Structure of the Exon Junction Core Complex with a Trapped DEAD-Box ATPase Bound to RNA

See allHide authors and affiliations

Science  29 Sep 2006:
Vol. 313, Issue 5795, pp. 1968-1972
DOI: 10.1126/science.1131981


In higher eukaryotes, a multiprotein exon junction complex is deposited on spliced messenger RNAs. The complex is organized around a stable core, which serves as a binding platform for numerous factors that influence messenger RNA function. Here, we present the crystal structure of a tetrameric exon junction core complex containing the DEAD-box adenosine triphosphatase (ATPase) eukaryotic initiation factor 4AIII (eIF4AIII) bound to an ATP analog, MAGOH, Y14, a fragment of MLN51, and a polyuracil mRNA mimic. eIF4AIII interacts with the phosphate-ribose backbone of six consecutive nucleotides and prevents part of the bound RNA from being double stranded. The MAGOH and Y14 subunits lock eIF4AIII in a prehydrolysis state, and activation of the ATPase probably requires only modest conformational changes in eIF4AIII motif I.

In higher eukaryotes, the pre-mRNA splicing reaction leads to the deposition of the exon junction complex (EJC) on mature mRNAs; the EJCs are deposited at a conserved position located upstream of exon junctions and exclusively dictated by the splicing machinery (1). The core of the EJC is a heterotetramer that contains four proteins: eIF4AIII, MLN51, MAGOH, and Y14. The EJC constitutes a central effector of mRNA functions; it forms a stable and sequence-independent grip on mRNA and provides an anchoring point for nuclear and cytoplasmic factors participating in mRNA transport, translation, and quality control (2). Reconstitution of the EJC core revealed that it is an ATP-dependent complex. Its stable association with mRNA is maintained by MAGOH and Y14 through inhibition of the DEAD-box protein eIF4AIII ATPase activity (3). DEAD-box proteins use the energy from ATP hydrolysis to unwind double-stranded nucleic acid or rearrange RNA-protein complexes at virtually every step of the gene expression pathway (46). The translation initiation factor eIF4A (we refer to eIF4AI and its isoform eIF4AII collectively as eIF4A) and its close but functionally distinct homolog eIF4AIII are considered archetypal members of the DEAD-box family. By comparison to eIF4A, eight eIF4AIII specific sequence motifs (patches A to H, fig. S1) were identified (7).

To shed light on both the EJC core architecture and the molecular mechanism of DEAD-box proteins, we determined the crystal structure of a minimal reconstituted EJC core assembled on a poly(U) oligonucleotide mimicking the mRNA at 2.3 Å resolution (8) (table S1). The EJC core is an elongated complex with overall dimensions of 99 by 67 by 54 Å organized around eIF4AIII (Fig. 1A). Extensive intersubunit interactions lead to burial of 23 to 36% of their surface areas (table S2). As in VASA (9)—a DEAD-box ATPase that regulates translation of specific mRNAs during early development in Drosophila—the two domains of eIF4AIII adopt a closed conformation forming composite binding sites for the 5′-adenylyl-β-γ-imidodiphosphate (ADPNP) and the RNA. In the open conformation, found in the crystal structure of free eIF4AIII that was determined at 3.3 Å resolution (Fig. 1B), the two domains had rotated by 160° relative to each other compared with the closed conformation. For the MLN51 fragment expressed for this study (residues 137 to 283), only residues 170 to 194 and 216 to 246 could be traced, and the fragment only contains limited secondary structure (Fig. 1A). Residues 216 to 246 contact patches C, D, and E in eIF4AIII domain 1 (Fig. 1C and fig. S2), explaining why the mutation of MLN51 Tyr240 and Gly241 disrupts formation of the EJC core (3). Residues 170 to 194 are located at the 5′ end of the bound RNA and contact eIF4AIII domain 2 (Fig. 1C). The side chain of MLN51 Phe188 stacks with the base of RNA U1 (Fig. 2E), confirming that MLN51 directly contacts RNA and increases RNA binding efficiency when bound to eIF4AIII (3). The flexible non-conserved linker (10) between the two ordered MLN51 fragments allows the protein to remain associated with rather different conformations of eIF4AIII. Small angle x-ray scattering (SAXS) indicates that, both alone and in complex with the MLN51 fragment, eIF4AIII in solution adopts a conformation that resembles the open conformations of free eIF4AIII or eIF4AI (11) more than the closed conformation (fig. S2, E and F).

