Structural determinants of an internal ribosome entry site that direct translational reading frame selection

The dicistrovirus intergenic internal ribosome entry site (IGR IRES) directly recruits the ribosome and initiates translation using a non-AUG codon. A subset of IGR IRESs initiates translation in either of two overlapping open reading frames (ORFs), resulting in expression of the 0 frame viral structural polyprotein and an overlapping +1 frame ORFx. A U–G base pair adjacent to the anticodon-like pseudoknot of the IRES directs +1 frame translation. Here, we show that the U-G base pair is not absolutely required for +1 frame translation. Extensive mutagenesis demonstrates that 0 and +1 frame translation can be uncoupled. Ribonucleic acid (RNA) structural probing analyses reveal that the mutant IRESs adopt distinct conformations. Toeprinting analysis suggests that the reading frame is selected at a step downstream of ribosome assembly. We propose a model whereby the IRES adopts conformations to occlude the 0 frame aminoacyl-tRNA thereby allowing delivery of the +1 frame aminoacyl-tRNA to the A site to initiate translation of ORFx. This study provides a new paradigm for programmed recoding mechanisms that increase the coding capacity of a viral genome.


INTRODUCTION
Protein synthesis is a highly accurate process where errors in translation occur at low frequency about 5 × 10 −5 per codon. Although we have a basic understanding of transfer ribonucleic acid (tRNA) discrimination by the ribosome, the maintenance of the open reading frame (ORF) during translation is still not well understood. Viral strategies have been particularly informative, as compact viral genomes have evolved mechanisms to increase their coding capacities for successful infection. In particular, programmed frameshifting can increase coding capacity in viral genomes and in some cellular messenger RNAs (mR-NAs) (1)(2)(3).
We recently discovered a novel recoding mechanism found within a subset of the Dicistroviridae family (4). Di-cistroviruses possess a positive-sense, single-stranded RNA (ssRNA) genome that contains two main ORFs encoding the non-structural and structural viral proteins, respectively (5,6). Different internal ribosome entry sites (IRESs) direct translation of the two main ORFs (7). The intergenic (IGR) IRES has several unique properties; the IRES can recruit the ribosome without the need of initiation factors and initiates translation from a non-AUG codon in the ribosomal A site (8). The IGR IRES adopts a structure containing three pseudoknots (PKI-III). PKII/PKIII form a compact structure that is responsible for ribosome recruitment and binding (9)(10)(11)(12). Structural and biochemical studies have revealed that the PKI domain adopts a tRNA anticodon-codon-like interaction that occupies the ribosomal P site to start translation by delivery of the first aminoacyl-tRNA to the A site (13,14). A recent cryo-electron microscopy (EM) structure of the CrPV IGR IRES bound to the yeast ribosome at 3.7-3.8-Å resolution has provided additional insights into the IGR IRES mechanism: the PKI domain first occupies the ribosomal A site and translocation of the IRES by eEF2 occurs prior to delivery of the first aminoacyl-tRNA (15). Subsequently, the PKI domain occupies the ribosomal P site to drive translation from the ribosomal A site using a non-AUG codon. Thus, the IRES hijacks and manipulates the ribosome by functionally mimicking a tRNA. After ribosome recruitment, eEF1A mediates delivery of the first aminoacyl-tRNA to the A site and eEF2 catalyzes the initial translocation step in the absence of peptide bond formation, termed pseudotranslocation (8,16,17).
Bioinformatics studies have revealed a hidden gene called ORFx which is downstream of the IRES and overlaps with the structural protein ORF within a subset of dicistrovirus genomes including the fire ant virus, Solenopsis invicta virus-1 (SINV-1) and the honey bee viruses, Israeli acute paralysis virus (IAPV), Acute bee paralysis virus (ABPV) and Kashmir bee virus (KBV) (18,19). We previously showed that the IAPV IGR IRES directs translation of ORFx in the +1 frame that is dependent on a U6562/G6618 base pair adjacent to the PKI tRNA-like domain ( Figure 1) (4). Moreover, an ORFx peptide was detected in virally infected honey bees by mass spectrometry analysis, suggesting a role of ORFx during virus infection (4). This work demonstrated a novel recoding mechanism utilizing  The overlapping +1 frame ORFx, within the 0 frame ORF2, is shown. Conserved nucleotides among type II IGR IRESs are in capital letters. The CCU triplet, in part, mediates PKI base pairing and occupies the P site, whereas the first codon of the 0 frame ORF2 is the adjacent GGC codon in the A site. Translation of the +1 frame ORFx is directed by a U6562/G6618 base pair adjacent to PKI (in red nucleotides). The start codon in the +1 frame is a GCG alanine codon.
an IRES to initiate translation in the 0 and +1 frames to produce two overlapping gene products and suggested a 'yardstick' model in which the adjacent U-G base pair shifts the reading frame by +1. In this study, we expand on this model and provide extensive analyses using mutagenesis and SHAPE to show that the IRES adopts distinct conformations to direct reading frame selection. In addition to the U-G adjacent base pair, we have uncovered novel classes of mutations that direct only 0 or +1 frame translation. Finally, ribosome positioning experiments suggest that the viral IRES directs reading frame selection after ribosomes have assembled on the IRES. This work describes a novel strategy to alter the ribosome reading frame to increase the coding capacity of a viral genome.

