Translational control of aberrant stress responses as a hallmark of cancer

Transcription

1 Journal of Pathology J Pathol 2018; 244: Published online 20 February 2018 in Wiley Online Library (wileyonlinelibrary.com) DOI: /path.5030 INVITED REVIEW Translational control of aberrant stress responses as a hallmark of cancer Amal M El-Naggar 1,2,3 and Poul H Sorensen 1,2 * 1 Department of Pathology and Laboratory Medicine, University of British Columbia, Vancouver, Canada 2 Department of Molecular Oncology, British Columbia Cancer Research Centre, Vancouver, Canada 3 Department of Pathology, Faculty of Medicine, Menoufia University, Egypt *Correspondence to: Poul H Sorensen, Department of Molecular Oncology, British Columbia Cancer Research Centre, 675 West 10th Avenue, Vancouver, BC V5Z 1L3, Canada. psor@mail.ubc.ca Abstract Altered mrna translational control is emerging as a critical factor in cancer development and progression. Targeting specific elements of the translational machinery, such as mtorc1 or eif4e, is emerging as a new strategy for innovative cancer therapy. While translation of most mrnas takes place through cap-dependent mechanisms, a sub-population of cellular mrna species, particularly stress-inducible mrnas with highly structured 5 -UTR regions, are primarily translated through cap-independent mechanisms. Intriguingly, many of these mrnas encode proteins that are involved in tumour cell adaptation to microenvironmental stress, and thus linked to aggressive behaviour including tumour invasion and metastasis. This necessitates a rigorous search for links between microenvironmental stress and aggressive tumour phenotypes. Under stress, cells block global protein synthesis to preserve energy while maintaining selective synthesis of proteins that support cell survival. One highly conserved mechanism to regulate protein synthesis under cell stress is to sequester mrnas into cytosolic aggregates called stress granules (SGs), where their translation is silenced. SGs confer survival advantages and chemotherapeutic resistance to tumour cells under stress. Recently, it has been shown that genetically blocking SG formation dramatically reduces tumour invasive and metastatic capacity in vivo. Therefore, targeting SG formation might represent a potential treatment strategy to block cancer metastasis. Here, we present the critical link between selective mrna translation, stress adaptation, SGs, and tumour progression. Further, we also explain how deciphering mechanisms of selective mrna translation occurs under cell stress holds great promise for the identification of new targets in the treatment of cancer. Copyright 2018 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd. Keywords: selective mrna translation; microenvironmental stress; stress granules; metastasis Received 15 November 2017; Revised 21 December 2017; Accepted 22 December 2017 No conflicts of interests were declared. Introduction Dynamic interactions between tumour cells and their surrounding microenvironments play critical roles in conferring aggressive tumour behaviour. Tumours are characterized by high rates of growth and metabolism due to genetic alterations that underlie dysregulated growth placing increased metabolic and other demands on tumour cells. While primary tumour cells initially depend on their local microenvironments for pro-growth stimuli, tumours will eventually outgrow their local blood supply, leading to focal and eventually diffuse regions of reduced oxygen and nutrient levels. These and other microenvironmental stresses such as oxidative stress and genotoxic insults are potentially lethal to tumour cells unless they can acutely adapt through the emergence of stress-resilient clones [1 3]. Stress adaptation has previously been ascribed to transcriptional, epigenetic, and mutational changes generating clonal populations with increased capacity for stress adaptation [4,5]. However, emerging evidence suggests that adaptation also occurs through selective translation of major stress adaptive mrnas whose protein products are key to tumour cell survival under stress [6 10]. In fact, proteomic changes through reprogramming of mrna translation appear to constitute a major mechanism of tumour cell adaptation during the stress response [11 13]. Moreover, exposure of tumour cells to stress and the adaptive pathways that are activated to acclimatize to this adverse milieu, such as to preserve energy and ensure survival, can markedly influence their therapeutic response and metastatic potential (Figure 1) [14 21]. In this review, we highlight the critical role of stress adaptation in tumour progression, and how tumour cells hijack specific translational networks to support survival and metastatic capacity. In particular, we discuss selective mrna translation under stress and how it confers tumour cells with aggressive phenotypes, and how this might be exploited therapeutically.

