The Binding Interface Between Human APOBEC3F and HIV-1 Vif Elucidated by Genetic and Computational Approaches

APOBEC3 family DNA cytosine deaminases provide overlapping defenses against pathogen infections. However, most viruses have elaborate evasion mechanisms such as the HIV-1 Vif protein, which subverts cellular CBF-β and a polyubiquitin ligase complex to neutralize these enzymes. Despite advances in APOBEC3 and Vif biology, a full understanding of this direct host-pathogen conflict has been elusive. We combine virus adaptation and computational studies to interrogate the APOBEC3F-Vif interface and build a robust structural model. A recurring compensatory amino acid substitution from adaptation experiments provided an initial docking

To counteract the deleterious effects of these enzymes, HIV-1 encodes a protein called virion infectivity factor (Vif) that scaffolds the formation of an E3 ubiquitin ligase complex that directly binds restrictive A3 enzymes and targets them for proteasome-mediated degradation [reviewed by (Desimmie et al., 2014;Harris and Dudley, 2015)]. The Vif protein of primate lentiviruses HIV-1, HIV-2, and simian immunodeficiency virus (SIV) has evolved to dimerize with the transcription cofactor CBF-β [ (Anderson and Harris, 2015) and references therein]. This complex enables Vif to additionally bind ELOC and CUL5 and, through these partners, also to ELOB and RBX2. Upon the addition of NEDD8 to CUL5, this E3 ligation assembly becomes activated to transfer ubiquitin from an E2 enzyme to Vifengaged A3 substrates. Proteomic approaches have been instrumental in identifying Vif ligase components, and genetic studies have demonstrated the essential nature of all of the subunits. A crystal structure of HIV-1 Vif-CBF-β-ELOB-ELOC-CUL5 has helped rationalize prior results (Guo et al., 2014).
A3F is a notable HIV-1 restriction factor for multiple reasons. First, overexpression causes a dose-responsive drop in the infectivity of Vif-defective viruses and inflicts mutations in a GA-to-AA context (minus strand TC-to-TT) (Bishop et al., 2004;Liddament et al., 2004;Wiegand et al., 2004;Zheng et al., 2004). Second, knockout and knockdown experiments demonstrate that endogenous levels of this enzyme contribute to HIV-1 restriction and hypermutation in a CD4-positive T cell line and, together with A3D and A3H, account for the GA-to-AA hypermutation signature observed in patient-derived viral sequences (Refsland et al., 2012). Third, HIV-1 restriction activity is conserved between human and rhesus macaque A3F (Hultquist et al., 2011;Virgen and Hatziioannou, 2007;Zennou and Bieniasz, 2006). Fourth, both Vif-binding and DNA cytosine deaminase activities reside within a single zinc-coordinating domain, unlike A3G where these activities are split between the N-and C-terminal domains, respectively (Haché et al., 2005;Russell et al., 2009). Fifth, multiple high-resolution structures and mutagenesis studies have indicated that the Vif-binding surface of A3F is similar to that of A3D and APOBEC3C (A3C), but at least partially distinct from those of A3G and A3H (Albin et al., 2010b;Bohn et al., 2013;Kitamura et al., 2012;Kouno et al., 2015;Land et al., 2014;Russell et al., 2009;Schröfelbauer et al., 2006;Siu et al., 2013;Smith and Pathak, 2010). Sixth, HIV-1 Vif separation-of-function mutants demonstrated the importance of A3F in humanized mouse models for viral pathogenesis (Krisko et al., 2013;Sato et al., 2014). Finally, in long-term cell culture experiments, A3F is sufficiently potent to suppress Vif-deficient HIV-1 replication and select for Vif restoration-of-function mutants (Albin et al., 2010a). This final point is important because it indicates that HIV-1 requires Vif function in order to antagonize A3F and preserve the capacity to replicate. It also implies that HIV-1 cannot easily evolve a Vif-independent mechanism to evade A3F, which further distinguishes this enzyme from multiple Vif-independent mechanisms that the virus may employ to escape A3G restriction (Haché et al., 2008). In other words, the A3F-Vif interaction is essential from an HIV-1 perspective.
Despite these and other advances in the overall understanding of Vif and A3 biology, it is still not clear how Vif actually binds to an A3 substrate. Here, we leverage the selective potential of A3F and virus adaptation studies to identify A3F-interacting amino acid residues within Vif. A recurring genetic contact provided an anchor point for docking the two proteins and deriving a physical interaction model. Molecular dynamics simulations led to a refined Vif-A3F interaction model that predicted a durable anchoring electrostatic interaction. Charge reversing substitutions in either A3F (E289K) or Vif (R15E) abrogated the interaction, but the combined amino acid charge changes enabled the interaction to be restored as evidenced by A3F degradation and high viral infectivity. Genetic gain-offunction data were therefore instrumental in both informing the initial model as well as validating the refined model. Our Vif-A3F interaction model also explains prior loss-offunction observations. Overall, our multidisciplinary approach has yielded a comprehensive three-dimensional structural model of this critical pathogen-host interface and also inspired a general "wobble model" to explain the rapid evolution of this and other pathogen-host interactions.

