59-UTR recruitment of the translation initiation factor eIF4GI or DAP5 drives cap-independent translation of a subset of human mRNAs

Solomon A. Haizel, Usha Bhardwaj, Ruben L. Gonzalez, Jr. , Somdeb Mitra, and Dixie J. Goss* From the Ph.D. Program in Biochemistry, The Graduate Center of the City University of New York, New York, New York, USA, Department of Chemistry, Hunter College, New York, New York, USA, Department of Chemistry, Columbia University, New York, New York, USA, Department of Chemistry, New York University, New York, New York, USA, and Ph.D. Program in Chemistry, The Graduate Center of the City University of New York, New York, New York, USA

Translation of mRNAs into proteins is the most energy-consuming process in the cell (1,2) and plays a major role in the regulation of gene expression. In eukaryotes, initiation of cellular mRNA translation generally occurs via a cap-dependent pathway in which eukaryotic initiation factor 4E (eIF4E) binds to the N7-methylguanosine-triphosphate (m 7 GpppN, where N is any nucleotide) cap at the 59 end of the mRNA to be translated (3,4). Cap-bound eIF4E subsequently recruits eIF4G, which, together with eIF4A, recruits a ribosomal 43S preinitiation complex (PIC) composed of the 40S ribosomal subunit, a methionylated initiator tRNA (Met-tRNA i Met ), and additional eIFs to the m 7 GpppN cap. Subsequent scanning of the resulting 48S PIC to find the AUG start codon on the mRNA and joining of the 60S ribosomal subunit to the 48S PIC results in the for-mation of an elongation-competent 80S IC that can go on to translate the mRNA.
In addition to undergoing cap-dependent initiation, many cellular mRNAs can also initiate translation via cap-independent pathways in response to changes in cellular conditions (5). The ability of these mRNAs to switch from cap-dependent to cap-independent modes of translation initiation plays an important role in maintaining normal cellular physiology (6) as well as in the cellular response to diseases such as cancer, diabetes, and, possibly, neurological disorders (7)(8)(9)(10)(11)(12). A subset of cellular mRNAs, for example, has been shown to successfully bypass a global suppression of translation initiation that is caused by stress conditions such as tumor hypoxia, viral infection, and nutrient deprivation (3,5,13) and that is driven by the sequestration of eIF4E by hypophosphorylated 4E-binding proteins (4E-BPs). Whereas translation initiation of these mRNAs under these conditions is often referred to as cap independent, it may be more accurately described as eIF4E independent. Nonetheless, we will use the term cap-independent here to refer to translation initiation that does not involve eIF4Ebased recruitment of other eIFs to the m 7 GpppN cap. Many of the stress conditions that result in the suppression of cap-dependent initiation also result in increased expression levels of eIF4GI (7,8) and/or death-associated protein 5 (DAP5) (also called p97, NAT1, or eIF4G2) (14,15), suggesting that these two proteins are involved in a cap-independent initiation mechanism(s).
The subset of mRNAs that is translated cap independently under stress conditions as described above are predicted to contain highly stable structures in their 59 untranslated regions (UTRs) that may act as internal ribosome entry sites (IRES) or cap-independent translation enhancers (CITEs) (5,16). IRESlike mechanisms involve direct recruitment of the ribosome to structured IRES that are close to the AUG start codon (5,17), whereas CITE-like mechanisms involve direct recruitment of eIFs to structured CITEs near the 59 end of the mRNA, where the eIFs are then thought to initiate cap-independent translation via 48S PIC scanning to the AUG start codon (18,19). Although chemical and enzymatic probing of cellular mRNAs thought to contain IRESs/CITEs have revealed stem loops, pseudoknots, and other structures (20,21), no common sequence or structural motifs have been identified to allow This article contains supporting information. * For correspondence: Dixie J. Goss, dgoss@hunter.cuny.edu. prediction of cellular IRES/CITEs from mRNA sequence data. Regardless of whether their structured 59 UTRs act as IRESs or CITEs, the ability of these mRNAs to bypass the 4E-BPmediated global suppression of cap-dependent initiation has been linked to enhanced tumor development and cancer progression (2).
The overexpression of eIF4GI and DAP5 during stress conditions in which global cap-dependent translation initiation is suppressed implicates these two factors in cap-independent translation. DAP5 is a member of the eIF4G family that is homologous to the C-terminal two-thirds of eIF4GI (Fig. 1A) and that interacts with other known eIFs in manners that are both similar to and distinct from that of eIF4GI. Specifically, DAP5 and eIF4GI share 39% sequence identity in the central core region that comprises the eIF4A-, eIF3-, and RNA-binding domains (22) (Fig. 1B). Consistent with its similarity to the domain structure of eIF4GI, DAP5 interacts with eIF4A and the eIF3 component of the 43S PIC (22). Notably, however, DAP5 lacks the N-terminal eIF4E-and polyadenine binding protein (PABP)-binding domains that are present in eIF4GI and, consequently, does not interact with eIF4E or PABP (Fig. 1A). In addition, the b subunit of eIF2 (eIF2b) that interacts with the ternary complex formed by eIF2, GTP, and Met-tRNA i (23) binds to the C-terminal domain of DAP5 but not to the corresponding domain in eIF4GI. Collectively, these differences between DAP5 and eIF4GI suggest differences in translation initiation mechanisms involving these two eIFs (14). For example, rather than interacting with eIF4E to recruit factors to the m 7 GpppN cap in cap-dependent initiation of mRNAs, DAP5 has instead been shown to mediate the cap-independent initiation of a subset of mRNAs, including those encoding Bcl2, Apaf-1, p53, and DAP5 itself (24,25). With the exception of p53 mRNA, which contains a structural element that functions as an IRES and has been shown to bind DAP5 using electrophoretic mobility shift assay studies (24), there have not yet been any studies aimed at investigating whether mRNAs translated cap independently directly recruit eIF4GI and/or DAP5 to the 59 UTRs and how such recruitments might drive the switch from cap-dependent to cap-independent initiation in response to cellular stress.
