Search

National audience

International audience; The integration of genetic information in the cellular and nuclear environments is crucial for deciphering how the genome functions in physiological conditions. By combining 3D nuclear mapping, high-flow transcriptomic data analyses, and statistical methods for the development of co-regulated gene networks, it becomes possible to develop an integrated approach to depict the regulation of gene expression. For this purpose, we focused on the mechanisms involved in the transcriptional regulation of genes expressed in muscle during late fetal development in pig (90 and 110 days), a critical period for survival. We published a muscle transcriptomic analysis performed during this perinatal period (Voillet et al. 2014). Data from this previous study obtained from two extreme genetic lines in terms of mortality at birth (Large White and Meishan), were used to construct networks of differentially co-expressed genes. As co-expressed genes are not necessary related to a common biological process, we used information of gene co-localizations (3D DNA FISH) to reinforce observed links in the co-expressed gene network. The innovative network inference method developed, sequentially incorporates biological knowledge on gene spatial co-localization to construct robust networks gathering co-regulated genes. Clustering of nodes (genes) becomes more and more biologically consistent in each iteration. Interestingly, by means of the final network, we particularly uncovered unexpected gene associations in the nuclear space between IGF2 and MYH3 suggesting that they could be subject to similar transcriptional regulation.

National audience; Avec l’environnement sanitaire, les paramètres climatiques sont les principaux facteurs environnementaux affectant les performances des porcs dans les grands bassins de production mondiaux localisés en particulier dans les régions tropicales (TROP) mais également en zones tempérées (TEMP). Les réponses de thermorégulation varient d’une race à une autre mais également d’un animal à l’autre à l’intérieur d’une même population. Dans le futur, cette variabilité pourrait être exploitée pour améliorer la robustesse des systèmes de production porcine face aux risques climatiques. Pour cela, il existe un enjeu fort au développement d’indicateurs pertinents pour quantifier précisément et, si possible rapidement, le niveau de sensibilité ou de robustesse à la chaleur. L’objectif de l’étude était d’utiliser le métabolome plasmatique (RMN) de porc mesuré en conditions TEMP pour prédire la sensibilité des descendants issus 10 familles de pères Large-White x Créole elevés en conditions TROP. La valeur génétique de chaque famille a été évaluée à partir de la mesure des performances (consommation d’aliment, poids vif et température rectale) de leurs descendants produits dans les deux environnements (#60 descendants/famille/ environnement). L’analyse du métabolome a été réalisée par spectroscopie RMN sur des échantillons prélévés à 23 semaines d’âge sur des animaux élevés en conditions TEMP (n=560). L’indice de sensibilité au climat TROP (Isens) a été évalué en calculant la valeur du t de Welsh de la variation de la valeur de chaque famille entre TEMP et TROP. Le modèle de prédiction a été établi sur les 4 familles extrêmes (2 familles sensibles et 2 familles robustes) avec Isens comme variable à prédire et le métabolome des descendants (n=223 porcs) comme variables prédictrices. Pour construire le modèle, nous avons utilisé la méthode de régression des moindres carrés partiels (PLS-DA) associée à une étape de sélection en utilisant la méthode LASSO. Le modèle de prédiction a été construit avec 19 prédicteurs sélectionnés par les 445 variables du métabolome et permettait de prédire la sensibilité au climat TROP avec un taux d’erreur de 22%. Ce modèle a été utilisé pour prédire le pourcentage de descendants classés comme sensibles (Psens) dans chacune des 10 familles de père. La très bonne corrélation entre Isens et Psens (r = 0,97) a validé notre modèle de prédiction. Sur la base de données utilisées dans cette étude, il semble être possible de classer les porcs selon leur sensibilité à la chaleur à partir d’une simple prise de sang réalisée dans des conditions TEMP. Samir Dou a bénéficié d’un contrat post-doctoral financé par la région Bretagne (Dispositif « Stratégie d'attractivité durable »). Le projet a été conduit dans le cadre du programme ANR PigHeaT (ANR-12-ADAP-0015). Les auteurs remercient le personnel de l’UE PTEA (Petit Bourg, Guadeloupe) et de l’UE GenESI (Magneraud) pour leur aide dans la réalisation de ce projet.

The Self-Organizing Maps (SOM) is a very popular algorithm, introduced by Teuvo Kohonen in the early 80s. It acts as a non supervised clustering algorithm as well as a powerful visualization tool. It is widely used in many application domains, such as economy, industry, management, sociology, geography, text mining, etc. Many variants have been defined to adapt SOM to the processing of complex data, such as time series, categorical data, nominal data, dissimilarity or Kernel data. However, so far SOM had suffered from a lack of rigorous results on its convergence and stability. This article presents the state-of-art on the theoretical aspects of SOM, as well as several extensions to non numerical data and provides some typical examples of applications in different real-world fields.

