Work

Simulating Plant Metabolic Pathways with Enzyme-Kinetic Models

Year: 2010
Type: article
Abstract: The complexity of metabolic networks and their regulation renders an intuitive analysis of these biological systems a difficult task. Mathematical modeling approaches help to deal with this complexity, making them an important tool for metabolic engineering. Different methods were developed, ranging from basic stoichiometric models up to fine-grained kinetic models. Kinetic modeling is the most detailed and complex mathematical description of a metabolic network and constitutes an important branch in the growing fields of systems biology. In this update, we provide a guide for the construction, simulation, and analysis of kinetic metabolic models in general, before we describe some recently published models of plant metabolic pathways, giving an overview of the opportunities and challenges of this mathematical method. Furthermore, we evaluate the current strategies and take an outlook to possible and necessary future developments of kinetic modeling. A metabolic network can be translated in mathematical terms by relatively easy means: The concentration of a metabolite is described by a variable S i. As a result of mass balances (no matter can appear or disappear), the change of this variable over time (dS i/dt) is given by the sum of the rates of the enzymes synthesizing the metabolite minus the sum of the rates of the enzymes utilizing the metabolite. The rate of each enzymatic step can be described by enzyme-kinetic rate laws, such as the Michaelis-Menten equation, as a function depending on metabolite concentrations and parameters such as the maximal velocity of a reaction, or binding constants. This process yields a system of ordinary differential equations (ODEs) in which dS i/dt is on one side and the metabolite-dependent rate laws are on the other side of the equations (see Eqs. 4 and 5 below). With this system of differential equation, the metabolic network can be simulated, and by solving the system of ODEs the steady state can be calculated, in which all reaction rates and metabolite concentrations are constant. The process described above is simplified by considering the cell as a homogenous volume. While for many scenarios this is a valid simplification, for other instances, ODE-based kinetic models will not be the method of choice. For example, if spatial gradients are of importance to address a certain question, partial differential equations would be used instead, which increases the complexity of a model dramatically. Furthermore, if metabolites with very low concentrations are considered (e.g. hormones), stochastic effects are coming into play, so that the model would include stochastic components rather than being deterministic. More details on different methods for metabolic modeling are given in a recent comprehensive overview of computational models of metabolism (Steuer and Junker, 2009). Deterministic, ODE-based enzyme-kinetic models have the longest history in the area of metabolic pathway modeling. The local complexity of kinetic models leads to the fact that for reasons of feasibility usually single pathways or even just some reactions are modeled. For decades kinetic models have been the most frequently used mathematical approach to metabolism, starting with the work of B. Chance who published the first numerical simulation of a biochemical system in 1943, solving the equations for the behavior of a simple enzymatic system using a mechanical differential analyzer (Chance, 1943). Before starting the construction of a kinetic model it is necessary to define the scope of the model. This definition must be as exact as possible, because it will have a profound impact on all further steps of the modeling process. Once this purpose has been defined, the borders of the model can be set accordingly. Due to the connection between model size and processing complexity, a kinetic model will normally describe just some reactions up to a few metabolic pathways. We will illustrate the generation of a kinetic model by the example of the Higgins-Sel'kov oscillator (Higgins, 1964; Sel'kov, 1968), a simple kinetic model that has been thoroughly analyzed over time and that has been used as a teaching example earlier (Klipp et al., 2005). The modeling process is summarized in Figure 1. Construction of an exemplary kinetic model. A, Information about basic connectivities. B, Stoichiometric data adds numerical dependencies between the metabolites. C, Rate laws defining the reaction rate of each enzymatic step. D, Combining the stoichiometric data with the rate laws into a system of ODEs describing the change in metabolite concentrations over time. [See online article for color version of this figure.] For establishing a detailed kinetic model, a variety of data sources have to be consulted to obtain detailed and reliable information. The stoichiometry of a reaction is usually well established and can be taken from standard biochemistry textbooks or from online databases such as the examples listed in Table I. The structure of the rate law depends on the stoichiometry of the reaction and on the enzyme's mechanism and may be derived from enzyme kinetics textbooks (Segel, 1975). The maximal velocity of the reaction (V max) is dependent on the concentration of the enzyme in the cell or the extract, but