In vitro translation assays
Bicistronic luciferase plasmids, linearized with XbaI, were incubated in Sf21 extract (Promega) with an additional 40-mM potassium acetate and 0.5-mM magnesium chloride (final concentration) for 2 h at 30 • C in the presence of [ 35 S]-methionine (PerkinElmer). Reactions were analyzed by sodium dodecyl sulphate-polyacrylamide gel electrophoresis (SDS-PAGE) and by phosphorimager analysis (Typhoon, Amersham).
Monocistronic luciferase-containing plasmids were linearized with SpeI. IAPV IGR IRES RNAs were transcribed in vitro using a bacteriophage T7 RNA polymerase reaction and purified with an RNeasy kit (Qiagen). The integrity and purity of the transcribed RNAs were confirmed by gel analysis.

In vivo translation assay
We previously demonstrated that the fusion of ORFx inframe with the Firefly luciferase (FLuc) ORF inhibits FLuc enzymatic activity and the inclusion of the Thosea asigna virus 2A peptide (T2A) into the bicistronic reporter system increases the sensitivity of the luciferase assay (20). Capped reporter RNAs from bicistronic T2A containing plasmids were generated by in vitro transcription in the presence of a cap analog [m 7 G(5 )ppp(5 )G] (Ambion) at a 5:1 ratio to GTP. Drosophila S2 cells were grown and passaged in M3+BPYE medium plus 10% fetal bovine serum (FBS) at 25 o C. Capped bicistronic reporter RNAs (2 g) were transfected into 3 × 10 6 S2 cells with Lipofectamine 2000 (Invitrogen). After 6 h of transfection, cells were harvested, lysed and luciferase activity was measured by dual-luciferase reporter assay (Promega) as previously described (20).

SHAPE probing
SHAPE probing was performed as described (21). An amount of 20 pmol of the RNA was heated to 85 • C for 2 min, followed by the addition of Buffer E (final concentration of 20-mM Tris, pH 7.5, 0.1-M KCl, pH 7.0, 2.5-mM MgOAc, 0.25-mM spermidine and 2-mM dithiothreitol (DTT) and incubated at 30 • C for 20 min. To modify the RNA, N-methylisatoic anhydride (NMIA) dissolved in dimethyl sulfoxide (DMSO) was added (final concentration 3 mM) and incubated for 1.5 h at 30 • C. Control reactions containing only DMSO (no NMIA) were performed in parallel. The modified RNA was recovered by ethanol precipitation, followed by primer extension analysis. Reactions were resolved on 8% (w/v) denaturing polyacrylamide sequencing gels at 60 W. The gels were dried and analyzed by phosphorimager analysis. Individual band intensities were quantitated using semi-automated footprinting analysis (SAFA) (22) and normalized as described (23). Band intensity variations for each nucleotide were normalized by dividing all intensities by the average value of the next 10% of reactive nucleotides, after excluding the top 2% of the reactive nucleotides (23).
For SHAPE analysis of IRES/ribosome complexes, 20 pmol of the RNA was heated to 85 • C for 2 min, followed by the addition of Buffer E (final concentration of 20-mM Tris, pH 7.5, 0.1-M KCl, pH 7.0, 2.5-mM MgOAc, 0.25-mM spermidine and 2-mM DTT) and incubated at 30 • C for 20 min. 80S/IRES complexes were formed by adding 20 pmol of purified 80S Hela ribosomes to the folded RNA for 10 min at 30 • C. To modify the RNA, NMIA was added to the reaction at a final concentration of 3 mM and incubated for 1.5 h at 30 • C. IRES-ribosome complexes were separated by ultracentrifugation through 10-35% (w/v) sucrose gradients containing 20-mM Tris, pH7.5, 0.1-M KCl, pH 7.0, 5-mM MgOAc, 0.25-mM spermidine, 2-mM DTT. After fractionation, IRES-80S complexes were collected, ethanol-precipitated and resuspended in water. Primer extension was performed and reactions were resolved on 8% (w/v) denaturing polyacrylamide sequencing gels at 60 W. The gels were dried and analyzed by phosphorimager analysis. Individual band intensities were quantitated using SAFA and normalized as described (22). Control experiments including reactions in the absence of NMIA or ribosomes were performed in parallel. The difference in SHAPE reactivities of nucleotides between wild-type and mutant IRES/ribosome complexes was calculated. The averages and standard deviations were calculated from at least three independent experiments.

Purification of 40S and 60S ribosomal subunits
Ribosomal subunits were purified from HeLa cell pellets (National Cell Culture Center). HeLa cells were lysed in Triton-X lysis buffer (15-mM Tris-HCl (pH 7.5), 300-mM NaCl, 1% (v/v) Triton X-100, 6-mM MgCl 2 , 1-mg/ml heparin) and the supernatant was subjected to brief centrifugation to remove cellular debris. The supernatant was applied to a 30% (w/w) sucrose cushion containing 0.5-M KCl and centrifuged at 100,000 g to pellet ribosomes. Ribosomes were resuspended in buffer B (20-mM Tris-HCl (pH 7.5), 6-mM magnesium acetate, 150-M KCl, 6.8% (w/w) sucrose, 2-mM DTT) and subsequently treated with puromycin to dissociate ribosomes from the mRNAs. KCl was added to obtain a final concentration of 0.5 M. The dissociated ribosomes were resolved on a 10-30% (w/w) sucrose gradient where the peaks corresponding to the free 40S and 60S subunits were detected by measuring the absorbance at 260 nm. The corresponding fractions were collected, pooled and concentrated in buffer C (20-mM Tris-HCl (pH 7.5), 0.2-mM ethylenediaminetetraacetic acid (EDTA), 10-mM KCl, 1-mM MgCl2, 6.8% (w/w) sucrose) using Amicon Ultra spin concentrators (Millipore). The resultant concentrations of the ribosomal subunits were determined by spectrophotometry using the conversions 1 A260 nm = 50 nM and 1 A260 nm = 25 nM for the 40S and 60S subunits, respectively.

