Structural basis for the rescue of hyperexcitable cells by the amyotrophic lateral sclerosis drug Riluzole

For this work, we utilised NavMs, a prokaryotic VGSC (BacNav) from the bacterium Magnetococcus marinus, which we have previously demonstrated to be a good model for drug binding to eukaryotic VGSCs (eNavs)³⁵, and for which we have successfully determined crystal structures in apo^36,37 and drug-bound forms^37,38. The apo-NavMs structure was initially reported to be in the open state due to its wide pore gate. However, recent work has shown this channel model to be non-conductive and also convincingly proposes that π-helix driven transitions in the channel pore (not present in the NavMs crystal structure) are required for Na⁺ conduction, redefining this structure as representing an inactivated channel state³⁹. BacNavs are homotetramers of ∼30 kDa monomeric subunits containing 6 transmembrane helices (Fig. 1a, left), (designated S1-S6), which, when associated, form channels with the basic structure of four peripheral voltage-sensing domains (VSDs), containing helices S1-S4, each independently connected to a Na⁺conducting pore module (PM) in a domain-swapped arrangement that leads to the PMforming structures from one subunit packing against the VSD of a neighbouring subunit⁴⁰, (Fig. 1b). The PM houses the intramembranous Na⁺-conducting channel and the extracellular Na⁺ selectivity filter (SF), with the channel being formed by S5 and S6 helices from each subunit, and the SF contained within their interconnecting P-loops, which each contain selectivity filter residues sandwiched between membrane descending (P1) and ascending (P2) helices. Lateral fenestrations in the PM penetrate the channel and form intramembrane tunnels that connect the membrane core to the channel pore. This fundamental architecture is maintained in eNavs which create the channels from a single amino acid chain (∼210 kDa), where four individual domains (D_I-D_IV) replace subunits to form pseudo-tetrameric structures (shown for the human Nav1.4 isoform in Fig. 1a, right and Fig. 1c).

Riluzole binding to NavMs

To establish if riluzole binds to NavMs we used the nuclear magnetic resonance (NMR) technique Saturation Transfer Difference NMR (STD-NMR), which identifies protein-ligand interactions by reporting on internuclear saturation transfer from protein to ligand in close contact, (distances 41. We used a fluorine-adapted method (¹⁹F-¹⁹F STD-NMR)⁴², biosynthetically labelling NavMs with 4-fluorophenylalanine to make NavMs_F-Phe. NavMs contains 16 phenylalanine residues distributed throughout the voltage sensor and pore domains common to eNavs, but none in its cytoplasmic domain found only in BacNavs. Irradiation at the NavMs_F-Phe fluorine resonance (-119 ppm) produced time-dependent saturation transfer to the riluzole trifluoro-group resonance (-58.4 ppm) when the experiment was performed in the presence of NavMs_F-Phe (Fig. 2a, Supplementary Information Fig. 1a), but not in its absence, (Supplementary Information Fig. 1b), identifying direct interaction between NavMs and riluzole.

a ¹⁹F-¹⁹F STD build-up over lengthening saturation times produces a curve for the interaction of riluzole with NavMs_F-Phe with an STDmax at plateau ∼3.3% (inset, riluzole coloured by heteroatom, with numbering as used throughout this study). b Riluzole binding site in NavMs produced from x-ray crystal analysis. Two views of the rilzuole binding site in NavMs (pink and yellow denoting sidechains from interacting residues from neighbouring domains) with 2Fo-Fc map (contoured at 1σ, grey), and sulphur anomalous signals (contoured at 3.5σ, red). c ¹⁹F-¹⁹F STD NMR using a site specific probe orientates riluzole in its binding site in NavMs. Subtraction of riluzole ¹⁹F signal produced from irradiation at the NavMs-BTFA fluorine resonance, (ON RESONANCE left, middle) from that produced from irradiation at the control resonance, (OFF RESONANCE left, bottom) results in a large STD peak ( ∼11% top, left) indicating strong saturation transfer between NavMs-BTFA and positioning the CF3-group of riluzole at the membrane side of the fenestration binding site (right panel). The grey sphere represents the radius of 7 Å which is the maximum distance for the STD effect from the fluorines on the C204-BTFA probe (shown as the modified sidechain at the centre of the sphere) (d) The binding site of riluzole within NavMs. (left), The complete structure of NavMs is shown cut through at fenestration depth with riluzole bound. (left,) with zoomed in view (right). e Ligplot analysis of the riluzole binding site with hydrophobic contacts from protein residues shown as red eyelashes. Interacting residues are labelled for their corresponding domains if transposed onto the D_III-D_IV fenestration of eNavs (f) Surface view of the fenestration binding site for riluzole coloured by hydrophobicity scale (low, white to high, red) looking outward from the pore.