HIV-1 Adapts to Replicate in the Presence of Vif-Resistant APOBEC3F
The lentiviral Vif protein is thought to have evolved to selectively neutralize the A3 repertoire of cognate host species [reviewed by (Desimmie et al., 2014;Harris and Dudley, 2015)]. Consequently, virus replication is typically restricted by A3s of non-cognate species. For example, the replacement of human A3F (huA3F) residues E324 with the corresponding rhesus macaque A3F (rhA3F) residues K324 renders the human enzyme resistant to degradation by HIV-1 Vif [ (Albin et al., 2010b;Land et al., 2014) and below]. We therefore hypothesized that cells expressing this simianized form of huA3F or the naturally restrictive rhA3F (Hultquist et al., 2011;Virgen and Hatziioannou, 2007) would provide selective pressure to force HIV-1 to adapt by making compensatory changes in Vif, analogous to the way that ancestral SIV strains may have adapted in order to transmit into the human population ( Figure 1A).
To test this hypothesis, HIV-1 IIIB and HIV-1 LAI-GFP stocks were used in graded selection experiments with each passage containing increasing proportions of SupT11 clones stably expressing either huA3F-QE323-324EK or rhA3F (i.e., permissive T cells rendered nonpermissive through stable expression of these restrictive A3 enzymes). A3F levels in low expressing T cell lines approximated those in primary CD4-positive T cells (Land et al., 2014), whereas those in medium and high expressing lines were intentionally superphysiological ( Figure 1B). This approach allows viral sequence diversity to accumulate while selecting for compensatory changes. In total, four rounds of step-wise selection were performed, followed by four additional rounds of MOI controlled passages under 100% nonpermissive conditions to select high fitness viruses (Experimental Procedures).
Dozens of independent adapted viral populations were obtained as evidenced by similar replication kinetics in permissive SupT11 control cells and nonpermissive A3F expressing derivatives. Adapted variants were identified by sequencing vif from proviral DNA, and a clear hotspot emerged with Vif G71D dominating both selective conditions ( Figure 1C). Although other amino acid substitutions occurred, none was as prominent as Vif G71D and none apart from Vif G71D yielded a clear phenotype in the context of an otherwise clean molecular clone (G71D data below and additional data not shown; the identities of all amino acid changes occurring in ≥2 independent cultures are listed in Figure 1C relative to previously implicated interaction motifs in HIV-1 Vif).

HIV-1 Vif G71 Influences the Interaction with APOBEC3F
To determine whether HIV-1 Vif G71D overcomes restriction barriers imposed by huA3F QE323-324EK and rhA3F, single-cycle infectivity experiments were done with Vif G71D versus wild-type huA3F, huA3F E324K, and rhA3F. As shown in Figure 2A, G71D mutants displayed modest loss-of-function in neutralizing wild-type huA3F but gained significant activity against huA3F E324K and rhA3F. Spreading infection data corroborated these results as Vif G71D engineered into the parental HIV-1 IIIB molecular clone, with no other amino acid changes, became attenuated in cells expressing medium and high levels of huA3F but clearly gained the capacity to replicate in the presence of huA3F E324K ( Figure  2B). Peak spreading infection titers did not appear to be affected but a kinetic delay was observed suggesting that the single G71D change is sufficient to overcome restriction but not optimal for virus replication. Analogous results were obtained for spreading infection experiments with HIV-1 Vif G71 versus D71 molecular clones in SupT11 cells stably expressing rhA3F ( Figure S1).

