|Home | About | Journals | Submit | Contact Us | Français|
The RNA genomes of plus-strand RNA viruses have the ability to form secondary and higher-order structures that contribute to their stability and to their participation in inter- and intramolecular interactions. Those structures that are functionally important are called cis-acting RNA elements because their functions cannot be complemented in trans. They can be involved not only in RNA/RNA interactions but also in binding of viral and cellular proteins during the complex processes of translation, RNA replication and encapsidation. Most viral cis-acting RNA elements are located in the highly structured 5′- and 3′-nontranslated regions of the genomes but sometimes they also extend into the adjacent coding sequences. In addition, some cis-acting RNA elements are embedded within the coding sequences far away from the genomic ends. Although the functional importance of many of these structures have been confirmed by genetic and biochemical analyses their precise roles are not yet fully understood. In this review we have summarized what is known about cis-acting RNA elements in nine families of human and animal plus-strand RNA viruses with an emphasis on the most thoroughly characterized virus families, the Picornaviridae and Flaviviridae.
Most life forms in nature contain dsDNA as their genetic material but viruses frequently possess RNA genomes. Viruses that replicate their genomes entirely through a RNA intermediates are classified into plus-strand, minus-strand and double-stranded RNA viruses. Of these three classes, the most abundant are plus-strand RNA viruses, which use their genomic RNA sequence as mRNA to directly encode proteins. Although these viruses vary widely in their virion structures (icosahedral, spherical), mechanisms of RNA translation (cap-dependent, cap-independent), RNA replication (primed or de novo initiated), and encapsidation, they are similar in the basic steps that make up their life cycles. Following binding and entry of the viral RNA into a permissive cell, the incoming mRNA is translated. The products are either the mature viral proteins or a polyprotein, which is subsequently processed into smaller precursor and mature polypeptides. These products include both structural and nonstructural components. The structural proteins are the building blocks of the progeny virion particles. The nonstructural polypeptides, including the RNA-dependent RNA polymerase, are primarily involved in RNA replication, which takes place in replication complexes associated with altered intracellular membranes. First the parental plus strand RNA is converted into a minus strand copy, which then serves as the template for the synthesis of progeny plus strands. Finally, the plus strand genomes are encapsidated and the newly assembled virion particles are usually released from the host cells. All of these steps in the viral life cycle are strictly regulated and the RNA structures present in the genomic RNA have important roles in these processes.
The genomes of plus strand RNA viruses are able to form secondary and higher-order RNA structures that contribute to their stability and to their participation in inter- and intramolecular interactions. The presence of RNA structures in viral genomes can be predicted by computational procedures that combine structure prediction and nucleotide sequence comparison. Subsequently, the biological importance of these predicted structures can be tested by genetic analyses.
Those structures that are functionally important are called cis-acting RNA elements because their functions cannot be complemented in trans. They can be involved not only in RNA/RNA interactions but also in binding viral and cellular proteins required for the complex processes of the viral life cycle.
Most viral cis-acting RNA elements are located in the 5′ and 3′-nontranslated regions (NTRs) of the genomes but now evidence is accumulating for the widespread existence of functional structures embedded within protein-coding sequences. Some of these internal RNA elements comprise individual functional units far away from the genomic ends. Others are simply extensions into the down-stream or up-stream coding sequences of functional RNA elements that are already present in the NTRs. In this review we will attempt to summarize what is known about cis-acting RNA elements in nine families of human and animal plus-strand RNA viruses with an emphasis on the most thoroughly characterized virus families, the Picornaviridae and Flaviviridae. Unfortunately, due to the limited scope of this review article, it is not possible to include a discussion of all of the RNA structures present in all of the viruses that make up these virus families. However, we hope that the representative members of each virus family that we have selected for our review will provide the reader with an overview of the nature and functions of the cis-acting RNA elements in the life cycle of these viruses.
Members of the Picornaviridae virus family include a large number of medically important human and animal pathogens, which cause a wide variety of illnesses. The plus-strand genomes of picornaviruses vary in length from 7209 to 8450 nt and encode a single polyprotein that consists of one structural (P1) and two nonstructural domains (P2 and P3) (Figs. 1–4). The genomic RNA is covalently linked at the 5′-end to the hydroxyl group of tyrosine in the terminal peptide VPg [1–3] and the 3′-end is polyadenylated [4, 5]. There are 12 genera in the Picornaviridae family: Enterovirus, Cardiovirus, Aphtovirus, Hepatovirus, Parechovirus, Kobuvirus, Erbovirus, Teschovirus, Sapelovirus, Senecavirus, Tremovirus, and Avihepatovirus [6, ICTV 2009]. It should be noted that recently the genus Rhinovirus was incorporated into the Enterovirus genus (www.picornavirus.com). For this review we have divided picornaviruses into 3 major groups based on similarity in their 5′ NTRs. In addition, we have briefly discussed the RNA structures present in the genomes of three of the less well-known genera, Kobuvirus, Parechovirus and Teschovirus.
Based on RNA folding analyses and sequence comparisons Witwer et al. 2001  have predicted the presence of numerous conserved RNA structures in the 5′NTRs and 3′NTRs of enterovirus genomes, including those of human rhinoviruses (HRVs) (Figs. 1A–1C). In addition, internal conserved structures were identified in their 3Dpol domains, in the 2CATPase coding sequence of poliovirus (PV) and VP1 or 2A sequences of HRVs.
The 5′ noncoding region or oriL (origin of replication at the left) of enteroviruses is highly structured and contains two functional domains, which are involved in translation and RNA replication. The first domain in enteroviruses forms a cloverleaf (CL) structure  that carries signals to control both translation and RNA replication (Fig. 1A, 1B) . The PV CL (88 nt) contains four stem loops (SL-A to SL-D), which is followed by a spacer region (nt 89-123) located between the cloverleaf and the IRES (Fig. 1A). The second domain of the enteroviral 5′NTR is the internal ribosomal entry site (IRES) that promotes translation.
Numerous studies have shown that the formation of a ribonucleoprotein (RNP) complex, consisting of cloverleaf RNA, cellular protein PCBP and viral protein 3CDpro, is required for viral RNA replication and virus proliferation [10–14]. Protein PCBP2, with its KH1 domain, was shown to bind to stem loop B of the cloverleaf [13, 15, 16]. The primary PCBP2 binding sites on the CL are located on 3C residues in the loop of SL-B (nt 23-25) and on two Cs (C94 and C95) in the spacer region [15, 17]. The other partner in the formation of the CL RNP complex is proteinase and RNA binding protein 3CDpro for PV [12, 14, 18] or 3Cpro for rhinoviruses . PV proteins 3Cpro/3CDpro interact with a tetra loop (UGCG) in SL-D of PV (Fig. 1B) but do not bind to the HRV14 CL, which contains only 3 nt (UAU) . Interestingly, the formation of another RNP complex that included CL RNA, and viral proteins 3AB and 3CDpro, was also shown to be important for PV RNA replication [12, 14]. Herold and Andino  have proposed a model in which the PV genome circularizes prior to minus strand RNA synthesis by way of an interaction between cellular poly(A) binding protein with the poly(A) tail of PV RNA on the one hand and PCBP2/3CDpro/CL on the other. This model was based on an observation that PABP physically interacts with PCBP2, 3CDpro and the poly(A) tail both in vivo and in vitro.
The sequence complementary to the plus-strand CL also forms a similar structure. Using an in vitro translation/RNA replication system Sharma et al.  have shown that the duplex structure of SL-A in the PV CL (Fig. 1B) was required for minus-strand but not for plus-strand RNA synthesis. On the other hand altering the primary sequence at the 5′-terminal end of SL-A resulted in a striking reduction of plus-strand RNA synthesis. These results suggested that the 3′-terminal end of minus strands is required for the initiation of plus-strand RNA synthesis. The specific binding of cellular proteins (p36 and p38) and of viral protein 2CATPase to the 3′-end of minus strand PV RNA has also been reported [23–25].
The highly structured heteropolymeric regions of picornaviral 3′NTRs are very diverse and their functions are unknown although genetic evidence supports a role in RNA replication [26, 27]. The entero- and rhinoviral 3′NTRs can be grouped into 3 types: single stem loop (Y) in rhinoviruses; two stem loops (X and Y) (65 nt) in PVs; 3 stem loops (X, Y and Z) (100 nt) in coxsackieviruses . In PV the poly(A) tail is a part of the overall 3′-terminal structure with poly(A) hybridized to sequences in stem X, just upstream from the termination codon (Fig. 1B). The predicted structures were confirmed by enzymatic and chemical probing [28, 29]. It was proposed that the enteroviral 3′NTRs together with the poly(A) tail form the origins of replication oriRs for minus strand RNA synthesis, which serve as specific binding sites for viral and/or cellular proteins.
