Joint modeling of social genetic effects in mono- and pluri-specific groups: case study in intercrops

This paper addresses the gap in breeding frameworks for interspecific groups by proposing and implementing a novel quantitative genetic model in R/C++ that jointly analyzes direct, intraspecific, and interspecific social genetic effects, thereby enabling accurate estimation of breeding values and simultaneous genetic gains in both sole crops and intercrops.

The Gist

Imagine you are a chef trying to create the perfect dish. For decades, you've only cooked single-ingredient meals: a bowl of pure rice, a plate of pure chicken. You've become an expert at breeding the best rice and the best chicken separately.

But now, you want to start making stir-fries (mixing rice and chicken together). You quickly realize that the rice you grew for a plain bowl doesn't taste the same when it's cooked with chicken. The chicken changes the rice's flavor, and the rice changes the chicken's texture. They influence each other.

This is exactly the problem farmers and plant breeders face with intercropping (growing two different crops, like wheat and peas, in the same field). While mixing crops is great for the environment (it uses less fertilizer and fights pests), we don't know how to breed the "perfect mix." We've been breeding plants to be the best solo performers, not the best team players.

This paper is a new recipe for solving that problem. Here is the breakdown in simple terms:

1. The Problem: The "Solo" vs. "Team" Personality

In traditional farming, breeders pick the "star players"—the wheat plants that grow tallest and produce the most grain when they are all identical clones growing together.

But in a mixed field (intercrop), a plant's success depends on two things:

• Its own genes (The Solo Skill): How well does this specific wheat plant grow on its own?
• Its neighbors' genes (The Team Skill): How does this wheat plant react to the peas next to it? Does it get jealous and stop growing? Or does it help the peas, which in turn helps the wheat?

The paper argues that we can't just look at the "Solo Skill." We need to measure the "Team Skill" (called Social Genetic Effects) to breed better crops for mixed fields.

2. The Solution: A New "Scorecard"

The authors created a new mathematical model (a fancy scorecard) that breaks a plant's performance down into three parts:

• The Direct Value (DBV): The plant's natural talent. (e.g., "This wheat is naturally tall.")
• The Social Value (SBV): How much this plant helps or hinders its neighbors. (e.g., "This wheat is great at shading out weeds, which helps the peas.")
• The "Intra-Group" Value (SIGV): This is a tricky new concept. It accounts for how a plant behaves when it's surrounded by its own kind (like a wheat field) versus how it behaves in a mix. It's like realizing a person might be a quiet leader in a small group of friends but a loud show-off in a large crowd.

The Analogy: Think of a basketball player.

• Direct Value: How many points they score.
• Social Value: How well they pass the ball to teammates.
• The Paper's Insight: You can't just look at the points. If you want to win a game (intercrop), you need a player who scores and passes well. But if you only train for points (sole crop), you might accidentally train a player who hoards the ball, which ruins the team game.

3. The Experiment: The "Tasting Menu"

The researchers ran computer simulations to test different ways of testing these plants. They compared three "menus":

1. The Solo Menu: Testing only pure wheat fields.
   • Result: Great at finding the best solo wheat, but terrible at predicting how that wheat will do in a mix.
2. The Mix-Only Menu: Testing only mixed wheat/pea fields.
   • Result: Great at finding the best mix, but you lose information about the plant's base talent.
3. The Hybrid Menu (The Winner): Testing 50% of the plants in solo fields and 50% in mixed fields.
   • Result: This was the magic combination. By seeing the plants in both environments, the model could figure out the "Direct Value" and the "Social Value" separately.

4. The Big Takeaway: You Don't Have to Choose

For a long time, breeders thought they had to choose: "Do we breed for solo crops OR mixed crops?"

This paper says: No. You can do both at the same time.

By using this new "Hybrid Menu" and the new math model, breeders can:

• Keep breeding high-yield crops for traditional farms (which still dominate the market).
• Simultaneously breed crops that are "team players" for intercropping.
• Do this without needing to test every single possible combination of plants (which would be impossible and too expensive).

Why This Matters

Climate change is making weather more extreme. Mixing crops (intercropping) is a natural way to make farms more resilient and sustainable. But we can't switch to this system if we don't have the right seeds.

This paper provides the blueprint for seed companies to start breeding "team-player" crops without abandoning their current "solo-player" breeding programs. It's like teaching your athletes to be great soloists and great band members at the same time, ensuring that no matter what stage they play on, they will succeed.

Technical Summary

1. Problem Statement

Modern crop breeding has historically focused on selecting genotypes for sole cropping (SC) systems (monospecific, monogenotypic stands). However, intercropping (IC) (simultaneous cultivation of multiple species) offers significant agroecological benefits, such as improved resource use efficiency and disease resistance.

• The Gap: There is a lack of breeding frameworks that integrate the genetics of social interactions (Indirect Genetic Effects, or IGEs) across both SC and IC systems.
• The Challenge: Phenotypic performance in IC is often poorly correlated with performance in SC due to G×G interactions (genotype-by-genotype interactions) and phenotypic plasticity induced by the presence of a second species.
• The Limitation: Existing models often treat SC and IC as independent breeding programs or fail to decompose the specific genetic components (direct vs. social effects) that drive performance in mixed stands. Furthermore, no open-source software existed to fit the complex mixed models required to jointly analyze these systems.

