A Functional –1 Ribosomal Frameshift Signal in the Human Paraneoplastic Ma3 Gene*

A bioinformatics approach to finding new cases of –1 frameshifting in the expression of human genes revealed a classical retrovirus-like heptanucleotide shift site followed by a potential structural stimulator in the paraneoplastic antigen Ma3 and Ma5 genes. Analysis of the sequence 3′ of the shift site demonstrated that an RNA pseudoknot in Ma3 is important for promoting efficient –1 frame-shifting. Ma3 is a member of a family of six genes in humans whose protein products contain homology to retroviral Gag proteins. The –1 frameshift site and pseudoknot structure are conserved in other mammals, but there are some sequence differences. Although the functions of the Ma genes are unknown, the serious neurological effects of ectopic expression in tumor cells indicate their importance in the brain.

A bioinformatics approach to finding new cases of ؊1 frameshifting in the expression of human genes revealed a classical retroviruslike heptanucleotide shift site followed by a potential structural stimulator in the paraneoplastic antigen Ma3 and Ma5 genes. Analysis of the sequence 3 of the shift site demonstrated that an RNA pseudoknot in Ma3 is important for promoting efficient ؊1 frameshifting. Ma3 is a member of a family of six genes in humans whose protein products contain homology to retroviral Gag proteins. The ؊1 frameshift site and pseudoknot structure are conserved in other mammals, but there are some sequence differences. Although the functions of the Ma genes are unknown, the serious neurological effects of ectopic expression in tumor cells indicate their importance in the brain.
Although nonoverlapping triplet reading of mRNA codons is the essence of genetic decoding, dynamic nonstandard events at specific sites can permit product diversity and provide regulatory options. One of these recoding events involves shifting to an alternative reading frame by a proportion of ribosomes, thereby changing the linearity of readout. Such utilized frameshifting commonly features both a "shifty site" in the mRNA and an appropriately positioned structural feature in the mRNA that acts to enhance the level of frameshifting, i.e. it is programmed. Although the shift involved can be to either alternative frame, a shift to the Ϫ1 frame often involves tRNAs in both the A and P sites detaching from their codons and re-pairing to mRNA at the two overlapping Ϫ1 frame codons (1,2). This tandem detachment and re-pairing usually occurs on a "slippery" heptanucleotide sequence that follows the general pattern of X XXY YYZ where the A and P site tRNAs detach from the zero frame codons XXY YYZ and re-pair after shifting 1 nucleotide to XXX YYY. The spacer region between the shift site and the 3Ј structural element is commonly 6 -8 nt 2 (3). The most common 3Ј recoding signals for Ϫ1 frameshifting are pseudoknots (4,5) with distinct features (6,7). A less common type of pseudoknot is the kissing stem-loop found in the human coronavirus, HCV229E (8), but bioinformatic studies suggest that it may be involved for transmissible gastroenteritis virus frameshifting also (9).
Examples of genes that require Ϫ1 frameshifting for expression are well known in viral genomes. There is only one demonstrated case of Ϫ1 frameshifting in a mammalian cellular gene (10,11), but it does not display the phylogenetic breadth of the ϩ1 frameshifting utilized in decoding antizyme mRNA, conserved from mammals to yeasts (12). Frameshifting near the end of a coding sequence often allows a proportion of ribosomes to access an ORF that extends beyond the zero frame terminator so that the transframe (coupled ORF) protein is longer than that of standard decoding. The easiest class of Ϫ1 programmed frameshift events to search for is one that has conserved architecture of two ORFs that partially overlap and contains a defined shift site. In these cases conservation of an independent ORF2 helps to distinguish significant candidates from a genomic background that contains sequencing errors and pseudogenes. However, cases such as the Escherichia coli dnaX gene, which utilizes a Ϫ1 frameshift to cause a proportion of ribosomes to access a stop codon in the new frame near the frameshift site, demonstrate that frameshifting can be utilized to yield an additional product that lacks a carboxyl-terminal domain of the product of standard decoding. In attempting to identify cases of this nature, searches for two long and partially overlapping ORFs will be fruitless because ORF2 will be wholly contained within the sequence encoding ORF1 and can be small or nonexistent. In addition to identifying an appropriate pair of ORFs that will support a given type of recoding, a specific frameshift site must be identified. The classic heptanucleotide Ϫ1 shift site followed by a pseudoknot, as described above, provides a straightforward target for bioinformatic searches. Other possible shift sites are less well defined and thus more complicated to identify. For example, the potential for both A and P site tRNAs to re-pair in the new frame can be difficult to discern (13). In yet other cases, our current knowledge does not adequately predict the likelihood of frameshifting (14,15). A recent case in point is that reported for the generation of polyalanine runs from long CAG tracts in the MJD-1 transcript (related to a late onset neurodegenerative disorder) where there is no conventional heptanucleotide frameshift site (16). Diverse approaches are needed to discern the extent to which programmed Ϫ1 frameshifting is utilized in decoding eukaryotic cellular genes. Here we report a functional Ϫ1 frameshifting site in a mammalian cellular gene, Ma3, identified by a bioinformatic analysis of mRNA sequences searching for a classic heptanucleotide shift site followed by a pseudoknot and contained in the overlap between two ORFs. A counterpart exists in the human Ma5 gene, but the 3Ј stimulator was not analyzed in detail.
