q-bio0602002/Analysis03.tex
1: %%%%%%%%%%%%%%%%%%%%%%% file Analysis03.tex %%%%%%%%%%%%%%%%%%%%%%%%%
2: \documentclass[12pt,a4paper]{article}
3: 
4: %----- Globale Optionen setzen ------------------------------------------------%
5: \parindent 0mm 
6: \oddsidemargin -10mm
7: \evensidemargin -10mm
8: \textwidth 180mm
9: \textheight 230mm 
10: \topmargin -10mm
11: 
12: 
13: 
14: % Remove option referee for final version
15: %
16: % Remove any % below to load the required packages 
17: \usepackage{graphics}
18: \usepackage{epsfig}     
19: 
20: \usepackage[dvips]{color}
21: \usepackage{color}
22: \usepackage{amssymb}       
23: \usepackage{times}     
24: \usepackage{bm}
25: 
26: \begin{document}
27: 
28: 
29: \author{
30: Ralf Steuer$~^\mathrm{a,c}$\footnote{steuer@agnld.uni-potsdam.de}, Thilo Gross$^\mathrm{a}$, Joachim Selbig$^\mathrm{b,c}$, and Bernd Blasius$^\mathrm{a}$ \vspace{1cm} \\ 
31: {\small \em $^\mathrm{a}$University Potsdam, Institute for Physics, Nonlinear Dynamics Group,} \\
32: {\small \em                 Am Neuen Palais~10, 14469~Potsdam, Germany } \\
33: {\small \em $^\mathrm{b}$Max-Planck-Institute for Molecular Plant Physiology,} \\ 
34: {\small \em Am M\"uhlenberg 1, 14476 Potsdam-Golm, Germany }\\
35: {\small \em $^\mathrm{c}$University Potsdam, Institute for Biochemistry and Biology,} \\
36: {\small \em  Karl-Liebknecht-Strasse 24-25, Haus 20, 14476 Potsdam, Germany }
37: }
38: 
39: \title{Structural Kinetic Modeling of Metabolic Networks}
40: \date{\today}
41: 
42: 
43: \maketitle
44: 
45: \abstract{ \bf
46:  To develop and investigate detailed mathematical models of cellular metabolic 
47:  processes is one of the primary challenges in systems biology.
48:  However, despite considerable advance
49:  in the topological analysis of metabolic networks,
50:  explicit kinetic modeling based on differential equations is still 
51:  often severely hampered by inadequate knowledge of the 
52:  enzyme-kinetic rate laws and their associated parameter values.
53:  Here we propose a method that aims to give a detailed and quantitative account of
54:  the dynamical capabilities of metabolic systems, 
55:  without requiring 
56:  any explicit information about the particular functional form of the 
57:  rate equations.
58:  Our approach is based on constructing 
59:  a local linear model at each point in parameter space, 
60:  such that each element of the model
61:  is either directly experimentally accessible, 
62:  or amenable to a straightforward biochemical interpretation.
63:  This ensemble of local linear models, encompassing all possible explicit kinetic models,
64:  then allows for a systematic statistical exploration 
65:  of the comprehensive parameter space. 
66:  The method is applied to two paradigmatic examples: 
67:  The glycolytic pathway of yeast and a realistic-scale representation 
68:  of the photosynthetic Calvin cycle. 
69: }
70: 
71: 
72: 
73: \small
74: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
75: \section*{Introduction}
76: Cellular metabolism constitutes a complex dynamical system and gives 
77: rise to a wide variety of dynamical phenomena, including multistability and temporal oscillations.
78: To elucidate, understand and eventually predict the behavior of
79: metabolic systems 
80: represents one of the primary challenges in the postgenomic era~\cite{Palsson00,Kell_2004,FTKW04,WP04}.
81: To this end, substantial effort was dedicated in the recent years 
82: to develop and investigate detailed kinetic models of cellular metabolic processes~\cite{HS96,THT99}. \\  
83: Once a mathematical model is established, it can serve a multitude of purposes:
84: It can be regarded as a ``{\em virtual laboratory}'' that allows to build up a 
85: characteristic description of the system,
86: irrespective of experimental
87: restrictions, and gives insights into 
88: fundamental design principles of cellular functions, such as adaptability, 
89: robustness and optimality. 
90: Important questions often concern the existence and size of regions in the space of parameters
91: with qualitatively different behavior, 
92: such as multiple steady states (multistability) or autonomous oscillations~\cite{MWBB02,SSS04,AFS04}.
93: Likewise, mathematical models of cellular metabolism serve as a basis to  
94: investigate questions of major biotechnological importance, such as the 
95: effects of directed modifications of enzymatic or regulatory activities to
96: improve a desired property of the system~\cite{SAM04}. \\
97: However, while there has been a formidable progress in the structural (or topological) analysis
98: of metabolic systems~\cite{SFD00,FFN03}, 
99: and despite the long history of metabolic modeling,
100: dynamic models of cellular metabolism incorporating a 
101: realistic complexity are still scarce. \\
102: This is owed to the fact that the construction of such models 
103: encompasses a number of profound difficulties.
104: Most importantly, the construction of kinetic models relies on the precise knowledge of the
105: functional form of all involved enzymatic rate equations and their associated parameter values.
106: Furthermore, even if both is available from the literature, parameter values may (and usually do)
107: depend on many factors such as tissue type, or experimental and physiological conditions.
108: Likewise, most enzyme-kinetic rate laws have been determined {\em in vitro} and often there is only 
109: little guidance available whether a particular rate function is still appropriate {\em in vivo}. \\
110: In this work, we aim to overcome some of these difficulties and 
111: propose a bridge between structural modeling, which is based on the stoichiometry alone~\cite{SFD00,FFN03,Bai01}, 
112: and explicit kinetic models of cellular metabolism. 
113: In particular, we demonstrate that it is possible to acquire an exact, detailed 
114: and quantitative picture of the bifurcation structure of a given metabolic system, 
115: without explicitely referring to any particular set of differential equations.  \\
116: Our approach starts with the observation that in most
117: circumstances an explicit kinetic model is not necessary.
118: For example, to determine under which conditions 
119: a steady state looses its stability, only a local linear approximation of the 
120: system at this respective state is needed, i.e.  
121: we only need to know the eigenvalues of the associated Jacobian matrix. 
122: Note that by saying this, and unlike related approaches to qualitative modeling~\cite{GC05,Bai01},
123: we do not aim at an approximation of the system. 
124: The boundaries of an oscillatory region in parameter space that arise
125: out of a Hopf bifurcation are actually and exactly determined by 
126: the eigenvalues of the Jacobian. 
127: Likewise, other bifurcations, including bifurcations of higher codimension, can
128: be deduced from the spectrum of eigenvalues and give rise to specific dynamical behavior. \\
129: The basis of our approach thus consists of giving a parametric representation
130: of the Jacobian matrix of an arbitrary metabolic system at each possible point in parameter space,
131: such that each element is accessible even without explicit 
132: knowledge of the functional form of the rate equations.
133: Once this parametric representation of the Jacobian is obtained, 
134: it allows to give a detailed statistical account 
135: of the dynamical capabilities of a metabolic system, 
136: including the 
137: stability of steady states, 
138: the possibility of sustained oscillations, as well as the 
139: existence of quasiperiodic and chaotic regimes.
140: Moreover, the analysis is quantitative, i.e. it allows to deduce
141: specific biochemical conditions under which a certain dynamical behavior 
142: occurs and allows to assess the plausibility or robustness 
143: of experimentally observed behavior by
144: relating it to a quantifiable region in parameter space. 
145: 
146: 
147: 
148: \section*{Structural Kinetic Modeling}
149: The temporal behavior of an arbitrary  metabolic reaction network, consisting
150: of $m$ metabolites and $r$ reactions,  can be
151: described by a set of differential equations~\cite{HS96},  
152: \begin{equation} \label{eq:Metabolism}
153: \frac{\mathrm{d}\mathbf{S}(t)}{\mathrm{d}t} = \mathbf{N} \, \bm{\nu}(\mathbf{S},\mathbf{k}) 
154: \end{equation}
155: where $\mathbf{S}$ denotes the $m$-dimensional vector of biochemical reactants and
156: $\mathbf{N}$ the $m \times r$ stoichiometric matrix. %~\cite{HS96}. 