To address these gaps in our understanding, we used a fluorescence anisotropy-based equilibrium binding assay to measure the affinities with which two variants of human eIF4GI, one that lacks the N-terminal, eIF4E-binding domain (eIF4GI 682-1599 ) and one that contains the eIF4E binding domain (eIF4GI 557-1599 ), as well as full-length human DAP5, bind to RNA oligonucleotides corresponding to the 59 UTRs of a representative set of mRNAs. These mRNAs bypass (2) the 4E-BP-mediated global suppression of capdependent initiation and encode HIF-1a, FGF-9, and p53. These mRNAs were chosen because HIF-1a and FGF-9 have been shown to be translationally upregulated in hypoxic conditions where cap-dependent translation is suppressed and where eIF4GI and 4EBPI levels are overexpressed (2,26). p53 mRNAs were chosen because these mRNAs are known to interact with DAP5 as part of a cap-independent translation mechanism (24). Complementing these binding assays, a luciferase-based gene expression reporter assay was used to characterize whether and to what extent binding of eIF4GI 682-1599 , eIF4GI 557-1599 , and/or DAP5 promotes the translation of luciferase-encoding mRNAs containing these same 59 UTRs (UTR-Luc mRNAs) in a rabbit reticulocyte lysate-based cap-dependent in vitro translation system. We further used this assay with UTR-Luc mRNA constructs that either lack or have a highly stable, engineered hairpin (stem-loop) near the 59 cap of the 59 UTR that blocks initiation and scanning from the 59 end. These experiments have further allowed us to investigate whether these mRNAs use an IRES-like or CITE-like mechanism of translation initiation. The results of our experiments demonstrate that eIF4GI 682-1599 , eIF4GI 557-1599 , and DAP5 exhibit differential binding affinities to the 59 UTRs of the mRNAs encoding HIF-1a, FGF-9, and p53; that the eIF4E-binding domain of eIF4GI confers additional binding affinity and specificity; and that binding affinity positively correlates with the abilities of these eIFs to drive translation of these same UTR-Luc mRNAs. Moreover, the inhibition, or lack thereof, of translation initiation by the engineered hairpin suggests that some of these mRNAs use an IRES-like mechanism of translation initiation that does not require an exposed 59 end, whereas others use a CITE-like mechanism (27). Based on our observations, we propose that unlike cap-dependent translation, where binding of eIF4E to the m 7 G-cap is the initial event in protein synthesis, an important initial event in cap-independent initiation of eukaryotic cellular mRNAs with structured 59 UTRs is the binding of eIF4GI or DAP5. These 59 UTR structured mRNAs selectively recruit eIF4GI or DAP5 to drive expression of the proteins they encode, similar to the cap-independent initiation of CITE-or IREScontaining viral mRNAs (28)(29)(30).

Results
eIF4GI and DAP5 bind specifically and with differential affinities to the 59 UTRs of a subset of cellular mRNAs To characterize the binding of eIF4GI and DAP5 to the 59 UTRs of the mRNAs encoding HIF-1a, FGF-9, and p53, we used a fluorescence anisotropy-based equilibrium binding assay developed in our laboratories ( Fig. 2A). This assay produces binding curves from fluorescence anisotropy changes that arise as titrated proteins bind to RNA oligonucleotides that are covalently 59-end labeled with fluorescein and that lack a m 7 GpppN cap. Four RNA oligonucleotides comprising the 59 UTRs of mRNAs encoding HIF-1a, FGF-9, and the two 59 UTRs of p53 were assayed. The two p53 59 UTRs represent the two 59 UTRs involved in translation of the two distinct isoforms of p53. Each p53 59 UTR contains elements that select for translation initiation at either an upstream start codon (p53 A ) or a downstream start codon (p53 B ), which produce the full-length p53 (FL-p53) or an N-terminal-truncated isoform (DN-p53), respectively (31). When eIF4GI 682-1599 , eIF4GI 557-1599 , or DAP5 binds to one of the fluorescein-labeled 59 UTRs, there is an increase in molecular weight of the 59 UTR that increases the rotational restriction as the system becomes more rigid, resulting in a decrease in the rotational speed of the fluorescein-labeled 59 UTR. The slower rotation is observed as increased fluorescence anisotropy of the fluorescein reporter. To quantify the affinity of eIF4GI 557-1599 , eIF4GI 557-1599 , and DAP5 binding to the four 59 UTRs, we recorded the change in the fluorescence anisotropy of each fluorescein-labeled 59 UTR as a function of increasing concentrations of eIF4GI 682-1599 , eIF4GI 557-1599 , and DAP5 and fitted the resulting data points with a single-site, equilibrium-binding isotherm (Fig. 2B, C, and D). The results of these experiments (Table 1) demonstrate that eIF4GI 682-1599 , eIF4GI 557-1599 , and DAP5 bind to the four 59 UTRs with equilibrium dissociation constants (K d s) ranging from 12-290 nM ( Table 1).
Comparative analyses of our results demonstrate that eIF4GI 682-1599 , eIF4GI 557-1599 , and DAP5 exhibit differential binding affinities among the 59 UTRs. Specifically, eIF4GI 682-1599 binds to the p53 A 59 UTR with a K d that is ;2-fold higher (i.e. binding that is ;2-fold weaker) than that with which it binds to the other 59 UTRs (Table 1). DAP5 exhibited an even greater difference in binding affinity among the 59 UTRs, with a K d for the p53 B 59 UTR that is more than 2.5-fold lower than the K d for the p53 A 59 UTR. It is important to note, however, that the trend of the differences in binding affinities between the two translation factors (eIF4GI 682-1599 and DAP5) for the 59 UTRs is similar. However, dramatic differences in binding among the 59 UTRs were observed for eIF4GI 557-1599 . Inclusion of the eIF4E-binding domain in eIF4GI 557-1599 relative to eIF4GI 682-1599 increased the binding affinity of eIF4GI 557-1599 from 1.8-fold (p53 B 59 UTR) to 10-fold (FGF-9 59 UTR) compared with those measured for eIF4GI  . Further, the differences in binding affinity among the 59 UTRs were as much as 7-fold different compared with an ;2-fold difference among the 59 UTRs for eIF4GI  . Taken together, these results suggest much of the specificity of eIF4GI binding to the 59 UTRs is conferred by the eIF4E-binding domain (i.e. residues 557-682) of eIF4GI.
