The Crystal Structure of the
Exon Junction Complex Reveals How
It Maintains a Stable Grip on mRNA
Fulvia Bono,1Judith Ebert,1Esben Lorentzen,1and Elena Conti1,2,*
1European Molecular Biology Laboratory (EMBL), Meyerhofstrasse 1, D-69117 Heidelberg, Germany
2Max-Planck-Institute of Biochemistry, Am Klopferspitz 18, D-82152 Martinsried, Germany
The exon junction complex (EJC) plays a major
role in posttranscriptional regulation of mRNA
in metazoa. The EJC is deposited onto mRNA
during splicing and is transported to the cyto-
plasm where it influences translation, surveil-
lance, and localization of the spliced mRNA.
The complex is formed by the association of
four proteins (eIF4AIII, Barentsz [Btz], Mago,
and Y14), mRNA, and ATP. The 2.2 A˚resolution
structure of the EJC reveals how it stably locks
onto mRNA. The DEAD-box protein eIF4AIII
encloses an ATP molecule and provides the
binding sites for six ribonucleotides. Btz wraps
around eIF4AIII and stacks against the 50nucle-
otide. An intertwined network of interactions
anchors Mago-Y14 and Btz at the interface be-
tween the two domains of eIF4AIII, effectively
stabilizing the ATP bound state. Comparison
with the structure of the eIF4AIII-Btz subcom-
plex that we have also determined reveals that
large conformational changes are required
upon EJC assembly and disassembly.
The expression of protein-encoding genes in eukaryotes
involves a series of steps, starting from transcription,
through processing of the precursor mRNA (pre-mRNA),
to translation, and ultimately to decay in the cytoplasm.
Although these steps are mechanistically distinct, they
are functionally coupled, contributing to the efficiency
and fidelity of gene expression (reviewed in Moore,
2005). A well-known example in this context is the influ-
ence of nuclear splicing on translational yield, stability,
and localization of mRNAs, events that occur long after
the intron lariat has been excised (reviewed in Le Hir
et al., 2003). How the molecular memory of introns is car-
ried forward to the cytoplasm had been a puzzling ques-
tion until the discovery that splicing not only alters the
nucleotide sequence of the nascent transcript but also
deposits a protein complex, the exon junction complex
(EJC), as a landmark on the transcript (Le Hir et al., 2000).
In human cells, the EJC is assembled during the second
step of splicing, when the lariat has been formed and the
exons are being ligated (Reichert et al., 2002) (Shibuya
quence-independent manner at a fixed distance (20–24
nucleotides) upstream of the exon-exon junction (Le Hir
as the mature messenger ribonucleoprotein particle
(mRNP) is exported to the cytoplasm. The stable core of
the complex contains four proteins: Mago, Y14, eIF4AIII,
and Barentsz (Btz, also known as MLN51). As it proceeds
through the different cellular environments, the EJC core
acquires and loses additional factors. Peripherally associ-
ated proteinsincludesplicing coactivators (Srm160,Pinin,
RNPS1), mRNA export factors (REF/Aly, TAP-p15) and
factors involved in cytoplasmic mRNA surveillance (UPF3
and UPF2) (reviewed in Tange et al., 2004). Thus, the EJC
provides a platform for anchoring proteins that function at
different steps of mRNA metabolism. The complex is be-
lieved to be eventually removed by ribosomes in the first
or ‘‘pioneering’’ round of translation (Dostie and Dreyfuss,
2002; Lejeune et al., 2002).
Surveillance of mRNAs by the nonsense-mediated
mRNA decay (NMD) pathway has long been known to de-
pend on exon-exon boundaries in human cells (Cheng
et al., 1994). NMD elicits rapid degradation of aberrant
mRNAs containing a premature stop codon (PTC) if an
exon-exon junction is present at least 50 nucleotides
downstream of a stop codon (reviewed in Conti and Izaur-
ralde, 2005; Lejeune and Maquat, 2005). NMD can be re-
capitulated by tethering either Mago, Y14, eIF4AIII, or Btz
within the 30UTR of a reporter mRNA in human cells, as in
the case of the bona fide NMD factor UPF3 (Ferraiuolo
et al., 2004; Fribourg et al., 2003; Gehring et al., 2003;
Lykke-Andersen et al., 2000; Palacios et al., 2004; Shi-
buya et al., 2004). This indicates that when present down-
stream of a translation termination site, the EJC coordi-
nates a series of interactions necessary for NMD, thus
providing the direct molecular link for splicing-dependent
PTC recognition. When present upstream of the stop
Cell 126, 713–725, August 25, 2006 ª2006 Elsevier Inc. 713
codon (i.e., in a normal situation), the EJC increases the
translational yield of the mRNA. Tethering EJC core com-
enhances polysome association (Nott et al., 2004). The
ing on mRNA translation (Lu and Cullen, 2003; Nott et al.,
2003; Wiegand et al., 2003).
The core components of the human EJC are conserved
in other eukaryotes, with the notable exception of S. cere-
visiae. Drosophila Mago, Y14, eIF4AIII, and Btz are all
2001; Palacios et al., 2004; van Eeden et al., 1991). Oskar
mRNA localization is required for germ line and abdomen
formation in Drosophila and is dependent on splicing
(Hachet and Ephrussi, 2004). In contrast to what is ob-
served in human cells, neither splicing nor the EJC are re-
quired for NMD in Drosophila (Gatfield et al., 2003). These
data suggest that although a structurally homologous EJC
complex exists in Drosophila, the EJC splicing ‘‘mark’’ im-
plasmic events in this organism. In human cells, Btz has
also been found as a component of dendritic mRNPs in
hippocampal neurons, suggesting that in humans the
EJC might also be involved in either localization or regu-
lated expression of specific mRNAs (Macchi et al., 2003).
EJC assembly is splicing dependent in vivo but can be
recapitulated in vitro when all the constituent components
of the complex are simultaneously present (Ballut et al.,
2005; Tange et al., 2005). EJC assembly not only requires
the protein components but also depends on the pres-
ence of ATP and single-stranded RNA, both of which are
an integral part of the complex (Ballut et al., 2005). The
in vitro reconstituted EJC protects eight to nine nucleo-
tides from nuclease digestion, corresponding to the foot-
print characteristic of the EJC assembled in a splicing re-
action (Ballut et al., 2005; Le Hir et al., 2000).
While Btz displays no homology to proteins of known
function, eIF4AIII belongs to the well-studied family of
DEAD-box proteins (reviewed in Cordin et al., 2006). Pro-
teins belonging to this family have RNA-dependent
ATPase activity and range from acting as ‘‘molecular
motors’’ in unwinding nucleic acid duplexes (hence the
name ‘‘helicase’’) to RNA ‘‘place holders’’ as in the case
of eIF4AIII (Shibuya et al., 2004; Tanner and Linder,
2001). In the EJC, the ATPase activity of eIF4AIII is in-
hibited by the presence of Mago and Y14 (Ballut et al.,
2005). EJC incorporation is specific for eIF4AIII and does
not occur with the highly similar translation initiation factor
eIF4AI (Shibuya et al., 2006).
