Translation acrobatics: how cancer cells exploit alternate modes of translational initiation

Abstract

Recent work has brought to light many different mechanisms of translation initiation that function in cells in parallel to canonical cap‐dependent initiation. This has important implications for cancer. Canonical cap‐dependent translation initiation is inhibited by many stresses such as hypoxia, nutrient limitation, proteotoxic stress, or genotoxic stress. Since cancer cells are often exposed to these stresses, they rely on alternate modes of translation initiation for protein synthesis and cell growth. Cancer mutations are now being identified in components of the translation machinery and in cis‐regulatory elements of mRNAs, which both control translation of cancer‐relevant genes. In this review, we provide an overview on the various modes of non‐canonical translation initiation, such as leaky scanning, translation re‐initiation, ribosome shunting, IRES‐dependent translation, and m⁶A‐dependent translation, and then discuss the influence of stress on these different modes of translation. Finally, we present examples of how these modes of translation are dysregulated in cancer cells, allowing them to grow, to proliferate, and to survive, thereby highlighting the importance of translational control in cancer.

Glossary

AMPK

AMP‐activated protein kinase

ATF4

Activating transcription factor 4

CaMV

Cauliflower mosaic virus

CDKN1B

Cyclin‐dependent kinase inhibitor 1B

CDKN2A

Cyclin‐dependent kinase inhibitor 2A

CEBPA

CCAAT/enhancer‐binding protein A

CITE

Cap‐independent translation enhancer

DAP5

Death‐associated protein 5

DENR

Density‐regulated re‐initiation and release factor

eIFs

Eukaryotic initiation factors

FGF2

Fibroblast growth factor 2

FTO

Fat mass and obesity‐associated protein

GCN2

General control nonderepressible 2

HIF1α

Hypoxia‐inducible factor‐1 alpha subunit

hnRNP

Heterogeneous nuclear ribonucleoprotein A1/C1/C2/Q/Q1

HRI

Heme‐regulated eIF2α kinase

HSPA1A

Heat shock protein family A (Hsp70) member 1A

HSPA5

Heat shock protein family A (Hsp70) member 5

HuR

Human antigen R

IRES

Internal ribosome entry site

ITAF

IRES trans‐acting factor

m⁶A

N₆‐methyladenosine

m⁷G

7‐Methylguanosine

MCTS1

Malignant T‐cell‐amplified sequence 1

METTL3

Methyltransferase‐like 3

MNK1/2

MAP kinase‐interacting serine/threonine kinase 1/2

mTORC1

Mammalian target of rapamycin complex 1

NF‐Kβ

Nuclear factor kappa‐light‐chain‐enhancer of activated B cells

NRF2

NF‐E2‐related factor 2

ODC

Ornithine decarboxylase

ORF

Open reading frame

p38MAPK

p38 mitogen‐activated protein kinase

PEK

Pancreatic eIF2 kinase

PERK

Protein kinase RNA‐like endoplasmic reticulum kinase

PIC

Pre‐initiation complex

PKR

Protein kinase R

PTBP1

Polypyrimidine tract‐binding protein 1

RBM3

RNA‐binding protein 3

RHA

RNA helicase A

TCP80

Translational control protein 80

TP53

Tumor protein p53

uORF

Upstream open reading frame

UPR

Unfolded protein response

UTR

Untranslated region

VEGF

Vascular endothelial growth factor

XIAP

X‐linked inhibitor of apoptosis

YTHDF1/2

YTH N₆‐methyladenosine RNA‐binding protein ½

Introduction

Almost every biological process is tweaked by cancer in one way or the other. This includes the most “basic” and “constitutive” cellular processes required for viability, such as cell metabolism or protein biosynthesis. Here, we provide an overview on how mRNA translation is altered in cancer cells.

Canonical, cap‐dependent translation has been extensively studied in the past few decades 1. This has led to the identification of the molecular factors involved in cap‐dependent translation, and to the delineation of the succession of events involving ribosome recruitment, scanning, initiation, elongation, and finally termination of translation 2, 3, 4, 5, 6, 7. Although this mode of mRNA translation is the most studied and best understood, alternate modes of translation initiation also exist, such as IRES‐dependent, m⁶A‐dependent, or re‐initiation‐dependent translation 8. These alternate modes of translation initiation can become as important, if not more important, than the canonical cap‐dependent mode under particular circumstance 9, 10. Since canonical cap‐dependent initiation is inhibited by stresses such as nutritional stress, hypoxia, proteotoxic stress, or genotoxic stress, stressed cells are forced to rely on alternate translation modes for their survival 11, 12, 13. Of particular note, cancer cells are often subjected to these types of stress, and therefore likely rely on, and exploit, these alternate modes of translation for their survival and proliferation 14. In this review, we first briefly highlight the different modes of translation initiation. We do not aim to go into great depth into any one mode of translation initiation, but wish to provide the reader with an overview, and will refer whenever possible to the many excellent recent reviews that elaborate on individual translation modes in detail. Then, we discuss the influence of stress on these different modes of translation and, finally, will conclude by describing how alternate modes of translation are exploited by cancer cells. Of note, regulation of translation is quite complex, with many interconnected mechanisms. Therefore, we try to explain translational control as simply as possible by presenting only the main concepts.

