Repeatability of adaptation in interacting species

This study uses an NKC model to demonstrate that intergenomic epistasis in coevolving species drives highly repeatable adaptation patterns that diverge from intragenomic expectations due to fitness trade-offs, often resulting in evolutionary cycling and increased fitness loads.

The Gist

Imagine two dancers, let's call them Focal and Partner, trying to learn a routine together. They are on a giant, invisible dance floor (the Fitness Landscape). Their goal is to find the highest point on the floor, where they can spin the fastest and look the best (maximum Fitness).

In the real world, these "dancers" are species like a host and a parasite, or a flower and a bee. They evolve together, constantly reacting to each other's moves.

This paper asks a big question: If we reset the game and let them dance again from the start, will they end up in the exact same spot every time? Or will they end up in different places? This is called the repeatability of adaptation.

Here is the breakdown of what the researchers found, using simple analogies:

1. The Two Types of "Tangling" (Epistasis)

The researchers looked at how the dancers' moves are connected. They found two ways the dancers get tangled up:

• Intra-genomic Epistasis (Internal Tangles): Imagine Focal's left foot is tied to their own right arm. If they move their arm, their foot has to move too. This is a connection within one dancer's own body.

  • The Result: This creates a bumpy floor with lots of small hills and valleys. It makes it hard to find the one best spot because you might get stuck on a small hill. It makes the outcome less predictable because there are many different small hills to get stuck on.

• Inter-genomic Epistasis (Partner Tangles): Now imagine Focal's left foot is tied to Partner's right arm. If Partner moves, Focal must move, even if it feels weird. This is a connection between the two dancers.

  • The Result: This is the paper's big discovery. When they are tied to each other, the dance floor changes shape dynamically. Surprisingly, this makes the outcome highly repeatable. No matter where they start, they almost always end up in the same specific spot (or the same loop). The "tangle" between them forces them into a very specific pattern.

2. The "Red Light, Green Light" Effect

The researchers found that when the dancers are tied to each other (Inter-genomic epistasis), they often get stuck in a loop.

• The Analogy: Imagine Focal moves forward to get a better view. But because they are tied to Partner, Partner gets pushed backward into a bad spot. Partner then has to move to fix their spot, which pushes Focal back to where they started.
• The Cycle: They keep chasing each other in circles forever. They never reach the "perfect" high point because every time one gets better, the other gets worse. This is called Coevolutionary Cycling. It's like an arms race where neither side ever truly wins; they just keep running in place.

3. The "Cost of Dancing Together" (Fitness Load)

You might think that dancing together helps both partners get to the top. But the paper found that often, both dancers end up lower than they could have been.

• The Analogy: Imagine there is a perfect spot on the dance floor where both could spin at 100 mph. But because they are tied together, they can't both get there. They have to compromise. They end up spinning at 60 mph.
• The "Loser" and "Winner": Sometimes, the compromise hurts one dancer more than the other. One might be at 70 mph while the other is at 50 mph. The paper calls this the Fitness Load. It's the "cost" of having to adapt to your partner.

4. The Shape of the Connection Matters

The researchers tested different ways the dancers could be tied together:

• Specific Ties: If Focal's foot is tied only to Partner's left hand, the dance is very predictable. They always end up in the same spot.
• Messy Ties: If Focal's foot is tied to Partner's left hand, right hand, and head, the dance becomes chaotic. They might end up in many different places, or get stuck in a wild loop.

The Big Takeaway

The main lesson is that how species are connected matters more than how fast they move.

• If you just look at one species alone, you might think evolution is messy and unpredictable.
• But once you add the "partner" into the mix, the rules change. The connection between them creates a rigid structure that forces them into very specific, repeatable patterns.
• However, this repeatability often comes at a price: they might end up in a "good enough" spot rather than the "perfect" spot, or they might get stuck in an endless cycle of chasing each other.