Understanding how the EJC is formed by the coopera-
tive action of all its constituents requires knowledge of
its three-dimensional structure, which we report here at
2.2 A˚resolution. The crystal structure of the Mago-Y14-
the binding of RNA within the EJC is stable and specific,
structure of the eIF4AIII-Btz subcomplex reveals a dra-
matic conformational change of the helicase domains as
compared to EJC bound eIF4AIII. More generally, we pro-
pose that similarly to what observed for protein kinases,
members of the DEAD-box family of proteins are likely to
adopt a similar structure in their ‘‘on’’ state (when bound
to ATP and RNA) despite having different conformations
in their ‘‘off’’ states.
Structure Determination and Quality
The human EJC was reconstituted with the components
previously described (Ballut et al., 2005). Crystals that
diffract to 3.2A˚resolution were obtained with thecomplex
of Mago (full length), Y14 (residues 66–174, referred to as
Y14DN), eIF4AIII (full length), Btz (the SELOR domain,
residues 137–286, Degot et al., 2004), a nonhydrolyzable
ATP analog (AMPPNP), and U15RNA. A crystal form that
diffracts to 2.2A˚resolution wasobtained byreconstituting
the EJC with a Y14 construct that also lacks the low-com-
plexity C terminal region (Y14DCDN, residues 66–154)
The 2.2 A˚structure of the EJC (containing Y14DCDN)
was determined by molecular replacement and refined
reochemistry, with 91.4% of the residues lying in the most
favored regions of the Ramachandran plot (see Table 1 for
data collection and refinement statistics). The final model
includes six bases of polyU RNA, AMPPNP, essentially all
of Mago-Y14 (Mago residues 4–146 and Y14 residues
66-154), most of eIF4AIII (except for the 20 N-terminal
residues that are disordered), and two ordered stretches
of Btz (residues 168–196 and 215–230) (Figure 1). The
3.2 A˚structure of the EJC containing Y14DN does not
show ordered electron density for the C-terminal region
of Y14, but in this crystal form additional residues of Btz
can be modeled (residues 168–196 and 214–248; Rfreeof
27.7%, R factor of 23.3%).
Overall Architecture of the EJC: A Close-Knit
Assembly Centered around eIF4AIII
The EJC structure resembles an L-shaped molecule, with
eIF4AIII-Btz at the base of the assembly and Mago-Y14
arranged almost perpendicularly to it (Figure 2). The core
structure of Mago-Y14 in the EJC is essentially identical
to that observed for the heterodimer in isolation (Fribourg
etal., 2003;Lau etal.,2003; Shiand Xu,2003). Magofolds
into a flat antiparallel b sheet flanked on one side by two
parallel a helices (blue, Figure 2). The a-helical surface of
Mago interacts with the RNA binding domain (RBD) of
Y14 (magenta, Figure 2). This protein-protein interaction
is mediated by the b sheet surface of the RBD that is
instead generally used in this type of folds to bind RNA
(reviewed in Hall, 2002). The b sheet surface of Mago is
entirely exposed to solvent (Figure 2), but part of it is
expected to interact with the N-terminal a helix of Y14
(Fribourg et al., 2003) that is not present in the construct
714 Cell 126, 713–725, August 25, 2006 ª2006 Elsevier Inc.
The domain organization of eIF4AIII (yellow in Figure 2)
is typical of the DEAD-box family of proteins, with two
RecA-like domains (reviewed in Cordin et al., 2006). The
two domains are arranged to form a deep interdomain
cleft where ATP binds (gray, Figure 2). Single-stranded
RNA (black, Figure 2) binds with an overall bent conforma-
tion in a shallow cleft across the two RecA-like domains
and is flanked on one side by Btz. Btz (red, Figure 2)
does not have a globular fold; instead it extends with
two separate stretches over both domains of eIF4AIII
and also contacts Mago. The Mago-Y14 heterodimer in-
teracts mainly with domain 2 of eIF4AIII but also docks
into the interdomain cleft. Thus, the components of the
EJC engage one another in a network of interactions
that are centered at eIF4AIII (summarized in Figure 1).
EJC Bound eIF4AIII Has a Closed Conformation
In eIF4AIII, the RecA-like domain 1 (residues 38–240) is
joined to domain 2 (residues 251–411) by a 10-residue
linker and is flanked by an N-terminal sequence that is
characteristic of eIF4AIII (Figure 1A). Each RecA-like do-
main is formed by a parallel b sheet with loops on both
loops contain the motifs that are characteristic of the
DEAD-box family of proteins (Cordin et al., 2006; Gorbale-
nyaandKoonin, 1993)and areinvolvedin eitherATPbind-
ing or RNA binding (shown in green for domain 1 and in
orange for domain 2 in Figures 3A and 3B).
Within the EJC, the two domains of eIF4AIII are in close
proximity and are oriented in such a way that the b sheets
lie almost perpendicular to each other. The overall confor-
mation of eIF4AIII in the EJC structure is similar to that ob-
(Figure 3C) (Sengoku et al., 2006). The two proteins super-
pose with a root mean square deviation (rmsd) of 1.7 A˚
over 369 a-carbon atoms, distributed over the entire mol-
ecule. This similarity is striking considering that previous
structural studies of DEAD-box proteins in the absence
of RNA have shown a variety of conformations. However,
the comparison between eIF4AIII and Vasa is the first for
two different DEAD-box proteins locked in the same on
state, i.e., in a productive complex with ATP and nucleic
ATP Is Buried within eIF4AIII and Is
Approached by Mago
The ATP binding site is at an interface between the two
RecA-like domains of eIF4AIII and is formed by the loops
atthe C-terminal ends of the domain1b strands(motifs Q,
I, II, and III) and loops at the N-terminal ends of the domain
2 b strands (motifs V and VI) (Figure 3A). Motif I is the so-
called Walker A motif (or P loop) and binds the ATP phos-
phate groups, in particular with Lys88 bridging the b and g
phosphates. Motif II is the Walker B motif (the Asp-Glu-
Ala-Asp sequence from which the name of the DEAD-
box family derives). Asp187 of motif II coordinates the
magnesium ion together with Thr89 from motif I (Figure 3E
and Figure S1). From domain 2, Asp342 of motif V con-
tacts the ribose (at the 30OH), and Arg367 and Arg370 of
motif VI interact with the g phosphate of ATP. Of the three
water molecules at the active site, one is the likely candi-
date for being activated by a general base for the in-line
nucleophilic attack in the ATPase reaction (Figure 3E).
This water is at 3.7 A˚from the ATP g-phosphate and is co-
ordinated by Glu188 (motif II) and His363 (motif VI), which
in turn is hydrogen bonded to Thr220 (motif III).
drogen bonds (with Gln65 in the Q motif) and by stacking
interactions (with Phe58 in the Q motif and with Tyr371 in
motif VI). The C-terminal residue of Mago (Ile146) makes
van der Waals contacts with Tyr371 of eIF4AIII, capping
the adenine binding site (Figure 3E). Comparison with
the structure of Vasa shows that ATP is held in place by
a similar set of interactions. The only interaction with
ATP that is not present in Vasa is the contribution of the
C terminus of Mago to the hydrophobic environment of
the adenine base. Instead, in Vasa a hydrophobic residue
from the N-terminal flanking region (Phe225) is positioned
at a similar structural position to Ile146 of Mago.
