New frontiers in translational control of the cancer genome

Abstract

The past several years have seen dramatic leaps in our understanding of how gene expression is rewired at the translation level during tumorigenesis to support the transformed phenotype. This work has been driven by an explosion in technological advances and is revealing previously unimagined regulatory mechanisms that dictate functional expression of the cancer genome. In this Review we discuss emerging trends and exciting new discoveries that reveal how this translational circuitry contributes to specific aspects of tumorigenesis and cancer cell function, with a particular focus on recent insights into the role of translational control in the adaptive response to oncogenic stress conditions.

Alterations in translational control reflect a powerful means of adjusting the development^1–5, maintenance^6–9 and overall fitness^10–13 of cancer cells. Seminal studies have revealed that nearly all major oncogenic signalling pathways found to be deregulated in human cancers — including RAS–MAPK, PI3K–AKT–mTOR, MYC and WNT–β-catenin — ultimately lead to reprogramming of the genome at the translation level^1–5,10 (FIG. 1a). Oncogenic signalling pathways enhance translation initiation largely through stimulation of the eukaryotic translation initiation factor 4F (eIF4F) complex, either directly by altering the expression and phosphorylation of eIF4F complex members^3,14–17 or indirectly through regulation of eIF4F complex formation and activity^4,9,18–22 (FIG. 1a). In particular, the eIF4E initiation factor has emerged as a crucial node of translational control (BOX 1) that is hyperactivated downstream of oncogenic signalling pathways through enhanced transcription^16,17, via phosphorylation downstream of the MAPK-interacting serine/threonine kinases (MNKs)^3,14,15 and by mTOR-dependent inactivation of its negative regulators, the tumour-suppressive eIF4E-binding proteins (4EBPs)^{4,9,18–20,23}. Although most translational control is typically attributed to modulations at the level of initiation, substantial regulation also occurs at the stage of translation elongation, and recent work suggests that oncogenic activation of eukaryotic translation elongation factor 2 (eEF2)-dependent translation elongation may be crucial for some cancers^5,24 (BOX 1; FIG. 1a).

Figure 1. Oncogenic activation of mRNA translation.

a | A central output of oncogenic MYC, RAS–MAPK, PI3K–AKT–mTOR and WNT–β-catenin signalling pathways is the aberrant activation of mRNA translation at the initiation and elongation steps. Translation initiation, the first step in this process, is considered the primary rate-limiting step of protein synthesis and typically proceeds in a ‘cap-dependent’ manner that relies on the ability of the eukaryotic translation initiation factor 4F (eIF4F) complex to bind to the 5′ 7-methylguanosine cap present on mature mRNAs⁶². Oncogenic signalling promotes translation initiation predominantly through alterations in the eIF4F complex, which comprises the major cap-binding protein eIF4E, the scaffolding protein eIF4G and the helicase eIF4A. The eIF4F complex drives translation initiation through the ability of eIF4E to bind to the 5′ cap and interact with eIF4G, which recruits the 43S ribosomal pre-initiation complex (comprising a 40S ribosomal subunit, the eIF2–GTP–Met-tRNA_i^Met ternary complex, eIF3 and several additional accessory factors). Oncogenic signalling can hyperactivate eIF4E through enhanced transcription^16,17, through phosphorylation of eIF4E at serine 209 by the MAPK-interacting serine/threonine kinases (MNKs)^3,14,15 and through mTOR complex 1 (mTORC1)-dependent phosphorylation and inactivation of the eIF4E inhibitors eIF4E-binding proteins (4EBPs)^{4,9,18–20,23}. Upon recruitment of the 43S complex to the 5′ untranslated region of mRNA, it scans in the 5′ to 3′ direction until reaching a start codon, a process facilitated by eIF4A helicase unwinding of secondary structures and promoted by ribosomal protein S6 kinase (S6K)-dependent stimulation of eIF4A activity through inhibition of programmed cell death protein 4 (PDCD4) and activation of eIF2B^21,22. Start codon recognition by the 43S complex is followed by GTP hydrolysis within the ternary complex and joining of the 60S ribosomal subunit to form a translationally competent ribosome. Translation can also be regulated at the elongation stage by oncogenic signalling, largely through S6K-dependent inhibition of eukaryotic translation elongation factor 2 (eEF2) kinase (eEF2K)^24,232,233. Phosphorylation of eEF2K by S6K relieves its suppression of eEF2 (REF. 205), promoting the codon by codon translocation of the ribosome along the mRNA. Dashed arrows indicate indirect activation. b | Oncogenic activation of translation initiation and elongation supports tumorigenesis in part by driving selective changes in the translation of specific mRNA transcripts independently of alterations in transcript levels or global increases in protein synthesis. This selective translational control of specific mRNA transcripts underlies the acquisition and execution of distinct cancer cell behaviours central to the transformed phenotype, such as increased cell growth and proliferation⁵, altered metabolism²⁸, enhanced angiogenesis¹⁷⁵, proper reactive oxygen species (ROS) control¹⁰, immune cell recruitment³, and invasion and metastasis^15,26. CCL, C-C motif chemokine ligand; CCND3, cyclin D3; FTH1, ferritin heavy polypeptide 1; GCLC, glutamate–cysteine ligase catalytic subunit; MTA1, metastasis-associated 1; PRPS2, phosphoribosyl pyrophosphate synthetase 2; RAPTOR, regulatory associated protein of mTORC1; VEGFA, vascular endothelial growth factor A; YB1, Y-box binding protein 1.

Box 1. New insights into translation initiation and elongation in cancer.

Translation initiation is typically considered the primary rate-limiting step of protein synthesis and usually proceeds in a cap-dependent manner that relies on the ability of the eukaryotic initiation factor 4F (eIF4F) initiation complex to bind to the 5′ 7-methylguanosine cap on mature mRNAs⁶². Historically, eIF4E was identified as the quantitatively limiting factor in the eIF4F complex^200,201. As such, even modest increases or decreases in eIF4E expression were expected to profoundly influence mRNA translation and cellular function. In line with this, seminal studies revealed that overexpression of eIF4E is sufficient to drive transformation in cell lines^202,203 and spontaneous tumorigenesis in mice²⁰⁴. However, the recent generation of an eIF4E-haploinsufficient mouse has created a paradigm shift in our understanding of the role of eIF4E levels in development and tumorigenesis¹⁰. Unexpectedly, eIF4E-haploinsufficient mice were found to be physiologically normal, yet strikingly resistant to tumour formation. This study further revealed that eIF4E is present at levels that exceed those required for normal translational control and development and instead become limiting specifically for expression of the ‘oncogenic translation programme’. These findings challenge the nearly 30-year-old dogma that eIF4E dose is limiting for normal protein synthesis and cellular homeostasis and further delineate how crucial eIF4E expression levels are for cancer development.

Recent studies suggest that translation elongation, like translation initiation, can also be a crucial node of hyperactivation downstream of oncogenic signalling in cancer cells²⁴. Translation elongation is regulated largely at the level of eukaryotic translation elongation factor 2 (eEF2) kinase (eEF2K), which blocks ribosomal translocation along the mRNA by phosphorylating and inhibiting eEF2 (REF. 205). Exciting new work has revealed that inhibition of eEF2K is essential for cellular proliferation in a mouse model of intestinal tumorigenesis driven by oncogenic WNT signalling⁵. In particular, this study demonstrated that the ability to enhance translation elongation was a crucial output of mTOR complex 1 (mTORC1) signalling, which is predominantly thought to exert translational control at the level of translation initiation. Paradoxically, although this work demonstrates a pro-tumorigenic role for the inhibition of eEF2K activity, certain cancers instead show increased eEF2K activity^206,207. This is likely to be due to the crucial role that eEF2K plays in adapting to cellular stress conditions frequently found in tumours (see the section on translational adaption to stress).

This link between translational control and oncogenic signalling has been echoed by an increased appreciation, spanning both basic and translational biology, of a great disconnect between protein abundance and transcriptional activity within cells (reviewed in REF. 25). In this regard, translational control has emerged not merely as a layer of ‘fine-tuning’ but also as a central regulator of gene expression in cancer. However, a crucial question has been to what extent oncogenic signalling pathways modulate protein synthesis in order to cause cellular transformation. Key insights have come from pioneering studies using mice with spontaneous mutations that decrease overall levels of protein synthesis and thus serve as genetic tools to dampen elevated protein synthesis rates in mouse models of cancer development^2,4. This work elegantly demonstrated that genetically constraining the ability of oncogenic MYC or AKT signalling to enhance global protein synthesis rates is sufficient to block cell growth, proliferation and tumorigenic potential in mouse models of lymphomagenesis.

