Page 40 - YORAM RUDY BOOK FINAL

P. 40
extracellular domain). These motions cause pore opening by moving the S4-S5 linker (which con-

nects the voltage sensor to the pore) upward and away from the pore, providing space for the S6
tails to move away from each other, opening the pore (Fig.2.23). Simulations of the voltage sensor
gating revealed that it is constantly moving within the configuration space (along stochastically

different trajectories) rather than transitioning among few stable conformations. Therefore, the
notion of stable conformations may refer to continuous regions in the configuration space where
the voltage sensor can reside with high likelihood. The computed dynamics are visualized for one
representative motion trajectory in Movie 2 (https://youtu.be/6lwrXsCzY-8). Single-channel current
traces, computed from the simulated motion trajectories during gating, have stochastic properties

similar to experimentally recorded traces. Macroscopic current through an ensemble of channels
displays an initial delay of activation, as in experimentally recorded macroscopic current traces.

Atomistic-Scale Simulations of Ion-Channel Structure – Function

A voltage-gated ion channel undergoes atomistic structural changes that are stochastic
in nature during gating. Relating these dynamic changes to the channel function as a
transmembrane charge carrier in atomistic detail remains a formidable computational challenge.

The physiological time scale of most ion-channels function is in the range of milliseconds to
seconds. Even with customized hardware, simulations of 1 ms gating dynamics take months of
computer time and require nonphysiological conditions, e.g., a membrane voltage of 500 mV 118,119 .

Yet, understanding the structure – function relationship is crucial for elucidating normal and
disease mechanisms and for developing genetics-based precision medicine. In the previous
section, a reductionist approach was used to compute I gating, considering only large backbone
Ks
movements and reduced number of degrees of freedom, and assuming tetrameric symmetry of
the four voltage-sensors movement and concerted (cooperative) gating for channel activation

(i.e., all four voltage sensors must be in their activated position for the channel to open). The
simulations included only the transmembrane segments of the proteins. In later simulations, we
relaxed these simplifying assumptions and simulated I activation at the atomistic scale 120,121 ; these
Ks
simulations are summarized below.

To overcome the impossible computational challenge, we employed an artificial
intelligence (AI) machine learning (ML) approach in the simulations . Figure 2.24 provides a
120
schematic diagram of the simulation strategy. First, a library of I structures (~ 3,000,000
Ks
conformations) was constructed (using Kv1.2 crystal structure as homologous template ) to
122
sample the conformational space that I spans during gating. Energy was minimized and steric
Ks
clashes were removed. Many conformations in the library had small deviations from the

recently published cryogenic electron microscopy (cryo-EM) KCNQ1 structure . Selected I
123
Ks
attributes (features) that alter protein energy (e.g., voltage-sensor height and pore diameter)
were extracted from the library, along with the computed energy of each corresponding structure.
Using these data, the ML algorithm was trained to predict I energy of structures outside the
Ks
library. With this approach, the energy landscape covered the entire I conformational space
Ks