IntroductionRelated WorkModeling Molecular Motions with PRMsRigidity-Based SamplingAutomatic Roadmap ConstructionMapping Specified TransitionsResults and DiscussionConclusionSimulating Protein Motionswith Rigidity AnalysisShawna Thomas, Xinyu Tang, Lydia Tapia, and Nancy M. AmatoParasol Lab, Dept. of Comp. Sci., Texas A&M University,College Station, TX 77843Abstract. Protein motions, ranging from molecular flexibility to large-scale conformational change, play an essential role in many biochemi-cal processes. Despite the explosion in our knowledge of structural andfunctional data, our understanding of protein movement is still very lim-ited. In previous work, we developed and validated a motion planningbased method for mapping protein folding pathways from unstructuredconformations to the native state. In this paper, we propose a novelmethod based on rigidity theory to sample conformation space moreeffectively, and we describe extensions of our framework to automatethe process and to map transitions between specified conformations.Our results show that these additions both improve the accuracy ofour maps and enable us to study a broader range of motions for largerproteins. For example, we show that rigidity-based sampling results inmaps that capture subtle folding differences between protein G and itsmutations, NuG1 and NuG2, and we illustrate how our technique canbe used to study large-scale conformational changes in calmodulin, a 148residue signaling protein known to undergo conformational changes whenbinding to Ca2+. Finally, we announce our web-based protein foldingserver which includes a publically available archive of protein motions:http://parasol.tamu.edu/foldingserver/1 IntroductionProtein motions, ranging from molecular flexibility to large-scale conformationalchange, play an essential role in many biochemical processes. For example, con-formational change often occurs in binding. While no consensus has been reachedregarding models for protein binding, the importance of protein flexibility in theprocess is well established by the ample evidence that the same protein can existin multiple conformations and can bind to structurally different molecules.Our understanding of molecular movement is still very limited and has notkept pace with the explosion of knowledge regarding protein structure and func-tion. There are several reasons for this. First, the structural data in repositoriesSupported in part by NSF Grants EIA-0103742, ACR-0081510, ACR-0113971, CCR-0113974, ACI-0326350, and by the DOE. Thomas supported in part by an NSFGraduate Research Fellowship.A. Apostolico et al. (Eds.): RECOMB 2006, LNBI 3909, pp. 394–409, 2006.c Springer-Verlag Berlin Heidelberg 2006Simulating Protein Motions with Rigidity Analysis 395like the Protein Data Bank (PDB) [8] consists of the spatial coordinates of eachatom. Unfortunately, the experimental methods used to collect this data can-not operate at the time scales necessary to record detailed large-scale proteinmotions. Second, traditional simulation methods such as molecular dynamicsand Monte Carlo methods are computationally too expensive to simulate longenough time periods for anything other than small peptide fragments.There has been some attention focused on methods for modeling proteinflexibility and motion. One notable effort is the Database of MacromolecularMovements [15, 14]. They generate and archive protein ‘morphs’ that interpo-late between two different protein conformations. While the method used is morechemically realistic than straight-line interpolation (as described in Section 2),it was selected over other more accurate methods for computational efficiencyand is known to have problems for some kinds of large deformations.In previous work [3, 2, 42, 41], we developed a new computational technique forstudying protein folding that builds an approximate map of a protein’s potentialenergy landscape. This map contains thousands of feasible folding pathways tothe known native state enabling the study of global landscape properties. Weobtained promising results for several small proteins (60–100 amino acids) andvalidated our pathways by comparing secondary structure formation order withknown experimental results [3].Our Contribution. We augment our framework with three powerful new con-cepts that enable us to study a broader range of motions for larger proteins:– We propose a new method based on rigidity theory to sample conformations.– We generalize ourPRM framework to map specified transitions.– We present a new framework to automate the map building process.Our new rigidity-based sampling allows us to study larger proteins by moreefficiently characterizing the protein’s energy landscape with fewer, more re-alistic conformations. We exploit rigidity information by focusing sampling on(currently) flexible regions. This results in smaller, better maps. In one dramaticcase study, we show that rigidity-based sampling and analysis reveals the foldingdifferences between protein G and its mutants, NuG1 and NuG2, which is animportant ‘benchmark’ set that has been developed by the Baker Lab [36].Extending our framework to focus on particular conformations enables us toinvestigate questions related to the transition between particular conformations,e.g., when studying folding intermediates, allostery, or misfolding. We provideevidence that the transitions mapped by our approach are more realistic thanthose provided by the computationally less expensive Morph Server [14], espe-cially for transitions requiring large conformational changes.The accuracy of our approach heavily depends on how densely we sample theconformation space. Previously, this was user specified and fixed. Here, we use anextension of our basic technique which incrementally samples the conformationspace at increasingly denser resolution until our map of the landscape stabilizes.Finally, we announce our protein folding server which uses our techniqueto generate protein transitions to the native state or between selected396 S. Thomas et al.Table 1. Comparison of protein motion modelsApproach Landscape #PathsPath Quality Computation Native RequiredMolecular Dynamics No 1 Good Long NoMonte Carlo No 1 Good Long NoStatistical Model Yes 0 N/A Fast YesPRM-Based (Our Approach) Yes Many Approx Fast YesLattice Model Notusedonrealproteinsconformations. We invite the community to help enrich our publicly availabledatabase by
View Full Document
Unlocking...