Six Ribonucleotides Are Cradled by eIF4AIII and Btz
Single-stranded RNA binds eIF4AIII on the opposite sur-
face from the ATP binding site (Figure 2). Although the
complex that was crystallized contained a U15RNA, or-
dered electron density is observed only for six nucleotides
(Figure S2). The nucleotides bind with the 30end interact-
with domain 2, where it is stacked against Btz (Figure 3B).
Contacts are made primarily by the ribose-phosphate
backbone of the nucleic acid, with the bases facing
The ribose and phosphate moieties bind in consecutive
binding pockets formed by the sequential arrangement of
the loops atthe C-terminal endof the bsheets, whichcon-
tain conserved helicase motifs (Figure 3B). Proceeding
from U2 to U6, five phosphates contact sequentially mo-
tifs IV, V, QxxR (from domain 2), and Ia, GG, Ib (from do-
main 1). The fourth nucleotide (U4) is at the interface of
the two domains, with its phosphate moiety interacting
tively) (Figures3Dand3F).Four ofthesix nucleotidesform
hydrogen bonds between the 20-OH groups of the riboses
and eIF4AIII: U1 with Asn285 (motif IV), U3 with Pro114
(motif Ia), U4 with Gln200, and U5 with Asp169 (motif Ib)
The uracyl bases form only a few specific contacts with
eIF4AIII (Figure 3D). At the 50end, U1 stacks against the
aromatic ring of Phe188 of Btz (Figure 3F). The bases of
U1, U2, U3, and U4 also stack on each other. Between
U4 and U5, the RNA bends dramatically, probably due
to the position of motif Ib, which would sterically interfere
with a straight RNA geometry (Figure 3B). This bent con-
formation of the RNA is similar to that observed in the
Vasa helicase (Figure 3C), where it was suggested to
play a role in unwinding of nucleic acids (Sengoku et al.,
Cell 126, 713–725, August 25, 2006 ª2006 Elsevier Inc. 715
716 Cell 126, 713–725, August 25, 2006 ª2006 Elsevier Inc.
Btz Wraps around eIF4AIII and Mediates
The SELOR domain of Btz is molded on the surface of
eIF4AIII with two separate stretches interacting with two
distinct sites (Figure 2). The N-terminal stretch (residues
168–196) binds to domain 2 of eIF4AIII and the C-terminal
stretch (residues 214–248) binds to domain 1. The two
across species (Figure 1B). They are connected by a flex-
ible linker, as judged by the lack of ordered electron den-
sity (represented by a dotted line in Figure 2A).
The C-terminal stretch of Btz binds to the loops at
the N-terminal end of the domain 1 b sheet. In this area
there are extensive hydrophobic interactions: Btz Trp218
Figure 1. Sequence Conservation and Interactions in the EJC
Structure-basedsequencealignment.Thesecondarystructureelementsareshown below thesequences(hfora helices, bforbstrands, anddotsfor
extended/loop regions). Conserved residues are colored according to sequence alignments including orthologs from H. sapiens (Hs), D. mela-
nogaster (Dm), and C.elegans (Ce) (shown), and also from X. laevis (Xl) and D. rerio (Dr) (not shown).
(A) In the eIF4AIII alignment, conserved sequence motifs shared by all DEAD-box helicases are boxed in gray and labeled (Cordin et al., 2006; Sen-
goku et al., 2006). Highlighted in yellow are residues that are conserved specifically in eIF4AIII orthologs (Hs, Dm, Ce, Xl, Dr) and not in eIF4AI (Hs and
Scincludedinthealignment,andalso Dm,Ce,Xl,Dr andSc,notincluded).Abovethe sequences, fullycolored circles identifyresiduesinvolvedinthe
interaction with ATP (gray circles), with RNA (black circles), with Mago (blue circles), with Y14 (magenta circles), and with Btz (red circles). Residues
shown by mutagenesis studies to affect protein-protein interactions or NMD are surrounded by a black square or circle, respectively (Oberer et al.,
2005; Shibuya et al., 2006).
(B–D) Btz (MLN51), Mago, and Y14 alignments. Conserved residues are highlighted in red (Btz), blue (Mago) and magenta (Y14). Fully colored circles
above the sequences are as in panel (A). Yellow circles identify the interaction sites for eIF4AIII. Other interactions known from previous structural
studies are indicated as colored empty circles above the sequences. In panels (C) and (D), empty orange circles identify residues of Mago-Y14 in-
teracting with PYM (Bono et al., 2004), while empty blue and magenta circles identify residues involved in the interaction between Mago and the N-
terminal helix of Y14 (Fribourg et al., 2003) that is missing in the EJC construct crystallized. Residues shown by mutagenesis studies to affect protein-
protein interactions or NMD are surrounded by a black square (panel B [Ballut et al., 2005] or a circle in panels C and D [Fribourg et al., 2003]).
Table 1. Crystallographic Statistics
Data Set EJC (Y14 DNDC) EJC (Y14 DN) eIF4AIII-Btz
Unit cell (A˚)a = b = 169.44 c = 71.04a = 69.15 b = 161.24 c = 193.24a = 71.76 b = 107.19 c = 243.95
Resolution range (A˚)a
50-2.2 (2.3-2.2) 50-3.2 (3.4-3.2)15-3.0 (3.2-3.0)
Unique reflections58631 3492618221
98.9 (95.6) 95.4 (97.2)94.2 (86.4)
11.1 (2.5)4.8 (2.0) 4.3 (2.0)
8.7 (54.5)24.7 (68.3)22.0 (61.8)
Resolution range (A˚)a
50-2.2 (2.3-2.2)50-3.2 (3.3-3.2) 15-3.0 (3.1-3.0)
21.9 (36.3) 27.7 (41.0)32.5 (43.3)
18.8 (30.1)23.3 (37.0)27.1 (35.7)
Rmsd bond (A˚)0.0140.0090.011
Rmsd angle (?)
B factor protein (A˚2)
B factor ligands (A˚2) 51 56
Most favored (%)91.4 86.583.7
Additionally allowed (%) 7.8 12.614.8
Generously allowed (%) 0.3 0.7 0.9
Disallowed (%) 0.5 0.20.6
aValues in parentheses correspond to the highest resolution shell.