Méthodes pour l’apprentissage de données massives

International audience; El algoritmo de auto-organización (Self-Organizing Map, SOM), o Mapa de Kohonen, es un algoritmo muypopular, definido por Teuvo Kohonen al principio de los anõs 80. El actúa como un algoritmo de clasificación(clustering) no supervisado y almismo tiempo como una herramienta potente de visualización. El es ampliamenteusado para aplicaciones en muchos campos, tales como economía, industría, gestión, sociología,geografía, análisis de textos, etc. Muchas variantes han sido definidas para adaptar SOM al estudio de datoscomplejos, tales como series temporales, datos de categoría, datos nominales, datos de disimilaridades. Sinembargo, la convergencia y la estabilidad del algoritmo SOM no tienen pruebas rigurosas y completas hastaahora. Este papel presenta el estado-del-arte de los aspectos teoréticos de SOM, al mismo tiempo que algunasextensiones para datos no numéricos y ejemplos típicos de diferentes campos con datos reales.; The Self-Organizing Maps (SOM) is a very popular algorithm, introduced by Teuvo Kohonen in the early80s. It acts as a non supervised clustering algorithm as well as a powerful visualization tool. It is widelyused in many application domains, such as economy, industry, management, sociology, geography, textmining, etc. Many variants have been defined to adapt SOM to the processing of complex data, such astime series, categorical data, nominal data, dissimilarity or kernel data. However, so far, SOMhas sufferedfroma lack of rigorous results on its convergence and stability. This article presents the state-of-art on thetheoretical aspects of SOM, as well as several extensions to non numerical data and provides some typicalexamples of applications in different real-world fields.

Apprentissage connexionniste

In metagenomic analysis, the integration of various sources of information is a di cult task since produced datasets are often of heterogeneous types. These datasets can be composed of species counts, interaction networks or phylogenetic information. The combinations of all these types of data have been shown relevant to provide a better comparison between communities. However, standard integration methods (like PLS) can take advantage of external information but do not allow to analyse heterogenous multi-omics datasets in a generic way. We propose to use similarity functions, called kernels, to integrate multiple datasets of various types into a single exploratory analysis. Kernels can be computed for various data types, such as numerical vectors, phylogenetic trees but also any diversity indexes. They can also be combined into a single meta-kernel. In this work, we provide several solutions to learn either a consensual meta-kernel or a meta-kernel that preserves the original topology of the datasets. This kernel is subsequently used in kernel PCA to provide a fast and accurate visualisation of similarities between samples, in a non linear space and from the multiple source point of view. A generic procedure is also proposed to improve the interpretability of the kernel PCA in regards with the original data. We applied our framework to the multiple metagenomic datasets collected during the TARA Oceans expedition. We demonstrated that our method is able to retrieve previous ndings in a single analysis as well as to provide a new image of the sample structures when a larger number of datasets from di erent sources are included in the analysis.

Functional annotation of livestock genomes is a critical and obvious next step to derive maximum benefit for agriculture, animal science, animal welfare and human health. The aim of the Fr-AgENCODE project is to generate multi-species functional genome annotations by applying high-throughput molecular assays on three target tissues/cells relevant to the study of immune and metabolic traits. An extensive collection of stored samples from other tissues is available for further use (FAANG Biosamples ‘FR-AGENCODE’). From each of two males and two females per species (pig, cattle, goat, chicken), strand-oriented RNA-seq and chromatin accessibility ATAC-seq assays were performed on liver tissue and on two T-cell types (CD3+CD4+&CD3+CD8+) sorted from blood (mammals) or spleen (chicken). Chromosome Conformation Capture (in situ Hi-C) was also carried out on liver. Sequencing reads from the 3 assays were processed using standard processing pipelines. While most (50–70%) RNA-seq reads mapped to annotated exons, thousands of novel transcripts and genes were found, including extensions of annotated protein-coding genes and new lncRNAs (see abstract #69857). Consistency of ATAC-seq results was confirmed by the significant proportion of called peaks in promoter regions (36–66%) and by the specific accumulation pattern of peaks around gene starts (TSS) v. gene ends (TTS). Principal Component Analyses for RNA-seq (based on quantified gene expression) and ATAC-seq (based on quantified chromatin accessibility) highlighted clusters characterised by cell type and sex in all species. From Hi-C data, we generated 40kb-resolution interaction maps, profiled a genome-wide Directionality Index and identified from 4,100 (chicken) to 12,100 (pig) topologically-associating do- mains (TADs). Correlations were reported between RNA-seq and ATAC-seq results (see abstract #71581). In summary, we present here an overview of the first multi-species and -tissue annotations of chromatin accessibility and genome architecture related to gene expression for farm animals.

