Design of nucleotide-mimetic and non-nucleotide inhibitors of the translation initiation factor eIF4E: Synthesis, structural and functional characterisation

Eukaryotic translation initiation factor 4E (eIF4E) is considered as the corner stone in the cap-dependent translation initiation machinery. Its role is to recruit mRNA to the ribosome through recognition of the 5′-terminal mRNA cap structure (m7GpppN, where G is guanosine, N is any nucleotide). eIF4E is implicated in cell transformation, tumourigenesis, and angiogenesis by facilitating translation of oncogenic mRNAs; it is thus regarded as an attractive anticancer drug target. We have used two approaches to design cap-binding inhibitors of eIF4E by modifying the N7-substituent of m7GMP and replacing the phosphate group with isosteres such as squaramides, sulfonamides, and tetrazoles, as well as by structure-based virtual screening aimed at identifying non-nucleotide cap-binding antagonists. Phosphomimetic nucleotide derivatives and highly ranking virtual hits were evaluated in a series of in vitro and cell-based assays to identify the first non-nucleotide eIF4E cap-binding inhibitor with activities in cell-based assays, N-[(5,6-dihydro-6-oxo-1,3-dioxolo[4,5-g]quinolin-7-yl)methyl]-N′-(2-methyl-propyl)-N-(phenyl-methyl)thiourea (14), including down-regulation of oncogenic proteins and suppression of RNA incorporation into polysomes. Although we did not observe cellular activity with any of our modified m7GMP phosphate isostere compounds, we obtained X-ray crystallography structures of three such compounds in complex with eIF4E, 5′-deoxy-5′-(1,2-dioxo-3-hydroxycyclobut-3-en-4-yl)amino-N7-methyl-guanosine (4a), N7-3-chlorobenzyl-5′-deoxy-5′-(1,2-dioxo-3-hydroxy-cyclobut-3-en-4-yl)amino-guanosine (4f), and N7-benzyl-5′-deoxy-5′-(trifluoromethyl-sulfamoyl)guanosine (7a). Collectively, the data we present on structure-based design of eIF4E cap-binding inhibitors should facilitate the optimisation of such compounds as potential anticancer agents.


Introduction
In eukaryotes, most mRNAs are translated using a capdependent mechanism, which consists of three stages: initiation, elongation, and termination. The initiation step requires a number of eukaryotic translation initiation factors (eIFs) [1]. Initially, the 40S ribosomal subunit associates with Met-tRNAi and a group of initiation factors, including eIF1, eIF1A, eIF2, eIF3, and eIF5 to form the 43S pre-initiation complex [2]. This complex is then recruited to the 5 0 -end of capped mRNA by another complex of factors (known as eIF4F) consisting of eIF4E, eIF4G, and eIF4A. eIF4E is an indispensable element for cap-dependent translation initiation and plays a major role in recognition of the mRNA cap structure (m 7 GpppN, where N is any nucleotide). It binds to eIF4G, which serves as a scaffolding protein to gather eIF4A (ATP-dependent RNA Abbreviations: eIF, eukaryotic translation initiation factor; 4E-BP, eF4E-binding protein; FP, fluorescence polarisation; m 7 G, N 7 -methyl guanosine; RRL, rabbit reticulocyte lysate. helicase), eIF4B, eIF3, and poly-A-binding protein to form the 48S complex [3,4]. This complex then scans the mRNA for an initiation codon and, once this is located, the 60S ribosomal subunit joins to form the elongation-competent fully functional 80S.
Under typical circumstances, eIF4E is the least abundant of the cellular translation initiation components [5] and it represents the rate-limiting element for the initiation process [6,7]. Under resting conditions, eIF4E is kept in an inactive form bound to 4E-binding proteins (4E-BPs); once the latter have been phosphorylated through the PI3K-AKT-mTOR signalling pathway, eIF4E is released and can associate with eIF4G to launch translation. eIF4E is overexpressed in numerous human tumours, and it contributes to transformation, tumourigenesis, and progression of cancers [8,9]. eIF4E overexpression facilitates translation of weak and highly structured mRNAs that typically encode proteins involved in cancer pathology, such as proto-oncoproteins (e.g. cyclin D1, ornithine decarboxylase), angiogenesis factors (e.g. FGF-2, VEGF), and factors related to tumour invasiveness, such as MMP-9 [10], due to the ability of eIF4E to stimulate eIF4A [11]. For these reasons, eIF4E represents an attractive cancer drug target [12]. Moreover, a recent study suggests that targeting eIF4E for cancer treatment has minimal effects on growth of eand protein synthesis ine healthy cells [13].
Strategies to design eIF4EemRNA cap-binding antagonists have been based on nucleoside (purine N 7 -substitutions) [14e17] and nucleotide (altered phosphate groups) [17e19] modifications of 7methylguanosine (m 7 G) nucleotides (Fig. 1) [12], including recent work affording for the first time m 7 G monophosphate (m 7 GMP) nucleotide mimetic compounds with high affinity for eIF4E [20]. However, to date no cell-permeable small-molecule eIF4E capbinding antagonists have been reported. As with antiviral agents derived from nucleotides, prodrugs that mask the ionic nature of phosphate groups, such as phosphate esters of phosphoramidates, may offer an avenue for the design of cell-permeable eIF4E inhibitors, but this strategy has not progressed very far as yet [21].
Here we report on new methods to design cap antagonists and we show a group of nucleotide mimetic compounds with phosphate group isosteres and various purine N 7 -substituents. The biological activities of these compounds were assessed using a range of techniques and the eIF4E-binding modes of three compounds were determined experimentally using X-ray crystallography. However, despite phosphate group modifications, these nucleotide mimetics still exhibit poor cellular bioactivity. Therefore, a computer-aided drug design method was exploited to design a set of non-nucleotide compounds to provide the first smallmolecule eIF4E inhibitor possessing cellular activity consistent with blocking of eIF4E-mediated initiation of translation.