Modes of translation initiation

Canonical cap‐dependent translation initiation

Translation can be split into four main steps, namely initiation, elongation, termination, and recycling 2, 5, 6, 7, 15. When eukaryotic cells are growing in non‐stressed conditions, most mRNAs are translated via a cap‐dependent and scanning‐dependent initiation mechanism (Figs 1 and 2i) 7, 16. The first step of initiation occurs when the eIF4F complex binds the m7G cap structure at the 5′ end of mRNAs 3, 15, 17. The eIF4F complex is composed of eIF4E, which recognizes the mRNA 5′ cap structure; the RNA helicase eIF4A, which helps unwind the 5′ region of the mRNA; eIF4B, which stimulates the activity of eIF4A; and the main scaffolding subunit eIF4G, which recruits the ribosome by binding eIF3 and circularizes the mRNA by binding poly(A)‐binding proteins 18, 19. In parallel, a key initiation step is the binding of initiator tRNA and GTP to the eIF2 complex (eIF2α, β, and γ) 20, yielding the so‐called ternary complex 15, 21. As discussed below, this is one of the key regulatory steps in canonical cap‐dependent translation, because many stresses inactivate eIF2. The ternary complex then binds the 40S small ribosomal subunit together with other initiation factors, yielding the 43S pre‐initiation complex (PIC), which is then recruited to the activated mRNA 21, 22, 23. The resulting complex scans along the mRNA in a 5′‐to‐3′ direction to identify an AUG start codon in an appropriate sequence context. After start codon recognition by the Met‐tRNAi, scanning is arrested, and eIF2 and other initiation factors are released and recycled for translation of other mRNAs. This complex then binds the 60S ribosome to form the elongation‐competent 80S ribosome 2, 5, 24, 25 (Fig 1).

Figure 1. The canonical pathway of eukaryotic translation initiation.

The figure shows a schematic representation of the canonical pathway of eukaryotic translation initiation, including the canonical initiation factors and signaling pathways that regulate these initiation factors.

Figure 2. Modes of translation.

An overview of the various modes of mRNA translation is shown. For details, please refer to the main text.

Leaky scanning

Leaky scanning (Fig 2ii) occurs when several AUG start codons are present on a transcript, and the first AUG is in a sequence context of intermediate strength, so that ribosomes sometimes initiate on this AUG, and sometimes scan past it to initiate on another AUG downstream 26, 27, 28. Various configurations are possible 29, 30. If the upstream AUG is part of a small upstream open reading frame (uORF) which does not code for a protein or peptide of biological function, this mechanism essentially blunts translation of the main downstream ORF. If the two AUG codons are in frame with each other, this leads to translation of two isoforms of the same protein, one with an N‐terminal extension. If the two AUG codons are part of two overlapping ORFs but with different reading frames, this leads to expression of two different proteins from the same mRNA. In these latter two cases, the frequency of leaky scanning affects the relative amounts of the short and long protein isoforms or of the two proteins being made. As discussed below, all of these mechanisms have been found mutated in cancers. In addition, the frequency of leaky scanning depends on several factors such as cellular levels of eIF1 and eIF5 31, 32, 33, 34, which affect the stringency of start codon selection by the ribosome. Hence, leaky scanning could also be regulated physiologically or pathophysiologically. Interestingly, eIF1 and eIF5 auto‐regulate their own translation levels by leaky scanning, thereby establishing a feedback loop to modulate the stringency of start codon selection 31, 32.

Translation re‐initiation

In addition to the canonical cap‐dependent mode of translation, several non‐canonical yet cap‐dependent translation modes have been discovered. Forty‐nine percent of all human mRNAs contain uORFs 35. In many cases, these uORFs contain AUG codons in a strong sequence context, causing ribosomes to initiate translation there, as observed recently by ribosome footprinting experiments 36, 37. In these cases, ribosomes need to terminate translation of the uORF and then re‐initiate translation further downstream on the main ORF. This process of re‐initiation (Fig 2iii) has been known to exist for a while, but is only recently starting to be molecularly characterized 38, 39, 40, 41. The canonical translation initiation factor eIF3h is involved in this process 42, 43, as well as the non‐canonical initiation factors eIF2D, DENR, and MCTS1 40, 44, 45, 46, 47, 48. One key step in this process is likely the de novo recruitment of an initiator tRNA after termination of translation of the uORF, and these non‐canonical factors can recruit tRNA in an eIF2‐independent manner 45.

Ribosome shunting

Another cap‐dependent but non‐canonical translation mode is ribosome shunting (Fig 2iv), which was first observed on mRNAs from viruses such as cauliflower mosaic virus (CaMV) and Sendai virus. After recruitment to the cap, a non‐linear ribosome migration or shunt is mediated by an extended hairpin structure in the 5′ UTR, facilitating migration of the ribosome directly to the start site for initiation, bypassing the large internal leader region 49. This mechanism has also been observed on cellular mRNAs. The inhibitor of apoptosis protein 2 (cIAP2) mRNA has structures similar to those of the CaMV mRNA which allow ribosomes to shunt past upstream AUGs for efficient translation of the cIAP2 protein 50.

eIF4E‐independent cap recognition

The c‐Jun mRNA was recently found to be well translated under cellular conditions where recruitment of the 43S PIC by eIF4E is inhibited. This led to the discovery that c‐Jun translation can be partly attributed to an alternate cap‐dependent but eIF4E‐independent mechanism. Indeed, the c‐Jun mRNA was found to have an inhibitory RNA element that blocks eIF4E recruitment. Translation of c‐Jun mRNA may occur via binding of eIF3d directly to its cap 51. Future work will be required to further study this very interesting scenario.