Ma3 and Ma5 are members of a family of mammalian genes whose protein products are the target of immunity associated with paraneoplastic disorders whose symptoms, though associated with tumors elsewhere in the body, are not directly caused by the tumors or the treatment of them. These disorders are often the consequence of an autoimmune response to antigens ectopically expressed in the tumor cells that target the protein(s) normally expressed by the affected cells. Ma1 is expressed in the brain and testis, Ma2 in the brain, and Ma3 in the brain and testis and to a lesser extent in the trachea, kidney, and heart. Immunity to Ma2 and sometimes additional immunity to Ma1 and Ma3 are associated with brainstem encephalitis and cerebral degeneration (17). Our results indicate that the human paraneoplastic antigen Ma3 gene contains a functional heptanucleotide shift sequence and a classical H-type pseudoknot that promotes Ϫ1 frameshifting at ϳ20% efficiency both in vitro and in vivo.

EXPERIMENTAL PROCEDURES
Bioinformatic Analysis-Programs written in Perl were used to analyze 20,763 human mRNA sequences from the RefSeq data base at NCBI (www.ncbi.nlm.nih.gov/RefSeq/). The sequences were searched for ORF pairs in which ORF1 was the annotated coding sequence (CDS) and ORF2 extended at least 300 nt beyond ORF1 and was in the Ϫ1 frame relative to it. The overlapping sequences from these 5,709 ORF pairs were searched for the retrovirus-like heptanucleotide shift sites, A AAA AAC, U UUA AAC, G GGA AAC, and U UUU UUA, shown in the zero reading frame (18). For the 1,153 sequences where such a motif was found, a 50-nt region downstream from the shift site was analyzed with the zipfold server (19) to determine the minimum free energy of potential secondary structures. Sequences that had a ⌬G of Ϫ8 kcal/mol or less were analyzed with mfold (19,20) to generate graphical representations of all available folds. These RNA secondary structure graphics were inspected manually to search for stem-loops that could be extended to pseudoknots with similarity to known frameshift stimulators.
Protein Sequence Analysis-Searches with the Ma3 protein sequence were conducted with Psi-BLAST (21) in the nr data base at NCBI. Supplemental Table S2 shows the species and accession number of sequences considered to be homologs to Ma3. Sequences were searched for protein coding domains with the HMMER algorithm using the Pfam data base (22). Protein sequences from orthologs of the Ma3 gene in human, chimp, macaque, orangutan, mouse, and rat were aligned with ClustalX (23). Columns that contained any gaps were removed, and a phylogenetic tree was bootstrapped by the neighbor joining method also using ClustalX. The trees were prepared for publication with TreeView (24).

