Grain protein and yield stability study in rainfed durum wheat RILs Authors: Sina Ghanbari, Kianoosh Cheghamirza, Leila Zarei, Roghayeh Naseri, Francesca Desiderio, Nino Virzi, Fabiola Sciacca, Massimo Palumbo, Elisabetta Mazzucotelli Published: 2026 ABSTRACT AbstractDeveloping durum wheat cultivars with stable grain yield across diverse environments remains a key breeding objective. This study evaluated 118 recombinant inbred lines (RILs) derived from a cross between the drought-adapted cultivar ‘Zardak’ (Triticum durum) and the landrace ‘Iran-249’ (T. turanicum) with desirable seed characteristics, across four heterogeneous rainfed environments in Italy and Iran. The assessment focused on grain yield (GY) and grain protein content (GPC) stability. Combined analysis of variance revealed significant (p < 0.01) effects for genotype, environment, and their interaction for both traits. Line ZD-050 showed the highest GY (3.91 t ha⁻¹), while ZD-032 had the highest GPC (14.27%). Stability analysis using parametric and non-parametric methods, along with AMMI and GGE biplot modeling, identified ZD-050 as among the most promising genotypes according to yield-integrating and dynamic-stability approaches. This line showed high grain yield in methods such as the Superiority Index and Kang’s rank-sum, although stability rankings differed across the used methods. This line maintained superior yield, demonstrated broad adaptability across environments, and had moderate protein levels, identifying it as an optimal candidate for breeding programs targeting yield stability and wide adaptation under rainfed conditions.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1038/s41598-026-54808-2.Keywords: AMMI method, Genotype × Environment interaction, Stability analysis, Triticum turgidumSubject terms: Biotechnology, Genetics, Plant sciences Introduction Durum wheat (Triticum turgidum L. subsp. durum Desf.) represents approximately 5% of global wheat production. It is cultivated on about 17 million hectares worldwide, making it the second most important wheat species1,2. A major challenge in durum wheat cultivation is yield instability, which contributes significantly to the gap between potential and actual yield, particularly under stress conditions3. For breeders it is necessary to maintain the level of genetic gain and the diversity of gene pools to select for field adaptation traits4. This challenge is further exacerbated by climate change and increasing environmental variability. Therefore, developing cultivars that maintain high grain yield across diverse environments has become a crucial breeding objective5. Indeed, potential yield and yield stability in different regions and for many years have always been used as essential criteria in selecting and introducing new varieties6. Inter-(sub) specific crosses may introduce new genetic diversity to explore in segregant populations in order to find genotypes with new allelic combinations which may result in yield potential and yield stability. Yield stability is inherently influenced by Genotype × Environment (G×E) interactions, where a genotype’s yield varies significantly across different growing conditions7. This interaction presents a major challenge for breeders, as high-yielding genotypes in one environment may yield poorly in another8,9. G×E interactions can be divided into two groups: cross-over or qualitative interactions and non-cross-over or quantitative interactions. Qualitative or cross-over interactions occur when the ranking of genotypes in environments is significantly different. In contrast, non-cross-over interactions occur when the yield of genotypes in environments is different, yet the ranking of genotypes in environments remains constant10. In the presence of cross-over interactions, genotypes that perform well in specific environments can be selected and recommended as cultivars or parents for breeding programs11. When the interaction is of the non-cross-over type, the emphasis can be on selecting genotypes with general adaptability, that is, genotypes that perform well in a wide range of environments. This can help breeders create more flexible and adaptable plant varieties that can perform well in diverse environmental conditions10. To increase crop production, one key factor is to enhance current knowledge about the genetic material used12. Multi-environmental trials are crucial for evaluating the yield of genotypes in various environments and for identifying stable or high-yielding genotypes as potential varieties or parents in breeding programs13. While combined analysis of variance can detect G×E interactions, it cannot elucidate their nature, necessitating specialized stability analyses to identify consistently high-yielding genotypes14. Several statistical models have been proposed to study G×E interactions and exploit their potential in the cultivar release process. These models are crucial for analyzing and interpreting complex data structures in multi-environment experiments, elucidating G×E interactions, and informing decision-making for cultivar release in breeding programs10. Two principal stability concepts guide this evaluation: biological stability (static), where genotypes show minimal yield variation across environments but limited response to improved conditions, and agronomic stability (dynamic), where genotypes demonstrate predictable, positive responses to environmental improvement - the preferred choice for commercial breeding15. These concepts are evaluated using various statistical approaches, including parametric methods (e.g., Shukla’s stability variance, Wricke’s ecovalence), non-parametric measures (e.g., Nassar and Huehn’s, Thennarasu’s statistics), and multivariate techniques such as the AMMI (additive main effects and multiplicative interactions) and GGE biplot (genotype + genotype × environment interaction) models16,17. In addition, the non-parametric Kang’s rank sum method is considered one of the appropriate methods in stability analysis due to the high accuracy of the results18, and many studies have been conducted using this method to determine the stability in different crop species19–22. Multivariate methods provide more comprehensive information about interaction and phenotypic variability16,23,24. Among these, the AMMI and GGE biplot models are useful for graphically analyzing G×E interactions and identifying stable, high-yielding genotypes17. The GGE biplot graphical method, which eliminates the effect of the environment while simultaneously combining the main effect of the genotype and the G×E interaction, leads to reliable results25. Using this method, stable and high-yielding genotypes with both general and specific adaptations, as well as favorable environments, can be identified. General adaptation is defined as selection based on the average performance of genotypes across all environments, and specific adaptation is defined as selection based on the average performance of genotypes in a subset of environments26. Numerous researchers have employed the GGE biplot method to evaluate the adaptation and stability of genotypes27–30. Recent applications of these methods in wheat research have demonstrated their effectiveness in identifying stable genotypes. Studies have employed AMMI analysis to evaluate yield stability in both bread wheat31 and durum wheat32,33, while integrated approaches combining parametric and non-parametric methods have successfully identified stable durum wheat genotypes34,35. About Recombinant Inbred Lines (RILs), the stability of 175 RILs belonging to a wider population originating from 26 diverse common wheat