Abstract
In salinity affected areas, variation in salinity level is the major cause of yield fluctuations in rice during the dry season (boro season). To sustain food security in Bangladesh, plant breeders need to develop new rice varieties which are higher yielding, salinity tolerant and stable across different environments. We evaluated the yield performance and stability of 51 rice elite genotypes including two salinity tolerant rice varieties (BRRI dhan67 and BINA dhan-10) and the dominant boro rice variety (BRRI dhan28) in three locations, including a salinity “hotspot”. Best linear unbiased predictions (BLUPs) were used to identify superior genotypes from multi-environment trials. Selection from a large set of candidates is required for evaluation and recommending a new variety for release. Estimated breeding values (EBVs) were used to select parents. Six parents with high EBVs (BR8982-5, IR 87870-6-1-1-1-1-B; BR8943-B-1-2-7, BR8940-B-17-4-7, A69-1 and BR8943-B-5-5-14) would be useful as parents to develop new breeding populations. To evaluate yield performance and stability across environments, we used additive main effects and multiplicative interaction (AMMI) model using a randomized complete block design with two replications. Genotype × environmental analysis was performed using GGE biplot analysis. Based on yield performance and stability, BR8982-5, IR 87870-6-1-1-1-1-B, BR8943-B-1-2-7, BR8940-B-17-4-7, A69-1 and BR8943-B-5-5-14 were promising across the tested environments, as they showed yield advantages over check varieties. The results in this study will be useful for selecting the elite lines suitable for salinity affected areas and parents for rapid breeding of salinity tolerance.
-
Key words: Genetic gain, AMMI Model, Genotype × environment (GE) analysis, Stability, Best Linear Unbiased Prediction (BLUP), Estimated Breeding Value (EBV)
INTRODUCTION
Rice production is critical to food security in Bangladesh. The dry or
boro season is important for rice production, because the highest yields are obtained during this season. However, Bangladesh is facing an enormous threat of salinity intrusion in arable land because of global warming. The salinity stress affected land area and human population size are increasing every day. To feed the rapidly growing population of Bangladesh, salinity affected areas must be used for rice cultivation, as rice is the staple food (
Jagadish et al. 2012;
Rahman et al. 2016;
Rahman et al. 2019). The present trend of genetic gain is incremental and may not be sufficient meet the requirement of the continuously increasing population. Innovation in rice breeding will be required to improve the current relatively low rate of genetic gain for rice yield (BRRI 2018).
Genetic gain is a concept in breeding and classical biometrical genetics where the rate of performance is compared to a baseline. It is usually evaluated after advancing one generation to the next generation that is done by artificial selection (
Falconer and Mackay 1996). However, long term estimates of genetic gain are measured by comparing varieties together, including base or dominant varieties (
Peng et al. 1999).
Yield is a complex trait and depends on a number of genetic and environmental factors. Estimation of the heritable component can be done by using a technique called best linear unbiased prediction (BLUP) that maximizes the association between true and predicted breeding values through minimizing predictive error variance (
Robinson 1991;
Rutkoski et al. 2014). BLUPs are used to select the best inbred lines to form new breeding populations when combined with genetic relatedness or kinship information (i.e. coefficient of coancestry), sometimes designated as pedigree BLUPs which can be used to provide estimated breeding values (EBVs). This is a measure of the genetic capability of a parent to produce better progeny based on the parent’s own performance, pedigree records and progeny data. EBVs are extremely useful for parent selection for increasing genetic gain (
Bernardo 2010). Surprisingly though, the use of pedigree information has been extremely limited in public rice breeding programs in Asia to improve complex traits like yield.
Crop yield variation results from environmental effects and fluctuations in soil attributes and the inherent potential of the genotypes. The evaluation of genotype × environment (G × E) interaction is important in plant breeding and new crop variety introductions (
Prasad and Singh 1990;
McLaren and Chaudhary 1994). The selection process becomes difficult when genotype × environment (G × E) interactions reduce associations between phenotypic and genotypic factors (
Al-Naggar et al. 2018). Among the methods to evaluate G × E interactions, additive main effects and multiplicative interaction (AMMI) model (
Zobel et al. 1988;
Crossa et al. 1990;
Ebdon and Gauch 2002) and GGE-biplot have been widely-used to explore G × E (
Yan et al. 2007). AMMI separates the genotype and environment main effects and the GEI effects (
Gauch et al. 2008) and provides insight into GEI (
Crossa et al. 1990). The GGE biplot shows the relationship between genotypes and environments for selected traits graphically by use of a genotype plus genotype by environment (GGE) biplot that allows visual assessment of genotype by environment interaction (GEI) pattern of multi-environment data (
Yan et al. 2000;
Yan 2001;
Yan and Hunt 2002;
Yan and Rajcan 2002). The model produces a “which-won-where” pattern and mean performance and stability of genotypes, discriminating ability, mega-environment investigation, and representativeness of environments. Recently, this analysis has been used to identify mega-environments for salinity stress in India (
Krishnamurty et al. 2017).
Currently, breeders’ efforts are focused on developing new elite breeding lines that can replace the existing varieties in terms of superiority in salinity stress tolerance, higher yield with good stability. Furthermore, more rapid and effective breeding methods are needed to mitigate adverse effects from climate change (
Atlin et al. 2017). In a previous study, a set of five to eight released Bangladeshi varieties were extensively tested for their adaptation to coastal regions of Southern Bangladesh across six locations for four years (
Islam et al. 2015). Our study aimed to identify high yielding, stable genotypes under salinity stress and estimate genetic gain of the current breeding program and estimated breeding values to make selection decisions. We tested a set of advanced breeding lines and varieties for grain yield in a range of salinity stress environments including a ‘hotspot’ and a favorable environment to identify high yielding and stable rice genotypes adapted to diverse environments. The salinity hotspot was representative of vulnerable rice production areas in southern coastal zones of Bangladesh where salinity gradients vary from low (EC: 3.5 dS/m) to very high (EC: > 15.0 dS/m) covering > 1.0 M ha. These areas have limited varietal options and so require improved, higher-yielding varieties to boost national production.
MATERIALS AND METHODS
Plant materials
A total of 48 recent, elite breeding lines (18 lines from IRRI and 30 BRRI developed breeding lines) along with two BRRI released varieties (BRRI dhan28: released in 1994, BRRI dham67: released in 2014) and one BINA released variety (BINA dhan-10: released in 2012) were evaluated in the boro season. Genotypes with diverse pedigrees were selected from previous observational trials based on yield performance and salinity tolerance data that were included in this study. Many common salinity tolerant parents appear in the parentage.
