Inter-residue, inter-protein and inter-family coevolution: bridging the scales

Abstract

Interacting proteins coevolve at multiple but interconnected scales, from the residue-residue over the protein-protein up to the family-family level. The recent accumulation of enormous amounts of sequence data allows for the development of novel, data-driven computational approaches. Notably, these approaches can bridge scales within a single statistical framework. Although being currently applied mostly to isolated problems on single scales, their immense potential for an evolutionary informed, structural systems biology is steadily emerging.

Figure 1. Schematic overview over the multiple scales of PPI, which can be addressed by coevolutionary analysis

Starting from two MSA of homologous protein families (upper left: paralogs inside the same species are symbolized by identical colors, different species by distinct colors), coevolutionary analysis addresses the following questions: (i) Do the two families interact (upper right)? (ii) If yes, which specific protein pairs do interact (lower right)? (iii) How do these proteins interact; which residues are in contact across the PPI interface (lower left)? Even if each question requires the previous one to be answered first, recent coevolutionary studies have approached them mostly in the opposite order, by using input derived from other approaches and not based on inter-protein coevolution.

Figure 2. Coevolutionary prediction of inter-protein residue-residue contacts and specific protein-protein interaction partners

The left panels show the first 15 inter-protein residue-residue correlations (A) and direct couplings (B) for the HK/RR interaction (green – true positive, i.e. in contact across the interface, red – false positive, i.e. distant across the interface). While it is impossible to bring all predicted pairs in Panel A into simultaneous contact, Panel B suggests a quaternary structure bringing also the active sites (yellow) into spatial vicinity, as needed for phosphotransfer. Panel C (from [26]) shows the result of the progressive matching procedure between paralogs applied to 8,998 TCS from 712 species: the red line follows, from the lower left to the upper right corner, the progressive inclusion of more and more species into the paralog matching. It shows the fraction of correctly matched pairs (the so-called cognates) as a function of the currently matched sequences. While a perfect algorithm would follow the dashed diagonal (all included sequences are correctly matched), the algorithm of [26] finally matches 86% of all 8,998 pairs correctly. A random algorithm would correctly match, on average, one protein pair per species, corresponding to only about 8% of all pairs. Contact predictions at different stages of the matching procedure are shown (green – true positive, red – false positive), for 59 sequences (seed MSA), for about 1000 sequences, for the fully matched MSA, and for the correct MSA of cognate pairs (cf. grey arrows). The quality of the contact prediction for the full matching is very close to the one used the correct cognate pairing.

Figure 3. Coevolutionary prediction of PPI networks for the small (A) and large (B) ribosomal subunit in the bacteria

Each panel shows the 10 pairs of highest inter-protein coevolutionary score. True positive predictions are drawn in green, false positives in red. Grey lines stand for existing but not predicted interactions. The width of the grey and green lines is proportional to the interface size. We observe that most large interfaces are found, but small ones are typically missed. Figures from [40].