Not sorcery after all: Roles of multiple charged residues in membrane insertion of gasdermin-A3

Gasdermins execute programmatory cell death, known as pyroptosis, by forming medium-sized membrane pores. Recently, the molecular structure of those pores as well as the diversity in their shape and size have been revealed by cryoTEM and atomic force microscopy, respectively. Even though a growth of smaller to larger oligomers and reshaping from slits to rings could be documented, the initiation of the gasdermin pore formation remains a mystery. In one hypothesis, gasdermin monomers insert into membranes before associating into oligomeric pores. In the other hypothesis, gasdermin oligomers preassemble on the membrane surface prior to membrane insertion. Here, by studying the behavior of monomeric membrane-inserted gasdermin-A3 (GSDMA3), we unveil that a monomeric gasdermin prefers the membrane-adsorbed over the membrane-inserted state. Our results thus support the hypothesis of oligomers preassembling on the membrane surface before membrane penetration. At the same time, our simulations of small membrane-inserted arcs of GSDMA3 suggest that the inserting oligomer can be small and does not have to comprise a full ring of approximately 26–30 subunits. Moreover, our simulations have revealed an astonishingly large impact of salt-bridge formation and protein surroundings on the transmembrane passage of charged residues, reducing the energetic cost by up to 53% as compared to their free forms. The here observed free energy barrier of mere 5.6 kcal/mol for the membrane insertion of monomeric GSDMA3 explains the surprising ability of gasdermins to spontaneously self-insert into cellular membranes.

1981) with τ p = 1 ps kept the pressure of the semiisotropic system (compressibility = 4.5 · 10 −5 bar −1 ) at 1 bar. Hydrogen atom vibrations were constrained by the LINCS algorithm Hess (2008), far-reaching electrostatics interactions (beyond 1.2 nm) were modelled using particle-mesh-Ewald Darden et al. (1993) and the potential-shift-Verlet modifier while van-der-Waals interactions between 0.8 and 1.2 nm were shifted to zero by the potential-switch method. The buffer tolerance was set to 0.005 kJ/(mol·ps·atom) and the neighbor list updated every 0.02 ps by the Verlet cut-off scheme Páll and Hess (2013). The production simulations were run for five to 12.8 µs (all individual runtimes are listed in the Supplementary Table S1). In all all-atom simulations the integration timestep of 0.002 ps was used. All simulations were carried out in GROMACS (2018GROMACS ( .x-2021x) Abraham et al. (2015). In simulations of monomeric GSDMA3, placed in a hexagonal box of 13.5 × 11.7 × 13.5 nm one copy of GSDMA3 was surrounded by 491 E. coli polar lipid extract (PLE) lipids Pluhackova and Horner (2021), about 50 000 water molecules and neutralized by 171 Na + and 30 Cl − ions. In simulations of 7-arc GSDMA3, seven copied of GSDMA3 were placed into an E. coli PLE bilayer comprising 1344 lipids, solvated by about 100 000 water molecules and 383 Na + , resulting in a hexagonal box of 23.0 × 19.9 × 11.0 nm.  Webb and Sali (2016) as loops, d started from 5µs of the uuu simulation in which the β-hairpins slipped out of the bilayer (described in b ), e "cleaved" model, i.e. missing residues in the N-terminus of the full GSDMA3 from the crystal structure 5B5R Ding et al. (2016) were modelled as loops by Modeller Webb and Sali (2016), f simulations were performed within the framework of our previous publication Mari et al. (2022). Total simulation time 80 µs.

