Interactome-wide prediction of protein-protein binding sites reveals effects of protein sequence variation in Arabidopsis thaliana

Abstract

The specificity of protein-protein interactions is encoded in those parts of the sequence that compose the binding interface. Therefore, understanding how changes in protein sequence influence interaction specificity, and possibly the phenotype, requires knowing the location of binding sites in those sequences. However, large-scale detection of protein interfaces remains a challenge. Here, we present a sequence- and interactome-based approach to mine interaction motifs from the recently published Arabidopsis thaliana interactome. The resultant proteome-wide predictions are available via www.ab.wur.nl/sliderbio and set the stage for further investigations of protein-protein binding sites. To assess our method, we first show that, by using a priori information calculated from protein sequences, such as evolutionary conservation and residue surface accessibility, we improve the performance of interface prediction compared to using only interactome data. Next, we present evidence for the functional importance of the predicted sites, which are under stronger selective pressure than the rest of protein sequence. We also observe a tendency for compensatory mutations in the binding sites of interacting proteins. Subsequently, we interrogated the interactome data to formulate testable hypotheses for the molecular mechanisms underlying effects of protein sequence mutations. Examples include proteins relevant for various developmental processes. Finally, we observed, by analysing pairs of paralogs, a correlation between functional divergence and sequence divergence in interaction sites. This analysis suggests that large-scale prediction of binding sites can cast light on evolutionary processes that shape protein-protein interaction networks.

Figure 1. SLIDERBio strategy to predict protein-protein binding sites.

(A–B) SLIDERBio follows the assumption that interfaces can be represented by short sequence motifs: (A) Interaction sites (spacefill) are continuous patches of amino acid residues in the 3D structure of a protein, while in a protein sequence (B) the interface is composed of scattered short motifs (regions highlighted in red and green). In (A–B), protein structure and sequence of the Mms2/Ubc13 heterodimer (PDB id 1jat) are used as illustration. (C–D) SLIDERBio predicts interaction sites by finding motif pairs that are overrepresented in pairs of interacting proteins in an interaction network. (C) illustrates a protein-protein interaction network in which the proteins are represented by nodes and the interactions represented by connecting edges; (D) illustrates the protein sequences and their short motifs (regions highlighted in colored bars; same colors represents similar motifs). In this example, the motif pair [grey-orange] is overrepresented compared to the motif pair [red-green]. To calculate the degree of overrepresentation of a motif, the method verifies in how many sequences of interacting proteins a certain motif is found. Originally, SLIDER considered a motif present in a sequence if a perfect match was found between motif sequence and a region in the protein sequence. In contrast, SLIDERBio makes use of a substitution matrix to calculate the similarity between the motif and the sequence. If the degree of similarity between a motif and a sequence is greater than a threshold, SLIDERBio considers that the sequence contains the motif. In addition, SLIDERBio verifies whether the conservation score and the surface accessibility score of the motifs are greater than pre-defined thresholds. These three thresholds are based on the average value per residue over the length of the motif (E).

Figure 2. Overall performance of the SLIDERBio algorithm in different datasets.

(A–C) Coverage of protein-protein interfaces and Accuracy of predicted motifs. Each dot represents the result of SLIDERBio using one of the 180 tested sets of parameters, for (A) human, (B) yeast and (C) Arabidopsis structurally mapped subsets. The grey arrows indicate the dot corresponding to the result of the previous SLIDER algorithm. (D–F), Correlation of the performance for each of the SLIDERBio parameter settings is compared among datasets of different species: (D) human vs. yeast; (E) human vs. Arabidopsis; and (F) yeast vs. Arabidopsis. Pearson Correlation Coefficient (PCC) is indicated.

Figure 3. Overall description of the predicted binding sites in the Arabidopsis interactome.

(A) Network representation of the Arabidopsis interactome and predicted interaction sites. The vertices and edges in black represent, respectively, the 985 proteins and the 1498 interactions to which predicted motifs are mapped. (B) Degree distributions from the complete protein-protein interaction dataset (grey) and from the subset with only proteins and interactions that have a predicted motif (black). A and B suggest that our method is not biased to predict motifs that can be mapped only to proteins with high degree (i.e. number of interactions); moreover, the proteins with predicted motifs are distributed in different positions in the network. (C) Percentage of residues in the interfaces, either in the predicted interfaces or those observed in the structurally mapped dataset. Standard deviation is indicated.

Figure 4. Putative molecular mechanisms underlying effects of amino acid mutagenesis.

A, C and E show the interacting partners of the proteins ZTL, CXIP1 and SHY2, respectively (interactions shown as dashed lines are not covered in the Arabidopsis Interactome data). B, D and F show a schematic representation of the sequences of the three proteins, including predicted binding sites (coloured box, using same colour as the proteins predicted to bind to it), mutagenesis sites (triangles for experimental mutagenesis sites, circles for naturally occurring sequence variants) and their positions, and residue surface accessibility (RSA) and conservation (bar plots) as predicted based on the sequence. A–B, in the protein ZTL, alanine mutagenesis of the residues 200 and 213 independently eliminate the interaction with ASK1; for ZTL, the stretch of residues from 208 to 220 is predicted as interaction site for binding with ASK2 and ASK4. This leads to the hypothesis that mutation on ZTP, specifically on the residue Leu213, would not only disrupt its interaction with ASK1, but also with other SKP1-like proteins, such as ASK2 and ASK4. C–D, In CXIP1, alanine mutagenesis of two highly conserved motifs (residues from 133 to 137; and residues from 97 to 100) leads to loss of ability to activate CAX1. For CXIP1, the stretch of residues from 125 to 136 was predicted as binding site, which overlaps the mutated motif SNWPT. The interaction of CXIP1 and the other interacting partners identified in the Arabidopsis interactome, i.e. AT5G09830, AT3G50780, AT1G70410 and TCP13 (AT3G02150), may also be mediated by the same motif. E–F, in the sequence of SHY2, three motifs were predicted as binding sites. The first (residues from 59 to 69; represented in grey) overlaps the position of two naturally occurring mutations (residues 67 and 69) and is predicted to be responsible for binding of TOPLESS (TPL, AT5G27030). A second motif (residues from 180 to 187; represented in brown) is predicted to be responsible for the interactions of SHY2 with six other IAA proteins. This leads to the hypothesis that two known mutations disrupt the interaction of SHY2 with TPL, but the same mutations do not impede its interaction with other IAA proteins.

Figure 5. Binding sites contain signal about functional divergence.

Distributions of sequence identity values are shown for paralogous pairs classified as having “no” (red), “low” (black) or “high” (blue) functional divergence. The x-axis represents the sequence identity of paralogous pairs. For each paralogous pair, the sequence identity was calculated using either (A) the whole protein sequences, or (B) just the sequence of predicted binding sites. The better separation between the curves for no functional divergence vs. high functional divergence when using predicted interaction sites indicates that these contain signal related to functional divergence.