2 Translational control of the tumour stress response 651 Figure 1. Selective mrna translation under cell stress. Tumour cells are continually exposed to various stress forms. Tumour cells respond by altering their transcriptional, epigenetic, and translational programmes to minimize the deleterious effects of these potentially lethal stresses. The process of mrna translational reprogramming represents the quickest defence strategy. To minimize stress-mediated damage, and to conserve energy for initiating prompt repair strategies, cap-dependent translation is blocked to mediate global inhibition of protein synthesis, with only a subset of mrnas still being selectively translated, including those encoding proteins required for repair and adaptation processes, including DNA repair and chaperone proteins, metabolic enzymes, HIF1α, and other stress-adaptive regulators. mrna translation: a central regulator of gene expression under cell stress It is well established that control of mrna translation is central to the coordinated gene expression that is fundamental to cancer development and progression [22 24]. This has been further refined in recent years, utilizing new techniques such as polysome and ribosome profiling as well as mass spectrometry approaches [25 29]. Translation comprises three distinct phases, namely initiation, elongation, and termination; recently, ribosome recycling has been considered as a fourth step in this process (Figure 2A D) [30,31]. Translation initiation Most studies of stress-associated selective translation have focused on the initiation phase. In eukaryotes, translation initiation of most mrnas utilizes an m 7 GTP cap-dependent mechanism in which binding of the translation machinery to the m 7 GTP cap is the critical rate-limiting step. This is driven by availability of the cap-binding protein, eif4e, which recruits eif4g and eif4a proteins to form the so-called eif4f initiation complex at the 5 -UTR cap to initiate translation [31]. Notably, the expression of many eifs is deregulated in cancer. For instance, eif4e is highly expressed and promotes tumour progression and therapy resistance in multiple cancers such as prostate cancer [32], oesophageal squamous cell carcinoma [33], a subset of acute myeloid leukaemias [34], and lymphoma [35]. Further, eif4g is overexpressed in several tumour subtypes and is linked to aggressive tumour behaviour in lung and breast cancer [36,37]. Once formed, the eif4f complex then recruits the 43S preinitiation complex (PIC), consisting of the 40S ribosomal subunit and associated initiation factors such as eif1, along with the ternary complex eif2 GTP Met-tRNAi and eif3, to the cap region. This initiates ribosomal scanning along the 5 -UTR of mrna until an AUG start codon is recognized by this complex, placing the Met-tRNAi in the P site as the first amino acid of the nascent polypeptide. GTP hydrolysis then takes place, releasing the eif2 GDP binary complex and other initiation factors, promoting recruitment of the 60S ribosomal subunit, with assistance from eif5, to form the 80S ribosome which then binds eif5b, marking the start of the elongation phase[38] (see below). Of critical importance, activity of the mrna translational machinery is not constant within tumour cells, but is tightly coupled to cellular energy and oxygen

3 652 AM El-Naggar and PH Sorensen Figure 2. Regulation of mrna translation. (A) In eukaryotes, recognition and binding to the cap domain at the 5 end of mrnas by eif4e marks the beginning of mrna translation initiation. The binding of eif4e is tightly regulated at different levels, with mtorc1 representing a crucial upstream regulator. Then eif4g and eif4a are recruited to the eif4e-cap structure to form the eif4f cap-binding complex. (B) Once formed, the eif4f complex recruits the 43S preinitiation complex (PIC), consisting of the 40S ribosomal subunit and associated initiation factors eif1 and eif1a, along with the ternary complex eif2 GTP Met-tRNAi and eif3, to the cap. (C) The PIC moves along the mrna chain (5 3 ), scanning along the 5 -UTR of mrna until an AUG start codon within a Kozak consensus sequence is recognized by the complex, placing the Met-tRNAi in the ribosomal P site as the first amino acid of the nascent polypeptide. (D) GTP hydrolysis then takes place, releasing the eif2 GDP binary complex and other initiation factors, recruiting the 60S ribosomal subunit, with assistance from eif5, to form the 80S ribosome, which then binds eif5b, marking the start of the elongation phase. The latter is tightly controlled by eef2k. (E) As shown in E, some mrnas, notably stress-adaptive transcripts, have complex stem loop structures within their 5 -UTRs, making them less amenable to translation under ambient conditions. Under stress, these transcripts adopt alternative mechanisms to initiate mrna translation; e.g. the cold shock domain (CSD) family member YB1 binds putative IRES-like elements within the 5 -UTRs, melting their secondary structure through its helicase activity, leading to translational activation. Examples of these stress-associated mrnas include HIF1A, G3BP1, SNAIL, TWIST, and others, as discussed in the text.

4 Translational control of the tumour stress response 653 levels, as well as other factors determining cellular fitness. Indeed, several highly conserved signalling pathways couple overall translation rates to rapid changes in the extracellular milieu, which tumour cells hijack to adapt to stress. These pathways collectively support and orchestrate adaptive responses to stress by restraining overall translation to save energy, while simultaneously stimulating selective synthesis of stress adaptive proteins [13,31]. As is extensively reviewed elsewhere [3,39], upstream activators RAS ERK, PI3K AKT, and MYC family proteins are often dysregulated in human malignancies. These pathways work together to drive mrna translation initiation when conditions are favourable for cell proliferation, and couple overall translation rates to acute changes in the extracellular milieu (reviewed in refs 10, 31, and 40 43). Indeed, the 4EBP1/2 arm of the mtorc1 pathway can itself regulate selective translation, as 4EBP1/2 activation inhibits the translation of messages involved in cell growth [44], translation regulation [45], and mitochondrial functions [46]. PI3K AKT and Ras ERK pathways are stimulated by exogenous growth factors and hormones [41,47], while mtor requires specific amino acids (Leu and Arg) for activation [48], reviewed in ref 49. These pathways work together to activate the mtorc1 complex [50], which induces phosphorylation of 4EBP1/2 and programmed cell death 4 (PDCD4), which are inhibitory binding proteins of eif4e and eif4a, respectively. Inactivation of these factors releases eif4e and eif4a, which can then join the eif4f complex to activate cap-dependent