Comparative evolutionary analysis of protein complexes in E. coli and yeast

Abstract

Background: Proteins do not act in isolation; they frequently act together in protein complexes to carry out concerted cellular functions. The evolution of complexes is poorly understood, especially in organisms other than yeast, where little experimental data has been available.

Results: We generated accurate, high coverage datasets of protein complexes for E. coli and yeast in order to study differences in the evolution of complexes between these two species. We show that substantial differences exist in how complexes have evolved between these organisms. A previously proposed model of complex evolution identified complexes with cores of interacting homologues. We support findings of the relative importance of this mode of evolution in yeast, but find that it is much less common in E. coli. Additionally it is shown that those homologues which do cluster in complexes are involved in eukaryote-specific functions. Furthermore we identify correlated pairs of non-homologous domains which occur in multiple protein complexes. These were identified in both yeast and E. coli and we present evidence that these too may represent complex cores in yeast but not those of E. coli.

Conclusions: Our results suggest that there are differences in the way protein complexes have evolved in E. coli and yeast. Whereas some yeast complexes have evolved by recruiting paralogues, this is not apparent in E. coli. Furthermore, such complexes are involved in eukaryotic-specific functions. This implies that the increase in gene family sizes seen in eukaryotes in part reflects multiple family members being used within complexes. However, in general, in both E. coli and yeast, homologous domains are used in different complexes.

Figure 1

Generation of MCL-GO complex datasets. (a) For IntAct data, rendering TAP-MS data using the spoke model rather than the matrix model gave improved performance. (b) Combining IntAct and MINT datasets and weighting interactions with GOSS scores gives greater accuracy over either resource alone and without weighting. (c) Accuracy of MCL-GO complexes (using MINT+IntAct and edge weighting) in capturing MIPS yeast complexes and EcoCyc E. coli complexes. 'Random' lines show mean accuracy achieved over 10000 sets of randomised clusters. Error bars show one standard deviation either side of the mean. (d) Size distribution of E. coli and yeast MCL-GO complexes.

Figure 2

Coverage of complexes by single functional terms. This figure shows the percentage of proteins in yeast MCL-GO complexes which could be annotated with the most common term in each complex. Complexes were classified using FunCat terms. Complexes with <2 annotated proteins were excluded.

Figure 3

Principal function of MCL-GO complexes in each species. Complexes were classified using FunCat terms. Complexes with <2 annotated proteins were excluded.

Figure 4

Distribution of CATH superfamilies in MCL-GO complexes. This figure shows the number of CATH superfamily members versus number of complexes containing members of that superfamily for E. coli and yeast MCL-GO complexes.

Figure 5

Percentage of MCL-GO complexes containing homologous pairs. Homologous pairs are defined here as either homologous domains shared between two proteins or proteins sharing a common domain architecture. All observed values were significantly larger than expected (p < 0.01).

Figure 6

Percentage of complexes containing homologous pairs for alternative datasets. Asterisks show observed values which were significantly above random.