Molecular view of ER membrane remodeling by the Sec61/TRAP translocon

Abstract Protein translocation across the endoplasmic reticulum (ER) membrane is an essential step during protein entry into the secretory pathway. The conserved Sec61 protein‐conducting channel facilitates polypeptide translocation and coordinates cotranslational polypeptide‐processing events. In cells, the majority of Sec61 is stably associated with a heterotetrameric membrane protein complex, the translocon‐associated protein complex (TRAP), yet the mechanism by which TRAP assists in polypeptide translocation remains unknown. Here, we present the structure of the core Sec61/TRAP complex bound to a mammalian ribosome by cryogenic electron microscopy (cryo‐EM). Ribosome interactions anchor the Sec61/TRAP complex in a conformation that renders the ER membrane locally thinner by significantly curving its lumenal leaflet. We propose that TRAP stabilizes the ribosome exit tunnel to assist nascent polypeptide insertion through Sec61 and provides a ratcheting mechanism into the ER lumen mediated by direct polypeptide interactions.

increased to keep the simulation model faithful to our cryo-EM model.Moreover, the backbone restraints were maintained throughout the equilibration steps.Finally, the protein backbone atoms were restrained by a force constant of 1000 kJmol −1 nm −2 for production runs of 100 ns at 37°C with both force Uield combinations.The simulation parameters recommended for CHARMM or Amber force Uields in GROMACS were used (Lee et al, 2016).Since these parameters are consistent between all simulations performed using the same force Uield, and they are listed below in a separate subsection.
The hydrogen bonding partners between the different TRAP subunits, between Sec61 and TRAP, between Sec61 and the LSU of the ribosome, and between TRAP and the LSU of the ribosome were calculated from the backbone-restrained simulations, in which the side chains were free to adapt to the environment.A hydrogen bond was deUined by a donor-acceptor distance of 3.5 A_ and a hydrogen-donor-acceptor angle of less than 30 • .The calculations was performed using the HBonds tool in Visual Molecular Dynamics (Humphrey et al, 1996).Key hydrogen bonds are listed in Appendix Tables S4, S5, and S7, and the key ones are highlighted in the structural snapshots in the main text.The last 90 ns were included in the analyses.
The simulations with a restrained backbone were also used to extract other key interactions between the lumenal domains of TRAP subunits.The rerun functionality of gmx mdrun was used to extract the short-range (no long-range electrostatics were included) Coulombic and van der Waals (Lennard-Jones potential) interactions.The major contributors to these energies are listed in Appendix Table S6.

