Search

Recently, digestive efficiency (DE) was proposed as a trait of interest to improve feed efficiency (FE) in pigs, especially when they are fed with alternative feeding resources. Both are influenced by the host genetics, and also by the gut microbiota composition. The goal of this study was to quantify the impact of faecal microbial information on the prediction accuracies of genomic estimated breeding values (GEBVs) of FE and DE traits for pigs fed conventional or fiber diets. For DE traits, gains in prediction accuracy of GEBVs were increased by about 18% when microbial information was included in linear mixed models. In addition, these gains of prediction accuracy were very similar in both diets. For FE traits, no improvement was observed. Thus, the addition of microbial information in breeding programs is promising to better estimate GEBVs for DE traits.

Classical methods as genomic BLUP performs well for genomic prediction of polygenic trait, but does not consider interaction between genes or between genes and other information such as host genetic or microbial data. This study aims at comparing several methods including parametric and machine learning methods to predict digestive coefficient using genomic, microbial and both genomic and microbial information. Considering only microbial data led to the best prediction accuracies for digestive coefficients, whereas considering only genomic data performed worst. BLUP, RKHS and GSVM gave the best prediction accuracies except when combined genomic and microbial data was used. Combining microbial and genomic data did not improve prediction accuracies for all traits and methods considered in this study. Thus, considering microbial information is crucial to predict digestive efficiency and interactions between host genetic and faecal microbial information seem to be limited.

International audience; Breeding pigs that can efficiently digest alternative diets with increased fibre content is a viable strategy to mitigate the feed cost volatility in pig production. This study aimed at determining the relative contribution of microbiome and genetics to the phenotypic variability of three digestibility coefficients (DC). Faecal samples were collected at sixteen weeks of age to sequence the V3-V4 regions of the 16S RNA gene and to predict DC with near-infrared spectrometry for energy, organic matter, and nitrogen. Data were available for 1,564 Large White pigs, and 14,366 identified Operational Taxonomic Units (OTUs) were identified. The proportions of phenotypic variance explained by the microbiome, or ‘microbiability’, and by additive genetic effects were first estimated independently and then jointly using a Bayesian approach. The microbiability was estimated under three scenarios: without filtering the OTUs, by filtering the OTUs presented in more than five samples and with an average abundance higher than 0.001% (2,399 OTUs) or greater than 0.01% (803 OTUs). Estimates of microbiability for the traits were reduced with increasingly stringent filtering, from 54±2 to 66±2% in the first scenario and from 25±1 to 35±1% for 803 OTUs. With intermediate filtering, the estimated microbiability ranged from 37±1 to 46±2% and was higher than heritability (from 26±1 to 29±1%) when estimated independently from genetics effects. The joint estimation of microbiability and heritability will point out how the two random effects overlap in explaining the trait variability. To conclude, our results show that the filtering of OTUs is important and influences microbiability. The microbiome explained a significant proportion of the phenotypic variance of digestive efficiency, which was even larger than heritability estimates.

International audience; The use of early biomarkers is an interesting approach to improve selection efficiency for traits that are difficult to measure. The aim of this study was to validate juvenile IGF1 and leptin as biomarkers of growth and feed efficiency in Large White (LW), Duroc and Piétrain purebred lines. To determine plasma concentrations of IGF1 and leptin, blood samples were obtained from 8-9-week-old piglets at the end of the post-weaning phase. After data curation, 337 LW, 192 Duroc and 189 Piétrain pigs had both plasma concentrations and production performance recorded during the growing-finishing phase, namely average daily gain (ADG), daily feed intake (DFI) and feed conversion ratio (FCR). Nested linear mixed models including successively the biomarker concentration and the weight at bleeding (BBW) as covariables in addition to the usual effects were compared using a likelihood ratio test. In the three lines, adding IGF1 concentration in prediction models of growth and feed efficiency traits increased the goodness-of-fit, with the covariable being significant (P < 0.001), except for FCR in the Piétrain line. Augmenting this model with BBW increased the model likelihood in the LW and Duroc breeds for any traits, as BBW was highly significant (P < 0.001) although IGF1 was not significant anymore. These results suggested that IGF1 is a biomarker of growth and feed efficiency that captured the same information as BBW. In the Piétrain line, IGF1 contributed to increase the likelihood of the model along with BBW for ADG and DFI. Hence, IGF1 seems to be a relevant biomarker for the Piétrain breed but to have limited utility for the LW and Duroc breeds if pigs can be weighed in the post-weaning phase. The same approach was applied to juvenile leptin, but it was found not to be a useful biomarker of growth or feed efficiency for the three lines.