parents was investigated using the Finlay-Wilkinson regression coefficient method, and seven lines were identified as the most stable about grain yield36. Beyond grain yield, stability analysis has also proven valuable for quality traits such as protein content, with several studies demonstrating its utility in selecting genotypes with stable quality characteristics37,38. For instance, a stability analysis of wheat genotypes was done to identify the most stable wheat genotype for qualitative protein and gluten content traits using the AMMI method37, while the univariate parameters coefficient of stability and deviation from regression were used to evaluate 11 wheat genotypes and identify stable genotypes for grain yield, protein content, and gluten content39. This study conducted a comprehensive stability analysis on a RIL population derived from a cross between the drought-adapted cultivar ‘Zardak’ (T. turgidum ssp. durum) and the landrace ‘Iran-249’ (T. turgidum ssp. turanicum). The latter contributes favorable alleles for grain characteristics and quality traits40. The study aimed to evaluate the stability of RILs for both grain yield and protein content across multiple environments, identify genotypes combining high yield with stability, and select promising lines for rainfed durum wheat breeding programs targeting wide adaptation and consistent quality. Results Combined analysis of variance and calculation of stability parameters The phenotypic yield of RILs and parents across the four environments is summarized in Table 1A. Before conducting stability analyses, we verified the fundamental assumptions for parametric tests. The Kolmogorov-Smirnov test confirmed normality of residuals, while Bartlett’s test established homogeneity of error variances across environments for both grain yield (GY) and grain protein content (GPC). Subsequently, combined analysis of variance revealed highly significant (p < 0.01) effects for environment, RILs, and their interaction for both traits (Table 1B). Table 1Summary of phenotype data (A) and combined analysis of variance (B) for grain yield (GY) and grain protein content (GPC) in 118 durum wheat RILs along with parents across the four environments.A)EnvironmentGrain Yield (t ha − 1 )Parent MeanGrain protein content (%)Parent MeanMeanMinMaxZardakIran-249MeanMinMaxZardakIran-249Libertinia 2013–20142.351.92.971.260.7611.4111.112.912.0511.9Libertinia 2014–20152.72.52.972.082.4613.1712.514.513.1313.4Fiorenzuola d’Arda 2014–20154.1744.334.593.58-----Kermanshah 2020–20211.921.91.933.13315.5115.515.515.3515.7Average across environments2.782.63.052.772.4513.361314.313.5113.67B)Source of variationGYGPCdfMean of squaresdfMean of squaresEnvironment3343.60**2775.56**Block (Environment)812.58**614.12**Genotype1191.68**1191.34**Genotype×Environment3570.70**2381.32**Experimental Error9520.227140.61C.V %-17.19-5.67**: Significant at 0.01 level of probability. Summary of phenotype data (A) and combined analysis of variance (B) for grain yield (GY) and grain protein content (GPC) in 118 durum wheat RILs along with parents across the four environments. **: Significant at 0.01 level of probability. The significant environment effect reflected substantial differences among test locations, while the significant Genotype effect indicated considerable genetic diversity within the population. Most importantly, the significant G×E interaction demonstrated differential responses of genotypes across environments, confirming the necessity of stability analysis. Among the evaluated lines, ZD-050 line showed the highest mean GY (3.91 t ha⁻¹) compared to the population average of 2.78 t ha⁻¹ (Table 2). For GPC, ZD-032 achieved the highest value (14.27%) against a population mean of 13.36% (Table 3). A table containing the grain yield and grain protein content mean values of all lines across the evaluated environments was provided as Supplementary Table 1 S. Table 2Values of parametric statistics of grain yield stability are reported for the top 10 durum wheat lines for each calculated parameter.RIL CodeGY(t ha− 1) R 2 RIL CodeEVRIL CodeCVRIL CodeShuklaRIL CodeWiRIL CodeDiRIL CodePiRIL CodeBiRIL CodeS2di50*3.910.98820.118214.321000.0021000.01249*2.48500.21322.29530.002116*3.880.76650.227815.69770.02770.05Zardak2.461160.24242.02820.0021323.820.221360.245416.65690.02690.061322.10780.40511.61290.003783.750.781030.2813617.2060.0260.06732.041320.43211.601000.0071043.590.67330.286518.04520.02520.07192.031040.49611.58520.008493.560.79540.2910320.17810.02810.081181.99490.51581.54650.008113.510.921260.327220.24430.03430.09711.97250.51121.54370.009253.460.771200.3310420.70530.03530.09921.95110.59151.501270.0091333.430.97780.3513221.181050.031050.101451.941330.65451.48120.010193.370.46760.3911621.33560.03560.101371.921310.68681.48770.013*: code 249 = Iran-249; code 50 = ZD-050; Code 116 = ZD-116 and etc. Values of parametric statistics of grain yield stability are reported for the top 10 durum wheat lines for each calculated parameter. *: code 249 = Iran-249; code 50 = ZD-050; Code 116 = ZD-116 and etc. Table 3Values of parametric statistics of grain protein content stability for the top 10 durum wheat lines for each calculated parameter.RIL CodeGPC (%) R 2 RIL CodeEVRILCodeCVRILCodeShuklaRIL CodeWiRIL CodeDiRIL CodePiRIL CodeBiRIL CodeS2di32*14.270.9950.0451.4479−0.002790.0011452.71990.48191.97790.0000119*14.210.921110.43824.85940.001940.011332.54980.66301.90720.000123014.070.96820.451114.8840.00240.01512.48800.731171.88450.0001611714.070.98730.49735.221350.0021350.01992.461010.77321.871160.00022214.030.81560.54565.41380.002380.011212.421270.88521.69750.000415214.020.991340.611285.56890.003890.02282.41190.89241.67250.000479914.010.621280.621345.761050.0071050.03192.40280.97181.64290.000638013.970.97350.67436.17230.013230.05362.36300.98201.63710.000959813.950.89430.68356.29450.014450.051372.351420.99171.521470.0011711213.940.99620.80626.551030.0201030.071272.311121.031471.48550.00152*: code 32 = ZD-032; Code 19 = ZD-019 and etc. Values of parametric statistics of grain protein content stability for the top 10 durum wheat lines for each calculated parameter. *: code 32 = ZD-032; Code 19 = ZD-019 and etc. Parametric stability analysis The different stability parameters computed in this study are reported in Tables 2 and 3 for GY and GPC, respectively, for the 10 best RILs for each parameter and in Table 2S and 3S for all RILs. For GY stability assessment, Roemer’s Environmental Variance method identified ZD-005, ZD-111, ZD-082, ZD-078, and ZD-104 lines as having the lowest variance. Among these, ZD-078 and ZD-104 demonstrated both stability and above-average yield (3.75 and 3.59 t ha⁻¹, respectively). The highest amount of EV was related to lines ZD-032, ZD-024, and ZD-061 (Table 2S), which indicated the fluctuation of their GY in the tested environments. The Coefficient of Variation analysis revealed ZD-082, ZD-078, ZD-054, and ZD-136 lines as having the lowest values, suggesting biological stability. However, ZD-082’s below-average yield precludes its classification as agronomically stable. According to Wricke’s ecovalence and Shukla’s stability variance, ZD-100, ZD-077, ZD-069, ZD-006, and ZD-052 lines had the lowest values, with ZD-077 and ZD-052 combining stability with favorable yield. The Superiority Index identified ZD-050, ZD-116, ZD-078, ZD-132, and ZD-104 lines as the most stable, while regression analysis revealed ZD-032, ZD-024, ZD-051, and ZD-061 (bi > 1) as specifically adapted to favorable environments. Lines ZD-053, ZD-082, ZD-029, ZD-100, and ZD-052 showed the lowest deviation from regression, indicating biological