Introduction
Soybean is an important grain legume cultivated worldwide for food, feed and industrial applications. Waterlogging in soybean during the vegetative stage reduces yield by approximately 17–43% (Adegoye et al. Reference Adegoye, Olorunwa, Alsajri, Walne, Wijewandana, Kethireddy, Reddy and Reddy2023; Ramlal et al. Reference Ramlal, Lal and Vidyasagar2023; Rajendran et al. Reference Rajendran, Ramlal, Sarkar, Agasti, Rajarajan, Lal, Raju and Subramaniam2024, Reference Rajendran, Ramlal, Harika, Subramaniam, Raju and Lal2025). Multi-environment testing (MET) is essential for characterizing germplasm and selecting broad- and specifically adapted genotypes. Most studies on soybean flooding tolerance focus on the late vegetative or reproductive stages, with limited attention to pre-germination anaerobic stress under field conditions (Sayama et al. Reference Sayama, Nakazaki, Ishikawa, Yagasaki, Yamada, Hirota, Hirata, Yoshikawa, Saito, Teraishi and Okumoto2009; Yu et al. Reference Yu, Chang, Lv, Sharmin, Wang, Kong, Bhat and Zhao2019; Sharmin et al. Reference Sharmin, Karikari, Chang, Al Amin, Bhuiyan, Hina, Lv, Chunting, Begum and Zhao2021). Also, earlier germination-stage screenings were largely conducted in single environments or controlled conditions, restricting the assessment of genotype stability and adaptation in a balanced trial (Rajendran et al. Reference Rajendran, Lal, Raju and Ramlal2022, Reference Rajendran, Lal, Raju, Mallikarjun, Ramlal and Sharma2023). Although GEI has been extensively analyzed for yield traits using AMMI and GGE models (Yan et al. Reference Yan, Hunt, Sheng and Szlavnics2000; Yan and Rajcan Reference Yan and Rajcan2002; Nataraj et al. Reference Nataraj, Pandey, Ramteke, Verghese, Reddy, Onkarappa, Mehtre, Gupta, Satpute, Mohan, Shivakumar, Chandra and Rajesh2021), its influence on pre-germination anaerobic tolerance across contrasting agro-ecologies remains poorly understood. A genetically diverse soybean germplasm panel, comprising predominantly exotic accessions from the national germplasm repository, including landrace-derived material, across multiple environments, was used in the study to characterize natural variation for pre-germination anaerobic tolerance. This study aimed to evaluate soybean genotypes across four diverse locations to identify and rank tolerant and stable accessions suitable for further breeding.
Materials and methods
Biophysical inherent seed traits like electrical conductivity (
${{\mu S c}}{{\text{m}}^{ - 1}}{{\text{g}}^{ - 1}}$), hundred-seed weight, delay in germination (days) and germination percentage (GP) were computed from 2018 to 2023 to shortlist the genotypes for next-level field screening for transient waterlogging (Rajendran et al. Reference Rajendran, Lal, Raju and Ramlal2022) (Table S1). Fifty soybean genotypes (G1–G50) along with the check variety DS9712 were shortlisted and evaluated for pre-germination anaerobic stress tolerance, performed in both 2022 and 2023 seasons at four major soybean-growing regions of India: Umiam (Meghalaya), New Delhi (Delhi), Kota (Rajasthan) and Dharwad (Karnataka) (Figure S1; Table 1; Table S1). The passport information was retrieved from http://pgrinformatics.nbpgr.ernet.in/akmu/Default.aspx. These locations represent the north-eastern hill, northern plains, central and southern agro-ecological zones of soybean cultivation. Trials were conducted during locally recommended on-season and off-season sowing seasons, yielding eight locations × season environments. Experiments followed a randomized complete block design with three replications. For stress treatment, plots were flooded with 3–5 cm of standing water for nine consecutive days immediately after sowing to simulate pre-germination waterlogging. Control plots were maintained under well-drained conditions. Tolerance was assessed in terms of GP recorded after removing waterlogging for nine days, continuing for an additional 20 days.
Agro-climatic characteristics of the testing environments
Table 1 Long description
Four major soybean-growing regions of India: Umiam (Meghalaya), New Delhi (Delhi), Kota (Rajasthan) and Dharwad (Karnataka) with their geographical attributes (latitude, longitude and altitude) with soil type and maximum and minimum temperatures.
