Structural basis for targeting avian sarcoma virus Gag polyprotein to the plasma membrane for virus assembly

For most retroviruses, including HIV-1, binding of the Gag polyprotein to the plasma membrane (PM) is mediated by interactions between Gag's N-terminal myristoylated matrix (MA) domain and phosphatidylinositol 4,5-bisphosphate (PI(4,5)P2) in the PM. The Gag protein of avian sarcoma virus (ASV) lacks the N-myristoylation signal but contains structural domains having functions similar to those of HIV-1 Gag. The molecular mechanism by which ASV Gag binds to the PM is incompletely understood. Here, we employed NMR techniques to elucidate the molecular determinants of the membrane-binding domain of ASV MA (MA87) to lipids and liposomes. We report that MA87 binds to the polar head of phosphoinositides such as PI(4,5)P2. We found that MA87 binding to inositol phosphates (IPs) is significantly enhanced by increasing the number of phosphate groups, indicating that the MA87–IP binding is governed by charge–charge interactions. Using a sensitive NMR-based liposome-binding assay, we show that binding of MA87 to liposomes is enhanced by incorporation of PI(4,5)P2 and phosphatidylserine. We also show that membrane binding is mediated by a basic surface formed by Lys-6, Lys-13, Lys-23, and Lys-24. Substitution of these residues to glutamate abolished binding of MA87 to both IPs and liposomes. In an accompanying paper, we further report that mutation of these lysine residues diminishes Gag assembly on the PM and inhibits ASV particle release. These findings provide a molecular basis for ASV Gag binding to the inner leaflet of the PM and advance our understanding of the basic mechanisms of retroviral assembly.

also been shown that the MA domain binds to specific tRNA in the cytosol (34) to prevent Gag from interacting with intracellular membranes (15)(16)(17)35). Incorporation of PI(4,5)P 2 in membranes inhibits the interaction between MA and cellular RNA (15)(16)(17). Therefore, RNA is considered as a negative regulator of Gag-membrane binding.
NMR studies of HIV-1 MA binding to PI(4,5)P 2 , PS, PC, and phosphatidylethanolamine (PE) containing truncated (tr) acyl chains as well as membrane mimetics such as bicelles and micelles have been reported (36 -38). It was shown that binding of tr-PI(4,5)P 2 to HIV-1 MA induced a conformational change that promoted myr exposure (37). The structure of MA bound to tr-PI(4,5)P 2 showed that both the polar head and the truncated 2Ј-acyl chain are involved in binding (37). tr-PS, tr-PC, and tr-PE also bind to HIV-1 MA via a distinct second hydrophobic pocket on the protein (38). Based on these studies, we proposed a model for HIV-1 MA bound to the membrane (38). Recent studies have provided an alternative membranebinding interface of HIV-1 MA based on NMR and biochemical studies with liposomes containing lipids with native acyl chains (39). It was concluded that acyl chains are not involved in MA binding and that Gag-membrane interaction is mediated predominantly by dynamic, electrostatic interactions between conserved basic residues of MA and PI(4,5)P 2 /PS (39).
Despite the great similarity in the structural domains and their functions, retroviruses have distinct mechanisms for Gag assembly. The Gag proteins of avian sarcoma virus (ASV), also known as Rous sarcoma virus (RSV), and equine infectious anemia virus (EIAV) are naturally unmyristoylated, and binding to the PM is thought to occur mainly via electrostatic contacts (40 -48). Interestingly, it has been shown that EIAV Gag is detected both on the PM and in compartments enriched in phosphatidylinositol 3,5-bisphosphate (PI(3,5)P 2 ) (49). Treatment of cells with a kinase inhibitor that blocks production of PI(3,5)P 2 from phosphatidylinositol 3-phosphate (PI(3)P) caused Gag to co-localize with aberrant compartments and inhibited virus-like particle (VLP) release (49). NMR studies have shown that EIAV MA bound PI(3)P with higher affinity than PI(4,5)P 2 (49). In contrast to HIV-1, it has been shown that release of EIAV VLPs was not significantly diminished by depleting the PI(4,5)P 2 pool from the PM (49) and that assembly of EIAV particles occurred on interior cellular membranes (49,50). In vitro studies revealed that EIAV Gag interacts with electrically neutral membranes as well as negatively charged membranes, suggesting that other factors may also contribute to binding (40).
Pioneering studies have shown that neutralization of basic residues in the MA 87 domain led to severe reduction in ASV Gag assembly on the PM and subsequent virus release (42). In vitro flotation studies of ASV MA binding to liposomes have indicated that interaction is electrostatic in nature and depends upon the presence of a biologically relevant concentration of negatively charged lipids (44). The role of PI(4,5)P 2 in ASV Gag interactions with membranes has also been examined, but the findings of several studies are not in agreement (35,43,51). Chan et al. (43) have provided evidence that neither membrane localization of ASV Gag nor release of VLPs was affected by phosphatase-mediated depletion of PI(4,5)P 2 from the PM in transfected avian cells. In liposome flotation assays, ASV Gag showed no specificity to PI(4,5)P 2 , although it required acidic lipids for binding (43). On the other hand, Parent and co-workers (51) have shown that depletion of intracellular PI(4,5)P 2 and phosphatidylinositol 3,4,5-trisphosphate (PI(3,4,5)P 3 ) levels impaired ASV Gag PM localization. It was suggested that differences in experimental design may have led to these discrepancies (51). Later, it was shown that in vitro membrane binding of ASV Gag leucine zipper (GagLZ) is both PI(4,5)P 2dependent and susceptible to RNA-mediated inhibition (35). The PM-specific localization and VLP release of these GagLZ proteins are severely impaired by overexpression of 5ptaseIV. At physiological ionic strength, it was found that ASV Gag binds strongly to liposomes containing acidic lipids. This interaction is enhanced by physiological levels of PI(4,5)P 2 and by cholesterol (Chol) (47). By employing giant unilamellar vesicles as membrane models, it has been shown that membrane association of ASV Gag followed the overall membrane charge, independent of membrane order, confirming that electrostatic interactions are the primary force for ASV Gag-membrane association (47).
In summary, the mechanisms of retroviral Gag assembly appear to be complex and require more detailed investigation at the molecular level. Our current understanding of ASV Gag-PM interaction is incomplete because of the lack of molecular details on how membrane components contribute to the overall binding. Herein, we employed NMR, isothermal titration calorimetry (ITC), and computational approaches to investigate how ASV MA interacts with lipid mimetics and liposomes.