157: The $r$-dimensional vector of reaction rates $\bm{\nu}(\mathbf{S},\mathbf{k})$ consists 
158: of nonlinear (and often unknown) functions,
159: which depend on the substrate concentrations $\mathbf{S}$, 
160: as well as on a set of (often unknown) parameters $\mathbf{k}$. \\
161: In the following, we will not assume explicit knowledge of the functional
162: form of the rate equations, but instead aim at a parametric 
163: representation of the Jacobian of the system.
164: As the only mathematical assumption about the system, we require
165: the existence of a positive state $\mathbf{S}^0$ that fulfills the steady state 
166: condition $\mathbf{N} \bm{\nu}(\mathbf{S}^0,\mathbf{k})=\mathbf{0}$. 
167: Importantly, the state $\mathbf{S}^0$ is neither required to be unique, nor stable. \\
168: %
169: Using the definitions
170: \begin{equation} \label{eq:rates}
171: x_i(t) := \frac{S_i(t)}{S_i^0}, 
172: \quad 
173: \Lambda_{ij} := N_{ij}   \frac{\nu_j(\mathbf{S^0})}{S_i^0}
174: \,\,\, \mbox{and} \,\,\, \mu_j(\mathbf{x}) :=  \frac{\nu_j(\mathbf{x})}{\nu_j(\mathbf{S^0})} 
175: \end{equation}
176: and following the normalization procedure proposed in~\cite{Gross04,GF05},
177: the system can be straightforwardly rewritten in terms of new variables $\mathbf{x}(t)$
178: \begin{equation}
179: \frac{\mathrm{d}\mathbf{x}}{\mathrm{d}t} = \bm{\Lambda} \, \bm{\mu}(\mathbf{x}) 
180: \qquad \mbox{.}
181: \end{equation}
182: The corresponding Jacobian of the normalized system at the steady state $\mathbf{x}^0=\mathbf{1}$ is
183: \begin{equation} \label{eq:Jacobian}
184: \mathbf{J}_\mathbf{x} = \bm{\Lambda} \, 
185: \left. \frac{\partial \bm{\mu}(\mathbf{x})}{\partial \mathbf{x}} \right|_{\mathbf{x}^0=\mathbf{1}}
186: =: \bm{\Lambda} \, \bm{\theta}_\mathbf{x}^\mathbf{\bm \mu}
187: \end{equation}
188: As the new variables $\mathbf{x}$ are related to $\mathbf{S}$ by a simple
189: multiplicative constant, $\mathbf{J}_\mathbf{x}$ can be straightforwardly transformed 
190: back into the original Jacobian. \\
191: Any further discussion now rests crucially on the interpretation of the
192: terms in Eq.~(\ref{eq:Jacobian}). Once these coefficients are known, 
193: the Jacobian of the system can be evaluated. 
194: We begin with an analysis of the matrix $\bm{\Lambda}$.
195: Its elements $\Lambda_{ij}$ have the units of an inverse time and
196: consist of the elements of the stoichiometric matrix $\mathbf{N}$,
197: the vector of steady state concentrations $\mathbf{S}^0$, a
198: nd the steady state fluxes $\bm{\nu}(\mathbf{S}^0)$.
199: Provided a metabolic system is designated for mathematical modeling, 
200: we can safely assume that there exists some knowledge about
201: the relevant concentrations,  
202: i.e. for each metabolite, we can specify an interval 
203: $S_i^- \;  \leq  \;  S_i^0 \;  \leq \;  S_i^+$
204: which defines a physiologically feasible range of the respective concentration.
205: Furthermore, the steady state fluxes $\bm{\nu}(\mathbf{S}^0)$ are subject to the 
206: mass-balance constraint $\mathbf{N}\bm{\nu}(\mathbf{S}^0) = \mathbf{0}$, leaving only
207: $r-\mathrm{rank}(\mathbf{N})$ independent reaction rates~\cite{HS96}. 
208: Again, an interval $\nu_i^- \leq \nu_i^0 \leq \nu_i^+$ can be 
209: specified for all independent reaction rates, defining a physiologically admissible flux-space. \\
210: In the following, we denote $\mathbf{S}^0$ and $\bm{\nu}(\mathbf{S}^0)$, usually 
211: corresponding to an experimentally observed state of the system, 
212: as the {\em operating point} at which the Jacobian is to be evaluated. 
213: This information, together with the stoichiometric matrix $\mathbf{N}$, fully specifies 
214: the matrix $\bm{\Lambda}$. \\
215: %
216: %
217: The interpretation of the matrix $\bm{\theta}_\mathbf{x}^\mathbf{\bm \mu}$ 
218: in Eq.~(\ref{eq:Jacobian}) is slightly more subtle
219: since it involves the derivatives of the unknown functions $\bm{\mu}(\mathbf{x})$ 
220: with respect to the new normalized variables $\mathbf{x}$ at the point $\mathbf{x}^0 = \mathbf{1}$. 
221: Nevertheless, an interpretation of these parameters is possible and
222: does not rely on the explicit knowledge of the detailed functional form of the rate equations:
223: Each element $\theta_{x_i}^{\mu_j}$ of the matrix $\bm{\theta}_\mathbf{x}^\mathbf{\bm \mu}$ measures 
224: the normalized degree of saturation %, or likewise, the effective order 
225: of the reaction $\nu_j$ with respect to a substrate $S_i$ at the operating point $\mathbf{S^0}$. \\
226: In particular, the dependence of almost all biochemical rate laws $\nu_j(\mathbf{S})$ 
227: on a biochemical reactant $S_i$ 
228: can be written in the form $\nu_j(\mathbf{S},\mathbf{k})=k_\mathrm{v} S_i^n/f_m(\mathbf{S},\mathbf{k})$, where
229: $f_m(\mathbf{S},\mathbf{k})$ denotes a polynomial of order $m$ in $S_i$ with positive coefficients $\mathbf{k}$. 
230: All other reactants have been absorbed into $\mathbf{k}$~\cite{HS96}. 
231: After applying the transformation of Eq.~(\ref{eq:rates}), 
232: we obtain 
233: \begin{equation} \label{eq:GeneralRate}
234: \theta^{\mu_j}_{x_i} = 
235: \left. \frac{\partial\mu_j(\mathbf{x})}{\partial x_i} \right|_{\mathbf{x}^0=\mathbf{1}}  = n - \alpha \, m
236: \end{equation}
237: with $\alpha \in [0,1]$ denoting a free variable in the unit interval.
238: The limiting cases are {\em always} $\lim_{S^0_i\rightarrow 0} \alpha= 1$ and $\lim_{S^0_i\rightarrow \infty} \alpha= 0$.
239: To evaluate the matrix $\bm{\theta}_\mathbf{x}^\mathbf{\bm \mu}$  we thus 
240: restrict each saturation parameter  
241: to a well-defined interval, specified in the following way:
242: As for most biochemical rate laws $n=m=1$, the partial derivative usually takes a value 
243: between zero and unity, determining the degree of saturation of the respective reaction. 
244: In the case of cooperative behavior with a Hill coefficient $n = m \geq 1$, 
245: the normalized partial derivative lies in the interval $[0,n]$  and, analogously, 
246: in the interval $[0,-m]$ for inhibitory interaction with $n=0$ and $m \geq 1$. 
247: For examples and proof of Eq.~(\ref{eq:GeneralRate}) see {\em Materials and Methods}. \\
248: The matrices $\bm{\theta}_\mathbf{x}^\mathbf{\bm \mu}$ and $\bm{\Lambda}$, defined above,
249: now fully specify the Jacobian of the system.  
250: In the following, both quantities are treated as free parameters, 
251: defining the physiologically admissible {\em parameter space} of the system.  