Although global translation rates are generally enhanced in cancer cells, it has become increasingly clear that oncogenic signalling also induces transcript-specific changes in translation^{2,4–7,10,26}. For example, the acute inhibition of KRAS and AKT signalling in glioblastoma cells has been shown to cause a rapid decrease in the translation of specific subsets of mRNAs²⁷. Interestingly, no major changes in total transcript levels were found at these early time points, highlighting that altered translational control is a fundamental and immediate response to oncogenic signalling that may precede changes in transcriptional control.

Accumulating data are revealing that selective mRNA translation has a key role in steering distinct aspects of the transformed phenotype. Indeed, recent studies using unbiased translational profiling techniques have identified large groups of functional gene classes that are translationally regulated downstream of oncogenic signalling pathways^6,7,10,26. This work has demonstrated that oncogenic regulation of translation supports not only cellular proliferation and growth⁵, phenotypes strongly associated with increased global protein synthesis rates, but also specific gene expression programmes that drive a diverse array of cancer cell behaviours, including altered cellular metabolism²⁸, metastasis^15,26 and resistance to oxidative stress¹⁰ (FIG. 1b).

In this Review, we highlight exciting new studies that have begun to unravel the molecular mechanisms underlying translational specificity in expression of the cancer genome and describe an expanding code of cis-regulatory elements that govern post-transcriptional expression of key mRNAs deregulated in cancer. In particular, we discuss regulatory elements embedded in mRNA untranslated regions (UTRs), including both structural and sequence-specific motifs, that have been identified as crucial mediators of translational control in cancer. In addition, we review how codon usage choice and tRNA expression have been shown to be highly biased and distinct for genes with different functions and expression levels in transformed cells. Importantly, although it touches on many different and new aspects of mRNA translation, this Review is not intended to provide a comprehensive or historical account of translational control in cancer, as several excellent reviews have already been published on the subject^29–31. Instead, it focuses on recent insights into the mechanistic basis by which cancer cells co-opt specific translational networks that support tumorigenesis, discusses emerging roles for translation in mediating oncogenic cellular stress responses and highlights technological advances that have progressed our understanding of how the landscape of mRNA translation is specifically tuned in cancer cells.

Structural RNA regulatory elements

Some of the first insights into selective mRNA translation came from classic studies demonstrating that the presence of complex secondary structures within the 5′UTR inhibits mRNA translation, which provided a conceptual framework for the mechanistic basis of selective mRNA translation in cancer^32,33. Subsequent studies revealed that the overexpression of eIF4E was sufficient to overcome translational limitations imposed by 5′UTR secondary structures³⁴, mainly through the ability of eIF4E to recruit the eIF4A helicase as part of the eIF4F complex and unwind these structures^35–38 (FIG. 2a). Indeed, a large body of work has demonstrated a key role for eIF4A in promoting the translation of mRNAs with structured 5′UTRs and uncovered many underlying principles of eIF4A regulation and activity (reviewed in REF. 39). Importantly, complex 5′UTR structures have been identified in several pro-tumorigenic mRNAs that are translationally induced in an eIF4E-dependent manner, including the metabolic enzyme ornithine decarboxylase (ODC), as well as fibroblast growth factor 2 (FGF2) and the pro-angiogenic factor vascular endothelial growth factor A (VEGFA)^40–42. In line with this, pharmacological inhibitors of eIF4A, such as silvestrol, display potent antitumour activity in preclinical mouse models of cancer^6,43–45. However, the full impact of 5′UTR structures on translation of the cancer genome has remained largely unresolved.

Figure 2. mRNA regulatory elements direct specialized translation of the oncogenic programme.

Structural and sequence-specific regulatory elements underlie the code for selective control of the translation of specific pro-tumorigenic mRNAs. a | General secondary structure in the 5′ untranslated region (5′UTR) can steer selective translation of mRNAs, such as the metabolic enzyme ornithine decarboxylase (ODC)⁴⁰, by conferring enhanced sensitivity to eukaryotic initiation factor 4F (eIF4F) complex formation, which promotes unwinding of these structures through the eIF4A helicase. b | Likewise, mRNAs with G-quadruplexes in their 5′UTR, such as the super-enhancer-associated transcription factor ETS1, selectively require eIF4A helicase activity for their translation⁶. c | Transcripts containing a structural element called an internal ribosome entry site (IRES), such as the anti-apoptotic BCL2 (REF. 50), can be selectively translated in a cap-independent manner by directly recruiting the ribosome to the 5′UTR, frequently with the aid of IRES trans-acting factors (ITAFs). d | Many structural elements confer translational specificity through interactions with distinct RNA-binding proteins (RBPs). For example, the epithelial–mesenchymal transition regulator disabled homologue 2 (DAB2) contains a 3′UTR transforming growth factor-β (TGFβ)-activated translation (BAT) element that recruits RBPs to form a complex that blocks translation elongation^59,60. e | Structural elements in the 5′UTR have even been shown to interact with eIF3, which non-canonically binds RNA in a manner dependent on the presence of distinct stem–loops. For example, direct interaction of eIF3 with a stem–loop in the 5′UTR of the mRNA encoding the mitotic regulator JUN promotes its translation⁶¹. f | One of the shortest sequence-specific elements in the 5′UTR is an alternative translation start site, typically comprising an upstream AUG (uAUG) or, more rarely, a non-AUG codon. These upstream start codons are in-frame with the primary open reading frame (ORF) and can drive the expression of protein isoforms with novel functions, such as the long form of the tumour suppressor PTEN (PTEN-long), the translation of which starts from an upstream CUG codon⁶⁴. g | Upstream ORFs (uORFs) are discontiguous from the primary ORF and typically act to inhibit translation, as has been shown for the growth factor receptor HER2 (REF. 74). However, under certain conditions uORFs can act to promote translation (see FIG. 4). h | MicroRNAs (miRNAs) recognize complementary sequences in target mRNAs and guide transcript-specific translational repression as part of the miRNA-induced silencing complex (miRISC), as has been shown for let-7 regulation of the RAS oncogenes⁹⁸. Cancer cells can reduce their 3′UTR length through alternative cleavage and polyadenylation (APA) to avoid miRNA-mediated regulation⁸⁵. i | Many RBPs recognize sequence-specific RBP-binding domains (RBDs) in the 3′UTR to regulate transcript-specific translation, as has been shown for the RBP Musashi homologue 2-induced translation of the self-renewal factor Ikaros family zinc finger protein 2 (IKZF2)¹¹⁰. j | There is recent and growing appreciation of the presence of numerous sequence-specific elements in the 5′UTR, including the translation inhibitory element (TIE)¹¹⁸, pyrimidine-rich translational element (PRTE)^26,28 and cytosine-enriched regulator of translation (CERT)¹⁰, all of which can direct selective cap-dependent translation of distinct transcripts that influence tumour development and progression. Although the underlying molecular mechanisms by which these elements exert translational control remains to be elucidated, they may function in part by recruiting specific trans-acting factors (light blue circles labelled ‘?’) that interact directly or indirectly with the translation initiation complex. FTH1, ferritin heavy polypeptide 1; PRPS2, phosphoribosyl pyrophosphate synthetase 2.

Recent progress in understanding how 5′UTR structure influences genome-wide translation in cancer has been made by ribosome-profiling experiments surveying the effects of eIF4A inhibition by silvestrol in tumour cells^6,7. Strikingly, this work has not only validated the well-established ability of eIF4A to direct the translation of mRNAs with generally complex 5′UTR secondary structures but has also identified a role for smaller structural domains of 9–12 nucleotides, including the G-quadruplex, in conferring sensitivity to eIF4A helicase activity (FIG. 2b). Intriguingly, eIF4A inhibition in breast cancer cells was largely found to affect the translation of cell cycle, cell morphology and cell survival genes, whereas T cell acute lymphoblastic leukaemia (T-ALL) cells displayed reduced translation of mRNAs functionally enriched for transcription factors, oncogenes and super-enhancer-associated genes^6,7. Although the broader functional role of the G-quadruplex and related structural motifs remain to be validated, these findings suggest that smaller structural domains within 5′UTRs may also steer specific cancer cell behaviours.