translation initiation [40,51 53]. Activated mtorc1 also induces p90 ribosomal S6 kinase (RSK) to mediate p70 ribosomal S6 kinase (S6K)-dependent eif4b phosphorylation; the latter is essential for enhancing eif4a helicase activity, itself a critical step in translation initiation [40,54 56]. Moreover, the MAPK-interacting kinases 1 and 2 (Mnk1/2), activated by ERK and p38 MAPK, bind to the eif4g carboxy-terminus and phosphorylate neighbouring eif4e at Ser209, which is fundamental for the tumourigenic activity of eif4e [57 60]. Finally, high expression of the MYC transcription factor such as under pro-growth conditions also promotes translation by transcriptionally inducing the expression of many key components of the mrna translational machinery, including eif4e, eif4ai, and eif4g1 [61]. Therefore, enhanced expression of and/or augmented MYC activity in tumour cells can markedly impact mrna translation. When nutrient or O 2 levels are scarce, mtorc1 is inhibited, either directly through reduced amino acids or by upstream pathway blockade, thus inhibiting cap-dependent mrna translation initiation [10,50]. For example, the energy-sensing kinase AMP kinase (AMPK), which is conformationally activated when AMP/ATP ratios are increased, represses mtor signalling by directly phosphorylating and activating the mtor negative regulators TSC1 and TSC2. The latter are GTPase activating proteins (GAPs) of the mtor activator RHEB, and inactivated in many human tumours. Further, AMPK phosphorylates the mtorc1 subunit Raptor, leading to its sequestration by proteins [50,62], and reviewed in ref 63. On the other hand, AMPK is highly expressed in various tumours [64 66] and inhibits mtor activity in tumour cells [67]. This has led to the notion that enhanced AMPK activity provides tumour cells with the ability to shut down mtorc1-mediated translation when necessary, such as under energy-replete conditions [68]. These pathways thus support and orchestrate adaptive responses to stress by restraining overall translation initiation to save energy, while simultaneously stimulating selective synthesis of stress-adaptive proteins, as discussed below [13,31]. Oxidative stress is also linked to attenuation of translation. It was recently shown that loss of NRF2, the master antioxidant response regulator, leads to oxidation of cysteine residues in cysteine-rich translation regulatory proteins in pancreatic cancer cells, resulting in marked attenuation of both cap-dependent and cap-independent translation [69]. Another mechanism to limit translation under stress is through inhibition of eif2, which consists of α, β, and γ3 subunits. Scanning and recognition of the AUG start codon enabled by the eif2 GTP Met-tRNAi ternary complex is a prerequisite step in protein synthesis [70]. The best-established mode of regulation of eif2 is through stress-induced eif2α phosphorylation. The eif2α subunit cycles between active (GTP-bound) and inactive (GDP-bound) forms by the activity of eif2b, which acts as a guanine nucleotide exchange factor (GEF) for eif2α [71]. Under different forms of stress, such as ER stress, eif2α is phosphorylated on Ser51, resulting in sequestration by eif2b and reducing GTP-bound eif2 levels to block translation [72]. Four kinases phosphorylate eif2α, including protein kinase RNA-activated (PKR; activated by double-stranded RNA), protein kinase RNA-like endoplasmic reticulum kinase [PERK; activated by endoplasmic reticulum (ER) stress], general control non-derepressible 2 (Gcn2; induced by amino acid deprivation), and haem-regulated inhibitor (HRI; activated by haem deficiency). Phospho-eIF2α blocks recruitment of the methionyl initiator trna to the 40S ribosomal subunit, thereby blocking assembly of the translation initiation complex [10,30]. While associated with a block in global protein synthesis and apoptosis if sustained [73], eif2α phosphorylation also triggers selective translation of stress-related mrna subsets. For example, under ER stress and as part of the integrated stress response, phospho-eif2α is associated with selective translation of messages required for the synthesis of DNA damage repair proteins, chaperones such as BIP and heat shock proteins, ATF4 and CHOP transcription factors, and other ER stress mitigating factors [74 77]. These mrnas typically contain 5 -UTR upstream open reading frames (uorfs), as will be discussed below. Stress-induced translational reprogramming drives cell plasticity, invasion, and progression in melanoma via ATF4 induction as a result of p-eif2α-dependent eif2b inhibition [78]. Further, genetic signatures associated

5 654 AM El-Naggar and PH Sorensen with eif2b inhibition correlate with poor responses to adoptive T-cell and anti-pd-1 therapy. Therefore, targeting translation reprogramming may overcome resistance to T-cell-mediated therapies [78]. Other regulators of translation initiation under cell stress Several alternative mechanisms of translation initiation have been described that circumvent stress-induced translation inhibition, while boosting the selective synthesis of stress adaptive proteins [13,31,79], and which are exploited by tumour cells to enhance their protection against stress [39,80]. Approximately 10% of cellular mrnas, including many stress-induced mrnas with highly structured 5 -UTR regions, initiate translation through a so-called cap-independent mechanism, thereby bypassing general translation inhibition exerted by environmental cues such as nutrient deprivation or hypoxia [81 83]. First, structural and sequence-specific RNA regulatory elements, particularly in the 5 -UTRs of transcripts, play a critical role in selective mrna translation under stress [22]. One well-known example is the internal ribosome entry site/segment (IRES), first identified in the RNAs of poliovirus (PV) and encephalomyocarditis virus (EMCV) viruses as a mechanism to facilitate translation of viral messages in infected cells [84,85]. IRES elements are present in 5 -UTRs of certain transcripts, for example those encoding the anti-apoptotic protein BCL-2 [86] and the tumour suppressors p53 and p27 [87,88]. These elements lack dependency on the eif4f complex for translation initiation. Instead, augmented by IRES trans-acting factors (ITAFs) such as PTB, YB-1, and Unr, which act as RNA chaperones or adaptor proteins [89], IRES elements directly recruit ribosome subunits and certain eifs, eif2, eif3, eif5, and eif5b, to initiate translation [31,90,91]. In addition to its role in cap-dependent translation, eif3 possesses two RNA recognition motifs (RRMs) as well as other RNA-binding domains, facilitating its direct interaction with IRES elements [92 94]. Interestingly, eif3 shows binding specificity to METTL3-mediated N 6 -methyladenosine (m 6 A) RNA modifications within 5 -UTRs, enabling 43S complex recruitment to initiate translation