Atomistic Simulations of Ion Channels - Computational Neuroscience: A Comprehensive Approach - page 41

so far have been useful for a number of problems:

1. MD simulations have shown how the properties of water and ions in narrow
channels change significantly compared to bulk. In particular, in many cases
water molecules are strongly oriented due to local electric fields from the pro-
tein. Examples of this may be found in porin [112] and channels formed by
parallel helix bundles [111]. In addition, water and ion diffusion coefficients
are significantly lower than in bulk, which is relevant for coarse-grained sim-
ulations.

2. MD simulations have given insight into the actual process of ion motion in
potassium channels, as well as into local structural changes that may explain
the experimentally observed differences between e.g. sodium and potassium in
the potassium channel, or different types of ions in gramicidin A (see below).

3. MD simulations have been useful to construct models of channels for which
the structure is not known, when such simulations are combined with other
modelling techniques and experimental data (see [25] for a review).

4. MD simulations have begun to give detailed insight into the interactions of
small molecules and toxins with ion channels (e.g. [36, 46]).

5. MD simulations can be used to make models of states of ion channels that are
not present in crystal structures. The open channel models for KcsA of Biggin
and Sansom are a good example of this [15].

6. MD simulations give insight into the effect of the environment (lipids) on the
channel protein and vice versa (e.g. [85, 113]). This is an important aspect
because there are few other techniques available to study this directly.

7. MD simulations can be used to provide parameters and other information for
more course-grained simulations. This is potentially a very powerful use of
molecular dynamics simulations that has been applied in a number of cases.
Two recent examples can be found in interesting studies of OmpF [61] and
KcsA [23].

Nonetheless, there are several important caveats and limitations that have to be
taken into account. An obvious limitation is the combination of system size and
simulation length, which is mainly determined by available computer power and
software efficiency. In particular, the maximum time scale of ca. 100 ns is not enough
to accurately determine the average number of ions passing through a channel, except
for very wide channels such as porins or simplified geometrical models. This means
that by timescale alone one of our primary objectives, connecting atomic models with
current-voltage curves, is still mostly out of reach of MD simulations. A second
limitation is inherent in the specific choice of algorithms used. For example, it is
now quite clear that electrostatic interactions must be accurately calculated, but due
to their 1 r dependence (long-ranged compared to the system size) this entails a

certain degree of approximation. Several methods have been proposed (e.g. [114, 23]

/