The binding of eIF4GI and DAP5 is specific to the 59 UTRs of the selected mRNAs To further investigate whether binding of eIF4GI and DAP5 to RNA depends on structural features within the RNA and to contextualize and validate the results of the binding studies described above, we assessed the binding of eIF4GI 682-1599 , eIF4GI 557-1599 , and DAP5 to a presumably unstructured, 101nucleotide poly(UC) RNA oligonucleotide and an oligonucleotide encompassing the 59 UTR of the mRNA encoding b-actin, which has been reported to utilize a cap-dependent mechanism for translation initiation (32). The results of these experiments demonstrate that eIF4GI 682-1599 , eIF4GI 557-1599 , and DAP5 do not exhibit appreciable binding to the polyUC oligonucleotide ( Fig. 3A, B, and C and Table S1) or to the b-actin 59 UTR, findings consistent with the hypothesis that structural features within the 59 UTRs of our subset of RNAs act as specific recognition elements and binding sites for eIF4GI and DAP5.
Previously, we have shown that eIF4F (a complex composed of eIF4GI, eIF4A, and eIF4E) binds to the 30-nucleotide ironresponsive element (IRE) stem-loop within the 59 UTR of the mRNA encoding ferritin with a K d of 9 nM, a binding interaction that stimulates ferritin mRNA translation in response . Briefly, 10 nM (B) or 100 nM (C and D) fluorescein-labeled uncapped RNA oligonucleotides were titrated with increasing concentrations of eIF4GI 557-1599 , eIF4GI 682-1599 , or DAP5 in the titration buffer at 25°C, and the anisotropy at each titration point was measured using excitation and emission wavelengths of 495 nm and 520 nm, respectively. Data points correspond to the average from three independent anisotropy measurements, and the curves represent the nonlinear fits that were used to obtain the averages and standard deviations for the corresponding K d values presented in Table 1.  eIF4GI or DAP5 drives cap-independent translation initiation to elevated cellular concentrations of iron (33). Using the fluorescence anisotropy-based equilibrium binding assay, here we show that eIF4GI 682-1599 , eIF4GI 557-1599 , and DAP5 bind to the IRE with K d s of 18 nM, 15 nM, and 35 nM, respectively ( Fig. 3 and Table S1). Similarly, we performed experiments in which we quantified the affinity of eIF4GI 682-1599 , eIF4GI 557-1599 , and DAP5 for the J/K domain of the IRES in the 59 UTR of the positive-strand genomic RNA, encoding the encephalomyocarditis virus (EMCV) polyprotein ( Fig. 3 and Table S1). These experiments show that eIF4GI 682-1599 , eIF4GI 557-1599 , and DAP5 bind to the EMCV J/K IRES with K d s of 175 nM, 63 nM, and 519 nM, respectively. The K d for binding of eIF4GI 682-1599 to EMCV J/K IRES RNA is similar to what has been previously reported for the binding of human eIF4GI (643-1076) to the EMCV J/K IRES using electrophoretic mobility shift assay (170 nM [34]) as part of the mechanism through which eIF4GI drives expression of the EMCV polyprotein in human cells (35). The same report showed that a construct containing an N-terminal portion of DAP5 (62-330) did not bind to the EMCV J/K IRES, consistent with our much higher K d (519 nM) for full-length DAP5 than for eIF4GI construct binding.

The 59 UTRs of a subset of cellular mRNAs that bind eIF4GI and DAP5 can drive cap-independent translation
Having demonstrated that eIF4GI and DAP5 bind specifically and with relatively high affinity to the 59 UTRs of HIF-1a, FGF-9, p53 A , and p53 B , we examined the cap-independent translation of a set of reporters containing these 59 UTRs. To quantify the cap-independent activities of our selected mRNAs, the 59 UTRs of the HIF-1a, FGF-9, p53 A , and p53 B mRNAs and, as a control, the 59 UTR of b-actin mRNA were cloned upstream of a luciferase reporter gene (Fig. 4A). The resulting capped, polyadenylated UTR-Luc mRNAs were translated using a nuclease-treated, rabbit reticulocyte (RRL)-based, capdependent, in vitro translation system containing a natural abundance of the eIFs. To assess cap-independent luciferase expression, a nonfunctional cap analog (ApppG) was used to cap the UTR-Luc mRNAs (here referred to as the ApppGcapped transcripts). The cap-independent expression levels of the ApppG-capped transcripts were compared with the expression observed when these transcripts were capped with a functional m 7 GpppA cap. We observed that the activities of the ApppG-capped transcripts ranged from ;70% for FGF-9 UTR-Luc mRNA to 8% for p53 B UTR-Luc mRNA compared with their corresponding m 7 GpppA-capped transcripts (Fig. S1). Expression of the ApppG-capped b-act-UTR-Luc mRNA was reduced to less than 2% of the corresponding m 7 GpppAcapped construct (Fig. S1A). These results further support the notion that the 59 UTR of a subset of cellular mRNAs can drive cap-independent translation, albeit at a relatively lower efficiency than cap-dependent translation. During stressful conditions in which cap-dependent translation is inhibited and the expression of eIF4GI and DAP5 are elevated, expression from these 59 UTRs can help cells mitigate the effects of the stressor (2).
