Crystal structure of human eIF5A1: Insight into functional similarity of human eIF5A1 and eIF5A2

Eukaryotic initiation factor 5A (eIF5A) plays an essential role in the viability of eukaryotic cells.1-3 eIF5A is known to act as a translation initiation factor specific for a small number of mRNAs,4-6 a cellular target of HIV-1 REV protein,7 and an exportin-4-dependent nuclear export cargo.8 eIF5A is also involved in mRNA turnover9, 10 and the establishment of actin polarity.11 A single lysyl residue of eIF5A is post-translationally modified to the unusual amino acid hypusine, involving two enzymes12, 13: deoxyhypusine synthase (DHS) catalyzes the transfer of the 4-aminobutyl moiety of spermidine to the lysyl residue's ε-amino group to form deoxyhypusine, and deoxyhypusine hydroxylase oxidizes the 2′ position to yield the hypusyl residue. Knock-out of either the eif5a or dhs gene is fatal in yeast.2, 3, 14 Two eIF5A isoforms, eIF5A1 and eIF5A2, occur in humans and have 84% identity in their amino acid sequence.15-18 The eif5a1 gene is ubiquitously expressed, whereas eif5a2 is highly expressed in the testis and at moderate levels in the brain.18 The eIF5A2 protein is also detectable in colorectal and ovarian cancer-derived cell lines.18-20 Although the cancer-specific eif5a2 expression implies that eIF5A2 plays a role distinct from eIF5A1 in cancer cells, either human gene can restore the viability of a yeast eif5a mutant, suggesting functional similarity of the two human isoforms in eukaryotic cell survival.17 The amino acid sequences of the human isoforms are conserved near the hypusination site,17 suggesting that their function in cell survival might be linked to the hypusine modification. Structures of eIF5A from five different archaeal and protozoan species have been solved.21-23 There have been attempts at homology modeling of the 3D structure of human eIF5A.24, 25 The known limitations of homology models26 have prompted us to determine the structure of human eIF5A experimentally. The crystallographic model of human eIF5A1 at 2.5 Å resolution demonstrates the flexibility of the hypusination loop. Crystallographic mapping of amino acid residues that vary between eIF5A1 and eIF5A2 reveals that these variable sites are located away from the hypusination site. Finally, we propose a putative common binding mode of both eIF5A isoforms to DHS that involves the N-terminal domain of eIF5A. Residues 15–151 of the gene encoding eIF5A1 was PCR-amplified from its Mammalian Genome Collection cDNA clone (BC000751), and sub-cloned into the BseRI sites of a modified pET vector pET28-mhl (gi:134105571) using the In-Fusion PCR cloning system (Clontech). Protein was expressed in Escherichia coli strain BL21-CodonPlus(DE3)-RIL (Stratagene) by induction using 0.5 mM isopropyl-β-D-thiogalactopyranoside (IPTG). The harvested cell pellets were resuspended in a lysis buffer (20 mM HEPES pH 7.5, 300 mM NaCl, 5% Glycerol, 2 mM BME and 5 mM Imidazole) and disrupted using freeze-thaw and micro-fluidizer. The protein was affinity-chromatographically purified using Ni-NTA (Qiagen) beads. The Ni-NTA washing buffer consisted of 20 mM HEPES pH 7.5, 300 mM NaCl, 5% Glycerol, 2 mM BME and 30 mM to 75 mM of Imidazole, whereas the elution buffer contained 20 mM HEPES pH 7.5, 300 mM NaCl, 5% Glycerol, 2 mM BME and 300 mM Imidazole. The poly-His tag of the protein was removed using TEV protease, followed by gel filtration on a superdex-75 column (GE Healthcare). The gel filtration buffer consisted of 20 mM HEPES pH 7.5, 150 mM NaCl and 1 mM TCEP. Fractions containing the protein were collected and concentrated with an Amicon centrifugal filter (Millipore). The final protein sample was of better than 95% purity, as judged by SDS-PAGE. The crystallization sample had a protein concentration of 30 mg/mL in 20 mM HEPES pH 7.5, 150 mM NaCl, and 1 mM TCEP. A diffraction-quality crystal was grown at 18°C using 22.0% PEG 3350, 0.2M ammonium sulfate, 0.1M sodium cacodylate, pH 5.5, and a hanging-drop setup. It was briefly incubated in a cryo solution containing 20% PE3350 and 20% ethylene glycol and frozen in a nylon loop by rapid immersion into liquid nitrogen.27, 28 A data set of 332 0.5° oscillation images was collected on a FR-E (Rigaku) copper rotating anode equipped with Osmic confocal optics and a R-Axis IV++ detector and processed using HKL2000.29 Molecular replacement with PHASER30 placed one copy of the search model, (PDB code 1x6o) in the asymmetric unit, but failed to position the expected second molecule. Visually, the N-terminal and C-terminal domains of a second protein molecule were aligned consecutively with partial model-phased difference density. Interactive model rebuilding was done in COOT.31 CNS32 and REFMAC33 were both used for coordinate and temperature factor refinement. TLS groups were defined based on calculations on the TLSMD server.34 Model geometry was validated on the MOLPROBITY server.35 This model was deposited in the PDB as entry 3cpf. Diffraction data collection and refinement statistics are given in Table I. Diffraction data were of moderate quality. Data in the highest resolution shell were complete with 〈I〉/〈σI〉 of 3.3 while they exhibited high scaling residuals (Table I). The Wilson B factor39 was determined to be 66 Å2, suggesting significant thermal disorder in the atomic coordinates. As a result, the definition of side chain and peptide bond features were poor in extended areas of the electron density map. “Free” reflections for cross-validation38 were selected in thin resolution shells with the program SFTOOLS (B. Hazes, unpublished results) to reduce the potential for model bias during refinement in the presence of noncrystallographic symmetry (NCS). We attempted the imposition of NCS restraints during restrained coordinate and B-factor refinement. Initially, we refined the structure without imposing the NCS restraints, and then analyzed the RMS deviations between the Cα positions of two NCS-related molecules to notice significant differences in the relative orientations of the N-terminal domain (residues 15–83) to the C-terminal domain (residues 84–150). For example, least squares superposition of residues 15 through 150 resulted in a RMS deviation of 1.4 Å, whereas as the N- and C-termini were superimposed with deviations of 0.6 and 0.5 Å, respectively. However, when the domain-based NCS restraints of various strengths were imposed during model refinement, Rfree increased by several tenths of a percent. On the basis of this finding, we decided not to impose even the domain-based NCS restraints in the refinement. Other aspects of refinement parameterization, such as the number of TLS groups, were also guided by the response of Rfree in numerous trials. The application of TLS refinement resulted in a significant (>3%) drop in Rfree, at the cost of a small number of additional refinable parameters.40 Nevertheless, risks associated with the use of Rfree in decision-making during model refinement were discussed previously.41 Therefore, the TLS parameters should be interpreted with caution in this case. Our model of human eIF5A comprises amino acid residues 15 through 151. Two domains with an approximate inter-domain boundary at residue 83 can be distinguished. The N-terminal domain is dominated by β-strands where strands β1 and β6 form a double-stranded, anti-parallel sheet and a four-stranded, twisted, anti-parallel sheet is formed by strands β2 through β5 at the opposite side of the domain. The loop connecting β1 and β2 includes a one-turn 310-helix (η1, residues 22–24). The hypusination site, Lys50, is situated on the loop connecting β3 and β4. This hypusination loop, including the sequence 49GKHG52, is disordered. The C-terminal domain consists of a three-turn α-helix α2 and five strands β7-β11 [Fig. 1(a)]. The C-terminal half of β7 is part of the β7-β8-β9 anti-parallel sheet to which β9 is only loosely associated. β7's N-terminal portion contributes to the β7-β10-β11 anti-parallel sheet that is oriented approximately perpendicular to the β7-β10-β11 sheet, owing to a bend in β7. β9 and β10 are connected by three-turn helix α2 (residues 117–129). Structure of human eIF5A1 (a) Structural gallery of eIF5As. The structures are from human (left), Leishmania braziliensis (PDB code 1x6o), Leishmania mexicana (1xtd), Methanococcus jannaschii (1eif, 2eif), Pyrobaculum aerophilum (1bkb), and Pyrococcus horikoshii (1iz6). The two domains of the human structure are colored differently and its secondary structural elements are labeled. The location of the hypusination loop is indicated (black stick). Varied conformations of the other structures from the human structure are shown in orange. (b) Structure-based sequence alignment of two human eIF5As. Amino acid sequences of human eIF5A1 and eIF5A2 are aligned on the basis of the human eIF5A1 structure42 and their divergent residues are indicated (conservative changes in green and nonconservative changes in orange). The designation of each structural element is given above the amino acid sequence. Dotted line indicates the disordered hypusination loop, whereas x indicates the truncated residues, not included in the final structure. A blue arrowhead under the sequence indicates Lys50, the hypusination site, and two orange arrowheads indicate Gly52 and Lys55, of which mutations reduced the deoxyhypusine modification of Lys50.43 (c) Structural mapping of divergent residues between two human eIF5As. The divergent residues have the same color coding as in Figure 1b and are located at the C-terminal domain or away from the disordered hypusination loop at the bottom of the picture. The figure was obtained with PYMOL.44 A query on the secondary structure matching server45 against PDB structures found similarities between human eIF5A and its orthologues from Leishmania braziliensis (PDB code 1x6o), L. mexicana (1xtd), Methanococcus jannaschii (1eif, 2eif), Pyrococcus horikoshii (1iz6) and Pyrobaculum aerophilum (1bkb), and fungal Hex1 (1khi). Both the SCOP and CATH fold databases classify the N- and C-terminal domains of eIF5A as SH3-related barrel and OB-fold, respectively.46 Human eIF5A is the most similar to the structures from L. braziliensis (1x6o) and L. mexicana (1xtd) (RMSD of 1.28Å for 129 a.a. residues and RMSD of 1.28Å for 131 a.a residues, respectively) [Fig 1(a)]. Within the respective N-terminal domains, all contain a one-turn 310 helix and a long, protruding loop containing the hypusination site, although the hypusination loop is only fully visible in the crystal structure of L. mexicana. Within their respective C-terminal domains, all contain the three-turn α-helix and a similar β-strand layout. However, the β7-β8 loop is shorter in human eIF5A1 than in its Leishmania orthologs. Human eIF5A is also similar to the structures of two crystal forms of eIF5A from M. jannaschii (PDB code 1eif: tetragonal, 2eif: monoclinic; RMSD of 1.72Å for 117 a.a. residues and RMSD of 2.02 Å for 117 a.a. residues, respectively) [Fig. 1(a)]. When the N-terminal domains are compared, the difference is minimal. It is of interest that the tetragonal crystal form contained a partially defined hypusination loop including the hypusination site followed by two disordered residues while the monoclinic crystal included a completely ordered hypusination loop. Within the C-terminal domains, the differences are more obvious; for instance the three-turn α-helix found in human eIF5A is replaced by a loop in M. jannaschii. Human eIF5A is also similar to its P. horikoshii (1iz6) and P. aerophilum (1bkb) orthologs. In the C-terminal domains of these structures, single-turn rudiments of helix α1 can still be found. Taken together, the comparison of human eIF5A with the previously studied eIF5As revealed that the longer three-turn α-helix in the C-terminal domain found earlier in protozoan eIF5As is conserved in human eIF5A. A sequence comparison between the human eIF5A1 and eIF5A2 isoforms revealed several divergent residues [Fig. 1(b)]. Mapping these sites to the eIF5A1 structure, it became apparent that the majority of them are located on the C-terminal domain, whereas only five divergent residues occur in the N-terminal domain that included the hypusination site. Of those five residues at the N-terminal domain, two, A16S and F18Y, are remote to the hypusination site and can be considered conservative modifications [Fig. 1(b,c)]. These substitutions are therefore unlikely to cause a conformational change of eIF5A2 near the hypusination loop. Three additional variable sites on the N-terminal domain were not part of the truncated protein construct prepared for crystallization because this truncated region was predicted to be unstructured [Fig. 1(b)]. However, it is rare for the disordered N-terminal region to affect protein folding and stability.47 In the C-terminal domain, variable residues that have nonconservative alterations are mostly exposed, minimizing their impact on protein folding [Fig. 1(b,c)]. Some of the conservative variable residues are buried and could destabilize the eIF5A2 structure, but are limited to the C-terminal domain and are therefore unlikely to interfere with the hypusination of Lys50 on the N-terminal domain [Fig. 1(b,c)]. This finding is consistent with kinetic data showing that the Km values for association with DHS differ only slightly between the two human isoforms.17 Either eifa1 or eifa2 can complement growth of a yeast eif5a mutant,17 possibly because the amino acid sequences of the human eIF5A isoforms are conserved near the hypusination site.17 A previous mutational analysis approximates the minimal set of determinants for the modification of Lys50 of human eIF5A1.43 G52A and K55A mutations impair the deoxyhypusine modification of Lys50 but the mutations of residues surrounding Lys50 (i.e., K47A, K47D, K47R, G49A, H51A) do not affect it,43 suggesting a role of Gly52 and Lys55 in deoxyhypusine modification. In the human eIF5A1 structure, Gly52 is part of the disordered hypusination loop, and Lys55 is located toward the N-terminus of β4 [Fig. 1(b)]. Crystallographic mapping of divergent sites between human eIF5A1 and eIF5A2 suggests that the N-terminal domains of the two isoforms, including the N-terminal segment of β4 with residues Gly52 and Lys55 near the hypusination loop, are similar to each other. The DHS active site is a deep tunnel at the homodimer interfaces.48 It is likely that the β4 segment of eIF5A1 and the corresponding region of eIF5A2 bind to the active site tunnel of DHS in a similar manner, consistent with the Km values of the two isoforms.17 A crystal structure of the complex between DHS and eIF5A should reveal the binding mode between eIF5A and DHS.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.