Computational protocol: Routine phasing of coiled-coil protein crystal structures with AMPLE

Similar protocols

Protocol publication

[…] AMPLE (Bibby et al., 2012) was used to generate 1000 cheap ab initio predictions (decoys) using Rosetta (Simons et al., 1997, 1999, 2001), which were clustered by r.m.s.d. using SPICKER (Zhang & Skolnick, 2004). The appearance of large clusters is considered to be an indication of reliable modelling, so up to 200 decoys from the largest cluster were selected and the structural variation amongst the aligned Cα atoms was calculated using THESEUS (Theobald & Wuttke, 2006), allowing the residues to be grouped according to their structural similarity. Structural variance along the chain predicts deviation from the true structure (Bibby et al., 2012; Qian et al., 2007), so AMPLE constructed a range of search models by pruning back residues at a series of 20 variance levels (truncation thresholds). This generated a pool of truncated and fragmented models that served as the basis for the assembly of search ensembles.The models under each truncation level were structurally clustered again in a ‘subclustering step’. The centroid representative model of each truncation step served as the basis for the assembly of three increasingly diverse clusters containing structures superimposable on the centroid with overall Cα r.m.s.d. values within 1, 2 or 3 Å using MaxCluster (http://www.sbg.bio.ic.ac.uk/~maxcluster). These clusters were limited to at most 30 structures and were structurally aligned with THESEUS to create an ensemble.To generate the final search models, these ensembles were subjected to three different side-chain treatments: polyalanine, where all the side chains were stripped back to Cβ; reliable side chains, where only those side chains that are usually well modelled (Shapovalov & Dunbrack, 2007) were kept, with others being stripped back to alanine; and all-atom, where all side chains were kept.The ensemble search models were processed with MrBUMP (Keegan & Winn, 2008), which performed MR using Phaser (Storoni et al., 2004; McCoy et al., 2005; 2007), main-chain tracing by SHELXE (Sheldrick, 2008; Usón et al., 2007; Thorn & Sheldrick, 2013) and automatic rebuilding of the SHELXE trace with ARP/wARP (Cohen et al., 2008; Langer et al., 2008) and Buccaneer (Cowtan, 2006). [...] AMPLE v.0.1.0 (Bibby et al., 2012) in the CCP4 suite v.6.4.0-008 (Winn et al., 2011) was used for this study. The study excluded fragments homologous to the target from the ab initio model building in order to treat each case as if it were a novel fold. REFORIGIN (Winn et al., 2011) was used to assess the accuracy with which search models were placed with respect to the crystal structure. However, this tool captures only the in-register placements that lie at the heart of conventional MR. In this work, it rapidly became clear that out-of-register overlaps between search models and crystal structures could contribute partially or fully to successful structure solution. We therefore developed the residue-independent overlap (RIO) score to quantify these successes. The RIO score counts the number of Cα atoms in the MR result that are within 1.5 Å of a Cα atom in the crystal structure (regardless of how the Cα atoms are related in sequence), including only stretches where at least three consecutive Cα atoms of the MR result overlie matching Cα atoms in the deposited structure. This quantifies how well the MR result maps onto the crystal structure, without making any assumptions about the correctness of the model or the placement with respect to register. The total RIO score is composed of an in-register (i.e. residues in the placed MR search model correctly overlay their counterparts in the crystal structure) RIO_in component and a second RIO_out component summing out-of-register overlays. Scripts for calculating RIO scores are available on request from the authors: they require both CCP4 and PHENIX (Adams et al., 2010) installations. A detailed description of the RIO score calculation pipeline is included in the Supporting Information. […]

Pipeline specifications