All-atom simulations for protein dynamics and membrane-protein interactions
We studied the behavior of the complex formed by Sec61, TRAP, and the ribosome as well as its multiple sub-complexes.To this end, we embedded the atomistic models of 1) Sec61 alone, 2) TRAP alone, 3) Sec61 with TRAP, or 4) Sec61 with TRAP and parts of the ribosome into a lipid bilayer.Additionally, we performed a simulation of a protein-free bilayer as an additional control.The bilayer composition was set to mimic that of the ER membrane (Bollen & Higgins 1980;Colbeau et al, 1971;Davison & Wills 1974;Casares et al, 2019;Van Meer et al, 2008), and it contained 54% phosphatidylcholine, 21% phosphatidylethanolamine, 10% phosphatidylinositol, 4% phosphatidylserine, 4% sphingomyelin, and 7% cholesterol.Since no information on the acyl chain pairing with the different lipid classes is available in the literature, we modeled them as palmitate and oleate, except for sphingomyelin, which had a palmitate chain (Keenan & Morre 1970).For systems containing ribosome, the proteins and the parts of the RNA strands that were located in the vicinity of Sec61 or TRAP were included in the model.The sizes of the lipid membranes were adapted to the lateral extent of the protein, whereas the number of water molecules was adjusted to solvate the entire protein.The system dimensions and molecule counts are provided in Appendix Table S1.
For the simulations containing parts of the ribosome, the atoms of the ribosomal proteins and RNA that lie far away from the Sec61 and TRAP were restrained to avoid having to model the entire ribosome, yet capturing the key interaction sites and the ribosomal anchoring effect due its large size.The systems were generated in CHARMM-GUI (Jo et al, 2008;Wu et al, 2014), downloaded in GROMACS-formats, and subjected to the standard equilibration protocol.We simulated all systems using the CHARMM family of force Uields, namely CHARMM36m (Huang et al, 2017) for the protein, CHARMM36 for lipids (Klauda et al, 2010;Wang & Klauda 2017;Lim et al, 2012) and RNA (Denning et al, 2011), and the CHARMM-compatible TIP3P model (Jorgensen et al, 1983;Durell et al, 1994) S1: The summary of unrestrained atomistic MD simulations of the Sec61/TRAP/ribosome complex and its various sub-complexes.The membrane compositions between the CHARMM and Amber force Uield families differ somewhat depending on the availability of certain lipid species in their lipid libraries (see text for detailed compositions).The dimensions (in A_ ) are given in the membrane plane (x/y) or perpendicular to it (z).Additionally, the system containing all components (Sec61, TRAP, and parts of the ribosome) was also simulated using the Amber force Uields, namely FF19SB (Tian et al, 2019) for the protein, Lipid21 (Dickson et al, 2022) for the lipids, OL3 force Uield for RNA (Zgarbová et al, 2011), and the standard TIP3P model (Jorgensen et al, 1983) for water.For the simulation with Amber force Uields, not all lipid types found in the ER membrane were available in the lipid library, so for these simulation we adapted the composition to contain 62.75% phosphatidylcholine, 24.5% phosphatidylethanolamine, 4.75% phosphatidylserine, and 8% cholesterol.
The membrane systems with proteins were simulated for 2 µs each.The protein-free membrane system was simulated for 500 ns.The simulation parameters were consistent within simulations using the same force Uield, and are listed below.
The protein stability was evaluated by calculating the root mean squared deviation (RMSD) of the protein backbone, after Uitting the backbone structure Uirst.This analysis was performed separately on Sec61 (all subunits together) or TRAP (all subunits together).
From these simulations involving full protein and lipid dynamics, we studied the membrane perturbations by the different protein assemblies.In long simulations, the membrane proteins rotate and their local effects on the local membrane properties are smeared out.To account for this rotation, we Uirst centered the protein, then RMSD-Uitted the protein to a Uixed orientation in the plane of the membrane.As such rotations would cause the membrane to cross the edges of the simulation box, the simulation box was simultaneously enlarged.We then performed the analyses with g_lomepro Gapsys et al, (2013) on this larger system, and the last 1.5 µs were included in these spatial analyses.We extracted the leaUlet shapes, the local membrane thickness, and the local order of the palmitate chains in each leaUlet.Snapshots demonstrating the effect of the protein on the leaUlet shape and membrane thickness were rendered using the tachyon renderer in Visual Molecular Dynamics (Humphrey et al, 1996).Additionally, twodimensional proUiles of the leaUlet position, membrane thickness, and acyl chain order were resolved by projecting the 3-dimensional position, thickness, and order parameter maps onto a line running parallel or perpendicular to the axis connecting Sec61 and TRAP.An area of 120 A_ by 90 A_ covered the entire extent of the protein, and the averaging of the projected 3D proUiles was done over these extents of the chosen axes.
Validation of the spatial analyses was performed by comparing the thickness and order parameter maps to those resolved from a protein-free system.The values in the map were histogrammed and Uitted with a single (protein-free system) or a double (protein-containing systems) Gaussian.
We analyzed the openness of the lateral gate in Sec61 as a distance between the centers of mass of TM helices 2 and 7.This analysis was performed using gmx distance from the GROMACS simulation software (Páll et al, 2020).

All-atom simulations of bicelle curving
The complex formed by Sec61 and TRAP, and maintained in a certain conformation by ribosomal anchoring, seemed to induce local curvature in the membrane simulation.However, as the Ulat membrane cannot bend signiUicantly due to the periodic simulation box, we repeated this simulation in a bicelle model.To this end, Sec61, TRAP, and parts of the ribosome were inserted in a POPC membrane and hydrated.We chose this single-component membrane to avoid any lipid demixing due to the bicelle edges.The system was generated in CHARMM-GUI (Jo et al, 2008;Wu et al, 2014), and GROMACS-compatible simulation Uiles were downloaded (Lee et al, 2016).Then, we carved out a circularly shaped region with a diameter of ∼210 A_ from the membrane, and subjected it to the standard CHARMM-GUI equilibration protocol.Then, the system was simulated for 1 µs using the suggested simulation parameters for CHARMM with GROMACS (Lee et al, 2016) with one exception: The compressibility of the barostat in the plane of the bicelle was set to 0 so the area in that plane was kept constant.The other used simulation parameters are listed below in detail.The bicelle system was simulated for 1 µs.The bicelle simulations were analyzed in the same manner as the membrane ones described above with g_lomepro (Gapsys et al, 2013).

Coarse-grained simulations
Atomistic simulations are limited in size and time scale by the amount of computing power available.In a smaller membrane, signiUicant curvature cannot build up due to the system periodicity.In the bicelle system, the bicelle will eventually drift away from the protein, limiting the achievable time scale.Thus, we also set up a coarse-grained simulation model for the Sec61α/TRAP complex.The coarse-grained protein was embedded in a lipid bilayer formed solely by POPC.This simple composition was chosen, as the recent Martini 3 force Uield (Souza et al, 2021) was used, and it currently lacks a vast and veriUied lipid library.The membrane contained a total of 3514 POPC molecules and it was solvated by ∼195,000 water beads.Neutralizing ions and ∼157 mM of NaCl were included.The system dimensions were ∼340 × 340 × 240 A_ 3 .The system was simulated for 20 µs with the protein backbone restrained.These restraints locked the Sec61/TRAP complex into the conformation observed in the presence of ribosomal anchoring.The perturbations caused by the presence of the Sec61/TRAP complex were again evaluated using g_lomepro.Due to the restraints, no additional centering or alignment steps were necessary.
The ability of Sec61, TRAP, the different TRAP subunits, and the Sec61/TRAP complex to increase membrane permeability was probed by analyzing the phospholipid Ulip-Ulops.To this end, we also simulated Sec61 (all subunits included), TRAP (all subunits included), as well as the four TRAP subunits separately in a POPC membrane for 20 µs.In these simulations, the protein backbone structure was restrained.Details on all simulations are shown in Appendix Table S2.The Ulip-Ulops were analyzed based on the position of the lipid phosphate beads.A lipid was assigned to the upper (lower) leaUlet, when this phosphate bead was at least 4 A_ above (below) the membrane midplane.The coordinates were processed every 100 ns, and change in the leaUlet identity was considered a Ulip-Ulop.S2: Summary of the coarse-grained simulations systems performed using the Martini 3 force Uield.The dimensions (in A_ ) are given in the membrane plane (x/y) or perpendicular to it (z)

