A Divergent Sm Fold in EDC3 Proteins Mediates DCP1 Binding and P-Body Targeting

Abstract

Members of the (L)Sm (Sm and Sm-like) protein family are found across all kingdoms of life and play crucial roles in RNA metabolism. The P-body component EDC3 (enhancer of decapping 3) is a divergent member of this family that functions in mRNA decapping. EDC3 is composed of a N-terminal LSm domain, a central FDF domain, and a C-terminal YjeF-N domain. We show that this modular architecture enables EDC3 to interact with multiple components of the decapping machinery, including DCP1, DCP2, and Me31B. The LSm domain mediates DCP1 binding and P-body localization. We determined the three-dimensional structures of the LSm domains of Drosophila melanogaster and human EDC3 and show that the domain adopts a divergent Sm fold that lacks the characteristic N-terminal α-helix and has a disrupted β4-strand. This domain remains monomeric in solution and lacks several features that canonical (L)Sm domains require for binding RNA. The structures also revealed a conserved patch of surface residues that are required for the interaction with DCP1 but not for P-body localization. The conservation of surface and of critical structural residues indicates that LSm domains in EDC3 proteins adopt a similar fold that has separable novel functions that are absent in canonical (L)Sm proteins.

Proteins of the Sm and Sm-like family [conjointly referred to as (L)Sm proteins] are found in all domains of life and play important roles in RNA processing and decay (reviewed in references 21 and 46). They share the Sm fold, which comprises an N-terminal α-helix stacked on top of a strongly bent, five-stranded antiparallel β-sheet, which forms a barrel-like structure. The fold can be divided into two segments corresponding to the highly conserved Sm1 and Sm2 motifs, where Sm1 comprises β-strands 1 to 3 (β1-3) and Sm2 comprises β-strands 4 and 5 (β4-5). The two motifs are joined by a nonconserved linker (L4) of variable length (18, 20, 34) (see Fig. 4 and 5).

FIG. 4.

Structures of the HsEDC3 LSm and DmEDC3 LSm domains. (A) Crystal structure of HsEDC3 LSm. (B) NMR structure of DmEDC3 LSm. (C) Crystal structure of HsSmD3 (PDB ID, 1d3b-A) according to reference 20. β-strands belonging to the Sm1 motif are colored in red and, β-strands belonging to the Sm2 motif are colored in yellow. Shown in gray are the N-terminal α-helix (which is absent in EDC3), the extended loop L4 from canonical (L)Sm proteins, and the iβ4 insertion that is unique to EDC3. Left, plain views of the open β-barrel with the open side on the top. Center, front views. Right, edge views.

FIG. 5.

Structure-based alignment of the EDC3 LSm domain. (A and B) Superposition of the structures of HsEDC3 LSm (blue), DmEDC3 LSm (lime), and HsSmD3 (gray) represented as tubes. (A) Plain view. (B) Front view (stereo) illustrating the different orientations of loops L3 and L4. (C) Alignment of EDC3 LSm domains with HsSmD3. Secondary structure elements are colored as in Fig. 4. The Sm1 and Sm2 motifs are indicated together with the Sm1 and Sm2 signatures (x, any amino acid; h, hydrophobic; p, polar; +, charged). Residues from the hydrophobic core that define the Sm1 and Sm2 motifs are shaded blue; conserved glycines are shaded gray. Red letters indicate invariant amino acids in EDC3 and SmD3. Gray letters indicate amino acids absent in the presented structures. Functionally interesting surface residues are shaded in yellow, and those shown to mediate DCP1 binding are shaded in green. Relevant backbone elements that are structurally conserved in EDC3 are boxed in orange. Abbreviations and accession numbers: Hs, Homo sapiens (gi:18204641 for EDC3 and gi:74007795 for SmD3); Dm, Drosophila melanogaster (gi:24665977); Gg, Gallus gallus (gi:71896289); Xl, Xenopus laevis (gi:82180413); Dr, Danio rerio (gi:125853827); Ag, Anopheles gambiae (AGAP003131); Ce, Caenorhabditis elegans (gi:17508551); Cb, Caenorhabditis briggsae (gi:39595594).

The (L)Sm domains often oligomerize to form hexameric or heptameric rings that stably or transiently bind single-stranded RNA. The major contacts between the subunits of the ring are mediated by antiparallel interactions between the backbones of strand β4 of one subunit and strand β5 of the adjacent subunit. RNA binding is mediated mainly by residues in loops between strands β2 and β3 and between strands β4 and β5 (loops L3 and L5, respectively), which face the lumen of the ring (7, 20, 26, 33, 40, 41).

The eubacterial and archaeal genomes encode from one to three (L)Sm paralogs that form homohexameric or homoheptameric rings, while eukaryotes encode more than eighteen (L)Sm paralogs that assemble into heteroheptameric rings of different composition and function (reviewed in references 1, 2, 21, and 46).

Seven of the eukaryotic proteins (SmB, SmD1, SmD2, SmD3, SmE, SmF, and SmG) form a ring that stably associates with RNA polymerase II-transcribed uridine-rich small nuclear RNAs (i.e., U1, U2, U4, and U5), and functions in uridine-rich snRNP biogenesis and mRNA splicing (18, 21, 34, 46). In addition to the Sm ring, at least two LSm rings have been described (1, 2, 18, 21, 34, 46). One consists of LSm2 to -8, localizes to the nucleus, and participates in splicing, rRNA processing, and maturation of polymerase III-transcribed RNAs (1, 2, 18, 21, 34, 46). Another LSm ring consists of LSm1 to -7, interacts transiently with cytoplasmic mRNAs, and was implicated in mRNA decapping (3, 38, 39).

In addition to the above-mentioned (L)Sm proteins that consist of a single Sm domain, several LSm proteins which have long C-terminal extensions were recently identified; LSm12 to -16 belong to the latter group (1, 2). LSm12 orthologs are characterized by a C-terminal protein methyltransferase domain. LSm13 to -16 share a divergent form of the Sm domain and a central FDF domain of unknown function (which includes a conserved phenylalanine and aspartic acid motif [1, 2]). The FDF domains of LSm13 to -15 are preceded and followed by low-complexity linker regions rich in glycine and charged residues (1, 2). In contrast, in LSm16 (also called enhancer of decapping 3 [EDC3] and referred as such hereafter) the FDF domain is followed by a conserved C-terminal YjeF-N domain with a predicted Rossman fold as found in the N-terminal domain of the protein YjeF (1, 2) (Fig. 1A).

