Model structure of APOBEC3C reveals a binding pocket modulating ribonucleic acid interaction required for encapsidation

Comparative Modeling of A3C.

Structures of APOBEC2 (A2) and A3G, C-terminal domain (A3G-CD) have been solved experimentally (20–23). A2 crystallizes as a homotetramer [Protein Data Bank (PDB) identifier 2NYT, 2.5 Å resolution] composed of 2 outer and 2 inner monomers, forming a dimer of dimers, each of whose β-strands form an extended β-sheet. Each monomer possesses 1 copy of the conserved deaminase motif H-X-E-X_23–28-C-X_2–4-C coordinating 1 catalytic Zn²⁺ ion. Whereas the overall conformations of the inner and outer monomers—chains A and C, B and D—differ only slightly [0.2–1.1 Å pairwise root mean square deviation (RMSD)], the orientation of E60 with respect to the Zn²⁺ differs remarkably, possibly representing a molecular switch between the active (outer monomers) and inactive (inner monomers) conformation (22). We thus chose chains B and D as possible templates for comparative modeling. Because several residues are not resolved in chain D, chain B was chosen as the template for the A3C model.

Recently, 2 solution structures (PDB identifiers 2JYW and 2KBO) and a crystal structure (PDB identifier 3E1U, 2.3 Å resolution) of A3G-CD have been obtained (20, 21, 23). Superposition of A2 and the crystal structure of A3G-CD (21) reveal the common fold of the 2 polypeptides despite their relatively low sequence identity (<30%) (Fig. 1). Notably, in the solution structure of A3G-CD, one β-strand adopts a loop conformation (20). Because this segment is anticipated to be involved in domain dimerization in native full-length A3G, it has a limited suitability as a template structure for comparative modeling of A3C. Stereochemical quality was higher in the A2 structure [supporting information (SI) Fig. S1]. Because we anticipate A3C to form oligomers and A3G-CD was crystallized as a monomer, whereas A2 has been crystallized as a tetramer, oligomerization properties might be better represented by native A2 as a template than by A3G-CD. Modeling of A3C on both templates at the same time led to poor stereochemistry of the resulting models, which could not be resolved by geometry optimization.

Alignments of A2 and A3G-CD to A3C were carried out using MODELLER (24), explicitly considering structural information of the templates (Fig. S2). We yielded favorable BLAST (25) e-values (A3C to A2: 6 × 10⁻²³, A3C to A3G-CD: 10⁻³⁰; BLOSUM62). All residues involved in the Zn²⁺ coordination are conserved in both alignments. Insertions and deletions were placed in loop regions. Alignments are supported by matching secondary structure predictions [PSIPRED (26)] to those from the crystal structures. For both models, there is no correspondence in the templates for the N-terminal amino acids of A3C, so this region had to be constructed without template. Residue conservation was mapped back to the templates (Fig. 1 A and C) and is substantially higher in the protein core and around the Zn²⁺-coordinating center, showing a striking pattern of alternating conserved/nonconserved residues in buried/exposed parts of both α-helices (conserved: i, i+3, …) and β-strands (conserved: i, i+2, … ).

Ten initial models were built for each template, energy-minimized, and evaluated for robustness (27). For each of the templates, the model with the fewest violations of the stereochemistry was comparable to the quality of the template structures (Figs. S1 and S3) and subjected to 3 independent runs of 20 ns molecular dynamics (MD) simulations (Fig. S4). Convergence of RMSD and energy parameters suggested the fold to be stable (Fig. S5). Identical protocols were applied for the template structures, also showing convergence. Simulated B-factors were calculated from the trajectories for the template structures as described previously (28), averaged for each system, and compared with the experimental B-factors reported for the x-ray structures (r = 0.68 for A2, r = 0.76 for A3G-CD; Fig. S5), suggesting the dynamics of the structure to be well captured by our MD simulation (29). Taken together, these results suggest that both models of A3C are valid from a structural point of view and thus useful to deduce further hypotheses.

Although the 2 minimized models show a moderate pairwise RMSD of 2.7 Å (without the N-terminal amino acids in the loop region preceding the β-sheet), all residues considered in this study are located at overlapping positions (RMSD 0.7–1.4 Å) in the 2 models within precision expected from given levels of sequence identity and thus are practically equivalent (Fig. 2). Here, only the model structure of A3C based on the structure of A2 is shown. All experiments conducted in this study have been replicated with the A3G-CD–based model without significant change of predictions.

A3C Functions as a Dimer.

For A3C, which possesses only 1 domain, it is reasonable to assume a mode of dimerization analogous to that of A2, whereby the β-strands of 2 monomers build a single extended sheet. We would assume a similar mode of domain interaction in full-length A3G. We posed the questions whether (i) A3C oligomerizes, and (ii) there is differential activity between the monomeric and oligomeric forms.