Delineation of the Vif-A3F Interface
The gain-of-function amino acid substitution G71D selected in adaptation experiments with both huA3F-E324K and rhA3F suggested that these two residues are physically interacting. This possibility is consistent with the crystal structure of HIV-1 Vif ligase complex, where G71 is located within a solvent exposed loop on the same surface as the α-helical D14-R15-M16-R17 motif previously implicated in interacting with A3F (Russell and Pathak, 2007;Russell et al., 2009;Smith and Pathak, 2010) ( Figure 3A). It is also consistent with huA3F E324 being located within the conserved α4-helix and likewise accessible for direct interaction ( Figure 3B). In addition, E324 is part of the larger α3-α4 region of huA3F and rhA3F implicated by genetic studies as interacting with HIV-1 Vif (Albin et al., 2010b;Kitamura et al., 2012;Land et al., 2014;Russell and Pathak, 2007;Russell et al., 2009;Smith and Pathak, 2010).
We therefore used HIV-1 Vif G71 and huA3F E324 as anchoring points to generate a structural interaction model that obeys physical constraints and best explains prior genetic studies. The ClusPro protein-protein docking web server was used to generate 20 Vif-huA3F interaction models, and one model with Vif G71 and A3F E324 in close proximity was selected for further computational studies ( Figure 3C). In this model the main chain amide of Vif G71 is within bonding distance of the side chain of A3F E324 (ca. 3 Å). Additional features of this model are extensive interactions between the G71 loop and the DRMR motif of HIV-1 Vif with the α3 and α4 helices of A3F. In particular, Vif R15 is predicted to form a direct electrostatic interaction with A3F E289 ( Figure 3C).
To optimize the predicted Vif-huA3F interface, the docked complex was subjected to three independent 1-µs MD simulations ( Figure 3D; Figures S2 & S3; Movies S1-S3). The first notable observation was the relative fragility of the interaction between Vif G71 and huA3F E324, which was lost rapidly in two of the simulations (persistence times of 70-ns and 1-ns in Movies S1 & S2; inter-residue distances plotted in Figure S2). Second, the electrostatic interaction between Vif R15 and huA3F E289 was stable through large proportions of each independent 1-µs simulation (Movies S1-S3; inter-residue distances plotted in Figure S3; representative pose in Figure 3D). Moreover, all three MD simulations arrived at a single preferred interaction conformation anchored by an interaction between Vif R15 and huA3F E289.
An additional appealing feature of these MD refinements was an approximately 90° rotation in the Vif interaction to include a hydrophobic pocket of huA3F between the α2 and α3 helices ( Figure 3D; Movie S2). Specifically, Vif W79, which is located in the G71 loop, docked into the α2 and α3 hydrophobic pocket of huA3F. This predicted interaction is concordant with prior mutagenesis experiments, which had implicated this huA3F region in the Vif interaction (Kitamura et al., 2012). In addition, this MD-optimized pose still maintains interactions between Vif and the α3 and α4 regions of huA3F, including the electrostatic interaction between Vif R15 and huA3F E289. Altogether, the MD simulationoptimized Vif interaction model buries approximately 1300 Å 2 surface area between the two proteins and involves huA3F helices α2, α3, and α4, as well as the hydrophobic pocket between α2 and α3 ( Figure 3D; Movie S2).