IRES‐dependent translation initiation

Since cap‐dependent translation is inactivated in animal cells as a protective mechanism against viral infections, translation of viral proteins is often initiated via an RNA element called an internal ribosome entry site (IRES) (Fig 2v) 52. The term IRES was coined after discovering that 40S ribosomes can directly bind to specific internal structures of the mRNA without the necessity of 5′ cap‐dependent initiation 53. IRESs were first discovered in picornaviruses (hepatitis A virus, human rhinovirus, poliovirus, and foot‐and‐mouth disease virus), which lack a 5′ cap and a poly(A) tail 52, 54, 55, 56, 57. Detailed structural analysis of viral IRES elements led to the discovery of mRNA secondary and tertiary structures that are responsible for the direct recruitment of the canonical translation machinery 58, 59. Viral IRESs thereby allow a bypass of the cellular block in cap‐dependent translation initiation, allowing production of viral proteins and viral replication 60, 61, 62, 63. Analogous IRESs have also been found in cellular mRNAs, with 50 viral IRESs and 70 cellular IRESs reported to date 64, 65. The validity of many cellular IRESs, however, is unclear 66, 67, 68, 69. Cellular IRESs are often assayed using a bicistronic reporter with an upstream ORF that is translated in a cap‐dependent manner, and a downstream ORF that is poorly translated unless an IRES is placed between the two ORFs. IRES activity, however, could be confused with other activities of this intervening sequence, such as a cryptic promoter, or a splice acceptor that causes splicing out of the upstream ORF, as both of these would also lead to expression of the downstream ORF 70. Such concerns, however, were assuaged by experiments using RNA reporters which were transcribed in vitro and then transfected into cells, thereby bypassing the nucleus where transcription and splicing occur 71. Recent work suggests that 10% of cellular mRNAs have the potential to be translated via an IRES 72, 73. Cellular IRES elements are thought to recruit components of the translational machinery including both canonical and non‐canonical initiation factors, and consequently the ribosome 72, 73, 74, 75, 76. A class of proteins has been identified, termed IRES trans‐acting factors (ITAFs), which bind cellular IRESs and help recruit the ribosome 77, 78. The field has grouped cellular IRESs into two classes, based on the mechanism of translation initiation: (i) The “land and scan” class recruits ribosomes near to a start codon and the 40S then scans for the appropriate AUG. This is utilized by the IRESs of c‐Myc and l‐Myc 79. (ii) A second class containing a sequence similar to the bacterial Shine‐Dalgarno sequence which helps recruit the ribosome to the mRNA. This is utilized by cellular IRES of the insulin‐like growth factor 1 receptor (IGF1R) and GTX mRNAs 80, 81. Cellular IRESs are thought to serve two major functions. Firstly, they allow cells to adapt to stress. As described below, cap‐dependent translation is inhibited in response to stress. Hence, mRNAs essential for cell survival and for stress response are translated via an IRES‐mediated mechanism 82. Cellular IRESs thereby allow cells to go through apoptosis, mitosis, or angiogenesis 83, 84, 85, 86. Additionally, IRESs also aid in the translation of proteins encoded by mRNAs with highly structured 5′ UTRs, where the accessibility of the cap‐binding complex is limited 87 (Fig 2). Very nice reviews describing IRES‐mediated translation in detail, including the trans‐acting factors, have recently been published 9, 10, 88, 89.

m⁶A‐dependent translation initiation

N₆‐methyladenosine (m⁶A) modification is the most abundant RNA modification, comprising roughly 80% of all RNA base modifications. N₆‐methyladenosine was first discovered to be most prevalent in the 3′ UTR of mRNAs and was hence hypothesized to be responsible for the recruitment of RNA‐binding proteins for mRNA stability 90. However, in recent years attention has also been placed on m⁶A modifications in the 5′ UTR of mRNAs where it plays a role in mRNA translation initiation 91 (Fig 2vi). Indeed, uncapped mRNAs can be translated in cell‐free extracts if they contain m⁶A, and translation initiation complexes can assemble on such mRNAs in reconstituted systems in the absence of eIF4F complex, thereby providing convincing evidence that m⁶A modification can promote translation initiation 92, 93. Additionally, 5′ UTR m⁶A modification can function to directly bind eIF3, which then recruits the 43S complex for translation initiation 92. It has been proposed that m⁶A‐dependent translation could be used by cells under stress, as the abundance of m⁶A modification increases in response to heat shock 94. Recent work has found that m⁶A methylation of mRNA occurs predominantly in the nucleus prior to mRNA splicing 95, suggesting that such regulation in response to stress would need to happen during mRNA biogenesis. Depletion of the methyltransferase METTL3 or of the m⁶A reader protein YTHDF1 selectively inhibits the translation of mRNAs bearing m⁶A modification, indicating that m⁶A modification plays an important role in mRNA translation 96, 97, 98.

Cap‐independent translation enhancers (CITEs)

Cap‐independent translation enhancers were first discovered in plant viruses over two decades ago, where the CITEs are used for the translation of naturally uncapped mRNAs. They are most prevalent in the 3′ UTR of plant viral mRNAs, where they assume various three‐dimensional T‐shaped, Y‐shaped, and I‐shaped forms and pseudoknotted structures in order to facilitate the recruitment of the components of the eIF4F complex or the ribosome and ribosomal subunits 99. Similar to IRESs, CITEs act in a cap‐independent manner. Unlike IRESs, however, they need a free mRNA 5′ end to function. Hence, CITEs are cap independent but 5′ end dependent. Recently, the 5′ UTR of the APAF‐1 mRNA was shown to have CITE‐like properties (Fig 2vii) 100, in addition to an IRES 101. Future work will be needed to elucidate whether additional cellular mRNAs have CITEs.

How cellular stress affects translation

In this section, we describe how cellular stress affects translation. This is critical for understanding how these mechanisms are bypassed or hijacked in cancer cells to survive and proliferate. Firstly, this is because many cancer cells are exposed to stresses such as genotoxic stress, hypoxia, or nutrient limitation, which normally would inhibit translation, and thereby cell growth and proliferation 102, 103, 104. Hence, mutations accumulate in cancer cells that lead to inactivation of these translational mechanisms 105. Secondly, cancer mutations can emulate some of these regulatory mechanisms to provide an advantage to the cells, for instance, reducing translation of tumor suppressors or increasing translation of oncogenes 105, 106.

Stress inactivates canonical cap‐dependent translation

Many types of cellular stress inactivate cap‐dependent translation as a cellular defense mechanism 107. In response to double‐stranded RNA, cells block cap‐dependent translation in an attempt to inhibit viral replication 63. In response to the unfolded protein response in the endoplasmic reticulum, cells block cap‐dependent translation to reduce bulk protein biosynthesis rates, thereby granting the folding machinery time and capacity to fold proteins properly 108, 109. In response to limiting nutrients or oxygen, cells block cap‐dependent translation to adapt to the lower amino acid or energy availability, both of which are needed for translation 105, 110.

