View metadata, citation and similar papers at core.ac.uk brought to you by CORE provided by Elsevier - Publisher Connector Structure, Vol. 13, 473–482, March, 2005, ©2005 Elsevier Ltd All rights reserved. DOI 10.1016/j.str.2005.01.006 Software Extensions to UCSF Chimera for Interactive Visualization of Large Molecular Assemblies Thomas D. Goddard, Conrad C. Huang, semblies (Figure 1). The icosahedral capsid contains 3 and Thomas E. Ferrin1,* million atoms and cannot be opened in existing desk- Department of Pharmaceutical Chemistry top analysis software due to insufficient memory. The University of California, San Francisco capsid can be described by using only 50,000 atom po- San Francisco, California 94143 sitions and the 60-fold icosahedral symmetry, but most existing software is not able to read and use these sym- metry matrices. Current low-resolution display styles such as ribbons and molecular surfaces are too de- Summary tailed for making an informative display of this large system, and the excessive detail prevents smooth in- Many structures of large molecular assemblies such teractive rotation of the model because of insufficient as virus capsids and ribosomes have been experi- graphics rendering speed. The atomic contacts of viral mentally determined to atomic resolution. We con- RNA with the capsid are of interest for this system. The sider four software problems that arise in interactive simplest algorithm for finding these contacts, calculat- visualization and analysis of large assemblies: how to ing distances between all pairs of atoms, requires sev- represent multimers efficiently, how to make cartoon eral minutes of computation and thus is not fast enough representations, how to calculate contacts efficiently, for interactive exploration of the contacts. In summary, and how to select subassemblies. We describe tech- analysis of this large system runs into problems of lim- niques and algorithms we have developed and give ited computer resources that require better data repre- examples of their use. Existing molecular visualiza- sentations and algorithms and new features such as low-resolution display styles. tion programs work well for single protein and nucleic Virus capsids are the largest structures in the PDB. acid molecules and for small complexes. The meth- All other PDB structures of cellular machinery are sig- ods presented here are proposed as features to add nificantly smaller in number of atoms, and they can be to existing programs or include in next-generation opened in current-generation molecular analysis pro- visualization software to allow easy exploration of as- grams. Yet, software designed for small complexes is semblies containing tens to thousands of macromole- cumbersome to use on these larger systems. An exam- cules. Our approach is pragmatic, emphasizing sim- ple is provided by a recent paper describing the in- plicity of code, reliability, and speed. The methods teractions of 27 ribosomal proteins with the 23S ribo- described have been distributed as the Multiscale somal RNA (Klein et al., 2004)(Figure 2). Each of the extension of the UCSF Chimera (www.cgl.ucsf.edu/ intermolecular interfaces between the 27 proteins and chimera) molecular graphics program. the 101 helices of the 23S RNA are individually ana- lyzed and illustrated in that work, and differences be- tween two different organisms are noted. Interactively Introduction exploring this large set of interfaces would benefit from the ability to switch views from the full assembly to indi- Advances in electron cryomicroscopy and X-ray crys- vidual proteins and neighboring helices with minimum tallography have enabled structures of many cellular effort. Low-resolution depiction of the whole assembly machines to be determined. Structures such as ribo- and methods of treating the 101 helices comprising the somes, virus capsids, molecular motors, cytoskeletal ribosomal RNA as natural structural units are useful ca- filaments, proteasomes, nucleosomes, chaperonins, pabilities for facile navigation of this assembly. transmembrane channels, and pumps, composed of The virus and ribosome studies described above tens to thousands of macromolecules, have been de- were carried out without the benefit of interactive analysis termined to atomic resolution. The inventory of large software designed for handling these large complexes. Such studies can be done laboriously with software de- molecular assemblies is growing rapidly. At the start of signed for small complexes, as well as with nonin- 2000, the Protein Data Bank (PDB) (Berman et al., 2000) teractive software and custom-written analysis pro- contained 137 structures with 10 or more chains, while grams. For virus capsids, files containing small pieces at the start of 2004, there