Mutational analyses of the PV and coxsackie B3 virus (CVB3) 3′NTRs have provided convincing evidence for a “kissing” interaction between 5 nt in the loops of stem loops X and Y, which is functionally important (Fig. 1B) [29, 30]. Three-dimensional structures, derived by molecular modeling, displayed the PV and CVB3 oriRs as quasi-globular multi-domain structures [29, 30].
Unexpectedly, it was observed that the PV 3′NTR (2 stem loops) could be replaced by the 3′NTRs of HRV14 (1 stem loop) or of coxsackievirus B4 (CVB4) (3 stem loops) . Even more surprising was the finding that the PV or HRV14 3′ NTRs could be deleted without loss of viability in HeLa cells [31, 32]. Subsequently it was shown that the removal of the PV 3′NTR restricted plus-strand RNA synthesis, particularly in cells of neuronal origin .
The poly(A) tail of PV  and of all other enteroviruses (and of picornaviruses whose genomes have been analyzed) is about 90 nt long and it is genetically encoded . Interestingly, the complementary poly(U) is only 20 nt . These results are in agreement with previous studies showing that a poly(A) tail of 20 nt is sufficient for the replication of PV  and for the binding of poly(A) binding protein to the 3′NTR-poly(A) in vitro . The poly(A) tail of CVB3 is markedly shortened in the absence of the OriR, indicating a role for this cis-replicating element in poly(A) synthesis .
Among the best-characterized functional RNA structures are the picornavirus cres (cis-replicating RNA elements), which are small hairpins of varying nucleotide sequence located in the viral ORFs [36, 37], and in one case so far reported, in the 5′NTR of FMDV . This type of cre was first discovered by McKnight and Lemon  in the VP1 coding sequence of HRV14 and was shown to be required for RNA replication . Since then similar cres have been identified in the 2CATPase coding domains of PV (Fig. 1B)  and CVB3 . In rhinoviruses the corresponding cre elements are found in the 2A ORF of species A rhinoviruses , the VP1 ORF of species B rhinoviruses (Fig. 1C) , the VP2 ORF of species C rhinoviruses . The precise function of the oriI is to template the linkage of two UMPs to VPg, to yield VPgpU and VPgpUpU, the primers for both plus and minus strand RNA synthesis [45–47]. According to current models of PV RNA replication the VPgpUpU made on the oriI is used for the initiation of both plus and minus strand RNA synthesis [34, 37].
The ability of the enteroviral cre elements to support RNA replication is dependent both upon the specific RNA structure, in particular the upper of the part of the stem, and specific sequences within the loop [46, 48]. The cres of enteroviruses share a conserved nucleotide sequence motif R1XXXA5A6R7XXXXXXR14 (R=purine) in the loop of the hairpin (Fig. 1B) [47, 48]. Within this motif, the A5 residue templates the linkage of both UMPs to VPg by a “slide back mechanism” in a reaction catalyzed by RNA polymerase 3Dpol and stimulated by viral proteinase 3CDpro [37, 45–47, 49]. Based on biochemical studies Pathak et al.  proposed a model for the assembly of the PV VPg uridylylation RNP complex. According to this model the first step is the binding of a 3CDpro dimer to the upper part of the oriI stem, which is unwound. In the next step 3Dpol associates with the complex via an interaction between the back of the thumb subdomain of 3Dpol and a convex surface formed by the top of the subunits of the 3Cpro dimer .
The function of the enteroviral cre is independent of its position within the genome [40, 48]. When the endogenous cre element of PV is inactivated by mutation, cre function can be rescued by the insertion of a second cre (PV or HRV14) into the 5′NTR [40, 48]. Interestingly, in vivo studies indicated that the PV oriI inhibited PV replication in a trans-dominant manner .
Recently a second type of internal RNA element was discovered in the coding sequence of protein 3Cpro of PV and of several coxsackie A viruses (CAVs) (Fig. 1) . This RNA element potently inhibits the activity of a cellular protein, RNase L, a latent endoribonucloease in an interferon-regulated, dsRNA-activated pathway. The RNA structure consists of 4 stem loops and of these stem loops 1 and 4 is important for function. These are involved in a putative “kissing loop” interaction . Interestingly, RNase L activity was activated late during the course of PV replication in HeLa cells as virus assembly neared completion. This activity did not diminish virus production; rather, it was associated with larger plaques and increased cell-to -cell spread. Han et al.  suggested the possibility that RNase L, which is activated late in infection and is proapoptotic, facilitates cythopathic effect and virus release at the end of the PV growth cycle. We have included this RNA structure in our discussion although currently its exact function in the PV life cycle is not known. It is possible, however, that a biological function for it will be discovered in the future but only under certain conditions, such as in different cell lines.
The cardio- and aphtoviral 5′NTRs differ from those of enteroviruses in two respects. First, their 5′ terminal RNA elements have a less-defined structure than the corresponding enteroviral CL . Second, the 5′NTRs contain a long poly(C) tract of about 130 nt in encephalomyocarditis virus (EMCV) (Fig. 2A) and 150–250 nt in foot-and-mouth disease virus (FMDV) (Fig. 2B). EMCV genomic RNA contains a 5′-terminal hairpin (84 nt), linked to two pseudoknots (Pk, Fig. 2A) , which is connected to the poly(C) tract, followed by the IRES. In FMDV the 5′NTR (>1300 nt) begins with a very large (360-bp) stem loop structure (S fragment) , then the poly(C) tract, 3–4 tandemly repeated pseudoknots of unknown function , the cre element  and finally the IRES (Fig. 2B). The S fragment is predicted to function in RNA replication although direct evidence for this is lacking. Interestingly, Serrano et al.  has identified an interaction in vitro between the 3′NTR-poly(A) tail and the S fragment (Fig. 2B).
The 3′NTR of FMDV is 90 nt long and consists of two stem loops, linked to a genetically encoded poly(A) tail (Fig. 2B). In contrast to PV or HRV14 the 3′NTR of FMDV could not be deleted without loss of virus replication . Recent studies indicated that each of the stem loops of the FMDV 3′NTR interacted in vitro with the S fragment of the 5′ NTR and this interaction was dependent on the presence of the poly(A) tail (Fig. 2B) . Similarly, the 3′NTR was also found to establish a long range RNA/RNA interaction with the IRES.
Duque and Palmenberg  have identified three phylogenetically conserved stem loops in mengovirus (cardiovirus) 3′NTR (126 nt). Deletion of stem loop I was found to be dispensable for virus growth while the deletion of stem loop III was lethal. The deletion of stem loop II led to an intermediate growth phenotype. In other studies the 3′NTR of EMCV was predicted to form a single stem loop . Mutational and deletion analyses studies indicated that a U-rich stretch in the loop interacts with the poly(A) tail (Fig. 2A).
In cardioviruses, Theiler’s virus and mengovirus, oriIs were identified in the coding sequence of capsid protein VP2  and these were found to be essential for RNA replication. These stem loop structures had a conserved sequence of 9 nt in mostly unpaired regions of the structure that also contained the AAACA sequence, characteristic of the enterovirus cres. The cres of TMEV and mengovirus could be functionally exchanged but they could not be replaced with the cre of HRV14.
Of all the picornaviruses analyzed so far, only FMDV contains a cre element outside its protein coding sequence, just upstream of the IRES (Fig. 2B) . Mutations in the conserved AAACA sequence severely reduced RNA replication and yielded quasi-infectious viruses. The structure of the stem was also important for replication. Interestingly, and unlike other picornaviral cres, the FMDV cre could be complemented in trans in infected cells .
Early computational studies by Witwer et al.  have predicted the presence of a large number of conserved RNA structures in the genome of FMDV, both in the 5′ and 3′NTRs and in the coding sequences. Recently, large-scale thermodynamic prediction methods have detected a genome-scale ordered RNA structure in FMDV [64, 65]. Interestingly, in FMDV, like in other mammalian plus strand RNA viruses, these GORS are associated with a persistent growth phenotype. Although the reason for the association between GORS and persistence is not yet understood, Simmonds et al.  suggested the possibility that the formation of an extensive RNA a secondary structure might play a role in the evasion of cell defenses.
Hepatitis A virus, a member of the hepatovirus genus, differs from other picornaviruses in the details of the organization and function of its polyprotein. As shown below, however, it also shares many features with other picornaviruses in the overall genome organization and replication strategy (Fig. 3).
Structural predictions and enzymatic probing have shown that the 5′NTR of hepatitis virus A has several motifs common with aphto- and cardioviruses . The 5′NTR contains 6 major structural domains. Domains I and II (1–95 nt) contain a 5′-terminal hairpin and two small stem loops (Fig. 3). This is followed by a pyrimidine-rich tract (96–154 nt). The remainder of the 5′NTR, domains III to VI (nt 155-734), contains several stem loops and includes the IRES (starting nt 355). The interaction of cellular protein glyceraldehyde-3-dehydrogenase (GAPDH) with the 5′NTR was proposed to modulate protein translation . The binding of viral protein precursor 3ABC in vitro to the 5′NTR was also observed .
