Eur. J. Biochem.

270, 415421 (2003)  FEBS 2003

doi:10.1046/j.1432-1033.2003.03357.x

Stoichiometric network theory for nonequilibrium biochemical

systems
Hong Qian1,2, Daniel A. Beard2 and Shou-dan Liang3
Departments of 1Applied Mathematics and 2Bioengineering, University of Washington, Seattle, USA;
3
NASA Ames Research Center, Moett Field, CA, USA

We introduce the basic concepts and develop a theory for

nonequilibrium steady-state biochemical systems applicable
to analyzing large-scale complex isothermal reaction
networks. In terms of the stoichiometric matrix, we demonstrate both Kirchhos ux law R J 0 over a biochemical species, and potential law R l 0 over a
reaction loop. They reect mass and energy conservation,
respectively. For each reaction, its steady-state ux J can
be decomposed into forward and backward one-way uxes
J J+ J, with chemical potential dierence Dl RT
ln(J/J+). The product JDl gives the isothermal heat
dissipation rate, which is necessarily non-negative accord-

ing to the second law of thermodynamics. The stoichiometric network theory (SNT) embodies all of the relevant
fundamental physics. Knowing J and Dl of a biochemical
reaction, a conductance can be computed which directly
reects the level of gene expression for the particular
enzyme. For suciently small ux a linear relationship
between J and Dl can be established as the linear ux
force relation in irreversible thermodynamics, analogous to
Ohms law in electrical circuits.

With the completion of the Human Genome Project,

understanding of complex biochemical systems is entering a
new era that emphasizes engineering approaches and
quantitative analysis [1]. In order to develop a comprehensive theory for biochemical networks, Palsson and colleagues have utilized ux balance analysis (FBA) which is
based on the fundamental law of mass conservation [24].
In terms of general network theory, ux balance is
Kirchhoffs current law [5,6]. We recently augmented the
FBA with Kirchhoffs loop law [7] for energy conservation
[8,9], as well as the second law of thermodynamics. An
analysis combining FBA and energy balance analysis of
Escherichia coli central metabolism has provided signicant
insights into the regulation and control mechanism of the
biological system and improved the computational predictions from FBA alone [7].
The objective of the present work is to provide the energy
balance analysis a sound basis in terms of biophysical
chemistry. The earlier work provides one simple and one
realistic example for the general theory developed in the
present paper. Establishing a rigorous thermodynamic
theory for
living metabolic networks [10] challenges the
current theories of nonequilibrium steady-state systems. The
classic nonequilibrium physics, culminated in the work of
R. Kubo and L. Onsager, deals mainly with transient
processes and transport properties [11,12]. A living meta-

bolic network is sustained under a nonequilibrium steady

state [13,14] via active chemical pumping and heat dissipation; it is far from equilibrium [15]. The approach of
I. Prigogine and the Brussels group is formal and its
application to chemical reactions has been difcult and
controversial. Balancing chemical energy pumping and heat
dissipation in terms of the stoichiometry of a general
nonlinear reaction network is the focus of the present work.
In rigorous physical chemistry, one of us recently has
developed a nonequilibrium statistical theory which addresses particularly the stochastics as well as thermodynamics of
isothermal nonequilibrium steady states [1619]. In the
work of Katchalsky et al. [9], the thermodynamics of
biochemical networks have been explored. In particular,
Oster and Perelson [20] have developed an algebraic
topological theory of Kirchhoffs laws in terms of chains,
cochains, boundary and coboundary operators. But in
terms of the stoichiometric matrix, no explicit, practically
useful formulae have been obtained. The objective of the
current work is to present the mathematical basis of energy
balance analysis for complex reaction networks with
nonlinear graphs. (A simple linear graph can be represented
sufciently and necessarily by an incidence matrix,
nodes edges, which has 0 and 1 as elements and exactly
two 1s per column. Nonlinear reaction networks cannot be
represented by a simple linear graph; rather they have to be
represented by nonlinear graphs, also known as hypergraphs.) The essential idea is to introduce the metabolite
chemical potentials into the biochemical network analysis.
While the theory of basic isothermal nonequilibrium
processes can be found in Hill 1974, and Qian 2001, 2002
[13,16,18], its application to metabolic networks, combining
with FBA, has led to a simple and powerful optimization
approach [7]. In addition, we also establish the connection
between the laws of physical chemistry and Kirchhoffs laws

Correspondence to H. Qian, University of Washington, Seattle,

98195-2420, WA, USA.
E-mail: qian@amath.washington.edu
Abbreviations: SNT, stoichiometric network theory; FBA, ux balance
analysis; cmf, chemical motive force; hdr, heat dissipation rate.
(Received 1 July 2002, revised 4 September 2002,
accepted 7 November 2002)

Keywords: biochemical network; chemical potential; ux;

nonequilibrium thermodynamics; steady-state.