stability with adaptation to unfavorable conditions. Line ZD-039, with a regression coefficient equal to one and an average grain yield higher than the total average, was adapted to all environments (Table 2S). For GPC stability, different lines emerged as stable across various parameters. ZD-079, ZD-072, and ZD-045 lines showed minimal deviation from regression, while ZD-099, ZD-098, and ZD-080 had the lowest Superiority Index values. Non-parametric stability analysis Non-parametric stability measurement methods do not require assumptions about variance distribution and uniformity. In addition, they are based on phenotype-based genotype ranking instead of phenotypic data. Some non-parametric methods have been used to evaluate the stability of lines, and results are reported in Tables 4 and 5 for GY and GPC and also in Table 4S and 5S, and ZD-069 lines as the most stable, though ZD-031, ZD-100, and ZD-069 exhibited below-average yields. Additional non-parametric indices (Si⁽³⁾, Si⁽⁶⁾, Ni⁽²⁾, Ni⁽³⁾, Ni⁽⁴⁾) identified ZD-003 and ZD-043 as stable, but their low yields limited practical utility. Table 4Values of non-parametric statistics of grain yield stability for the top 10 durum wheat lines for each calculated parameter.RIL CodeGY(t/ha)RIL CodeSi1RIL CodeSi2RIL CodeSi3RIL CodeSi6RIL CodeNi1RIL CodeNi2RIL CodeNi3RIL CodeNi4RIL CodeRank sum50*3.91312.83314.92310.06310.13311.75310.01310.18310.025021116*3.88504.175010.921000.231001.06502.7530.051000.511000.09116271323.821005.3310020.3330.23692.001003.251000.05690.7130.17849783.751167.1711636.92690.2532.131165.25430.0630.73690.149621043.596996958.00430.30433.21525.5690.06430.9430.1213370493.5652105261.67160.31163.44695.5160.08160.93160.141171113.51310.67377.6760.3763.8336.2560.0960.9860.1513272253.464312.1743106.251050.45524.35437.3890.11521.041050.18104741333.431613.3316111.3390.501055.85167.51050.111051.2190.212579193.377714.1777122.92520.52506.24778.75520.13501.25520.2475114*: code 50 = ZD-050; Code 116 = ZD-116 and etc. Values of non-parametric statistics of grain yield stability for the top 10 durum wheat lines for each calculated parameter. *: code 50 = ZD-050; Code 116 = ZD-116 and etc. Table 5Values of non-parametric statistics of grain protein content stability for the top 10 durum wheat lines for each calculated parameter.RIL codeGPC (%)RIL codeSi1RIL CodeSi2RIL CodeSi3RIL CodeSi6RIL CodeNi1RIL CodeNi2RIL CodeNi3RIL CodeNi4RIL CodeRank sum32*14.27211.56211210.03210.04210.83210.01210.12210.0198431914.21898.22792.33890.08890.28891.83890.02890.31790.0399483014.07798.67892.33790.12790.38792790.03790.35890.03806711714.07135261354940.171351.131353.5230.041350.61230.05272214.03434.8944.33230.17231.1444.17940.04230.62940.06127745214.022336.22234.671350.20941.51234.171350.04940.711350.0671809914.0110540.221055810.281052.781054.33810.061050.96810.08112858013.973846.22385.331320.30812.99944.51320.078111320.14919813.9594549461050.4043.45384.671050.0941.071050.12779111213.949880.89457.33630.411323.61985.6740.11321.1630.1312391*: code 32 = ZD-032; Code 19 = ZD-019 and etc. Values of non-parametric statistics of grain protein content stability for the top 10 durum wheat lines for each calculated parameter. *: code 32 = ZD-032; Code 19 = ZD-019 and etc. For GPC, non-parametric methods consistently identified ZD-021, ZD-089, and ZD-079 lines as the most stable across multiple statistics (Si⁽¹⁾, Si⁽³⁾, Si⁽⁶⁾, Ni⁽¹⁾-Ni⁽⁴⁾). Kang’s rank-sum method, which integrates mean yield and stability, identified ZD-050 and ZD-116 lines as the most agronomic stable for GY, and ZD-098, ZD-099 for GPC. AMMI analysis The AMMI model was employed to investigate the nature of the G×E interaction. The analysis of variance based on the AMMI model revealed that the first two principal components (IPCA1 and IPCA2) were significant for grain yield, while only IPCA1 was significant for grain protein content (Table 6). This predominance of the first axis indicates that a major environmental gradient, likely related to differences in rainfall, temperature, and other site-specific conditions among the test environments, contributed substantially to the observed interaction. Because disease incidence and other biotic stresses were not systematically recorded in this study, their possible contribution to genotype × environment interaction cannot be excluded. Table 6Analysis of variance for the AMMI model for grain yield (GY) and grain protein content (GPC) in 118 durum wheat RILs along with parents across four and three environments.Source of variationGYGPCdfMean of squaresdfMean of squaresRIL1190.561190.44Environment3114.532258.521RIL × Environment3570.232380.44IPCA 11210.32**1200.60**IPCA 21190.23**1180.27n.sIPCA 31170.14n.s--n.s: Not significant.**: Significant at 0.01 level of probability. Analysis of variance for the AMMI model for grain yield (GY) and grain protein content (GPC) in 118 durum wheat RILs along with parents across four and three environments. n.s: Not significant. **: Significant at 0.01 level of probability. The first principal component (IPCA1) accounted for 80.4% and 98.2% of the total G×E interaction sum of squares for grain yield and grain protein content, respectively. Previous studies have indicated that the most appropriate AMMI model is one where the first two principal components are significant and capture the majority of the interaction variance41,42. In our study, this criterion was met for grain yield analysis. Analysis of IPCA1 scores identified ZD-099, ZD-056, ZD-040, ZD-066, and ZD-145 lines as having the lowest absolute values for grain yield (Table 7). However, ZD-099 and ZD-040 were excluded from consideration due to their below-average yields. For grain protein content, lines ZD-090, ZD-141, and ZD-063 showed the lowest IPCA1 values (Tables 6 and 7S). Table 7Values of IPCA1 and IPCA2 for the top 10 durum wheat RILs according to IPCA1 value.RILscodeGY(t ha− 1)IPCA1IPCA2RILscodeGPC(%)IPCA1IPCA299*2.21−0.970.309013.267−0.930.14562.84−0.96−0.9114113.359−0.910.14402.48−0.890.166313.09−0.820.61663.08−0.880.216513.331−0.820.131452.84−0.85−0.40Iran-24913.415−0.81−0.36842.79−0.81−0.244913.309−0.8−0.22942.60−0.770.211914.215−0.790.161382.40−0.77−0.343113.605−0.72−0.23Zardak2.77−0.760.523014.075−0.690.35183.21−0.73−0.279713.018−0.670.17*: code 99 = ZD-099, code 56 = ZD-056 and etc. Values of IPCA1 and IPCA2 for the top 10 durum wheat RILs according to IPCA1 value. *: code 99 = ZD-099, code 56 = ZD-056 and etc. For the trait grain yield, the lines ZD-010, ZD-128, ZD-016, ZD-048, and ZD-090 had the lowest AMMI Stability Value (ASV), which integrates information from significant IPCAs, as having the lowest values (Table 8). Among these, ZD-016 was removed from further consideration due to its below-average yield. For grain protein content, lines ZD-008, ZD-059, and ZD-046 demonstrated the lowest ASV values (Tables 7 and 8S). Table 8Values of ASV for the top 10 durum wheat RILs.RILsGY(t ha−)ASVRILsGPC(%)ASV10*2.940.18813.5160.171283.120.185913.6090.17162.270.194613.2090.18482.820.195013.3310.20902.930.197413.0650.20972.680.2011213.9410.21632.860.224813.1540.24262.560.237813.0690.251172.970.244213.2070.27592.950.2410013.1450.28*: code 10 = ZD-010, code 128 = ZD-128 and etc. Values of ASV for the top 10 durum wheat RILs. *: code 10 = ZD-010, code 128 = ZD-128 and etc. GGE biplot analysis The first two components of the GGE biplot accounted for 74.23% of the total line effects and line × environment interaction (Fig. 1), indicating the adequacy and high discriminatory power of this method for simultaneously evaluating the grain yield and stability of durum wheat lines. The placement of PC1 as a representative