Effects of genotype × environment interactions on plant height traits in soybean
In order to evaluate whether G × E impact plant height in a natural soybean population, 308 representative cultivars from a core soybean germplasm collection  were selected and planted at two distinct experimental sites, Boluo (BL, 114.29°E, 23.17°N) and Hainan (HN, 109.48°E, 18.31°N). Three traits related to plant height (SH, shoot height; SNN, stem node number; and AIL, average internode length) were determined from field samples. In these tests, the mean values of SH, SNN and AIL were 81.46%, 34.05% and 36.24% higher, respectively, in BL than that in HN (P value < 0.001) (Fig. 1a, b and c). This demonstrated that the plant height in soybean significantly varied between two distinct environments. Furthermore, genetic analysis suggested that the distributions for the three tested traits measured in two environments were approximately normal according to Kurtosis and Skewness values calculated over three replicates (Table 2). Broad-sense heritability (h2b) for all the traits under the tested environments varied from 0.74 to 0.92, with generally higher values being observed for SH than for the other two traits (Table 2). Regardless of these relatively small differences among traits, the results herein clearly suggest that variation in SH, SNN and AIL depend mainly on genotypic effects in a single environment. Across locations, however, values of h2b for SH, SNN and AIL ranged between 0.38 and 0.40, all of which were significantly lower than in individual environments. Taken together, these results strongly suggest that SH, SNN and AIL are all greatly affected by both genotype and environment. In order to further determine G × E, two-way ANOVA was performed. As expected, the results showed that SH, SNN and AIL were significantly all affected (P value < 0.001) by environment, genotype and G × E (Table 3). However, the environment itself consists of many factors, including temperature, day-length, precipitation, soil properties and so on. To sort through these myriad environmental influences, we further evaluated the effects of several primary environmental factors, along with QTLs and QTL × environmental (QTL × E) on the tested traits. Analyzing specific environmental factors in this way might contribute to breeding soybean with shoot architectures optimized for specific sets of environmental conditions.
Phenotypic variation among recombinant inbred lines
Given the prevalence of G × E identified for soybean in the plant height experiments above, two representative soybean accessions were, therefore, selected for developing a RIL population designed to explore QTL × E more fully in soybean. In addition, field characterizations were performed in an expanded set of four geographically distinct growth environments. In these trials, plant height traits of the parental lines, BX10 with the genotype of E1E2E3E4E9dt1dt2tof11Tof12J and BD2 with the genotype of E1E2E3E4E9Dt1dt2Tof11tof12J, significantly varied across the four tested environments, with observed ranges falling between 33.56 and 122.00 for SH, 9.63 and 23.00 for SNN, and 3.43 and 5.27 for AIL (Table 4). Although there were no significant differences observed between parental lines within individual environments, data from the RIL population exhibited maximum and minimum values beyond the parental extremes, and most of the distributions for traits tested across four environments were approximately normal according to Kurtosis and Skewness values calculated over three replicates (Fig. 2). These results suggest that soybean plant height traits are typical quantitative traits and both parents contain one or more genes contributing additively towards the tested traits. When sites were observed separately, the mean values of SH, SNN and AIL significantly varied in the ranges of 33.20 - 112.39, 10.07 - 22.70 and 3.36 - 5.06, respectively (Fig. 1d, e, f and Table 4), implying large impacts of environmental factors on the tested traits. Furthermore, ANOVA results revealed that the variation observed for SH, SNN and AIL among RILs was significantly affected by environment and genotype, individually or in interaction terms (P value < 0.001) (Table 5). This was consistent with the results obtained from using the core collection germplasm cultivars (Table 3). Overall, the results herein demonstrate that the observed RIL population was suitable for further analysis.
Identification of QTLs contributing to plant height traits
A high-density genetic linkage map consisting of 3319 recombinant bin markers had been constructed using the RIL population developed in a previous study . In order to identify significant QTLs, trait mean values were calculated for each RIL line. Subsequent QTL analysis identified a total of 19 significant loci containing 51 QTLs for the three tested traits, with 23, 13 and 15 QTLs being associated with SH, SNN and AIL, respectively. The LOD values of these QTLs ranged from 2.50 to 16.46, and explained 2.80% - 26.10% of phenotypic variation (Additional file 1: Table S1). Within environments, 13, 16, 13 and 9 QTLs were identified at the Zhao County (ZC, 114.48°E, 37.50°N), Hangzhou (HZ, 120.69°E, 30.51°N), Yangzhong (YZ, 118.20°E, 26.17°N) and BL field sites, respectively. However, only two loci, Loc11 and Loc19-1, containing a total of 20 QTLs, were identified in each of the four distinct environments. Interestingly, the additive effect of Loc11 was derived from BX10 and BD2 as determined in the two southern (including YZ and BL) and two northern (ZC and HZ) experimental stations, respectively. In addition, seven loci (QTLs) were significant only for single trait observed within one of the four tested environments. Other loci contributed to variation in two or more traits and/or at least two environments (Additional file 1: Table S1). The variation in significant QTL numbers and the extent of the additive effects of these QTLs suggests that soybean height QTLs might depend in part on specific environmental conditions present within individual sites, resulting in plant height influenced by genotype, environment, and G × E.