252: Importantly, our representation of the Jacobian fulfills three essential conditions:
253: {\em i)} The reconstructed Jacobian represents the exact Jacobian at this
254: point in parameter space. There is no approximation involved.
255: {\em ii)} Each term in the Jacobian is either directly experimentally accessible, 
256: such as flux or concentration values, or has a well-defined biochemical 
257: interpretation, such as a normalized degree of saturation of a given reaction. 
258: {\em iii)} The Jacobian does not depend on any particular choice 
259: of specific rate functions. Rather, it encompasses {\em all} possible 
260: kinetic models of the system that are consistent with the considerations above.
261: In particular, any specific kinetic model, involving a specific choice of biochemical rate functions, 
262: can be mapped onto a particular point or region of the generalized parameter space. 
263: In this sense, the reconstructed Jacobian is exhaustive. 
264: 
265: 
266: 
267: \subsection*{An illustrative Example}
268: Prior to an application to more detailed biochemical models, we
269: exemplify our approach using a simple hypothetical pathway.
270: Suppose the reaction scheme depicted in Fig.~\ref{fig:BierModel},
271: consisting of  $m=2$ metabolites and $r=3$ reactions, 
272: is designated for mathematical modeling. 
273: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
274: \begin{figure}
275: \centering{\includegraphics[width=0.3\textwidth]{steuer_figure01.eps}}
276: \begin{displaymath}
277: \frac{\mathrm{d}}{\mathrm{d}t} 
278: \left[ \begin{array}{c} G \\ T  \end{array}    \right] =
279: \left[ \begin{array}{rrr} 1 & -1 & 0 \\ 0 & 2 & -1  \end{array}    \right]
280:  \left[ \begin{array}{l} \nu_1 \\ \nu_2(G , T) \\ \nu_3(T)  \end{array}    \right] 
281: \end{displaymath}
282: \caption{\label{fig:BierModel} \small
283: A simple hypothetical pathway, reminiscent of a minimal model of yeast glycolysis~\cite{BBW00}: 
284: One unit of glucose (G) is converted into two units of ATP (T),
285: with ATP exerting a positive feedback on its own production.
286: The lower panel depicts the corresponding set of differential equations with the as yet
287: unspecified rate functions $\nu_1=const, \nu_2(G,T)$ and $\nu_3(T).$
288: }
289: \end{figure}
290: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
291: The starting point of our analysis is then a particular
292: operating point, characterized by the metabolite
293: concentrations $\mathbf{S}^\mathbf{0}=(G^0,T^0)$ and 
294: flux values $\bm{\nu}^\mathbf{0}=(\nu_1^0,\nu_2^0,\nu_3^0)$. 
295: Within a reasonably realistic scenario, we can assume that  
296: the average concentrations of both metabolites have been determined experimentally.
297: Furthermore, an analysis of the stoichiometric matrix $\mathbf{N}$ 
298: reveals that there is only one independent steady state reaction rate $c$, 
299: with $\nu_1^0 = \nu_2^0 =c$ and $\nu_3^0=2c$.
300: Thus we only require knowledge
301: of the average overall flux through the pathway, specifying the value $c$. \\
302: This information already enables the construction of the matrix $\bm{\Lambda}$, 
303: which defines the (usually experimentally observed) operating point 
304: at which the system is to be evaluated.
305: \begin{equation}
306: \bm{\Lambda} = \left[ 
307: \begin{array}{ccc} 
308: c/G^0  & -c/G^0  & 0 \\
309:    0   & 2c/T^0  & -2c/T^0 \\
310:  \end{array} \right] 
311: \end{equation}
312: The only remaining parameters are now the elements of 
313: the matrix $\bm{\theta}_\mathbf{x}^\mathbf{\bm \mu}$.
314: Starting with the dependence of each reaction upon its substrate and
315: assuming conventional biochemical rate laws, we obtain
316: $\theta^{\mu_2}_\mathrm{G}\in [0,1]$, specifying the  degree of 
317: saturation of $\nu_2$ with respect to its substrate G.
318: Furthermore, $\theta^{\mu_3}_\mathrm{T}\in [0,1]$ specifies the degree of 
319: saturation of $\nu_3$ with respect to T.
320: Additionally, the known regulatory feedback of the metabolite T upon the reaction~$\nu_2$ is 
321: incorporated by $\theta^{\mu_2}_\mathrm{T}\in [0,n]$, 
322: where $n \geq 1$ denotes a positive integer (Hill coefficient). 
323: The matrix $\bm{\theta}_\mathbf{x}^\mathbf{\bm \mu}$ thus contains three  
324: nonzero values, each restricted to a well-defined interval
325: \begin{equation}
326: \bm{\theta}_\mathbf{x}^\mathbf{\bm \mu} = 
327: \left[
328: \begin{array}{cc}
329: 0 & 0 \\
330: \theta^{\mu_2}_\mathrm{G} & \theta^{\mu_2}_\mathrm{T} \\
331: 0         & \theta^{\mu_3}_\mathrm{T}  
332:  \end{array} \right] ~~.
333: \end{equation}
334: We emphasize that the three elements of $\bm{\theta}_\mathbf{x}^\mathbf{\bm \mu}$ 
335: represent {\em bona fide} parameters of the system,
336: specifying the Jacobian matrix   
337: no less unique and quantitative than 
338: a corresponding set of Michaelis constants, 
339: albeit without referring to the explicit functional form of any rate equation.  
340: Given the elements of $\bm{\theta}_\mathbf{x}^\mathbf{\bm \mu}$ 
341: as adjustable parameters, we 
342: have thus obtained a parameteric representation of the Jacobian matrix
343: which encompasses all possible kinetic models consistent with the experimentally
344: observed operating point.
345: In the remainder of this paper, 
346: we utilize our approach to evaluate the dynamical capabilities of two more
347: complex examples of metabolic system.
348: 
349: 
350: 
351: 
352: \section*{The Glycolytic Pathway}
353: Among the most classical and probably best studied example of a 
354: biochemical oscillator is the breakdown of sugar via glycolysis in yeast.
355: Damped and sustained glycolytic oscillations have been observed for several decades 
356: and have triggered the development of a large variety of
357: kinetic models, ranging from simple minimal models~\cite{BBW00} 
358: to more elaborate representations of the pathway~\cite{WPSS00,HDS01}. \\ 
359: In the following, we will address some of the characteristic questions that
360: led to the development of those earlier models, and show that these can
361: be readily answered using the concept of  structural kinetic modeling. 
362: %%BEGIN FIGURE 02 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%5
363: \begin{figure}
364: \centering{ \includegraphics[width=0.49\textwidth]{steuer_figure02.eps}}
365:  \caption{\label{fig:Glycolysis} \small
366: A medium-complexity representation of the yeast glycolytic pathway,
367: adapted from earlier kinetic models~\cite{WPSS00}.
368: The system consists of $8$ metabolites and $8$ reactions.
369: The crucial regulatory step is the phosphofructokinase (PFK), here 
370: combined with the hexokinase (HK) reaction into the reaction rate $\nu_1$. 
371: Though PFK is known to have several effectors, we only consider the inhibition by
372: its substrate ATP, again following earlier kinetic models~\cite{WPSS00}. 
373: For metabolite abbreviations and the observed metabolite concentration  
374: see the Supplementary Information accompanying this article.
375: }
376: \end{figure}
377: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
378: Given a schematic representation of the pathway, as depicted in Fig.~\ref{fig:Glycolysis},
379: the first and foremost problem is to establish whether the proposed
380: reaction mechanism indeed facilitates  
381: sustained oscillations at the experimentally observed operating point. 
382: And, if yes, what are the specific kinetic conditions and requirements 
383: under which sustained oscillations can be expected.   \\
384: We start out by constructing the matrix $\bm{\Lambda}$
385: using the experimentally observed state $\mathbf{S}^0$ and $\bm{\nu}^0$,
386: identified here with the average concentration and flux values reported in~\cite{WPSS00,HDS01}. 