Cellular stresses mainly inhibit cap‐dependent translation via two protein complexes—the eIF4F cap‐binding complex and the eIF2 ternary complex. One major stress‐sensing signaling hub is the anabolic kinase mTORC1, which becomes inactivated in response to limiting amino acids, oxygen, ATP, growth factors, nutrients, or the unfolded protein response 111, 112, 113. Inactivation of mTORC1 leads to de‐phosphorylation of the translational repressor eIF4E‐BP, which binds the cap‐binding factor eIF4E, thereby preventing it from recruiting the rest of the translational machinery 114. Other stress conditions such as heat, cold, or hypoxia inactivate cap‐dependent translation by activating p38MAPK and consequently MNK1 and MNK2 115. These kinases then bind to eIF4G, which binds to eIF4E. This brings the kinases in proximity to their substrate eIF4E, which facilitates the phosphorylation of eIF4E, markedly reducing its affinity to the mRNA cap and thereby inhibiting translation initiation 116, 117. The physiological significance of eIF4E phosphorylation is still a matter of debate, with some reports indicating that eIF4E phosphorylation promotes translation initiation 118, 119, 120, 121.

Multiple stresses impinge on the eIF2 ternary complex which is responsible for recruiting the initiator tRNA to the 40S small ribosomal subunit prior to cap binding. Amino acid deprivation, ER stress, the presence of toxic heavy metals, heme levels, UV, hypoxia, osmotic shock, heat shock, viral infection, or dsRNA induces phosphorylation of eIF2α by a battery of kinases, each sensing a different subset of stresses: GCN2, PERK, HRI, PEK, and PKR 122, 123, 124, 125, 126. Sub‐stoichiometric phosphorylation of the eIF2α subunit of the eIF2 complex leads to strong binding and titration of the GDP/GTP exchange factor eIF2B, thereby inactivating all the remaining eIF2 complexes in the cell, resulting in globally reduced cap‐dependent translation initiation 127.

Of note, when a cell responds to stress, it is not beneficial for the cell to simply turn off all translation. Instead, it needs to produce the proteins necessary to maintain viability and to counteract the stress 11. For this reason, in addition to inactivating canonical cap‐dependent translation, cellular stresses usually also activate or modify the activity of other modes of translation 128, as described below.

Effect of stress on usage of alternate AUG start codons

In response to stress, the usage of uORFs changes, thereby affecting translation of the main ORF coded by the corresponding transcripts 129. A typical example of this is translation of ATF4. ATF4 is a transcription factor that promotes transcription of cellular stress response genes 130. The mouse ATF4 mRNA contains two uORFs, and the second one overlaps the main ATF4 coding sequence 131, 132. Traditionally, ATF4 translation was thought to be regulated as follows: When eIF2α is active, translation commences from the first uORF, a small peptide is translated, and then, the scanning ribosome recognizes the second uORF. Owing to the availability of the ternary complex, translation from the second uORF is also initiated. Since the second uORF overlaps with the main ORF, the translating ribosome bypasses the start codon of the main ORF. On the other hand, under stress conditions such as amino acid removal, hypoxia, or growth factor deprivation, activity of the ternary complex is low due to eIF2α phosphorylation. In this case, after terminating translation of the first uORF, the scanning ribosome cannot recruit the ternary complex quickly enough to initiate on the second uORF. Hence, the ribosome skips the second uORF, leading to initiation on the main ATF4 ORF 131. Recent data, however, have called into question whether this traditional model of ATF4 regulation is correct. For instance, uORF2 translation does not appear to be suppressed in response to stress 133. This suggests that additional mechanisms are at play. Recently, m⁶A modifications in the 5′ UTR of the ATF4 mRNA have been discovered. These modifications regulate ATF4 translation by stalling scanning ribosomes on uORF2, thereby reconfiguring translation in response to low ternary complex availability 134, 135. Although this is a specific, highly characterized example of alternate uORF usage, it illustrates the general principle that uORF usage can change in response to physiological conditions. Indeed, uORF translation changes dramatically when comparing mitotic versus meiotic yeast cells 136, or mouse stem cells versus differentiated cells 36.

Effect of stress on usage of non‐AUG start codons

Recent ribosome footprinting experiments have revealed that thousands of non‐AUG codons are used as start codons for translation initiation genome‐wide 137, 138, 139. Selection of the appropriate start codon for initiation is a regulated process, involving eIF1, eIF1A, eIF2, eIF3, and eIF5 140. Alterations in levels, stoichiometry, or activity of these factors can alter start codon selection. For instance, increased levels of eIF1 lead to more stringent selection of AUG as a start codon, and therefore, initiation on CUG start codons is reduced 141. Cells appear to alter their selection of start codons in response to stress. For instance, c‐Myc is translated at higher levels from an upstream CUG start codon when blood cells or fibroblasts reach high density in culture. This effect is mediated by methionine limitation and leads to the translation of a longer isoform of c‐Myc 142, 143, 144, 145.

IRES‐mediated translation during cellular stress

One mechanism by which cells continue to translate proteins required for stress responses is via IRESs. Many genes of the integrated stress response, such as HSPA1A, HSPA5, NRF2, and HIF1α, are thought to contain IRESs 11, 146, 147, 148, 149. These are translated when cells experience transient stress. Under conditions of chronic stress where the cell has reached an irreparable state, IRES‐mediated translation of apoptosis genes is required to induce cell death and avoid metastatic transformation 9. In many cases, IRES‐mediated translation initiation requires the activity of IRES trans‐acting factors (ITAFs), which bind the IRESs and modulate the recruitment of other initiation factors. Activation and expression of ITAFs is also controlled by various stress conditions 150. Below we will illustrate how the various types of stress—ER stress, hypoxia, nutrient deprivation, genotoxic stress, and thermal stress—affect IRES‐dependent translation by providing a few examples.

