A central paradigm within structural biology is the concept of domains. In past definitions, domains would be linked to units of compact structure, evolution and folding, and/or function. With the advent of modern bioinformatics, the conservation of domains throughout evolution became instantly recognisable, leading to the established view that domains can be represented by a specific sequence of secondary structure elements that adopt a canonical form when in solution. Much less appreciated but still important is that a significant number of domains have additional elements of structure that lie almost immediately before or after the canonical domain, extending the domain. The presence of these extensions and their impact on folding, structure, dynamics, and function of the domain to which it is attached is of particular significance for the PDZ domain (which was named after the three proteins - PSD95, DLG1, and ZO1 - that led to its discovery).
Figure 1: Canonical and extended PDZ structures. Panel A shows a ribbon representation of the structure of PSD-95 PDZ2. The canonical fold contains six β-sheets and two α-helices. A peptide ligand (orange) is shown bound to the PDZ. Panel B shows a ribbon representation of the structure of PSD-95 PDZ3, which is an example of an extended PDZ domain. Notice the additional helix (pink) at the C-terminal end of the canonical PDZ domain.
We searched for extended PDZ domains using bioinformatics, using the definition that extensions are structured regions outside the canonical PDZ domain boundary. To perform this search, we initially needed sequences and domain boundary information for PDZ-containing proteins. We also needed methods of predicting structure from sequence; we used several programs to predict secondary structure and to predict disorder. All predictions have been collated into this database, which uses PHP to display the results and MySQL to store the data. Figure 2 illustrates the overall procedure used to generate the predictions, highlighting five main steps:
Figure 2: Flowchart of the process used to search for structured PDZ extensions. The general procedure is shown on the left, and a more specific example, which uses only the C-terminal extension region as input for prediction by the program PSIPRED, is shown on the left.
We chose UniProt, a source of curated protein entries, as the source of our protein sequences and domain boundary definitions. We extracted sequences that contained at least one PDZ domain and were complete (i.e. not a fragment) and reviewed because the quality of the annotation was important for this study. From UniProt, we obtained 154 human and 128 mouse sequences that satisfied our search criteria. (We did not look at other species in detail because our search criteria gave us very few results; e.g. 12 from zebrafish, and 13 from fruitfly.)
Each reviewed protein entry in UniProt contains sequence annotation information, such as the start and end position of any constituent domains. We used these annotations to define the location of the PDZ domains (and other domain types) in protein sequences we extracted. The canonical PDZ domain contains six β-sheets and two α-helices; often, domain boundary definitions in other databases, such as SMART, ignores the first β-sheet, causing the predictions for the N-terminal extension region to be biased. We believe that the curated entries of UniProt would provide the most accurate domain boundary definitions compared to other available databases.
All prediction programs require an input sequence. The results from these prediction programs can be sensitive to the composition of the input sequence. Therefore, we decided to use different input sequences to give us an idea of how the predictions varied. Figure 3, panel B in particular, shows the different input sequences that were used for the predictions. Each input sequence contained at least a 50 amino acid long extension sequence (extending either the N- or C-terminus of the PDZ domain), but differed by their start or end position.
Figure 3: Input Sequences for Predictions. Panel A shows the overall region that was analysed. We focused on sequence regions that are 50 amino acid residues in length on either side of the PDZ domain. The canonical secondary structure content of a PDZ domain is also illustrated. Panel B shows the different input boundaries used for predictions. The dotted box highlights the regions used for analysis of the prediction results.
We were interested in looking for structured regions in the input sequences. To do this, we used the programs PSIPRED, PROFPHD, and PREDATOR to predict secondary structure and DISEMBL and DISOPRED to predict disorder. We used the default parameters for each program.
After the predictions were made, we applied several metrics to analyse the 50 amino acid extension region on both ends of the PDZ domain. Details of each prediction can be viewed by clicking on the appropriate entry. We summarized the predictions in several plots and they can be viewed in more detail by clicking the statistics links on the left side menu.