First, protein–protein interaction interfaces were predicted using ProMate (30). Amino acids corresponding to the A2 dimerization interface were highlighted as potentially involved in the interaction. The protein docking and clustering technique ClusPro (31) accurately reproduced the A2 dimer (RMSD = 1.7 Å) and was applied to generate an A3C dimer model in silico (Fig. 3A). The overall topology of both the A2 and A3C dimer models is similar.

Selected amino acids in the predicted dimerization interface of A3C were mutated. Because exposed aromatic amino acids often participate in protein–protein interaction (32), 2 prominent aromatic amino acids (F55 and W74) and K51, possibly contributing to electrostatic interactions, were mutated to alanine (Fig. 3B). All constructs (K51A, F55A, and W74A) were expressed, showing WT-like degradation in presence of Vif (Fig. 3C). To determine whether these A3C mutants display antiretroviral activity, virions were generated by cotransfection with A3C expression plasmids. In transduced cells, A3C and K51A reduced the infectivity of the Δvif SIV by approximately 90–120-fold, and F55A and W74A showed a clearly diminished inhibitory activity (approximately 2–4-fold inhibition of Δvif viruses) (Fig. 3D). In contrast, WT SIV was not inhibited by any of the A3C mutants. Inhibition of Δvif SIV by WT A3C and K51A was shown to be dose dependent, whereas F55A and W74A were still inactive although using highest amounts of expression plasmid (Fig. S6).

Packaging of A3 into the virion is crucial for its antiretroviral activity (4, 8, 9). To determine whether missing encapsidation of the inactive mutants is responsible for the absence of inhibition, virions were generated and analyzed for A3C content by immunoblot analysis. Fig. 4A shows that all mutants were efficiently packaged, comparable to WT A3C.

To determine whether oligomerization of the mutant proteins correlates with antiretroviral activity, expression vectors for V5-tagged WT A3C were cotransfected with expression plasmids for HA-tagged A3Cs (WT or mutants). F55A and W74A barely precipitated V5-tagged WT A3C. In contrast, K51A and HA-tagged WT A3C were able to precipitate V5-tagged WT A3C (Fig. 4B). Quantification of the precipitated V5-tagged WT A3C in the presence of WT A3C compared with W74A resulted in a significantly higher binding efficiency (approximately 10-fold; Fig. S7). W74A-HA coprecipitated the V5-tagged WT A3C only very weakly. Crosslinking of WT A3C in total cell lysate of transfected cells showed monomeric, dimeric, and tetrameric forms, whereas W74A only formed monomers and dimers (Fig. S7). We postulate that the W74A mutation in the dimerization interface prevents the formation of the inner dimer but does not influence the formation of the outer dimer. This correlates with the observation of a weaker binding activity seen in immunoprecipitation, assuming that the inner dimer is more stable, whereas the outer dimer gets disrupted during the washing steps. These results indicate that F55 and W74 participate in dimerization of A3C and that oligomerization is crucial for antiviral activity of the enzyme. It is noteworthy that the dimer mutants did not exhibit DNA editing activity in an Escherichia coli mutation assay (compared with WT A3C; Fig. S8), which is in perfect agreement with the requirement for dimeric protein for enzymatic activity.

A Cavity of A3C Mediates its Encapsidation.

Enzyme–substrate interactions typically occur over well-defined protein binding pockets, where the active site typically is one of the largest cavities of the protein (33). Our software PocketPicker (33) was used to identify potential ligand binding pockets in the A3C model. The presumed active site was found to be the 5th biggest cavity in the A2-based structure (A3C/A2) (≈80 ų). The largest cavity (≈200 ų) is found 15 Å apart from the Zn²⁺ ion and has not yet been described in the literature (Fig. 5A). The A3G-CD–based model structure of A3C contains a similar binding pocket (Fig. 5B).

Four residues near this pocket were mutated to alanine: K22, T92, R122, and N177 (Fig. 5C). Coexpression of these mutants with Vif demonstrated protein degradation similar to that of WT A3C (Fig. 6A). To test whether mutant A3C proteins inhibit virus replication, WT and Δvif SIV were generated by cotransfection with A3C expression plasmids. WT A3C, as well as its mutants K22A, T92A, and N177A, inhibited Δvif SIV (approximately 60–84-fold) but not WT SIV (Fig. 6B). In contrast, the mutant R122A lost the inhibitory activity. Immunoblot analysis of A3C content in viral particles showed a greatly reduced packaging of R122A compared with A3C WT and K22A, T92A, and N177A (Fig. 6C), although it was detectable in cell lysates (Fig. 6A). By fusing viral protein R (Vpr) to R122A the mutant could be retargeted into viral particles and showed antiviral activity (Fig. 6 D and E). In addition, R122A showed DNA editing activity in bacteria (Fig. S8). We conclude that R122 is critically relevant for particle packaging but not for antiviral activity of A3C.