Structure determination of ASV MA
ASV MA is a 155-residue domain; however, only the first 87 residues (MA 87 domain) are indispensable for virion budding and infectivity (52). Therefore, previous biochemical and NMR studies have been conducted with the MA 87 domain (53). Although the structure of MA 87 has been determined previously by NMR methods (PDB code 1A6S) (53), the chemical shift assignments are not available in the Biomolecular Magnetic Resonance Bank. To be able to characterize MA 87 interactions with lipids by NMR spectroscopy, we first generated nearly complete NMR assignments for backbone and sidechain resonances using a standard set of triple-resonance, TOCSY, and NOESY experiments. In the course of the assignment process, we identified numerous unambiguous NOEs, of which some are shown in Fig. S1, that were inconsistent with the existing structure of ASV MA 87 . Given the remarkable inconsistency and relatively low quality of the structure, as judged from its PROCHECK scores (data not shown), we used signal assignments together with 15 N-resolved NOESY-HSQC and 13 C-resolved HMQC-NOESY spectra to determine the structure of MA 87 de novo in Unio'10. The final ensemble of 20 structures is based on 1628 distance restraints (18.7 restraints/ residue), from which more than half were medium-and longrange interactions (Table S1). Superposition of the 20 lowest-

Characterization of ASV matrix-membrane interactions
penalty MA 87 structures is shown in Fig. S2 (also see Table S1). The MA 87 protein consists of four ␣-helices and a 3 10 -helix. Whereas the original and new models share the general topology of ␣-helices, which is a canonical fold of retroviral matrix proteins, (54) significant differences in interhelical angles, points of contacts of ␣-helices, and structures of loops are noticeable between the two structures ( Fig. 1). Furthermore, significant differences in orientations and packing of numerous side chains are observed (Fig. S1). The positional root mean square deviation (RMSD) of backbone N, C ␣ , and CЈ atoms is 4.1 Å (residues 3-85), illustrating the nontrivial extent of differences between the two structures. In summary, herein we provide a significantly improved NMR structure of the ASV MA 87 protein that we believe is important for a detailed characterization of its interaction with lipids and membrane mimetics.
During the preparation of this manuscript, Doktorova et al. (48) reported the X-ray structure of full-length ASV MA (155 amino acids) at 2.8 Å resolution. In this structure, residues 1-102 were resolved, but there was no interpretable electron density for residues 103-155, consistent with an unstructured conformation. A truncated variant (residues 2-102) crystallized in the same form as the full-length molecule (space group I4122) and also in an alternate form (space group I41); structures were determined at 3.2 and 1.8 Å resolution, respectively. For crystals in the space group I4122, the C-terminal helix (residues 89 -102) interdigitates with the corresponding helix from a neighboring molecule, forming a symmetric dimer (48). In contrast, in crystals with space group I41, the C terminus is disordered, and neither helix 6 nor the dimer are observed. It was concluded that the crystallographic dimer is not biologically relevant because deletions in the C-terminal region of MA (amino acids 87-155) have no effect on viral budding and infectivity (52) and because there is no evidence that MA dimerizes in solution (46) or when membrane-associated (45). An overlay of residues 2-87 revealed that the X-ray structure (PDB code 5KZ9) is almost identical to our NMR structure (Fig. S3), further demonstrating that our NMR model represents an accurate and physiologically relevant form of the membrane binding domain of ASV MA.

PI(4,5)P 2 binding to ASV MA 87
Because the addition of native PI(4,5)P 2 to MA proteins often led to severe broadening in the 1 H-15 N HSQC NMR spectra (12,37), our first attempt to characterize binding properties of PI(4,5)P 2 to MA 87 involved titration with soluble analogs (see Fig. S4 for chemical structures of lipids used in this study). Binding studies were first conducted with dibutanoyl PI(4,5)P 2 (diC 4 -PI(4,5)P 2 ). As shown in Fig. 2, a subset of backbone 1 H and 15 N resonances exhibited significant chemical shift perturbations (CSPs) upon titration of MA 87 with diC 4 -PI(4,5)P 2 . The observed changes in chemical shifts in response to increasing concentrations of diC 4 -PI(4,5)P 2 indicate that the free and bound forms are in fast exchange on the NMR time scale. Signals that exhibited the most substantial CSPs correspond to residues located in ␣-helix 1 (Ala 3 , Val 4 , Ile 5 , Lys 6 , Val 7 , Ser 9 , Ser 10 , and Lys 13 ) and Lys 23 at the tip of ␣-helix 2 (Fig. 2, A and  B). The MA 87 structure revealed that Lys 6 , Lys 13 , and Lys 23 form a well-defined basic patch, flanked by Lys 18 and Lys 24 (Fig.  2C). Because the binding of lipid most likely involves direct interactions with the affected basic residues, it is reasonable to suggest that the largest CSPs observed for residues at the beginning of ␣-helix 1 are induced either by the lipid acyl chains or indirectly by conformational changes upon lipid binding to the basic patch. The dissociation constant (K d ) was determined by fitting chemical shift changes as a function of diC 4 -PI(4,5)P 2 concentration, yielding a value of 0.85 Ϯ 0.2 mM (in the absence of NaCl) based on averaging residue-specific individual values. The affinity of diC 4 -PI(4,5)P 2 to MA 87 is ϳ5-fold weaker than those previously observed for HIV-1 and HIV-2 MA proteins (K d ϳ0.15 mM) (12,37). These results demonstrate that diC 4 -PI(4,5)P 2 binds directly to MA 87 , albeit with a weaker affinity than observed for other retroviral MA proteins.