In short: Evolution isn't just about climbing a mountain alone. When you are climbing a mountain while holding hands with someone else, you are forced to take a very specific path. You might end up in the same place every time you try, but you might also end up stuck in a loop, or lower on the mountain than you could have been if you were climbing alone.

Technical Summary

1. Problem Statement

The paper addresses a fundamental question in evolutionary biology: How repeatable is the adaptive process when species coevolve?

• Context: While intragenomic epistasis (genetic interactions within a single genome) is known to constrain mutational paths and increase the repeatability of adaptation, the role of intergenomic epistasis (genetic interactions spanning across species, e.g., host-parasite or mutualist relationships) remains unclear.
• Gap: Previous studies have examined species interactions and genetic interactions separately, but few have investigated how the joint structure of intra- and intergenomic epistasis shapes the repeatability of adaptation, the emergence of evolutionary outcomes (endpoints vs. cycling), and the associated fitness costs.
• Hypothesis: The authors investigate whether intergenomic epistasis constrains adaptation similarly to intragenomic epistasis or if the dynamic nature of coevolution (fitness trade-offs between species) leads to distinct patterns, such as cycling and increased fitness loads.

2. Methodology

The study employs a theoretical modeling approach using a modified NKC model (Kauffman & Johnsen, 1991) to simulate coevolutionary fitness landscapes.

• Model System:

  • Two haploid species: a focal species ($F$) and a partner species ($P$).
  • Genome Structure: Each species has $N=2$ loci with $A=2$ alleles (diallelic).
  • Fitness Calculation: The fitness of a species depends on the "metagenotype" (the combined genotype of both species). Fitness is the average of contributions from each locus.
  • Epistasis Parameters:
    • $K$: Number of interacting loci within the same genome (intragenomic epistasis).
    • $C$: Number of interacting loci across genomes (intergenomic epistasis).
    • Interaction Networks: The specific structure of interactions is defined by interaction matrices (IM), varying in symmetry (symmetric vs. asymmetric) and specificity (specific vs. non-specific).
  • Fitness Landscape Generation: Fitness contributions are drawn from a normal distribution $N(0, 0.25)$, creating probabilistic, rugged landscapes. 200 landscape replicates were generated for various parameter combinations.

• Simulation of Adaptation:

  • Adaptive Walks: Simulated under a Strong Selection, Weak Mutation (SSWM) regime. Species take alternating turns to mutate to a fitter neighbor.
  • Walk Types: Primarily "random" adaptive walks (equal probability of fixing any beneficial mutation), with comparisons to "greedy" and "true" walks in supplements.
  • Mutation Asymmetry: The parameter $R$ was introduced to simulate different mutation rates, allowing one species to take multiple steps before the other.
  • Termination: Walks stop when a joint fitness peak is reached (no beneficial mutations for either species) or when the system enters a cycle (no joint peak is accessible, leading to perpetual oscillation between metagenotypes).

• Metrics:

  • Repeatability: Measured by the number of realized endpoints and the frequency distribution of these endpoints across landscapes.
  • Fitness Load: The difference in fitness rank between the realized outcome and the global fitness peak (rank 1).
  • Interaction Outcomes: Categorized as Win-Win, Win-Lose, Lose-Lose, or Cycling.
  • Landscape Properties: Ruggedness and correlation of fitness ranks between species.

3. Key Contributions

• Decoupling Epistasis Types: The study explicitly separates the effects of intragenomic ($K$) and intergenomic ($C$) epistasis on adaptation repeatability, revealing they have opposing effects.
• Identification of Cycling: The model demonstrates that intergenomic epistasis uniquely enables evolutionary cycling (Red Queen dynamics) even in the absence of population dynamics, a phenomenon not observed in static single-species landscapes.
• Landscape-Specific Repeatability: The authors show that while intragenomic epistasis reduces repeatability within a landscape (due to local peaks), intergenomic epistasis creates highly repeatable, landscape-specific outcomes.
• Emergent Interaction Dynamics: The study links abstract genetic parameters to emergent ecological outcomes (win-win, win-lose, cycling) based on the correlation of fitness ranks between species.