Experimental design and cultural practices
The experiments were carried out at BRRI headquarters, Gazipur and farmer’s fields at Assasuni Upazila of Satkhira and Koyra Upazila of Khulna district in Bangladesh. Gazipur, Assasuni and Koyra are favourable, medium-stress and high-stress sites, respectively (
Table 1). The trials were conducted during the dry season (boro 2016–17), and information for each trial is presented in
Table 1. The experimental design used was a randomized complete block design (RCBD) with two replications. The plot size was 5.4 m length × 10 rows. (10.8 m
2). Forty-day-old seedlings were transplanted @ 2–3 seedlings per hill with 20 × 15 cm spacing. Fertilizers dose 120:19:60:20:3.6 kg NPKSZn/ha (260-97-120-110-11 kg/ha Urea-TSP-MoP-Gypsum-ZnSO
4, respectively) were applied with split application of N at 15, 30, 50 days after transplanting (DAT). Total P K S Zn were applied at the time of final land preparation. Irrigation and weed control strategies were applied according to the requirement.
Analysis of variance of the RCBD was performed for each of the three environments on the basis of individual plot observation using the STAR 2.0.1 software. Combined analysis of variance across the three environments was also performed using the STAR 2.0.1 software. Genotypes were considered as random effects in order to calculate BLUPs for yield using PBTools 1.4. software. Both software programs are open-source and were developed by the IRRI biometrics team (
http://bbi.irri.org/products).
Due to the parentage of the germplasm used, we were able to determine relatedness based on pedigrees. We used R (version 3.5.1) to analyze the relationship matrix, inbreeding coefficient for each individual and estimated breeding values and plots. Additive relationship matrix/kinship coefficient (or coefficient of parentage: CoP) was calculated using pedigree information. The ‘editPed’ function used to construct pedigrees using R package ‘pedigreemm’ (
Vazquez et al. 2010;
Bates and Vazquez 2014). Data was coded following the animal model of dam (female/mother) and sire (male/father). We know coefficient of relationship (
r) is double the coefficient of kinship (i.e.
r =2* the coefficient of kinship). For example, the value of CoP is 0 (sparse matrix or sparse array) when crossed between unrelated parents/individuals and similarly this will be 0.5 when no inbreeding occurred for a parent/individual with themselves, 0.25 between mother and child, 0.125 between an uncle and niece, etc.
Inbreeding coefficients were measured by subtracting additive relationship matrix from diagonal value. Estimated breeding values assessed on the basis of own performance of individual plant/parent, pedigree records and progeny data. We calculated the reliability of EBV derived from prediction error variance (PEV) as square root of 1 minus the ratio between PEV and additive genetic variance (VA) i.e. rel < –sqrt(1-PEV/VA).
We also estimated the predicted genetic gain or expected genetic advance under selection (
Yan 2014). Genetic gain of a crop breeding program is calculated by the following equation:
Where i = selection differential, h2 is the heritability and σP is the phenotypic variance.
We used META-R (Multi Environment Trial Analysis with R for Windows), Version 6.0 (2016) to calculate h2 and σP.
AMMI and GGE biplot analysis
Grain yield data for dry season in 2016–17 at three locations were used for combined analyses via additive main effect and multiplicative interaction (AMMI) analysis of variance and genotype effect and genotype × environment interaction effect (GGE). AMMI uses ANOVA to analyze the main effects (additive part) and principal component analysis (PCA) to analyze the non-additive residuals by the ANOVA (
Gauch 1993). The factor explained (%) was calculated comparing sum of square (SS) from AMMI ANOVA. When a genotype and environment have the same sign on their respective first PCA axis, their interaction is positive; if different, their interaction is negative (
Tariku et al. 2013). Genotypes or environments with large IPCA1 scores, either positive or negative had large interactions, whereas genotypes with IPCA1 score of zero or nearly zero had smaller interactions (
Crossa et al. 1990). The biplot of the first two IPCA axes demonstrates the relative magnitude of the GEI for specific genotypes and environments.
The GGE concept was used to visually analyze the yield from multi-environment trial (MET) data. This methodology uses a biplot to show the factors (G and GE) that are important in genotype evaluation and that are also sources of variation in GEI analysis of MET data (
Yan et al. 2000;
Yan 2001). GGE biplot symmetric view was used in this study to explain the ‘which-won-where’ patterns for genotypes and environments. Different polygons composed of one or several environment(s) and one or more genotype(s) can be used to determine which genotypes perform best in specific environments.
AMMI stability values (ASVs) were used to calculate yield stability following the method described by
Purchase et al. (2000). ASVs were derived from the AMMI model and calculated for each genotype and each environment according to the relative contribution of IPCA1 to IPCA2 using the formula:
The larger the ASV value, either negative or positive, the more specifically adapted a genotype was to a certain environment. A smaller ASV value indicates a more stable genotype across environments (
Purchase et al. 2000).
RESULTS
Salinity and trial locations
The level of salinity was measured at seven-day-intervals starting from the date of transplanting to flowering stage of the genotypes and the degree of salinity varies depending on season, temperature and precipitation. Higher salinity levels were observed in Koyra than Assasuni. The salinity level of Koyra experimental plot ranged from 7.5 dS/m to 13.2 dS/m whereas, 3.73 dS/m to 4.6 dS/m was observed in Assasuni (
Supplementary Fig. S1). All the three locations were positively correlated. Strong positive correlation (0.64) observed between E1 and E2 however correlation between E1 and E3 (0.15); E2 and E3 (0.12) are also positive but weak (
Supplementary Table S1) indicating that both E1 and E2 are low saline environments and E3 (high saline) is different from E1 and E2. It was somewhat unexpected that the correlation between E1 and E2 was so low, given that these trial locations were in adjacent districts.
Genotype number, designation and parentage information are shown in
Table 2. Moreover genotypes were ranked based on predicted mean values of yield (
Table 2). The top 10 genotypes based on BLUP values for yield were BR8982-5 (G7), IR87870-6-1-1-1-1-B (G3), BR8943-B-1-2-7 (G4), BR8940-B-17-4-7 (G11), BR8987-6 (G9), BR 8943-B-1-1-2 (G12), A69-1 (G2), BR8943-B-5-5-14 (G6), BINA dhan-10 (G51), IR58443-6B-10-3 (G1). BINA dhan-10 was the highest ranking check variety. The ranks for the other check varieties, BRRI dhan67 and BRRI dhan28, were 14 and 37 respectively.
