1: %% LyX 1.3 created this file. For more info, see http://www.lyx.org/.
2: %% Do not edit unless you really know what you are doing.
3: \documentclass[english]{article}
4: \usepackage[T1]{fontenc}
5: \usepackage[latin1]{inputenc}
6: \usepackage{geometry}
7: \geometry{verbose,letterpaper,tmargin=1in,bmargin=1in,lmargin=1in,rmargin=1in}
8: \usepackage{subfigure}
9: \usepackage{amsmath}
10: \usepackage{graphicx}
11: \usepackage{amssymb}
12: \usepackage[numbers]{natbib}
13:
14: \makeatletter
15: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Textclass specific LaTeX commands.
16: \newcommand{\lyxaddress}[1]{
17: \par {\raggedright #1
18: \vspace{1.4em}
19: \noindent\par}
20: }
21:
22: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%% User specified LaTeX commands.
23: \usepackage{subfigure}
24: \usepackage{graphicx}
25: %\bibpunct{(}{)}{,}{a}{}{;}
26: %\date{}
27:
28: \usepackage{babel}
29: \makeatother
30: \begin{document}
31:
32: \title{Comparative analysis of some models of mixed-substrate microbial
33: growth}
34:
35:
36: \author{Atul Narang%
37: \thanks{Email: \texttt{narang@che.ufl.edu}%
38: }}
39:
40: \maketitle
41:
42: \lyxaddress{Department of Chemical Engineering, University of Florida, Gainesville,
43: FL~32611-6005. }
44:
45: \noindent \begin{flushleft}Keywords: Mathematical model, mixed substrate
46: growth, diauxic growth, lac operon, Lotka-Volterra model.\end{flushleft}
47:
48: \begin{abstract}
49: Mixed-substrate microbial growth is among the most intensely studied
50: systems in molecular microbiology. Several mathematical models have
51: been developed to account for the genetic regulation of such systems,
52: especially those resulting in diauxic growth. In this work, we compare
53: the dynamics of three such models (Narang, Biotech.~Bioeng., 59,
54: 116, 1998; Thattai \& Shraiman, Biophys.~J, 85, 744, 2003; Brandt
55: et al, Water Research, 38, 1004, 2004). We show that these models
56: are dynamically similar --- the initial motion of the inducible enzymes
57: in all the models is described by Lotka-Volterra equations for competing
58: species. In particular, the prediction of diauxic growth corresponds
59: to {}``extinction'' of one of the enzymes during the first few hours
60: of growth. The dynamic similarity occurs because in all the models,
61: the inducible enzymes possess properties characteristic of competing
62: species: Their synthesis is autocatalytic, and they inhibit each other.
63: Despite this dynamic similarity, the models vary with respect to the
64: range of dynamics captured. The Brandt et al model captures only the
65: diauxic growth pattern, whereas the remaining two models capture both
66: diauxic and non-diauxic growth patterns. The models also differ with
67: respect to the mechanisms that generate the mutual inhibition between
68: the enzymes. In the Narang model, the mutual inhibition occurs because
69: the enzymes for each substrate enhance the dilution of the enzymes
70: for the other substrate. In the Thattai \& Shraiman model, the mutual
71: inhibition is entirely due to competition for the phosphoryl groups.
72: Elements of all the models appear to be necessary for quantitative
73: agreement with data.
74: \end{abstract}
75:
76: \section{Introduction\label{s:Introduction}}
77:
78: When microbial cells are grown in a batch culture containing a mixture
79: of two carbon sources, they often exhibit \emph{diauxic} growth, which
80: is characterized by the appearance two exponential growth phases separated
81: by a lag phase called \emph{diauxic lag}~\citep{monod47}. The most
82: well-known example of this phenomenon is the batch growth of \emph{E.
83: coli} on a mixture of glucose and lactose (Figure~\ref{f:DiauxicGrowth}a).
84: Early studies by Monod showed that in this case, the two exponential
85: growth phases reflect the sequential consumption of glucose and lactose~\citep{monod1}.
86: Moreover, only glucose is consumed in the first exponential growth
87: phase because the synthesis of the \emph{peripheral} enzymes for lactose
88: (the enzymes that catalyze the transport and peripheral catabolism
89: of lactose) is somehow abolished in the presence of glucose. During
90: this period of preferential growth on glucose, the peripheral enzymes
91: for lactose are diluted to very small levels: 6--7 generations of
92: growth on glucose reduce the enzyme levels to $\sim1$\% of their
93: initial values. Thus, the diauxic lag reflects the time required for
94: the cells to build up the peripheral enzymes for lactose to sufficiently
95: high levels. After the diauxic lag, one observes the second exponential
96: phase corresponding to consumption of lactose.
97:
98: %
99: \begin{figure}
100: \begin{center}\subfigure[]{\includegraphics[%
101: width=7cm,
102: height=5cm]{Figures/InadaWTdata.eps}}\hspace{0.1in}\subfigure[]{\includegraphics[%
103: width=7cm,
104: height=5cm]{Figures/cAMPlevelDuringtheDiauxie.eps}}\end{center}
105:
106:
107: \caption{\label{f:DiauxicGrowth}Diauxic growth of \emph{E. coli} on a mixture
108: of glucose and lactose (from~\citep{inada96}): (a) The optical density~($\square$)
109: shows two exponential growth phases separated by a diauxic lag ($60\lesssim t\lesssim160$~min).
110: The levels of $\beta$-galactosidase~($\bigcirc$), a peripheral
111: enzyme for lactose, remain low until the beginning of the diauxic
112: lag. (b)~Evolution of the intracellular cAMP levels during the experiment
113: shown in (a). The intracellular cAMP levels~($\bullet$) during the
114: first phase of exponential growth on glucose ($t\lesssim60$~min)
115: are similar to the intracellular cAMP levels during the second phase
116: of exponential growth on lactose ($t\gtrsim160$~min).}
117: \end{figure}
118:
119:
120: The key to the resolution of the glucose-lactose diauxie is clearly
121: the molecular mechanism by which the synthesis of lactose-specific
122: enzymes is abolished in the presence of glucose. The first inroads
123: into this problem were made by Monod and coworkers who discovered
124: the mechanism for synthesis (induction) of the lactose-specific enzymes
125: in the presence of lactose~\citep{jacob61}. It was shown that the
126: genes corresponding to the peripheral enzymes for lactose are contiguous
127: on the DNA, an arrangement referred to as the \emph{lac} operon. In
128: the absence of lactose, transcription of the \emph{lac} operon is
129: prevented by a repressor molecule, called the \emph{lac} repressor,
130: which is bound to a specific site on the \emph{lac} operon. In the
131: presence of lactose, transcription of the \emph{lac} operon is triggered
132: because allolactose, a product of $\beta$-galactosidase, sequesters
133: the repressor from the operon, thus liberating it for transcription.%
134: \footnote{A similar mechanism serves to induce the genes for glucose transport~\citep[Figure~4]{plumbridge03}.