Reconstitution of IRES-ribosome translation
To reconstitute translocation on IRES-ribosome complexes, 150-ng bicistronic wild-type or mutant RNAs were annealed to primer PrEJ761 as described above. Following primer annealing, the RNAs were incubated with 100-nM or 150-nM purified HeLa 40S and 60S subunits, respectively, in the presence of 1-mM ATP, 0.4-mM GTP and 0.5-mg/ml cycloheximide. Purified yeast elongation factor 1A (30 ng/l) and elongation factor 2 (50 ng/l) and bulk aminoacyl-tRNAs were added to promote translocation. Reverse transcription and visualization of cDNA products were performed as above.

Base pairing between nucleotides 6562 and 6618 is important but not absolutely necessary for +1 frame translation
Previously, we demonstrated that the U6562/G6618 base pair adjacent to the IAPV PKI domain (see Figure 1) directs +1 frame translation of ORFx (4). To examine in detail whether the nucleotide identities of U6562 and G6618 are important for 0 or +1 frame translation, we generated mutant IRESs containing all 16 base permutations at positions 6562 and 6618 ( Figure 2B-E). IRES-mediated 0 and +1 frame translation were assessed by using a bicistronic reporter construct, in which the downstream +1 frame ORFx Relative Translational Activity 0.5 is fused in-frame with FLuc to produce an ORFx-FLuc fusion protein (∼72 kDa) ( Figure 2A). This fusion also results in a shortened 0 frame ORF2 (Figure 2A, sORF2). Thus, scanning-dependent Renilla luciferase (RLuc) and IGR IRES-mediated 0 and +1 frame translation are monitored simultaneously from the same construct. The translational activities were analyzed by incubating the reporter constructs in an insect Sf21 coupled transcription/translation extract containing [ 35 S]-methionine, followed by quantifying the radiolabeled protein products by SDS-PAGE analysis. As shown previously, Northern blot analyses showed that the bicistronic RNAs synthesized in these extracts are intact and full-length ( Figure 2C) (4). The translational activities of the 16 different permutations are summarized in Figure 2D and Supplementary Figure S1 in two graphs that are organized based on the base identities at positions 6562 and 6618. Figure 2D shows the relative translational ratio of +1 frame to 0 frame, whereas Supplementary Figure S1 normalizes the translational activities to both the wild-type 0 and +1 frame translation set as 1.
Depending on the identity of the bases at positions 6562 and 6618, the efficiency of translation in either 0 or +1 frame as well as the ratio of efficiencies varied considerably. G6618 mutated to C or U but not to A abolished +1 frame translation, consistent with the model that base pairing between nucleotides 6562 and 6618 is necessary for +1 frame translation ( Figure 2C, lanes 2, 4-6; Figure 2D) (4).
In support of this model, combinations that maintained base pairing such as U6562A/G6618U, U6562C/G6618, U6562G/G6618C and U6562G/G6618U ( Figure 2C, lanes 10, 11, 17, 18) rescued +1 frame translation albeit to different extents (4), whereas mutation combinations that do not base pair, such as U6562A/G6618C, U6562C/G6618C, and U6562/G6618U, inhibited +1 frame translation, thus reinforcing the importance of base pairing between nucleotides 6562 and 6618. However, surprisingly, Watson-Crick base pairing is not a strict requirement for mutant IRESs with position 6618 as G or A. For example, mutation of U6562 to G or A ( Figure 2C, lanes 7 and 15) or a combination of U6562G/G6618A ( Figure 2C, lane 16), all of which should not form a Watson-Crick 6562/6618 base pair, resulted in a significant level of +1 frame translation. Interestingly, specifically mutating U6562 to G regardless of the nucleotide identity of 6618 significantly increased +1 frame translation ( Figure 2D, left graph, and Supplementary Figure S1B). This is most evident when the ratio of +1/0 frame translation is plotted ( Figure 2E). The mutant IRES U6562G, which increased +1 frame translation by ∼4.5-fold as compared to the wild-type IRES, displayed similar levels of 0 frame translation ( Figure 2D and Supplementary Figure S1). Altering the first +1 frame GCG codon to a stop UAG codon in combination with the U6562G mutation eliminated +1 frame translation, confirming that translation starts at the +1 frame GCG codon (Qian,R., unpublished data). As shown previously, mutating CC6615-6 to GG which disrupts PKI base pairing eliminated both 0 and +1 frame translation ( Figure 2B; Figure 2C, lane 3), whereas compensatory mutations that restore base pairing rescued translation in both frames (Supplementary Figure  S1B and C), demonstrating that 0 and +1 frame translation are dependent on the integrity of PKI and are IGR IRES-dependent (4). In summary, these results demonstrate that +1 frame translation mediated by the wild-type IRES (for simplicity, we will refer to the IAPV IGR IRES as IRES unless specifically stated) is dependent on base pairing between nucleotides 6562 and 6618. However, in specific mutant IRES contexts, the adjacent base pairing is not absolutely necessary for +1 frame translation. The wild-type IAPV IRES directs +1 frame translation at ∼20-25% of 0 frame translation in vitro ( Figure 2C-E and Supplementary Figure S1) (4). We reasoned that this may be due to differences in the levels of or in the delivery efficiency of the incoming aminoacyl-tRNA to the 0 or +1 frame initiating codon. To address the latter possibility, we mutated the +1 frame GCG alanine codon to a GGG glycine codon (C6620G), such that both the 0 and +1 initiate with the GGG glycine codon and should utilize the same Gly-tRNA for delivery to the A site ( Figure 3). Compared to the wild-type IRES, the C6620G mutant IRES displayed a lower level of translation in both frames, which is likely due to the particular Gly-tRNA that is delivered to the GGG codons in the 0 and +1 frames ( Figure 3). Comparing the ratio of +1/0 frame translation, +1 frame translation occurred at 30% frequency of 0 frame translation for the C6620G mutant ( Figure 3, 10% versus 29%), thus reflecting the true ratio of +1 to 0 frame translation mediated by the IAPV IRES.

