Adaptive and agronomic performances of soybean genotypes derived from different genealogies through the use of several analytical strategies

The objectives of the present study were: (i) to evaluate the genotypic performances of 45 soybean genotypes with the future finality of recommendation of varieties for the State of São Paulo, Brazil; (ii) to determine the stability and adaptability of the genotypes and compare the performance and accuracy of the Wricke’s ecovalence, additive main effects and multiplicative interaction analysis (AMMI), GGE-Biplot and harmonic mean of the relative performance of genotypic values (MHPRVG) methods; (iii) to evaluate the phenotypic, genotypic and environmental correlations among the traits of 45 genotypes in three environments. The exploration of genotype-by-environment interaction (GEI) allowed the identification of 21 genotypes with high mean grain yield, representing different relative maturity groups and stability levels to the environments. This group was subdivided by crop cycle, in which the genotypes 18, 36, 20, 34 and 33 were early cycles (108 to 125 days), while genotypes 11, 22, 44 (CD 219), 24, 23, 14, 32, 1, 12, 39, 30, 38, 7 and 26 were medium cycles (126 to 135 days) and genotypes 25 and 37 were late cycles (≥ 136 days). The interpretations obtained from the ecovalence, AMMI and GGE-biplot methods were more similar than those from the MHPRVG method. This was due to the method’s properties, which assigns more weight to grain yield and little weight to the adaptability and stability parameters. The genotypic and environmental correlations among traits enhanced the interpretations of the genotype x environmental interactions.


INTRODUCTION
Soybean [Glycine max (L.) Merrill] is the most important crop in Brazil due to its large cultivation area in different regions of the country. During the 2012/2013 agricultural year, the soybean sown area was 27.7 million hectares while total soybean grain production was 81.5 million tons. In the State of São Paulo in 2012/2013 agricultural year, the soybean harvested area was 637 thousand hectares with average grain yield of 3,220 kg ha -1 and average grain production of 2.15 million tons (CONAB, 2013a). São Paulo is also the biggest sugarcane producer with 51.7% (4,552 thousand hectares) of planted area in Brazil, in 2013/2014. The sugarcane renewal area is expected to be approximately 630 thousand hectares, an area that usually could be occupied by soybeans (CONAB, 2013b).
The early and medium maturing varieties are more flexible regarding farming operations, and they also have the potential to avoid drought and pest. These soybean varieties are viable alternatives for replacing sugarcane in the renewal areas. Soybean is able to fix a large quantity of atmospheric nitrogen. The nitrogen fixed by the soybean roots, leaves that are shed during harvest contribute greatly to improving the chemical, physical and biological properties of soil. Such improvements in soil properties help to attain high yields in the succeeding crops after the rotation (Singh and Shivakumar, 2010).
Furthermore, demand for soybean remains strong and continues to grow because it is used as an ingredient in the formulation of a multitude of foods and industrial products. The demand for soybean is expected to increase by approximately 140% by 2050 (Bruinsma, 2009). In order to cover this demand and attenuate the effects of climate change on soybean grain yield, it is necessary to improve resource utilization efficiency and grain yield in all regions, whether susceptible to droughts or not.The soybean genetic improvement program of the FCAV/UNESP -Jaboticabal, evaluated a set of 45 genotypes in three environments. The genotypes were selected due to high productivity, desirable agronomic traits and their tolerance/resistance to the principal diseases found in São Paulo. These genotypes belong to different relative maturity groups and their crop cycles make them suitable for the sugarcane-soybean rotation system.
Diverse methodologies were used to evaluate soybean genotypes in multi-environmental trials (MET) (Cruz and Carneiro, 2003;Balzarini and Di Rienzo, 2013;Resende, 2007a). Here, the genotypes were evaluated by four different methodologies as follows: the ecovalence method (Wricke, 1965), which only measures the stability; the AMMI (Gauch and Zobel, 1996) and GGEbiplot (Yan and Kang, 2002) methods which consider simultaneously the grain yield trait as well as the stability and adaptability parameters. And finally, the harmonic median of relative performance of genotypic values (MHPRVG) method, which is strongly correlated with the grain yield trait and little correlated with the stability and adaptability parameters (Resende, 2007b). The last methodology has not been reported yet in previous publications about soybean. Therefore, the genotypes with high grain yield and stability are expected to attain the greatest commercial success.
The objectives of the present study were: (i) to evaluate the genotypic performances of 45 soybean genotypes with the future finality of recommendation of varieties for the State of São Paulo, Brazil; (ii) to determine the stability and adaptability of the genotypes and compare the performance and accuracy of the Wricke's ecovalence, additive main effects and multiplicative interaction analysis (AMMI), GGE-Biplot and harmonic mean of the relative performance of genotypic values (MHPRVG) methods; (iii) to evaluate the phenotypic, genotypic and environmental correlations among the traits of 45 genotypes in three environments.

