Ligand Binding Site Structure Influences the Evolution of Protein Complex Function and Topology

Abstract

It has been suggested that the evolution of protein complexes is significantly influenced by stochastic, non-adaptive processes. Using ligand binding as a proxy of function, we show that the structure of ligand-binding sites significantly influences the evolution of protein complexes. We show that homomers with multi-chain binding sites (MBSs) evolve new functions slower than monomers or other homomers, and those binding cofactors and metals have more conserved quaternary structure than other homomers. Moreover, the ligands and ligand-binding pockets of homologous MBS homomers are more similar than monomers and other homomers. Our results suggest strong evolutionary selection for quaternary structure in cofactor-binding MBS homomers, whereas neutral processes are more important in complexes with single-chain binding sites. They also have pharmacological implications, suggesting that complexes with single-chain binding sites are better targets for selective drugs, whereas MBS homomers are good candidates for broad-spectrum antibiotic and multitarget drug design.

Keywords: drug design; heteromers; homomers; ligand binding; neutral evolution; polypharmacology; protein complex evolution.

Graphical abstract

Figure 1

Examples of SBSs and MBSs in Homomers and Ligand Distributions (A) Human ketohexokinase complexed with pyrimidopyrimidine (PDB ID: 3Q92). The protein forms a dimer, and each subunit has an independent binding pocket, containing a ligand (red). Residues binding the ligand are highlighted in yellow. (B) Anemone STING (stimulator of interferon [IFN] genes) protein, complexed with cyclic diguanosine monophosphate (PDB ID: 5CFL). The homodimer has only one binding site, which is formed by both protein chains. Residues binding the ligand are highlighted in yellow. (C) The composition of biologically relevant ligands in monomers, SBS homomers, MBS homomers, SBS heteromers, and MBS heteromers. In all cases, the majority of ligands are small organic molecules, but MBS complexes are characterized with higher fraction of cofactors and fewer metal ions. Nucleic acids and peptides were not used in the analyses. (D) Boxplots of the number of ligands in the five protein categories, excluding outlayers. The number of different ligands per protein follows an exponential-like distribution, with the majority of proteins having one or two different ligands. (E) Outline of the ligand binding pocket searches: for each pair of homologous proteins, we performed an exhaustive search, i.e., we searched all structures of the target protein with all ligand binding pockets of all structures of the query protein, using both proteins as target and query.

Figure 2

Evolution of Ligand Binding and QS We use the ability to bind ligands of homologous proteins (scaled with sequence similarity) as the measure of their functional similarity. (1) General relationships between ligand binding and QS in homomers, heteromers, and monomers. (A) Homomers with MBSs show a much slower functional divergence than monomers. This pattern is caused by two separate processes: (1) the binding pockets of small-molecule ligands diverge significantly less rapidly than the pockets of monomers and (2) MBS homomers have a higher fraction of cofactors, which are frequently identical and have highly conserved binding pockets plus fewer metal ions. (B) In homomers with binding sites restricted to a single chain, the ability to bind the ligands of homologs changes qualitatively similarly to monomers. (C) MBS heteromers do not show the same pattern as MBS homomers. (D) In SBS heteromers, similarly to SBS homomers, the ability to bind the ligands of homologs changes qualitatively similarly to monomers. In the case of SBS complexes, stochastic processes (i.e., drift) are likely to play a significant role in shaping the number of subunits or their topology. (2) Evolution of ligand binding of different ligand categories in homomers (seeFigure S1for heteromers). (E and F) The binding pockets of small molecules show much higher conservation in (E) MBS homomers than in monomers or (F) SBS homomers. (G and H) In the case of cofactors, the binding pockets are highly conserved and show little difference between QS types (G, MBS homomers; H, SBS homomers). (I and J) The metal binding pockets are the less conserved and show no qualitative differences between the groups, most likely due to their small size (I, MBS homomers; SBS homomers). (3) Evolution of quaternary structure. (K) In homomers that bind cofactors or metal ions, the structure of binding pockets have fundamental consequences for the evolution of quaternary structure (see also Figure S2). Homomers with multi-chain biding sites have significantly lower variability in quaternary structure than complexes with SBSs, suggesting that the structure of the binding pocket is an important determinant of the evolution of their quaternary structure. (L) In the case of heteromers with cofactors and metals, we see the opposite pattern to homomers in the evolution of QS: MBS heteromers appear to change faster than SBS heteromers. On all panels, bars represent proportions, whiskers 95% confidence intervals. ^∗∗p < 0.005; ^∗p < 0.05; tests of proportions, with Benjamini-Hochberg correction for multiple testing.

