Genotype network intersections promote evolutionary innovation

Abstract

Evolutionary innovations are qualitatively novel traits that emerge through evolution and increase biodiversity. The genetic mechanisms of innovation remain poorly understood. A systems view of innovation requires the analysis of genotype networks-the vast networks of genetic variants that produce the same phenotype. Innovations can occur at the intersection of two different genotype networks. However, the experimental characterization of genotype networks has been hindered by the vast number of genetic variants that need to be functionally analyzed. Here, we use high-throughput sequencing to study the fitness landscape at the intersection of the genotype networks of two catalytic RNA molecules (ribozymes). We determined the ability of numerous neighboring RNA sequences to catalyze two different chemical reactions, and we use these data as a proxy for a genotype to fitness map where two functions come in close proximity. We find extensive functional overlap, and numerous genotypes can catalyze both functions. We demonstrate through evolutionary simulations that these numerous points of intersection facilitate the discovery of a new function. However, the rate of adaptation of the new function depends upon the local ruggedness around the starting location in the genotype network. As a consequence, one direction of adaptation is more rapid than the other. We find that periods of neutral evolution increase rates of adaptation to the new function by allowing populations to spread out in their genotype network. Our study reveals the properties of a fitness landscape where genotype networks intersect and the consequences for evolutionary innovations. Our results suggest that historic innovations in natural systems may have been facilitated by overlapping genotype networks.

Fig 1. The experimental fitness landscape at the intersection of two genotype networks.

(A) Overlay of the HDV and Ligase genotype networks. Nodes represent individual genotypes that are connected by an edge if they are different by a single nucleotide change. Nodes are colored based on their dominant activity (red = HDV; blue = Ligase). For each genotype, “ribozyme fitness” is defined as the relative ribozyme activity determined by high-throughput sequencing and is indicated by the size of the node and the color saturation. Genotypes with fitness below 0.07 are excluded for visualization purposes. Boxes on the left (HDV reference) and right (Ligase reference) show the secondary structure for the reference genotypes and all the mutational changes that were analyzed. The mutations in blue boxes convert the HDV reference to the Ligase reference. The mutations in red boxes convert the Ligase reference to the HDV reference. (B) Distance-based layout of the two fitness landscapes. Each sequence is positioned on the x-axis according to its mutational distance from the HDV reference genotype. HDV fitness (red) and Ligase fitness (blue) are indicated by the y-axis value. The number of genotypes (n) increases in the middle of the plot, and the total number of genotypes at each position is indicated about the graph. The number of dual-function intersection sequences (i) at each mutational distance is also indicated. Inset text “peaks” and “ruggedness” describe quantitative characteristics of the landscapes. Data and Python scripts used to construct fitness landscapes can be found on GitLab. (C) Distributions of the magnitude of epistatic values found in each landscape. Data and Python scripts used to calculate and graph epistasis can be found on GitLab. (D) Representation of the chemical reactions catalyzed by each ribozyme. HDV, Hepatitis Delta Virus.

Fig 2. Proximity and overlap of the two genotype networks.

(A) Distributions of shortest mutational distance (x-axis) between genotypes on different networks as a function of fitness cutoff (y-axis; blue = Ligase to HDV distances; red = HDV to Ligase distances). For each genotype with a fitness above the cutoff value for one function, the distance to the nearest genotype with the other function was determined. The distribution of these distances determined for all genotypes are plotted as violin plots. The diagram (above, left) illustrates the measurement of distance between the two functions. Inset (above, right) shows the distribution at fitness cutoff = 1.3 as histograms, and dashed lines indicate the sample means. Data and Python scripts used to plot distributions can be found on Gitlab. (B) Intersection sequences with detectable activity for both functions. For each genotype, the HDV fitness is plotted on the x-axis, and Ligase fitness is plotted on the y-axis. Color indicates the ratio of Ligation fitness (blue) to HDV fitness (red). The size of the node is scaled to the higher of the two fitness values. Fitness values are log10 transformed. Dashed lines indicate wild-type level activity with fitness = 1 (log10 fitness = 0). Data and Python scripts used to plot intersection sequences can be found on Gitlab. HDV, Hepatitis Delta Virus.

Fig 3. Periods of evolutionary stasis revealed by computational simulation of evolutionary innovation.