The 3′NTR of hepatitis A virus is 63 nt long and terminates with a poly(A) tract of 40–80 nt. There are two alternate models for the structure of the 3′NTR. The first consists of two stem loops (X and Y) and a putative “kissing” interaction between the stem loops has been predicted . The second model suggests that that the 3′NTR exists in a pseudoknot-like structure and this model was favored by enzymatic probing experiments (Fig. 3). The specific binding of GADPH to the 3′NTR and to a part of the upstream 3Dpol coding region was also reported .
The cre element of hepatitis A virus, recently identified, differs from those of other picornaviruses in location, size and stability . It is located near the 5′ end of the 3Dpol coding sequence and both the stem (35 bp), which is rich in A-U pairs, and the loop (18 nt) are large. Like other picornaviral cres the HAV cre also contains an AAACA/G motif and mutations that disrupted the structure of this element ablated the replication of a subgenomic replicon.
Compared to other picornaviruses, relatively little is known about the Kobuvirus (Aichi virus and Porcine kobuvirus), Parechovirus (Human parechovirus [HpeV1], Ljungan virus) and Teschovirus (Porcine teschovirus) genera of Picornaviridae, which have been recently recognized as important human pathogens. The VPg-linked genomes of Aichi virus and of Porcine kobuvirus are 8280 and 8210 nt long, respectively . They encode a single polyprotein that contains a leader protein followed by the capsid proteins and the nonstructural proteins (Fig. 4A). The 3′ end of the genomes is terminated with a poly(A) tail.
The 5′NTRs of Aichi virus (744 nt) and of HPeV resemble that of cardio- and aphtoviruses (Figs. 4A, 4B). In Aichi virus the first 115 nt of the 5′NTR are predicted to form 3 stem loops (SL-A to SL-C) and a pseudoknot . Disruption of these RNA elements impaired minus strand RNA synthesis while the stem of SL-A was found to be important for plus strand RNA synthesis. Nagashima et al.  has observed a specific binding of viral protein 3ABC to the 5′NTR of Aichi virus and this binding was required for minus strand RNA synthesis. Recent studies by Sasaki and Taniguchi  indicated that a 7 nt long segment (UCCCACU and its complementary sequence) in the stem of SL-A of Aichi virus has an important role in encapsidation. The 5′NTR of Porcine kobuvirus (S-1-HUN) is 576 nt long and is predicted to form 7 stem loops . The first three of these stem loops are similar to those of Aichi virus. A polypyrimidine tract is located between nt 459 and 465, within a predicted IRES element (nt 284-576), which has a strong sequence similarity to that of Porcine teschovirus and belongs to the type IV hepaci/pestivirus IRES group (Fig. 4A, ,5E).5E). The 3′NTR of Porcine kobuvirus is 167 nucleotides long and is predicted to form 3 stem loops (Fig. 4A) . Using bioinformatics analysis methods Simmonds et al.  have predicted the presence of genome-scale ordered RNA structures in the genomes of kobuviruses.
In HpeV1 the 5′NTR also contains the same functional stem-loops (SL-A to SL-C) as in Aichi virus (Fig. 4B) . The 3′NTR of HpeV1 is made up of two highly conserved tandem repeats, which is predicted to form a single stem loop structure with extensive base-pairing between the poly(A) tail a U -rich region of the 3′NTR [76, 77]. The biological function of this 3′-terminal structure is not yet known. The oriI of HpeV1 has been identified in the coding sequence of VP0 as a hairpin structure of about 50 nt containing a large loop (23 nt). The loop contains the CAAAC motif, typical of picornaviruses . In this same study the cre element of Ljungan virus was predicted to exist in the VPg coding region.
Computer analysis has predicted 5 highly conserved RNA structures in the 5′NTR and six such structures in the noncoding regions of Teschoviruses . The 5′ NTR of Porcine teschovirus is 412 nt long and contains an HCV-like IRES element (nt 122-411) (Fig. 5D) .
The discovery of IRESes in picornaviruses has fundamentally changed the perception of gene expression by translation in eukaryotic cells. Since until 1988 a single mechanism of cap-dependent initiation of translation was accepted almost as dogma, the experiments illustrating IRES-mediated translation (translation of artificially engineered dicistronic mRNAs) [79–82] was received with skepticism. Two experiments can be cited that lent unambiguous support for IRES function. The first provided proof in vivo through the construction and propagation of a dicistronic poliovirus whose translation of the segmented polyprotein was controlled by two different IRES elements and whose infection of cells followed single hit kinetics . The second experiment described the construction of a circular mRNA containing a picornavirus IRES element. This circular mRNA was efficiently translated in vitro . In the following section, the structural and functional hallmarks of picornavirus IRESes will be summarized from several recent general reviews [85–87].
Picornavirus IRESes that all reside in the 5′NTR, are 290 to 450 nt long. They are preceded by RNA structures of varying length, the nature of the structures depending upon the picornavirus genus (Fig. 5). In the case of the cardio- and aphthoviruses, IRESes are preceded by long poly(C) tracts. In the case of the aphthoviruses the critical cre element maps between the poly(C) and the IRES (Figs. 2A, B). In hepatitis A virus, the polypyrimidine tract consists of C and U residues (Fig. 3).
Based on the primary sequence as determinants of structures, the IRESes of enteroviruses, cardio- and aphthoviruses, and hepatitis A virus have been classified as type I, type II, and type III IRESes, respectively  (Fig. 5A–C). The nucleotide sequence homology between the three IRES types is barely 50%. In spite of the general sequence differences, all three IRES types fold roughly into domains dominated by a large domain in the center (Fig. 5). In addition, representatives of type I and type II IRESes carry tetra loops (GNRA or GNAA; N, any nucleotide; R, purine) in the loops of their domains (Fig. 5A, B). The integrity of the tetra loops is required for IRES function.
Sequence analyses of Porcine teschoviruses, which are members of the new picornavirus genus Teschovirus, have revealed that their highly conserved 5′NTRs exhibited no apparent similarity with picornavirus 5′NTRs known at the time . Closer analyses of the 5′NTR RNA of Porcine teschovirus-1 (PTV-1) led to the unexpected discovery that in structure and function the IRES of this virus is more related to the IRES of hepatitis C virus (HCV) than to the type I–III IRESes discussed above (Fig. 5E) . As can be seen, the PTV-1 IRES (Fig. 5E) consists only of two major stem-loop domains (domains II and III), forming a pseudoknot at the base of domain III, and it is only 290 nt long (between the base of domain II and the initiating AUG at nt 412) . The small size, the architecture and genetic properties resemble the IRES of HCV. Indeed, it now appears that HCV-related IRES elements are found in genomes of many genera of picornaviruses . The HCV-like IRES elements of picornaviruses are designated as type IV IRESes. For a more detailed discussion of the HCV IRES, see the section on Flaviviridae. A recent report by Reuter et al.  shows the putative IRES element (nt 284-576) of Porcine kobuvirus (Fig. 5D), which has a 74% sequence similarity to the IRES of Porcine teschovirus (Fig. 5E). The predicted structure of this IRES element (nt 122-411) is also that of type IV IRESes, similar to those of hepaci- and pestiviruses.
Type I–III IRES elements contain a highly conserved motif consisting of an oligo pyrimidine tract (Yn), followed by a tract of an unspecified sequence of 15–20 nucleotides (Xm), followed by an AUG . The AUG codon in type II IRES elements is the initiating codon (in aphthoviruses, a downstream codon may also be used) whereas in type I IRESes, the motif is upstream of the initiating codon and its AUG codon is, therefore, cryptic. Genetic analyses have shown that the integrity of the motif is essential for IRES function. Eliminating Yn or shortening or enlarging Xm, etc. have resulted in severe deficiencies in translation of PV RNA; the same is true for the EMCV type II IRES [93–96]. It has been speculated that the YnXmAUG motif may constitute the “landing pad” for the small ribosomal subunit but direct evidence for this hypothesis is still lacking. In contrast to enteroviruses and EMCV, Pilipenko et al  reported that the YnXmAUG motif in Theiler’s virus, another species of the genus Cardiovirus, is dispensable for translation in vitro and for replication of the virus in tissue culture . These authors found, however, that elimination of the motif strongly influenced the pathogenesis of Theiler’s virus in mice. Meanwhile, available evidence has clearly indicated that IRES elements do not function by the same mechanism in recruiting the ribosomal subunits, followed by selecting the initiating AUG codon. Indeed, the HCV-related type IV IRESes of picornaviruses, discussed above, lack the YnXmAUG motif altogether.
