Search

Recently, it has been proposed to switch molecular markers to near-infrared (NIR) spectra for inferring relationships between individuals and further performing phenomic selection (PS), analogous to genomic selection (GS). The PS concept is similar to genomic-like omics-based (GLOB) selection, in which molecular markers are replaced by endophenotypes, such as metabolites or transcript levels, except that the phenomic information obtained for instance by near-infrared spectroscopy (NIRS) has usually a much lower cost than other omics. Though NIRS has been routinely used in breeding for several decades, especially to deal with end-product quality traits, its use to predict other traits of interest and further make selections is new. Since the seminal paper on PS, several publications have advocated the use of spectral acquisition (including NIRS and hyperspectral imaging) in plant breeding towards PS, potentially providing a scope of what is possible. In the present chapter, we first come back to the concept of PS as originally proposed and provide a classification of selected papers related to the use of phenomics in breeding. We further provide a review of the selected literature concerning the type of technology used, the preprocessing of the spectra, and the statistical modeling to make predictions. We discuss the factors that likely affect the efficiency of PS and compare it to GS in terms of predictive ability. Finally, we propose several prospects for future work and application of PS in the context of plant breeding.

Crop breeding involves two selection steps: choosing progenitors and selecting offspring within progenies. Genomic prediction, based on genome-wide marker estimation of genetic values, could facilitate these steps. However, its potential usefulness in grapevine (Vitis viniferaL.) has only been evaluated in non-breeding contexts mainly through cross-validation within a single population. We tested across-population genomic prediction in a more realistic breeding configuration, from a diversity panel to ten bi-parental crosses connected within a half-diallel mating design. Prediction quality was evaluated over 15 traits of interest (related to yield, berry composition, phenology and vigour), for both the average genetic value of each cross (cross mean) and the genetic values of individuals within each cross (individual values). Genomic prediction in these conditions was found useful: for cross mean, average per-trait predictive ability was 0.6, while per-cross predictive ability was halved on average, but reached a maximum of 0.7. Mean predictive ability for individual values within crosses was 0.26, about half the within-half-diallel value taken as a reference. For some traits and/or crosses, these across-population predictive ability values are promising for implementing genomic selection in grapevine breeding. This study also provided key insights on variables affecting predictive ability. Per-cross predictive ability was well predicted by genetic distance between parents and when this predictive ability was below 0.6, it was improved by training set optimization. For individual values, predictive ability mostly depended on trait-related variables (magnitude of the cross effect and heritability). These results will greatly help designing grapevine breeding programs assisted by genomic prediction.

Viticulture has to cope with climate change and decrease pesticide inputs, while maintaining yield and wine quality. Breeding is a potential key to meet this challenge, and genomic prediction is a promising tool to accelerate breeding programs, multivariate methods being potentially more accurate than univariate ones. Moreover, some prediction methods also provide marker selection, thus allowing quantitative trait loci (QTLs) detection and allowing the identification of positional candidate genes. We applied several methods, interval mapping as well as univariate and multivariate penalized regression, in a bi-parental grapevine progeny, in order to compare their ability to predict genotypic values and detect QTLs. We used a new denser genetic map, simulated two traits under four QTL configurations, and re-analyzed 14 traits measured in semi-controlled conditions under different watering conditions. Using simulations, we recommend the penalized regression method Elastic Net (EN) as a default for genomic prediction, and controlling the marginal False Discovery Rate on EN selected markers to prioritize the QTLs. Indeed, penalized methods were more powerful than interval mapping for QTL detection across various genetic architectures. Multivariate prediction did not perform better than its univariate counterpart, despite strong genetic correlation between traits. Using experimental data, penalized regression methods proved as very efficient for intra-population prediction whatever the genetic architecture of the trait, with accuracies reaching 0.68. These methods applied on the denser map found new QTLs controlling traits linked to drought tolerance and provided relevant candidate genes. These methods can be applied to other traits and species.

Genetic selection in grapevine and many similar species is a challenge due to their perennial status and outbred nature. Marker-assisted selection and genomic selection can hence be useful methods to ease and speed up breeding. In Coupel-Ledru et al (2014), a 191 progeny of Syrah x Grenache was phenotyped in two successive years for several ecophysiological traits under two conditions, well-watered and water deficit, on a high-throughput phenotyping platform coupled to a controlled–environment chamber. As offsprings were previously genotyped at 153 SSR markers, several QTLs were found for each trait separately, differing across years and conditions.But do these differences reflect biological processes or contrasted power between conditions? And how accurately do they allow predicting phenotypes depending on the conditions? To answer such questions, our aim is to explore the ability of sparse and regularized multivariate regression models and algorithms to select QTLs based on their predictive properties. In the present study, we perform variable selection with various flavours of the LASSO method (group Lasso, fused Lasso) adapted for multiple responses, extending the model and algorithm from Chiquet et al. (2017). We apply these methods on simulated data and on real data fromCoupel-Ledru et al. (2014).

