📄 evolutionary biology

LAVA: a method for identifying local and global adaptation in structured populations

The paper introduces LAVA, a Bayesian R-based method that improves the detection of local and global adaptation in structured populations by comparing ancestral additive genetic variance estimates to overcome the high false positive rates of traditional Qst-Fst comparisons caused by unequal subpopulation relatedness.

Original authors: do O, I., Bachmann Salvy, M., Gaggiotti, O. E., Goudet, J., de Villemereuil, P.

Published 2026-04-16

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: do O, I., Bachmann Salvy, M., Gaggiotti, O. E., Goudet, J., de Villemereuil, P.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine you are a detective trying to solve a mystery: Why do two groups of the same animal look different?

Maybe one group of fish has bigger fins than the other. Is it because they evolved to swim faster in their specific river (a sign of adaptation)? Or is it just because they got separated for a long time, and random chance made them different (a sign of genetic drift)?

For a long time, scientists had a standard tool to answer this, but it was like using a map of a flat, featureless plain to navigate a mountain range. It worked okay in simple cases, but in the real, messy world, it often gave false alarms.

This paper introduces a new, smarter tool called LAVA. Here is the story of what it does, explained simply.

The Old Problem: The "Flat Map" Mistake

The old method (called QST-FST) assumed that all groups of animals were equally related to each other, like neighbors in a perfect circle where everyone is the same distance from everyone else.

The Analogy: Imagine a classroom where the teacher assumes every student is equally close to every other student.

Reality: In a real classroom, some students sit in tight little cliques (close relatives), some are isolated in the back, and some are best friends with people in other rows.
The Mistake: If you assume everyone is equally close, you might think a student is acting differently because they are "special" (adaptation), when really they just happen to sit next to a group of friends who all look alike (drift). The old method often confused these two, leading to false accusations of "specialness."

The New Solution: LAVA

The authors created LAVA (which stands for Log Ancestral Variance Analysis, though you can just think of it as the "Smart Detective").

Instead of assuming everyone is equally related, LAVA looks at the actual family tree and the history of the group. It uses a sophisticated computer model to ask: "Given how these groups are actually related, how much difference should we expect just by random chance?"

The Core Idea:
LAVA compares two different ways of measuring the "genetic potential" of the ancestors:

The "Between" View: How different are the groups from each other?
The "Within" View: How much variation exists inside each group?

Under normal, random circumstances (neutrality), these two views should balance out perfectly. If they don't balance, it's a sign that something else is going on.

Two Types of "Specialness"

LAVA can detect two different scenarios:

Local Adaptation (The "Specialist" Case):
- What it is: Groups are becoming more different than random chance would predict because they are adapting to different local environments (e.g., one group gets cold, the other gets hot).
- LAVA's Signal: The "Between" differences are huge. The groups are pulling apart.
Global Adaptation (The "Uniform" Case):
- What it is: All groups are facing the same pressure (e.g., a new predator everywhere), so they are becoming more similar than random chance would predict.
- LAVA's Signal: The "Between" differences are tiny. The groups are converging.

The Secret Weapon: Adding the Environment

The coolest part of LAVA is that it can take environmental data (like temperature, rainfall, or soil type) and plug it directly into the math.

The Analogy:

Old Method: You see the fish have big fins. You guess, "Maybe it's the water temperature?" but you can't prove it.
LAVA: You feed the computer the water temperature data. It says, "I see that in the hot water, the fins are big. Even after accounting for their family history, this link is strong. It's not just luck; it's the heat!"

This allows scientists to test specific hypotheses: "Is this trait changing because of the rain, or the soil, or the predators?"

Why Does This Matter?

The paper tested LAVA against the old methods using thousands of computer simulations (like running a video game with different rules).

The Result: The old methods were like a smoke detector that goes off every time you cook toast (false alarms) in complex, real-world scenarios.
LAVA: It stays quiet when it's just toast (random drift) and only screams when there's a real fire (actual selection). It is more accurate and doesn't get confused by complex family trees or isolation.

In a Nutshell

LAVA is a new, flexible, and highly accurate tool for evolutionary biologists. It stops them from blaming "nature" for differences that are actually just "family history," and it helps them pinpoint exactly which environmental factors are driving animals to evolve in specific ways. It's the upgrade from a blurry, old map to a high-definition GPS for understanding how life adapts.

1. Problem Statement

Demonstrating that local adaptation drives phenotypic divergence in quantitative traits requires distinguishing selection from neutral genetic drift. The classical approach, $Q_{ST} - F_{ST}$ comparison, relies on the assumption of isotropic population structure (i.e., that all subpopulations are equally related).

The Flaw: Real-world metapopulations often exhibit complex demographic histories, non-uniform gene flow, and hierarchical structures (e.g., isolation-by-distance, stepping-stone models). In these scenarios, the assumption of equal relatedness is violated.
The Consequence: When population structure is non-uniform, $Q_{ST} - F_{ST}$ methods suffer from inflated false positive rates, incorrectly inferring selection where only neutral drift is present.
Existing Alternatives: Methods like Driftsel attempt to address this using coancestry matrices but still rely on specific assumptions (e.g., the admixture $F$ -model) that can produce biased estimates and remain sensitive to complex hierarchical structures.