in an eif4e-independent manner [95]. Notably, the stress-induced molecular chaperone Hsp70, which blocks tumour cell proteotoxic stress, inhibits apoptosis, and modulates tumour cell responses to cytotoxic agents, among many other functions [96], is translated through cap-independent mechanisms, potentially due to increased m 6 A modifications in its mrna [95]. Further, exposure to various cellular stress forms drives widespread redistribution of m 6 A, eventually resulting in an increase in the number of m 6 A-containing mrna transcripts [95]. Targeting the methyltransferase METTL3 in glioblastoma multiforme (GBM) in vitro and in vivo inhibited tumour growth and enhanced the survival of tumour-bearing mice, reflecting its promising utility as a therapeutic target [95,97,98]. Therefore, targeting stress adaptation by reducing m 6 A modifications warrants further investigation in targeted cancer therapy, such as through inhibition of adenosine methylation, which decreases m 6 A-containing mrna transcripts [95,99]. A second mechanism of selective mrna translation initiation under stress is through short uorfs located in the 5 -UTRs of mrnas. It is estimated that over 50% of mammalian mrnas contain uorfs, suggesting critical roles for these structures in regulating translation [ ]. Indeed, uorfs appear to play opposing roles in translation initiation under different conditions. These structures actually inhibit translation under ambient conditions by blocking the ability of ribosome complexes within the translation machinery to scan across the 5 -UTR from the cap to the translation start site of the main ORF [100,103]. However, uorfs enhance translation initiation of a subset of uorf-containing mrnas under stress through a process called reinitiation. Following uorf translation, scanning ribosomes exploit the extended distance spanning between uorf stop codons and the optimal initiation codon of a protein coding sequence (CDS) to become reloaded with initiation factors and the eif2 ternary complex [74,100,104]. This mechanism, a component of the integrated stress response as mentioned above, plays an important role in the translation of certain stress-ameliorating transcripts such as ATF4, CHOP, ATF5, and C/EBPα transcription factors that modulate gene expression in response to stress [100]. Indeed, AUG codons within the optimal sequence for translation initiation, called the Kozak consensus sequence (GCCA/GCCAUGG), notably the nucleotides at positions 3 and +4, mark the strength of initiation codon recognition [105]. For some stress-mitigating transcripts featuring weaker initiation codon context-specific sequences, another mechanism has been employed to ensure their translation, which is called leaky scanning [105,106]. This process supports recognition of the initiation codon irrespective of surrounding sequences, or bypasses the first start codon and instead initiates mrna translation at a downstream initiation codon. For example, this mechanism facilitates translation of mrnas encoding GADD34, which controls dephosphorylation of phosphorylated eif2α by activating the catalytic subunit of the eif2α protein phosphatase 1, PP1c [100,101,107]. Additional mrna structural elements or regulatory motifs have been identified such as RNA G-quadruplexes (G4s), G-rich RNA sequences found in both protein coding sequences and non-coding RNAs [108], which impact translation initiation of tumour-associated messages [109]. Moreover, ribosome composition within translating ribosomes has recently been shown to contribute to the translatability of specific subsets of mrnas, in which ribosomes exploit the heterogeneity of its core components to drive selective mrna translation [110]. Intriguingly, mutations affecting ribosome components, and hence the cells translational landscape, have been implicated in tumourigenesis [3,111,112]. For instance, it has

6 Translational control of the tumour stress response 655 been shown that patterns of ribosomal protein expression are distinct in colorectal carcinoma (CRC) compared with the non-malignant counterpart, with some components being particularly high in CRC [ ]. Further, mutations in the dyskerin (DKC1) gene, associated with the ribosomopathy X-linked dyskeratosis congenita (X-DC), characterized by cutaneous lesions and bone marrow failure, have also been shown to increase susceptibility to carcinomas and haematopoietic malignancies [112]. Translation elongation The elongation phase, in which additional amino acids are continuously added to the nascent polypeptide [116], has only recently received attention in the context of selective mrna translation. In brief, once the initiator Met-tRNA has been delivered to the P site of the ribosome, the latter moves along the mrna codon by codon. The next aminoacyl-trna then comes into the A site on ribosomes to match the corresponding mrna codon with the amino acid that it codes for, to be linked to Met by a peptide bond to extend the growing polypeptide chain in a series of reactions orchestrated by eef1 and eef2 elongation factors [117]. This is repeated until the stop codon is reached, at which point the termination phase is reached and the nascent polypeptide, ribosomes, and mrna are released [117]. It is clear that the rate of translation elongation, and the emerging concept of codon optimality, in which certain codons may be more optimal for elongation under different rates of elongation than other, particularly under different conditions [118], will impact selective translation. Elongation is highly energy-demanding, consuming one ATP and GTP for each additional amino acid added to the growing polypeptide [116]. Moreover, recent studies have linked eukaryotic elongation factor 2 kinase (eef2k) to altered translation under stress. The cytoplasmic enzyme eef2k is a master regulator of translation elongation, as it induces inhibitory phosphorylation of eukaryotic elongation factor eef2, the major rate-limiting driver of translation elongation [ ]. This protein is activated under diverse stresses including nutrient deprivation [7], oxidative stress [124], DNA damage responses [125], and ribosomal stress such as altered ribosomal biogenesis [ ]. Although not proven for each stress form, eef2k activation is presumably linked to reduced elongation and overall translation. In contrast, eef2k is inactivated in cells with defective ribosomal RNA processing, potentially compensating for chronically reduced translation [129]. Growing evidence implicates eef2k in a wide range of pathological conditions including cancer [7,130], neurodegenerative diseases [131], and cardiovascular diseases [132]. Activation of eef2k is tightly regulated by signalling fluxes, which coordinate protein synthesis rates to the energy state of cells. Ca 2+ /calmodulin (CaM) as well as AMPK induces eef2k activation, while