Effect of population structure and stabilizing selection on quantitative genetic variation

This paper demonstrates that in a subdivided population under stabilizing selection, a critical migration threshold exists where genic variance is maximized and the genetic basis of trait variation across subpopulations is most similar, a phenomenon accurately predicted by an analytical mean-field approach using effective parameters.

The Gist

Imagine a vast archipelago of islands, each inhabited by a small group of people. Everyone on every island shares a common goal: to be the "perfect height." If you are too tall or too short, you struggle to survive and reproduce (this is stabilizing selection). However, there is a twist: being tall isn't just about one gene; it's the result of hundreds of tiny genetic switches scattered throughout your DNA.

Now, imagine boats occasionally travel between these islands, carrying people (and their genes) from one place to another. This is migration.

This paper asks a fascinating question: How does the movement of people between islands affect the variety of heights we see, both on a single island and across the entire archipelago?

Here is the story of what the researchers found, told through simple analogies.

1. The "Goldilocks" Zone of Migration

The researchers discovered that the amount of travel between islands creates three distinct scenarios, but the most interesting one happens at a specific "Goldilocks" level of traffic.

• Too Little Travel (Isolation): If the islands are completely cut off, each island eventually settles on a different "recipe" for the perfect height. Island A might use mostly "tall" switches, while Island B uses mostly "short" switches. Both groups are the right height, but they got there using different genetic tools. The total variety across the whole archipelago is huge, but each individual island is very uniform.
• Too Much Travel (The Melting Pot): If boats are constantly shuttling people back and forth, the islands become a giant, mixed-up crowd. Everyone starts using the same genetic recipe. The variety within each island drops because everyone is becoming the same.
• The "Just Right" Level (The Critical Threshold): The paper found a critical tipping point. When migration is at a specific, moderate level, the variety within a single island actually explodes. It's like a busy marketplace where people are bringing in new ingredients from everywhere, but not so much that the local recipe gets completely overwritten. This is the point where a single island holds the maximum amount of genetic diversity possible.

2. The "Hidden" Variety (Cryptic Variation)

Here is a counter-intuitive finding: Even though the islands look different genetically (Island A has "tall" genes, Island B has "short" genes), the actual height of the people doesn't vary much between islands. They all hit the "perfect height" target.

Think of it like two different teams building a bridge to the same height. Team A uses red bricks; Team B uses blue bricks. The final bridge height is identical, but the materials are totally different.

• The Takeaway: The genetic differences are "cryptic" (hidden). If you just looked at the people's heights, you'd think the islands were identical. But if you looked at their DNA, you'd see they are using completely different blueprints. This hidden variety is a "reserve" that could be unlocked if the environment changed.

3. The "Outlier" Problem (Why GWAS is Tricky)

The paper also tackles a real-world problem for scientists: Genome-Wide Association Studies (GWAS). These are studies where scientists scan the DNA of many people to find which genes cause a trait (like height or disease risk).

• The Problem: If you study people from Island A, you might find that "Gene X" is the main driver of height. But if you take that same finding and apply it to Island B, it might fail completely. Why? Because Island B uses "Gene Y" to achieve the same height.
• The Paper's Insight: The researchers found that the "portability" of these studies depends heavily on how much the islands mix.
  • If islands are too isolated, they use totally different genes.
  • If they mix too much, the unique "outlier" genes disappear.
  • Surprisingly, the genetic basis of the trait is most similar between islands at that "Goldilocks" level of migration we mentioned earlier. At this specific level, the same genes tend to be the "stars" in multiple places, making it easier to predict traits across different populations.

4. The "Effective" Migration Rate

Finally, the paper introduces a clever mathematical trick. They realized that because the islands are all trying to hit the same "perfect height" target, the mixing of genes isn't as simple as just counting boats.

If a person from Island A (who uses "tall" genes) marries someone from Island B (who uses "short" genes), their children might have a weird mix of genes that makes them the wrong height, even if they have the right number of "tall" and "short" switches. This creates a hidden penalty for mixing.

• The Metaphor: It's like trying to mix oil and water. Even if you stir them together (migration), they naturally want to separate because they don't blend well chemically. The researchers calculated an "Effective Migration Rate"—a number that tells us how much mixing actually happens after accounting for this genetic friction. It turns out that even a little bit of "friction" can drastically slow down how quickly islands become genetically similar.