One of the most extensively studied 5′UTR structural elements able to direct transcript-specific translation is the internal ribosome entry site (IRES). IRES-dependent translation is thought to fine-tune gene expression by promoting selective cap-independent mRNA translation (FIG. 2c), particularly during distinct cellular conditions in which cap-dependent translation is inhibited, such as during mitosis⁴⁶, megakaryocyte differentiation⁴⁷ and cellular stresses such as DNA damage⁴⁸. A role for IRES-dependent translation in cancer has been best characterized for tumour-associated stress conditions such as hypoxia (discussed further below); however, several studies suggest a more complex role for IRES-dependent translation during cancer. For example, IRES-dependent translation has been shown to drive the expression of proteins with contrasting cellular functions, such as the pro-apoptotic protein apoptotic peptidase-activating factor 1 (APAF1)⁴⁹ and the anti-apoptotic protein BCL-2 (REF. 50), as well as oncoproteins such as MYC⁵¹ and tumour suppressors such as p53 and p27 (REFS 52–54). Although approximately 10% of all mRNAs have been predicted to contain IRES elements⁵⁵, the functional role of IRES elements in many cellular mRNAs remains to be studied. Moreover, the impact of IRES-driven translation on gene expression in cancer is largely unknown. The potential diversity of cellular functions regulated by IRES-dependent translation may reflect a role for distinct structural features of the IRES as well as associated IRES trans-acting factors (ITAFs) in directing the expression of specific mRNA classes. New technologies to study these structural features (for example, selective 2-hydroxyl acylation analysed by primer extension (SHAPE) in the case of IRES and photoactivatable-ribonucleoside-enhanced crosslinking and immunoprecipitation (PAR-CLIP) for ITAFs; TABLE 1) are likely to fuel future progress in understanding their role during tumorigenesis.

Table 1.

Technologies for studying post-transcriptional control of the cancer genome

Technology	Variations	Description
Translational profiling	Polysome microarray	Translational profiling that uses classic microarray technology to analyse mRNAs associated with actively translating ribosomes isolated by sucrose gradient fraction234
	Ribosome profiling	Deep sequencing of ribosome-protected mRNA fragments as a genome-wide measure of translation with subcodon resolution235
	Proximity-specific ribosome profiling	Ribosome profiling using a spatially restricted biotin ligase and biotin-acceptor-tagged ribosomes as a strategy to look at proximity-specific translation236
RNA–protein interaction network analysis	Crosslinking and immunoprecipitation followed by high-throughput sequencing (CLIP-Seq)	Also known as high-throughput sequencing of RNA isolated by crosslinking immunoprecipitation (HITS-CLIP), a method that uses high-throughput sequencing of RNAs isolated by UV crosslinking and immunoprecipitation to map genome-wide binding sites of a given protein within a 30-nucleotide region of RNA237
	Photoactivatable ribonucleoside-enhanced crosslinking and immunoprecipitation (PAR-CLIP)	An alternative to CLIP-Seq that uses PAR analogues, such as 4-thiouridine, added to cell culture to enable enhanced crosslinking and precise mapping of RBP-binding sites by scoring for crosslinking-induced thymidine to cytidine transitions238
	Individual-nucleotide-resolution UV crosslinking and immunoprecipitation (iCLIP)	An adaptation of CLIP-Seq that enables improved individual-nucleotide resolution by using an adaptor that exploits the tendency of cDNAs to prematurely truncate before the crosslinked nucleotide239
	RBP interactome capture	A technique using either UV or PAR crosslinking of RNA–protein interactions followed by oligo(dT) purification of mRNAs and mass spectrometry (MS) to identify the full repertoire of RBPs within a cell113
	Comprehensive identification of RBPs by MS (ChIRP-MS)	A method for identifying the specific subset of proteins associated with a particular RNA using formaldehyde crosslinking of protein–RNA interactions, followed by RNA purification using tiled biotinylated probes specific for the RNA of interest, and MS to identify bound proteins240
	RNA antisense purification with MS (RAP-MS)	An approach similar to ChIRP-MS that uses UV crosslinking and stable isotope labelling by amino acids in culture (SILAC) to perform quantitative MS of RNA–protein interactions241
RNA–RNA interaction network analysis	Crosslinking, ligation and sequencing of hybrids (CLASH)	A high-throughput method to identify RNA–RNA interactions associated with mRNPs based on purification of UV-crosslinked complexes, intra- and intermolecular ligation of RNA–RNA interactions, and sequencing of hybrid RNAs242
RNA structure analysis	Parallel analysis of RNA structure (PARS)	A method for analysing RNA structure genome-wide at single-nucleotide resolution by parallel deep sequencing of RNA fragments generated by treatment with various structure-specific endonucleases243
	Fragmentation sequencing (Frag-Seq)	Genome-wide analysis to identify RNA structure based on high-throughput sequencing of RNA fragments produced by the cleavage of single-stranded nucleic acids by nuclease P1 (REF. 244)
	Selective 2-hydroxyl acylation analysed by primer extension (SHAPE)	This standard analysis of RNA secondary structure, which uses primer extension to measure 2′-hydroxyl group reactivity with chemical probes, has recently been improved by the development of new probes enabling its application in living cells245
	SHAPE-Seq	A high-throughput version of SHAPE that enables genome-wide analysis of RNA structure by using multiplexed paired-end deep sequencing of primer extension products246
	In vivo click-SHAPE (icSHAPE)	An adaptation of SHAPE that combines genome-wide profiling with in vivo click selective 2′-hydroxyl acylation to overcome previous limitations in chemical probes that precluded analysis of all four nucleotides247
	RNA hybrid and individual-nucleotide-resolution UV CLIP (hiCLIP)	A modified iCLIP procedure that is similar to CLASH but which uses an adaptor in the ligation step enabling it to better resolve RNA–RNA interactions and predict secondary structure248
RNA modification profiling	m6A profiling	Techniques such as Me-RIPseq134 and m6A-Seq136 measure N6-methyladenosine modifications on RNA by high-throughput sequencing of transcripts captured using an antibody specific for the modification
	m5C profiling	The recent development of affinity purification-based high-throughput sequencing methods for profiling RNA 5-methylcytosine modifications — such as 5-azacytidine RNA immunoprecipitation (Aza-IP)249 and methylation individual-nucleotide-resolution CLIP (miCLIP)250 — provide some advantages over bisulfite sequencing approaches251, such as the enhanced ability to detect modification on less abundant RNAs
	Pseudo-Seq, PSI-Seq and Ψ-Seq	High-throughput sequencing methods for genome-wide identification of RNA pseudouridylation using CMC to modify pseudouridines and generate a block in reverse transcription137,139,140
	N3-CMC-enriched pseudouridine sequencing (Ceu-Seq)	An approach similar to Pseudo-Seq that uses a CMC derivative modified by click chemistry to add biotin, enabling the enrichment of pseudouridine-containing RNAs by biotin pulldown before sequencing138
	Quantitative MS	Optimized MS approaches have been developed for measuring pseudouridine levels at select rRNA sites226, for determining global changes in the full range of tRNA modifications252 and for evaluating global changes in nucleoside modifications across all species of RNA253
tRNA profiling	tRNA microarray	Chemical ligation-based microarrays have been developed for profiling tRNA isoacceptors at single-nucleotide resolution141, improving on previous hybridization-based microarrays that can only distinguish tRNAs differing by at least eight nucleotides150
	tRNA high-throughput sequencing	Recent adaptations of RNA-Seq such as: DM-tRNA-seq254 and ARM-seq162 have been developed to overcome limits in sequencing caused by RNA modifications and structure; these techniques not only enable quantitative analysis of tRNA expression but can also provide dynamic measurements of specific tRNA methylations

ARM-seq, AlkB-facilitated RNA methylation sequencing; CMC, N-cyclohexyl-N′-(2-morpholinoethyl)carbodiimide methyl-p-toluenesulfonate; DM-tRNA-seq, demethylase-thermostable group II intron RT tRNA sequencing; mRNP, messenger ribonucleoprotein; RBP, RNA-binding protein.

Another major class of UTR structural elements that can direct translational control in cancer comprises motifs recognized by RNA-binding proteins (RBPs)⁵⁶. Although most RBP recognition motifs are thought to be sequence specific⁵⁷ (discussed further below), structural RNA elements can also bind to RBPs. For example, the interferon-γ (IFNγ)-activated inhibitor of translation (GAIT) element is a structural RBP recognition motif that directs translational suppression of pro-inflammatory molecules in response to IFNγ signalling⁵⁸. Another recent example of a structural RBP motif implicated in tumorigenesis is the transforming growth factor-β (TGFβ)-activated translation (BAT) element, a 3′UTR stem–loop region with an asymmetrical bulge that mediates TGFβ-driven translation of select mRNA transcripts⁵⁹ (FIG. 2d). In the absence of TGFβ, the BAT element inhibits translation elongation by binding to both the eEF1A1 elongation factor and the RBP heterogeneous nuclear ribonucleoprotein E1 (hnRNPE1; also known as PCBP1) to form a messenger ribonucleoprotein (mRNP) complex that blocks translation elongation⁶⁰. TGFβ signalling relieves this translational suppression by promoting AKT2-dependent phosphorylation of hnRNPE1 on serine 43, which prevents the formation of the BAT–mRNP complex. Importantly, the BAT element seems to be crucial for selective translational control of specific mRNAs involved in TGFβ-induced epithelial–mesenchymal transition (EMT) and metastasis, such as disabled homologue 2 (DAB2) and interleukin-like EMT inducer (ILEI; also known as FAM3C), highlighting a crucial role for RBP-directed translation in steering specific cancer cell behaviours.