387: Furthermore, the matrix of saturation coefficients $\bm{\theta}_\mathbf{x}^\mathbf{\bm \mu}$
388: has to be specified.
389: For simplicity, we assume that all reactions are 
390: irreversible and depend on their respective substrates only, resulting
391: in $13$ free parameters.
392: Based on our discussion of conventional biochemical rate laws above, 
393: the saturation coefficients are restricted to the unit interval $\theta_{S}^{\mu} \in [0,1]$.  \\
394: For the dependence of the PFK-HK reaction on ATP, we follow a previously proposed
395: kinetic model~\cite{WPSS00} and assume a linear activation
396: due to its effect as a substrate and a saturable inhibition involving a 
397: positive Hill coefficient $n \geq 1$. 
398: The corresponding parameter is 
399: thus $\theta^{\mu_1}_\mathrm{ATP}=1-\xi$, with $\xi \in [0,n]$.
400: No further assumptions about the detailed functional form of 
401: any of the rate equations are necessary. 
402: For an explicit representation of both matrices $\bm{\Lambda}$ and 
403: $\bm{\theta}_\mathbf{x}^\mathbf{\bm \mu}$ see the Supplementary Information. \\
404: To investigate the possibility of sustained oscillation, we 
405: begin with the most simplest scenario and set $\theta_{S}^{\mu}=1$ for all reactions, 
406: corresponding to bilinear mass-action kinetics. 
407: Note, however, that the inhibition term is still assumed to be an unspecified nonlinear saturable function. 
408: Fig.~\ref{fig:LargestEV} shows the largest eigenvalue of the resulting Jacobian at the 
409: experimentally observed operating point as a function of the feedback strength $\xi$.
410: Indeed for sufficient inhibition, the spectrum of eigenvalues passes 
411: through a Hopf bifurcation and the system facilitates sustained oscillations.
412: %%% BEGIN FIGURE 03 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
413: \begin{figure}
414: \centering{\includegraphics[width=0.49\textwidth]{steuer_figure03_up.eps}}
415: \centering{
416: \resizebox{0.5\textwidth}{!}{
417:            \includegraphics[width=0.12\textwidth]{steuer_figure03a.eps}
418:            \includegraphics[width=0.12\textwidth]{steuer_figure03b.eps}
419:            \includegraphics[width=0.12\textwidth]{steuer_figure03c.eps}
420:            \includegraphics[width=0.12\textwidth]{steuer_figure03d.eps}
421: }}
422:  \caption{\label{fig:LargestEV} \small
423: {\em Upper plot:} The eigenvalue with the largest real part 
424: as a function of the inhibitory feedback strength $\xi$ of ATP on the combined PFK-HK reaction~$\nu_1$.
425: All other saturation parameter are $\theta^\mu_x=1$. 
426: Shown is the real part $\lambda_R^\mathrm{max}$ (solid line) together 
427: with the imaginary part $\lambda^\mathrm{max}_I$ (dashed line). 
428: At the Hopf bifurcation a complex conjugate pair of eigenvalues 
429: $\lambda^\mathrm{max}=\lambda_R^\mathrm{max} \pm i \lambda^\mathrm{max}_I$
430: crosses the imaginary axis. 
431: {\em Lower plots:} 
432: A varying feedback strength $\xi$ allows for $4$ different dynamical regimes.
433: Shown are time courses of FBP using an an explicit kinetic model 
434: at the points (a,b,c,d) indicated above.  
435: {\em a)} Slow relaxation to the stable steady state.  
436: {\em b)} Optimal response to perturbations, as determined by the minimal 
437: largest eigenvalue $\lambda_R^\mathrm{max}$.
438: {\em c)} Oscillatory return to the stable steady state.
439: {\em d)} Sustained oscillations.
440: All different regimes can be deduced solely from the Jacobian 
441: and are only exemplified using the explicit kinetic model.
442: For rate equations and kinetic parameters see Supplementary Information.
443: }
444: \end{figure}
445: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
446: Importantly, for a Hopf bifurcation to occur at the observed operating point, a Hill
447: coefficient $n \geq 2$ is needed, irrespective of the detailed functional form of the rate equation. \\
448: We have to highlight one fundamental aspect of our analysis:
449: Given our parametric representation of the Jacobian, the impact of the inhibition is
450: decoupled from the steady state concentrations and 
451: flux values the system adopts (the latter being solely determined by the matrix $\bm{\Lambda}$).
452: Thus, with Fig.~\ref{fig:LargestEV}, we specifically ask whether the assumed
453: inhibition is indeed a necessary condition for the 
454: observation of oscillations {\em at the experimentally observed operating point}.
455: In contrast to this, using a conventional kinetic model and reducing the influence of the regulation, i.e. 
456: by increasing the corresponding Michaelis constant, would concomitantly result in 
457: altered steady state concentrations -- thus not straightforwardly contributing to this question. \\  
458: Furthermore, as glycolytic oscillations have no obvious physiological role and are only
459: observed under rather specific experimental conditions, 
460: some questions concerning their possible functional significance have been raised.
461: One assertion is that the observed oscillations might 
462: only be an unavoidable side effect of the regulatory interactions, optimized for other purposes~\cite{HS96}.
463: Indeed, as shown in Fig.~\ref{fig:LargestEV}, a varying feedback strength $\xi$ allows for
464: different dynamical regimes. In particular, an intermediate value
465: speeds up the response time with respect to perturbations, as 
466: also frequently observed in explicit models of cellular regulation~\cite{REA02}. 
467: 
468: 
469: \subsection*{Statistical Analysis of the Parameter Space}
470: Going beyond the case of bilinear kinetics, 
471: we now evaluate the properties of Jacobian at the most general level.
472: All saturation coefficients $\theta^\mu_S \in (0,1]$ are allowed 
473: to take arbitrary values in the unit interval, encompassing all possible explicit kinetic models 
474: of the pathway shown in Fig.~\ref{fig:Glycolysis}. 
475: The steady state concentrations and flux values are again restricted 
476: to the experimentally observed operating point. 
477: To assess the robustness of the system at this operating point,
478: the saturation coefficients $\theta^\mu_S \in (0,1]$ are 
479: repeatedly sampled from a uniform distribution. 
480: For each random realization the Jacobian is evaluated and 
481: the largest real part $\lambda^\mathrm{max}_R$ of its eigenvalues is recorded. 
482: Fig.~\ref{fig:EV_histogram} shows the histogram of the largest real part within
483: the spectrum of eigenvalues, with $\lambda^\mathrm{max}_R>0$ implying instability of the operating point.
484: In the absence of the inhibitory feedback $\xi=0$ the 
485: operating point is likely to be unstable, 
486: i.e. most realizations result in a spectrum of eigenvalues with at least one positive real part. \\
487: Two ways to circumvent this inherent instability are conceivable: 
488: First we can ask about the dependence on particular ~ reactions, that is,
489: whether the saturation (or non-saturation) of a specific reaction contributes to an 
490: increased stability of the system.
491: To this end, the correlation coefficient between $\lambda^\mathrm{max}_R$, reflecting the stability
492: of the system, and the saturation parameters $\theta^\mu_S$ was estimated. 
493: Indeed, several parameters $\theta^\mu_S$ show a strong correlation with $\lambda^\mathrm{max}_R$, 
494: indicating that their value essentially determines the stability of the system (for data see Supplementary Information).  
495: Fig.~\ref{fig:EV_histogram}a depicts the distribution of $\lambda^\mathrm{max}_R$ under the assumption that 
496: these reactions are restricted to weak saturation. 
497: In this case, the resulting distribution is shifted towards negative values, 
498: corresponding to an increased probability of the system to operate at a stable steady state. \\
499: The second option to ensure stability of the system arises from the negative feedback of ATP
500: upon the combined PFK-HK reaction. 
