Synergy and allostery in ligand binding by HIV-1 Nef

Nef is an accessory protein that enhances the virulence and pathogenesis of human and simian immunodeficiency viruses (HIV and SIV, respectively) [1,2]. HIV viruses that lack functional Nef protein fail to cause acquired immunodeficiency syndrome in infected persons [3–6]. Nef has therefore been suggested to be a promising drug target.

Nef contributes to disease progression via its capacity to disrupt several functions of the host cell. Nef alters the protein composition of the host cell surface by down-regulating the transmembrane proteins involved in immune signalling and viral entry, including CD3, CD4, CD8, CD28, CXCR4, CCR5, SERINC3, SERINC5, and antigen-loaded MHC-II [reviewed in [7]]. Additionally, Nef up-regulates the surface expression of other factors, including TNF, DC-SIGN, and LIGHT [8,9]. Other functions of Nef include promoting T-cell activation [10,11] and lymphocyte chemotaxis [12]. Nef also disrupts intracellular signalling primarily through its interactions with cytoplasmic kinases, including members of the Src and Tec/Bt kinase families, PAK2, and PI3 kinase [13–18]. Collectively, these actions promote the immune evasion of infected cells and enhance viral release, spread, and infectivity.

Nef is a noncatalytic protein that causes these effects by interacting with a multitude of cellular proteins. The molecular bases for some of these interactions have been identified. Initial NMR and crystallographic studies have shown that Nef comprises a myristoylated flexible N-terminal arm (residues 1–70; we use the numbering of the LAI isolate unless stated otherwise), and a folded core domain (residues 71–206) that harbours a long flexible loop (residues 148–179) (Figure 1A) [19–23]. Both the core domain and flexible regions contain important ligand-binding sites [1,7].

Subsequent structural studies have revealed the molecular determinants of Nef's capacity to disrupt Src kinases by interacting closely with their SH3 domains [20,22,24]. In addition to its canonical ‘linear' proline-rich motif P₇₂xxPxR (where x is any amino acid), Nef uses an extended surface provided by its folded core to bind to Src family SH3 domains. This resulting ‘tertiary' binding mode increases the interaction surface along with affinity and selectivity for SH3 domains, compared with the canonical linear interaction between proline-rich motifs and SH3 domains [1,20,22]. One SH3 residue, located in the arginine (R) position of the arginine–threonine (RT) loop (residue number 96 in Fyn, a kinase from the Src family), was shown to primarily affect the binding affinity by interacting with a pocket on the core surface of Nef [25]. This pocket is hydrophobic in HIV-1 Nef, and more hydrophilic in SIV Nef. Consequently, HIV-1 Nef binds stronger to those kinases whose SH3 domain has a hydrophobic residue at the R position, whereas SIV Nef prefers charged or polar residues [20,26].

Additionally, the rigidity of the SH3 RT loop also influences the affinity to Nef [27]. The RT loop needs to bend to bind to Nef. Therefore, SH3 domains where the RT loop is stabilized by an extensive hydrogen-bond network (such as that seen in Fyn) need to pay a higher enthalpic cost for binding to Nef than SH3 domains with a more flexible RT loop (as present in Hck, another member of the Src family).

SH3 binding by Nef activates Hck by outcompeting those intramolecular interactions that maintain Hck in its catalytically inactive ‘closed’ conformation [28,29]. SH3-dependent activation by Nef has also been observed for Lyn and c-Src in a Saccharomyces cerevisiae assay [30], but not in an in vivo transformation assay [31]. The yeast assay also failed to show activation of Fgr, Fyn, Lck, or Yes by Nef. Hence, although Nef-targeted kinases have been shown to play roles in HIV replication, infectivity, MHC-1 down-regulation, and pro-inflammatory vesicle release [16,17,32,33], the molecular mechanism underlying these effects remains to be fully elucidated.

Structural studies have also revealed the mechanism through which Nef molecules hijack the host cell's trafficking machinery by serving as an adaptor between AP1 and MHC-I or between AP2 and CD4 [7,34–37]. Similar to its interactions with Src family SH3 domains, Nef uses its core domain along with the linear recognition motifs that are located on its flexible regions to firmly bind to AP1 or AP2. However, Nef binds AP1 differently than AP2. Nef binds to the µ1 subunit of AP1, and a part of the MHC-I cytoplasmic tail is sequestered in a groove formed between AP1 and Nef. In the case of AP2, Nef binds to the ɑ2, β2, and σ2 subunits of AP2. In contrast, Nef binds to the ɑ2, β2, and σ2 subunits of AP2. In this association, Nef uses its E₁₆₀xxxLL motif, located in the core domain loop, to mimic acidic dileucine motifs of cellular AP2 cargos. In parallel, Nef uses a core domain binding pocket to engage a part of the cytoplasmic CD4 tail [34,37], thus connecting CD4 to AP2. Despite these differences, the binding sites of CD4 and MHC-I on Nef show partial overlap [37].

Collectively, these studies show that Nef's strong polyvalent interactions result from a combination of the following two factors: (1) the multivalency that is intrinsic to flexible protein–protein interaction motifs, and (2) the high specificity and affinity resulting from the large binding surface offered by a 3D folded domain. Nef's composition of 50% tertiary structure and 50% flexible regions constitutes an optimal structural framework for this strategy. However, the resulting capacity of Nef to bind to a high number of cellular targets, raises the question of whether binding to one target affects Nef's associations with other targets. More specifically, has Nef evolved allosteric mechanisms for creating synergies between sets of ligands that contribute to the same cellular function while excluding others? In this study, we combined structural and functional studies to identify crossreactivity in selected intra- and intermolecular associations.