This is one of the few attempts to estimate pedigree-based breeding values in rice. Ancestry information (parentage) and pedigree record (number of generations selfed) of each elite breeding line and variety was required for pedigree file preparation. Kinship/relationship matrix/ coefficient was estimated based on related, unrelated individuals/parents and their foundation/original parents. Also genetic, phenotypic, and pedigree information of IRRI germplasm was combined and stored in database and provided a specific GID for each germplasm accession including all pedigree information (pedigree file;
http://irri.org/). This GID linked to phenotypic data using R package and estimate breeding values. Moreover, we started with specific elite line/variety and traced it to foundation/parents based on the available ancestry information and selfed generations and prepared pedigree file information. This pedigree file used for kinship matrix analysis and along with phenotypic data used in breeding value estimation.
We found that the studied genotypes had estimates of inbreeding coefficients (CoI) from 0.001 to 0.97 (
Supplementary Fig. S2). Lower the CoI give higher genetic distance. Genetic distance based on inbreeding coefficient between the parents may be used as one of criteria for parent selection. The magnitude of the additive relationship matrix/kinship matrix/coefficient-CoP is 0 (i.e. off-diagonal value) indicating crossing between unrelated individuals and the EBV may be predicted only from the performance of each individual (inbred) by itself. However, the diagonal elements in the matrix have values more than 1.0 suggesting breeding between related parents/same ancestor (consanguineous mating) in the kinship. The non-zero off-diagonal values are the information of kinship (CoP) ranging from 0.125–1.9375 (
Supplementary Table S2). The additive relationship matrix shows a number of small groups of closely related individuals, indicating family structure (
Supplementary Fig. S3). Based on pedigree data, 51 individuals were derived from 24 full-sib families. Individuals resulting from the same full-sib family formed a group together based on the relationship matrix (
Supplementary Fig. S3). The heatmap that was generated can be helpful in selection of distantly related parents for breeding.
Estimated breeding values for all genotypes ranged from –1.00 to 1.36 t/ha. The highest estimated breeding values were obtained for G3: IR 87870-6-1-1-1-1-B (1.36 t/ha) and the lowest was for G36: IR93915-82-CMU2-2-CMU3- AJYB (–1.00 t/ha) with reliability of 65–75% (
Table 2,
Fig. 1). The EBVs for 15 genotypes was between 0.1 to 0.5 and was between –0.1 to 0.5 for 14 other genotypes. The heritability for yield was estimated and the genetic correlations are also important. Reliability is an important factor for measuring accuracy of estimated breeding values of each genotype (
Gorjanc et al. 2015). Accuracy (r) is correlation between EBV and true breeding value (TBV) and reliability is the squared figure of accuracy (r
2). Accuracy was moderate since reliability is 65–75%. Accuracy (%) refers to yield performance for the rice genotypes and their close relatives.
Genetic gain predicted for grain yield were used to explore the amount of genetic gain expected solely from selection. This refers to selecting and advancing the top proportion of lines (out of 51). For prediction of genetic gain is the calculations was as follows:
For our test population predicted genetic gain at 10% selection intensity is calculated.
AMMI analysis
The AMMI analysis of variance for grain yield of rice genotypes across environments showed highly significant main effects (
P < 0.01) of genotypes, environments and GE interaction (
Table 3). Of the total variation, 6.71% contributed by the environment and GE interaction contributed for 39.2% of the total variation. Genotypic main effect accounted for 43.6% and the residual variance (i.e., error variance) captured 10.2% of the total variation. The first two significant IPCAs accounted for 100% of the total GE interaction (
Table 3). The breeding line BR8982-5 (G7), followed by IR87870-6-1-1-1-1-B (G3), BR8943- B-1-2-7 (G4), G11 had the highest overall yields across environments, and showed generally adaptation to the test environments because the PCA1 values were close to zero (
Fig. 2). In contrast G39 followed by G35, G36, G38 and G32 had the lowest overall yield across environments and were inferior to the check varieties (
Table 2,
Fig. 2).
Selection of genotypes based on AMMI stability value (ASV) is important because the most stable genotypes may not always produce the highest yield conversely genotypes selected based on ASV may show consistent performance and general adaptability. The AMMI stability values (ASVs) of 51 genotypes and IPCA1, IPCA2 scores are presented in
Tables 2 and
3, respectively. Genotypes with the lowest ASV and IPCA scores were considered the most stable. An ideal genotype should have a high mean grain yield and small ASV. Accordingly, G19 and G48, showed the lowest ASVs (0.11 and 0.14), respectively and moderate grain yield (mean grain yield: 5.37 and 5.38 t/ha), respectively (
Table 2). Furthermore, G3 was the highest yielding genotype (6.80 t/ha) with a relatively low ASV (0.36). These results revealed that those genotypes are showing relatively better stability than the rest of genotypes. The genotypes G7, G9, G12, G51 were among the top 10 high yielding genotypes (6.82, 6.16, 6.05, 5.87 t/ha, respectively), but had high ASV (1.42, 2.91, 2.19, 0.96, respectively) were identified as potentially useful genotypes for the
boro season.
The GGE biplot analysis clearly indicated relationship between locations, which was consistent with the correlation analysis that E1 and E2 were highly correlated. E2 and E3 had longer vectors, whereas E1 has a shorter vector (
Fig. 3). The average environment has the average coordinates of all environments. The average-environment axis (AEA) is the line that passes through the average environment and the biplot origin (
Fig. 4). Among the three environments, E1 was the most representative environment, but it is not a good test environment. E2 is more discriminating and representative compared to E3 and E1.
One of the most useful features of the GGE biplot analysis was the “which-won-where” pattern. Genotypes were located in seven sectors of which our three test environments located in three separate sectors. It indicated that our test environments were different from one another, but did not form groups. Grouping environments is one of the useful features of GGE biplot analysis. Genotypes placed on the edges of the polygon are the best yielding genotypes in one or multiple environments situated within a specific sector (
Fig. 5). G7, G3, G4 and G11 were the winning genotypes in environment E1, G7 and G9 were the “winners” in E2, and G31 and G3 were the “winners” in E3.
DISCUSSION
The
boro or dry season for rice production in Bangladesh is extremely important because of the high yields during this season. Of the relevant environmental factors, soil and water salinity are the most important and there is an obvious negative correlation between yield and salinity (
Islam et al. 2015). Higher yields were clearly obtained in the non-salinity or trace saline levels (EC: 0.7–1.0 dS/m in Gazipur), 3.73 dS/m to 4.6 dS/m salinity at Assasuni and 7.5 dS/m to 13.2 dS/m salinity at Koyra. Rigorous evaluation and adaptation to hotspot areas were recommended to test the suitability of new genotypes across the regions (
Islam et al. 2015).