SHAPE analysis of the wild-type IAPV IGR IRES
Our mutagenesis analyses suggest that domains within the IRES other than base pairing between nucleotides 6562 and 6618 may contribute to +1 frame translation. By extension, we hypothesize that the IRES may adopt different conformations to direct either 0 or +1 frame translation. To address this, we probed the structure of the wildtype IAPV IRES using selective 2 -hydroxyl acylation analyzed by primer extension (SHAPE) analysis. Briefly, reactivity of nucleotides to chemicals such as NMIA reveals the flexibility of nucleotides within the RNA structure (21). Using primer extension analysis, the location and extent of modification of each nucleotide are determined by comparing band intensities from the modification reaction to those of an unmodified control. We focused our attention on the PKI domain as this region likely directs +1 frame translation. For the wild-type IRES, nucleotides that were reactive to NMIA were observed within single-stranded regions of the Variable Loop Region (VLR), the loop of SLIII, the unpaired A6554 bulge and U6562 ( Figure 4A and Supplementary Figure S2). In general, nucleotides within predicted helical stems and base pairs were not reactive to NMIA (Figure 4). The one exception is within PKI. Specifically, nucleotides G6563, GCA6565-7 and U6613 within the wild-type IRES were reactive to NMIA ( Figure 4A). These results are in agreement with previous reports that the PKI domain may be dynamic and is reactive toward chemicals and enzymes that recognize both ssRNA and doublestranded RNA (9,10).
We next interrogated the structure of the IRES bound to the ribosome. Previous studies have shown that the PKI domain of the related dicistrovirus, CrPV, undergoes subtle conformational changes upon ribosome binding (13). The IAPV IGR IRES was complexed with salt-washedpurified human ribosomes, subjected to NMIA, followed by sucrose gradient fractionation to isolate IRES-ribosome complexes. The IRES RNA was purified and sites of modification were analyzed by primer extension. The differences in NMIA reactivities of nucleotides of IRES alone versus IRES-ribosome complexes are shown in Figure 4B. In general, nucleotides that were NMIA-reactive within the IRES in solution such as A6554, U6613, PKI and VLR showed decreases in NMIA reactivity upon IRES binding to the ribosome. This suggests that these nucleotides become less flexible upon ribosome binding. In contrast, C6603 and CA6611-2 were more reactive in IRES-ribosome complexes than in the isolated IRES, indicating that these nucleotides become more flexible when the ribosome is assembled on the IRES.

Novel mutant IAPV IGR IRESs that direct 0 and +1 frame translation
The SHAPE analysis of the wild-type IAPV IRES suggests that the IRES may undergo conformational changes upon ribosome binding, especially within ssRNA regions and the PKI base pairing domain. To investigate this more closely, we systematically engineered mutations within the IRES guided by the SHAPE analysis. IRES-mediated 0 and +1 frame translation were monitored using the bicistronic reporter construct described in Figure 2A. Figure 5 summarizes the data by calculating the percentage of 0 and +1 frame translation normalized to the wild-type IRES (given as 100%).
PKI base pairing is required for 0 and +1 frame translation. We first focused on the anticodon-like PKI domain since the majority of changes in SHAPE reactivities occur in this region. The nucleotides on the upper strand (residues 6563-6567) within PKI were sensitive to NMIA reactivity. We reasoned that strengthening or weakening the base pairs in PKI may differentially affect 0 or +1 frame translation. We chose to mutate the bottom strand in order to not disrupt the anticodon-like loop of PKI. Mutating U6613A had a deleterious effect on +1 frame translation (∼60% decrease) but did not significantly affect 0 frame translation (Figure 5, blue box). Further disruption of PKI by changing UG6613-4 to AC, abolished both 0 and +1 frame translation, indicating that a minimum of four base pairs is required for IRES activity and translation in either frames ( Figure 5, blue box). To further confirm this result, we mutated U6617, which is adjacent to the GGC start codon, to the other bases ( Figure 5, red box). Mutating U6617 to G, at the most right base pair of PKI, decreased 0 and +1 frame translation by ∼85 and 60%, respectively ( Figure 5, red box). In contrast, strengthening the base pair by changing U6617 to C increased 0 frame translation by 2-fold and decreased +1 frame translation by 70%. However, changing U6617 to A, which should disrupt the same base pair, increased 0 frame translation but decreased +1 frame translation ( Figure 5, red box). The latter result suggests that the nucleotide identity may be important in certain contexts. In general, U6617 is absolutely necessary for optimal +1 frame translation. Moreover, strengthening the right most base pair from G6563/U6617 to G6563/U6617C increases 0 frame translation. To complement our investigation of the right side of PKI, we introduced mutations into position G6563. Mutating G6563 to A, allowing base pairing in PKI, increased 0 frame translation and decreased +1 frame translation ( Figure 5, yellow box). In contrast, mutating G6563 to C or U eliminated both 0 and +1 frame translation ( Figure 5, yellow box), which is in agreement that base pairing between nucleotides 6563 and 6617 is important for IRES-mediated 0 frame translation. In summary, these results suggest that all five base pairs in PKI are required for IRES-mediated +1 frame translation. In contrast, 0 frame translation can tol- The difference in NMIA reactivity (radioactivity) of each nucleotide within the PKI domain is shown. To the right, secondary structure of the PKI domain is shown. Nucleotides that showed a significant increase (red) or decrease (blue) (greater than 1 SD) in NMIA reactivity are labeled. Data represent the average of at least three independent experiments ± SD. erate modest disruption of PKI. Finally, our data rule out the hypothesis that +1 frame translation is due to a realignment of base pairs in PKI to shift the reading frame.