mtorc1 impairs its activity [133]. In cancer, eef2k confers survival advantages under energy stress, leading to the emergence of clones that are resistant to nutrient deprivation [7]. Moreover, it has been shown that eef2k is required for adaptation to low availability of nutrients in MYCN-amplified neuroblastoma cells [130]. Further studies to link this kinase to selective translation are therefore warranted. Codon bias and cellular trna pools Newly anointed mechanisms of translation regulation include codon bias and levels of cellular trnas. While synonymous codons can code for the same amino acid, they are non-randomly and differentially expressed, with preferences for particular codons, generating so-called codon bias. Importantly, the decoding rate of synonymous codons is not equal, thus providing another level of translational regulation by adjusting the rate of the elongation phase [134,135]. Contributing to this effect are differences in the availability of cellular trna pools, which dramatically impacts translation rates. Interestingly, using human cell lines and tissues of both normal and malignant counterparts, Gingold et al characterized two trna signatures, one associated with a proliferative programme and another with differentiation, with distinct codon usage linked to each condition [136]. Moreover, recent studies highlight the potential impact of codon bias and trna on cancer progression [137,138]. Expanding roles of RNA-binding proteins in selective mrna translation under cell stress: YB-1 as a paradigm Under changing conditions requiring a rapid cellular response, such as in the face of acute microenvironmental stress, new DNA transcription may fail to deliver a timely molecular response, as it generally takes hours to generate new proteins. In contrast, the availability of ready-to-use pools of translationally competent mrnas poised for translational activation has evolved as an efficient strategy for cells to rapidly respond to their changing environments [139]. So, how can cells rapidly toggle between translational activation and repression of critical mrnas? One mechanism, known for both growth-promoting and stress-inducible mrnas with highly structured 5 -UTR regions, is for mrnas to be bound to proteins as mrna protein complexes (mrnps) [81,82]. Indeed, RNA-binding proteins (RBPs) are emerging as critical determinants of appropriate and timely gene expression under microenvironmental stress, cell differentiation, and cell death [140]. These mrnps can either activate or repress translation. One example is the highly expressed RBP Y-box-binding protein 1 (YB-1). While YB-1 is expressed in a wide variety of normal tissues, enhanced YB-1 expression has been linked to malignant transformation, cell invasion, and drug resistance in a large spectrum of human cancers including breast and prostate carcinoma, glioblastoma multiforme, and

7 656 AM El-Naggar and PH Sorensen sarcomas [ ]. YB-1 is an integral component of mrnps and can bind up to 20% of the total cellular mrnas through its cold shock domain (CSD), which is a highly conserved nucleic acid-binding domain [146]. Effects on translation depend on the YB-1/mRNA ratio for a given message. At high ratios, YB-1 competes with eif4e and eif4g for binding to the 5 -cap structure, therefore inhibiting cap-dependent protein synthesis very early in translation initiation [146,147]. Under stress conditions such as hypoxia or nutrient deprivation, YB-1 globally suppresses cap-dependent translation to block cell growth and proliferation, a crucial energy-conserving response. YB-1 binds and suppresses the translation of many growth-related messages such as D-type cyclins and CDKs, allowing for an appropriate cellular adaptive response to prevail before proliferation is restored [146,148]. Under these conditions, non-translating mrnas are temporarily stored in cytoplasmic structures called stress granules (SGs), as discussed later, for silencing and to prevent degradation, so they can re-associate with polysomes after stress termination to resume active translation [75 77]. In this way, YB-1 acts as a safeguard of repressed mrnas, maintaining their competence and availability, allowing for restoration of their cap-dependent translation under more suitable conditions [81,149]. Regulation of this function occurs at least in part by YB-1 post-translational modifications (PTMs). YB-1 phosphorylation by AKT/PKB at YB-1 residue Ser102 provides a potential mechanism for releasing stored mrnas from its association with YB-1 [149,150], in response to activation of PI3K AKT [81]. However, at lower YB-1/mRNA ratios, YB-1 can enhance the translational efficiency (TE) of specific transcripts, potentially by binding to and melting of RNA secondary structures within their 5 -UTRs, either directly [ ] or via its association with other RNA helicases that can unwind nucleic acids [155]. By melting 5 -UTR stem loop structures, YB-1 allows for more efficient scanning of the translational machinery from the cap to the translational start site. However, IRES-dependent YB-1-mediated translation activation has also been reported [141,156]. Moreover, under certain stress conditions, a fraction of trnas are cleaved by the RNase activity of angiogenin (ANG) into 5 - and 3 -trna fragments called tirnas [157]. This induces SG formation and mrna translation inhibition in a phospho-eif2α-independent manner, by displacing eif4f from the cap, in a process mediated by binding the YB-1 cold shock domain (CSD) to G-quadruplex-like structures within tir- NAs [158]. In our own work, we showed that YB-1 induces an epithelial-to-mesenchymal transition (EMT) in breast cancer cells and increases their metastatic ability in vivo through IRES-dependent YB-1-mediated translation activation of Snail1 and Twist mrnas. YB-1 binds putative IRES-like elements within the 5 -UTRs of SNAIL1 and TWIST mrnas, melting their secondary structure through YB-1-associated helicase activity, leading to their translational activation (Figure 2E) [141]. Notably, physiological processes such as morphogenesis and pathological ones such as EMT both require translational reprogramming to inhibit growth and proliferation, but while enhancing translation mrnas that facilitate cellular migration necessary for each of these processes [159,160]. Moreover, YB-1 translationally activates HIF1A mrnas under hypoxia, increasing HIF1α synthesis and driving childhood sarcoma metastasis [6]. Stable YB-1 knockdown severely reduced sarcoma cell line metastatic capacity in vivo, which was rescued by ectopic HIF1α expression in YB-1-deficient sarcoma cells [6], thus linking both HIF1α and YB-1 to sarcoma progression. YB-1-deficient sarcoma xenografts showed reduced VEGF and microvessel formation, suggesting possible roles for the YB-1/HIF1α axis in sarcoma angiogenesis. We have yet to identify