Design and synthesis of nucleotide monophosphate mimetic eIF4E inhibitors
One of the main challenges in the design of inhibitors of eIF4E cap binding has been to achieve membrane permeability and thus cellular bioavailability [20]. This is due to the fact that the affinity of eIF4E for the RNA m 7 GpppN cap structure derives in large part from polar interactions between the ligand triphosphate group and the receptor protein (Fig. 1a) [15,25]. In order to address the permeability problem, as well as the intrinsic hydrolytic and enzymatic lability of nucleotides, we aimed to design nucleoside monophosphate mimetics [25,26]. Several phosphate group replacements were investigated, such as squaramides, sulfonamides, and tetrazoles. Since e.g. m 7 GMP has much lower affinity for eIF4E than m 7 GTP [25], we expected to compensate the loss in affinity with nucleoside monophosphate mimetics through optimal purine N 7 -substituent modifications, based on previous studies demonstrating that replacing the N 7 -methyl group with bulkier groups contributed significantly to ligand affinity (Fig. 1b) [27,28].
Squaric acid possesses similar charge distribution, polarity, and acidic properties as phosphoric acid, hence the squaramide group could represent an isostere for the phosphate group [29,30]. Similarly, sulfonamide derivatives have been reported as phosphate mimics in the design of tyrosine phosphatase inhibitors [31]. The third candidate we considered for phosphate group replacement was a tetrazole [32], as this system mimics the acidic features of the phosphate group, provided the proton attached to one of the tetrazole ring nitrogens is not replaced [33].  7 GTP shown as yellow CPK sticks) and binding of eIF4G or 4E-BPs (represented by a peptide derived from 4E-BP1 shown as a cyan cartoon) occurs on opposite faces of eIF4E (green cartoon). Apart from direct cap-binding antagonists, allosteric inhibitors (binding pose of 4-EGI1 [22] shown as magenta sticks) and inhibitors derived from eIF4G and 4E-BPs [23] are being developed. Whereas m 7 GTP-binding is dominated by polar interactions between the cationic N-methylpurine system and eIF4E residues W56, W102, and E103 (cationep interaction and H-bonds), as well as the phosphate groups with residues R157, K159, and K162, (b) GMP derivatives with N 7 -substituents other than methyl, such as 4-fluorobenzyl [24] (green) or (4-chlorophenoxy)ethyl [20] (cyan), also make hydrophobic contacts with two concave lipophilic pockets (surface representations) behind the W56eW102 stack. The key synthetic precursor for the preparation of our nucleotide cap mimetics was the protected guanosine derivative 1e (Scheme 1) [34], which was prepared by first blocking the 2 0 ,3 0vicinal glycol of guanosine as the acetonide 1a [35], followed by dimethoxytritylation of the exocyclic amine group (1b) [36], and transformation of the 5 0 -hydroxyl group to the amine via successive conversion to the iodide 1c and azide 1d, and reduction of the latter [37]. The amine 1e was condensed either with dimethyl squarate [29] to afford the squaramide derivative 2 or with polar sulfonyl chlorides to furnish the sulfonamide derivatives 5. These intermediates were N 7 -alkylated using a range of alkyl or aryl bromides [14] to provide the methyl squaramides 3 and sulfonamides 6. The methyl squaramide groups of the former were first demethylated with sodium iodide in acetone [38], followed by treatment with aqueous formic acid to afford the squaramides 4. The acetonide group of precursors 6 was removed similarly with formic acid to provide the sulfonamides 7 [39]. The tetrazole derivatives 11 were also elaborated from 1e, by Boc-protection of the 5 0 -amino group (8), followed by N 7 -alkylation (9), global deprotection (10), and acylation of the 5 0 -amino group with 1H-tetrazole-5-acetic acid.