Fyn SH3 (Fyn_R96, Fyn_R96I, and Fyn_R96W) and Nef (SF2, LAI, ΔN, ΔNΔL, and WLΔN) constructs were expressed in Escherichia coli Bl21 (DE3) cells using a pGEX-2T expression vector containing an N-terminal GST tag with a thrombin cleavage site (Fyn_R96W; Nef WLΔN), a pET42a expression vector containing a C-terminal hexa-His tag (Fyn_R96 and Fyn_R96I), or a pET23d expression vector containing a C-terminal hexa-HIS tag (SF2). Nef ΔN, Nef ΔNΔL, and LAI were expressed using a pJEx411c expression vector (pJExpress411, originally from DNA 2.0, CA [38]) containing an N-terminal GST tag with 3C protease cleavage site. The bacterial cells were cultured at 37°C in 2xYT broth containing 100 μg/ml ampicillin (all Fyn, Nef WLΔN, and SF2) or 50 μg/ml kanamycin (Nef ΔN, Nef ΔNΔL, and LAI). When the optical density at 600 nm reached 0.6–0.8, protein expression was induced using 0.2 mM IPTG (Fyn constructs) or 0.5 mM IPTG (Nef constructs) and proceeded overnight at 18°C. The cells were then harvested and resuspended in 50 mM Tris (pH, 8.0), 200 mM NaCl, 2 mM EDTA (excluded in His tag purification), 1 mM DTT, one tablet of EDTA-free protease inhibitor (Roche) per 50 ml of resuspension, 1% triton, and 0.4 mg/ml lysozyme. Cells were lysed via sonication and pelleted by centrifugation at 87 207×g for 45 min at 4°C.

All proteins were purified using HIS-affinity columns (Fyn_R96, Fyn_R96I, and SF2), or GST-affinity columns (Fyn_R96W and LAI, WLΔN, ΔN, ΔNΔL) followed by mono Q 10/100 GL (GE Healthcare) anion exchange chromatography and HiLoad Superdex75 16/60 (GE Healthcare) size-exclusion chromatography . Protein purity was analyzed using SDS–PAGE gels. The buffer used were 50 mM Tris (pH, 8.0), 200 mM NaCl, 2 mM EDTA (excluded for HIS column), and 1 mM DTT. Proteins were dialyzed in the same buffer but with 100 mM NaCl for the Mono Q column.

For the initial crystallization experiments for the apo Fyn_R96I or Fyn_R96W SH3 domains, the domains were dialyzed against 20 mM Tris (pH 8.0), 150 mM NaCl, and 1 mM EGTA and concentrated by ultrafiltration to 5–8 mg/ml. Crystallization was performed by vapour diffusion (Fyn_R96I, sitting-drop at 18°C; Fyn_R96W, hanging-drop at 4°C) where 1 µl of protein solution was mixed with 1 µl of well solution (0.2 M ammonium acetate, 0.1 M sodium citrate [pH, 5.6], and 1.0 M lithium sulfate for Fyn_R96I [crystal form I]; 0.2 M ammonium acetate, 0.1 M Tris–HCl [pH, 5.6], and 30% v/v MPD for Fyn_R96I [crystal form II]; and 2 M ammonium sulfate, 2% PEG 400, and 0.1 M Tris [pH, 9.1] for Fyn_R96W SH3). PEG 200 (25%) was used as a cryoprotectant for Fyn_R96I (crystal form I). Data were recorded at 100 K (Fyn_R96I) or 295 K (Fyn_R96W) at λ = 1.54 Å, using the Rigaku 200 X-ray generator, confocal multilayer mirrors (Osmic), and a MAR300 image plate detector. Data were integrated, merged, and scaled using Mosflm [39] and Scala [40] (Supplementary Table S1). All data were checked using ContaMiner [41], and the initial phases were determined by molecular replacement using Molrep [42]. For Fyn_R96I (crystal form I) and Fyn_R96W SH3, Fyn_R96 SH3 was used as a template (Protein Data Bank PDB ID 1SHF) [43]. Fyn_R96I (form I) was used as a template for Fyn_R96I (crystal form II). The structures were refined using Refmac [44] and COOT [45] (Supplementary Table S1; PDB IDs, 3HOF, 3HOH, and 3HOI).

In a second round of crystallization experiments, Fyn_R96I with a C-terminal hexahistidine tag (see [46] for cloning details) was crystallized in 0.1 M citric acid at a pH of 3.5 and 25% w/v polyethylene glycol 3,350. The crystals were cryoprotected with 25% glycerol and diffracted to 1.34 Å resolution at the SOLEIL synchrotron beamline PROXIMA 2A, France. The hexahistidine tag stabilized the crystal packing (Supplementary Figure S1A, Supplementary Table S1; PDB ID 6IPY). Fyn_R96W SH3 was crystallized in 0.08 M sodium acetate trihydrate at a pH of 4.6, 1.6 M ammonium sulfate, 20% v/v glycerol, and 50 mM 18-crown-6. Crystals diffracted to a 1.57 Å resolution at PROXIMA 2A. The CCP4 online version of MoRDa was used to determine the structures [47], followed by rebuilding through BUCCANEER [48], and manual and automated refinement (REFMAC5, PHENIX) [49,50]. Two 18-crown-6 molecules were identified (Supplementary Figure S1B, Supplementary Table S1; PDB ID 6IPZ).

Crystals of the HIV-1 Nef_58-205–Fyn_R96W SH3 complex were obtained via the hanging-drop vapour diffusion method at 18°C–20°C. Before crystallization, stoichiometric amounts of Nef_58–205 and Fyn_R96W SH3 were mixed, concentrated to obtain 0.35 mM of Nef–SH3 complex, and filtered. Then, 1 µl samples of this solution were mixed with 1 µl of the reservoir buffer containing 0.25–0.3 M sodium potassium tartrate, 0.5 M bicine buffer (pH, 8.4), and one critical micelle concentration β-d-octylglucopyranoside. Crystals were cryoprotected by rapid transfer into 28% glycerol, 0.3 M sodium potassium tartrate, and 0.5 M bicine buffer (pH, 8.4). The crystals were then flash-cooled in liquid nitrogen, and data were collected at 100 K at the FIP beamline at the European Synchrotron Radiation Facility (Grenoble, France) using the MAR345 image plate detector. The images were processed and scaled using Mosflm [39] and Scala [40]. Crystal phases were obtained by molecular replacement using LΔN–Fyn_R96 and WLΔN–Fyn_R96I SH3 complex structures as templates (PDB IDs 1AVZ and 1EFN). The structure was refined using Refmac [49] and Phenix [50] (Supplementary Table S2; PDB ID 4D8D).

