Fitness landscapes for species interactions: when do population genetics and adaptive dynamics diverge?

This study employs a data-driven microbial model to demonstrate that population genetics and adaptive dynamics often yield divergent eco-evolutionary predictions for species interactions, particularly regarding coexistence, due to the critical influence of finite timescales, mutation supply, effect sizes, and species asymmetries.

The Gist

Imagine you are watching a race between two runners, but instead of running on a track, they are running up a giant, invisible mountain called the "Fitness Landscape." The higher they climb, the better they survive and reproduce.

This paper is a detective story comparing two different ways scientists try to predict how these runners will behave: Adaptive Dynamics and Population Genetics.

The Two Predictors

1. Adaptive Dynamics (The Smooth Map):
   Think of this as a GPS that assumes the runners are perfect, infinite machines. It assumes mutations (changes in the runner's shoes or muscles) happen one tiny step at a time, very rarely. It draws a smooth, continuous path up the mountain. It tells you, "If they keep going, they will definitely reach the very peak (the best possible spot) and stay there forever." It's great for seeing the big picture, but it ignores the bumps, the stumbles, and the fact that real runners get tired.

2. Population Genetics (The Real-Time Camera):
   This is a high-definition camera filming a real race with a limited number of runners. It knows that mutations can be huge leaps or tiny steps, that there are only so many runners (finite population), and that sometimes a runner might trip and fall off the mountain entirely. It simulates the messy, chaotic reality of evolution over a specific amount of time.

The Big Question

The authors asked: "Do these two methods agree?"
They used a computer model based on real bacteria (like yeast in sourdough bread) to see if the smooth GPS map matches the messy real-time camera footage.

What They Found (The Plot Twists)

1. The "Time Limit" Problem

The GPS (Adaptive Dynamics) says, "You will reach the peak."
The Camera (Population Genetics) says, "Maybe, but only if you have enough time and energy."

• The Analogy: Imagine trying to climb a mountain. The GPS says, "You'll get to the top." But the camera shows that if you only have a few hours (finite time) and you don't have many energy bars (mutations), you might get stuck halfway up.
• The Result: If mutations are rare or small, the bacteria might never reach the "perfect" peak the GPS predicted, even after thousands of generations. They get stuck in the "foothills."

2. The "Trade-Off" Trap

In biology, you can't be good at everything. If you run fast (high growth rate), you might carry less fuel (lower yield). This is a trade-off.

• The Analogy: Think of a car. You can tune it for speed or for fuel efficiency, but rarely both perfectly. The GPS assumes the car is forced to stay on a specific road that balances speed and fuel.
• The Result: The researchers found that if the "road" (the trade-off curve) is too strict, or if the car doesn't have enough fuel (mutations) to get there, the car might end up in a ditch below the road. The GPS predicted a perfect balance, but the real simulation showed the car crashing because it couldn't make the turn in time.

3. The "Uneven Race" (Two Species)

This is where it gets really interesting. They simulated two species competing.

• The Analogy: Imagine two runners, Alice and Bob. The GPS says, "They will run side-by-side to the top forever."
• The Reality: What if Alice has a super-powerful engine (lots of big mutations) and Bob has a weak engine (few mutations)?
  • Alice zooms ahead.
  • Bob lags behind.
  • Because Bob is slow, he gets pushed off the mountain by Alice.
  • The Twist: The GPS predicted they would coexist (run together). The Camera showed that because Alice evolved faster, she accidentally (or intentionally) pushed Bob off the cliff. Coexistence failed.

Why Does This Matter?

This paper is a wake-up call for scientists doing evolution experiments (like growing bacteria in a lab for months).

• Don't trust the smooth map blindly: Just because a theory says "these two species will live together forever" doesn't mean they will in a real, finite experiment.
• Time matters: If you stop the experiment too early, you might think evolution failed, when really it just needed more time.
• Luck matters: Sometimes, a species goes extinct not because it's "bad," but because it didn't get the right "lucky" mutations fast enough to keep up with its rival.

The Bottom Line

Evolution isn't just a smooth march toward perfection. It's a chaotic, messy race where the size of the crowd, the speed of the mutations, and the time you have to watch the race all decide who wins, who loses, and who survives. The "perfect" predictions of the past need to be adjusted to account for the messy reality of the present.

Technical Summary

1. Problem Statement

The paper addresses a fundamental disconnect between two major frameworks used to study the evolution of species interactions: Adaptive Dynamics (AD) and Population Genetics (PG).

• Adaptive Dynamics: Relies on deterministic differential equations, linear stability analyses, and the assumption of infinite population sizes, rare mutations of small effect, and clonal interference-free evolution. It predicts long-term outcomes like Evolutionarily Stable Strategies (ESS) and stable coexistence using invasion fitness landscapes.
• Population Genetics: Mechanistically models finite populations, stochastic processes, and specific mutation supplies over finite timescales.
• The Gap: While AD is powerful for predicting qualitative outcomes (e.g., coexistence vs. exclusion), it is unclear if these predictions hold in realistic, finite populations with specific mutation rates, effect sizes, and initial conditions. Furthermore, AD assumptions (e.g., traits constrained strictly to trade-off curves) may conflict with mechanistic realities where populations evolve toward performance bounds rather than being constrained to them. The authors aim to determine under what conditions these frameworks diverge, particularly regarding the timescale of evolution and the stability of species coexistence.

2. Methodology

The authors employed a hybrid approach combining analytical theory, data-driven modeling, and stochastic simulations.