FIG. 1.

EDC3 interacts with DCP1, DCP2, and Me31B. (A) Domain architecture of EDC3. EDC3 orthologs contain three classified globular domains: an LSm domain, an FDF domain, and a YjeF-N-type Rossman fold domain. The LSm and FDF domains are connected by a low-complexity linker region. Numbers above the protein outline represent amino acid positions at fragment boundaries for the D. melanogaster protein. The protein domains sufficient for the localization to P bodies and the interaction with DCP1 (red), DCP2 (orange), and Me31B (blue), as well as for self-association (green), are indicated. (B) Epitope HA-tagged versions of MBP, DCP1, Me31B, or Tral were transiently expressed in S2 cells. Cell lysates were immunoprecipitated using a monoclonal anti-HA antibody. Inputs (10%) and immunoprecipitates (IP) (25%) were analyzed by Western blotting using a polyclonal anti-HA antibody. The presence of endogenous EDC3 in the immunoprecipitates was tested by Western blotting with an anti-EDC3 antibody. (C) Epitope HA-tagged versions of MBP or EDC3 were transiently coexpressed in S2 cells with GFP-DCP1 and/or GFP-DCP2 as indicated. Cell lysates were immunoprecipitated using a monoclonal anti-HA antibody. Inputs (10%) and immunoprecipitates (25%) were analyzed by Western blotting using polyclonal anti-HA and anti-GFP antibodies.

The biological functions of LSm12 and LSm13 remain unclear. LSm14 (also known as RAP55) and Drosophila melanogaster LSm15 (DmLSm15) (also known as Trailer Hitch [Tral]) have been implicated in translational regulation (37, 45, 47). Saccharomyces cerevisiae EDC3 was shown to stimulate the mRNA-decapping activity of the DCP2 enzyme (23).

In human cells EDC3 is a component of a multiprotein complex that includes the decapping enzyme DCP2 and several decapping activators. These are DCP1, the RNA helicase RCK/p54 (a human homolog of D. melanogaster Me31B), and Ge-1 (also called human enhancer of decapping large subunit [Hedls]) (16, 48). These proteins are conserved in eukaryotes, with the exception of Ge-1 (which is not present in S. cerevisiae). They colocalize in discrete cytoplasmic foci named mRNA-processing bodies (P-bodies) (12, 31) and have been implicated in microRNA-mediated gene silencing (13). The molecular basis of their interactions and the mechanism underlying decapping activation remain to be established.

To begin to elucidate the molecular function of EDC3, we determined the solution and crystal structures of the N-terminal domains of the D. melanogaster and human proteins, respectively. The structures reveal a divergent Sm fold that does not form oligomers and that lacks residues that in canonical (L)Sm domains mediate RNA binding. Instead, on the opposite side of the canonical RNA-binding surface, a patch of conserved surface residues mediates the interaction with DCP1. Furthermore, the DmEDC3 LSm domain is necessary and sufficient to localize the protein to P-bodies. This localization is independent of the interaction with DCP1. Thus, the divergent LSm domain of EDC3 has acquired novel functionalities that are conserved among members of this protein family.

MATERIALS AND METHODS

DNA constructs and transfection of S2 cells.

cDNAs encoding full-length DCP1, DCP2, EDC3, Me31B, Tral, and LSm1-7 proteins or protein domains were amplified with primers containing appropriate restriction sites, using a (dT)₁₅-primed S2 cDNA library as the template. The amplified cDNAs were cloned into a vector allowing the expression of green fluorescent protein (GFP) or λN-hemagglutinin (HA) peptide fusions (pAc5.1B-EGFP or pAc5.1B-λN-HA, respectively) as described before (14, 32). Transfections were performed in 6-well or 24-well dishes using Effectene transfection reagent (QIAGEN).

Fluorescence microscopy.

Three days after transfection, S2 cells were allowed 15 min to adhere to poly-d-lysine-coated coverslips, washed once in serum-free medium, and fixed with 4% paraformaldehyde in phosphate-buffered saline (PBS) for 15 min, followed by 5 min of incubation in methanol at −20°C. After fixation, cells were washed in PBS, permeabilized for 5 min with PBS containing 0.5% Triton X-100, and washed again with PBS. Cells were stained with affinity-purified anti-Tral antibodies (14) diluted 1:250 in PBS containing 1% bovine serum albumin. Tetramethyl rhodamine isocyanate-coupled goat secondary antibody (Molecular Probes) was used at a dilution of 1:250. Cells were mounted using Fluoromount-G (Southern Biotechnology Associates, Inc.). Images were acquired using a Leica TCS SP2 confocal microscope.

Coimmunoprecipitation assay and Western blotting.

Antibodies to DmEDC3 were raised in rats immunized with an EDC3 protein fragment encompassing residues 339 to 440 (FDF domain) expressed in Escherichia coli as a glutathione S-transferase fusion. For Western blots, the polyclonal antibodies were diluted 1:1,000. Bound primary antibodies were visualized with alkaline phosphatase-coupled secondary antibodies (Western-Star kit from Tropix).

For coimmunoprecipitations, S2 cells were collected 3 days after transfection, washed with PBS, and lysed for 15 min on ice in NET buffer (50 mM Tris [pH 7.4], 150 mM NaCl, 1 mM EDTA, and 0.1% NP-40 or Triton X-100) supplemented with protease inhibitors. Cells were spun at 16,000 × g for 15 min at 4°C. Anti-HA antibodies (Covance Research Products) were added to the supernatants (2.5 μl/2 × 10⁶ cells). After 1 h at 4°C, 25 μl of protein G-agarose (Roche) was added, and the mixtures were rotated for 1 hour at 4°C. Beads were washed three times with NET buffer and once with PBS. Bound proteins were eluted with sample buffer.