In the search for potential ligands of this presumed binding pocket, it was compared with pockets extracted from the PDBBind (34) collection with known protein function and ligands. The 4 hits that were identified as most similar stem from human papillomavirus (HPV) type 11 E2 transactivation domain (TAD) complex (PDB identifier 1R6N), a DNA binding protein of HPV; and 3 holo-structures of bovine RNase A (PDB identifiers 1QHC, 1JN4, and 1O0M), cocrystallized with different nucleotides. For the A3C/A3G-CD pocket, 4 holo-structures of bovine Rnase A (PDB identifiers 1U1B, 1W4P, 1O0M, and 1QHC) were among the 8 top scoring pockets. The natural substrates of these pockets are nucleic acids: ds DNA for HPV 11 E2 TAD, and ss and dsRNA for bovine RNase A. Steric and electrostatic properties of the hypothesized pocket in A3C would allow for nucleic acids as a substrate, possibly by R122 interacting with the negatively charged sugar–phosphate backbone.

To test for RNAs interacting with this binding pocket, WT A3C and R122A were precipitated from transfected cells, and interacting RNA was isolated and detected by ³²P labeling (Fig. 6F). Mutation of R122 resulted in strongly decreased amounts of RNA bound to the protein compared with WT A3C or a C98S active site mutation. The isolated RNA was further subjected to RT-PCR to amplify 7SL or 5.8S RNA (Fig. 6G). A3C WT protein showed an interaction with 7SL and 5.8S RNA, whereas the mutant R122A lost the binding to 5.8S RNA. Furthermore, RNA binding was shown to be crucial for interaction of A3C to SIV nucleocapsid (NC). A3C WT interacts in a RNA-dependent manner with SIV-NC, whereas R122A exhibits no binding activity (Fig. S9).

We conclude that the large pocket detected on the surface of A3C plays a key role in incorporation of A3C into viral particles, mediated by RNA-dependent interaction with SIV-NC.

Model Building.

Target A3C sequence was retrieved from National Center for Biotechnology Information accession number: NP_055323. Chain B of the tetrameric crystal structure of A2 [PDB identifier 2NYT, 2.5 Å resolution (22, 42)] and catalytic domain of A3G [PDB identifier 3E1U, 2.3 Å resolution (21)] served as template. The initial sequence alignment of A3C to monomeric A2, chain B, and the catalytic domain of A3G was performed by the align2d function of MODELLER 9v4 (24, 43). Both target–template alignment and structural coordinates of A2, chain B, and A3G-CD were used to build 10 initial models by satisfaction of spatial restraints, subjected to energy minimization and MD simulation (see SI Text).

Characterization of Binding Pockets.

PocketPicker (33) was used to automatically identify potential binding pockets and encode them as correlation vectors as described previously (33, 44). Each vector was then compared with 1,300 binding pockets with annotated function from the “refined set” of the PDBBind data set (34) using the Euclidean distance metric.

Mapping of Functional Residues.

MOE 2006.08 (Chemical Computing Group) was used to calculate the solvent accessible surface of the protein model and map electrostatic properties by a Poisson-Boltzmann potential. Putative protein–protein interaction interfaces were selected manually by looking for solvent-exposed hydrophobic patches, and fully automated using ProMate (30). Protein–protein docking and clustering of docking poses was carried out with ClusPro (31).

Plasmids.

C-terminally HA-tagged A3C expression plasmid has been described (45). Viral vectors were produced by cotransfecting pSIV_agm Δvif luc (4), pMD.G, a VSV.G expression plasmid, and supplemented by pcVif-SIV_agm-V5 (46). For coimmunoprecipitation studies pcDNA3.1-APOBEC3C-V5–6xHis (47) was used. HA-tagged mutated A3C constructs were derived by fusion PCR and cloned into pcDNA 3.1(+) (Invitrogen) using BamHI and NotI restriction sites. The pcVPR-A3C-HA expression plasmid was generated by fusion of SIVagm-Vpr cDNA to the N-terminus of HA-tagged WT or mutant A3C with a Gly₄-Ser-linker and cloned into pcDNA 3.1(+) using the same restriction sites. The premature stop codon in Vpr of SIVagm_TAN-1 was rectified by site-directed mutagenesis. Sequences of primers are given in Table S1. PCRs were performed with Phusion DNA Polymerase (Finnzymes): 1 cycle at 98 °C for 3 min; 35 cycles at 98 °C for 15 s, 65–71 °C for 30 s, and 72 °C for 20 s; and 1 cycle at 72 °C for 10 min.

Immunoblot Analysis.

