The malignant phenotype is largely the consequence of dysregulated gene expression. Transformed cells depend upon not just a global increase in protein synthesis but an altered translational landscape in which pro-oncogenic mRNAs are translationally up-regulated. Such mRNAs have been shown to possess longer and more structured 5′-UTRs requiring high levels of eukaryotic initiation factor 4A (eIF4A) helicase activity for efficient translation. As such there is a developing focus on targeting eIF4A as a cancer therapy. In order for such treatments to be successful, we must develop a detailed understanding of the mechanisms which make specific mRNAs more dependent on eIF4A activity than others. It is also crucial to fully characterize the potentially distinct roles of eIF4A1 and eIF4A2, which until recently were thought to be functionally interchangeable. This review will highlight the recent advances made in this field that address these issues.
Dysregulation of mRNA translation is a hallmark of many cancers and it must be better understood if it is to become a viable therapeutic target. Increased cellular proliferation demands an increase in both global protein production and synthesis of specific influential pro-proliferative proteins. Translation initiation is the step at which most control is exerted and is the focus for most studies of translational dysregulation in cancer [1,2]. It has been suggested that malignant cells become ‘addicted’ to this altered translational landscape .
Translation initiation is facilitated by the eukaryotic initiation factor (eIF) 4F heterotrimeric complex (Figure 1). The eIF4F complex contains the cap-binding protein eIF4E, the scaffold protein eIF4G and the ATP-dependent DEAD box RNA helicase eIF4A. Before ribosome recruitment, eIF4A is delivered to the m7G cap structure of mRNAs as a subunit of eIF4F; unwinding of mRNA secondary structure by eIF4A then allows the 43S pre-initiation complex (40S ribosome and associated factors) to bind. The 5′-UTR is then scanned by the 43S complex, which also requires eIF4A to unwind secondary structures and upon recognition of the initiation codon the 60S subunit joins allowing elongation to proceed .
The role of eIF4A during translation initiation
eIF4A isoforms and their regulation
There are two mammalian isoforms of eIF4A involved in translation: eIF4A1 and eIF4A2 . These are highly conserved and murine homologues share 91% identity at the amino acid level . In an in vitro assay, both eIF4A1 and eIF4A2 are able to assemble into the eIF4F complex , which is also seen when these proteins are overexpressed in HeLa cells . A third isoform of eIF4A, eIF4A3, has a role in nonsense-mediated decay as a component of the exon junction complex .
Expression of eIF4A1 and eIF4A2 varies between different types of tissue, but eIF4A1 is generally more abundant than eIF4A2 . eIF4A1 is expressed at higher levels in dividing cells, whereas eIF4A2 is up-regulated in growth-arrested states . The expression of eIF4A1 is up-regulated in a range of malignancies [11–14] (Table 1); in contrast, high eIF4A2 levels in human tumours are associated with better outcome [15,16]. This variable requirement of the two isoforms suggests differences in their roles in proliferation.
eIF4A1 suppression leads to an increase in eIF4A2 transcription; however, eIF4A2 is unable to rescue the inhibition of translation or cellular proliferation resulting from eIF4A1 suppression . The transcription of eIF4A1, but not eIF4A2, is driven by the oncogenic protein MYC  and eIF4A1 and eIF4A2 also show distinct and opposite regulation during the differentiation of muscle cells . The amount of free functional eIF4A1 is regulated by a tumour suppressor product, programmed cell death 4 (PDCD4), the abundance of which is itself regulated by the mammalian target of rapamycin (mTOR) and the pro-oncogenic miR-21; however, it is not clear if the inhibitory effect of PDCD4 also affects eIF4A2 [20,21].
These divergent relationships with cell behaviour and clinical outcomes and distinct patterns of regulation suggest distinct biological roles for eIF4A1 and eIF4A2. Recently, it has been shown that miRNA-mediated gene regulation is dependent upon eIF4A2 activity but not eIF4A1 . This exciting finding offers a possible explanation for this diversity in roles and control; eIF4A1 may be truly pro-proliferative by its effects upon translation initiation, whereas eIF4A2 might have more unpredictable but generally anti-proliferative effects via global miRNA functions.
Targeting eIF4A as a therapeutic modality
The recent observation that increased eIF4A1 activity can drive the malignant phenotype  suggests that mRNAs that are most dependent on eIF4A1 activity will be involved in cellular transformation. Indeed, this and other genome-wide studies analysing the eIF4A-dependent translatome have shown that eIF4A-dependent mRNAs include those that contribute to cell proliferation, cell cycle progression, cell survival and angiogenesis [14,23,24]. Unsurprisingly, these mRNAs were shown to contain longer 5′-UTRs with a greater degree of predicted secondary structure (Table 2). Thus, the effect of eIF4A up-regulation upon transformed cells appears to act via specific messages, perhaps in addition to a global up-regulation of translation, making eIF4A an attractive target for therapeutic intervention. Several natural compounds have been characterized that inhibit cap-dependent translation by specifically inhibiting eIF4A activity. All appear to act non-specifically upon eIF4A1 and eIF4A2.
