Conformational Sampling And Structural Analysis For Protein Conformational Transitions

Proteins have the intrinsic ability in undergoing transitions among different confor-mational states, such as open and closed states, which plays important roles in biologicalfunctions, including enzyme catalysis, force generation in motor proteins, allosteric com-munication in proteins. How to understand the molecular origins of these functional proteintransitions is a key factor in regulating biological functions. Although experimental methodscan be used to investigate conformational states and their corresponding distributions duringconformational transitions, they can't provide enough intermediate conformers to explore thedetailed molecular mechanisms underlying these functional transitions. Then computationalapproaches emerge to be an important alternative.Computational study on protein conformational transition is a very broad area. In thisdissertation, our computational study for protein conformational transitions just covers threecoupled aspects: conformational sampling, protein structural similarity and structural statis-tical analysis. The sampling of protein conformations can yield the conformational transitionpathways among functional states, providing the necessary data for investigating molecularmechanisms and biological functions. The protein structural similarity is a kind of reactioncoordinate, which can be used to monitor conformational transition pathways and identifyfunctional substates. This monitoring and identification not only play key role in understand-ing conformational sampling, but also provide important information for structural statisti-cal analysis in next step. The suitable structural analysis methods can be utilized to revealthe molecular mechanisms and biological functions underlying protein conformational tran-sitions, based on the grouped conformers from sampling procedure and protein structuralsimilarity. Based on experimental evidence, we proposed a computational framework to in-vestigate protein conformational transitions, including sampling scheme, structural similaritymetric and structural analysis approaches. In detail, this framework includes the followingthree aspects: 1. In protein conformational sampling scheme, the long timescale overall transition isseparated into several parallel and independent local sub-transitions from substate tosubstate, if there exist these substates which can be identified. Under the assumptionthat there is a hierarchy of space and time in protein conformational transition, wepropose to separate the long timesacle overall transition into several parallel and in-dependent local sub-transitions, and then we use a bunch of parallel short timescalemolecular dynamics (MD) simulations to replace the traditional single long timescaleMD simulation to sample the protein configurational space. This strategy not onlyimproves the efficiency of computational resource usage, but also helps protein fastescape from local energy minimum. In this way, it can speed up the conformationalsampling.2. In reaction coordinate, the normal mode structural similarity based on elastic networkmodel (ENM) is used as a protein structural similarity metric to measure the proteinconformational changes. This protein structural similarity is the connecting link be-tween the proceeding sampling approach and following structural analysis methods,and it is also the novel innovation in this dissertation. For every sampled conformerfrom the parallel MD simulations, we use ENM to calculate its low-frequency normalmodes, and then project them to the corresponding normal modes of reference struc-tures to define protein structural similarity. This protein structural similarity can beused to identify the new substates in simulation trajectories, which can be the seed ofthe next parallel MD simulations. In addition, the identification of sampled proteinconformers provides the basis for structural analysis in next step.3. In structural analysis, statistical methods are utilized to analyze the sampled confor-mations to reveal the molecular mechanisms and biological functions underlying pro-tein conformational transitions. Based on the grouped conformations by normal modestructural similarity, we can do statistics on each groups (or substates), such as corre-lation analysis, principal component analysis, clustering, and network analysis. Thenthe differences and similarities among different substates can be compared to uncoverthe molecular origins of protein conformational transitions. It is worth to mention thatthese structural analysis methods are not proposed by us, we just make use of them inthe right way in our framework base on our sampling scheme.This computational framework were applied to different protein systems. First, we stud-ied the template protein: adenylate kinase (AdK). There are abundant both experimental and computational data in this system, making it a good candidate to validate our computationalframework and show its advantages. As a result, the molecular mechanisms explored by ourcomputational framework are in highly agreement with the biological functions of AdK. Inaddition, our approach can uncover some new insights in the conformational transition ofAdK. Compared to experimentally measured timescale, our approach reduces timescale bytwo orders of magnitude.Second, we used this framework to investigate a hot but unknown topic, the dynamicalmechanism of auto-inhibition of AMP-activated protein kinase (AMPK), a highly conservedenzyme in eukaryotic cells that regulates cellular and whole-body energy homeostasis. Dueto the incomplete structural information and unclear biological mechanisms, it is hard tostudy its dynamics. By using our computational framework, we successfully revealed thedynamical mechanism of AMPK on how protein conformational transition plays role in reg-ulating biological functions. It is worth to mention that our work is the first research work instudying the dynamical mechanism of AMPK.Finally, because of the key role of normal mode structural similarity in our computa-tional framework, we investigated the reason why it works . By using different methods, weshowed that, except the robustness, the low-frequency normal modes are also quite sensitiveto important structural rearrangement during conformational transitions, partially explainingwhy normal mode structural similarity can monitor transition pathways and identify the func-tional substates in our framework. This also provides the insights for further understandingour computational framework.