Design of non-nucleotide eIF4E inhibitors
As an alternative to nucleoside monophosphate mimetics we considered structure-based design of eIF4E cap-binding antagonists. For this purpose we docked multi-conformer libraries of leadlike molecules from the ZINC [40] database to a model of the Bn 7 GMP-binding site of eIF4E derived from a complex crystal structure we reported earlier (PDB entry 2V8Y) [15] using shapebased fast rigid exhaustive docking (FRED) [41,42]. We applied a combination of scoring functions, including functions based on Poisson-Boltzmann electrostatic potentials, in order to ensure that interactions of virtual hits mimicking the cationep interaction [43,44] of the N 7 -methylguanine system in the mRNA cap with the eIF4E W56eW102 indole side chains (Fig. 1a) would be rewarded. The virtual screening protocol was validated by docking and scoring a set of 500 randomly-chosen diverse decoy molecules from the ZINC database spiked with a set of N 7 -alkylated capderived nucleotides known to bind eIF4E, as well their non-Scheme 1. Preparation of nucleotide mimetics 4ae4i, 7ae7c, and 11ae11b a . a Conditions: (i) DMTr-Cl, Me 3 SiCl; (ii) (PhO) 3 P þ MeI À , THF, À78 C; (iii) NaN 3 , DMF, 85 C; (iv) Ph 3  alkylated counterparts, which are known not to interact with eIF4E [25]. This analysis revealed that the N 7 -alkylated cap-derived nucleotides populated the top 12% of the hit list from the combined scoring protocol. Based on this result we proceeded to screen the entire lead-like subset of the ZINC database. We acquired several hundred of the top-ranking virtual hits and submitted these to the mass spectrometry-based assay described earlier [45], as well as the fluorescence polarisation (FP) assay described below. The chemical structure and predicted properties of the most highly scored virtual hits are shown in Table 1.