Construction of Clones for in Vitro and in Vivo Frameshifting Assays-
The 435-nt region of Ma3 containing the shift site and potential kissing loops sequence was amplified by PCR using primers ATATTAGGT-CACCAGGCTGCAGTTGAGTCGGGAAAC and ATATTAGG-TACCAGGGGTGGTTGGATGAGCAGGAC (BstEII and KpnI restriction sites underlined) for the GST-␤-globin vector, pGB01 (25). Either a human brain cDNA library (Invitrogen) or human genomic DNA (Promega) was used as template for the PCRs. Following amplification, the products were restricted and run on an agarose gel. The correct sized fragments were isolated from a gel slice and ligated into BstEII/KpnI-digested pGB01. A series of deletions was generated using primers to the appropriate 3Ј sequence and cloned as above. For the constructs containing mutations in the human Ma3 pseudoknot sequence or WT mouse Ma 3 and human Ma 5, complementary oligonucleotides were synthesized containing the overhangs for BstEII and KpnI.
The polylinker region of the dual luciferase vector, p2luc (26) was modified to introduce additional cloning sites. A region of p2luc was amplified by PCR using the following primers: ATATAAGTCGACTT-GGATCCCCCGGGGAGCTCAGATCTACGGGCCCTCTCGAGG-AAGACGCCAAAAACATAAAGAAAGGC (SalI, BamHI, SmaI, SacI, BglII, ApaI, and XhoI sites underlined) and CCAGAGGAATTCC-ATTA TCAGTGCAATTGTTTTGTC (EcoRI site underlined). The 692-bp PCR fragment was digested with SalI and EcoRI and ligated into p2luc, also digested with SalI and EcoRI, generating the vector, p2lucj.
The human Ma3 pseudoknot sequence was introduced on complementary oligonucleotides containing the overhangs for SalI and XhoI that were cloned into the SalI/XhoI-digested p2lucj. All of the sequences were verified by automated dideoxy sequencing.
In Vitro Frameshifting Assays-100 g of plasmid DNA from the GST-␤-globin constructs was added to a 10-l TNT T7-coupled transcription/translation rabbit reticulocyte lysate (Promega). Incubations were carried out at 30°C for 1 h. Following treatment with 100 g/ml RNase A at room temperature for 10 min, 30 l of SDS sample buffer was added. The products were separated by 15% SDS-PAGE, and the gels were fixed, dried, and exposed to a phosphorimaging screen. The data were collected on a Molecular Dynamics phosphorimaging instrument and analyzed by ImageQuant software. The frameshifting efficiencies were determined by normalizing for the number of methionine residues in each product and calculating the percentage of frameshifting [frameshift/(termination ϩ frameshift)] ϫ 100%. The WT human Ma3 pseudoknot sequence promotes frameshifting at ϳ20%. The efficiencies of the mutant, mouse Ma3, and human Ma5 sequences are reported relative to WT Ma3 (100%). All of the constructs were assayed at least three times, and the average is reported.
In Vivo Frameshifting Assays-The dual luciferase assays were performed as previously described (9).

RESULTS
A putative Ϫ1 frameshift site was identified in the human paraneoplastic antigen Ma3 gene. The original accession for Ma3 was NM_013364.1, and this sequence was used for the bioinformatic analysis. Shortly after work began on the experimental analysis of Ma3, NM_013364.1 was updated by NCBI staff to a new version, NM_013364.2, in which the sequence was revised. An internal region of 121 nt was deleted removing the heptanucleotide shift site. To resolve the conflict between these two sequences, diagnostic PCRs of human brain cDNAs with primers flanking the reported deletion were performed to determine whether the transcript was spliced as indicated by the revised sequence. No evidence for splicing was observed (data not shown). PCR products were also generated from the same cDNA library using primers specific to the 5Ј coding sequence of the initiating Ma3 reading frame and the 3Ј coding sequence of the Ϫ1 reading frame. These PCR fragments were cloned, and several clones were sequenced. There were no spliced variants, although there was evidence of editing in regions distant from the frameshift site in a minority of cases (data not shown).
The potential frameshift site in Ma3, G GGA AAC, is followed by a 6-nt spacer and a predicted classical H-type pseudoknot. The Ma3 sequence further 3Ј of the predicted pseudoknot was inspected, and two additional potential stem-loop structures were discovered, the loops of which could interact with the first stem-loop to form kissing loops (Fig.  1A). Therefore, the region tested encompassed the potential kissing loops sequence (411 nt 3Ј of the putative frameshift site) that included the potential pseudoknot.