Simulation parameters
The simulation parameters used with the different force Uields are listed below in detail.

All-atom CHARMM family
The simulations were performed using the recommended simulation parameters for the CHARMM36 force Uield in GROMACS (Lee et al, 2016).Namely, buffered Verlet lists were used to keep track of neighbour atoms (Páll & Hess 2013).The Lennard-Jones potential was cut off at 1.2 nm, and the forces were switched to zero between 1.0 nm and the cut-off distance.The smooth particle mesh Ewald (PME) algorithm was used for the calculation of long-range electrostatics (Darden et al, 1993;Essmann et al, 1995).The temperature of the protein (and RNA), the membrane, and the solvent were separately maintained at 37°C using the stochastic velocity rescaling thermostat (Bussi et al, 2007) with a time constant of 1 ps.The pressure along the membrane plane and normal to it were coupled to a semi-isotropic Parrinello-Rahman barostat (Parrinello & Rahman 1981) with a target pressure of 1 bar, compressibility of 4.5×10 −5 bar −1 , and a time constant of 5 ps.The bonds involving hydrogen atoms were constrained using P-LINCS (Hess et al, 1997;Hess 2008), whereas water structure was constrained using SETTLE (Miyamoto & Kollman 1992).
All-atom Amber family The system was downloaded in GROMACS-compatible formats from CHARMM-GUI (Lee et al, 2016;Lee et al, 2020), and subjected to the standard equilibration protocol of CHARMM-GUI.The system was simulated for 2 µs with an integration time step of 2 fs using GROMACS 2021 (Páll et al, 2020).
The simulation parameters provided by CHARMM-GUI were used (Lee et al, 2020).Namely, buffered Verlet lists were used to track the neighbouring atoms for non-bonded interactions (Páll & Hess 2013).The Lennard-Jones potential was cut off at 0.9 nm, and a plain cutoff was used.Corrections due to the cutoff were performed to both energy and pressure (Shirts et al, 2007).The smooth particle mesh Ewald algorithm was used to calculate long-range electrostatics (Darden et al, 1993;Essmann et al, 1995).The temperatures of the protein (with RNA), the lipids, and the solvent were maintained at 37°C by coupling them to a Nose-Hoover thermostat (Nosé 1984;Hoover 1985) with a time constant of 1 ps.The pressure was maintained at 1 bar with a semi-isotropic Parrinello-Rahman barostat (Parrinello & Rahman 1981) with a time constant of 5 ps and a compressibility of 4.5×10 −5 bar −1 .The bonds involving hydrogen atoms were constrained using P-LINCS (Hess et al, 1997;Hess 2008).Waters were constrained with SETTLE (Miyamoto & Kollman 1992).

Coarse-grained Martini 3
The coarse-grained simulation systems were generated with the CHARMM-GUI Martini maker (Qi et al, 2015) and downloaded in GROMACS-compatible formats.The latest version 3.0 (Souza et al, 2021) of the Martini force Uield was used for the protein.All simulations were run for 20 µs with a time step of 20 fs using GROMACS 2021 (Páll et al, 2020).We used the recently suggested "New-RF" simulation parameter set (de Jong et al, 2016).The Lennard-Jones potential was cut off at 1.1 nm, a distance at which the potential was shifted to zero.For electrostatics, a reaction Uield approach with a cutoff of 1.1 nm and a dielectric constant of ∞ was used for efUiciency (de Jong et al, 2016).The stochastic velocity rescaling thermostat (Bussi et al, 2007) with a time constant of 1 ps was applied separately to the protein, the lipids, and the solvent.A semi-isotropic Parrinello-Rahman barostat (Parrinello & Rahman 1981) with a time constant of 12 ps, compressibility of 3×10 −4 bar −1 , and a target pressure of 1 bar was applied semi-isotropically.Electrostic interactions were screened by a dielectric constant of 15.Constraints present inherently in the force Uield were handled by P-LINCS (Hess et al, 1997;Hess 2008) S5: Key hydrogen bonds between TRAP subunits in the backbone-restrained simulations using two complementary sets of protein and lipid force Uields.The occupancies observed with both force Uields are also listed.
for water. .