The eIF4E-eIF4GI interaction allows recruitment of the eIF4F complex to the m 7 G cap of mRNAs (36). 4EGI-1 is a small molecule that binds to the same site on eIF4E that interacts with eIF4GI and inhibits the interaction of eIF4GI and eIF4E (37,38). To further support the idea that translation . Briefly, 10 nM (A) or 100 nM (B and C) fluorescein-labeled uncapped RNA oligonucleotides were titrated with increasing concentrations of eIF4GI 557-1599 , eIF4GI 682-1599 , or DAP5 in the titration buffer at 25°C, and the anisotropy at each titration point was measured using excitation and emission wavelengths of 495 nm and 520 nm, respectively. Averages from 3 independent experiments were performed, and the curves represent the nonlinear fits that were used to obtain the averages and standard deviations for the corresponding K d values given in Table S1. eIF4GI or DAP5 drives cap-independent translation initiation initiation of these mRNAs was independent of eIF4E, we determined the extent to which translation of the m 7 GpppA-and ApppG-capped transcripts was affected by 4EGI-1. Cap-dependent initiation was significantly suppressed through the addition of 4EGI-1 to m 7 GpppA-capped b-act-UTR-Luc mRNA but had no effect on the ApppG-capped transcripts of our selected 59 UTRs, with the exception of the ApppG-capped HIF-1a transcript, which exhibited an ;44% reduction in translation (Fig. S2). These results show that, to a large extent, translation of the ApppG-capped transcripts proceeds via an eIF4E-independent mechanism. The observation that the ApppG-capped HIF-1a transcript expression is partially reduced suggests that eIF4E does not function exclusively during cap-dependent translation initiation but also plays an important role in the capindependent translation initiation of some mRNAs. For example, eIF4E has been shown to stimulate the helicase activity of , and DAP5 (bar 9-10). Relative luciferase activity was normalized to the respective controls (ApppG-capped UTR-Luc mRNA) for each reporter performed in nondepleted RRL. Bar heights and error bars correspond to the average and standard deviations, respectively, from three independent luciferase activity measurements. Data were analyzed by two-tailed unpaired Student's t test: **, p = 0.002; ***, p , 0.001. eIF4GI or DAP5 drives cap-independent translation initiation eIF4A in the case of some highly structured 59 UTRs (39), such as HIF-1a, and/or to induce conformational changes in eIF4G that enhance its binding to mRNAs (40). eIF4GI and DAP5 stimulate and restore cap-independent in vitro translation of the selected 59 UTR-Luc mRNAs Having established that eIF4GI and DAP5 bind specifically and with relatively high affinity to the HIF-1a, FGF-9, p53 A , and p53 B 59 UTRs and that the corresponding UTR-Luc mRNAs are translated through a cap-independent pathway, we sought to establish the extent to which this cap-independent translation initiation of these mRNAs depended on eIF4GI and DAP5. To accomplish this, we used our luciferase-based reporter assay and first determined the effects on translation of depletion of either eIF4GI or DAP5. We measured the luciferase activity produced by each ApppG-capped transcript in an RRL that had been depleted of either endogenous eIF4GI [RRL (-)4GI ] or DAP5 [RRL (-)DAP5 ] using antibodies directed against these proteins. Western blots confirmed the successful depletion of eIF4GI or DAP5 (Fig. 4B) from the RRLs. As a control to test for the indirect depletion of other eIFs, eIF4E, eIF2b, and eIF4A levels were tested in RRL (-)4GI or RRL (-)DAP5 and were found to be similar to, or only slightly reduced from, those seen in RRL (Fig. S3). Using either RRL (-)4GI or RRL (-)DAP5 , we measured the luciferase activity produced by each translation reaction of the ApppG-capped transcript and compared it to RRL translation assays of the same transcript without depletion of endogenous eIF4GI or DAP5. To confirm the eIF4GI or DAP5 dependence on the translation initiation of these UTR-Luc mRNAs, we added increasing concentrations of exogenous eIF4GI 557-1599 , eIF4GI 682-1599 , or DAP5 to the corresponding depleted RRL. Translation of UTR-Luc mRNAs was significantly reduced in the depleted RRLs compared with those of the RRL control (Fig. 4, C-F). Specifically, the translation output of HIF-1a, FGF-9, p53 A , and p53 B ApppG-capped UTR-Luc mRNAs were reduced by 70%, 67%, 51%, and 70%, respectively, in RRL (-)4GI (Fig. 4, C-F). Addition of eIF4GI 557-1599 or eIF4GI 682-1599 significantly rescued and stimulated the cap-independent translation initiation of all four UTR-Luc mRNAs (Fig. 4, C-F). For RRL (-)4GI , eIF4GI 557-1599 restored translation of the ApppG-capped transcripts to ;85-100% of the levels observed in RRL, whereas eIF4GI 682-1599 was slightly less effective, restoring levels to ;70-90% of the RRL levels for the same transcripts (Fig. 4, C-F). In the case of RRL (-)DAP5 , the translation outputs of HIF-1a, FGF-9, p53 A , and p53 B ApppG-capped UTR-Luc mRNAs were reduced by 83%, 67%, 63%, and 81%, respectively. Addition of DAP5 was also able to rescue cap-independent translation initiation of all four UTR-Luc mRNAs, although with somewhat less efficiency than the eIF4GI constructs. DAP5 restored translation to ;45-65% of the levels for the same transcripts in RRL (Fig. 4, C-F). These data follow the same overall trend as our fluorescence anisotropy-based binding assay data, where eIF4GI 557-1599 showed the highest binding affinity and DAP5 the lowest affinity to the 59 UTRs (Table 1).
Further demonstrating the cap dependence of translation, the ApppG-capped b-act-UTR-Luc mRNA showed less than 2% of the translation output of that of the corresponding m 7 GpppA-capped transcript (Fig. S2). In agreement with our binding results, eIF4GI 682-1599 , eIF4GI 557-1599 , and DAP5 did not show any significant stimulation of the cap-independent translation of this ApppG-capped b-act-UTR-Luc mRNA (data not shown). Significantly, although depletion of eIF4GI and DAP5 from the lysate decreased the translation of m 7 GpppAcapped b-act-UTR-Luc mRNA to ;18 and ;67%, respectively, only the eIF4GI 557-1599 construct containing the eIF4E binding domain rescued its translation (Fig. S4), results that are consistent with the previous and widespread use of this mRNA as a control for cap-dependent initiation and translation (32). As expected, neither eIF4GI 682-1599 nor DAP5, which lack the eIF4E binding domain, rescued the translation of the control m 7 GpppA-capped b-act-UTR-Luc mRNA (Fig. S4), as demonstrated by the absence of any significant change in luciferase activity, indicating the specificity of the stimulation.
