"There is more virus in us than us in us". John Coffin's famous sentence illustrates that particular nucleic acid sequences related to exogenous viruses, called retrotransposons, constitute almost half of the human genome and largely exceeds the amount of protein-coding DNA (section 1.4). At the beginning of the past century, biologists realized the size of the genomes of various eukaryotes did not correlate with the level of complexity of these species. In extreme cases, certain species such as the fish L.paradoxa had 30-times more genomic DNA than do primates. In 1980, Francis Crick and Carmen Sapienza independently postulated that part of this once-called "junk DNA" must stem from self-replicating units, categorized as "selfish DNA" to distinguish them from more "altruistic" cellular genes. Indeed, while genes amplify in a population because of the selective advantage they confer to their host, the selfish DNA amplifies without making any contribution to the host phenotype and survives rather as a not-too-deleterious parasite. In early 2000's, the accomplishment of the human genome sequencing projects gave formal credence to Crick's and Sapienza's initial hypothesis. Retrotransposons replicate in the genome of their host by a copy-and-paste mechanism. Accordingly, any new retrotransposition event causes a genetic alteration potentially detrimental to the host. On an evolutionary perspective however, accumulation of retrotransposon-derived sequences can be beneficial to the species after these once "selfish" elements become "domesticated" to deserve more physiological roles. In human or mouse for instance, retrotransposons are known to cover a large spectrum of cellular functions, from gene regulation to antiviral defense (section 1.6). In order to maintain under tight control the balance between the good and adverse effects of retroelements, eukaryotes have evolved cellular defenses aimed at inhibiting the replication of such elements (section 1.7). Gene silencing by DNA methylation is a mechanism widely used in plants, fungi or vertebrates to control the expression of self-replicating elements. For instance in mammals, specific KRAB-zinc finger proteins can target retroviral-derived sequences and mediate their transcriptional silencing (section 3). The first aim of this study was to characterize the sensitivity of retroviruses to KRAB-mediated epigenetic silencing in varied chromosomal contexts (section 5). Since some retroviruses like HIV favor integration within transcriptionally active regions, these viruses may escape blockade by integrating chromosomal area refractory to epigenetic silencing. We found that instead, KRAB-mediated repression mechanism was not dependent upon integration site of a retroviral vector, hence the emergence of viruses escaping KRAB-mediated repression would be unlikely. Mammalian cells are also able to control the replication of retrotransposons at a post-transcriptional level via the APOBEC3 family of cytidine deaminases (section 2). Since certain retrotransposons are related to exogenous retroviruses, it is perhaps not surprising that APOBEC3 proteins can also inhibit a range of such elements including HIV. While there is only one APOBEC3 gene in the mouse genome, the family considerably expanded in other species such as primates were it comprises seven members (section 2.1). Although it is tempting to postulate that the expansion of the family in primates was driven by the strong selective pressure imposed by various genetic threats growing in number and complexity, it is not understood how only a few homologous proteins can target such a disparate group of replicating elements. The second aim of this study was to understand the mechanism by which different APOBEC3 proteins can target and inactivate particular self-perpetuating elements (section 4). We launched an integrative study focused on two human APOBEC3 homologues with distinct antiviral functions. APOBEC3A, a single-domain cytidine deaminase, can block efficiently the parvovirus adeno-associated virus type 2 (AAV-2) or the retrotransposon LINE-1 but cannot inhibit HIV replication. APOBEC3G is formed of two independent cytidine deaminase domains, only the C-terminal of which is catalytically active. APOBEC3G is endowed with restriction activity against HIV, but is largely ineffective against AAV-2 or LINE-1. We combined functional studies with structural modeling and phylogenetic analyses and found that restriction in each case involved similar regions at the surface of the protein and nearby the catalytic center, yet with distinct shapes and nucleic-acid interacting properties between the homologous proteins. While on APOBEC3G, the amino-acids in this region mediate intersubunit contacts between monomers and the predicted dimer interface shapes a groove that may facilitate binding to single-stranded RNA, on APOBEC3A however, it is corresponding residues on the monomer that form a DNA-binding groove important for editing. In summary, this study describes the nucleic-acid-interacting domains of APOBEC3A and APOBEC3G essential for their intrinsic restriction activities, and further suggests that structural variations in this domain may contribute to define the sets of targets inhibited by each APOBEC3 protein. APOBEC3 proteins may be considered as modular units, whose restriction function can evolve with emerging selective pressures by specific variations in their nucleic-acid binding domain, alternatively by fusion of two APOBEC3 proteins endowed with different functionalities. APOBEC3 proteins would efficiently protect the genome integrity of their host against highly variable genetic threats or "selfish DNA", in combination with other restriction factors such as the KRAB-zinc finger proteins. Contrary to epigenetic silencing however, post-transcriptional restriction as provided by APOBEC3 proteins may prohibit the deleterious effect of retrotransposition, without altering other retroelement functions putatively important for the physiology of the cell.