🧬 genetics

Leveraging synthetic environmental covariates to explore crop genotype by environment interactions

This study extends Finlay-Wilkinson regression using synthetic environmental covariates to dissect genotype-by-environment interactions at individual developmental windows in barley, revealing that early grain fill drives yield variance and introducing growth-stage-specific stability as a novel breeding target.

Original authors: Brunner, S., Piepho, H.-P., Chenu, K., Heidariask, B., Moody, D., Baraibar, S., Ganesalingam, D., Hickey, L., Voss-Fels, K., Robinson, H. M.

Published 2026-10-16

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Brunner, S., Piepho, H.-P., Chenu, K., Heidariask, B., Moody, D., Baraibar, S., Ganesalingam, D., Hickey, L., Voss-Fels, K., Robinson, H. M.

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 ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed.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

The weather is the great variable in farming. A farmer can prepare the soil, choose the seed, and tend the crop, but they cannot control the rain, the heat, or the wind. For decades, plant breeders have tried to solve the puzzle of how different plant varieties respond to these shifting conditions. They look for plants that perform well everywhere, a trait known as stability. However, the environment is not a single, static thing; it changes hour by hour and day by day. A plant might thrive in a cool spring but struggle in a hot summer, or vice versa. This creates a complex interaction between the plant's genetic code and the specific conditions it faces, a phenomenon scientists call genotype-by-environment interaction. Understanding this interaction is critical because as the climate becomes more unpredictable, the ability to predict which crops will succeed in which fields becomes the difference between a harvest and a failure.

A team of researchers, led by Stephanie Brunner and colleagues, has developed a new way to look at this problem. Instead of treating a growing season as one long, undifferentiated block of time, they broke it down into specific developmental windows, much like looking at a movie frame by frame rather than just the final scene. They studied 961 different barley varieties tested across twelve different field trials in Australia over four years. These trials covered a wide range of conditions, with average yields varying from 2.5 tons per hectare in one location to 5.5 tons per hectare in another. The researchers wanted to know not just which plants were the most stable overall, but exactly when during the growing season the weather mattered most for yield, and how different plants reacted to those specific moments.

To do this, the team created a statistical pipeline that acts like a filter for environmental data. They gathered detailed weather information for each trial, including rainfall, temperature, humidity, and soil moisture. Instead of trying to analyze every single day of weather separately, which would be too messy to interpret, they grouped the weather data into six distinct periods based on the plant's life cycle: emergence, development, flowering, early grain fill, late grain fill, and maturity. They then used a mathematical method to combine the various weather factors within each period into a single, simplified score. This allowed them to see how the plants reacted to the "weather story" of each specific stage of growth.

The results revealed a clear hierarchy of importance. The researchers found that the environmental conditions during the early grain fill stage contributed the most to the differences in yield between the varieties. This makes biological sense, as this is the period when the plant is actively filling the grain, and stress here directly reduces the final weight of the crop. Surprisingly, the conditions at the very end of the season, during maturity, also played a significant role, largely due to rainfall. The team noted that heavy rain at this late stage can delay harvest or cause the crop to fall over, known as lodging, which reduces yield. Interestingly, the study showed that the specific weather factors driving these differences changed depending on the growth stage. For example, while minimum temperatures (nighttime lows) were a major driver of variation during most of the season, radiation (sunlight) was a key factor during the early grain fill.

Perhaps the most significant finding was that stability is not a single, fixed trait. A plant that is stable across all environments might not be stable at every stage of its life. The researchers discovered that some varieties were highly unstable during the early growth stages but remarkably stable during the grain fill, while others showed the opposite pattern. One variety, RGT Planet, was unstable when the plants were young but became one of the most stable performers during the critical early grain fill and maturity stages. This suggests that breeders could stop looking for a "perfectly stable" plant and instead target stability for specific, high-impact windows. If the early grain fill is the most critical time for yield, as this study suggests, breeders could select for plants that are specifically robust during that window, even if they are less stable at other times.

The study also looked at the genetic building blocks behind these traits, examining small chunks of DNA called haplotypes. They found that the stability of these genetic blocks varied across different environments and growth stages. In some cases, the genetic variants that produced the highest yields were also the most unstable, creating a trade-off that breeders must navigate. However, they also found instances where high-performing genetic blocks were also stable, offering a clear target for future selection. By mapping these interactions, the researchers showed that it is possible to identify exactly which parts of the genome are responsible for stability during specific environmental stresses.

This approach offers a new lens for understanding crop performance. Rather than treating the growing season as a black box where inputs go in and a yield comes out, the researchers have shown that the season is a series of distinct chapters, each with its own environmental demands and genetic responses. The method they used allows scientists to dissect the complex web of interactions between plants and their environment, revealing that the "best" plant depends entirely on the specific conditions it faces and the timing of those conditions. While the study was limited to barley and a specific set of Australian trials, the framework they developed can be applied to other crops and regions. It provides a tool for breeders to move beyond simple rankings and start designing crops that are resilient to the specific, shifting challenges of a changing climate, ensuring that the right plant is matched to the right time and place.