Model Validation
MD simulations predicted a direct electrostatic interaction between Vif R15 and A3F E289 ( Figure 3D; Figures S2 & S3; Movies S1-S3). Prior work showed that Vif R15A generated a separation-of-function variant defective for huA3F degradation but not for A3G degradation, and also that huA3F E289K becomes resistant to HIV-1 Vif-mediated degradation (Russell and Pathak, 2007;Russell et al., 2009;Smith and Pathak, 2010). Thus, we predicted that reciprocal charge changes at these amino acid positions in Vif and huA3F, respectively, would singly destroy Vif degradation activity but together restore the functional interaction. Indeed, each of these charge changing amino acid substitutions alone compromised Vif degradation function but in combination the capacity for HIV-1 Vif to degrade huA3F was restored ( Figure 4A). This result is evident through the decreased huA3F band intensities in immunoblot images (indicating degradation) and through significantly increased viral infectivity in single cycle infection experiments. Additionally, this predicted gain-of-function pairing enabled HIV-1 with Vif E15 to undergo a robust spreading infection in SupT11 cells stably expressing the restrictive A3F K289 enzyme ( Figure 4B). Taken together, these data provide strong genetic evidence in support of a direct physical interaction between Vif R15 and A3F E289 and the Vif-huA3F interaction model derived above.

DISCUSSION
The Vif-A3F interaction is essential for HIV-1 pathogenesis in humanized mice, rhesus macaques, and most likely also humans (Desrosiers et al., 1998;Farrow et al., 2005;Krisko et al., 2013;Rangel et al., 2009;Sato et al., 2014;Schmitt et al., 2010;Simon et al., 2005). Here, we used human A3F E324K and rhesus A3F (naturally K324), which both resist degradation by HIV-1 Vif (Albin et al., 2010b;Hultquist et al., 2011;Virgen and Hatziioannou, 2007;Zennou and Bieniasz, 2006), as tools to select for viral variants that regain the capacity to replicate in long-term cell culture experiments. Vif G71D was the major adaptive change that emerged in these studies and the only one that reconfirmed in infectivity experiments. This infectivity restoring amino acid change suggested that a new, stabilizing physical interaction had been established between A3F K324 and Vif D71, and provided a critical constraint for docking studies. However, structural optimization using MD simulations suggested that the interaction between A3F E324 and Vif G71 is unlikely to be robust in the wild-type setting, and instead revealed a considerably more stable proteinprotein interface dominated by a persistent electrostatic interaction between A3F E289 and Vif R15. Remarkably, the reciprocal charge changes of A3F E289K and Vif R15E resulted in A3F degradation and HIV-1 infectivity phenotypes similar to a wild-type scenario. These gain-of-function results provide strong genetic evidence in favor of A3F E289 and Vif R15 constituting an integral part of the physical interaction. The best-fit model presented here provides a comprehensive structural view of the A3F-Vif interaction and, by homology modeling, also of the entire A3F-bound ubiquitin ligase complex ( Figure S5).
Prior mutagenesis studies have implicated multiple amino acids in the interaction between human A3F and HIV-1 Vif (Albin et al., 2010b;Kitamura et al., 2012;Land et al., 2014;Russell and Pathak, 2007;Russell et al., 2009;Smith and Pathak, 2010). Most of these prior results can be re-interpreted using our best-fit model and grouped into those that participate directly or indirectly in the actual interaction. Prior separation-of-function data include A3F amino acid substitutions C259K, IL262/3AA, S264D, and Y269A, which are located between the α2 and α3 helices, and E289K and E324K, which are located on the solvent exposed surfaces of the α3 and α4 helices, respectively. Each of these A3F amino acid substitutions confer resistance to Vif-mediated degradation and do not affect the enzyme's HIV-1 restriction capabilities [ (Albin et al., 2010b;Kitamura et al., 2012;Land et al., 2014;Russell and Pathak, 2007;Russell et al., 2009;Smith and Pathak, 2010) and this study]. These A3F residues are all located within the interface of our best-fit model, best highlighted by A3F E289 forming a persistent electrostatic interaction with Vif R15.
We propose a 'wobble' model to explain the dynamic nature of the Vif-A3 interaction ( Figure 5). In this model, Vif forms a network of interactions with a given A3 substrate such as A3F. If the overall interaction is destabilized, for instance, due to an immune escape variation in Vif or a naturally occurring variation in a host A3 enzyme, a compensatory change in Vif can restore a fully functional interaction. Such a compensatory change may occur at the original interacting residue, nearby, or even at an edge of the interface. In support of this idea, the negatively charged mutation G71D was selected in multiple independent adaptation experiments in order to counteract a positively charged E324K substitution in human A3F or the naturally occurring K324 in rhesus macaque A3F, yet this interaction was weak in MD simulations. On an evolutionary time scale, an extensive series of wobbles could enable the Vif-A3 interface to drift to a completely different solventexposed surface ( Figure 5). This model provides an attractive explanation for how HIV-1 Vif could have evolved to use distinct surfaces to bind each of the restrictive human A3 enzymes [e.g., A3F and A3G in Figure 5; data in this study and (Albin et al., 2010b;Kitamura et al., 2012;Kouno et al., 2015;Land et al., 2014;Ooms et al., 2013;Refsland et al., 2014;Russell and Pathak, 2007;Russell et al., 2009;Smith and Pathak, 2010)]. It also provides a molecular explanation for the evolutionary theory known as the Red Queen Hypothesis (Van Valen, 1973). Moreover, based on the relatively facile cross-species adaptation of HIV-1 from human A3F-to rhesus macaque A3F-expressing cells, this model accounts for how an ancestral lentivirus may have adapted from a host with a simple A3 repertoire (most nonprimate mammals) to be able to replicate in the presence of a more complex, multi-protein A3 repertoire as exists in modern primates including humans ( Figure  5). This wobble model may also explain the evolutionary dynamics of a wide range of pathogen-host interactions.