BRRI dhan28 has been a dominant variety for the dry season in Bangladesh since its release several decades ago. The effect of salinity on yield performance of BRRI dhan28 was clearly visible. Surprisingly, the yield of BRRI dhan28 (which is sensitive to salinity) was not the lowest yielding genotype in the salinity affected areas. BRRI dhan28 produces much lower yield in these areas, however farmers cultivate this variety for its good grain quality (i.e. medium slender grains), shorter growth duration, taste and market value for a long time. It is possible that its earliness may enable this variety to partially avoid saline stress during booting/flowering stage.
The most important finding from this study was the identification of several elite breeding lines that out-yielded BRRI dhan28 and other salinity tolerant check varieties BRRI dhan67 and BINA dhan-10. These genotypes included BR8982-5 (G7), IR 87870-6-1-1-1-1-B (G3) and BR8943-B-1-2-7 (G4) that produced the best overall yield. Interestingly, IR87870-6-1-1-1-1-B was developed using rapid generation advance (RGA) or single seed descent, providing further evidence that the RGA method is an effective rice breeding method (
Collard et al. 2017).
Stability in addition to high grain yield is important for selection (
Naroui Rad et al. 2013). We used the ASV measure of stability described by
Purchase et al. (2000) because of the simplicity and suitability of this stability measure compared to other measures, especially since it is derived from AMMI information. Higher stability is observed when ASVs are smaller and is useful to combine with overall yield performance, as has been performed in bread wheat (
Farshadfar 2008). In multi-location yield trials, finger millet genotypes with high grain yield but high ASV were recommended for specific adaptability (
Lule et al. 2014).
AMMI and GGE biplot analysis were used to explore G × E interactions. Although the site mean yields differed across the three locations used in this study, our results indicated that there were no G × E crossover interactions. A large percentage of G × E interaction was explained by IPCA-1 (75.5%) followed by IPCA-2 (24.5%) (
Table 3). The (G × E) interaction effect was rich in the first two IPCA scores (100%) interpreting the magnitude of (G × E) interaction effect on yield. Two principal components (IPCA-1 and IPCA-2) were used to produce a 2-dimensional GGE biplot. The most accurate AMMI model can be enumerated by utilizing the first two IPCAs (
Gauch and Zobel 1996). In most cases, the first IPCA explains most of the variation such as for maize (
Ndhlela et al. 2014), bread wheat (
Asnake et al. 2013), common bean (
Abeya et al. 2008) and finger millet (
Lule et al. 2014).
To break the yield barrier of Bangladeshi rice varieties in the future, we need to implement new methods to develop new varieties that can out-yield the current dominant varieties. One novel approach we used in our study was to incorporate pedigree information to calculate EBVs. EBVs are based on additive effects and provide an indication of the best lines to use for crossing to develop new breeding populations. The high EBVs of genotypes G7 (BR8982-5), G3 (IR87870-6-1-1-1-1-B), G4 (BR8943-B-1-2-7), G11 (BR8940-B-17-4-7), G2 (A69-1), G6 (BR8943-B-5-5-14) in our study indicate that these would be ideal parents for crossing. In this study, we derived pedigree information by manually coding pedigree relationships. This emphasizes the importance for breeding programs to maintain accurate and extensive pedigree records. However, for routine implementation, it would be more efficient if pedigree relationships could be generated from a database system. For example, most database systems have tools to generate coefficient of parentage (
McLaren et al. 2005).
An even more accurate and efficient approach would be to use DNA markers to estimate genetic relatedness and implement genomic selection (GS;
Heffner et al. 2009;
Heffner et al. 2010;
Heslot et al. 2015). The use of DNA markers permits the calculation of genomic-estimated breeding values (GEBVs) and subsequent selection of un-phenotyped material (also referred to as “genome-wide prediction), permits substantial gains in the efficiency of selection. GS may be more effective for quantitative traits such as yield, for which MAS has not been effective. The other advantage of using genomic selection is that the breeding cycle can be shortened, thereby increasing the rate of genetic gain. The first GS pilot in rice utilized elite indica breeding lines from IRRI’s irrigated breeding program was recently reported (
Spindel et al. 2015). Prediction accuracies were moderate for plant height (0.34) and flowering time (0.63), and moderately-low for grain yield (0.31). Another empirical test of GS in rice was reported by
Grenier et al. (2015) who used inter-related synthetic populations developed using recurrent selection in an upland rice breeding program. Four agronomic traits (flowering time, plant height, grain yield and panicle weight) were evaluated, and the accuracy of GEBVs ranged from 0.12 to 0.54. Recently, GS has also been applied for investigating the effectiveness to predict rice blast disease resistance (
Huang et al. 2019). Although there is considerable further optimization and validation required for GS in rice, this approach represents an exciting new area of applied breeding research.
However, despite the promise of this new technology, there are numerous challenges (
Nakaya and Isobe 2012) including lack of operating funds for genotyping and insufficient database systems with analytical tools to implement routine GS in public rice breeding programs in Asia. Until GS can be routinely implemented, the most effective breeding strategy would be to continue to use pedigree information and faster breeding methods such as rapid generation advance, which has already been initiated at BRRI (
Collard et al. 2017).
In conclusion, we have identified several promising new breeding lines for the boro rice season. The best breeding lines will be further tested across seasons and locations and undergo further grain quality evaluation. The use of EBVs and estimation of predicted genetic gain represent new a new awareness regarding the importance of genetic gain. Decisions about parent selection based on EBVs and conducting advanced trials in salinity-affected hotspots represent significant improvements to BRRI’s rice breeding program that will ultimately enable new, higher-yielding rice varieties to be developed for Bangladesh.
Supplementary Information
ACKNOWLEDGEMENTS
We thank Dr. Mohammad Rafiqul Islam and Md. Abdul Qayum, BRRI (
www.brri.gov.bd), Gazipur 1701 for their useful suggestions. This work was supported by TRB-BRRI Project study on “Product development for replacing ruling rice variety in coastal saline zone using cutting-edge breeding” funded by the Bill and Melinda Gates Foundation (BMGF).
Fig. 1Histogram showing frequency distribution of 51 rice genotypes for estimated breeding values (EBV) where zero (0.0) indicates population mean (μ) and each positive value denotes the amount of yield (t/ha) will be increased (μ + EBV) in their progeny in the next generation if these parents use in cyclic breeding program.