ER stress leads to the unfolded protein response (UPR), which signals to the cell that its protein synthesis rate has surpassed its protein folding and secretion capacity. An early event in response to ER stress is the activation of PERK which phosphorylates eIF2α, thereby reducing global translation rates and hence the accumulation of misfolded proteins 151. Under these conditions, a number of UPR mRNAs are translated in an IRES‐dependent fashion. For instance, during ER stress, caspase‐12‐mediated cleavage of the ITAF eIF4G2 (DAP5/p97) produces a cleaved fragment termed p86 152. The p86 fragment promotes translation of HIAP2 (human inhibitor of apoptosis protein 2) via its IRES. This consequently attenuates apoptosis and allows UPR to cope with the stress 153. PERK activation also leads to the up‐regulation of IRES‐dependent translation of an isoform of TP53 154. Two isoforms of the TP53 protein have been shown to be translated from two different IRESs in the same mRNA, namely TP53 and TP53/47 155, 156. Upon ER stress, an increase in IRES‐dependent translation of the TP53/47 isoform induces cell cycle arrest at G2, unlike the full‐length TP53 which induces G1 arrest 154.

Analogously to ER stress, genotoxic stress leads to expression of stress response genes via IRES‐dependent mechanisms. In response to doxorubicin‐induced genotoxic stress, translation of both isoforms of TP53 (TP53 and TP53/47) is increased in order to improve DNA repair mechanisms. This increase is dependent on the relocalization of the ITAF PTBP1 from the nucleus to the cytoplasm 157.

Hypoxia decreases the ATP/AMP ratio in cells 158, leading to AMPK (AMP‐activated protein kinase) activation, which in turn represses mTORC1 signaling, and thereby cap‐dependent translation 159. Proteins essential for survival under hypoxic conditions still need to be translated, and this occurs at least in part by an IRES‐mediated mechanism. One major regulator of the hypoxic response is the transcription factor HIF1α 160. Translation of HIF1α under hypoxia is largely dependent on an IRES which is recognized by the ITAF HuR (human antigen R) 161.

Nutrient deprivation also causes inactivation of mTORC1 leading to reduced cap‐dependent translation 162. Hence, some proteins required to fight this stress are thought to be translated in an IRES‐dependent manner 149, 163, although this has been debated 164. Under conditions of glucose deprivation, cytoplasmic localization of SMAR1 (scaffold/matrix attachment region binding protein 1) is increased. This protein functions as an ITAF for the translation of both the IRESs in the TP53 mRNA 165. Analogously, IRES‐dependent translation of XIAP is also increased under serum starvation, which leads to inhibition of apoptosis, improving cell survival 166.

Heat stress causes proteins to misfold and to accumulate, causing a feedback to halt protein synthesis through phosphorylation of eIF2α 167. Under such conditions, heat stress is alleviated in part by expression of the chaperone HSPA1A. Translation of HSPA1A is facilitated by a 216‐nt‐long IRES element in its mRNA 147, as well as via a m⁶A‐dependent mechanism (described below). Hypothermia also induces a physiological response which includes a prolonged G2 phase of the cell cycle 168. Cold‐stress‐induced protein RBM3 (RNA‐binding protein 3), which aids in mitigating hypothermic stress, is also translated via an IRES element in its mRNA 169.

Stress and m⁶A‐dependent translation

5′ UTR m⁶A methylation is very dynamic and induced by stress conditions such as heat shock, UV, and by interferon‐gamma 10. During heat shock, nuclear localization of the protein YTHDF2 (YTH domain‐containing family protein 2) aids in preserving methylation of stress‐induced transcripts, by inhibiting demethylation by FTO (fat mass and obesity‐associated protein) 94. Preservation of this m⁶A modification promotes cap‐independent translation initiation of selected mRNAs to relieve heat shock stress. For example, m⁶A modification in the 5′ UTR of HSPA1A mRNA 170 enables its cap‐independent translation initiation 94, in addition to the IRES‐dependent mechanism described above.

Conclusion

The various examples provided here illustrate how different modes of translation are regulated in healthy cells in response to physiological stimuli (Table 1). These mechanisms essentially provide the molecular “toolbox” available for cancer cells to alter gene expression, as described in the next section.

Table 1.

Effect of stress on different modes of translation

Mode of translation	Stress	Example mechanism	References
Canonical cap‐dependent translation	Hypoxia/hypothermia/hyperthermia	p38MAPK activation → MNK1/2 activation → eIF4E phosphorylation → Translation inhibition	115, 116, 117
	Nutrient deprivation	AKT inhibition → mTOR inhibition → 4E‐BP activation → Translation inhibition	162
	Amino acid deprivation	GCN2 activation → eIF2α phosphorylation → Translation inhibition	122
	ER stress	PERK activation → eIF2α phosphorylation → Translation inhibition	151, 267
	Heavy metal toxicity	HRI activation → eIF2α phosphorylation → Translation inhibition	124
uORF translation	Amino acid deprivation	eIF2α phosphorylation → Increased translation of ATF4	131
Alternate start codon usage	Methionine limitation	eIF2α phosphorylation → Elevated alternate start codon usage of oncogenes → cMYC CUG isoform translation	142, 143, 144, 145
IRES‐mediated translation	ER stress	PERK activation → eIF4G2 cleavage → IRES‐dependent translation of HIAP2 → Inhibition of Apoptosis	152, 153
	Hypoxia	AMPK activation → HuR (ITAF) upregulation → IRES‐dependent translation of HIFα → Stress response	159, 160, 161
	Nutrient deprivation	Cytoplasmic relocalization of SMAR1 → IRES‐dependent translation of TP53 and TP53/47 → Stress response	165
	Genotoxic stress	Re‐localization of PTBP1 (ITAF) to the cytoplasm → IRES‐dependent translation of TP53 and TP53/47 → Promotes DNA repair	157
	Hyperthermia	Upregulation of HSPA1A IRES‐dependent translation → Heat shock response	147, 167, 268
m6A‐dependent translation	Hyperthermia	YTHDF2 nuclear localization → Preserves m6A modification of HSPA1A → Increased HSPA1A cap‐independent translation → Heat shock response	94, 96, 170, 269