of the average grain yield and PC2 as an indicator of genotype instability is fully consistent with the theoretical framework of the GGE biplot. The concentration of the majority of lines around the origin of the coordinates in this study indicates their moderate yield and relative stability. In contrast, the distance of lines ZD-116, ZD-104, ZD-078, ZD-132, ZD-054, and ZD-050 in the positive quadrant of PC1 from the origin indicates the higher yield potential of these lines. On the other hand, the considerable distance of lines ZD-032, ZD-024, and ZD-061 from the biplot origin, characterized by strongly to moderately negative scores on both PC1 and PC2 confirms their instability and lower yield across most environments. Biplot of line-environment interaction based on 118 durum wheat RILs grain yield in four environments. (E1 = Libertinia 2013–2014, E2 = Libertinia 2014–2015, E3 = Fiorenzuola d’Arda 2014–2015, E4 = Kermanshah 2020–2021). The first two components of the GGE biplot collectively explained 83.04% of the grain protein content variation attributed to line (L) effects and line-by-environment (L×E) interaction (Fig. 2). This high explained variance demonstrates the adequacy and robustness of the GGE biplot method for evaluating the stability of grain protein content in durum wheat lines. Most lines clustered near the origin of the biplot, indicating their average grain protein content and relative stability across the test environments. In contrast, lines ZD-019, ZD-098, ZD-099, and ZD-127 were positioned farther along the positive direction of the first principal component (PC1). This statement suggests their superior potential for higher grain protein content compared to the overall mean. Biplot of line-environment interaction based on 118 durum wheat RILs grain protein content in three environments. (E1 = Libertinia 2013–2014, E2 = Libertinia 2014–2015, E3 = Kermanshah 2020–2021). Which-won-where model of the GGE biplot The which-won-where model of the GGE biplot is one of the most effective tools for examining the structure of G×E interaction in multi-environment trials. In this model, a polygon is drawn by connecting the lines that are at the farthest distance from the biplot origin, known as the vertices. In this study, lines ZD-082, ZD-031, ZD-032, ZD-024, ZD-050, and ZD-132 are located at the apex of the polygon (Fig. 3). Being located at the apex of the polygon indicates that these lines have shown much higher or lower grain yield than the other genotypes in some environments. Drawing lines perpendicular to the sides of the polygon also divides it into distinct sections, each section associated with a vertex genotype. The environments in the same section are those in which the vertex genotype of that section performed well. Which-won-where model of line-environment interaction based on 118 durum wheat RILs grain yield in four environments. (E1 = Libertinia 2013–2014, E2 = Libertinia 2014–2015, E3 = Fiorenzuola d’Arda 2014–2015, E4 = Kermanshah 2020–2021). Based on grain protein content, lines ZD-021, ZD-133, ZD-005, ZD-127, ZD-099, ZD-058, ZD-017, ZD-032 and ZD-019 were located at the vertices of the polygon (Fig. 4). The location of these lines at the vertices of the polygon indicates that they exhibited the highest or lowest grain protein content values compared to other lines in at least one environment. This pattern suggests the existence of strong line-by-environment interactions and demonstrates the potential of these lines to be used for targeted selection of high-protein genotypes in specific environments. Which-won-where model of line-environment interaction based on 118 durum wheat RILs grain protein content in three environments. (E1 = Libertinia 2013–2014, E2 = Libertinia 2014–2015, E3 = Kermanshah 2020–2021). Relationships between environments and lines Although multi-environment experiments are primarily designed to evaluate genotypes, the data obtained from these experiments also allow for the evaluation of the quality and efficiency of experimental environments. In this framework, two key concepts are discriminative power and representativeness. In the GGE biplot model, these characteristics are visualized through the relationships between the discriminative power and representativeness of environments. In this analysis, a longer environment vector indicates greater discriminative power for distinguishing genotypes, while a smaller angle between the environment vector and the mean axis indicates better representativeness of the average conditions (Fig. 5). In this study, the E1 environment has the longest vector and a very low angle with the mean axis of the environments, indicating high discriminative power coupled with high representativeness. The E4 environment has a long vector but forms a larger angle with the mean axis, indicating high discriminative power but only moderate representativeness. The E3 environment, despite its long vector, has a large angle with the mean axis of the environments, indicating a strong G×E interaction and poor representation of average conditions. In contrast, the E2 environment has a short vector and a small angle, indicating low resolution and limited information for distinguishing genotypes. The concentric circles in the biplot indicate the distance of the genotypes from the ideal genotype (highest yield and highest stability). Accordingly, lines such as ZD-116, ZD-104, ZD-078, ZD-050, and ZD-132 can be introduced as superior options with relative general adaptability. In contrast, lines such as ZD-024 and ZD-032, although they may perform well in some environments, exhibit less stability due to their distance from the ideal line. Relationships between environments and lines based on 118 durum wheat RILs grain yield in four environments. (E1 = Libertinia 2013–2014, E2 = Libertinia 2014–2015, E3 = Fiorenzuola d’Arda 2014–2015, E4 = Kermanshah 2020–2021). Based on the distance of each line from the ideal line, lines ZD-127, ZD-019, ZD-098 and ZD-099 were identified as superior lines with relative overall adaptability in terms of grain protein content (Fig. 6). In contrast, lines such as ZD-058, ZD-060 and ZD-029, despite having good yield in some environments, showed less stability for this trait due to their greater distance from the ideal genotype and are not recommended for general cultivation. Relationships between environments and lines based on 118 durum wheat RILs grain protein content in three environments. (E1 = Libertinia 2013–2014, E2 = Libertinia 2014–2015, E3 = Kermanshah 2020–2021). Correlation analysis To assess the consistency among different stability evaluation approaches, Pearson correlation coefficients were calculated between various parametric and non-parametric stability indices for both grain yield and grain protein content (Figs. 7 and 8). Heatmap drawn based on Pearson’s correlation coefficients between grain yield and stability indices for 118 durum wheat RILs evaluated across four environments. Heatmap drawn based on Pearson’s correlation coefficients between grain protein content and stability indices for 118 durum wheat RILs evaluated across three environments. Strong positive correlations were observed between conceptually related methods. The coefficient of variation (CV) and environmental variance (EV) showed a high positive correlation for both traits. Similarly, Wricke’s ecovalence (Wi) and Shukla’s stability variance demonstrated an almost perfect correlation, confirming their mathematical equivalence in measuring a genotype’s contribution to G×E interaction. Non-parametric measures developed by Nassar and Huehn (Si⁽¹⁾, Si⁽²⁾, Si⁽³⁾) and