Nuclear Magnetic Resonance Structure of the N-Terminal Domain of Nonstructural Protein 3 from the Severe Acute Respiratory Syndrome Coronavirus

Full-length nsp3a (consisting of residues 1 to 183) and a construct devoid of residues 113 to 183, nsp3a(1-112), were cloned into the expression vector pMH1F (His₆ tag; pBAD derivative) and expressed in DL41 Escherichia coli cells with induction at 14°C in 2× YT (yeast extract and tryptone) medium. Each of the two constructs was shown, by one-dimensional (1-D) ¹H NMR, to form a folded globular domain (data not shown). To facilitate expression of samples suitable for NMR structure determination, both constructs were subcloned into pET-25b (Novagen). These plasmids were used to transform E. coli strain BL21-CodonPlus (DE3)-RIL (Stratagene). The expression of uniformly ¹³C,¹⁵N-labeled nsp3a(1-112) was carried out by growing freshly transformed cells in M9 minimal medium containing 1 g/liter ¹⁵NH₄Cl and 4 g/liter d-[¹³C₆]glucose as the sole nitrogen and carbon sources. Cell cultures were grown at 37°C with vigorous shaking to an optical density at 600 nm of 0.8 to 0.9. The temperature was then lowered to 18°C, and after induction with 1 mM isopropyl-β-d-thiogalactopyranoside, the cell cultures were grown for 18 h. The cells were harvested by centrifugation, resuspended in extraction buffer (50 mM sodium phosphate at pH 6.5, 150 mM NaCl, 0.1% Triton X-100, and Complete protease inhibitor tablets [Roche]), and lysed by sonication. The cell debris was removed by centrifugation (20,000 × g for 20 min). For the first purification step, the soluble protein was loaded onto an anion exchange column (HiTrap Q FF; Amersham) equilibrated with 50 mM sodium phosphate buffer at pH 6.5 containing 150 mM NaCl. The proteins were eluted with a 150 to 1,000 mM NaCl gradient. Fractions containing nsp3a(1-112) were pooled and concentrated to a volume of 10 ml using centrifugal ultrafiltration devices (Millipore). Subsequently, the sample was loaded onto a size exclusion column (Superdex 75; Amersham) equilibrated with 50 mM sodium phosphate buffer (pH 6.5) containing 150 mM NaCl and eluted with the same buffer. The fractions containing nsp3a(1-112) were again pooled and concentrated to a final volume of 550 μl, for a final protein concentration of 1.8 mM.

nsp3a prepared as described in the preceding section copurifies with nucleic acids, as was readily observed in the 1-D ¹H NMR spectrum (see Fig. 8a). Nucleic acid-free samples were obtained by the following modification of the purification procedure. After the anion-exchange chromatography, the sample was kept at 25°C for 18 h. The protein solution was subsequently loaded onto a size exclusion column (Superdex 75; Amersham) equilibrated with 50 mM sodium phosphate buffer (pH 6.5) containing 150 mM NaCl and eluted with the same buffer. Under these conditions, the protein and the nucleic acid eluted separately. The fractions containing nucleic acid-free nsp3a(1-112) were again pooled and concentrated to a final volume of 550 μl, for a final protein concentration of 1 to 2 mM. The 1-D ¹H NMR spectrum of the sample used for the NMR structure determination (see Fig. 8b) confirms the absence of nucleic acids.

NMR measurements were performed at 298 K with Bruker Avance 600, DRX 700, and Avance 800 spectrometers (Bruker BioSpin, Billerica, MA), equipped with TXI-HCN-z- or TXI-HCN-xyz gradient probe heads. Proton chemical shifts were referenced to internal 3-(trimethylsilyl)-1-propanesulfonic acid sodium salt (DSS). The ¹³C and ¹⁵N chemical shifts were referenced indirectly to DSS, using the absolute frequency ratios (42). The following NMR spectra were used to obtain sequence-specific backbone and side chain resonance assignments: 2-D [¹⁵N,¹H]-heteronuclear single-quantum coherence (HSQC), 2-D [¹³C,¹H]-HSQC, 3-D HNCA, 3-D HNCACB, 3-D CBCA(CO)NH, 3-D HNCO, 3-D HC(C)H-total correlation spectroscopy, 3-D ¹⁵N-resolved [¹H,¹H]-total correlation spectroscopy, and 2-D [¹H,¹H]-nuclear Overhauser effect spectroscopy (NOESY).