Cell 126, 713–725, August 25, 2006 ª2006 Elsevier Inc. 717
and His220 are in van der Waals contact with Tyr205 and
Phe232 of eIF4AIII, while Btz Phe223 flanks eIF4AIII
Pro210 (Figure 4A). Hydrophilic contacts are made
by Btz Asp221 and Gln228, which are adjacent to
eIF4AIII Arg206 and Thr178, respectively. In addition,
eIF4AIII Asp154 and Tyr155 are in contact with Btz Tyr240
and Arg245, in a region of Btz that could only be modeled
in the 3.2 A˚crystal form. Mutation of Btz residues Trp218
and Asp221 impairs the interaction with eIF4AIII and the
localization of the Btz SELOR domain at nuclear speckles
(corresponding to the W49 and D52 mutations in Ballut
et al., 2005). Many of the helicase residues in this region
are conserved within eIF4AIII orthologs but are not pres-
ent in other highly similar proteins such as eIF4AI (see
yellow shading in Figure 1A), suggesting that this inter-
action site is important to confer specificity to eIF4AIII
A Network of Interactions between Btz, Mago,
Y14, and eIF4AIII
The N-terminal stretch of Btz binds domain 2 of eIF4AIII
with the main chain winding in between RNA and Mago
(Figure 4B). Several interactions within Btz appear impor-
tant to maintain such a conformation (in particular at
Arg184 and Phe188, Figure 4B). A subset of conserved
residues of Btz contact eIF4AIII: Btz Pro179 makes van
der Waals contacts with eIF4AIII Trp292, and Btz Arg184
and Arg194 form salt bridges with eIF4AIII Asp385 and
Asp391, respectively. A cluster of electrostatic inter-
actions is formed between Btz Glu190, Mago Lys41,
eIF4AIII Glu224, and Glu359 (Figure 4B). The side chains
of Mago, Lys41, Tyr40, Phe17, and Lys16 participate in
stacking interactions that also involve the aliphatic portion
of eIF4AIII, which in turn forms a salt bridge with Mago
Glu20 (Figure 4B).
Overall, this complex network of interactions results in
the insertion of Btz and of the two long loops of Mago
(spanning residues 14–19 and 37–45) into the interface
between the two domains of eIF4AIII. The importance of
these contacts is underscored by the effect of mutations
in the 16-17 loop of Mago that impair both NMD in human
cells and oskar mRNA localization in Drosophila embryos
(Fribourg et al., 2003; Newmark and Boswell, 1994). An-
other cluster of interactions is centered at the C-terminal
helix of eIF4AIII (residues 384–396) (Figure 4C). On one
side, Arg390 and Asp401 of eIF4AIII form a salt bridge
and together with Glu402 interact with Mago Tyr123 and
Gln127 and with Y14 Arg108. On the other side, the C-ter-
minal helix interacts via Asp391 with Mago Lys103 and
Btz Arg194 and contacts Asp43 of Mago via Tyr395.
Thus, the C-terminal helix of eIF4AIII is at the intersection
with the other three proteins of the complex.
Mago-Y14 Shields the Linker between
the eIF4AIII Domains
Adjacent to the C-terminal helix of eIF4AIII, we find the
linker that connects the two RecA-like domains wedged
in between the b sheet and the long a helix of Mago (Fig-
ure 4D). Overall, the linker region between the two RecA-
like domains is anchored with extensive interactions.
Figure 2. Structure of EJC
View of the human EJC in two orientations related by a 180?rotation about a vertical axis. In the complex, Btz (shown in red) stretches around the
of eIF4AIII. ATP (in gray) binds at an interface between the two domains of eIF4AIII, distinct from the RNA binding cleft. The other two protein com-
ponents of the EJC, Mago (blue), and Y14 (magenta), bind mainly to domain 2 of eIF4AIII, but the interaction surface also extends over to the interface
with domain 1. The dotted line in red shows the approximate path of a portion of Btz not present in the electron density (residues 198–213; Figure 1).
The helix at the C-terminal stretch of Btz is present in the 3.2 A˚resolution structure (shown), while it is partially disordered in the 2.2 A˚structure. The
two EJC structures are otherwise virtually identical. All ribbon drawings were rendered using PyMOL (DeLano, W.L., 2002, http://www.pymol.org).
718 Cell 126, 713–725, August 25, 2006 ª2006 Elsevier Inc.
The linker has a pronounced kink at residues 242–250.
Within the kinked region, Asp244 and Glu245 forma direct
salt bridge with Lys48 from the b sheet of Mago, and
Arg243 interacts with Asp43 from one of the two long
loops of Mago (the 37–45 loop, which in turns contacts
Btz, discussed above).
Proceeding from the kink toward domain 1, the linker
Lys242 engages in a salt bridge with Glu29, within the
N-terminal flanking region of eIF4AIII. From the kink to-
ward domain 2, the linker is involved in interactions with
both Mago and Y14 (Figure 4D): eIF4AIII Glu249 is flanked
by Mago Lys144 and contacts Asn101 and His103 of Y14.
This cluster of interactions holds the C-terminal end of
Mago so as to position Mago Ile146 at the ATP binding
site. Interestingly, in the case of Vasa, the linker follows
a similar path and is flanked by an insertion (residues
224–232) that occupies a similar structural position as
the C-terminal region of Mago.
Large Conformational Change between EJC Bound
and Unbound eIF4AIII
Thetwodomains ofeIF4AIII inthe EJCarehighlyintercon-
nected with ATP, RNA, and Mago-Y14. This prompted us
to ask whether the same interdomain orientation is pres-
ent in the absence of these factors. To determine the
structure of the eIF4AIII-Btz subcomplex, we used an eI-
F4AIII construct (residues 38–411) lacking the N-terminal
sequence that flanks the helicase domains and a Btz con-
struct trimmed to residues 137–250. The structure is re-
fined at 3.0 A˚resolution to an Rfreeof 32.5% and R factor
of 27.1%. The relatively high Rfreeis likely due to a com-
bination of weak diffraction data (Table 1) and of the sig-
nificant proportion of poorly ordered residues of Btz.
Nevertheless, the electron density map shows a good
agreement with the model, which includes residues 38–
408 of eIF4AIII and 217–225 of Btz (Figure S2).
The structure of the eIF4AIII-Btz reveals a striking differ-
ence in the conformation of the helicase as compared
to the conformation observed in the EJC. Superposition
of the two structures at domain 1 of eIF4AIII shows that
domain 2 occupies diametrically opposite positions (Fig-
ure 5A). In the eIF4AIII-Btz subcomplex, motifs Q, I, II,
and III of domain 1 face the solvent rather than facing
site is completely open and the RNA binding site is disrup-
ted. The pivot point for the swiveling motion between the
two domains appears to reside in the linker region, which
bends away with a roughly 180?angle from the path
observed in the EJC (Figure 5B). The diversion occurs at
residue 242 of eIF4AIII. In particular, Asp244 and Leu246
in the linker interact with motif VI, whereas in the EJC
structure Asp244 points toward Mago (Figure 4D) and
Leu246 points toward motif I instead (Figure 5B).
The contacts between the two domains of eIF4AIII are
EJC conformation. Contacts between domain 1 and 2 in
the eIF4AIII-Btz subcomplex include Gln81 (at the edge
of motif I) with Arg370 (in motif VI) and Arg238 with
Gln345 (at the edge of motif V). In contrast, within the
EJC, Arg238 packs against Phe28 from the N-terminal
flanking sequence of eIF4AIII (Figure 5B). Interestingly, in
both conformations the residues of eIF4AIII correspond-
ing to the main interaction site of eIF4AI with the middle
domain of eIF4G (Asp270, Asp273, Asn301, and Thr303;
Figure 1A and Oberer et al., 2005) are exposed to solvent
and accessible. It is known that eIF4AIII can also bind
eIF4G (Li et al., 1999), but whether the conformation it
acquires upon eIF4G binding is compatible with the
closed EJC bound conformation and whether binding
can also occur in the context of the EJC is currently
ATPase Inhibition in the EJC
The EJC has been proposed to function as a place holder
on mRNA (Shibuya et al., 2004), onto which it is locked by
inhibition of eIF4AIII ATPase activity (Ballut et al., 2005).
One possible mechanism for the inhibition is that in the
context of the EJC the ATPase site of eIF4AIII is disrupted.
box protein (Vasa) bound to ATP and RNA shows that in
the EJC eIF4AIII has a similar mode of ATP and RNA bind-
ing. All the residues at the ATP site appear to be in the
proper conformation for catalysis, including a water mole-
cule that is at the appropriate distance and with the cor-
rect geometry to carry out the nucleophilic attack required
for the ATPase reaction. Thus, the mechanism by which
the ATPase activity is inhibited in the EJC does not reside
in disruption of the active site. Rather, the ATPase inhibi-
tion appears to be due to locking the RNA bound confor-
mation of eIF4AIII. This is achieved by positioning Mago-
Y14 and Btz at the ‘‘closed’’ crevice that forms between
the two domains of eIF4AIII when ATP and RNA are
bound. Such positioning is mediated by an extensive
and intertwined set of interactions between all compo-
nents of the complex.
This mechanism of ATPase inhibition is consistent with
the results of previous studies. Mutations that impair the
Walker A and B motifs of eIF4AIII have no effect on either
EJC formation or NMD (Shibuya et al., 2006), in line with
the observation that the ATPase active site isnot involved.
Mutations that have been shown to affect NMD map at
residues that participate in crucial protein-protein interac-
terface between Mago and Y14 (at Y14 Leu118 and
Mago Leu136; Fribourg et al., 2003) and residues at the
interface between Mago, eIF4AIII, and Btz (Mago Lys16
lix of eIF4AIII, Shibuya et al., ). All constituents of the
complex are required to lock the RNA in the EJC. Indeed,
mutations in eIF4AIII that interfere with either protein inter-
actions (at the C-terminal region of eIF4AIII), withthe bind-
ing of the g-phosphate of ATP (at motif VI) or with RNA (at
motif Ib), have similar consequences in impairing EJC for-
mation in vitro and NMD in vivo (Shibuya et al., 2006).
Cell 126, 713–725, August 25, 2006 ª2006 Elsevier Inc. 719
Figure 3. ATP and RNA Bind at the Interface of the Two Helicase Domains of eIF4AIII and Are Flanked by Mago and Btz
(A)ATP(showninstickformat)issandwiched between conservedhelicase motifsfrombothdomain1ofeIF4AIII(motifsQ,I,II,andIII,showningreen)
and domain 2 (motifs V and VI, shown in orange). The rest of eIF4AIII is shown in gray, Mago is in blue. The molecule is viewed in a similar orientation
and colors as in Figure 2B (as are all other panels in this figure).
by Btz (in red). eIF4AIII is shown in gray, with the helicase motifs from domains 1 and 2 highlighted in green and orange, respectively, and labeled.
(C) Overlay of the structure of eIF4AIII when in the EJC (yellow, with RNA and ATP in black) with the structure of Vasa (light blue, with RNA and ATP in
cyan).ThetwoDEAD-boxproteinsassume asimilarconformation,haveasimilarmodeofATPbinding,andasimilarbend oftheRNA.Theother com-
ponents of the EJC are also partially visible in this view.
720 Cell 126, 713–725, August 25, 2006 ª2006 Elsevier Inc.
Mode of RNA Binding
The mode of RNA binding we observe in the EJC struc-
ture explains several biochemical observations. First,
eIF4AIII and Btz are the only components that contact
RNA directly in the complex, as predicted from UV-
crosslinking experiments (Ballut et al., 2005; Shibuya
et al., 2004). It also explains why the complex is resistant
to RNase A (Ballut et al., 2005), as the RNA is cradled in
such a way that the backbone is inaccessible from the
solvent. While we observe direct interactions with six
nucleotides, biochemically eight to nine nucleotides are
protected in the EJC from nuclease digestion (Ballut
et al., 2005; Le Hir et al., 2000). It is possible that the re-
maining two to three nucleotides do not bind tightly to
the complex but are nevertheless sterically inaccessible
The extensive contacts we observe with the 20-OH
in the assembly of the EJC, which does not form with DNA
(data not shown). The paucity of specific interactions with
Figure 4. Interaction Networks between the Protein Components of the EJC
of Btz residues contacting a region of the DEAD-box protein that is conserved in eIF4AIII orthologs but not in paralogs such as eIF4AI.
(B) Group of interactions between Mago (blue), Btz, and eIF4AIII. Mago and Btz protrude into the cleft that is formed between the two domains of
(C) The C-terminal helix of eIF4AIII engages in a cluster of interactions between Y14 (magenta) and Mago.
into Mago-Y14. It interacts on one side with the loops of Mago shown in panel (B) and on the other side with Y14 and with the C-terminal region of
Mago (see Ile146 in Figure 3E).
(D) Schematic drawing of the interactions that mediate RNA binding within the EJC. Residues boxed in green belong to domain 1 of eIF4AIII, residues
boxed in orange belong to domain 2, and in red to Btz.
(E) Close-up view of the ATP binding site (same view as panel A), showing a subset of conserved residues recognizing the phosphates, ribose, and
adenine moieties of ATP (shown in sticks), or the magnesium ion (shown as a yellow sphere) and associated water molecules (shown as light blue
spheres). A dotted line identifies the water molecule likely to carry out the nucleophilic attack on the ATP g-phosphate. Colors are as in Figure 1.
Ile146 of Mago is highlighted in blue. See also Figure S1.
Cell 126, 713–725, August 25, 2006 ª2006 Elsevier Inc. 721
the bases provides a structural rationale for how the EJC
can form in a sequence-independent manner. When re-
constituted with recombinant proteins, the EJC does not
form at a specific position on the RNA. Why the EJC is de-
posited in a position-dependent manner upon splicing
ing machinery that is required to assemble the complex
Figure 5. Large Conformational Change of eIF4AIII
(A) The structure of eIF4AIII in the EJC (eIF4AIII in yellow, Btz in red, and Mago-Y14 in light gray) is overlayed to the structure of the eIF4AIII-Btz sub-
complex (eIF4AIII in green and Btz in purple). The two structures are shown with domain 1 superposed. Domain 2 is swiveled away from its position in
(B) On the left is the structure of the EJC bound state of eIF4AIII and on the right of the Btz bound subcomplex. The molecules are viewed in the same
orientation of panel (A). A subset of eIF4AIII residues are involved in different interactions in the EJC bound conformation (on the left) and in the Btz
bound subcomplex (on the right).
722 Cell 126, 713–725, August 25, 2006 ª2006 Elsevier Inc.
The RNA conformation has a pronounced bend, similar
to that observed in the structure of Vasa, an unwinding
DEAD-box helicase (Sengoku et al., 2006). Given that
the EJC functions as a place holder on RNA rather than
as an unwinding motor, the RNA bending we observe in
the complex is unlikely to function in strand separation.
It is possible that RNA bending has a role in unwinding
nucleic acids when eIF4AIII is in isolation or that it has a
different role in the EJC (for example, in positioning pro-
teins bound upstream and downstream on the mRNA
with a particular geometry that might be important for in-
teractions). It is also possible that RNA bending is simply
the outcome of the geometry of secondary structure ele-
ments in DEAD-box helicases and has no direct function.
In general, the similarity between eIF4AIII and Vasa in the
on state (bound to ATP and RNA) suggests that other
DEAD-box proteins will adopt a similar conformation to
satisfy the chemical constraints of the active state. In con-
trast, DEAD-box proteins are not subject to such con-
straints in the off state and therefore do not necessarily
adopt a particular conformation. This situation is highly
reminiscent to what observed for protein kinases (re-
viewed in Huse and Kuriyan, 2002).
Implications for EJC Assembly and Disassembly
At the biochemical and structural level, the EJC can
be dissected into two subcomplexes: Mago-Y14 and
eIF4AIII-Btz. Biochemically, they form stable hetero-
dimers (see Experimental Procedures and also Fribourg
et al., 2003; Shibuya et al., 2006). Structurally, the inter-
actions within each heterodimer are mediated by signifi-
cant hydrophobic contacts. In contrast, the interactions
between Mago-Y14 and eIF4AIII-Btz in the EJC are mainly
hydrophilic and are cushioned by a layer of water mole-
cules. This is consistent with the notion that unzipping
of macromolecular complexes occurs through interfaces
held together by hydrogen bonds, suggesting that the
two subcomplexes might be an intermediate step in
EJC assembly/disassembly (Figure S3). In this model,
upon EJC incorporation, while Mago-Y14 behaves as
a rather rigid unit. The transition between the open off
state of eIF4AIII (in the eIF4AIII-Btz subcomplex) and the
closed on state (in the EJC) is likely mediated by the flex-
ible linker of the two RecA-like domains and coordinated
by the binding of all constituents of the complex.
A similar conformational change would also occur upon
EJC disassembly. A possible scenario for EJC disassem-
bly is the dissociation of Mago-Y14 by the interaction of
translating ribosomes and/or additional factors. The cyto-
solic protein PYM is known to be connected to EJC com-
ponents and involved in NMD, as shown by the effect of
mutating the interface between PYM and Mago (Bono
et al., 2004). The binding mode of PYM to Mago-Y14 ob-
served in the structure of the PYM-Mago-Y14 ternary
complex (Bono et al., 2004) would not be compatible in
the EJC, as PYM Tyr33 would clash with Pro404 and
Met405 of eIF4AIII (Figure S4). This steric clashing would
be even more dramatic in the context of full-length PYM,
raising the interesting possibility that the function of
PYM in NMD might reside in a disassembly step of the
EJC in the cytosol. Upon dissociation of Mago-Y14, hy-
drolysis of ATP would occur promptly since the ATPase
active site of eIF4AIII is already correctly positioned for
catalysis. Release of the g-phosphate of ATP would in
turn destabilize the RNA bound conformation.
This model does not exclude that another intermediate
tween eIF4AIII and Btz. While Mago-Y14 is likely a consti-
tutive heterodimer, the structure of eIF4AIII-Btz has a
rather small interaction site, suggesting that this subcom-
plex might be more labile. This is consistent with localiza-
tion data: at steady state, Mago, Y14, and eIF4AIII localize
2001; Kim et al., 2001; Le Hir et al., 2001; Palacios et al.,
2004). Incontrast, Btzislocalized mainly in thecytoplasm,
while the nuclear speckles localization of its SELOR do-
main depends on the interaction with eIF4AIII (Ballut
et al., 2005; Shibuya et al., 2004; Palacios et al., 2004).
Btz appears to function with an ‘‘anchoring’’ site and an
‘‘effector’’ site. The anchoring site (the C-terminal stretch)
is essential and sufficient to bind to domain 1 of eIF4AIII,
even in the absence of other EJC components. The effec-
tor site (the N-terminal stretch) interlocks the complex
when the other EJC components are present by binding
simultaneously to eIF4AIII, RNA, and Mago-Y14. The an-
choring is achieved on a surface of eIF4AIII that is con-
ralogs, including eIF4AI. This is likely to provide the means
of discriminating eIF4AIII from closely related DEAD-box
proteins, many of which are present in the splicing ma-
chinery and are not incorporated into the EJC. On the
other hand, eIF4AIII does not enhance translation initia-
tion, in contrast to eIF4AI (Li et al., 1999). Structural stud-
ies will ultimately be required to understand what is key to
the precision of the translation initiation machinery in dis-
criminating against eIF4AIII. Finally, the mechanism by
which the EJC orchestrates the downstream networks of
interactions leading to NMD, mRNA localization, and en-
hanced incorporation into polysomes are open questions
for future studies.
Full-length (fl) human eIF4AIII (residues 1–412; untagged) and the
SELOR domain of Btz (residues 137–286; TEV-cleavable GST tagged)
were cloned into a bicistronic vector (pETMCN-GST, gift from
C. Romier, IGBMC, Strasbourg) and coexpressed in E. coli. Cells
were lysed in 20 mM Tris-HCl pH 7.5, 300 mM NaCl, 1 mM DTT,
10% glycerol. The complex was purified by Glutathione Sepharose
4B (GE Healthcare), cleaved with TEV protease, and further purified
on a Heparin Sepharose CL-4B column (GE Healthcare). The complex
The complexes of fl human Mago (residues 1–147) with Y14DN
(residues 66–174) or with Y14DNDC (66–154) were prepared by coex-
pression from a bicistronic vector (pETMC-His, gift from C. Romier,
Cell 126, 713–725, August 25, 2006 ª2006 Elsevier Inc. 723
IGBMC, Strasbourg) and purified with a similar protocol as previously
described for the Drosophila orthologs (Fribourg et al., 2003). For EJC
complex formation, eIF4AIII(fl)-Btz(137–286) was incubated with 2 mM
AMP-PNP, 2 mM MgCl2, and a 1.53 molar excess of a U15RNA (or an
A20RNA) (Dharmacon) to which was added Mago-Y14DN (or Mago-
Y14DNDC). The mixture was purified using size-exclusion chromatog-
raphy (Superdex 200) in the presence of 3 mM MgCl2. Although the
EJC is stable in the presence of ATP (Ballut et al., 2005), the nonhydro-
lysable AMP-PNP analog was used to ensure sample homogeneity
Crystallization and Data Collection
AMPPNP grew at 18?C in 8% PEG 6000, 100 mM MgCl2, and
100 mM MES pH 6.0, a condition of the ‘‘Complex screen’’ (Radaev
et al., 2006). Smaller crystals were also obtained with the EJC recon-
stituted with the A20RNA. The U15RNA-containing crystals were opti-
mized by microseeding and grew to a size of 80 3 80 3 200 mm. They
diffract to 2.2 A˚resolution and contain one EJC per asymmetric unit.
obtained at 18?C in 10% PEG 4000, 200 mM ammonium-acetate,
10 mM CaCl2, and 50 mM Na cacodylate pH 6.5. The crystals grew
as needles to a maximum size of 20 3 20 3 150 mm. They diffract to
3.2 A˚resolution and contain two EJCs per asymmetric unit. Crystals
of eIF4AIII(38–412)-Btz(137–250) grew at 18?C in 700 mM diammo-
nium tartrate, 100 mM Na acetate pH 4.6 as needles of 15 3 15 3
150 mm dimensions. They diffract to 3 A˚resolution and contain two
molecules per asymmetric unit.