MATERIALS AND METHODS
The experiment investigated soybean genotypes previously developed in the genetic improvement program of the FCAV/UNESP -Jaboticabal, S. P., Brazil. The genotypes were originated from different types of crosses and then belonging to different relative maturity groups (Kaster and Farias, 2012). In turn, four varieties (CD219, CD216, MG/BR-46 and V-MAX) were used as checks (Table 1). The trials were conducted in three environments in the State of São Paulo, Brazil, in Jaboticabal (2011) and Piracicaba (2012. In Jaboticabal, soybean was sown on Piracicaba (22°42'S latitude; 47°37'W longitude) is located in the soybean macroregion 2, Mid-South, in the edaphoclimatic region 203 while Jaboticabal (21°14'S latitude; 48°18'W longitude) is in macroregion 3, Southeast in the edaphoclimatic region 302 (Kaster and Farias, 2012).
The experimental design was a randomized block with three replications per experiment. The plots consisted of four 5-m long rows spaced 0.5 m. The plot useful area included only the plants in the two central rows. The standard agro-technical practices were applied following the recommendations for the soybean crop (Embrapa, 2011).
The following agronomical traits were evaluated: grain yield (GY; in kg ha -1 ), number of days to maturity (NDM; in days), number of days to flowering (NDF; in days), plant height at maturity (PHM; in cm), plant height at flowering (PHF; in cm), height of first pod insertion (HFPI; in cm), lodging (L; on a visual scale), varying from 1 (all plants of a plot were erect) to 5 (all plants of a plot were lodged) and agronomic value (AV; on a visual scale), ranging from 1 (plants with bad agronomic characteristics) to 5 (plants with excellent agronomic characteristics).
After data acquisition, we performed a combined analysis of variance to examine the main effects of the environment (E) and genotypes (G) and (GE) interaction effects. Where the F-statistics indicated significance, the means were separated using the Least Significant Difference (LSD) test at P = 0.05 in the Info-Gen software (Balzarini and Di Rienzo, 2013;Steel et al., 1997).
In the principal component analysis (PCA), standardized data of quantitative traits (NDF, NDM, PHF, PHM, HFPI and GY) were used. The eigenvalues associated with each eigenvector were represented as the variance of each principal component. Also, the PCA was performed to explore genotypic and environment correlations among quantitative traits of 45 genotypes in three environments (Johnson and Wichern, 2002;Balzarini, 2003). Wricke's (1965) ecovalence (W) measured stability on the basis of the contribution of a genotype to the GEI sums of squares. The more stable genotypes are associated with smaller W i values (Cruz and Carneiro, 2003). Grain yield data were subjected to Wricke's Table 1. Genealogy of the 45 soybean genotypes evaluated by the genetic improvement program of FCAV/UNESP, Jaboticabal.

Genotype
Nomenclature     (Cruz, 2006). The second method called AMMI was used to evaluate the stability and adaptability of grain yield trait (Shafii and Price, 1998;Zobel et al., 1988;Dias et al., 2013). The AMMI-1 analysis was carried out according to Vargas and Crossa (2000) using the SAS software (SAS INSTITUTE, 2003). Subsequently, the GGE-biplot methodology was performed to explore the stability and adaptability of the grain yield trait of the 45 genotypes to all three environments. It removed the environmental main effect (E), focusing the response on the genotype (G) + genotype-by-environment interaction (GEI) (Yan and Hunt, 2002;Yan and Kang, 2002). This analysis was carried out using the Info-Gen software (Balzarini and Di Rienzo, 2013).