Subsequent crystallisation trials performed with NavMs in the presence of DMSO (the riluzole vehicle solvent, 141 mM, 1%) with and without riluzole (∼0.5 mM) yielded pyramidal-shaped crystals that diffracted to high resolution ( 37) and is attributable to the hydrocarbon tails of the detergent Hega-10 which replaces lipid during purification. However, NavMs-RIL consistently had density in this region containing small additional features in their Fo-Fc difference maps, consistent with the four symmetry-related subunits producing overlapping electron density in this region reflecting partial occupancy of Hega-10 and riluzole, but with no way to deconvolute the density and unambiguously place the small (234 Da) planar riluzole molecule into this site (Supplementary Information Fig. 2a and b).

To address this, we attempted to locate the riluzole sulphur atom by collecting single-wavelength anomalous diffraction (SAD) data close (see “methods”) to the sulphur absorption K-edge. Both DMSO and riluzole contain sulphur so we collected data on DMSO-only and NavMs-RIL crystals, and the anomalous density present in the X-ray data was determined. In all crystals, we successfully identified sulphur signals for 10 of the 11 methionine residues of NavMs (not M90 which is in an unstructured loop), and for its single cysteine, C52. For NavMs-RIL, an additional peak in the anomalous density map was present in the region of the extra electron density found in the fenestrations, identifying the riluzole sulphur. Collection from the first crystal (crystal 1), gave anomalous density for this peak with an intensity of ∼5σ (Fig. 2b), indicative of a riluzole occupancy of ∼20% relative to the strongest anomalous density peak in the data. Four separate collections, performed on a second crystal (crystal 2), each at a different crystal orientation, all produced this extra peak with the first collection producing a peak intensity comparable to crystal 1 (Supplementary Information Fig. 2c). In the subsequent collections this peak was maintained but with reduced intensities, consistent with expectations for a genuine peak while accounting for increased radiation damage impacting the signal as collections proceeded (Supplementary Information Fig. 2d). The DMSO-only crystals contained no corresponding anomalous peak even at the level of signal noise, (Supplementary Information Fig. 2e and f). All of the crystals analysed exhibited two anomalous sulphur peaks for M204, a bottleneck residue located at the membrane entrance to the fenestration. The major peak corresponds to the open fenestrations modelled in our structure. However, the movement of the methionine sidechain to match the second peak (shown as the minor anomalous sulphur peak in Fig. 2b) closes the fenestrations (Supplementary Information Fig. 2g and h). These closed fenestrations would be empty of detergent tails in the crystals, (and in vivo, lipid tails), as they could not access the channel. Additionally, this sidechain movement would clash with the trifluoromethoxy-group of riluzole as positioned in our structure. However, an alternative riluzole orientation achieved by rotating the molecule 180° around the sulphur atom to reposition the drug with this group in the channel pore, although unlikely from the electron density maps, could be accommodated if riluzole bound to this closed fenestration state.

To eliminate any ambiguity concerning the position of riluzole in the binding site, we again, performed ¹⁹F-¹⁹F STD-NMR, but this time, we used it to look for saturation transfer between NavMs, specifically fluorine-labelled at a single site in the protein, and riluzole. A trifluoromethyl group was introduced onto the cysteine of a NavMs C52A/M204C double mutant by chemical conjugation using the alkylating agent 3-bromo-1,1,1-trifluoroacetone (BTFA), to produce NavMs-BTFA. This introduced a trifluorinated probe into NavMs at a specific location near the membrane-facing opening to the fenestrations. Selective irradiation of the ¹⁹F resonance of NavMs-BTFA (-83.8 ppm) produced robust STD at the riluzole trifluoro-group resonance (-58.4 ppm, Fig. 2c, left), placing the fluorines of riluzole within the 7 Å radius delineated by the probe (Fig. 2c, right), and confirming the original placement of riluzole in the structure as the correct one.