135: In the absence of glucose, transcription of the \emph{ptsG} gene,
136: which codes for the transport enzyme, IIBC$^{{\rm glc}}$, is inhibited
137: because a repressor called Mlc is bound to a regulatory site \emph{}on
138: the gene. Upon entry of glucose, IIBC$^{{\rm glc}}$ is dephosphorylated.
139: Dephosphorylated IIBC$^{{\rm glc}}$ sequesters Mlc away from the
140: regulatory site on \emph{ptsG}, thus liberating the gene for transcription.%
141: }
142:
143: Given this mechanism for induction of the lactose-specific enzymes,
144: it seems plausible to hypothesize that the glucose-lactose diauxie
145: occurs because transcription of the \emph{lac} operon is somehow abolished
146: in the presence of glucose. These mechanisms are not fully understood~\citep{stulke99}.
147: Until recently, there were two models for inhibition of \emph{lac}
148: transcription in the presence of glucose
149:
150: \begin{enumerate}
151: \item \emph{cAMP activation:} This model postulates that a complex consisting
152: of cyclic AMP (cAMP) and catabolite repression protein (CRP) must
153: bind to a specific site on the \emph{lac} operon before it can be
154: transcribed. When glucose is added to a culture growing on lactose,
155: the cAMP levels somehow decrease, which reduces the binding of the
156: cAMP-CRP complex to the \emph{lac} operon, thus inhibiting its transcription
157: rate.
158: \item \emph{Inducer exclusion:} According to this model, enzyme IIA$^{{\rm glc}}$,
159: a peripheral enzyme for glucose, is dephosphorylated in the presence
160: of glucose. The dephosphorylated enzyme IIA$^{{\rm glc}}$ inhibits
161: lactose uptake by binding to the lactose permease, the transport enzyme
162: for lactose. This reduces the intracellular concentration of allolactose,
163: and hence, the transcription rate of the \emph{lac} operon.
164: \end{enumerate}
165: Experiments by Aiba and coworkers have shown that the cAMP activation
166: model is not tenable~\citep{inada96}. The cAMP levels are the same
167: during growth on glucose and lactose (Figure~\ref{f:DiauxicGrowth}b).
168: Moreover, the \emph{lac} operon is not transcribed even if cAMP is
169: added to a culture growing on glucose and lactose. It is now believed
170: that inducer exclusion alone is responsible for inhibiting \emph{lac}
171: transcription. But in \emph{E. coli} ML30, the activity of lactose
172: permease is inhibited no more than $\sim$40\% at saturating concentrations
173: of glucose~\citep[Table 2]{cohn59a}. Likewise, Saier and coworkers,
174: who discovered inducer exclusion in \emph{S. typhimurium}, found that
175: inducer exclusion by glucose inhibits the synthesis of the peripheral
176: enzymes for melibiose, glycerol, maltose, and lactose by 10--50\%~\citep[Figures~1--2]{Saier1972}.
177: This partial inhibition by inducer exclusion cannot explain the almost
178: complete inhibition of the genes for the {}``less preferred'' substrates.
179:
180: Although the diauxie has dominated the literature on mixed-substrate
181: growth, there is ample evidence of non-diauxic growth patterns. This
182: was already evident from Monod's early studies in which he classified
183: his mixed-substrate data into two categories~\citep{monod1,monod47}.
184: Growth on a particular mixture was called diauxic if the growth curve
185: showed the diauxic lag, and \emph{normal} if it showed no such lag.
186: Yet, the phenomenon of normal growth was virtually ignored until recently.
187: In the last few years, several studies have shown that both substrates
188: can be consumed simultaneously. Figure~\ref{f:NondiauxicGrowth}a
189: shows, for instance, that \emph{E. coli} consumes fumarate and pyruvate
190: simultaneously during batch growth. Egli has summarized all known
191: examples of simultaneous substrate utilization in a comprehensive
192: review article~\citep{egli95}. He notes that, in general, simultaneous
193: substrate utilization is observed when both substrates support low-to-medium
194: specific growth rates, and diauxic growth occurs when one of the substrates
195: supports a specific growth rate that is substantially higher than
196: the specific growth rate on the other substrate. In addition to simultaneous
197: substrate utilization, there is some evidence that the substrate utilization
198: pattern can depend on the history of the inoculum, one example of
199: which is shown in Figure~\ref{f:NondiauxicGrowth}b (see~\citep{narang98b}
200: for other examples).
201:
202: %
203: \begin{figure}
204: \begin{center}\subfigure[]{\includegraphics[%
205: width=7cm,
206: height=5cm]{Figures/fupy.eps}}\hspace{0.1in}\subfigure[]{\includegraphics[%
207: width=7cm,
208: height=5cm]{Figures/icdep.eps}}\end{center}
209:
210:
211: \caption{\label{f:NondiauxicGrowth}Nondiauxic growth of \emph{E. coli} (from~\citep{narang97a}):
212: (a) Simultaneous substrate utilization during batch growth of \emph{E.
213: coli} K12 on a mixture of fumarate (FUM) and pyruvate (PYR). The cell
214: density is denoted by $c$ (gdw/L) . This growth pattern is observed
215: with several pairs of organic acids~\citep{narang97a}. (b) Growth
216: pattern dependent on the history of the inoculum. When the inoculum
217: is grown on glucose ($\ast$), the specific growth rate on a mixture
218: of glucose and pyruvate is 0.74~1/hr. When the inoculum is grown
219: on pyruvate ($\blacksquare,\blacklozenge$), the specific growth rate
220: on the same mixture is 0.56~1/hr.}
221: \end{figure}
222:
223:
224: The phenomenon of mixed-substrate growth is of fundamental importance
225: in molecular biology as a paradigm of the mechanism by which the expression
226: of DNA is controlled. It also has profound implications for several
227: large-scale biotechnological processes.%
228: \footnote{The large-scale production of chemicals, such as bioethanol and biopolymers,
229: is economically feasible only if they are derived from cheap lignocellulosic
230: feedstocks~\citep{ingram99}. The pretreatment of these feedstocks
231: yields a mixture of hexoses (primarily, glucose) and pentoses (primarily,
232: xylose). The cells that ferment these sugars to useful products typically
233: exhibit diauxic growth with preferential consumption of hexoses.%
234: } This has spurred the development of several mechanistic models of
235: mixed-substrate growth. Some of these models are inspired by the detailed,
236: but constantly evolving, knowledge of the molecular mechanism for
237: the glucose-lactose diauxie~\citep{Santillan2004,vandedem73,wong97}.