Proteins were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred to nitrocellulose membranes. Membranes were blocked in PBS containing 5% fat-free milk powder and 0.3% Tween 20. Western blotting was performed with polyclonal anti-HA antibodies (Sigma) and anti-EDC3 or anti-GFP antibodies, using the CDP-Star chemiluminescent immunoblot system (Western-Star kit from Tropix), as recommended by the manufacturer.

Purification of EDC3 LSm domains.

The LSm domains of DmEDC3 (UniProtKB entry Q9VVI2; M1 to G101) and human EDC3 (HsEDC3) (UniProtKB entry Q96F86; M1 to T82) were amplified from (dT)₁₅-primed S2 cell and HEK293 cell cDNA libraries, respectively, and cloned into the pETM60 vector (derived from pET24-d; Novagen). The proteins were expressed in the E. coli strain BL21(DE3) Rosetta II at 20°C overnight. To uniformly label the DmEDC3 LSm domain with ¹⁵N/¹³C or ¹⁵N, cells were grown in M9 minimal medium supplemented with ¹⁵NH₄Cl with or without ¹³C₆-glucose. Cell lysates were purified by affinity chromatography using Ni-nitrilotriacetic acid HiTrap chelating HP columns (GE Healthcare), followed by cleavage of the tag with TEV protease overnight. The proteins were purified to homogeneity by two subsequent gel filtrations using a HiLoad 26/60 Superdex G75 preparative-grade column (GE Healthcare). The purity of the resulting proteins, consisting of the cloned sequences plus additional two residues at the N terminus, was confirmed by sodium dodecyl sulfate-polyacrylamide gel electrophoresis. Multiangle laser light scattering was done online with gel filtration using a Superdex S75 10/30 chromatography column, and molecular weights were calculated using instruments (MiniDawnTreos/OptiLabRex) and software (Astra) from Wyatt Technology. Samples for nuclear magnetic resonance (NMR) at 0.8 to 1.2 mM were prepared in PBS (pH 7.1) containing 0.02% Na-azide and 0.5 mM dithiothreitol.

Crystallization.

The HsEDC3 LSm domain was concentrated to 8 mg/ml in 20 mM Na-HEPES (pH 7.5), 50 mM NaCl, and 1 mM dithiothreitol. Diffraction-quality crystals were grown using hanging-drop vapor diffusion. One microliter protein solution plus 1 μl reservoir were equilibrated at 19°C over 0.5 ml of 40 mM Na-propionate, 20 mM Na-cacodylate, 40 mM bis-Tris propane (pH 4.1), and 32% polyethylene glycol 1500. Rhombic single crystals grew to an average size of about 80 μm over 4 days.

Crystals for the native data set were flash frozen directly in liquid nitrogen, which caused ice rings affecting local data quality. Cryoprotected crystals, however, never diffracted beyond 1.6 Å. For the platinum derivative, crystals were soaked overnight in mother liquor supplemented with 2 mM K₂PtCl₄. They were flash frozen after adding 8% glycerol and washing the crystals in mother liquor containing 16% glycerol.

Structure determination and analysis.

Diffraction data for the platinum derivative were collected on beamline PXII of the Swiss Light Source, Villigen, Switzerland. Diffraction images were recorded at a single wavelength of 0.976 Å on the high-energy remote side of the platinum L(III) edge and processed by MOSFLM (25) and SCALA (15). The structure was solved by single anomalous dispersion (SAD). We used autoSHARP (43) to search for a single platinum site. Assignment of the correct hand and solvent flattening (optimum contrast at 48.4%) was done automatically. In the resulting map ARP/wARP as implemented in autoSHARP was able to trace 74% of the final model and built 33% of the side chains.

Diffraction data for the high-resolution native set (1.3 Å) were collected on the MPG beamline BW6 at the DESY, Hamburg, Germany, and processed by the HKL program suite (29). After rigid-body refinement (REFMAC [27]), automated model building with ARP/wARP (5) was continued in the high-resolution data, which resulted in 89% of the final protein model. The model was completed manually in COOT (11), including alternative conformations. Refinement was done in REFMAC and COOT iteratively, using anisotropic B factors for protein atoms.

Solution structure of the LSm domain of DmEDC3.

All spectra were recorded at 298 K on Bruker DMX600 and DMX750 spectrometers. Backbone sequential assignments were completed using standard triple-resonance experiments implemented using selective proton flipback techniques for fast pulsing (10). Aliphatic side chain assignments and aromatic assignments were completed as described before (6). Stereospecific assignments and the resulting χ¹ rotamer assignments were determined for 20 of 53 prochiral C^βΗ₂ protons and for the C^γΗ₃ groups of 6 of 10 valine residues. Assignments of χ¹ rotamers were also available for 9 of 10 isoleucine residues and 3 of 5 threonine residues. Assignments of χ² rotamers were made for 9 of 10 isoleucine and 3 of 8 leucine residues.

Distance data were derived from a set of four three-dimensional nuclear Overhauser effect spectroscopy (NOESY) spectra, including the heteronuclear edited NNH and CNH NOESY spectra (9) in addition to conventional ¹⁵N and ¹³C heteronuclear simple quantum coherence NOESY spectra and a 2D NOESY spectrum recorded on an unlabeled sample. Distance restraints, dihedral angle restraints (applied for the 61 high-confidence predictions found by the program TALOS [8]), 61 direct coupling constant restraints (included for the backbone φ angles [44]), and 27 hydrogen bond restraints were derived as detailed elsewhere (42).

Refinement was carried out by comparison of experimental and back-calculated ¹⁵N heteronuclear simple quantum coherence NOESY, CNH NOESY, and NNH NOESY spectra. This process resulted in adjustment of side chain rotamers for several residues. As a crystal structure was available by this stage, back calculation was simultaneously used to justify any differences between the solution and crystal structures.

Structures were calculated with XPLOR (NIH version 2.9.3) using standard protocols with modifications as described before (6). For the final set, 50 structures were calculated and 21 chosen on the basis of lowest restraint violations. An average structure was calculated and regularized to give a structure representative of the ensemble (used here for all NMR figures).

Accession numbers.

