Flexible backbone sampling methods to model and design protein alternative conformations

Abstract

Sampling alternative conformations is key to understanding how proteins work and engineering them for new functions. However, accurately characterizing and modeling protein conformational ensembles remain experimentally and computationally challenging. These challenges must be met before protein conformational heterogeneity can be exploited in protein engineering and design. Here, as a stepping stone, we describe methods to detect alternative conformations in proteins and strategies to model these near-native conformational changes based on backrub-type Monte Carlo moves in Rosetta. We illustrate how Rosetta simulations that apply backrub moves improve modeling of point mutant side-chain conformations, native side-chain conformational heterogeneity, functional conformational changes, tolerated sequence space, protein interaction specificity, and amino acid covariation across protein-protein interfaces. We include relevant Rosetta command lines and RosettaScripts to encourage the application of these types of simulations to other systems. Our work highlights that critical scoring and sampling improvements will be necessary to approximate conformational landscapes. Challenges for the future development of these methods include modeling conformational changes that propagate away from designed mutation sites and modulating backbone flexibility to predictively design functionally important conformational heterogeneity.

Figure 1. The backrub move and its applications in Rosetta

(A) The Richardson group originally described the “Backrub” move asa rotation around the Cα_i-1 and Cα_{i +1}axis by τ_disp, along with simultaneous peptide plane rotations (τ₁ and τ₂), without disturbing other surrounding atom coordinates. (B) By changing the position of the Cα_i-Cβ_i bond vector, this move can couple side chain rotameric changes with small local backbone adjustments.(C)In Rosetta, the generalized backrub move isa single rotation that can also include longer intervals and other backbone atom types as pivots for the rotations. (D) Implementing backrubs as a Monte Carlo move in Rosetta enables a variety of flexible backbone prediction and design applications that are described in this paper: predicting mutant conformations (Figure 2), modeling alternative conformations (Figures 3 and 4), coupling conformational and sequence plasticity (Figure 5), and designing amino-acid co-variation at protein interfaces (Figure 6).

Figure 2. Backrub sampling improves the prediction of mutant side chain conformation compared to fixed backbone simulations

(A) Example predictions of mutant conformations given the wild type structure, with the indicated PDB codes and point mutations. The mutant crystal structure is shown in yellow compared to the wild type crystal structure (green, left), prediction based on fixed backbone simulations (magenta, center), or prediction based on backrub flexible backbone sampling (cyan, right). (B) The overall quantification of the results of fixed backbone and backrub predictions over a set of 136 buried (SASA <5%) side chains with conformations differing by more than 0.2 Å between mutant and wild type. The median RMSD decreases from 1.17 Å to 0.98 Å. Shown are box plots with the median as a black line and the 25-75^th percentiles in the shaded box with outlier-corrected extreme values as dashed lines. (C) A scatter plot representation of the data in (B) shows that for many mutant structure predictions backrub leads to large improvements compared to fixed backbone simulations.

Figure 3. Backrub sampling improves the prediction of alternative side chain conformations observed in protein crystal structures

(A) Electron density sampling by Ringer around the χ1of R29 from PDB 1KWN reveals high electron density for the primary conformation 60° and a secondary peak (indicated by the black arrow), above the 0.3σ threshold that enriches for alternative conformations (shaded green area), near the 180° rotameric bin. (B)2mFo-DFc electron density surrounding R29 from PDB 1KWN contoured at 1σ (blue mesh) and 0.3σ (cyan mesh). The original PDB model is shown in yellow, with an alternative conformation identified by Ringer and modeled with qFit at 25% occupancy shown in green. (C) Example predictions with Rosetta, with the indicated PDB codes and residues. Sampling of side chain conformations (yellow) starting from alternative conformations (green, right) is improved by flexible backbone backrub moves (cyan, right) compared to fixed backbone side chain only sampling (magenta, center). (D) The overall quantification of the results, showing that backrub sampling increases identification of discrete side chain local minima modeled as alternative conformations by qFit compared to fixed backbone models over a set of 152 side chains with solvent accessibility less than 30%. The median RMSD decreases from 0.47 to 0.33. Box plots are shown as in Figure 2C. (E) A scatter plot representation of the data in (D) shows that backrub leads to large improvements compared to fixed backbone for many alternative conformation predictions.

Figure 4. Backrub sampling can be used to model functionally relevant alternative conformations

(A) NMR relaxation experiments detect that residues in a dynamic network (cyan transparent surface) undergo a collective exchange between a major and minor conformation with and without substrate present (Eisenmesser et al., 2002; Eisenmesser et al., 2005; Fraser et al., 2009). Room temperature X-ray data collection and qFit multiconformer refinement identify a major (green) and minor (yellow) conformation providing a structural basis for the NMR observations. Additional alternative conformations are shown in orange. (B) Rosetta simulations can access the alternative conformation starting from either state (yellow/green) using backrub (right, cyan), but not fixed backbone (middle, magenta) sampling methods.

Figure 5. Rosetta generates near native ensembles using backrub sampling

(A) Cα cartoons of Rosetta generated conformational ensembles using backrub sampling at different temperatures, compared to the fixed backbone (top left). Higher temperatures increase the conformational diversity and can increase agreement with experimental data. The PDZ domain structure is shown in white and peptide in grey. (B) Example results from the sequence tolerance protocol to predict peptide specificity for 4 PDZ domains (DLG1-2, MPDZ-12, MPDZ-13 and Erbin) and 1 PDZ domain point mutant (Erbin V83K); peptide positions are indicated using the standard nomenclature for PDZ domain motifs, with 0 denoting the C-terminal residue, followed by -1, -2, etc. Without backbone flexibility Rosetta fails to predict important residue preferences observed in experimental phage display selections, such as valine at the 0 position or tryptophan at the -1 position for DLG1-2 and Erbin.

Figure 6. Rosetta backrub design methods capture features of evolutionary amino-acid co-variation

(A) Comparison between designed and natural intermolecular amino acid co-variation for histidine kinases (HK) and their cognate response regulators (RR). Each point represents a pair of amino acid positions. Natural co-variation was quantified using a mutual information based metric for all pairs of positions in a multiple sequence alignment of HKs concatenated to their cognate RRs. A backrub ensemble of 500 structures was generated for a HK/RR complex (PDB ID 3DGE) and RosettaDesign was used to predict one low energy sequence for each structure in the ensemble. Designed co-variation was quantified for all pairs of positions in the resulting multiple sequence alignment of 500 sequences. The red lines indicate the threshold cut-off for the top 30 designed co-varying intermolecular pairs (horizontal) and the top 30 natural co-varying pairs (vertical). The 12 intermolecular pairs of positions that are highly co-varying in both designed and natural sequences are highlighted in green. (B) The structure of a HK/RR complex with amino acids that are involved in highly co-varying intermolecular pairs in both natural and designed sequences are shown in green and stick representation. (C) Close-up of the 12 intermolecular co-varying pairs. Each of these 12 pairs of amino acids forms a physical interaction across the interface of the complex.