501: %%% BEGIN FIGURE 04 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
502: \begin{figure}
503: \centering{\includegraphics[width=0.234\textwidth]{steuer_figure04a.eps}
504:            \includegraphics[width=0.230\textwidth]{steuer_figure04b.eps}
505: }
506:  \caption{\label{fig:EV_histogram} \small
507: The distribution of the largest real part $\lambda^\mathrm{max}_R$ within the spectrum of eigenvalues 
508: for $10^5$ realizations of the Jacobian matrix at the operating point. 
509: For each realization, the $12$ saturation parameters $\theta^\mu_S$ were sampled randomly
510: from a uniform distribution in the unit interval. 
511: A value $\lambda^\mathrm{max}_R>0$ implies instability.
512: In the absence of the regulatory feedback ($\xi=0$, blue solid line), 
513: the observed operating point of the system
514: is likely to be unstable. 
515: {\em Left:} Influence of specific reaction rates. 
516: The saturation parameters showing the largest impact on $\lambda^\mathrm{max}_R$ 
517: are restricted to weak saturation: 
518: $\theta^{\mu_8}_\mathrm{ATP}=0.9$, $\theta^{\mu_6}_\mathrm{Pyr}=0.9$, and $\theta^{\mu_7}_\mathrm{TP}=0.9$.
519: The probability of finding $\lambda^\mathrm{max}_R<0$ is markedly increased (red dashed line). 
520: {\em Right:} For a nonzero feedback strength $\xi \approx 0.92$ the distribution of $\lambda^\mathrm{max}_R$
521: is shifted towards negative values, i.e. most realizations of the Jacobian give rise to a stable
522: steady state (red dashed line).
523: }
524: \end{figure}
525: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
526: Fig. \ref{fig:EV_histogram}b shows the distribution of the 
527: largest real part $\lambda^\mathrm{max}_R$ 
528: of the eigenvalues for a nonzero feedback strength~$\xi>0$.
529: Again, the distribution is markedly shifted towards negative values, increasing the
530: probability of a stable steady state. \\
531: In more detail, Fig.~\ref{fig:EV_Stability} depicts the distribution of $\lambda^\mathrm{max}_R$ 
532: as a function of the feedback strength $\xi$.
533: %%% BEGIN FIGURE 05 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%5
534: \begin{figure}
535: \centering{\includegraphics[width=0.23\textwidth]{steuer_figure05a.eps}
536:            \includegraphics[width=0.23\textwidth]{steuer_figure05b.eps}
537: }
538:  \caption{\label{fig:EV_Stability} \small
539: The distribution of the largest real part $\lambda^\mathrm{max}_R$ 
540: of the eigenvalues as a function of the feedback strength $\xi$. 
541: All other saturation parameters are sampled randomly from a uniform distribution.
542: {\em Left:} Color-coded visualization of the resulting distribution of~$\lambda^\mathrm{max}_R$ (red: large, blue: small). 
543: {\em Right:} The relative fraction of models with $\lambda^\mathrm{max}_R>0$, implying the 
544: instability of the observed operating point. 
545: The solid line (SN) denotes the case of a single positive real part $\lambda^\mathrm{max}_R$ within the spectrum of eigenvalues,
546: indicating a saddle-node bifurcation.
547: The dashed line (HO) denotes the case of a pair of complex conjugate eigenvalues with positive real parts.
548: Note that, strictly speaking, this does not necessarily imply sustained oscillations.
549: However, it indicates the existence of a nearby Hopf bifurcation, thus constituting prima facie evidence
550: for oscillatory behavior.
551: }
552: \end{figure}
553: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
554: As can be observed, in the absence of the regulatory feedback the system is prone to instability,
555: i.e. it is not possible (or rather unlikely) for the observed operating point to exist as a stable
556: steady state. 
557: Subsequently, as the feedback strength is increased, the probability of obtaining a stable steady state increases.
558: For an intermediate value $\xi=1$ the system is fully stable: Any realization of
559: the Jacobian will result in a stable steady state, independent of the detailed functional form of
560: the rate equations or their associated parameters.    
561: However, as the feedback is increased further, the operating point again looses its stability.
562: This time the instability arises out of a Hopf bifurcation, indicating the presence
563: of sustained oscillations. \\
564: Based on these findings, we can summarize some essential properties 
565: of the pathway depicted in Fig.~\ref{fig:Glycolysis}: 
566: Given the experimentally observed metabolite concentrations and flux values, 
567: our results show that
568: in the absence of the assumed regulatory interaction 
569: it would not be possible (or, at least highly unlikely) to observe either 
570: sustained oscillations or a stable steady state.
571: However, for sufficiently large inhibitory feedback, the system will inevitably
572: exhibit sustained oscillations. 
573: Furthermore, as the feedback strength $\xi \in [0,n]$ is bounded by the  
574: Hill coefficient $n$ of the (unspecified) rate equation, $n \geq 2$ is required for the existence of sustained oscillations. \\
575: We emphasize again that these results do not rely on any explicit kinetic model 
576: of the system. 
577: As demonstrated, our method allows to derive the likeliness or plausibility of 
578: the experimentally observed oscillations, as well as the specific 
579: kinetic requirements for oscillations to occur, 
580: without referring to the detailed functional form of the rate equations. 
581: 
582: 
583: 
584: \section*{The photosynthetic Calvin Cycle}
585: The CO$_2$ assimilating Calvin Cycle, taking place in the chloroplast stroma of plants, 
586: is a primary source of carbon for all organisms and of central  
587: importance for many biotechnological applications. 
588: However, even when restricting an analysis to the core pathway shown in Fig.~\ref{fig:CalvinCycle},
589: the construction of a detailed kinetic model already entails considerable challenges with respect 
590: to the required rate equations and kinetic parameters~\cite{PRP88,POL01}. \\
591: In the following, we thus use the scheme depicted in Fig.~\ref{fig:CalvinCycle} to
592: demonstrate the applicability of our approach to a system of a reasonable complexity.  
593: In particular, we seek to describe a general strategy to extract information about the dynamical capabilities of the system, without
594: referring to an explicit set of differential equations.
595: Our agenda focuses on 
596: {\em (i)} the stability and robustness of the experimentally observed operating point,
597: {\em (ii)} the relative impact or importance of each reaction upon the dynamical properties of the system,
598: {\em (iii)} the existence and quantification of different dynamical regimes, 
599: such as oscillations and multistability, and 
600: {\em (iv)} the possibility of complex or chaotic temporal behavior. \\
601: Starting point is again a particular observed state, characterized by the 
602: vector of metabolite concentrations $\mathbf{S}^\mathbf{0}$ and flux values~$\bm{\nu}^\mathbf{0}$. 
603: %%% BEGIN FIGURE 06 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%5
604: \begin{figure}
605: \centering{\includegraphics[width=0.45\textwidth]{steuer_figure06.eps}}
606:  \caption{\label{fig:CalvinCycle} \small
607: The photosynthetic Calvin Cycle, adapted from earlier kinetic models~\cite{PRP88,POL01}.
608: The systems consists of $18$ metabolites, subject to two conservation relations,
609: and $20$ reactions, including three export reactions, starch synthesis and regeneration of ATP.
610: The rank of the stoichiometric matrix is $\mathrm{rank}(\mathbf{N})=16$, leaving $4$ independent
611: steady state reaction rates.
612: Throughout this section, the steady state concentrations and flux values are as 
613: reported by Petterson and Ryde-Petterson~\cite{PRP88},
614: describing the pathway under conditions of light and CO$_2$ saturation.
615: For metabolite abbreviations see the Supplementary Information.}
616: \end{figure}
617: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
618: Although additional knowledge on the reactions is often available,
619: for the moment we assume that all reactions depend
620: only on their substrates and products, with parameters~$\theta^{\mu}_S \in (0,1]$ and 
621: $\theta^{\mu}_P \in [0,-1]$, respectively.   