• Theoretical Framework (Adaptive Dynamics):

  • Utilized the Generalized Lotka-Volterra (gLV) model to describe species growth and interactions.
  • Defined growth parameters: $r_i$ (basal growth rate) and $a_{ii}$ (intraspecific competition coefficient).
  • Assumed $r_i$ and $a_{ii}$ evolve, while interspecific coefficients ($a_{ij}$) are fixed.
  • Derived invasion fitness ($s$) by calculating eigenvalues of the stability matrix for rare mutants.
  • Identified candidate ESSs using pairwise invasibility plots (PIPs) and second-order derivatives, imposing exponential trade-off functions ($a_{ii} = C e^{Cr_i}$) to ensure convergence stability.

• Simulation Framework (Population Genetics):

  • Used SLiM v4.3 to perform forward-in-time, stochastic simulations of haploid, clonal populations.
  • Key Deviations from AD Assumptions:
    • Finite Population Sizes: Scaled to thousands (computationally feasible) rather than infinite.
    • Mutation Supply: Varied mutation rates ($\mu$), the proportion of trait-increasing mutations ($m_k$), and effect sizes (sampled from Gamma distributions).
    • Trade-off Interpretation: Treated trade-off curves as upper bounds on performance rather than strict paths. Populations could start "off" the curve and evolve toward it, accumulating mutations until bounded by the curve.
    • Finite Timescales: Simulations ran for fixed durations ($5 \times 10^4$ generations) or until reaching ESS/extinction.

• Experimental Design:

  • Single-Species: Analyzed the time to reach ESS as a function of mutation supply and effect sizes.
  • Two-Species: Investigated coexistence stability under varying mutation asymmetries between species.
  • Initial Conditions: Tested populations initialized both on and off the trade-off curve, and at varying distances from the ESS.

3. Key Contributions

• Bridging Frameworks: The study provides a rigorous quantitative link between the deterministic predictions of AD and the stochastic reality of PG, identifying specific parameter regimes where they diverge.
• Reconceptualizing Trade-offs: The authors propose modeling trade-offs as performance bounds (constraints imposed by physics/energetics) rather than strict evolutionary paths, allowing for more realistic modeling of populations starting in maladapted states.
• Quantifying Timescales: They derive analytical expressions for the expected time to reach an ESS, demonstrating that this time is inversely proportional to the mutation supply and the square of the effect size.
• Asymmetry in Coexistence: They demonstrate that coexistence predictions are highly sensitive to asymmetries in mutation supply between competing species, a factor often ignored in standard AD models.

4. Key Results

A. Single-Species Evolution

• Timescale Dependence: The time required to reach the ESS is heavily dependent on mutation supply ($L\mu$) and effect size ($\Delta r$). Specifically, $Time \propto 1/(\mu \cdot \Delta r^2)$.
• Failure to Converge: In finite time simulations, populations with low mutation rates, low supplies of beneficial mutations, or small effect sizes frequently fail to reach the predicted ESS within experimental timescales (e.g., 50,000 generations).
• Inference of Trade-offs: The ability to detect the underlying trade-off function from experimental data depends on the mutation supply. Low supply leads to populations remaining far from the trade-off curve, resulting in poor inference of the true trade-off shape, even after long durations.

B. Two-Species Coexistence

• Divergence from AD Predictions: While AD predicts stable coexistence for specific parameter sets, stochastic simulations often resulted in competitive exclusion (extinction) of one species.
• The "Intermediate" Danger Zone: Extinctions were most frequent at intermediate levels of mutation supply.
  • Low Supply: Both species evolve too slowly to significantly alter the competitive landscape; coexistence is maintained simply due to lack of change.
  • High Supply: Both species adapt rapidly to the ESS, maintaining coexistence as predicted.
  • Intermediate Supply: One species may adapt faster (due to stochasticity or slightly higher supply) and cross a threshold that pushes the system out of the coexistence region before the other species can catch up.
• Asymmetry: When mutation rates were asymmetric (one species high, one low), the species with the lower mutation rate was significantly more likely to go extinct, even if AD predicted coexistence.
• Initial Conditions: The probability of coexistence destabilization is highly sensitive to the initial distance from the ESS and the specific evolutionary trajectory taken. If the trajectory temporarily exits the coexistence region, extinction can occur even if the final ESS is stable.

5. Significance

• Interpretation of Experimental Evolution: The findings suggest that "failure" to observe predicted evolutionary outcomes (like coexistence or specific trade-off shapes) in lab experiments may not indicate a flaw in the theory, but rather a limitation of finite timescales and mutation supplies.
• Refining Predictive Models: The study highlights that for accurate predictions in microbial communities, models must account for finite population sizes and the specific distribution of mutation effect sizes, not just the existence of a trade-off.
• Mechanistic Insight: By treating trade-offs as performance bounds, the model better aligns with mechanistic constraints (e.g., enzymatic limits) and explains why populations might exhibit "runaway" adaptation or fail to detect trade-offs in short-term experiments.
• Future Directions: The authors call for empirical characterization of mutation supply rates and effect sizes in specific systems to better integrate invasion fitness landscapes with population genetics, moving beyond the assumption of infinite populations and infinitesimal mutations.

In conclusion, the paper demonstrates that while Adaptive Dynamics provides a robust qualitative map of evolutionary outcomes, Population Genetics reveals that the journey to those outcomes is stochastic and time-dependent, often leading to divergent results in finite, realistic scenarios.