SF2 Nef–Fyn_R96I SH3 complex crystals were obtained by the sitting-drop vapour diffusion method at 18°C–20°C. Before crystallization, stoichiometric amounts of full-length SF2 Nef and Fyn_R96I SH3 were mixed, concentrated to obtain 1 mM of Nef–SH3 complex, and filtered. Then, 1 µl of this solution was mixed with 1 µl of the reservoir buffer containing 0.10 M calcium acetate hydrate, 0.10 M sodium acetate buffer (pH, 4.5), and 10% w/v PEG 4000. The crystals were cryoprotected by a rapid transfer into 28% glycerol, 0.10 M calcium acetate hydrate, 0.10 M sodium acetate buffer (pH, 4.5), and 10% w/v PEG 4000. All data for SF2 Nef–Fyn_R96I SH3 were collected at 100 K at PROXIMA 2A with an EIGER 9 M detector (proposal numbers 2016 0098, 20161236, and 20170193). Data from three crystals were processed, scaled, and combined using XDS as implemented in the XDSme pipeline (unmerged data are available at https://doi.org/10.25781/KAUST-8CV15). STARANISO [51] was used for merging the three datasets and for establishing the anisotropic resolution limits. Initial phases were determined using MoRDa [47] with the PDB ID 4D8D as the molecular replacement template (Q score, 0.834). The structure was manually corrected using COOT and refined using LORESTR [52] (Ramachandran outliers/favourite 2.53/91.14%) (Supplementary Table S2; PDB ID 7D7S).

Molecular dynamics simulations were carried out in triplicates for apo-Nef (Nef-A) and Nef in a complex with Fyn_R96I (Nef-C), using the 2.5 Å resolution crystal structure of LAI Nef_T71R bound to Fyn_R96I (PDB ID 1EFN) as a template. MD simulations were performed using GROMACS 2018 [53] with the AMBER14SB force field [54]. Each protein was inserted into a cubic box filled with TIP3P water molecules. Box size was chosen to have a minimum distance of 10.0 Å from the protein edges to the box sides. The system was neutralized with Na⁺ and Cl⁻ ions. Energy minimization was performed followed by equilibration using isothermal ensemble dynamics (NVT) with a velocity-rescale thermostat [55] for computing velocities and atom positions. The structures were further equilibrated for 2 ns with isothermal–isobaric ensemble dynamics (NPT). Periodic boundary conditions were applied in the X, Y, and Z directions. Final production simulations were conducted using NPT for 200 ns in triplicate. The temperature was kept constant at 300 K using a velocity-rescale thermostat [55] (τT = 0.1 ps) and 1 bar pressure was maintained using a Parrinello–Rahman barostat [56] (τP = 2.0 ps). Electrostatic interactions beyond 12 Å were evaluated using the particle mesh Ewald method [57]. The LINear Constraint Solver algorithm (LINCS) [58] was used to constrain the bond lengths and the Output trajectories were analyzed using GROMACS 2018 and PyMol (www.pymol.org).

Proteins were labelled using the Monolith protein labelling kit RED-NHS 2nd Generation. The concentrations of the labelled protein ranged from 10 to 50 nM. The concentrations of unlabelled preincubated SH3 and H1 were 50 µM and 1 mM, respectively. Measurements were performed at 25%–50% LED and 40% microscale thermophoresis powers. Data were analyzed using the program provided by Nanotemper Technologies. All MST experiments were performed in 50 mM Tris (pH, 8.0), 200 mM NaCl, 2 mM EDTA (excluded in HIS tag purification), and 1–2 mM DTT at 20°C.

Isothermal titration calorimetry (ITC) experiments were performed on MicroCal PEAQ-ITC (Malvern Panalytical) using the 19-injection standard method at 25°C. All proteins were dialyzed and degassed in 20 mM sodium phosphate (pH, 7.5), 150 mM NaCl, 2 mM EDTA, and 1–2 mM TCEP. The protein concentrations in the cell ranged from 20 to 75 μM and were 10-fold higher in the syringe. The measurements and data were analyzed using the software provided by Origin.

Yeast two-hybrid (Y2H) screenings were performed as previously reported [59].

Bioinformatics analysis for Y2H was performed using RaptorX for property and structure predictions and the ELM database for the identification of SH3 binding motifs [60,61].

To assess the influence of the flexible regions of Nef on ligand binding, we recombinantly produced the following HIV-1 LAI Nef constructs: (i) full-length Nef, (ii) Nef with a deletion of the N-terminal arm while retaining the W₅₇L motif that is implicated in CD4 binding (residues 56–205; WLΔN); (iii) Nef with a deletion of the N-terminal arm that lost W₅₇ and retained L₅₈ (residues 58–205; LΔN), corresponding to the product of Nef cleavage by the HIV-1 protease [62]; (iv) Nef with a deletion of the N-terminal arm that lost the W₅₇L motif (residues 60–205; ΔN); and (v) a Nef ΔN construct with an additional truncation of the core loop to eliminate the E₁₆₀xxxLL endocytosis motif (residues 60–158;174–205; ΔNΔL) (Figure 1A). We also included the full-length SF2 Nef to test the isolate specificity of the effects. Compared with the LAI isolate, SF2 Nef has a four amino-acid insertion at the N-terminal region and a T₇₁R substitution that enhances binding to Src family SH3 domains [28,63] (Figure 1B).

To test the effect of the SH3 R position on Nef binding, we produced SH3 domain constructs that were engineered to have different side chains at this position. We used Fyn_R96, the Fyn SH3 wild-type where the R position is a long and positively charged arginine; Fyn_R96I, where the R position is filled by a medium-sized hydrophobic isoleucine as in Hck; and Fyn_R96W where the R position is occupied by a large and hydrophobic tryptophan, as in the Rous sarcoma virus v-Src (Figure 1C).