The Ϫ1 Frameshift Event Is Efficient in Reticulocyte Lysates-To determine whether the predicted frameshift site in Ma3 is functional, the region including 15 nt 5Ј and 411 nt 3Ј of the G GGA AAC frameshift site was amplified by PCR. The region was cloned between two ORFs, glutathione S-transferase, and rabbit ␤-globin, such that Ϫ1 frameshifting was required for expression of the 3Ј ORF. The vector, pGB01, contained a T7 promoter, and frameshifting activity was monitored in a coupled transcription/translation rabbit reticulocyte lysate system. The cloned region of the Ma3 sequence, ϩ411, promoted Ϫ1 frameshifting at a level of 20% (Fig. 1B). To investigate the importance of the potential RNA structures, a series of deletions was constructed that successively removed the predicted structural elements, and the effects on Ϫ1 frameshifting were measured. Removal of the potential distal and proximal kissing loop structures, as well as sequences 3Ј of the potential pseudoknot, constructs ϩ374, ϩ271, ϩ198, and ϩ122, had a small positive effect on the frameshifting ( Fig. 1B and Table 1). The potential pseudoknot sequence, ϩ53, supported frameshifting at the same level as the sequence encompassing both the potential proximal and distal kissing loops. In contrast, a construct containing only the first stem-loop of the predicted pseudoknot, ϩ37, showed a dramatic reduction in frameshifting to 25% of the entire pseudoknot sequence (WT). Interestingly, when U34 was changed to C in the construct containing only the first stem-loop (ϩ37 C34), allowing the formation of a G-C pair in place of a G:U wobble pair, frameshifting was observed at 54% of the wild type level ( Table 1). The basal level of frameshifting with the G GGA AAC shift site and the 6-nt spacer (ϩ6) was 4% of WT ( Fig. 1B and Table 1). Mutation of the frameshift site to C GGC AAC to prevent tandem tRNA slippage further reduced frameshifting to 1% of WT ( Fig. 1B and Table 1).
Testing the Potential Pseudoknot Structure in Vitro-Sequences containing the presumptive pseudoknot were sufficient to stimulate a high level of Ϫ1 frameshifting. The pseudoknot structure was tested by making disruptive mutations in stems 1 and 2 as well as double mutations that restored the base pairing in the stems (Fig. 2A). The disruptive mutations in stem 1, C8C9, C15C16, G35G36, and G28G29 gave drastic reductions in Ϫ1 frameshifting, to 4, 3, 8, and 2% of WT, respectively (Fig. 2B, lanes 2-5, and Table 1). In constructs where the compensatory mutations would restore base pairing in stem 1, Ϫ1 frameshifting was observed at 62 (C8C9::G35G36) and 42% (C15C16::G28G29) of WT (Fig. 2B, lanes 6 and 7, and Table 1).
Other features of the pseudoknot were tested by making mutations. Changing the sequence of loop 1 from CA to GU, G18U19, had a positive effect on frameshifting (120% of WT) (Fig. 2B, lane 9, and Table 1). The presence of a presumptive unpaired A26 residue between stems 1 and 2 is reminiscent of the "bulged " A that induces a kink between two stems and is required for efficient frameshifting at the gag-pro junction in mouse mammary tumor virus (29,30). When the A26 residue was changed to C, C26, frameshifting was reduced to 21% of WT (Fig. 2C, lane 13, and Table 1). More strikingly, when A26 was deleted, frameshifting was virtually eliminated, 3% of WT (Fig. 2C, lane 12, and Table  1), indicating the importance of the A residue and raising the possibility that it is responsible for causing a required kink between the stems.