Fig. 1.

Structures of the EJC and free eIF4AIII. (A) The EJC viewed from the ATP side (left) and the RNA side (right) with domains 1 and 2 of eIF4AIII colored blue and green, respectively. MLN51 is shown in purple, Y14 in yellow, and MAGOH in red. The dotted line connects the two ordered fragments of MLN51. (B) The open conformation of eIF4AIII with domain 1 in the same orientation as in the left panel of (A). (C) Surface representation of eIF4AIII, in which the three other subunits are shown as Cα skeletons. Conserved DEAD-box motifs (4) and eIF4AIII specific patches (7) (fig. S1) are mapped.

Fig. 2.

Intersubunit and RNA contacts within the EJC. (A) Interaction footprint of MAGOH (red) and Y14 (yellow) on eIF4AIII and MLN51. (B) Interaction footprint of eIF4AIII (green) and MLN51 (purple) on MAGOH-Y14. (B) is rotated 180° relative to (A) around a vertical axis located between MAGOH-Y14 and eIF4AIII-MLN51. Residue numbers in MLN51, Y14, or MAGOH are preceded by m, y, or a, respectively. (C) Interaction of the C-terminal (C-term) residues of MAGOH with conserved motifs at the ATP binding site. Water molecules are marked w. (D) Packing of the linker for eIF4AIII domains 1 and 2 between MAGOH and motifs I and III at the ATP site. (E) Stereoview of the RNA bound to eIF4AIII with MLN51 forming the 5′ boundary of the binding pocket. Motifs Ia, Ib, IV, and V in eIF4AIII contribute to the RNA binding pocket. Residues shown in gray also participate in RNA binding but are not part of the DEAD-box motifs. Amino acid residues are labeled (28).

The structure of the MAGOH-Y14 heterodimer located at the 5′ pole of the EJC (oriented relative to the bound RNA; Fig. 1A) displays only subtle conformational changes relative to the free heterodimer (1214). Residues from the two C-terminal helices and one end of the large β sheet in MAGOH contact eIF4AIII domain 2 (Figs. 1A and 2, A to D). In the free eIF4AIII, the linker (Lys242 to Gly250) between the two eIF4AIII domains is buried between motifs V and VI in eIF4AIII domain 2 (fig. S3). Within the EJC, conserved residues of the linker form salt bridges or hydrogen bonds with Tyr34, Lys48, and Lys142 in MAGOH (Fig. 2D). The second loop in the MAGOH β sheet contacts both the two eIF4AIII domains and their linker (fig. S2C). Two loops in the MAGOH β sheet interact with residues 190 to 194 in MLN51 (Fig. 2, A and B, and fig. S2D), explaining why the mutation of Lys16 to Phe17 in MAGOH disrupts the assembly of the EJC (15). The C-terminal extremity of MAGOH points toward the eIF4AIII ATP binding site in the EJC. The only Y14-eIF4AIII contact is a salt bridge between Y14 Arg108 and eIF4AIII Asp401 (fig. S3C). Both residues are strictly conserved, and the double mutation (Leu106→Glu106 and Arg108→Glu108) in Y14 prevents association of MAGOH-Y14 with the rest of the EJC (15).

