Dissecting oligogenic and polygenic indirect genetic effects through the lens of neighbor genotypic identity

This study presents a unified multi-kernel mixed model that integrates oligogenic and polygenic indirect genetic effects to effectively dissect the genetic architecture of group performance, revealing evidence of intergenotypic competition in woody plants and identifying specific competitive variants in apple trees.

The Gist

Imagine you are walking through a crowded garden. Your growth and health aren't just determined by your own seeds and soil; they are heavily influenced by the plants growing right next to you. If your neighbor is a giant oak tree, it might steal your sunlight, stunting your growth. If your neighbor is a friendly vine, it might offer support.

This paper is about a new, super-smart way to measure exactly how much our "neighbors" affect us, not just in plants, but in any group of living things. The authors call these Indirect Genetic Effects (IGEs).

Here is a simple breakdown of what they did and why it matters, using some everyday analogies.

1. The Problem: The "Blind Spot" in Genetics

For a long time, scientists looked at genetics like a solo act. They asked, "How does your DNA determine your height?" This is called a Direct Genetic Effect.

But in reality, we are all part of a group. Your height might be limited because your neighbor is tall and blocks the sun. That's an Indirect Genetic Effect. The old math tools were like a camera with a blind spot; they could see your DNA, but they struggled to separate "your DNA" from "the effect of your neighbor's DNA." They often got confused, mixing the two up.

2. The Solution: A "Team Score" Calculator

The authors built a new mathematical model (a "multi-kernel mixed model") that acts like a sophisticated referee. It can look at a group of individuals and say:

• "Okay, 60% of this tree's size is due to its own genes."
• "20% is due to the genes of the trees next to it."
• "And 20% is the complex interaction between the two."

They used a concept from physics called the Ising Model (originally used to explain how magnets align). Think of it like this:

• Magnets: Imagine each tree is a magnet. Some magnets want to face the same way (cooperation), and some want to face opposite ways (competition).
• The New Model: This model calculates how the "magnetic pull" of the neighbors changes the outcome for the individual. It simplifies a very complex physics problem into a tool that biologists can use easily.

3. The Simulation: The "Video Game" Test

Before using real trees, the authors tested their model in a computer simulation (like a video game). They created fake populations of plants with known rules:

• Scenario A: Neighbors help each other (Positive teamwork).
• Scenario B: Neighbors fight for resources (Negative competition).

They found that their new model was excellent at spotting the difference. It could tell if the group was working together or fighting, even when the data was messy. It also showed that when neighbors are in direct competition, the "team score" (the group's overall success) might actually go down, even if the individuals are trying their best.

4. The Real-World Test: Trees and Vines

The authors took their new tool out into the real world and tested it on three types of woody plants:

• Aspen Trees (The Competitive Neighbors): They found strong evidence that aspen trees compete with their neighbors. As the trees got bigger, the competition got fiercer. It's like a crowded apartment building where, as everyone gets taller, they start blocking each other's windows more aggressively.
• Apple Trees (The Growth Struggle): Similar to aspens, apple trees showed signs of competition. The study even found specific "bad neighbor" genes on the apple tree's DNA that seemed to make the tree grow slower when surrounded by certain other trees. It's like finding a specific gene that makes you sensitive to a noisy roommate.
• Grapevines (The Chill Neighbors): Surprisingly, grapevines didn't show much competition. Why? Because grapes are climbers! They grow up trellises, not out into the space next to them. They don't fight for horizontal space the way trees do. It's like comparing a crowded subway car (trees) to people standing on a ladder (grapes); the ladder climbers don't bump into each other as much.

5. Why This Matters

This isn't just about trees. This new method is a universal tool.