rather a result of regulated gene expression, and thus has to be measured for each situation again. We note here that the V max used in kinetic models is not the specific activity that is measured with the purified enzyme. Thus, V max is dependent on the enzyme's concentration and needs to be measured in vivo or in crude extracts. For reversible reactions the V max is different for the forward and reverse reaction, respectively. The binding constant (or K m) results from the structure of the enzyme, and thus is independent of the enzyme's concentration. We therefore argue that in many cases it is a legitimate simplification to take the K m values from the literature, also from related species if the data situation requires. Information for the model creation process is collected from databases, and the models are analyzed with simulators. To standardize models and simulations, data standards have been established for the exchange and description of models and simulations. Finally, there are public model repositories in which kinetic models are collected. Information for the model creation process is collected from databases, and the models are analyzed with simulators. To standardize models and simulations, data standards have been established for the exchange and description of models and simulations. Finally, there are public model repositories in which kinetic models are collected. The state variables of the system, v(S) and S, i.e. the fluxes and metabolite concentrations, respectively, can be measured for a specific state of the system, usually at (quasi) steady state, but sometimes also as time course. If enough parameters of the model are known, it is theoretically possible to estimate a limited number of missing parameters by fitting the system variables (metabolite concentrations) to data sets from steady-state or time-course measurements. Care has to be taken as the data used for the parameter estimation is not to be used in the model validation. Once the stoichiometry, the rate laws, and all parameters are collected, the equation system of the kinetic model is fully specified. Using specialized software (Table I), the model can thus be deterministically simulated in a stepwise manner. Starting from a defined point, the differential equations are solved to obtain the expected changes in metabolite concentrations over a short time period, the concentrations are then updated with these changes, and the process is repeated. Depending on the model and its parameters, different behaviors might be observed. Figure 2 shows the example model characterized by different sets of parameters that results in different behavior. In Figure 2A the system converges toward a steady state, which is the case for most biological systems if they are kept under constant conditions. At the steady state, all fluxes in this linear pathway are the same, so that the concentrations of the metabolites do not vary. In Figure 2B, a divergent behavior is depicted: one metabolite concentration is constantly increasing. A similar scenario is obtained if one metabolite is depleted with such a high rate that it eventually reaches zero. Consequently, the metabolic network might come to a complete hold. Changing the same parameter in the example model to yet another value results in oscillatory behavior of the system (see Fig. 2C). It has been shown that the probability of a model to exhibit oscillatory behavior or other instability actually increases with its size (May, 1973). Consequently, for large models it is an exception rather than common to reach a stable steady state. The reason why large biological systems nevertheless tend to steady states might be that they are the product of evolution. Concentration time course in kinetic model with different parameter sets, exhibiting converging (A), diverging (B), and oscillatory behavior (C). The model is described by Equations 1 to 5. Parameters are k 1 = k 2 = 1 and k 3 = 0.5 (A); k 1 = k 2 = 1 and k 3 = 1.5 (B); and k 1 = k 2 = k 3 = 1 (C). Units are arbitrary. Modeling tools or other mathematical frameworks are indispensable for building and interrogating a model. To simplify using different tools to analyze one model, data standards and exchange languages have been defined. Finally, models can be deposited in online databases to make them accessible for the scientific community (Table I). A steady state or stationary state of a metabolic system is defined by the fact that all of the state variables (i.e. metabolite concentrations and fluxes) stay constant in value over time, which means that the system will not leave steady state unless a change of these state variables, a perturbation occurs. A perturbation can be a change in metabolite concentration or enzyme activity. The behavior of a system after a perturbation differs if its steady state is stable (i.e. the perturbed system returns to a steady state), unstable (i.e. the perturbed system leaves steady state), or metastable (i.e. the perturbed system oscillates). A very useful visualization of steady-state conditions is the impact of metabolic fluxes on metabolite concentrations. A metabolite concentration (S) is defined to be changed by two fluxes, synthesis (v syn) and consumption (v con). If the steady state at a given metabolite concentration (S 0) of this system is stable, the metabolite concentration and fluxes