Role of lipid acyl chains and polar head in MA 87 binding
In previous NMR studies of HIV-1, HIV-2, and MPMV MA proteins (12,28,37) and surface plasmon resonance studies of HIV-1 Gag and MA proteins (55), it was reported that the acyl chains of phosphoinositides interact with the MA protein and that the affinity is increased upon increasing the length of the acyl chain. The involvement of acyl chains in MA binding in the context of native lipids and membrane liposomes has recently been evaluated (39). As shown above, several of the residues affected upon binding of diC 4 -PI(4,5)P 2 are hydrophobic (Ala 3 , Val 4 , Ile 5 , and Val 7 ). Their CSPs are either due to a direct contact with the acyl chains or are a result of allosteric conformational changes involving ␣-helix 1. To assess whether the acyl chains of PI(4,5)P 2 are involved in MA 87 binding, we have conducted NMR titrations with dihexanoyl PI(4,5)P 2 (diC 6 -PI(4,5)P 2 ). Titration of MA 87 with diC 6 -PI(4,5)P 2 resulted in chemical shift changes that are similar to those observed for diC 4 -PI(4,5)P 2 (Fig. S5). However, the binding affinity is 4-fold weaker than that of diC 4 -PI(4,5)P 2 ( Table 1), suggesting an opposing effect of the acyl chains. These results suggest that, in contrast to what was observed for HIV-1, HIV-2, and MPMV MA, increasing the length of PI(4,5)P 2 acyl chains seems to be unfavorable for MA 87 binding. To examine the role of the polar head of PI(4,5)P 2 in MA 87 binding, we conducted NMR titrations with inositol 1,4,5trisphosphate (I(1,4,5)P 3 ). As shown in Fig. S6, titration of I(1,4,5)P 3 into MA 87 led to CSPs that are very similar to those observed upon binding of diC 4 -PI(4,5)P 2 , which indicates that the polar head is sufficient for binding. The titration data afforded a K d value of 0.68 mM, which is also similar to that obtained for diC 4 -PI(4,5)P 2 (Table 1). Altogether, these results indicate that PI(4,5)P 2 interacts with MA 87 via the polar head.
To determine the specificity of binding, 1 H-15 N HSQC NMR titration experiments were conducted with inositol 1,3,5-trisphosphate (I(1,3,5)P 3 ), which differs from I(1,4,5)P 3 only in the placement of a single phosphate. I(1,3,5)P 3 is the polar head of PI(3,5)P 2 found on late endocytic compartments (56). Our NMR titration data show that I(1,3,5)P 3 binds to MA 87 in a manner that is very similar to I(1,4,5)P 3 but with a 2-fold weaker affinity ( Fig. S6 and Table 1), indicating that positioning of the phosphate group on the fourth carbon is of minimal significance for binding.

ASV MA 87 binding to PIPs and IPs is governed by charge-charge interactions
The results above indicate that the acyl chains do not contribute to productive phosphoinositide binding to MA 87 . To demonstrate that phosphoinositide-MA 87 interaction is governed by electrostatic forces, we conducted HSQC NMR titration experiments using phosphoinositides and IPs with varying numbers of phosphate groups. Phosphoinositides containing only one phosphate group on the third, fourth, or fifth carbon exhibited no detectable binding (data not shown). On the other hand, inositol 1,3,4,5-tetrakisphosphate (IP 4 ) and inositol hexakisphosphate (IP 6 ) bound MA 87 with substantially increased affinities (Table 1). Whereas the residues affected by titration with IP 4 and IP 6 are identical to those affected by diC 4 -PI(4,5)P 2 and I(1,4,5)P 3 , the CSPs observed upon titration of IP 6 are more dramatic than with any other PIP and IP tested ( Fig. 3 and Fig. S6). At 100 mM salt concentration, the affinity of IP 6  15 N chemical shift changes versus residue number calculated from the HSQC spectra for MA 87 upon titration with diC 4 -PI(4,5)P 2 . C, cartoon representation and electrostatic map of the MA 87 structure highlighting basic residues (blue) that exhibited substantial chemical shift changes upon binding of diC 4 -PI(4,5)P 2 . Of note, Lys 18 and Lys 24 (cyan) are in close proximity to the perturbed residues but do not appear to be perturbed by lipid binding.

Table 1 Dissociation constants (mM) for lipid and IP binding to ASV MA 87
Titrations were conducted in the absence of NaCl except when noted. ND, not detectable.

Characterization of ASV matrix-membrane interactions
binding to MA 87 decreased by 10-fold (Table 1), further demonstrating the electrostatic nature of the interaction.

Thermodynamics of IP binding to MA 87
Our NMR data provided compelling evidence for direct binding of PIPs and IPs to MA 87 and afforded binding affinities. However, these experiments do not provide information on the stoichiometry and thermodynamics of binding. To determine stoichiometry (n), enthalpy change (⌬H°) and entropic term (T⌬S°), we performed ITC experiments with IP 6 and MA 87 . We used IP 6 because it binds to MA 87 in a manner that is essentially identical to all IPs tested but with a much higher affinity. By fitting the data using a binding model for a single set of identical sites (Fig. 4), the following parameters were obtained: K d ϭ 9.9 Ϯ 1.8 M, n ϭ 0.93 Ϯ 0.02, ⌬H°ϭ Ϫ3.3 Ϯ 0.1 kcal/mol, and T⌬S°ϭ 3.75 kcal/mol. The K d value is very similar to that obtained by fitting the NMR titration data. ITC data confirm that IP 6 binds to MA 87 in a 1:1 mode. According to the enthalpy change, binding of IP 6 to MA 87 is exothermic, further indicating that formation of the IP 6 -MA 87 complex is governed by ionic forces.

