Figure 1.
An extended model of protein evolution incorporates the stability effect of mutations.
A Conventionally, protein evolution is described by the rates of non-synonymous substitutions dN, and synonymous substitutions dS, and their rate ratio ω = dN/dS. To accommodate the disparate effect of different amino acid changes on protein stability, the rate of non-synonymous substitutions dN is split into the rates of conservative substitutions dC, and non-conservative substitutions dNC. B Predicted stability effects of conservative (C) and non-conservative (NC) mutations. NC mutations are more likely destabilizing than C mutations. C NC mutations are much more likely highly destabilizing (ΔΔG<−2 kcal/mol) than C mutations. D The evolutionary rate dNC is on average lower than dC and dN, suggesting that NC substitutions are generally under stronger purifying selection. E The evolutionary rate ratios ω = dN/dS and λ = dNC/dC are not correlated, thus independent parameters that contribute orthogonal insights.
Figure 2.
λ describes the evolution of intrinsically disordered proteins.
Proteins are classified according to their percentage of residues that is predicted to be disordered (S: <10%, M: 10–30%, U: >30%). A The evolutionary rate ratio ω is largely independent of the level of protein structuredness for proteins sampled with similar levels of expression. B The evolutionary rate ratio λ increases with increasing protein disorder. C Comparison of the local sequence variability and conservation of the unstructured regions (stretch) to the immediately adjacent structured flanking regions (flank) with the Rate4Site algorithm [41] indicates compensatory selective pressure in the flanking regions. D Protein half-lives on average decrease with increasing protein disorder. Significance levels are indicated as n.s. (not significant), and *** (p<0.001).
Figure 3.
λ describes the re-wiring of protein-protein interactions.
A The most highly re-wired proteins are derived from a consensus ranking of nodes with the most non-orthologous neighbors in the protein interactions networks (PN) of S. cerevisiae (Scer) and S. pombe (Spom). B Highly re-wired proteins evolve at lower ω, thus are under stronger purifying selective pressure, but evolve at clearly higher λ compared to less rewired proteins. C NC mutations are much more likely to comprise changes between amino acids with very different interaction propensities than C mutations. Shown are the distributions of the changes in interaction propensity of NC and C mutations (lines), and a simpler binary classification into high and low differential interaction propensities (bars, more transparent in the background). D The number of protein-protein interactions (PPIs) negatively correlates more strongly with dC than dNC, and only weakly with λ, suggesting that λ is not systematically biased by high connectivity in protein networks. Statistical significance is indicated as ** (p<0.01).
Figure 4.
Weak chaperone dependence promotes NC substitutions.
A The over-expression of GroEL in the directed evolution of enzymes reported in [6] promotes NC substitutions, thus validation the definition of NC mutations and λ. B Native substrates of the E. coli chaperonin GroEL (n = 58) do not evolve at higher ω than non-substrates (n = 216). Shown is the partial regression plot of the contribution of chaperone dependence on ω, together with 5% confidence intervals. The confidence intervals reflect the limited power of predicting a continuous variable, e.g. ω, from chaperone dependence alone. The significance of chaperone dependence is assessed by non-parametric regression analyses that explicitly include expression and disorder as confounding factors. C Native substrates of the E. coli chaperonin GroEL evolve at significantly higher λ. D Human kinases that are substrates of Hsp90 do not evolve at significantly higher ω than kinases that are not Hsp90 substrates when expression levels as confounding factor are included in the analysis. E Both strongly (n = 71) and weakly (n = 70) Hsp90 dependent kinases exhibit a significantly higher λ than non-substrates (n = 61). F Strong substrates of the yeast Hsp70 SSB (n = 648) evolve at lower ω, weak substrates (n = 310) at similar ω compared to non-substrates (n = 721). G Strong SSB substrates also evolve at lower λ, but weak substrates evolve at higher λ. Significance levels are indicated as n.s. (not significant), * (p<0.05), ** (p<0.01), and *** (p<0.001).
Figure 5.
Selective protein quality control in the evolution of protein networks.
A Hubs comprise the most connected proteins in protein networks, while low connectivity bottlenecks are functionally important network nodes that connect network modules. B Hubs and bottlenecks as the most critical proteins in protein interaction networks are characterized by low ω, thus strong purifying selective pressure. Despite their generally high degree of conservation, both hubs and bottlenecks exhibit significantly higher distributions of λ. C Hubs compared to non-hubs are significantly enriched in chaperone substrates and proteins with long disordered stretches. D Deletion of the Hsp70s SSB1/2 (data obtained from [65]) results in 50% of the hubs, but less than 10% of the non-hubs to immediately aggregate. While SSB might not directly promote the evolution of hubs, it is instrumental in the homeostasis of intrinsically disordered hub proteins, thus serving as evolutionary potentiator. E Bottlenecks are significantly enriched in chaperone substrates and proteins with long disordered stretches compared to non-bottlenecks. F The Hsp70 SSB directly promotes non-conservative mutations in bottlenecks in addition to, but independent of intrinsic disorder. Significance levels are indicated as n.s. (not significant), * (p<0.05), and *** (p<0.001).
Figure 6.
Cost-benefit trade-off in protein network evolvability.
Complementary cellular quality control strategies promote non-conservative mutations, thus the evolvability of protein interactions. Intrinsically disordered proteins allow more non-conservative substitutions, but are subject to a more costly regulated turn-over to prevent their aggregation. While energetically expensive, molecular chaperones can promote non-conservative substitutions directly by buffering their destabilizing effect, or indirectly by stabilizing intrinsically disordered proteins. By conferring robustness to otherwise deleterious mutations, protein quality control mechanisms facilitate a higher number of non-conservative mutations, which increases the likelihood of evolving new protein interactions and functions.
Table 1.
Conservative and non-conservative amino acid substitutions.