Steady-state ¹⁵N{¹H}-NOEs were measured using transverse relaxation optimized spectroscopy-based experiments (32, 46) on a Bruker Avance 600 spectrometer with a saturation period of 3.0 s and a total interscan delay of 5.0 s. Diffusion experiments were recorded on a Bruker DRX700 spectrometer using a longitudinal eddy current delay pulse scheme (1), with a diffusion time of 50 ms and sine-shaped gradients of 4.5 ms. The data were processed with TopSpin software (Bruker BioSpin, Billerica, MA).

The interaction of nsp3a(1-112) with single-stranded RNA (ssRNA) was evaluated by comparison of the 2-D [¹⁵N,¹H]-HSQC spectra of nsp3a(1-112) recorded at four nsp3a(1-112):ssRNA2 ratios, i.e., 16:1, 8:1, 4:1, and 2:1. As controls, 2-D [¹⁵N,¹H]-HSQC spectra were obtained after addition of eight units of uridine (Octa-U) and Octa-A in fourfold excess with respect to the protein concentration, using otherwise identical conditions. The weighted average of the ¹H and ¹⁵N chemical shift differences, Δδ_av, was calculated as follows: Δδ_av = {0.5[Δδ(¹H^N)² + (0.2Δδ(¹⁵N))²]}^1/2 (28).

The structure calculation was based on a 3-D ¹⁵N-resolved [¹H,¹H]-NOESY spectrum and on two 3-D ¹³C-resolved [¹H,¹H]-NOESY spectra recorded with the carrier frequency in the aliphatic and the aromatic regions, respectively. All three data sets were recorded with mixing times of 60 ms. In the input for the stand-alone version of the software package ATNOS/CANDID (9, 10), these NOE data were supplemented with the amino acid sequence and the chemical shift lists from the independently obtained sequence-specific resonance assignment (36). Seven cycles of automated NOESY peak picking and NOE cross-peak identification with ATNOS (9), automated NOE assignment with CANDID (10), and structure calculation with the torsion angle dynamics algorithm of CYANA (8) were performed. In the second and subsequent cycles, the intermediate protein structure was used as an additional guide for the interpretation of the NOESY spectra. During the first six cycles, ATNOS/CANDID/CYANA uses ambiguous distance restraints. In the final cycle, only distance restraints which could be attributed to a single pair of hydrogen atoms were retained. The 20 conformers with the lowest residual CYANA target function values obtained from the seventh ATNOS/CANDID/CYANA cycle were energy minimized in a water shell with the program OPALp (18, 21), using the AMBER force field (5). The program MOLMOL (19) was used to analyze the ensemble of 20 energy-minimized conformers.

Analysis of the stereochemical quality of the molecular models was accomplished using the Joint Center for Structural Genomics Validation Central Suite (http://www.jcsg.org ), which integrates seven validation tools: Procheck, SFcheck, Prove, ERRAT, WASP, DDQ, and Whatcheck.

Perfluoro-octanoic acid-polyacrylamide gel electrophoresis (PFO-PAGE) was performed according to the method of Ramjeesingh et al. (30). Purified protein samples were mixed 1:1 with PFO loading buffer containing 8% (wt/vol) PFO, 100 mM Tris, 20% (vol/vol) glycerol, and 0.05% (wt/vol) orange G. Samples with protein concentrations of 250 μM, 500 μM, and 1 mM were loaded onto precast 4 to 20% Tris-glycine gels, and electrophoresis was performed with a standard Tris-glycine running buffer (Invitrogen) to which 0.5% (wt/vol) PFO was added. Protein was detected by SYPRO-ruby poststain (Invitrogen).