Technical Summary: Leveraging Synthetic Environmental Covariates to Explore Crop Genotype by Environment Interactions

Problem Statement
Plant breeders face significant challenges in predicting genotype performance due to the unpredictability of environmental conditions and the complexity of Genotype by Environment Interactions (GEI). While existing statistical approaches, such as Factor Analytic (FA) models, effectively capture GEI variance for prediction, they often operate as "black boxes" with latent factors that lack direct interpretability regarding specific environmental drivers. Conversely, reaction norm approaches using single environmental axes offer interpretability but struggle to scale to multiple covariates without becoming computationally unfeasible. Furthermore, traditional stability selection often sacrifices yield potential, and current methods rarely dissect GEI at the resolution of individual developmental windows, obscuring which specific growth stages drive yield instability. There is a need for a scalable, interpretable framework that links GEI to specific environmental covariates across the crop's phenological timeline.

Methodology
The authors developed a pipeline extending the Finlay–Wilkinson regression using synthetic environmental covariates, based on the framework by Piepho and Blancon (2023). The study utilized a barley breeding dataset comprising 961 genotypes tested across 12 field trials (six locations, four years) in Australia.

The methodology proceeded in four key steps:

Data Preparation: Phenotypic yield data was combined with genomic data (6,599 SNPs) to generate a Genomic Relationship Matrix (GRM). Environmental data (rainfall, temperature, VPD, radiation, day length, soil moisture) were extracted from SILO datasets. Phenology was aligned using flowering time (Spartacus CL) as a reference, with missing data simulated via the APSIM-Barley model.
Defining Developmental Windows: The growing season was divided into 11 thermal time windows (spanning -1100°Cd to +1000°Cd relative to flowering). An initial "Environmental Covariate Model" (an extended Finlay–Wilkinson regression) was fitted to each window to derive factor loadings ( $\lambda$ ). Windows with positively correlated loadings were grouped, resulting in six distinct developmental windows: emergence, development, flowering, early grain fill, late grain fill, and maturity.
Deriving Synthetic Covariates: For each of the six grouped windows, a linear combination of the seven environmental variables was derived to create a single "synthetic environmental covariate" ( $z_j$ ) per window. These covariates were weighted by their contribution to GEI variance.
Synthetic Covariate Model: A final mixed linear model was fitted incorporating the six synthetic covariates. This model included fixed genotype effects, genotype-specific slopes ( $\beta_{ih}$ ) for each synthetic covariate (representing sensitivity to environmental conditions at specific growth stages), and a residual GEI term. The model utilized an unstructured variance-covariance matrix constrained by the GRM for the genotype-by-covariate interactions.

Key Contributions

Developmental Resolution of GEI: The approach dissects GEI not just across environments, but across specific developmental windows, identifying which growth stages contribute most to yield instability.
Interpretability: By using synthetic covariates derived from observed environmental variables, the model retains the interpretability of genotype-specific sensitivity parameters (analogous to Finlay–Wilkinson slopes) while accommodating high-dimensional environmental data without latent confounding.
Stage-Specific Stability Metrics: The pipeline enables the calculation of yield stability metrics specific to individual growth stages, rather than a single aggregate measure across the entire season.
Haplotype-Level Analysis: The method was extended to analyze local genomic estimated breeding values (Local GEBV), allowing for the identification of haplotypes that exhibit stability or instability at specific developmental stages.

Results

Drivers of GEI: In the case study, environmental conditions during early grain fill contributed the most to the total GEI variance for barley yield, followed by maturity. The development stage contributed the least.
Environmental Drivers: Within the early grain fill window, all seven environmental covariates contributed to GEI variance. At maturity, rainfall was the dominant driver (accounting for ~47% of the variance attributed to that stage), likely reflecting management limitations (e.g., harvest delays) rather than direct physiological stress. Minimum temperature (night temperature) generally contributed more to GEI variance than maximum temperature across most stages.
Genotypic Stability: Stability was found to be dynamic. Varieties could be stable across environments at one growth stage but unstable at another. For example, variety RGT Planet showed high instability at early growth stages but high stability during early grain fill and maturity. Conversely, Oxford was generally unstable but showed high stability at late grain fill.
Haplotype Dynamics: Haplotype analysis revealed that the most beneficial haplotypes for yield were not always the most stable. In some genomic blocks (e.g., b001604), the top-performing haplotypes in one environment were the most unstable, while in others, high performance correlated with stability. This highlights trade-offs between effect magnitude and stability at specific chromosomal segments.
Model Performance: The synthetic covariate component captured approximately 99.1% of the GEI variance relative to the residual GEI term, demonstrating the model's capacity to explain genotype-environment interactions through environmental covariates.

Significance and Claims
The paper claims that this approach provides a "new, growth-stage-resolved measure of yield stability" and a means to dissect GEI in complex traits that is more biologically interpretable than purely latent models. By identifying that yield stability varies across development, the authors suggest that breeders can target stability specifically within the developmental windows that drive the most GEI variance (e.g., early grain fill), rather than selecting for broad, season-long stability which may compromise yield potential.

The authors position the method as a tool to move beyond "black box" predictions, offering a pathway to:

Identify target regions for breeders to improve stability or specificity at specific growth stages.
Facilitate environmental clustering based on stage-specific GEI patterns rather than just geographic or overall productivity metrics.
Potentially predict performance in new environments (provided covariate data is available) by leveraging the explicit link between model parameters and observable environmental variables.

The study concludes that while it does not replace mechanistic crop models, it offers a scalable statistical framework to bridge the gap between quantitative genetics and environmental characterization, enabling more strategic selection decisions in breeding programs.

More like this