All crystallographic data were measured at the PX beamline X06SA
at the Swiss Light Source (SLS), Switzerland, using the microdiffrac-
tometer in the case of the needle-like crystal form. For data collection,
all crystals were stabilized in a solution consisting of the mother liquor
supplemented with 30% glycerol and flash cooled in a nitrogen stream
at 100K. Data were indexed and scaled using XDS (Kabsch, 1993).
Crystal Structure Determination
Molecular replacement was carried out with PHASER (McCoy et al.,
2004) and refinement with REFMAC (Murshudov et al., 1997). The
search models for the EJC included the Mago-Y14 structure (PDB
entry 1P27) and the two separate domains of eIF4A, with which human
eIF4AIII shares 62% sequence identity (PDB entries 1FUU and 1FUK).
The Arp/Warp program for automated model building (Perrakis et al.,
1999) was used to trace the initial model in the 2.2 A˚resolution EJC
structure. The model was completed using iterative cycles of model
buildingin Coot(Emsley andCowtan, 2004)and restrained refinement.
The refinement was carried out by defining each domain/protein as
a TLS group in the modeling of anisotropy within REFMAC.
The refined 2.2 A˚resolution model of EJC containing Y14DNDC was
used as search model for the 3.2 A˚crystal form (with Y14DN). Refine-
ment in REFMAC was carried out with tight noncrystallographic sym-
metry restraints between the two copies of the EJC in the asymmetric
unit. For the binary complex, the two domains of eIF4AIII were used as
models for the molecular replacement. Refinement was carried out
with tight noncrystallographic symmetry restraints between the two
molecules, with each eIF4AIII domain defined as a TLS group in the
modeling of anisotropy in REFMAC.
Supplemental Data include four figures and can be found with this
article online at http://www.cell.com/cgi/content/full/126/4/713/DC1/.
We are grateful to C. Shulze-Briese, T. Tomizaki, and E. Pohl at the
Swiss Light Source (Zu ¨rich) for assistance during data collection,
D. Linder for technical assistance, J. Basquin and A. Scholz at the
Crystallization Facility (EMBL-Heidelberg), and C. Roome for com-
puter assistance. We thank P. Brick, E. Izaurralde, A. Cook, M. Jinek,
A. Jeyaprakash, and F. Glavan for discussions and critical reading of
the manuscript, and S. Fribourg for initial stages of this work. We
also thank A.V. Bono for the drawing in Figure S3. This study was sup-
ported by the Human Frontier Science Program Organization (HFSPO)
and F.B was supported by the EU contract HRPN-CT-00239.
Received: July 17, 2006
Revised: July 31, 2006
Accepted: August 9, 2006
Published: August 24, 2006
Ballut, L., Marchadier, B., Baguet, A., Tomasetto, C., Seraphin, B., and
Le Hir, H. (2005). The exon junction core complex is locked onto
RNA by inhibition of eIF4AIII ATPase activity. Nat. Struct. Mol. Biol.
Bono, F., Ebert, J., Unterholzner, L., Guttler, T., Izaurralde, E., and
Conti, E. (2004). Molecular insights into the interaction of PYM with
the Mago-Y14 core of the exon junction complex. EMBO Rep. 5,
Cheng, J., Belgrader, P., Zhou, X., and Maquat, L.E. (1994). Introns are
cis effectors of the nonosense-codon-mediated reduction in nuclear
mRNA abundance. Mol. Cell. Biol. 14, 8219–8228.
Conti, E., and Izaurralde, E. (2005). Nonsense-mediated mRNA decay:
molecular insights and mechanistic variations across species. Curr.
Opin. Cell Biol. 17, 316–325.
Cordin, O., Banroques, J., Tanner, N.K., and Linder, P. (2006). The
DEAD-box protein family of RNA helicases. Gene 367, 17–37.
Degot, S., Le Hir, Alpy, F., Kedinger, V., Stoll, I., Wendling, C., Sera-
phin, B.,Rio,M.C.,andTomasetto,C.(2004). Association of thebreast
cancer protein MLN51 with the exon junction complex via its speckle
localizer and RNA binding module. J. Biol. Chem. 279, 33702–33715.
Dostie, J., and Dreyfuss, G. (2002). Translation is required to remove
Y14 from mRNAs in the cytoplasm. Curr. Biol. 12, 1060–1067.
Emsley, P., and Cowtan, K. (2004). Coot: model-building tools for
molecular graphics. Acta Crystallogr. D Biol. Crystallogr. 60, 2126–
Ferraiuolo, M.A., Lee, C.S., Ler, L.W., Hsu, J.L., Costa-Mattioli, M.,
Luo, M.J., Reed, R., and Sonenberg, N. (2004). A nuclear translation-
like factor eIF4AIII is recruited to the mRNA during splicing and func-
tions in nonsense-mediated decay. Proc. Natl. Acad. Sci. USA 101,
Fribourg, S., Gatfield, D., Izaurralde, E., and Conti, E. (2003). A novel
mode of RBD-protein recognition in the Y14-Mago complex. Nat.
Struct. Biol. 10, 433–439.
Gatfield, D., Unterholzner, L., Ciccarelli, F.D., Bork, P., and Izaurralde,
E.(2003). Nonsense-mediated mRNA decayinDrosophila:attheinter-
section of the yeast and mammalian pathways. EMBO J. 22, 3960–
(2003). Y14 and hUpf3b form an NMD-activating complex. Mol. Cell
Gorbalenya, A.E., and Koonin, E.V. (1993). Helicases: amino acid
sequence comparisons and structure-function relationships. Curr.
Opin. Struct. Biol. 3, 419–429.
Hachet, O., and Ephrussi, A. (2001). Drosophila Y14 shuttles to the
posterior pole of the oocyte and is required for oskar mRNA transport.
Curr. Biol. 11, 1666–1674.
Hachet, O., and Ephrussi, A. (2004). Splicing of oskar RNA in
the nucleus is coupled to its cytoplasmic localization. Nature 428,
724 Cell 126, 713–725, August 25, 2006 ª2006 Elsevier Inc.
Hall, T.M. (2002). RNA-protein interactions. Curr. Opin. Struct. Biol.12,
Huse, M., and Kuriyan, J. (2002). The conformational plasticity of pro-
tein kinases. Cell 109, 275–282.
Kabsch, W. (1993). Automatic processing of rotation diffraction
data from crystals of initially unknown symmetry and cell constants.
J. Appl. Crystallogr. 26, 795–800.
Kataoka, N., Diem, M.D., Kim, V.N., Yong, J., and Dreyfuss, G. (2001).
Magoh, a human homologue of Drosophila mago nashi protein, is
a component of the splicing-dependent exon-exon junction complex.
