Protein-protein Interaction Networks of E. coli and S. cerevisiae are similar

Abstract

Only recently novel high-throughput binary interaction data in E. coli became available that allowed us to compare experimentally obtained protein-protein interaction networks of prokaryotes and eukaryotes (i.e. E. coli and S. cerevisiae). Utilizing binary-Y2H, co-complex and binary literature curated interaction sets in both organisms we found that characteristics of interaction sets that were determined with the same experimental methods were strikingly similar. While essentiality is frequently considered a question of a protein's increasing number of interactions, we found that binary-Y2H interactions failed to show such a trend in both organisms. Furthermore, essential genes are enriched in protein complexes in both organisms. In turn, binary-Y2H interactions hold more bottleneck interactions than co-complex interactions while both binary-Y2H and co-complex interactions are strongly enriched among co-regulated proteins and transcription factors. We discuss if such similarities are a consequence of the underlying methodology or rather reflect truly different biological patterns.

Figure 1. Enrichment of essential genes in different protein-protein interaction datasets of E. coli.

(A) Overlaps between sets of proteins that are involved in binary-Y2H, co-complex and literature curated binary protein-protein interaction data sets in E. coli, including the number of essential proteins that are involved in interactions (brackets). (B) Enrichment of essential bacterial proteins as a function of their number of interactions in E. coli and S. cerevisiae. Notably, we observed that binary-Y2H interactions failed to show an ascending trend compared to literature curated binary and co-complex interactions in both organisms.

Figure 2. Essential genes in the protein-protein interaction networks of E. coli.

(A) The Venn diagram shows overlaps between the binary-Y2H, co-complex and literature curated binary interaction networks in E. coli, as well as interactions between essential genes in these sets (brackets). (B) We grouped E. coli proteins that were placed a given distance away from essential proteins in the underlying interaction networks. The fraction of essential proteins is largest in the immediate vicinity of other essential proteins. Error bars indicate 95% confidence interval. (C) Observed and expected sizes of the largest connected component between essential genes in E. coli. As a null model we randomly sampled essential genes 10,000 times, indicating that the size of the largest component in co-complex and literature curated binary interaction sets was significantly larger than randomly expected (P < 10⁻⁴).

Figure 3. Interactions between functional classes.

Significant connections between functional classes are mediated by protein-protein interactions. For each dataset and each class combination a P-value was calculated, reflecting the significance of the interaction density between classes in an interaction dataset of certain size and class coverage. Functional groups that exhibit cross-talk are highlighted (dotted lines are a guide to the eye).

Figure 4. Protein complexes and protein-protein interactions.

(A) Schematic illustration of interactions that appear between and within complexes. In (B) E. coli and (C) S. cerevisiae we determined the number of binary-Y2H and co-complex interactions between proteins in the same complex as well as within complexes. As a random null model we resampled proteins in complexes 10,000 times. Our results in the upper panels indicate that interactions between complexes appear diluted in all interaction sets (P < 10⁻⁴) while interactions in the same complexes seem to be enriched (P < 10⁻⁴). Analogously, we investigated interactions between essential proteins in both organisms (lower panels), confirming an enrichment of PPIs within complexes compared to between-complexes in both organisms (P < 10⁻⁴). In (D) we calculated the fraction of essential genes in each complex. As a null-model, we randomly sampled essential genes 10,000 times, indicating that complexes generally do not randomly contain essential proteins. Notably, complexes with a very low/high fraction appear to show a significant enrichment of essential genes.

Figure 5. Topological features of protein-protein interaction sets in E. coli and S. cerevisiae.

In (A) we schematically illustrate interactions between co-regulated and co-regulating genes. In particular, co-regulated genes are considered a set of target genes that are controlled by a transcription factor. In turn, co-regulating genes refer to transcription factors that control the expression of the same genes. In (B) we calculated the number of interactions between co-regulated genes. In turn, we also calculated the enrichment of interactions between co-regulating genes in both E. coli and S. cerevisiae. As for their expected levels, we randomized the set of target genes of each transcription factor 10,000 times. While enrichments were significant in all datasets (P < 10⁻⁴) transcription factors tend to interact more frequently with each other than their target genes.

Figure 6. Bottleneck interaction sets in E. coli and S. cerevisiae.

(A) To illustrate the concept of bottleneck edges, we considered a toy network of 138 interactions. After calculating their betweeness centrality we defined the top 10% of edges with highest betweeness as bottleneck set. In (B) we calculated the edge betweeness of proteins in combined networks of binary-Y2H, co-complex and literature curated binary interactions in both E. coli and S. cerevisiae. Randomizing sets of such bottleneck edges 10,000 times, we calculated the ratio of observed and expected number of bottleneck edges that appeared in the binary-Y2H, co-complex and literature curated binary sets. While enrichments were significant in all cases (P < 10⁻⁴) we generally observed an enrichment of binary-Y2H interactions, while co-complex interactions appeared to be significantly diluted. However, E. coli and yeast differ in literature curated binary PPI data.