Using this combination of techniques, we could unambiguously place riluzole into the binding site in the open fenestration conformation of NavMs (Fig. 2d). No movements in protein structure were observed on riluzole interaction, with riluzole binding NavMs in a hydrophobic pocket within the fenestration and making contacts with S6 helix and SF residues from two consecutive domains and residues from the P1 helix from the first of these domains (Fig. 2e, f). As a true tetramer NavMs contains four equivalent fenestrations, whereas in pseudo-tetrameric eNavs individual fenestrations are markedly different in both size and sequence⁴³. The D_III-D_IV fenestration of eNavs contains a conserved phenylalanine residue on helix S6_IV at the pore lining of this fenestration, which forms part of the ‘local anaesthetic (LA) binding site’, that is key to the action of VGSC pore-blocking drugs that preferentially bind to and stabilise the inactivated channel state^44,45. This residue has also been implicated in the eNav-riluzole interaction¹⁹. The riluzole binding site in NavMs includes T207 from the S6 helix of the second binding domain which, if this fenestration was transposed onto the D_III-D_IV fenestration of eNavs, represents the position of this S6_IV residue, (D_III-D_IV equivalent labelling added to NavMs interacting residues in Fig. 2e). Additionally, the P1 helix and SF of eNavs, both containing residues forming part of the riluzole binding site in NavMs, are also involved in inactivation⁴⁶. As a model for eukaryotic inactivation, BacNavs only have slow inactivation processes reminiscent of slow inactivation in eNavs. However, it has been shown that riluzole modulates both fast and slow inactivation of eNavs with equivalent potency⁴⁷, indicative of a common binding event inhibiting both processes.

NavMs as a functional model for riluzole action on eNavs

At therapeutically relevant concentrations it has been shown that riluzole selectively stabilises eNavs in their inactivated state, which in whole-cell studies manifests as riluzole causing dose-dependent hyperpolarisation of steady-state inactivation (SSI) and slowed recovery from inactivation (RFI), without it affecting activation kinetics or producing channel block^9,10. We performed whole-cell patch clamp experiments using Human embryonic kidney 293 T (HEK293T) cells transiently transfected to express NavMs and found that this profile of riluzole effects on eNavs was replicated on NavMs (Supplementary Information Tables 2 and 3). Riluzole dose-dependently shifted SSI of these cells in the hyperpolarised direction (Fig. 3a), with the IC₅₀ of 2 µM for the shift (Supplementary Information Fig. 3a) being comparable to the IC50 values for similar shifts of SSI found with eNavs⁴⁷. Riluzole slowed the RFI of NavMs, with 2.5 µM riluzole causing a 3-fold greater delay in channel recovery compared to vehicle (Fig. 3b). Activation kinetics were unaffected by riluzole, with no change in steady-state activation or voltage-dependence of activation of NavMs up to 25 µM, (Supplementary Information Fig. 3b and c), and channel block was absent up to the same concentration (Fig. 3c, Supplementary Information Fig. 3d, e), although, similarly to eNavs^9,47, some block developed at high concentration (Fig. 3c, 100 µM). A riluzole binding site in NavMs that can produce the effects of the drug at therapeutically relevant concentrations is consistent with its location in the crystal structure (Fig. 3d) where binding does not occlude the Na⁺ conduction pathway, even if all four fenestrations are occupied (Fig. 3e). The channel block produced at high concentration results from riluzole either binding to a secondary binding site located in the pore or through it non-specifically occluding of the pore, with either scenario being functionally-irrelevant at therapeutic riluzole concentrations.

