We are glad to announce a symposium on genotype to phenotype modelling in relation to adaptation and GxE. The symposium is organized by WUR-Biometris and the EU-project Whealbi, with a program subdivided in three sessions.
The first session will focus on the importance of collecting, generating and characterizing the genetic variation for traits contributing to adaptation.
The second session focuses on strategies for phenotypic characterization across environments.
Finally, we discuss statistical and crop growth modelling approaches to connect genotypes and phenotypes across environments.
Session I: Genetic variation
The Whealbi (Wheat and Barley Legacy for Breeding Improvement) project, started beginning of 2014, stems from the strong conviction that to improve wheat and barley production in order to face severe global changes, we need to better exploit knowledge from basic science to develop new varieties and innovative cropping systems. It associates 18 partners in 9 countries.
WHEALBI combines genomics, genetics and agronomy to improve European wheat and barley production in competitive and sustainable cropping systems. It will generate original data from expressed genome sequences of 1000 wheat and barley genetic resources and provide models and tools to integrate these data in breeding programmes and crop management.
As first results, exome of 500 barley and 500 wheat accessions covering a wide range of diversity have been sequenced. For example in wheat, 680 000 SNP and structural variation have been found, of which # 600 000 in bread wheat. These data have been analysed to detect evolutionary pathways, signature of selection and quantitative trait loci.
The results will be disseminated to a broad user community, highlighting the benefits and issues associated with the adoption of what is considered sustainable and environmentally friendly wheat and barley crop production in a European context.
Making maize genetic resources amenable to elite germplasm improvement
Genome-enabled strategies for harnessing untapped allelic variation of landraces are currently evolving. The success of such approaches depends on the sampling of source material. Most studies capitalize on maximizing diversity by sampling few individuals from many landraces covering a wide range of geographic regions. For the improvement of elite germplasm, an alternative approach might be more suitable, namely sampling many individuals from few pre-selected landraces. We show the impact of different sampling strategies on diversity parameters and LD based on high-density genotypic data of 35 European maize landraces. First results from genome wide association mapping and genomic prediction for quantitative traits segregating in doubled haploid lines derived from pre-selected landraces will be presented.
Analyses of a set of Chromosome Substitution Lines in Arabidopsis thaliana
The importance of gene interactions (epistasis) is still not evident from current genetic approaches. Based on targeted approaches it is known that interacting loci can have large effects on the phenotype. Besides, with missing heritability’s and the complexity of genetic pathways it is not hard to assume there are many non-additive loci present in biological systems. We developed a set of chromosome substitution lines (CSLs) obtained via a reverse breeding approach that will provide a new tool to dissect epistasis. This new set of CSLs of Col-0 and Ler-0 Arabidopsis thaliana is used in a QTL mapping experiment, and used for identification of interacting chromosomes. Using different approaches of regression models and multivariate clustering we identify multiple two- and three-way interactions for different traits. Incorporating these interactions in the linear model allow prediction of the phenotype with much more accuracy and give insight into the complexity of quantitative traits.
Session II: Phenotypic characterization
Prediction of the genetic variability of yield in environmental scenarios: comparing statistical and dynamical approaches based on phenotyping
S Alvarez Prado, E Millet, S Lacube, B Parent, C Welcker, F van Eeuwijk, F Tardieu
We have developed two approaches to deal with the prediction of yield in a range of environmental scenarios in 29 experiments in Europe. Both of them consist in predicting sensitivity to environmental condition rather than yield or traits themselves. In a previous study, allelic effects at QTLs of yield were modelled as functions of environmental scenarios.
The first approach consists in extending this study towards genomic prediction of the sensitivities of yield to light, temperature and water deficit during periods of time corresponding to phenological stages of each genotype. Sensitivities are predicted using genetic markers (QTLs or GP). Yields are simulated in each site by combining calculated sensitivities with environmental conditions averaged during considered phenological stages.
The second approach consists in (i) phenotyping the genetic variability of traits (e.g. stomatal conductance, radiation use efficiency, growth), (ii) calculating trait values from genetic markers and environmental conditions and (iii) simulating yields based on these traits by using a dynamic crop model. Noteworthy, steps (ii) and (iii) are performed simultaneously by the crop model with an hourly or daily time step.
The likely conclusion is that the first approach may allow better prediction in a range of environmental scenarios whose characteristics are close to those in experiments. The second approach is more flexible and, at least potentially, would allow prediction in new combinations of environmental conditions (e.g. climate change).