Fig. 2Genotypes grain yield vs PC1 (AMMI plot).
Fig. 3GGE biplot (Environment view for yield).
Fig. 4GGE biplot (Environment view for yield).
Fig. 5Polygon view of the GGE biplot based on symmetrical scaling for ‘which-won-where’ pattern of dry season rice genotypes and environments.
Table 1Location of the experiment.
Table 1
|
Upazila |
District |
Longitude E |
Latitude N |
CV (%) |
Mean yield (t/ha) |
IPCA1 |
IPCA2 |
|
Assasuni (E1) |
Satkhira |
89°08′58.25″ |
22°34′37.51″ |
9.42 |
5.25 |
0.41 |
0.03 |
|
Gazipur Sadar (E2) |
Gazipur |
90°24′08.59″ |
23°59′ 25.01″ |
8.62 |
5.54 |
0.90 |
−0.14 |
|
Koyra (E3) |
Khulna |
89°19′46.80″ |
22°27′13.70″ |
8.39 |
4.90 |
0.12 |
0.99 |
Table 2Mean grain yield (t/ha), predicted mean, EBV and reliability of 51 rice genotypes under each of the three environments.
Table 2
|
Geno code |
Designation |
Parentage |
Rank based on BLUP |
Yield (t/ha) |
ASV |
IPCA1 |
IPCA2 |
EBV (t/ha) |
CoI |
Rel |
|
|
Assasuni (E1) |
Gazipur (E2) |
Koyra (E3) |
Predicted mean |
|
G1 |
IR58443-6B-10-3 |
AT 401/IR31868-64-2-3-3-3 |
10 |
6.43 |
6.81 |
4.21 |
5.55 |
1.70 |
0.54 |
0.35 |
0.85 |
0.47 |
0.70 |
|
G2 |
A69-1 |
BG 94-1/POKKALI |
7 |
5.58 |
7.05 |
5.52 |
5.68 |
0.79 |
0.24 |
−0.31 |
0.47 |
0.97 |
0.75 |
|
G3 |
IR87870-6-1-1-1-1-B |
AT 401/CSR 28 |
2 |
6.52 |
7.46 |
6.44 |
6.10 |
0.36 |
0.10 |
−0.19 |
1.36 |
0.71 |
0.75 |
|
G4 |
BR8943-B-1-2-7 |
BRRI dhan47/IR69337-AC2-2-2 |
3 |
6.35 |
7.67 |
6.04 |
6.03 |
0.85 |
0.27 |
−0.23 |
1.13 |
0.75 |
0.75 |
|
G5 |
BR8943-B-4-3-9 |
BRRI dhan47/IR69337-AC2-2-2 |
15 |
4.61 |
6.82 |
4.97 |
5.36 |
1.13 |
0.32 |
−0.56 |
0.002 |
0.50 |
0.75 |
|
G6 |
BR8943-B-5-5-14 |
BRRI dhan47/IR69337-AC2-2-2 |
8 |
5.62 |
6.71 |
5.55 |
5.63 |
0.48 |
0.14 |
−0.23 |
0.63 |
0.97 |
0.75 |
|
G7 |
BR8982-5 |
BR47/Pokkali 15661 |
1 |
6.52 |
8.14 |
5.81 |
6.11 |
1.42 |
0.46 |
−0.21 |
0.87 |
0.97 |
0.75 |
|
G8 |
BR8982-9 |
BR47/Pokkali 15661 |
17 |
5.61 |
6.54 |
4.16 |
5.34 |
1.46 |
0.47 |
0.08 |
0.58 |
0.97 |
0.75 |
|
G9 |
BR8987-6 |
BR29/FL478 |
5 |
5.99 |
8.32 |
4.19 |
5.74 |
2.91 |
0.94 |
−0.15 |
0.55 |
0.97 |
0.75 |
|
G10 |
BR8992-10 |
BR47/FL478 |
29 |
5.81 |
6.17 |
3.25 |
5.15 |
1.97 |
0.62 |
0.43 |
0.56 |
0.97 |
0.75 |
|
G11 |
BR8940-B-17-4-7 |
IR72593-B-3-2-2-2/BRRI dhan47 |
4 |
6.53 |
7.25 |
5.85 |
5.95 |
0.64 |
0.21 |
−0.03 |
1.18 |
0.48 |
0.75 |
|
G12 |
BR8943-B-1-1-2 |
BRRI dhan47/IR69337-AC2-2-2 |
6 |
6.52 |
7.44 |
4.21 |
5.68 |
2.19 |
0.71 |
0.26 |
0.90 |
0.90 |
0.75 |
|
G13 |
BR8943-B-20-9-22 |
BRRI dhan47/IR69337-AC2-2-2 |
33 |
5.74 |
4.82 |
4.41 |
5.1 |
0.48 |
−0.05 |
0.45 |
−0.201 |
0.90 |
0.75 |
|
G14 |
IR86385-85-2-1-B |
IRRI 149/IR61920-3B-22-2-1 (NSIC RC 106) |
27 |
5.66 |
4.1 |
5.57 |
5.16 |
1.78 |
−0.57 |
0.35 |
0.07 |
0.89 |
0.75 |
|
G15 |
IR83484-3-B-7-1-1-1 |
IRRI 113/BR 40 |
41 |
5.61 |
4.7 |
3.59 |
4.9 |
0.73 |
0.14 |
0.59 |
−0.13 |
0.97 |
0.75 |
|
G16 |
IR87872-7-1-1-2-1-B |
AT 401/IR73571-3B-14-1 |
39 |
5.13 |
4.29 |
4.67 |
4.94 |
0.88 |
−0.27 |
0.27 |
−0.33 |
0.97 |
0.75 |
|
G17 |
IR86385-117-1-1-B |
IRRI 149/IR61920-3B-22-2-1 (NSIC RC 106) |
43 |
4.99 |
3.72 |
4.78 |
4.83 |
1.44 |
−0.46 |
0.31 |
−0.45 |
0.97 |
0.75 |
|
G18 |
BR8992-B-18-2-26 |
BR47/FL478 |
12 |
5.86 |
5.28 |
5.73 |
5.45 |
0.91 |
−0.29 |
0.14 |
0.47 |
0.97 |
0.75 |
|
G19 |
BR9154-2-7-1-2 |
Bhojon/Nonabokra |
20 |
5.42 |
5.73 |
4.96 |
5.31 |
0.11 |
0.03 |
0.02 |
0.17 |
0.83 |
0.75 |
|
G20 |
BR9156-4-3-2-22 |
Bhojon/BRRI dhan47 |
13 |
4.83 |
5.85 |
5.97 |
5.41 |
0.80 |
−0.22 |
−0.45 |
0.39 |
0.91 |
0.75 |
|
G21 |
BR8964-3-2-3-12 |
BR28/Pokkali 15661 |
28 |