Protein samples (twofold dilutions from 128 μM to 1 μM) were mixed with 0.8 μg of RNA substrate in 20 μl of assay buffer containing 150 mM NaCl, 50 mM Tris (pH 8.0), and 5 mM CaCl₂. The RNA sequences used included ssRNA1, AAAUACCUCUCAAAAAUAACACCACACCAUAUACCACAU, and ssRNA2, GGGGAUAAAA. Samples were incubated at 37°C for 1 h and analyzed by native electrophoresis on precast 6% acrylamide DNA retardation gels (Invitrogen). RNA was detected by SYBR-gold poststain and photographed using a UV light source equipped with a digital camera. Protein was then detected by SYPRO-ruby poststain. Densitometric analysis was performed using a flatbed scanner with ImageJ software (NIH). The mobility shift of RNA at each protein concentration was calculated relative to the maximum shift observed in each experiment. K_d (dissociation constant) values were determined from the midpoints of the fitted titration data (37).

nsp3a(1-183) and nsp3a(1-112) were incubated with several different nucleases in order to characterize nucleic acids that copurified with both proteins. RNase-free DNase I (NEB), T7 endonuclease I (NEB), RNase I_f (NEB), RNase A (Invitrogen), and RNase T₁ (Ambion) cleavage assays were thus performed at 37°C for 1 h with the manufacturer's recommended buffer conditions. Digested samples were analyzed by native electrophoresis on precast 6% acrylamide DNA retardation gels (Invitrogen). Nucleic acid was detected by SYBR-gold poststain.

The ¹H, ¹³C, and ¹⁵N chemical shifts have been deposited in the BioMagResBank (http://www.bmrb.wisc.edu ) under accession number 7029 (36). The atomic coordinates of the bundle of 20 conformers used to represent the solution structure of nsp3a(1-112) and of the conformer closest to the mean coordinates of the ensemble have been deposited in the Protein Data Bank (PDB; http://www.rcsb.org/pdb/ ) under accession numbers 2GRI and 2IDY, respectively.

The NOE cross-peaks that were unambiguously assigned in the seventh cycle of the ATNOS/CANDID/CYANA calculation (see Materials and Methods for details) yielded 1,888 meaningful upper distance limits, which were used as input for the final structure calculation with the program CYANA. The residual CYANA target function value of 1.88 ± 0.28 Ų and the average global root-mean-square deviation (RMSD) value relative to the mean coordinates of 0.77 ± 0.09 Å calculated for the backbone atoms of residues 20 to 108 in the bundle of Fig. 2a (Table 1) represent a high-quality NMR structure determination.

nsp3a(1-112) exhibits a ubiquitin-like fold with four helices and four β-strands arranged in the sequential order β1-α1-β2-α2-3₁₀-α3-β3-β4 (Fig. 1 and 2). The long helix α2 and the presence of the α1- and 3₁₀-helices, which have not been observed in other ubiquitin-like proteins, make the overall structure more elongated than other ubiquitin-related folds. The strand β1 spans residues 20 to 24 and is connected via a well-defined nine-amino-acid linker to the helix α1 containing residues 34 to 37. A short turn then leads to β2 with residues 42 to 46. The helix α2 with residues 52 to 66 is followed by a short loop that leads to the 3₁₀-helix of residues 70 to 75, which is further connected by a short turn with the helix α3 of residues 79 to 84. The last two regular secondary structures, β3 with residues 89 to 91 and β4 with residues 101 to 106, form an antiparallel β-sheet, and they are connected to each other by a tight turn followed by an extended chain segment. The electrostatic potential surface of nsp3a(1-112) shows a pronounced polarity (Fig. 3), with the helices α2, α3, and 3₁₀ exhibiting mainly negative charges to the solvent while the strands β1 and β3 and helix α1 contain primarily positive or hydrophobic surface residues.

nsp3a(1-112) is the second domain with a ubiquitin-like fold found within full-length nsp3. Previously, the N-terminal 70-amino acid segment of the fourth domain of nsp3, nsp3d (or PLpro), was found to have a ubiquitin-like fold (31). In Fig. 4, regular secondary structure elements in the segment 20 to 104 of nsp3a have been superimposed with the corresponding polypeptide segments in the region of residues 725 to 777 of nsp3, which corresponds to the N-terminal domain of PLpro (31). In as far as they overlap, the two structures share the same topology as canonical ubiquitin-like proteins, such as ISG15 (24) and Bacillus subtilis YukD (41). However, nsp3a also displays unique features (Fig. 4, yellow ribbon); i.e., the connection between strands β1 and β2 in nsp3a is longer than that in nsp3d and includes helix α1, nsp3a has two additional helices inserted between strands β2 and β3, and helix α2 is much longer in nsp3a than in nsp3d.