a Riluzole produces a dose-dependent hyperpolarised shift in the SSI curve for NavMs. Voltage dependence of inactivation of HEK293T cells transfected with NavMS after perfusion of the vehicle (black circles) or increasing concentrations of riluzole. Data points were fit to a Boltzmann equation to calculate V_1/2 (Supplementary Information Table 2). The normalised current data is displayed as mean ± S.E.M. SSI V_1/2 after perfusion with the vehicle (black circles) was −84.5 ± 1.4 mV (n = 11), but was −86.9 ± 2.6 mV (n = 5), −94.1 ± 2.5 mV (n = 18), −95.5 ± 1.8 mV (n = 8), and −97.3 ± 3.2 mV (n = 6) after perfusion with 1 µM (blue triangles), 2.5 µM (red squares), 5 µM (green triangles), and 25 µM (purple squares) riluzole respectively. One-way ANOVA and post-hoc Tukey test found that 2.5, 5 and 25 µM riluzole significantly shifted SSI compared to vehicle (p = 0.025, 0.038, and 0.024 respectively). b Riluzole slows NavMs recovery from inactivation in HEK293T cells. The t_fast component of a biphasic recovery process was significantly increased by 3-fold. Data displayed as mean ± S.E.M, *p p fast after perfusion with the vehicle was 0.009 ± 0.002 s (n = 8), but was 0.034 ± 0.007 s (n = 8), 0.050 ± 0.004 s (n = 5), and 0.033 ± 0.008 s (n = 6) after perfusion with 2.5, 5 and 25 µM riluzole respectively. One-way ANOVA and post-hoc Tukey test found that that 2.5, 5 and 25 µM riluzole significantly shifted τ_fast compared to vehicle (p = 0.008, 0.0002, and 0.027 respectively). c Riluzole only blocks NavMs in HEK293T cells at the high concentration (100 µM) with 1–25 µM producing no significant occlusion of the channel pore. Data points shown with mean ± S.E.M indicated, ***p n = 11), and was −0.027 ± 0.021 (n = 5), −0.007 ± 0.024 (n = 18), 0.055 ± 0.068 (n = 8), −0.011 ± 0.050 (n = 6), and 0.239 ± 0.064 (n = 6) for 1 µM (blue triangles), 2.5 µM (red squares), 5 µM (green triangles), 25 µM (purple squares), and 100 µM (pink diamond) riluzole respectively. One-way ANOVA and post-hoc Tukey test found that only 100 µM riluzole significantly blocked NavMS compared to vehicle (p = 0.0001). d Top-down sliced view of NavMs at fenestration depth showing riluzole bound in all four fenestrations. e Side view from Hole2 analysis showing that even with all 4 fenestrations occupied, riluzole does not block Na⁺ conduction. A pore radius of > 2.3 Å is required for Na+ conduction and is represented by blue in the channel tunnel cartoon (left panel) and delineated by the broken vertical line in the plot of pore radius along the pore axis (right panel). The green shaded area in the plot (right panel) represents Na⁺ conduction pathway at fenestration depth.

Molecular dynamics simulations reveal the mechanism of riluzole interaction

To investigate how riluzole interacts with NavMs in a membrane environment we performed a 2µs unbiased molecular dynamics (MD) flooding simulation with NavMs embedded in a POPC lipid bilayer and riluzole molecules initially distributed in the aqueous phase. Due to the high concentration of drugs required in flooding experiments, combined with riluzole hydrophobicity, some riluzole molecules non-specifically aggregated on the aqueous-exposed surfaces of NavMs at the membrane interface at the beginning of this simulation and these were excluded from subsequent analysis. During the 2µs simulation, lipid tails quickly and dynamically occupied the fenestrations of NavMs and riluzole molecules rapidly partitioned into the lipid bilayer. The simulation, when assessed for contacts between any riluzole molecule and NavMs, found binding hotspots only in the NavMs fenestrations (Supplementary Movie 1). Spatial clustering was carried out over the length of the simulation for a typical binding riluzole molecule using an RMSD threshold of 2 Å, producing a primary cluster occupied by riluzole for 26% of the total simulation time (Fig. 4a). Superimposing this cluster onto the NavMs-RIL crystal structure showed that riluzole would be entirely nested within this cluster (Fig. 4b).

a Clustering analysis of riluzole and NavMs, all members of the primary cluster (depicted in grey) and NavMs WT (depicted as pink helices in a top-down view at the depth of the fenestrations. The primary cluster represents the riluzole location for 26% of the total simulation time, b The primary cluster overlays with the binding location of riluzole in the X-ray structure. c Contact map of riluzole atoms with T176 and M204 from NavMs calculated from the simulation, coloured by the average number of contacts per timestep, showing that riluzole orientation is the same in both structurally-determined and MD simulation-determined binding sites.

