Dans le contexte des changements globaux, un des enjeux majeurs en génomique écologique est d'estimer le potentiel adaptatif des populations naturelles. Répondre à cet enjeu nécessite 3 étapes: identification des agents sélectifs et de leurs échelles spatiales de variation, identification des bases génétiques de l'adaptation et étude de la dynamique adaptative sur une courte échelle de temps. Durant ma thèse, je me suis intéressé à étudier le potentiel adaptatif de la plante modèle Arabidopsis thaliana. A partir de 168 populations naturelles d'A. thaliana caractérisées pour 24 traits phénotypiques et 60 facteurs abiotiques (climat, sol) et biotiques (communautés végétales et microbiote), j'ai pu mettre en évidence que les communautés végétales étaient les principaux agents sélectifs associés à la fitness. Après avoir séquencé le génome de ces 168 populations (~ 4.8 millions de SNPs), j'ai effectué des analyses de type 'association génome-environnement' couplé à des scans génomiques de différenciation génétique spatiale. Ces analyses ont confirmé l'importance de considérer les interactions plante-plante dans l'étude de l'adaptation chez A. thaliana. Afin d'étudier le potentiel adaptatif d'A. thaliana sur le court terme dans le contexte d'un réchauffement climatique, j'ai combiné une étude de résurrection in situ avec une étude de Genome Wide Association mapping, à partir de 195 accessions locales caractérisées pour 29 traits phénotypiques et pour environ 1.9 million de SNPs. J'ai identifié une architecture originale de l'adaptation vers un nouvel optimum phénotypique combinant (i) de rares QTLs avec des degrés de pléiotropie intermédiaires fortement sélectionnés et (ii) de très nombreux QTLs spécifiques d'un micro-habitat et faiblement sélectionnés. A travers les différents projets abordés pendant ma thèse, j'ai pu suggérer qu'une architecture génétique flexible pouvait permettre à A. thaliana de s'adapter rapidement aux changements globaux, tout en maintenant de la diversité génétique au sein des populations naturelles d'A. thaliana. / In the context of global changes, one of the challenges in ecological genomics is to estimate the adaptive potential of natural populations. Three steps are requested to address this challenge: identification of the selective agents and their associated spatial grains, identification of the genetic bases of adaptation and monitoring the adaptive dynamics of natural population over a short time period. Here, I aimed at studying the adaptive potential of the model plant Arabidopsis thaliana. Based on 168 natural populations of A. thaliana characterized for 24 phenotypic traits and 60 abiotic (climate, soil) and biotic (plant communities and microbiota) factors, plant communities were found to be the main selective agents. Based on 4.8 million SNPs, I combined Genome Environment Association analysis with genome scans for signatures of selection. I confirmed the importance to consider plant-plant interactions when studying adaptation in A. thaliana. To monitor the adaptive dynamics of a natural population in the context of global warming, I combined an in situ resurrection study with an approach of GWA mapping based on 195 local accessions characterized for 29 phenotypic traits and 1.9 million SNPs. Adaptive evolutionary changes were largely driven by rare QTLs with intermediate degrees of pleiotropy under strong selection. In addition to these rare pleiotropic QTLs, weak selection was detected for frequent small micro-habitat-specific QTLs that shape single traits. Overall, I suggest that a rapid adaptive phenotypic evolution can be rapidly achieved in A. thaliana, while still maintaining genetic variation in natural populations.

<p> Despite the continuous breeding efforts towards improving yield, seed quality, and yield-related traits, there is still little understanding of several aspects of soybean breeding; however, crop breeding is ever-evolving, and plant breeding technologies offer immense potential for accelerating genetic improvement in soybeans. This thesis explores different frameworks to further characterize tradeoffs among seed quality traits, soybean maturity's genetic architecture, and selections for yield. We explored the interactions of carbohydrate traits with other seed traits, flowering, and maturity using data from a large panel of <em>G. max </em>accessions from the USDA soybean germplasm collection. We found a negative correlation between sucrose and protein and a negative correlation between protein and oil, representing a significant challenge for improving seed quality. In contrast to other well-documented correlations, such as protein and oil, correlations between raffinose and oil content seem more specific to populations and environments and are unlikely to generalize to the whole specie; however, the correlations of sucrose with protein and seed size appears to be more stable. In addition, we performed a genome-wide association analysis (GWA) to detect novel QTLs for flowering (R1) time, maturity (R8) time, and reproductive length (RL) using a soybean panel with the same genotype for major <em>E </em>genes (<em>e1-as/E2/E3). </em>While major maturity <em>E</em> genes are known to have pleiotropic effects on R1 and R8, we found two QTLs associated with R8 and RL that do not control R1, suggesting minor-effect, trait-specific loci are also involved in controlling R1 and R8. In addition, w<em>e identified six genes that may play essential roles in regulating R1, R8, and RL; however, further validation of the QTLs and f</em>ine mapping and map-based cloning studies of the candidate genes are necessary before they can be used in breeding programs. Lastly, we conducted a selection experiment in progeny row (PR) populations of four breeding programs to compare the agronomic performance of lines selected by breeders using their usual selection methods to lines selected through prediction of yield performance using new sources of data and information. Our results suggest that aerial average canopy coverage (ACC) used as a secondary trait in combination with field spatial variation adjustment is an efficient high throughput methodology to effectively select high-yielding lines from non-replicated experiments at the PR stage. </p>