We measured the thermodynamic parameters of the interactions between the three SH3 variants and different Nef constructs using ITC (Table 1). In agreement with previous studies [25,27], the hydrophobic substitutes Fyn_R96I and Fyn_R96W showed markedly higher affinities than Fyn_R96 by better matching the hydrophobic Nef pocket. Compared with LAI Nef, SF2 Nef showed a higher affinity for all SH3 variants. This observation can be explained, at least in part, by the additional hydrogen bond formed by R₇₁ in SF2 Nef [63]. However, our titrations also provided the following additional insights:

• (i)    In all of the LAI Nef titrations, the binding thermodynamics of Fyn_R96I were overall similar to those of Fyn_R96W; however, Fyn_R96W consistently had more favourable enthalpy (ΔH) and less favourable entropy (ΔS). The less favourable ΔS of the most hydrophobic R mutant, Fyn_R96W, was unexpected because hydrophobic interactions normally contribute favourably to ΔS [64]. This observation suggested that unknown additional factors influenced the thermodynamics of binding.

• (ii)   Truncated Nef constructs bound approximately four times more strongly to Fyn_R96W than to Fyn_R96I, whereas full-length LAI and SF2 Nef bound stronger to Fyn_R96I than to Fyn_R96W. The affinity to the Fyn_R96 SH3 domain was weak for all Nef constructs (K_d > 25 µM) and did not vary significantly. Thus, the presence of the N-terminal residues 1–55 increased the affinity of Nef specifically for the SH3 isoleucine variant.

• (iii) Compared with Fyn_R96I, Fyn_R96 bound ∼10 times more weakly to the truncated LAI constructs. When full-length Nefs were used, the difference in affinity between Fyn_R96 and the other SH3 variants increased to 50-fold (LAI) and 200-fold (SF2). Although precise thermodynamic parameters could not be established for Fyn_R96 due to the low C-values of the Fyn_R96 binding curves, this observation suggested that the presence of the N-terminal segment increased the selectivity between Fyn_R96 and the other two variants.

• (iv) WLΔN showed significantly different thermodynamics from the other LAI titrations. WLΔN showed favourable ΔH and ΔS, whereas in all other LAI titrations, an approximately two-fold more favourable ΔH compensated for an unfavourable ΔS. This WLΔN-specific irregularity suggested that the W₅₇L dipeptide promotes a thermodynamic contribution to SH3 binding that disappears once the full N-terminal segment is present.

Together, our ITC analysis showed that the presence of the flexible Nef regions affected the affinity and specificity of the association with SH3, and that these effects were specific to the motifs present in the flexible Nef regions.

Our ITC data suggested that the presence of the flexible N-terminal Nef residues influenced the association of Nef with SH3 domains. This Nef region contains an amphipathic helix (termed H1, comprising residues 15–22 in LAI; Figure 1A) that helps to anchor Nef to the charged lipid headgroups of the cellular membrane [23]. In structural studies of full-length Nef, this H1 helix has been observed to bind to the hydrophobic groove formed by the core helices ɑ1 and ɑ2 and the connecting residues (see PDB IDs 6CRI, 4EN2, and 6CM9). In this report, we refer to this region as the ɑ1–ɑ2 groove. This groove is adjacent to, but not overlapping with, the SH3-binding site [35,36].

We synthesized a Nef fragment containing H1 (residues 8–26) to test its affinity for the different Nef constructs in the presence or absence of the SH3 Fyn_R96I and Fyn_R96W variants. We used microscale thermophoresis (MST) for this experiment because it can measure low affinities as well as assess the effect of the presence of one ligand (preincubated at constant concentration with the labelled protein) on the binding of a second ligand (titrated as a dilution series). We preincubated fluorescently labelled Nef with SH3 domains before titrating H1. To ensure that most Nef bound to the SH3 domain, we used a constant concentration of 50 µM (∼10 × K_d) of SH3 domains. We did not include the Fyn_R96 SH3 here because its low affinity meant that we were unable to reach a concentration of 10 × K_d.

In MST experiments without SH3 domains, H1 bound to the core domain-only Nef ΔNΔL with a low affinity (K_d ∼ 300 µM). The presence of additional flexible regions further decreased this affinity. The core loop containing Nef ΔN bound to H1 with a K_d of 700 µM, Nef WLΔN bound H1 with K_d > 1 mM, and full-length LAI and SF2 showed no binding at all. The presence of either Fyn_R96I or Fyn_R96W SH3 domains increased the affinity of H1 for Nef ΔNΔL and ΔN by approximately two-fold (Figure 2A–F, Table 2). WLΔN also showed a two-fold increase in the affinity for H1 when preincubated with Fyn_R96I. However, the effect of Fyn_R96W could not be determined from the MST data due to the low amplitude of the signal change. However, preincubation with SH3 domains did not result in a significant binding signal of H1 to the full-length Nefs.

Finally, we investigated whether the effect observed was mutual, i.e. whether preincubation of Nef with H1 also increased the affinity of Nef for SH3 domains. We chose ΔNΔL for this experiment, as it had the highest affinity for H1. We observed that the presence of H1 significantly increased the affinity of ΔNΔL for Fyn_R96I (Figure 2G, Table 2).

These data demonstrated that the presence of SH3 domains can enhance the association of H1 with the hydrophobic ɑ1–ɑ2 groove and vice versa. However, the more flexible regions were included, the lower was the affinity of Nef for H1. Full-length Nefs did not show any binding, possibly because their own H1 motifs outcompeted the H1 peptide. The W₅₇L motif was found to back-bind to the ɑ1–ɑ2 groove in the apo-Nef NMR structure (2NEF) [19], which explains its competition with H1. However, the competition between H1 and the core loop was unexpected and suggested that the core loop also loosely associates with the same Nef region.

To determine whether a binding synergy also existed between SH3 domains and other Nef ligands, we tested the association between Nef and the C-terminal fragment of the p85 subunit of the PI3 kinase. The C-terminal region, which encompasses the inter-SH2 coiled-coil domain (iSH2), has previously been shown to bind to SF2 Nef [65–67]. However, the structural basis of this association is unknown. Our LAI and SF2 Nef constructs bound to the p85 nSH2–iSH2–cSH2 fragment (nicSH2) with K_ds in the range of 50–150 µM (Figure 2H–K, Table 2). We observed that the affinity tended to decrease with increasing length of the flexible regions and that the presence of R96I/W SH3 domains negatively impacted on the nicSH2 affinity. However, these differences were not significant given the large measurement uncertainties of these low-affinity interactions. We concluded that the presence of the SH3 domains did not markedly increase the binding between Nef and p85 nicSH2 in our assay.