Structures of MA 87 bound to inositol phosphates
The NMR titration data show that all IPs tested bind to a basic patch formed by several lysine residues, including Lys 6 , Lys 13 , and Lys 23 . CSPs observed for the several N-terminal residues in ␣-helix 1 can be caused by binding of the ligand in their vicinity or are a result of induced conformational changes. To precisely identify the lipid binding site and to identify structural changes in MA 87 , if any, we obtained multidimensional NMR data on MA 87 bound to IP 6 . Again, we chose IP 6 because it binds to MA 87 in a manner that is essentially identical to other IPs but with a higher affinity. Interestingly, no new intraprotein

Characterization of ASV matrix-membrane interactions
NOEs were detected that would be indicative of a different protein conformation in the complex. More than 97% of the backbone and side-chain chemical shifts were assigned for the MA 87 -IP 6 complex, and the structure of complexed MA 87 was determined in the same way as that of the free MA 87 (Table S2). The structure of MA 87 in the MA 87 -IP 6 complex is nearly identical to that of the free MA 87 protein (Fig. S7), demonstrating that binding of IPs does not induce any major conformational changes in the MA 87 protein. Based on these observations, it is conceivable that the large CSPs observed for residues 3-5 are caused by minor adjustments at the tip of the helix.
Several intermolecular NOEs were detected between the inositol H4 and H6 atoms and the side chains of Lys 13 and Lys 23 using 13 C-filtered NOESY. The structure of the MA 87 -IP 6 complex was determined using HADDOCK based on the unambiguous intermolecular NOEs and ambiguous interaction restraints derived for residues that showed significant 1 H-15 N CSPs (for more details, see "Experimental procedures"). Structure calculations of MA 87 -IP 6 yielded 200 final structures that were clustered based on the positional RMSD cut-off of 1 Å at the interaction interface (Fig. S8). A single cluster of 193 structures was obtained, revealing the preferred binding orientation of IP 6 (Table S2). In a representative model of the MA 87 -IP 6 complex, the binding site is clearly defined by the side chains of Lys 6 , Lys 13 , Lys 23 , and Lys 24 (Fig. 5). At least three phosphate groups of IP 6 are poised to form salt bridges with the lysine amine groups.
Similar HADDOCK calculations were performed for MA 87 -I(1,4,5)P 3 and MA 87 -I(1,3,5)P 3 that were based on the observed CSPs. A single cluster of 160 structures was obtained for MA 87 -I(1,4,5)P 3 ( Fig. S8 and Table S2), in which the 4-and 5-phosphate groups interact with the amine groups of Lys 23 and Lys 24 , respectively, whereas the 1-phosphate group is sandwiched between the amine groups of Lys 6 and Lys 13 (Fig. 5). On the other hand, for the MA 87 -I(1,3,5)P 3 complex, two clusters have been obtained (Fig. S8 and Table S2) in which the structures differ in the ring orientation (ring flipped by 180°). In the larger cluster of the two, the 3-phosphate group is sandwiched between Lys 6 and Lys 23 side chains, whereas the 5-phosphate group interacts with Lys 24 and the 1-phosphate group with Lys 13 (Fig. 5). Interestingly, although the Lys 24 amide signal exhibits only negligible chemical shift changes upon IPs binding, and thus no restraints involving this residue have been included in the HADDOCK calculations, the side chain of Lys 24 appears to be well positioned and flexible enough to reach out to the phosphate groups and become part of the binding surface. In summary, our docking studies revealed a preferred binding mode for the IPs in which all of them bind the same MA 87 residues in remarkably similar orientations.

Interactions of PC and PS with MA 87
We have previously shown that PC, PS, and PE lipids with truncated acyl chains are capable of binding to HIV-1 MA via a distinct site that is adjacent to the PI(4,5)P 2 binding site (38). Herein, we conducted NMR titration studies to characterize binding properties of diC 6 -PC and diC 6 -PS to MA 87 . As shown in Fig. S9, numerous signals exhibited CSPs upon the addition of diC 6 -PC to 15 N-labeled MA 87 . Three distinct regions appear to be largely affected by diC 6 -PC binding (residues 2-20, 45-47, and 72-86). Mapping the CSPs on the MA 87 structure revealed that the perturbed residues form an opening to a hydrophobic pocket surrounded by ␣-helices 1, 3, and 5 (Fig.  S9). Likewise, titration of diC 6 -PS to 15 N-labeled MA 87 induced spectral changes in a subset of signals similar to those observed upon diC 6 -PC binding. However, compared with diC 6 -PC, residues located in the first ␣-helix and the following loop were affected to a significantly larger degree. It appears that, while being restrained by the acyl chain interaction with MA 87 hydrophobic surface, the negatively charged diC 6 -PS polar head transiently binds to the basic patch, suggesting a synergy between PS and PI(4,5)P 2 that enhances affinity of MA 87 to membrane.

Identification of MA 87 residues critical for binding to inositol phosphates
The results above revealed a well-defined IP-binding basic surface consisting of Lys 6 , Lys 13 , Lys 23 , and Lys 24 . To probe residues critical for interaction, we generated single, double, and triple lysine-to-glutamate mutants and conducted NMR titrations with IP 6 and I(1,4,5)P 3 . Spectra of all mutants consisted of sharp signals, and only residues in the vicinity of the mutation sites exhibited changes in signal positions compared with the WT MA 87 , indicating that these mutations did not alter the structure of the protein. As shown in Table 2, binding of IP 6 to MA 87 K23E, K6E/K13E, and K18E/K23E mutants was greatly reduced (by ϳ100-fold), and binding of I(1,4,5)P 3 was undetectable. For K6E, K13E, K24E, and K18E/K24E, the IP 6 affinity decreased by ϳ10 -20-fold, whereas the affinity of I(1,4,5)P 3 decreased by severalfold for the single mutants and

Characterization of ASV matrix-membrane interactions
was undetectable for the double mutant. Binding of IP 6 and I(1,4,5)P 3 to the MA 87 K6E/K13E/K23E mutant was undetectable. On the other hand, the binding affinity was either unchanged or minimally decreased for K18E and K35E mutants, consistent with their exclusion from the IP binding site. These results further validate the identity of residues composing the IP binding site.

Interaction of MA 87 with liposomes
The interplay and/or synergy between membrane components is thought to influence affinity to proteins. By employing an in vitro liposome-pelleting assay, Vogt and co-workers (46 -48) have shown that increasing amounts of PS and inclusion of Chol enhanced ASV Gag binding to membranes. Of note, the most recent experiments were conducted at varying salt concentrations, with the best binding achieved at 50 mM NaCl (48).
In contrast to full-length Gag, the purified ASV MA protein bound only weakly to liposomes even in the presence of PI(4,5)P 2 (46 -48). Herein, we employed a sensitive NMRbased assay to characterize binding of MA 87 to liposomes containing native lipids. This NMR approach, which has been employed recently to characterize HIV-1 MA binding to liposomes (39), allows for measurement of the unbound protein population in solution under equilibrium conditions with the liposome-bound state (57,58). One-dimensional 1 H NMR experiments were conducted on samples with fixed protein and total lipid concentrations, while varying lipid composition within large unilamellar vesicles (LUVs) (Fig. 6A). No detectable binding was observed between MA 87 and LUVs containing only 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC).
Next, we conducted NMR titrations with LUVs containing POPC and increasing amounts of 1-palmitoyl-2-oleoyl-snglycero-3-phospho-L-serine (POPS). Binding was only detectable upon incorporation of Ն30% mol POPS (Fig. 6B). Fitting the binding isotherms with the Hill equation (Fig. 6C) yielded a microscopic K d of 186 M for POPS. Binding was highly cooperative, as indicated by the high n value (Fig. 6D), suggesting that MA 87 engages multiple PS molecules for an efficient membrane interaction. Titrations conducted with LUVs containing POPC and increasing amounts of PI(4,5)P 2 revealed a similar degree of cooperativity but 5 times stronger binding compared with POPS (K d ϭ 37 M; Fig. 6C), demonstrating that affinity of MA 87 to membrane is enhanced by PI(4,5)P 2 . Titration of LUVs containing POPC, fixed 20% mol POPS, and increasing amounts of PI(4,5)P 2 led to substantial NMR signal dampening

and I(1,4,5)P 3 binding to ASV MA 87 mutants
Titrations were conducted in the absence of NaCl. ND, not detectable.

Characterization of ASV matrix-membrane interactions
showing a synergistic effect of both acidic lipids in MA 87 binding to membrane (Fig. 6, B and D). We also compared the binding of MA 87 to POPC/POPS liposomes incorporating differently phosphorylated phosphatidylinositols. Binding of dipalmitoyl-phosphatidylinositol 3,5-bisphosphate (diC 16 -PI(3,5)P 2 )-containing liposomes yielded a microscopic K d that is slightly weaker than that for dipalmitoyl-PI(4,5)P 2 (diC 16 -PI(4,5)P 2 )-containing LUVs (Fig. 6, C and D). Altogether, these findings indicate that MA 87 binding to LUVs is sensitive to PI(4,5)P 2 , is significantly enhanced by inclusion of PS, and is not strongly dependent on the positioning of the phosphate group (fourth versus third carbon). Finally, to further validate the IP-binding site in the context of binding to membrane, we have employed an NMR-based assay to assess binding of LUVs to ASV MA 87 mutants described above. As shown in Fig. 7, the most significant reduction in binding of MA 87 to POPC/POPS/PI(4,5)P 2 (65:30:5) was observed for K23E, K24E, K6E/K13E, and K18E/K24E mutants. Binding was also compromised, albeit more moderately, for K6E, K13E, and K18E mutants. On the other hand, MA 87 K35E mutant bound to LUVs in a manner similar to the WT protein.
Notably, in contrast to the more conservative reduction in binding to I(1,4,5)P 3 observed in inositol-binding experiments, the K24E mutant displayed a highly attenuated signal in the LUV-binding experiments. This difference may indicate that Lys-24 plays a role in additional interactions with acidic membrane lipids near the primary PI(4,5)P 2 -binding site. Altogether, these results confirm that MA 87 binding to membrane is mediated by a basic patch formed by Lys 6 , Lys 13 , Lys 23 , Lys 24 , and, to a lesser extent, Lys 18 .

Discussion
Despite the great similarity in Gag structural domains and their functions, retroviruses have evolved distinct mechanisms for assembly that are incompletely understood. Because the ASV Gag protein is naturally unmyristoylated and because the structure of its MA domain has a canonical fold of retroviral MA proteins (54), it is often used as a comparative model for HIV-1. Previous studies have shed light on factors that regulate ASV MA and Gag binding to lipid membranes (40 -48), supporting a model in which ASV Gag assembly on the PM is driven by electrostatic interactions (42,44,47). However, recent findings on the role of PI(4,5)P 2 in ASV Gag targeting and localization on the PM are not in agreement (35,43,51).
In this study, we elucidated the molecular determinants of ASV MA interaction with lipids and membranes and provided a model for MA-membrane binding. We assessed the role of various elements, such as acyl chain, polar head, overall charge, lipid specificity, and membrane composition. Additionally, we complemented and validated our structural studies with in vivo assays as shown in an accompanying paper (79). Several important points have emerged from this study. (i) We obtained a significantly improved structural model of ASV MA 87 . Our structure revealed a basic surface that proved to be important for membrane binding (Fig. 8). This membrane-binding site was not obvious in the previously determined NMR structure (53). (ii) We provided compelling evidence for a direct interaction between MA 87 and PI(4,5)P 2 via a basic surface consisting of Lys 6 , Lys 13 , Lys 23 , and Lys 24 (Fig. 8). (iii) We found that the  Favorable electrostatic interactions occur between the acidic polar head of PI(4,5)P 2 and a basic patch formed by Lys 6 , Lys 13 , and Lys 23 (blue). HADDOCK calculations suggest a potential membrane interaction with Lys 24 due to the close proximity of the side chain to the inositol ring. Membrane bilayer was generated in the VMD membrane builder plug-in (78). PI(4,5)P 2 was generated in Avogadro (75). B, cartoon representation of the ASV and HIV-1 MA structures (PDB entries 6CCJ and 2H3I, respectively), highlighting the similar structural motifs and basic residues involved in membrane binding (blue sticks). The myr group, residues 2-3, and residues 115-132 of HIV-1 MA are not shown for clarity.

