The first 3′ CITE discovered was the translation enhancer domain (TED) of Satellite tobacco necrosis virus (STNV), a subviral RNA (1.2 kb) with a single open reading frame (ORF) that encodes its capsid protein. Two reports described a 120-nucleotide sequence (TED) near the 5′ end of the 3′ UTR that is capable of functionally replacing a 5′ cap (4, 56) (Figure 1). Deleting the TED reduces translation at least 20-fold and adding a 5′ cap, but not a poly(A) tail, restores translation. The TED binds to eIF4F with high affinity (11), is predicted to form a long stem loop with several bulges of six to eight bases, and has no strong uninterrupted helices (59). Because this structure has not been confirmed experimentally, the features that make it attractive to eIF4F remain unknown. The apical loop of the STNV TED contains a sequence that is complementary to the apical loop of a 5′ terminal hairpin, suggesting the formation of a long-distance, kissing-loop interaction (Figure 1). Two carmoviruses [Pelargonium line pattern virus (PLPV) and Pelargonium chlorotic ring pattern virus (PCLPV)] contain a similar TED-like structure within their 3′ UTRs and both have a sequence in their apical loops that can putatively form a kissing-loop interaction with a 5′ proximal hairpin (Figure 1). The core of the PLPV/PCLPV interacting sequence (YGCCA; Y is a pyrimidine) is conserved in other carmovirus 3′ CITEs that engage (or are predicted to engage) in similar long-distance interactions with 5′ proximal sequences (3) (Table 1). The STNV TED RNA-RNA interaction has not been confirmed, as mutating the terminal loop sequence reduced translation only modestly, and reestablishing potential base-pairing by mutating the 5′ UTR did not restore full translation (29). However, artificial constructs that do not contain STNV coding sequences were used in this study, and it is possible that additional sequences are necessary in vivo to support the formation or function of the kissing-loop interaction. Altogether, these observations suggest that TED participation in long-distance interactions is needed for optimal infectivity.

One of the best-characterized 3′ CITEs is the Barley yellow dwarf virus-like element (BTE) (14, 63) (Figure 2), which is present in all members of the Luteovirus, Dianthovirus, and Necrovirus genera and in some umbraviruses (30, 31, 52, 62, 64). The BTE is distinguished by the presence of a 17-nucleotide conserved sequence: GGAUCCUGGgAaACAGG (62), with the underlined bases pairing to form a small stem loop, designated SL-I (13, 64). The bases in lower case can vary, but the 5-nucleotide terminal loop always fits the consensus of a GNRNA pentaloop (N is any base, R is a purine) first identified in λ boxB RNA (25). Like the well-characterized GNRA tetraloop (16), the G and A at the base of the loop form a propeller-twisted base pair with the remaining bases (except for the fourth base of the pentaloop) stacked in a helical alignment with the adjoining stem, adding substantial stability to the structure (25). The fourth base of the GNRNA protrudes into the solvent, making it potentially accessible for interactions with protein (25). Whether SL-I of the BTE actually forms this particular structure has yet to be shown, but given that it fits the consensus, such a structure is likely.

The BTE, including the 17-nucleotide conserved sequence, forms a series of stem loops radiating from a central hub (64) (Figure 2). Together with the helix that connects the BTE to the rest of the viral genome, the Barley yellow dwarf virus (BYDV) BTE has four helices (13); the necrovirus BTE has three helices; and the dianthovirus BTE has six helices, giving it a two-dimensional asterisk-like structure (64). At the hub of the helices are unpaired bases (i.e., not engaged in any apparent Watson-Crick base pairing) that also play a critical role in function. Mutating almost any base in the hub greatly reduces activity of the BYDV BTE (13, 24), suggesting that there are critical non-Watson-Crick interactions that are not simple to predict around the helical junctions. Removal of SL-II from the BYDV BTE, making the two-dimensional structure more similar to a necrovirus BTE, rendered the BTE inactive (13).