The coordinates and structure factors for HsEDC3 LSm were deposited in the Protein Data Bank (PDB) under ID code 2vc8. The coordinates for the DmEDC3 LSm structure ensemble have been deposited in the Protein Data Bank under ID code 2rm4 and the chemical shifts in the Biological Magnetic Resonance Data Bank (accession code 11009).

RESULTS

A modular domain architecture enables EDC3 to participate in multiple protein-protein interactions.

Members of the EDC3 protein family have a modular domain organization made up of an N-terminal LSm domain, a central FDF domain and a C-terminal YjeF-N domain (Fig. 1A). The LSm and FDF domains are connected by a sequence rich in glutamine, asparagine, glycine, and arginine residues, which is likely to be unstructured (Fig. 1A, linker) (1, 2).

HsEDC3 is a component of a multimeric protein complex consisting of the decapping enzyme DCP2 and several decapping activators. These include DCP1, DCP2, Ge-1, and RCK/p54 (the human homolog of D. melanogaster Me31B) (16). Additional decapping activators include the heptameric LSm1-7 complex (reviewed in reference 30). To identify the components of the decapping machinery that are likely to interact with EDC3 directly, we investigated whether, in lysates of D. melanogaster Schneider cells (S2 cells), endogenous EDC3 could be coimmunoprecipitated with transiently expressed HA-tagged versions of DCP1, Me31B, LSm1, LSm4, LSm7, or Tral (LSm15). Our results show that endogenous EDC3 coimmunoprecipitated with HA-DCP1 and HA-Me31B (Fig. 1B, lanes 6 and 7) but not with the negative control (HA-tagged maltose-binding protein [HA-MBP]) (Fig. 1B, lane 5) or the additional proteins tested, such as HA-Tral (Fig. 1B and data not shown). Because HA-DCP2 was expressed at very low levels in S2 cells it could not be tested in this assay; however, an interaction between DCP2 and EDC3, as well as between DCP2, DCP1, and EDC3 was observed in independent experiments in which GFP fusions of DCP2 and/or DCP1 were coexpressed with HA-EDC3 (Fig. 1C, lanes 10 and 12).

To define which domains of EDC3 mediate the observed interactions, we coexpressed HA-tagged EDC3 protein fragments with GFP-tagged DCP1, DCP2 or, Me31B. We found that GFP-DCP1 interacted with full-length HA-EDC3 as well as with EDC3 fragments including the LSm domain but that it failed to interact with any of the protein fragments lacking the LSm domain (Fig. 2A, lanes 14 to 16). In fact, the EDC3 LSm domain alone was sufficient for this interaction (Fig. 2A, lane 11).

FIG. 2.

EDC3 interacts with DCP1, DCP2, and Me31B through specific domains. HA-tagged EDC3 or the indicated EDC3 protein fragments were cotransfected in S2 cells with GFP fusions of DCP1 (A), DCP2 (B), or Me31B (C) as indicated. Cell lysates were immunoprecipitated using a monoclonal anti-HA antibody. HA-tagged MBP served as a negative control. Inputs (10%) and immunoprecipitates (IP) (25%) were analyzed as described for Fig. 1C.

In contrast, GFP-DCP2 coimmunoprecipitated with full-length EDC3 and with fragments lacking the LSm domain, indicating that the LSm domain is dispensable for this interaction, although it may have a stimulatory effect (Fig. 2B, lane 12 versus lanes 14 and 15). Instead, a fragment encompassing the low-complexity linker region was sufficient for DCP2 interaction (Fig. 2B, lane 14). Finally, GFP-Me31B coimmunoprecipitated with full-length EDC3 and with fragments containing the FDF and YjeF-N domains (Fig. 2C, lane 16). The FDF domain (amino acids 331 to 440) was sufficient for this interaction, but the binding efficiency of this domain alone was slightly reduced (data not shown). The interactions of EDC3 with DCP1 or Me31B were insensitive to RNase A treatment (data not shown), suggesting that they are not mediated by RNA. Furthermore, these interactions were observed with recombinant proteins coexpressed in E. coli, although the full-length proteins are significantly degraded in bacteria (data not shown). Together these results indicate that EDC3 has a modular domain organization, with specific functionalities for each domain.

The LSm domain of EDC3 is necessary and sufficient for P-body localization.

Previously, we showed that DmEDC3 localizes to P-bodies (12, 14). However, the domains of the protein required for P-body localization were not defined. Therefore, we investigated the role of the protein domains in P-body localization.

N-terminal fragments of EDC3 comprising the LSm domain formed prominent cytoplasmic foci, similar to those observed with the full-length protein (Fig. 3 and data not shown) (12, 14). These foci correspond to endogenous P-bodies as judged by the staining with antibodies to Tral (i.e., LSm15, which is a P-body marker in D. melanogaster) (Fig. 3) (12, 14). In contrast, C-terminal fragments lacking the LSm domain spread throughout the cytoplasm and did not accumulate in P-bodies (Fig. 3). Because these fragments interact with DCP2 and Me31B (Fig. 2B and C), which both localize to P-bodies (12, 14), we concluded that these interactions are not sufficient for P-body targeting.

FIG. 3.

The LSm domain of EDC3 is necessary and sufficient for P-body localization. Confocal fluorescent micrographs of fixed S2 cells expressing GFP fusions of full-length EDC3 or the indicated protein fragments are shown. Cells were stained with affinity-purified anti-Tral antibodies. The merged images show the Tral signal in red and the GFP signal in green. Bar, 5 μm.

In summary, our results indicate that the LSm domain of EDC3 is necessary and sufficient to localize EDC3 to P-bodies. Thus, this domain is an independent folding unit with two distinct functionalities: DCP1 binding and P-body targeting. To delineate the molecular bases for these functions and to define domain boundaries, we determined the solution structure of the LSm domain of DmEDC3 by NMR as well as the crystal structure of the human homolog (Fig. 4).

Structure of the LSm domain of EDC3.

The structure of the HsEDC3 LSm domain (residues M1 to T82) was solved by SAD and was refined at 1.3-Å resolution with an R_free of 23.0% (Fig. 4A). The crystals contain one molecule per asymmetric unit. The model discussed here includes residues T3 to Q74 and 54 water molecules. The structure has good stereochemistry, with more than 96% of the residues lying in the most favored regions of the Ramachandran plot (phasing and refinement statistics are given in Tables 1 and 2). The C-terminal residues (H75 to T82) are not visible in the electron density and are likely to be disordered.

TABLE 1.

Data collection and phasing statistics

Parameter (unit)	Valuea for:
	Native data	K2PtCl4 derivative (SAD)
Data collection
Wavelength (Å)	1.05	0.97620
Resolution range (Å)	18.7-1.31	30-1.80
Space group	P3221	P3221
Cell dimensions, a/b/c (Å)	37.3, 37.3, 79.2	36.2, 36.2, 79.8
Rmerge (%)	5.3 (27.6)	4.4 (31.7)
Completeness (%)	98.6 (99.8)	99.6 (100.0)
Completeness (anomalous) (%)		99.7 (100.0)
Mean I/σ(I)	15.5 (4.7)	32.0 (6.6)
No. of unique reflections	15783 (625)	6009 (866)
Multiplicity	12.9 (3.4)	11.2 (11.5)
Multiplicity (anomalous)		6.2 (6.1)
Phasing
Rcullis		0.732
Phasing power		1.2
Mean figure of merit		0.18

^aValues in parentheses correspond to those in the outer resolution shell (1.33 to 1.31 Å and 1.9 to 1.8 Å, respectively).

TABLE 2.

Refinement statistics

Parameter (unit)	Value
Rcryst (%)	19.6
Rfree (%)	23.0
No. of:
Molecules in asymmetric unit	1
Protein atoms (T3-Q74)	1,250
Water molecules	54
Avg B factor (anisotropic) (Å2)	18.3
Ramachandran plot
Most favored regions (%)	96.6
Allowed regions (%)	3.4
RMSD from ideal geometry
Bond lengths (Å)	0.017
Bond angle (°)	1.716

The solution structure of the DmEDC3 LSm domain (residues M1 to G101) comprises a structured N-terminal region (residues P3 to A71) with domain boundaries similar to those of the human homolog (Fig. 4B). The remaining C-terminal extension (30 residues) is unstructured. The ensemble of 21 lowest-energy NMR structures is well defined (see Fig. S1 in the supplemental material): the root mean square deviation (RMSD) calculated over 55 structured residues (G10 to L48 and E54 to E69) is 0.19 Å for backbone atoms and 0.81 Å for all nonhydrogen atoms (Tables 3 and 4). The restraint violations are also very low, with the final set having on average 2.6 violations of distance restraints greater than 0.1 Å per structure and no dihedral restraint violation greater than 1°.

TABLE 3.

Structural statistics

Parameter (unit)a	Value (mean ± SD) for structureb:
	SA	<SA>r
RMSD from distance restraints (Å)
All (358)	0.021 ± 0.001	0.021
Intraresidue (73)	0.014 ± 0.004	0.012
Interresidue sequential (123)	0.020 ± 0.001	0.021
Medium range (29)	0.015 ± 0.001	0.013
Long range (106)	0.028 ± 0.002	0.028
H bond (27)	0.000 ± 0.000	0.000
RMSD from dihedral restraints (185)	0.13 ± 0.01	0.12
RMSD from J coupling restraints (Hz) (60)	0.82 ± 0.02	0.82
H bond restraints, avg (Å/°)c (27)	2.2 ± 0.1, 13.4 ± 5.8	2.2 ± 0.1, 13.6 ± 6.2
H bond restraints, minimum-maximum (Å/°)	1.94-2.47, 4.5-27.6	1.92-2.42, 2.6-27.2
Deviations from ideal covalent geometry
Bonds (Å [10−3])	5.81 ± 0.03	5.79
Angles (°)	0.644 ± 0.003	0.641
Impropers (°)	1.48 ± 0.03	1.48
Ramachandran map regionsd (%)	93.7/6.1/0.1/0.1	90.6/9.4/0.0/0.0

^aNumbers in parentheses indicate the number of restraints of each type.

^bSA, the set of 21 final simulated annealing structures; <SA>, the mean structure calculated by averaging the coordinates of SA structures after fitting over secondary structure elements; <SA>_r, the structure obtained by regularizing the mean structure under experimental restraints.

^cHydrogen bonds were restrained by treating them as pseudocovalent bonds (see Materials and Methods). The average and the minimum and maximum for distances and acceptor antecedent angles are stated for restrained hydrogen bonds.

^dDetermined using the program PROCHECK (24). Percentages are for residues in allowed/additionally allowed/generously allowed/disallowed regions of the Ramachandran map.

TABLE 4.

Atomic RMSD

Structure or comparisona	Atomic RMSDb (mean ± SD), Å
	SA vs <SA>		SA vs <SA>r
	Backbone	All	Backbone	All
Secondary structurec	0.19 ± 0.05	0.81 ± 0.07	0.23 ± 0.05	1.06 ± 0.09
<SA> vs <SA>rd	0.14	0.75

^aSA, the set of 21 final simulated annealing structures; <SA>, the mean structure calculated by averaging the coordinates of SA structures after fitting over secondary structure elements; <SA>_r, the structure obtained by regularizing the mean structure under experimental restraints.

^bBased on heavy atom superimpositions.

^cDefined as residues G12 to L50 and E56 to E71.

^dRMSD for superimposition over ordered residues.

The structured part of HsEDC3 LSm and DmEDC3 LSm corresponds to the core of canonical (L)Sm proteins as represented by HsSmD3 (Fig. 4C) (20) and as described in the SCOP database (http://scop.mrc-lmb.cam.ac.uk/scop/ [28]). The fold consists of a five-stranded open β-barrel (β1 to β5), where β-strands β2 and β3 are strongly bent by β-bulges. The edges of the barrel are formed by β-strands β4 and β5, which are connected by a structurally conserved helical turn (loop L5) that completes the barrel on the open side. The N-terminal α-helix, a hallmark of the (L)Sm protein family (Fig. 4C), is absent as such from both structures, as is a predicted sixth β-strand on the C-terminal side (predicted in reference 2). The two structures confirm and confine the EDC3 LSm domain as an independent folding unit and suggest that residues beyond HsP68 and DmA71 are part of the unstructured region of EDC3 which links the LSm domain to the predicted FDF domain.

The identity between HsEDC3 LSm and DmEDC3 LSm is 39% over the alignable region. This results in a high degree of structural similarity between the two proteins (Fig. 4 and 5), as reflected by the RMSD of 1.3 Å over 54 alignable C_α positions. The main differences between the two structures are found in loop L3 (linking β-strands β2 and β3) and in the insertion (iβ4) that interrupts β-strand β4 into strands β4a and β4b (Fig. 4A and B). Loop L3 is two amino acids shorter in DmEDC3 LSm, reducing it to a standard type I β-turn, while the insertion iβ4 is two amino acids longer in DmEDC3 LSm and rather flexible. Internal motions on the time scale of the chemical shifts result in a complete loss of information over residues DmR51 to DmN54. Finally, the NMR data support the existence of a short, unstable 3₁₀-helix at the N terminus of DmEDC3 LSm (residues DmQ6 to DmW8) that is not present in the X-ray structure.

The LSm domain of EDC3 adopts a divergent Sm fold.

A search of the Protein Data Bank using the DALI server (19) indicates that the closest structural neighbors to HsEDC3 LSm are the human Sm proteins SmD3 and SmB, together with the archaeal LSm protein SmAP1 from Pyrobaculum aerophilum (DALI Z scores of above 8.0 for PDB entries 1d3b-A, 1d3b-B, and 1i8f-A). The eubacterial (L)Sm protein Hfq from Staphylococcus aureus also scores with high significance (DALI Z score of 6.2 for PDB-entry 1kq1-A). This demonstrates that, from a structural point of view, the N-terminal domain of EDC3 clearly belongs to the Sm superfamily of proteins as described in SCOP.

A superposition of HsSmD3 with HsEDC3 LSm (19% identity) and DmEDC3 LSm (17% identity) yields low RMSD values of 2.0 and 1.8 Å, respectively, calculated from the positions of 54 alignable C_α atoms (Fig. 5A and B). A structure-based alignment (Fig. 5C), which includes HsSmD3 and EDC3 LSm domains from various species, illustrates that EDC3 contains the Sm1 (β1 to β3) and Sm2 (β4 and β5) motifs (18, 20, 34); these contribute a characteristic set of hydrophobic side chains that pack together to form the core of the protein.

The best-scoring non-Sm-like fold is the SH3-like barrel in the Tudor family (classification according to SCOP) as found in 53BP1 (DALI Z score of 4.9 for PDB-entry 2ig0-A). The Tudor domain lacks the N-terminal α-helix, similar to EDC3 LSm and in contrast to the canonical (L)Sm and eubacterial Hfq proteins.

Apart from the absence of the characteristic N-terminal α-helix, there are additional structural differences between EDC3 LSm domains and related (L)Sm structures. The most notable difference is an unusual insert (iβ4) of variable size that interrupts β-strand β4 and which is a unique feature of EDC3 (Fig. 4 and 5). Loop L4 of EDC3 LSm consists of a tight and highly conserved β-turn that is restricted to EDC3. Finally, β-strand β5 is unusually long on its C-terminal end (Fig. 5A to C, strand β5C), a feature that is likely to be functionally important, because this extension contributes to a conserved patch of surface residues involved in DCP1 binding (see below). These structural differences establish the EDC3 LSm domain as a rather divergent member of the Sm superfamily.

The LSm domain of EDC3 does not form multimeric (L)Sm rings.

Canonical (L)Sm and Hfq proteins multimerize in a head-to-tail fashion via an antiparallel arrangement of the β-strands β4 and β5. This leads to six- or seven-membered homo- or heteromeric rings with a continuous inner β-sheet (7, 20-22, 26, 33, 40, 41, 46). Similar multimerization properties were therefore predicted for the LSm domain of EDC3 (2). However, gel filtration chromatography and static light-scattering experiments indicated that HsEDC3 LSm is monomeric in solution (data not shown). The NMR measurements with DmEDC3 LSm showed no significant multimerization at concentrations of up to 1.2 mM, as reflected by a diffusion coefficient of 1.07 (± 0.07) × 10⁻¹⁰ m²/s. Considering that homomeric (L)Sm and Hfq rings are usually considerably stable (7, 41), this argues against the existence of such rings for EDC3 LSm.

In the crystal packing of HsEDC3 LSm, however, strands β4 and β5 from neighboring asymmetric units interact in a parallel fashion with a buried molecular surface of ∼700 Ų. This leads to a continuous β-sheet meandering through the crystal, with an alternating head-to-tail arrangement of the monomeric domains (see Fig. S2 in the supplemental material). This packing indicates that the edges of β-strands β4 and β5 are available for parallel or antiparallel main-chain interactions with β-strands of potential interacting partners, forming a continuous sheet.

We therefore tested in vivo whether EDC3 forms heteromeric associations with members of the LSm1-7 complex or with Tral (LSm15), which all colocalize with EDC3 to P-bodies (12, 14). As mentioned above, endogenous EDC3 did not detectably interact with HA-tagged LSm1, LSm4, LSm7, or Tral (LSm15) (Fig. 1B and data not shown). These results indicate that EDC3 is unlikely to represent an alternative monomeric unit in the assembly of a specialized LSm1-7-like ring, as proposed before (2). In agreement with this, HA-EDC3 did not interact with GFP-LSm1 (Fig. 6A, lane 8) under the conditions in which HA-LSm7 or HA-LSm4 coprecipitated GFP-LSm1 (Fig. 6A, lanes 6 and 7). Because LSm4 and LSm7 do not directly contact LSm1 within the LSm1-7 ring, the coimmunoprecipitation of GFP-LSm1 with HA-LSm4 or HA-LSm7 indicates that these overexpressed proteins assemble into LSm1-7 rings, together with the additional endogenous proteins. Consequently, the failure of EDC3 to interact with LSm1, LSm4, or LSm7 is unlikely to be caused by its overexpression.

FIG. 6.

