Unequal evolutionary conservation of human protein interactions in interologous networks

Abstract

Background: Protein-protein interaction (PPI) networks have been transferred between organisms using interologs, allowing model organisms to supplement the interactomes of higher eukaryotes. However, the conservation of various network components has not been fully explored. Unequal conservation of certain network components may limit the ability to fully expand the target interactomes using interologs.

Results: In this study, we transfer high quality human interactions to lower eukaryotes, and examine the evolutionary conservation of individual network components. When human proteins are mapped to yeast, we find a strong positive correlation (r = 0.50, P = 3.9 x 10(-4)) between evolutionary conservation and the number of interacting proteins, which is also found when mapped to other model organisms. Examining overlapping PPI networks, Gene Ontology (GO) terms, and gene expression data, we are able to demonstrate that protein complexes are conserved preferentially, compared to transient interactions in the network. Despite the preferential conservation of complexes, and the fact that the human interactome comprises an abundance of transient interactions, we demonstrate how transferring human PPIs to yeast augments this well-studied protein interaction network, using the coatomer complex and replisome as examples.

Conclusion: Human proteins, like yeast proteins, show a correlation between the number of interacting partners and evolutionary conservation. The preferential conservation of proteins with higher degree leads to enrichment in protein complexes when interactions are transferred between organisms using interologs.

Figure 1

Properties of PPI networks. (a) Co-expression of yeast 'high confidence' protein interactions (solid lines) and random protein pairs (dotted lines) using two microarray datasets. This network is enriched in stable complexes, represented by a high mean correlation. (b) Co-expression of the yeast 'kinome' [31], which is enriched for transient interactions. This type of interaction shows co-expression that is highly similar to the random distribution (dotted lines). (c) Distribution of clustering coefficients in stable and transient PPI networks. Complexes are represented by a high C_w (blue line), while the sparsely connected transient network is typified by a low C_w (green line). (d) The properties of the human interaction network. The clustering coefficients indicate that this network is more sparsely connected, with few protein complexes. The co-expression profile is only slightly higher than the randomly generated distribution, suggesting the presence of many transient PPIs.

Figure 2

Effect of interolog transfer across evolutionary distance. Interologous protein interactions were predicted from the known human PPI network. (a) The mean C_w for the predicted network in each model organism (mean ± standard deviation), averaged over all nodes with k > 1. P values indicate the significance of the difference from the human interactome. (b) The mean co-localization for each model organism network is shown, normalized against the number of PPIs with localization data for both proteins. (c) The Pearson correlation of genes encoding interacting proteins in each organism (mean ± standard deviation). In all cases, the average correlation is significantly higher than a randomized network (P << 0.001). In each plot, the dotted line indicates the average level for the human network.

Figure 3

Conservation of interacting proteins by degree. (a) Each protein in the yeast interaction network was examined for orthologous proteins in the five higher eukaryotes, and binned according to degree. The proportion of each bin with orthologous proteins is shown. The linear trend shows the strong positive correlation (Spearman's rank r = 0.52, P = 2.8 × 10^-11) between yeast and human proteins. (b) The proteins in the human interactome were compared against all five lower eukaryotes, and binned according to degree. This trendline also shows a strong correlation against yeast (Spearman's rank r = 0.50, P = 3.9 × 10^-4), which is similar for worm and rat, and there is a weak (non-significant) correlation to fly. There was a weak negative correlation in mouse (Spearman's rank r = -0.02); however, the overall conservation was high, likely biasing this measurement.

Figure 4

Yeast interactions transferred from the human interactome. The human interactome was used as a source to predict 750 yeast interactions, 405 of which are novel (red lines), while 345 overlap with previously known yeast PPIs. (a) The replisome, responsible for DNA replication, is enriched by the human interactome. (b) The yeast protein GCS1 is linked to retrograde transport between the Golgi and the endoplasmic reticulum through physical interactions with ERD2, ARF2, and the coatomer complex (COPA, COPB, COPB2, COPG) using human interactions. The node colors indicate the broad functional category of each protein as derived from GO annotations.