should return to their original status after an infinitesimal perturbation. In case of product inhibition, the synthesis will diminish with increasing product concentration, while the consumption is usually increased (see Fig. 3). Metabolic fluxes representing consumption and synthesis of a metabolite changing over time. The steady state is only reached at the point where synthetic flux (v syn) and consumption flux (v con) are equal. Adapted from Steuer and Junker (2009). After a perturbation the metabolite concentration will be either higher or lower than the concentration at steady state. This dictates the necessary net flux to reach steady-state conditions: If the concentration after perturbation is lower than before (S 0), the net flux has to be positive to reach the steady state. Accordingly, if the concentration is higher after perturbation than before (S 0), the net flux must be negative. In a system in which the net flux shows a more complicated course with the increase of metabolite concentration, more than one steady state can occur (e.g. Poolman et al., 2000). But, as mentioned before, not all steady states are stable after a perturbation. If the net flux is negative and the concentration at the same time point is lower than at the steady state (or vice versa), the system will further shift away from this steady state. A stable steady state represents a local or global minimum; therefore the gradients of all points close to the steady state must be negative. These steady states are comparable to the deepest point of a curved surface, so the system will return to this point after perturbation. Unstable steady states would correspond, for example, to saddle points on this surface. This information on the gradients in the vicinity of a steady state can be derived from the Jacobian matrix, which describes the flux changes occurring upon small perturbations of the metabolic system (Klipp et al., 2005; Steuer and Junker, 2009). After identifying a steady state, the response to a system perturbation can be simulated, e.g. by inducing a change of a metabolite concentration and observing the following return to a steady state (metabolite pulse). Further applications are in silico overexpression or silencing experiments. Effectively, the overexpression or silencing of an enzyme can be simulated by increasing or decreasing, respectively, the maximal activity (v max) of the enzyme in question. The expression of an additional enzyme working in the metabolic network is simulated in the same way, for example the simulated expression of Suc phosphorylase in potato (Solanum tuberosum) tubers (Junker, 2004). One of the long-term applications of metabolic analysis of plants is the change of metabolite levels for product improvement (i.e. increase of biosynthesis, reduction of toxic compounds, etc). For decades there has been the quest for rate-limiting steps or bottlenecks in metabolic pathways. It took a relatively long time to appreciate the theory of metabolic control analysis (MCA) for quantifying the impact of a change in enzymatic activity on a flux and/or a metabolite, which has been already developed in the 1970s (Kacser and Burns, 1973; Heinrich and Rapoport, 1974). The main point of the theory is that the control of metabolic fluxes and metabolite concentrations is distributed across many enzymes of a metabolic pathway, and that this control can be quantified for each enzymatic step. Nowadays it has been internalized that the strategy to overexpress one enzyme to increase the flux through an associated pathway will most likely fail (Morandini and Salamini, 2003). This is supported by a recent study in which was shown for a certain pathway that maximal catalytic activities (i.e. limiting rates) of most enzymes were in large surplus of the fluxes through these reaction steps (Junker et al., 2007). These coefficients are the quantitative connections between a change in the maximal catalytic activity (A) of an enzyme and the resulting change of flux (J) or metabolite concentration (S), respectively. So these relations compare the named factors between two different steady states. Control coefficients can be computed given a kinetic metabolic model, but they can also be measured by observing the changes in fluxes or metabolite concentrations obtained upon genetic modification of the activity of an enzyme. A flux control coefficient close to 1 indicates an enzyme with rate-limiting capabilities. However, such a value is rather unusual (Fell, 1992, 1997), meaning that the classic idea of a true rate-limiting enzyme is mostly not valid, as stated above. The generation of kinetic models using explicit kinetics is arguably the most detailed depiction of a metabolic system. Nevertheless it also requires values and numerous parameters that are usually not easy to determine. Thus, the process kinetic modeling suffers from a notorious lack of information. Several other approaches have been proposed to overcome this problem, usually by simplifying the underlying kinetic formulas without losing the physiological closeness to the in vivo situation. The performance of the lin-log kinetics is in certain cases better than power-law functions or linear approximation (Heijnen, 2005), because the elasticities and therefore the kinetic orders are not constant in the lin-log framework, but dependent on metabolite