238: The other models appeal to the fact that the phenomenon of diauxic
239: growth is ubiquitous --- it has been observed in diverse microbial
240: species on many pairs of \emph{substitutable} substrates (i.e., substrates
241: that satisfy the same nutrient requirements) including pairs of carbon~\citep{egli95,harder82,kovarova98},
242: nitrogen~\citep{Neidhardt1957}, and phosphorus~\citep{Daughton1979}
243: sources, and even among pairs of electron acceptors~\citep{Liu1998}.
244: Thus, it is conceivable there exist some general mechanisms driving
245: the dynamics of mixed-substrate growth. These general models, which
246: abstract features common to many, if not all, mixed-substrate systems,
247: include the cybernetic model~\citep{kompala84,ramakrishna96} and
248: several kinetic models~\citep{Brandt2003,narang97c,Thattai2003}.
249:
250: The original cybernetic model, which was analyzed in~\citep{narang97b},
251: cannot capture nondiauxic growth patterns. In this work, we compare
252: the general kinetic models developed by Narang et al~\citep{narang98b,narang97c},
253: Brandt et al~\citep{Brandt2003}, and Thattai \& Shraiman~\citep{Thattai2003}.
254: Hereafter, we shall refer to these as the N-, B- and T-models, respectively.
255: We show that
256:
257: \begin{enumerate}
258: \item All the models are similar inasmuch as they exhibit the same general
259: class of dynamics. More precisely, the equations describing the initial
260: evolution of the peripheral enzymes are special cases of the generalized
261: Lotka-Volterra model for competing species. This similarity arises
262: because all the models possess the two defining properties of the
263: Lotka-Volterra model for competing species: \emph{Autocatalysis} (the
264: synthesis of the peripheral enzymes for both substrates is autocatalytic),
265: and \emph{mutual inhibition} (the peripheral enzymes for each substrate
266: inhibit the synthesis of the peripheral enzymes for the other substrate).
267: The existence of this similarity implies that the dynamics of the
268: peripheral enzymes are analogous to the dynamics of the Lotka-Volterra
269: model. In particular, the prediction of diauxic growth by these models
270: corresponds to {}``extinction'' of one of the enzymes.
271: \item The models differ with respect to the predicted range of dynamics
272: and the mechanism by which they inherit the essential properties of
273: the Lotka-Volterra model.
274:
275: \begin{enumerate}
276: \item The B-model captures only diauxic growth patterns, whereas the N-
277: and T-models capture both diauxic non-diauxic growth patterns
278: \item In the N-model, mutual inhibition arises because each enzyme stimulates
279: the dilution of the other enzyme. On the other hand, in the T-model,
280: the mutual inhibition occurs because the sugar-specific enzymes of
281: the phosphotransferase system compete for phosphoryl groups.
282: \end{enumerate}
283: \end{enumerate}
284: Comparison with experiments suggests that elements of both all the
285: models are required for capturing the data.
286:
287:
288: \section{The models}
289:
290: Before describing the models, it is useful to mention a few points.
291:
292: \begin{enumerate}
293: \item Although all models contain more or less the same variables, the notation
294: varies considerably from one study to another. To facilitate comparison
295: between the models, we have used the same notation for the variables.
296: We denote the cells, exogenous substrates, inducers, and peripheral
297: enzymes by $C$, $S_{i}$, $X_{i}$ and $E_{i}$, respectively. The
298: concentrations of these entities are denoted by the corresponding
299: lower-case letters, $s_{i}$, $e_{i}$, $x_{i}$ and $c$, respectively.
300: \item All the models assume the existence of a small \emph{constitutive}
301: or background enzyme synthesis rate that persists even in the absence
302: of the inducer. We neglect this term since it is generally small compared
303: to the \emph{induced} enzyme synthesis rate.
304: \item The cell density and exogenous substrate concentrations are based
305: on the volume of the culture (gdw/L and g/L, respectively). In contrast,
306: the concentrations of intracellular variables, such as the enzymes
307: and inducers, are based on the dry weight of the cells (g/gdw).
308: \item The foregoing choice of units implies that if $Z$ is any intracellular
309: entity produced at the rate, $r_{z}^{+}$~g/gdw-hr, and degraded
310: at the rate, $r_{z}^{-}$~g/gdw-hr, then the mass balance for~$z$
311: (in g/gdw) is given by the equation\[
312: \frac{d(zc)}{dt}=\left(r_{z}^{+}-r_{z}^{-}\right)c\Rightarrow\frac{dz}{dt}=r_{z}^{+}-r_{z}^{-}-\left(\frac{1}{c}\frac{dc}{dt}\right)z.\]
313: Here, the last term reflects the dilution of $Z$ due to growth.
314: \end{enumerate}
315: We are now ready to describe the key features of the models.
316:
317: %
318: \begin{figure}
319: \begin{center}\subfigure[]{\includegraphics[%
320: width=7cm,
321: height=5cm]{Figures/SchemeNarang.eps}}\hspace{0.3in}\subfigure[]{\includegraphics[%
322: width=7cm,
323: height=5cm]{Figures/SchemeBrandt.eps}}\end{center}
324:
325:
326: \caption{\label{f:ModelSchemes1}Kinetic schemes: (a) N-model~\citep{kompala86}.
327: (b) B-model~\citep{narang98b}.}
328: \end{figure}
329:
330:
331:
332: \subsection{N-model}
333:
334: The kinetic scheme for the N-model is shown in Figure~\ref{f:ModelSchemes1}a.
335: It is assumed that~\citep{narang98b}
336:
337: \begin{enumerate}
338: \item Transport of $S_{i}$ is catalyzed by enzyme $E_{i}$. The specific
339: uptake rate of $S_{i}$, denoted $r_{s,i}$, follows the kinetics,
340: $r_{s,i}\equiv V_{s,i}e_{i}s_{i}/(K_{s,i}+s_{i})$.
341: \item Part of the internalized substrate, denoted $X_{i}$, is converted
342: to biosynthetic constituents such as amino acids and proteins, denoted
343: $C^{-}$. The remainder is oxidized to energy (${\rm CO_{2}}$).
344:
345: \begin{enumerate}
346: \item The specific rate of conversion of $X_{i}$ to $C^{-}$ and ${\rm CO_{2}}$
347: is $r_{x,i}\equiv k_{x,i}x_{i}$.
348: \item The fraction of $X_{i}$ converted to $C^{-}$ is a constant (parameter),
349: denoted $Y_{i}$. Thus, the specific rate of biosynthesis from $X_{i}$
350: is $Y_{i}r_{x,i}$.
351: \end{enumerate}
352: \item The internalized substrate induces the synthesis of $E_{i}$.