Translation dysregulation in cancer

In the previous sections, we introduced the many modes of translation to point out that translation is actually a combination of different mechanisms, each of which affects either a subset of mRNAs or even single mRNAs 171. Hence, translation is a key regulated step in the central dogma as information is converted from DNA to RNA to protein 172, 173. Indeed, a recent analysis across 95 colorectal tumors found that genome‐wide alterations in copy number abundance at the DNA level correlate well with changes in mRNA abundance, but changes in mRNA abundance do not correlate well with protein levels 174. Hence, the translation step is key in shaping the proteome of a cell, and hence its phenotype 175, 176. Cancer mutations impact on translation mainly for two reasons. Firstly, many cancer cells are exposed to stresses that cause cap‐dependent translation to be inhibited 14, 177, 178, yet cancer cells require high levels of protein synthesis to grow and proliferate. Hence, the mutations found in cancer cells either boost cap‐dependent translation 179 or activate alternate translation modes to bypass the block in cap‐dependent translation 77. Secondly, various modes of translation are affected by cancer mutations, thereby providing cancer cells with a competitive advantage, either by increasing translation of oncogenes or by reducing translation of tumor suppressors 105, 106. In addition, epigenetic and epitranscriptomic changes could also lead to translation dysregulation in cancer cells 180.

The various types of cancer mutations that have been discovered so far will be detailed below. In the past years, a more detailed understanding of translation dysregulation in cancer has emerged. These developments have broadened the scope of identifying targets of potential prognostic and diagnostic value 181.

Cap‐dependent translation

As mentioned above, cellular stresses present in many cancer cells would normally act to inactivate cap‐dependent translation via inhibition of the eIF4F complex and of the ternary complex. Indeed, increased eIF2α phosphorylation is a prominent feature of many cancer cells (Fig 3i) 77. Hence, some cancers have mutations that counteract this effect by hyperactivating cap‐dependent translation. Overexpression of components of the eIF4F complex is observed in various cancer entities—for instance, eIF4E is overexpressed in endometrial, head and neck, bladder, cervical, and prostate cancers, eIF4A is overexpressed in melanoma and hepatocellular carcinomas, and eIF4B is overexpressed in breast cancers 182, 183, 184, 185. Indeed, overexpression of eIF4E has been shown to transform fibroblasts 186. Phosphorylation of eIF4E at Ser209 is also elevated in some tumors, which increases the affinity of eIF4E for the mRNA cap (Fig 3i) 118, 119. The functional relevance of eIF4E phosphorylation in translation initiation, however, still remains unclear, with some reports finding that increased phosphorylation of eIF4E does not facilitate translation initiation 187, 188, 189, 190, 191, 192. The eIF3 complex contains 10‐13 proteins and is involved in the interaction between the 43S pre‐initiation complex and the eIF4F cap‐binding complex 193. Dysregulation of the eIF3 initiation complex is also involved in oncogenesis 194, 195. Overexpression of eIF3A, 3B, 3C, 3H, 3I, and 3M is observed in several human cancers 196, 197. Overexpression of eIF3A, 3B, 3C, 3H, and 3I causes an increase in global protein synthesis and transformation of NIH 3T3 cells 198. This is accompanied by elevated translation of specific oncogenes such as cyclin D1, c‐Myc, ODC1, and FGF2 (Fig 3i) 198. We refer the reader to very nice reviews summarizing the roles of various canonical initiation factors in tumor initiation and progression which have been published recently 185, 199, 200, 201, 202, 203, 204. In summary, these discoveries pave the way for the discovery of novel anti‐cancer drugs that target the components of the cap‐dependent translation machinery 205, 206, 207, 208, 209.

Figure 3. Translational dysregulation in cancer.

Non‐canonical translational mechanisms found in cancer cells are summarized. Red indicates down‐regulation, and blue indicates up‐regulation.

Cancer cells have also been shown to exploit a novel mechanism of cap‐dependent translation by recruiting alternative cap‐binding factors. Hypoxia, which is often found in cancer cells, stimulates the formation of an alternative translation initiation complex that contains HIF‐2α (hypoxia‐inducible factor 2α), RBM4 (RNA‐binding protein 4), and eIF4E2 210. This complex binds the 5′ cap and promotes translation of a wide array of mRNAs in an eIF4E‐independent manner, thereby circumventing hypoxia‐induced inhibition of protein synthesis and promoting tumor progression 210, 211, 212.

Upstream ORF translation

Upstream ORFs are most common in mRNAs of oncogenes and in mRNAs encoding for proteins involved in cancer‐relevant cellular processes such as differentiation and cell proliferation 213. Cancer mutations can affect the uORF of tumor suppressors and oncogenes, leading to decreased or increased translation initiation, respectively 214. A few examples of uORF regulation in various cancers are described below.

CDKN1B (cyclin‐dependent kinase inhibitor 1B), also known as p27^Kip1, is an important cell cycle inhibitor protein that prevents or slows down cell division. Due to its function in cell cycle inhibition, its expression is often reduced in tumor cells 215. In a recent study, a 4‐bp deletion that modifies the regulatory uORF in the 5′ UTR of the CDKN1B mRNA was discovered in patients with tumors in the pituitary gland or pancreas. This deletion shifts the uORF stop codon, leading to the translation of a longer uORF peptide and shortening of the intercistronic space between the uORF and the main ORF from 429 to 38 base pairs. Owing to these modifications, it is thought that the 40S ribosome cannot keep or re‐acquire the appropriate cofactors for translation initiation on the main start codon, leading to decreased translation of the protein and consequentially increased cell division and tumor progression (Fig 3ii) 216.

Cyclin‐dependent kinase inhibitor 2A (CDKN2A) translation initiation is another well‐described example of uORF dysregulation in cancer. Somatic mutations in CDKN2A have been discovered in a majority of human cancers, and CDKN2A is the second most commonly inactivated tumor suppressor after TP53 217. Two mutations in the CDKN2A mRNA, namely ‐34G to T and ‐21C to T, were identified in melanoma patients. Both of these mutations create a uORF which reduces the translation of the main ORF, leading to reduced cell cycle inhibition and tumor progression (Fig 3ii) 218. Even though until now there are only a few documented examples of mutations altering regulatory uORFs, further identification and characterization of these altered ORFs may be useful in the future as novel biomarkers 214.