2. Methodology

A. Genetic Model Decomposition

The authors propose a unified quantitative genetic model that decomposes the breeding value of a focal genotype ($i$) into three distinct components:

1. Direct Breeding Value (DBV): The additive effect of the genotype's own genes on its own phenotype. This is shared across both SC and IC.
2. Social Breeding Value (SBV): The effect of a genotype's genes on the phenotype of its neighbors.
   • In IC: Interspecific social effects (e.g., Wheat genotype $i$ affecting Pea genotype $j$).
   • In SC: Intraspecific social effects (e.g., Wheat genotype $i$ affecting other Wheat plants).
3. Social Intra-genotypic Value (SIGV): A novel component introduced to bridge SC and IC. Since SC data alone cannot separate the direct effect from the social effect of identical neighbors, the authors define $SIGV_i = SBV_{i|SC} + (DBV \times SBV)_{ii}$.
   • This allows the classical SC breeding value to be expressed as: $BV_{SC} = DBV + SIGV$.
   • The IC breeding value is expressed as: $BV_{IC} = DBV + SBV_{IC}$.

B. Statistical Framework

• Joint Linear Mixed Models (LMM): The study defines three simultaneous LMMs to fit data from:
  1. Sole cropping only.
  2. Intercropping only.
  3. A combined design (Sole + Intercrop).
• Random Effects: The model treats DBV, SBV, and interaction terms (DBV×SBV) as random effects following matrix-variate normal distributions, incorporating genomic relationship matrices (GRM).
• Software Implementation: Since standard R packages (e.g., lme4, ASReml) could not handle the specific correlation structures and response variable formats (vectors for SC, matrices for IC), the authors developed a custom implementation using R/C++ and the Template Model Builder (TMB) package. This leverages Automatic Differentiation (AD) and Laplace Approximation for efficient Restricted Maximum Likelihood (REML) estimation.

C. Simulation Design

• Scenario: A focal species (e.g., Wheat) and a tester species (e.g., Pea).
• Experimental Designs: Three designs were compared, each with 400 micro-plots:
  1. sole_only: All plots in sole cropping.
  2. inter_only: All plots in intercropping (sparse design).
  3. sole_inter_50: A joint design where 50% of genotypes are tested in sole cropping and 50% in intercropping (sparse design).
• Parameters: Simulations varied the variance of SIGV (10% to 50% of DBV variance) and the correlation between DBV and SBV (set to -0.8 based on empirical data).
• Evaluation: Accuracy was assessed via Normalized Bias Error (NBE) for variance parameters and Pearson correlation for breeding values. Selection accuracy was measured by the True Positive Rate (TPR) of selecting top genotypes.

3. Key Contributions

1. Unified Genetic Framework: The paper introduces a theoretical decomposition of breeding values that explicitly links SC and IC through the SIGV concept. This resolves the identifiability issue of separating direct and social effects in monovarietal stands.
2. Novel Statistical Implementation: The development of a custom R/C++ tool using TMB to fit complex, correlated random effects models for joint SC/IC analysis, filling a gap in open-source software for social genetic effects.
3. Optimal Experimental Design: The study demonstrates that incomplete, balanced designs (specifically the sole_inter_50 design) are sufficient to estimate all genetic (co)variances without the combinatorial explosion of full factorial designs.
4. Selection Strategy: The proposal of a weighted selection index ($Idx$) that allows breeders to gradually shift focus from sole cropping to intercropping by adjusting the weight ($w_{sole}$) of the selection criteria.

4. Key Results

• Parameter Estimation Accuracy:
  • The sole_inter_50 design was the only design capable of unbiasedly estimating all genetic parameters, including the elusive SIGV variance.
  • The sole_only design led to significant overestimation of IC breeding values and could not estimate social effects.
  • The inter_only design provided unbiased estimates for DBV and SBV but slightly underestimated the variance of SC breeding values.
• Breeding Value Accuracy:
  • DBV and SBV: Both sole_inter_50 and inter_only designs provided high accuracy for estimating DBV and SBV, significantly outperforming sole_only.
  • SIGV: Only the joint design allowed for the estimation of SIGV.
  • Genomic Relationships: Incorporating a Genomic Relationship Matrix (GRM) significantly improved the accuracy of social breeding value estimates (SBV and SIGV).
• Selection Efficiency:
  • Sole Cropping Target: sole_only performed best, but sole_inter_50 was nearly as effective, even as SIGV variance increased.
  • Intercropping Target: inter_only was optimal, but sole_inter_50 maintained strong performance.
  • Crucial Finding: At low SIGV variance, the inter_only design actually selected better sole-crop genotypes than the sole_only design because it provided more accurate estimates of the underlying DBV.
  • Joint Selection: The sole_inter_50 design enabled simultaneous genetic gain in both systems, proving that a single breeding program can serve both markets.

5. Significance and Implications

• Practical Breeding Strategy: The study provides a pragmatic pathway for breeders to transition from sole-crop-only programs to intercrop-inclusive programs without abandoning existing infrastructure. It validates the use of incomplete designs to reduce costs while maintaining statistical power.
• Economic Viability: By showing that joint analysis yields accurate estimates, the paper argues against the need for completely separate breeding programs for SC and IC, which would be prohibitively expensive.
• Broader Applicability: While focused on crops, the framework is applicable to any system involving social genetic effects, including animal breeding (e.g., group housing) and forest tree breeding.
• Agroecological Transition: The work supports the integration of intercropping into modern agriculture by providing the genetic tools necessary to breed varieties specifically adapted to mixed-species environments, addressing a major bottleneck in the adoption of sustainable farming practices.

In conclusion, Salomon et al. successfully demonstrate that by mathematically unifying direct and social genetic effects, breeders can efficiently select for both sole and intercropping systems simultaneously, maximizing genetic gain while minimizing experimental costs.