Mutation of unpaired nucleotides within the PKI domain.
Phylogenetic analysis of the Type II IGR IRES secondary structures predicts that the PKI anticodon-like loop contains four unpaired nucleotides (Supplementary Figure  S3A). Within the IRES, the unpaired nucleotides of PKI, C6561, U6562, G6568 and U6569 showed differential reactivities to NMIA by SHAPE probing (Figure 4). In particular, C6561 and G6568 were not reactive to NMIA ( Figure  4A). To examine these nucleotides further, we mutated them to other bases. Mutating G6568 to A, C or U inhibited +1 but not 0 frame translation ( Figure 5, orange box). In contrast, mutating C6561 to A, U or G decreased 0 frame translation by 60-75% and increased +1 frame translation up to 2.5-fold ( Figure 5, green box). These data show that the identities of C6561 and G6568 are important for the selection of 0 or +1 frame translation.
U6569 is predicted to be unpaired within the anticodonlike loop of PKI and is reactive to NMIA by SHAPE (Figure 4A). Deletion of U6569 ( U6569) surprisingly resulted in complete loss of 0 frame translation and a significant increase in +1 frame translation (∼3.5-fold of wild-type) (Figure 5, magenta box). Furthermore, mutating U6569 to A, C or G decreased 0 frame translation and increased +1 frame translation ( Figure 5, magenta box). Thus, U6569 is important for 0 but not +1 frame translation.
A6554 is predicted to be unpaired and is highly sensitive to NMIA ( Figure 4A). Mutating A6554 to U, C or G did not have an effect on +1 frame translation but increased 0 frame translation by 70% when mutated to C or G ( Figure 5, black box). Interestingly, deletion of A6554 ( A6554) abol- ished 0 frame but not +1 frame translation, suggesting that the unpaired A6554 is required for 0 frame translation (Figure 5, black box). Specifically, it is noteworthy that deletion of U6569 or A6554 essentially abolished 0 frame translation and maintained or increased the level of +1 frame translation. In contrast, mutating G6568 specifically inhibited +1 frame translation.
We wished to determine if this trend is specific to IAPV or whether the other Type II IGR IRESs like KBV, SINV-1 and ABPV, which can direct +1 frame translation (4), also exhibited similar nucleotide dependencies. As predicted, disruption of the PKI base pairing within each Type II IGR IRES eliminated both 0 and +1 frame translation (Supplementary Figure S3B and C), indicating that +1 frame translation is IGR IRES-dependent. Furthermore, we deleted unpaired nucleotides within the KBV, SINV-1 and ABPV IGR IRESs at equivalent positions to A6554 and U6569 of the IAPV IGR IRES (Supplementary Figure S3A). Deletion of C6566 or C6581 within the KBV IGR IRES stimulated +1 frame translation by ∼2-4-fold (Supplementary Figure S3B and C). Similar deletions within the SINV-1 ( A4361 and 4376) and ABPV IGR IRESs ( A6475 and U6490) resulted in either exclusive +1 frame translation or an increase in +1 to 0 frame translation ratio, but did not stimulate +1 frame translation (Supplementary Figure  S3B and C). In general, as observed with the IAPV IGR IRES, the deletion of these nucleotides results in a prefer-ential translation in the +1 frame translation over 0 frame translation, thus supporting the idea that a subset of Type II IGR IRESs uses similar determinants to direct reading frame selection.
Combinatorial mutations that direct exclusive 0 or +1 frame translation. Given that +1 frame translation is ∼20-25% of 0 frame translation, we propose that the wild-type IRES adopts distinct conformations in equilibrium prior to ribosome assembly and that specific mutations shift the equilibrium to favor an IRES conformation for 0 or +1 frame translation. To determine if a mutation that favors 0 or +1 frame translation is dominant over another, we created several double mutations. We combined G6618U mutation, which exclusively shows 0 frame translation, with either A6554 or U6569, both of which lead to +1 frame translation exclusively. The resultant double mutants conferred an intermediate level of 0 and +1 frame translation compared to that of the single mutation or deletion alone ( Table  1, lines 7 and 8). Similarly, combining mutation U6562G, which increases +1 frame translation ∼4.5-fold, with mutation G6568C, which leads to a ∼95% decrease in +1 frame translation, resulted in intermediate levels of both 0 and +1 frame translation (∼80-90% translation in both frames compared to the wild-type IRES; Table 1, line 10). These results indicate that +1 frame translation directed by IRES mutants U6569 or A6554 depends on the adjacent base pairing between U6562 and G6618. Thus, one class of mutation is not dominant over another.
Finally, we asked whether mutations that increase 0 or +1 frame translation are additive. Combining mutations U6562G and U6569 resulted in a slight increase in +1 frame translation compared to U6562G alone (Table 1, line  11). However, combinations of U6562G / A6554 did not result in an additive effect on +1 frame translation compared to that of U6562G mutant IRES alone (Table 1, line 12) but did lead to complete loss of 0 frame translation. Similarly, combining mutations G6568C and G6618U, both of which decrease +1 frame translation, resulted in a minor increase in 0 frame translation (Table 1, line 9). These results suggest that these classes of mutations are at or close to maximal activity for +1 frame translation.
Nucleotide identity within the VLR affects IRES translation. SHAPE analyses showed differential reactivities within the nucleotides of the VLR upon ribosome binding ( Figure  4). We asked whether mutations within the VLR affected IRES-driven translation. Altering AA6606-7 to UU or CC resulted in a general decrease in both 0 and +1 frame translation by ∼25-40% ( Figure 5, gray box). Similarly, mutating AUAC6607-10 also decreased 0 frame translation by 15-34% without an effect on +1 frame translation ( Figure 5, gray box). The overall decrease in translation of some mutations in the VLR is in agreement with previous results that the identities of specific nucleotides are important for IRES activity (24). It remains to be investigated how the VLR plays a role in IRES activity.