common structural features of SNAIL, TWIST, and HIF1A mrnas that could explain their binding to and translational activation by YB-1, other than complex secondary structures within their 5 -UTRs. It will be important to uncover how YB-1 activates multiple stress-adaptive mrnas. Stress granules: stress-associated mrna guardians Under stress, tumour cells must inhibit global protein synthesis to preserve energy, while maintaining selective translation of mrnas encoding stress-adaptive proteins that facilitate cell survival. One highly conserved mechanism to regulate stress-associated protein synthesis is to sequester mrnas into cytosolic aggregates called stress granules (SGs), where their translation is silenced [161]. SGs are distinct non-membranous cytoplasmic aggregates formed of mrnas (and likely other RNA species), different RBPs including YB-1, other proteins, and the 40S ribosome [162]. These structures are rapidly assembled, both in vitro and in vivo, in cells exposed to various types of stress such as oxidative stress, hypoxia, ER stress, and genotoxic stress (Figure 3A,B). SGs therefore form under diverse conditions that block translation, but only mrnas in which translation has already been initiated are recruited to SGs [161,163,164]. SG formation is in part triggered by phosphorylation of eif2α by eif2α kinases PERK, HRI, GCN2, and PKR. As discussed above, these kinases are activated by different (but potentially overlapping) stress forms to phosphorylate eif2α, thus blocking the initial steps of translation [72]. However, SG assembly can also occur independently of eif2α phosphorylation, largely by unknown mechanisms. One recently described mechanism involves phosphorylation of the critical SG nucleator proteins G3BP1 and its isoform G3BP2 by an as-yet unknown kinase, leading to dissociation of these proteins from USP10 and binding to another RBP, CAPRIN1, somehow acting as a switch for G3BP1/2-mediated SG formation [165]. SG components are highly dynamic structures, with rapid shuttling of RBPs, mrnas, and the 40S ribosome

8 Translational control of the tumour stress response 657 Figure 3. Stress granules and their emerging role in metastasis. (A) SGs are assembled in vitro in response to various stress forms, including oxidative stress (e.g. arsenite), ER stress (e.g. thapsigargin), heat stress, and hypoxia, as detected by immunofluorescence (IF) using antibodies against G3BP (green) and YB-1 (red) SG proteins. Draq5 was used as a nuclear stain (blue). (B) SGs are assembled in vivo in Ewing sarcoma xenografts, as detected by IF using antibodies against G3BP (green) and YB-1 (red) SG proteins. DRAQ5 was used as a nuclear stain (blue). Inhibiting YB-1 expression, a master regulator of mrna translation under stress, inhibits SG formation in vivo, as shown. (C) Top panels: Ewing sarcoma tumour xenografts competent for SG formation, implanted under mouse renal capsules, show highly invasive phenotype marked by loss of the demarcation between the tumour and normal kidney tissue, while xenografts lacking the ability to form SGs due to YB-1 knockdown (KD) show marked inhibition of its invasive phenotype marked by flattened tumour kidney interfaces (arrows). Lower panel: SG competent cells exhibit a highly metastatic phenotype (arrows point to lung metastases), while cells lacking SG formation fail to metastasize. between the cytoplasm and SGs in a process orchestrated in part by microtubules [75,166]. Transcripts recruited to SGs are translationally silenced and stabilized, in contrast to stress-induced cytoplasmic P-bodies (PBs), which also recruit mrnp complexes within the cytoplasm, but in contrast to SGs are associated with mrna Copyright 2018 Pathological Society of Great Britain and Ireland. degradation [163,164]. It remains poorly understood what governs the recruitment of mrnas to either SGs or PBs, but recent studies have begun to shed light on this question [167]. SG formation occurs within minutes of stress exposure, with rapid dissolution once the stress abates, and restoration of mrnas for polysomal J Pathol 2018; 244:

9 658 AM El-Naggar and PH Sorensen association and subsequent translation. This process therefore affords cells with a robust mechanism for rapid and reversible control of mrna silencing under adverse conditions. SGs are also involved as hubs for cellular signalling; for example, SGs are associated with inhibition of the mtor pathway by sequestration of components of mtorc1 complexes [168,169]. Several known SG components include proteins that are those involved in inflammation and apoptotic signalling, such as RACK1, and SGs promote cell survival [170]; interfering with SG assembly is associated with enhanced stress-induced cell death [171,172]. It is well established that many viruses inhibit SG formation through viral proteins that inactivate SG nucleators [85,173]; this likely prevents SG silencing of viral mrna translation. Numerous studies associate altered SGs with human disease. SG protein mutations are reported in multiple pathological conditions, including FUS and TDP43 [174], and TIA-1 mutations in amyotrophic lateral sclerosis (ALS) and frontotemporal dementia [175], and FMRP mutations in fragile X syndrome [176]. Accumulating studies implicate SGs in cancer biology, where they likely confer tumour cell fitness and chemotherapeutic resistance, such as in hypoxic tumour cells [163,164,177]. SGs may be critical for reprogramming translation under unfavourable conditions, such as by sequestering CCND1 and other growth-promoting transcripts to inhibit proliferation, to save energy or prevent accumulation of misfolded proteins. Many SG components are overexpressed in human tumours, where their levels are predictive of clinical outcome [6,8,178]. We recently uncovered an unexpected role for YB-1 in granule (SG) formation. YB-1 binds to G3BP1 5 -UTRs, which also contain predicted stem loops, leading to increased G3BP1 translation under oxidative stress and other adverse cellular conditions [8]. This suggests yet another level of stress-mediated translational control involving RBPs, through enhanced SG formation and translational silencing. Moreover, as discussed in the following section, SG formation correlates with metastatic capacity (Figure 3C). If SGs block translation of sequestered mrnas, then translation of stress-associated mrnas facilitating stress adaptation must by definition be excluded from SGs to maintain the synthesis of proteins essential for stress-adaptive responses [8,161,164]. Indeed, it is known that a subset of translationally active mrnas that are required for the synthesis of DNA damage repair proteins and chaperones, such as heat shock proteins, are excluded from SGs during stresses such ER stress [75 77]. A key deliverable in the field will be to determine what regulates the inclusion or