A newly expanding appreciation of the role of structural motifs in steering cancer-associated translation stems in part from technological advances in both transcript-specific and genome-wide analysis of mRNA structure as well as unbiased profiling of RNA–protein interactions (TABLE 1). For example, these technologies were recently used to reveal a novel functional role for eIF3 in interacting with structural domains in the 5′UTRs of select cancer-associated mRNAs⁶¹. eIF3 is a multimeric complex best known for the scaffolding-like role it plays in connecting the eIF4F and the 43S pre-initiation complexes⁶²; however, this function of eIF3 cannot explain why distinct eIF3 subunits are frequently overexpressed in human cancers²⁹. Using transcriptome-wide PAR-CLIP analysis to identify preferential interactions of eIF3 with specific subsets of mRNAs, this study uncovered eIF3-binding sites in approximately 3% of all expressed transcripts, predominantly in 5′UTRs⁶¹. In support of a putative role in cancer, eIF3-bound transcripts included mRNAs associated with cancer-relevant pathways, such as apoptosis, cell cycle and differentiation. SHAPE analysis on two eIF3 target mRNAs, the oncogene JUN and the tumour suppressor B cell translocation gene 1 (BTG1), showed that eIF3-binding sites map to distinct conserved stem–loop regions⁶¹ (FIG. 2e). Intriguingly, these two eIF3-binding structures were found to have opposing effects on mRNA translation, suggesting that other context-specific features of the 5′UTR may interact with eIF3 to determine translational outputs that may be important for cancer cell survival and proliferation.

Sequence-specific RNA elements

A growing realization of the discrepancy between mRNA transcript levels and protein expression has kindled interest in translational control mediated by sequence-specific elements in the 5′ and 3′ UTRs. Technological advances, such as genome-wide profiling of actively translating ribosomes at codon-by-codon resolution, are enabling the identification of an increasing number of sequence-specific elements enriched in distinct subsets of transcripts that control important cancer cell behaviours (FIG. 2).

Some of the shortest sequence-specific elements that can affect transcript-specific translation are upstream initiation codons⁶³. These can mark the presence of an alternative translation start (ATS) site that is in-frame and contiguous with the primary open reading frame (ORF) or can comprise an upstream open reading frame (uORF) that is distinct from the primary ORF. ATS sites can enable the expression of a unique protein isoform with different functions from those of the protein encoded by the primary ORF (FIG. 2f). For example, translation from a non-canonical CUG start codon in the PTEN 5′UTR was shown to drive expression of a long form of PTEN (with an additional N-terminal 173 amino acids) that is membrane permeable, can act in a non-cell-autonomous manner to inhibit PI3K signalling and is downregulated in samples from patients with breast cancer compared with matched normal breast tissue⁶⁴. Although the mechanisms underlying translation from upstream ATS sites remain poorly understood, these elements may be under selective control in cancer cells.

uORFs, instead, are typically thought to inhibit translation of the primary ORF⁶³ (FIG. 2g). Nearly half of all cellular mRNAs are predicted to contain one or more uORFs⁶⁵. In line with this, recent data from ribosome-profiling experiments have confirmed the presence of translating 80S ribosomes upstream of the primary ORF in a large portion of expressed transcripts^66–68. Despite the presence of uORFs across the genome, several studies have reported that genes encoding growth factors and cellular oncogenes are enriched for the presence of uORFs^69,70, suggesting that overcoming uORF-mediated translational repression could be important in activating tumorigenic pathways. One prominent mechanism by which cancer cells can overcome uORF-mediated translation suppression is through phosphorylation of the α-subunit of eIF2 during tumour-associated stress conditions (discussed further below). However, cancer cells have been shown to use a diverse array of additional mechanisms to overcome uORF-dependent translational inhibition of specific transcripts. Examples include the transcription of MDM2 from a cryptic promoter to produce a truncated transcript that lacks two inhibitory uORFs that are normally present in the 5′UTR^71–73 and the binding of the RBPs Hu antigen R (HUR; also known as ELAVL1) and hnRNPA1 to a translational derepression element (TDE) in the 3′UTR of HER2 (also known as ERBB2) that overrides the inhibitory effect of uORFs⁷⁴.

Conversely, it has been shown that uORF-mediated translational repression can also positively affect tumorigenesis through gain-of-function mutations, typically in tumour suppressor genes. Indeed, mutations creating a new inhibitory uORF in the 5′UTR of the cyclin-dependent kinase inhibitor 2A (CDKN2A) tumour suppressor gene (which encodes p16^INK4A and p14^ARF) have been identified in a subset of patients with an inherited predisposition to melanomas⁷⁵. Interestingly, some mRNAs even depend on uORFs for optimal translation. For example, genetic deletion of a uORF in the 5′UTR of the transcription factor CCAAT/enhancer-binding protein-β (CEBPB) in mice has revealed that this element is required for physiological in vivo expression of the C/EBPβ liver inhibitory protein (LIP) isoform⁷⁶, which mediates breast cancer cell resistance to the TGFβ cytostatic response⁷⁷ and increases tumour susceptibility when overexpressed in mice⁷⁸. Importantly, although these studies demonstrate the potential for uORFs to steer transcript-specific translation in cancer, the vast majority of predicted uORFs remain to be functionally validated.

Most of the sequence-specific elements identified to date that influence translational control of oncogenic gene expression programmes are in the 3′UTR of transcripts, which are typically longer than 5′UTRs and contain a wealth of microRNA (miRNA)-binding sites and RBP motifs that can directly affect mRNA translation⁷⁹. Importantly, 3′UTRs can be alternatively trimmed or lengthened to exclude or include these specific regulatory elements through a mechanism of alternative cleavage and polyadenylation (APA)⁸⁰, which is estimated to occur for roughly half of all human genes^81,82. Interestingly, proliferative cell states are typically associated with the selective shortening of 3′UTRs through APA⁸³. Moreover, 3′UTR shortening is further selected for in cancer cells and can help to promote the expression of proto-oncogenes such as insulin-like growth factor 2 mRNA binding protein 1 (IGF2BP1)⁸⁴.

mRNA isoforms with shortened 3′UTRs are typically associated with enhanced gene expression that can be attributed, in part, to the loss of miRNA-binding sites⁸⁰ (FIG. 2h). The relative contribution of translational repression and mRNA decay to miRNA-mediated control of gene expression has been thoroughly debated^85,86; however, recent work suggests that altered mRNA translation is one of the primary outputs of miRNA recognition^87–93. Multiple mechanisms have been proposed to explain miRNA-mediated translational repression; however, accumulating evidence suggests that inhibition of cap-dependent translation initiation is the major mechanism^{89,90,94–96}.

Interestingly, although numerous studies have demonstrated that specific miRNAs are frequently altered in human cancers, such as the let-7 family that regulates expression of the RAS oncogenes⁹⁷, it has also been reported that miRNA expression and processing are globally downregulated in human cancer^98,99. This raises an intriguing question of what advantage miRNA-mediated suppression through APA-dependent 3′UTR shortening may provide over direct control of miRNA expression and processing. As miRNAs typically recognize multiple mRNA targets within the cell, it is possible that APA-dependent 3′UTR shortening might enable cancer cells to selectively filter miRNA-mediated suppression to target distinct transcripts. Although recent studies have begun to identify specific molecular components required for 3′UTR shortening in cancer cells, such as the cytoplasmic polyadenylation element-binding protein 1 (CPEB1)^100,101 and the cleavage factor 1M complex member, cleavage and polyadenylation specificity factor 25 kDa subunit (CFIm25; also known as NUDT21)¹⁰², we still have an incomplete understanding of APA regulation and the extent to which it may promote the expression of distinct transcripts that steer cancer-relevant behaviours.