To understand the molecular interaction between NavMs and riluzole, distance-based intermolecular contacts were calculated to quantify the number and nature of the interactions within the binding site throughout the simulation. A contact was defined as when any atom of a riluzole molecule was within 6 Å of both the M204 residue, at the membrane-facing entrance to the fenestration, and the T176 residue, which is closer to the channel pore. This analysis allowed for a maximum residence time for riluzole binding to be determined to quantify the strength of the interaction, and preferred orientations to be calculated. Riluzole bound NavMs at this site with a maximum residence time of 92 ns, with the contact map produced for the interaction (Fig. 4c) showing that riluzole molecules bound to NavMs in the same orientation and with similar atomic contacts as those of the crystal structure.

Importance of the Local Anaesthetic (LA) binding site residue for riluzole action on VGSCs

Riluzole (100 µM) effectiveness in stabilising inactivation is diminished in an eNav (rat Nav1.4) by alanine substitution of the LA binding site phenylalanine residue¹⁹. Using whole-cell patch clamp experiments, we investigated the importance of this residue on VGSCs at clinically relevant riluzole concentrations, initially comparing how riluzole-affected HEK293T cells transiently transfected to express WT human Nav1.4 (hNav1.4) to those expressing the LA binding site mutant channel hNav1.4 F1586A. Riluzole (1 µM) stabilised the inactivation of the WT channel, producing a hyperpolarised shift in the SSI curve (Supplementary Information Fig. 4a). The same experiment produced no shift when WT hNav1.4 was replaced with the F1586A mutant (Supplementary Information Fig. 4c) showing that this residue was also vital for riluzole’s stabilisation of the inactivation of hNav1.4 at this low concentration. Consistent with other studies, riluzole did not affect the activation of either channel type (Supplementary Information Figs. 4b and 4d). To investigate this site for the riluzole effect on inactivation on NavMs, a similar patch-clamp experiment was performed as previously, but with cells expressing NavMs T207A. The addition of 2.5 µM riluzole, which had produced a ∼10 mV hyperpolarised shift in the SSI curve with the WT channel as characterised by the V_1/2 for the shift (V_1/2 = −84.5 ± 1.4 mV vehicle, −94.1 ± 2.5 mV riluzole), failed to produce a significant shift on T207A (V_1/2 = −84.7 ± 2.6 mV vehicle, −88.6 ± 1.6 mV riluzole) (Fig. 5a, Supplementary Information Table 3), showing that this site is also crucial for riluzole to stabilise the inactivated state of NavMs, even though the residue is a threonine and not a phenylalanine.

a LA binding site mutation T207A abrogates the riluzole effect on SSI of NavMs expressed in HEK293T cells. Data points for WT NavMs (circles) and NavMS T207A (squares) were fit to a Boltzmann equation to calculate V_1/2 (Supplementary Information Table 2). The normalised current data is displayed as mean ± S.E.M. Data points were fit to a Boltzmann function to calculate V_1/2. Unpaired Student’s t-test of the riluzole effect on SSI of NavMs T207A found no significant difference compared to NavMs T207A without riluzole (n = 5, p > 0.5), b Contact map for riluzole interaction with NavMs T207A over the course of the simulation showing that the average number of riluzole atoms which are in simultaneous contact with residues T176 and M204 are greatly reduced compared to WT c Top-down view at fenestration depth of all members of the primary cluster (grey) from the NavMs T207F MD simulation. d Riluzole binding site within this primary cluster represents riluzole binding in the eNav D_III-IV-mimicking fenestration of NavMs T207F with the additional π–π stacking interaction indicated by a broken yellow line e Bar chart of the maximum riluzole residence times observed in the binding sites of each NavMs variant (f) Pathway taken by a typical NavMs binding riluzole molecule during the MD simulations. Black lines represent the membrane boundary.