353:
354: \begin{enumerate}
355: \item The specific synthesis rate of $E_{i}$ is $r_{e,i}\equiv V_{e,i}x_{i}^{n_{i}}/(K_{e,i}^{n_{i}}+x_{i}^{n_{i}})$,
356: where $n_{i}=1\textnormal{ or }2$.%
357: \footnote{Enzyme induction can be hyperbolic ($n_{i}=1$) or sigmoidal ($n_{i}=2$),
358: depending on the number of inducer molecules that bind to a repressor
359: molecule~\citep{chung96,yagil71}.%
360: }
361: \item The synthesis of the enzymes occurs at the expense of the biosynthetic
362: constituents, $C^{-}$.
363: \end{enumerate}
364: \end{enumerate}
365: Thus, one obtains the equations\begin{align}
366: \frac{ds_{i}}{dt} & =-r_{s,i}c,\; r_{s,i}\equiv V_{s,i}e_{i}\frac{s_{i}}{K_{s,i}+s_{i}},\label{eq:NsO}\\
367: \frac{dx_{i}}{dt} & =r_{s,i}-r_{x,i}-\left(\frac{1}{c}\frac{dc}{dt}\right)x_{i},r_{x,i}\equiv k_{x,i}x_{i},\label{eq:NxO}\\
368: \frac{de_{i}}{dt} & =r_{e,i}-\left(\frac{1}{c}\frac{dc}{dt}\right)e_{i},\; r_{e,i}\equiv V_{e,i}\frac{x^{n_{i}}}{K_{e,i}^{n_{i}}+x^{n_{i}}},\label{eq:NeO}\\
369: \frac{dc^{-}}{dt} & =(Y_{1}r_{x,1}+Y_{2}r_{x,2})-(r_{e,1}+r_{e,2})-\left(\frac{1}{c}\frac{dc}{dt}\right)c^{-}.\label{eq:NcMO}\end{align}
370: These equations implicitly define the specific growth rate and the
371: evolution of the cell density. To see this, observe that since all
372: the intracellular concentrations are expressed as mass fractions (g/gdw),
373: their sum equals~1, i.e., $x_{1}+x_{2}+e_{1}+e_{2}+c^{-}=1$. Hence,
374: addition of equations (\ref{eq:NxO}--\ref{eq:NcMO}) yields\[
375: 0=\sum_{i=1}^{2}r_{s,i}-(1-Y_{i})r_{x,i}-\frac{1}{c}\frac{dc}{dt}\]
376: which can be rewritten in the more familiar form\begin{equation}
377: \frac{dc}{dt}=r_{g}c,\; r_{g}\equiv\sum_{i=1}^{2}r_{s,i}-(1-Y_{i})r_{x,i}\label{eq:NrG}\end{equation}
378: where $r_{g}$ denotes the specific growth rate.
379:
380: We can simplify the model by observing that $x_{i}\sim10^{-3}$~g/gdw~\citep{chung96}
381: and $r_{s,i},r_{x,i}\sim1$~g/gdw-hr. Thus, $x_{i}$ attains quasisteady
382: state on a time scale of $10^{-3}$~hr. Moreover, the dilution term
383: $r_{g}x_{i}\sim10^{-3}$~g/gdw-hr is negligibly small compared to
384: $r_{s,i},r_{x,i}$. Hence, within a few seconds, (\ref{eq:NxO}) becomes,
385: $0\approx r_{s,i}-r_{x,i}$, so that $r_{g}\equiv\sum_{i}r_{s,i}-(1-Y_{i})r_{x,i}\approx\sum_{i}Y_{i}r_{s,i}$.
386: Thus, we arrive at the equations \begin{align}
387: \frac{dc}{dt} & =(Y_{1}r_{s,1}+Y_{2}r_{s,2})c,\; r_{s,i}\equiv V_{s,i}e_{i}\frac{s_{i}}{K_{s,i}+s_{i}}\label{eq:Nc}\\
388: \frac{ds_{i}}{dt} & =-r_{s,i}c\label{eq:Ns}\\
389: \frac{de_{i}}{dt} & =r_{e,i}-(Y_{1}r_{s,1}+Y_{2}r_{s,2})e_{i},\; r_{e,i}\equiv V_{e,i}\frac{x_{i}^{n_{i}}}{K_{e,i}^{n_{i}}+x_{i}^{n_{i}}}\label{eq:Ne}\\
390: x_{i} & \approx\frac{V_{s,i}e_{i}s_{i}/(K_{s,i}+s_{i})}{k_{x,i}}\label{eq:Nx}\\
391: c^{-} & =1-x_{1}-x_{2}-e_{1}-e_{2}\label{eq:NcM}\end{align}
392: where (\ref{eq:Nx}) is obtained by solving the quasisteady state
393: relation, $r_{x,i}\approx r_{s,i}$, for $x_{i}$. Substituting (\ref{eq:Nx})
394: in the expression for $r_{e,i}$ yields\[
395: r_{e,i}=V_{e,i}\frac{[e_{i}s_{i}/(K_{s,i}+s_{i})]^{n_{i}}}{\bar{K}_{e,i}^{n_{i}}+[e_{i}s_{i}/(K_{s,i}+s_{i})]^{n_{i}}},\;\bar{K}_{e,i}\equiv K_{e,i}\frac{k_{x,i}}{V_{s,i}}\]
396: which shows that enzyme synthesis is autocatalytic: The larger the
397: enzyme level, the higher its synthesis rate. This is a consequence
398: of the cyclic structure associated with the kinetics of induction.
399: Figure~\ref{f:ModelSchemes1}b shows that the enzyme, $E_{i}$, promotes
400: the synthesis of the inducer, $X_{i}$, which in turn stimulates the
401: synthesis of even more $E_{i}$. This cycle of reactions implies that
402: enzyme synthesis is autocatalytic.