Biological activity of nucleotide monophosphate mimetic and non-nucleotide eIF4E inhibitors
In order to assess our m 7 GMP-derived compounds and virtual screening hits as mRNA cap-binding antagonists, we developed an FP assay (Fig. 2) based on displacement of a fluorescently labelled m 7 GDP derivative from eIF4E by cap-binding inhibitors. Assay optimisation showed that a tracer concentration of 10 nM and an eIF4E concentration of 90 nM was optimal. All test compounds were screened initially at 100 mM and actives were titrated to calculate K d values, summarised in Table 2.
As expected, m 7 GTP had almost 10-fold higher affinity for eIF4E in our FP assay than m 7 GMP. Although the majority of the squaramide (4), sulfonamide (7), and tetrazole (11) phosphate-mimetic analogues inhibited tracer binding in the FP assay at the prescreen concentration of 100 mM to some extent, only the sulfonamides 7a and 7b showed levels of inhibition that permitted construction of a sufficiently wide doseeresponse titration for accurate K d determination. Of the squaramides, all derivatives except 4a, 4c, and 4f showed insignificant inhibition (<10%) in the prescreen. Surprisingly, in the context of the squaramide phosphomimetic group, the expected enhancement of affinity upon replacing the N 7 -methyl group (4a) with a benzyl group (4b) was not observed. Furthermore, substitution of the phenyl group with small and electronically neutral groups (3-Me, 4c; 3-Cl, 4f) retained the modest activity of 4a and 4b, whereas introduction of larger and more polar substituents reduced or abolished activity. The trifluoromethyl-and the para-carboxyphenyl sulfonamides 7a and 7b were the only nucleoside monophosphate mimetic compounds for which K d values could be obtained, and these are significantly higher than that of m 7 GMP. The tetrazole derivatives 11 had very modest activity. Here the N 7 -benzyl derivative 11b was somewhat more active than the N 7 -methyl compound 11a.
Amongst the virtual screening hits, a cluster of 4,5-bisaryl-2carboxymethylthio-1,2,3-triazoles 12aec was identified but these showed low activity in the FP prescreen and K d values could not be determined. Of the somewhat structurally related quinolinones 13 and 14, the former was essentially without activity in the FP assay, whereas the latter was clearly active, at a level similar to the best of the nucleotide derivatives (7b). The thiazolone 15 and the pyridothiazolopyrimidine 16 showed no activity in the FP assay. Further to confirm inhibition of eIF4E cap binding, a radiometric competition binding assay was developed based on an in vitro UV cross-linking technique widely used to study proteineRNA interactions [48]. In this assay, a radiolabelled and capped RNA was incubated with eIF4E in the presence or absence of putative inhibitors 4e16. The RNAeeIF4E complexes formed were crosslinked, separated by SDS-PAGE, and quantitated by autoradiography. The assay was validated with positive controls (m 7 GMP and m 7 GTP), test compounds were pre-screened at 100 mM (Table 2), and for sufficiently active compounds half-maximal inhibitory concentration (IC 50 ) values were determined (Fig. 3).
As expected, m 7 GMP and m 7 GTP displayed similar activity levels in the UV cross-linking assay as in the FP assays. The squaramides 4 were inactive with the exception of the derivative 4h. All three sulfonamides 7 showed activity in the UV cross-linking assay and for 7a and 7b IC 50 values could be determine (Fig. 3). For the inhibitors from the virtual screen (11e16), the UV cross-linking assay results mirrored those from the FP assay but an IC 50 value could only be determined for quinolinone 14.
For the nucleoside monophosphate mimetic compounds, which we expected to be devoid of cellular bioactivity, we also sought to obtain a functional read-out of eIF4E inhibition. For this purpose we  employed a cell-free rabbit reticulocyte lysate (RRL) system to observe the translational consequences of eIF4E inhibition [49]. Test compounds m 7 GMP, m 7 GTP, as well as the sulfonamides 7a and 7b were incubated in the RRL system, which was primed with luciferase RNA, and expression of luciferase was then quantitated. From the results ( Table 2) it can be observed that all four test compounds inhibited luciferase synthesis. For m 7 GMP and m 7 GTP the IC 50 values in the RRL assay were significantly higher than those determined for competitive binding, perhaps due to limited stability of these nucleotides in RRL, whereas the drop-off in activity was much less pronounced for sulfonamides 7a and 7b. RRLs were then primed with bicistronic mRNAs where the initiation of translation of the upstream Renilla cistron occurs in a capdependent manner, while the translation of the downstream firefly luciferase cistron is driven by the HCV internal ribosome entry segment and is therefore independent of the cap structure at the 5 0 end of the mRNA (Fig. 4). This bicistronic RNA allowed the comparison of the effect of quinonlinone 14 and the para-carboxyphenyl sulfonamides 7a and 7b on cap-dependent and capindependent translation simultaneously. The data show that while, as expected, there was inhibition of synthesis from the capdependent Renillia luciferase mRNA, there was no effect on the translation initiation of the firefly luciferase mediated by the HCV-IRES, suggesting that sulfonamides 7a and 7b and quinolinone 14 inhibit cap-dependent translation (Fig. 4). In order to investigate the effect of the eIF4E inhibitors on overall protein synthesis in cells, we employed 35 S-Met labelling [50] to measure the amount of newly synthesised total protein in HeLa cells after incubation for 24 h with test compounds (the protein synthesis inhibitor puromycin was used as a positive control). None of the nucleoside phosphates (m 7 GMP and m 7 GTP) or phosphate mimetics (7a and 7b) showed measurable inhibition of cellular protein synthesis, indicating that membrane permeability of the nucleoside phosphate mimetic compounds, including the sulfonamides 7a and 7b, is still sub-optimal. Quinolinone 14, on the other hand, which was the only virtual screening hit with activity in the UV cross-linking assay, also inhibited protein synthesis in both HeLa and MCF-7 cells with IC 50 values similar to the activity levels observed in cell-free assays, suggesting that this compound is cellpermeable. Based on the predicted physicochemical properties of this compound (Table 2), this finding was not unexpected.
For quinolinone 14, which was observed to inhibit cellular protein synthesis, we investigated if this effect was indeed due to suppression of translation, as would be expected of an eIF4E capbinding inhibitor. For this purpose we used polysome analysis [51]. HeLa cells were treated for 2 h with test compound 14 or puromycin as a positive control, harvested, treated with cycloheximide in order to trap RNAeribosome complexes, and cell lysates were fractionated (Fig. 5a, ieiv). For cells treated with quinolinone 14, there was a decrease of the amount of RNA incorporated in multiple ribosomes (polysomes) compared to untreated cells and a concomitant increase in the sub-polysomal material, consistent with a block in initiation ( Fig. 5a and b), supporting our data which shows 14 to suppress cap-dependent translation in cells.
To examine whether there was an effect on the expression of proteins which are known to be eIF4E dependent and are associated with tumorigenesis, lysates of HeLa cells treated for 2 h with quinolinone 14 were probed by western blotting for c-myc and cyclin D1 protein, with actin used as a loading control. The data show that at 50 and 100 mM treatment concentrations of 14 there were decreases in levels of both proteins (Fig. 5c). To confirm that the decrease in protein levels corresponded with altered polysomal association, mRNAs were isolated from polysomal and subpolysomal regions of the gradients and RT-PCR was used to assess the relative amount of c-myc mRNA before and after treatment. The data show that following treatment with quinolinone 14 there was a small shift of the c-myc mRNA into the sub-polysomes, consistent with a decrease in its translation (Fig. 5 aiv).