Thennarasu (Ni⁽¹⁾, Ni⁽²⁾, Ni⁽³⁾, Ni⁽⁴⁾) showed significant positive inter correlations within their respective groups. These non-parametric indices also exhibited positive correlations with parametric measures of interaction, particularly Wricke’s ecovalence, Shukla’s variance, and deviation from regression (S²di). Notably, both the superiority index (Pi) and Kang’s rank-sum showed significant negative correlations for both traits, confirming that these indices effectively integrate high yield into their stability assessment. In contrast, the AMMI Stability Value (ASV) showed only weak negative correlations with the non-parametric indices Si⁽³⁾ and Ni⁽⁴⁾, suggesting that the multivariate AMMI approach captures distinct aspects of stability compared to univariate methods. To provide explicit quantitative support for line selection across different stability concepts, Table 9 summarizes the number of parametric and non-parametric methods that ranked each high-yielding line among the most stable, along with their ASV rank and Kang’s rank-sum values. Table 9Quantitative consensus among stability methods for high-yielding RILs based on Supplementary Tables 2–7 S.TraitRILMean valueRank (value)No. of parametric methods ranking1No. of non-parametric methods ranking 2ASV rank (lower is better)Kang’s rank-sum (lower is better)Overall consensus (%)3GYZD-0503.91 t/ha11/85/984216/17 (35%)GYZD-1163.88 t/ha22/84/920276/17 (35%)GYZD-0783.75 t/ha42/81/942493/17 (18%)GYZD-1323.82 t/ha33/81/980724/17 (24%)GPCZD-09914.02%72/83/978483/17 (18%)GPCZD-09813.95%91/82/984434/17 (24%)GPCZD-01914.22%23/80/91031303/17 (18%)GY: Grain Yield, GPC: Grain Protein Content.¹The eight parametric methods from Table 2S (GY) and Table 3S (GPC): EV, CV, Shukla, Wi, Di, Pi, bi, S²di.²The nine non-parametric methods from Table 4S (GY) and Table 5S (GPC): Si¹, Si², Si³, Si⁶, Ni¹, Ni², Ni³, Ni⁴, Rank sum.³Number of methods (out of 17) that ranked the line among top 10 for stability. Quantitative consensus among stability methods for high-yielding RILs based on Supplementary Tables 2–7 S. GY: Grain Yield, GPC: Grain Protein Content. ¹The eight parametric methods from Table 2S (GY) and Table 3S (GPC): EV, CV, Shukla, Wi, Di, Pi, bi, S²di. ²The nine non-parametric methods from Table 4S (GY) and Table 5S (GPC): Si¹, Si², Si³, Si⁶, Ni¹, Ni², Ni³, Ni⁴, Rank sum. ³Number of methods (out of 17) that ranked the line among top 10 for stability. Discussion Selection trials in multiple environments are essential to plant breeding, and applying appropriate statistical analysis is crucial to improve selection accuracy. In durum wheat breeding, much progress has been made regarding genetic improvement, but the reduction of genetic diversity within breeding germplasm pools is still a concern43. This study leveraged the complementary genetic backgrounds of the durum wheat cultivar ‘Zardak’ and the landrace ‘Iran-249’ to develop a population of 118 recombinant inbred lines (RILs). The objective was to exploit this diversity for enhancing yield stability, a critical breeding goal. ‘Zardak’ contributes favorable alleles for phenology and seed composition, whereas ‘Iran-249’ is a valuable source of alleles for spike architecture, grain morphology, and stress resistance44. This aligns with strategic pre-breeding approaches that utilize primary gene pool relatives to introgress novel traits and recover lost genetic diversity while minimizing linkage drag, thereby enriching the breeding germplasm45,46. Given several barriers associated with using species from secondary and tertiary gene pools, availability of crop relative germplasm within the primary genetic pool to use as parents is of great utility46,47. Our findings confirm the potential of this strategy, as the derived RILs exhibited a wide range of yields, including transgressive segregants that surpassed both parents, underscoring the value of ‘Iran-249’ for trait introgression40. A significant Genotype × Environment (G×E) interaction was observed for grain yield, a common phenomenon in durum wheat that complicates selection48–52. Breeding programs for the selection of superior durum wheat genotypes for grain yield in the past years have shown that the effect of genotype compared to the effects of the environment and the G×E interaction has a smaller role in justifying the variance50,53. The substantial environmental variance, likely driven by variations in rainfall distribution and other climatic factors across test sites, highlights that selection efficiency depends on accurately dissecting the G×E interaction to identify genotypes with reliable yield54,55. To this end, we employed a suite of univariate and multivariate stability parameters, acknowledging that no single method provides a complete picture56. Our approach was to distinguish between two fundamental concepts of stability: static (biological stability, characterized by minimal variance across environments) and dynamic (agronomic stability, characterized by a predictable and proportional response to improving conditions). Consistent with the static concept, parameters such as the environmental variance (σ²i) and coefficient of variation (CV) identified lines like ZD-082 and ZD-078 as stable under rainfed conditions. However, these genotypes generally exhibited lower mean yields, making them potentially suitable only for low-input, unpredictable environments57. In contrast, statistics highlighting dynamic stability are more aligned with the breeder’s objective of selecting high-yielding, responsive genotypes. The superiority measure (Pi) effectively identified high-yielding and agronomic stable RILs, which combine high yield with reduced variance. The parameter Pi makes yield a main influencer of the ranking across environments and should allow selection for favorable environment58. According to the Pi statistic, the high yielding ZD-050 and ZD-116 lines were defined as stable RILs, which indicates their dynamic stability. Furthermore, the high concordance between Shukla’s stability variance and Wricke’s ecovalence, as expected theoretically59, provided corroborative evidence for the dynamic stability of these top-performing lines. In this population and under our specific set of environments, the application of non-parametric statistics yielded mixed results. While Si¹, Si², Ni¹, and Kang’s rank-sum effectively identified high-yielding, dynamically stable lines (e.g., ZD-050 and ZD-116), most other non-parametric indices (Si³, Si⁶, Ni²–Ni⁴) were ineffective for selecting high-yielding, stable lines. This finding is specific to our study and could not be generalized to other genetic backgrounds or environmental networks without further validation. In the present study, these five indices (Si³, Si⁶, Ni²–Ni⁴) aligned with the static rather than the dynamic stability concept, consistent with previous findings in safflower60. However, this pattern may not hold for all datasets or crops, and caution is needed when generalizing these results to other populations or environments. Multivariate analysis via the AMMI model and its derived AMMI Stability Value (ASV) offered a more robust and agronomically meaningful assessment of stability. Analysis of AMMI is one of the multivariate stability methods that predict genotype compatibility and stability and has been widely used in the last two decades61. The AMMI analysis of variance (Table 6) revealed that the first two interaction principal component axes (IPCA1 and IPCA2) were significant for grain yield, explaining more than 80% of the total G×E interaction, while only IPCA1 was significant for grain protein content. This predominance