• For Farmers: It helps breed better crops. Instead of just picking the "best" individual plant, farmers can pick plants that are "good neighbors," creating a field where the whole group grows bigger and healthier together.
• For Animal Breeders: It could help manage livestock. If you know certain cows are "bad neighbors" (they stress out the herd), you can separate them to improve milk production for everyone.
• For Evolution: It helps us understand how species evolve. Sometimes, the pressure to get along with neighbors drives evolution just as much as the pressure to survive on your own.

The Bottom Line

The authors created a "genetic microscope" that finally lets us see the invisible social dynamics of nature. They proved that who your neighbors are is just as important as who your parents are. By understanding these interactions, we can grow better crops, manage animals more effectively, and understand the complex social web of life.

Technical Summary

1. Problem Statement

Individual phenotypes are often influenced not only by an individual's own genotype (Direct Genetic Effects, DGEs) but also by the genotypes of neighboring individuals (Indirect Genetic Effects, IGEs). While IGEs are critical for understanding evolutionary trajectories, intraspecific competition, and group performance in breeding, existing methods face significant limitations:

• Lack of Unified Framework: Previous approaches struggled to jointly model oligogenic (locus-specific) and polygenic (genome-wide) IGEs within a single statistical framework.
• Complexity of Interactions: Detecting IGEs requires accounting for population structure and spatial dependencies, which can lead to false positives in Genome-Wide Association Studies (GWAS) and reduced accuracy in Genomic Prediction (GP).
• Unknown Genetic Architecture: The specific functional traits and genetic variants driving IGEs (e.g., competition vs. facilitation) remain elusive, particularly in sessile organisms like plants.

2. Methodology

The authors developed a flexible Multi-Kernel Mixed Model that integrates oligogenic and polygenic IGEs, leveraging the Ising model from statistical physics to simplify locus-wise interactions.

A. Model Formulation

The core of the method is a linear mixed model (Eq. 1 in the paper) that decomposes the phenotype ($y$) of a focal individual $i$ into:

1. Fixed Effects (Oligogenic):
   • DGE ($\beta_{q,D}$): The effect of the focal individual's own genotype at SNP $q$.
   • IGE ($\beta_{q,I}$): The effect of the mean allelic similarity between the focal individual and its neighbors. This term is derived from the Ising model, representing frequency-dependent selection.
   • Interaction Term ($\beta_{q,D\times I}$): An interaction between the focal genotype and the neighbor genotype mean.
2. Random Effects (Polygenic):
   • Polygenic DGE ($u_D$): Captured via a kinship matrix ($K_1$) based on individual genetic similarity.
   • Polygenic IGE ($u_I$): Captured via a kinship matrix ($K_2$) weighted by global allelic similarity between local spaces.
   • Covariance ($\rho_{DI}$): A crucial parameter allowing the model to estimate the covariance between DGE and IGE. This is modeled via asymmetric cross-covariance matrices ($K_{12}$ and $K_{21}$).

The total phenotypic variance is partitioned into components explained by DGE, IGE, their covariance, and residuals (Eq. 2).

B. Implementation

• Algorithm: The model is implemented using the RAINBOW (Reliable Association INference By Optimizing Weights) algorithm, which optimizes weights ($w_D, w_I$) for the kernels to estimate variance components.
• Software: The authors utilized the RAINBOWR and rNeighborGWAS R packages. They introduced a new covariance parameter handling to allow for negative correlations between DGE and IGE.
• GWAS Strategy: The method performs GWAS by comparing models with and without specific SNP-wise fixed effects ($\beta_{q,D}, \beta_{q,I}, \beta_{q,D\times I}$) using Likelihood Ratio Tests (LRT) on a weighted kernel.

C. Validation

• Simulations: The authors conducted extensive simulations (30 iterations) with varying DGE-IGE covariance ($\rho_{DI} = -0.6, -0.3, 0, 0.3, 0.6$) to test parameter estimation, genomic prediction accuracy ($R^2$, MAE), and GWAS power (AUC, sensitivity).
• Real Data Application: The model was applied to three woody plant datasets:
  1. Aspen (Populus tremuloides): Field garden data (WisAsp).
  2. Apple (Malus × domestica): Multi-site European REFPOP data.
  3. Grape (Vitis vinifera): INNOVINE project data (climbing woody plant).