exclusion of mrnas from SGs under stress, as the proteins encoded by excluded messages could represent key targets in clinical oncology. Also critical will be to establish whether SG formation represents a general process for reprogramming translation under diverse stresses of the tumour microenvironment, or if SGs are cytoprotective under more specific conditions. Metastasis: a paradigm of tumour cell adaptation to stress Tumour metastasis is estimated to account for greater than 90% of cancer-related deaths [179]. For sarcomas such as Ewing sarcoma (EWS), modern multi-agent chemotherapy and radiation regimens have markedly improved the outcomes for localized forms of disease, but the prognosis for metastatic disease remains dismal, with almost no improvements in the outcome for the past 20 years [180]. From sequencing studies, it is generally postulated that mutational changes in tumours lead to clonal selection and acquisition of aggressive cellular phenotypes such as chemoresistance and metastatic capacity [98]. However, another (and not necessarily mutually exclusive) view is that stress adaptation is a major contributor to clonal selection, and that this occurs through acute changes in mrna translation and protein synthesis (Figure 4) [181]. This could either favour the survival of genetic clones that more readily adapt to stress, or alternatively, those cells with the capacity to enhance pro-survival translation might then lock in aggressive phenotypes through epigenetic remodelling. It is widely held that metastasis is an inefficient process, with only a tiny fraction of primary tumour cells surviving the many steps of the metastatic process to potentially form distant metastases [182]. The prevailing hypothesis to explain this observation is that metastatic dissemination is a highly stressful process, with each step continuously subjecting tumour cells to heterogeneous environments and different types of cell stress that can prevent metastasis [179]. For example, it was recently shown that oxidative stress and accumulation of reactive oxygen species (ROS) actually prevent melanoma metastasis if cells are not able to mount an anti-oxidant response [183,184]. As mentioned, we recently reported that YB-1 drives SG formation by activating G3BP1 translation. Moreover, this YB-1 G3BP1 SG axis appears to be critical for sarcoma metastasis [8]. Genetically blocking either YB-1 or G3BP1 in sarcoma cells dramatically inhibits SG formation in vitro in response to oxidative stress, and dramatically inhibits the invasive and metastatic capacity of sarcoma and other tumour cells in vivo, indicating that an adequate stress response through SG formation is critical for in vivo tumour invasion and metastasis. Therefore, targeting key players in stress adaptation, such as YB-1 or other components of SG assembly, offers novel strategies for therapy in metastatic disease. Targeting the translational machinery in cancer Dysregulated mrna translation is a common feature in human cancers, and components of the translational machinery are strong candidates for tailored cancer therapies [3,39]. Targeting upstream regulators of the mrna translational machinery has been thoroughly investigated, and extensive numbers of small

10 Translational control of the tumour stress response 659 Figure 4. Stress adaptation and tumour progression. Schematic illustration depicting genetic (bottom) and translational models (top) for tumour metastasis. In addition to the emergence of aggressive tumour cells as a result of accumulation of abnormal genetic events, altered mrna translation under stress, notably at early stages of tumour progression, confers tumour cells with adaptive plasticity and fitness, critical for metastatic spread. molecule inhibitors have emerged and many of them have already made their way to clinical trials. Targeting the mtor pathway with rapamycin and rapalogues as well as mtor kinase inhibitors including Torin is reviewed elsewhere [30, ]. The Ras Erk signalling pathway has also been extensively studied and many clinical trials are ongoing for MEK and RAF inhibitors, as reviewed in refs Similarly, targeting PI3K AKT [30,43,193] as well as MYC pathways [194,195] for cancer therapy has been reviewed elsewhere. Below, we will focus instead on specific examples pertaining to inhibition of the translation machinery. Targeting ternary complex formation Since eif2α phosphorylation inhibits global protein synthesis [74], targeting this process is under investigation as a strategy for cancer treatment. BTdCPU and N,N -diarylureas were found to induce HRI-dependent eif2α phosphorylation preclinically in MCF-7 human breast cancer cells [196]. Salubrinal, a phosphatase inhibitor that represses eif2α dephosphorylation [197], is currently in clinical trials in combination with carfilzomib for patients with multiple myeloma (NCT ). Further, salubrinal in combination with the mtor inhibitor rapamycin shows enhanced anti-tumour activity against cholangiocarcinoma cells [198]. On the other side of the coin, ISRIB (integrated stress response inhibitor) is a small molecule inhibitor identified by cell-based small molecule screens. ISRIB disrupts the interaction between phospho-eif2α and eif2b, central to the integrated stress response (ISR). ISRIB reverses stress-induced mrna translation repression and inhibits stress granule formation, enhancing stress-induced apoptosis as a result of a defective stress response in tumour cells [199,200]. Preclinical studies using a pancreatic ductal adenocarcinoma model showed promising anti-tumour activity for ISRIB [201]. Targeting specific components of the translation initiation machinery eif4e The eif4e protein is highly expressed in various tumours and the list of eif4e-addicted tumours is accumulating [202,203]. Interestingly, eif4e also contributes to translation of viral mrnas which lack a cap structure, and whose translation is driven by cap-independent mechanisms [204,205]. Several strategies have been utilized to target/reduce the levels of eif4e, such as clinical trials using antisense oligonucleotides (ASO) [206], or blocking its interaction with the cap using ribavirin [207,208]. Ribavirin, used for the treatment of hepatitis C, is a cap analogue in clinical trials for the treatment of relapsed and refractory acute myeloid leukaemia exhibiting high levels of eif4e [207,208], as well as for metastatic breast cancer (NCT ). Additional compounds targeting eif4f formation include 4E1RCat [209], 4E2Rcat [210], and 4EGI-1 or its isomers [211], which effectively inhibit persistent translation in breast cancer stem cells and suppress tumour progression in vivo [212,213]. For example, using an Eμ-MYC mouse lymphoma model, 4E1RCat was shown to enhance the response to chemotherapy and, when combined with doxorubicin, prolonged tumour-free survival [209].