Are Physiology and Phenomics Going to be Phruitful for Breeding
Scott C. Chapman, David R. Jordan, Bangyou Zheng, Andries Potgieter, Graeme L. Hammer, Wei Guo, Tao Duan, Barbara George-Jaeggli, James Watson, Pengcheng Hu, Matthew P. Reynolds and Greg J Rebetzke
Investment in field phenomics technologies is rapidly increasing - even some ex-molecular biologists are getting on the bandwagon. The capabilities of these technologies can be impressive, but how do we ensure that this research becomes more than a stamp-collecting exercise. Breeders need to balance resources across their programs and had major challenges in working out how to deploy molecular marker technology to accelerate gain. Statistical and logistical short-cuts (e.g. using pedigree info etc) have helped with this. What kinds of 'short-cuts' can we make in phenomics? For example, our experiences with ground and aerial vehicle phenotyping, mainly in wheat and sorghum, have indicated that we can take some short-cuts in data processing and by using 'in-field' calibration to gain precision that is hard to achieve by relying on better but expensive (in time and cost) methods. Estimating crop height is one example - how can you use UAVs to match the moderate precision required for this trait without taking days to geo-position and process the data. Extracting key adaptation traits using phenomics and using this information to drive crop simulation models can also be used to derive or compute traits that are difficult to measure. This paper explosres some of the ways that we can 'add biological value' to raw phenomics data.
Session III: Genotype to phenotype modelling
Phenotypic prediction augmented through crop model-whole genome prediction: Application to ARGOS
Phenotypic prediction accuracy for yield and transgenic effects is limited in agricultural systems where Transgene-by-Genotype-by-Environment-by-Management (T×G×E×M) interactions are frequent. The fusion of crop growth models (CGMs) with whole genome prediction (WGP) methods was demonstrated to improve prediction accuracy. We demonstrate the CGM-WGP methodology for the prediction of the effects of ARGOS, a transgene that can improve drought tolerance in maize.
Crop models: bridging the gap between genotype and phenotype
When QTL mapping and genomic prediction is conducted for yield components (model parameters) instead of yield itself, QTLs may explain a larger part of variation in yield and QTLs are less dependent on the environment. In this approach yield is dissected in yield components. Subsequently a statistical model to estimate these yield components is built from marker profiles by QTL mapping or genomic prediction. Finally, the yield components for a genotype are estimated based on the statistical model and inserted into a crop model to obtain predictions for yield. The models are validated for environments and genotypes not used for model calibration. Results show that crop models can help to bridge the gap between genotype and phenotype and help to disentangle genotype-by-environment (G×E) interactions for complex traits.
Multi-environment genomic prediction: a penalized regression perspective
Willem Kruijer, Emilie Millet, François Tardieu, Claude Welcker and Fred van Eeuwijk
Multi-environment trials are usually characterized by environmental covariates like maximum temperature. From a statistical perspective, these covariates are important for dissecting genotype by environment interaction, and genomic prediction in new environments. In view of the large numbers of covariates that are now often available, variable selection has become increasingly important. We show how this can be achieved with a variety of penalized regression approaches, using data from the DROPS network (Millet et al. 2016, Plant Phys.)
Considerations from statistical and crop growth models to design strategies to predict plant adaptation across environments
Daniela Bustos-Korts, Marcos Malosetti, Martin Boer, Fred van Eeuwijk
Selection processes in plant breeding depend critically on the quality of phenotype predictions. The phenotype is classically predicted as a function of genotypic and environmental information. One aspect that determines the quality of phenotype prediction is the set of genotypes used to train the prediction model, especially when populations are structured. A second aspect that influences the accuracy of phenotype predictions is the choice of environments used to train the prediction model, which should capture the heterogeneity in the target population of environments. Prediction accuracy can also benefit from modelling the target trait and its underlying physiological components, when components have a high heritability and they are correlated to the target trait. Trait correlations are modified by environmental conditions and by the genetic architecture. Therefore, additional phenotyping might be advantageous in some environments but not in others. A combination of crop growth models and statistical models can help breeders to identify when phenotyping additional traits would pay off by a larger accuracy for the target trait. In this presentation, we will discuss and illustrate considerations that need to be taken into account when deciding which genotypes, traits and environments are used to collect phenotypic information to predict adaptation across multiple environments.
Assessing the impact of phenotyping protocols
Fred van Eeuwijk, Daniela Bustos Korts, Marcos Malosetti, Willem Kruijer, Martin Boer
Biometris, Wageningen University
Genetic and breeding research can benefit in various ways from high precision phenotyping and high throughput phenotyping information collected by special purpose devices and procedures in growth chambers, greenhouses, managed stress environments and field experiments. In high precision phenotyping, traits are measured under highly controlled environmental conditions and with a better control of measurement errors. These traits as measured at platforms and controlled conditions are then used to predict a similar trait in the field or to predict a parameter in a model for a related trait in the field. In high throughput phenotyping, additional information is collected in field trials by sensors and imaging devices and this information is used to improve predictions for complex target traits across environments. One can say that traits measured as part of phenotyping efforts are used as inputs for models that predict correlated traits in single or multiple environmental configurations. However, these prediction models are often not made explicit in the phenotyping literature. We will present some suitable quantitative frameworks for evaluating the utility and impact of phenotypic traits as measured in high precision and high throughput phenotyping protocols. The most general frameworks are based on correlated response selection theory and multi-environment factorial regression in mixed model and penalized regression context. For the multi-environment models, we will look at QTL and genomic prediction models as such as well as integrations of those models with crop growth models.