EDC3 LSm is not incorporated into the LSm1-7 ring. (A) HA-tagged versions of MBP, LSm7, LSm4, and EDC3 were transiently coexpressed in S2 cells with GFP-LSm1. Cell lysates were immunoprecipitated using a monoclonal anti-HA antibody. (B and C) HA-tagged versions of MBP, EDC3, and the indicated EDC3 protein fragments were transiently coexpressed in S2 cells with full-length GFP-EDC3 (B) or an EDC3 protein fragment encompassing residues 331 to 680 (C). Cell lysates were immunoprecipitated using a monoclonal anti-HA antibody. Inputs (10%) and immunoprecipitates (IP) (25%) were analyzed as described for Fig. 1C.

We also tested whether EDC3 could homooligomerize in vivo. We observed that HA-EDC3 coimmunoprecipitated GFP-EDC3 (Fig. 6B, lane 10). However, this interaction is mediated not by the LSm domain but by the C-terminal domain comprising the FDF and YjeF-N domains (Fig. 6B and C lanes 15 and 16). Further analyses showed that the YjeF-N domain alone was sufficient for self-association and that this interaction is insensitive to RNase A treatment (data not shown). The observation that the LSm domain of EDC3 is not required for homooligomerization in vivo agrees with the structural and biochemical data showing that this domain does not form oligomers in vitro.

The LSm domain of EDC3 lacks (L)Sm-like RNA-binding properties.

RNA binding is another hallmark of canonical ring-forming (L)Sm and Hfq proteins. The crystal structures of the eubacterial Hfq hexamer from S. aureus (33) and the archaeal SmAP1 heptamer from Pyrococcus abyssi (40) show how uridine-rich RNA binds around the cationic pore on the inner surface side of the ring. The binding is mediated by conserved amino acids in loops L3 and L5 that face the pore. A model for the human heteroheptameric Sm ring (20, 40) shows a similar conservation pattern and pore charge.

In EDC3 LSm the sequences of loops L3 and L5 are not conserved, apart from the negatively charged aspartate HsD58/DmD61 in loop L5 and the following isoleucine HsI59/DmI62 (Fig. 5C). In DmEDC3 LSm, loop L3 even has a deletion of two amino acids, and the backbone conformations of that loop vary widely between HsEDC3 LSm, DmEDC3 LSm, and HsSmD3 (Fig. 5A and B). Furthermore, in size exclusion chromatography, HsEDC3 LSm did not detectably interact with an RNA oligonucleotide consisting of eight uridines (data not shown). Together with the apparent failure to form (L)Sm-like rings, this suggests that EDC3 has no (L)Sm-like or Hfq-like RNA-binding properties.

A patch of conserved surface residues on the EDC3 LSm domain mediates the interaction with DCP1.

The divergence of EDC3 LSm from the canonical (L)Sm-proteins is also reflected by the level of conservation of individual surface residues (Fig. 7A). Overall, EDC3 LSm shows much more surface variation than SmD3 or other canonical Sm proteins, but some surface residues are quite well conserved in EDC3 exclusively. In the Sm1 motif, at the bend of β-strand β2, the characteristic hxG signature (x, any amino acid; h, hydrophobic) (2) is strongly conserved as an (F/Y)QG sequence (Fig. 5C). In contrast, the +Gpph signature from the Sm2 motif located in loop L5 of the canonical (L)Sm proteins (+, charged; p, polar; h, hydrophobic) (2) is not conserved as such. It is replaced by variable sequences in EDC3 that do not correspond to a distinct signature, with the exception of the invariable aspartate and isoleucine (HsD58 to I59/DmD61 to I62) in positions 4 and 5 (Fig. 5C).

FIG. 7.

Localization of functionally relevant residues. (A) Surface representation (plain view) of the structures colored by sequence conservation, comparing eight species (Fig. 5). Color ramp by identity: orange (100%) to white (50% or less). (B and C) Tube representation with C_α carbons as spheres in plain view (B) and edge view (C). Structurally relevant backbone elements (the L4 β-turn and the β5 C-terminal extension in EDC3 versus loops L3 and L5 in canonical (L)Sm/Hfq proteins (represented by HsSmD3) are colored in orange. Functionally interesting surface residues are drawn as sticks with carbons in yellow, oxygens in red, and nitrogens in blue. Side chains (green carbons) known to mediate the interaction of EDC3 with DCP1 are located on a surface on the opposite side from the RNA-binding residues on loops L3 and L5 in canonical (L)Sm/Hfq proteins.

Strikingly, several invariant or strongly conserved side chains are located in proximity to each other on the β-sheet surface on the side of EDC3 LSm opposite to loops L3 and L5 [i.e., which form the RNA-binding site in canonical (L)Sm]. They form a highly conserved patch that includes the β-turn of loop L4, suggesting a conserved function of that region that is not found in canonical (L)Sm proteins (Fig. 7A to C). These surface residues comprise the invariant HsN43/DmN44 from the L4 β-turn, the hydrophobic HsF41/DmF42 on strand β3, the invariant HsQ22/DmQ25 from the Sm1 signature on strand β2, the hydrophobic HsV20/DmV23 on strand β2, the invariant HsS11/DmS14 on strand β1, and the charged HsK63/DmD66 and the hydrophobic HsL65/DmI68 on the extended C-terminal end of strand β5 (Fig. 5B and C and 7B and C).

To test whether residues in this conserved patch are involved in EDC3 LSm functions in vivo, we mutated them to alanines and examined their effect on DCP1 binding and P-body targeting (Fig. 8). Specifically, alanine substitutions of DmQ25 (Q25A), and of DmF42 and DmN44 (F42A,N44A), drastically reduced binding to GFP-DCP1 (Fig. 8A). The mutant proteins were expressed at a level comparable to that of the wild-type protein. Moreover the mutant proteins localized to P-bodies (Fig. 8B). These observations suggest that the mutations do not significantly affect the general folding or stability of the EDC3 protein. Thus, the conserved patch on the LSm domains of EDC3 is required for the interaction with DCP1 but is not necessary for P-body targeting.

FIG. 8.