4. Key Results

A. Repeatability of Adaptation

• Intragenomic Epistasis ($K > 0, C = 0$): Increases the number of local peaks, causing adaptive walks to get stuck at different local optima depending on the starting point. This decreases repeatability of endpoints on a single landscape.
• Intergenomic Epistasis ($C > 0$):
  • High Repeatability: Contrary to expectations, intergenomic epistasis leads to highly repeatable outcomes on individual landscapes. Because a joint peak requires both species to be at a peak simultaneously, the number of accessible equilibria is drastically reduced.
  • Landscape Specificity: While outcomes are repeatable within a specific landscape, the specific endpoint varies significantly between different landscape replicates.
  • Cycling: In many cases ($C > 0$), no joint peak is accessible, leading to perpetual cycling between metagenotypes. This is a unique outcome of intergenomic interactions.

B. Interaction Structure Matters

• The structure of the interaction network (symmetry and specificity) is more critical than the navigation method (adaptive walk type) in determining outcomes.
• Specific/Symmetric Networks: Tend to have fewer realized outcomes and higher repeatability across landscapes due to redundancy (some metagenotypes map to identical fitness).
• Asymmetric/Non-specific Networks: Create complex interaction webs, leading to high repeatability within a landscape but diverse outcomes across landscapes.

C. Fitness Load and "Winner-Loser" Dynamics

• Coevolutionary Fitness Load: Intergenomic epistasis often results in a fitness cost. Even when a joint peak is reached, it is frequently not the global peak for both species.
• Asymmetry:
  • Antagonistic Interactions: Strongly negative correlations in fitness ranks (antagonism) lead to high fitness loads and "Win-Lose" or "Lose-Lose" scenarios.
  • Mutualistic Interactions: Positive correlations can still result in fitness loads if a joint global peak does not exist due to genetic constraints.
• Cycling and Load: Cycling outcomes generally incur the highest community fitness load (lowest mean fitness). However, cycling tends to buffer asymmetry, meaning the fitness difference between the "winner" and "loser" species is smaller during cycles compared to static equilibria.

D. Robustness to Navigation

• The patterns of repeatability and fitness load are largely robust to the type of adaptive walk (random vs. greedy) and mutation rate asymmetry ($R$).
• The only significant change observed with mutation rate asymmetry was the elimination of cycling when one species mutated much faster than the other, suggesting cycling requires a balance in evolutionary potential.

E. Predictability

• Global landscape metrics (ruggedness, rank correlation) failed to predict the repeatability of outcomes or the occurrence of cycling. The specific topology of the joint fitness landscape is an emergent property that cannot be inferred from simple summary statistics.

5. Significance

• Theoretical Advancement: The paper challenges the view that epistasis always increases predictability. It shows that intergenomic epistasis creates a unique form of determinism where outcomes are highly repeatable within a specific coevolutionary context but highly unpredictable across different contexts.
• Ecological Implications: The findings suggest that coevolution often imposes a "fitness tax" (load) on interacting species. The existence of cycling implies that species may never reach a static optimal state, instead engaging in perpetual arms races that maintain genetic diversity but reduce overall population fitness.
• Methodological Insight: The study highlights that understanding adaptation in interacting systems requires analyzing the joint fitness landscape (metagenotype space) rather than treating species in isolation. Standard single-species fitness landscape models are insufficient for predicting coevolutionary dynamics.
• Future Directions: The authors propose that empirical validation could be achieved through co-evolutionary Genome-Wide Association Studies (co-GWAS) to map the complex fitness landscapes of real interacting species.

In summary, the paper demonstrates that intergenomic epistasis acts as a strong constraint that funnels coevolving species into highly repeatable, yet often suboptimal, evolutionary trajectories, frequently characterized by cycling and significant fitness trade-offs.