Potential of Mean Force Simulations
The frames used for the "umbrella" geometric perturbation sampling of GSDMA3 were extracted from a 5.8 µs long simulation in which gasdermin-A3 (uuu) spontaneously slipped out of the membrane.
Additional structures in the membrane-inserted state were taken from a 5 µs equilibrium simulation in the ccc state, and further conformations on the membrane surface stem from a 1 µs equilibrium ccc simulation. the overlap of the structures from these simulations along the reaction coordinate is visualized in Figure S10. The same GSDMA3 umbrella conformations were used for each charge state (ccc, uuu, cuc, and ucu) by adaptation of the protonation state and the number of Na + and Cl − ions, so that the system was always neutral and contained the same number of atoms for each state. The charge adaptation was followed by energy minimization. As a reaction coordinate ζ, the position of the C α of K97 relative to the center of mass (COM) of the membrane were chosen. The maximal spacing between the windows along ζ amounted to 0.05 nm. Each of the windows per charge state (see Table S1) was sampled for 50 ns, yielding a total simulation time of more than 3 µs with a timestep of 2 fs. A constraining potential of 500 or 1000 kJ was applied to each window to achieve the best possible overlap of the windows. At a temperature of 310 K, the Nosé-Hoover thermostat Evans and Holian (1985) with τ T = 0.5 was utilized along with a pressure of 1 bar and the semiisotropic pressure coupling with the Parrinello-Rahman barostat Rahman (1980, 1981) and τ p = 5.0. Intermolecular forces were simulated using a vdW-shift between 0.8 and 1.2Å and PME electrostatics (0.12Å grid) with a coulomb radius of 1.2Å. The potentials of mean force (PMFs) were extracted using the Weighted Histogram Analysis Method (WHAM) Kumar et al. (1992); Souaille and Roux (2001); Hub et al. (2010) implemented in GROMACS. The error bars were obtained from 100 bootstraps. Figure S1. Starting structures of the equilibrium simulations: (A) GSDMA3 in its pore conformation, (B) "preinserted" model (see Table S1), (C) started from 5 µs of the uuu simulation in which the β-hairpins slipped out of the bilayer, (D) "cleaved" model (see Table S1). GSDMA3 is shown as blue cartoon with the β-hairpins highlighted in cyan and pink, respectively. Molecular representation of (E) lysine, and (F) lysine side chain analog (n-butylamine).
Lysine and a lysine side chain analog (n-butylamine) were placed into the solvent in 0.6 nm distance to the E. coli PLE membrane in a preequilibrated rectangular simulation box spanning 11.1 × 11.1 × 7.8 nm and containing 386 lipids, 16365 water molecules, and a neutralizing NaCl concentration. The water molecules overlapping with the amino acid (analog) were deleted and the system was energy minimized and equilibrated for 10 ps with a timestep of 0.001 ps at a temperature of 310 K controlled by the v-rescale thermostat (τ T = 0.5 ps) and a pressure of 1 bar controlled by the Berendsen barostat Berendsen et al. (1984) (τ p = 5.0 ps) in a semiisotropic manner with a compressibility of 4.5 · 10 −5 bar −1 . Hydrogen bonds were restrained using the LINCS algorithm Hess (2008), electrostatics beyond 1.2 nm was modelled by particle-mesh Ewald Darden et al. (1993) and van-der-Waals interactions were shifted to zero between 0.8 and 1.2 nm by the potential-switch method. To obtain frames for geometric perturbation sampling, the pull code implemented in GROMACS was used to pull the aminoacid through the membrane in z-direction. 49 windows between 2.5 nm and -1.5 nm distance of the C α from the membrane's center of mass were selected for lysine. For its analog, 43 windows were selected. An NpT equilibration was run for 10 ns in each window using a timestep of 0.002 ps. All equilibrations were kept at a temperature of 310 K by the Berendsen thermostat Berendsen et al. (1984) (τ T = 0.5 ps) and coupled semiisotropically to the Berendsen barostat Berendsen et al. (1984) at 1 bar and τ p = 1.0 ps. The compressibility was set to 4.5 · 10 −5 bar −1 , hydrogen bonds were constrained using the LINCS algorithm Hess (2008). Neighbor search was cut off by the Verlet method, far-reaching electrostatics were modelled by particle-mesh Ewald Darden et al. (1993) beyond 1.2 nm. The actual geometric perturbation simulations were run for 50 ns using a timestep of 0.002 ps at a temperature of 310 K controlled by the Nosé-Hoover thermostat Evans and Holian (1985) (τ T = 0.5 ps) and coupled semiisotropically to the Parrinello-Rahman barostat Rahman (1980, 1981) at 1 bar and τ p = 5.0 ps at a compressibility of 4.5 · 10 −5 bar −1 . Hydrogen bonds were restrained using the LINCS algorithm Hess (2008), electrostatics beyond 1.2 nm were modelled by particle-mesh Ewald Darden et al. (1993) and van-der-Waals interactions were shifted to zero between 0.8 and 1.2 nm by the potentialswitch method. The shifts of the PMF curves were calculated using ∆G ≈ ±(2.303RT ) · (pKa − pH), pKa(lysine) = 10.53 at 311. 15 K O'Neil (2006), and pKa(glutamic acid) = 4.25 Hunt (2022).

Water defects
The water defect analysis in the bilayers was perfomed using Python 3.x Python (2020) and MDAnalysis Gowers et al. (2019);Michaud-Agrawal et al. (2011). The number of water molecules entering the bilayer due to the presence of the protein was estimated by selecting water molecules residing within the membrane (bordered by the z-coordinates of the phosphate headgroups in each leaflet, determined in each frame individually) and within a cylinder (height 20Å, radius 10Å) around the center of geometry of the β-hairpins (residues 92-102 and 176-186). For lysine, a cylinder around its center of geometry with a height of 14Å and a radius of 10Å was used, for the analog, a height of 10Å was chosen.