Other viruses have unusual BTE variants. The BTEs of Rose spring dwarf-associated luteovirus and two umbraviruses have many more unpaired bases around the central hub than those shown in Figure 2. In addition, the BTE of Groundnut rosette umbravirus has major deviations from the 17-nucleotide conserved sequence. These deviations from consensus may explain why these three BTEs are weaker translational enhancers than those that fit the consensus more closely (64).

The BTE binds eIF4G with unusually high affinity, which is sufficient to facilitate translation in the absence of eIF4E (57). eIF4E alone has no stimulatory effect, whereas eIF4G and eIF4E together (i.e., eIF4F) stimulate translation to levels that are 20% to 30% greater than eIF4G alone. Truncated eIF4G lacking the eIF4E- and PABP-binding sites stimulates translation as efficiently as full-length eIF4G (24, 57). Footprinting experiments revealed that eIF4G protects SL-I, which is within the 17-nucleotide conserved sequence, as well as additional bases around the hub (Figure 2). Addition of eIF4E enhanced the level of protection (24), consistent with binding constants for eIF4G (177 nM) and eIF4F (37 nM) (57). The authors proposed that SL-I interaction with eIF4F resembles boxB RNA of bacteriophage λ (which contains a well-characterized GNRNA pentaloop) interacting with λ protein N and host protein NusA. NMR of boxB in the presence of a fragment of the N protein and NusA revealed specific interactions between an arginine-rich region and the 5-bp boxB helix, with a tryptophan residue base-stacking on the helically arranged bases of the GNRNA pentaloop. This binding allows the boxB-bound N protein and the protruding fourth base of the GNRNA loop to interact with the NusA protein. Kraft et al. (24) proposed that a similar interaction may recruit eIF4G to the BTE SL-I, with the fourth base of the GNRNA loop enhancing binding of eIF4E [which does not bind in the absence of eIF4G (57)]. However, additional interactions around the helical junction are also necessary, as SL-I alone or with its flanking bases is insufficient to facilitate cap-independent translation.

A stable hairpin outside the 17-nucleotide conserved sequence in all BTEs contains an apical loop sequence complementary to a sequence in the 5′ UTR (64) (Table 1). A long-distance RNA-RNA interaction between these viral sequences is required for efficient translation by the BYDV BTE but can be replaced by nonviral sequences outside the BTE (44). The sole exception is the BTE of Red clover necrotic mosaic virus (RCNMV), as mutating several loops with sequences complementary to the 5′ UTR had no significant effect on translation (49). However, the loops were not mutated simultaneously, leaving open the possibility that redundant long-distance interactions exist. Alternatively, as discussed for the STNV TED, base-pairing may be biologically relevant but not crucial for the constructs and assays used. Base-pairing between the BTE and 5′ UTR led to a model in which eIF4F binds the BTE without obscuring the loop complementary to the 5′ UTR. However, available data do not distinguish between (a) the 40S subunit binding the eIF4F-BTE complex before the BTE base-pairs with the 5′ end, followed by delivery to the 5′ terminus, and (b) delivery of the bound translation factors to the 5′ end after which 40S is recruited. This is a crucial question that requires answers for all 3′ CITEs.

Viruses in the Panicovirus genus, along with some carmoviruses and the umbravirus Pea enation mosaic virus RNA 2 (PEMV2), contain a 3′ CITE termed the PTE (Panicum mosaic virus-like translational enhancer) based on its initial discovery in Panicum mosaic virus (PMV) (1). The PTE, unlike nearly all other known uncapped RNAs, binds eIF4E with high affinity (K_d = 58 nM), even in the absence of eIF4G (66). The only other natural RNA that binds eIF4E without 7-methylguanosine modification is the eIF4E-sensitive element (4E-SE) element of histone H4 mRNA (27).