4.89 |
5.62 |
4.81 |
5.16 |
0.20 |
0.04 |
−0.15 |
0.001 |
0.97 |
0.73 |
|
G22 |
BR8967-2-1-3-6 |
BR28/BR47 |
19 |
4.19 |
7.06 |
4.87 |
5.31 |
1.47 |
0.41 |
−0.76 |
0.25 |
0.97 |
0.75 |
|
G23 |
BR9144-4-3-2-17 |
BRRI dhan50/BRRI dhan47 |
42 |
4.9 |
5.31 |
3.61 |
4.89 |
0.91 |
0.29 |
0.16 |
−0.10 |
0.97 |
0.75 |
|
G24 |
BR9144-2-3-1-18 |
BRRI dhan50/BRRI dhan47 |
30 |
4.83 |
4.97 |
5.26 |
5.11 |
0.80 |
−0.26 |
−0.12 |
−0.035 |
0.97 |
0.75 |
|
G25 |
BR9145-5-2-7 |
BRRI dhan29/BRRI dhan47 |
21 |
4.93 |
4.89 |
6.14 |
5.28 |
1.61 |
−0.52 |
−0.24 |
0.11 |
0.97 |
0.75 |
|
G26 |
BR9152-B-2-3-1 |
BRRI dhan29/Nonabokra |
45 |
4.51 |
4.9 |
4.01 |
4.81 |
0.22 |
0.07 |
0.01 |
−0.52 |
0.97 |
0.71 |
|
G27 |
BR9152-1-3-1-8 |
BRRI dhan29/Nonabokra |
25 |
5.4 |
5.57 |
4.67 |
5.22 |
0.24 |
0.07 |
0.10 |
0.05 |
0.97 |
0.75 |
|
G28 |
BR9154-3-2-4-7 |
Bhojon/Nonabokra |
32 |
5.1 |
4.81 |
5.08 |
5.1 |
0.75 |
−0.24 |
0.06 |
−0.11 |
0.90 |
0.75 |
|
G29 |
BR9156-5-3-4-15 |
Bhojon/BRRI dhan47 |
24 |
4.91 |
6.31 |
4.44 |
5.22 |
1.04 |
0.33 |
−0.21 |
0.14 |
0.97 |
0.75 |
|
G30 |
IR89330-14-3-1-2-2-3 |
AT 401/2*IR 61920-3B-22-2-1 (NSIC RC 106) |
44 |
5.01 |
4.35 |
4.1 |
4.82 |
0.44 |
−0.10 |
0.32 |
−0.50 |
0.97 |
0.75 |
|
G31 |
IR89331-32-3-1-3-2-2 |
AT 401/2*IR 73571-3B-14-1 |
31 |
4.36 |
4.13 |
6.54 |
5.11 |
2.59 |
−0.83 |
−0.39 |
−0.34 |
0.97 |
0.75 |
|
G32 |
IR91715-8-1-1-1 |
A 69-1/4*IR03A477 |
47 |
4.42 |
3.9 |
4.99 |
4.79 |
1.45 |
−0.47 |
−0.01 |
−0.47 |
0.97 |
0.75 |
|
G33 |
IR91715-8-1-1-AJY1 |
A 69-1/4*IR03A477 |
46 |
4.95 |
3.87 |
4.55 |
4.8 |
1.12 |
−0.35 |
0.30 |
−0.50 |
0.97 |
0.65 |
|
G34 |
IR91820-25-BAY2-3 |
A 69-1/AGAMI MONT 1/A 69-1/IR05A125 |
35 |
4.90 |
5.21 |
4.81 |
5.09 |
0.21 |
−0.06 |
−0.06 |
−0.13 |
0.97 |
0.75 |
|
G35 |
IR92860-33-CMU1-1- CMU2-AJYB |
IR45427-2B-2-2B-1-1/3*IR61920-3B-22-2-1 (NSIC RC 106) |
50 |
4.44 |
3.88 |
4.08 |
4.62 |
0.71 |
−0.22 |
0.19 |
−0.82 |
0.97 |
0.75 |
|
G36 |
IR93915-82-CMU2-2- CMU3-AJYB |
IR03W134/PUSA BASMATI 1 |
49 |
3.94 |
3.54 |
5.22 |
4.68 |
1.95 |
−0.63 |
−0.18 |
−1.00 |
0.97 |
0.75 |
|
G37 |
IR12T198 |
IR 84089-35/IR72875-94-3-3-2/IR72875-94-3-3-2 |
38 |
4.86 |
3.88 |
5.86 |
5.03 |
2.19 |
−0.71 |
0.01 |
−0.15 |
0.97 |
0.75 |
|
G38 |
IR12T136 |
NERICA 2/POKKALI |
48 |
4.1 |
4.53 |
4.46 |
4.75 |
0.51 |
−0.16 |
−0.17 |
−0.60 |
0.97 |
0.75 |
|
G39 |
IR11T182 |
IRRI 149/IR61920-3B-22-2-1 (NSIC RC 106) |
51 |
3.72 |
3.85 |
4.31 |
4.53 |
0.93 |
−0.30 |
−0.15 |
−0.95 |
0.97 |
0.75 |
|
G40 |
IR11T219 |
IRRI 149/IR61920-3B-22-2-1 (NSIC RC 106) |
26 |
5.08 |
5.3 |
5.09 |
5.19 |
0.37 |
−0.12 |
−0.06 |
−0.04 |
0.97 |
0.75 |
|
G41 |
IR11T220 |
IRRI 149/IR61920-3B-22-2-1 (NSIC RC 106) |
36 |
5.12 |
5.03 |
4.55 |
5.05 |
0.18 |
−0.04 |
0.12 |
−0.24 |
0.97 |
0.75 |
|
G42 |
BR8980-4-6-5 |
BR45/BR47 |
11 |
6.47 |
6.08 |
4.84 |
5.54 |
0.66 |
0.17 |
0.40 |
0.84 |
0.97 |
0.75 |
|
G43 |
BR8981-1-6-3-14 |
BR47/Pokkali 8948 |
34 |
4.33 |
5.74 |
4.86 |
5.09 |
0.45 |
0.06 |
−0.42 |
0.05 |
0.97 |
0.75 |
|
G44 |
BR8987-2-4-6 |
BR29/FL478 |
40 |
5.39 |
4.88 |
3.8 |
4.93 |
0.57 |
0.13 |
0.42 |
−0.12 |
0.97 |
0.75 |
|
G45 |
BR8992-3-4-10 |
BR47/FL478 |
16 |
5.7 |
6.08 |
4.55 |
5.35 |
0.76 |
0.24 |
0.14 |
0.44 |
0.97 |
0.74 |
|
G46 |
BR8980-B-1-1-1 |
BR45/BR47 |
22 |
5.49 |
5.77 |
4.66 |
5.27 |
0.41 |
0.13 |
0.09 |
0.31 |
0.97 |
0.75 |
|
G47 |
BR8980-B-1-3-5 |
BR45/BR47 |
23 |
5.38 |
5.26 |
5.19 |
5.25 |
0.48 |
−0.15 |
0.06 |
0.19 |
0.97 |
0.75 |
|
G48 |
BR8980-3-4-1-3 |
BR45/BR47 |
18 |
5.18 |
5.79 |
5.18 |
5.31 |
0.14 |
−0.01 |
−0.14 |
0.27 |
0.97 |
0.75 |
|
G49 |
BRRI dhan28 (S. Ck) |