Alternative start site recognition

Several recent studies have discovered that during tumor initiation and progression, the translation apparatus recognizes unconventional upstream start sites for enhancing translation of oncogenic proteins 219, 220, 221. One study highlighted the non‐canonical initiation factor eIF2A as the factor responsible for this regulation. eIF2A has been implicated in leucine tRNA recruitment, initiating at a CUG start site 219, 222. Under normal conditions, this alternate initiation factor competes poorly with eIF2α for tRNA recruitment. However, when eIF2α is inhibited, eIF2A becomes more important for translation of certain mRNAs 219. Other studies have found that CUG‐initiated forms of FGF2 (fibroblast growth factor 2) and VEGF (vascular endothelial growth factor) are specifically synthesized in transformed cells, whereas only AUG‐initiated isoforms were found in normal cells. Expression of these isoforms leads to tumor vascularization and progression 220, 221. A nice review summarizing the usage of alternate start codons in cancer has been published 223. Since alternate initiation factors like eIF2A are required to initiate translation from alternate start sites, inhibiting these factors could prove to be a valuable therapeutic intervention in cancer.

Leaky scanning

Although not a lot is known about perturbed leaky scanning mechanisms in cancer, there have been a few cancer mutations discovered that modify leaky scanning and provide an advantage to cancer cells 30. Mutations of CEBPA (CCAAT/enhancer‐binding protein alpha) in patients with acute myeloid leukemia is one such example 224. An internal start codon within the main ORF which is accessed by leaky scanning was discovered. Initiation from this start codon produces a N‐terminally truncated protein of 30 kDa, whereas the full‐length protein translated from the main start codon is 42 kDa 225. In normal cells, a balanced production of both isoforms is maintained by leaky scanning; however, a variety of cancer mutations perturb this balance, increasing leaky scanning and tipping the balance toward the production the 30‐kDa isoform 224. The 30‐kDa isoform, unlike the 42‐kDa CEBPA isoform, does not inhibit cell proliferation 226 and also in some cases has been shown to exert a dominant negative effect toward the longer isoform, thus inhibiting its role in tumor suppression (Fig 3iii) 224.

IRES‐mediated translation and cancer

The translation of many oncogenes that control tumor initiation and progression is tightly regulated by IRESs. We present below selected proteins that are translated predominantly via IRESs that are responsible for tumor suppression, initiation, promotion, and progression (Fig 3iv). For more details and a more complete list of IRES‐containing cancer‐relevant genes, we refer the reader to a recent review 9.

TP53: One of the most established tumor suppressor proteins is TP53. TP53 initiates DNA repair in response to transient damage and induces apoptosis in response to chronic damage 227. Due to these properties, inactivation of TP53 is an important step in tumor progression and is found to be mutated in different types of human cancers 228. Two protein isoforms of TP53 are translated in normal cells from IRES elements (described earlier in the review), both of which play vital roles as tumor suppressors 229. In a recent study, a novel mechanism of TP53 inactivation in cancer cells was discovered. The study found that in various cancer cell lines, the IRES‐dependent translation of TP53 was inhibited due to reduced levels of ITAFs RHA and TCP80 230.

XIAP: One oncogene that is extensively studied for its tumor initiating properties is XIAP (X‐linked inhibitor of apoptosis protein), a protein that inhibits apoptosis. In the class of proteins known to inhibit apoptosis (IAPs), XIAP is the most potent inhibitor of apoptosis 231. Deregulation of XIAP has been linked to many types of human cancer. It is also used as a common marker for tumor cells 232, 233. Since XIAP activity is crucial, its expression levels are tightly controlled. The mRNA of XIAP consists of an exceptionally long 5′ UTR of 1.6 kb and is thought to contain an IRES element 162 nucleotides upstream of the start codon which is able to recruit ribosomes and initiate translation, generating a full‐length XIAP protein 234. In cancer cells, IRES‐dependent translation of XIAP is increased, in part due to increased expression of the ITAF MDM2, which positively regulates XIAP 235. Increased expression of XIAP increases resistance of cancer cells to radiation‐induced cell death 235, 236.

c‐Myc: c‐Myc is a transcription factor that regulates cell growth, cell cycle progression, cell transformation, and apoptosis 237. Even though the half‐life of c‐Myc protein is only 30 min 238, it is still expressed in apoptotic cells where cap‐dependent translation is suppressed. This can be attributed to an IRES element in the 5′ UTR of the c‐Myc mRNA 239, 240. In multiple myelomas, c‐Myc IRES activity is significantly increased due to a C>T mutation in the IRES structure in the 5′ UTR of the mRNA 241. This mutation increases the binding of activated ITAFs, increasing c‐Myc protein 242.

c‐Jun: c‐Jun is a transcription factor that promotes cell cycle progression and prevents apoptosis. c‐Jun activity was extensively studied in glioblastomas and contributes to a large extent to the malignant phenotype 243. In these cancers, c‐Jun protein levels are often found elevated, without an increase in c‐Jun mRNA levels, suggesting an increase in c‐Jun translation. Indeed, this was attributed to a 400‐nucleotide IRES element in its mRNA 243.

c‐Src: c‐Src is a tyrosine kinase that promotes cell proliferation, survival, and invasion 244. Hyperactivation of the c‐Src pathway has been observed in many tumors, especially lung, breast, pancreatic, and colon cancers 245. The c‐Src IRES is one of the most well‐defined IRES elements, which constitutes nucleotides 1–383 of its mRNA, overlapping the main ORF which harbors the initiation codon at nucleotide 351 246. The IRES structure of the c‐Src mRNA recruits the initiation factor eIF2A, which is required for c‐Src translation under stress conditions 247. Assembly of the 80S ribosome on the c‐Src IRES increases under stressed conditions when eIF2α and eIF4E are inhibited. This bypass in canonical initiation leads to less dependence on canonical initiation factors, ultimately leading to continued production of c‐Src protein also in stressed cells, thereby favoring cancer cell survival and proliferation 246.