Figure 3

Homomers with MBSs Are Characterized with Chemically More Similar Ligands (A) Example of ChEBI ontology similarity between two different chemical compounds (dopamine and guanine). Shared ontology terms are highlighted with red; dopamine and guanine specific terms are highlighted with blue and green, respectively. The difference in their chemical composition was measured as the proportion of the ligand specific and all terms, i.e., (blue + green)/all terms: (15 + 17)/49 = 65%. (B and C) The chemical properties of ligands of homomers with MBSs (B) change less with sequence divergence than the properties of ligands of monomers or other complexes, but not in heteromers (C). (D and E) Ligands of homomers with SBSs (D) show a weak but consistently higher variability in the chemical properties of ligands than monomers, while there is no clear difference between SBS heteromers (E) and monomers. Bars represent averages; whiskers 95% confidence intervals; ^∗∗p < 0.005; ^∗p < 0.05; t tests, with Benjamini-Hochberg correction for multiple testing. See also Figure S3.

Figure 4

Homologs of MBS Homomers Have More Similar Binding Site Structure Than SBS Homomers or Monomers (A–C) The comparison of binding sites of homologous proteins with 30%–50% sequence identity indicates that the structural variability of binding sites of (A) homomers (measured as the root mean square deviation [RMSD], in angstrom) with multi-chain binding pockets is significantly lower than the variability of single-chain binding pockets, whereas there is no such pattern in heteromers (B). Monomers (C) show a pattern comparable to SBS homomers. (D) The comparison of apo and holo structures of homomers indicates that the binding pockets MBS homomers do not show a larger structural variability than SBS homomers. ^∗∗p < 0.005; ^∗p < 0.05; Wilcoxon tests with Benjamini-Hochberg correction.

Figure 5

Functional Characteristics of MBS Homomers, in Relation to All Homomers included in the Analysis (A) Scatterplot of significantly enriched GO molecular function terms, summarized and visualized with REVIGO. Related terms form clusters that are labeled with the most significant term of the cluster. The size of the circles corresponds to the number of proteins in the term; colors indicate significance. (See also Figure S5 and Table S1 for all enriched terms and exact significances.) MBS homomers have diverse functions, including acyl-CoA dehydrogenase activity, transaminase activity, thiamine pyrophosphate binding, cofactor binding, transmembrane transport, ion binding, or transporter activity, which frequently involve binding of cofactors. (B) Graph of significantly enriched ChEBI structural ontology terms. White nodes are not significant; the intensity of red corresponds to significance (see Table S2 for exact p values). Most enriched structural terms are related to nucleobases/nucleotides. (C) Graph of significantly enriched ChEBI role ontology terms. (See Table S3 for exact p values.) The enrichment shows that the ligands of MBS homomers are typically involved in metabolism.

Figure 6

Structural and Evolutionary Characteristics of Single- and Multi-chain Binding Complexes (A) Subunits forming MBSs are significantly more flexible than subunits with single-chain sites (Wilcoxon tests). (B) The frequency of single- and multi-chain binding sites in different symmetry groups of homomers. The symmetry groups are A, asymmetric; C, cyclic; C2, two-fold dimeric; C2h, two-fold symmetric with >2 subunits; and D, dihedral (tests of proportions). (C) The frequency of homomers is higher among prokaryotes, whereas heteromers are more frequent in eukaryotes (tests of proportions). (D) The frequency of complexes with MBSs is higher in prokaryotes than in eukaryotes, particularly in the case of heteromers (tests of proportions). On (B)–(D), whiskers represent 95% confidence intervals. On all panels, ^∗∗p < 0.005 and ^∗p < 0.05.

Figure 7

The Frequency of Pathogenic Mutations Is Different in Different Complexes (A–D) Complexes with (A) small molecules, (B) cofactors, (C) metal ions, and (D) all ligands. The baseline level of pathogenic mutations is highest in MBS heteromers and lowest in monomers. In most complex types, the frequency of pathogenic mutations of the binding sites is higher than the baseline level, particularly in the case of heteromers. ^∗∗p < 0.005; ^∗p < 0.05; tests of proportions; whiskers represent 95% confidence intervals.