An exposed 59 end is important for the cap-independent translation activities of HIF-1a and p53 A UTR-Luc mRNAs but not FGF-9 and p53 B UTR-Luc mRNAs To gain a better mechanistic understanding of the cap-independent translation initiation of our selected mRNAs, we introduced a stable hairpin at the 59 end of our mRNAs (Fig. 5A). This same hairpin at this same position previously was used to block scanning from the exposed 59 end of an mRNA while having no effect on the internal translation initiation mediated by an IRES (25,27,41). We found that the hairpin structure significantly repressed translation of m 7 GpppA-capped b-act-UTR-Luc mRNA (Fig. 5B), which served as a positive control for inhibition of 59-end scanning-dependent translation initiation. Comparative analyses of our selected UTR-Luc mRNAs showed that the hairpin did not affect the translation initiation activities of FGF-9 UTR-Luc mRNA (Fig. 5D) and p53 B UTR-Luc mRNA (Fig. 5F), suggesting that internal initiation occurred on these UTR-Luc mRNAs. In contrast, the translation efficiencies of HIF-1a Luc mRNA (Fig. 5C) and p53 A UTR-Luc mRNA (Fig. 5E) were significantly reduced by the hairpin, suggesting that translation of these UTR-Luc mRNAs required an exposed 59 end for translation initiation and that these 59 UTRs contain structural elements that act as CITEs rather than IRESs.

Discussion
In this study, using a fluorescence anisotropy-based equilibrium binding assay and a luciferase-based gene expression reporter assay, we have demonstrated that eIF4GI and DAP5 directly bind to and stimulate cap-independent translation initiation of a subset of cellular mRNAs. The truncated eIF4GI proteins, eIF4GI 557-1599 and eIF4GI 682-1599 , as well as DAP5 bind to the 59 UTRs of this subset of mRNAs with K d s in the range of 12-290 nM ( Fig. 2 and 3 and Table 1) and stimulate cap-independent translation of the corresponding UTR-Luc mRNAs by 2-to 5-fold over lysates depleted of either eIF4GI or DAP5 (Fig. 4, C-F). These K d s are comparable with those observed for the binding of eIF4GI constructs to the EMCV J/K IRES ( Fig. 3 and Table S1), a well-characterized system in which it has been previously shown that binding of truncated eIF4GI or DAP5 drives cap-independent translation initiation eIF4GI 643-1076 to a well-defined structural feature within the 59 UTR of EMCV mRNA stimulates the cap-independent translation of this mRNA (34). In line with cellular data (2, 15), our observations strongly suggest that both eIF4GI and DAP5 recognize and bind to specific structural features within the 59 UTRs of our selected mRNAs. Moreover, the observation that eIF4GI 682-1599 and DAP5 bound to our selected 59 UTRs with similar binding affinities ( Fig. 2 and Table 1) suggests that eIF4GI and DAP5 recognize similar elements within the various 59 UTRs. Furthermore, the intriguing differences between the K d s for the larger construct of eIF4GI 557-1599 and DAP5 (Table 1) suggest that, to bring about translational outcomes, cellular mRNAs with structured 59 UTRs may preferably recruit eIF4GI, depending on its cellular availability, which may change according to the type of stress conditions, cell type, or stability of the proteins (2,42,43).
Several reports have demonstrated that a subset of cellular mRNAs possess IRES-like elements in addition to the cap structure, and that these mRNAs act to recruit ribosomal PICs and, presumably, key eIFs internally to the 59 UTR of these mRNAs to initiate cap-independent translation (2,5,44). However, no common structural motifs were identified among the cellular IRES elements, and, compared with viral IRES, cellular IRES elements appear to be much more diverse and less stable in terms of Gibbs free energy of folding (45). In contrast to these reports, it has recently been proposed that other cellular eIF4GI or DAP5 drives cap-independent translation initiation mRNAs recruit ribosomal PICs in a cap-independent manner to the 59 end using a CITE that recruits eIFs, and that the ribosomal PIC then scans from the 59 end rather than using an IRES to initiate translation internally (19). Given the fact that some 59 UTRs may possess either IRESs and/or CITEs, we were prompted to test the hypothesis that the exposed 59 end of our selected 59 UTRs is required for translation. Our data showed that the presence of a highly stable, scanning-inhibiting hairpin (41,46) at the 59 end of the FGF-9 and p53 B ApppG-capped hp-UTR-Luc mRNAs did not affect the absolute translation levels of these mRNAs (Fig. 5), demonstrating that they were cap-independently translated, presumably using IRES-like mechanisms. In contrast with this, the hairpin at the 59 end of the HIF-1a and p53 A ApppG-capped hp-UTR-Luc mRNAs repressed the cap-independent translation initiation activities of these UTR-Luc mRNAs (Fig. 5), suggesting that accessibility of the 59 end of these cellular mRNAs is necessary and, correspondingly, a CITE-like mechanism is important for the initiation of these mRNAs. Because the subset of cellular mRNAs we have investigated here employ different systems of cap-independent translation, we speculate that this phenomenon provides additional regulation for the expression of these mRNAs.