The PTE consists of a three-way branched helix with a large G-rich bulge (G domain) in the main stem (66) (Figure 3). Connecting the two helical branches at the branch point is a short C-rich or pyrimidine-rich bulge (C domain). In the G domain, one guanylate is hypersensitive to SHAPE reagents that modify the 2′-hydroxyl of non-Watson-Crick base-paired nucleotides. In the presence of mutations in the C domain or in the absence of magnesium ions, this hypersensitivity is lost and the G residues in the G domain are equally modified as expected for a uniformly single-strand RNA (65). These findings led Wang et al. (65) to propose the existence of a pseudoknot between the G domain and the C domain. Many different mutations designed to reestablish disrupted base-pairing between these regions were ineffective at restoring cap-independent translation (66), suggesting that the primary sequence is important. Footprinting experiments indicated that the region around this pseudoknot is protected by eIF4E (65). This finding led to a model positing that pseudoknot interaction forces the SHAPE-hypermodified G residue to protrude from the PTE into the cap-binding pocket of eIF4E (Figure 3b). eIF4E binding is proposed to be strengthened by interactions of the rim of the cap-binding pocket with portions of the PTE around the pseudoknot. This would compensate for the absence of the 7-methyl group on the RNA. However, this model has not been tested.

The 5′ side hairpin of all PTEs, except for the PEMV2 PTE, has an apical loop that is complementary to the apical loop of a hairpin at or near the 5′ end of the viral RNA (3) (Figure 3a). In panicoviruses, the sequence complementary to the PTE is in the apical loop of a 5′ terminal hairpin, whereas in the carmovirus Saguaro cactus virus (SCV), the 5′ complementary sequence is in the loop of a coding region hairpin (3) (Table 1). The carmovirus PTE sequence that engages in the RNA-RNA interaction has the same conserved motif (YGCCA/UGGCR) found in carmovirus TED-like elements and I-shaped structure (ISS) elements (see below), with either sequence present at either end (Table 1). The 5′ interacting sequence in general is located in the apical loop of hairpins located either at the 5′ terminus or in the coding region within 150 nucleotides of the 5′ end of the mRNA. The conservation of interacting sequences suggests a function for the paired sequences beyond simple kissing-loop formation, but this function remains unknown.

3′ CITEs with a common I-shaped structure (ISS) have been found in carmoviruses and in Maize necrotic streak virus (MNeSV), a unique member of the Tombusviridae (50). The MNeSV ISS contains a key central domain with a four-base helix and flanking bulged sequences (Figure 4a). Mutagenesis and structure probing found significant roles for non-Watson-Crick interactions, and possibly alternative structures, as few mutations in the core region are tolerated, even when conserving the proposed secondary structure (34, 35). Translational enhancement by the MNeSV ISS in factor-depleted wheat germ extracts requires intact eIF4F, unlike the BTE for which eIF4G was sufficient. The MNeSV ISS also binds only to intact eIF4F (K_d = 190 nM), unlike the BTE (binds eIF4G) or the PTE (binds eIF4E). As with other 3′ CITEs, base-pairing between the terminal loop and its complementary sequence in a 5′ UTR hairpin is necessary for efficient translation. The long-distance interaction and eIF4F binding can occur simultaneously, supporting the hypothesis that the long-distance pairing delivers eIF4F to the 5′ end. Toeprinting assays were consistent with ribosome scanning from the 5′ end to the start codon, which required the 3′ CITE (34).

Direct involvement of the ISS of the carmovirus Melon necrotic spot virus (MNSV) and eIF4E was inferred genetically, because a single point mutation in eIF4E provides recessive resistance to MNSV in cucumber (nsv gene), including strain MNSV-Mα5 (37). A resistance-breaking strain, MNSV-264, contains mutations in the central portion of the bulged stem loop in the ISS that are sufficient to break resistance (58). In wheat germ extracts, uncapped reporter constructs containing the cucumber nsv resistance-breaking MNSV-264 ISS in the 3′ UTR translated five times more efficiently than those containing the noninfectious MNSV-Mα5 ISS. However, addition of eIF4E from cucumber susceptible to both viruses allowed the MNSV-Mα5 ISS to translate nearly as efficiently as the construct containing the MNSV-264 ISS (58).