Summary

In simple terms, this paper shows that population structure (islands) and selection (the pressure to be perfect) dance together in a complex way.

1. Moderate travel creates the most diversity within a single group.
2. Isolation creates huge genetic differences between groups, even if they look the same on the outside.
3. Predicting traits across different groups is hardest when they are either totally isolated or totally mixed, but works best at a specific middle ground.

This helps scientists understand why genetic studies in one country don't always work in another, and how nature maintains a massive "library" of genetic options that we might not even see until we look closely.

Technical Summary

1. Problem Statement

The paper investigates how spatial population structure (subdivision into demes) interacts with stabilizing selection to maintain quantitative genetic variation in polygenic traits. While it is well-established that large, panmictic populations maintain variation through a balance of mutation and selection, the dynamics in subdivided populations are less understood.

Key questions addressed include:

• How much genetic variation (both genic and phenotypic) can be maintained in a subdivided population compared to a panmictic one?
• How is this variation partitioned within subpopulations versus among them?
• Does population structure alter the genetic architecture of traits (i.e., which loci contribute most to variance) and the similarity of this architecture across subpopulations?
• How do these factors impact the "portability" of Genome-Wide Association Studies (GWAS) and genomic predictions across populations?

The study focuses on the simplest scenario: a subdivided population under spatially homogeneous stabilizing selection (the same trait optimum across all demes), where variation is maintained by mutation, drift, migration, and selection.

2. Methodology

The authors employ a combination of analytical theory and individual-based simulations within an infinite-island model framework.

• Model Setup:

  • Population: $D$ demes (islands), each with $N$ diploid individuals. Migration occurs at rate $m$ per individual.
  • Trait: An additive quantitative trait influenced by $L$ unlinked, bi-allelic loci with effect sizes $\alpha_i$.
  • Selection: Stabilizing selection with a Gaussian fitness function $W(z) = e^{-(z-z_{opt})^2 / 2V_s}$.
  • Evolutionary Forces: Mutation ($\mu$), genetic drift, migration, and selection.

• Analytical Approach:

  • Single-Locus Approximation: The authors first treat the polygenic trait as a set of independent underdominant loci (heterozygote disadvantage) because stabilizing selection on the trait mean generates selection against variance. They use the diffusion approximation (Kimura, 1964) to derive the equilibrium distribution of allele frequencies ($p$) in a deme conditioned on the global mean frequency ($\bar{p}$).
  • Linkage Equilibrium (LE) vs. Linkage Disequilibrium (LD):
    • LE Predictions: Assume loci evolve independently.
    • LD Predictions (Mean-Field): Account for statistical associations (LD) between loci using effective parameters.
      • Effective Selection ($s_{eff}$): Adjusts for negative LD generated by stabilizing selection (Bulmer effect), which reduces trait variance.
      • Effective Migration ($m_{eff}$): Adjusts for indirect selection against migrant alleles due to their association with less fit genetic backgrounds (reduced fitness of F1 hybrids and backcrosses).
  • Quantitative Genetic (QG) Predictions: Derive analytical expressions for $F_{ST}$ (allelic differentiation) and $Q_{ST}$ (phenotypic differentiation) and total genetic variance ($V_T$) based on the effective parameters.

• Simulations:

  • Discrete-generation, individual-based simulations with $D=100$ demes and $N=200$ individuals.
  • Tested scenarios with equal-effect loci and loci with exponentially distributed effect sizes.
  • Compared simulation results against LE and LD analytical predictions.

3. Key Contributions

1. Identification of a Critical Migration Threshold ($N m_{crit}$): The paper derives an analytical approximation for a critical migration rate below which population structure drastically inflates genic variance. This threshold depends on the ratio of mutation to selection ($\mu/s$) and the effect size of loci.
2. Effective Parameter Framework: The authors successfully demonstrate that the complex effects of Linkage Disequilibrium (LD) in a polygenic system can be accurately captured by simple effective selection and effective migration parameters. This allows for tractable analytical predictions that match simulations across a wide parameter space.
3. Non-Monotonic Effects of Migration: The study reveals that within-deme genetic variation is maximized at intermediate migration rates (near $N m_{crit}$), rather than monotonically increasing with migration.
4. GWAS Portability and Genetic Architecture: The paper provides a theoretical basis for understanding why GWAS results may not transfer well between populations. It shows that the genetic basis of trait variation is most similar across subpopulations near the critical migration threshold, but becomes divergent at both very low and very high migration rates.