of the first axis indicates that a single major environmental gradient—likely associated with differences in rainfall and temperature among test sites—was the main source of interaction in this study. The non-significance of higher-order components suggests that most of the G×E variation was systematic rather than random, making the AMMI1 biplot adequate for visual interpretation. Similar patterns have been reported in durum wheat and barley, where IPCA1 typically explains 60–85% of the total interaction variance33,62. Accordingly, lines with near-zero IPCA1 scores, such as ZD-099 and ZD-056 for yield and ZD-090 and ZD-141 for protein content, can be considered the most stable across environments. The AMMI stability value (ASV) was used to simultaneously apply the information of the two principal components. According to this, stable genotypes have the lowest ASV. Many researchers have introduced the ASV parameter as an essential parameter in the study of G×E interaction due to more accurate results63,64. The ASV parameter has been widely used in G×E studies65 and has performed reliably in several durum and bread wheat datasets20,33,66. In our study, ASV also provided meaningful rankings, though it did not favor ZD-050 (ranked 84th), highlighting that even well-established methods may yield divergent results depending on the dataset. ASV statistics have been used to identify stable varieties in many different studies on durum wheat20,33–35,67 and bread wheat31,66,68. In our study, AMMI analysis reinforced the identification of dynamically stable genotypes, confirming that methods capturing the agronomic concept of stability are more pertinent for modern breeding programs targeting variable but improvable environments. The AMMI model provided the results that were consistent with the agronomic concept of stability. The some researchers have suggested that AMMI is a powerful method compared to certain alternatives54,69,70, although the relative yield of stability methods may vary depending on the dataset. Therefore, our finding that AMMI performed well in this specific context could be taken as an evidence. Several studies have reported that in GGE biplot analysis, when the cumulative variance explained by PC1 and PC2 exceeds 60%, the interpretation of genotype compatibility and stability patterns is reliable71. Therefore, the value obtained in this study falls within the desired range and is even higher than that reported in many similar studies on durum wheat. Studies on durum wheat have shown that genotypes with large positive PC1 values and close to zero PC2 values are the best options for breeding programs, as they exhibit high yields and a more uniform response to environmental changes71. This pattern is typically observed in advanced breeding collections and indicates the gradual elimination of highly unstable genotypes that occurred in previous breeding stages. Similar results have been reported in recent studies of durum wheat, in which durum lines with a greater distance from the origin in the direction of PC1 were introduced as high-yielding and desirable genotypes72. Previous studies have emphasized that such genotypes, even in the case of relative stability, have less value for introduction or use in breeding programs71,72. But nevertheless, it may be reasonable to eliminate or use these genotypes in later stages of breeding. This view is particularly used in plant breeding studies to plan targeted selection of genotypes in specific or general environments73. These genotypes show strong G×E interactions and are specifically adapted to certain environments. That is, they have strong environment-specific responses that can aid in selecting genotypes suitable for specific environments. Such a structure provides clear evidence for the existence of distinct genotype × environment sections and indicates that the interaction of genotypes with environments is qualitative and environment-specific, such that some genotypes are superior to others only in specific environments. Genotypes within the polygon generally display more uniform and stable performance across environments and are considered to have general adaptation73. The grain protein content in durum wheat is strongly influenced by environmental conditions. In the GGE biplot, PC1 serves as the main axis for distinguishing genotypes in terms of protein content. The E1 environment, being a high-discrimination environment, has enabled the identification of superior lines such as ZD-099 and ZD-127. This could be attributed to favorable environmental conditions, such as moisture status or soil characteristics suitable for the expression of genes affecting protein content. The stability of genotypes can be assessed based on their distance from the biplot origin; where lines closer to the center (e.g., ZD-021) exhibit higher stability but moderate protein content, while lines farther away (e.g., ZD-099) show high protein content coupled with greater sensitivity to environmental changes. These results are consistent with the principles of GGE biplot interpretation, particularly when the sum of PC1 and PC2 explains more than 80% of the variance, which allows for the interpretation of interaction effects and the identification of environmental patterns. The present results agree with the findings of74,75, who reported the stability of genotypes close to the average axis and the effective role of discriminating environments. The which-won-where pattern in GGE biplot cannot only identify superior genotypes in specific sections of environments, but also determine the presence or absence of mega-environments65,76. In our study, combining GGE biplot with the ‘which-won-where’ perspective proved useful for identifying superior genotypes in specific environments. This approach may provide valuable guidance for breeding decisions in similar multi-environment trials, though its effectiveness should be confirmed with independent validation sets. The results of this study showed that the test environments could be separated into two main groups in terms of grain protein content: a highly discriminatory environment, E1, which facilitated the identification of superior lines in terms of protein, and a more representative environment, E3, in which genotypes responded more consistently. Among them, ZD-099 was identified as the superior line in the E1 environment, which could indicate a specific genetic adaptation of this line to specific environmental conditions such as temperature or soil fertility. The absence of superior genotypes in the negative sectors of the biplot indicates that none of the environments were entirely unfavorable, although the magnitude of the line × environment interaction is significant. The present results are consistent with the report by77, who identified several mega-environments for wheat protein content using the Which-won-where” view. Discriminative power refers to an environment’s ability to detect genetic differences between genotypes. In contrast, representativeness refers to the extent to which an environment accurately represents the average conditions of all environments under study. From a breeding perspective, the ideal environment should combine both characteristics to properly facilitate the selection of stable and high-yielding genotypes78. Therefore, the ideal environment has a long vector and a low angle with the mean axis of the environments, allowing it to simultaneously distinguish superior genotypes and evaluate their overall yield (general adaptability)79,80. From a breeding perspective, such an environment is ideal for selecting high-yielding genotypes and assessing their overall yield. Similar results have been reported in multi-environment studies on various crops, where environments with these characteristics have been introduced as reference environments for selecting stable genotypes79. In the context of our four test environments, sites with long vectors but large angles to the mean axis played a relatively minor role in identifying generally stable lines. Researchers have suggested that such environments may be candidates for exclusion or merging with genetically similar environments, as they contribute little to identifying superior genotypes in some multi-environment trials80. However, this decision should be made cautiously and depends on the specific breeding objectives and the number of environments available. These environments are usually suitable for evaluating genotypes under stress conditions or specific environments. They are used more to identify specific adaptations than to select generally stable genotypes, a result that is fully consistent with the theoretical framework of the GGE biplot79. Genotypes positioned in the positive direction of PC1 and near the center of the concentric circles combine high yield with good stability. In the ideal genotype view, the E1 environment served as an ideal environment for discriminating lines in terms of grain protein content, while the E3 environment showed less discriminatory power, which could be due to the relative uniformity of its environmental conditions. This finding indicates the importance of using discriminatory environments in breeding programs. Genotypes located near the E1 environment vector, although less stable, show high potential for increasing grain protein content and are suitable for specific breeding programs. The results of this study are consistent with the findings of75,81, who reported that more than 89% of the variance was explained by the first two components and highlighted the role of representative environments in assessing stability. Furthermore, the similarity of the environmental discrimination pattern with the results of74 indicates that different environments can play distinct roles in assessing qualitative and quantitative traits. The correlation between the environmental variance and the regression coefficient value for selecting suitable agronomic stability has been confirmed. Other studies have shown a variety of and sometimes contradictory relationships among stability statistics14,56,82–84. The frequent lack of correlation, or even the observation of contradictory relationships, among different stability parameters in our study and others14,84 is not unexpected. This is attributable to the fundamentally different statistical foundations of each parameter some are variance-based, others regression-based, and others rank-based. This confirms that high yield and stability are not inherently correlated, and the ideal genotype must be selected based on a combination of mean yield and multiple stability metrics85,86. Different methods are recommended to evaluate stability and decide after checking the results. Indeed, it is not possible to recommend a stability analysis method for all crops and in all conditions. Different reports have been presented on the correlation between stability parameters and high grain yield. A review of the discussed parameters reported that only the detection coefficient was correlated with grain yield86. The results of our study using 17 stability methods (Supplementary Tables 2–7 S) demonstrated that ZD-050 was identified as stable by 6 methods (35%), including all yield-integrating measures such as Pi (Superiority Index = 0.21, the lowest among all lines) and Kang’s rank-sum (21, the lowest). In comparison, ZD-116 showed agreement across 6 methods (35%), and ZD-078 across 3 methods as well (18%). However, it is important to note that ZD-050 received a poor ASV rank (84th out of 120) from the AMMI model, suggesting that the AMMI method, which emphasizes interaction patterns rather than mean yield, does not favor this line. This divergence highlights the necessity of using multiple stability concepts in breeding programs, as no single method provides a complete picture. Specifically, ZD-050 and ZD-116 consistently yielded above the population mean, exhibited low variance, and were ranked stable by several key non-parametric and multivariate indices. These lines represent a valuable genetic resource, as they possess the resilience to maintain yield in stressful seasons and the capacity to respond to favorable conditions. The successful identification of such lines from a cross with a landrace validates the strategic use of genetic resources from the primary gene pool to simultaneously enhance productivity and stability, a phenomenon linked to the introduction of adaptive alleles for diverse environments87–90. The present study has several limitations that should be considered when interpreting the results. First, the evaluation was conducted across only four environments, which may limit the robustness of conclusions regarding mega-environment classification and broad adaptability. Second, the environments were not only geographically but also temporally separated, with trials conducted in Italy (2013–2015) and in Iran (2020–2021). In fact, in this study was evaluated the response of genotypes to different environments, which included the confounding location and year effects. Third, although several promising RILs were identified, no subsequent validation trials or clear pathway toward cultivar release were included in this study. Finally, the results are based on a single biparental population, which may limit the generalizability of the findings to other genetic backgrounds. Despite these limitations, the identification of stable, high protein content and high-yielding RILs across contrasting environments provides a hopeful foundation for future breeding programs. With further validation, these genotypes could contribute to the development of durum wheat cultivars with both broad and specific adaptability, particularly for rainfed and semiarid regions. Conclusion This multi-method stability analysis identified several promising durum wheat recombinant inbred lines for rainfed environments. The fact that line ZD-050 was among the highest-yielding lines across most dynamic stability parameters justifies its selection as the most promising line under specific stability criteria rather than as the universally most stable genotype for rainfed durum wheat breeding programs. Other lines, including ZD-116, ZD-132, and ZD-078, also showed acceptable yield and could be considered valuable candidate materials for further breeding and validation. In contrast, while line ZD-024 possesses high yield and protein content, its strong specific adaptation limits its recommendation to rainfed environments matching its preferred conditions. Therefore, the lines identified in this study could be considered as candidate materials for further evaluation rather than as final selections based solely on yield stability. In addition, although grain yield and grain protein content were the main traits examined here, further assessment of other quality-related traits is needed before breeding recommendation or possible cultivar development. In this study, parametric and multivariate approaches proved more effective than most non-parametric indices for selecting high-yielding, agronomically stable lines. However, the suitability of these methods may depend on the specific dataset, and we do not claim that they are universally indispensable across all breeding contexts. Conversely, several