Mobility in the two nsp3a protein constructs was investigated by heteronuclear ¹⁵N{¹H}-NOE experiments (Fig. 5 and 6). Figure 5 shows the values of the steady-state ¹⁵N{¹H}-NOE for each ¹⁵N-¹H moiety in nsp3a(1-112). Residues 20 to 108, for which the mobility of the backbone ¹⁵N-¹H moieties is essentially limited to the overall tumbling of the molecule, have positive NOE values of about 0.8. In contrast, residues 1 to 19 and 110 to 112 have values in the range of −0.4 to 0.5, indicating increased mobility for these polypeptide segments, which are also visibly less well defined in the structure (Fig. 2a).

In order to investigate the structural role of the Glu-rich subdomain of residues 113 to 183, two nsp3a variants were generated which differ in the presence or absence of the C-terminal Glu-rich region, and the 2-D [¹⁵N,¹H]-HSQC spectra of the two proteins were then compared (Fig. 6a). There are no significant changes in the chemical shifts of the resonances of residues 2 to 112 in the two proteins, which indicates that both variants contain a similarly structured globular domain. These data also show for the full-length nsp3a (Fig. 6a, blue peaks) that most of the peaks from residues 113 to 183 are in the random coil chemical shift region (¹H shifts between 7.5 to 8.5 ppm). These chemical shifts and the high intensity of these resonances compared with the peaks from the globular region (Fig. 6a and b) are indicative of a flexibly extended polypeptide segment. This is confirmed by the fact that the ¹⁵N{¹H}-NOE values for most of the peaks corresponding to residues 113 to 183 are negative (Fig. 6c, pink peaks). Thus, the C-terminal Glu-rich subdomain is best described as a flexible tail of residues 113 to 183 attached to the globular domain of residues 1 to 112.

During the purification of nsp3a(1-112), we noticed that the retention volume of nsp3a by size exclusion chromatography (Superdex 75; Amersham) was lower than expected for a globular protein with a molecular mass of 12.6 kDa. In view of the implications for the structure determination and the biological activity of the protein, we decided to further investigate the oligomeric state of nsp3a(1-112) in solution using NMR diffusion experiments and PFO-PAGE.

In diffusion NMR experiments, the decay of the signal intensity versus the square of the magnetic field gradient was used to estimate the translational diffusion properties of the proteins (40). In Fig. 7a we compare data obtained for 1 mM solutions of nsp3a(1-112), RNase A, and chymotrypsinogen, which have molecular masses of 12.6 kDa, 13.7 kDa, and 25.0 kDa, respectively. The nsp3a(1-112) intensity decay curve is located between the two standards, which is indicative of the presence of the monomeric form, since the elongated shape of nsp3a(1-112) should result in a lower diffusion coefficient than near-spherical proteins with similar molecular masses. A PFO-PAGE gel also indicates that nsp3a(1-112) exists predominantly in the monomeric form at room temperature. The assays performed at the three different protein concentrations of 1 mM, 500 μM, and 250 μM (Fig. 7b) show that even at a 1 mM concentration the monomeric form predominates, and only a small amount of the dimeric form can be observed.

In the initial nsp3a(1-112) purification assays (see Materials and Methods), the protein copurified with small fragments of ssRNA. These fragments were readily detected in the 1-D ¹H NMR spectra (Fig. 8a) and were subsequently also observed by native PAGE analysis. In addition to preparing the nucleic acid-free protein for the NMR structure determination (described in Materials and Methods), we also investigated the nature of the copurifying nucleic acids. To this end a sample of nucleic acid-loaded nsp3a(1-112) was unfolded in 6 M guanidinium-HCl solution, and the mixture was subsequently loaded onto a Superdex 75 size exclusion column (Fig. 8c). The mass spectrometry analysis of the isolated fragments allowed us to identify an RNA component with a molecular weight of 1327.3 (Fig. 9). The different peaks found in this spectrum are consistent with the sequences AU, GAU, and GAUA, with the longest component corresponding to GAUA.