QTL contributions to soybean plant height traits under varied environmental conditions
In order to explore the stability of detected QTL contributions to plant height traits, QTL and plant height data from the four tested environments were subjected to principal components analysis (PCA). In this case, the first two principal components accounted for 44.3 and 25.7% of the total trait variation and QTL additive effects, respectively (Fig. 3a). Traits associated with plant height (SH, SNN and AIL) tended to group together, indicating a high correlation among them. In contrast, the total additive QTL effects for plant height traits (i.e. qSHt, qSNNt and qAILt) tended to group separately, to the extent that nearly 90° angles were observed among the directional vectors (Fig. 3a), which is indicative of these effects acting independently. These results suggest that the detected QTLs do not fully explain the extent of variation in plant height traits observed across varied environments, with the fact that most of these 51 QTLs were not significant in one or more tests reinforcing the conclusion that site specific conditions significantly influenced soybean height outcomes. To test this hypothesis, qSHt, qSNNt and qAILt were replaced by total additive QTL effects (qSHs, qSNNs and qAILs) from the corresponding environments in further PCA. Consistent with the previous PCA results, the first two principal components in this test accounted for 59.2% and 16.8% of the total variation, respectively (Fig. 3b). Besides the vector for qSNNs, the other 5 vectors grouped closely together (Fig. 3b), which suggests, consistent with our hypothesis, that the studied traits are highly correlated. On the other hand, the unexpected PCA results for qSNNs, the vector of which deviated considerably from the vector for SNN, strongly implied that environment differences greatly affected the QTLs for SNN. To minimize environment effects, plant height trait data (SH, SNN and AIL) were replaced by corrected data (SHc, SNNc and AILc) and subjected to PCA again. As expected, the first two principal components accounted for most of the variation, in this case, 42.9% and 24.5% of total variation, respectively (Fig. 3c). Additionally, all three vectors of additive effects (qSHs, qSNNs and qAILs) were relatively close to their corresponding traits (SHc, SNNc and AILc). Taken together, all of the results above strongly indicate that both G × E and QTL × E contribute to plant height phenotypes in the tested soybean population.
Genotype × environmental factor interaction effects on plant height traits expressed in RILs
In order to further evaluate the effects of the main environmental factors on soybean plant height traits, correlation analysis and PCA were conducted with data collected for the tested traits, agro-meteorological factors and basic soil chemical properties. Results from PCA clearly showed that the first two principal components accounted for more than 88% of the total variation, and the vectors of AD and AMaT grouped closely with the vectors of SH, AIL and SNN (Fig. 4a). This suggests that both AD and AMaT contribute to enhance SH, SNN and AIL. Although, AMiT, EAT and AT grouped separately from most of the other vectors, their placement below 90°, implies that these three environmental factors might also enhance SH, SNN and AIL (Fig. 4a). This was further supported by the results from Pearson correlation analysis, in which significant correlations were identified among tested traits and agro-meteorological factors and correlation coefficients varied between 0.220 - 0.827 (P value <0.01) (Table 6). Contrasting results were obtained when no vectors for soil factors grouped closely with SH, SNN or AIL (Fig. 4b). Except for the angle between pH and AN, all other angles between the AP and AK vectors and plant height traits were larger than 90°, which suggests that there were positive or negative interaction effects of pH and AN, or AP and AK on plant height traits (Fig. 4b). This was further confirmed in Pearson correlation analysis, in which significant positive correlations were established for pH and AN, and negative correlations for AP and AK with SH, SNN and AIL (Table 6). These results strongly demonstrate that both agro-meteorological and soil properties influence plant height traits, but the agro-meteorological factors largely predominate.
QTL × environmental factor interactions in RILs
In order to further explore the main factors imparting QTL additive effects, Pearson correlation analysis and PCA were also performed for agro-meteorological factors, soil properties and QTLs additive effects. Here, AD and AMaT closely grouped with qSHs and qAILs, while, AMiT, EAT and AT distributed separately (Fig. 5a), which is consistent with the relationships obtained in PCA of environmental factors and plant height traits (Fig. 4a). Interestingly, qSNNs aligned very closely with AMiT, yet were far from AMaT, suggesting that the additive effects of qSNNs increased with either increases in AMiT or reductions in AMaT. The positive relationship between qSNNs and AMiT, as well as, the negative relationship between qSNNs and AMaT were further confirmed by correlation analysis, in which the Pearson correlation coefficient was 0.491 between qSNNs and AMiT, or -0.263 between qSNNs and AMaT (P value < 0.01) (Table 6). Further evaluation of soil properties and plant height traits showed that qSHs were significantly negatively correlated with AP, but positively correlated with pH. Meanwhile, qSNNs exhibited significant negative correlations with AN, and positive correlations with AK, while qAILs had significant positive correlations with two soil factors (pH and AN), but was negatively correlated with AK (Fig. 5b, Table 6). Taken together, these results demonstrate that both agro-meteorological factors and soil properties can significantly affect the additive effects of QTLs in regulating soybean plant height.