In the context of the FAANG pilot project ‘FR-AgENCODE’ to improve the functional annotation of livestock genomes, we characterised the transcriptome and the chromatin accessibility of pig hepatocytes and two types of lymphocytes. More specifically, CD3+CD4+ (‘CD4’) and CD3+CD8+ (‘CD8’) T-cells were sorted from the blood of two male and two female Large White adult pigs. These samples, along with liver samples from the same animals, were processed by strand-oriented RNA-seq and ATAC-seq experiments. Principal Component Analyses on log-transformed TMM-normalized read counts in genes (from RNA-seq) and regions of chromatin accessibility (from ATAC-seq) consistently highlighted the variability between liver and the T-cells, and to a lesser extent within T-cells (CD4 v. CD8), as well as between the male and female samples. Comparative analyses identified differentially expressed genes between cell types as well as potential regulatory sites from differentially accessible chromatin regions. As expected, ontology annotations of differentially expressed genes were enriched for either immunity- or metabolism-related terms. Interestingly, correlations between gene expression and promoter accessibility across samples were enriched for both extreme positive and negative values, which suggests that ATAC-seq can efficiently capture distinct regulatory mechanisms of gene expression. Candidate enhancers and repressors were identified by comparing ATAC-seq regions with predicted binding sites of 500+ transcription factors. By integrating these results with those from Hi-C chromosome con-formation capture on the liver samples, we further characterised the differences between ‘active’ and ‘repressed’ topological domains in terms of functional features, including gene density and general chromatin accessibility. Altogether, these results lead to a better un-derstanding of the molecular mechanisms involved in pig immune and metabolic functions, and illustrate a useful contribution to the functional annotation effort of the FAANG initiative.

La reprise de cyclicité post partum a été décrite chez 760 vaches laitières Prim’Holstein françaises à partir de données collectées en 2015-2016 dans 21 élevages laitiers en traite robotisée (système Herd NavigatorND). Sur les 140 premiers jours de lactation, 38539 dosages de progestérone, 10380 dosages de béta hydroxybutyrate ont été analysés ainsi que les résultats de réussite de 1911 inséminations artificielles. Dès le premier mois de lactation, plus de la moitié des vaches ont déjà ovulé. 23% souffrent d’anoestrus post partum (reprise de cyclicité après 45 jours de lactation) et leur intervalle vêlage-insémination fécondante a été augmenté de 23 jours. 29% des vaches ont eu au moins une interruption de cyclicité, 36% au moins une lutéolyse précoce et 27% au moins une phase lutéale prolongée. Ces anomalies ont été associées à une augmentation de l’IVIAf de respectivement 21 jours, 6 jours et 23 jours. La durée de la phase lutéale précédant l’IA n’a pas montré d’impact sur le taux de réussite de l’IA, au contraire de la durée de la phase folliculaire : lorsque les IA sont réalisées lors d’une phase folliculaire longue, le taux de réussite chute de 20 points par rapport à des IA réalisées lors d’une phase folliculaire normale. La prévalence de la cétose était de 17%, plus élevée chez les vaches ayant eu au moins une anomalie (19%) par rapport à celles n’ayant jamais eu aucun trouble (11,5%). Ce travail montre l’importance et la fréquence des interruptions de cyclicité, assez peu considérées en pratique.

Genome sequence alone is not sufficient to explain the overall coordination of nuclear activity in a particular tissue. The nuclear organisation and genomic long-range intra- and inter-chromosomal interactions play an important role in the regulation of gene expression and the activation of tissue- specific gene networks. Here we present an overview of the pig genome architecture in muscle at two late developmental stages. The muscle maturation process occurs between the 90th day and the end of gestation (114 days), a key period for survival at birth. To characterise this period we profiled chromatin interactions genome-wide with in situ Hi-C (High Throughput Chromosome Conformation Capture) in muscle samples collected at 90 and 110 days of gestation, specific moments where a drastic change in gene expression has been reported. About 200 million read pairs per library were generated (3 replicates per condition). This allowed: (a) the design of an experimental Hi-C protocol optimized for frozen fetal tissues, (b) the first Hi-C contact heatmaps in fetal porcine muscle cells, and (c) to profile Topologically Associated Domains (TADs) defined as genomic domains with high levels of chromatin interactions. Using the new assembly version Sus scrofa v11, we could map 82% of the Hi-C reads on the reference genome. After filtering, 49% of valid read pairs were used to infer the genomic interactions in both developmental stages. In addition, ChIP-seq experiments were performed to map the binding of the structural protein CTCF, known to regulate genome structure by promoting interactions between genes and distal enhancers. The Hi-C and ChIP-seq data were analysed in combination with the results of a previous transcriptome analysis, focusing on the hun-dreds of genes that were reported as differentially expressed during muscle maturation. We will report the observed general differences between both developmental stages in terms of transcription and structure.