Source: UNIV OF WISCONSIN submitted to
RESOURCE ALLOCATION OPTIMIZATION FOR MULTI-ENVIRONMENT TRIALS AND GENOMIC SELECTION IN OATS
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
NEW
Funding Source
HATCH
Reporting Frequency
Annual
Accession No.
1012486
Grant No.
(N/A)
Project No.
WIS01984
Proposal No.
(N/A)
Multistate No.
(N/A)
Program Code
(N/A)
Project Start Date
Mar 29, 2017
Project End Date
Mar 31, 2020
Grant Year
(N/A)
Project Director
Gutierrez Chacon, LU, .
Recipient Organization
UNIV OF WISCONSIN
21 N PARK ST STE 6401
MADISON,WI 53715-1218
Performing Department
Agronomy
Non Technical Summary
Plant breeding is probably the single most important activity in human history allowing civilization by initiating agriculture where people went from hunter-gatherers to farmers creating food surplus that gave rise to modern civilization. Modern plant breeding activities consists in evaluating the genetic merit of lines discerning genetic from environment and noise components. Therefore, controlling micro and macro environmental (i.e. genotype by environment interaction, GxE) variability is fundamental for breeding success. Furthermore, modern plant breeding tools such as genomic selection (GS) has been proved successful to increase the rate of genetic gain in plants. However, the best way to allocate testing resources within and among environments remains largely unsolved. The purpose of this research is to optimize resource allocation for plant breeding. First, we will compare strategies to optimize resource allocation for genotypic evaluation for the Multi-Environment Trials (MET) in the Wisconsin Oat Breeding Program (WOBP). We will compare experimental design strategies based on both micro-environmental variation (local control of field heterogeneity with experimental designs), and macro-environmental variation (GxE). Second, we will optimize an oat training population set using genomic prediction that model GxE to increase prediction accuracy and to target local adaptation. We will compare strategies for phenotyping and envirotyping and how to incorporate them into genomic prediction models to increase prediction accuracy as well as strategies for predicting genotypic performance for local adaptation.
Animal Health Component
0%
Research Effort Categories
Basic
10%
Applied
60%
Developmental
30%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
2011560108180%
2021560108010%
2051560108010%
Knowledge Area
202 - Plant Genetic Resources; 205 - Plant Management Systems; 201 - Plant Genome, Genetics, and Genetic Mechanisms;

Subject Of Investigation
1560 - Oats;