622: This information, embedded within the matrices~$\bm{\Lambda}$ and $\bm{\theta}_\mathbf{S}^\mathbf{\bm \mu}$,
623: constitutes the structural kinetic model of the Calvin cycle at the observed operating point. \\ 
624: As a first approximation, we commence with global saturation parameters, $\theta^{\mu}_S$ and $\theta^{\mu}_P$,
625: set equal for all reactions.
626: Though clearly oversimplified, the resulting bifurcation diagram, 
627: depicted in Fig.~\ref{fig:CalvinBifurcation},
628: already reveals some fundamental dynamical properties of the system:
629: First, the observed operating point is indeed a 
630: stable steady state for most parameters $\theta^{\mu}_S$ and $\theta^{\mu}_P$. 
631: Interestingly, however, in the absence of product inhibition $\theta^{\mu}_P=0$,  a steady state is no longer feasible. 
632: In particular, for pure irreversible mass-action kinetics ($\theta^{\mu}_S=1$, $\theta^{\mu}_P=0$), 
633: corresponding to a non-enzymatic chemical system, the pathway could not operate at the observed
634: steady state. 
635: Second, for low product saturation ($\theta^{\mu}_P$ close to zero), a Hopf bifurcation occurs. 
636: While this does not necessarily imply that this region within parameter space is 
637: actually accessible under normal physiological conditions,
638: it shows the dynamical capability of the system to generate sustained oscillations, i.e. there
639: exists a region in parameter space that allows for oscillatory behavior. 
640: Additionally, for low values of the substrate saturation $\theta^{\mu}_S$, a saddle-node bifurcation occurs.
641: %%%% BEGIN FIGURE 07 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%5
642: \begin{figure}
643: \centering{\includegraphics[width=0.3\textwidth]{steuer_figure07.eps}}
644:  \caption{\label{fig:CalvinBifurcation} \small
645: The bifurcation diagram of the Calvin cycle at
646: the observed operating point with respect to the two global saturation 
647: parameters~$\theta^{\mu}_S \in (0,1]$ and $\theta^{\mu}_P \in [0,-1]$.   
648: }
649: \end{figure}
650: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
651: This shows that the observed steady state will eventually loose its stability, i.e. there are 
652: conditions under which the observed steady state is no longer stable. 
653: Noteworthy, both dynamical features have been 
654: observed for the Calvin cycle: 
655: Photosynthetic oscillations are known for many decades
656: and have been subject to extensive experimental and numerical studies~\cite{RP91}. 
657: Furthermore, multistability was recently found in a detailed kinetic model of the 
658: Calvin cycle and verified {\em in vivo}~\cite{POL01}. \\
659: %
660: To proceed with a systematic analysis, the next step is to drop the assumption of
661: global saturation parameters. 
662: All individual parameters $\theta^{\mu}_S \in (0,1]$ are now allowed
663: to take arbitrary values in the unit interval, reflecting the full spectrum of
664: possible dynamical capabilities of the metabolic system.    
665: For simplicity, all reactions are still restricted to weak saturation by their products~$\theta^{\mu}_P = 1/3$. 
666: Of foremost interest is again the robustness of the experimentally observed operating point:
667: Evaluating an ensemble of $5 \cdot 10^5$ random realizations of the Jacobian at this operating 
668: point, the system gives rise to a stable steady state in $\approx 94.3\%$ of all cases (see Supplementary
669: Information for convergence and dependence on ensemble size).
670: Thus the stability of the observed operating point is indeed generic and does not
671: rely on a specific choice of the kinetic parameters. \\
672: As for the remaining $\approx 5.7\%$ of models, corresponding to the case 
673: where the observed operating point is instable, about $5.1\%$ give
674: rise to a single positive eigenvalue. Only $\approx 0.6\%$ correspond to a more
675: complex situation, with two or more real parts larger than zero. 
676: The latter case, though only restricted to a small region within parameter space,
677: holds profound implications for the possible dynamics of the system. 
678: As a further step within our approach, 
679: the existence of certain bifurcations of higher codimension allows to
680: predict the possibility of specific dynamics (see {\em Materials and Methods}).  
681: Fig.~\ref{fig:Calvin_Bifdiagram} shows a bifurcation diagram of
682: the Calvin cycle within a particular region of parameter space where such bifurcations occur.  
683: %%%% BEGIN FIGURE 08 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%5
684: \begin{figure}
685: \centering{\includegraphics[width=0.5\textwidth]{figure_small08.eps}}
686: %\centering{
687: %\begin{minipage}{0.3\textwidth}
688: %\includegraphics[width=1.0\textwidth]{steuer_figure08_left.eps}
689: %\end{minipage} \hspace{0.05cm}
690: %\begin{minipage}{0.155\textwidth}
691: %\includegraphics[width=0.99\textwidth]{steuer_figure08A.eps} 
692: %\includegraphics[width=0.99\textwidth]{steuer_figure08B.eps}
693: %\end{minipage}
694: %}
695: \caption{\label{fig:Calvin_Bifdiagram} \small
696: The bifurcation diagram of the Calvin cycle as a function of the saturation of the rubisco
697: reaction with respect to RuBP and the saturation of the Aldolase reaction with respect to GAP,
698: while all other saturation parameters are fixed to specific values. Bifurcation lines are depicted in blue, 
699: numerals indicate the number the positive real parts within the spectrum of eigenvalues. 
700: {\em Inset A:} The system gives rise to a  Gavrilov-Guckenheimer (GG) bifurcation, formed by the interaction
701: of a Hopf (HO) and a saddle-node (SN) bifurcation.
702: {\em Inset B:} The interaction of two Hopf (HO) bifurcations gives rise to a double Hopf (DH) bifurcation.
703: }
704: \end{figure}
705: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
706: Here, the system gives rise to a Gavrilov-Guckenheimer (GG) bifurcation, implying
707: the existence of quasiperiodic dynamics and making the existence of chaotic dynamics likely. 
708: In close vicinity of the GG bifurcation, we also find a double Hopf (DH) bifurcation, 
709: formed by the interaction of two codimension-1 Hopf bifurcations.  
710: The generic existence of a chaotic parameter region close to the DH bifurcation can
711: be proven~\cite{Kuz95,GEF05}.  \\
712: Thus our results demonstrate the possibility of quasiperiodic 
713: and chaotic dynamics for the model of the photosynthetic Calvin cycle shown in Fig.~\ref{fig:CalvinCycle}, 
714: without relying on any particular assumptions about the functional form of the kinetic rate equations. 
715: Furthermore, being a quantitative method, we can assert that complex dynamics at the 
716: operating point are confined to a rather small region in parameter space and that the experimentally
717: observed steady state is generically stable. 
718: 
719: 
720: 
721: \section*{Discussion and Conclusions}
722: We have presented a systematic approach to explore and quantify 
723: the dynamic capabilities of a metabolic system, 
724: without requiring to specify the detailed functional form of 
725: any of the involved rate equations.
726: Starting with a parametric representation of the Jacobian matrix, 
727: constructed in such a way that each element is either directly experimentally
728: accessible or amenable to a clear biochemical interpretation,
729: we look for characteristic bifurcations that give insight into the possible dynamics of the system.
730: Our method then builds upon the construction of a large ensemble of models, 
731: encompassing {\em all possible} explicit kinetic models,  
732: to statistically explore and quantify the parameter region associated 
733: with a specific dynamical behavior. \\
734: One of the primary advantages is that all results relate to
735: a particular experimentally observed operating point of the system.
736: In this respect, the method contrasts with the trivial alternative 
737: of drawing all (known) nonzero elements of the Jacobian from a random distribution.   
738: While the latter would likewise allow to indicate e.g. the possibility of oscillatory behavior,
739: it fails to actually quantify the associated parameter region at a particular observed state. 
740: Only by means of our parametric representation of the system, we are in the position to
741: identify crucial reaction steps that predominantly contribute to
742: the stability, and thus robustness, of an experimentally observed state and can give 
743: explicit biochemical conditions
744: for which a specific dynamical behavior can be expected. 