403:
404:
405: \subsection{B-model}
406:
407: The B-model is similar to the N-model, the only difference being that
408: the intracellular substrate, $X_{i}$, not only stimulates the induction
409: of $E_{i}$, but also inhibits the induction of $E_{j},j\ne i$ (shown
410: in Figure~\ref{f:ModelSchemes2}b as arrows with a bar at one end).%
411: \footnote{Brandt et al refer to the intracellular substrate (inducer) and the
412: enzyme induction machinery as \emph{signal molecule} and \emph{synthesizing
413: unit} (SU), respectively (see Figure~1 of~\citep{Brandt2003}).%
414: } Assuming that $x_{i}$ rapidly attains quasisteady state, Brandt
415: et al arrive at the equations~\citep{Brandt2003} \begin{align}
416: \frac{dc}{dt} & =r_{g}c,\; r_{g}=r_{g,1}+r_{g,2},\; r_{g,i}\equiv V_{g,i}e_{i}\frac{s_{i}}{K_{s,i}+s_{i}}\label{eq:Bc}\\
417: \frac{ds_{i}}{dt} & =-r_{s,i}c,\; r_{s,i}=\frac{r_{g,i}}{Y_{i}}\label{eq:Bs}\\
418: \frac{de_{i}}{dt} & =r_{e,i}-r_{g}e_{i},\; r_{e,i}\equiv r_{g}\left[\frac{p_{i}e_{i}s_{i}/(K_{s,i}+s_{i})}{p_{1}e_{1}s_{1}/(K_{s,1}+s_{1})+p_{2}e_{2}s_{2}/(K_{s,2}+s_{2})}\right]\label{eq:Be}\end{align}
419: where $r_{g,i},r_{s,i},Y_{i}$ denote the specific growth rate, the
420: specific substrate uptake rate, and the yield of biomass on the $i^{{\rm th}}$
421: substrate, $r_{g}$ denotes the total specific growth rate, and $0\le p_{i}\le1$
422: are parameters called \emph{substrate preference coefficients}.
423:
424: Evidently, enzyme synthesis is autocatalytic because of the positive
425: feedback from $X_{i}$ to $E_{i}$. The appearance of the specific
426: growth rate, $r_{g}$, in the expression for $r_{e,i}$ stems from
427: an additional assumption. It is argued that the specific enzyme synthesis
428: rate should be proportional to the specific growth rate to ensure
429: that the specific enzyme synthesis rate grows in proportion to the
430: specific growth rate.
431:
432: We note finally that Brandt et al scaled the enzyme level, $e_{i}$,
433: with the level, $e_{i}^{*}$, that would be observed during exponential
434: growth on $S_{i}$ alone. Thus, the variable $e_{i}$ shown in equation
435: (\ref{eq:Be}) corresponds to the scaled variable, $\kappa_{i}\equiv e_{i}/e_{i}^{*}$
436: in~\citep{Brandt2003}.
437:
438:
439: \subsection{T-model}
440:
441: %
442: \begin{figure}
443: \begin{center}\subfigure[]{\includegraphics[%
444: width=7cm,
445: height=3cm]{Figures/PTSpathway.eps}}\hspace*{0.2in}\subfigure[]{\includegraphics[%
446: width=7cm,
447: height=5cm]{Figures/SchemeThattai.eps}}\end{center}
448:
449:
450: \caption{\label{f:ModelSchemes2}T-model~\citep{Thattai2003} (a) The phosphotransferase
451: system (PTS) for carbohydrate uptake. (b) The kinetic scheme.}
452: \end{figure}
453:
454:
455: The T-model is aimed at describing the evolution of the peripheral
456: enzymes belonging to the \emph{phosphotransferase system} (PTS), which
457: catalyzes the uptake of various sugars in certain bacteria~\citep{Postma1993}.
458: The uptake of PTS sugars is coupled to their phosphorylation. This
459: is mediated by a cascade of 5~phosphorylation reactions involving
460: the successive transfer of a phosphoryl group from PEP to the sugar
461: (Figure~\ref{f:ModelSchemes2}a). The first two steps, involving
462: phosphorylation of enzyme~I (EI) and HPr, are common to all the PTS
463: sugars. The last three steps are mediated by sugar-specific enzymes,
464: enzyme~IIA (EIIA) and the enzyme~IIBC complex (EIIBC), which ultimately
465: transfer the phosphoryl group to the sugar during its translocation
466: across the membrane. The sugar-specific enzymes are inducible, and
467: their synthesis is coupled since they lie on the same operon.
468:
469: Figure~\ref{f:ModelSchemes2}b shows the kinetic scheme of the T-model.
470: It is assumed that
471:
472: \begin{enumerate}
473: \item There is a maximum flux, $J$, of phosphoryl groups through the common
474: enzymes, EI and HPr, and the sugar-specific enzymes of PTS compete
475: for these phosphoryl group. It turns out that at quasisteady state,
476: the specific phosphorylation rate of $i^{{\rm th}}$ substrate, $J_{i}$,
477: (which is equal to the specific substrate uptake rate, $r_{s,i}$)
478: is given by\[
479: r_{s,i}=J_{i}=J\frac{\tau_{i}}{1+\tau_{1}+\tau_{2}},\;\tau_{i}\equiv\frac{e_{i}^{2}}{\beta_{i}}\frac{s_{i}}{s_{i}+e_{i}},\]
480: where $\tau_{i}$ can be interpreted as the \emph{demand} for phosphoryl
481: groups by the $i^{{\rm th}}$ sugar --- it is an increasing function
482: of the exogenous sugar concentration, $s_{i}$, and the sugar-specific
483: enzyme level, $e_{i}$. Consistent with this interpretation, $J_{i}<J$
484: due to competing demands for phosphoryl groups imposed by the substrates.
485: \item The specific rate of enzyme synthesis, $r_{e,i}$, is proportional
486: to the concentration of the intracellular substrate (inducer), $x_{i}$,
487: which in turn is proportional to the specific substrate uptake rate,
488: $r_{s,i}$. Hence\[
489: r_{e,i}\propto r_{s,i},\]
490: which implies that enzyme synthesis is autocatalytic.
491: \item The substrate concentrations and the specific growth rate are constant
492: --- they are treated as parameters.
493: \end{enumerate}
494: Thus, the evolution of the sugar-specific enzymes for the $i^{{\rm th}}$
495: substrate, when appropriately scaled, is given by the equations~\citep{Thattai2003}
496:
497: \begin{equation}
498: \frac{de_{i}}{dt}=\frac{\tau_{i}}{1+\tau_{1}+\tau_{2}}-e_{i},\;\tau_{i}\equiv\frac{e_{i}^{2}}{\beta_{i}}\frac{s_{i}}{s_{i}+e_{i}}.\label{eq:Te}\end{equation}
499: Note that the specific growth rate does not appear in the equations,
500: since time is scaled by the parameter,~$1/r_{g}$.
501:
502:
503: \section{\label{s:Results}Results}
504:
505: We wish to compare the dynamics of the three models described above.
506: At first sight, this seems impossible since the T-model describes
507: the dynamics of the enzymes only, whereas the N- and B-models describe
508: the dynamics of the enzymes, substrates, and cells. It turns out,
509: however, that the dynamics of the substrates and cells are irrelevant
510: on the time scale of interest. Indeed, insofar as the dynamics of
511: mixed-substrate growth are concerned, the asymptotic dynamics ($t\rightarrow\infty$)
512: of the N- and B-models is of little interest. Much more revealing
513: are their dynamics during the first exponential growth phase, since
514: it is these finite-time dynamics that determine the substrate utilization
515: pattern. Specifically, diauxic growth will occur if the peripheral
516: enzymes for one of the substrates vanishes during the first exponential
517: growth phase. In contrast, simultaneous substrate utilization will
518: be observed if the enzymes for both substrates persist during the
519: first exponential growth phase. We show below that
520:
521: \begin{enumerate}
522: \item In the N-, and B-models, the motion of the enzymes during the first
523: exponential growth phase can be described by a reduced \emph{}system
524: of two equations that are formally similar to the equations of the
525: T-model. This makes it possible to compare the N- and B-models with
526: the T-model.