We also probed for a synergy in binding between SH3 domains and the CD4-dileucine motif. A previous NMR study found that the presence of the Hck SH3 domain, which carries an isoleucine on the R position, enhances the affinity of Nef for a CD4-dileucine motif-encompassing peptide by a factor of two (from K_d ∼ 1 mM to ∼500 µM) [68]. We attempted to reproduce these data using MST and our Nef constructs with and without SH3 domains. However, the CD4 affinities in all cases were below the detection limit of our MST assay (Supplementary Figure S3).

In summary, our data established that SH3 binding to Nef, in particular to Fyn_R96I, facilitated the interaction of H1 with the hydrophobic ɑ1–ɑ2 groove of Nef. In turn, binding of H1 also enhanced the affinity of SH3 domains to Nef. The ɑ1–ɑ2 groove is located next to the SH3-binding site, but the two do not overlap; hence, the binding of one ligand to one site appears to allosterically stabilize the binding site of the other ligand. However, under the conditions used, SH3 binding did not enhance the association of Nef with the p85 C-terminal region or the CD4 dileucine motif. This indicates that the affinity-enhancing effect of SH3 domains does not cover all Nef associations. Moreover, our observation that the Nef core loop and the W₅₇L motif can compete with the H1 motif suggested that these flexible regions can loosely associate with the ɑ1–ɑ2 groove of Nef.

Our assays showed that the binding of the SH3 domain increased the affinity of Nef for H1 and vice versa, which suggests the existence of allosteric cooperativity. To further investigate this phenomenon, we performed MD simulations comparing the Nef core domain in its apo state (denoted Nef-A) with its Fyn_R96I-complexed state (Nef-C). We derived Nef-A and Nef-C from the 2.5 Å resolution crystal structure of LAI Nef_{54–205,T71R} bound to Fyn_R96I (PDB ID 1EFN [20]). This crystallographic Nef model lacked residues 54–70 and 149–177, making it akin to the molecule we defined as ΔNΔL. In MD simulations, Nef-C had lower calculated van der Waals and electrostatic energy, and showed decreased overall mobility compared with Nef-A [root mean square fluctuations (RMSF) were 2.52 ± 1.88 Å for Nef-A and 2.16 ± 1.32 Å for Nef-C] (Figure 3A–D), Supplementary Data SDS1.S1), indicating that the SH3 domain stabilized and rigidified Nef. The RMSF differences were greatest in the residues of the N-terminal arm and the core domain loop (Figure 3B). Within the core domain, residues 75–80 showed the strongest stabilization in Nef-C, as expected from their direct interaction with the SH3 domain (Figure 3B,E). However, this SH3-induced stabilization also extended to the underlying residues 111–126, including the C-terminal part of the ɑ2 helix. Additionally, Nef residues 101 to 102 and residue 138 showed fewer fluctuations in the SH3 domain complex (Figure 3B,E). Thus, a large part of the H1 binding site was stabilized by the SH3 domain as was a region that interacts with CD4 [37] and contributes to Nef–Nef crystal contacts (observed in PDB IDs 1EFN, 1AVV, 1AVZ, 4USW, 3REA, and 3D8D) and possibly to dimerization in vitro [69] (Figure 3F,G). These results suggested that SH3 binding selectively decreases the dynamics of Nef core regions involved in protein–protein interactions.

Our ITC data showed clear differences in the thermodynamic signatures of the SH3 domain associations with Nef WLΔN compared with the associations of these SH3 domains with other LAI fragments. Our MST data indicated that the binding of H1 to Nef is reduced by the presence of the core loop in the W₅₇L motif and by the additional presence of the N-terminal residues 1–55. Together, these observations suggested that the core loop and the W₅₇L motif compete with H1 for binding to the ɑ1–ɑ2 groove.

Indeed, previous structural analyses found the hydrophobic ɑ1–ɑ2 groove to be occupied by intramolecular protein regions. In NMR studies of a truncated apo-Nef (equivalent to Nef WLΔN plus a deletion of the E₁₆₀xxxLL endocytosis motif), W57 from the N-terminal extension was found to bind back to the ɑ1–ɑ2 groove (2NEF) [19]. W₅₇L was also associated with the ɑ1–ɑ2 groove, albeit in a different position, in a crystal structure of SF2 Nef bound to an engineered high-affinity SH3 domain (3REA [70]). However, other crystallographic studies of the core domains of HIV-1 and SIV Nef show the ɑ1–ɑ2 groove occupied by the endocytosis motif from the core domain loop of an adjacent molecule in the crystal lattice (PDB IDs: SIV, 5NUH and 5NUI; HIV, 3RBB). The binding positions of the endocytosis motifs in HIV and SIV Nef in these crystals were again different [71,72].

Investigating the LAI Nef_54-205,T71R–Fyn_R96I structure (PDB ID 1EFN), we found strong, continuous residual density in the ɑ1–ɑ2 groove in both Fyn_R96I-bound Nef molecules of the asymmetric unit (Supplementary Figure S1). In both molecules of the asymmetric unit, the electron density region closest to W₅₇ in the NMR model provided a good fit for a tryptophan side chain whereas the adjacent density fits a leucine side chain. These observations suggest that the W₅₇L motif binds to the ɑ1–ɑ2 groove in these crystal structures (Supplementary Figure S1). In agreement, the electron density for W₅₇L was absent from two LAI Nef LΔN–SH3 complexes crystallized in the same space group as 1EFN, namely, 1AVZ (Nef_58–205–Fyn_R96) and 4D8D (Nef_58–205–Fyn_R96W produced for this study; Supplementary Figure S1, Supplementary Table S2).