To investigate how this mutation affects NavMs on a structural level, we ran an analogous MD simulation to that performed previously, replacing WT NavMs with T207A. In this simulation, riluzole entered both membrane and fenestration similarly to the WT simulation. However, once in the fenestration, sustained binding was not seen, with riluzole molecules continuing to be mobile in the pore. This simulation produced no majority spatial cluster (ie., no cluster > 5% occupancy). Consequently, riluzole contacts at the binding site were much fewer (Fig. 5b), resulting in a greatly reduced maximum residence time for T207A (36 ns, Supplementary Movie 2) compared to WT (Fig. 5e, compare Supplementary Movies 1 (WT) and 2 (T207A)), showing that riluzole and NavMs interactions were less frequent and weaker compared to WT.

Considering the importance of the LA binding site for the action of riluzole, we used the same MD simulation to explore the impact of introducing a ‘humanising’ T207F mutation into one subunit of the NavMs tetramer. In this simulation, there was a clear preference for binding in this eNav D_III-D_IV interface-mimicking fenestration compared to the native fenestrations, resulting in it containing the primary cluster, representing 32% occupancy of simulation time (Fig. 5c). This cluster produced a binding site location slightly closer to the pore compared to WT, but riluzole bound in the same orientation and maintained a non-pore-blocking pose (Supplementary Information Fig. 5a and b). The introduction of F207 resulted in a binding interaction where the benzothiazole ring of riluzole π–π stacked with F207 (Fig. 5d, Supplementary Movie 3), stabilising the interaction, and resulting in riluzole having a higher maximum residence time in this fenestration (234 ns) compared to WT (Fig. 5e).

A hydrophobic pathway links VGSC binding to riluzole potency

The MD simulations with NavMs reveal the basis for how riluzole can potently affect VGSC function. In these simulations, riluzole rapidly partitioned out of the aqueous phase to form a highly concentrated repository of riluzole in the membrane (all riluzole molecules entered the membrane within 100 ns in a membrane-only simulation). Therefore, riluzole action is not reflective of its measured aqueous concentration but results from a much higher concentration in the membrane. Membrane partitioning was facilitated by the highly lipophilic trifluoromethyl group of riluzole. Once in the membrane riluzole (which often non-specifically sampled NavMs) enters the fenestration, amine group first, to access its binding site, (complete pathway taken by a typical riluzole molecule shown in Fig. 5f and dynamically in (Supplementary Movie 4).

MD simulations with riluzole and human Nav1.4

Eukaryotic VGSCs are incompatible with the structural techniques used in this study, and cryoEM has only revealed a pore-blocking pose for the drug⁴⁸, consistent with a second binding site causing the channel block produced by riluzole at high non-therapeutic concentrations. However, considering the accuracy of MD simulations in reproducing the interaction of riluzole with NavMs found in our structural study, we sought to use it to investigate the effect of riluzole on a human VGSC (hNav). We used the same MD simulation as before but replaced NavMs with hNav1.4 (the first available cryoEM structure for an hNav, PDB ID: 6AGF⁴⁹, also captured in the inactivated state). Here, after ∼500 ns, riluzole bound in the D_IV-D_I fenestration where it remained for the rest of the 2µs simulation. Riluzole bound in a cluster further away from the channel pore compared to the NavMs simulations (Supplementary Information Fig. 6a), binding in a hydrophobic pocket, with π–π staking between the benzothiazole group of riluzole and F432 of hNav1.4 (Supplementary Information Fig. 6b). However, this interaction does not involve the S6_IV residue F1586 which we have shown to be crucial to the effect of riluzole on hNav.1.4 and therefore lacks relevance for any known therapeutic effect of the drug. In this simulation, lipids extended deeper into the D_III-D_IV fenestration compared to that of D_IV-D_I, visibly interacting with the pore-lining F1586 residue, making riluzole interaction less likely to occur in this fenestration during the 2µs of simulation.