745: Furthermore, by taking bifurcations of higher codimension into account, we go
746: beyond the usually considered case and are able to predict the 
747: possibility of complex or chaotic dynamics - often 
748: a nontrivial task, even if an explicit kinetic model is available. \\
749: We emphasize that our approach is not restricted to an analysis of the
750: bifurcations and stability properties of metabolic systems.
751: Once the parametric representation of the Jacobian is obtained, 
752: it can serve a multitude of purposes.
753: The Jacobian holds a wealth of information, including the systems response to (small) perturbation, 
754: the hierarchy of characteristic timescales~(Modal Analysis)~\cite{HS96},
755: as well as the possibility to deduce the flux and concentration control coefficients, 
756: defined in the realm of Metabolic Control Analysis~\cite{HS96}.   
757: Along similar lines, it is thus possible to explore the influence 
758: of particular reactions and their associated saturation parameters
759: upon more general features of the system.
760: In this respect, structural kinetic modeling also serves as a prequel to
761: explicit mathematical modeling, aiming to identify 
762: crucial reaction steps and parameters in best time. 
763: 
764: 
765: 
766: 
767: 
768: 
769: 
770: 
771: 
772: \appendix
773: \section*{\sc Materials and Methods}
774: {\small
775: \subsection*{\sc The Interpretation of the Saturation Matrix}
776: Our approach relies crucially on the interpretation of the elements of 
777: the matrix $\bm{\theta}_\mathbf{x}^\mathbf{\bm \mu}$.
778: As a simple example, consider a single bilinear 
779: reaction rate of the form $\nu(S_1,S_2)=v_\mathrm{max}S_1 S_2$.
780: Then, according to Eq.~(\ref{eq:rates}), the normalized rate is $\mu(x_1,x_2)=x_1 x_2$, thus
781: \begin{equation}
782: \theta_{x_i}^{\mu} = 
783: \left. \frac{\partial\mu}{\partial x_1} \right|_{\mathbf{x}^0=\mathbf{1}} =
784: \left. \frac{\partial\mu}{\partial x_2} \right|_{\mathbf{x}^0=\mathbf{1}} = 1 
785: ~~.
786: \end{equation} 
787: In the case of Michaelis-Menten kinetics $\nu(S)=v_\mathrm{max}S/(K_\mathrm{M}+S)$, 
788: depending on a single substrate $S$, we obtain 
789: \begin{equation}
790: \mu(x) = x \, \frac{K_M+S^0}{K_M + x S^0} \quad \Rightarrow \quad
791: \theta_{x}^{\mu} 
792: = \frac{1}{1+S^0/K_\mathrm{M}} \in [0,1]
793: \end{equation}
794: Clearly, the partial derivative $\theta_{x}^{\mu} \in [0,1]$ measures the degree of saturation or, likewise, 
795: the effective order of the reaction at the steady state~$S^0$.
796: The limiting cases are $\lim_{S^0\rightarrow 0}\theta_{x}^{\mu}=1$ (linear regime)
797: and $\lim_{S^0\rightarrow\infty}\theta_{x}^{\mu}=0$ (full saturation). 
798: This implies that the saturation parameter indeed covers the full interval, 
799: which holds likewise for the general case of Eq.~(\ref{eq:GeneralRate}).
800: For additional instances of specific rate functions, as well as a proof of Eq.~(\ref{eq:GeneralRate}), 
801: see the Supplementary Information. \\
802: Note that, except for the change in variables, the 
803: saturation parameters $\bm{\theta}_\mathbf{x}^\mathbf{\bm \mu}$  
804: are reminiscent of the scaled elasticity coefficients, as defined in the 
805: realm of Metabolic Control Analysis~\cite{HS96}.
806: However, for our reasoning to hold, the analysis is restricted to unidirectional reactions, i.e.
807: in the case of reversible reactions, forward and backward terms have to be treated separately. 
808: As the denominator is usually preserved for both terms, this does 
809: not give rise to additional free saturation parameters. \\
810: %
811: Another close analogy to the saturation parameters is found 
812: within the power-law approximation~\cite{HS96},
813: where each enzyme kinetic rate law is replaced by a 
814: function of the form $\nu_j(\mathbf{S})=\alpha_j \prod_i S_i^{g_{ij}}$.
815: In fact, the power-law formalism can be regarded as the simplest 
816: possible way to specify explicit nonlinear functions that are consistent with a given Jacobian.
817: Applying the transformation of Eq.~(\ref{eq:rates}), we obtain $\mu_j(\mathbf{x}) = \prod_i x_i^{g_{ij}}$,
818: thus $\theta_{x_i}^{\mu_j} = g_{ij}$.
819: However, beyond the properties of the Jacobian itself, only little confidence can
820: be placed in an actual numerical integration of these functions~\cite{HS96}. 
821: Generally, it is possible to specify several classes of explicit functions that, by construction,
822: result in a given Jacobian, but have no, or only little, biochemical justification otherwise.   
823: Consequently, within our approach, we opt for utilizing the properties of the parametric representation
824: of the Jacobian directly, instead of going the loop way via auxiliary ad-hoc functions.
825: 
826: 
827: \subsection*{\sc Dynamics and Bifurcations}
828: One of the foundations of our approach is the fact that 
829: knowledge of the Jacobian matrix alone is sufficient to deduce certain
830: characteristic bifurcations of a metabolic system. 
831: In general, the stability of a steady state is lost either in a {\em Hopf bifurcation} (HO) 
832: or in a bifurcation of {\em saddle-node} (SN) type, both of codimension-1. 
833: %In the bifurcations of saddle-node type a single zero eigenvalue of the Jacobian 
834: %appears as the number or stability of steady states changes. 
835: %Bifurcations of the SN type often indicate the presence  of multiple steady states. \\
836: %%are often associated with the existence of multiple
837: %%steady states. \\% (bi- or multistability). \\
838: %As the only other local codimension-1 bifurcation, 
839: %a Hopf bifurcation occurs as a complex conjugate pair of eigenvalues 
840: %crosses the imaginary axis towards positive real parts. 
841: %This gives rise to (at least transient) oscillations as the stability of the steady state is lost.
842: %Note that it is not possible to distinguish between a sub- and supercritical Hopf bifurcation
843: %solely on basis of the Jacobian. \\
844: Of particular interest to reveal insights about the dynamical behavior of systems are
845: also bifurcations of higher codimension, such as the 
846: {\em Takens-Bogdanov} (TB), the {\em Gavrilov-Guckenheimer} (GG) 
847: and the {\em double Hopf} (DH) bifurcation~\cite{Gross04,Kuz95}. %, all of codimension-2. \\
848: Each of these local bifurcations of codimension-2 arises out of
849: an interaction of two codimension-1 bifurcations and 
850: has important implications for the possible dynamical behavior.  
851: %The codimension-1 bifurcation surfaces of Hopf and Saddle-node bifurcation type form 
852: %hypersurfaces in the parameter space. 
853: %Where these surfaces coincide codimension-2 bifurcations like Takens-Bogdanov (TB), 
854: %Gavrilov-Guckenheimer (GG) and double Hopf bifurcations are formed. 
855: %Although these bifurcations can not be observed directly in experiments 
856: %there presence has many important implications. 
857: For instance the TB bifurcation %, corresponding to a situation where a SN and a Hopf bifurcation coincide,
858: indicates the presence of a homoclinic bifurcation 
859: and therefore the possibility of spiking or bursting behavior. 
860: The presence of a Gavrilov-Guckenheimer bifurcation shows that complex (quasiperiodic or chaotic) 
861: dynamics exist generically in a certain parameter space. 
862: In the same way the double Hopf bifurcation indicates 
863: the generic existence of a chaotic parameter region. 
864: For details see~\cite{Gross04,Kuz95}
865: and the {\em Supplementary Information}.
866: 
867: }
868: {\footnotesize
869: \begin{thebibliography}{10}
870: 
871: \bibitem{Palsson00}
872: Palsson, B.~O. (2000)
873: %\newblock The challenges of in silico biology.
874: \newblock {\em Nat. Biotechnol.}, {\bf 18}, 1147--1150.