BR 6/PURBACHI |
37 |
4.66 |
5.93 |
3.79 |
5.04 |
1.01 |
0.32 |
−0.17 |
−0.006 |
0.97 |
0.75 |
|
G50 |
BRRI dhan67 (Ck) |
IR 61247-3B-8-2-1/BR 36 |
14 |
5.5 |
5.71 |
5.42 |
5.4 |
0.30 |
−0.10 |
−0.04 |
0.33 |
0.97 |
0.75 |
|
G51 |
BINA dhan-10 (Ck) |
IR 42598-B-B-B-B-12/NONA BOKRA |
9 |
5.87 |
5.61 |
6.12 |
5.58 |
0.96 |
−0.31 |
0.00 |
0.59 |
0.98 |
0.75 |
|
Mean |
|
|
5.25 |
5.54 |
4.9 |
5.23 |
|
|
|
|
|
|
|
H2
|
|
|
0.76 |
0.92 |
0.86 |
0.55 |
|
|
|
|
|
|
Table 3Additive main effects and multiplicative interaction analysis of variance for grain yield (t/ha) of 51 rice genotypes across three environments.
Table 3
|
Sources of variation |
DF |
Sum of Square |
Mean Square |
F Value |
Pr |
Variability explained |
|
|
% TSS |
% GE |
|
Env |
2 |
21.05 |
10.53 |
40.36 |
0.0068 |
6.71 |
|
|
Genotype |
50 |
136.88 |
2.74 |
12.79 |
<0.0001 |
43.60 |
|
|
Env: Genotype |
100 |
123.19 |
1.23 |
5.76 |
<0.0001 |
39.23 |
|
|
IPCA1 |
51 |
92.98 |
1.82 |
8.58 |
<0.0001 |
|
75.5 |
|
IPCA2 |
49 |
30.20 |
0.62 |
2.90 |
<0.0001 |
|
24.5 |
|
Pooled Error |
150 |
32.10 |
0.21 |
|
|
10.22 |
|
|
Total |
305 |
313.99 |
|
|
|
|
|
References
- Abeya T, Chemeda D, Girma M, Dagnachew L, Negash G. 2008. Regression and additive main effects and multiple interactions (AMMI) in common bean (Phaseolus vulgaris L.) genotypes. Ethiop J Biol Sci. 7: 45-53.
- Al-Naggar AMM, El-Salam RMA, Asran MR, Yaseen WYS. 2018. Yield adaptability and stability of grain sorghum genotypes across different environments in Egypt using AMMI and GGE-biplot models. Annu Res Rev Biol. 23: 1-16.
- Asnake W, Henry M, Temesgen Z, Girma T. 2013. Additive main effects and multiplicative interactions model (AMMI) and genotype main effect and genotype by environment interaction (GGE) biplot analysis of multienvironmental wheat variety trials. Afr J Agric Res. 8: 1033-1040.
- Atlin GN, Cairns JE, Das B. 2017. Rapid breeding and varietal replacement are critical to adaptation of cropping systems in the developing world to climate change. Glob Food Sec. 12: 31-37.
- Bates D, Ana Ines Vazquez AI. 2014. Pedigree-based mixed-effects models Version 0.3-3.
- Bernardo R. 2010. Breeding for Quantitative Traits in Plants. 2nd edition. Stemma Press.
- Collard BCY, Beredo JC, Lenaerts B, Mendoza R, Santelices R, Lopena V, et al. 2017. Revisiting rice breeding methods–evaluating the use of rapid generation advance (RGA) for routine rice breeding. Plant Prod Sci. 20: 337-352.
- Crossa J, Gauch HG, Zobel RW. 1990. Additive main effects and multiplicative interaction analysis of two international maize cultivar trials. Crop Sci. 30: 493-500.
- Ebdon JS, Gauch HG Jr. 2002. Additive main effect and multiplicative interactions analysis of national turf grass performance trials: I. Interpretation of Genotype × Environment Interaction. Crop Sci. 42: 489-496.
- Falconer DS, Mackay TFC. 1996. Introduction to quantitative genetics. 4th Edition. Addison Wesley Longman. Harlow.
- Farshadfar E. 2008. Incorporation of AMMI stability value and grain yield in a single non-parametric index (GSI) in bread wheat. Pak J Biol Sci. 11: 1791-1796.
- Gauch HG, Piepho HP, Annicchiarico P. 2008. Statistical analysis of yield trials by AMMI and GGE: Further considerations. Crop Sci. 48: 866-889.
- Gauch HG, Zobel RW. 1996. AMMI analysis of yield trials. Kang MS, Gauch HG, editors. Genotype by Environment Interaction. CRC Press. Boca Raton: pp. 85-122.
- Gauch HG. 1993. Matmodel Version2.0. AMMI and related analysis for two-way data matrices. Micro Computer Power. Ithaca, NY, USA.
