Biology Reference
In-Depth Information
Two such properties are discussed in this and the following
section.
The first is that a given phenotype is typically not just
formed by one or few genotypes, but by astronomically
many genotypes [54,56,84,86,87] . In other words, vast sets
of genotypes share the same phenotype. In some systems,
such as metabolism, it is possible to characterize such sets of
genotypes through Markov chain Monte Carlo sampling of
genotype space [54,56] . This involves carefully designed
random walks through genotype space. Briefly, one starts
from a specific metabolic network (metabolic genotype)
with a given number of reactions and a given phenotype.
(This starting genotype can be viewed as a single point in
metabolic genotype space.) Techniques such as flux-balance
analysis allow one to compute this metabolic phenotype
from the network's metabolic genotype [14,97] . One then
either eliminates a specific, randomly chosen reaction from
this starting genotype, or one adds a reaction chosen at
random from the known universe of biochemical reactions.
After this change to the network, one computes the pheno-
type of the changed network. If this phenotype is the same as
before the change
of metabolic networks. I will keep these two meanings of
a network distinct.
Genotype networks exist for metabolic phenotypes that
can sustain life on many different sole carbon sources, as
well as on multiple carbon sources (when each source is
provided as the sole carbon source). Even metabolic
networks that can sustain life on up to 60 carbon sources
form genotype networks that can still differ in 75% of their
reactions. Extended genotype networks also exist for
metabolic networks that contain different numbers of
reactions, and for phenotypes that involve sources of
chemical elements other than carbon [57] . Thus, the most
basic properties of genotype networks are not highly
sensitive to the phenotype one considers. Based on what we
know, they appear to be generic features of metabolic
genotype space.
To explore the genotype space of regulatory circuits one
can use similar sampling approaches [83,84] , and one finds
a similar organization of this space. Two circuits with the
same gene expression phenotype can have a genotype
distance between D
1, that is, they may
differ in 75 e 100% of their regulatory interactions. What is
more, circuits with very different genotypes can typically
be connected through a sequence of steps, each of which
changes a single regulatory interaction, but none of which
alters the circuit's gene expression phenotype. These
observations hold for broadly different gene expression
phenotypes, and regardless of the number of genes or
regulatory interactions in a circuit, except possibly for the
smallest circuits [84,98] .
Comparative data on the tertiary structure phenotypes
of proteins demonstrates the existence of genotype
networks here as well. Although exceptions exist [99] ,
proteins with the same structure and/or function can differ
in most of their amino acids [100
¼
0.75 and D
¼
that is, if this change has not altered the
ability of the network to sustain life on a given spectrum of
carbon sources
e
then the altered network is kept. Other-
wise, the genotypic change is discarded and one reverts to
the initial metabolic network. One then applies a second
change (reaction deletion or addition), evaluates the
phenotype, and keeps the altered network if it is unchanged,
and so on, in a long sequence of
e
10 5 reaction changes, each
of which has to keep the network's phenotype unchanged.
This approach can not only sample sets of genotypes with
a given phenotype uniformly, that is, in an unbiased manner,
it also resembles the process by which metabolic networks
evolve through the deletion and the addition of reactions to
a network, for example through horizontal gene transfer.
Using this approach one can ask how different two
metabolic genotypes that have the same phenotype can
become? The answer is that they can become very different.
For example, metabolic networks that have the same
number of reactions as the E. coli network, and that can
synthesize all E. coli biomass precursors in a minimal
environment that contains glucose as the sole carbon
source, can differ in more than 75% of their reactions.
Moreover, any two such metabolic networks can typically
be connected to one another. This means that sequences of
reaction changes exist that can convert one metabolic
network into the other, such that no individual change alters
the phenotype [54,56] . In other words, metabolic genotypes
with the same phenotype form extended networks
>
103] . Examples include
oxygen-binding globins. These proteins occur both in
animals and plants, probably share a common ancestor, but
are extremely diverse in their genotypes. For example, no
more than four of their more than 90 amino acids are
absolutely conserved. Despite this genotypic divergence,
globins have largely preserved their tertiary structure and
their oxygen-binding ability [104
e
106] .
Globins are not unusual in this regard. Other proteins
with preserved phenotype are even more diverse. Take
triose phosphate isomerase (TIM) barrel proteins. These
proteins have preserved their tertiary structure but can
differ in every single amino acid [107,108] .More
generally, proteins with highly diverged genotype yet
highly conserved phenotype are the rule rather than the
exception [100
e
e
genotype networks
in metabolic genotype space. Note
the distinction between a metabolic network and a geno-
type network: a metabolic network corresponds to a single
point, a single genotype in genotype space; a genotype
network is a network of such genotypes, and thus a network
103] . Such proteins form vast genotype
networks that extend far into genotype space. Phyloge-
netic analyses of related proteins from different organ-
isms reveal a reflection of these networks in the tree of
life [106] .
e
e
Search WWH ::




Custom Search