To alleviate this constraint, we performed a slightly modified MD simulation, placing a riluzole molecule at the entrance to the D_III-D_IV fenestration at the beginning of the simulation, allowing it to compete with lipids for binding from the start. In this simulation, riluzole immediately moved into the fenestration (within the first nanosecond), where it stayed for the remainder of the simulation. This simulation produced two major riluzole binding clusters, both contained within overlapping hydrophobic pockets inside the D_III-D_IV fenestration, with both involving interaction with F1586. Comparably with NavMs, these clusters produced binding sites that included residues from the S6 helices, P1 helix, and SF contained within the fenestration (Supplementary Information Fig. 7). Cluster 1 (Fig. 6a, c) sits centrally to the fenestration and produces a binding site almost identical in its alignment to that of the WT NavMs (Supplementary Information Fig. 7, top alignment) Here, riluzole binds almost exclusively through hydrophobic interactions, although a weak hydrogen bond exists between the amine of riluzole and the side chain hydroxyl of S1283 on S6_III (Fig. 6b). Cluster 2 (Fig. 6d, f) lies slightly closer to the pore of the channel and produces a binding site almost identical in alignment to that produced in the simulation of NavMs T207F (Supplementary Information Fig. 7, bottom alignment) where riluzole binds exclusively by hydrophobic interactions and features a π–π stacking interaction between the benzothiazole ring of riluzole and F1586 (Fig. 6e).

a Cluster 1 (grey) for riluzole interaction in the hNav1.4 D_III-D_IV fenestration. b Interacting residues at binding site 1 showing a hydrogen bond between riluzole and S1283. c Surface view  of riluzole bound in the fenestration at binding site 1 looking outward from the pore. d Cluster 2 (grey) for riluzole interaction. e Interacting residues at binding site 2 features a π–π stacking interaction between the benzothiazole group of riluzole and F1586. f Surface view of riluzole in the fenestration at binding site 2 looking outward from the pore. Figure 6c, f surfaces are coloured for hydrophobicity on the scale shown in Fig. 2f.

To investigate the importance of F1586 to the interaction, we performed the same simulation with hNav1.4 F1586A. Here, riluzole again entered the fenestration but, similarly to the simulation involving NavMs T207A, showed higher mobility, lower occupancy, and greatly reduced binding site contacts relative to WT (Supplementary Information Fig. 6c, d), consistent with our functional experiments showing the diminished effect produced by riluzole on the F1586A channel compared to WT.

Riluzole normalises the pathological I_NaL produced by a Nav1.4 disease variant

Humans have nine VGSC subtypes differentially expressed across excitable tissues of the body. The hNav1.4 used in this study is a VGSC isoform expressed predominantly in skeletal muscle, and mutation within this channel is causative to many muscular diseases, including myotonias and/or periodic paralyses^11,50. Myotonias are hyperexcitable states that are associated with altered inactivation kinetics and elevated I_NaL^11,50, and it has been reported that riluzole has high potency against cells expressing WT hNav1.4 when mimicking a hyperexcitable myotonia-like state, (IC50 ∼0.9μM), compared to normal states, (IC50 ∼50 µM)³³. We have previously shown that the inactivation-impaired hNav1.4 variant P1158S⁵¹, which causes myotonia and periodic paralysis, has a pathologically increased I_NaL (∼5% of peak), compared to the WT channel (∼0.75% of peak)⁵². The addition of 1 µM riluzole to HEK293T cells transiently transfected to express hNav1.4 P1158S stabilised the inactivation in this variant, significantly shifting the SSI curve to more negative potentials, compared to vehicle (Fig. 7a, p NaL to WT levels (Fig. 7b). No significant additional change to I_NaL occurred on the addition of 5μM riluzole (Fig. 7b).

a Effect of riluzole (1 µM) or its vehicle solvent (DMSO) on the SSFI of hNav1.4 P1158S expressed in HEK293T cells. Data is plotted as average ( ± S.E.M) with the insert showing the protocol and representative currents (n = 5, each). Data points were fit to a Boltzmann function. b Representative I_NaL showing the effect of riluzole (1 or 5 µM) or its vehicle solvent on hNav1.4 P1158S expressed in HEK293T cells. The inset bar graph shows the effect of riluzole on the percentage ( ± S.E.M) of I_NaL (n = 5, each). One-way ANOVA and post-hoc Tukey test found that hNav1.4 P1158S significantly increased I_NaL (p NaL (p NaL levels produced on addition of 1 and 5 µM riluzole to hNav1.4 P1158S were not significantly different from WT vehicle (p > 0.9 for both).