Field Of Science
1080 - Genetics; 1081 - Breeding;
Goals / Objectives
The purpose of this research is to optimize resource allocation for plant breeding. First, we will compare strategies to optimize resource allocation for genotypic evaluation for the Multi-Environment Trials in the Wisconsin Oat Breeding Program. We will compare experimental design strategies based on both micro-environmental variation, and macro-environmental variation (GxE). Second, we plan to optimize an oat training population set using genomic prediction that model GxE to increase prediction accuracy and to target local adaptation. We will compare strategies for phenotyping and envirotyping and how to incorporate them into genomic prediction models to increase prediction accuracy as well as strategies for predicting genotypic performance for local adaptation.
Project Methods
Objective 1: To compare strategies and optimize resources allocation for genotypic evaluation for the MET in the Wisconsin Oat Breeding Program. We will compare experimental design strategies based on both micro-environmental variation (local control), and macro-environmental variation (genotype-by-environment interaction).Rationale: Before a line can be released to the farmers, multi-environment evaluations need to be completed in order to evaluate the line's performance across several environmental conditions. The WOBP has historically relied on the evaluation of approximately 40 genotypes in randomized complete block designs with three replications evaluated in six locations. This requires approximately 720 plots yearly to evaluate the performance of 40 genotypes. New experimental designs have been proposed to increase the number of genotypes evaluated without increasing the resources for field evaluation. This would allow the genotypic evaluation of larger population sizes, and therefore, a larger selection intensity could be achieved providing better locally adapted varieties.Methods for evaluating micro-environmental control (i.e. experimental design efficiency): Yield data from cultivars will be simulated using real field variability, genotypic effects and a number of experimental designs. Each realization will be analyzed using a series of models with different levels of correction for spatial variability. Models will be compared by goodness of fit, model efficiency, and recovery of the original ranking of the genotypes. The best experimental design will be recommended.Oat yield data and spatial variability (i.e. micro-environmental) from yield monitors: Two fields will be planted with the oat cultivar 'Ron' at the West Madison and Lancaster Agricultural Research Station during year 1 and 2. Management will be homogenous within and across the fields. The fields will be harvested by a yield monitor in rectangular plots with GPS coordinates. Yield maps will be created for each field evaluating linear and quadratic trends in both directions, as well as different semi-variograms. Selection of the best spatial model will be conducted using the Error Sum of Squares for the weighted least square fit of the semi-variogram model.Genotypic effects: Genotypic effect for the simulation process will be obtained from yield mean data of 15, 50, and 200 oat cultivars from the reported historical nursery evaluations of the WOBP.Experimental designs: Four experimental designs will be simulated at each field: a Completely Randomized Design, a Randomized Complete Block Design, an Incomplete Block - Alpha Design, and a Partially Replicated design.Simulation procedure: A vector of yield plots will be obtained according to the following procedure: first, treatments will be assigned to plots according to one of the four experimental designs described above. Second, yield of each plot will be simulated.Analysis models: Each vector of phenotypic yield will be analyzed according to the following models: no spatial correction, spatially correlated error model with one-dimensional auto-regressive process, and a spatially correlated error model with two-dimensional auto-regressive process.Estimation method and statistics for model comparison: Design and models will be compared by fitness and accuracy. The AIC will be used as a model fit statistic. The standard error of the difference between cultivar means, the number of times that the null hypothesis of equal treatment effects was rejected, and the correlation between simulated and predicted genotypic effects will be used as accuracy statistics.Methods for evaluating macro-environmental control: The use of small balanced experiments will be compared to the use of large purposefully designed unbalanced experiments. This objective will be accomplished both theoretically and empirically.Theoretical simulation for resource allocation among environments: We will simulate four resource allocation strategies that cover a range of 40 advanced inbred lines in a RCBD with replications in six environments to genotypes unreplicated within environments but replicated across environments. Allocation strategies will be evaluated based on model fit and accuracy. Finally, the use of genomic information to model the correlation among genotypes for borrowing information from relatives and models without genomic information will be evaluated.Field based approaches for resource allocation among environments: The same four resource allocation strategies will be simulated based on re-sampling of actual field-evaluation experiments. Large field based experiments will be conducted at two locations to evaluate GxE and field heterogeneity using resolvable incomplete block designs with three full replications to evaluate 312 genotypes.Resource allocation strategy based on expected selection gain: We will compare the four resource allocation strategies based on the expected selection gain from each approach.Objective 2: To optimize an oat training population set using genomic prediction that model GxE to increase prediction accuracy and to target local adaptation. We will compare strategies for phenotyping and envirotyping and how to incorporate them into genomic prediction models to increase prediction accuracy as well as strategies for predicting genotypic performance for local adaptation. Rationale: The use of genomic prediction models and GS strategies has been documented elsewhere as highly successful in breeding programs. Furthermore, modeling the GxE has demonstrated its potential in some crops. Using genomic information and modeling the GxE could increase the response to selection and locally adapted genotypes could be better identified.Methods for using historical information from the breeding program to build a training population: We will apply the covariance-among-location method, the use of mega-environments, a simple factorial regression method that incorporates weather-based stress covariates (Heslot et al., 2014), and new strategies to incorporate environmental covariates based on the crop physiology to model GxE.Plant Material and Phenotyping: Approximately 300 oat advanced inbred lines from the WOBP were extensively phenotyped and also genotyped and will be used as a training set for genomic prediction models. These genotypes were chosen to have been evaluated in at least three environments in the breeding program in the last 20 years. Phenotypic evaluation from six locations in Wisconsin evaluated in twenty years will be used.Genotypic Data: Half of the oat lines were genotyped by sequencing of DNA extracted from a single plant. The SNP identification was conducted using the Haplotag pipeline to deliver 241,210 markers. The remaining individuals will be genotyped during the first year of the proposal as part of the POGI initiative that is fully supported.Phenotypic Data Analysis: Phenotypic best linear unbiased estimation will be obtained for all genotypes present in each trial.Characterization of GxE: Several strategies will be used to characterize GxE including variance component estimation, correlation between environments, AMMI and GGE models, and graphical representation through augmented biplots.Genomic Predictions Models: To compare strategies to deal with GxE, we will partition the data sets of environments: by year, location, and ME. We will use two general approaches to make genomic predictions: overall predictions for the mean performance within each set of environments, and predictions by environment. We will integrate weather variables that allow predictions for unobserved locations, assuming their expected weather is known and we will explore novel strategies to incorporate environmental covariates based on crop physiology to model GxE for new environments. The accuracy of genomic estimated breeding values will initially be estimated with a CV1 and CV2.