Grapevine breeding needs to address two main issues over the next few years: reducing pesticide use and adaptating to climate change. If the selection of new resistant grapevine cultivars has been accelerated, breeding remains a long and costly process for a perennial species such as grapevine. That is why I tested and compared different methodologies for optimizing the breeding of new grapevine varieties. The first is genomic prediction (GP), which relies on the use of molecular markers to train a model to predict genetic values. The second is phenomic prediction (PP), which relies on the use of spectra measured on plant tissues, which is cheaper and more high-throughput than genotyping. I used GP and PP under different configurations to evaluate their interest in breeding programs. For that, I used three grapevine populations with contrasted relatedness, both genotyped and phenotyped. First, I compared univariate and multivariate GP models in a bi-parental population (N=188), on 14 traits related to drought. Multivariate methods did not perform better than univariate ones, and ranking between methods depended on the genetic architecture and heritability of the trait. Secondly, I tested across-population GP (a more applicable configuration for breeding) for 15 traits, by training the prediction model in a diversity panel (N=277) and using 10 bi-parental families of a half-diallel (N=622) as validation sets. For that, I first predicted the average genetic value of each family (for the first selection step of future crosses to be made) and then the genetic values of individuals within each family (for the second selection step of offspring within crosses once they have been realized). Prediction accuracy for these two steps appeared to be satisfying for the application of GP in breeding programs, compared to within-population prediction accuracy. Finally, I tested for the first time the application of PP in grapevine, by using near-infrared spectra measured on wood and leaves. Prediction accuracies were encouraging, despite a suboptimal framework because phenotypes and spectra were not measured in the same years.In conclusion, these results provide a good overview of methods and configurations that could be applied to optimize the selection of new grapevine varieties. For this optimization to be effective, it will be necessary to define the varietal ideotype and to have a reference population genotyped and phenotyped in several environments. My results allowed me to propose a new breeding scheme that comprises two steps, the selection of crosses to be made with GP and the selection of individuals within crosses with GP or PP. Such a scheme would allow to increase genetic gain while decreasing the cycle length from 15 to 9 years.In conclusion, these results provide a good overview of methods and configurations that could be applied to optimize the selection of new grapevine varieties. For this optimization to be effective, it will be necessary to define the varietal ideotype and to have a reference population genotyped and phenotyped in several environments. My results allowed me to propose a new breeding scheme that comprises two steps, the selection of crosses to be made with GP and the selection of individuals within crosses with GP or PP. Such a scheme would allow to increase genetic gain while decreasing the cycle length from 15 to 9 years.In conclusion, these results gave us a good overview of what could be applied to optimize the selection of new grapevine varieties. For this optimization to be effective, it will be necessary to define the varietal ideotype and to have a reference population genotyped and phenotyped in several environments. The new breeding scheme that I propose comprises two steps, selection of crosses to realize by GP and the selection of individuals within a cross with GP or PP. Such a scheme would allow to increase genetic gain while decreasing the cycle length from 15 to 9 years., La sélection génétique de la vigne dois répondre à deux enjeux majeurs dans les prochaines années : la réduction de l'utilisation de produits phytosanitaires et l'adaptation au changement climatique. Si la sélection de variété résistantes a été grandement accélérée, la sélection de caractères d'intérêt complexes reste un processus long et coûteux pour une plante pérenne comme la vigne. C'est pourquoi j'ai testé et comparé différentes méthodologies pour optimiser la sélection de nouvelles variétés de vigne. La première est la prédiction génomique (PG), qui repose sur les marqueurs moléculaires pour entraîner un modèle de prédiction des valeurs génotypiques. La seconde est la prédiction phénomique (PP), qui repose sur l'utilisation de spectres mesurés sur les tissus de la plante, plus rapides et moins chers à obtenir que le génotypage. J'ai utilisé la PG et la PP dans différentes configurations pour mesurer leur intérêt pratique dans les programmes d'amélioration. Pour cela, je me suis basée sur trois populations de variétés de vigne avec des apparentements génétiques contrastés, préalablement génotypées et phénotypées. J'ai d'abord comparé des modèles de PG univariés et multivariés dans une population bi-parentale (N = 188), sur 14 caractères relatifs au stress hydrique. Les précisions de prédiction étaient bonnes pour les caractères héritables, les méthodes multivariées n'étaient pas meilleures que les méthodes univariées, et le classement entre les méthodes dépendait de l'architecture génétique du caractère. Dans un second temps, j'ai testé la PG inter-population (configuration plus applicable en sélection) pour 15 caractères, en entraînant le modèle de prédiction dans un panel de diversité (N = 277) pour prédire dans 10 familles bi-parentales connectées en demi-diallèle (N = 622). Pour cela, j'ai prédit d'abord la moyenne des familles (pour la sélection des futurs croisements à réaliser), puis la valeur des individus au sein de chaque famille (pour la sélection d'individus au sein des croisements un fois qu'ils ont été réalisés). Pour ces deux étapes, les précisions de prédiction en inter-population restaient satisfaisantes (par rapport à l'intra-population) pour l'application de la PG dans les programmes d'amélioration. Finalement, j'ai testé la PP pour la première fois chez la vigne, en utilisant des spectres dans le proche infrarouge obtenus sur bois et sur feuilles. Les précisions de prédiction en PP se sont révélées encourageantes, malgré un cadre suboptimal car les spectres et les phénotypes n'ont pas été mesurés les mêmes années.En conclusion, ces résultats donnent un bon aperçu de ce qui pourrait être appliqué pour optimiser la sélection de nouvelles variétés de vigne. Pour que cette optimisation soit effective, il faudra cependant bien définir l'idéotype variétal et avoir à disposition une population de référence génotypée et phénotypée dans plusieurs environnements. Le nouveau schéma de sélection que je propose sur la base de mes résultats comprend deux étapes : la sélection des croisements à réaliser par PG et la sélection des individus au sein des croisements par PG ou PP. Un tel schéma pourrait permettre d'augmenter le gain génétique par cycle et de diminuer la durée du cycle de 15 à 9 ans.