Ma5 and other Ma3 genes contain a putative Ϫ1 frameshift site, G GGA AAC, followed by a potential pseudoknot (see below). The frameshift regions from human Ma5 and mouse Ma3 were tested for activity in reticulocyte lysates. The human Ma5 sequence was 54% as efficient as the human Ma3 sequence, whereas the mouse Ma3 sequence was 75% as efficient (Table 1). There are five nucleotide differences between the mouse and human Ma3 pseudoknot regions. One change, U34 3 C replaces a G:U wobble pair in stem 1 in human Ma3 with a G-C pair, which should increase the thermodynamic stability of the stem. The effect of the G10-C34 base pair in the context of the human Ma3 pseudoknot is to increase frameshifting to 150% of WT (Fig. 2B, lane 8, and Table 1). Two of the other nucleotide differences between the mouse and human sequences occur in the spacer region. Replacement  ؊1 Frameshifting in Paraneoplastic Antigen Ma3 MARCH 17, 2006 • VOLUME 281 • NUMBER 11 of G1 by A or A5 by G in the human sequence resulted in 52 and 115% of WT frameshifting ( Table 1). The remaining two nucleotide differences occur in the loop regions of the pseudoknot. Replacing C18 in loop 1 with G or G40 in loop 2 with A in the human sequence increased frameshifting to 108 and 120% of WT, respectively (Table 1).
Testing the Pseudoknot Sequence in Vivo-The human Ma3 pseudoknot sequence promotes efficient Ϫ1 frameshifting in vitro. To test whether frameshifting is functional in vivo, the pseudoknot region was cloned into a dual luciferase reporter where expression of the 3Ј reporter, firefly luciferase, is dependent on Ϫ1 frameshifting. The plasmids were transfected into HEK 293 cells, and Renilla and firefly luciferase activities were measured in cell lysates. The Ma3 pseudoknot promoted Ϫ1 frameshifting at a level of 18%, comparable with the in vitro level of 20% (data not shown).
Orthologs in Other Organisms-Homology searches with BLAST (21) using human paraneoplastic Ma3 sequence revealed similar Ma3 genes in mouse, rat, chimpanzee, and macaque. The alignment of the pseudoknot sequences shows that the stem regions are highly conserved, whereas there are some differences in the spacer and loops (currently the only chimp sequence available ends at the zero frame terminator codon within the pseudoknot). In the rodents, stem 1 is strengthened by the replacement of the G-U wobble pair (in primates) with a G-C pair. Although these species are closely related, the fact that mutations in the pseudoknot region appear to be accumulating preferentially in the loop and spacer regions suggests selection for the pseudoknot.
Family of Ma Genes-There are currently six genes in the human genome annotated as being part of the paraneoplastic Ma family. These are designated Ma1, Ma2, Ma3, MAOP-1 (Ma4), Ma5, and PNMA6A. We will use Ma6 to refer to PNMA6A in this paper. Each of the sequences was analyzed for frameshifting features and protein coding domains. All of the Ma genes show homology to the same region of various retrotransposon Gag proteins (Fig. 3), although only the Ma2 and Ma3 genes have homology that are assigned e values of less than 0.001 by the HMMER algorithm (22). Ma3 is the only member of the family to contain a CCHC zinc finger. As noted above, Ma5 has the same G GGA AAC shift site and a pseudoknot, distinct but similar in both sequence and structure to the Ma3 pseudoknot. Although Ma6 sequence has the potential to form a strong stem-loop or pseudoknot near the end of ORF1, it is quite different in both structure and sequence from the pseudoknot in Ma3. None of the other Ma genes contain a classical retroviral shift site including the G GGA AAC motif found in Ma3.
Phylogenetic analysis of the members of the Ma gene family from various mammals shows that the Ma3 genes from orangutan and crabeating macaque (NCBI GenBank TM accession numbers CR861370.1 and AB062932, respectively) cluster with the Ma2 genes from the other species (Fig. 4). Closer analysis of these gene sequences reveals that they have clearly been incorrectly annotated and are, in fact, Ma2 genes.

DISCUSSION
In this systematic search for new cases of recoding, the first mammalian cellular gene identified with a classic Ϫ1 programmed shift site and stimulatory pseudoknot was human Ma3. The Ma3 gene has many features similar to retroelements including conservation of a well defined zinc finger and other regions of the Gag protein from several retrotrans-  posons. However, none of the Ma family of genes code for domains in the Pol protein, i.e. aspartyl protease, reverse transcriptase, or integrase, and thus, they are certainly not associated with any active retroelements. The only previously known mammalian gene to utilize Ϫ1 programmed frameshifting in its expression, edr (embryonal carcinoma differentiation regulated) in mice (10,11) and PEG10, the homolog in humans, resembles Ma3 in having homology to retroelement gag genes. The CCHC-zinc fingers are similar in Ma3 and edr, but the other retrovirus-like feature in edr, a putative aspartyl protease catalytic site, is absent in Ma3.