Our findings are reminiscent of the manner in which viruses use highly structured IRESs to directly recruit eIFs, the 40S subunit, and/or the 43S PIC to drive the cap-independent translation of viral mRNA (5,47,48). Although the mechanisms through which viral mRNAs use IRESs to drive cap-independent initiation have been extensively characterized using genetic, biochemical, and structural approaches (16,49,50), the mechanisms through which eukaryotic cellular mRNAs drive cap-independent initiation remain largely unknown. Nonetheless, an increasing body of evidence suggests eukaryotic cellular mRNAs can employ noncanonical initiation mechanisms that are distinct from those employed by IRES-containing viral mRNAs. For example, Cate and coworkers have recently shown that the d subunit of eIF3 (eIF3d) targets mRNAs encoding proteins involved in cell proliferation and serves as a transcriptspecific, cap-binding protein (51). Using the mRNA encoding the c-Jun transcription factor as an example, these authors showed that, under conditions in which an RNA structure in the 59 UTR blocks eIF4E from binding to the 59 cap, eIF3d binds directly to the 59 cap and serves as an alternative capbinding factor (52). Even more recently, it has been shown that DAP5 is a direct binding partner of eIF3d, and it has been proposed that DAP5 uses eIF3 to direct the eIF4E-independent, cap-dependent translation of a subset of cellular mRNAs when cellular stress conditions lead to the inactivation of eIF4E (53). In addition to these mechanisms in which RNAs utilize IRES elements, eIF3d, or other eIFs that function as alternative capbinding proteins, the methylation of adenosine residues in the 39 and 59 UTRs of eukaryotic cellular mRNAs have been shown to stimulate translation by an unknown mechanism (54,55). The cellular IRES-and CITE-based mechanisms we have investigated here are distinct from these other types of mechanisms that make use of an alternative cap-binding protein or posttranscriptional modification of the mRNA to be translated. Here, eIF4GI or DAP5 completely bypasses any cap-dependent processes and is instead directly recruited to IRESs or CITEs within the mRNAs. Indeed, because the binding studies we present here employ purified eIF4GI 557-1599 , eIF4GI 682-1599 , and DAP5 and uncapped RNA oligonucleotides, our results show these translation factors can bind directly to 59 UTRs of select mRNAs in a completely cap-independent manner and without the need of alternative cap-binding proteins or mRNA methylation. Similarly, because the gene expression assays we present here also employed purified mRNAs with nonfunctional cap analogs and are performed in RRL, RRL (-)4GI , and RRL (-)DAP5 , we can be confident that the stimulation of translation we observe is because of the direct interaction of eIF4GI or DAP5 with the IRES or CITE elements in the 59 UTRs of our selected mRNAs rather than to the indirect effects of alternative capbinding proteins or methylation of the mRNA.
Based on our observations, we propose that an important initial event in cap-independent initiation of eukaryotic cellular IRES-or CITE-containing mRNAs is binding of eIF4GI or DAP5, similar to the cap-independent initiation of CITE-or IRES-containing viral mRNAs (28)(29)(30). Specifically, we propose that eIF4GI or DAP5, perhaps with the aid of additional factors, bind to elements within the 59 UTRs and subsequently recruit additional eIFs, the 40S subunit, and/or the 43S PIC to these mRNAs. In analogy to the cap-independent initiation of several CITE-containing viral mRNAs (28,30,56), we propose that the resulting 48S PIC scans to the AUG start codon. Our data suggest that this is the case for the HIF-1a and p53 A mRNAs. Alternatively, these factors may be bound to elements near the AUG, acting as IRES-like structures, as appears to be the case for FGF-9 and p53 B (Fig. 6). It is not surprising that multiple mechanisms and eIF requirements may be present, with different and possibly dynamic RNA structures in the 59 UTR having different requirements. This is certainly the case for viral capindependent translation (48).
The transition from cap-dependent to cap-independent translation of cellular mRNAs under stress conditions has important physiological consequences, particularly in disease states, such as diabetes, cancer, and possibly neurological disorders (7)(8)(9)(10)(11)(12). Under stress conditions, the levels of 4EBP1, eIF4GI, and DAP5 are elevated (2,7,8,14,15). Here, we have identified key interactions of eIF4GI and DAP5 through which cellular mRNAs containing either IRESs or CITEs may bypass the 4EBP1-mediated depletion of eIF4E and undergo cap-independent initiation and expression. Identification of specific interactions that facilitate the transition from cap-dependent to cap-independent translation promises to facilitate a better understanding of noncanonical translation mechanisms and to inform the potential development of novel therapies that modulate this transition to regulate the cellular response to stress conditions.  (Table S2) were purchased from Integrated DNA Technology (IDT), and the corresponding RNAs were synthesized via in vitro transcription using the HiScribe TM T7 quick high-yield RNA synthesis kit (New England Biolabs, Inc.) by following the manufacturer's protocol. RNAs from transcription reactions were purified using the RNA Clean and Concentrator kit from Zymo Research by following the manufacturer's protocol. Purified RNA transcripts were labeled with fluorescein at their 59 termini using the 59 EndTag DNA/RNA labeling kit from Vector Laboratories by following the manufacturer's protocol. RNA concentrations were determined using a nano-drop UVvisible spectrometer, and integrity was verified by 1.5% agarose gel electrophoresis.

