Biomedical Engineering Reference
In-Depth Information
processes, no robust expression profiles, and hence no differentiated cell types.
In reality, genes interact with each other through the regulatory proteins they
encode. Each gene (or its encoded protein) has a very specific set of interaction
partners based on its molecular structure, and each interaction has a distinct
mode (e.g., stimulatory, inhibitory). Thus, the genome contains a hard-wired
interaction network. Interactions between genes therefore introduce constraints
in the whole network, such that many gene activation profiles become unstable
and are never realized. It is this collapse of the vast space of theoretically possi-
ble configurations of gene activity combinations that leads to distinct dynamics
and the robustness of a limited number of cell phenotypes (43).
In technical terms, constraint of the dynamics by these molecular interac-
tions means that the
high-dimensional state space
of gene activation is highly
structured. One can define a state space as the
N
-dimensional space in which
every point represents a different network state defined by a distinct gene activa-
tion profile, where
N
is the number of genes. Now assume that gene
A
uncondi-
tionally inhibits the expression of gene
B
; then all network states in which both
A
and
B
are active will be unstable, thus forcing the network to "move" in state
space until it hits a stable state. The network may also cycle between a few
states. Thus, taking all the interactions into account, it can be shown that the
network can change its activity profile in only a few directions (following stable
trajectories) until it reaches a stable state, the so-called "
attractor state
," which
can be a fixed-point or cycling attractor (42,43). The existence of unstable re-
gions and of multiple stable attractors impose a substructure to the state space,
which might be imagined as an "
attractor landscape
," as shown in Figure 4.
Accordingly, the state of the network (and hence of the cell) can be viewed as a
marble on that landscape: it is forced to roll along valleys (trajectories) into the
pits (attractors) (Figure 4). This attractor landscape therefore captures the con-
strained, global dynamics of cell fate switching (i.e., phenotypic control). In
fact, Waddington, Delbrück, Monod, Jacob, Kauffman, and others have all pro-
posed (in various forms) that the distinct, phenotypic differentiation states that
we observe in living systems correspond to attractors in the state space defined
by the molecular activities of the underlying network (15,42,48,68). Thus, at-
tractors in the state space map into stable phenotypic states (differentiation to
distinct cell types, cell proliferation, programmed cell death, etc.), and the trajec-
tories represent directed developmental processes.
In the landscape of a real gene regulatory network, the attractors would
represent cell states that are stable to many random perturbations. At the same
time, the network would allow the cell to switch to other attractors given the
appropriate sets of conditions, such as the presence of external regulatory signals
that promote a particular cell fate (30). This highly structured landscape with
latent, "preexisting" possibilities creates the stage on which the developmental
program is played out. Interestingly, Waddington similarly proposed an
"epigenetic landscape," with a marble whose position represents developmental
s
t
a
t
e
r
o
l
l
i
n
g