 FEBS 2003

416 H. Qian et al. (Eur. J. Biochem. 270)

for circuits. Finally, the practical aspects of such analysis in

biochemistry is discussed.
The paper is organized as follows. Basic concepts pertinent
to the steady-state stoichiometric network theory (SNT) such
as biochemical reaction ux, chemical potential, conductance, one-way uxes and heat dissipation are introduced
using simple examples. In particular, we discuss nonlinear
relationships between reaction ux and chemical potential
and its linear approximation. The latter is analogous to the
Ohms law in electrical circuits and is widely known as the
linear ux-force relationship in irreversible thermodynamics.
The materials in this section are closely related to the work of
T. L. Hill [13,14] and Westerhoff and van Dam [10]. In the
following section the loop law is introduced via an example of
a simple nonlinear reaction. Next we present the general
proof of the loop law in terms of an arbitrary stoichiometric
matrix. Combining the ux and loop laws, as well as an
inequality for heat dissipation, the thermodynamically
feasible null space of a stoichiometric matrix can be
signicantly restricted. In the case of no external ux
injection and concentration clamping, a zero null space
consistent with chemical equilibrium is uniquely determined
from the three constraints alone. The last section gives a
brief discussion of SNT and its future direction.

Basic concepts
Nonlinear flux-potential relationship
Most of the basic concepts used in the SNT can be found in
classic treaties [10,13,14]. For readers unfamiliar with the
work on nonequilibrium steady-state biochemical reactions,
we introduce some of the necessary essentials using simple
examples. We begin our analysis with the simplest possible
uni-molecular biochemical reaction with balanced input and
output steady-state uxes J:
k1

J
J

*
)

 B

!


! A 
k1

Conceptually, the SNT is the generalization of this simple

example to complex nonlinear reaction networks in terms of
their stoichiometric matrices. The steady-state solution to
reaction (1) is:
cA

k1 cT J
;
k1 k1

cB

k 1 cT  J
k1 k1

which can be obtained from the kinetic equation in terms of

the law of mass action:
dcA
k1 cA k1 cB J;
dt

dcB
k1 cA  k1 cB  J
dt
3

cA + cB cT is the total concentration for molecules in A

and B states. In steady state the total number of A and B
molecules is conserved and the uxes are balanced.
Without ux J, the reaction in Eqn (1) approaches
chemical equilibrium with cB/cA k1/k)1. We focus on the
energetics of reaction (1) in the nonequilibrium steady state
with a nonzero external ux J (also known as boundary
ux). In this case, the chemical potential difference between
states A and B is:

 
 
cB
J
Dl DloAB RT ln
RT ln
cA
J

where T is the temperature, R is the gas constant, and DloAB is

the reaction chemical potential in standard state. The ux is
decomposed into forward and backward components
J J+ ) J where J+ k1 cA, J k)1 cB. Hence
J
k1 cB
J k1 cA which is known as the mass-action ratio.
RT 2.48 kJmol)1 at room temperature of 298.16 K
and Dl have units of kJmol)1. Units for concentrations c,
ux J, and rst-order rate constant k are M, Ms)1, and s)1,
respectively. Substituting Eqn (3) into Eqn (4) gives:


k1 cT Jk1
DlAB J RT
;
k1 cT  Jk1
 Dl =RT

5
k1 k1 cT e AB
1
J
k1 k1 eDlAB =RT
We see that the DlAB is a nonlinear function of J. However,
when J 0, Dl 0 and vice versa. More importantly,
JDl is always non-negative and equals zero if and only if
J Dl 0. JDl is the rate of heat dissipation of the
reaction in nonequilibrium steady state [13,18]. When
J 0, the reaction is at its thermodynamic chemical
equilibrium.
For small J we can take rst-order Taylor expansion for
Eqn (5) at J 0 and obtain a linear relation


RT 1
1
DlAB J 
J;
6

cT k1 k1
which is analogous to Ohms law. The linearity is only valid
when J/cT  k1, k)1. According to OnsagerHills theory
on uni-molecular cycle kinetics [13,21], the linear
Ohms
resistance rAB DlAB/J, is directly related to the equilibrium one-way ux in the absence of nite J. That is, in the
absence of J, the equilibrium probabilities pAeq and pBeq of
a single molecule are in detailed balance, the one-way ux
J k1 pAeq J k1 pBeq (Onsagers reciprocal relation),
and


