The Computational Core research plan has been revised to address specific questions and concerns raised by the reviewers, and to emphasize the principle focus of this core. The primary goals of the Computational Core are 1) to develop a set of computational tools and protocols to facilitate the analysis and interpretation of EPR spectral data, including distance measurements obtained from DEER experiments for doubly spin-labeled proteins, and 2) provide basic computational support for the individual research projects. For Project 1, basic computational support entails a series of equilibrium MD simulations to support EPR spectral calculations. In Project 2, this computational support includes detailed equilibrium MD simulations for CDB3 to explore possible conformational changes triggered by the P327R point mutant, and preliminary results are described above in the Project 2 Research Plan. Basic computational support for Project 3 includes routine structure refinement calculations for conventional 2D-NMR experiments and paramagnetic resonance enhancement NMR experiments, as well as MD simulations to explore conformational trends for spin labels introduced in the amyloid-beta peptides. This conformational analysis will be important to address distance dependencies on spin label side chain conformational behavior in both EPR experiments and paramagnetic resonance enhancement NMR studies. The development of practical computational tools and protocols to facilitate EPR data analysis depends crucially on data obtained in Project 1, and requires several discreet steps. First, it is important to establish that we can use conventional equilibrium MD simulations that describe spin label side chain dynamics and protein backbone dynamics, coupled with Brownian dynamics calculations that model global protein tumbling, to compute EPR spectra directly for singly labeled proteins. As the reviewers noted, previous published attempts to exploit this type of strategy have not been completely satisfactory or convincing. However, these previous studies were based on rather limited MD simulations, and possibly suffered from some other issues that we address in more detail in the Research Plan below. It is essential to establish that a simulation strategy can be used to compute EPR spectra, in order to establish that we can capture the important features and behavior of spin-labeled proteins that give rise to unique EPR spectra for different samples (e.g., the sharp, distinct spectral signal typical of a completely mobile spin label versus the broader, more complex signals representative of partially immobilized spin labels). As discussed in the Project 1 Research Plan, we now have preliminary results that indicate we can compute EPR spectra more accurately and reliably than has been reported previously. There is still need for improvement, and we present detailed analysis of current MD-based EPR spectral simulations below that highlight possible inadequacies in the current methodology, and discuss specific strategies and tests to address these problems. Only after we have established convincingly that we can calculate EPR spectra directly with the combined MD/Brownian dynamics simulation protocol can we address seriously the calculation of spin label pair distances obtained in EPR DEER experiments, or pursue development of simpler computational strategies that do not require multiple, lengthy MD simulations with explicit solvent to estimate these distances. A number of issues impact the reliable MD simulation of spin label pair distances, including several raised by the reviewers for Project 1 (E.g., potential function parameters, electrostatics treatment, periodic boundary effects, etc.) We present preliminary data in the revised Research Plan below that addresses these issues and other important factors, and the strategies to achieve improved EPR spectral calculations and DEER distance estimates are presented in the context of a new Specific Aim 1.