IAPV IGR IRES-mediated 0 and +1 frame translation in cells
To determine if the observed uncoupling of 0 and +1 frame translation directed by mutant IRESs occurs in vivo, we transiently transfected in vitro transcribed capped bicistronic mRNAs containing either wild-type or mutant IRESs into Drosophila S2 cells and followed the translation of Renilla (cap-dependent) or Firefly (IRES-dependent) luciferase protein by measuring their enzymatic activities ( Figure 6). We used a novel bicistronic reporter construct that increases the sensitivity of detection of IRES-mediated FLuc activity (20). As shown previously, the PKI mutation, which disrupts PKI base pairing, significantly decreased FLuc activity in both 0 and +1 frames as compared to that of the wild-type IRES ( Figure 6A and B) (20). We previously showed that the mutant IRES G6618U results in a decrease in +1 frame translation in Drosophila S2 cells (Wang,Q. and Jan,E., unpublished data). Transfection of reporter RNAs containing mutant IRESs A6554, U6569 and U6562G resulted in ratios of 0 to +1 frame translation that are similar to those observed in in vitro experiments ( Figure 6). For example, U6569 generated an ∼3-fold increase in +1 frame translation and an ∼90% decrease in 0 frame translation, which are similar to the effects observed in vitro (Figures 5 and 6). Consistent with these in vitro data, A6554 maintained similar levels of +1 frame translation as the wild-type version and a decrease in 0 frame translation. U6562G showed an increase in +1 frame translation and ∼30% decrease in 0 frame translation (Supplementary In vitro transcribed bicistronic reporter RNAs containing the wild-type or the indicated mutant IRESs that drive either (A) 0 frame or (B) +1 frame translation were transfected into Drosophila S2 cells. At 6 h after transfection, cells were harvested and lysed. Renilla (cap-dependent) and firefly (IRESdependent) luciferase activities were measured and normalized to that of the wild-type IRES. The Fluc ORF was fused in the 0 or +1 frame to monitor IRES translation. Shown are averages from at least three independent experiments ± SD. Figures S1, S5 and S6). In summary, these results show that 0 and +1 frame translation mediated by the IRES can be uncoupled in vitro and in Drosophila S2 cells.

SHAPE analysis of mutant IRESs
The mutational analyses reveal that IAPV IRES-mediated 0 and +1 frame translation can be uncoupled. We hypothesize that the IRES can assume a range of conformations, a subset of which favors one or the other specific reading frames. To address this possibility, we performed SHAPE analysis on mutant IRESs including G6618U and G6568C that lead to loss of +1 frame translation and U6562G and U6569 that lead to an increase in or exclusive +1 frame translation. Supplementary Figure S4 shows the relative SHAPE reactivities of the mutant IRESs and in Figure 7, we superimpose the relative SHAPE reactivities of the mutant IRES onto that of the wild-type IRES. For comparison, SHAPE analysis of the mutant IRES PKI, which disrupts PKI base pairing, resulted in changes in SHAPE reactivities within the PKI base pair region and VLR ( Figure 7A, PKI). For the G6618U mutant IRES, SHAPE reactivities were observed in the same regions in the wild-type IRES but to varying extents. Notably, we noted an increase in SHAPE reactivities at nucleotides 6605 to 6608 of the VLR and nucleotides 6564-5 of the PKI base pair in G6618U. Conversely, SHAPE reactivity decreased at nucleotides 6610  Quantitation of mutant IAPV IRES-mediated 0 and +1 frame translation in Sf21 extracts using bicistronic reporter constructs. The percent translation is normalized to the 0 and +1 frame translation (both as 100%) mediated by the wild-type IAPV IGR IRES. The results are from at least three independent experiments ± SD. and 6612 of the VLR, at U6613 and UG6562-3 of the PKI base pair and at U6618 ( Figure 7A). In contrast, the U6562G mutant IRES resulted in increased SHAPE reactivity at G6585 in the loop of SLIII, at G6564 and AG6567-8 of the loop of PKI and at G6574. The pattern of SHAPE reactivities within the VLR of the U6562G more closely resembled that of the wild-type IRES than G6618U, which may reflect the increased +1 frame translation observed with the U6562G mutant IRES (Supplementary Figure  S4). SHAPE reactivities of the mutant IRESs U6569 and G6568C differed from those of the wild-type IRES but, in general, the differences were subtle (Figure 7 and Supplementary Figure S4). For example, in U6569, nucleotides GGCA6564-7 and U6570 of the PKI anticodon-like loop and the bulge A6554 showed increased SHAPE reactivities compared to the wild-type IRES (Figure 7). In summary, we observed changes in SHAPE reactivities between individual mutants and wild-type IRESs, suggesting that a subset of conformations, albeit subtle, is associated with 0 or +1 frame translation.