Six uracil nucleotides of RNA are tightly bound to the EJC (Fig. 2E and fig. S4) in agreement with the seven to nine nucleotides that are protected from ribonuclease (RNase) treatment (1, 3). Nucleotides U1 and U2 at the RNA 5′ end are in contact with domain 2 of eIF4AIII, U3 interacts with both domains, and U4 to U6 bind domain 1 of eIF4AIII. The conformation of the bound RNA and the protein-RNA interactions in the EJC are notably similar to those observed in VASA (fig. S4). Residues in the helicase motifs Ia, Ib, IV, and V in eIF4AIII exclusively contact the sugar-phosphate backbone of all the RNA residues (Fig. 2E). Similar to VASA (9), residues outside these motifs also interact with the RNA. All RNA phosphates are recognized by hydrogen bonds or salt bridges, and four of the six 2'OH groups are recognized by eIF4AIII (Fig. 2E), which explains how eIF4AIII can bind RNA in a sequence-independent manner (1). The conformation of nucleotides U1 to U4 is very close to that of one strand in double-stranded A-form RNA (A-RNA) (fig. S4). Between U4 and U5, there is a sharp kink in the ribose-phosphate backbone induced by eIF4AIII motif Ib. The conformation of nucleotides U5 and U6 also resembles that of A-RNA, and these nucleotides could potentially be part of double-stranded RNA (dsRNA), as could nucleotides upstream of U1. Based on the structure of VASA, it was suggested that the four nucleotides 5′ of the RNA kink (U1 to U4 in the EJC) could be part of dsRNA if minor structural rearrangements took place (9). Ourmodelingofa double-stranded A-RNA into both the EJC and VASA structures shows that binding of A-form dsRNA would require a large rearrangement in the highly conserved residues 192 to 200 (the post-II region; fig. S4). Therefore, in the closed conformation of eIF4AIII, both motif Ib and the post-II region constitute macromolecular wedges forcing strand separation within dsRNA, but nucleotides both upstream and downstream of U1 to U4 may still be base paired. These features explain how eIF4A can bind to dsRNA (16) and suggest a partial mechanism for unwinding and messenger ribonucleoprotein particle (mRNP) remodeling in the DEAD-box helicases. The requirement of 12 to 18 nucleotides for efficient eIF4A ATPase activation (17), which contrasts the 6 to 7 nucleotides found in the EJC and VASA structures (9), can be explained by assuming that initial RNA binding occurs to the open conformations of eIF4A (11) and eIF4AIII, which have been observed by crystallography and SAXS.

As for the RNA recognition, almost identical binding sites for ATP are observed in VASA (9) and the EJC (Fig. 3 and fig. S5). The ADPNP is sandwiched between domains 1 and 2 of eIF4AIII, interacting with the conserved motifs F, Q, I, II, V, and VI. The base is stacked between Phe58 and Tyr371. MAGOH interlocks several of the eIF4AIII sequence motifs at the ATP site (Figs. 2C and 3A). MAGOH Ile146 stabilizes the position of eIF4AIII Tyr371 after motif VI, and MAGOH Ile143 makes a hydrogen bond through a water molecule with Gln83 in eIF4AIII motif I. Finally, MAGOH Pro145 forms a hydrogen bond with Ala63 in the Q motif of eIF4AIII. In eIF4AIII motif III, Thr220 links Asp190 and His363 from motifs II and VI, respectively, explaining the importance of eIF4A motif III and Asp190 in linking ATP hydrolysis and RNA unwinding (18). In addition, Asp190 interacts electrostatically with Arg339 in motif V. This interaction is probably required for positioning motif V, which links the RNA and ATP binding sites. This motif undergoes a large conformational change as a consequence of the EJC formation, where it interacts with RNA, ATP, motif II, and motif VI, whereas in the free eIF4AIII it interacts with motif VI and the linker between eIF4AIII domains 1 and 2 (fig. S3). These differences suggest motif V as an additional important mediator of cooperativity between ATP and RNA binding in eIF4AIII (19) and eIF4A (20).

Fig. 3.

The ATP hydrolysis site in DEAD-box proteins. (A) The eIF4AIII nucleotide binding site with residues involved in nucleotide binding colored according to the conserved motifs. Tyr371 is shown in gray. Water molecules are shown as red spheres. wc, the catalytic water; wr, a water likely to function in a proton relay; wa, a water not found in VASA; w, water molecules coordinating Mg2+. (B) Close-up of the surroundings of the γ-phosphate in the EJC. (C) Overlay of the EJC and VASA (9) structures. The wt water is not present in the EJC but occupies the coordination position taken by eIF4AIII Thr89 in the EJC. Carbon atoms in VASA are colored orange; oxygen atoms, including water molecules, are pink. Residues are labeled with the eIF4AIII number followed by the VASA number. The phosphates are labeled gP, bP, and aP. (D) Close-up of motif I (also called P-loop or Walker A motif). The reorientation of the γ-phosphate in VASA compared with the EJC causes the geometry for the inline attack of the catalytic water to become more favorable in VASA.

The VASA helicase core structure represents an active state of the DEAD-box ATPase (9). Although the use of ADPNP in both the EJC and the VASA structures may cause subtle differences to the experimentally inaccessible ATP-bound states, by comparing the two otherwise almost identical structures of DEAD-box ATPases bound to RNA and ADPNP, we can examine the molecular basis for the inhibition of the eIF4AIII ATPase activity by MAGOH-Y14. The γ-phosphate is shielded from bulk water in the EJC with three water molecules trapped in its vicinity. One of these (wc, Fig. 3) is a likely candidate for a nucleophilic water molecule attacking the γ-phosphate during ATP hydrolysis. A slight reorientation of the γ-phosphate between the EJC and the VASA structures probably explains the different states of the two ATPases. In the EJC, the distance between the catalytic water and the γ-phosphorus is 3.52 Å, and the inline attack angle between wc, the phosphorus atom, and the nitrogen linking the γ- and β-phosphates is 160°, whereas the corresponding values in VASA are 3.25 Å and 176°. Thus, the geometry between the nucleophilic centers is more favorable for attack in VASA (Fig. 3D). The suggested importance of the γ-phosphate reorientation is supported by an unexpected homology to the mitochondrial F1-ATPase (fig. S5), which also suggests that a second water molecule (wr, Fig. 3) functions in a proton relay during ATP hydrolysis.

The γ-phosphate is coordinated by motifs I and VI, and given that there are no obvious differences in motif VI between the structures of EJC and VASA, its reorientation between the two structures must be caused by differences in motif I. The most notable change occurs at eIF4AIII Thr89. In the EJC, its side chain coordinates the Mg2+ ion, but with a distorted octahedral geometry. In VASA, after a 1.6 Å shift, it is no longer a Mg2+ ligand; instead, a water molecule (wt) coordinates the ion in a regular octahedral geometry (Fig. 3C). Considering the otherwise almost perfect agreement between the two structures, this shift most likely forms a major component in the transition from the inhibited ATPase in the EJC to the active ATPase in VASA. The absence of the threonine in the Mg2+ coordination sphere in VASA is unusual, but the imperfect coordination of Mg2+ in the EJC and the demonstrated ATPase and unwinding activity of the VASA fragment (9) strongly suggests its physiological relevance. In addition, the difference in the backbone around Ser84 in eIF4AIII and Thr291 in VASA (Fig. 3D) may also contribute to reorientation of the γ-phosphate. This difference may be partially a result of the interaction of Gln83 with the linker and through water with MAGOH (Fig. 2C). In contrast, we find it unlikely that the difference in side chain at this position could contribute significantly to the ATPase inhibition mechanism because (i) a serine in this position is also found in eIF4A, the ATPase activity of which is not inhibited by associated subunits; (ii) in eIF4AIII, the wa water (Fig. 3) partially compensates for the Thr methyl group found in VASA; and (iii) a model with a threonine side chain in the same position in the EJC as in VASA and with the γ-phosphate in the same location as observed in the EJC does not result in unfavorable van der Waals interactions, which could induce movement of the γ-phosphate.

To further investigate how MAGOH contributes to the inhibition of the eIF4AIII ATPase, we mutated residues in the region Lys41 to Asp43 of MAGOH, which forms contacts with both domains in eIF4AIII, and deleted or mutated MAGOH residues Pro145 to Ile146, which contact the Q motif and motif VI (Fig. 2, C and D). These mutants fall in two groups. Those that prevent EJC core assembly completely abolish ATPase inhibition by MAGOH as expected. However, some of our mutants from both regions still support EJC assembly but partially release the eIF4AIII ATPase inhibition (fig. S6), demonstrating that residues in both MAGOH regions, and perhaps in others as well, are important for ATPase inhibition.

The participation of the EJC in multiple mRNA metabolic events is made possible by its core complex constituting a dynamic binding platform for diverse processing factors. This is exemplified by the role played by the EJC in the surveillance pathway of nonsense-mediated mRNA decay (NMD). By recruiting the NMD factors Upf3b and Upf2 to newly synthesized mRNAs, the EJC serves to distinguish premature from normal translation termination codons (21, 22). Because eIF4AIII patches A and G are not involved in intermolecular contacts within the EJC core (Fig. 1C), these patches may, alone or in combination with other conserved surface areas on eIF4AIII or other core EJC subunits (Fig. 4), be part of potential binding sites for peripheral EJC subunits such as the export factors REF and TAP/p15 and the NMD factor Upf3b (2). Given that Upf3b has been proposed to associate with the EJC through its conserved basic C-terminal domain (23), the acidic patch A on the surface of eIF4AIII may interact with the C-terminal domain of Upf3b.

Fig. 4.

Conserved surface patches of the core EJC. Conserved areas in eIF4AIII are colored green, in MLN51 purple, in Y14 yellow, and in MAGOH red. The N-terminal helix (residues 3 to 35) of Y14 from RCSB entry 1HL6 (13) has been docked on our structure and partially covers a highly conserved red area visible in the right panel. The ordered fragment of Pym (28) fits nicely into a cleft between the C-terminal helix of eIF4AIII and Y14 without overlapping with eIF4AIII and MLN51. Whether Pym contributes (29) or competes (15) for the association with the EJC core of other EJC subunits like the NMD factor Upf3b remains to be elucidated. The orientations are the same as in Fig. 1A.

The similarity between eIF4AIII and VASA is notable at all levels. Globally, 336 Cα atoms from both domains of eIF4AIII can be superimposed to their equivalents in VASA with a root mean square deviation of only 1.1 Å. This implies that the closed conformation of all DEAD-box proteins could be similar, although the closed conformation of other subgroups of these ATPases may be more variable. At the level of RNA binding, the almost identical recognition made by VASA and eIF4AIII is remarkable and is likely to be universal for all DEAD-box ATPases. In addition, the presence of MLN51 shows how the fundamental RNA recognition provided by the ATPase can be supplemented and regulated by additional domains in the helicases or associated proteins.

DEAD-box proteins cycle between an ATP state with high affinity for RNA and an ADP/apo state with low affinity (4). In both the VASA and the EJC structures, hydrogen bonds and salt bridges to RNA are formed by both domains of the ATPase, explaining the high RNA affinity of the closed conformation. In the open conformation of the adenosine diphosphate (ADP)–bound state, or the apo state, the cooperative binding between the two domains will be absent in agreement with the low RNA affinity. Translation is required to displace the EJC from the mRNA (24, 25), and this probably requires ATP hydrolysis by eIF4AIII. Hence, the function of the eIF4AIII ATPase activity appears to be remodeling of the EJC mRNP particle, given that the structure and biochemical data clearly shows that ATP hydrolysis will induce dissociation of the EJC from RNA. The mechanism by which association between MAGOH-Y14 and eIF4AIII-MLN51 (3) causes ATPase inhibition can now be rationalized by our structure. Globally, MAGOH-Y14 stabilizes the closed conformation of eIF4AIII by embracing the ATPase and interacting with MLN51. A similar function has been suggested for eIF4G with respect to eIF4A (26), which must bind RNA and ATP in a manner almost identical to eIF4AIII. In addition, through interaction with the Q motif and the eIF4AIII domain linker, which both interact with motif I (Figs. 2C and 3A), MAGOH apparently stabilizes these regions in conformations freezing motif I and thereby the γ-phosphate. A function of the Q motif for regulating ATP binding and hydrolysis is consistent with mutational studies of the DEAD-box helicase DED1 (27).

Supporting Online Material

Materials and Methods

Figs. S1 to S6

Tables S1 and S2


References and Notes

View Abstract

Navigate This Article