Research Article: Improving the efficiency of molecular replacement by utilizing a new iterative transform phasing algorithm

Date Published: September 01, 2016

Publisher: International Union of Crystallography

Author(s): Hongxing He, Hengrui Fang, Mitchell D. Miller, George N. Phillips, Wu-Pei Su.


An iterative transform algorithm is proposed to improve the conventional molecular-replacement method for solving the phase problem in X-ray crystallography. Several examples of successful trial calculations carried out with real diffraction data are presented.

Partial Text

Despite the success of molecular replacement (MR) as a major tool for new macromolecular structure determinations (Rossmann, 1972 ▸, 1990 ▸, 2001 ▸; Scapin, 2013 ▸), it requires a high degree of similarity between the template and target structures, a condition that is not met for many unknown structures. In particular, even for a properly placed template, there could be errors in the calculated density map that prevent further crystallographic model building.

After rotation and translation with PHENIX Phaser-MR (McCoy et al., 2007 ▸; Adams et al., 2010 ▸), the template structure was placed in the unit cell. The initial phase estimate was calculated from the positioned template structure. An electron-density map was calculated using the observed Fourier magnitudes of the target crystal with the initial phases estimated from the Fourier transform of the placed template model. This type of map is generally not ready for model building due to ambiguity and lack of connectivity. Thus density modification is necessary. A weighted-average electron-density map was calculated to identify a protein mask from the calculated electron-density map. The HIO method has proved to be a very effective solvent-flattening and phase-recovery technique (Liu et al., 2012 ▸; He & Su, 2015 ▸; Ayyer et al., 2016 ▸) by consistently applying constraints in real space and Fourier space (Marchesini, 2007 ▸). Like real-space phasing methods (Su, 2008 ▸), the protein mask serves as a high-density support. Solvent occupies the region outside the support and the calculated electron density in the solvent region is flattened progressively via a negative feedback scheme. In addition to HIO, conventional histogram matching (HM) (Zhang & Main, 1990a ▸,b ▸) was also employed to modify the calculated electron density inside the protein mask. Fig. 1 ▸ shows the flowchart of the iterative method starting from phases estimated from the placed template model.

We selected three known trial structures from those that were used in the Rosetta MR blind tests (DiMaio et al., 2011 ▸) for our initial evaluation. These structures have also been solved and analyzed in several subsequent MR methods development studies (Brunger et al., 2012 ▸; Terwilliger, Dimaio et al., 2012 ▸; Terwilliger, Read et al., 2012 ▸; DiMaio, 2013 ▸; Terwilliger et al., 2013 ▸; Carrozzini et al., 2015 ▸). All template models used in our trial calculations were the truncated and pruned models that were used for those tests and were downloaded from These are listed in Table 1 ▸.

The success of the MR method depends significantly on the similarity between the target and template structures (Scapin, 2013 ▸; Schwarzenbacher et al., 2004 ▸). As sequence identity between the target and template structures decreases, the increased likelihood of significant structural differences (Chothia & Lesk, 1986 ▸; Gan et al., 2002 ▸; Krissinel, 2007 ▸) makes finding an MR solution more difficult. Even when a distant homolog can be placed, the poor quality of the starting map prevents rebuilding and refinement. For example, when the sequence identity is between 30% and 40%, it is usually possible to solve the phase problem using Phaser-MR, but sometimes it is more difficult. If the sequence identity is between 20% and 30%, it usually becomes difficult to retrieve the phase using Phaser-MR. If it is possible then one needs to search and prepare a proper model carefully. MR-Rosetta has made further progress towards MR with lower sequence similarity templates. It is usually possible to solve the target structure when the sequence identity is between 20% and 30%, but it still requires careful model search and preparation. When the sequence identity is below 20%, Phaser-MR is unlikely to work and it is difficult for MR-Rosetta to work, if at all possible. We have provided a supplementary approach which may be critically needed for MR when the sequence identity is below 30%.

The HIO iterative phasing approach is capable of improving the MR method when the sequence identity between the target and the template structures is below 30%. The MR-placed template structure provides an initial phase estimate and an initial protein mask. It is simple and straightforward to apply this approach and it requires less computing resources than MR-Rosetta. Our calculations typically took less than 1 h on a standard laptop computer. Because no model building is involved during the iteration and the molecular mask evolves, the final density map has less model bias. The results of our trial calculations show that the final density map is ready for automated model building.




Leave a Reply

Your email address will not be published.