Repression of viral expression is a major strategy developed by retroviruses to escape from the host immune response. The absence of viral proteins (or derived peptides) at the surface of an infected cell does not permit the establishment of an efficient immune attack. Such a strategy appears to have been adopted by animal oncoviruses such as bovine leukemia virus (BLV) and human T-cell leukemia virus (HTLV). In BLV-infected animals, only a small fraction of the infected lymphocytes (between 1 in 5,000 and 1 in 50,000) express large amounts of viral proteins; the vast majority of the proviruses are repressed at the transcriptional level. Induction of BLV transcription involves the interaction of the virus-encoded Tax protein with the CREB/ATF factors; the resulting complex is able to interact with three 21-bp Tax-responsive elements (TxRE) located in the 5' long terminal repeat (5' LTR). These TxRE contain cyclic AMP-responsive elements (CRE), but, remarkably, the "TGACGTCA" consensus is never strictly conserved in any viral strain (e.g.,AGACGTCA, TGACGGCA, TGACCTCA). To assess the role of these suboptimal CREs, we introduced a perfect consensus sequence within the TxRE and showed by gel retardation assays that the binding efficiency of the CREB/ATF proteins was increased. However, trans-activation of a luciferase-based reporter by Tax was not affected in transient transfection assays. Still, in the absence of Tax, the basal promoter activity of the mutated LTR was increased as much as 20-fold. In contrast, mutation of other regulatory elements within the LTR (the E box, NF-B, and glucocorticoid- or interferon-responsive sites [GRE or IRF]) did not induce a similar alteration of the basal transcription levels. To evaluate the biological relevance of these observations made in vitro, the mutations were introduced into an infectious BLV molecular clone. After injection into sheep, it appeared that all the recombinants were infectious in vivo and did not revert into a wild-type virus. All of them, except one, propagated at wild-type levels, indicating that viral spread was not affected by the mutation. The sole exception was the CRE mutant; proviral loads were drastically reduced in sheep infected with this type of virus. We conclude that a series of sites (NF-B, IRF, GRE, and the E box) are not required for efficient viral spread in the sheep model, although mutation of some of these motifs might induce a minor phenotype during transient transfection assays in vitro. Remarkably, a provirus (pBLV-21-bp) harboring only two TxRE was infectious and propagated at wild-type levels. And, most importantly, reconstitution of a consensus CRE, within the 21-bp enhancers increases binding of CREB/ATF proteins but abrogates basal repression of LTR-directed transcription in vitro. Suboptimal CREs are, however, essential for efficient viral spread within infected sheep, although these sites are dispensable for infectivity. These results suggest an evolutionary selection of suboptimal CREs that repress viral expression with escape from the host immune response. These observations, which were obtained in an animal model for HTLV-1, are of interest for oncovirus-induced pathogenesis in humans.