SHAPE analysis of mutant IRES-ribosome complexes
We next interrogated the structure of the mutant IRESs bound to the ribosome by SHAPE analysis focusing on mutant IRESs that uncouple 0 and +1 frame translation. NMIA-mediated reactivities of nucleotides were compared between mutant IRES-ribosome and wild-type IRESribosome complexes. Nucleotides within a mutant IRES that showed an increase or decrease in SHAPE reactivity by more than two standard deviations compared to the wildtype IRES are shown in Figure 8. In general, the overall NMIA profiles of the mutant and wild-type IRESs bound to the ribosome are not significantly altered, indicating that the overall structures are similar in the IRES-ribosome complexes. However, subtle differences in SHAPE reactivities were observed within the PKI domain of each mutant IRES. For example, compared to the wild-type IRES bound ribosome, the PKI mutant IRES displayed a significant increase in SHAPE reactivities at nucleotides 6604-8 within the VLR, at UG6562-3 and at U6569 and a decrease in reactivity at nucleotides 6609 and 6611-2 ( Figure 8). The pattern of altered reactivities appears to be associated with an IRES that cannot position the ribosome correctly for translation ( Figure 8). Interestingly, mutants G6618U and G6568C, which display exclusive IRES-mediated 0 frame translation, exhibited different NMIA-reactive profiles. G6618U showed a significant increase in reactivity at nucleotides 6605-8 within the VLR, at A6554, and at G6565 and U6569 within the PKI anticodon-like domain ( Figure 8). In contrast, the G6568C mutant showed a general decrease in NMIA sensitivity at several nucleotides within the PKI base pairing domain (Figure 8), suggesting that the PKI base pairing domain is less flexible, which may contribute to IRES-mediated 0 frame translation. Though both direct exclusive 0 frame translation, the NMIA-reactive modification patterns between mutant IRESs G6618U and G6568C are different (Figure 8), which may reflect the specific type and location of the mutation within the PKI domain. Alternatively, the differences in 0 frame translation directed by G6618U and G6568C mutant IRESs may be associated with the distinct modification profiles ( Figures 2D and 5). For the mutant IRESs U6562G and U6569 that showed an increase or exclusive +1 frame translation, the SHAPE patterns are distinct ( Figure 8). The interpretation of the SHAPE modifications of U6562G is complicated by the fact that this mutant IRES still directs significant 0 frame translation. For the U6569, which displays a 3.5-fold increase in +1 frame translation, we observed specific changes in SHAPE reactivity at nucleotides G6568, U6613, and at a few nucleotides within the VLR. Cumulatively, the mutant IRESs bound to the ribosome displayed specific NMIA-reactive profiles that differ from those of the wildtype IRES.

Ribosome positioning on IRESs by toeprinting analysis
The IAPV IRES directs ORFx translation by delivery of an Ala-tRNA to the ribosomal A site in the +1 frame (4). To determine whether 0 or +1 reading frame is selected upon IRES binding to the ribosome, we performed toeprinting analysis of ribosome/IRES complexes. Toeprinting analysis is a primer extension-based approach in which cDNA products can in principle be used to infer the nucleotides that occupy the ribosomal P and A sites and thus the starting translational reading frame of the ribosome on the IRES. We monitored the toeprints of purified salt-washed human ribosomes assembled on wild-type and mutant IAPV IGR IRESs. Assembly of ribosomes on the wild-type IGR IRES resulted in a strong toeprint at A6628 ( Figure 9A, lane 2), which represents proper positioning of the ribosome on the IRES. Based on the original interpretation of these toeprints, toeprint A6628 is 14 nucleotides downstream of the CCU triplet given that the first C of the CCU is +1 and that the CCU triplet and the adjacent GGC glycine codon occupy the ribosomal P and A sites. However, in light of the recent cryo-EM studies showing that the PKI domain of the CrPV IGR IRES bound to the ribosome initially occupies the A site (15), the A6628 toeprint may be reinterpreted as 14 nucleotides downstream of the CCU triplet occupying the ribosomal A site. As expected, the PKI mutant IRES eliminated the A6628 toeprint ( Figure 9A, lane 4), in agreement with previous reports that the integrity of PKI is required for proper ribosome positioning on IGR IRESs (9,20). We first assessed the toeprints of assembled ribosomes on mutant IRESs G6618U, which abolishes +1 frame translation, and U6562G/G6618U and U6562G, which increase +1 frame translation. In each case, a strong toeprint at A6628 is reproducibly observed ( Figure 9A, lanes 6, 8 and 10). Weaker toeprints are observed in some experiments at C6627 and A6629; however, the presence of these toeprints did not correlate with the ability of the IRES to direct 0 or +1 frame translation. Because +1 frame translation is weaker than 0 frame translation, it is possible that the toeprint for the +1 frame is masked. To circumvent this, we generated a mutant IRES where the starting 0 frame GGC glycine codon is replaced with a UAG stop codon, which should abolish both 0 and +1 frame translation ( Figure 9A, right diagram). To restore +1 frame translation with this mutant IRES, we mutated U6562 to G, which mediates significant +1 frame translation (∼3.5-fold of wild-type IRES) ( Figure 2D) (4). Both of these mutant IRESs produced a strong toeprint at A6628 similar to that observed with the wild-type IRES ( Figure 9A, lanes 12 and 14). These results indicate that the position of the ribosome on the mutant IRESs is not altered and suggest that the reading frame is not differentially selected when ribosomes assemble on the IRES.
To confirm that the ribosome/IRES complexes are functional, we added purified yeast elongation factors, eIF1A and eEF2, and bulk aminoacyl-tRNAs to the reaction and monitored translocation by toeprinting (16). Addition of the factors to reactions containing the ribosome/wildtype IRES complex in the presence of cycloheximide produced two new toeprints at A6631 and AA6634-5 (Figure 7A, lane 3). The AA6634-5 toeprint is +6-7 nucleotides downstream of A6628 indicating that the ribosome has translocated by two codons, in agreement with previous reports showing that the ribosome can translocate two cycles of elongation in the presence of cycloheximide (8,16). The toeprint at A6631 is likely representative of an intermediate translocation event (16). Upon initiating translocation with ribosome/mutant IRES complexes, +6-7 toeprints were also observed at A6631 and AA6634-5 for mutants G6618U, U6562G and U6562G/G6618U, indicative of translocation events similar to that observed with the ribosome/wild-type IRES complex ( Figure 9A, lanes 7, 9 and 11). As expected, the translocated toeprints were not observed with the mutant IRES containing the stop codon in the 0 frame (GGC6618-20UAG) ( Figure 9A, lane 13), which is consistent with the model that the presence of the stop codon disrupts the adjacent U6562/G6618 base pairing. In contrast, the mutant IRES (U6562G/GGC6618-20UAG) that restores +1 frame translation displayed a weak yet reproducible toeprint at A6635 ( Figure 9A, lane  15). The toeprint at A6635, which is +7 nucleotides downstream of A6628, represents ribosomes that have translocated through two cycles of elongation in the +1 frame. These results demonstrate that the toeprinting technique is sensitive enough to detect differences in 0 and +1 reading frames and that +1 frame translation mediated by the IAPV IRES can be reconstituted using minimal factors.
To further examine the reading frame selection by the IAPV IRES, we analyzed toeprints of ribosome assembled on mutants A6554, G6568C and U6569 that specifically uncouple 0 and +1 frame translation. Ribosomes assembled on these IRESs produced a strong toeprint at A6628 similar to that observed with the wild-type IRES ( Figure 9B, lanes 6, 8 and 10). These results further support the conclusion that the +1 reading frame is not selected upon assembly of ribosomes on the IAPV IGR IRES.