527: \item The reduced equations of all the models are different realizations
528: of the generalized Lotka-Volterra model for two competing species.
529: Thus, in all the models, the enzymes behave like two competing species.
530: In particular, they coexist or become extinct, and these dynamics
531: have meaningful biological interpretations in the context of mixed-substrate
532: growth.
533: \item The B-model can never capture non-diauxic growth patterns.
534: \end{enumerate}
535: Finally, we compare the mechanisms underlying the dynamics of the
536: N- and T-models.
537:
538:
539: \subsection{All the models are dynamically similar to the Lotka-Volterra model}
540:
541: We begin by showing that in the N- and B-models, the dynamics of the
542: enzymes during the first exponential growth phase can be described
543: by a reduced system of two equations. To see this, observe that during
544: this period, both substrates are in excess, i.e., $s_{i}\gg K_{s,i}$.
545: Hence, even though the exogenous substrate concentrations are changing,
546: the transport enzymes remain saturated ($s_{i}/(K_{s,i}+s_{i})\approx$1).
547: Now, the cells sense the environment through the transport enzymes.
548: Since these enzymes see a quasiconstant environment during the first
549: exponential growth phase, they approach quasisteady state levels.
550: It follows that in the N- and B-models, the motion of the enzymes
551: from any initial conditions to the quasisteady state levels can be
552: obtained from~(\ref{eq:Ne}) and~(\ref{eq:Be}) by replacing $s_{i}/(K_{s,i}+s_{i})$
553: with~1.%
554: \footnote{We have reduced the equations by appealing to intuitive arguments.
555: This reduction can be justified rigorously by appealing to the theorem
556: of continuous dependence on initial conditions (see~\citep{narang97c}
557: for details). %
558: } Thus, we arrive at the reduced equations\begin{align}
559: \frac{de_{i}}{dt} & =V_{e,i}\frac{e_{i}^{n_{i}}}{K_{e,i}+e_{i}^{n_{i}}}-\left(Y_{1}V_{s,1}e_{1}+Y_{2}V_{s,2}e_{2}\right)e_{i}\label{eq:NeR}\\
560: \frac{de_{i}}{dt} & =r_{g}\left(\frac{p_{i}e_{i}}{p_{1}e_{1}+p_{2}e_{2}}-e_{i}\right).\label{eq:BeR}\end{align}
561: Since these equations are formally similar to equation~(\ref{eq:Te}),
562: we can compare the dynamics of all three models.
563:
564: It turns out that the equations of all three models are dynamical
565: analogs of the generalized Lotka-Volterra model for two competing
566: species, which is given by the equations~\citep[Chapter 12]{hirsch}\begin{equation}
567: \frac{dN_{i}}{dt}=f_{i}(N_{1},N_{2})N_{i}\label{eq:GLK}\end{equation}
568: where $N_{i}$ and $f_{i}(N_{1},N_{2})$ denote the population density
569: and specific growth rate of the $i^{{\rm th}}$ species, respectively,
570: and $f_{i}(N_{1},N_{2})$ satisfies the properties
571:
572: \begin{enumerate}
573: \item $\partial f_{1}/\partial N_{2},\partial f_{2}/\partial N_{1}<0$,
574: i.e., each species inhibits the growth of the other species.
575: \item $f_{i}(N_{1},N_{2})<0$ for sufficiently large $N_{1},N_{2}>0$, i.e.,
576: at sufficiently large population densities, the specific growth rates
577: are negative.
578: \end{enumerate}
579: The standard Lotka-Volterra model for competing species is a special
580: case of the generalized model with \[
581: f_{i}(N_{1},N_{2})=r_{i}(1-a_{i1}N_{1}-a_{i2}N_{2})\]
582: where $r_{i}$ is the unrestricted specific growth rate of the $i^{{\rm th}}$
583: species in the absence of any competition, and $a_{i1},a_{i2}$ are
584: coefficients that quantify the reduction of the unrestricted specific
585: growth rate due to intra- and inter-specific competition~\citep{murray}.
586: The analogy between the generalized Lotka-Volterra model and equations
587: (\ref{eq:Te}--\ref{eq:BeR}) becomes evident if we rewrite the latter
588: in the form\begin{align}
589: \frac{de_{i}}{dt} & =f_{i}^{N}(e_{1},e_{2})e_{i},\; f_{i}^{N}(e_{1},e_{2})\equiv V_{e,i}\frac{e_{i}^{n_{i}-1}}{\bar{K}_{e,i}^{n_{i}}+e_{i}^{n_{i}}}-\left(Y_{1}V_{s,1}e_{1}+Y_{2}V_{s,2}e_{2}\right)\label{eq:NeR1}\\
590: \frac{de_{i}}{dt} & =f_{i}^{B}(e_{1},e_{2})e_{i},\; f_{i}^{B}(e_{1},e_{2})\equiv r_{g}\left(\frac{p_{i}}{p_{1}e_{1}+p_{2}e_{2}}-1\right)\label{eq:BeR1}\\
591: \frac{de_{i}}{dt} & =f_{i}^{T}(e_{1},e_{2})e_{i},\; f_{i}^{T}(e_{1},e_{2})\equiv\frac{e_{i}s_{i}/(\beta_{i}+e_{i})}{1+\tau_{1}+\tau_{2}}-e_{i},\;\tau_{i}\equiv\frac{e_{i}^{2}}{\beta_{i}}\frac{s_{i}}{s_{i}+e_{i}}\label{eq:TeR1}\end{align}
592: One can check that the functions, $f_{i}^{N},f_{i}^{B},f_{i}^{T}$
593: satisfy the properties 1 and 2 above. Thus, in all the models, the
594: dynamics of the enzymes during the first exponential growth phase
595: are analogous to the dynamics of the generalized Lotka-Volterra model
596: for two competing species.