For analysis of expression of A3C constructs, 293T cells were transfected with 1 μg A3C expression plasmid and 2 μg of Vif expression plasmid, using Lipofectamine LTX (Invitrogen). Two days after transfection, cells were harvested and lysed using Western lysis buffer [100 mM NaCl, 20 mM Tris (pH 7.5), 10 mM EDTA, 1% sodium deoxycholate, 1% Triton X-100, and complete protease inhibitor (Roche)]. Lysates were cleared by centrifugation and subjected to SDS-PAGE followed by transfer to a PVDF membrane. A3C-HA constructs were detected using an anti(α)-HA antibody (1:10⁴ dilution; Covance) and α-mouse horseradish peroxidase (1:7,500 dilution; Amersham Biosciences). For detection of Vif-V5 or A3C-V5, an α-V5 antibody (1:4,000 dilution; Serotec) was applied. Alpha-tubulin was detected using an α-tubulin antibody (1:10⁴ dilution; Sigma). Signals were visualized by ECL plus (Amersham Biosciences).

Packaging of A3C.

To detect A3C in virions, A3C expression constructs were cotransfected with pSIV_agm Δvif luc and pMD.G in 293T cells, as described above. Particles were precipitated by ultracentrifugation over a 20% sucrose cushion, normalized for activity of reverse transcriptase, and lysed using Western lysis buffer. Lysates were directly subjected to SDS-PAGE and transferred to a PVDF membrane. p27 was detected using a p24/p27 monoclonal antibody AG3.0 (1:250) (48).

Coimmunoprecipitation.

To detect protein interaction, expression plasmids of the respective proteins were cotransfected into 293T cells. Cells were lysed in ice-cold lysis buffer [25 mM Tris (pH 8.0), 137 mM NaCl, 1% glycerol, 0.1% SDS, 0.5% Na-deoxycholat, 1% Nonidet P-40, 2 mM EDTA, and complete protease inhibitor mixture (Roche)]. The cleared lysates were incubated with 30 μL α-HA Affinity Matrix Beads (Roche) for 60 min at 4 °C. The samples were washed 5 times with ice-cold lysis buffer. Bound proteins were eluted by boiling the beads for 5 min at 95 °C in SDS loading buffer. Immunoblot analysis and detection was done as described. Light units of the elution fractions were directly quantified from membranes incubated with ECL Plus using Lumianalyst 3.0 software (Roche).

Reporter Virus Assay.

To measure antiretroviral activity of A3C constructs, expression plasmids were cotransfected to 293T cells with pSIV_agmΔvif luc, pMD.G, in the presence and absence of a Vif SIV_agm expression plasmid. Two days after transfection, supernatants were harvested, normalized by RT activity, and transduced 2 × 10³ HOS cells in a 96-well dish. Three days after transduction, intracellular luciferase activity was quantified using Steady Lite HTS (Perkin-Elmer). Data are presented as the average counts per second of the triplicates ± SD. RT activity was quantified using the Lenti-RT Activity Assay (Cavidi Tech). The cell lines 293T and HOS were cultured in Dulbecco's high-glucose modified Eagle's medium (Invitrogen), 10% FBS, 0.29 mg/mL L-glutamine, and 100 U/mL penicillin/streptomycin at 37 °C with 5% CO₂.

A3C–RNA interaction.

To detect protein–RNA interaction, expression plasmids of the respective proteins were transfected to 293T cells as described above. Cells were lysed in ice-cold lysis buffer [PBS with 1% Triton X-100, 16 U/mL RiboLock RNase Inhibitor (Fermentas), and complete protease inhibitor mixture (Roche)]. The cleared lysates were incubated with 50 μL α-HA Affinity Matrix Beads (Roche) for 60 min at 4 °C. The samples were washed 5 times with ice-cold lysis buffer. RNA bound to immobilized proteins was extracted from HA beads using TRIzol reagent (Invitrogen), according to the manufacturer's instructions. RNAs coprecipitated with A3C proteins were dissolved in diethyl pyrocarbonate–treated H₂O, and equal amounts were used for RT using random hexamer primers (Fermentas) in the presence of α-³²P-dATP (Hartmann Analytic). Radioactive labeled cDNA was separated on a 12% TBE 8M Urea PAA Gel and visualized with Amersham Hyperfilm (GE Healthcare). Specific RT-PCR on A3C-bound RNAs was performed using Revert Aid First Strand cDNA synthesis kit (Fermentas) and random hexamer primers.

Protein–Protein Crosslinking.

For crosslinking experiments, expression plasmids of the respective proteins were transfected to 293T cells; 2 days later cells were lysed, and whole-cell lysate was incubated with 50 mM N-ethylmaleimid (Calbiochem) for 2 h at room temperature. Immunoblot analysis was performed as described above, but no DTT or β-mercaptoethanol was added to the SDS loading buffer during electrophoresis.

For additional information see SI Text.