1
1
RT
eq :

7
rAB RT
k1 k1
p A k1
For biochemical network analysis, it is interesting and
important to note that the conductance, i.e. 1/rAB, is
linearly proportional to the concentration of the particular
enzyme which catalyzes the A B reaction, i.e. k1 and k)1
are, at rst-order approximation, proportional to the
expression level of the enzyme [E]. A gene regulation or
enzyme activation changes a particular network resistance
but does not directly effect the chemical potential difference
of the reaction per se!
It is also important to keep in mind that beyond the linear
regime, the biochemical resistance is not symmetric. In Eqn
(5) DlAB (J) DlAB (J). Such nonlinear behavior is
similar to that of diodes in electric circuits, which have
played a pivotal role in electronic circuit technology.
We now use a second example to demonstrate how the
linear result can be useful in analyzing networks with
balanced inux and efux. We consider the more complex
situation of two reactions in series:

 FEBS 2003

Nonequilibrium stoichiometric biochemical network theory (Eur. J. Biochem. 270) 417

k1

k2

k1

k2

J
J

*

*
)

 B 
)

 C

!


! A 

We have in the steady-state:

k1 k2 cT k1 k2 k2 J
k1 k2 k1 k2 k1 k2
k1 k2 cT k1  k2 J
cB
k1 k2 k1 k2 k1 k2
k1 k2 cT  k1 k1 k2 J
cC
k1 k2 k1 k2 k1 k2
and in the linear regime:


RT k1 k2 k1 k2 k1 k2
J
DlAB 
cT
k1 k1 k2
RT
 eq J rAB J
p A k1
cA

and



RT k1 k2 k1 k2 k1 k2
J
cT
k1 k2 k2
RT
 eq J rBC J
p B k2

DlBC 

In equilibrium thermodynamics, the chemical potential

Dl of a system is related to its partial molar enthalpy
h and entropy s: l h ) Ts. While the enthalpy
observes the law of conservation of energy, the l does
not [22]. In the equilibrium Dl is zero for each and every
reaction in the system. In a nonequilibrium steady-state,
the l, h, and s for each species do not change. However,
Dl for each reaction is no longer zero. These chemical
potential differences are maintained by an external
chemical pumping (in terms of the J in Eqn 1). The
work done by an external agent (e.g. a battery), which
equals precisely the chemical potential difference of the
reaction in a steady state, is the amount of heat
dissipated in the steady state. Therefore, the energy
conservation is between the chemical work done to a
system and the heat dissipated by the system [22]. The
amount of energy dissipated can be computed in terms
of the chemical potential differences in the system. We
emphasize the difference here between the equilibrium
Gibbs free energy of a reaction and the chemical
potential of its species in an nonequilibrium isothermal
steady-state.

10

in which the rAB and rBC are the linear resistances

introduced in Eqn (7), as expected. Therefore, in the linear
regime, the biochemical ux in a particular reaction and the
chemical potential difference across the reaction are linearly
related via the biochemical resistance.
Flux decomposition and biochemical heat dissipation
In FBA, the net ux in each reaction is determined based on
the Kirchhoffs ux law as well as certain optimization
criterion [2]. While the net ux is extremely important for
mass conservation, it does not provide sufcient information on biochemical energetics. For each reaction in a
biochemical network:

External flux injection and internal flux distribution

When a throughput ux is injected into a biochemical
system, how is it distributed throughout the entire
network? What is the most probable pathway? These
problems are well understood in terms of linear network
theory, but require further analysis for nonlinear biochemical networks. In this section we discuss these basic
questions using simple examples and suggest that some
of the results from linear analysis can be applicable to
nonlinear systems. A comprehensive study of this subject
will be published elsewhere.
We start with the simple three-state kinetic cycle with
detailed balance:

k1


*
A 
)

 B
k1

with nonequilibrium steady-state concentration cA and cB,

the heat dissipation rate (hdr) of this reaction is [13,17]


k1 cA
11
JDl RTk1 cA  k1 cB ln
k1 cB
in which the net ux J k1cA k)1cB is the turnover per
unit time, and the Dl RT ln(k)1cB/k1cA) the chemical
potential change of turnover per mole. One can decompose
J into J+ and J, which can provide information on its
energetics in biochemical network analysis [13,17]:
 
J
;
J J  J ; Dl RT ln
J
 
12
J
hdr RTJ  J ln
0
J
The last inequality is the second law of thermodynamics:
One cannot derive useful work entirely from a single
temperature bath. By summing over all the reactions in a
biochemical network, this energy formula can be used to
compute the total heat dissipation rate [18].

(13)

in which a nonzero throughput ux J is introduced. In

steady-state:
cA

cB

cC

CT
k2 k2 k3
J

k1 k2
D
1 kk11 k1
k2
k1
k1 CT
k1
k1 k2
k1 k1 k2

k2 k3 k3
J
D

k1 k2
k1 k2 CT
k1 k2
kk11 k1
k2

k2  k3
J;
D

where
D k1 k3 k3 k1 k1 k3 k1 k3 k3 k2
k1 k1 k3 k2 :

 FEBS 2003

418 H. Qian et al. (Eur. J. Biochem. 270)

The uxes are given by

k1 k2 k2 k3 k1 k2 k3 k3
J
D
k2 k3 k2 k3 k2 k3
J:
JCB
D

JAB
JAC

When J 0, all uxes are zero. The ratio JAB/JAC

JAB k1 k2 k2 k3 k1 k2 k3 k3

k2 k3 k2 k3 k2 k3

JAC
k1 k2 k1 k3 rAC rCB

k2 k3
rAB
again as expected from the Ohms law. This result is
surprising, since this equality is true even for large J.

k1 k2 k3
1
k1 k2 k3
known as the thermodynamic box in chemistry or detailed
balance in physics. The steady state is then a chemical
equilibrium with zero ux: k1cA ) k)1cB k2cB )
k)2cC k3cC ) k)3cA 0. Therefore, DlAB DlBC
DlCA 0, as expected for three resistors in a loop with
no battery (neither current source or voltage source).
Now let us assume that the system is open and the
reaction from B to C is really a second order with
charging:
k02 D


*
B 
)

 C
k02 E

Kirchhoffs loop law for chemical potentials:

an example
We now introduce the loop law for chemical potentials.
This is parallel to the Kirchhoffs voltage law for electrical circuits. Note that the Kirchhoffs voltage and current
laws are independent from the Ohms law which assumes
a linear relationship between the current and voltage.
Kirchhoffs laws are much more fundamental than that
of Ohms.
The Kirchhoffs voltage law in electrical circuit is due to
the fact that electrical energy has a potential function and
no curl. This is also the case for chemical reaction
network: for each species in the network, it has a uniquely
dened chemical potential. This is the origin of our loop
law. The loop in a network (a graph) is formally
equivalent to a closed curve in Euclid space. This
equivalence is best seen between a continuous physical
model and a lattice model. A graph is a generalization of
a high-dimensional irregular lattice. In fact, the boundary
ux and clamped concentration could be considered as
analogies to inhomogeneous Dirichlet and Newmann
boundary conditions, respectively.
To demonstrate the essential idea, we rst use simple
cyclic chemical reactions as examples. Both unimolecular
and more importantly nonunimolecular reactions will be
considered. A general proof for arbitrary topology (stoichiometric matrix) will be given later.
We rst consider the simplest cyclic, unimolecular
reaction with three states:

(14)

When the system is closed and there is no external

ux injection or concentration clamping, the microscopic
reversibility dictates that:

in which the cofactors D and E have xed concentrations,

o
[E] as k2 and
[D] and [E]. (One can treat the k2o [D] and k2
k)2 if the bimolecular reaction is rate limiting.) An example
of such a situation is a reaction accompanied by ATP
hydrolysis with D and E representing ATP and ADP,
o
respectively [23,24]. In this case k2o and k2
are second-order
rate constants. The equilibrium constant for the reaction
D $ E is
Keq

E
eq
D
eq

ko2 B
eq
ko2 C
eq

ko2 B
eq A
eq
A
eq ko2 C
eq

k1 ko2 k3
k1 ko2 k3

When [D] and [E] are not at their equilibrium, the amount of
energy in this reaction with xed concentrations, or
equivalently the amount of work needed to maintain the
concentrations, is


Keq D

DlDE RT ln
E

Now again consider the cyclic reaction in Eqn (14), with

pseudo-rst-order rate constants k2 and k)2, we have:
DlAB DlBC DlCA  DlDE 0

15

in which the rst three terms are chemical potential

differences due to a nonzero ux J, and the last term is
the energy, or more precisely
chemical motive force (cmf),
of a biochemical battery. If the ux J is running from
A ! B ! C ! A then the rst three Dl are negative. We
call Eqn (15) the law of energy balance. It is formally
analogous to Kirchhoffs loop law. Multiplying J throughout Eqn (15), we have energy conservation: chemical
work dissipated heat.
As a function of the driving force DlDE, the steady-state
cycle ux in the cyclic reaction is [23,24]



k1 ko2 k3 E
eDlDE =RT  1
:
k1 k3 k3 k1 k1 k3 k1 k3 k3 ko2 D
k1 k1 k3 ko2 E

16

 FEBS 2003

Nonequilibrium stoichiometric biochemical network theory (Eur. J. Biochem. 270) 419

This relationship is highly nonlinear. However for small



 
1
1
k2
1
1
k1
1
1
k2 1 DlDE
J

k1 k1 k1 k2 k2 k2 k1 k2 k3 k3 k2 k3

RT
rAB rBC rCA 1 DlDE

17

in which the rAB, rBC and rCA are the linear resistances, as in
Eqns (9) and (10). Eqn (17) observes the law of serial
resistors.

Energy balance analysis in terms of

stoichiometric matrix for nonlinear
reaction networks
Generalization of the above results on energy balance to
networks with arbitrary topology is not trivial. While it is
straightforward to identify reaction cycles in a system of
unimolecular reactions [13,14], it is not clear how to dene
loops for networks of multispecies biochemical reactions.
Here we discuss the methodology for imposing energy
balance in complex biochemical networks [7].
Consider a system of N + N metabolites Xi
(i 1,2,. . .,N
are
dynamic
concentrations
and
i N + 1,N + 2,. . .,N + N are clamped concentrations) with M + M biochemical reactions (j 1,2,. . .,M
for internal reactions and j M + 1,M + 2,. . .,M + M
for external boundary uxes). The jth internal reaction is
characterized by a set of stoichiometric coefcients mji and jji
in the form
jj1 X1

jj2 X2

   jjNN0 XNN0

kj

j1

is an overall reaction with the left and right sides identical:

M
N
M
N
X
X
X
X
mj
jji Xi 
vj
mji Xi
j1

k

i1

N
X

j1

Xi

M 
X

i1

20

21
loi

RT ln[Xi] is the chemical potential of ith

Dlext
j

NN
X0

S^ij li

iN1

is the cmf for the jth reaction. Combining Eqns (20) and (21)
and recalling S~ v 0, we reach the conclusion that:
X
vj Dlj  Dloj 0
j
ext

if Dl 0, i.e. when there is no externally clamped

concentration. Furthermore in equilibrium, according to the
thermodynamic box expression, for equilibrium constants,
P
it can be shown that j1 mj Dloj 0 (since for any steadyP
state j mj Dlj  Dloj 0; it has to be valid for equilibrium in which all Dlj 0) Hence we have:

..

..


jji  mji vj 0:

j1

18

in which some of the integers m and j can be zero. If Xn is an

m
enzyme for the reaction m, then mm
n jn 1. More
complex MichaelisMenten kinetics can also be expressed
in terms of Eqn (18).
The jth boundary ux is characterized by the same
Eqn (18) but all m and j are zero except one. A nonzero
m(j) corresponds to an inux (efux).
The stoichiometry of this set of reactions can be
mathematically represented by the (N + N) (M + M)
incidence matrix S fjji  mji g [5,6,2527]:

0
 . 1
..
..
 ..
.
.

C
B
B    S~NM     SNM0 C
B
 . C
..
..
SB
 .. C
C
B
.
.
C
B
A
@    S^ 0

NM
j 0

i1

The chemical potential difference of the jth reaction is

expressed as:
!
QNN0
jji
N
X
o
i1 Xi

Dlj Dlj RT ln Q
S~ij li Dlext
j
j ;
0
NN
mi
X

i1
i
i1
where li
species and

mj1 X1 mj2 X2    mjNN0 XNN0

balance. All internal uxes are necessarily cycles [16,29],

although these cycles may not be intuitively obvious for
nonlinear reactions involving many species.
We denote the expression of jth reaction in Eqn (18) by
(Rj). For each vector belonging to the cycle space
v (v1,v2,. . .,vM) in the null space NS~, the expression:
M
X
mj Rj
19

The lower-right 0 block indicates that there should be no

boundary ux to or from a clamped species. Matrix S is the
starting point of the FBA (which also assumes no clamped
metabolites, S^ 0 [2,28]) as well as other modeling
approaches such as metabolic control analysis (MCA).
The matrix S~ contains only the dynamic species and internal
uxes. The null space of the NS~, consists of all the
possible internal ux distributions which satisfy ux

M
X

vj Dlj 0

22

for any v in the null space NS~.

In general, in addition to the external boundary uxes
being held constant, a steady-state network can also have
chemical energy, i.e. cmf, supplied through clamped
concentrations of certain species. In this case, Dlext 0
and Eqn (22) becomes v(Dlint Dlext) 0 over the
reaction loop v (e.g. DlDE in Eqn (15) is external).
The law of energy balance restricts the NS~ to a smaller,
thermodynamically feasible subspace [7]. Now let

 FEBS 2003

420 H. Qian et al. (Eur. J. Biochem. 270)

v1, v2,. . ., vm be the m linearly independent vectors of the

thermodynamically feasible null space, and dene the
matrix K vT1 ; vT2 ; . . . ; vTm where vT denotes the transpose
of the row vector in a column form. Then we obtain an
novel algebraic structure for the biochemical network
theory:
S~K 0;

S~Jint bext ;

Dlint K pext ;

23

in which matrices S~ and K are known as incidence and loop

matrices in graph theory [5,30]. Vectors Jint and Dlint are
internal uxes and chemical potentials, both M dimensional.
bext SJext is a N-dimensional external ux (typically with
many zero components) and Jext is M-dimensional.
pext lext S^K is the external cmf on reaction loops, and
N-dimensional lext is determined by the externally
clamped species. The second and third equations in (23) are
Kirchhoffs ux law and potential law, respectively. Finally,
int
in steady-state, heat dissipation rate Jint
j Dlj  0 for
int
int
individual jth reaction and hdr J Dl  0 for an
entire network, where the equals sign holds true if, and only
if, the reactions are in chemical equilibrium. This is the
second law of thermodynamics [18].
Eqn (23) indicates that, in a biochemical network analysis
that avoids detailed reaction rate constants, the steady-state
ux and potential are on an equal footing. In optimizing
uxes, an idea originated by Palsson and his colleagues, one
should proceed both analyses in parallel and enforce the
second law. The classical FBA does not determine J+ and
J separately. Interesting biological questions concerning
the nature of biochemical control follow from the above
analysis: are metabolic networks sustained in nonequilibrium steady state by constant boundary ux, or by clamped
concentrations? We believe that providing answers to such
engineering questions will further deepen our understanding
of the regulation and control of metabolism and other
complex biochemical processes.
The practical value of the energy balance relation and nonnegative hdr is to further provide thermodynamic constraints
in the FBA with optimization. The introduction of chemical
potential signicantly restricts the null space of S~, NS~. In
the case when a network without any external ux and
clamped species: S~ Jint 0 ) lS~ Jint DlJint 0. Since
every term Dlj Jj 0, one has Dlj Jj 0 for all j. Hence
Dl Jint 0. Therefore, the only null space vector J that
satises ux balance, energy balance, and non-negative hdr is
zero, as expected for a chemical equilibrium.

Discussion
We have presented the concepts of SNT which serves the
foundation for analyzing nonequilibrium steady-state uxes
and energetics in biochemical systems. Cornerstone concepts of this theory are ux balance and energy balance, or
equivalently mass and energy conservation. While ux
balance can provide useful predictions of biochemical uxes
[4] it alone cannot sufciently restrict the solution space to
guarantee thermodynamically feasible uxes [7]. Energy
balance introduces proper thermodynamics into network
analysis while simultaneously providing quantitative information on control and regulation [7]. Theoretical tools such
as these, which are rmly rooted in rigorous biophysical

chemistry, are essential to the development of computational and bioinformatic protocols for analysis, simulation,
and design of complex biochemical systems.
The SNT developed in this paper provides a unique
conceptual framework for network analysis of large-scale
metabolic reaction systems. We expect tools from electrical
circuit analysis and nonlinear graph theory will soon
signicantly enhance the practical usefulness of this
approach. On the theoretical side, an integration of SNT
with existing theories on metabolic system analysis might
also be possible. We give several possible directions for the
future development of SNT.
Modular analysis of interactions between passive
and active subnetworks
Engineering analysis of large-scale complex systems requires
that one understands such systems in modular terms [31].
Toward this end, we dene the basic concepts of passive and
active biochemical subnetworks and examine the consequences of interactions between such subnetworks.
By a passive subnetwork, we mean the collection of the
reactions in the network that contain neither boundary uxes
nor clamped concentrations. All the species in the passive
subnetwork are dynamic, and all the internal uxes are
balanced. In this case, by a simple analogy with a subnetwork
of resistors, there should be no current loops in this
subnetwork. When all the connectivity between this subnetwork and the remaining network is severed, the subnetwork
approaches an equilibrium with zero uxes. Such a subnetwork is a passive component in a nonequilibrium steady
state. In contrast, an active subnetwork has to involve either
xed concentration or inux and efux. Such a subnetwork is
an open system with energy utilization and heat dissipation.
When a passive subnetwork is coupled to an active one,
the uxes pass through the former. A fundamental result
from Hills theory on cycle kinetics states that that there
should be no cycle ux in the passive subnetwork. Hence, an
active subnetwork cannot induce cycle ux in a passive one
[13,16]. A passive subnetwork can support only a transit ux
distribution.
SNT with MichaelisMenten kinetics
In the present analysis, we have assumed that the metabolic
kinetics follow the law of mass action. In cells, metabolic
reactions that involve enzymes can all be represented by a
stoichiometric matrix. (MichaelisMenten kinetics is an
approximated solution to the general enzyme kinetic models
based on the law of mass action.)
SNT analysis which takes the enzymatic reaction into
specic consideration is currently in progress. However, it is
important to rmly establish the physiochemical foundation
of the SNT rst based on the complete chemical kinetics.
Relation to MCA
MCA [3236] is a systematic approach to measure, both
theoretically and experimentally, the control imposed by a
metabolic network upon a particular ux. This evaluation is
done in terms of the concept of ux control coefcients,
which are dened as the fractional change in a ux induced

 FEBS 2003

Nonequilibrium stoichiometric biochemical network theory (Eur. J. Biochem. 270) 421

by a fractional change in an enzyme activity. A second type

of quantities central to MCA is the elasticity coefcient, the
fractional response of the rate of a reaction to a fractional
change in concentration of a metabolite in steady state.
We are currently developing the relations between these
important quantities and SNT.
In closing, we shall comment on the two alternative but
complementary approaches to metabolic network modeling. The traditional approach is based on rate constants
and kinetic equations. FBA with optimization, recently
introduced by Palsson and his colleagues and is now
integrated with energy balance in SNT, articulates an
optimization approach based on a (or several) biological
objective functions. The two approaches could be viewed
from a historical perspective as
Newtonian and
Lagrangian, respectively [37]. The challenge for the reductionistic
former is to obtain the detailed information on rate
constants and kinetic equations, while for the integrative
latter it is to discover cellular principles in terms of
optimalities, if they indeed exist.

Acknowledgements
H. Q. thanks J. S. Oliveira for an enlightening conversation and V. Hsu
for many helpful discussions. This work is supported in part by
National Institutes of Health grants NCRR-1243 and NCRR-12609,
and National Aeronautics and Space Administration grant NCC25463.

References
1. Stephanopoulos, G. (1994) Metabolic engineering. Curr. Opin.
Biotechnol. 5, 196200.
2. Edwards, J.S. & Palsson, B.O. (1998) How will bioinformatics
inuence metabolic engineering? Biotechn Bioeng. 58, 162169.
3. Schilling, C.H. & Palsson, B.O. (1998) The underlying pathway
structure of biochemical reaction networks. Proc. Natl Acad. Sci.
USA 95, 41934198.
4. Edwards, J.S. & Palsson, B.O. (2000) The E. coli MG1655 in silico
metabolic genotype: its denition, characteristics, and capacities.
Proc. Natl Acad. Sci. USA 97, 55285533.
5. Balabanian, N. & Bickart, T.A. (1981) Linear Network Theory:
Analysis, Properties, Design and Synthesis. Matrix, Beaverton, OR.
6. Strang, G. (1986) Introduction to Applied Mathematics. WellesleyCambridge Press, Wellesley, MA.
7. Beard, D.A., Liang, S.-D. & Qian, H. (2002) Energy balance
for analysis of complex metabolic networks. Biophys. J. 83,
7986.
8. Feynman, R.P., Leighton, R.B. & Sands, M.L. (1963) The Feynman Lectures on Physics. Addison-Wesley, Redwood City, CA.
9. Oster, G.F., Perelson, A.S. & Katchalsky, A. (1973) Network
thermodynamics: the analysis of biophysical systems. Quart. Rev.
Biophys. 6, 1134.
10. Westerho, H.V. & van Dam, K. (1987) Thermodynamics and
Control. of Biological Free-Energy Transduction. Elsevier,
Amsterdam.
11. Katzir-Katchalsky, A. & Curran, P.F. (1965) Nonequilibrium
Thermodynamics in Biophysics. Harvard University Press,
Cambridge.
12. Kubo, R., Toda, M. & Hashitsume, N. (1978) Statistical Physics
II: Nonequilibrium Statistical Mechanics. Springer-Verlag, New
York.

13. Hill, T.L. (1974) Free Energy Transduction in Biology: the SteadyState Kinetic and Thermodynamic Formalism. Academic Press,
New York.
14. Hill, T.L. (1989) Free Energy Transduction and Biochemical Cycle
Kinetics. Spinger-Verlag, New York.
15. Nicolis, G. & Prigogine, I. (1977) Self-Organization in
Nonequilibrium Systems. Wiley Intersci., New York.
16. Qian, H. (2001) Mathematical formalism for isothermal linear
irreversibility. Proc. R. Soc. Lond. A. 457, 16451655.
17. Qian, H. (2001) Nonequilibrium steady-state circulation and heat
dissipation functional. Phys. Rev. E. 64, 022101.
18. Qian, H. (2002) Entropy production and excess entropy in a
nonequilibrium steady-state of single macromolecules. Phys. Rev.
E. 65, 021111.
19. Qian, H. (2002) Equations for stochastic macromolecular
mechanics of single proteins: equilibrium uctuations, transient
kinetics, and nonequilibrium steady-state. J. Phys. Chem. 106,
20652073.
20. Oster, G.F. & Perelson, A.S. (1974) Chemical reaction dynamics.
Part II: Reaction networks. Arch. Ration. Mech. Anal. 57, 3198.
21. Hill, T.L. (1982) The linear Onsager coecients for biochemical
kinetic diagrams as equilibrium one-way cycle uxes. Nature 299,
8486.
22. Crabtree, B. & Nicholson, B.A. (1988) Thermodynamics and
Metabolism. In Biochemical Thermodynamics (Jones, M.N., ed.),
Elsevier Scientic, Amsterdam, pp. 359373.
23. Qian, H. (1997) A simple theory of motor protein kinetics and
energetics. Biophys. Chem. 67, 263267.
24. Qian, H. (2000) A simple theory of motor protein kinetics and
energetics II. Biophys. Chem. 83, 3543.
25. Peusner, L. (1986) Studies in Network Thermodynamics. Elsevier,
New York.
26. Poland, D. (1991) Indicator matrices for potential instabilities in
open systems. J. Chem. Phys. 95, 79847997.
27. Alberty, R.H. (1991) Equilibrium compositions of solutions of
biochemical species and heats of biochemical reactions. Proc. Natl
Acad. Sci. USA 88, 32683271.
28. Oliveira, J.S., Bailey, C.G., Jones-Oliveira, J.B. & Dixon, D.A.
(2001) An algebraic-combinatorial model for the identication
and mapping of biochemical pathways. Bullet. Math. Biol. 63,
11631196.
29. Schuster, S. & Schuster, R. (1991) Detecting strictly balanced
subnetworks in open chemical-reactions networks. J. Math. Chem.
6, 1740.
30. Kalpazidou, S. (1994) Cycle Representation of Markov Process.
Springer-Verlag, New York.
31. Hartwell, L.H., Hopeld, J.J., Leibler, S. & Murray, A.W. (1999)
From molecular to modular cell biology. Nature 402, C47C52.
32. Kacser, H. & Burns, J.A. (1973) The control of ux. Symp. Soc.
Exp. Biol. 27, 65104.
33. Heinrich, R. & Rapoport, T.A. (1974) A linear steady-state
treatment of enzymatic chains. General properties, control and
eector strength. Eur. J. Biochem. 42, 8995.
34. Fell, D.A. & Sauro, H.M. (1985) Metabolic control and its analysis. Additional relationships between elasticities and control
coecients. Eur. J. Biochem. 148, 555561.
35. Giersch, C. (1988) Control analysis of metabolic networks. 1.
Homogeneous functions and the summation theorems for control
coecients. Eur. J. Biochem. 147, 509519.
36. Kholodenko, B.N. & Westerho, H.V. (1995) The macroworld
versus the microworld of biochemical regulation and control.
Trends Biochem. Sci. 20, 5254.
37. Goldstein, H. (1980) Classical Mechanics. Addison-Wesley,
Reading, MA.