597:
598: The dynamics of the generalized Lotka-Volterra model for competing
599: species are well understood~\citep[Chapter 12]{hirsch}. Specifically,
600: the model entertains no limit cycles, so that all solutions ultimately
601: converge to some steady state. Despite the absence of limit cycles,
602: the model has a rich spectrum of dynamics. Even in the case of the
603: standard Lotka-Volterra model, one can get 4~different types of dynamics
604: depending on the parameter values~\citep{murray}. Indeed, if we
605: define the dimensionless variables, $u_{i}\equiv a_{ii}N_{i}$ and
606: $\tau\equiv r_{1}t$, we obtain the dimensionless equations \begin{align*}
607: \frac{du_{1}}{d\tau} & =(1-u_{1}-b_{12}u_{2})u_{1},\; b_{12}\equiv\frac{a_{12}}{a_{11}}\\
608: \frac{du_{2}}{d\tau} & =\rho(1-b_{21}u_{1}-u_{2})u_{2},\;\rho\equiv\frac{r_{2}}{r_{1}},\; b_{21}\equiv\frac{a_{21}}{a_{22}}.\end{align*}
609: The steady states of the model are completely determined by the parameters,
610: $b_{ij}$, which may be viewed as a measure of the extent to which
611: the $j^{{\rm th}}$ species inhibits the $i^{{\rm th}}$ species.
612: Figure~\ref{f:GlobalDynamics}a shows the bifurcation diagram of
613: the scaled standard Lotka-Volterra model. The bifurcation diagram
614: shows that when neither species inhibits the other species strongly
615: ($b_{12},b_{21}<1$), the two species coexist; when the cross-inhibition
616: is asymmetric ($b_{12}>1,b_{21}<1$ or $b_{12}<1,b_{21}>1$), one
617: of the species is rendered extinct; when both species inhibit each
618: other strongly ($b_{12},b_{21}>1$), the outcome of the competition
619: depends on the initial condition.
620:
621: Given the dynamical analogy between the Lotka-Volterra model and equations~(\ref{eq:NeR1}--\ref{eq:TeR1}),
622: it is reasonable to expect that the peripheral enzymes would yield
623: similar dynamics during the first phase of exponential growth. Importantly,
624: these dynamics have simple interpretations in terms of the substrate
625: utilization pattern. Indeed, extinction of one of the enzymes during
626: the first phase of exponential growth corresponds to diauxic growth;
627: coexistence of the enzymes during this period is the correlate of
628: simultaneous substrate uptake; and bistability reflects a substrate
629: utilization pattern which varies depending on the manner in which
630: the inoculum has been precultured.
631:
632:
633: \subsection{The B-model cannot capture non-diauxic growth patterns}
634:
635: %
636: \begin{figure}
637: \begin{center}\subfigure[]{\includegraphics[%
638: width=7cm,
639: height=7.5cm]{Figures/BifurcationDiagramLotkaVolterra.eps}}\hspace{0.3in}\subfigure[]{\includegraphics[%
640: width=7cm,
641: keepaspectratio]{Figures/BifurcationDiagramBrandtModel.eps}}\end{center}
642:
643:
644: \caption{\label{f:GlobalDynamics}Classification of the global dynamics: (a)
645: The bifurcation diagram of the standard Lotka-Volterra model. The
646: full and open circles show stable and unstable steady states, respectively.
647: (b) The bifurcation diagram of the B-model. When $p_{1}>p_{2}$ (resp.,
648: $p_{2}>p_{1}$), $E_{2}$ (resp., $E_{1}$) is rendered extinct during
649: the first exponential growth phase. The full and dashed lines show
650: the nullclines for $e_{1}$ and $e_{2}$, respectively. The full and
651: open circles show stable and unstable steady states, respectively.
652: The arrows show the orientation of the vector, $(de_{1}/dt,de_{2}/dt)$,
653: in the regions between the nullclines.}
654: \end{figure}
655: The N- and T-models can capture diauxic growth, simultaneous substrate
656: utilization, and bistable growth (see~\citep{narang98b,Thattai2003}
657: for details). However, the B-model always exhibits diauxic growth.
658: Indeed, the dynamics are completely determined by the substrate preference
659: coefficients, $p_{1}$ and $p_{2}$ (Figure~\ref{f:GlobalDynamics}b).
660: If $p_{1}>p_{2}$, $E_{2}$ becomes extinct during the first exponential
661: growth phase, which corresponds to preferential consumption of $S_{1}$.
662: Conversely, if $p_{2}>p_{1}$, $E_{1}$ becomes extinct during the
663: first exponential growth phase, which corresponds to preferential
664: consumption of $S_{2}$.
665:
666: To see this, it suffices to consider the nullclines of equation~(\ref{eq:BeR1}),
667: i.e., the curves along which $de_{i}/dt=0$. These curves, which separate
668: the $e_{1}e_{2}$-plane into regions in which $de_{i}/dt$ is nonzero,
669: are given by the equations\[
670: e_{i}=0\textrm{ or }f_{i}^{B}(e_{1},e_{2})\equiv\frac{p_{i}}{p_{1}e_{1}+p_{2}e_{2}}-1=0,\; i=1,2.\]
671: Evidently, $de_{1}/dt=0$ along the $e_{2}$-axis and the straight
672: line, $p_{1}e_{1}+p_{2}e_{2}=p_{1}$; we shall refer to the latter
673: curve as $\mu$. Likewise, $de_{2}/dt=0$ along the $e_{1}$-axis
674: and the straight line, $p_{1}e_{1}+p_{2}e_{2}=p_{2}$; we shall refer
675: to the latter curve as $\nu$. The steady states of (\ref{eq:BeR1})
676: lie at the intersection points of the nullclines for $e_{1}$ and
677: $e_{2}$. Since $\mu$ and $\nu$ are parallel, there are no {}``coexistence''
678: steady states. There are {}``extinction'' steady states at$(1,0)$
679: and $(0,1)$.%
680: \footnote{There is no steady state at $(0,0)$ since the model is undefined
681: (discontinuous) at this point.%
682: } One can check that the stability of $(1,0)$ and $(0,1)$ is determined
683: by the disposition of $\mu$ and $\nu$. If $p_{1}>p_{2}$, then $\mu$
684: lies above $\nu$, and $(1,0)$ is stable, while $(0,1)$ is unstable.
685: Conversely, if $p_{1}<p_{2}$, then $\nu$ lies above $\mu$, and
686: $(0,1)$ is stable, while $(1,0)$ is unstable. Thus, we conclude
687: that the B-model entertains only the diauxic growth pattern.
688:
689:
690: \subsection{The N- and T-models have different mechanisms of mutual inhibition}
691:
692: We have shown above that all the models are different realizations
693: of the generalized Lotka-Volterra model for two competing species.
694: Furthermore, the B-model cannot capture the non-diauxic growth patterns.
695: In what follows, we consider the similarities and differences between
696: the N- and T-models.
697:
698: We can develop a better appreciation of the similarities and differences
699: by examining the manner in which these models acquire the properties
700: of the generalized Lotka-Volterra model. The latter is characterized
701: by two essential properties.