Experimental procedures
Preparation of eIF4GI 557-1599 , eIF4GI 682-1599 , and DAP5 The plasmids for expression of eIF4GI 557-1599, and eIF4GI 682-1599 were a generous gift from Dr. Christopher Fraser (University of California at Davis). Full-length eIF4GI purified from cells as part of the eIF4F complex is easily degraded during purification and is contaminated with various amounts of eIF4E and/or eIF4A. Therefore, we used the stable, functional constructs described here (59)(60)(61). The codon-optimized eIF4GI 682-1599 construct (34) includes the minimal sequence for IRES-mediated cap-independent translation initiation (31) and is similar in domain structure to full-length DAP5. This construct has an N-terminal, 63 histidine tag followed by a Flag tag in the pET28c vector. We introduced a tobacco etch virus (TEV) protease cleavage site following the Flag tag using the Q5 ® sitedirected mutagenesis kit from New England Biolabs, Inc., by following the manufacturer's protocol. The eIF4GI 557-1599 construct was in a fastback vector. It was PCR amplified using a forward primer containing an NcoI site and a 63 histidine tag followed by a TEV protease site and a reverse primer containing an XhoI site. The amplified PCR product was subcloned in pET28c at the NcoI-XhoI site and used to express and purify the eIF4GI 557-1599 protein. The plasmid encoding full-length human DAP5 with an N-terminal 63 histidine tag was purchased from GenScript (Piscataway, NJ). All of the proteins were recombinantly expressed in Escherichia coli BL21-CodonPlus (DE3)-RIL Figure 6. Cartoon summarizing models for canonical (cap-dependent) and proposed cap/eIF4E-independent translation initiation of the mRNAs containing structured 59 UTRs. The top panel depicts normal physiological conditions where activated mTOR (mTOR) hyperphosphorylates 4EBP1, making it inactive. eIF4E binds to the m 7 G-capped mRNA to recruit other eIFs and the 43S PIC to initiate canonical cap-dependent translation. In the lower panel, inactive mTOR leads to hypophosphorylation of 4EBP1 which it blocks eIF4E's interaction with eIF4GI. The 4EBP1-eIF4E complex may interact with the m 7 G-capped mRNA but is unable to interact with eIF4GI. Our model shows eIF4GI or DAP5 binding the 59 UTR and recruiting the 43S PIC by either a CITE-like mechanism or an IRES-like mechanism. eIF4GI or DAP5 drives cap-independent translation initiation cells (Agilent) and were purified using a combination of Ni 21nitrilotriacetic acid (Ni-NTA) affinity and heparin affinity columns, as previously described (25,34). Briefly, the proteins were first purified from bacterial cell lysates using His-Trap HP (Ni-NTA) columns (GE Healthcare Life Sciences) per the manufacturer's instructions. The purified 63 histidine-tagged proteins were dialyzed overnight against storage buffer (20 mM HEPES-KOH, pH 7.6, 200 mM KCl, 10 mM b-mercaptoethanol, 10% glycerol) in the presence of TEV protease to cleave off the tags. The untagged proteins were further purified and concentrated using 1 ml HiTrap TM heparin HP columns (GE Healthcare Life Sciences). The eluted proteins were analyzed on 10% SDS-PAGE gels, and pure fractions (.95% purity) were pooled and dialyzed overnight against storage buffer. The concentrations of the purified and concentrated proteins were quantified using Coomassie protein assay reagent (Thermo Scientific) and were aliquoted and stored at 280°C.

Fluorescence anisotropy-based equilibrium binding assays
Fluorescein-labeled RNAs were diluted to 100 nM using folding buffer (20 mM HEPES-KOH, pH 7.5, and 100 mM KCl), heated to 90°C for 2 min, and slowly cooled over 1 h to room temperature. MgCl 2 was then added to the solution to a final concentration of 1 mM, and the solution was gently mixed and incubated on ice for about 1 h. Fluorescence anisotropy measurements for assessing the binding of eIF4GI constructs or DAP5 to the fluorescein-labeled RNAs were performed using the equilibrium titration module of an SF-300X stopped-flow fluorimeter (KinTek Corporation, Austin, TX). Fluorescein-labeled RNAs were excited at 495 nm, and emission was detected using a 515-nm high-pass filter (Semrock, Rochester, NY). Equilibrium binding titrations began with a 200-ml sample of either 10 nM (Fig. 2B and 3A) or 100 nM (Fig. 2, C-D, and Fig. 3, B-C) fluorescein-labeled RNA in the titration buffer (20 mM HEPES-KOH, pH 7.6, 100 mM KCl, and 1 mM MgCl 2 ), and 20-50 data points were collected for each anisotropy measurement by automated continuous injection of 20 ml of 2.5 mM eIF4GI 557-1599 , 10 mM eIF4GI 682-1599 , or 10 mM DAP5 over a period of 30 min at a temperature of 25°C. Note that the first reading is taken in the absence of protein. Using the Origin 2018b software package, the data were fitted to a nonlinear, single-site equilibrium binding equation of the form: where r obs is the observed anisotropy value, r min is the minimum anisotropy value in the absence of eIF4GI or DAP5, r max is the final saturated anisotropy value, [eIF4GI or DAP5] is the concentration of eIF4GI 557-1599 , eIF4GI 682-1599 , or DAP5, and K d is the equilibrium dissociation constant. The chi-squared values (x 2 ) that represented the statistical goodness of fit were always close to 1 and are reported in Table 1. Fitting data to a two-site model did not improve the fit, as judged by x 2 values. The equilibrium binding titration of each 59 UTR was performed three times and fit independently for K d . The fitted K d s were then averaged and the standard deviations were calculated ( Table 1).

Preparation of UTR-Luc reporter mRNAs for luciferase-based gene expression reporter assays
The UTR-Luc mRNA constructs for the luciferase gene expression reporter assays were generated from the BlucB plasmid (62), which contains a firefly luciferase gene flanked by 59and 39 UTR sequences of the barley yellow dwarf virus genomic RNA. To generate each UTR-Luc mRNA reporter construct, a sequence containing the T7 promoter followed by the target 59 UTR was cloned into the BlucB plasmid vector upstream of the firefly luciferase coding region, after removing the barley yellow dwarf virus 59 UTR as follows. Briefly, all UTR-Luc reporter constructs were PCR amplified with the forward primers containing a NotI site and reverse primers containing a BssHII site, as shown in Table 2, and then digested with NotI and BssHII. All the restriction enzymes were purchased from New England Biolabs (NEB), and restriction digestions were performed according to the manufacturer's instructions. The PCR-amplified sequences were ligated into a NotI-and BssHII-digested BlucB plasmid using DNA ligase (NEB) and transformed into E. coli DH5-a competent cells. Five colonies were selected, grown overnight in Luria-Bertani (LB) growth medium supplemented with 100 mg/ml ampicillin, and used to isolate plasmid DNA eIF4GI or DAP5 drives cap-independent translation initiation using the QIAprep ® Spin miniprep kit from Qiagen by following the manufacturer's protocol.