2. Methodology: LAVA

The authors present LAVA (LogAV), an R implementation of the LogAV test, designed to detect signatures of local and global adaptation in spatially structured populations without assuming isotropic structure.

Core Concept

LAVA compares two distinct estimates of ancestral additive genetic variance ( $V_A$ ):

$V_{A,B}$ : Derived from between-population divergence.
$V_{A,W}$ : Derived from within-population variation.

Under neutrality, both estimates should reflect the same ancestral variance ( $V_{A,B} = V_{A,W}$ ). Significant deviations indicate selection:

Local Adaptation: $V_{A,B} > V_{A,W}$ (Phenotypic divergence exceeds neutral expectations).
Global Adaptation: $V_{A,B} < V_{A,W}$ (Phenotypic similarity exceeds neutral expectations due to uniform selection).

Statistical Framework

Model: LAVA uses a Bayesian linear mixed-effects model (implemented via the brms R package).
Equation: $z = \mu + X\beta + a_p + a_i + u + \epsilon$ $z = μ + X β + a_{p} + a_{i} + u + ϵ$
- $a_p$ : Population-level random effect (modeled via a population-level relatedness matrix $\Theta_p$ ).
- $a_i$ : Individual-level random effect (modeled via an individual-level relatedness matrix $M$ ).
- $X\beta$ : Fixed effects, which can include environmental covariates (e.g., temperature, precipitation) to test specific ecological drivers.
Relatedness Matrices:
- $\Theta_p$ and $M$ are estimated using the allele-sharing method of moments based on neutral genetic markers (SNPs/microsatellites) from parental and offspring generations.
- This allows the model to capture complex, non-uniform genealogical relationships without assuming equal relatedness.
Test Statistic: The method calculates the log-ratio of the variance estimates:
$\text{LogAV} = \log\left(\frac{V_{A,B}}{V_{A,W}}\right)$
A Bayesian $p$ -value is derived from the posterior distribution of this log-ratio.

Data Requirements

Parental Genetic Data: Neutral markers from wild populations to estimate $\Theta_p$ .
Offspring Phenotypic Data: Quantitative traits measured in a common garden (to control for plasticity).
Offspring Genetic/Pedigree Data: To estimate individual relatedness ( $M$ ).

3. Key Contributions

Robustness to Structure: LAVA is the first method shown to maintain proper calibration (correct Type I error rates) across diverse population structures, including Stepping-Stone and Hierarchical models where $Q_{ST} - F_{ST}$ and Driftsel fail.
Environmental Integration: Unlike traditional $Q_{ST}$ methods, LAVA can directly incorporate environmental covariates into the mixed model. This allows researchers to test specific hypotheses about ecological drivers of divergence while simultaneously controlling for population structure.
Flexibility: The framework supports various breeding designs (e.g., North Carolina Design II), non-Gaussian traits (via GLMMs), and complex experimental blocks.

4. Results

The authors evaluated LAVA against $Q_{ST} - F_{ST}$ and Driftsel using extensive simulations (500 replicates per scenario) across three population structures: Island Model, Stepping-Stones, and Hierarchical.

Calibration (False Positive Rates):
- Island Model: All three methods were well-calibrated.
- Stepping-Stones & Hierarchical: $Q_{ST} - F_{ST}$ showed systematic miscalibration (inflated false positives). Driftsel was inconsistent and sensitive to structure. LAVA remained perfectly calibrated in all scenarios, regardless of selection strength or environmental correlation patterns.
Statistical Power:
- In uniform structures (Island), all methods performed similarly.
- In complex structures, LAVA generally maintained or exceeded the power of other methods.
- Environmental Covariates: Incorporating environmental data into LAVA ("LAVA w/ environment") significantly increased power to detect selection, often outperforming standard approaches even when environmental gradients were correlated with population structure.
Sampling Design: Increasing the number of sampled subpopulations improved power for all methods. However, LAVA was more robust to reduced sampling designs (e.g., fewer families per population) in complex structures compared to $Q_{ST} - F_{ST}$ .

5. Significance

Reliability in Real-World Systems: Natural populations rarely fit the idealized Island Model. LAVA provides a statistically rigorous tool for detecting adaptation in systems with isolation-by-distance, hierarchical colonization, or varying effective population sizes, preventing the misinterpretation of neutral drift as selection.
Ecological Insight: By integrating environmental variables directly into the selection test, LAVA moves beyond simply detecting that adaptation occurred to helping identify what environmental factors drove it, all while correcting for the confounding effects of demographic history.
Practical Implementation: As an R package, LAVA offers a user-friendly, flexible, and computationally efficient solution for evolutionary biologists and conservation geneticists working with common garden experiments and genomic data.

In conclusion, LAVA represents a significant methodological advancement by resolving the calibration issues of previous methods in structured populations, offering a unified framework to disentangle neutral drift from local and global adaptation.