Moreover, the MNSV-Mα5 reporter construct translated efficiently only in virus-susceptible melon protoplasts, whereas the MNSV-264 construct translated most efficiently in resistant protoplasts (58). Based on Mfold predictions, Nicholson et al. (34) proposed that the MNSV-Mα5 ISS forms an inverted version of the ISS, with the 3′ and 5′ ends reversed, relative to the MNSV-264 ISS, MNeSV ISS, and other ISS. However, the inverted functional orientation does not conform well with the alternative ISS that these authors predicted on the basis of more recent in vivo evolution studies (34, 35) (Figure 4a). The ISS from either strain interacts with the MNSV 5′ UTR in a strain-independent fashion. The apical loop of the MNSV ISS contains the conserved carmovirus motif (Table 1), and its 5′ partner hairpin is in the 5′ coding sequence.

The 3′ CITE of viruses within the genus Tombusvirus has a conserved Y-shaped structure (YSS) with three major helices that are all substantially longer than the three helices of the PTE (8, 9, 67) (Figure 4b). The 5′ side hairpin engages in a required long-distance kissing-loop interaction with a 5′ UTR hairpin loop. The base-pairing sequence (5′-GGUCG) shares only limited sequence similarity with the carmovirus consensus sequence. Mutations that maintain the structure of the three stems in the YSS of Tomato bushy stunt virus (TBSV) do not have a substantial impact on translation, unlike alterations in junction residues between helices and in a large asymmetric bulge in the major supporting stem (9). Addition of translation factors to factor-depleted wheat germ extracts revealed that the YSS of Carnation Italian ringspot virus (CIRV) requires eIF4F or eIFiso4F for efficient translation (35). As with translation associated with all of these 3′ CITEs, the YSS facilitates ribosome scanning from the 5′ end.

3′ CITEs that assume a three-dimensional T-shaped structure (TSS) were first discovered in the carmovirus Turnip crinkle virus (TCV) (28). Although all carmoviruses share similar 3′ end structural organization [stable 3′ terminal hairpin and an upstream hairpin with a large internal symmetrical loop (H5) that is linked to the 3′ terminus through a pseudoknot] (73), carmoviruses have a surprising variety of CITEs immediately upstream of the conserved 3′ terminal structures (3, 38, 54). TCV and the related carmovirus Cardamine chlorotic fleck virus (CCFV) contain a unique set of three hairpins (including H5) and two pseudoknots that fold into a TSS that is topologically similar to a tRNA, as predicted by molecular modeling and confirmed by small-angle X-ray scattering (SAXS) and NMR (28, 74) (Figure 5a). Unlike the Tymo, Rubi, and Tobamo lineages of Supergroup 3 viruses that have 5′ caps and 3′ terminal, aminoacylated tRNA-like structures (6, 7), TCV and CCFV lack 5′ caps and the 3′ TSS are present in an internal location, precluding aminoacylation. The TCV TSS binds to yeast and Arabidopsis 80S ribosomes and 60S subunits, and binding is important for CITE activity (54). The TCV TSS competes for 80S ribosome binding with a P-site-binding tRNA, and recent cryo-electron microscopy of 80S ribosomes bound to the TSS places the TSS in the P site, interacting with peptidyltransferase center helices H89/H90, helices H69 and H95 of 25S rRNA, and proteins L10 and L11 of the 60S subunit (J. Pallesen & J. Frank, unpublished data). The P-site proximal placement was confirmed by high-throughput SHAPE (hSHAPE) structure mapping of in situ ribosomal RNAs, which reports on residues with altered flexibility in the presence of the TCV TSS (J. Leschine, J. Dinman & A.E. Simon, unpublished data).