The other major group of sequence-specific elements in the 3′UTR that can direct translational control of oncogenic gene expression programmes comprises a growing number of motifs recognized by RBPs⁵⁷ (FIG. 2I). Although the selective loss of RBP recognition elements may be another driver of APA-induced 3′UTR shortening in cancer cells, their role in this process remains enigmatic, in part owing to the multifunctional nature of many RBPs, which can influence nearly every aspect of RNA metabolism¹⁰³. For example, the RBP HUR, which recognizes U- and AU-rich elements in the 3′UTR and has been well-studied in the context of cancer (reviewed in REF. 56), can influence gene expression by promoting mRNA splicing¹⁰⁴, stability¹⁰⁵ and translation¹⁰⁶. At the translational level, RBPs can regulate every step of protein synthesis from initiation to termination⁵⁶. Indeed, RBPs have even been shown to promote programmed translational readthrough (PTR) of stop codons. For example, the binding of hnRNPA2B1 to a 63-nucleotide element in the 3′UTR of VEGFA mRNA was recently shown to promote PTR and drive expression of a novel antiangiogenic isoform VEGFAx that can inhibit the growth of human colon carcinoma xenografts in nude mice¹⁰⁷. Importantly, although RBPs are typically expressed at much higher levels than most cellular proteins, contributing up to 20% of the expressed protein-coding transcriptome¹⁰⁸, specific RBPs are frequently altered in human cancers and can associate with poor prognosis¹⁰⁹. For instance, the RBP Musashi homologue 2 (MSI2), which promotes the leukaemia stem cell self-renewal programme in part through sustained translation of key self-renewal factors such as Ikaros family zinc finger protein 2 (IKZF2) and MYC¹¹⁰, was recently shown to be upregulated in human samples from patients with chronic myeloid leukaemia (CML) who have progressed from the chronic phase of disease to blast crisis^111,112.

Remarkably, several studies suggest that we have only just begun to uncover the full repertoire of cellular RBPs involved in translational control of oncogenic gene expression programmes. Indeed, novel approaches using quantitative mass spectrometry (MS) in combination with ultraviolet (UV) or PAR crosslinking to identify the complete RBP–mRNA interactome have uncovered an unexpected number and diversity of RBPs within the cell, including a large group of non-canonical RBPs not predicted to bind RNA^113–115. Moreover, recent work demonstrating that the tumour suppressor adenomatous polyposis coli (APC) can act as an RBP, recognizing mRNAs associated with its primary functions in regulating microtubule organization, motility and cancer¹¹⁶, suggests that even well-known oncogenic players could exert some of their tumorigenic potential through direct mRNA binding and translational control.

Numerous studies are also identifying an expanding list of sequence-specific elements in the 5′UTR of transcripts that are translationally regulated during tumorigenesis, including known RBP-binding motifs and sequence-specific elements for which regulatory mechanisms remain elusive^{10,26,117,118} (FIG. 2j). One of the oldest-known sequence-specific elements associated with oncogenic signalling pathways is the 5′-terminal oligopyrimidine tract (5′TOP), which consists of a 5′-terminal cytosine residue followed by a stretch of 7–13 pyrimidine nucleotides and was first identified as a translational regulatory motif controlling ribosomal protein expression associated with cellular growth^119,120. Subsequent studies demonstrated that rapamycin could selectively repress the translation of 5′TOP-containing transcripts¹²¹, linking mTOR signalling to this sequence-specific translational element.

Although 5′TOP-mediated translation was initially attributed to ribosomal protein S6 kinase (S6K)-dependent phosphorylation of 40S ribosomal protein S6 (RPS6), elegant studies genetically inactivating S6K signalling or phosphorylation of RPS6 revealed that perturbing this pathway has no impact on translational control of 5′TOP-containing transcripts^122,123. Recent work has now demonstrated that the ability of mTOR to direct the translation of 5′TOP-containing transcripts stems from its capacity to negatively regulate 4EBP-dependent inhibition of eIF4E as part of mTOR complex 1 (mTORC1)^26,117. However, evidence exists in some settings, such as during amino acid stimulation, that translational control of 5′TOP-containing transcripts may be independent of 4EBP^117,124 and that neither mTORC1 nor mTORC2 signalling alone can fully account for the control of 5′TOP-mediated translation by mTOR^125,126.

Interestingly, landmark studies using ribosome profiling to identify the genome-wide repertoire of mRNAs that are translationally sensitive to mTOR ATP-active site inhibitors^127,128 found that approximately 60% of mTOR-sensitive mRNAs also contain a 5′TOP-like element termed the pyrimidine-rich translational element (PRTE)^26,117. Importantly, the PRTE diverges from the 5′TOP in both sequence and topology, as it consists of an invariant uridine at position 6 flanked by pyrimidines and it does not reside at the +1 position in the 5′UTR. PRTE-mediated translation has been shown to be crucial for expression of a group of pro-invasion genes, including Y-box binding protein 1 (YB1; also known as YBX1) and metastasis-associated 1 (MTA1), which steer cellular metastasis downstream of oncogenic mTOR signalling²⁶, as well as for MYC-driven expression of the rate-limiting nucleotide biosynthesis enzyme phosphoribosyl pyrophosphate synthetase 2 (PRPS2), which supports anabolic metabolism of nucleotides, nucleic acids and proteins necessary for tumorigenesis²⁸. Moreover, it has recently been demonstrated that protein-level expression of the PRTE-containing transcripts YB1 and MTA1 serves as a novel biomarker for prediction of outcomes for patients with prostate cancer, further highlighting the clinical importance of this translational regulatory element¹²⁹. Importantly, unbiased translational profiling techniques, such as those used to identify the PRTE, could reveal many previously unknown translational elements that guide transcript-specific translation in cancer cells through distinct molecular mechanisms. For example, a functional element termed the cytosine-enriched regulator of translation (CERT) was recently discovered in mRNAs identified by polysome microarray analysis as being translationally sensitive to eIF4E dose during cellular transformation¹⁰.

Outstanding questions remain of how these sequence-specific elements, either through formation of higher-order structures or through recruitment of trans-acting factors, steer transcript-specific translation in cancer cells. Although work investigating the underlying mechanism of PRTE-mediated and CERT-mediated translation is currently under way, some progress has already been made with respect to 5′TOP-containing transcripts with the identification of La-related protein 1 (LARP1) as a 5′TOP-binding protein that can mediate mTOR-dependent translational control¹³⁰. However, LARP1 activity by itself seems insufficient to explain 5′TOP-mediated translation, as a recent study found that LARP1 interacts with a broad class of approximately 3,000 mRNAs that include most but not all 5′TOP-containing transcripts¹³¹. Moreover, there is evidence that LARP1 binding to the 5′TOP may actually repress translation rather than activate it¹³². Further studies using unbiased approaches, such as comprehensive identification of RBPs by MS (CHIRP-MS) to identify physiologically relevant RBPs and in vivo click-SHAPE (icSHAPE) to identify RNA structure, are needed to fully elucidate the underlying mechanisms of this and other translational elements and to move towards a comprehensive understanding of how mRNA structure and sequence synergize to write the translational code. Importantly, this work will need to determine how the expanding list of mRNA modifications (including methylation^133–136 and pseudouridylation^137–140) and alterations in ribosome biogenesis and function (BOX 2) affect interactions of the translation machinery with regulatory elements in mRNA 5′ and 3′ UTRs to influence translational control during tumorigenesis.

Box 2. The cancer cell ribosome: regulated at its core.

Alterations in ribosome synthesis and activity are a defining feature of cancer cells²⁰⁸. Indeed, a primary output from the activation of oncogenic signalling pathways, such as the MYC, RAS and AKT pathways, and from the loss of tumour suppressors, such as p53 and RB, is to drive increased ribosome biogenesis by enhancing transcription of genes encoding ribosomal proteins and ribosomal RNA (rRNA)^{145,209–218} (see the figure). Although ribosomes have historically been perceived as ‘housekeeping’ complexes that act as ‘backstage’ participants, supporting protein synthesis without having an active role in directing functional expression of the genome, recent studies have revealed that ribosome biogenesis and activity are in fact highly regulated to steer specific translational outputs²¹⁹. For example, mutations in the gene encoding ribosomal protein L38 (RPL38) have been shown to cause tissue-specific patterning defects during development as a result of impaired internal ribosome entry site (IRES)-dependent translation of select homeobox (HOX) mRNAs containing an upstream translation inhibitory element^118,220.