Progress 03/29/17 to 09/30/17

Outputs
Target Audience: Midwest farmers Crop Improvement groups Changes/Problems:There were no major changes or problems. We are on track for most objectives, and slightly ahead for objective 1.2. The only change in the approach is that because of the timing for the start of the project and a change in management at the Lancaster research station, we were unable to conduct the uniformity trial experiment during the 2017 growing season. We were still able to do the evaluations at the West Madison Research Station. We will do the evaluation in 2018. What opportunities for training and professional development has the project provided?A graduate student is being trained under this project. Several graduate student participated in research related activities during the 2017 growing season, and one student was emploed longer to also help in field preparation. How have the results been disseminated to communities of interest?Results from this project were presented in several instances including field days for farmers and stakeholders, and academic presentations: Gutierrez, L.* 2017. Cereals Breeding Program Update. Annual Meeting of the Wisconsin Crop Improvement Association. Madison, WI, USA, November 28, 2017. Gutierrez, L.* 2017. Quantitative Genetics Deployed in Breeding Programs. XXI Symposium on Genetics and Plant Breeding: Quantitative Genetics and its relationship to plant breeding. Part of the DuPont Plant Sciences Symposia Series. Universidad Federal do Lavras, Lavras, Brazil, November 8-10, 2017. Gutierrez, L.* 2017. Quantitative Genetics and Cereals Breeding at UW-Madison. Midwest Extended Rotation council meeting. Practical Farmers of Iowa. Ames, IA, USA, August 17, 2017. Kucek, L.K.*, Dawson, J.*, Gutierrez, L.* 2017. Organic Wheat Breeding. OGRAIN field day. Wisconsin, July 20, 2017. Gutierrez, L.* 2017. Research update from the cereals breeding and quantitative genetics group. Small grains field day. WCIA field day. Arlington, WI, USA, July 7, 2017. Gutierrez, L.* 2017. Modeling genotype-by-environment interaction to map and to predict complex quantitative traits in plants. 8th International Triticeae Symposium, Wernigerode/Gatersleben, Germany, June 12-16, 2017. Results will be presented at the Plant and Animal Genome Meetings in January 2018. Gutierrez, L.* 2018. Genomic Selection Addresses Genotype by Environment Interaction. Plant and Animal Genome Conference. San Diego, CA, USA, January 13-17, 2018. (Invited) What do you plan to do during the next reporting period to accomplish the goals?We will continue to work on objectives 1 and 2.

