Widespread evolutionary crosstalk among protein domains in the context of multi-domain proteins

Abstract

Domains are distinct units within proteins that typically can fold independently into recognizable three-dimensional structures to facilitate their functions. The structural and functional independence of protein domains is reflected by their apparent modularity in the context of multi-domain proteins. In this work, we examined the coupling of evolution of domain sequences co-occurring within multi-domain proteins to see if it proceeds independently, or in a coordinated manner. We used continuous information theory measures to assess the extent of correlated mutations among domains in multi-domain proteins from organisms across the tree of life. In all multi-domain architectures we examined, domains co-occurring within protein sequences had to some degree undergone concerted evolution. This finding challenges the notion of complete modularity and independence of protein domains, providing new perspective on the evolution of protein sequence and function.

Fig 1. Representations of a two-domain architecture.

A: Example structure of a two-domain protein. Chain A in Protein Data Bank [18] entry 1WAK contains two PF00069 (protein kinase) domains, shown here in blue and red. B: Correlated mutations in multi-domain proteins. Blue and red areas schematically depict individual domains involved in an interaction. Dots connected by solid lines represent pairs of coevolving positions. Coevolving positions within individual domains are shown in yellow. Coevolving positions localized in different domains are shown in black and brown. These can involve both positions forming physical contacts at the inter-domain interface (black) as well as positions separated by large distances (brown). C: General structure of a mutual information matrix for a two-domain architecture AB. White A and B labels denote individual domains. Diagonal dark blue and dark red elements describe the entropy of individual positions within domain A or B, respectively. Off-diagonal light blue and light red elements describe the mutual information between pairs of positions within domain A or B, respectively. Off-diagonal pink elements describe the mutual information between pairs of positions belonging to different domains.

Fig 2. Mutual information matrix for the native PF00069–PF00069 (protein kinase–protein kinase) two-domain architecture.

Each protein kinase domain contains 264 residues. A total of 8,753 URPs sequences have this architecture. The respective MSA consists of sequences of the two domains found within each of these URPs sequences. Note the non-linear color scale and the positive values of mutual information between positions corresponding to different domains (highlighted with the blue square). Values of entropy and mutual information are in bits.

Fig 3. Mutual information matrix for the perturbed PF00069–PF00069 (protein kinase–protein kinase) domain architecture.

Sequences of individual domains were randomly shuffled and rejoined. Note how the mutual information between positions corresponding to different domains (blue square) has vanished. Values of entropy and mutual information are in bits. The color scale is the same as in Fig 2.

Fig 4. Distributions of the values of nMI¯ for various domain pairings.

A: Distribution of the values (N = 5, 205) of $\mathit{n}\overline{\text{MI}}$ for domain pairs in native (non-disrupted) multi-domain protein sequences. B: Distribution of the values (N = 5, 205) of $\mathit{n}\overline{\text{MI}}$ for domain pairs after intra-architectural domain sequence shuffling. C: Distribution of differences (N = 5, 205) between the values of $\mathit{n}\overline{\text{MI}}$ for individual domain pairs before and after intra-architectural domain sequence shuffling. D: Distribution of the values (N = 2, 599) of $\mathit{n}\overline{\text{MI}}$ for domain–random sequence pairs.

Fig 5. Correlation between the values of nMI¯ before and after intra-architectural domain sequence shuffling.

Number of domain pairs (data points) N = 5, 205. The value of the Pearson correlation coefficient r ≈ 0.345; the value of the Spearman’s rank correlation coefficient ρ ≈ 0.432.

Fig 6. Correlation between the non-normalized values of MI¯ and average domain pair entropies.

Number of domain pairs (data points) N = 5, 205. The value of the Pearson correlation coefficient r ≈ 0.464; the value of the Spearman’s rank correlation coefficient ρ ≈ 0.460.