Residue-lipid headgroup contacts
To find out whether the lysines 97, 100, and 102 were snorkeling into the headgroup region of the membrane, contacts (i.e. distance smaller than 5.5Å) between the lysines' N atom of its NH 2 or NH + 3 group and phosphate groups' oxygen atoms were counted in each frame.

Situation at ζ = 2.5 nm
To ensure that shifting the PMF curves at ζ = 2.5 nm by the experimental free energy cost of (de)protonating in bulk water is valid, the titratable residues have to be surrounded by water so that they can be (de)protonated. Therefore we have measured the distance between the titratable residues and the membrane at ζ = 2.5 nm, and in case of the membrane-near K97 we have estimated the number of water molecules in the first solvation shell of the aminogroup, i.e. within 4Å of the amino hydrogens. According to the results summarized in Table S2, the residues E94 and K100 are found far away from the bilayer. K97 is well hydrated and despite its membrane proximity, allowing its (de)protonation. Table S2. Solvation of the amino group of K97 (estimated as the average water molecules that surround the hydrogens of K97's -NH + 3 /-NH 2 group within a 4 A radius) and the distances between the titratable groups of residues K97, E94, and K100 and the membrane surface (represented by phosphate oxygens) at ζ = 2.5 nm.  Figure S2. GSDMA3, simulated cryoTEM conformation Ruan et al. (2018), shown as blue cartoon inside an E. coli membrane. Polar amino acids are highlighted in green and charged amino acids in yellow. Inset: Sequence of GSDMA3's β-hairpins.    Figure S6. Fluctuating distance between K97's C α and the membrane's center of mass (ccc). Figure S7. Fluctuating distance between K97's C α and the membrane's center of mass (cuc). Figure S8. Fluctuating distance between K97's C α and the membrane's center of mass (uuu). At the end of the uuu1 simulation, the β-hairpins slip out of the membrane. Figure S9. GSDMA3 colored by B-factor (0-200, blue to teal). The blue parts move the least over the course of the simulation, the teal parts show the most movement. Left: ccc2 (average structure). Center: cuc (average structure over cuc 5-10 µs). Right: uuu1 (average structure). Figure S10. PMFs of four different charge states of the GSDMA3 monomer relative to the ucu state. uuu has been shifted at ζ = 2.5 nm by 5.03 kcal/mol (the free energy cost of deprotonating lysine in bulk water), cuc by -3.90 kcal/mol (the free energy gain of deprotonating glutamic acid in bulk water), and ccc by -5.03 kcal/mol and -3.90 kcal/mol (free energy gains of protonation of a lysine and deprotonation of a glutamic acid in bulk water, respectively). The PMFs of the GSDMA3 monomer with the charged pair E94 and K100 (cuc and ccc) lay lower in energy and do not intersect with the ones of the uncharged pair (uuu and ucu). Therefore, E94 is highly unlikely to be protonated and K100 to be deprotonated. ccc is found lowest in energy, along the whole reaction coordinate, however, around the energy barrier at ζ ≈ 0.7 nm the cuc and ccc states are found very close in energy. Thus spontaneous protonation change of K97 in this position can't be excluded. The error bars show standard deviations estimated over 100 bootstraps. The bars at the bottom visualize the span of the reaction coordinate covered by structures extracted from the following simulations: purple: 1 µs equilibrium ccc simulation in the membrane-adsorbed state, light grey: a 5.8 µs long simulation in which gasdermin-A3 (uuu) spontaneously slipped out of the membrane, and dark grey: 5 µs equilibrium simulation of the membrane-inserted conformation in the ccc state.       Figure S23. Number of contacts of K97 with phosphate headgroups of lipids averaged over each entire trajectory from the geometric perturbation. Figure S24. Number of contacts of K100 with phosphate headgroups of lipids averaged over each entire trajectory from the geometric perturbation. Figure S25. Number of contacts of K102 with phosphate headgroups of lipids averaged over each entire trajectory from the geometric perturbation. Figure S26. Water defect formed by lysine at ζ = 0.05 nm. Corresponds to a defect containing 18 water molcules at 27.3 ns. Figure S27. Water defect formed by the lysine analog at ζ = 0.03 nm. Corresponds to a defect containing 9 water molcules at 8.95 ns. Figure S28. RMSD of the 7-mer arc simulations. hairpin1 = residues 80-113, hairpin2 = residues 164-198, hairpin1-tips = residues 95-99, hairpin1-betas = 80-94 and 100-113, hairpin2-tips = residues 178-184, and hairpin2-betas = residues 164-177 and 185-198. The seven capital letters correspond to the seven chains. Figure S29. Water defects present at the beginning of the 4µs simulation with the protomers restrained in the pore conformation. This frame shows a defect containing 346 water molecules. Figure S30. Reduced water defects after 4µs with the relaxed 7-mer arc protomers. This frame shows a defect containing 268 water molecules.