- Gorjanc G, Bijma P, Hickey JM. 2015. Reliability of pedigree-based and genomic evaluations in selected populations. Genet Sel Evol. 47: 65
- Grenier C, Cao TV, Ospina Y, Quintero C, Châtel MH, Tohme J, et al. 2015. Correction: Accuracy of genomic selection in a rice synthetic population developed for recurrent selection breeding. PLoS One. 11: e0154976
- Heffner EL, Sorrells ME, Jannink JL. 2009. Genomic selection for crop improvement. Crop Sci. 49: 1-12.
- Heffner EL, Lorenz AJ, Jannink JL, Sorrells ME. 2010. Plant breeding with genomic selection: Gain per unit time and cost. Crop Sci. 50: 1681-1690.
- Heslot N, Jannink JL, Sorrells ME. 2015. Perspectives for genomic selection applications and research in plants. Crop Sci. 55: 1-12.
- Huang M, Balimponya EG, Mgonja EM, McHale LK, Luzi-Kihupi A, Wang GL, et al. 2019. Mol Breed. 39: 114
- Islam MR, Sarker MRA, Sharma N, Rahman MA, Collard BCY, Gregorio GB, et al. 2015. Assessment of adaptability of recently released salt tolerant rice varieties in coastal regions of South Bangladesh. Field Crops Res. 190: 34-43.
- Jagadish SVK, Septiningsih EM, Kohli A, Thomson MJ, Ye C, Redoña E, et al. 2012. Genetic advances in adapting rice to a rapidly changing climate. J Agron Crop Sci. 198: 360-373.
- Krishnamurthy SL, Sharma PC, Sharma DK, Ravikiran KT, Singh YP, Mishra VK, et al. 2017. Identification of mega-environments and rice genotypes for general and specific adaptation to saline and alkaline stresses in India. Sci Rep. 7: 7968
- Lule D, Fetene M, DeVilliers S, Tesfaye K. 2014. Additive main effects and multiplicative interactions (AMMI) and genotype by environment interaction (GGE) biplot analyses aid selection of high yielding and adapted finger millet varieties. J Appl Biosci. 76: 6291-6303.
- McLaren CG, Chaudhary C. 1994. Use of additive main effects and multiplicative interaction models to analyse multilocation rice variety trials. In: Paper presented at the FCSSP Conference; Puerton Princesa, Palawan, Philippines.
- McLaren CG, Bruskiewich RM, Portugal AM, Cosico AB. 2005. The International Rice Information System. A platform for meta-analysis of rice crop data. Plant Physiol. 139: 637-642.
- Nakaya A, Isobe SN. 2012. Will genomic selection be a practical method for plant breeding? Ann Bot. 110: 1303-1316.
- Naroui Rad MR, Kadir MA, RafiiHawa MY, Naghavi MRJ, Ahmadi F. 2013. Genotype × environment interaction by AMMI and GGE biplot analysis in three consecutive generations of wheat (Triticum aestivum) under normal and drought stress conditions. Aust J Crop Sci. 7: 956-961.
- Ndhlela T, Herselman L, Magorokosho C, Setimela P, Mutimaamba C, Labuschagne M. 2014. Genotype × environment interaction of maize grain yield using AMMI biplots. Crop Sci. 54: 1992-1999.
- Peng S, Cassman KG, Virmani SS, Sheehy J, Khush GS. 1999. Yield potential trends of tropical rice since the release of IR8 and the challenge of increasing rice yield potential. Crop Sci. 39: 1552-1559.
- Prasad KV, Singh RL. 1990. Stability analysis of yield and yield components and construction of selection indices of direct seeded rice in frost season. In: Annual Review conference Proceeding; 20–23 October 1992; National Agril. Res. Inst. Caribbean Agricultural Research and development Institute. Guyana, pp 63-71.
- Purchase JL, Hatting H, Van Deventer CS. 2000. Genotype × environment interaction of winter wheat in South Africa: II. Stability analysis of yield performance. S Afr J Plant Soil. 17: 101-107.
- Rahman MA, Thomson MJ, Shah-E-Alam M, de Ocampo M, Egdane J, Ismail AM. 2016. Exploring novel genetic sources of salinity tolerance in rice through molecular and physiological characterization. Ann Bot. 117: 1083-1097.
- Rahman MA, Thomson MJ, De Ocampo M, et al. 2019. Assessing trait contribution and mapping novel QTL for salinity tolerance using the Bangladeshi rice landrace Capsule. Rice (N Y). 12: 63
- Robinson GK. 1991. That BLUP is a good thing: The estimation of random effects. Statist Sci. 6: 15-32.
- Rutkoski JE, Poland JA, Singh RP, Huerta-Espino J, Bhavani S, Barbier H, et al. 2014. Genomic selection for quantitative adult plant stem rust resistance in wheat. Plant Genome. 7:
- Spindel J, Begum H, Akdemir D, Virk P, Collard B, Redoña E, et al. 2015. Genomic selection and association mapping in rice (Oryza sativa): Effect of trait genetic architecture, training population composition, marker number and statistical model on accuracy of rice genomic selection in elite, tropical rice breeding lines. PLoS Genet. 11: 1-25.
- Tariku S, Lakew T, Bitew M, Asfaw M. 2013. Genotype by environment interaction and grain yield stability analysis of rice (Oryza sativa L.) genotypes evaluated in north western Ethiopia. Net J Agric Sci. 1: 10-16.
- Vazquez AI, Bates DM, Rosa GJM, Gianola D, Weigel KA. 2010. Technical Note: An R package for fitting generalized linear mixed models in animal breeding. J Anim Sci. 88: 497-504.
- Yan WK. 2001. GGE biplot - A windows application for graphical analysis of multienvironment trial data and other types of two-way data. Agron J. 93: 1111-1118.
- Yan WK, Kang MS, Ma B, Woods S, Cornelius PL. 2007. GGE Biplot vs. AMMI analysis of genotype-by-environment data. Crop Sci. 47: 643-655.
- Yan WK, Rajcan I. 2002. Biplot analysis of test sites and trait relations of soybean in Ontario. Crop Sci. 42: 11-20.
- Yan W, Hunt LA. 2002. Biplot analysis of multienvironment trial data. pp. 289-303. Kang MS, editor. Quantitative Genetics, Genomics and Plant Breeding. CABI Publishing. New York.
- Yan W, Hunt LA, Sheng Q, Szlavnics Z. 2000. Cultivar evaluation and mega-environment investigation based on the GGE biplot. Crop Sci. 40: 597-605.
- Yan W. 2014. Crop variety trials: Data management and analysis. Wiley-Blackwell.
- Zobel RW, Wright MG, Gauch HG. 1988. Statistical analysis of a yield trial. Agron J. 80: 388-393.