As has been pointed out before, the nt sequences and apparent higher order structures of the four picornavirus IRES types vary widely. On the other hand, the overall function of IRES elements–internal initiation of translation–is the same, regardless of the underlying mechanism. Would an exchange of these different IRES elements between different picornaviruses yield viable viruses? The answer is yes. This was surprising since it was originally expected that these genetic elements that are huge in relation to the small viral genomes would contribute to functions other than translational control. Many IRES chimeras were analyzed using PV as the backbone. They included the exchange of the PV IRES with other type I IRESes such as CVB3 , HRV2 and HRV14 , with type II IRESes such as EMCV , and even with the IRES of HCV , a virus belonging to a different family. In HeLa cells at 37° C, these chimeric viruses replicated with wt kinetics, an observation suggesting that surrounding sequences did not influence the function of the IRES elements. Chimeras, however, may express interesting phenotypes, for example host range phenotypes, when assayed under different conditions. For example: PV carrying the HRV2 IRES [“PV(RIPO)”] is highly attenuated in cells of neuronal origin [99, 102].
These studies allow two conclusions. First, the IRESes of poliovirus, HRV, CVB3, EMCV, or HCV carry no essential signals necessary for PV genome replication. Second, IRESes are defined solely on the basis of function rather than on the basis of a specific structure. This notwithstanding, Le and Maizel  have suggested that picornavirus IRESes type I–III have evolved from a common structural core.
The family Flaviviridae includes the genera of Hepacivirus [HCV genotypes 1–7], Pestivirus [bovine viral diarrhea virus (BVDV) and classical swine fever virus (CSFV)] and Flavivirus [dengue virus (DENV), Japenese Encephalitis virus (JEV), Kunjin virus (KUNV), tick borne encephalitis virus (TBEV), yellow fever virus (YFV), West Nile virus (WNV)]. The members of this family are enveloped virions composed of a lipid bilayer with 2 or more species of envelope glycoproteins surrounding a nucleocapsid. The nucleocapsid contains the plus-strand RNA genome complexed with multiple copies of a small capsid protein. The RNA genomes (9–11 kb) of Flavivirus are capped at the 5′-end, that of the Hepacivirus and Pestivirus genera are terminated with p(pp)pGCC… Translation of members of the Hepacivirus and Pestivirus is controlled by IRES elements; the first of these was discovered for HCV by Tsukiyama-Kohara et al. . All genomes of the Flaviviridae encode an open reading frame (ORF) specifying a polyprotein with structural (SP) and nonstructural (NSP) domains (Fig. 6A). The polyprotein is processed by cellular and viral proteinases. The HCV genome contains, in addition, a small extra-frame ORF in the 5′ end of the coding region. The viral RNA lacks a 3′-terminal poly(A).
Hepatitis C virus (HCV) (genotypes 1-7) is the only member of the Hepacivirus genus. A comprehensive computational survey of conserved structure motifs has predicted a large number of structural elements both in the nontranslated regions and the ORF of the HCV genome .
The 5′NTR of HCV folds into a complex structure containing 4 distinct domains (domains I to IV)  of which domain II to IV belong to the IRES (Fig. 5A), altogether 341 nt long. Using chimeric subgenomic replicons of HCV, Kim et al.  have demonstrated that the first 40 nt of HCV RNA (including domain I) as well as domain II of the IRES are essential for HCV genome replication whereas the other two IRES domains III and IV aid in, but are not essential for, efficient RNA replication. In other studies Friebe et al.  demonstrated that although the first 125 nt of the 5′NTR are sufficient for RNA replication the efficiency could be greatly increased by the presence of the complete HCV 5′NTR. A short and highly conserved sequence mapping next to domain I binds the liver-specific micro-RNA, miR122, an interaction that appears to be required for efficient HCV RNA replication .
Numerous mutational studies of the HCV IRES [109, 110] and analyses by NMR and X-ray crystallography [111–113] have led to its apparent secondary structure (Fig. 6A), and subsequently to higher order structures in the absence or presence of the ribosomal subunit .
Domain II is organized into basal domain IIa and apical domain IIb of which IIa appears to provide the proper configuration for the apical domain to interact with the ribosomal subunit. Domains III and IV, in turn, provide the platform for binding the very limited menu of canonical translational factors plus the ribosomal subunit. In contrast to the picornavirus type I and II IRESes, none of the canonical translation factors eIF4A, eIF4B and eIF4G are required for HCV function [85, 114].
Note that the secondary structure of domain IIa of the IRES was solved by genetic analysis in 2001  and it was subsequently confirmed in studies of the entire domain II by NMR . Surprisingly, nearly all research publications up to the time of writing this review, and also most review articles, reproduce domain IIa with the wrong, pre-2001 secondary structure.
The downstream boundary of the HCV IRES extends unexpectedly into the core-coding region [101, 115–117]. In the HCV core-coding region, adjacent to the 5′NTR, there is evidence for both phylogenetically conserved RNA structures and an overlapping reading frame (ARF) in the 1+ frame, the latter encoding a putative polypeptide of approximately 124 aa . Antibodies to polypeptides from this reading frame have been found in HCV infected patients, but the significance of its expression remains unclear [119–122]. Two RNA stem loop structures SL-V and SL-VI have been predicted to exist in the region containing the ARF and were confirmed by enzymatic structure probing (Fig. 6A) . This region was found to be dispensable for the replication of the subgenomic HCV replicon in cell culture . It was proposed that these stem loops might be involved in the stimulation of HCV IRES function by reducing inhibitory interactions between the 5′ NTR and the core region [116, 123]. Using a chimpanzee model McMullan et al.  have also demonstrated that the ARF protein was not essential for HCV genotype 1a H77 RNA replication. However, the ARF region contains a functionally important RNA element (SL-VI) whose role remains unknown.
A recent study by Vassilaki et al.  has examined the roles of 4 predicted RNA stem loops in the core-encoding region (SL47, SL87, SL248 and SL 443) and also evaluated the function of the core+1 ORF. In agreement with the results of McMullan et al. , the expression of the core+1 ORF exhibited no role in the replication of HCV JFH1 isolate either in tissue culture or in xenografted mice. Using a mutational analysis it was observed that SL47 and SL87 (corresponding to SL-V and S-VI, respectively)  are important RNA elements contributing to HCV genome translation and replication both in cell culture and in vivo but SL248 and SL443 have no function. In addition, an interaction between a sequence at the 5′ end of SL87 (nt 428-442) and nt 24-38 of the 5′NTR was predicted to be detrimental to IRES-dependent translation .
The 3′NTR of HCV contains 3 domains, namely, the variable region (VR), the poly(U/UC) tract, and the highly conserved 3′X tail, which contains 3 putative stem loop structures SL1, SL2, and SL3  (Fig. 5A). Using deletion mutants of the HCV-N subgenomic replicon RNA, Yi and Lemon  demonstrated that the 3′-most 150 nt [the 3′X tail and the 3′ 52 nt of the poly(U/UC) tract] contained essential signals for RNA replication. The remaining segment of the 3′NTR enhanced replication but was not absolutely required. Similarly, in another study the 3′X tail and the poly(U/UC) tract were found to be required for infectivity of genome length RNA inoculated into chimpanzees .
Several independent groups have predicted the existence of extensive RNA secondary structure within the C-terminal NS5B encoding region of HCV, just upstream from the 3′NTR, suggesting the possibility that these have a function in vivo [120, 127–130]. The C-terminal coding region (nt 9126-9374) of NS5B contains 4 highly conserved and stable stem loops (SL-IV to SL-VII) (Fig. 6A) . A mutational analysis of these stem loops, in the context of the subgenomic HCV replicon, revealed that SL-V (nt 9262-9311) and SL-VI (nt 9215-9260) are essential for RNA replication in Huh-7 cells while the integrity of SL-IV (nt 9318-9355) and SL-VII (9129-9189) is less important. In vitro gel shift and filter-binding assays have shown that purified RNA polymerase NS5B specifically binds to SL-V. A similar study by You et al.  has also identified SL-V, designated as 5BSL3.2, as an essential cis-acting RNA for the replication of the HCV replicon in tissue culture. This hairpin is about 50 nt in length consists of an 8-bp lower helix, a 6-bp upper helix, a 12 bp terminal loop and an 8-bp internal loop. Primary sequences in the loops and RNA structure of the upper helix were found to be important for function. Interestingly, this RNA stem loop could be functionally moved to the 3′NTR, albeit with a reduction in replication efficiency. This same hairpin 5BSL3.2 was found to be involved in a “kissing” interaction between the upper loop and the loop of another hairpin SL2, located in the 3′NTR (Fig. 6A) . The importance of this tertiary RNA structure was confirmed by the rescue of RNA replication by compensating changes in both stem loops. Recent studies with the HCV cell culture system (HCVcc), using fully infectious HCV, demonstrated that both the “kissing” loop interaction and the length and composition of the poly(U/UC) tract were critical important for the replication in the genotype 2a HCVcc context . A model was proposed in which one or more trans acting factors interact with the highly conserved X tail to aid the formation of the “kissing” loop interaction. Using various bioinformatics methods to detect phylogenetically conserved RNA secondary structures, Diviney et al.  have predicted an additional long range interaction between the bulge loop of 5SBL3.2 (also known as SL9266) and an unpaired sequence located about 200 nucleotides upstream (around nt 9100) in the NS5B coding sequence. Mutational analyses of the two interacting sequences provided genetic support for this interaction. Fig. 6 illustrates the proposed structure of this pseudoknot with long-range interactions of SL9266 with 3′ and 5′ sequences. Despite extensive studies in several laboratories the precise role of this stem loop structure (SL9266) in HCV RNA replication remains unknown.