EMBO J. 20, 6424–6433.
(2001).The Y14protein communicatestothecytoplasm theposition of
exon-exon junctions. EMBO J. 20, 2062–2068.
Lau, C.K., Diem, M.D., Dreyfuss, G., and Van Duyne, G.D. (2003).
Structure of the Y14-Magoh core of the exon junction complex. Curr.
Biol. 13, 933–941.
Le Hir, H., Izaurralde, E., Maquat, L.E., and Moore, M.J. (2000). The
spliceosome deposits multiple proteins 20-24 nucleotides upstream
of mRNA exon-exon junctions. EMBO J. 19, 6860–6869.
Le Hir, H., Gatfield, D., Izaurralde, E., and Moore, M.J. (2001). The
exon-exon junction complex provides a binding platform for factors in-
volved in mRNA export and nonsense-mediated mRNA decay. EMBO
J. 20, 4987–4997.
Le Hir, H., Nott, A., and Moore, M.J. (2003). How introns influence
and enhance eukaryotic gene expression. Trends Biochem. Sci. 28,
Lejeune, F., and Maquat, L.E. (2005). Mechanistic links between non-
sense-mediated mRNA decay and pre-mRNA splicing in mammalian
cells. Curr. Opin. Cell Biol. 17, 309–315.
Lejeune, F., Ishigaki, Y., Li, X., and Maquat, L.E. (2002). The exon junc-
tion complex is detected on CBP80-bound but not eIF4E-bound
mRNA in mammalian cells: dynamics of mRNP remodeling. EMBO J.
Li, Q., Imataka, H., Morino, S., Rogers, G.W., Richter-Cook, N.J., Mer-
rick, W.C., and Sonenberg, N. (1999). Eukaryotic translation initiation
factor 4AIII (eIF4AIII) is functionally distinct from eIF4AI and eIF4AII.
Mol. Cell. Biol. 19, 7336–7346.
Lu, S., and Cullen, B.R. (2003). Analysis of the stimulatory effect of
splicing on mRNA production and utilization in mammalian cells.
RNA 9, 618–630.
Lykke-Andersen, J., Shu, M.D., and Steitz, J.A. (2000). Human Upf
proteins target an mRNA for nonsense-mediated decay when bound
downstream of a termination codon. Cell 103, 1121–1131.
Macchi, P., Kroening, S., Palacios, I.M., Baldassa, S., Grunewald, B.,
Ambrosino, C., Goetze, N., Lupas, A., St Johnston, D., and Kiebler,
M.A. (2003). Barentsz, a new component of the Staufen-containing ri-
bonucleoprotein particles inmammaliancells, interacts with Staufenin
an RNA-dependent manner. J. Neurosci. 23, 5778–5788.
McCoy, A.J., Storoni, L.C., and Read, R.J. (2004). Simple algorithm
for a maximum likelihood SAD function. Acta Crystallogr. D60, 1220–
Mohr, S.E., Dillon, S.T., and Boswell, R.E. (2001). The RNA-binding
protein Tsunagi interacts with Mago Nashi to establish polarity and lo-
calize oskar mRNA during Drosophila oogenesis. Genes Dev. 15,
Moore, M.J. (2005). From birth to death: the complex lives of eukary-
otic mRNAs. Science 309, 1514–1518.
Murshudov, G.N., Vagin, A.A., and Dodson, E.J. (1997). Refinement of
macromolecular structures by the maximum-likelihood method. Acta
Crystallogr. D53, 240–255.
Newmark, P.A., and Boswell, R.E. (1994). The mago nashi locus
encodes an essential product required for germ plasm assembly in
Drosophila. Development 120, 1303–1313.
Nott, A., Meislin, S.H., and Moore, M.J. (2003). A quantitative analysis
of intron effects on mammalian gene expression. RNA 9, 607–617.
Nott, A., Le Hir, H., and Moore, M.J. (2004). Splicing enhances trans-
lation in mammalian cells: an additional function of the exon junction
complex. Genes Dev. 18, 210–222.
Oberer, M., Marintchev, A., and Wagner, G. (2005). Structural basis for
the enhancement of eIF4A helicase activity by eIF4G. Genes Dev. 19,
Palacios, I.M., Gatfield, D., St Johnston, D., and Izaurralde, E. (2004).
An eIF4AIII-containing complex required for mRNA localization and
nonsense-mediated mRNA decay. Nature 427, 753–757.
Perrakis, A., Morris, R., and Lamzin, V.S. (1999). Automated protein
model building combined with iterative structure refinement. Nat.
Struct. Biol. 6, 458–463.
Radaev, S., Li, S., and Sun, P.D. (2006). A survey of protein-protein
complex crystallizations. Acta Crystallogr. D62, 605–612.
Reichert, V.L., Le Hir, H., Jurica, M.S., and Moore, M.J. (2002). 50exon
interactions within the human spliceosome establish a framework for
exon junction complex structure and assembly. Genes Dev. 16,
Sengoku, T., Nureki, O., Nakamura, A., Kobayashi, S., and Yokoyama,
drosophila Vasa. Cell 125, 287–300.
Shi, H., and Xu, R.M. (2003). Crystal structure of the Drosophila Mago
nashi-Y14 complex. Genes Dev. 17, 971–976.
Shibuya, T., Tange, T.O., Sonenberg, N., and Moore, M.J. (2004).
eIF4AIII binds spliced mRNA in the exon junction complex and is
essential for nonsense-mediated decay. Nat. Struct. Mol. Biol. 11,
Shibuya, T., Tange, T.O., Stroupe, M.E., and Moore, M.J. (2006).
Mutational analysis of human eIF4AIII identifies regions necessary
for exon junction complex formation and nonsense-mediated mRNA
decay. RNA 12, 360–374.
Tange, T.O., Nott, A., and Moore, M.J. (2004). The ever-increasing
complexities of the exon junction complex. Curr. Opin. Cell Biol. 16,
Tange, T.O., Shibuya, T., Jurica, M.S., and Moore, M.J. (2005). Bio-
chemical analysis of the EJC reveals two new factors and a stable tet-
rameric protein core. RNA 11, 1869–1883.
Tanner, N.K., and Linder, P. (2001). DExD/H box RNA helicases: from
generic motorstospecificdissociationfunctions.Mol. Cell8,251–262.
van Eeden, F.J., Palacios, I.M., Petronczki, M., Weston, M.J., and
St Johnston, D. (1991). Barentsz is essential for the posterior localiza-
tion of oskar mRNA and colocalizes with it to the posterior pole. J. Cell
Biol. 154, 511–523.
Wiegand, H.L., Lu, S., and Cullen, B.R. (2003). Exon junction com-
plexes mediate the enhancing effect of splicing on mRNA expression.
Proc. Natl. Acad. Sci. USA 100, 11327–11332.
Coordinates and structure factors have been deposited in the Protein
Data Bank under ID codes 2j0s (EJC at 2.2 A˚), 2j0q (EJC at 3.2 A˚), and
2j0u (eIF4AIII-Btz subcomplex).
Cell 126, 713–725, August 25, 2006 ª2006 Elsevier Inc. 725