DISCUSSION
Recoding mechanisms expand the coding capacity of viral genomes and some cellular mRNAs. Most of these mechanisms involve RNA structures that interact with elongating ribosomes to affect the reading frame. For example, −1 and +1 programmed frameshifting utilizes pseudoknots, stem-loops and modified tRNAs that interact with the ribosome to shift the reading frame (2,25). Unlike mechanisms that recode during translational elongation, the IAPV IGR IRES is unique in that it directly recruits the ribosome and initiates translation in either the 0 or +1 reading frame, thus providing a simple yet elegant model for understanding how an RNA structure manipulates the ribosome and sets the reading frame. In this study, we provide extensive mutagenesis and biochemical analysis to investigate a new paradigm for recoding where the tRNA anticodon-like PKI domain of the IAPV IGR IRES adopts structural conformations that mediate reading frame selection.
Our previous findings demonstrated that the U6562/G6618 base pair adjacent to the anticodonlike PKI domain directs IAPV IRES-mediated +1 frame translation (4). These findings suggested a model where base pairing of the PKI domain measures out the reading frame similar to a 'yardstick model' to initiate in either the 0 or +1 frame depending on the frequency of base pairing of U6562/G6618. In this study, we demonstrate that in specific contexts mutating U6562 to other bases can still support +1 frame translation ( Figure 2) (4), indicating that the adjacent base pairing is not absolutely necessary for +1 frame translation. At first glance, this appears to be contradictory to our previous findings. However, we argue that mutating G6618 does not alter the native conformation of the PKI anticodon-like loop, whereas mutating U6562 to the other bases likely alters the conformation of the PKI anticodon-like loop and thus affects reading frame  10,12,14) or with purified 80S subunits, elongation factors eEF1A, eEF2 and bulk-aminoacyl-tRNAs (lanes 3,5,7,9,11,13,15). Cycloheximide is incubated with reactions to block elongation. Reactions were analyzed by primer extension analysis and separated by denaturing polyacrylamide gels. The gels were dried and exposed by autoradiography. The location of major toeprint at A6628 and the translocated toeprints are noted. Sequencing ladders for the wild-type selection. For instance, mutating U6562 to G resulted in a dramatic increase in +1 frame translation and induces subtle conformational changes of the mutant IRES in solution (Figures 2 and 7). In contrast, the wild-type IAPV IRES, which contains a U at nucleotide position 6562, adopts a native conformation that still requires base pairing at U6562/G6618 to direct +1 frame translation. Alternatively, we cannot formally rule out the possibility that a non-Watson-Crick interaction between nucleotides 6562 and 6618 contributes to +1 frame translation.
Our results revealed that novel mutations within singlestranded regions of the IGR IRES can uncouple IRESmediated 0 and +1 frame translation ( Figure 5). One class of mutants including G6618U, G6618C and G6568 to other bases results in exclusive 0 frame translation while the other class including A6554 and U6569 leads primarily to +1 frame translation (Figures 2 and 5). We predicted that each class of mutant IRES would have similar structural conformations to direct 0 or +1 frame translation. However, the differences among the SHAPE reactivity profiles of the wild-type and mutant IRESs are subtle, in solution and