X-ray crystallography
eIF4E was co-crystallised with either an eIF4G-or a 4E-BPderived peptide (designated 4G and 4EBP, respectively) and one of three of the m 7 GMP-derived analogues (squaramides 4a and 4f, sulfonamide 7a). All complex structures showed the characteristic cupped hand structure of eIF4E [52] with the ligands occupying the cap-binding site.
The initial eIF4E:4G:7a complex structure ( Fig. 6b and c) was solved to a resolution of 1.71 Å (Table 3) with the ligand 7a residing in the cap-binding site, accompanied by a single sulfate ion (from the crystallisation buffer). The guanine of 7a makes three H-bonds, two with the side chain of E103 and one with the backbone carbonyl of W102. The sulfate ion interacts directly with eIF4E through R157 and makes additional water-bridged interactions with R112, N155, R157, and K162, as well as to the sulfonamide group of 7a. The ligand 7a itself interacts with four neighbouring or bridging waters. The unusual binding mode observed was unpredicted as it was expected that the guanine would make the same or similar cation-p stacking interaction as seen with all other capbased analogues to date (compare Fig. 6a) [12,15,20]. However, the presence of the sulfate ion allows 7a to 'fold 0 back on itself and the positively charged guanine forms a stabilising intramolecular electrostatic interaction with the electronegative sulfonamide group. In order to accommodate this ligand conformation, W102 is reoriented out of the binding pocket from its normal cation-p interaction position. W56 remains in the usual position, providing a stacking interaction with the guanine. In comparison to other complex structures with cap derivatives (Fig. 6a) the sulfate ion occupies the same position as the a-phosphate. Presumably the unusual bound conformation of 7a observed was favoured due to high sulfate concentration in the crystallisation buffer.
In order to test this hypothesis, a crystal from these conditions was transferred to an identical crystallisation drop but lacking  Cap-dependent translation drives expression from the Renilla luciferase transcript, whereas cap-independent translation drives protein synthesis from the Firefly luciferase transcript. (b) Rabbit reticulocyte lysates were primed with pRHCVF RNA and treated with compound 14 (i) or 7a or 7b (ii) at the doses shown. The compounds were dissolved in DMSO, which is known to inhibit the reticulocyte lysate system when compared to the control without DMSO. Therefore a 1% solution of DMSO was used as an additional control and the degree of inhibition measured relative to this value. Experiments were performed on three independent occasions, significance (***P < 0.001) was calculated using Student's t-test (error bars S.D.). Black bars denote Renilla luciferase and grey bars Firefly luciferase.    (Fig. 6d), previously occupied by sulfate. As expected, the displacement of the sulfate was accompanied by repositioning of W102 to form the usual cation-p 'sandwich 0 interaction with the positively charged guanine moiety and W56. The squaramide complex structures eIF4E:4EBP:4a and eIF4E:4f were solved to 3.61 Å and 2.40 Å, respectively (Table 3). Despite low overall resolution, the eIF4E co-crystal structure with ligand 4a clearly shows that the squaramide moiety acts as a phosphate mimic by making a similar interaction with R157 despite an altered conformation (Fig. 7a). The positively charged N 7 -methyl guanine system forms the characteristic cation-p interaction as expected. Fig. 7b shows compound 4f bound to the cap-binding site of eIF4E. The guanine group is located in its normal binding position, stacking with the side-chain of W56. As predicted, the squaramide occupies the a-phosphate position (Fig. 7c), making interactions with the side-chain of R157. The chloro substituent of the N 7 -benzyl group causes a rotation of approximately 90 in relation to other N 7 -benzyl substituted cap analogues (Fig. 7c). In turn, to accommodate the extra space required by the chloro group, the side chain of W102 is re-oriented away from making the usual cation-p stacking interaction to a position similar to that seen with 7a in the presence of sulfate (compare Figs. 6b and 7b).
We were unable to obtain complex crystal structures with unambiguous ligand density of the eIF4E inhibitors 12e16 from the virtual screen. However, it is interesting to compare the predicted binding modes of these compounds with the experimental eIF4E complexes of the m 7 GMP-derived derivatives (Fig. 8). Re-docking of Bn 7 GMP to the receptor structure of the Bn 7 GMP complex using the protocol used for virtual screening showed (Fig. 8a) virtually identical positioning of the N 7 -benzylguanine portion of the ligand, whereas the phosphate group in the re-docked pose bound more deeply into the cavity behind W56 and making polar interactions with the side chains of R157, W56, and D90, compared to the experimental binding pose, in which the phosphate group only interacts with R157. As a result of the different positioning of the phosphate group, the ribose systems in the experimental and modelled poses are also positioned somewhat differently.
The top-ranking docking solutions for the 4,5-bisaryl-2carboxymethylthio-1,2,3-triazoles 12aec all showed similar poses, with the substituted phenyl groups stacking between W56 and W102, the unsubstituted aryl group inserting into the cavity behind W102, and the carboxymethyl substituents occupying the polar pocket behind W56 (Fig. 8b). Quinolinones 13 and 14 are predicted to stack between W56 and W102 in a similar manner, disposing the remaining substituents into the subpockets behind the W residues as shown ( Fig. 8c and d). Thiazolone 15 and the pyridothiazolopyrimidine 16 again show stacked aromatic systems involving W56 and W102 but these ligands do not show good interactions with the other subpockets of the cap-binding site.

Conclusions
In the present study we report a new set of nucleotide-based cap analogues with phosphate modifications. Crystal structures of sulfonamide 7a and squaramides 4a and 4f in complex with eIF4E were obtained and show similar binding modes to cap derivatives co-crystallised with eIF4E previously. However, our structures highlight the flexibility of the m 7 G-binding pocket, especially residue W102 that forms part of the cation-p interaction characteristic of cap-recognition by eIF4E. This finding suggests that cap-binding inhibitors of eIF4E devoid of a permanent cationic structure may be possible. All of the confirmed eIF4E inhibitors (12a, 14, 16) from our alternative design strategy based on virtual screening of lead-like non-nucleotide compounds support this suggestion. We present likely eIF4E-binding modes for these compounds but the possibility that these induce conformational changes in the cap-binding pocket of eIF4E cannot be excluded.
As has been found elsewhere [20,25], imparting sufficient membrane permeability in eIF4E cap-binding inhibitors for such compounds effectively to block cellular eIF4E function is difficult. Despite modulation of the ionic and polar nature of the phosphate group in m 7 GMP derivatives through replacement with phosphate isosteres, none of our nucleoside monophosphate mimetic compounds showed cellular activity. The situation was different with the non-nucleotidic eIF4E inhibitors identified through virtual screening. These compounds possess physicochemical properties (Table 1) consistent with good permeability and we show that one of these compounds, the quinolinone 14, not only inhibited eIF4E cap-binding in a biochemical and cell-free functional assay, but also acted as a protein synthesis inhibitor at the level of RNA translation in cellular assays. Although compound 14 is not particularly leadlike due to excessive ring count and lipophilicity, as well as low intrinsic solubility (Table 1), our results confirm for the first time that cellular eIF4E can be targeted with non-nucleotidic compounds and inherently drug-like compounds.