875: 
876: \bibitem{Kell_2004}
877: Kell, D.~B.  (2004)
878: %\newblock Metabolomics and systems biology: Making sense of the soup.
879: \newblock {\em Curr. Opin. Microbiol.} {\bf 7}, 296--307.
880: 
881: \bibitem{FTKW04}
882: Fernie, A.~R., Trethewey, R.~N., Krotzky, A.~J. \& Willmitzer, L. (2004)
883: %\newblock Metabolite profiling: from diagnostics to systems biology.
884: \newblock {\em Nat. Rev. Mol. Cell Biol.} {\bf 5}, 1--7.
885: 
886: \bibitem{WP04}
887: Westerhoff, H.~V. \& Palsson, B.~O. (2002)
888: %\newblock The evolution of molecular biology into systems biology.
889: \newblock {\em Nat. Biotechnol.} {\bf 22} 1249--1252.
890: 
891: \bibitem{HS96}
892: Heinrich, R. \& Schuster, S. (1996)
893: \newblock {\em The Regulation of Cellular Systems}.
894: \newblock (Chapman \& Hall, New York).
895: 
896: \bibitem{THT99}
897: Hashimoto, K., Tomita, M., Takahashi, K., Shimizu, T.~S., Matsuzaki, Y., Miyoshi, F.,
898: Saito, K., Tanida, S., Yugi, K., Venter, J.~C. \& Hutchison III, C.~A. (1999)
899: %\newblock E-cell: Software enviroment for whole-cell simulation.
900: \newblock {\em Bioinformatics} {\bf 15}, 72--84.
901: 
902: 
903: \bibitem{MWBB02}
904: Morohashi, M., Winn, A. E., Borisuk, M. T., Bolouri, H., Doyle, J., \& Kitano, H. (2002)
905: %\newblock Robustness as a measure of plausibility in models of biochemical
906: %  networks.
907: \newblock {\em J. Theor. Biol.} {\bf 216}, 19--30.
908: 
909: \bibitem{SSS04}
910: Stelling, J., Sauer, U., Szallasi, Z., {Doyle III}, F. J. \& Doyle J. (2004)
911: %\newblock Robustness of cellular function.
912: \newblock {\em Cell}, {\bf 118}, 675--685.
913: 
914: \bibitem{AFS04}
915: Angeli, D., {Ferrell, Jr.}, J. E. \& Sontag, E. D. (2004)
916: %\newblock Detection of multistability, bifurcations, and hysteresis in a large
917: %  class of biological positive-feedback systems.
918: \newblock {\em Proc. Natl. Acad. Sci. USA} {\bf 101}, 1822--1827.
919: 
920: \bibitem{SAM04}
921: Stephanopoulos, G.~N., Alper, H. \& Moxley, J. (2004)
922: %\newblock Exploiting biological complexity for strain improvement through
923: %  systems biology.
924: \newblock {\em Nat. Biotechnol.} {\bf 22}, 1261--1267.
925: 
926: \bibitem{FFN03}
927: Famili, I., F\"orster, J., Nielsen, J., \& Palsson, B.~O. (2003)
928: %\newblock {\em Saccharomyces cerevisiae} phenotypes can be predicted by using
929: %  constraint-based analysis of a genome-scale reconstructed metabolic network.
930: \newblock {\em Proc. Nat. Acad. Sci. USA} {\bf 100}, 13134--13139.
931: 
932: \bibitem{SFD00}
933: Schuster, S., Fell, D.~A. \& Dandekar, T. (2000)
934: %\newblock A general definition of metabolic pathways useful for systematic
935: %  organization and analysis of complex metabolic systems.
936: \newblock {\em Nat. Biotechnol.} {\bf 18}, 326--332.
937: 
938: \bibitem{Bai01}
939: Bailey, J.~E.(2001)
940: %\newblock Complex biology with no parameters.
941: \newblock {\em Nat. Biotechnol.} {\bf 19} 503--504.
942: 
943: \bibitem{GC05}
944: Gagneur, J. \& Cesari, G. (2005)
945: %\newblock From molecular networks to qualitative cell behavior.
946: \newblock {\em FEBS Lett.} {\bf 579}, 1867--1871.
947: 
948: \bibitem{Gross04}
949: Gross, T. (2004)
950: %\newblock {\em Population Dynamics: General Results from Local Analysis}.
951: \newblock Der Andere Verlag T\"onning, Germany.
952: 
953: \bibitem{GF05}
954: Gross, T. \& Feudel, U. (2006)
955: %\newblock Generalized models as an universal approach to the analysis of
956: %  nonlinear dynamical systems.
957: \newblock {\em Phys. Rev. E } {\bf 73}, 016205--14 .
958: 
959: %\bibitem{SH83}
960: %M.~Schauer \& R.~Heinrich.
961: %\newblock Quasi-steady-state approximation in the mathematical modeling of
962: %  biochemical reaction networks.
963: %\newblock {\em Math. Biosci.}, 65:155--171, 1983.
964: 
965: \bibitem{BBW00}
966: Bier, M., Bakker, B.~M. \& Westerhoff, H.~V. (2000)
967: %\newblock How yeast cells synchronize their glycolytic oscillations: A
968: %  pertubation analytic treatment.
969: \newblock {\em Biophys. J.} {\bf 78} 1087--1093.
970: 
971: \bibitem{WPSS00}
972: Wolf, J., Passarge, J., Somsen, O.~J.~G., Snoep, J., Heinrich, R. \& Westerhoff, H.~V. (2000)
973: %\newblock Transduction of intracellular and intercellular dynamics in yeast
974: %  glycolytic oscillations.
975: \newblock {\em Biophys. J.} {\bf 78}, 1145--1153.
976: 
977: \bibitem{HDS01}
978: Hynne, F., Dan\o, S. \& {S\o rensen}, P.~G. (2001)
979: %\newblock Full-scale model of glycolysis in {\em saccharomyces cerevisae}.
980: \newblock {\em Biophys. Chem.} {\bf 94}, 121--163.
981: 
982: \bibitem{REA02}
983: Rosenfeld, N., Elowitz, M. \& Alon U. (2002)
984: %\newblock Negative autoregulation speeds the response times of transcription
985: %  networks.
986: \newblock {\em J. Mol. Biol.} {\bf 323}, 785--793.
987: 
988: \bibitem{PRP88}
989: Petterson, G. \& Ryde-Petterson, U. (1998)
990: %\newblock A mathematical model of the calvin photosynthesis cycle.
991: \newblock {\em Eur. J. Biochem.} {\bf 175}, 661--672.
992: 
993: 
994: \bibitem{POL01}
995: Poolman, M.~G., \"Olcer, H., Lloyd, J.~C., Raines, C.~A. \& Fell D.~A. (2001)
996: %\newblock Computer modelling and experimental evidence for two steady states in
997: %  the photosynthetic calvin cycle.
998: \newblock {\em Eur. J. Biochem.} {\bf 268}, 2810--2816.
999: 
1000: \bibitem{RP91}
1001: Ryde-Petterson, U. (1991)
1002: %\newblock Identification of possible two-reactant sources of oscillations in
1003: %  the calvin photosynthesis cycle \& ancillary pathways.
1004: \newblock {\em Eur. J. Biochem.} {\bf 198} 613--619.
1005: 
1006: \bibitem{Kuz95}
1007: Kuznetsov, Yu.~A. (1995)
1008: \newblock {\em Elements of Applied Bifurcation Theory},
1009: \newblock (Springer, Berlin).
1010: 
1011: \bibitem{GEF05}
1012: Gross, T., Ebenh\"oh, W. \& Feudel, U. (2005)
1013: %\newblock Long food chains are in general chaotic.
1014: \newblock {\em Oikos} {\bf 109}, 135--144.
1015: 
1016: \end{thebibliography}
1017: 
1018: }
1019: 
1020: \end{document}
1021: 
1022: 
1023: 
1024: 
1025: 
1026: 
1027: 
1028: 
1029: