Supplementary information for

“Ion conduction and conformational flexibility of a bacterial voltage-gated sodium channel”.

Céline Boiteux1, Igor Vorobyov2 and Toby W. Allen1,2

1School of Applied Sciences & Health Innovations Research Institute, RMIT University, GPO Box 2476, Melbourne, Vic, 3001, Australia

2Department of Chemistry, University of California, Davis. Davis, CA 95616, USA.

Supplementary Methods Systems

The NavAb protein (residues 1 to 221, X-ray crystallographic coordinates from pdb 3RVY (1)) was embedded in a bilayer of 336 (167 top and 169 bottom) dipalmitoylphosphatidylcholine (DPPC) lipids (chosen because it is the best studied lipid bilayer in Molecular Dynamics (MD) simulations (2)), with explicit water molecules (~20,200 molecules) and ions (described below), to form a simulation box of 125x125x76 Å containing 119,259 atoms. Residue C217 was mutated back to the original isoleucine on all four monomers.

All possible protonation states were considered for the EEEE ring formed by residues E177 in the selectivity filter (SF; Fig.1c), generating 6 different systems named PS0 (no glutamates were protonated), PS1 (one of the glutamates protonated), PS2 (A/B) (2 glutamates protonated on adjacent/opposite monomers respectively), PS3 (3 of the glutamates protonated) and PS4 (all glutamates protonated). These 6 systems were built in the presence of 150 mM NaCl. Additionally, PS0 and PS1 were simulated in mixed solutions: NaCl/KCl at 300/300 mM and NaCl/CaCl2 at 150/300 mM (see Table S1). A ratio of 2 between the concentrations of Na+ and Ca2+ was applied in order to obtain similar ionic activity for both species in bulk electrolyte to allow approximate matching of free energy profiles (with reduced activity coefficient for the divalent ion estimated approximately using Debye-Huckel theory, in agreement with experimentally reported values (3, 4)). Simulation Details

All systems were built and pre-equilibrated with CHARMM (version c34a1) (5), using the C36 lipid (2) and C22 protein (6) parameters with CMAP corrections (7), TIP3P water (8) and modified ion parameters (described below). After 1000 (4x250) steps of steepest descent minimization, MD simulations commenced with a timestep of 1 fs and initial harmonic restraints (10 kcal/mol/Å2) applied on all heavy atoms. These restraints were slowly released over the course of 2.5 ns, followed by 5 ns of simulation without any restraints, using a timestep of 2 fs. The systems were then equilibrated for an additional 20 ns using NAMD (version 2.9) (9) and the same force field. All simulations were performed at constant pressure (1 atm) (10, 11), with fixed lateral area (125.4x125.4Å2 in order to maintain the right area per lipid obtained after CHARMM equilibration) and constant temperature of 323K (chosen to avoid the gel

phase transition of DPPC lipids), using a Nosé-Hoover thermostat (12, 13). All bonds to hydrogen atoms were maintained using the SHAKE algorithm (14). Electrostatic interactions were computed using Particle Mesh Ewald (15), with grid spacing of 1 Å and 6th order B-spline for mesh interpolation. Non-bonded pair lists were updated every 20 steps with a cutoff distance of 15 Å and a real space cutoff distance of 12 Å was used with energy switch (switching distance of 10 Å). Anton software version 2.7.2 from D. E. Shaw Research was used for production runs of all systems, using the purpose-built Anton supercomputer (16). These simulations were carried out using tetragonal periodic boundary conditions in the NVT ensemble at 323 K, a 2 fs time step with non-bonded long-range interactions computed every 6 fs using the RESPA multiple time step algorithm (17). The Nose-Hoover algorithm (12, 13) was used for temperature coupling, whereas the Gaussian split Ewald (GSE) method (18) was used for handling long-range electrostatic interactions with Gaussian RMS width parameters σ and σs used for Ewald splitting and charge spreading, with the grid size and spreading radius optimized for each simulation using Guesser scripts for an initial structure. For Lennard-Jones (LJ) interactions, a 12.72 Å cutoff was used, optimized by Anton simulation scripts for these systems. A long-range LJ correction (beyond cutoff) was also used as default in the Anton software. A detailed description of the simulation methodology employed in this study can be found at http://www.deshawresearch.com/downloads/download_desmond.cgi/Desmond_Users_Guide-0.5.3.pdf.

Modified LJ parameters were used to describe the interactions between the ions Na+ and K+ and carbonyl oxygens of residues 175-178 to reproduce their correct free energies of solvation in N-methylacetamide (19-21). While this correction is important to ensure correct solvation free energies in narrow protein pores, such as found in K+ channels, they are expected to have only small influence on ion energetics in the wider NavAb SF, due to reduced coordination by backbone carbonyl oxygen atoms. Standard CHARMM ion LJ interactions were otherwise used, leading to hydration free energies of -103.1 ± 0.1 kcal/mol for Na+ and -82.5 ± 0.1 kcal/mol for K+, differing by -20.6 ± 0.2 kcal/mol. We note that modified ion LJ parameters exist (-18.6 kcal/mol; in better agreement with experiment (22)), differing by 2 kcal/mol. These parameters will also influence interactions with protein, and thus need not necessarily change the free energies of binding to the same extent.

Mixed solutions of ions were simulated in two stages, due to the closed gate, to allow ions to occupy all sites. We simulated with 2 different initial occupancy states: NaIN-KHFS/KIN-NaHFS and NaIN-CaHFS/CaIN-NaHFS, for different lengths of time (0.5 to 0.9 µs) for both PS0 and PS1, resulting in 8 simulations of similar weight, combined to produce 4 simulations of 1.1 to 1.4 µs (see Table S1). Analysis Ion occupation, distributions and free energy surfaces:

To calculate channel ion occupation (Fig.S1), we computed the mean and standard deviation of the number of ions within the limits of the pore defined by: xy-radial position ≤ 10Å, and z-axial position between -15 and +12.5 Å; where z is defined relative to the center of mass (COM) of the backbone of the filter (atoms N, C, O and Cα of residues 175 to 178).

Each free energy surface (Fig.2&3 and Fig.S3&6), or potential of mean force (PMF), was calculated from unbiased simulation as 𝑊 𝑅! =  −𝑘!𝑇𝑙𝑛 𝜌 𝑅! + 𝐶 , where 𝜌 is the unbiased probability distribution as a function of reaction coordinates Ri, being axial or radial positions of one or more ions, and C is a constant. Errors on the free energy maps were estimated as the RMS between maps built from first

and second halves of the sampled trajectory. The free energy barrier between two states was defined by the maximum along the path of lowest free energy connecting the states. To determine the path of lowest energy, the space between two states was divided into slabs of 0.2 Å along x and y. The local minima were identified for each of these slabs. The pathway was then constructed by joining the closest minimum in the next slab.

The 2D PMFs for Na+ permeation for the PS1 protonation state, not described in the main text, are shown in Fig.S6. In the presence of 2 ions for the simulation of PS1 (Fig.S6a), three states A, B2/B′2 and F2 can be identified, with 2 additional states E2 and H2. This map shows a transition from A (with one ion in high-SHFS and one ion in SIN) to F2 (where the top ion is in SCEN and the bottom ion has left to the cavity), via B2/B′2 (with one ion in low-SHFS and one ion in SCEN or SIN, respectively). The SHFS site can still accommodate 2 ions at the same time (state H2), but is less likely (barrier of 1.9 kcal/mol vs. 0.4 in PS0); a feature that seems characteristic of the fully charged state. Instead, the ions may pass each other down at the level of SCEN-SIN (state E2, barrier 1.1 kcal/mol). However, ions are bounded from the top at z=6Å essentially ruling out a 2-ion mechanism, even for the PS1 state. The 3-ion state (Fig.S6b) is once again represented by z1 (bottom ion) and z23 (COM of the middle and top ions). Three states A, B3 and F3 can be identified, where A is common to both occupancy states. During the transition from A to B3, ion 3 enters the upper part of the filter and pushes ion 2 down, leading to state F3, with ion 1 in the cavity, via a knock-on mechanism. Fig.S6c shows the 2D PMF in the presence of 2 ions before significant deformation of the SF (t < 1 µs). It is comparable to the map obtained for the whole simulation (panel a), with similar stable states and transitions.

Movies S1 and S2 show exchanges of Na+ ions within the SF during the simulations of PS0 and PS1, respectively. They illustrate the remarkable flexibility of the NavAb SF, in particular the rotational isomerizations of residues E177, as well the varying occupancy and Na+ coordination by protein and water molecules.

Hydrogen bonding in the SF: Fig.S2 shows the H-bonding changes observed during simulation with protonation states PS2A and PS2B. In PS2A, protonated E177 of adjacent monomers A and B form stable H-bonds with unprotonated E177 (C) during the second half of the simulation (t > 400 ns; measured as the distance between carboxylic acid H and carboxylate O atoms), effectively blocking the access of the lower part of SF to Na+ ions. In PS2B, protonated residue E177(C) forms a stable H-bond with adjacent unprotonated residue E177(D) throughout the whole simulation. Additionally, H-bond E177-S178 of monomer B becomes very stable during the second part of the simulation (t > 350 ns; shown in the lower panel of Fig.S2). These two H-bonds create a partially blocked SF, with no Glu side chain left free to create a stable site for Na+ ions, and is unlikely to conduct. Fig.S3b and c show unbiased PMFs for PS2A and PS2B respectively during the first (conductive) and second (non-conductive) parts of the simulations (with panel a showing the case for PS1, discussed in the main text). In both cases, the second panel displays a reduction in ion density indicating a loss of conduction.

Fig.S4a shows time series for the H-bonds between carboxylate oxygens of E177 and the hydroxyl hydrogen of S178 of the same subunit, in PS0 and PS1 simulations, revealing stages of breaking of H-bonds that instigated an increase in ion occupancy in PS0, and higher mobility of E177 side chains in PS1. Fig.S4b shows the normalized distributions of H-bonds, as well as the distribution of dihedrals χ1, χ2 and χ3 of residues 177 (first, second and third side-chain dihedrals of residue E177, formed by atoms N-CA-CB-CG,

CA-CB-CG-CD and CB-CG-CD-OE1 respectively) during the simulations of PS0 and PS1. According to these distributions, the filter of PS0 is more stable and the side chain of residues E177 less mobile than in PS1, maintaining H-bonds with S178 and presenting few isomerizations. In contrast, the four E177 side chains display additional orientations for χ1 and χ2, and a significantly higher mobility of the carboxyl group (χ3) in the PS1 simulation. Fig.S4c compares the distribution of side chain dihedrals as a function of ion occupancy in PS0. Fig.S5 shows changes in conformations for PS0 and PS1 simulations, involving discrete changes in H-bonding within the SF as described in Fig.S4.

Competitive binding in the SF:

Fig.S7 compares time series of the position along the z-axis of ions in the SF (-15≤ z ≤15 and r ≤ 10) for PS0 and PS1 in pure NaCl (a) and in mixed solutions of NaCl/KCl (b) and NaCl/CaCl2 (c). Free energy profiles were calculated separately for simulations 1 and 2 for each species (Na+, K+ and Ca2+), and for each protonation state PS0 (Fig.4) and PS1 (Fig.S8). Fig.S9 shows the convergence of these PMFs by cumulative block data analysis. For each curve in Fig.S9, the simulation time analyzed was 0.2, 0.4, 0.6, 0.8 and 1.0 of the total simulation time (listed in Table S1).

Ion coordination numbers (Fig.S10) were calculated as the number of water or protein oxygen atoms within 3.2 Å of Na+ and Ca2+ ions, and within 3.6 Å for K+, determined from the position of the first minima in the respective radial distribution functions for those ions in bulk electrolyte, and were then averaged over trajectory frames where an ion existed in a particular site of interest. An ion was considered as occupying a site if its position along the z-axis was within 1.5 Å of the center of the site (zNa+=3.0, -1.0, -3.5 and zK+=2.0, -1.0, -4.0 in SHFS, SCEN and SIN respectively, and zCa2+=1.5 in SCa2+, identified from Figure 4). The error in each value corresponds to one standard deviation. Protein structural changes and fluctuations: Protein structure was analyzed via root mean square deviation (RMSD) in structure and represented as state maps of the pore domain (PD) and SF. The state maps (Fig.5) represent the RMSD between all pairs of conformations obtained during the simulations after alignment of the part of the protein considered: backbone of residues 175 to 178 for the SF and backbone of residues 130 to 220 for the PD. A measure of fluctuation in dihedral angles φψ is defined at any time t for each residue as ∆𝜑𝜓 𝑡 =𝜑 𝑡 − 𝜑 !

!+ 𝜓 𝑡 − 𝜓 !!, with 𝜑 ! and 𝜓 ! being average values of the dihedral angles during

the simulation for each residue. Fig.S11a shows the value of that deviation for residues 196 to 208 of a single monomer during simulations of PS0 and PS1. In PS0, residues V196, F198 and P200 appear more flexible, because of the helix-breaking behavior of the proline, which destabilizes the V196-P200 portion of S6. I202 is also slightly shifted from ideal values. PS1 is the representation of a more extreme case, and while it displays the same characteristics than PS0 at the beginning of the simulation, it soon shows some drastic deviations not only around V196-P200 block, but also at the level of F203, where the bending of helix S6 happens.

Fig.S11b shows the average fluctuations over all monomers for residues 196 to 208 and for each protonation state. It shows that the behavior described in the time-series of one monomer of PS0 can be generalized, with the larger fluctuations occurring at the level of residues 196 to 200 and 202. The larger deviation observed in PS1 at the level of F203 is also reproduced in PS2B. These fluctuations appear to be inherent to the protein and not random movements.

θS6 refers to the bend of helix S6 around residue F203 and is defined as the angle between vectors ut

and ub, where ut characterized the orientation of the top of the helix and is defined as the vector between the COM of the Cα of residues 194 to 196 and 202 to 204, and ub characterized the orientation of the bottom part of the helix and is defined as the vector between the COM of the Cα atoms for residues 202 to 204 and 210 to 212 (see Fig.S11d). H-bonds I202-206 and I202-T206 (SC) refer to the carbonyl oxygen of I202 and the amine or hydroxyl (SC) hydrogen of T206 within the same monomer. Fig.S11a shows the correlation between the length of H-bond I202-T206 (SC) and the value of θS6, and Fig.S12 compares the geometric characteristics of H-bonds I202-T206 and I202-T206 (SC) in PS0 to those of a reference, taken at 3 different points along a stable transmembrane helix S5 (α-helix forming H-bonds between residues 133-137, 135-139 and 141-145 in PS0). It shows that the presence of the additional side chain H-bond perturbs the α-helix forming H-bond: its length and angle are slightly shifted toward less favorable values compared to the reference, which would affect the stability of the helix at that level. Movie S3 illustrates S6 bending around F203 in the PS1 simulation.

SF structural changes during the PS1 simulation are associated with a bend of helix S6, as shown in Fig.S13. Residues I202 and V205 change their orientation and press on the base of helix P1 via residues V173 and M174 (Fig.S13b&c), while a tilt of helix P2 (calculated as the angle between the vector formed by Cα atoms of residues 181 to 183 and 188 to 190 and the xy plane) is correlated with the degree of bending of S6 in the same monomer. These two effects contribute to the dislocation of a subunit of the SF (see main text).

Gate asymmetry was measured as the difference between the distances separating the COM of the Cα of residues 215 to 218 on pairs of opposite monomers. Fig.S14 represents the distribution of the asymmetry of the permeation gate for all protonation compared to the asymmetry exhibits by the X-ray structure of the proposed inactivated state (pdb 4ekw) (23). It shows that the asymmetry displayed in the X-ray structure is within the fluctuations observed during all our simulations, asserting the potential significance of those fluctuations for NavAb.

Phe203 Isomerizations and fenestration size:

During all simulations, the tails of some of the lipids were seen entering the cavity fenestrations for extended periods of times, supporting the role of those openings as binding sites or pathways toward the cavity for hydrophobic drugs.

The radius of the fenestration was measured using CAVER 3.0 (24). The starting point for the identification of portals was placed in the hydrophobic cavity, defined as the COM of residues 203 and 204 of all four monomers, and a spherical probe with a minimum radius of 1 Å was use to explore the opening. The system was limited to the pore domain (residues 130 to 221) and the structures were aligned using the backbone of the PD as reference. Analysis was repeated every 20 frames in PS0 and PS1, with the fenestration size corresponding to the radius at the bottleneck for each frame (Fig.S15a). The size of the opening is modulated by the bending of helix S6 but is mostly linked to the isomerizations of dihedral angle χ2 (CA-CG-CD-CG1) of residue F203, even in the absence of bending (Fig.6). Angle χ2 was chosen rather than χ1 as the former offers a better sampling of the orientations of the phenyl group and its position relative to the opening.

The placement of a phenytoin molecule in the opening (Fig.S15c) was realized manually, with the purpose of comparing sizes of the opening and the molecule only. A comparison of the radius of a benzene molecule (2.47 Å, based on Pauling atomic radii and bond lengths (25)) and the distributions of the fenestration radius demonstrates that an aromatic ring will fit within the natural variations of the size of the

opening. The fenestration is larger than an aromatic ring 9.5 and 12.5% (PS0 and PS1 respectively) of the time, corresponding to around 1.2-1.3 standard deviations, or just 0.7-0.8 kBT in energy, assuming an elastic model. Thus, the fluctuations would potentially allow the passage of drugs such as benzocaine, lidocaine or phenytoin into the pore. Further studies will be required to quantify drug binding and pathways.

Supplementary Tables

Simulations in NaCl (150 mM) Simulation PS0 PS1 PS2A PS2B PS3 PS4

Time (µs) 2.5 2.5 0.7 0.9 1.1 1.2

Simulations in mixed solutions Simulation PS0 K+/Na+ PS0 Ca2+/Na+ PS1 K+/Na+ PS1 Ca2+/Na+

Concentration (mM) 300/300 300/150 300/300 300/150

Time (µs) 0.5, 0.6 0.7, 0.6 0.5, 0.9 0.5, 0.5

Table S1: Ion concentrations and simulation times for each Anton simulation involving different protonation states (defined in the main text), in pure Na+ and mixed ionic solutions. Where two independent simulations were carried out, their times are comma-separated.

Supplementary Figures

Figure S1: Occupancy of the SF by Na+ ions. Number of Na+ ions in the filter during the simulations of protonation states PS0 (solid light blue) and PS1 (dashed dark blue) (top). Average number of ions in the filter for each protonation state of the EEEE ring (bottom).

Figure S2: Representations of SF conformations during simulations of NavAb for protonation states PS2A (top) and PS2B (bottom) at t~0.45 and t~0.4 µs respectively (H-bonds indicated with red dotted lines; not all H atoms are shown for clarity) contributing to the transition of the SF to non-conductive states. On the right, times series corresponding to the H-bonds are shown (OE1-blue dots, OE2-green dots).

Figure S3: Sample time breakdown of radial-axial Na+ free energy surfaces for protonation states PS1 (a), PS2A (b) and PS2B (c). For PS1, the two panels show the PMFs before (1/2) and after (2/2) significant change of conformation in the SF (t~1µs). For PS2A and PS2B, the two panels show the PMFs before (1/2) and after (2/2) the stabilization of a network of H-bonds that causes the SF to transition into non-conductive conformations (transitions at t~400 and 350 ns respectively), as described in the main text.

Figure S4: SF fluctuations. a) Sample time series of E177-S178 H-bonding for each monomer in the PS0 (left) and PS1 (right) simulations (OE1-blue dots, OE2-green dots). (b) Normalized distributions of the length of H-bond E177-S178 and of the value of dihedral angles χ1, χ2 and χ3 of residue E177 for each monomer A, B, C and D during simulations of PS0 and PS1. The protonated residue (PS1 A) is represented in red and the charged ones in shades of blue. (c) Distribution of the dihedral angle χ2 of residue E177 for the 2- (red) and 3-ion (green) occupancy states, with a representation of the orientation of the side-chain of E177 at the main peaks, corresponding to initial “up” orientation where the H-bond between E177 and S178 is maintained (χ2 ~ -70 deg) and the flipped “down” orientation toward the center of the filter (χ2 ~ 60 deg).

Figure S5: Representations of SF conformations during the simulations of NavAb in protonation states PS0 (top) and PS1 (bottom). For clarity, only 3 monomers are shown in most representations, except for PS1 t~1.2 µs and 2.1 µs, where all monomers are represented (with 2 monomers B and C at t~1.2 µs, and 3 monomers D, A and B at t~2.1 µs of interest in color). Hydrogen bonds are indicated with grey dashed lines, with A, B, C and D indicating which monomer the residue belongs to (clockwise from top).

Figure S6: 2D PMF projections for Na+ ions in PS1, when 2 (a) or 3 (b) ions occupy the SF. Each contour corresponds to 0.5 kcal/mol. Positions z1, z2 and z3 of ions 1, 2 and 3 (bottom to top) along the axis of the pore are relative to the SF backbone COM (residues 175 to 178). Reaction coordinate z23 corresponds to the position of the COM of the upper two ions (2 and 3). Stable occupancy states are labeled A, B2, B2’, E2, F2 and H in the 2-ion occupancy state, and A, B3 and F3 in the 3-ion occupancy state. The insets show the states (A, B2, B2’, F2, E2, H2, B3 and F3) identified on the maps, with their associated free energies relative to conformation A, common to both occupancy states. Panel (c) shows the 2D PMF projection when 2 ions occupy the SF, before any significant deformation of the filter (t<1µs). In this period, occupation by 3 ions was reduced, such that only the 2-ion PMF is replotted.

c

Figure S7: (a) Time series showing the z-position of Na+ ions for PS0 (left) and PS1 (right). Panels (b) and (c) show times series for mixtures of NaCl with KCl and CaCl2 solutions, respectively. Here, Na+ (yellow, orange and red lines) and K+ or Ca2+ (green, blue and purple lines) are plotted for PS0 (top) and PS1 (bottom), for independent simulations 1 and 2, as indicated in each panel.

Figure S8: Superimposed PMFs for Na+ (magenta), K+ (green) and Ca2+ (blue) ions during simulations 1 (solid lines) and 2 (dashed lines) of PS1 in mixtures of Na+/K+ (right) and Na+/Ca2+ (left), with binding sites indicated. Initial ion positions for the independent simulations are indicated with insets.

Figure S9: PMFs for Na+ (magenta), K+ (green) and Ca2+ (blue) ions during independent simulations 1 and 2 (as indicated in each panel) for PS0 (a, right) and PS1 (b, left) in mixtures of Na+/K+ (top) and Na+/Ca2+ (bottom), showing their convergence by cumulative block data analysis (see SI Text).

Figure S10: Coordination of ions. Average number of oxygen atoms (NOx) from water molecules and protein found within the coordination sphere of Na+ (magenta), Ca2+ (blue) and K+ (green) ions in the binding sites they occupy during the simulations of PS0 (top) and PS1 (bottom) in mixed solutions (Ca/Na, left, and K/Na right).

Figure S11: Bending of helix S6. (a) Times series showing the φψ deviation from average values for residues 196 to 208 (helix S6) of representative monomers during the simulation of protonation states PS0 and PS1. (b) Average φψ deviation from average values for residues 196 to 208 (helix S6) during the simulation of each protonation state PS0 to PS4. PD fluctuations. (c) Bending angle θS6 as a function of the length of the hydrogen bond between residues I202 and T206 in the bent S6 helix during the simulations of PS1 (blue squares), PS2 (purple circles) and PS4 (orange triangles), with corresponding representations of the monomer shown. (d) Representation of angle χ2(203) (red), angle θS6 (blue) and H-Bond I202-T206 (green), as described in the SI methods.

c d

a b

Figure S12: (a) Residue T206 (red) and its environment of non-polar side chains (green and blue residues), orienting its polar side-chain toward the backbone of residue I202 (orange). Normalized distribution of the length (b) and angle (c) of the H-bonds formed by reference residues on S5 in PS0 (red), and I202-T206 carbonyl-amide (green) and carbonyl-hydroxyl (blue) in PS1.

Figure S13: Structural changes associated with S6 bending. A) Superposition of the maximally bent monomer from PS1 (yellow) with the X-ray conformation (gray) reveals bending near residue F203 (orange sticks), and is associated with a tilt of the top pore helix, P2, and an upward displacement of the SF subunit (red coil). Note that the particular graphical representation used exaggerates S6 deformation for illustration purposes. Panels (b) and (c) show the arrangement of residues supporting this displacement in X-ray (gray) and bent (yellow) conformations. Panels (d) and (e) show times series of the tilt of P2 (left) and the bend of S6 (right) during PS0 and PS1 simulations, respectively, with raw data as green points and smoothed fit as solid red lines.

d e

Figure S14: Asymmetry of the activation gate. Normalized distribution of the difference of the distances between the COM of CA of residues 215 to 218 on opposite monomers. The black dashed line shows the corresponding value from pdb 4EKW (chains AB), a potentially inactivated structure (23).

Figure S15: (a) Radii of the 4 inter-subunit fenestrations during PS0 (left) and PS1 (right) simulations. (b) Illustration of a closed fenestration showing the opening formed by segment A (pink) and D (grey) in PS0 at t ~ 1.6 µs (as indicated in a), closed by the side chain of F203 (yellow) on segment A. (c) Representation of an open fenestration formed by segments D (pink) and C in PS1 at t ~ 1.1 µs. The side chain of F203 is flipped down, and the portal is large enough to accommodate a phenytoin molecule (colored sticks).

Supplementary References

1. Payandeh J, Scheuer T, Zheng N, Catterall WA (2011) The crystal structure of a voltage-gated sodium channel. Nature 475(7356):353–358.

2. Klauda JB et al. (2010) Update of the CHARMM all-atom additive force field for lipids: validation on six lipid types. J Phys Chem B 114:7830-43.

3. Butler JN (1968) The thermodynamic activity of calcium ion in sodium chloride-calcium chloride electrolytes. Biophys J 8:1426-33.

4. Nesbitt HW (1981) pH-electrode measurements of single-ion activity coefficients consistent with a pH convention. Chemical Geology 32:207-219.

5. Brooks BR et al. (2009) CHARMM: the biomolecular simulation program. J Comput Chem 30:1545-614. 6. MacKerell Jr AD et al. (1998) All-atom empirical potential for molecular modeling and dynamics

studies of proteins. J Phys Chem B 102:3586-3616. 7. Mackerell AD, Feig M, and Brooks CL (2004) Extending the treatment of backbone energetics in

protein force fields: limitations of gas-phase quantum mechanics in reproducing protein conformational distributions in molecular dynamics simulations. J Comput Chem 25:1400-15.

8. Jorgensen WL, Chandrasekhar J, Madura JD, Impey RW, and Klein ML (1983) Comparison of simple potential functions for simulating liquid water. J Chem Phys 79:926.

9. Phillips JC et al. (2005) Scalable molecular dynamics with NAMD. J Comput Chem 26:1781-1802. 10. Andersen HC (1980) Molecular dynamics simulations at constant pressure and/or temperature. J Chem

Phys 72:2384. 11. Feller SE, Zhang Y, Pastor RW, and Brooks BR (1995) Constant pressure molecular dynamics

simulation: the Langevin piston method. J Chem Phys 103:4613. 12. Nosé S (1984) A unified formulation of the constant temperature molecular dynamics methods. J Chem

Phys 81:511. 13. Hoover WG (1985) Canonical dynamics: Equilibrium phase-space distributions. Phys Rev A 31:1695-1697. 14. Ryckaert J-P, Ciccotti G, and Berendsen HJ (1977) Numerical integration of the cartesian equations of

motion of a system with constraints: molecular dynamics of< i> n-alkanes. J Comput Phys 23:327-341. 15. Darden T, York D, and Pedersen L (1993) Particle mesh Ewald: An N log (N) method for Ewald sums

in large systems. J Chem Phys 98:10089. 16. Shaw DE et al. (2008) Anton, a special-purpose machine for molecular dynamics simulation.

Communications of the ACM 51:91-97. 17. Tuckerman M, Berne BJ, and Martyna GJ (1992) Reversible multiple time scale molecular dynamics. J

Chem Phys 97:1990. 18. Shan Y, Klepeis JL, Eastwood MP, Dror RO, and Shaw DE (2005) Gaussian split Ewald: A fast Ewald

mesh method for molecular simulation. J Chem Phys 122:54101. 19. Bernèche S, and Roux B (2001) Energetics of ion conduction through the K+ channel. Nature 414:73-77. 20. Noskov SY, Im W, and Roux B (2004) Ion permeation through the alpha-hemolysin channel: theoretical

studies based on Brownian dynamics and Poisson-Nernst-Plank electrodiffusion theory. Biophys J 87:2299-309.

21. Allen TW, Andersen OS, and Roux B (2006) Molecular dynamics - potential of mean force calculations as a tool for understanding ion permeation and selectivity in narrow channels. Biophys Chem 124:251-67.

22. Noskov SY, and Roux B (2008) Control of ion selectivity in LeuT: two Na+ binding sites with two different mechanisms. J Mol Biol 377:804-18.

23. Payandeh J, Gamal El-Din TM, Scheuer T, Zheng N, and Catterall WA (2012) Crystal structure of a voltage-gated sodium channel in two potentially inactivated states. Nature 486:135-9.

24. Chovancova E et al. (2012) CAVER 3.0: a tool for the analysis of transport pathways in dynamic protein structures. PLoS Comput Biol 8:e1002708.

25. Pauling L (1960) The Nature of the Chemical Bond (Cornell Univ. Press, N.Y.).