Chemistry
Flash chromatography was carried out using silica gel cartridges (Sorbsil ® and ISOLUTE ® ). TLC was performed using Merck silica gel 60 F254 plates and visualisation by UV irradiation at 254 nm or by staining with KMnO 4 or ninhydrin solutions, as appropriate. 1 H-13 C-, 19 F-, and 31 P NMR spectra were obtained at room temperature using a Bruker 400 Ultrashield spectrophotometer at 400.13 MHz, 100.61 MHz, 376.50 MHz, and 161.97 MHz, respectively. Samples were prepared in the deuterated solvents Me 2 SO-d 6 (unless otherwise indicated). Chemical shifts (d) are expressed in ppm relative to SiMe 4 . Mass spectra were obtained using a Waters 2759 spectrometer using both positive and negative electrospray ionisation. Infrared spectra were recorded on a Nicolet IR-200 FT-IR spectrophotometer using the KBr disc technique. Analytical RP-HPLC was performed with a Shimadzu UFLCXR system coupled to a mass spectrometer (Applied Biosystems API2000). Phenomenex columns thermostated at 40 C were used: C 8 -110A (method 1); Luna 3u (PFP2) 110A, (method 2), and C 18 (method 3); all with dimensions 50 Â 2 mm. The flow rate was 0.5 mL/min and UV detection was at 220 and 254 nm. Gradient elution: preequilibration for 1 min with 10% solvent B in solvent A, 10e98% solvent B over 2 min, 98% B for 2 min, 98 to 10% B over 0.5 min, and finally 10% B for 1 min (solvent A: 0.1% HCOOH in H 2 O, solvent B: 0.1% HCOOH in MeCN). Preparative RP-HPLC was carried out using a Waters automated LC system, with Waters 2525 pump, Waters 2767 sample manager, and Waters 2487 dual wavelength absorbance detector system or a Dionex ICS-3000 HPLC system with an Ultimate-3000 detector, using Phenomenex Luna PFP or HILIC (5 mM) columns (250 mm Â 10 mm). The purity of all compounds screened in the biological assays was examined by LCÀMS analysis and was found to be ! 95%.  A stirred solution of 1b (1 g, 1.6 mmol) in anhydrous THF (100 mL) was cooled to À78 C. (PhO) 3 P þ MeI À (1.45 g, 3.2 mmol) was added and the solution was allowed to reach r.t. over 30 min. THF was evaporated and the residue was diluted with EtOAc (50 mL). This solution was washed with Na 2 S 2 O 3 (3 Â 30 mL of a 25% solution in H 2 O). The organic phase was treated with MgSO 4 , filtered, and evaporated. The residue was dissolved in EtOAc (15 mL) and precipitated with hexane (100 mL). The precipitate was chromatographed (SiO 2 ; CH 2 Cl 2 eMeOH, 9.2:0.8). The product was again precipitated from EtOAcehexane to afford the title compound as a white solid (70%). Mp 148e150 C. 1 To a stirred solution of 1d (1 g, 1.5 mmol) in dry pyridine (15 mL  To a stirred solution of 1e (300 mg, 0.48 mmol) in MeOH (25 mL) iPr 2 NEt (40 mL, 1.92 mmol) was added. After 5 min, dimethyl squarate (68 mg, 0.48 mmol) was added to the reaction mixture, which was stirred for 4 h. It was then evaporated and the residue was chromatographed (Si 2 O; CHCl 3 eMeOH, 9:1) to afford the title compound as a white solid (61%). Mp 170e172 C. 1   To a solution of compound 2, 5a, 5b, or 8 (1 eq) in DMF, the appropriate alkyl or aryl halide (10 eq) was added and the mixture was stirred for 24e48 h, with monitoring by TLC (CHCl 3 eMeOH, 17:3) and LC-MS. After completion of reaction, the mixture was evaporated in vacuo and the residue was triturated with Et 2 O with the aid of sonication. The precipitate was filtered and chromatographed (SiO 2 ; CHCl 3 eMeOH, 9:1) to afford the pure title compound.     Off-white solid (67%). 1   Off-white solid (54%). 1

General method 2: deprotection of 3ae3i
To a stirred solution of 3 (1 eq) in Me 2 CO, NaI (2 eq) was added and the mixture was heated under reflux for 48 h. After completion of the reaction, the solvent was evaporated and the residue was treated with HCOOH (1 mL/0.1 mmol of 80% aq soln) for 12 h. After cooling, the mixture was evaporated and the residue was triturated with Me 2 CO with the aid of sonication. The precipitate was filtered, washed successively with Me 2 CO and hexane, and was dried.

