Biology Reference
In-Depth Information
a protein
[34,35]
. The effects of such interventions on the
network components reflect the causal connectivity among
them. A perturbation in a single node propagates
throughout the network following its topology. By
measuring the response of other nodes, this propagator can
be determined. In perturbation analysis, the network
components (or a linearly independent set of components)
are sequentially perturbed to probe the network. By
combining the propagators obtained from independent
perturbations, the intrinsic structure of the system can be
used to predict the system response to different input data.
Whereas in correlative studies the size of the dataset
scales linearly with the network size, in perturbation studies
this relationship is quadratic, as the number of possible
connections scales in this way. Effective implementation of
network reverse engineering methods to resolve network
structure therefore demands a high-throughput automated
system for data acquisition. Recently developed fluores-
cence lifetime imaging microscopy on cell arrays
(CA-FLIM) is one such method
[32]
. Cells are plated on
microscopy-compatible chambers where chemical (inhibi-
tors) or genetic (siRNA, cDNA) perturbations have been
spotted in a high-density grid (up to 25 spots/mm
2
). Cells
take up the material locally from the spot, creating an
addressable array of perturbed samples. Fully automated
frequency domain FLIM is used to traverse the array
accurately determining in each spot the post-translational
modification state with subcellular resolution. As was
shown for the epidermal growth factor tyrosine phosphor-
ylation network, CA-FLIM can provide correlative data on
signal propagation. As FLIM measures the post-trans-
lational modification extent with spatial resolution,
CA-FLIM is the basis to resolve the local structure of
signaling networks in spatially regulated cellular processes.
The current ability to measure 10
4
samples per day
provides a way to derive causality in large networks (10
2
components) by perturbation analysis.
of large populations and their variation. If the underlying
agents and their interaction are known, a cellular automaton
offers a rapid approach to recreate the observed behavior in
silico on a single cell level. However, the prime challenge is
currently the inverse process. This means, can we compute
the network's causality if we know the topography of the
systems and the players involved, even from single cell
measurements? This would mean exploring the potential
behaviors of all possible network motifs in simulation and
then looking for distinguishing features that allow the
exclusion of certain motifs. For the example of an RTK
interacting with PTPs given above, that would mean
deducing a double-negative feedback from an observed
global activation pattern and excluding that for an observed
formation of activity hotspots.
As a further example of deducing a network architecture
from observed patterns, let us consider cell division in
Escherichia coli. To measure the size of the cell and at the
same time provide a spatial cue onwhere to divide, theMinD,
MinC and MinE proteins self-organize into a cell-spanning
oscillation of aMinDwave frompole to pole. A reconstituted
system of MinD, MinE and ATP proved sufficient to generate
2D-patterns in vitro
[36]
. This biological example is akin to
the BZ reaction and can be solved similarly to demonstrate
the power of modeling in deriving a network topology from
scant experimental knowledge. The reconstituted system in
vitro exhibits waves of membrane-bound MinD/MinE with
the following simplified characteristics:
MinD can bind to the membrane, slowing its diffusion.
l
MinD density increases slowly to a peak and drops more
steeply as seen from the direction of the traveling wave.
l
MinE density increases almost linearly towards the
trailing edge of the wave and drops off very sharply at
the end.
l
In this system we have two observables (MinD and MinE)
in two states (free and bound). In trying to model this
system, let us consider these two species in two states (free,
bound; in case of MinD to the membrane, in case of MinE
to MinD). This simple network therefore has four nodes
pairwise-linked by binding reactions, resulting in four
reaction rate constants, where each reaction rate can be
influenced by each node in a positive or negative way
(
Figure 17.3
). As long as no node influences its attached
reaction rate, this system comprises linearly coupled
differential equations, and a systematic check of these
possibilities shows that no combination leads to traveling
waves. Knowledge of the system further limits the amount
of sensible connections: e.g., the ATPase MinD is known to
bind to the PM in ATP-bound form and MinE mediates the
MinD-ATPase activity, thereby increasing MinD-PM
dissociation, and favorably binds to PM-bound MinD
dimers. By this reasoning, MinE can be considered an
antagonist to MinD-PM binding.
MODEL-DRIVEN EXPERIMENTATION AND
EXPERIMENTALLY DRIVEN MODELING
From the causality maps, further insight into the spatio-
temporal dynamics of biological processes can be obtained
by iteratively integrating experiments with simulations.
Partial information about the reactions involved in a bio-
logical process is used to simulate the biological process of
interest and the resulting insights inspire new experimental
observations. Biological parameters are obtained by fitting
the data to hypothetical models, and new experiments that
resolve the uncertainties/degeneracy of the models can be
then performed.
As described earlier, information about causation and
the agents involved can be extracted from the measurement