Recently, large-scale ordered RNA structures (GORS) were detected in the genomes of HCV by advanced computational methods [64, 65]. The presence of these large RNA structures was confirmed by hybridization accessibility assays. In addition, using atomic force microscopy HCV was visualized as tightly compacted spheroids. HCV belongs into a group of plus strand RNA virus where the presence of GORS is associated with viral persistence in the host.
Pestiviruses such as BVDV and CSFV are the causative agents of important livestock diseases.
The 5′NTRs of pestiviruses and hepaciviruses have similar structural and functional organization. Just as the HCV genome, the pestivirus genome is not capped. It starts with the tetra nucleotide pGUAU whose integrity is required for efficient genome replicaton . The 5′NTR (nt 385) of BVDV (NADL strain) contains stem loops designated as Ia, Ib, IIa, IIb, and IIIa-IIIe (Fig. 6B). The IRES resides within domain II (near nt 75) through nt ~310; that is, the 3′-end of the IRES extends into the ORF just as in the case of HCV . Domain Ia is a bifunctional structure element in that it modulates both translation and RNA replication [138, 139]. Mutations within the highly conserved 5′-terminal 4 nt (GUAU) of the genomic BVDV RNA resulted in severely impaired replication phenotypes .
The 3′NTRs of pestiviruses are about 190 to 270 nt long. According to computer analyses the 3′-most 70 nt represent a highly conserved element of which the last 56-60 nt form a stable stem loop structure (SLI) [139, 141]. The remaining region, with variable sequences, is predicted to form two less-stable stem-loop structures (SLII, SLIII), designated as 3′V  (Fig. 6B). Studies with subgenomic BVDV replicon RNAs indicated that the 3′-terminal SLI and part of the single stranded region between SLI and SLII were essential for RNA replication . The same results were obtained in the context of an infectious full-length BVDV cDNA clone . The deletion of both SLII and SLIII were also found to result in a lethal growth phenotype . However, deletions of either SLII or SLIII did not have a significant effect on viral replication. Interestingly, using a replicon system, Isken et al.  observed that the proper conformation of 3′V is required for accurate termination of translation at the stop codon of the viral ORF and this is essential for efficient RNA replication. However, this observation could not be confirmed in the context of the full length BVDV genome . The binding of the so-called NFAR proteins both to the 3′V domain and to the 5′NTR was observed suggesting the possibility of an interaction between the 5′ and 3′-terminii of the viral RNA .
The Flavivirus genus includes the medically important arthropod-borne viruses such as the mosquito transmitted DENV, YFV, KUNV and WNV or the tick borne encephalitis virus (TBEV). Just like HCV or pestiviruses, the flaviviruses are enveloped plus- strand RNA (11kb) viruses whose genome encodes a single ORF and lacks a 3′-terminal poly(A). In contrast to the two other genera, however, the flavirus genome lacks an IRES. Instead it controls its translation via a type 1 cap at the 5′-end. It remains a puzzle why within the family of Flaviviridae, the member viruses of the genus Flavivirus use a cap-dependent mechanism of initiation of translation whereas member viruses of the Hepacivirus and Pestivirus use IRES elements that are structurally related to each other and to the above-mentioned picornavirus type IV IRESes. It has been speculated that “… since all picornaviruses contain IRES elements but only certain flaviviruses contain such elements then it more likely that the flaviviruses acquired IRES elements (by recombination?) from picornaviruses” .
The 5′NTRs of flaviviruses, which are about 100 nt in length, exhibit a high degree of sequence conservation among different strains of the same virus but less conservation among the different members of the genus . The predicted secondary structures all contain a large stem loop (SLA) (~70 nt), which for DENV was proposed to function as the promoter for negative strand RNA synthesis (Fig. 6C) [147, 148]. Two helical regions were identified in DENV SLA, a side stem loop, a top loop and an U bulge as critical for RNA replication . A conserved oligo(U) track present downstream of SLA was found to modulate RNA synthesis in transfected cells . A smaller hairpin (SLB) is located at or near the translation initiation codon. This RNA element contains a 16 nt long stretch, called the 5′ upstream AUG region (5′UAR), which is complementary to a sequence in the 3′NTR (3′UAR). This pair of complementary sequence was named cyclization sequence 5′-3′ UAR and the importance of this long-range interaction was shown to be essential for RNA replication of both DENV and WNV [149–151]. Additional complementary sequences have been identified between the coding sequence of protein C (5′CS) and a sequence just upstream of the 3′SL in the 3′NTR (3′CS). The requirement for complementarity between 5′CS and 3′CS for RNA replication was demonstrated for KUNV, DENV and WNV [150, 152, 153]. Long range interactions between the ends of flavivirus RNA were confirmed by visualization of individual molecules using atomic force microscopy .
In many mosquito borne flaviviruses the start codon of the polyprotein is in a poor Kozak initiation context suggesting that this start codon would be utilized inefficiently for the initiation of translation. A small RNA hairpin (cHP) located 14 nt into the capsid coding region of DENV and WNV was found to be involved both in the recognition of the 5′ C protein start codon and also in RNA replication [154, 155] (Fig. 6C). Among different flaviviruses cHP sequences are poorly conserved. Interestingly, no sequence-specific element has been identified in cHP but an intact structure is required for function.
The 3′NTRs of flaviviruses are relatively large, and exhibit extensive heterogeneity in size and sequence. The 3′NTR of DENV can be divided into three domains. Domain I of is located immediately downstream from the stop codon and is variable in sequence (Fig. 6C) . Domain II contains moderately conserved sequences and two stem loops (A2 and A3), including a dumbbell structure with a conserved sequence motif (CS2) [105, 157]. Deletions of A2 or A3 resulted in reduced RNA replication . The most conserved domain (III) of the 3′NTR includes a stable stem loop at the terminus (3′SL), whose structure was confirmed by biochemical probing and which is absolutely required for RNA replication [150, 158]. Upstream of the 3′SL there is the conserved cyclization sequence (CS1 or 3′CS) (Fig. 6C) .
The 3′NTR of tick-borne flaviviruses, exemplified by TBE (tick-borne encephalitis virus), is subdivided into a variable domain, just downstream from the stop codon, and a 3′-terminal core element [159–161]. The variable region ranges in size from 50 nt to 400 nt and in some cases includes an internal poly(A) stretch. The core element, which is about 350 nt in length, is mostly conserved in sequence and structure. The very 3′-terminal 90-100 nt forms a highly conserved structural element, characteristic of all members of the Flaviviridae. Recently cyclization sequences (5′-CS-A and 3′-CS-A) have been identified for TBEV that are unrelated to the mosquito-borne flavivirus CS elements and are located at different genomic positions . The 5′-CS-A is located in the 5′NTR upstream of the AUG start codon rather than down stream. The 3′-CS-A is situated in the stem of the 3′-SL structure. In contrast to mosquito-borne flaviviruses, the C protein-coding region is not required for RNA replication.
The family of Picornaviridae contains vertebrate viruses, while plus-strand RNA viruses of invertebrates that have properties similar to those of picornaviruses have been termed as “picorna-like” viruses . Several of these have genomes that are organized differently than those of picornaviruses. They have two large, nonoverlapping ORFs, and have been classified into a new family Dicistroviridae. The 5′-proximal ORF1 encodes the nonstructural proteins and the 3′-proximal ORF encodes the capsid precursor. ORF1 contains motifs for “picorna-like” replicases (helicase, protease, polymerase). Some dicistroviruses contain 2A-like sequences but others do not. Interestingly, they contain repeated VPg coding sequences, just like FMDV. The second ORF is composed of coding sequences for capsid proteins VP2, VP4, VP3 and VP1 . VP4 is cleaved from the VP0 precursor (VP4-VP3).
The best-characterized member of the family is Cricket paralysis virus (CrPV) belonging to the Cripavirus genus. Other well-known members of the genus include Drosophila C virus (DCV) , and Plautia stali intestine virus (PSIV) . The genome of Cricket paralysis virus is about 8900 nt long, and contains two nonoverlapping ORFs translated from two different IRES elements (Fig. 7A). The 5′NTR (708 nt) contains the first IRES element  and an intergenomic region (IGR) IRES is located between nt 6022 and 6214 (Fig. 7B). The length of the 3′NTR in dicistroviruses varies from 150-500 nt .