702:
703: \begin{enumerate}
704: \item The growth of each species is autocatalytic, i.e., $dN_{i}/dt=0$
705: whenever $N_{i}=0$.
706: \item The interaction between the two species is \emph{mutually inhibitory},
707: i.e, $\partial f_{1}/\partial N_{2},\partial f_{2}/\partial N_{1}<0$.
708: \end{enumerate}
709: It is clear that all the models satisfy the first property precisely
710: because enzyme synthesis is autocatalytic ($r_{e,i}=0$ whenever $e_{i}=0$).
711: The mechanism that ensures that existence of this property is also
712: identical in all the models. It stems from the fact that the enzyme
713: promotes the formation of the internalized substrate (inducer) which
714: in turn stimulates the synthesis of even more enzyme.
715:
716: The difference between the models lies the mechanism(s) leading to
717: the second property, namely, mutual inhibition. In the N-model, each
718: enzyme inhibits the other enzyme by stimulating growth, and thus increases
719: the rate of \emph{dilution} of the other enzyme (e.g., $\partial f_{1}^{N}/\partial e_{2}<0$
720: precisely because $e_{2}$ appears in the dilution term for $e_{1}$).
721: On the other hand, in the T-model, there is no mutual inhibition due
722: to dilution --- in fact, the specific growth rate is assumed to be
723: a constant parameter. Instead, each enzyme inhibits the rate of \emph{synthesis}
724: of the other enzyme (e.g., $\partial f_{1}^{T}/\partial e_{2}<0$
725: precisely because $e_{2}$ appears in the synthesis term for $e_{1}$),
726: and this inhibition occurs due to competition for the phosphoryl groups.
727:
728: The N-model has two advantages over the T-model.
729:
730: \begin{enumerate}
731: \item It is more general than the T-model since it applies to any pair of
732: inducible substrates, as opposed to PTS sugars only. Indeed, the N-model
733: appeals to the two processes --- enzyme induction and growth --- that
734: occur in every system involving inducible substrates.
735: \item It explains an important empirical correlation observed in mixed-substrate
736: growth. Based on a comprehensive review of the experimental literature,
737: Harder \& Dijkhuizen~\citep{harder82} and Egli~\citep{egli95}
738: have observed that in general, both substrates are consumed simultaneously
739: when they support low-to-medium single-substrate growth rates. On
740: the other hand, diauxic growth is typically observed when one of the
741: substrates supports a much higher specific growth rate. In this case,
742: the substrate supporting the higher specific growth rate is usually
743: the {}``preferred substrate.'' \\
744: This can be understood in terms of the N-model, wherein each enzyme
745: inhibits the other enzyme by enhancing the latter's dilution rate.
746: Thus, enzymes for two substrates that support low-to-medium growth
747: rates will coexist since they will not inhibit each other significantly.
748: However, if the two substrates, say $S_{1}$ and $S_{2}$, support
749: high and low specific growth rates, respectively, then $E_{1}$ will
750: strongly inhibit the synthesis of $E_{2}$, but $E_{2}$ will have
751: little inhibitory effect on synthesis of $E_{1}$. Consequently, $E_{1}$
752: will drive $E_{2}$ to {}``extinction,'' resulting in preferential
753: utilization of $S_{1}$.
754: \end{enumerate}
755: The disadvantage of the N-model is that, unlike the T-model, it does
756: not account for inhibition of enzyme synthesis. It is conceivable
757: that this occurs by competition for phosphoryl groups. Another mechanism,
758: well-documented in the experimental literature, is inducer exclusion~\citep{stulke99}.
759: The latter is not accounted for by the N- and T-models. Indeed, neither
760: one of these models accounts for direct interaction between the enzymes
761: for the two substrates. The effect of the enzymes belonging to the
762: other substrate are exerted indirectly by influencing the specific
763: growth rate or demand for the phosphoryl groups. Thus, the N-model
764: can be viewed as a general model which is true of every pair of substrates
765: with inducible peripheral enzymes. However, for quantitative agreement,
766: it must be modified along the lines of the T-model by accounting for
767: specific mechanisms, such as inducer exclusion and competition for
768: phosphoryl groups, that inhibit enzyme synthesis.
769:
770: It is striking that all the models can predict diauxic growth despite
771: the absence of direct inhibitory interactions such as inducer exclusion.
772: These dynamics occur precisely because enzyme synthesis is autocatalytic
773: --- it is this property that makes it feasible for enzymes to become
774: {}``extinct'' during the first exponential growth phase. Thus, the
775: models imply that diauxic growth would not be observed if autocatalysis
776: were destroyed. This is consistent with the experimental data. Constitutive
777: mutants, in which synthesis of lactose-specific enzymes persists even
778: in absence of the inducer concentration ($\left.r_{e,i}\right|_{e_{i}=0}>0$),
779: do not display the diauxie~\citep[Figure~6]{inada96}. Similarly,
780: the glucose-lactose diauxie is not observed if the medium contains
781: IPTG, an inducer of the \emph{lac} operon that can enter the cell
782: even in the absence of the lactose permease~\citep[Figure~7]{inada96}.
783:
784:
785: \section{\label{s:Conclusions}Conclusions}
786:
787: We compared the similarities and differences between three kinetic
788: models of mixed-substrate growth. We showed that
789:
790: \begin{enumerate}
791: \item In all three models, the dynamics of the peripheral enzymes are formally
792: similar to the generalized Lotka-Volterra model for competing species.
793: This similarity occurs because the peripheral enzymes mirror the two
794: essential properties of the Lotka-Volterra model: (a)~Synthesis of
795: the peripheral enzymes for both substrates are autocatalytic (b)~The
796: peripheral enzymes for the two substrates inhibit each other.
797: \item The model by Brandt et al~\citep{Brandt2003} cannot capture non-diauxic
798: growth patterns. For all parameter values, the peripheral enzymes
799: for one of the substrates becomes extinct during the first exponential
800: growth phase, thus resulting in diauxic growth.
801: \item The models in~\citep{narang98b} and~\citep{Thattai2003} capture
802: both diauxic and non-diauxic growth patterns. Both models are identical
803: with respect to the mechanism that ensures that peripheral enzyme
804: synthesis is autocatalytic --- the peripheral enzymes promote the
805: synthesis of the inducer, which in turn stimulates the synthesis of
806: even more enzyme. However, they differ with respect to the mechanism
807: that produces mutual inhibition. In the Narang model, the mutual inhibition
808: occurs because each enzyme stimulates the dilution rate of the other
809: enzyme. In the Thattai \& Shraiman model, which applies to PTS sugars
810: only, the mutual inhibition stems from competition for phosphoryl
811: groups.
812: \end{enumerate}
813: \bibliographystyle{plainnat}
814: \bibliography{compAnalysis}
815:
816: \end{document}
817: