The Cost of Modularity 261
components, for example, enzymes and metabolites, may occur in almost even distribution
within the cell (though there will in fact be considerable heterogeneity and even compart-
mentalization). Such a network can nevertheless be decomposed into modules, either
according to structural or to functional criteria. The older method is functional decomposi-
tion: cell metabolism is decomposed into capacities that are brought about by metabolic
pathways such as glycolysis, the citric acid cycle, beta-oxidation, catabolic and anabolic
pathways in amino acid metabolism, and so forth. The metabolic pathways, then,
are regarded as functional units or modules. Some of these pathways may be regarded
equally as structural modules in the sense introduced in section 15.1: the strength of
internal interactions and the weakness of external interactions allows for their separation.^3
However, this does not hold true in general, since functional relations do not guarantee
independency of the pathways in the sense of near-decomposability, as the many interac-
tions among the pathways prove. This so-called crosstalk with other pathways often
turns out to encompass stronger interactions to the metabolic surroundings than can be
found within the pathway under consideration. The functional delineation of pathways
within a metabolic network need not coincide with structural modules.^4 Consequently,
functional analysis as a method to identify modules in biological systems is not undisputed
and often regarded as biased (Rohwer, Schuster, and Westerhoff 1996; Koza et al. 2002;
Friedman 2004; Papin, Reed, and Palsson 2004; for a discussion see Krohs and Callebaut
2007).
To avoid functional bias, and to end up with a picture that accounts for the structural
near-decomposability of the system, metabolic networks have been delineated more
recently according to the strength and relevance of static relations, as well as to dynamic
interactions. Both together constitute the structure of a network. The importance of the
dynamic dimension for the structural picture was emphasized by Simon (1969: 198):
the short-run behavior of each of the component subsystems [i.e., modules] is approximately inde-
pendent of the short-run behavior of the other components;... in the long-run the behavior of any
of the components depends in only an aggregate way on the behavior of the other components.
Though this delineation criterion looks as if it were straightforwardly applicable, its opera-
tionalization is diffi cult. Sophisticated mathematical methods had to be developed to allow
for a breakdown of a network into structural modules that satisfy the criterion. They are
neutral with respect to functional considerations and often seem to end up with a picture
of the organization of a network that differs signifi cantly from the results of functional
decomposition.
Only the results of structural decomposition are generally regarded as being capable of
delivering an authentic picture of a network (Bechtel and Richardson 1993; Schaffner
1998; Onami et al. 2002; Papin, Reed, and Palsson 2004; Palsson 2006).^5 I therefore stay
with structural modules for the main part of my argument, returning to the possible rele-
vance of functional modules in evolution in sections 15.6 and 15.7.