Virtual screening
An eIF4EeBn 7 GMP complex crystal structure (PDB code 2V8X) [15] was used for docking purposes, and docking was based on the lead-like subset of compounds in the ZINC database [40]. Receptor preparation was performed using the Biopolymer function in SYBYL 8.0 (Tripos International, 1699 South Hanley Rd., St. Louis, MO, 63144, USA). The entire binding site occupied by Bn 7 GMP in the eIF4E template structure was defined as the docking site. Ligand 3D co-ordinates were generated using CONCORD (R. S. Pearlman, "Concord," distributed by Tripos International, St. Louis, MO, 63144, USA). Conformer (up to 200 per compound) generation (OMEGA version 2.4.6) [54,55], as well as docking and scoring (FRED module of the OEDocking version 3.0.0) [41,42], were carried out using OpenEye Scientific Software (Santa Fe, NM, USA; http://www. eyesopen.com). Scoring of docked poses was performed using a combination of the OEDocking scoring functions Chemgauss3 and Zapbind (using Poisson-Boltzman electrostatics approximation), together with a semi-empirical method using the LocalSCFv2.0 application (Fujitsu FQS Poland Sp. z o. o., Krakow, Poland) [56]. Virtual hit compounds were acquired from vendors (Table 1) and were characterised by LC-MS. Two different batches of compound 14 with slightly different appearance and aqueous solubility were obtained. However, these were indistinguishable by LC-MS and NMR, and spectral data was in accord with the structure shown.

Recombinant eIF4E
Expression and purification of eIF4E followed a modified procedure of that already published [15]. Briefly, the procedure was modified to allow for rapid, scale-up production of eIF4E and will be published separately (manuscript in preparation). Following purification, the protein concentration was adjusted to 1 mg mL À1 (Vivaspin concentrator, 5 kDa cut-off) and the solution was flashfrozen in 200-mL aliquots, which were stored at À80 C.

Fluorescence polarisation assay
A fluorescent probe (EDA-m 7 GDPeATTO-465; NU-827-465) was obtained from Jena Biosciences; 96-well black plates (Optiplate ® ) were from Perkin Elmer. Recombinant eIF4E and the probe were diluted (20 mM HEPES, pH 7.4, 100 mM KCl, 1 mM DTT) to obtain the required concentration. m 7 GTP (Sigma-Aldrich) and m 7 GMP were used as positive controls. The final volume in each well was 100 mL. Compound screening was performed using a final DMSO concentration of 1%. Fluorescence polarisation measurements were recorded on a Perkin-Elmer Envision 2104 plate reader. For K d calculation, logarithmic serial dilutions of test compounds were prepared. Background-corrected readings were plotted against test compound concentration in GaphPad Prism (Version 5.0) in order to obtain K d values.

Crystallisation of ligand-bound eIF4E
A 3.2-mM solution of 4G-I (eIF4G-derived peptide, KKRY-DREFLLGFQF) or 4EBP-I (4E-BP-derived peptide, RIIYDRKFLMECRN) in crystallisation buffer (50 mM Tris, pH 8.0, 150 mM KCl, 1 mM DTT) was added to an equal volume of a solution of recombinant human eIF4E (8 mg mL À1 ). A ligand solution (50 or 100 mM) in DMSO was added to a final DMSO concentration of 2.5%. Thin needle-like crystals grew over a period of 3e7 d by the hanging drop method (1:1 proteinemother liquor) over 21e31% PEG-8,000, 1e3% (NH 4 ) 2 SO 4 , 100 mM HEPES, pH 7.5. Drops containing crystals were added to 50 mL mother liquor vortex-mixed for 3 min. This was used as a seed stock (at 1:1 and 1:10 dilution) for seeding back into the same conditions. Needle-like crystals grew in 1e7 d and were frozen directly in liquid N 2 . For back-soaking experiments with ligand 7a complex crystals, crystals were transferred to a drop containing a 1:1 ratio of crystallisation buffer and 100 mM HEPES, pH 7.5, 21% PEG-8,000, supplemented with 7a. The drop was equilibrated over 100 mM HEPES, pH 7.5, 21% PEG-8000 for 5 min before flash-freezing the crystals in liquid N 3 .
X-ray crystallography data collection and structure refinement. Data were collected at Diamond Light Source, UK, from single crystals using beamlines I02, I04, and I04-1, at 100 K, and were processed using either iMOSFLM [57] or XDS [58] using scaling with SCALA [59]. The structures were solved with the AutoMR command in the PHENIX suite [60], which uses the programme PHASER [61]. A single eIF4E and 4E-BP chain from the unit cell of PDB entry 2V8Y (ligand removed) was used as the search model and all structures were subjected to several rounds of refinement in PHENIX and manual refinement in COOT [62], including ligand fitting and a rebuilding of the 4G-I peptide chain (where necessary).