Assuming this W₅₇L position in the ɑ1–ɑ2 groove, the remaining free electron density in 1EFN cannot be attributed to the residues of the N-terminal region between W₅₇ and T₇₁ and might correspond to the core loop. A W₅₇L-stabilized association of this loop with the Nef core domain is supported by a relative protection of this loop region from hydrogen exchange [73]. A molecular model for such a ‘closed' form, built based on the unattributed residual electron density in 1EFN, was stable in MD runs, which indicates that this conformation is stereochemically possible (Supplementary Figure S1). Moreover, in MD simulations that started with the core loop and N-terminal arm (residues 57–71) in a dissociated ‘open' conformation these flexible elements tended to regroup towards the ɑ1–ɑ2 groove during the simulation (Figure 3A). Hence, our simulations suggested that these flexible residues have a tendency to adapt a ‘closed' conformation. Thus, we speculate that in the W₅₇L -containing Nef forms, W₅₇L motif, and the core loop jointly occupy the ɑ1–ɑ2 groove, with the W₅₇L motif directly interacting with the core loop and stabilizing it.

These observations suggest a model that explains our ITC and MST data. In this model, the core loop alone and the W₅₇L motif alone only associate very loosely and dynamically with the ɑ1-ɑ2 groove. When both regions are available this association with the core becomes more stable. However, both can be outcompeted by H1. Further evidence for the capacity of H1 to outcompete other intramolecular regions comes from our survey of the available experimental structures (PDB IDs 6CRI, 4EN2, and 6CM9) (Supplementary Table S3). This survey showed that whenever the H1 motif is present in Nef, it occupies the ɑ1–ɑ2 groove rather than W₅₇ or the core loop. In the 3.5 Å crystal structure of full-length LAI bound to Fyn_R96I that we determined in this study (PDB ID 7D7S; Supplementary Table S2), the ɑ1–ɑ2 groove was partly occluded by adjacent molecules in the crystal lattice, further supporting the hypothesis that also the H1–core association has a low affinity and can be easily displaced.

In our ITC titrations, we observed that increasing the hydrophobicity of the SH3 R position from isoleucine to tryptophan resulted in a less favourable entropy contribution, which suggests that the incorporation of tryptophan caused additional loss of degrees of freedom upon binding (Table 1). To compare the RT loop features of apo- and Nef-bound Fyn and Hck SH3 domains, we produced the missing crystal structures of the ligand-free Fyn_R96I and Fyn_R96W (PDB IDs 3HOF, 3HOH, 3HOI, 6IPY, and 6IPZ; Supplementary Table S1). We observed that the apo Fyn_R96 and Fyn_R96I structures had a similar strong RT loop-rigidifying network formed by six hydrogen bonds (Figure 4). Conversely, the NH group of Fyn_R96W disrupted the RT loop hydrogen-bond network and produced an RT loop that was more flexible in its free state. Consequently, the Fyn_R96W variant lost more entropy upon fixation of the RT loop in the Nef interface than did Fyn_R96I, explaining the less favourable entropy contribution to binding of Fyn_R96W. The poor stabilization of the Fyn_R96W RT loop is similar to that of Hck SH3, which was previously shown to have a less favourable enthalpy contribution to LAI Nef LΔN binding than those of other SH3 domains, including Fyn_R96 [27]. Yet, fewer hydrogen bonds are needed to be broken in Fyn_R96W upon Nef binding than in Fyn_R96I or Fyn_R96, which explained the observed greater favourable enthalpy contribution of Fyn_R96W.

Our ITC binding studies suggested that the presence of the Nef N-terminal H1 helix has specific effects on SH3 domain binding. The presence of H1 markedly increased the binding of Fyn_R96I more than that of Fyn_R96W with no significant effects on Fyn_R96. Our comparison of all available Nef–SH3 complexes revealed some flexibility in the relative orientation of the SH3 domains with respect to Nef, much of which can be attributed to the crystallographic lattice contacts (Supplementary Data SDS1.S2; Supplementary Figure S1H). However, the R residue position relative to Nef and the canonical P₇₂xxPxR–SH3 interactions are well conserved (Figure 5A). Maintaining these key interactions is possible primarily because of the malleability of the Nef P₇₂xxPxR motif region that, compared with the rest of the Nef core, acts as a buffer for positional differences (Supplementary Figure S1H). We selected the Nef–SH3 complexes that appeared not to be disturbed by the crystallographic lattice and inspected the interactions between the R position and the underlying Nef hydrophobic pocket.

The SH3 R position interacts with the side chains of Nef residues W113 and F90. These side chains have been termed the ‘gatekeeper' [72], because they separate the Nef SH3 binding surface from the hydrophobic ɑ1–ɑ2 groove. This groove was observed to bind to H1 and CD4 in previous structural studies [35,68]. Hence, a change in the gatekeeper residue side-chain position resulting from an interaction with the SH3 domain will also affect the shape and size of the hydrophobic groove. The R position of the SH3 domain might therefore influence the ligand interactions of the ɑ1–ɑ2 groove.

We noted that the side-chain position of W113 and, to a lesser extent, F90 changes in response to the different SH3 orientations and different residues in the R position. In cases where the R position is an isoleucine (Hck and Fyn_R96I), W113 is rotated towards the SH3 domain and away from the hydrophobic crevice compared with apo-Nef (Figure 5B). Conversely, W113 is pushed into the hydrophobic groove in Fyn_R96W-bound Nef structures, whereas W113 adopts an apo side-chain orientation when Nef is in a complex with Fyn_R96. Thus, binding to SH3 domains with isoleucine in the R position increases the adjacent ɑ1–ɑ2 groove by 2.3 Å, whereas tryptophan decreases the groove by 0.9 Å; and an arginine leaves it relatively unchanged.

Next, we superimposed the available experimental structures of HIV-1 Nef bound to its N-terminal H1 (EM: PDB IDs 6CM9, 6CRI; X-ray diffraction: 4EN2) with the Nef–SH3 complexes. In the Nef–H1 structures, the malleable N terminal of H1 partly occupies the same Nef region as the R position in the Nef–SH3 complexes. Tryptophan or arginine at the R position clashed more with H1 than isoleucine, suggesting a greater competition of the former with H1. The more stable helical part of H1 fills the ɑ1–ɑ2 groove. In this position, V16 and W13 of H1 are in close contact with W113 (Figure 5C,D). As observed in complexes with Fyn_R96I and Hck, an inward-oriented Nef W113 would leave sufficient space for H1, whereas the outward pointing W113 as seen in Fyn_R96W complexes would produce mild clashes. From these analyses we concluded that the R position–specific effects of H1 on SH3 binding result from small spatial obstructions that can be direct, or indirect, i.e. mediated by the gatekeeper. These R position–specific effects of H1 are convoluted with the stabilizing effect that H1 has on the Nef core and that non-specifically enhances SH3 interactions.