11 660 AM El-Naggar and PH Sorensen eif4g Another approach to block the interaction between eif4e and eif4g to inhibit eif4f complex assembly is to directly target eif4g. Recently, screening efforts identified a small molecule inhibitor targeting eif4g1, SBI-756, which was shown to inhibit the growth of chemoresistant melanoma cells harbouring NRAS, BRAF, and NF1 mutations [214]. eif4a This protein is an ATP-dependent DEAD-box RNA helicase recruited to the cap following eif4a eif4g interactions, where it binds eif4g [215,216]. Its helicase activity, augmented by the auxiliary factors eif4b and eif4h, is essential for the unwinding of RNA duplexes [217]. Indeed, eif4a exists in two forms, the more abundant eif4ai and its paralogue, eif4aii. Despite more than 90% identity, each is believed to have distinct functions. While eif4ai is critical for translation, due to its helicase activity directed against 5 -UTRs of many mrnas, eif4aii appears to mediate mirna-mediated gene silencing [218], although this is contentious [219]. A number of eif4a inhibitors targeting its helicase activity, such as hypericin, or functioning as allosteric inhibitors by locking it into an off (inactive) state, such as hippuristanol, have been identified. Further, pateamine A and silvestrol were found to augment eif4a helicase activity and induce dimerization, resulting in eif4a-deficient eif4f complexes lacking translational activity [30,215]. Rocaglamide A (RocA) is another eif4a inhibitor which induces ATP-independent clasping of eif4a onto mrna polypurine sequences, hindering mrna scanning by 43S, resulting in aborted translation [220]. These eif4a inhibitors have shown promising results in preclinical settings [221,222]. Emerging translation regulators for targeted cancer therapies MNK inhibitors MAPK-interacting kinases1/2 (MNK1/2) are tightly linked to mrna translation regulation. By binding the eif4g carboxy-terminus, they are recruited to eif4f to phosphorylate eif4e at Ser209, which enhances eif4e tumourigenic activity [57,58]. Several compounds that inhibit MNK-mediated eif4e phosphorylation have been described, including CGP052088, CGP57380, and cercosporamide. However, these molecules were found to target other kinases in addition to MNK1/2, as reviewed in ref 223. CGP dramatically inhibits eif4e Ser209 phosphorylation in cell lines, while CGP57380 targets eif4g, reducing levels of the eif4f complex, in addition to inhibiting eif4e phosphorylation [224,225]. Cercosporamide also effectively inhibits eif4e Ser209 phosphorylation and shows anti-tumour activity in vitro and in vivo [223,226,227]. Another category of MNK inhibitors is 5-(2-(phenylamino)pyrimidin-4-yl)thiazol-2(3H)-one and its derivatives [228]. However, lack of specificity and development of resistance remain roadblocks for these agents. Several new generation compounds have recently emerged, including eft508 and BAY , which are currently under intense investigation [229]. One of these, eft508, is currently in clinical trials (NCT ) in colorectal carcinoma, in combination with avelumab, an anti-pd-l1 checkpoint inhibitor. Stress granules As mentioned, SGs are assembled under diverse stress conditions and play critical roles in stress adaptation. Further, the link between SGs and metastasis has been recently revealed, as targeting SG formation significantly attenuates sarcoma metastatic capacity [8] and induces apoptosis in tumour cells [177]. One interpretation is that SGs provide tumour cell fitness and emergence of newly aggressive tumour clones. Therefore, inhibiting upstream components regulating SG assembly and/or targeting critical SG nucleators such as G3BP1/2 might be expected to impinge on both cap-dependent and cap-independent translation, and therefore represents exciting new therapeutic strategies. Several chemotherapeutic compounds are reported to increase SGs, including vincristine and related compounds [230] and bortezomib [231], while the microtubule destabilizing drug nocodazole inhibits SGs [232]. It will be important to test the effects of combining SG inducers and inhibitors to assess the potential for synthetic lethality of such combinations. Interestingly, several viruses inhibit SG formation by directly binding and inactivating SG proteins, as SGs otherwise sequester viral mrnas and block their translation [233,234]. Largely unexplored, this could potentially be exploited for targeting SGs in cancer. eef2k As mentioned, eef2k confers survival advantages to tumour cells under energy stress, leading to the emergence of clones that are resistant to nutrient deprivation, particularly in the context of high MUC-expressing tumours. Therefore, targeting eef2k represents a promising strategy for the treatment of cancer. Several eef2k inhibitors have been identified but challenges have been reported, such as a lack of specificity observed for rottlerin, induction of eef2 phosphorylation rather than suppression, as with NH125, or weak activity as with A , all of which are reviewed in ref 133. Therefore, targeting eef2k will likely require new drug-screening efforts or other inhibitory strategies. Summary Central dogma teaches us that gene expression leading to protein synthesis is primarily controlled through regulation of DNA transcription to generate an