We have performed a mutational study of residues in the N- and C-CDA domains of A3G whose counterparts in A2 are involved in key interactions that support oligomerization of A2. Our results show that mutations in the N-CDA, but not the C-CDA, are associated with reductions in A3G RNA-dependent oligomerization and packaging into virions. Upon modelling of an A3G dimer onto the template A2 crystal structure, we identified a positively charged pocket at the oligomer interface formed between two N-CDAs that bore the hallmarks of a nucleic acid binding site (). Indeed, mutation of basic residues within this pocket also resulted in losses of antiviral function, packaging into virus particles, oligomerization and association with cellular RNA ( and ). Consistent with previous work showing that only the C-CDA of A3G is responsible for DNA editing 
, this attribute was unaffected by disruption of this basic pocket. Thus, our findings demonstrate further segregation of functions between the N- and C-CDA domains of A3G.
Throughout our chemical crosslinking experiments, we consistently observed the generation of oligomeric A3G migrating at ~80 kD, which is twice the relative molecular mass of untreated A3G, which migrates at ~40 kD ( and ). This result suggests that A3G oligomerizes as a discrete dimer, an assertion that is further supported by the fact that we did not detect slower migrating species at ~120 (trimer) or ~160 kD (tetramer). We note, however, that we have not formally demonstrated dimerization of A3G, as attempts to perform analytical ultracentrifugation were unsuccessful owing to the poor solubility of purified full-length A3G at high concentrations (results not shown). Nonetheless, dimerization of A3G via the N-terminal CDA domain remains the simplest model to explain our current results. Although this conclusion is at odds with a recent study proposing oligomerization of A3G via the C-CDA domain 
, that study is also contradicted by the observations that the C-CDA of A3G appears as a monomer by both ultracentrifugation 
and crystallography 
Importantly, we have furthermore demonstrated that the oligomerizaton of A3G is dependent on the presence of RNA, as evidenced by the disruption of oligomers upon treatment with RNase (). These observations are explained by our combined modelling and structure-function analyses, which predict that the oligomer interface between the A3G N-terminal CDA domains produces a positively charged pocket that requires occupation by RNA to allow effective oligomerization. Thus, we propose that the formation of A3G oligomers requires hydrophopic and basic residues that mediate protein-protein interactions between the A3G subunits and/or protein-RNA interactions, in a manner similar to that proposed for PKR and RIG-I 
. We acknowledge, however, that the precise contribution of these residues to RNA-dependent oligomerization of A3G must await advances in the biochemical characterization of this protein.
An additional piece of evidence supporting the interdependence between the oligomerization of A3G and the association with RNA comes from the analysis of Y1, Y4 and 7SL RNA in A3G RNPs (). In general terms, we observed that oligomerization-impaired mutants of A3G exhibited much reduced co-immunoprecipitation of these RNA molecules. Indeed, correlations between oligomer formation and RNA interaction were excellent in that the R30A and Y124A mutants displayed partial A3G-A3G interactions as well as intermediate levels of RNA interactions (, , and ). An additional instructive observation was made upon velocity sedimentation of cell lysates with oligomerization-impaired A3G, which demonstrated that assembly into RNase-sensitive high molecular weight RNP complexes was not noticeably affected by these mutations (Figure S3
). This demonstrates that A3G's assembly into at least two intermolecular complexes is RNA-dependent: the oligomerization of A3G and its recruitment into RNP complexes. Importantly, our mutational analysis shows that oligomerization can be disrupted selectively without grossly preventing RNP association. This suggests either that (1) there is a certain degree of specificity to the identity of RNAs that are required for A3G oligomerization, but not to the RNAs that promote RNP association, or that (2) recruitment of A3G to RNase-sensitive RNP complexes is driven predominantly by protein-protein interactions. Specifically, RNA-dependent A3G RNP formation through protein-protein interactions could be mediated by proteins that bind A3G directly and additionally bind RNA.
Throughout these studies, we have highlighted a tight correlation between the packaging into HIV-1 virions and the RNA-dependent oligomerization of wild type and mutant A3G proteins. Here, we have presented a structure model of an A3G dimer that readily accommodates these attributes. Indeed, the packaging of A3G into virus particles has been reported to require binding to RNA and this has been interpreted as reflective of an RNA-dependent interaction between the HIV-1 Gag protein and A3G 
. Although the identity of RNA required for packaging of A3G into HIV-1 virions remains debated 
, specificity with regards to the RNA molecules that mediate oligomerization of A3G may impart some of the selectivity for the establishment of an A3G-Gag interaction and virion packaging.
The structure of A3G is also of considerable interest with regard to the binding of the HIV-1 Vif protein and efforts to manipulate this interaction therapeutically. Previous analyses have shown that Vif interacts with a three amino-acid core motif in A3G at residues 128–130 
, which is directly adjacent to residues Y124 and W127. This would position the residues of A3G that interact with Vif in close proximity to the oligomer interface. In our previous study, we found that mutant proteins with substitutions at position Y124 or W127 remain responsive to regulation by the Vif protein 
, suggesting that oligomerization is not a prerequisite for binding of Vif. Indeed, the interaction of Vif with A3G in co-immunoprecipitation experiments is resistant to treatment with RNase 
. Similarly, mutations at residues 128–130 in A3G affect the interaction with Vif but not packaging into virus particles 
or generation of the dimeric species by chemical crosslinking (results not shown). Thus, while the residues that mediate Vif-binding and RNA-dependent oligomerization are in close proximity, they appear to be functionally distinct.
We have presented evidence for the RNA-dependent oligomerization of A3G via its N-CDA domain. A structure model of an A3G dimer based on the A2 crystal structure readily rationalizes the RNA-dependency of oligomerization as it revealed a clustering of positive charge near the predicted dimer interface. Furthermore, the model proved consistent with the contribution of basic residues at the interface to RNA-dependent oligomerization and packaging of A3G into virus particles. We thus propose that this model can serve as a guide for the further dissection of the structure-function relationships of domains and motifs within A3G. Ultimately, this may help endeavours aimed at therapeutic intervention with the interaction between the HIV-1 Vif protein and A3G. In particular, such efforts should strive to preserve the antiviral functions of A3G by interrupting the interaction with Vif, while maintaining the interactions that mediate association with RNA, oligomerization and virion packaging.