This model can explain why full-length Nefs discriminate more strongly between Fyn_R96 and the other two SH3 variants as well as have the highest affinity for Fyn_R96I. The additional differences between the SH3 affinities of LAI and SF2 Nef may arise due to the T71R substitution in SF2 Nef, which adds an additional hydrogen bond to the SH3 interaction [28,63]. Our SF2 Nef–Fyn_R96I crystal structure did not show any evidence of additional interactions between the flexible regions and the SH3 domain. However, the H1 region of SF2 Nef is characterized by the duplication of the RAEP motif situated at the C-terminus of H1, and even though this motif has not been predicted to contribute to the intramolecular association, it may have subtle effects on the interaction.

In the context of previous work, our analysis of the ‘tertiary' association between an SH3 domain and Nef demonstrated that the combination of a linear interaction motif and a folded core domain can create opportunities for synergy and allostery in addition to enhancing affinity and selectivity. Given the promiscuity of SH3–PR motif interactions [74,75], these ‘tertiary' features might also be useful for increasing the selectivity in cellular SH3-based signalling.

To investigate whether a Nef-like tertiary SH3 association is a common mode of binding in cellular SH3 associations, we designed a yeast two-hybrid (Y2H) screen. As a high-throughput compatible proxy for this binding mode, we selected cellular proteins that could distinguish between isoleucine and arginine at the R position on the Src family SH3 domains. We used the SH3 domains of Hck, Fyn_R96, and Fyn_R96I as bait for the Y2H analysis. First, cotransformation of cells with Nef resulted in β-galactosidase production and yeast colony outgrowth on medium lacking uracil with Hck and Fyn_R96I as baits but not with Fyn_R96 (Figure 6A). These results were confirmed using the LacZ and HIS3 reporters for Nef–Hck and Nef–Fyn_R96. Because Fyn_R96I alone was able to transactivate the LacZ and HIS3 reporters, Fyn_R96I could not be used with these reporter systems. Unlike Nef, the Src-associated in mitosis 68 kDa (SAM68) protein interacted with comparable strength with every SH3 domain (Figure 6A), indicating that SAM68 cannot select for the RT-loop R position.

To identify selective cellular binding partners for the SH3 domains of Hck, Fyn_R96, and Fyn_R96I, we performed Y2H screening using a human fetal library (Figure 6A). To test for a Nef-like differential association, we performed two coupled serial screens. An initial screen, using the human Hck or Fyn_R96I SH3 domain as a bait, yielded 422 and 100 clones, respectively, after gap repair (Supplementary Tables S4, S5). These clones were included in a secondary screen by gap repair against the Hck, Fyn_R96, and Fyn_R96I SH3 domains, and the clones displaying different strengths of β-galactosidase staining and yeast colony outgrowth on selective medium were selected and sequenced.

We found that only 10 clones — nine from the screen using Hck and three from that using Fyn_R96I, representing 2% and 3% of total clones, respectively — had substantially different affinities when Hck and Fyn_R96I SH3-mediated interactions were compared with Fyn_R96-mediated interactions (Table 3). We then used bioinformatics analysis to identify putative SH3-binding motifs in these sequences and to predict the fragments’ secondary structures, solvent exposure, and structural disorder (Supplementary Figure S4).

Five of the ten clones were predicted to be almost entirely disordered, and hence could not produce a tertiary binding site (light red in Table 3). Three clones, ABI1, GFAP and ALIX, possessed a proline-rich motif adjacent to a coiled-coil domain (light blue in Table 3). However, these proline-rich motifs were separated from the folded part by 70, 10 and 20 residues in ABI1, GFAP and ALIX, respectively [46]. We previously reported that ALIX has the capacity to distinguish between arginine and isoleucine in the SH3 R position through a simple linear mechanism rather than a 3D motif [46,59]. Hence, it is likely that also ABI and GFAP achieve this selectivity through a linear mechanism. Finally, only two clones, ARS2 and HNRPK, harboured PR motifs within or close to a globular folded protein core (yellow in Table 3). Given this organization, they might provide a 3D environment capable of a tertiary Nef-like SH3 recognition (Figure 6B). We conclude that only a few cellular ligands can select SH3 domains of the Src family according to the same RT-loop residues as used by HIV-1 Nef. In most of these cases, however, this specificity was likely due to a linear binding mode.

Despite lacking catalytic activity, SIV and HIV Nef molecules have a remarkably large range of functions. Just as remarkable is the structural malleability of the flexible regions of Nef, which make up ∼50% of the Nef sequence. We investigated whether the combination of interactions established jointly by Nef's folded core and flexible regions creates synergy and allostery in ligand binding. To overcome the difficulties posed by the structure–function analysis of Nef's flexible regions, we combined structural, biophysical, and computational analyses.

Based on our data, we state that the flexible regions can adopt at least three different conformational states (states I–III; Figure 7). If the N-terminal helix H1 is available, then W₅₇L and the E₁₆₀xxxLL endocytosis motif are exposed (state I). If H1 is engaged, for example with the membrane, then the combined interaction of W57 and E₁₆₀xxxLL with the core conceals these motifs (state II). If H1 and W₅₇L are engaged or cleaved, then the E₁₆₀xxxLL motif only weakly associates with the core (state III).

We demonstrated that there is crosstalk between the flexible regions in the different states and the binding of some, but not all, Nef ligands. We showed that the binding of SH3 domains to the tertiary Nef site affects the binding of H1 to the ɑ1–ɑ2 groove and vice versa. Our analysis indicated that this influence results from the combination of a nonspecific dynamic coupling, wherein the binding of one ligand stabilizes the Nef surface for the other ligand, together with a steric interference that is specific for the SH3 R position. This specific interference results from the size of the R position side-chain and its interactions with the Nef gatekeeper residues. Hence, state I binds most tightly to SH3 domains that have a mid-sized hydrophobic isoleucine in the R position, such as Hck, Lyn and possibly Blk (where R is a methionine). States II and III preferred the SH3 domain with a bulky hydrophobic tryptophan in the R position, as in v-Src.

