Cassava stability study pinpoints broadly adapted genotype across 66 Thai environments

Cassava breeders face a persistent challenge: a variety that thrives in one field can falter in another, because the performance of a genotype depends on the environment in which it is grown. A decade-long multi-environment trial network in Thailand has now quantified how strongly this genotype × environment interaction (G×E) shapes cassava traits, and which varieties hold up across heterogeneous and climate-variable production conditions. Eight cassava genotypes were evaluated for eight agronomic and adaptive traits across 66 environments between 2011 and 2020, using linear mixed models to estimate variance components, best linear unbiased predictors (BLUPs), and entry-mean broad-sense heritability.

The analysis showed that environmental effects were significant for every trait measured. Yield- and growth-related traits exhibited substantial G×E, including pronounced crossover responses, meaning that the ranking of genotypes changed from one environment to another rather than simply rising or falling in parallel. AMMI and GGE biplot analyses, complemented by regression-based stability measures and multi-trait stability indices, were used to characterize the magnitude and structure of these interactions. Entry-mean heritability proved strongly trait-dependent, ranging from low to moderate values (H² = 0.10–0.47), with higher estimates for starch content and harvest index and lower values for yield and establishment traits.

Stability analyses consistently identified the genotype HB80 as the most broadly adapted entry, combining high fresh and dry root yield with a low magnitude of interaction across environments. This performance was associated with a relatively stable harvest index, which the researchers interpret as indicating efficient assimilate partitioning. Clone #21 showed a similar pattern, whereas the genotypes R1 and R5 displayed greater environmental responsiveness and instability, changing their relative standing more sharply as conditions shifted.

Beyond classical stability statistics, the study tested whether machine-learning models could predict trait performance from the trial data. Predictive accuracy was low to moderate across traits. Dry root yield was predicted most accurately, reaching a maximum cross-validated R² of 0.52, followed by starch content at 0.44. Germination and harvest index showed limited predictability, indicating that these traits remain difficult to forecast from the available environmental and genotypic information.

The authors argue that integrating mixed-model inference, G×E analysis, stability assessment, and machine learning provides a robust framework for cassava selection under heterogeneous and climate-variable production systems. Distinguishing genotypes that genuinely perform well everywhere from those that merely excel in favorable sites is a practical breeding priority, and the trait-specific patterns documented here offer a quantitative basis for that distinction.

Subject of Research: Trait-specific genotype × environment interaction, stability, and predictability of cassava in multi-environment trials

Article Title: Trait-specific genotype × environment interaction, stability, and predictability of cassava in multi-environment trials

Article References: Phumichai, C., Boonma, S., Aiemnaka, P., Nathaisong, P., Vichukit, V., Sarobol, E., & Rojanaridpiched, C. (2026). Trait-specific genotype × environment interaction, stability, and predictability of cassava in multi-environment trials. BMC Plant Biology. https://doi.org/10.1186/s12870-026-10227-z

Image Credits: AI Generated

DOI: 10.1186/s12870-026-10227-z

Keywords: cassava, genotype by environment interaction, multi-environment trials, stability analysis, AMMI, GGE biplot, heritability, machine learning, plant breeding, Thailand, root yield, starch content