International audience

International audience

Accuracy of genomic selection to improve litter traits in the French Landrace pig population. 67. Annual Meeting of the European Federation of Animal Science (EAAP)

Released in the 1980's and in the early 1990's, the statistical models used for genetic evaluations of beef cattle breeding stocks with on-farm performance recording (IBOVAL) and in central and progeny test stations rely on simple hypotheses which induce biases in the prediction of breeding values. These biases can imply a non optimal choice of breeding stocks and therefore reduce the short-term selection efficiency. They can also reduce long-term selection efficiency because they tend to designate breeding stocks to select among well-known parental lineages. Genetic diversity of the three main French beef cattle populations (Blonde d'Aquitaine, Charolais and Limousin) was also described using pedigree analyses. This study showed that genetic diversity is still large in these populations and sufficient to guarantee genetic gain in the next generations. The work achieved in this thesis also contributes to the improvement of beef cattle selection schemes by proposing new statistical models to eliminate some biases in beef cattle genetic evaluations., Mis en place dans les années 1980 et au début des années 1990, les modèles d'évaluation génétique des reproducteurs bovins allaitants en ferme (IBOVAL) et en stations de contrôle individuel (CI) et sur descendance (CD) reposent sur des hypothèses simples dont le non respect est à l'origine de biais dans la prédiction des valeurs génétiques. Ces biais peuvent induire un choix non optimal des reproducteurs et réduire ainsi l'efficacité de la sélection à court terme, mais aussi à plus long terme en privilégiant les reproducteurs issus de lignées familiales bien connues. Le travail présenté dans cette thèse contribue à l'amélioration des programmes de sélection allaitants en proposant de nouveaux modèles statistiques pour quantifier et éliminer certaines sources de biais des évaluations génétiques des reproducteurs bovins allaitants. Tout d'abord, un bilan détaillé de la diversité génétique des trois principales populations bovines allaitantes françaises a été réalisé à l'aide d'analyses de pedigree des animaux des bases de sélection Blonde d'Aquitaine, Charolaise et Limousine. Cette étude a montré que la diversité génétique est encore importante dans ces populations et largement suffisante pour garantir des marges de progrès génétique dans les générations futures. Elle a également permis de caractériser les populations d'animaux nés dans les élevages sélectionneurs, vendeurs de taureaux de monte naturelle ou approvisionnant en mâles les programmes de sélection des taureaux d'insémination artificielle (IA). Cette caractérisation a jeté les bases pour proposer de nouveaux modèles d'évaluation génétique des reproducteurs de bovins allaitants. Tout d'abord, le modèle IBOVAL actuel ne prend pas en compte correctement les différences de longueurs de pedigree existant dans les populations bovines allaitantes. En effet, il suppose que les parents inconnus proviennent d'une unique population de fondateurs génétiquement homogènes et défavorise ainsi les animaux aux généalogies les plus courtes en fixant leur valeur génétique prédite sur ascendance au niveau génétique de l'ensemble des fondateurs de la race. L'introduction de groupes de parents inconnus (GPI) dans le modèle d'évaluation permet de mieux intégrer les différences de niveau génétique existant dans la population des fondateurs. Une méthode a été élaborée et validée en race Charolaise pour définir des GPI robustes et homogènes d'après les flux de reproducteurs observés au sein de la population évaluée sur les performances au sevrage des veaux. L'inclusion de GPI dans le modèle IBOVAL s'avère utile pour améliorer l'efficacité à court terme de la sélection en race Charolaise, principalement sur la voie femelle. Elle contribue à améliorer sensiblement le choix des vaches pour le renouvellement des troupeaux ayant un taux élevé d'animaux nés de parents inconnus. En revanche, cela impacte peu le choix des reproducteurs mâles destinés à une large diffusion par IA parce que ces mâles sont essentiellement issus d'élevages spécialisés dans la vente de reproducteurs où les généalogies des animaux sont bien connues sur plusieurs générations. Les programmes de sélection des taureaux d'IA s'appuient sur une succession de trois étapes séquentielles d'évaluation et sélection, d'abord en ferme puis en stations de CI et de CD. A chaque étape, une évaluation génétique est réalisée à l'aide d'un modèle unicaractère ignorant les données qui ont servi à la sélection aux étapes précédentes. Un modèle multicaractère, combinant performances enregistrées en stations et performances homologues contrôlées en ferme dans les élevages approvisionnant le schéma IA, a été proposé pour éliminer les biais dus à la sélection des mâles entrant en stations. L'utilisation de ce modèle permet d'augmenter l'efficacité de la sélection par une meilleure précision des index, mais surtout par l'élimination des,biais dans la prédiction des valeurs génétiques en stations de CI et de CD et l'amélioration de la connexion entre séries évaluées au cours du temps. Si les reclassements des reproducteurs évalués en stations sont limités intra-série, ils peuvent être en revanche très sensibles entre séries évaluées en race Blonde d'Aquitaine ou Limousine, modulant l'utilisation des reproducteurs sélectionnés intra-série et donc la diffusion à large échelle de leur semence.

