Probably the most well-known use of 5′-to-3′ communication is to promote or regulate translation of eukaryotic mRNAs. Although some have urged caution in broadly accepting that the 3′ end of an mRNA affects initiation from the 5′ end ¹ , continued experimentation supports the existence of this communication. To explain how 5′-to-3′ communication can occur, we first outline the canonical initiation mechanism. The eukaryotic initiation factor (eIF) 4F complex, which contains the cap-binding protein eIF4E and scaffold protein eIF4G, binds at the 5′ end ^{2– 4} . Bound eIF4F recruits the 43S complex, which contains the 40S ribosomal subunit ³ . This complex then scans the 5′ UTR to find the appropriate start codon ⁴ , where GTP hydrolysis, release of eIFs, and 60S subunit joining create a translationally competent 80S ribosome ^{5– 7} (reviewed in 8– 10). As an alternative to this canonical cap- and scanning-dependent pathway, some mRNAs and viral RNAs contain an internal ribosome entry site (IRES) that directs translation initiation using a 5′ end-independent mechanism. IRES RNAs often use a subset of the canonical eIFs and RNA structures to drive translation initiation (reviewed in 11, 12).

A ubiquitous and important form of mRNA 5′-to-3′ communication depends on the 5′ cap and poly(A) tail ( Figure 1). The presence of these two signals on opposite ends of an mRNA synergistically enhances the rate of translation of that mRNA ^{13, 14} . The model is that eIF4F binds to the cap at the 5′ end while multiple poly(A)-binding proteins (PABPs) bind the poly(A) tail and eIF4G, physically linking the 5′ and 3′ ends ( Figure 1b) ^{15, 16} . Functionally, the PABP-eIF4G interaction is associated with favorable subunit binding at the 5′ end ^{17, 18} , and it increases the affinity of cap-binding protein eIF4E for the cap ¹⁹ . The PABP-eIF4G interaction can be regulated, supporting the importance of the 5′-to-3′ communication. For example, PABP-interacting proteins 1 and 2 (Paip1 and 2) either up- or down-regulate translation of an mRNA message by modulating the 5′-to-3′ communication ^{20– 22} .

The PABP-eIF4G interaction within a single mRNA is the “closed loop model” ²³ , and it is proposed that the physical proximity between the 5′ and 3′ ends favors the transfer of terminating ribosomes from the 3′ end of the ORF to reinitiate at the 5′ end of the mRNA ^{14, 15, 24} . One piece of evidence in support of this model comes from the observation that ribosomes within mature polysomes have a rate of exchange with surrounding free ribosomes that is slow enough to suggest reinitiation on an mRNA ²⁵ . In addition, the 5′-to-3′ interaction may serve as a quality-control mechanism to favor translation of full-length, mature mRNAs ²⁶ .

The canonical PABP-eIF4G circularization strategy appears to be a generally important mechanism, but care must be taken not to oversimplify its details or nuances. For example, the synergistic increase in translation efficiency conferred by the cap and poly(A) tail is not strictly contingent upon circularization. Specifically, addition of a poly(A) RNA to translationally competent extract in trans stimulates translation of a capped and non-polyadenylated mRNA to a similar degree as when the poly(A) tail is present in cis; this may be due to increased affinity of eIF4E for the cap ²⁷ . Also, in a series of recent publications, it has been observed that mRNAs go through changes in their global architectures over time and as the number of loaded ribosomes increases—first forming circles, then “double-row” structures, then complex “helical” structures where the ends may no longer interact—and there is evidence of similarly dynamic mRNA “closed-loops” in living cells ^{25, 28– 30} . Even more interesting, when an mRNA is altered to prevent the PABP-eIF4G interaction, only initial rounds of translation initiation were slowed; the formation of higher-order polysome structures eventually occurred ^{25, 31} . This suggests that the PABP-eIF4G interaction might be important for “kick-starting” the initial rounds of translation but is less important during steady-state translation on large polysomes, where other forces dictate higher-order architecture. This in turn suggests a dynamic model for 5′-to-3′ communication linked to the maturation of polysomes and perhaps other yet-to-be-determined signals. Clearly, there is much more to be learned even in regard to the canonical eIF4G-PABP interaction, especially surrounding events in live cells and how this relates to other events in the mRNA life, such as decay.

Exploration of 5′-to-3′ communication starts with the canonical PABP-eIF4G interaction ( Figure 2a) but does not end there (pun intended). Other strategies include those used by IRES RNA-containing transcripts of viral origin that operate without a cap. For example, the viral RNA of poliovirus (and related viruses) is not capped, but translation of the viral genome is still enhanced by a 3′ poly(A) tail ³² . The poliovirus IRES RNA directly binds eIF4G, which could interact with PABP bound to the poly(A) tail ( Figure 2b) ^{33– 36} . However, during infection, a viral protease cleaves eIF4G and while a fragment of the factor binds the IRES, it cannot bind PABP. Thus, an alternative mechanism includes 5′-to-3′ communication between poly(rC)-binding protein (PCBP) bound to the 5′ cloverleaf structure, upstream of the IRES, and PABP bound to the 3′ poly(A) tail ( Figure 2b) ^{37– 39} . In addition, it has been shown that sequences upstream of the poly(A) tail, within the 3′ UTR, can enhance translation in the absence of the poly(A) tail, perhaps by using other proteins ⁴⁰ . Thus, even within this one IRES-containing virus, there may be multiple ways that 5′-to-3′ end communication can be achieved. In addition, viral RNAs with a 5′ viral protein of the genome (VPg) in place of a 5′ cap structure, such as members of the Calciviridae family, might use VPg recruitment of the eIF4F complex (in particular, eIF4G, along with 4A and PABP) to mediate end-to-end communication ⁴¹ . These different strategies may be used during different stages of viral infection, suggesting temporally dynamic interactions that respond to changing cellular conditions and help coordinate different viral processes, such as replication (discussed in more depth later in this review).