Nuclease digestion assays of protein samples containing copurifying nucleic acids further revealed that the major species associated with nsp3a(1-183) was DNA, which could be completely removed by DNase I treatment (Fig. 10a), and that the shorter form of the protein retained a much smaller nucleic acid species that was partly susceptible to RNase A digestion and was not susceptible to RNase I or T1 digestion (Fig. 10a). The incomplete digestion by RNase A and the lack of cleavage by RNase I or T1 were interpreted as an indication of the formation of a robust protein-RNA complex.

We then went on to study the binding of exogenous ssRNA substrates to nsp3a(1-112), starting from the aforementioned observation that the endogenous RNA contained the predominant trinucleotide sequence AUA. We thus designed two AUA-containing ssRNA fragments for further studies, ssRNA1 with the sequence AAAUACCUCUCAAAAAUAACACCACACCAUAUACCACAU and ssRNA2 with the sequence GGGGAUAAAA. The binding of nsp3a(1-112) to these two ssRNAs was assessed by EMSA, using RNA-free protein prepared as described in Materials and Methods. The EMSA showed that nsp3a(1-112) bound to the two ssRNA substrates with similar affinity (Fig. 10b). Measurement of the percentage of bound ssRNA1 at variable concentrations of nsp3a(1-112) (Fig. 10c) allowed us to estimate the dissociation constant of the nsp3a(1-112)-ssRNA1 complex to be approximately 20 μM. In control experiments, no binding was observed with several single-stranded DNA (ssDNA) sequences, or with double-stranded RNA sequences containing the motif AUA (Fig. 11). Furthermore, no binding to smaller ssRNA forms, such as fragments containing only G and U, and Octa-A, Octa-C, Octa-G, and Octa-U could be detected. Thus, RNA binding by nsp3a is consistent with the profile of a sequence-sensitive ssRNA-binding protein.

Following up on these results, NMR chemical shift perturbation studies were performed in order to map the regions of nsp3a(1-112) that are affected by the interaction with ssRNA2. Figure 12a and b show the effect of the addition of ssRNA2 on the chemical shift for each residue in nsp3a(1-112). The residues with large chemical shift perturbations are all located on the same surface area of the protein. It is rather surprising that this contact region comprises the two loops linking β3 and β4, β1 and α1, and the helices α1 and 3₁₀, which contain a surplus of negatively charged amino acid side chains (Fig. 3). There is thus an indication that these chemical shift perturbations might result primarily from long-range effects on the protein conformation rather than from direct protein-RNA contacts.

nsp3a did not interact with other ssRNA species tested. For example, the superposition of the [¹⁵N,¹H]-HSQC spectra of nsp3a in the presence and absence of Octa-U (Fig. 12c) does not show any significant chemical shift differences, indicating that Octa-U does not bind to the protein and supporting the idea that the interaction of nsp3a(1-112) with ssRNA is sequence specific.

Many of the structural homologues of nsp3a interact with other polypeptides to regulate processes such as protein degradation, cell signaling (12), and antiviral response (24). It seems significant that five of them are Ras-interacting proteins. Based on the primary sequences, the ubiquitin α/β-roll superfold comprises five families (14). Members of three of these families, RA (RalGDS/AF6 Ras-association domain), RBD (Raf-like Ras-binding domain) and PI3K_rbd (Ras-binding domain of phosphatidylinositol 3-kinase-like proteins) interact with Ras (14). A large fraction of the structural homologues of nsp3a(1-112) identified using the software DALI are members of these families. The protein nsp3a(1-112) has the highest structural similarity with the Ras-interacting domain (RID) of RalGDS, a member of the RA family with which it shares the topology of the ubiquitin-like fold (Fig. 13d). This effector of Ras is a stimulator of the guanine nucleotide dissociation mechanism specific for Ral. RID-RalGDS binds Ras through its C-terminal domain and presents low sequence identity with other Ras-interacting proteins but similar hydrophobic profiles (12). The superposition of the 3-D structures of RID-RalGDS and nsp3a(1-112) reveals a region with conserved residues located in strand β1 of nsp3a(1-112) (Fig. 13a and b) that is intimately involved in the Ras contact interface. Similarly, the Ras-binding domain of the AF6 protein (29), which is also a member of the RA family, shows 3-D structural homology with nsp3a(1-112) (Fig. 13d) and similar residues located in the β1 region (Fig. 13b). Both RalGDS and AF6 are known as Ras effectors. Similar patterns are also found in other RA domains with significant levels of structural homology with nsp3a(1-112), e.g., the human Grb7 protein and the guanine nucleotide exchange factor for Rap1 (25).