International audience

Chantier qualité GA; In genetic evaluations, the definition of unknown parent groups (UPG) is usually based on time periods, selection path and flows of foreign founders. The definition of UPG may be more complex for populations presenting genetic heterogeneity due to both, large national expansion and coexistence of artificial insemination (AI) and natural service (NS). A UPG definition method accounting for beef bull flows was proposed and applied to the French Charolais cattle population. It assumed that, at a given time period, unknown parents belonged to the same UPG when their progeny were bred in herds that used bulls with similar origins (birth region and reproduction way). Thus, the birth period, region and AI rate of a herd were pointed out to be the three criteria reflecting genetic disparities at the national level in a beef cattle population. To deal with regional genetic disparities, 14 regions were identified using a factorial approach combining principal component analysis and Ward clustering. The selection nucleus of the French cattle population was dispersed over three main breeding areas. Flows of NS bulls were mainly carried out within each breeding area. On the contrary, the use and the selection of AI bulls were based on a national pool of candidates. Within a time period, herds of different regions were clustered together when they used bulls coming from the same origin and with an estimated difference of genetic level lower than 20% of genetic standard deviation (σg) for calf muscle and skeleton scores (SS) at weaning. This led to the definition of 16 UPG of sires, which were validated as robust and relevant in a sire model, meaning numerically stable and corresponding to distinct genetic subpopulations. The UPG genetic levels were estimated for muscle and SS under sire and animal models. Whatever the trait, differences between bull UPG estimates within a time period could reach 0.5 σg across regions. For a given time period, bull UPG estimates for muscle and SS were generally larger by 0.30 to 0.75 σg than those of cows. Including genetic groups in the evaluation model increased the estimated genetic trends by 20% to 30%. It also provoked re-ranking in favor of bulls and cows without pedigree.

Chantier qualité GA; Pedigree collected by the Interbeef service allowed genetic diversity to be assessed by using pedigree analyses for the European Charolais (CHA) and Limousin (LIM) cattle populations registered in national herdbooks in Denmark (DNK), France (FRA), Ireland (IRL), Sweden (SWE), and, solely for the LIM breed, the United Kingdom (UK). The CHA data set included 2,563,189 calves with weaning performance, of which 96.1% were recorded in FRA, 3.0% in SWE, 0.5% in IRL, and 0.4% in DNK. The LIM data set included 1,652,734 calves with weaning performance, of which 91.9% were recorded in FRA, 4.9% in UK, 1.8% in DNK, 0.9% SWE, and 0.5% in IRL. Pedigree files included 3,191,132 CHA and 2,409,659 LIM animals. Gene flows were rather limited between populations, except from FRA toward other countries. Pedigree completeness was good in all subpopulations for both breeds and allowed the pedigree to be traced back to the French population. A relatively high level of genetic diversity was assessed in each CHA and LIM subpopulation by estimating either effective population sizes (Ne >244 and Ne >345 in the CHA and LIM subpopulations, respectively), relationship coefficients within subpopulations (