(A) A landscape visualization of the two genotype networks. The height of each node (z-axis) indicates the relative fitness for the HDV phenotype (red) and the Ligase phenotype (blue). Nodes represent genotypes and are connected by an edge if they are different at one nucleotide position. Fitness is indicated by the height (z-axis), the size of the node, and the color saturation. Fitness values are normalized so that both graphs are similar heights. Genotypes used to start evolutionary simulations are labeled with lower case for the genotypes with the highest HDV fitness (a–p) and capital letters for genotypes with the highest Ligase fitness (A–Q). (B) Frames from simulations of evolving populations. Several examples are shown to illustrate different rates of increase of “population fitness” over simulation time (“generations”). Each row shows the progress of a single simulation. The starting genotype is indicated to the left. Each plot shows the genotypes present in the population with the number of generations of evolution labeled at the bottom. Genotypes present in the population are indicated by yellow nodes and edges. The corresponding mean fitness of each population over time is shown in the plots to the right. During simulations, the population size (N = 1,000) and mutation rate (μ = 0.01) were constant. HDV, Hepatitis Delta Virus.

Fig 4. Starting genotypes result in different rates of evolutionary adaptation.

(A) Rates of Ligase adaptation from a single HDV genotype. Each trace shows the average population fitness as a function of generation time for a separate simulation of 1,000 individuals each. The traces from 100 separate simulations are shown. Inset shows minor fluctuations during periods of stasis. Data and Python scripts for evolutionary simulations can be found on GitLab. (B) Average rates of evolutionary adaptation of Ligase activity starting from 17 genotypes. Each trace represents a different starting genotype (a–p and HDV reference) and shows the mean fitness of 100 simulations as a function of time (“generation”). The y-axis is scaled to the maximum fitness on this landscape (“summit,” horizontal dashed line). The vertical dashed line marks generation 200. Data and Python scripts for evolutionary simulations can be found on GitLab. (C) Rates of HDV adaptation from a single Ligase genotype. Data and Python scripts for evolutionary simulations can be found on GitLab. (D) Average rates of evolutionary adaptation of HDV activity starting from 17 genotypes. Each trace represents a different starting genotype (A–Q) and shows the mean fitness of 100 simulations as a function of time (“generation”). The y-axis is scaled to the maximum fitness on this landscape (“summit,” horizontal dashed line). Data and Python scripts for evolutionary simulations can be found on GitLab. (E) Distributions of initial rates of adaptation during simulations on the Ligase landscape. Initial rate is determined as the population fitness divided by the generations at 200 generations. Each violin plot represents the distribution of 100 simulations starting from the same genotype, which is indicated on the x-axis. Maximum growth rate, determined from a cubic spline regression, is also reported. Growth rate calculations and plots are reported in S10 Fig. Data and Python scripts for evolutionary simulations can be found on GitLab. (F) Sign epistasis in the local fitness landscape of genotypes that cause periods of stasis in the Ligase landscape. The fitness of the stasis genotype is plotted at mutations = 0, and this starting fitness is marked with a dashed line. The fitness of neighboring genotypes that differ by 1 or 2 mutations are shown. Distributions of initial rates of adaptation during simulations on the HDV landscape. Data and Python scripts for plotting local fitness landscapes can be found on GitLab. (G) Distributions of initial rates of adaptation during simulations on the HDV landscape. Growth rate calculations and plots are reported in S16 Fig. Data and Python scripts for evolutionary simulations can be found on GitLab. (H) Sign epistasis in the local fitness landscape of genotypes that cause periods of stasis in the HDV landscape. Data and Python scripts for plotting local fitness landscapes can be found on GitLab. HDV, Hepatitis Delta Virus; REF, reference.

Fig 5. The effects of neutral evolution on evolutionary adaptation.

(A) Average rates of evolutionary adaptation of Ligase activity starting from the summit genotype of the HDV landscape. Each trace represents a different number of generations of neutral evolution and shows the mean fitness of 100 simulations as a function of time (“generation”). The y-axis is scaled to the maximum fitness on this landscape (“summit”). The vertical dashed line marks generation 200. Data and Python scripts for evolutionary simulations can be found on GitLab. (B) Average rates of evolutionary adaptation of HDV activity starting from the summit genotype of the Ligase landscape. Data and Python scripts for evolutionary simulations can be found on GitLab. (C) Distributions of rates of adaptation, final population fitness, and the number of unique genotypes explored following generations of neutral evolution during simulations on the Ligase landscape. Each violin plot represents the distribution of 100 simulations following the same length of neutral evolution, which is indicated on the x-axis. Adaptation rate is determined as the rate of population increase for the first 100 generations following the neutral evolution. Final fitness is the mean population fitness at the end of 1,000 generations of evolution. Maximum growth rate, derived from a cubic spline regression, is also reported. Growth rate calculations and plots are reported in S19 Fig. Data and Python scripts for evolutionary simulations can be found on GitLab. (D) Distributions of rates of adaptation, final population fitness, and the number of unique genotypes explored following generations of neutral evolution during simulations on the HDV landscape. Growth rate calculations and plots are reported in S20 Fig. Data and Python scripts for evolutionary simulations can be found on GitLab. HDV, Hepatitis Delta Virus.