How the Structure of Signaling Regulation Evolves: Insights From an Evolutionary Model

Abstract

To monitor environmental changes, signaling pathways attenuate their activity with negative feedback loops (NFLs), where proteins produced upon stimulation downregulate the response. NFLs function both upstream of signaling to reduce input and downstream to reduce output. Unlike upstream NFLs, downstream NFLs regulate gene expression without the involvement of intermediate proteins. Thus, we hypothesized that downstream NFLs evolve under more stringent selection than upstream NFLs. Indeed, genes encoding downstream NFLs evolve at a slower and more consistent rate than upstream genes, suggesting that the latter may be under weaker or more context-specific selection. This suggests that downstream NFLs evolve more robustly, whereas upstream NFLs are more susceptible to changes in signaling proteins and stimuli. We tested these assumptions using a minimal model of immune signaling, which predicts robust evolution of downstream NFLs to changes in model parameters. This is consistent with their critical role in regulating signaling and the conservative rate of evolution. Furthermore, we show that the number of signaling steps needed to activate a downstream NFL is influenced by the cost of signaling. Our model predicts that upstream NFLs are more likely to evolve under a shorter half-life of signaling proteins, absence of host-pathogen co-evolution, and a high infection rate. Although it has been proposed that NFLs evolve to reduce the cost of signaling, we show that a high cost does not necessarily predict the evolution of upstream NFLs. The insights from our model have broad implications for understanding the evolution of regulatory mechanisms across signaling pathways.

Keywords: downstream regulation; multilevel regulation; negative feedback; upstream regulation.

Fig. 1.

Genes encoding downstream NFLs evolve at a slower and more consistent rate comopared to genes encoding upstream NFLs. In the left panel $\omega =\mathrm{dN}/\mathrm{dS}$ values for upstream and downstream NFLs are shown and the biological pathway is indicated in the parentheses. The ω values are shown for 20 downstream and 25 upstream NFLs. The ω values for human genes (H−) are from Liu et al. (2014), Drosopila melanogaster (D−) genes are from flyDIVaS_v1.2 (Clark et al. 2007; Stanley and Kulathinal 2016), mouse-rat (M−) and zebra finch-chicken (Z−) data are from BioMart Ensembl 99 (http://jan2020.archive.ensembl.org/biomart/martview). Statistical analyses using bootstrapping while accounting for species-specific ω and the effect of orthologous genes shows that downstream genes evolve at slower and more consistent rates (see supplementary methods, Supplementary Material online and supplementary fig. S1, Supplementary Material online for full details). The function of five upstream and five downstream NFLs are summarized in the right panel.

Fig. 2.

The model of the immune signaling network, capturing the core elements shared across immune signaling pathways. A host has five proteins: R (receptor), A (activator), I (immunity), U (upstream regulator), and D (downstream regulator). The host is exposed to a pathogen (P) with a probability θ, and the pathogen replicates at a rate π. The dashed arrows show potential interactions that can be formed during the evolution of the network. The arrows point from protein j to i and ${\mu }_{i,j}$ is the strength of the interaction. The three domains of each protein (receiver, neutral, and sender) are shown in the top left. The neutral domain does not participate in protein–protein interactions. For a detailed description of the model see Materials and Methods. A list of all variables and values can be found in the supplementary table S1, Supplementary Material online.

Fig. 3.

The evolution of immune signaling when the protein degradation rate is 0. The Y-axis in the left panel shows the coefficient of interaction $({\mu }_{i,j})$ from $-1$ (complete inhibition) to $+1$ (complete activation). The interaction does not evolve when μ fluctuates around 0. The X-axis shows the generation time. The right panel shows the evolved signaling pathway after 20,000 generations. The color of the arrows in the right panel corresponds to the color of the plots in the left panel. The oscillating lines represent co-evolution between the host and the pathogen. The dashed arrows show the interactions that did not evolve. Here, the infection rate $\theta =1$; thus, all individuals are infected. The initial condition ($P_0$) for the pathogen, its replication rate (π), and mutation rate for the host ($M_H$) and pathogen ($M_P$) are shown in the right panel.

Fig. 4.

A high cost of immunity results in the evolution of two types of signaling networks with a downstream NFL. a) Shows whether the first signaling step (activation of A by R) evolves. The Y-axis shows the coefficient of interaction (μ) and the X-axis is the generation time. The plots in b) (10 plots showing 10 replicate simulations) show the effect of D (red) and A (blue) on I. In 20% of the simulations, the activator (A) activates the immune response and the downstream regulator (D) acts as an NFL. These are annotated as A+ . In 60% of the simulations the downstream regulator activates immunity (D+) and the activator protein (A) functions as an NFL. In $20\text{\%}$ of the simulations the receptor does not activate the activator protein (gray trajectories in a); thus, signaling is not transmitted to the host proteins. These two simulations are marked with gray crosses in b). The two types of evolved signaling networks are shown in panels c1 and c2), where the numbers indicate the number of steps needed to downregulate I.

Fig. 5.

The evolution of the upstream NFL is sensitive to changes in the parameter values. In a), nonzero degradation rate for the host protein $(\varphi =0.15)$ and the absence of co-evolution between the receptor and pathogen $({\mu }_{R,P}=1)$ give rise to the evolution of the upstream NFL in most simulations along with the downstream NFL. In b), a more robust evolution of the upstream NFL is observed when amongst the host proteins only the receptor degradation rate is set to 0 $({\varphi }_{i\ne R}=0.15$ and ${\varphi }_R=0$). This condition is shown in green and is repeated in c) and d). c) Shows the combined effect of infection rate $(\theta )$ and population size $(n)$. When the infection rate is <1 $(\theta =0.7)$, the upstream NFL does not evolve in some simulations. Increasing the population size $(n=5,000)$ ensures the evolution of the upstream NFL even when the infection rate $\theta =0.7$. In a–c) we set cost of infection $\alpha =2$ and cost of immunity $\beta =1$. d) Shows the effect of the cost of the immune response under the conditions that favor the evolution of the upstream NFL. When the cost of immunity is high ($\beta =2$) the upstream NFL does not evolve in some simulations. Instead, the effect of A on $I$ is reduced to attenuate the immune response. In all simulations presented here, D evolves as an NFL.

Fig. 6.

The downstream regulator evolves under conservative evolution, whereas the upstream regulator is under weak selective pressure. a) Shows a typical rate of change for the neutral domain ($k_s$) of all host proteins across simulations. b) Shows the distribution of pooled ω values across three simulations for all proteins of the host. c) Shows the evolution of downstream and upstream regulators. The rate of evolution (ω) is calculated every five generations and only when $k_s<0.5$. The X-axis shows the instances of calculation ω through the evolution, and the Y-axis shows the rate of evolution for the two domains (rows) of D (left column) and U (right column). The trajectories show the average rate of evolution across 10 simulations. The shaded regions specify the standard deviation. Positive values on the Y-axis show positive selection, 0 indicates neutral evolution and negative values specify negative selection. The evolution of D and U are shown under three parameter regimes. Under one regime D evolves as an NFL (purple). Under the second regime, there is polymorphism in the choice of the downstream NFL (either A or D), which is shown by the yellow trajectories. The cyan trajectories specify the parameter regime under which both U and D evolve as NFLs.