4. Key Results

A. Genic Variance and the Critical Threshold

• Below $N m_{crit}$: Different demes fix alternative alleles at high frequencies (bistability). The total genic variance ($V_T^{gen}$) across the whole population is significantly higher than in a panmictic population because different subpopulations occupy different "adaptive peaks" (different genetic combinations yielding the same phenotype).
• Above $N m_{crit}$: Migration homogenizes allele frequencies. The system behaves like a single large panmictic population, and genic variance drops to panmictic levels.
• Within-Deme Variance ($V_W^{gen}$): This is maximized when migration is just strong enough to introduce new variants but not strong enough to drive fixation ($N m \approx N m_{crit}$). At this point, within-deme variance can be 5 times higher than in a panmictic population for loci with moderate-to-strong selection.

B. Genetic vs. Genic Variance (The Role of LD)

• Negative LD: Stabilizing selection generates negative LD (alleles with opposite effects are associated), which reduces the actual genetic variance ($V$) relative to the genic variance ($V^{gen}$).
• Total Population Variance: At low migration, while $V_T^{gen}$ is high, the total genetic variance $V_T$ is often much lower than $V_T^{gen}$ (i.e., $V_T / V_T^{gen} \ll 1$). This is because different demes are fixed for different specific combinations of alleles that all produce the optimal phenotype, constraining the overall phenotypic variance.
• Within-Deme Variance: The reduction of $V_W$ relative to $V_W^{gen}$ is modest (Bulmer effect), but significant.

C. $F_{ST}$ and $Q_{ST}$

• $F_{ST}$ (Allelic): For loci under strong selection, $F_{ST}$ is higher than neutral expectations at low migration but drops below neutral expectations at intermediate/high migration ($4Nm > 1$).
• $Q_{ST}$ (Phenotypic): $Q_{ST}$ is consistently lower than neutral $F_{ST}$ due to stabilizing selection constraining trait means. The paper provides a simple analytical formula:
  $$Q_{ST} \approx \frac{1}{1 + 4Nm + 4N(V_W/V_s)}$$
  This indicates that $Q_{ST}$ is low when migration is high or when the mutation target size ($2N\mu L$) is large relative to migration.

D. Distribution of Variance Contributions (Outliers)

• Effect Size Dependence: In highly structured populations (low migration), loci with large effect sizes contribute disproportionately to total genic variance.
• Migration Impact: As migration increases, the contribution of large-effect loci drops sharply once $Nm$ exceeds their specific $N m_{crit}$.
• GWAS Portability: The genetic basis of variation (which loci are "outliers" with large variance contributions) is most similar between demes at intermediate migration rates. At very low migration, different demes rely on different large-effect loci; at very high migration, all loci contribute similarly, but the "outlier" signal is diluted. This implies that intermediate gene flow maximizes the portability of GWAS findings between populations.

5. Significance

• Theoretical Advancement: The paper bridges the gap between complex individual-based simulations and tractable analytical theory for polygenic traits in structured populations. The use of effective parameters to model LD is a powerful tool for future evolutionary genetics research.
• Empirical Relevance:
  • GWAS and Prediction: The findings explain why polygenic scores often fail to transfer across populations with different ancestries. It suggests that the "portability" of genetic predictions is non-monotonic with respect to migration history.
  • Interpretation of $Q_{ST}$: The study refines the interpretation of $Q_{ST} < F_{ST}$, showing it is a robust signature of spatially uniform stabilizing selection, but its magnitude depends on migration rates and mutation input.
  • Cryptic Variation: It highlights that structured populations harbor significant "cryptic" genetic variation (high genic variance but constrained phenotypic variance) that could be unlocked by crossing individuals from different subpopulations, with implications for conservation and breeding.
• Limitations & Future Directions: The model assumes unlinked loci and no mutation bias. The authors note that extending this to continuous space, multiple traits, and linked loci is a necessary next step to fully capture natural population dynamics.

In summary, this work demonstrates that population structure does not merely add noise to quantitative genetics; it fundamentally reshapes the maintenance of variation, the distribution of effect sizes, and the predictability of traits across populations, with a critical migration threshold acting as a pivot point for these dynamics.