What is unusual about the IGR-IRESes of Dicistroviridae is that they do not require any of the canonical translation initiation factors to recruit the ribosome to the viral RNA, initiation is from a non-AUG start codon and from the ribosome’s A site rather than P-site [167–170]. The cryo-EM structure of the CrPV IRES, bound to ribosome, was determined by Schuler et al. (Fig. 7B). The structure indicates the presence of three pseudoknots (PKI, PKII and PKIII). Domain 1 of the IRES interacts with the 60S ribosomal subunit in the E and P-site regions, domain 2 interacts with the 40S subunit in the E-site region, and domain 3 places the alanine start codon into the A-site.
The Togaviridae family of animal viruses contains two genera, the Alphavirus and Rubivirus. These enveloped viruses contain an icosahedral nucleocapsid consisting of the viral RNA and multiple copies of a single capsid protein (C). The RNA has a capped 5′-end and a poly(A) tail at the 3′ end. The genome of these viruses is 9–12 kb long and it contains two ORFs. The 5′-terminal two thirds of the genome is translated into nonstructural proteins. The 3′-proximal ORF encodes the structural proteins, which is translated from a subgenomic RNA (SG) (Fig. 8). The synthesis of both genomic (G) RNA and SG RNA is templated by the minus strand RNA. However, the mechanism is different because the synthesis of G RNA begins at or near the 3′ end while synthesis of SG RNA begins with internal initiation on the minus strand template [171, 172].
The alphavirus genus contains about 30 members with Sindbis virus (SIN) as the prototype. The majority of alphaviruses are transmitted by mosquitoes to birds and mammals that serve as hosts. In cultured mosquito cells they establish persistent infections but infection of vertebrate cells leads to progressive cythopathic effect and cell death.
The 5′-terminal 200 nt of SIN RNA contains two cis-acting elements, one located in the 5′NTR and the other in the ORF, called conserved sequence element, CSE (Fig. 8A). The 5′NTR (nt 1-59) contains a small (SL1) and a large stem loop structure (SL2). The 51-nt long CSE is located in the nsP1 coding sequence (nt 155-205), that is predicted to form 2 smaller stem loop structures (SL3 and SL4) . Mutations in SL1 (nt 1-44) were either lethal or resulted in poor growth phenotypes  and were essential for efficient minus strand RNA synthesis [173, 175, 176]. The effect of the mutations differed in mosquito cells and in mammalian cells. SL2-SL4 comprises a replication enhancer but it is not essential for replication. The 51 nt CSE also enhances RNA replication and its integrity is more important for SIN replication in mosquito cells than in mammalian cells [173, 174, 177]. The complement of the 5′-terminal 200-250 nt was found to be required for the synthesis of viral genome RNAs from minus strand intermediates [173, 178].
It is interesting to note that some DI particles of Sindbis virus, obtained after serial high multiplicity passaging on cultured cells, were found to contain a cellular tRNAASP sequence (nt 10-75) at their 5′-ends [179, 180]. This RNA sequence could be folded into a cloverleaf like structure but missing the 5′-terminal stem. In the two DI RNAs examined, the 3′ end of the cellular tRNAASP was covalently attached to viral sequences and therefore did not possess amino acid accepting activity.
The 3′NTR of alphaviruses varies from 121-524 nt in length and that of SIN is 323 nt long (Fig. 8A). Immediately preceding the poly(A) tail a highly conserved 19 nt long CSE, with no discernable secondary structure, was identified. This sequence was postulated to play a key role in the initiation of minus strand RNA synthesis from a plus strand genomic template . Upstream of the 3′-terminal CSE the 3′NTR contains an AU rich segment and repeated sequence elements 25-72 nt in length. In the 3′NTR of SIN there are 3 copies of a 40 nt long repeated element. Deletion of the repeated elements resulted in a viable virus but the growth of the virus was impaired . It is not yet known whether the 3′ and 5′ ends of the genomic RNA interact to facilitate minus strand RNA synthesis. Panhandle structures have been visualized by electron microscopy  but complementary sequences near the 5′ and 3′ ends are not evident.
In SIN the initiation of nucleocapsid assembly begins with the recognition and binding of the encapsidation signal by capsid protein. The encapsidation signal of SIN genomic RNA was localized to a 132 nt long segment of the nsp1 coding sequence, spanning nt 944-1076, which is predicted to form 4 stem loop structures connected by single stranded RNA sequences (Fig. 8A) [183–185]. The capsid protein of SIN is 264 aa long and of these amino acids 81–112 are involved in the recognition of the encapsidation signal . Two purine-rich loops were found to be essential for the binding of the capsid protein to the RNA.
A second internal cis-acting RNA element in the SIN genome is located in the “junction region”, which includes nucleotide sequences of the genomic RNA preceding and including the beginning of the 26S SG RNA. This region contains the promoter for the synthesis of the subgenomic RNA (26S), in the context of the minus strand. This is a highly conserved sequence in alphaviruses that encompasses 19 nt upstream and 2 nt downstream from the initiation site (Fig. 8A) . Subsequent studies with DI particles indicated that the minimal subgenomic promoter for SIN was 18-19 nt upstream and 5 nt downstream from the start of the subgenomic RNA .
Rubella virus (RV) is the only member of the Rubivirus genus. An interesting feature of rubella virus genotypes is their genomic uniformity . A comparison of 8 genotypes showed that 78% of nucleotides in the genomes were conserved. In addition, the viruses in all 8 genotypes had the same number of nucleotides in each of the 2 ORFs and NTRs.
The 5′-end of RV genomic RNA contains a 14 nt single-stranded leader followed by a stem loop structure [5′ (+)SL] (nt 15-65), which contains the AUG (nt 41) for initiation of translation of the nonstructural proteins (Fig. 8B). Mutations of an AA dinucleotide at positions nt 2 or 3 were lethal but no other mutation within the leader or the 5′ (+)SL was lethal . Some mutations in the leader or the stem of 5′ (+)SL resulted in viruses that grew to a low titer. Interestingly, mutations in the 5′ (+)SL resulted in a significant reduction in the synthesis of nonstructural proteins indicating that this structure is important for translation and not for plus-strand RNA synthesis.
The complementary negative strand equivalent of the 5′ SL structure of RV [3′ (-)SL] was found to bind specifically 3 cellular proteins (p56, p79, p97). Altering the SL structure in either one of the two predicted loops has abolished the binding interaction .
Four stem loop structures were thermodynamically predicted to exist within the 3′-terminal 305 nt (3′CSE) of the RV RNA (SL1to SL4) (Fig. 8B) . SL1 and SL2 are located in the E1 coding region while SL3 and SL4 are within the 59 nt long 3′NTR. Mutational analyses indicated that most of the 3′NTR is required for RNA replication except for the 3′-terminal 5 nt and the poly(A) tail. Maintenance of the SL3 structure, an 11 nt long single stranded sequence between SL3 and SL4, and the sequences forming SL4 were all important for virus viability . A cellular protein, calreticulin (CAL), was found to bind to SL2 but this was independent of virus viability. Subsequent studies by Chen et al.  showed that the part of the 3′CSE, overlapping the E1 coding region, affected plus-strand but not minus-strand RNA synthesis.
The two ORFs of RV are separated by an untranslated region known as the J-UTR (junction UTR), which contains on the negative strand the promoter for subgenomic (SG) RNA synthesis. The secondary structure predicted for the junction region contains two stem loop structures (SLI and SLII) (Fig. 8B)  of which SLII is within the subgenomic promoter (SGP). The minimal SGP starts from nt −26 through the SG start site and extends to nt +6, although a larger region is required for the generation of virus with wt phenotype.
The family Coronaviridae, in the order of Nidovirales, includes the largest plus-strand RNA viruses with 5′-capped and 3′-polyadenylated genomes ranging from 27.3 kb [human coronavirus (hCoV)] to 31.3 kb [mouse hepatitis virus (MHV)] . At the 5′-end there is a 60-80 nt long leader sequence followed by a 200-500 nt long nontranslated region. Coronavirus genomes contain multiple ORFs with the replicase gene occupying the 5′-most two-thirds of the genome followed by the nonstructural proteins (Fig. 9). Additional accessory genes are interspersed among the structural proteins. There are 5–7 overlapping, 3′-coterminal subgenomic mRNAs, each of which is capped, polyadenylated, and contain a common leader sequence with the genomic RNA.
Most studies of coronavirus cis-acting RNA elements were conducted with DI RNAs because of the large size of the genomic coronavirus RNAs. DI RNAs, which contain extensive deletions, are not viable but they can propagate by using the RNA synthetic machinery provided by a homologues helper virus.
