Anal Biochem. 1995, 232: 98-106. 10.1006/abio.1995.9954.View ArticlePubMedGoogle Scholar
- Buziol S, Bashir I, Baumeister A, Claaßen W, Noisommit-Rizzi N, Mailinger W, Reuss M: New bioreactor-coupled rapid stopped-flow sampling technique for measurements of metabolite dynamics on a subsecond time scale. Biotechnol Bioeng. 2002, 80: 632-636. 10.1002/bit.10427.View ArticlePubMedGoogle Scholar
- Hairer E, Wanner G: Solving ordinary differential equations II: Stiff and differential-algebraic problems. 1996, Berlin: Springer-Verlag, 2View ArticleGoogle Scholar
- Conn AR, Gould NIM, Toint PL: Trust-region methods. MPS/SIAM Series on Optimization. Society for Industrial Mathematics. 1987Google Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.