non-parametric indices were found to be more aligned with the static concept of stability, effectively selecting low-yielding, non-responsive lines. Therefore, we conclude that leveraging a cross with a genetically diverse landrace like ‘Iran-249’ could be a highly effective pre-breeding strategy. This approach simultaneously enriches the gene pool and enables the development of lines that successfully balance high productivity with the stability required for rainfed cultivation in unpredictable and variable environments. Materials and methods Plant materials and experimental design This study utilized 118 recombinant inbred lines (RILs) from the F₇₋₉ generations, developed from a cross between the local drought-adapted cultivar ‘Zardak’ (Triticum turgidum ssp. durum), and ‘Iran-249’, a T. turgidum ssp. turanicum landrace with desirable seed characteristics collected from Kermanshah province, Iran. ‘Zardak’ was used as the cultivated parent because of its adaptation to rainfed conditions and agronomic suitability, whereas ‘Iran-249’ was selected as a donor parent because of its distinctive seed morphology and favorable grain-related traits. Both ‘Zardak’ and ‘Iran-249’ have a winter growth habit. The RILs were advanced by selfing to the F₇₋₉ generations, resulting in near-homozygous lines appropriate for replicated field evaluation. The same seed generation and seed source were used for all environments. The RILs were designated with ZD-number codes (e.g., ZD-050) and evaluated alongside their parents under field conditions. Field experiments were conducted during four growing seasons across distinct locations (thereafter called environments): in 2013–2014 and 2014–2015 at CREA-Research Centre for Cereal and Industrial Crops in Libertinia (Southern Italy); in 2014–2015 at CREA-Research Centre for Genomics and Bioinformatics in Fiorenzuola d’Arda (Northern Italy); and in 2020–2021 at Razi University Research Farm in Kermanshah, Western Iran (Table 10). The trial sites represented contrasting climatic conditions relevant to the evaluation of genotype × environment interaction under rainfed production. Libertinia, located in Sicily (Southern Italy), is characterized by a Mediterranean climate with mild, Wetter winters and warmer, drier conditions during the later part of the growing season. Fiorenzuola d’Arda (Northern Italy) represents a cooler and more humid environment with more evenly distributed rainfall during the growing season. In contrast, Kermanshah (Western Iran) represents a semi-arid continental rainfed environment, characterized by lower and less predictable rainfall, greater temperature fluctuation, and a higher likelihood of terminal drought stress. These climatic contrasts were expected to contribute substantially to differences in grain yield and grain protein content across environments. All trials followed a randomized complete block design (RCBD) with three replications under rainfed conditions. Table 10Geographic and environmental characteristics of field trial areas.EnvironmentGPS coordinatesSowing dateTemperature (growing season1)RainfallRelativehumidity(%)Mean1Min2Max3Growing season4Libertinia 201437.529345, 14.576105December 17th12.9 °C6.1 °C19.7 °C327 mm77.6Libertinia 201537.529345, 14.576105December 22th9 °C6.2 °C20 °C335 mm79.7Fiorenzuola d’Arda 201544.928084, 9.91072180ctober 29th8.8 °C4.2° C13.8 °C724 mm86.1Kermanshah 202147.107783, 34.390833November 11th20.4 °C5.5 °C21.7 °C305.6 mm46.71 average temperature; 2mean of Min temperatures; 3mean of Max temperatures; 4November -May. Geographic and environmental characteristics of field trial areas. 1 average temperature; 2mean of Min temperatures; 3mean of Max temperatures; 4November -May. Each plot consisted of five 3-meter rows spaced 25 cm apart, with seeds sown at 350 seeds m⁻². Standard agronomic practices were implemented, including mechanical weed control. Soil management followed conventional practices for each location, though detailed soil characteristics were not the focus of this investigation. At physiological maturity, grain yield (GY) was measured from harvested plot areas, and grain protein content (GPC) was determined using Near Infrared (NIR) spectroscopy on whole-grain samples. Grain yield was evaluated across all four environments, whereas grain protein content was available for only three environments (Libertinia 2013–2014, Libertinia 2014–2015 and Kermanshah 2020–2021), because protein measurements were not obtained for Fiorenzuola d’Arda 2014–2015 experiment. Therefore, stability analyses for grain protein content were based on a smaller environmental dataset than those for grain yield. Also, no systematic disease scoring was performed across environments; therefore, the possible contribution of biotic factors, including fungal disease, to genotype × environment interaction could not be quantified in the present study. Statistical analyses Prior to combined analysis, data assumptions were verified through the Kolmogorov-Smirnov test for normality and Bartlett’s test for homogeneity of variances. A combined analysis of variance (ANOVA) was performed using a mixed model with environments as random effects and RILs as fixed effects. The significance of the RIL × Environment interaction prompted further stability analysis using multiple approaches. Stability indices were calculated using the mean value of each genotype in each environment: Parametric stability statistics included: Environmental Variance (EV)91, Coefficient of Variation (CV)92, Desirability Index (Di)93, Superiority Index (Pi)94, Shukla’s Stability Variance95, Wricke’s Ecovalence (Wi)96, Finlay and Wilkinson’s Regression Coefficient (bi)97, Coefficient of Determination (R²)98, and Eberhart and Russell’s Deviation from Regression (S²di)99. Non-parametric measures comprised: Nassar and Huehn’s statistics (Si⁽¹⁾, Si⁽²⁾, Si⁽³⁾)100, Huhn and Leon’s Si⁽⁶⁾101, Thennarasu’s parameters (Ni⁽¹⁾-Ni⁽⁴⁾)102, and Kang’s Rank Sum41, all calculated from genotype rankings within environments. Multivariate analysis was conducted using the GGE biplot and AMMI models, with genotype stability assessed through the AMMI Stability Value (ASV). Finally, Pearson correlation coefficients were computed between the stability parameters. Combined ANOVA and related parametric and non-parametric analyses were performed in SAS 9.1103,104, descriptive and supplementary analyses were carried out in Minitab105, multivariate analyses including AMMI and GGE biplot were conducted in R 4.5.2106, and Cropstat 7.2107. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements The authors acknowledge financial support from Razi University as part of the PostDoc project. Author contributions S.GH. Analyzed the data, wrote a manuscript draft. K.CH. Supervised the development of work, designed research, and helped in data interpretation and manuscript evaluation. L.Z., R.N., F.D., N.V., F.S., M.P. and E.M. Conducted experiments, contributed new reagents and drafted the manuscript. All authors read and approve the final manuscript. Funding This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Data availability The data that support this study are available in the article and accompanying supplementary tables, and in the requesting the raw data supporting the conclusions will be made available by the authors without undue reservation. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The data that support this study are available in the article and accompanying supplementary tables, and in the requesting the raw data supporting the conclusions will be made available by the authors without undue reservation.