Rabbit reticulocyte lysate in vitro translation assays
PCR (hot KOD polymerase kit, Novagen ® ; Qiaquick PCR purification kit, QIAGEN) was used to amplify the luciferase DNA sequence from a pGl3 vector with the primers GCGGCCGGCCGCCCCGACTCTAGAA and GGGGTAATACGACTCAC TATAGGGTCC-ACCTCGAATCACTAGTCAGCTG. Alternatively, the plasmid pRHCVF, which contains Renilla luciferase as the upstream cistron and firefly luciferase as the downstream cistron preceded by the HCV IRES, was digested with HpaI and use to prime in vitro transcription reactions. In vitro RNA synthesis was performed on the resulting DNA using a mMessage mMachine ® T7 kit (Ambion). The resulting capped RNAs were purified using microSpin G-50 columns (GE Healthcare), analysed by agarose gel electrophoresis, and were stored in aliquots at À80 C. An in-vitro translation Rabbit Reticulocyte Lysate System kit (Promega) was used as per the manufacturer's instructions. Briefly, 20 ng capped RNA was incubated with rabbit reticulocyte lysate in a final reaction volume of 20 mL. The mixture was incubated at 30 C for 90 min and then placed on ice for immediate analysis. Diluent (H 2 O) was replaced by test compound solutions in the course of compound screening and the final DMSO concentration was limited to 1%. Assays were performed with m 7 GTP as a positive control and a DMSO negative control (which was assigned a 100% value).

Ultraviolet cross-linking assay
In a way similar to RNA luciferase synthesis, a 392-base coding region of the pGL3 vector was amplified and the corresponding RNA was generated using the T7 RiboMax large-scale production system (Promega) according to the manufacturer's instructions. The resulting transcript was then cap-radiolabelled with 32 P-[a-GTP] (specific activity 400 Ci mmol À1 , Hartmann Analytic GmbH) using a vaccinia virus capping system (New England Biolabs) and purified using a microSpin G-50 column. A 1-mL aliquot of the labelled RNA was transferred to a scintillation vial with 2 mL scintillation liquid (Ecoscint A, National Diagnostics), and was read in a scintillation counter (Wallac Winspectral 1414 liquid scintillation counter). The disintegration per minute readings were converted to Ci units, and the concentration of radiolabelled capped RNA was then calculated based the original 32 P-GTP stock concentration after correction based on decay time. Assay procedures: to work out suitable concentrations of RNA and eIF4E to be used in the assay, a titration experiment of eIF4E against a constant concentration of labelled RNA was carried out. The assay was performed using 96-well flat-bottom plates (Corning, Costar ® ), in a volume of 20 mL for each reaction (in a ratio of 10:5:5 for buffer, RNA, eIF4E, respectively). Plates were kept on ice and were exposed to a UV radiation source of 275 J m 2 at a distance of 3 cm for a period of 30 min. The plate wells were then treated with a mixture of RNase (0.1 mg mL À1 RNase A, 0.1 mg mL À1 RNase T1, 0.1 mg mL À1 RNase V1) at 37 C for 30 min. Samples were then separated by SDS-PAGE and the gels were dried and exposed to autoradiography film for 24 h, before being read using a Typhoon 9500 FLA scanner. Band intensities were quantified using image analysis software (Image J, Version 1.46) relative to a DMSO controls.

35 S-methionine incorporation assay
HeLa cells were seeded into 12-well plates at 1 Â 10 5 cells per well for 2-h and at 6 Â 10 4 cells per well for 24-h treatment experiments. Cells were incubated for 24 h until they reached 70% confluency and were then treated with test compounds. Cells were methionine-starved for 30 min by changing to a Met-free DMEM medium, with renewed test compound addition. A 1.5-mL aliquot of 35 S-methionine (30 mCi mL À1 , Hartmann Analytic) was added to each well of the plates, which were incubated for 30 min. The plates were placed on ice and medium was removed. The cells were washed with warm phosphate-buffered saline (PBS) twice and lysed by storing at À80 C for 30 min in lysis buffer (Promega). For quantification, protein was precipitated with trichloroacetic acid, collected, dried, transferred to scintillation liquid, and read in a scintillation counter.

Polysome profiling assay
HeLa cells were seeded in 15-cm diameter dishes and were left for 24 h to reach 70% confluency. Cells were treated with test compound for 2 h and cycloheximide (0.1 mg mL À1 ) was added. Cells were harvested on ice and then lysed using lysis buffer 300 mM NaCl, 15 mM MgCl 2 , 15 mM Tris (pH 7.5) containing 1 mg mL À1 heparin sulfate and 0.1 mg mL À1 cycloheximide, supplemented with 0.1% (v/v) Triton X-100. Cell lysates were loaded on sucrose gradients (10e60% sucrose in 12-mL Sorvall centrifuge tubes) made with the same buffer, and were ultracentrifuged using a SW40Ti (Beckman Coulter) centrifuge at 38,000 r.p.m. for 2 h. The gradients were then separated into fractions of 1 mL with continuous UV absorbance recording.

Author contributions
The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.