concentrations. This approach is not restricted to an analytical solution, but maintains the interpretability of the involved parameters in the metabolic context. However, the lin-log function shares the problem of a required reference state with the BST as well as inaccuracies at small concentration values. However, the latter is sometimes considered to be a lesser problem due to the limited concentration ranges in metabolic pathways (Heijnen, 2005). Structural kinetic modeling is a recent approach to close a gap between pure structural modeling derived from stoichiometric data and kinetic modeling, depicting the dynamic values of a metabolic system (Steuer et al., 2006; Steuer and Junker, 2009). The main difference to other approaches is the parametric representation of the Jacobian matrix for all possible parameters. This gives the possibility to access certain dynamic behaviors (e.g. oscillation) without knowing the exact functional form of the rate laws or the values of the parameters. It is not astonishing that the first models of plant metabolic pathways were built for the Calvin cycle, arguably the most important metabolic pathway in plants (for review, see Poolman et al., 2001). Comprehensive lists of kinetic models of any plant metabolic pathway have been published in recent reviews (Morgan and Rhodes, 2002; Rios-Estepa and Lange, 2007). Furthermore, various models can be found in model repositories (see Table I) such as Java Web Simulation Online (Olivier and Snoep, 2004), which also allows amateur users to explore the modeling and simulation process. The first model of parts of photosynthesis linked equations for Rubisco kinetics with stoichiometric equations of the photosynthetic carbon reduction cycle and the photorespiratory carbon oxidation cycle (Farquhar et al., 1980). In a recent approach, Zhu and coworkers constructed a large compartmented model of leaf carbon metabolism comprising more than 40 reactions and transporters (Zhu et al., 2007). To simulate the effect of evolutionary adaptation to higher CO2 concentrations, they allowed the model to adapt by redistributing the amount of protein nitrogen between the enzymes by means of an evolutionary algorithm. They found a substantial increase in photosynthesis, suggesting that the typical partitioning in C3 leaves might be suboptimal for maximizing the light-saturated rate of photosynthesis. The Suc-to-starch has been the of two kinetic models of potato one to study the effect tubers are of the plant and 2005), and the other one to simulate the effect of increased activities in enzymes on metabolic fluxes (Junker, 2004). In cases the models the in In to potato metabolism in in the of Suc In an earlier model of Suc in and each reaction step was with kinetics by using detailed and The between and simulation a of The that overexpression of the or or the Suc as well as reduction of levels appear to be the most for genetic The version of the model et al., also information about so that the model describes the metabolic behavior as in different of the One of the established kinetic models the pathway in et al., 2009). This model was on in kinetic and in vivo data metabolite concentrations and The pathway branch points and many enzyme to the simulation, these to the regulation of the fluxes and are therefore not This is a of the complex kinetic model in the synthesis of et al., which are important metabolites in The model the leaves with of and These enzymes define the of through control of the number of The of the in the simulation with the measured Furthermore, the model a connection between the of the two enzyme in so as the overexpression of is to the measured in of plant metabolic pathways were already built long before the systems was and will to an important in approaches to plant The increase in biochemical data and will these approaches in the However, the of the data will increase through such as and online measurements. A that has to be in the future is the of data et al., 2004). Furthermore, a standard for modeling for complex systems into and needs to be established et al., 2009). The connection of metabolic models with other of regulation is another example for future It is useful to have a at kinetic models of other such as and to estimate future developments in plant kinetic modeling. the metabolic networks in are well characterized in to plant In an approach for modeling et by modeling the effect on metabolic fluxes, by the of this information into the rate of a single the of a of was simulated to estimate the concentration et al., 2009). This might be to such as plants by a of rather than a single cell that is usually the better in systems and the generation of metabolic networks from data it is necessary to a kinetic model from a given metabolic network et al., 2009). This process is further complicated by the of the the main for the future in all the fields with systems is the reliable and or from established see Table I) and of data into metabolic increasing the of the derived models. steps in the of biological systems from an point of were recently for synthetic et al., and systems et al., 2009). the of metabolic modeling all other systems depends on the of and in all fields such as systems but not biology.
Language: English
Open Access status: green
OpenAlex ID: W2171783541
FWCI: 1.88
Cites: 60
Cited by: 83
Related to: 10