were 517 such structures. with just the needed asymmetric units can be created Existing software to interactively display and analyze with specialized programs and then explored inter- macromolecules is difficult to use on these large as- actively. For the helices of the 23S rRNA, many exact semblies. residue sequence ranges can be entered by hand to A published study of viral RNA bound to the outside specify regions of interest during interactive analysis. of bluetongue virus capsid (Diprose et al., 2002) il- Software that provides more efficient ways of exploring lustrates several limitations with existing interactive large assemblies will speed up the elucidation of the molecular analysis programs when applied to large as- functions of these systems. Past efforts to develop analysis software for large as- semblies have addressed some of the limitations. A *Correspondence: [email protected] study done 6 years ago (Macke et al., 1998), when 1Lab address: http://www.cgl.ucsf.edu many fewer experimental structures were available, fo- Structure 474 Figure 1. Bluetongue Virus Capsid with Bound Viral RNA (A) Two protein layers of bluetongue virus capsid with viral RNA bound to the surface (PDB models 2btv and 1h1k). Each molecule is displayed by using a low-resolution sur- face. RNA contacts seven protein monomers (shown in blue) on the surface. The model contains approximately 3 million atoms (no hydrogens) and 900 molecular components. (B) Atomic contacts between RNA and cap- sid proteins within a 5 Å range shown in black. cused on building molecular assemblies by using sym- (Humphrey et al., 1996) provide many of these capabili- metry operations and displaying them with variable res- ties. Analysis of large assemblies requires the same olution surfaces using spherical harmonics (Duncan tools to study many local regions in atomic detail. On and Olson, 1995). Other work (Bajaj et al., 2004) has top of these, better facilities to navigate to small re- explored computer graphics techniques to efficiently gions of interest, restrict attention to them, and show display molecular properties at various resolutions by them in the context of the larger assembly are needed. using texture maps on surfaces and volumetric render- The main virtue of the tools we have developed and ing. Another group has developed a web service to ap- describe here is that they interoperate within a program ply symmetry matrices to generate atomic coordinates that provides the rich set of analysis capabilities al- of crystal packings and visualize them (Hussain et al., ready created for studying smaller systems. To suc- 2003). Most macromolecular structures are determined cessfully augment an already complex software pack- by crystallography, so this type of large molecular as- age, we have favored simple code, reliable algorithms, sembly is common. and speed. Our aim is to develop the breadth of capa- The distinctive feature of the work presented here is bilities necessary for effective analysis of large assem- our focus on software system integration. Many meth- blies, rather than maximal performance of any one soft- ods exist for analyzing single molecules and small com- ware component. plexes, for example, structure and sequence align- The following sections address the most essential ments, molecular dynamics, binding site prediction, software capabilities needed for exploring large assem- electrostatics calculations, hydrogen bond identifica- blies. We describe how to efficiently represent multi- tion, superimposing homologous or mutant structures, meric molecular complexes in computer memory, an testing conformational changes, and fitting experimen- algorithm for creating low-resolution surfaces to repre- tal density maps. Programs such as UCSF Chimera sent molecules, an algorithm for finding atomic con- (Pettersen et al., 2004), PyMol (DeLano, 2002), and VMD tacts between large sets of atoms, and a hierarchical Figure 2. Protein-RNA Contacts in Large Ri- bosomal Subunit (A) Arrangement of 27 ribosomal proteins contacting 23S ribosomal RNA (gray) and 5S ribosomal RNA (black) (PDB model 1s72). Each protein and RNA is represented by a low-resolution surface. This coarse view is used to select and focus on individual pro- tein-RNA interactions. (B) Displaying atoms as spheres creates a pebbly surface that reduces the effec- tiveness of the three-dimensional cues pro- vided by object illumination techniques. (C) Ribbon display style permits seeing through the structure. Clear display of pro- tein-RNA interactions requires restricting the view to small pieces of the assembly. Extensions to UCSF Chimera 475 representation of quaternary structure useful for select- column. Most molecular visualization software is un- ing pieces of a large assembly.