One of the major mechanisms by which tumour cells alter ribosome biogenesis to drive selective translation of the cancer genome is through rRNA modifications, predominantly methylation and pseudouridylation on nucleotides in functionally important regions. Most rRNA methylation is catalysed by the rRNA 2′-O-methyltransferase fibrillarin as part of a ribonucleoprotein (RNP) complex guided by the C/D class of small nucleolar RNAs (snoRNAs). Fibrillarin overexpression has been linked to p53 loss and can drive specific alterations in mRNA translation, including increased IRES-dependent translation of insulin-like growth factor 1 receptor (IGF1R) and fibroblast growth factor 2 (FGF2)²²¹. rRNA pseudouridylation is catalysed by the pseudouridine synthase dyskerin as part of an RNP complex guided by the H/ACA class of snoRNAs. Dyskerin mutations have been identified in human cancers and are invariably associated with the syndrome X-linked dyskeratosis congenita, an inherited human cancer susceptibility disorder that belongs to a broader class of ribosomopathies²⁰⁸. Decreases in dyskerin function impair IRES-dependent translation of key tumour suppressor genes, such as those encoding p27 and p53, and can cause changes in the expression of distinct subsets of snoRNAs, leading to site-specific alterations in rRNA pseudouridylation^222–226. Recently, a growing body of work has also revealed that deregulated expression of C/D and H/ACA snoRNAs is a common feature of human cancers^227–231, suggesting a potentially broader role for selective modification of rRNA in cellular transformation and tumour development.

tRNA function and codon usage bias

The human genome contains a total of 61 different sense codons, many of which are synonymous and encode the same amino acid. These 61 codons are translated by 49 different tRNA ‘isoacceptor species’, so-called because they are charged with identical amino acids but have unique anticodon sequences. New unbiased approaches for profiling tRNAs have unexpectedly revealed that the expression of tRNA isoacceptor species is enhanced in several human cancers (including breast cancer, multiple myeloma and ovarian cancers)^141–143, raising the question of whether such changes may have functional implications for expression of the cancer genome.

tRNAs have typically been thought to exist in excess of cellular demand; however, exciting new work suggests that this is not always the case. For example, the overexpression of methionine initiator tRNA (tRNA_i^Met) in human mammary epithelial cells has been shown to drive enhanced proliferation and cellular metabolism¹⁴⁴. Unexpectedly, this also caused broad increases in other tRNA levels, calling into question the relative contributions of global and specific changes in tRNA expression to translational control of these cellular processes. Given that tRNA expression can be limiting for cellular growth and proliferation, it is perhaps not surprising that many oncogenic lesions that drive these processes also promote tRNA expression. For example, activation of the AKT–mTOR, RAS–MAPK and MYC oncogenic signalling pathways and loss of the tumour suppressors p53 and RB have been shown to influence RNA polymerase III (Pol III)-dependent transcription of tRNA genes^145–148.

How might increases in tRNA expression influence translational control to steer cellular transformation and tumorigenesis? Although it is possible that general increases in tRNA expression are needed simply to meet the enhanced global protein synthesis demands of tumour cells, numerous studies suggest that cancer-associated increases in tRNA expression are not equally distributed across tRNA species^141,143. In particular, specific increases in tRNA_i^Met expression could not only augment global protein synthesis rates by enhancing translation initiation rates through increased ternary complex formation but also simultaneously inhibit the translation of select mRNAs containing multiple uORFs, such as activating transcription factor 4 (ATF4), translation of which is favoured by low levels of the ternary complex (discussed further below).

Additionally, changes in tRNA expression could influence translational outputs by enhancing translation elongation for distinct subsets of mRNAs based on their codon usage. For example, it has been shown that bias in codon usage can contribute significantly to differences in expression between genetic paralogues with similar amino acid identity, such as the KRAS and HRAS oncogenes¹⁴⁹. In this scenario, the presence of ‘rare’ codons in KRAS that are generally underrepresented in the genome has been found to limit KRAS expression at the translational level compared with HRAS. Strikingly, replacing rare KRAS codons with more commonly used codons is sufficient not only to enhance KRAS expression but also to promote KRAS-driven tumorigenesis, explaining, at least in part, the inherent differences in tumorigenicity observed between KRAS and HRAS oncogenes.

Although there initially seemed to be a lack of evidence for a broader role of genome-wide codon usage bias in mammalian systems, this has been at least partially reconciled by work demonstrating that codon selection is biased in a highly tissue-specific manner mirrored by tRNA isoacceptor expression^150,151. Importantly, these studies suggest that codon usage in mammalian systems could support the translation of specific gene expression programmes (FIG. 3). Indeed, a novel extension of this concept has recently been described, whereby function-related codon usage was found to be hardwired into transcriptional programmes underlying cellular proliferation in both normal cells and cancer cells¹⁵². In particular, this study identified selective and divergent expression of tRNA pools between states of cellular proliferation and differentiation that matched well with the respective codon usage bias in cell-autonomous genes driving proliferation and supporting differentiation. Importantly, this work highlights a direct correlation between codon content, tRNA levels and gene expression that comprises a novel mode of translational control underlying transformed cellular phenotypes such as enhanced proliferation. In line with this, the overexpression of specific tRNA isoacceptors in breast cancer cell lines has also been found to correlate positively with codon usage patterns in cancer-related genes involved in the cell cycle, extracellular matrix and transcriptional control¹⁴¹. Collectively, these studies suggest a crucial role for altered tRNA expression in driving specific pro-tumorigenic programmes.

Figure 3. Codon usage: a new layer of control in translation of the cancer genome.

Oncogenic signalling can drive changes in the transcription of tRNA genes and their modifications (denoted here as red asterisks in the anticodon loop), leading to the expression of a specific repertoire of tRNAs in the cancer cell. This can in turn support the translation of pro-tumorigenic mRNAs based on their codon usage bias, whereby mRNAs with codons better matched to the tRNA pool (depicted here by matched codon and tRNA colour) are more likely to be translated. For example, the cancer cell tRNA pool is well matched to the codon usage of transcripts involved in cellular proliferation, such as cell cycle genes, but not to those mRNAs involved in differentiation, such as pattern specification genes^141,152. Pol, RNA polymerase.

In addition to being transcriptionally controlled, tRNAs are one of the most heavily modified RNA species in the cell, with more than 90 different modifications that occur after processing and splicing of the initial transcript¹⁵³. Most of these modifications are thought to have crucial roles in tRNA structural conformation and stability; however, specific modifications in the anticodon loop have been shown to affect tRNA-directed translation¹⁵³. Recently, several studies on human cancers have identified quantitative changes in tRNA modifications and the enzymes that catalyse them, such as tRNA-isopentenyltransferase¹⁵⁴. Although it largely remains to be determined how changes in tRNA modifications contribute functionally towards tumorigenesis, emerging evidence has highlighted a role for tRNA modifications in the regulation of distinct translational outputs in response to stress conditions relevant to the transformed phenotype, including oxidative stress and DNA damage^155,156. An interesting example of this comes from the methyltransferase alkylated DNA repair protein alkB homologue 8 (ALKBH8), which is required for specific modifications to the wobble position of select tRNAs, including selenocysteine tRNAs^157–159. Selenocysteine is an amino acid uniquely ‘coded’ for by UGA stop codons that lie upstream of a structural motif known as a selenocysteine insertion sequence (SECIS) element, a common feature of several mRNAs that encode antioxidant proteins¹⁶⁰. Importantly, the ability of selenocysteine tRNA to recode UGA stop codons requires ALKBH8-mediated tRNA modifications, which are induced downstream of oxidative stress conditions, in order to support the translation of reactive oxygen species (ROS)-detoxifying enzymes, such as glutathione peroxidase 1 (GPX1)¹⁶¹. Importantly, future studies are needed to fully characterize the number and extent of changes in tRNA modifications that occur during tumorigenesis and to elucidate how tumour-associated stress conditions, such as oxidative stress, influence these modifications to control tRNA function. This work will require the continued development of new technologies that enable quantitative transcriptome-wide assessment of tRNA expression and modifications, such as AlkB-facilitated RNA methylation sequencing (ARM-Seq)¹⁶² (TABLE 1), and has the potential to provide a new understanding of how tRNA expression, modification and codon usage bias cooperate to drive the translation of gene expression programmes that underlie specific cancer cell behaviours.

Translational adaptation to stress

A key feature of cancer cells that enables them to survive and proliferate is their capacity to adapt to multiple cellular stresses encountered during tumorigenesis, including replicative, oxidative, genotoxic, metabolic and proteotoxic stress, such as endoplasmic reticulum (ER) stress¹⁶³. These cellular stresses are inherently connected to the activation of oncogenic signalling pathways and can profoundly influence tumour development and progression. For example, the ability of oncogenic signalling to promote enhanced cellular proliferation is directly linked to the induction of replicative stress and genomic instability during early stages of tumorigenesis¹⁶⁴. Numerous oncogenes have also been shown to drive altered mitochondrial activity that leads to the generation of ROS, which can be either pro-tumorigenic or antitumorigenic depending on the level and duration of their production^165,166. Although ROS can contribute to DNA mutations and act as signalling mediators that support cellular transformation at low levels, high levels of ROS are toxic and invariably associated with decreased cancer cell survival^165,166. In addition, the tumour microenvironment can dramatically influence cellular stress states. In particular, insufficient vascularization can contribute to tumour hypoxia and nutrient deprivation, causing metabolic, oxidative and ER stress¹⁶³. In order to overcome these blocks to tumorigenesis, cancer cells co-opt cytoprotective pathways that enable them to weather and resolve specific stress conditions. Importantly, an increasing number of studies are demonstrating a vital role for translational control in establishing such adaptive stress responses, whereby cancer cells redirect translation initiation and elongation to drive the synthesis of specific proteins that fuel their survival (FIG. 4).