Interestingly, searches for homology to the Ma3 protein using Psi-BLAST (21) reveal that the best homology hits in the sequence data bases are to retroelements from Japanese rice. In these retroelements, however, a single frame encodes the Gag-Pol polyprotein (as is the case with most retroelements in plants (31)), whereas Ma3 clearly has a retrovirus-like Ϫ1 frameshifting mechanism. This similarity could have resulted from an artifact of the Psi-BLAST search algorithm; inferring phylogeny from homology is not always accurate. It could also suggest that the history of the Ma family of genes is more complex than straightforward descent from a single retroelement.
Proteins encoded by mobile genetic elements occasionally evolve to assume cellular roles (32,33). It is perhaps not surprising that a proportion of them also utilize frameshifting, which is well known in the decoding of retroviruses and, to an increasing extent, in retroelements (31,34), including in mammalian cells (35), yeast (36), and Drosophila (37).
There is a family of Ma genes that includes Ma1, Ma2, Ma3, Ma4 (MOAP-1), Ma5, and Ma6. All of the genes identified thus far in this family are single copy. Ma1 and Ma4 are on chromosome 14, Ma2 is on chromosome 8, and Ma3, Ma5, and Ma6 are on the X chromosome. Although little work has been done on any of these genes, they are known to be expressed in brain as well as other tissues and are also expressed in a range of tumors. Serious neurological phenotypes are associated with tumor expression of the these proteins. Paraneoplastic Ma3 genes have been found in human, chimpanzee, macaque, mouse, and rat. All five of these sequences preserve the G GGA AAC motif and the 3Ј pseudoknot (except for the incomplete chimp sequence as noted above). There is variation between the primate and rodent sequences in the region of the pseudoknot, but these variations lie within presumptive single-stranded regions with the exception of a G:U wobble pair in the primate stem 1 that is replaced by a G-C pair in rodents. Interestingly, although Ma3 and Ma5 genes have active Ϫ1 frameshifting sites, the retroelements to which they are most closely related have Gag-Pol polyproteins encoded by a single frame. The other Ma genes do not contain a retroviral shift motif or a pseudoknot.
Among the Ma3 gene sequences from various species, the protein sequence of the putative ORF2 appears to be much less conserved than ORF1. The ORF2 protein has no similarities to any known protein in public sequence data bases. Although long evolutionary conservation is a hallmark of functional importance, the products of certain ORFs, including those involved in human brain function, may be subject to recent selection and be less matched than average to mouse counterpart genes.
A classical H-type pseudoknot has been shown to promote efficient Ϫ1 frameshifting at a retrovirus-like shifty heptanucleotide sequence in the human Ma3 gene. It is very similar to the well characterized infectious bronchitis virus (IBV) pseudoknot in the lengths of stems 1 and 2 (6,38). Notably, the base of IBV stem 1 is a G-C pair, whereas it is a U-A pair in Ma 3. Loop 2 in the Ma3 pseudoknot is 10 nt, shorter than the 32-nt loop 2 in its IBV counterpart, but reduction of IBV loop 2 to 8 nt does not affect its frameshifting efficiency (39).
Although the potential exists for interactions between the first stemloop of the Ma3 pseudoknot and distant sequences 3Ј to form kissing loops, these interactions are not necessary to promote the Ϫ1 frameshift event as shown by a 3Ј-deletion series. In the human Ma3 pseudoknot, mutations at the top of stem 1 (C15C16 and G28G29) are more detrimental than those at the base of stem 1 (C8C9 and G35G36). The two constructs designed to restore base pairing in stem 1 gave lower levels of frameshifting than the wild type pseudoknot sequence. This has been observed for other recoding stimulatory pseudoknots and may indicate a primary sequence requirement or reflect subtle structural or thermodynamic effects. In the Ma3 pseudoknot, the disruptive mutations in stem 2 were asymmetric in that mutations to the 3Ј part of the stem were less detrimental than those to the 5Ј part of the stem. However, the restorative mutations resulted in frameshifting activities closer to the wild type level. Although only one mutant loop 1 sequence was tested, it had a slight positive effect on frameshifting, indicating at least some flexibility in sequence requirements.