Characterization of ASV matrix-membrane interactions
acyl chains of PI(4,5)P 2 are not involved in binding to MA 87 , as binding occurs exclusively via the polar head. (iv) We have shown that binding of IPs does not induce any structural changes in the MA 87 protein.
(v) We found that PI(3,5)P 2 binds to MA 87 with only slightly lower affinity than PI(4,5)P 2 . (vi) We found that MA 87 affinity to liposomes is enhanced by PS and PI(4,5)P 2 incorporation, indicating that binding of MA 87 to membrane is exclusively electrostatic. ASV MA 87 binding to LUV was only detectable upon incorporation of Ն30% POPS (Fig. 6, B and D). Interestingly, the MA fraction bound to LUV at 7% PI(4,5)P 2 is approximately equal to that at 50% POPS. Given that the negative charge is Ϫ1 for PS and Ϫ5 for PI(4,5)P 2 , the total negative charge at 50% PS is similar to 20% PS ϩ 7% PI(4,5)P 2 , which further indicates that ASV MA binding to membrane is governed by charge-charge interactions.
The finding that the phosphoinositide acyl chains are not involved in binding is interesting because previous structural and biophysical studies of retroviral (HIV-1, HIV-2, and MPMV) MA interaction with tr-PI(4,5)P 2 indicated a role of acyl chains in binding (12,28,37,55). We have shown previously that binding of tr-PI(4,5)P 2 to HIV-1 MA induced a conformational change that triggered myr exposure (37). The structure of MA bound to tr-PI(4,5)P 2 revealed that both the polar head and the truncated 2Ј-acyl chain of the lipid are involved in MA binding and that increasing the length of tr-PI(4,5)P 2 acyl chains also increased the lipid affinity to MA (37). These findings were recapitulated for HIV-2 MA (12). On the contrary, our current data show a weakening effect of longer acyl chains on binding of tr-PI(4,5)P 2 to ASV MA 87 . We show that, unlike HIV-1 MA, MA 87 possesses a hydrophobic pocket that is well separated from the basic residues at the MA N terminus that bind the PI(4,5)P 2 polar head (Fig. S9). Therefore, the weakening effect of longer acyls on tr-PI(4,5)P 2 binding to ASV MA can be explained by competition between two independent binding events (i.e. binding of the lipid polar head versus acyl chains). The differential roles of acyl chains may also explain the observed 5-fold difference in affinities to diC 4 -PI(4,5)P 2 between ASV MA 87 and HIV MAs.
Structural studies of MPMV MA bound to diC 8 -PI(4,5)P 2 revealed that the acyl chains penetrate deeply into a hydrophobic pocket (11,28). However, the position of the myr group of MPMV MA was not affected by lipid binding, as it remained sequestered in the protein core. Interestingly, no binding was observed between the unmyristoylated MA protein of MPMV and soluble PI(4,5)P 2 analogs (28), which was attributed to the structural differences between MA and unmyristoylated MA. More recently, studies conducted with native PI(4,5)P 2 ruled out a direct contact between the acyl chains and the HIV-1 MA protein. It was suggested that binding is mainly governed by electrostatic forces (39).
A much higher binding affinity of IP 6 to MA 87 compared with I(1,4,5)P 3 allowed for structural determination of the complex, which revealed the interaction interface. Interestingly, our data showed that NMR signals corresponding to Lys 24 were unaffected upon titration of MA 87 with any of the tested lipids and IPs, including I(1,4,5)P 3 and IP 6 . However, HADDOCK structural analysis shows that the side chain of Lys 24 is flexible enough and is capable of reaching to the IP phosphate groups to form a salt bridge. Indeed, binding of I(1,4,5)P 3 , IP 6 and liposomes to MA 87 is significantly reduced for the K24E mutant ( Fig. 7 and Table 2), suggesting that Lys 24 plays a role in lipid binding. It is conceivable that the proximity of Lys 24 to Lys 23 increases the local positive electrostatic surface potential, thereby enhancing binding affinity. Substitution of Lys 24 to glutamate decreases the positive electrostatic surface potential and hence affinity to acidic lipids. Of note, previous studies have shown that mutations of Lys 6 , Lys 13 , Lys 23 , and Lys 24 reduced virus release by ϳ70%. Gag proteins with double mutations of these residues blocked localization of Gag on the PM and led to nearly complete abrogation of virus release (42). In an accompanying paper (79), we show that disruption of the IP-binding site inhibits Gag localization at the cell periphery in DF-1 and COS-1 cell lines and severely reduces particle production. These studies are fully consistent with the structural findings as represented in our MA-membrane model (Fig. 8A).
Neutron scattering experiments of ASV MA bound to liposomes containing Chol revealed preferential binding with preexisting POPS-rich clusters formed by nonideal lipid mixing. It was reported that binding occurs via the lipid headgroups with minimal perturbation to the bilayer structure (48). Molecular dynamics simulations revealed a stronger MA-bilayer interaction in the presence of Chol and that the enhanced lipid packing and membrane surface charge density are largely driven by Chol (48). Molecular dynamics simulations suggested two models for ASV MA membrane interaction. In the Chol-free model, MA interaction is mediated mainly by Lys 23 and Lys 24 . However, in the Chol-enriched membrane, several other basic residues, such as Lys 13 , Lys 18 , Lys 67 , and Lys 72 , also become attracted to the PS cluster. It was concluded that Chol enhances ASV MA-membrane association by increasing negative charge density on the bilayer surface due to increased lipid packing. Our structural data clearly show a unique and continuous basic surface with high potential that is capable of binding efficiently to a highly acidic surface (Figs. 2 and 8). Our data obtained with lipids and LUVs lacking cholesterol indicate that Lys 6 , Lys 13 , Lys 23 , and Lys 24 are key determinants of ASV MA-membrane interaction.
The role of PI(4,5)P 2 in ASV Gag targeting and localization on the PM membrane has been investigated independently by three research groups (35,43,51). However, their findings were not in agreement on whether PI(4,5)P 2 was absolutely required for Gag binding to the PM. A study by Chan et al. (43) indicated that neither localization of ASV Gag nor release of VLPs was affected by phosphatase-mediated depletion of PI(4,5)P 2 from the PM in transfected avian cells. Liposome flotation assays of ASV Gag showed no specificity to PI(4,5)P 2 , although acidic lipids were required for binding (43). On the other hand, a study by Parent and co-workers (51) has shown that depletion of intracellular PI(4,5)P 2 and PI(3,4,5)P 3 levels impaired ASV Gag PM localization. It was suggested that these discrepancies may stem from differences in experimental design (51). A subsequent study by Ono and co-workers (35) has shown that the PM localization and VLP release of ASV Gag leucine zipper (GagLZ) proteins are severely impaired by overexpression of 5ptaseIV. Here, we have shown a slight preferential binding of PI(4,5)P 2 to MA 87 compared with PI(3,5)P 2 (1.5-2-fold), indicating that the position of the phosphate groups is not a key