Figure 4. Translational responses and cancer cell adaptation to tumour-associated stress.

The ability of cancer cells to adapt to stresses encountered during tumorigenesis is fundamental for tumour growth and survival¹⁶³. One of the major cellular responses to stress conditions is global inhibition of protein synthesis, which acts to conserve energy and prevent the accumulation of damaged proteins. This is mediated largely at the level of translation initiation through decreases in eukaryotic initiation factor 4E (eIF4E) activity downstream of mTOR–eIF4E-binding protein (4EBP) signalling¹⁶⁸ and inhibition of eIF2A by the eIF2A kinase family, which includes PRKR-like endoplasmic reticulum (ER) kinase (PERK) and general control non-derepressible 2 (GCN2)¹⁷¹. Stress conditions can also block translation elongation through AMP-activated protein kinase (AMPK)-dependent activation of eukaryotic translation elongation factor 2 (eEF2) kinase (eEF2K), which inhibits eEF2 (REF. 11). Although oncogenic signalling frequently suppresses the AMPK–eEF2K pathway, cancer cells selectively reactivate this pathway to promote survival during nutrient deprivation. Cancer cells can also adapt to stress by increasing the translation of selective mRNA transcripts that promote cell survival and resolve stress conditions. This adaptive stress response is driven in part by internal ribosome entry site (IRES)- and upstream open reading frame (uORF)-dependent translational mechanisms that can be maintained or even favoured under conditions in which global protein synthesis is inhibited and can promote the translation of pro-survival genes, such as BCL2 (REF. 50) and X-linked inhibitor of apoptosis protein (XIAP)¹⁷⁷, and stress response genes such as excision repair cross-complementation group 5 (ERCC5)¹⁸¹. Likewise, it has been revealed that under hypoxic conditions that inhibit global protein synthesis the eIF4E homologue 4EHP can promote the translation of select mRNAs involved in the adaptive response to hypoxia, including epidermal growth factor receptor (EGFR) and platelet-derived growth factor receptor-α (PDGFRA)¹⁹¹. Recently, exciting new studies have revealed that even eIF4E can act to promote the adaptive response to stress, for example, through enhanced translation of reactive oxygen species (ROS) regulators such as ferritin heavy polypeptide 1 (encoded by FTH1) and glutamate–cysteine ligase catalytic subunit (encoded by GCLC)¹⁰. As cancer cells do not always globally downregulate protein synthesis in response to stress, this new-found role for eIF4E may be especially relevant for tumorigenesis.

Translational control provides an unparalleled mechanism for cells to rapidly and precisely regulate gene expression during stress conditions¹⁶⁷. One of the major cellular responses to a wide range of stress conditions is the global inhibition of protein synthesis, which not only acts as a mechanism of energy conservation but can also prevent the accumulation of damaged and misfolded proteins during specific stress conditions, such as oxidative or ER stress¹⁶⁷. Global inhibition of protein synthesis is thought to occur largely at the level of translation initiation through modulation of eIF4E and eIF2 activity. Most cellular stress conditions block eIF4E activity by suppressing mTOR-dependent phosphorylation of 4EBP1, which is thought to be mediated by AMP-activated protein kinase (AMPK)-dependent activation of the mTOR inhibitor TSC2 (also known as tuberin)¹⁶⁸, although REDD1 (also known as DDIT4) and RAS GTPases, respectively, have also been implicated in mTOR control during hypoxia and amino acid deprivation^169,170.

eIF2, on the other hand, is regulated by a family of kinases that are activated under various stress conditions to phosphorylate serine 51 of the eIF2A subunit, thus blocking the recycling of eIF2 for delivery of tRNA_i^Met to the 40S ribosome. Perhaps the best characterized eIF2A kinase is PRKR-like ER kinase (PERK; also known as eIF2AK3), which regulates one of the three effector arms of the unfolded protein response (UPR) during ER stress, oxidative stress and hypoxia¹⁷¹. Three other eIF2A kinases — general control non-derepressible 2 (GCN2; also known as eIF2AK4), IFN-induced double-stranded RNA-activated protein kinase (PKR; also known as eIF2AK2) and haem-regulated inhibitor (HRI; also known as eIF2AK1) — have also been identified, although these are typically thought to be more restricted to specific cell types or certain stress conditions. In particular, GCN2 is activated by amino acid deprivation directly in response to the accumulation of uncharged tRNAs¹⁷², PKR is classically activated in response to double-stranded RNA during antiviral immune responses¹⁷³ and HRI expression is restricted to erythroid cells, where it regulates protein synthesis in a haem-dependent manner¹⁷⁴.

Traditionally, the inhibition of eIF4E and eIF2 under stress conditions is thought to cause global decreases in protein synthesis while simultaneously enabling selective translation of mRNAs essential to the adaptive stress response¹⁶⁷. These mRNAs possess distinct cis-regulatory elements in their 5′UTRs, such as IRES elements and uORFs, that direct transcript-specific translation in a manner compatible with reduced eIF4E and eIF2 activity. For example, hypoxia has been shown to induce a switch from cap-dependent to IRES-dependent translation that promotes tumour survival and angiogenesis through the ability of eIF4G to engage with IRES elements in specific mRNAs¹⁷⁵, and IRES-mediated translation of select transcripts (for example, those encoding anti-apoptotic factors such as X-linked inhibitor of apoptosis protein (XIAP) and pro-angiogenic factors such as VEGFC) has been shown to support tumour growth and survival under stress conditions such as DNA damage and hypoxia^176–179.

Transcripts containing uORFs can also be selectively translated under stress conditions. For example, a recent study found that nearly all transcripts efficiently translated under the acute induction of oxidative stress contain at least one uORF¹⁸⁰. Normally, the presence of one or more uORFs inhibits downstream translation at the major protein-encoding ORF; however, stress-induced phosphorylation of eIF2A can relieve this suppression. For example, DNA damage-induced phosphorylation of eIF2A has been shown to promote translation of a common polymorphic variant of the ERCC5 nucleotide excision repair gene that contains a new uORF in the 5′UTR, and, strikingly, the presence of this uORF polymorphism in paediatric patients with ependymoma was found to confer resistance to platinum-based DNA-damaging agents in association with significantly reduced progression-free survival¹⁸¹. One way that eIF2A phosphorylation can promote such uORF-dependent translation is through a unique ribosome scanning and re-initiation mechanism that relies on a specific configuration of two uORFs, as occurs for the transcript encoding ATF4 (REF. 182). In this scenario, after translating the first uORF, the 40S ribosome remains associated with the mRNA and continues scanning for additional start codons. Normally, eIF2 efficiently delivers tRNA_i^Met to the scanning 40S ribosome so that it re-initiates translation at the second uORF, which overlaps with the major ORF and blocks its translation. However, under stress conditions in which eIF2 is inhibited, the 40S ribosome bypasses the second uORF before tRNA_i^Met is delivered, enabling translation re-initiation at the major ORF to occur¹⁸². This mechanism promotes the selective translation of ATF4 under stress conditions in order to coordinate the transcription of numerous stress response genes, including genes involved in the production of glutathione, which detoxifies ROS, and autophagy genes such as autophagy related 5 (ATG5) and microtubule-associated protein 1 light chain 3β (MAP1LC3B) that promote cell survival¹⁸³.

Although translational control during stress has largely been attributed to enhanced IRES- and uORF-dependent translation that is coupled with decreased global protein synthesis, accumulating evidence suggests that this paradigm is insufficient to explain translational regulation of the adaptive stress response during tumorigenesis. For example, several studies have demonstrated that not all cancer cells globally downregulate protein synthesis in response to stress conditions^184–186. In fact, ER-stress-dependent activation of the pro-survival PERK–eIF2A pathway in MYC-driven lymphomas is inherently linked to the ability of MYC to enhance global protein synthesis¹⁸⁶. Moreover, it is apparent that our understanding of the regulation of IRES- and uORF-containing transcripts and their contribution to the stress response is incomplete. For example, although the ability of eIF4G to promote selective translation under stress conditions has largely been ascribed to its capacity to influence IRES-dependent translation, many eIF4G-dependent transcripts seem to lack functional IRES elements¹².