The 5′-terminal leader sequence of BCoV contains a stem loop structure that is involved in DI particle replication . In groups 1 and 2 coronaviruses the 5′NTR ranges in size from 210 nt [MHV, hCoV and bovine corona virus (BCoV)] to 314 nt [transmissible gastroenteritis virus (TGEV)]. The minimal 5′ cis-replicating element for RNA replication usually extends into the replicase ORF. In the case of group 3 coronaviruses such as IBV the 5′NTR (528 nt) appears to contain the entire cis-acting RNA element. The 5′-genomic region of BCoV contains at least 6 stem loop structures, numbered SL-I to SL-VI in the 5′ to 3′ direction, whose structures were confirmed by enzymatic probing (Fig. 9) [195–199]. Stem loops I and II were found to be essential for RNA replication. SL II contains a “transcription regulating sequence” (TRS), which is involved in the transcription of SG RNAs. SL-III and SL-IV were identified as cis-acting elements for DI RNA replication. In these stem loops the structure of the stem rather than the primary sequences are important RNA replication. SL-IV contains the start codon for the replicase ORF in the downstream arm of its stem. The binding of purified nsp1, the N-terminal ORF 1a protein, to SLIII and its flanking sequences was also demonstrated . A recent study suggests that the base of SL-I mediates a long range interaction between the 5′NTR and the 3′NTR that is an essential step in the transcription of SG RNAs but is not required for genomic minus-strand synthesis .
Studies with BCoV DI particles have determined that a 186 nt long sequence from the nsp1 coding region of the DI RNA ORF is required for RNA replication . This sequence contains 2 stem loops (SL V and SL VI) whose structures were confirmed by RNase probing. SL VI maps within the nsp1 coding region (ORF 1a) found in all naturally occurring DI particles (Fig. 9) . Mutational analyses have indicated that SLVI is required for DI RNA replication and is likely to be also involved in the replication of the full-length viral genome . Recently, SLV was also identified as a cis-acting RNA element . In SLV the overall higher-order structure and the integrity of the lower stem contributed to optimal RNA replication. Two cellular proteins were found to bind SLV and SLVI in vitro but the biological relevance of these interactions is not yet proven .
The 3′NTR of coronaviruses ranges in size from 273 nt in TGEV to about 505 nt in infectious bronchitis virus (IBV) and is followed by a poly(A) tail. For MHV the domain required for RNA replication not only encompassed the 3′NTR but also ran into the adjacent N ORF . In contrast the required cis-acting signals for coronaviruses group 1 and 2 did not contain any of the N coding sequences . The 3′NTR of MHV and BoCV upstream end contains a bulged stem loop  and a pseudoknot (Fig. 9) . The bulged stem loop and the pseudoknot overlap and they cannot form simultaneously. This has led to a proposal that these structures are part of a molecular switch that regulates different steps of replication. The pseudoknot is followed by a variable region containing nt 46-156 from the 3′-end of MHV RNA (HVR), within which a highly conserved octanucleotide (GGAAGAGC) is located. Interestingly, the MHV HVR is required for DI RNA replication but not for the replication of full length genomic RNA [206, 207]. However, the HVR has a significant role in pathogenesis . The very 3′-end of the genome (~55 nt) is required for MHV minus strand RNA synthesis, as determined by DI RNA analysis, in conjunction with a poly(A) tail of 5-10 nt [205, 208, 209]. Recently, a direct interaction was demonstrated between loop 1 of the pseudoknot and the extreme 3′ end of the MHV genome (Fig. 9) . In the same study an interaction between the pseudoknot and gene 1a replicase products nsp8 and nsp9 was also reported.
An internal cis-acting RNA element is located in the ORF1b of group 2 coronavirus MHV. A deletion analysis of MHV DI RNAs has identified a 190 nt long sequence within which a 69 nt bulged stem loop was required for packaging of DI RNAs into particles (Fig. 9) . RNA structures similar to that found in MHV are also found in the ORF 1b regions of BCoV  and SCoV but their role as encapsidation signals remains to be demonstrated . It should be noted that in other coronaviruses, such as IBV and TGEV, packaging signals are located at the genome ends .
The two replicase proteins of SARS-CoV, which are coterminal with their N-terminal ends, are produced by ribosomal frameshifting [215, 216]. This process is dependent upon a slippery heptanucleotide sequence (UUUAAAC) and a closely spaced pseudoknot, located in the replicase ORF (Fig. 9).
Members of the Arteriviridae family are also called “small nidoviruses” because they have smaller genomes (13–16 kb) than the other virus families in the Nidovirales order. They are enveloped animal viruses that cause asymptomatic, persistent or acute infections. The nidovirus genomes contain a 5′ cap structure, 5′ and 3′ NTRs, and a poly(A) tail. Their RNA genomes encode multiple (4–14) ORFs (Fig. 8). There are two large replicase ORFs and a set of downstream ORFs, which encode structural proteins from SG RNAs . Not much is known about RNA structures in Arteriviridae because DI particles have not been observed for any arteriviruses. Recently, a reverse genetic system was used to study RNA elements in Porcine reproductive and respiratory syndrome virus (PRRSV), which uses an infectious cDNA as a bacterial artificial chromosome .
The 5′NTRs of arterivirus genomic and subgenomic RNAs contain a leader sequence (nt 1-211) . For Equine Arteritis Virus (EAV) the 5′-terminal 313 nt were shown to be sufficient for genome translation, RNA replication and SG RNA synthesis . RNA folding of this region, containing the leader and part of ORF1a, predicted 5 stem loops (SL A–E), a pyrimidine rich segment followed by an additional 5 stem loops (SL F–J) (Fig. 10A) . In stem loop G, contained in the leader sequence, there is a conserved transcription regulating sequence (TRS), which base pairs with downstream complementary TRS sequences (“body” TRSs). This interaction is essential for SG mRNA synthesis [220–222].
Both full-length and SG minus-strand RNA synthesis are initiated at the 3′-end of the EAV genomic RNA. Using computer aided analysis and chemical and enzymatic probing of the 3′-terminal region (200 nt) of EAV RNA two domains were identified that are required for RNA synthesis . The first domain, directly upstream of the 3′NTR (nt 12610-12654), contains one small stem loop (SL IV) and a single stranded region (Fig. 10A). The second domain is located within the 3′NTR (nt 12661-12690) and is predicted to fold into a prominent stem loop (SL V) with a large loop. Two cellular proteins (PTB and aldolase) were found to specifically interact with the 3′NTR but the biological significance of these interactions has not been established .
ORF7 of PRRSV, which encodes the nucleocapsid protein, is located just upstream of the 3′NTR. It contains a 34 nt long sequence that is required for minus-strand RNA synthesis . This sequence is highly conserved among PRRSV isolates and is predicted to form a hairpin (Fig. 10B). A 7 nt long sequence within the loop of this structure was predicted to form a “kissing interaction” with the loop of a hairpin in the 3′ NTR. Mutational analyses have confirmed that the “kissing interaction” was required for replication but the ability of the two loops to base pair and not their sequence was functionally relevant.
The arterivirus genomes contain two large replicase ORFs (ORF1a and ORF1b). ORF1b translation requires a ribosomal frameshift just before ORF1a translation is terminated. The ORF1a/1b overlap region contains two signals that are believed to promote this function. The first is a “slippery” sequence (GUUAAC) that is the actual frame shift site and a downstream pseudoknot structure (Fig. 10A) [217, 226].
The family Caliciviridae contains small plus-strand RNA viruses, which are members of 4 genera: Norovirus, Sapovirus, Vesivirus, and Lagovirus. Human noroviruses and sappoviruses are important medical pathogens of humans, which cause gastroenteritis. Murine norovirus (MNV), the most thoroughly studied calicivirus, causes diarrhea and lethality in mice deficient in components of the innate immune system. Vesiviviruses such as feline calicivirus (FCV) cause respiratory disease in cats and lagoviruses such as rabbit hemmorhagic disease virus (RHDV) causes hemmorhagic disease in rabbits. The genome of caliciviruses is 7.3–8.5 kb in length and is organized into 2 or 3 ORFs (Fig. 11). The nonstructural proteins are encoded near the 5′-end and the structural proteins near the 3′-end of the genome. The viral RNA is covalently linked to a terminal protein (VPg) at the 5′-end and the 3′-end is polyadenylated. So far very limited amount of information is available about cis-acting RNA structures in caliciviruses. Using a variety of bioinformatics methods Simmonds et al.  have recently searched for conserved RNA structures in the genomes of different human and animal caliciviruses.
Caliciviruses have short 5′ NTRs of 5-20 nt preceding their coding regions. Simmonds et al.  have predicted the presence of RNA structures in the first 150 nt of calicivirus genomes but there was no similarity in shape or position between groups or even within a genus. Nucleotide substitutions introduced into the 5′-terminal two stem loops starting at nt 8 and 29 of the murine norovirus (MNV) RNA genome resulted in a 20 fold reduction in virus titer and no significant inhibition of translation (Fig. 11) . In the 5′NTR of Norwalk virus (NV) (norovirus genus) the existence of a double stem loop was predicted (nt 20-98) but its biological function was not tested . Interestingly, caliciviruses use a novel protein-directed translation initiation, which relies on the interaction of eIF4E and VPg-linked to the RNA . The 5′-terminal RNA structures in MNV RNA do not appear to have a significant role during the translation process .