EXPERIMENTAL PROCEDURES Plasmids
Human A3F matching NM_145298 and A3F QE323-324EK have been described (Albin et al., 2010b;Liddament et al., 2004). Rhesus A3F matching NM_001042373.1 was provided by T. Hatziioannou (ADARC, New York) and cloned into a pcDNA3.1-derived 3xHA vector as described (Liddament et al., 2004). A V5-tagged derivative of rhA3F was made by subcloning from the 3xHA vector as described (Albin et al., 2010b). Specific point mutants were made by QuikChange site-directed mutagenesis (Stratagene). Full-length proviral HIV IIIB and HIV LAI-GFP plasmids have been described, and the HIV IIIB sequence is identical to EU541617 except for an A200C nucleotide change to interrupt an atypical upstream open reading frame (Albin et al., 2010b;Haché et al., 2009;Haché et al., 2008). Proviral mutant plasmids mutants were assembled in a TOPO shuttle vector containing the vif-vpr region and then subcloned back into the full-length context via the unique sites SwaI/ SalI in HIV-1 IIIB or PshAI/SalI in HIV-1 LAI-GFP .

Cell Culture
T cell lines were maintained in RPMI supplemented with 10% fetal bovine serum, penicillin/streptomycin and β-mercaptoethanol. 293T cells were maintained in DMEM medium supplemented with 10% fetal bovine serum and penicillin/streptomycin. Stable APOBEC3-expressing cell lines have been described and new lines were established using the same methods (Albin et al., 2010a;Albin et al., 2010b;Hultquist et al., 2011).

HIV-1 Adaptation and Infectivity Studies
The selection protocol was carried out as described in the text and depicted in Figure 1A. Adapted viruses were selected for molecular characterization if 4 th passage infectivity exceeded 2% absolute CEM-GFP infectivity for viruses adapted to A3F QE323-324EK or at least 5% for viruses adapted to rhA3F. Adapted vif genotypes were identified by preparing genomic DNA from infected CEM-GFP reporter cells (PureGene), amplifying the vif-vpr region as described (Albin et al., 2010a), and sequencing each PCR product (Genewiz). Chromatograms were inspected manually to identify mutations. Spreading infections were performed as described (Albin et al., 2010a). Single cycle infectivity studies were done as described using an HIV-1 IIIB Vif-deficient proviral plasmid, APOBEC3F-V5 or an A3Funtagged constructs, and codon-optimized HIV-1 IIIB Vif-HA constructs (Albin et al., 2010a;Albin et al., 2010b;Hultquist et al., 2011). Membranes were imaged using commercial chemiluminescence reagents (Denville Scientific) and film or a Licor Odyssey instrument.