Recently, unanticipated layers of translational control, beyond IRES and uORFs, have been revealed as integral to the adaptive stress response in cancer cells. One unexpected finding came when 4EHP (also known as eIF4E2), a cap-binding homologue of eIF4E best characterized for its role as a translational suppressor crucial for embryonic development and stem cell function^187–189, was implicated in the hypoxic stress response. Although eIF4E is typically thought to be the primary cap-binding protein under hypoxic conditions¹⁹⁰, 4EHP was recently identified as a central component of a hypoxia-induced translation initiation complex that includes oxygen-regulated hypoxia-inducible factor 2α (HIF2α) and the RBP RBM4¹⁹¹. PAR-CLIP analysis revealed that under hypoxic conditions RBM4 recruits HIF2α to a specific RNA element in the 3′UTR of select mRNAs involved in the adaptive response to hypoxia, including epidermal growth factor receptor (EGFR), platelet-derived growth factor receptor-α (PDGFRA) and insulin-like growth factor 1 receptor (IGF1R), promoting their translation through interaction with cap-bound 4EHP. Although this work describes a novel role for 4EHP in promoting translation under hypoxic stress conditions, the full scope of mRNAs regulated by the 4EHP hypoxia complex and the underlying mechanism by which it promotes translation remain outstanding questions. Interestingly, even the major cap-binding protein eIF4E, which is typically thought to be inhibited under stress conditions, has emerged as a surprise player in the response to cellular stress during tumorigenesis. Indeed, eIF4E dose was recently identified as a crucial determinant of the translation of oxidative stress response genes essential for cellular transformation and in vivo lung tumorigenesis, such as ferritin heavy polypeptide 1 (FTH1) and the rate-limiting enzyme for glutathione synthesis glutamate–cysteine ligase catalytic subunit (GCLC)¹⁰. Similarly, increased eIF4E dose was shown to rescue human mammary epithelial cells from oncogene-induced replication stress through the activation of a pro-survival DNA damage response¹⁹². In line with these studies, eIF4E phosphorylation has also been implicated in resistance to oxidative stress, starvation and DNA-damaging agents in cancer cells¹⁹³.

Although most work has focused on translation initiation, translation elongation can also be a major node in regulation of the adaptive stress response in cancer cells. Normally, translation elongation is blocked under metabolic stress conditions by AMPK-dependent activation of eEF2 kinase (eEF2K), which phosphorylates and inhibits eEF2 to promote cell survival¹⁹⁴. However, oncogenic signalling can suppress eEF2K activity and sensitize cells to nutrient deprivation-induced apoptosis¹¹. Therefore, in order to adapt to metabolic stress, cancer cells selectively reactivate AMPK–eEF2K signalling to inhibit translation elongation and exploit this survival pathway¹¹. Remarkably, eEF2K-dependent inhibition of translation elongation does not seem to influence global protein synthesis rates during nutrient deprivation, suggesting that eEF2 blockade may instead promote cell survival through selective effects on mRNA translation.

Collectively, these findings highlight a newly uncovered vulnerability of cancers that can be exploited therapeutically. Indeed, as tumours face numerous stresses during their development and progression, targeting translational control of the adaptive stress response provides a unique strategy by which to selectively kill cancer cells. Moreover, translation inhibitors have the potential to potently synergize with pharmacological inducers of cellular stress in novel combinatorial approaches. In this regard, it has recently been shown that genetically targeting eIF4E-dependent control of the oxidative stress response sensitizes tumours to small-molecule inducers of ROS¹⁰. Similarly, the inhibition of MNK-dependent eIF4E phosphorylation in cancer cells has been found to cooperate with both ROS induction and DNA-damaging agents in inducing apoptosis¹⁹³. As current research continues to reveal a more detailed understanding of how the translational circuitry is rewired to weather cellular stresses encountered during tumorigenesis, new therapeutic targets and approaches for selective blockade of these adaptive stress responses will undoubtedly be discovered.

Conclusion and future perspectives

In recent years, many studies have revealed crucial new insights into the regulatory circuitry governing translation of the cancer genome, including cis-regulatory elements present in the 5′ and 3′ UTRs of mRNAs and codon usage bias embedded within the mRNA-coding region. This regulatory logic provides the basis for the selective translation of key mRNA networks governing distinct tumour cell behaviours, such as proliferation, cell survival, metastasis and adaptation to stress conditions. Although considerable gains have been made in understanding the molecular basis underlying translational specificity in cancer cells, we still lack a complete map of this regulatory landscape. In particular, major gaps exist in our understanding of how structural and sequence-specific cis-regulatory elements in mRNAs, such as IRESs and PRTEs, interface in trans with specific RBPs and with components of the translational machinery. Exciting technologies have recently been developed (TABLE 1), such as individual-nucleotide-resolution UV CLIP (iCLIP) and icSHAPE, that will enable future studies to address these crucial questions and open new doors into our understanding of cancer aetiology.

Importantly, translation deregulation is emerging not only as a primary downstream output of oncogenic signalling but also as a central mediator of cancer resistance to several clinical therapies^13,195. These findings, along with a growing body of work that has shifted the view of translation from an ‘undruggable’ housekeeping function to a regulated network that is co-opted to support tumour growth and function, are driving increasing interest in targeting translational control as a means of selectively killing cancer cells³⁰. Indeed, therapeutic approaches to inhibit the translation initiation machinery^26,196 and even ribosome biogenesis itself ^197–199 are already being tested in clinical trials for their efficacy against various cancers. Ultimately, as research continues to decipher the regulatory language underlying translation of the cancer genome, this will pave the way for new strategies that target even more refined levels of translation, providing novel therapeutic approaches to disrupt the translational control networks that cancer cells tailor to support the transformed phenotype.

Glossary

mTOR

A serine/threonine kinase that forms two distinct molecular complexes (mTORC1 and mTORC2) and acts as a master regulator of protein synthesis, largely through mTORC1-dependent phosphorylation of eukaryotic translation initiation factor 4E-binding proteins (4EBPs) and ribosomal protein S6 kinase

tRNA

Non-coding RNA with a unique L-shaped tertiary structure that contains a ‘charged’ amino acid covalently linked to the tRNA by aminoacyl transferases at one end and the tRNA anticodon loop that recognizes distinct mRNA codons through base-pairing interactions at the other end

Internal ribosome entry site (IRES)

A complex structural element first discovered and characterized in viruses that facilitates translation initiation by recruiting the 40S ribosome to the mRNA in a cap-independent manner frequently aided by specific RNA-binding proteins known as IRES trans-acting factors (ITAFs)

Messenger ribonucleoprotein (mRNP) complex

A complex of mRNA and RNA-binding proteins that can vary throughout the life of the mRNA and act to either promote or inhibit mRNA splicing, stability and translation depending on the nature of the bound proteins

Upstream open reading frame (uORF)

An ORF comprising a start codon upstream of the primary ORF with an in-frame stop codon that can lie either 5′ or 3′ of the primary start codon. uORFs typically block translation of downstream ORFs, as ribosomes generally initiate translation at the first start codon they encounter and then disassociate from the mRNA upon translation termination

Alternative cleavage and polyadenylation (APA)

A mechanism for generating mRNAs with different lengths of 3′ untranslated region that relies on the recognition of alternative poly(A) signals in association with U- and GU-rich downstream sequence elements by the cleavage and polyadenylation specificity factor (CPSF) complex and the cleavage stimulating factor (CSTF) complex

Programmed translational readthrough (PTR)

A phenomenon that enables translation to continue past the normal stop codon in favour of termination at a downstream stop codon, potentially generating an elongated protein with a novel 3′ extension

5′-terminal oligopyrimidine tract (5′TOP)

A sequence-specific element located at the +1 position of the 5′ untranslated region that consists of a 5′ cytosine residue followed by a stretch of 7–13 pyrimidine nucleotides

Pyrimidine-rich translational element (PRTE)

A 5′ untranslated region (5′UTR) sequence-specific motif related to the 5′-terminal oligopyrimidine tract that is not restricted to the +1 position of the 5′UTR and contains an invariant uridine at position 6 flanked by pyrimidines

Ribosomopathies

Inherited human cancer susceptibility disorders — such as cartilage–hair hypoplasia syndrome, Shwachman–Diamond syndrome, 5q deletion syndrome, dyskeratosis congenita and Diamond–Blackfan anaemia — that are characterized by mutations in distinct components of the translational machinery, including enzymes involved in the modification and processing of rRNA, ribosome assembly factors and ribosomal proteins

Methionine initiator tRNA (tRNA_i^Met)

A distinct species of tRNA charged with methionine that interacts with the ribosome P site as part of the eIF2–GTP–Met-tRNA_i^Met ternary complex that initiates translation at start codons

Wobble position

The first position in a tRNA anticodon loop that can enable non-Watson–Crick base interactions with the third nucleotide in the codon triplet and be a site for tRNA modifications

Unfolded protein response (UPR)

A conserved pathway activated by the accumulation of unfolded proteins in the endoplasmic reticulum (ER) that signals through three downstream effector arms (PRKR-like ER kinase (PERK), inositol-requiring enzyme 1α (IRE1α) and activating transcription factor 6 (ATF6)) to either relieve protein misfolding stress in the ER through increased production of chaperone proteins, enhanced protein degradation and dampened protein synthesis or commit the cell to apoptosis if stress is unresolvable