A growing number of cellular mRNAs are thought to possess internal ribosome entry sites (IRESs), sequences that permit translation of a transcript independent of its 5 end and cap structure. Although dicistronic assays are the canonical method of testing sequences for IRES activity, they may produce false-positive results if unanticipated monocistronic RNAs arise from the dicistronic construct used. Using a dicistronic reporter system and a green fluorescent protein-tagged retrovirus to evaluate six previously reported cellular IRESs, we found that four contain 3 splice sites whose activity was required for apparent IRES function and which resulted in formation of monocistronic transcripts by splicing. Bioinformatic analysis revealed that the 3 splice sites identified in three of these putative IRESs are used in their native mRNAs and that the fourth is likely an artifactual sequence created during cDNA cloning. Our findings demonstrate a need for reexamination of other reported cellular IRESs by using careful RNA structural analysis to rule out splicing as the source of perceived IRES activity.T he vast majority of eukaryotic mRNAs are translated via a mechanism in which the 40S ribosomal subunit engages the mRNA at its methylguanosine-capped 5Ј end (1). Upon associating with the transcript, these subunits are believed then to scan in the 5Ј to 3Ј direction for an appropriately situated AUG at which to begin translation (2, 3). A smaller number of mRNAs are translated by a 5Ј end-and cap-independent mechanism wherein ribosomes are recruited to the transcript at an interior location through an internal ribosome entry site (IRES).IRESs were first discovered in the picornaviruses encephalomyocarditis virus (EMCV) and poliovirus (4, 5). The RNA of these viruses possesses very long 5Ј UTRs bearing many unutilized upstream AUGs (uAUGs) and, unlike cellular mRNA, is uncapped (6). Soon after the identification of picornaviral IRESs, a number of cellular mRNAs were also reported to contain IRESs. To date, at least 85 cellular IRESs have been described (7). The experimental grounds on which proof of most cellular IRESs rest, however, has been the subject of dispute (8-10).A primary criticism of the data presented as establishing the existence of cellular IRESs concerns the plasmid-based dicistronic assay, the standard method of ascertaining IRES activity. In this assay, the candidate sequence is inserted between two reporter genes (5) so that both the upstream and downstream cistron are transcribed on the same RNA. If the test insert causes increased expression of the downstream cistron relative to the upstream cistron, the result is considered evidence for internal ribosome entry. However, the generation of even low levels of monocistronic RNAs from dicistronic constructs has the potential to falsely indicate IRES activity (8, 9, 11). One way that such RNAs could arise is through splicing of the dicistronic transcript due to the presence of a 3Ј splice site (ss) in the test sequence [see supporting information (SI) Fig....

SummaryRetroviruses are well known for their ability to incorporate envelope proteins from other retroviral strains and genera and even from other virus families. This characteristic has been widely exploited for the generation of replication-defective retroviral vectors, including those derived from murine leukemia virus (MLV), bearing heterologous envelope proteins. We desired to investigate the possibility of "genetically" pseudotyping replication-competent MLV by replacing the native env gene in a full-length viral genome with that of another gammaretrovirus. We previously developed replication-competent versions of MLV that stably transmit and express transgenes inserted in the 3′ untranslated region of the viral genome. In one such tagged MLV expressing green fluorescent protein, we replaced the native env sequence with that of gibbon ape leukemia virus (GALV). Although the GALV Env protein is commonly used to make high titer pseudotypes of MLV vectors, we found that the env replacement greatly attenuated viral replication. However, passage of cells exposed to the chimeric virus resulted in selection of mutants exhibiting rapid replication kinetics and different variants arose in different infections. Two of these variants had acquired mutations at or adjacent to the splice acceptor site and three others had acquired dual mutations within the long terminal repeat. Analysis of the levels of unspliced and spliced viral RNA produced by the parental and adapted viruses showed that the mutations gained by each of these variants functioned to reverse an imbalance in splicing caused by the env gene substitution. Our results reveal the presence of previously unknown cis-acting sequences in MLV that modulate splicing of the viral transcript and demonstrate that tagging of the retroviral genome with an easily assayed transgene can be combined with in vitro evolution to efficiently generate and screen for replicating mutants of replicationimpaired recombinant viruses.