In general, Ras domains contain a combination of hydrophobic and acidic residues that interact with hydrophobic and positive groups on RIDs. Both nsp3a and the different, aforementioned Ras-interacting proteins exhibit these characteristics (Fig. 13c). In nsp3a, the conserved basic residue R23 is located in strand β1, which exhibits a high degree of consensus with the RA family sequence (Fig. 13b). This suggests that nsp3a could interfere in biological processes that involve Ras. Given its high degree of similarity with RA family proteins, there might be a potential interaction of nsp3a with human Ras proteins during SARS-CoV infection. Ras family proteins act as molecular switches that cycle between inactive GDP- and active GTP-bound states. In this manner, Ras family proteins control cell growth, motility, intracellular transport, and differentiation. The fundamental role of Ras in the cell cycle progression from phase G₀ to G₁ has been extensively reported (6, 27). Molecular interactions that result in Ras inactivation prevent cell progression to the G₁ phase. In this context, murine hepatitis virus is able to induce cell cycle arrest in the G₀/G₁ phase during the lytic infection cycle (4). It has also been shown that some SARS-CoV proteins are able to induce apoptosis or G₀/G₁ arrest in transfected cells (17, 43, 44). Overall, the structural similarity of nsp3a(1-112) and RIDs thus leads us to hypothesize that nsp3a may have a physiological role in cell cycle arrest.

The Glu-rich C-terminal polypeptide segment 113 to 183 of nsp3a shows less than 25% sequence identity with the corresponding acidic regions in other CoV genomes, whereas the SARS-CoV protein contains overall a somewhat higher percentage of acidic residues than the acidic regions of other CoVs. Similar motifs are found in some eukaryotic proteins (11, 23). In mammals, these acid-rich polypeptide segments are mainly involved in the transport of mRNA from the nucleus to the cytoplasm by association with RNA binding proteins. For example, the pp32/leucine-rich acidic protein associates with HuR, which binds to AU-rich elements of mRNAs to export mRNAs from the nucleus to the cytoplasm (11). Interestingly, Higashino et al. reported that several viruses interfere with this transport in order to increase the production of their virions by the cellular machinery (11).

The NMR data of Fig. 5 and 6 now show that the Glu-rich region of nsp3a forms a flexible tail attached to the globular region of residues 1 to 112. Although sequence similarity to other proteins is not identifiable, several cellular proteins also contain regions with high percentages of acidic residues. The homopolymer of glutamic acid, poly-l-Glu, is unstructured at pH 8 but can adopt helical structures at pH 5 (15). Some acidic regions of polypeptides exhibit a well-defined regular secondary structure when interacting with other proteins. The structure of the RanGAP (35) complex with RanBP1 and RanGAP presents examples of both situations. Ran is a nuclear Ras-related protein that regulates both transport between nucleus and cytoplasm and the formation of the mitotic spindle or nuclear envelope in dividing cells (35). The C-terminal region of RanGAP, which is important for binding affinity, exhibits an acidic motif that is flexibly disordered in both the complexed and the uncomplexed forms of RanGAP, whereas other acid-rich segments of this protein comprise folded secondary structure elements. Given the high degree of similarity of nsp3a(1-112) with some of the other polypeptides involved in protein-protein interaction processes, as well as its location in the large multidomain protein nsp3, the long, flexibly extended Glu-rich segment could have an important role in interactions with other SARS-CoV or host cell molecules, and this domain might adopt a well-defined fold during interactions with other polypeptides. Overall, the structural data reported in this paper indicate that the globular and nonglobular subdomains of nsp3a are important for SARS-CoV infection and persistence and thus represent new potential targets for therapeutic intervention.