3. Key Contributions

• Unified Modeling: First integration of oligogenic (locus-specific) and polygenic IGEs into a single multi-kernel mixed model with a single covariance parameter.
• Ising Model Integration: Successfully adapted the Ising model of ferromagnetics to quantify frequency-dependent selection and neighbor genotypic interactions in a biological context.
• Covariance Estimation: Developed a method to explicitly estimate the DGE-IGE covariance ($\rho_{DI}$), allowing researchers to detect trade-offs (negative covariance) or synergies (positive covariance) between individual and group performance.
• Open Source Tools: Provided an R package (rNeighborLMM) and GitHub repositories to facilitate the application of these complex models by the broader community.

4. Key Results

A. Simulation Performance

• Variance Partitioning: The model accurately estimated the sign and magnitude of the covariance parameter ($\rho_{DI}$), though negative covariances showed slightly higher estimation variance than positive ones.
• Genomic Prediction (GP):
  • When DGE and IGE were positively correlated, including the covariance term slightly improved prediction accuracy ($R^2$).
  • When negatively correlated, the inclusion of the covariance term did not significantly improve (and sometimes slightly worsened) prediction, likely due to challenges in ensuring the positive semi-definiteness of the weighted kernel matrix.
• GWAS Power: The model maintained moderate power (AUC > 0.6) to detect causative IGE SNPs regardless of the covariance sign, confirming its utility for association mapping.

B. Application to Real Data

• Aspen (Populus):
  • Found significant negative DGE-IGE covariance for basal area increment (growth), indicating intergenotypic competition.
  • Detected significant IGE SNPs for bud burst and size growth.
• Apple (Malus):
  • Growth traits (trunk diameter and increment) showed large IGE influence, while flowering intensity was dominated by DGE.
  • Competition Intensity: The negative covariance ($PVE_{cov}$) for trunk increment became more pronounced as trees grew larger, suggesting competition intensifies with tree size.
  • GWAS Findings: Identified two significant IGE SNPs on chromosome 7 (positions ~8.7 Mb) associated with trunk increment. These SNPs showed negative effects, implying that allelic mixtures reduce growth. The region contains a protein kinase superfamily protein and homologs of MYB33 and SUCROSE PHOSPHATE SYNTHASE 1F.
• Grape (Vitis):
  • Limited Evidence: Most traits showed low IGE influence ($w_I < 0.1$).
  • Positive Covariance: Traits like pruning weight and citric acid showed positive DGE-IGE covariance (synergy), but GWAS results were inflated (false positives).
  • Interpretation: The lack of strong competition signals in grapes is attributed to their climbing growth habit (vertical growth) and high phenotypic plasticity, reducing horizontal neighbor competition compared to aspen and apple.

5. Significance

• Breeding Applications: The study provides a robust tool for Group Performance Breeding. By quantifying the trade-off between individual growth (DGE) and competitive ability (IGE), breeders can select genotypes that optimize group yield rather than just individual performance, mitigating intraspecific competition.
• Evolutionary Insights: The ability to detect negative covariance supports the hypothesis of frequency-dependent selection and competitive exclusion in natural and managed populations.
• Methodological Advancement: By resolving the technical challenge of modeling asymmetric neighbor effects and DGE-IGE covariance, this framework bridges the gap between quantitative genetics and spatial ecology, applicable to both sessile plants and social animals.
• Candidate Genes: The identification of specific genomic regions (e.g., in apple) associated with competitive IGEs opens new avenues for functional validation of genes controlling plant-plant interactions.

In conclusion, this paper offers a flexible, statistically rigorous framework to dissect the genetic architecture of group performance, moving beyond simple additive models to capture the complex dynamics of indirect genetic effects in diverse species.