Statistical Analysis
GraphPad Prism 5.0 was used for statistical analyses (GraphPad Software, La Jolla, CA). Quantitative data are presented as mean ± SEM. Statistical significance was determined by two-way ANOVA followed by a post hoc Bonferroni test to determine significance between or within testing groups. A p-value of less than 0.05 was considered statistically significant. (A) A schematic depicting the co-culture selection strategy used to adapt HIV-1 to Vifresistant A3F expressing cells as the selective pressure. Viruses are passaged step-wise approximately every 8-11 days from permissive to increasingly non-permissive cultures as shown. Upon completing a round of selection from fully permissive to fully non-permissive cultures, portions of each culture were cycled back to the beginning of the process for another round of selection. (B) Anti-A3F immunoblot of SupT11-derived T cell lines stably expressing human A3F, A3F QE323-324EK, or empty vector (relative protein expression levels: L, low; M, medium; H, high; VH, very high; V, empty vector). A3F levels in HIV-1 infected primary T-cell and H9 lysates are shown for comparison. (C) Bar graphs indicating the number of independent times that each of the indicated HIV-1 IIIB or LAI Vif amino acid changes were observed. The x-axis is a to-scale depiction of Vif residues 1-192 with previously reported motifs indicated below for reference. In most instances, HIV-1 IIIB and HIV-1 LAI had the same amino acid change at a given position with the exception of HIV-1 IIIB E117K and HIV-1 LAI D117N for both A3F QE323-324EK and rhA3F selective conditions (shown explicitly in the figure). Richards   D14 is the first residue of an α-helix containing Vif residues 14-31, which includes the DRMR motif. Vif G71 is located in a nearby loop on the same surface of the structure (PDB: 4N9F). Vif is colored cyan, and a faded surface representation of CBF-β is shown in green to facilitate visualization. (B) A ribbon schematic of the Vif-binding domain of huA3F highlighting the α3 and α4 helices and the position of residue E324 near the end of the α4-helix (PDB: 4IOU). (C) A model of the complex generated by docking A3Fctd onto Vif. In this initial docked model, direct interactions occur between A3F E324 and Vif G71 as well as A3F E289 and R15.
(D) An MD simulation-optimized model of the A3F-Vif macromolecular complex. Residues E289 and R15 form a strong persistent interaction, and residues within the Vif G71 containing loop are interacting with A3F residues between helices α2 and α3. See the main text for details.

Figure 4. HIV-1 Vif-A3F Interface Interactions Validated by Gain-of-Function Viral Infectivity Experiments
(A) Single-cycle infectivity data for Vif-null HIV-1 IIIB produced in the presence of huA3F, huA3F E324K, or huA3F E289K and the indicated amounts of Vif R15 (wild-type) or Vif E15 expression constructs. Each histogram bar shows the average infectivity from three independent experiments (mean ± SEM). Statistical significance was determined by twoway ANOVA followed by a post hoc Bonferroni test to determine significance between or within testing groups. P values of less than 0.05 are considered significant. Immunoblots are Richards et al. Page 17 Cell Rep. Author manuscript; available in PMC 2015 December 18.

Figure 5. Wobble Model to Explain the Evolution of the Vif-A3 Interaction
The schematic shows the wobble model for adaptation of the Vif side of the Vif-A3 interface. Each hexagon of the lattice depicts a potential interaction: ancestral, dark gray; potential, light blue; attenuated, light gray; A3F, orange; A3G, purple. The model predicts that the ancestral Vif-A3 interaction was strong and consisted of six interaction points (arbitrary number for discussion purposes). Viral transmission to a new host with a larger A3 repertoire results in diminished but still partly functional interactions. A series of rapid adaptations (possibly most or all in the original new host) restores the strong interaction and results in partly overlapping interaction surfaces. Then, over a much longer evolutionary period, the combined effect of many independent wobbles triggered by virus or host genetic changes could result in the present day non-overlapping surfaces of Vif that interact with A3F and A3G. Similar rules would apply to other HIV-1 restrictive A3s (not shown for simplicity). A strong interaction may be disrupted by a viral or host amino acid substitution mutation, but the overall interaction can be restored by a compensatory change in Vif at the same site, nearby, or even to extend the edge of an interaction surface. If this change occurs at a new position, it can be considered a 'wobble'. A series of wobbles over evolutionary time can account for a shift in the entire interaction surface. Richards et al. Page 19 Cell Rep. Author manuscript; available in PMC 2015 December 18.