In this section, we illustrate how POLYVIEW-3D server can be used for protein structure analysis in the context of protein-protein interactions. In particular, we show several specific examples of structure rendering and annotation for a homodimeric complex of regulatory units of the transcriptional antiterminator protein LicT, which regulates the expression of Bacillus subtilis
operons involved in beta-glucoside metabolism [35
]. All options discussed here can be specified using the POLYVIEW web-interface. We would also like to point out that animated versions of the images shown here can be easily generated using the server.
The regulatory units of LicT consist of two five helical bundle domains called PRD1 and PRD2, which adopt dramatically different relative orientations in inactive and active forms of the protein (PDB: 1TLV, 1H99, respectively). In the activated state, each PRD forms a dimeric unit with its counterpart in the other chain, burying, at the dimer interface, phosphorylation sites that are critical for regulation (conserved histidine residues). In the inactive state, a wide swing movement of PRD2 results in partial opening of the dimer, making the phosphorylation sites accessible on the protein surface [35
]. This inactive form of the dimer, with essentially only PRD1 and PRD1' domains involved in the formation of the interaction interface [35
], was used in the CAPRI assessment as Target09 [36
], and is shown here in Figure .
As can be further seen from Figures , and , protein chains can be rendered using different models, including available surface representations with different coloring schemes. In particular, we illustrate various structural annotations that can be automatically generated and displayed, including the identification of interacting residues within the complex (Figures and ), visualization of (two) large pockets partially overlapping with the observed interface (shown in blue and cyan in Figure ) that were identified using CASTp, coupled with the analysis of evolutionary conservation and putative functional hotspots carried out using ConSurf (Figure ), as well as estimated flexibility due to thermal motions, as encoded by temperature factors (Figure ).
Figure 2 Chain A of the regulatory domain of LicT protein shown alone, with two largest pockets on the surface that partially overlap with the interaction interface (shown in magenta), as identified using CASTp, highlighted in blue and cyan (the latter for residues (more ...)
Figure 3 Chain A of the regulatory domain LicT protein shown alone, with surface exposed residues colored according to their evolutionary conservation, as assessed by the ConSurf server (residues that are highly conserved are shown in deep purple, whereas highly (more ...)
Figure 4 Same as Figures 2 and 3, with colors representing B-factors this time (red corresponding to highly flexible, and white to relatively rigid parts of the structure, respectively), and with the semitransparent rendering of the surface (residues forming pockets (more ...)
We would like to comment that a relatively large, rather hydrophobic with some hydrophilic hot spots (data not shown) and relatively rigid (in this bound structure) pocket that involves some conserved residues, is identified in the central groove of the structure, essentially in between two distinct patches forming the interface. Given the importance of the dimerization state and interface formation for the function of the protein, this pocket may represent a valuable target for ligand design. We would also like to point out that POLYVIEW-3D can be used for additional analysis of conformational changes between inactive and active forms by generating animated movies and individual snapshots of putative trajectories representing transition between the two structures, as generated, e.g., by the AD-ENM server [21
], or obtained from the Database of Macromolecular Movements [22
In Figures and , we further illustrate the analysis and scoring of protein docking models, using an approach that was described in the previous section. Putative dimer models were generated for this system by the ClusPro server, and were submitted to POLYVIEW-3D. Two different models (ranked by ClusPro as number one and nine, respectively) are shown for comparison. The first model (Figure ) is qualitatively consistent with the inactive form of LicT regulatory module, with PRD1 domains forming most of the interface, and PRD2 domains in an open orientation (although somewhat different than observed experimentally – see Figure ). The other model chosen here for illustration purposes, and shown in Figure , is characterized by a very different (and incorrect) orientation of the monomers, and the resulting interaction interface.
Figure 5 Application of POLYVIEW-3D to the analysis and further assessment (in terms of the overlap between predicted and observed interaction interfaces) of protein docking models, generated for LicT using the ClusPro server (see text for details). An overall (more ...)
Visualization of an alternative (and qualitatively incorrect) ClusPro model for the LicT regulatory domain complex, using the same color scheme as in Figure 5. Note the lack of overlap with SPPIDER predictions in this case.
The correct ranking of these two models by ClusPro is certainly very encouraging. However, in general, multiple and often vastly different models from protein docking simulations are difficult to assess. Therefore, contrasting the results of protein docking approaches with predicted functional hot spots and interaction interfaces, provides a complementary approach to further improve model ranking and confidence in the models selected as top candidates. Using POLYVIEW-3D greatly facilitates such comparative analysis. In this particular case, as can be seen from the figures, only the first model shows significant overlap with predicted (from unbound structures) interaction sites, which are highlighted in red (for residues observed within the interface in a given model, and predicted as interaction sites) and yellow (for the remaining predicted interaction sites).
We would like to comment that SPPIDER predicts in this case two distinct interaction interfaces in the N- and C-terminal regions of the regulatory domain, coinciding with PRD1 and PRD2 subdomains. While only one of these interfaces is present in the inactive form analyzed here in detail, due to the rearrangement of the structure in terms of the relative orientation of PRDs, the other predicted patch overlaps, in fact, with the alternative interface observed in the active form of LicT. The latter can readily be verified using POLYVIEW-3D and the mapping of interaction sites from multiple complexes involving close homologs of the chain of interest, as described in the previous section.
While this is just an example of an application of POLYVIEW-3D to simplify the analysis and visualization of protein docking models, and caution should be exerted to avoid over-interpretation of such results in any particular case, on average, similar level of accuracy was observed for other CAPRI targets [24
]. In the future, we are planning to expand this option to include other types of potentially useful scoring functions, including user provided contact potentials and other measures for model assessment (see, e.g., [37