To assess whether the availability of a 59 end is required for the translation of these reporter mRNAs, a highly stable hairpin (41) was inserted at the 59 end of each UTR-Luc reporter (hp-UTR-Luc mRNA) using site-directed mutagenesis with the primers in Table 1. All clones were confirmed by sequencing (Genewiz). To generate a linearized plasmid DNA template for in vitro transcription, plasmid DNAs were linearized using KpnI so as to remove the 39 UTR from the UTR-Luc mRNA reporter construct. The resulting linearized DNA was purified using the GeneJET gel extraction and DNA cleanup micro kit from GeneJET per the manufacturer's instructions. DNA templates were in vitro transcribed using a T7 RiboMax large-scale RNA production kit (Promega) by following the manufacturer's protocol. ApppG (NEB) or Ribo m 7 GpppA cap analog (Promega) was added to the transcription mix in an ApppG or Ribo m 7 GpppA:GTP ratio of 10:1 to get mRNA transcripts with nonfunctional and functional caps, respectively. Capped RNAs were poly(A) tailed using the poly(A) tailing kit (Invitrogen) by following the manufacturer's protocol. The resulting capped and polyadenylated mRNAs were then purified using an RNA Clean and Concentrator kit (Zymo) by following the manufacturer's protocol. RNA concentrations were determined using a nano-drop UV-visible spectrometer, and integrity was verified by 1.5% agarose gel electrophoresis.
Western blots and depletion of eIF4GI and DAP5 from the rabbit reticulocyte lysate Monoclonal mouse primary antibodies for DAP5 (catalog number sc-137011), eIF4GI (catalog number sc-373892), eIF4E (catalog number sc-9976), eIF2b (catalog number sc-9978), and eIF4AI/II (catalog number sc-377315) (Santa Cruz Antibodies) were used for Western blotting and/or immunoprecipitation experiments. For the Western blotting experiments, the specific antibodies were diluted to 1:1000 in PBST (PBS-Tween) buffer containing 1% BSA. The blots were incubated overnight in the primary antibodies at 4°C with constant shaking. The membranes were washed three times with PBST and then incubated with horseradish peroxidase-goat anti-mouse secondary antibody (1:5000 dilution, Invitrogen, catalog number 31430) for 1 h at room temperature. After three subsequent washes in PBST, the membrane was developed using enhanced chemiluminescent substrate (SuperSignal TM West Femto, ThermoScientific, catalog number 34094). For the depletion of eIF4GI and DAP5 from the RRL, 20 ml of either eIF4GI or DAP5 capture antibodies were incubated with PureProteome TM protein A/G mix magnetic beads (EMD Millipore Corporation) at room temperature with continuous mixing for 1 h. The bead-antibody complexes were washed for 10 s with 500 ml PBS. The bead-antibody complexes were captured using a magnetic stand (Promega), and the suspensions were removed. The washing step was repeated 2 more times with PBS and then with the wash buffer (20 mM HEPES-KOH, pH 7.5, 50 mM KCl, 75 mM KOAc, 1 mM MgCl 2 ). The nuclease-treated RRLs (Promega) were incubated with these preformed bead-antibody complexes at 4°C with continuous shaking for 1 h. The magnet was reengaged to capture the bead-antibody-protein complex, and the resulting depleted RRL [RRL (-)4GI or RRL (-)DAP5 ] was collected and used for translation. To determine the extent of depletion, Western blotting assays were performed after resolving the samples on a 4-15% Tris-HCl gradient gel (Bio-Rad Laboratories).

Luciferase-based gene expression reporter assays
Gene expression was achieved by translating the UTR-Luc mRNAs in vitro, using the nuclease-treated RRL in vitro translation system from Promega. The RRL was made more cap dependent by addition of 75 mM KCl (63). Each 25-ml reaction mixture contained 70%, v/v, RRL, RRL (-)4GI , or RRL (-)DAP5 (Promega) supplemented with 0.5 mM MgCl 2 , 0.02 mM amino acid mixture, 10 units/ml RiboLock RNase inhibitor (Thermo-Scientific), and various concentrations of purified eIF4GI constructs or DAP5, as indicated in the figure legends. 4EGI-1 chemical (10 mM stock in DMSO) (Selleck Chemicals, catalog number S7369) was added to the translation mix when indicated at a concentration of 0.2 mM. Briefly, 1 mg of UTR-Luc mRNA was added to the RRL, RRL (-)4GI , or RRL (-)DAP5 in vitro translation mixture that had been preincubated at 30°C for 10 min following the addition of the specified concentration of eIF4GI constructs or DAP5. The resulting in vitro translation reaction then was incubated at 30°C for 1 h and stopped by the addition of 60 mM puromycin. Firefly luciferase activities were then assayed using a GloMax 96 microplate illuminometer (Promega). To achieve this, 3 ml of translation reaction mixture was added to 30 ml Bright-Glo luciferase assay reagent (Promega), and the resulting luminescence was measured in the illuminometer over a spectral wavelength of 350-650 nm and an integration time of 10 s at room temperature. After subtracting the background, measured using an in vitro translation reaction to which no UTR-Luc mRNA had been added, the luminescence data were analyzed and plotted using the Prism 8 software package. At least 3 different batches of RRLs were used. The translation data for each UTR-Luc mRNA were reported as an average from three independent experiments. Each independent experiment was done in triplicate, and the means 6 S.D. were calculated using GraphPad Prism 8. Statistical significance between the mean values was analyzed using two-tailed unpaired Student's t test (GraphPad Prism 8 software). The statistical significance was set at a p value of ,0.05, and the p values were calculated. The calculated p values for the analyses are indicated above the bar graphs ( Fig. 4 and 5 and Figs. S1, S2, and S4, n.s. [nonsignificant], p = 0.12; *, p = 0.033; **, p = 0.002; ***, p , 0.001).

Data availability
All data are presented in the manuscript.