Perhaps the most striking similarity with a subset of frameshift stimulating pseudoknots is the critical importance of the A residue at the junction of stems 1 and 2. Deletion of this residue effectively eliminated frameshifting, whereas changing the A to a C reduced frameshifting by 5-fold. The importance of an A residue at the stem junction indicates that its function may be to induce a required kink between the two stems as is the case in the mouse mammary tumor virus gag-pro (30,40) and feline immunodeficiency virus (41) pseudoknots.
A minority of the known cases of programmed Ϫ1 frameshifting, especially those in bacteria, utilize a simple or elaborated stem-loop rather than a pseudoknot as their 3Ј recoding stimulator. The list includes IS911 (42), human immunodeficiency virus gag-pol (43)(44)(45), an RNA polymerase-encoding sequence of a human astrovirus (46), and E. coli dnaX, which encodes DNA polymerase III subunits (47). However, in general, the investigated Ϫ1 frameshift stimulatory pseudoknots are not effectively replaceable by even strengthened versions of their component stem 1 (6). In contrast, the first stem-loop alone of the Ma3 pseudoknot shows considerable activity, and when the stem-loop is strengthened by replacing a G:U wobble pair with a G-C pair, the structure is 50% as efficient as the pseudoknot. Without detailed structural knowledge of the pseudoknot, it is difficult to understand how mutations that disrupt base pairing in stem 2 can have more severe detrimental effects on frameshifting than the absence of stem 2. Perhaps there are tertiary interactions involving stem 2 and other regions of the pseudoknot that directly or indirectly affect stem 1. A mutation in stem 2 could then theoretically interfere with the structure of stem 1 and eliminate its stimulatory activity observed when stem 2 is not present. (A related finding was reported for a stem-loop sequence within loop 2 of the stimulatory pseudoknot for Ϫ1 frameshifting in SARS-CoV. The deletion of the stem-loop sequence had no appreciable effect, whereas mutations within the stem-loop sequence reduced frameshifting significantly (9,48,49).) Alternatively, the activity of the stem-loop may be artificially high because two sequences, GGUUC and GGCUC, further 3Ј in the vector sequence can potentially base pair with the stem-loop to form pseudoknots with large loops 2 (131 or 195 nt, respectively). Previous work has shown that the stability of stimulatory stem-loops is correlated with the level of frameshifting (47,50) and that flanking sequences, including the spacer, can influence the level of frameshifting mediated by particular structures or sequences (51,52) The 3Ј stimulator cannot be considered in isolation. This is exemplified by comparison of the mouse and human Ma3 pseudoknot sequences, which differ by 5 nt. One of these differences creates the potential for a stronger stem 1 in the mouse pseudoknot because of a G10-C34 base pair replacing G:U. The prediction was that the mouse Ma3 pseudoknot would be more efficient at promoting Ϫ1 frameshifting than the human version, but unexpectedly, the mouse pseudoknot was only 75% as efficient. The most likely explanation for the reduced activity is the A residue immediately 3Ј to the shift site, because this change made alone in the human pseudoknot reduced frameshifting by ϳ50%.
The degree of importance of frameshifting for the expression of transframe products in human PEG10, Ma3, Ma5, and other genes that may utilize programmed Ϫ1 frameshifting is of great relevance when considering the utility of compounds being sought to affect frameshifting efficiency as pharmaceuticals (53). For instance, although this approach could be important in altering the critical Gag to Gag-Pol ratio of retroviruses (54) and in the treatment of human genetic disorders that result from frameshift mutations, an understanding of the extent and importance of frameshifting in the human genome is critical because of the possible side effects of such therapeutics. The search for new cases of recoding in this report is based on finding examples of the well established Ϫ1 frameshifting model exhibited in retroviruses. Many other types of recoding are known to be utilized by living systems in a host of functional ways. What more will we discover as we continue to learn how eukaryotes utilize translational recoding as a rich source of proteomic diversity?