Impacts
What was accomplished under these goals? Impact Plant breeding has been historically essential for providing food and fibers for humankind. Every year, billions of dollars are invested to provide farmers with better, more adapted, and resilient cultivars. The process of selecting the best genotypes requires highly trained breeders who can distinguish the genetic merit of the individuals from the specific noise of environmental influence that they experience due to the field or environment in which it was grown. Our research focuses on optimizing this process to improve plant breeding efficiently. Goals The purpose of this research is to optimize resource allocation for plant breeding. First, we would like to compare strategies to optimize resource allocation for genotypic evaluation for the Multi-Environment Trials (MET) in the Wisconsin Oat Breeding Program (WOBP). We will compare experimental design strategies based on both micro-environmental variation (local control of field heterogeneity with experimental designs), and macro-environmental variation (GxE). Second, we would like to optimize an oat training population set using genomic prediction that model GxE to increase prediction accuracy and to target local adaptation. We will compare strategies for phenotyping and envirotyping and how to incorporate them into genomic prediction models to increase prediction accuracy as well as strategies for predicting genotypic performance for local adaptation. Accomplishments Obj1.1. Micro-environment (experimental design). Uniformity trials with yield monitors in West Madison and Lancaster We planted the uniformity trials at West Madison during the 2017 growing season. Growing conditions and management were standard for the crop in terms of planting date, fertilization and weed control. We harvested the experiment with a yield monitor driven at a 1.5 miles per hour average speed, with a 25 feet width and yield evaluated every second. Lancaster trials were not planted due to a change in management in the farm but they will be conducted during the 2018 growing season. Obj1.1. Micro-environment (experimental design). Uniformity trials analysis: yield maps, spatial models, experimental design simulation, model comparison. Yield maps were obtained with the yield monitors and data was curated and smoothed. Several spatial models were compared and the best model was used for the final map. Experimental designs were simulated using the real field heterogeneity and average yield performance of genotypes in two locations (i.e. West Madison Agricultural Research Station and Arlington Agricultural Research Station). Experimental designs and spatial corrections were compared and the best one was selected. The correlation between simulated and predicted yield performance for different combinations of experimental designs (i.e. completely randomized design, CRD; randomized complete block design, RCBD; alpha-designs, ALPHA; and partially replicated designs, PREP) using different levels of spatial corrections (i.e. no spatial correction, NSC; an autoregressive of order one spatial correction in one dimension; and an exponential spatial correction in two dimensions, EXP) for two locations, Arlington, WI, and Madison, WI showed that the ALPHA design was the best one. Obj1.1. Micro-environment (experimental design) report and paper publishing. We have started preparing the paper for publication which we anticipate might be ready before the summer in 2018. Obj.1.2. Macro-environment (experimental design with GxE). Historical data curation. The historical data set has been curated. Obj.1.2. Macro-environment (experimental design with GxE). Large balanced yield trials at West Madison and Lancaster locations. This objective was planned for the 2018 growing season. Obj.1.2. Macro-environment (experimental design with GxE). Experimental design analysis: yield maps, spatial models, experimental design simulation, model comparison, genetic gain. Although, we planned this activity for 2018, we started building the models and working on the theoretical approach using simulated data to have an idea of model performance. We simulated four resource allocation strategies using information from the uniformity experiments and the WOBP historical database. In all the strategies, six locations with 60 experimental units (plots) per location were be used. Strategy 1 uses the current experimental design used in most breeding programs with a randomized complete block with 3 replications repeated at each location. Strategy 2 uses another common strategy for genomic studies where each location consist of a partially replicated experiment but the same genotypes are evaluated in all locations. Strategy 3 and 4 uses an extreme strategy where not all genotypes are evaluated in all the locations creating purposefully unbalanced designs. The difference is that in strategy 3, less overlap among locations is used than in strategy 4. The red cells indicate the presence of checks replicated 3 times and green cells represent genotypes partially replicated 3 times within each location. Additionally, the study of GxE for the historical data set is also ready. Obj.1.2. Macro-environment (experimental design with GxE) report and paper publishing. We began work on the paper using the theoretical approach and it will be ready before the summer of 2018. We will probably publish it separated from the empirical approach that will be published later, after the 2018 growing season data is ready. Obj.2. Genomic selection. Genotyping. All the genotypes were grown in the greenhouse and tissue was collected from each individual and sent to the USDA-ARS genotyping lab for analysis. Results will be ready before the summer of 2018. Obj.2. Genomic selection. Data curation and phenotypic analysis, GxE characterization, GS model comparison for GxE strategies. This activity was planned for 2018. We will start working on it once we have the genotypic data and have planned the large phenotyping experiment for 2018. Obj.2. Genomic selection report and paper publishing. This is planned for 2019.

Publications

  • Type: Conference Papers and Presentations Status: Other Year Published: 2018 Citation: Gutierrez, L.* 2018. Genomic Selection Addresses Genotype by Environment Interaction. Plant and Animal Genome Conference. San Diego, CA, USA, January 13-17, 2018. (Invited)
  • Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: Gutierrez, L.* 2017. Modeling genotype-by-environment interaction to map and to predict complex quantitative traits in plants. 8th International Triticeae Symposium, Wernigerode/Gatersleben, Germany, June 12-16, 2017.