The 3′NTRs of caliciviruses vary in length from about 45 to 100 nt. Conserved secondary structures were reported in the 3′NTRs of MNV, human noroviruses, vesiviruses and sapoviruses [66, 229]. Two large stem loops were identified in MNV sequences, a terminal hairpin from nt 7330 to the end of the genome and another large stem loop (164 nt) centering around position 7239 (Fig. 11) . Mutational analysis of the terminal stem loop confirmed its essential role in RNA replication. The presence of 3′-terminal RNA stem loop structures was also reported in FCV genomes, both in the 3′NTR and in the adjacent coding sequences of ORF3 . Binding of cellular proteins La, PTB, and PAB to the 3′NTR-poly(A) of Norwalk virus was reported but the biological significance of this observation is not yet established .
A highly structured region was observed at the junction of the nonstructural and structural proteins in human norovirus, vesivirus, lagovirus and sapovirus . A stem loop of varying sizes was located in the minus strand just 6 nt upstream from the transcription site of the SG RNA in all of the caliciviruses tested indicating putative roles as promoters for SG RNA synthesis (Fig. 11). Nucleotide substitutions and repair mutations in the stem of the MNV stem loop resulted in a loss of viral viability and confirmed the importance of the RNA structure but not of the sequence.
The family Astroviridae includes nonenveloped human and animal plus strand RNA viruses, which usually cause gastroenteritis. The plus strand RNA genome of astroviruses (6.4–7.3 kb) is linked to a terminal protein (VPg) at the 5′-end  and has a poly(A) tail at the 3′ end (Fig. 12). The viral genome contains 3 overlapping ORFs. ORF1 (1a and 1b) encodes the nonstructural proteins while ORF 2, at the 3′ end of both genomic and SG RNAs, encodes the structural proteins.
A short 5′NTR (11-85 nt) precedes ORF1, which has not yet been characterized. The size of the 3′NTR varies from 80 nt human astroviruses up to 305 nt in avian astroviruses. The genome is terminated in a poly(A) tail of about 30 nt in human astrovirus . About 110 nt (80 from 3′NTR and 19 from ORF2) adjacent to the poly(A) tail of human astroviruses are very well conserved and are predicted to fold into two RNA stem loops (Fig. 12) .
A ribosomal frame shift site was identified in the overlap region (70 nt) of ORF1a and 1b, consisting of a heptanucleotide and an RNA stem-loop structure, which may form a pseudoknot with a downstream sequence (Fig. 12) . At the junction of ORF1b and ORF2 a 120 nt long sequence, highly conserved among astroviruses, was predicted to form 3 hairpins (Fig. 12) . The structure, which contains the start codon for the capsid gene as well as the stop codon for the polymerase gene, was proposed to be part of the promoter for sgRNA synthesis.
Recent bioinformatics analyses by Davis et al.  have predicted the presence of genome-scale ordered RNA structures in the genome of MNV, which was confirmed by probe hybridization accessibility assays. MNV is another example of mammalian viruses where the presence of GORS is correlated with the persistence of the virus in its host.
Nodaviruses are nonenveloped, icosahedral viruses. They are divided into two genera, α and β, that infect insects and fish, respectively. The most thoroughly studied insect viruses in this family are flock house virus (FHV) and blackbeetle virus (BBV). In tissue culture these viruses can replicate in cells derived from a variety of organisms such as yeast, mammals and plants.
The nodavirus genome consists of two plus stranded RNAs, RNA1(3.1 kb) and RNA2 (1.4 kb), that are capped at the 5′-end and do not contain a poly(A) tail at the 3′-end. The combined length of RNA1 and RNA2 is 4507 nt, one of the smallest known viral genomes for animal viruses. Both RNAs are required for infectivity and are packaged within the same virion. RNA1 encodes the multifunctional replication protein A (112 kDa), which serves as the RNA polymerase for the synthesis of both genomic RNAs and of a subgenomic RNA, RNA3 (387 nt). Protein A also contains signals for targeting and insertion into outer mitochondrial membranes where RNA replication takes place . RNA3, which corresponds to the 3′-end of RNA1, is not packaged into virions. It encodes two ORFs for the expression of proteins B1, whose function is unknown, and B2 (12 kDa), which is a suppressor of host-mediated silencing. RNA2 encodes the capsid protein precursor α (43 kDa) , whose subunits assemble into a precursor viral particle, the provirion. Interestingly, RNA3 also acts as a transactivator of RNA2 replication . At the onset of RNA2 replication the synthesis of RNA3 is suppressed .
Various RNA dimers can be observed during FHV replication, which appear to have functional roles in RNA replication . These include dimeric RNAs covalently linked, head to tail monomers of RNAs 1, 2, and 3, as well as heterodimers of RNA2 and RNA3.
FHV RNA replication is regulated by cis-acting RNA elements both at the 5′- and 3′-end of the genome . The 5′NTR of FHV is 39 nt in length and the 3′NTR is 71 nt long . Minus strand RNA synthesis depends on the 3′-proximal 108 and 50 nt of RNA1 and RNA2, respectively. The 3′-terminal RNA element of RNA2 is predicted to form two stem loop structures . In contrast, plus-strand RNA synthesis requires only about 3-14 nt at the 3′-terminus of minus strands .
There are also internal cis-acting RNA elements on the plus strand of both RNA1 (intRE; nt 2322-2501) and RNA2 (nt 520-720) . A second region with a 5′ boundary between nt 2735-2755 (3′RE) is also important for RNA1 replication. Studies with a DI particle, isolated from persistently infected Drosophila cells, suggested that a 32-bp region of RNA2 (nt 186-217) is required for packaging of RNA2 into virions .
A recent report described the identification of a cis-acting element in FHV RNA1 (nt 68-205), which is predicted to form two stem loops with nearly identical loop sequences . Mutational analyses of the stem loops showed that both are required for the recruitment of RNA1 to the outer mitochondrial membranes, and for minus and plus strand RNA synthesis. The 5′NTR of FHV, comprising the 5′-terminal 39 nt, had no effect on RNA1 recruitment to the membranes.
RNA3 synthesis requires two cis-acting RNA elements on RNA1, which take part in a long distance base-pairing, involving about 6 base pairs (ACCGGU/UGGCCA) . The proximal subgenomic control element (PSCE) (nt 2272-2282) is located just upstream of the RNA3 start site. A short distal subgenomic control element (DSCE) (nt 1229-1239) is 1.5 kb upstream from the RNA3 start site. Disruption of the DSCE-PSCE interaction abolished both negative and positive strand RNA3 replication. RNA3 can also replicate independently of RNA1 and this process requires the 3′-terminal 540 nt of plus strand RNA3 .
From the nine different virus families discussed in this review, those that have been studied in detail, all possess cis-acting RNA elements in the 5′- and 3′-terminal regions of their genomes. These RNA structures are involved in the complex process of RNA replication and in some cases also in promoting the initiation of translation. The virus families differ from each other greatly in the total number of functional RNA structures in their genomes and also in the lengths of their 5′ and 3′ NTRs. For example some viral RNAs contain large IRES elements in their 5′NTRs that are required for the translation of the polyprotein while other viruses with shorter 5′NTRs use cap-dependent translation. In general, where the 5′ or 3′ NTRs are too short to include all of the required functional elements, the RNA structures extend into the downstream or upstream coding sequences, respectively. In some viral genomes functional structural elements are also located in the coding sequences far away from the genomic ends that are not an extension of the 5′ and 3′ NTRs. These “internal” cis-acting RNA structures are involved in a variety of processes such as ribosomal frameshifting, RNA replication or encapsidation. Although the functional importance of many of these structures have been confirmed by genetic and biochemical studies their precise roles are not yet fully understood.
Recently, a less well-characterized, but much more extensive set of RNA secondary structures, designated as genome-scale ordered structure (GORS), have also been identified in the coding sequences of Picornaviridae (FMDV, kobuvirus, cardiovirus), Caliciviridae (OG caliciviruses), and Flaviviridae (HCV, HGV/GBV-C) genomes [64, 65]. Interestingly, there was remarkable variability between genera of each family that exhibited this characteristic. GORS were not associated with translation or replication strategies of the various genera but they were strikingly correlated with the ability of each genus to persist in their natural host. The mechanism that underlines the association between GORS and host persistence is not yet understood. Davis et al.  suggested the possibility that the role of GORS is in the modulation of innate intracellular defense mechanisms, and the acquired immune system triggered by dsRNA. GORS may delay or prevent interferon induction early in viral infection or replication.
The accumulation of knowledge about RNA structures in the genomes of plus-strand RNA viruses during the last decade not only enhances our understanding of the proliferation of individual viruses but also reveal certain unifying principles that link many aspects of translation, RNA replication and encapsidation. Future studies in this field will surely extend and refine our knowledge of how these structures function at the various stages of the viral life cycle and will have an impact on our ability to limit the toll of viral diseases.
We thank SK Jang for helpful discussions. This work was supported by grant AI15122 from the National Institute of Allergy and Infectious Diseases.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.