Characterization of ASV matrix-membrane interactions
determinant for binding to ASV MA. Interestingly, this result is similar to that obtained for HIV-1 MA, in which the affinity to PI(3,5)P 2 was found to be ϳ2-fold weaker than that of PI(4,5)P 2 (39). It has been suggested that because PI(3,5)P 2 abundance in cells is significantly lower than that of PI(4,5)P 2 (ϳ100-fold lower (56)), PM targeting of HIV-1 Gag results from the high relative concentration of PI(4,5)P 2 rather than differences in affinity of MA for these phosphoinositides (39). It is also conceivable that the small difference in the affinities of MA to PI(4,5)P 2 versus PI(3,5)P 2 becomes exponentially enhanced when multimeric Gag engages in numerous PI(4,5)P 2 interactions on the PM. In the accompanying paper (79), to test the importance of PI(4,5)P 2 -targeted localization, we employed Sprouty2 (Spry2) and the pleckstrin homology domain of phospholipase C␦ (PH-PLC) proteins that localize to membranes enriched in PI(4,5)P 2 . Spry2 has been shown to inhibit localization of HIV-1 Gag on the PM and reduce VLP release (59). We show that co-expression of Spry2 or PH-PLC affects ASV Gag trafficking to the PM and budding.
Because retroviral MA proteins share similar structural features and because ASV MA is often used as a comparative model for HIV-1 MA studies, we analyzed the similarities of ASV and HIV-1 MA structures and their membrane-interacting motifs. Our NMR structure afforded a better structural alignment with the HIV-1 MA protein. As shown in Fig. 8B, both proteins are strikingly similar, and more so, they clearly share very similar membrane-interacting regions. A plethora of structural, biochemical, biophysical, in vitro, and in vivo assays support a model in which Arg 22 , Lys 26 , Lys 27 , Lys 30 , and Lys 32 of HIV-1 MA form a membrane-interacting surface (8,16,17,37,39,60,61). For both proteins, the membrane-interacting residues reside on similar motifs (␣-helices 1 and 2 and the loop connecting them; Fig.  8B). Hence, these findings indicate that despite the lack of an N-terminal myristoyl group, ASV MA utilizes a membrane-interacting motif very similar to that of HIV-1 MA, which proved to be sufficient for efficient Gag-membrane binding.
In summary, we demonstrated that ASV MA binding to membranes is governed by electrostatic interactions and that affinity of MA binding to membrane is enhanced by acidic lipids, such as phosphoinositides and PS. We have shown that ASV MA contains a single IP-binding site formed by Lys residues that are also implicated in PM localization of Gag and virus release. These findings provide a molecular basis for ASV Gag binding to the inner leaflet of the PM and advance our understanding of the basic mechanisms of retroviral assembly.

Sample preparation
A plasmid encoding full-length ASV Gag (Prague C isolate; amino acids 1-701) was used as a DNA template. The MA gene encoding for amino acids 1-87 was inserted into pET28 vector at its XhoI and BamHI sites, yielding a construct that is fused to a His 6 -SUMO tag gene on the 5Ј-end. The His 6 -SUMO-MA 87 protein was overexpressed in Escherichia coli BL21-Codon-Plus-RIL cells (Agilent Technologies). A starter culture was prepared by inoculating 20 ml of Luria-Bertani broth medium containing 60 mg/liter kanamycin with the transformed cells and grown overnight at 37°C (220 rpm). The next day, cells were spun down, and the pellet was resuspended in 1 liter of Luria-Bertani broth medium. Cells were grown at 37°C (250 rpm) until A 600 reached 0.6 -0.8, induced with 1 mM isopropyl ␤-D-1-thiogalactoside, and grown for 4 h. Cells were harvested by centrifugation and stored at Ϫ80°C or used directly for protein purification. Cell pellet was resuspended in a lysis buffer containing 50 mM sodium phosphates (pH 7.4), 300 mM NaCl, and 2 mM ␤-mercaptoethanol (␤ME). Cells were sonicated, and cell lysate was spun down at 17,000 rpm for 30 min. The protein-containing supernatant was purified on cobalt affinity resin (Thermo Fisher Scientific). The His 6 -SUMO-MA 87 protein was eluted via a gradient using a buffer containing 50 mM sodium phosphate (pH 7.4), 300 mM NaCl, 300 mM imidazole, and 2 mM ␤ME. Fractions containing the MA 87 protein were pooled, mixed with SUMO protease (1:1000 protease/protein), and dialyzed overnight in 50 mM sodium phosphates (pH 7.8), 150 mM NaCl, and 2 mM ␤ME. His 6 -SUMO was separated from MA 87 with fresh cobalt resin, and fractions containing the protein were concentrated using 3-kDa cut-off Amicon filters. Uniformly 15 N-and 15 N, 13 C-labeled MA 87 samples were prepared by growing cells in M9 minimal medium containing glucose-13 C 6 and 15 NH 4 Cl. Protein purification was performed as described above.

