Atomistic Simulations of Ion Channels - Computational Neuroscience: A Comprehensive Approach - page 52

Figure 2.6
Models and a simulation system for alamethicin K18. A, B: starting model; C: full
simulation system, including salt, lipids and water.

Reproduced with permission

from [115].

The maximum anion selectivity of the putative octameric conducting state is ob-
tained at pH 7 or lower. Since no change in selectivity is seen between pH 7 and
pH 3, and since protons are expected to be in equilibrium with the open state of the
channel during a selectivity measurement, the channel should be fully charged (i.e.,
all 8 lysines protonated) at pH 7. An example of single channel I-V measurements
is shown in Figure 2.5. This work poses several questions that simulations might be
able to address. First, what is the structure of these channels? Because of the sim-
ple sequence of alamethicin we can be somewhat more confident of models of this
channel than of models of complex physiological ion channels. A modeled structure
is shown in Figure 2.6.
Second, can we link the models to the measured pH-dependent selectivities? Third,
why is the channel not more selective for anions, even with 8 charged lysines point-
ing into the channel? To address these questions a number of computer simulations
of the system has been performed, of 10 ns each of the octameric bundle in a lipid
bilayer environment, with either 0, 4, or 8 lysines charged in the absence of salt, and
with 8 lysines charged in the presence of 0.5 M or 1 M KCl. Without salt present
and with all lysines charged, on average 1.9 Cl − ions are inside the channel and the
channel significantly deforms. With 0.5 M KCl present, 2.9 Cl − ions are inside the
channel. With 1 M KCl present, 4 Cl − ions are present and the channel maintains
a regular structure. Poisson-Boltzmann calculations on the same system showed the
effect of ionic strength on the calculated electrostatic potential in the channel ( Figure
2.7) . The barriers in these graphs can be linked to a conductance through the Nernst-
Planck equation. Clearly, the results are rather sensitive to the exact algorithm used