Mutations of conserved surface residues abolish DCP1 interaction but not P-body localization. (A) HA-tagged MBP, EDC3, or the indicated EDC3 mutants were cotransfected in S2 cells with GFP-DCP1. Cell lysates were immunoprecipitated using a monoclonal anti-HA antibody and analyzed by Western blotting as described for Fig. 1C. (B) Confocal fluorescent micrographs of fixed S2 cells expressing GFP fusions of full-length EDC3 (wt) or EDC3 mutants. Cells were stained with affinity-purified anti-Tral antibodies. The merged images show the Tral signal in red and the GFP signal in green. Bar, 5 μm.

DISCUSSION

The LSm domain of EDC3 mediates DCP1 binding and P-body targeting.

From a structural point of view, the N-terminal domain of EDC3 belongs to the (L)Sm family of proteins. Indeed, this domain shows the classical Sm topology. The closest structural relatives of the HsEDC3 LSm domain are the human Sm proteins SmD3 and SmB, followed by the archeal LSm proteins. In agreement with this, the Sm1 and Sm2 motifs are present in the EDC3 proteins from various species, although the general conservation of the EDC3 proteins is much lower than for the classical (L)Sm-like proteins.

Canonical (L)Sm proteins multimerize via an antiparallel arrangement of the β-strands β4 and β5 (7, 20, 21, 26, 33, 40, 41, 46). Similar multimerization properties were predicted for EDC3 (1, 2). However, our results indicate that in solution the human and D. melanogaster EDC3 LSm domains are monomeric. Moreover, we did not observe heteromeric associations in vivo with LSm1-7 or Tral (LSm15). Therefore we have no evidence that the EDC3 LSm domain forms Sm-like rings. Nevertheless, the crystal packing indicates that the edges of β-strands β4 and β5 are available for parallel or antiparallel main-chain interactions with β-strands of potential interacting partners, an association that would not require extensive sequence conservation.

RNA binding is another hallmark of classical ring-forming (L)Sm proteins (reviewed in references 21 and 46). The respective RNA-binding residues are located in loops L3 and L5 and line the pore of the ring (33, 40, 41). In EDC3 these residues are not conserved, and loop L3 has distinct conformations even between HsEDC3 LSm and DmEDC3 LSm. Furthermore, in gel filtration chromatography HsEDC3 LSm failed to interact with an RNA oligomer of eight uridines. Together with the lack of evidence supporting ring formation, this indicates that EDC3 has no (L)Sm-like RNA-binding properties.

Instead, we propose that DmEDC3 LSm is a protein-protein interaction domain that is necessary and sufficient both for binding to DCP1 and for targeting EDC3 to P-bodies. These are two separable functions, since if the surface residues that form the conserved patch are replaced with alanine, then DCP1 binding is abolished, although the protein still localizes to P bodies. The conservation of the mutated residues indicates that the interaction with DCP1 is a conserved function of LSm domains in metazoan EDC3 proteins. Although (L)Sm-like domains have not been described as monomeric interacting platforms before, it is possible that several novel LSm domains (1, 2) located near the N termini of larger multidomain proteins serve such a function.

Apart from DCP1, none of the tested P-body components (DCP2, Me31B, LSm1-7, and Tral) coimmunoprecipitated with the DmEDC3 LSm domain, and the interaction with DCP1 was dispensable for localization to P-bodies. This suggests that P-body targeting of EDC3 is mediated by a novel, as-yet-unidentified factor that might interact with a second site on EDC3 LSm (e.g., the edges of β-strand β4 or β5). This factor could be an integral P-body component, an mRNP component, or a specific P-body targeting factor.

An important observation from our studies is that an interaction with P-body components is not sufficient for proteins to accumulate in P-bodies. Indeed, EDC3 protein fragments lacking the LSm domain do not localize to P bodies, even though these fragments interact with DCP2 and/or Me31B. Conversely, EDC3 is not required for P-body integrity in S2 cells (12, 14), although it can bridge multiple P-body component interactions. This suggests that other features influence P-body targeting and assembly or that P-body formation occurs through redundant pathways.

Role of EDC3 in mRNA decapping.

A major mRNA decay pathway in eukaryotes is initiated when the poly(A) tail is degraded by deadenylases (30). Following deadenylation, the cap structure is removed by the decapping enzyme DCP2, rendering the mRNA susceptible to exonucleolytic degradation by the cytoplasmic 5′-to-3′ exonuclease XRN1 (30).

DCP2 activity is enhanced by several proteins generically termed decapping coactivators, though they may activate decapping by different mechanisms (30). In S. cerevisiae, several decapping activators have been identified: DCP1, which forms a complex with DCP2 and is required for decapping in vivo; EDC1 to -3; the heptameric LSm1-7 complex; the DExH/D box RNA helicase 1 (Dhh1, RCK/p54 in mammals, or Me31B in D. melanogaster); and Pat1, a protein of unknown function that interacts with the LSm1-7 complex, Dhh1, and XRN1 (30). Similarly, in human cells, DCP1 and DCP2 are part of a multimeric protein complex that includes RCK/p54, EDC3, and Ge-1 (also known as RCD-8 or Hedls), a protein that is absent in S. cerevisiae (16, 48). All of these proteins colocalize in P bodies (12-14, 16, 31, 48). Although structural information is available on the conserved domains of DCP1, DCP2, and Dhh1p (4, 35, 36), the molecular basis of their interaction and the mechanism of decapping activation remain unknown.

In this paper, we show that the LSm domain of DmEDC3 interacts with DCP1, the linker region with DCP2, and the FDF domain with Me31B and that the YjeF-N domain mediates self-association. This suggests that the role of EDC3 in decapping may be to provide a binding platform for components of the decapping complex. Although we do not know whether all these interactions are direct or take place simultaneously, the modular domain organization of EDC3 is well suited for a role as a molecular platform. The resulting proximity of the decapping factors on EDC3 could enhance the rate of decapping, providing one explanation for the observed activity of EDC3.

This role of EDC3 as a molecular platform is likely to be conserved in animals and fungi, because databases of protein-protein interactions show that S. cerevisiae EDC3 interacts with DCP1, DCP2, Dhh1, Pat1, LSm1-7, and XRN1 (17; also see reference 23 and references therein). It would therefore be of great interest to determine the molecular bases for the specificity of the interaction between EDC3 domains and components of the general mRNA decapping machinery and how these interactions affect DCP2 activity.