* End date is the observation of 2 mm emergence or the 29th day (9 days of waterlogging and 20 days of recovery).
The dataset retained genotype (G), location/environment (L), year/season (Y) and replication (R). Normality and homogeneity were tested using Shapiro–Wilk and Levene’s tests. As heteroscedasticity was detected, Kruskal–Wallis tests were used to validate main effects. A linear mixed model fitted by restricted maximum likelihood (REML) was used to estimate variance components. Genotype and location/environment were treated as fixed effects, while year and its interactions were random. Variance components for G, Y, G × Y, L × Y, G × L × Y and residuals were estimated. Combined ANOVA partitioned total variation into genotype (G), environment (E = location/environment × year) and G × E components. AMMI analysis decomposed the interaction into principal component axes (IPCA). GGE biplots were constructed to visualize ‘which-won-where’ patterns, mean vs stability relationships and environmental representativeness. Stability parameters were estimated using Eberhart–Russell and Finlay–Wilkinson regression models (bi, S2di), Shukla’s stability variance and Lin–Binns superiority index. Best linear unbiased predictions (BLUPs) were obtained, and the weighted average of absolute scores from BLUPs (WAASB) and Rank Summation Index (RSI) were calculated. Analyses were performed in R, SPSS and ViTSel (2020). The statistical analysis was carried out using R (Supplementary Material).
Results
Based on the laboratory-based experiment, the genotypes EC471920, EC471972 and EC472119 are the most promising donor genotypes for breeding soybean adapted to waterlogging-prone environments (Table S1). Significant genetic variability was observed for GP under pre-germination waterlogging stress. Levene’s test indicated heterogeneity of variances (p < 0.05), and the Kruskal–Wallis test confirmed highly significant genotypic differences (p < 0.001), whereas location main effects were comparatively small (Table S2, S3). Mixed-model analysis showed that genotypic variance accounted for 73.99% of total phenotypic variation, while G × E interaction contributed 11.97%. Residual variance was low (1.4%), indicating high experimental precision. The mixed-model analysis confirmed highly significant genotypic and genotype × environment variance components, indicating substantial genetic variability and differential genotype responses across environments. AMMI analysis revealed highly significant G, E and G × E effects. IPCA1 and IPCA2 explained 86.0% and 13.9% of interaction variance, respectively, and together accounted for 99.9%, indicating that the first two components best captured the interaction structure. GGE biplot analysis identified three vertex genotypes: EC471972, EC472119 and EC471920 (Fig. 1). The GGE biplot analysis, grounded in the methodology proposed by Yan and Kang (Reference Yan and Kang2003), demonstrates that EC471920 possesses superior mean performance with specific stability, EC472119 offers stability with moderate performance, and K.2023 (Kota location 2023) serves as a highly discriminating environment. The high explanatory power of the first two principal components (PC) strengthens the reliability of these interpretations (Table S4; Fig. 1).
GGE biplot analysis. The first two principal components (PC1 = 95.38% and PC2 = 3.78%) together accounted for approximately 99% of the total G + GE variation, indicating excellent goodness of fit for the two-dimensional biplot for visual interpretation. The green line represents the environment, the blue dot represents genotypes, and the concentric semicircles represent the ranking of genotypes based on their distance from the ‘ideal genotype’. Each semicircle is a ranking scale. The first ring contains genotypes closest to the ideal. Outer rings represent progressively lower desirability.
Figure 1 Long description
GGE biplot analysis. The first two principal components (PC1 = 95.38% and PC2 = 3.78%) together accounted for approximately 99% of the total G + GE variation, indicating excellent goodness of fit for the two-dimensional biplot for visual interpretation. The green line represents the environment, the blue dot represents genotypes, and the concentric semicircles represent the ranking of genotypes based on their distance from the ‘ideal genotype’. Each semicircle is a ranking scale. The first ring contains genotypes closest to the ideal. Outer rings represent progressively lower desirability.