For the TSS to function as a 3′ CITE, additional elements are required, including an adjacent A-rich repeat, the terminal and internal loops of upstream hairpin H4, and portions of a largely unstructured region upstream of H4 (54). The A-rich repeat is currently thought to interact with the adjacent downstream pseudoknot Ψ₃, as disruption of either the A-rich region or Ψ₃ enhances flexibility in the residues of the other’s structure, as assayed by in-line RNA structure probing (70).

All these TCV 3′ UTR elements, together with additional coding region sequences, form an integrative network of noncanonical interactions that are important for virus accumulation (71, 72). In the presence of the RNA-dependent RNA polymerase (RdRp), this higher-order structure is substantially disrupted, including extensive alteration of the region of the TSS that mimics the tRNA acceptor arm (H4a/Ψ₃) (70). Because the tRNA acceptor arm and its TSS mimic are ribosome-binding sites (51, 54), it was proposed that the RdRp-mediated conformational switch arrests 3′ ribosome binding and CITE function, thus allowing transcription to initiate at the 3′ end without ribosome-bound impediments (70).

Because TCV lacks a detectable RNA-RNA interaction that links sequences proximal to the 5′ and 3′ ends to circularize the genome (55), a key question is whether and how 60S subunits bound to the 3′ TSS are recruited to the 5′ end. The discovery of a 40S-binding site in the 5′ UTR of TCV led to the hypothesis that circularization of the template may result from the 40S bound to the 5′ end combining with 60S bound to the 3′ proximal TSS to form 80S ribosomes (55). In support of this hypothesis, structural probing of the TCV 5′ UTR in the presence of 80S ribosomes revealed substantive differences in nucleotide flexibility when the 3′ UTR of TCV was included in the reaction. These differences were interpreted to result from 80S ribosomes binding simultaneously to both 5′ and 3′ TCV sequences, which altered ribosome interaction with the 5′ end (55). While it is currently unknown whether additional translation factors interact with the TSS or the TSS-ribosome complex, translation of TCV constructs containing the genomic or subgenomic RNA 5′ UTR and 3′ UTR is reduced in Arabidopsis cells deficient in eIF4G (68, 69).

A second TSS with 3′ CITE activity was recently discovered in the central region of the 3′ UTR of PEMV2, 9 nucleotides upstream of the PTE 3′ CITE (Figure 5b,c). The kissing-loop TSS (kl-TSS) comprises a three-way branched structure with a 5′ side hairpin (3H1) that engages in a long-distance kissing-loop interaction with a coding sequence hairpin (Figure 3a) that uses the conserved carmovirus interacting motif (10) (Table 1). The kl-TSS differs from the TCV TSS in that it binds to 40S and 60S ribosomal subunits and to 80S ribosomes (10). The kl-TSS does not compete with the TCV TSS for 80S binding, suggesting a binding site outside the P-site (F. Gao & A.E. Simon, unpublished data). Ribosome binding is compatible with the long-distance interaction (F. Gao & A.E. Simon, unpublished data), suggesting that the binding site within the ribosome must allow for the presence of upstream and downstream viral sequences as well the interacting 5′ region hairpin. The orientation of the kl-TSS with respect to a tRNA is currently unknown, and two possible orientations are shown in Figure 5b,c. The downstream proximity of the PTE 3′ CITE suggests that the kl-TSS provides the missing bridge to the 5′ end in this unusual PTE. Alternatively, interaction of the nearby PTE bound to eIF4E may enhance ribosome binding to the kl-TSS. It is currently unknown why 3′ CITEs capable of sequestering ribosomes or ribosomal subunits exist in the 3′ UTR, but the most likely explanation is that TSS-type 3′ CITEs enhance the recycling of posttermination ribosomes, thereby increasing the rate of ribosome reinitiation at the 5′ end. The three-way branched organization of the kl-TSS is a common RNA structure, which could suggest that ribosome recognition of elements other than a 5′ cap (via initiation factors) or IRESs in coding RNAs may be more widespread than currently thought.