RESULTS AND DISCUSSION
Combined analysis of variance identified significant differences (p<0.05) among genotypes and the genotype-by-environment interaction for all traits. Furthermore, significant differences (p<0.05) were evident for the following traits: NDF, NDM, PHF, PHM, HFPI, GY and AV among the different environments. For the lodging trait, the environment effect was not significant (p<0.05) by F-test. In the three environments, the 45 genotypes displayed the average values of 54 days, 127.5 days, 65 cm, 83 cm, 16 cm, 2920 kg ha-1, 1.6 and 3.2 for the traits, NDF, NDM, PHF, PHM, HFPI, GY, L and AV, respectively. The relationship between the largest and the smallest mean square was 7.9 for the grain yield trait (Table 2).
In the principal components analysis, for the environment-genotype combinations, two eigenvalues were greater than one, explaining 69% of the variance contained in the six traits. The first principal component (PC1) kept 39% of the original variance. The NDF and NDM traits, especially, explained this retention of variance with principal component correlation values of 0.56 and 0.55, respectively (Table 3). The second principal component (PC2) retained 30% of the original variance. The PHM and GY traits explained this variance retention with principal component correlation values of 0.65 and 0.59, respectively (Table 3).
In the dispersion graph of the first two components, the traits NDF, NDM and HFPI had a greater inertia on the right, displaying positive correlations amongst them. In Piracicaba, located at higher latitude, the genotypes evinced greater average NDM and NDF while the cycles of theses genotypes were smaller at lower latitude in Jaboticabal. On the other hand, the PHF and PHM traits were positively correlated. Majority of the genotypes exhibited determine growth habits which support the correlation amongst traits (Figure 1).
The GY trait was positively correlated with PHM but not with PHF. Also, GY trait was negatively Table 2. Mean squares derived from ANOVA of measured traits 1 of 45 soybean genotypes in the three environments.
The polygon view of the GGE-biplot showing the three environments, explains 94% of genotype and genotype x environment (G and GE) variation. The vertex genotypes were 3, 16, 27, 9, 18, 11, 24 and 40. In the sector which contained E2 environment, Genotypes, 11, 24 and 18 were vertexes of the polygon. In the sector which contained environment E1, genotype 40 was the vertex of the polygon. Finally, the sector which contained E3 environment, genotype 26 was the one which exhibited high grain yield. Here, GE interaction component was greater than the G component, all environments being in different sectors ( Figure 5). The genotypic correlations among traits of 45 genotypes in the three environments were studied by principal component analysis. The PC1 (65.1%) separated four genotypes (18, 34, 33 and 36) with early cycles of 17 genotypes (11,22,44,24,32,1,39,30,38,12,26,37,23,14,25 and 7) with medium and late cycles. The early cycle Genotype 20 remained on the right side of the graph possibly due to the highest plant height at maturity. The traits HFPI, NDF and GY were separated from traits NDM, PHF and PHM by PC2 (13.9%). GY and NDF traits showed positive genotypic correlation. While NDM trait showed positive genotypic correlation with PHF and PHM traits. Among the medium and late cycle genotypes, nine genotypes (11,22,44,24,32,1,39,30,38) displayed higher mean values for NDF, HFPI and GY than the other eight genotypes (12, 26, 37, 23, 14, 25 and 7). These last genotypes showed higher average values for NDM, PHF and PHM than the first genotypes ( Figure 6).
The environmental correlations among traits of the 45 genotypes in the three environments were also studied by principal component analysis. PC1 (78.9%) separated the E2 environment under good water availability from the E1 and E3 environments under drought stress. Also, PC1 separated GY, PHM and PHF traits from NDF, NDM and HFPI traits. PC2 (21.1%) separated NDF, HFPI, PHM and GY traits from NDM and PHF traits. The E2 environment, with greater water availability had positive environmental correlation with GY and PHM traits, while E3 environment, at higher latitude, had positive environmental correlation with NDM trait and to a smaller degree with the PHF trait. Finally, E1 environment with  drought stress had positive environmental correlation with NDF and HFPI traits (Figure 7). Based on the REML/BLUP mixed model, the contribution of the variance components confirmed that 68.9% was attributed to the environment, 21.1% to the GxE interaction and 10% to the evaluated genotypes. The average genotypic value for the GY trait was μ = 2920 while for E1, E2 and E3 environments the respective values were μ 1 = 2758, μ 2 = 4126 and μ 3 = 1876, respectively. In turn, Neto et al. (2013) reported non-significant variation for rice genotypes.
The MHPRVG method classified the genotypic values of the 21 high average grain yield genotypes in the following descending order (show in parentheses), genotypes 11 (1), 22 (2) Among the twenty high mean grain yield genotypes, 16 resulted from single crosses; two others, from four-way crosses; and two others, from eight-way crosses. Still, a great number of genotypes remained related by two common parents, Embrapa-48 and MG/BR-46. The Embrapa-48 variety belongs to the 6.8 maturity group with the following genealogy (Davis x Paraná) x (IAS 4 x BR5). While MG/BR-46 variety belongs to the 8.1 maturity group and derived from the following genealogy Lo76-4484/Numbaíra.
The narrow genetic base of genotypes was supported by other studies which evaluated soybean genotypes of different improvement programs and releasing periods (Priolli et al., 2004;Priolli et al., 2010) as well as those studies based on the coefficient of percentage (Vello et al., 1988;Miranda et al., 2007).
Regarding the methodologies used, the interpretations obtained from the ecovalence, AMMI and GGE-biplot methods were more similar than those from the MHPRVG analysis. This was due to the method's properties, which assigns more weight to grain yield and little weight to adaptability and stability parameters. The results clearly show the need for further assessments of genotypes in more environments to decrease the coefficient of variation and increase the heritability of traits. More studies should also better define performances such as stability and adaptability of the genotypes, in addition to assess other important agronomic parameters, e.g., seed weight, seed oil and protein content.

Conclusion
The study of the G×E interaction allowed identifying 21 genotypes with high grain productivity, different relative maturity groups and stability levels to different environments. The interpretations obtained from the ecovalence, AMMI and GGE-biplot methods were more similar than the interpretations obtained from the MHPRVG analysis. This was due to the properties of the method which assigns more weight to grain yield and little weight to adaptability and stability parameters. The genotypic and environmental correlations among traits enhanced the interpretations of the genotype × environmental interactions.