The combined fixed-effect ANOVA detected significant genotypic variability and a significant genotype × environment interaction. The AMMI model revealed that nearly all interaction variation was captured by the first two PC, indicating a structured interaction pattern. The WAASB analysis, based on a REML/BLUP framework, confirmed significant genotypic and interaction variance components and enabled robust stability ranking. Across all indices – including Shukla’s variance, WAASB and RSI – EC471920 consistently ranked highest for combined performance and stability. EC471972 and EC472119 exhibited moderate performance with favourable adaptation across waterlogging conditions, where plants still perform relatively well in moderate stress. Integrated ranking through combined ANOVA, AMMI analysis, regression stability parameters (bi and S2di), Wricke’s ecovalence, Shukla’s stability variance, AMMI-based stability measures (ASTAB and ASI) and WAASB-based AMMI confirmed three accessions, EC471972, EC472119 and EC471920, as superior and potential donor lines (Tables S5–S13).
Discussion
The study emphasizes germplasm characterization and utilization for pre-germination anaerobic tolerance through balanced multi-environment evaluation. The predominance of genotypic variance and high heritability of genotype means indicate strong genetic control of this trait. Similar observations have been reported in recent flooding tolerance studies, where genetic effects dominate phenotypic variation under hypoxic conditions (Abebe et al. Reference Abebe, Adewumi, Adebayo, Shaahu, Mushoriwa, Alabi, Derera, Agbona and Chigeza2024; Mullualem et al. Reference Mullualem, Tsega, Mengie, Fentie, Kassa, Fassil, Wondaferew, Gelaw and Astatkie2024). Although environmental effects were substantial, the moderate magnitude of G × E interaction suggests that intrinsic seed traits largely govern germination under flooding stress. Field and laboratory screening approaches were highly complementary; membrane stability represented an important physiological determinant of tolerance and integrated physiological and stability analyses provided robust donor identification.
The consistent performance of certain accessions across environments indicates inherent tolerance mechanisms. This could be because under oxygen deprivation, tolerant genotypes maintain ATP production through enhanced fermentative metabolism and efficient carbohydrate mobilization (Nonogaki Reference Nonogaki2019; Wijewardana et al. Reference Wijewardana, Reddy, Krutz, Gao and Bellaloui2019). The significant G × E interaction justifies the use of stability-based selection. BLUP-based approaches such as WAASB, combined with GGE biplots, provide reliable genotype evaluation across environments (Verbrigghe et al. Reference Verbrigghe, Muylle, Pegard, Rietman, Đorđević, Ćeran and Roldán-ruiz2025).
The high proportion of interaction variance explained by the first two PCs confirms structured interaction patterns and facilitates identification of mega-environments. Such differentiation supports targeted breeding strategies, allowing deployment of stable donors for broad adaptation and responsive genotypes for specific agro-ecologies. The integration of multi-criteria decision tools (RSI; ViTSel) also facilitates donor identification. Pre-germination anaerobic tolerance in soybean was found to be a genetically controlled and moderately heritable trait suitable for multi-environment characterization. Integrated MET analysis enabled the identification of both broadly and specifically adapted accessions.
The study demonstrated that only a limited number of accessions possessed meaningful anaerobic germination tolerance. The combined GGE biplot and stability analyses indicate that EC471920 exhibited the highest mean performance along with moderate stability, indicating its potential for broad adaptation across environments. EC472119 showed stress-induced germination, high predictability and minimal contribution to genotype × environment interaction, suggesting stable performance and possible specific adaptation. In contrast, EC471972 displayed better germination under stress, greater responsiveness to environmental changes and comparatively lower stability.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S147926212610077X.
Acknowledgements
The authors are grateful to the Indian Council of Agricultural Research, IARI, New Delhi, India, for providing the facilities for this work.
Author contributions
Am.R. and D.R. conceived the idea. Am.R., P.B. and Ay.R. wrote the original manuscript. Am.R., D.R., M.S., B.P.M., K.S., A.K., P.W.M. and M.Y. performed the experiments. R.B. performed the statistical analysis. S.K.L. and A.T. revised the draft manuscript and supervised the study. Am.R., D.R., Ay.R., D.E. and R.B. were involved in the reviewing and editing of draft versions. All authors read and approved the final manuscript.
Data Availability Statement
All data generated or analyzed during this study are included in the article or as supplementary files.
Funding statement
Not applicable.
Competing interests
The authors declare no competing interests.
Ethical standards
Not applicable.
