![](https://csdnimg.cn/release/download_crawler_static/88358541/bg11.jpg)
between evolution rate and dispensability becomes insignificant. These
observations have been taken to falsify the fitness-rate conjecture and have
led to the notion that, apart from a gene's specific biological function, expres-
sion level is the principal determinant of its sequence evolution rate (Pal et
al. 2003; Rocha and Danchin 2004).
However, this was not the end of the road for the intuitively attractive,
more or less natural notion that important genes (i.e., those that cannot be
knocked out without impairing the organism) should evolve slowly. Using
more extensive and refined omic data, four independent studies have re-exam-
ined the cOlmections between gene expression, dispensability, and evolution
rate (Drummond et al. 2005; Lemos et al. 2005a; Wall et al. 2005; Zhang and
He 2005). The first three studies analyzed substantially improved, genome-
wide data sets on expression levels and dispensability of yeast genes. These
studies also employed genome sequences from closely related yeast species
to obtain accurate estimates of short-term evolution rates approximating the
desired instantaneous rate. The fourth group performed similar analyses on
the omic data from another model organism, Drosophila (Lemos et al. 2005a).
Using these improved data sets and applying sophisticated statistical tech-
niques to account for the experimental noise, all four groups concluded that
there was a relatively weak but highly significant correlation between the evo-
lution rate and dispensability, even after controlling for the expression level.
Notably, the effect of dispensability on the evolution rate diminished rap-
idly with increasing evolutionary distance (Wall et al. 2005; Zhang and He
2005). The reverse test for the effect of dispensability yielded a significantly
stronger dependence of evolution rate on expression level (Drummond et
al. 2005; Wall et al. 2005; Zhang and He 2005). Three groups concluded that
the effects of expression level and dispensability on gene evolution rate were
independent, and that both were substantial (Lemos et al. 2005a; Wall et al.
2005; Zhang and He 2005). The contribution of expression level appeared to
be greater than that of dispensability, but it has been posited that uncertain-
ties regarding the noise level in the data made definitive conclusions on the
relative importance of the two variables premature (Wall et al. 2005). By con-
trast, Drummond and coworkers held that the correlation between gene dis-
pensability and evolution rate still might be spurious, depending on the noise
levels for different variables (Drummond et al. 2005; Drummond et al. 2006).
Thus, although the latest analyses with updated data sets generally seem to
uphold the fih1ess-rate conjecture, a shadow of doubt persists.
We addressed the controversy around the connection between gene dispen-
sability and evolution rate in some detail in order to demonstrate the complex
state of affairs that is characteristic of current research in evolutionary systems
biology. Very similar developments have taken place with regard to other cor-
relations between omic variables (see Table 2.2). For example, the relation-
ship between node degree (the number of nodes with which a given node is
connected) in protein-protein interaction networks and evolution rate became
the subject of an even more confounding debate following the original claim
that highly connected genes evolved slowly (Fraser et al. 2002). A re-examina-
tion of the data suggested that the link between connectivity and evolution
rate, even if real, was extremely weak; only the most prolific interactors (net-
work hubs) were characterized by low evolution rate (Jordan et al. 2003). An
analysis of an updated data set by the authors of the original observation
appeared to vindicate the link (Fraser et al. 2003), but then further reanalysis
showed that the correlation disappeared altogether after correcting for protein
abundance (Bloom and Adami 2003; Bloom and Adami 2004; Drummond et
al. 2006). Nevertheless, an independent, detailed study of protein-protein inter-
action networks from three species seemed to provide support for the notion
that a gene's centrality in a network (defined as betweenness, i.e., the frequency
with which a network node lies on the shortest path between pairs of other
nodes, a quantity that strongly correlates with connectivity) constrains its evo-
lution (Hahn and Kern 2005). A similar analysis of gene coexpression networks
(see a detailed discussion of these networks later in this chapter) revealed a
sh'ong correlation between node degree and evolution rate (Jordan et al. 2004).
The structure of the yeast metabolic network has been reported to similarly
affect evolution of enzymes (Vitkup et al. 2006). Thus, despite the general attrac-
tiveness of the idea that network hubs are biologically important, slowly evolv-
ing genes, the validity of this connection remains an open issue.
A different, more biologically inspired way to address the relationships
between genome-related variables is to step back from technical problems
and consider the general structure of the correlations (see Table 2.2). The
genome-related variables naturally fall into two classes that can be roughly
defined as phenomic (expression level, protein abundance, dispensability,
centrality in networks, and so forth) and evolutionary (sequence evolution
rate and propensity for gene loss). Considering this simple classification,
the pattern of correlations is perfectly clear: the correlations within a class
are consistently positive whereas those between classes are typically nega-
tive (see Table 2.2 and Figure 2.2) (Wolf et al. 2006a,b). The phenomic vari-
ables show a "the bigger the better" pattern whereas, with the evolutionary
variables, it is "the slower the better": Genes with high expression levels, a
large number of interactions, and a drastic knockout effect typically evolve
slowly and are not lost during evolution. In itself, this coherence does not
exclude the possibility that some of the variables have no independent sig-
nificance but are, instead, fully determined by one or more genuinely impor-
tant quantities (e.g., gene expression level, which is most strongly correlated
with the rate of evolution, might induce most, if not all, of the correlations
between other phenomic and evolutionary variables). Alternatively, the pat-
tern depicted in Figure 2.2 could reflect a genuine synergistic effect of the
different variables within each class. In particular, it stands to reason that
distinct characteristics of gene function might combine to determine the rate
of evolution. This notion has been encapsulated in the so-called extended
complexity concept, according to which the rate of a gene's evolution is, to
a large extent, determined by the complexity of that gene's functional inter-
actions: Genes that are involved in complex processes and multiple interac-
tions tend to evolve slowly (Aris-Brosou 2005).