Viral IRESs are not the only places where alternative methods of 5′-to-3′ communication are found, as cellular mRNAs with IRESs also exhibit different strategies. An interesting example is found in the cellular c-myc mRNA, which contains an IRES that binds initiation factors, including eIF4G, and translation is enhanced by the poly(A) tail but without needing PABP ⁴² . In this case (and many others), the precise mechanism of 5′-to-3′ communication is mysterious. Another intriguing example is the cellular mRNA for serine hydroxymethyltransferase 1, which does not require eIF4G or PABP. Rather, proposed structures in the 5′ and 3′ UTRs bind to heterogeneous nuclear ribonucleoprotein H2 (hnRNP H2) and CUG-binding protein 1 (CUGBP1), respectively, which mediate IRES-driven translation initiation via direct protein-protein contact during states of decreased cap-driven translation initiation ( Figure 2c) ⁴³ .

Although IRESs are logical places to use alternative modes of 5′-to-3′ communication, non-canonical mechanisms are also found in viral and cellular transcripts that do not contain IRESs. Rotaviruses are capped but do not have a poly(A) tail; rather, viral non-structural protein 3 (NSP3) binds near the 3′ end and interacts with eIF4G ( Figure 2d) ^{43– 46} . Whereas earlier studies concluded that NSP3 was not involved in translation based on a partial knockdown ⁴⁷ , another interpretation of these data is that NSP3 is a potent translation enhancer, such that a 10-fold decrease in the amount of NSP3 levels still supports viral protein synthesis ⁴⁸ . Likewise, the cellular mRNAs that encode for histones have a highly conserved stem-loop structure at their 3′ ends rather than a poly(A) tail. This structure binds stem-loop-binding protein (also called hairpin-binding protein), which facilities interactions between initiation factors at the 5′ end ⁴⁹ . Clearly, the advantageous effect of 5′-to-3′ communication can be achieved with a variety of combinations of mRNA intrinsic elements (5′ cap, poly(A), and RNA structure), used with canonical factors (eIF4G and PABP), or several “non-canonical” proteins, or both. These mRNAs and viral RNAs illustrate how, despite the existence of the general eIF4G-PABP strategies, there are idiosyncratic methods of achieving 5′-to-3′ communication that add layers of regulatory complexity, which may be linked to cell type, cell conditions, and so on. No doubt there are many more strategies to be discovered and understood.

The above examples illustrate how 5′ and 3′ ends communicate through various combinations of RNA-encoded signals and proteins. Interesting variations on this theme are RNA signals near the 3′ end of an RNA that bind translation-essential proteins to be used at the 5′ end ( Figure 3). Examples of this phenomena are found in plant-infecting viruses, whose genomic RNAs are often not capped or poly(A)-tailed. Rather than evolving an IRES, these viruses have structured RNA elements in their 3′ UTRs called cap-independent translation enhancers (CITEs) (reviewed in 50, 51). CITEs are distinct from IRESs in that CITEs themselves do not bind ribosomes in the proper location or context for translation initiation. The structures of CITEs and the mechanisms by which they operate are diverse and fall into several classes. This has been well reviewed recently ⁵¹ ; therefore, we present just a few examples.

CITEs generally eschew the use of proteins to achieve long-range communication, relying instead on direct RNA-RNA base-pairing between distal sequences. This streamlined approach often consists of a hairpin structure in the 3′ UTR that is complementary to four to eight bases in another short hairpin structure in the 5′ UTR (reviewed in 52). One of the best-characterized examples is from the non-capped and non-polyadenylated Barley yellow dwarf virus (BYDV), whose 3′ UTR contains a proposed cruciform-like secondary structure referred to as a BYDV-like element (BTE). One stem-loop of the BTE can base-pair to five nucleotides in a short 5′ UTR hairpin, creating a “kissing” interaction ⁵³ . Because the BTE binds to eIF4G directly, this provides a way to deliver eIF4G to the 5′ UTR ^{54, 55} and also possibly binds directly to the 18S rRNA through base-pairing to recruit the ribosome to the 5′ end ( Figure 3) ⁵⁶ . In fact, direct RNA-ribosome interaction is an emerging mechanism in viral and cellular mRNA translation ⁵⁷ . Another class of CITEs are Panicum mosaic virus-like translational enhancers (PTEs), which functionally replace a cap structure by directly binding eIF4E ^{58– 60} . Although the high-resolution structure of a PTE-eIF4E complex has not been solved, it has been proposed that these RNAs form a compact fold in which a conserved guanine base is extruded from the structure to be recognized by the initiation factor ⁶¹ . Consistent with the trend of many 3′ CITEs, there are sequences in the PTE that are complementary to sequences in the 5′ end, and these are likely part of the mechanism for bringing eIF4E to the 5′ end to be used in translation initiation.

The CITEs mentioned above have a fairly well-established strategy to link their 5′ and 3′ ends, but the mechanism for other CITEs is less straightforward. For example, the T-shaped structure (TSS) CITEs found in the genome of several viruses, including Turnip crinkle virus, are proposed to fold into a three-dimensional structure that resembles a tRNA ^{62, 63} which binds to 80S ribosomes and 60S subunits ⁶⁴ . Unlike the case in many other CITEs, there is no clear sequence complementarity between the TSS and sequence in the 5′ end; the current model is that 5′-to-3′ communication occurs through an unusual ribosome subunit-subunit interaction of the 3′ bound 60S subunit with a 40S subunit bound to the 5′ end ⁶⁵ . Clearly, the details of these interactions deserve continued exploration.