Hippuristanol is isolated from the coral Isis hippuris . It inhibits activity of both eIF4A1 and eIF4A2 without affecting the activity of helicases outside of the eIF4A family . By binding to the C-terminal domain of eIF4A, directly adjacent to the ATP-binding site and distant from the RNA-binding site, it acts as an allosteric inhibitor of eIF4A . Recently, Sun et al.  have demonstrated that it blocks eIF4A1 helicase activity by locking it into a ‘closed’ conformation. Hippuristanol has not only been shown to suppress tumour growth in a mouse model of T-cell leukaemia, but also chemosensitizes Myc-driven tumours to DNA damaging agents in the Eμ-Myc mouse model [28,29].
Silvestrol belongs to a class of compounds called rocaglamides isolated from species of the Aglaia genus of plants . Silvestrol induces binding of eIF4A to mRNA in a sequence-independent manner and has been shown to target the activity of eIF4A both in vitro and in vivo [23,28]. Rocaglamides also bind to prohibitin 1 and 2, which inhibits CRaf/MEK/ERK/Mnk1 signalling pathway, making it difficult to relate the activity of these compounds to direct inhibition of eIF4A . Silvestrol has been shown to significantly inhibit synthesis of proteins that drive tumour growth such as CCND1, BCL2, MYC and NOTCH in tumour cell lines [23,24].
Pateamine A is a metabolite isolated from the sea sponge Mycale hentscheli and was first characterized as a selective agent against tumour cells . It was later characterized as an inhibitor of cap-dependent translation initiation that specifically targets eIF4A [33,34]. Pateamine A increases the affinity of eIF4A for ATP and RNA by inducing dimerization, which allows non-specific interaction of eIF4A–RNA . As a result, it sequesters eIF4A from the eIF4F complex . The anti-tumour activity of Pateamine A and its simplified analogue has been demonstrated in human cancer cell lines  as well as in xenograft models  respectively.
eIF4A, therefore, is tractable to small-molecule inhibition and this is a promising avenue for novel cancer therapies. However, the non-specific inhibition of eIF4A1 and eIF4A2 by existing agents may prove to be disadvantageous, especially if eIF4A2’s putative anti-proliferative/anti-malignant properties are substantiated.
Which mRNAs are eIF4A dependent?
Although these eIF4A inhibitors and other anti-eIF4A approaches have potential as cancer therapies, their full biological affect can only be understood when we have a full understanding of which mRNAs are dependent upon eIF4A activity for their translation. The helicase activity of eIF4A (generally meaning eIF4A1) has been shown to be vital to overcome classical Watson–Crick base-paired secondary structure within 5′-UTRs [37,38]; however, additional structured elements have recently been described that may also be important (Figure 1).
A role of eIF4A activity in unwinding G-quadruplex structures within 5′-UTRs
G-quadruplexes are four-stranded structures comprised of stacks of Hoogsteen hydrogen bonded G-tetrads and can form within G-rich regions of both DNA and RNA. There is growing interest in the role of RNA G-quadruplexes in RNA biology as they have been implicated in splicing, polyadenylation, mRNA stability and translation [39,40]. Several mRNAs have been shown experimentally to contain G-quadruplexes within their 5′-UTRs, including genes of clinical interest  such as the known oncogenes NRAS  and BCL2  and G-quadruplex structures have been shown to inhibit cap-dependent translation in cellular reporter systems [43–45]. Putative quadruplex-forming sequences (PQS) were shown to be enriched within 5′-UTRs of human genes compared with the whole transcriptome and this enrichment was strongest towards the 5′-end .
Recent studies have globally identified the mRNAs requiring eIF4A function for their efficient translation [14,23,24] (Table 2). In addition to the expected findings that eIF4A-dependent mRNAs contained longer 5′-UTRs with a predicted greater degree of secondary structure [14,23,24], both Modelska et al.  and Wolfe et al.  also found 5′-UTRs of eIF4A-dependent mRNAs to be enriched for motifs with G-quadruplex forming potential. For example, Wolfe et al.  described an enrichment of a 12 nt (CGG)4 motif, which they showed by circular dichroism could form a G-quadruplex structure in K+ buffer . Modelska et al.  reported an enrichment of a remarkably similar GC(GGC)3G motif, which within the right context also has G-quadruplex forming potential. Indeed PQS were also shown to be enriched within the 5′-UTRs of eIF4A1-dependent mRNAs in this study.