ITAFs in tumor initiation and progression

ITAFs often relocalize from the nucleus to the cytoplasm in response to stress, and generally function to increase the interaction between the IRESs and translation initiation factors 248. Various ITAFs are thought to act via multiple different mechanisms. In some cases, the ITAFs bind and remodel the IRES increasing or decreasing the affinity for the translation machinery. In other cases, it has been shown that ITAFs act as non‐canonical initiation factors, orchestrating the recruitment of ribosomes for initiation and scanning 150.

ITAFs have also been shown to play a role in cancer. In some cases, the expression or the subcellular localization of various ITAFs has been found to be misregulated in cancer cells 88, 230, 249. ITAF activity could also be misregulated in cancers via post‐translational modifications. For example, AKT represses activity of the cyclin D1 IRES, by phosphorylating and deactivating the ITAF hnRNP A1 250.

Heterogeneous nuclear ribonucleoproteins (hnRNPs) play a vital role in mRNA processing, splicing, stability, and translation. There are various hnRNPs that have been shown to act as ITAFs and to play a role in IRES‐mediated translation 251, 252, 253, 254, 255. For example, hnRNP Q binds to the 5′ UTR of TP53 and regulates its translation 256. The ITAF hnRNP Q1 is overexpressed in colorectal cancer, and when overexpressed, it switches from nuclear to cytosolic localization. hnRNP Q1 binds the IRES element in the 5′ UTR of the Aurora‐A kinase and regulates its expression, which is also consequently up‐regulated in colorectal cancer 252. Another well‐studied hnRNP class protein is the hnRNP A1. In response to stress, hnRNP A1 relocalizes to the cytoplasm from the nucleus. hnRNP A1 specifically binds the region ‐34 to ‐62 upstream of the XIAP start codon and inhibits IRES‐dependent and also cap‐dependent translation of XIAP. This mechanism has been extensively explored in prostate and lung cancers. This is also one of the few rare cases where ITAFs can inhibit mRNA translation (also cap‐dependent), not only in cancers but also in normal cells 257.

Polypyrimidine tract‐binding protein (PTBP1) belongs to the subfamily of heterogeneous nuclear ribonucleoproteins (hnRNPs). PTBP1 has a distinct RNA‐binding capacity which aids in IRES binding, and it is one of the few ITAFs that can shuttle from the nucleus to the cytoplasm under stress 258. PTBP1 is the ITAF responsible for increased c‐Myc expression that results from the cancer C>T mutation in the c‐Myc IRES described above 242, 259.

It is evident that IRES trans‐acting factors play a critical role in tumor initiation and progression (Table 2). Misregulation of ITAFs has been shown to regulate expression of many oncogenic mRNAs involved in proliferation, metabolic remodeling for cell survival, cell cycle regulation, and also migration and invasion of tumor cells 88. By regulating expression of these oncogenes, ITAFs impinge on several hallmarks of cancer 12. The examples presented above suggest that pharmacological interventions that increase or decrease expression or activity of ITAFs could be a potential targeted therapy in certain types of cancers 260, 261, 262, 263, 264, 265. However, since most IRESs that aid in translating oncogenic mRNAs appear to be recognized by single ITAFs 266, one valid avenue of drug therapy could be to disrupt the interaction or recognition of IRESs by their respective ITAF. Recently, several small‐molecule inhibitors have been identified to block translation of IRES‐containing transcripts, without blocking global cap‐dependent translation 261. Understanding the mechanism of action for these inhibitors will aid in uncovering valuable signaling mechanisms.

Table 2.

Role of ITAFs in cancer

ITAF	IRES	ITAF in cancer cells	References
TCP80 and RHA	TP53	Down‐regulated	230
MDM2	XIAP	Up‐regulated	235
hnRNP Q	TP53	Up‐regulated	256
hnRNP A1	CCND1, XIAP, MYC	Activated by phosphorylation	250, 257, 270
hnRNP Q1	AURKA	Up‐regulated	252
YB‐1	MYC	Up‐regulated	242
HuR	HIF1A	Up‐regulated	161

Conclusion

We have illustrated that protein synthesis is not a single process, but rather a combination of multiple different processes that converge at the level of the elongating 80S ribosome, but have distinct initiation mechanisms. These various mechanisms regulate translation of subsets of mRNAs or even individual mRNAs. Hence, both physiological input and cancer mutations can modulate translational mechanisms to sculpt the cellular proteome, and thereby cellular phenotypes. Indeed, cancer mutations have been found which influence some, but not all, of these translational mechanisms. Since this is currently a burgeoning field (see also Box 1), it will be exciting to see if future studies find cancer mutations affecting the remaining translational mechanisms.

Box 1: In need of answers.

1. Which cancer mutations affect translation of cancer‐relevant mRNAs?

   • As discussed in this review, cancer mutations impact translational mechanisms to up‐ or down‐regulate translation of oncogenes or tumor suppressors, respectively. This is likely critical for cancer cells, since protein levels determine cell phenotypes. Unlike mutations that change amino acid sequence, however, the mutations that affect translational cis‐regulatory elements are more difficult to spot. Hence, future work will be needed to identify which cancer mutations affect translation of tumor‐relevant genes.

2. Which alterations in protein synthesis are key?

   • Since thousands of mutations are accumulated in cancer cells, some of these are critical drivers of oncogenesis and tumor progression, whereas others provide slight or no benefits to the tumor cells. Understanding which cancer mutations cause large changes in translation of key target genes will be important to identify which mutations are critical for the disease, and hence targetable for therapy.

3. Understanding the basic mechanisms of non‐canonical translation.

   • As discussed in this review, cancer cells rely on non‐canonical modes of translation for protein synthesis and growth. Unlike canonical cap‐dependent translation, however, the non‐canonical modes of mRNA translation are not yet well understood. Research in the basic mechanisms of mRNA translation, and its regulation, will be key for understanding how these mechanisms are exploited by cancers.

Conflict of interest

The authors declare that they have no conflict of interest.

EMBO Reports (2018) 19: e45947