Preparation of LUVs
POPC, POPS, porcine brain PI(4,5)P 2 (Avanti Polar Lipids), diC 16 -PI(4,5)P 2 , and diC 16 -PI(3,5)P 2 (Echelon Biosciences) were used as received. Lipid concentrations in stocks were determined by a molybdate phosphorus assay (62). Lipids were mixed in appropriate ratios, and solvent was evaporated under a stream of air and further under high vacuum for at least 2 h. Lipids were then resuspended in 50 mM sodium phosphates (pH 6.5), 50 mM NaCl, and 2 mM TCEP by repeated brief vortexing and passed 29 times through a 100-nm pore filter in an extruder (Avanti Polar Lipids). Final total lipid concentration in LUV stocks corrected for extruder dead volume was 4.7 mg/ml for titrations and 7.7 mg/ml for the MA 87 mutant-binding assay.

Characterization of ASV matrix-membrane interactions
iments (150-ms mixing time); signals were assigned after correcting chemical shifts of lysine 13 C ⑀ and 13 C ␦ atoms for the deuterium isotope effect (67). 1 H-15 N HSQC NMR titrations were conducted with 100 M samples of 15 N-labeled MA 87 in 50 mM sodium phosphate (pH 6.0) and 2 mM TCEP. Stock solutions of lipids and IPs were prepared in water at 10 -50 mM. The pH of IP 6 stock solution was adjusted to 6.0 by using NaOH prior to titrations. CSPs were calculated as follows,

Lipid NMR titrations
where ⌬␦ H and ⌬␦ N are 1 H and 15 N chemical shift changes, respectively. Dissociation constants were calculated by the nonlinear least-squares fitting algorithm in the gnuplot software using the equation, where ⌬␦ HN max is the chemical shift difference between complex and free protein, [L] 0 is total concentration of lipid, and [P] 0 is total concentration of protein.

LUV NMR titration
Individually prepared samples for NMR titration contained 50 M MA 87 in 50 mM sodium phosphates (pH 6.5), 50 mM NaCl, 2 mM TCEP, 188 g of LUVs with varying PS or phosphoinositide concentrations, and 4% D 2 O (v/v) in a total volume of 500 l. 1 H NMR spectra with excitation sculpting water suppression were recorded for each sample, and integral intensity was measured in the region 9.5-8.0 ppm. The amount of protein bound to LUVs was determined as the difference between integrals of samples without and with LUVs. The binding data (averages of three experiments) were fitted in Matlab 2015b (MathWorks) using the Hill equation, ϭ [L] n /(K d n ϩ [L] n ), where is the fraction of bound protein, [L] is the concentration of available lipid in LUV, K d is the microscopic dissociation constant, and n is the cooperativity constant.

Isothermal titration calorimetry
Thermodynamic parameters of IP 6 binding to MA 87 were determined using a MicroCal PEAQ-ITC system (Malvern Instruments). ITC experiments were conducted on protein samples in a buffer containing 25 mM MES (pH 6.5) and 2 mM TCEP. IP 6 prepared at 2.5 mM in the same buffer was titrated into 192 M MA 87 . Heat of reaction was measured at 35°C for 19 injections. Heat of dilution was measured by titrating IP 6 into buffer and was subtracted from the heat of binding. Data analysis was performed using PEAQ analysis software. The thermodynamic parameters were determined by fitting baseline-corrected data by a binding model for a single set of identical sites.

Structure calculations
Structure calculations were performed using Unio'10 software (68) that utilizes Atnos/Candid functionality for auto-mated iterative peak picking of raw NOESY spectra, peak assignments, and calibration, in conjunction with the CYANA structure calculation engine (69,70). Backbone and dihedral angle constraints were generated by Unio based on the chemical shifts. Default settings were used for calculations. NOESY spectra were converted to XEASY format using CARA (71). Visualization of structures was performed using PyMOL (version 1.5.0.2, Schrödinger, LLC, New York). Electrostatic potential maps were generated using PDB entry 2PQR and APBS software (72,73).

HADDOCK docking
The structures of MA 87 -IP 6 , MA 87 -I(1,4,5)P 3 , and MA 87 -I(1,3,5)P 3 complexes were calculated using HADDOCK software (74). The input structures of IP 6 (12 negative charges) and IP 3 (5 negative charges) were built and energy-minimized in Avogadro (75) using universal force field (76), and their CNS topology and parameters files were generated with ACPYPE (77). The structure of MA 87 bound to IP 6 was used as an input model. All calculations were based on identical ambiguous distance restraints that were derived for MA 87 residues displaying CSP larger than average ϩ 1 S.D. of all CSPs (Ala 3 , Val 4 , Ile 5 , Lys 6 , Ser 10 , Lys 13 , and Lys 23 ). Neighboring residues with Ͼ50% side-chain solvent accessibility (http://wolf.bms.umist.ac.uk/ naccess) 3 were then defined as passive residues (Glu 2 , Ala 3 , Ser 10 , Thr 14 , Lys 18 , Ser 22 , and Lys 24 ). Structure calculation of MA 87 -IP 6 included several additional unambiguous intermolecular NOE restraints. The calculations performed with settings recommended for protein-ligand docking yielded 200 final structures that were clustered based on the positional RMSD cut-off of 1 Å at the interaction interface. The 20 lowestenergy structures were deposited in PDB for each complex.