Hybrid Refinement of Heterogeneous Conformational Ensembles Using Spectroscopic Data

Abstract

TheoryA more complex formulation of bias-resampling ensemble refinement (BRER) may be used to perform more advanced sampling. Rather than draw each conformation x from the previous conformational estimate "#$ , a history is maintained of k refinement rounds, so that the conformation x is drawn from the union of "#$ , "#& , … , "#( . The conformational estimate " is then obtained as before: the conformations are updated using a biased MD simulation such that the updated estimate $…" will optimally reproduce *++, ( ). Just as with the formulation provided in the main text, over the course of multiple rounds of refinement, the conformational estimate { } should yield a distribution 2 ( ) that converges on *++, ( ). Methods Molecular dynamics simulations Setup and equilibration of syntaxin-1aIn order to best demonstrate the ability of our method to sample backbone conformational change and rare conformational states, we started all simulations of syntaxin-1a from its closed state. We obtained an initial structure of closed syntaxin by extracting the soluble domain from the crystal structure of syntaxin in complex with Munc-18 (PDB ID 3C98) 1 . The system was solvated with approximately 90,000 TIP3P water molecules and ions were added to obtain a system with 150 mM NaCl and no net charge. Simulations were run in Gromacs 2 using the CHARMM36 3 force field. The system was energy minimized using the steepest-descent integrator for 5000 steps or until the largest force was less than 500 kJ/mol/nm 2 , whichever came first. A brief 100 ps equilibration was run using NPT conditions using the velocity-rescaling thermostat 4 at 310 K with a 2-ps time constant and pressure maintained at 1 bar using the Parrinello-Rahman barostat 5 with a 10-ps time constant. Covalent bonds were constrained using LINCS, and long-range electrostatics were treated using Particle Mesh Ewald 6 . For each set of ensemble simulations, we generated 50 identical replicas from the equilibrated structure and used these replicas as initial states for production runs. Production simulationsAll production simulations were run under the same NPT conditions described above. DEER-derived distance distributions were smoothed with a Gaussian filter. The smoothing parameter σ was chosen to reflect the experimental uncertainty in the fine modes of the DEER-derived distance distributions, 2 Å for all three distributions. Histograms were calculated using 1 Å bins. These distributions were then incorporated into MD simulation using each of three ensemble methods, detailed below. Production simulations were carried out using 50 ensemble members and 5 μs of simulation data were collected for each refinement method except EBMetaD. The reason for this exception is described in "EBMetaD simulations." Simulations were run using Gromacs 2 and the gmxapi Python API 7 , which permits introduction of user-defined biasing potentials. BRER simulationsTo sample the syntaxin conformational ensemble, we performed five iterations of BRER for each of 50 ensemble members. Each iteratio...

gmxapi: a GROMACS-native Python interface for molecular dynamics with ensemble and plugin support

“…We illustrate this using a starting crystal structure for the BG505 SOSIP of HIV gp41[29] and previously reported DEER spectroscopy data of 5 different spin labels positions on this SOSIP[30]. The gp41 trimer was asymmetrically restrained (each of the three monomer-monomer distances was sampled separately), and the previously reported BRER algorithm for refining heterogeneous conformational ensembles[28] was applied. This involves a custom force plugin for GROMACS, and gmxapi was used to run an array of 250 refinement simulations, each randomly sampling target distance values from the experimental distribution for each measured spin-label pair.…”

“…The Python interpreter serves as a convenient tool for dynamically loading object libraries at run time. Gmxapi establishes a protocol 18 for providing MD extension code objects to the GROMACS simulator during launch. This allows researcher code to be executed during the MD integration loop with minimal overhead.…”

“…Gmxapi establishes a facility[28] for providing MD extension code objects to the GROMACS simulator during launch. Binary objects can be loaded at runtime via the Python interpreter.…”

Methodology for rigorous modeling of protein conformational changes by Rosetta using DEER distance restraints

“…Several algorithms have recently been developed to refine ensembles of spin label rotamers by employing multilateration [16,[32][33][34][35][36]. Multilateration refers to the determination of an object's position in three-dimensional space given its distance from a constellation of points; common applications include the positioning of electronic devices using the Global Positioning System and of earthquakes epicenters using time-of-arrival data [37].…”

“…Molecular dynamics simulations have been used to determine a set of optimized rotamers from explicitly modeled spin labels restrained by experimental distance distributions [14,21,38,39]. Alternatively, rotamer libraries have been precomputed and reweighed using either Monte Carlo [32,35], singular value decomposition [34], or nonlinear least-squares minimization [33]. The positions of these labels can, in turn, be used to more precisely locate paramagnetic ligands or metal ions [35,36,40,41], as well as make small-scale refinements to protein structures [16,42].…”