Aim 1 in the original proposal (now renumbered Specific Aim 2) contained a detailed discussion of previous studies designed to explore the impact of (limited) long-range distance constraints on 3D structural model generation. Reviewer #1 noted that the general strategy outlined in this Aim was reasonable, but rather timeconsuming. We note below some specific efficiency improvements for certain steps that reduce the overall computational expense for this protocol (although this is still a non-trivial computational task). Reviewer #1 also noted several specific concerns or suggestions related to this aim. Alternate metrics, such as backbone torsion angles rather than protein backbone RMSD values, were suggested for structural comparisons and clustering. This is certainly a reasonable recommendation, and we have explored some simple alternative comparison metrics. Backbone torsion angle comparisons, or other simple quantitative assessments such as volumetric or shape descriptors are intrinsically appealing, although those metrics are somewhat less """"""""intuitive"""""""" for structural comparison (at least for us at this stage). We discuss below the use of backbone torsion angles as a potentially quite useful and efficient comparison metric in new work proposed. We have also discussed this issue with several colleagues who focus on protein structure prediction and thus perform these types of calculations routinely. Interestingly, we were referred back to the SUPPOSE algorithm for backbone RMSD comparisons by these groups (this program has clearly become more popular than we realized). Reviewer #1 also recommended that we consider alternate programs for the actual clustering process, and this is most reasonable. Nothing in our protocol commits us to use Jeff Barton's """"""""OC"""""""" program, and it is straightforward to integrate alternate clustering algorithms in our job control scripts, so we will explore other algorithms after we have established the applicability and scope of our protocol. Reviewer #1 also suggested that we consider strategies to enhance the structural """"""""diversity"""""""" in our relatively small 3D model datasets;this suggestion is closely coupled to the concern raised by reviewer #2 that 10,000-20,000 trial structures per run will be inadequate to sample 3D structural space adequately. It is our belief that an appropriate set of long-range distance constraints will limit the feasible 3D structural solution space sufficiently to reduce the severity of this problem. Our previous results, as well as those of several other research groups, have shown clearly that a small number of long-range distance constraints can dramatically reduce the 3D conformational search space for protein model construction, although there is no guarantee that any arbitrary set of long-range distance constraints will achieve this goal, and we must perform additional tests outlined in Specific Aim 2 to better understand how effective a relatively small collection of long-range distance constraints might be in reducing the search space. We also describe a new strategy to improve the structural """"""""diversity"""""""" of the trial structures, which utilizes 3D model generation techniques incorporated in Rosetta (Wollacott, et al., 2007;Rohl, et al., 2004). Both reviewers expressed concerns regarding the scoring functions used to """"""""rank"""""""" structural solutions. There is no easy or obvious answer here, and we can only pursue the strategies outlined in the Research Plan below. Our real solution to this problem is to use an iterative process of model generation and additional DEER distance measurements to systematically reduce the number of acceptable structural models. We now provide a more detailed discussion of the strategy we use for selection of additional labeling sites to illustrate more clearly how we expect this process will work, as requested by Reviewer #1. We also provide a more detailed explanation for how we have coupled the 3D model generation protocol with motif identification and homology modeling techniques for the test systems we have studied to date. Finally, we discuss in the revised Aims 2 and 3 ways to include additional EPR experimental data beyond inter-residue distances in the model generation and refinement procedures. We should reemphasize that the goal for calculations outlined in Specific Aim 2 is generation of low- to intermediate-resolution 3D models. It is inappropriate at this stage to talk about true 3D structure refinement from EPR DEER distance measurements in the same context as, for example, conventional NMR or x-ray structure refinement procedures. A more realistic goal at this point is structural motif identification for previously uncharacterized proteins, and our previous studies for test systems presented below suggest that this is feasible.
Aim 2 in the original proposal (now Specific Aim 3) focused primarily on development of tools for analysis of inter-residue distances obtained from DEER measurements. This section has been modified significantly to better describe the tight integration of this work with Project 1, as well as to address various concerns raised by the reviewers. More methodological detail is provided for various strategies, and the planned implementation of coarse-grained models is discussed in greater detail. We discuss issues related to the adequacy of conformational sampling in depth, and criteria for """"""""validation"""""""" of computed results such as distance distributions. Reviewer #1 also raised a question regarding constraint quality, and this is a rather tricky issue with EPR distance measurements. In some contexts, a measured distance that exhibits a large distance distribution might be classified as a lesser-quality data point (at least in the context of 3D model construction or refinement). However, many in the EPR field would take exception to such a characterization, arguing correctly that a large distance distribution is itself an important and informative piece of data. We discuss this issue in more detail in the new Specific Aim 3. Original Aim 3 (now Aim 4) entails primarily """"""""toolkit"""""""" design and application to specific tasks in Projects 1-3, followed by packaging for wider dissemination to the general user community. These goals are unmodified from the original proposal. Major revisions in the Research Plan are demarcated by bold square brackets [] around the relevant text.
Showing the most recent 10 out of 22 publications