In contrast with DNA G-quadruplexes, RNA quadruplexes form exclusively parallel structures and, in general, are more stable than their DNA counterparts [47,48]. The single stranded nature of RNA is also thought to increase the likelihood of these structures forming in vivo. Increasing the number of G-tetrads [47,49,50] and shortening loop length [45,48] leads to the formation of more stable structures with a lower dependence on monovalent cations such as K+ . Unsurprisingly, the predicted prevalence of PQS within 5′-UTRs depends largely on the sequence parameters imposed on the computational search. Initial bioinformatic analysis whereby a PQS was defined as G3+-N1–7-G3+-N1–7-G3+-N1–7-G3+, where the number of G-tetrads is three or more and loop length is 1–7 nts, identified 2034 5′-UTRs (6.2% of those tested) that contained one or more PQS in the human genome . Unsurprisingly, if these parameters are relaxed to allow PQS with only two stacks of G-tetrads and/or longer loop lengths, many more PQS would be identified. However, although G-quadruplexes with just two stacks of G-tetrads are able to form under the correct conditions [24,49,50], the structural preferences of CGG repeats remains under debate. Malgowska et al.  investigated G-quadruplex formation for CGG, AGG and UGG trinucleotide repeats in various conditions using multiple techniques. Although all molecules tested were able to form intermolecular G-quadruplex structures, those formed from CGG repeats were shown to compete with duplex or hairpin formation. Furthermore, G(CGG)2C oligos were shown to form G-quadruplexes in K+ buffer, yet in HeLa cell extracts duplex formation was shown to occur almost exclusively . This polymorphic nature of CGG repeats could therefore be a crucial layer of regulation, allowing these structures to only form in certain conditions such as cellular stress. For example, Mullen et al. [50,52] propose that in plants, these two stacked G-quadruplex structures may only form under conditions of drought, in which levels of intracellular K+ ions increase.
It is therefore interesting that only motifs composed of (CGG)n repeats rather than (AGG)n or (UGG)n repeats (or PQS for triple stacked G-quadruplexes) were found to be enriched in the 5′-UTRs of eIF4A dependent mRNAs [14,24]. A regulatory role of CGG repeats for translation is supported by the fact the CGG repeats are over-represented in 5′-UTRs . In addition to the properties of the PQS, the adjacent sequence also seems to be of importance, in that increased cytosine content is able to favour stem loop structures in direct competition with G-quadruplex formation . Furthermore, concentrations of K+ and Mg2+ have been shown to be crucial in determining which of these two structures is the predominant form . It will therefore be crucial to determine whether these motifs do indeed form G-quadruplex structures under cellular conditions and, if not, why these CGG repeat motifs confer eIF4A dependence.
Does eIF4A have a role in unwinding mRNA–rRNA duplexes?
It is clear from the substantial body of work investigating the role of eIF4A during translation that it is necessary to unwind internal Watson–Crick interactions within the 5′-UTR of mRNA for efficient translation initiation to occur and recent evidence has also implicated eIF4A activity in unwinding G-quadruplex structures. We would also like to postulate an additional role of eIF4A in the unwinding of mRNA–18S rRNA hybrids during scanning (Figure 1C). The concept of direct base pairing between sequences of mRNA and rRNA is not new and was first shown experimentally over 20 years ago . Work by Mauro and Edelman, who termed the ‘ribosome filter’ hypothesis , has provided the bulk of evidence to support the concept that mRNA-rRNA duplex formation is able to occur in vivo. The majority of this work has focused on the 9-nt element found in the Gtx mRNA, which has been shown to promote translation initiation via its interaction with 18S rRNA [57,58].
A recent bioinformatic study demonstrated an evolutionary conserved pattern of complementarity between 18S rRNA and mRNA 5′-UTRs , providing weight to the argument that such interactions could be biologically significant. Furthermore, the rRNA regions showing significant complementarity to 5′-UTRs were found to be clustered within the evolutionarily recent expansion segments of 18S rRNA, which are exposed on the surface of the ribosome. This would also provide a gratifying evolutionary hypothesis to explain the role of the expansion regions for translational regulation. Panek et al.  hypothesize that interactions between mRNA exiting the scanning 40S ribosome form hybrid structures with complementary sequences in the 18S rRNA expansion regions and that these ‘sticky elements’ could impede scanning during translation initiation. eIF4A activity could therefore be required to unwind these duplexes during scanning and we feel this certainly deserves further investigation.
The emerging theme that eIF4A1 and eIF4A2 could have distinct biological roles during translation has not only challenged the previously accepted notion of their interchangeable roles, but also raised questions about the way eIF4A activity could be targeted for cancer therapy. The contribution of these two isoforms in regulating the translation of eIF4A-dependent mRNAs needs to be properly understood and in the context of cellular transformation. In accordance, the differential functions of eIF4A1 and eIF4A2 in translation need to be further probed to better understand the regulation of translation and how this affects cellular growth and survival.
The advancement of next-generation sequencing technology has allowed the eIF4A-dependent translatome to be determined for the first time. These genome-wide studies have clarified that length and secondary structure are important determinants of eIF4A dependency. However, these studies also discovered that sequence motifs with G-quadruplex forming potential were enriched in eIF4A-dependent 5′-UTRs, suggesting that G-quadruplex structures could confer an additional layer of regulation. This suggests that perhaps there could be more to eIF4A dependency than simple Watson–Crick base-paired structures and we believe this warrants further investigation.
This work was supported by JLQs MRC programme funding.
Translation UK 2015: Held at the University of Aberdeen, U.K., 7–9 July 2015.
1These authors contributed equally to the present work.