Our analysis corroborated and extended previous findings that the affinity of SH3 domains towards Nef is influenced by the extent of the hydrogen-bond network that stabilizes the RT loop in apo-SH3 [27]. We showed that the incorporation of a tryptophan in Fyn_R96W led to a loss of RT-loop hydrogen bonds, making it intrinsically flexible like Hck SH3. We also showed that state I shielded the Nef ɑ1–ɑ2 groove most strongly from other ligands. However, the conformational states and presence of SH3 domains did not have a measurable effect on the binding of the p85 C-terminal fragment to Nef. Our MD simulations explain this selective allostery, at least in part, by showing that the tertiary binding of SH3 domains stabilizes several interaction surfaces of Nef, including the one used for H1 binding. Thus, our analysis showed that the combination of a tertiary SH3-binding site with large flexible regions can provide an additional level of regulation of ligand binding. In vivo, Nef is myristoylated on its N-terminal glycine [21,85]. The presence of this fatty acid chain allows the N-terminal flexible region of Nef (including H1) to associate with charged lipid headgroups of the cell membrane. In the absence of a suitable cellular membrane, the myristoyl is expected to further increase the ways in which the flexible N-terminal arm can associate with the core domain through a myristoyl-switch mechanism [85]. The resulting crosstalk between the conformational states of Nef and its ligands may help to synergize certain subsets of ligands while excluding others with potential functional implications for viral replication.

Based on our findings we make the following predictions and speculations: Due to the mutual reinforcement between the state I conformation and Fyn_R96I SH3 interactions, this state may promote the Nef function of activating Hck and Lyn. State II may present a membrane-associated and partly autoinhibited form, in which concealing the W₅₇L and E₁₆₀xxxLL motifs prevents premature down-regulation in absence of CD4 or other surface receptors, such as CD3. Once a specific ligand binds to the ɑ1–ɑ2 groove, W₅₇L and the endocytosis motif would get exposed to trigger down-regulation of Nef and its cargo. State III would represent the conformation obtained after proteolytic cleavage by the viral protease between W₅₇ and L₅₈ [62]. Cleaved Nef is predominantly found inside HIV-1 virions [62]. Hence, state III might facilitate tighter packing of Nef molecules within the virion. Upon release into infected cells, state III Nef molecules would remain mostly cytoplasmic, where they easily bind ligands with their poorly concealed ɑ1–ɑ2 groove, while relatively freely displaying their endocytosis motif.

Our findings also raise the possibility that additional subtle allosteric mechanisms are in effect. We noted that the SH3 R position affects the Nef W113 orientation, which, in turn, might affect other ligands. For example, the SH2–SH3 fragments from Fyn or v-Src would not be able to form the same dimeric complex with Nef as Hck SH2–SH3, because of clashes between Fyn R96 and W113, or R96 and Fyn E93 (in Hck the corresponding glutamic acid interacts with R106 of an adjacent Nef molecule in the dimer) [24]. This effect might contribute to the differential outcomes between Nef interactions with Hck and other Src kinases [24,31]. Similarly, the R position of a bound SH3 domain might favour or disfavour specific ligands to the ɑ1–ɑ2 groove. An additional allosteric regulation mechanism might arise from the stabilizing effect of the SH3 domain on selected Nef regions. Thus, SH3 binding could promote certain Nef interactions, including Nef self-associations. SH3 binding also stabilizes the Nef region involved in CD4 binding. However, the presence of an SH3 domain did not enhance binding of a CD4 peptide in our studies, and the structural models available for this interaction show the proline-rich motif of Nef in a conformation that is incompatible with SH3 binding [37]. Further research is needed to establish which ligands, if any, can bind Nef simultaneously with Src kinases or other SH3-domain containing ligands, such as Tec family kinases [16]. Although our study did not include cell-based functional assays, it provides a new viewpoint through which the different functions and interactions of Nef can be analyzed and better understood.

Considering the versatility and potency of the combination of flexible and folded regions in directing Nef's interactions, we considered whether a Nef-like tertiary interaction is common among cellular SH3 interactions. Our Y2H screen revealed that only very few human SH3-binding proteins distinguish SH3 domains according to the same RT-loop residue as Nef. Our bioinformatics analysis and the experimentally established example of ALIX show that the RT-loop selectivity of these cellular proteins does not necessarily require a Nef-like tertiary interaction but can be obtained through a linear peptide-like binding mode. Although our study could not rule out the presence of supplementary SH3 associations in general, the results imply that a Nef-like tertiary, and possibly allosterically active, SH3 binding mode is not common among cellular proteins. Hence, this binding mode appears to be a de novo adaptation that is specific to this viral protein. It is possible that Nef's affinity and selectivity characteristics are intrinsically unsuitable for mediating SH3 binding in cellular signalling. The tertiary recognition mode of Nef results in strong SH3 binding, which not only activates Src kinases that contain a favoured isoleucine in the R position, such as Hck or Lyn, but also interferes with the function of those with less favourable R residues, such as Lck, which has a serine, or Fyn and Src, which have an arginine [30,31,86,87]. The Nef–SH3 domain binding mode may therefore be too strong to be widely used in cells where it might disrupt several Src kinases. The uniqueness of the tertiary SH3-binding site of Nef supports the identification of this site as a selective antiviral drug target without an apparent risk of crossreactions with other cellular proteins [88].

All crystallographic data are deposited in the Protein Data Bank (Fyn SH3 domain mutants, 3H0F, 3H0H, 3H0I, 6IPY, and 6IPZ; Nef–Fyn complexes, 4D8D and 7D7S). Additionally, unmerged crystallographic data are available as PDB ID 7D7S at https://doi.org/10.25781/KAUST-8CV15.

The authors declare that there are no competing interests associated with the manuscript.

This research was supported by funding from King Abdullah University of Science and Technology through the baseline fund and award no. FCC/1/1976-25 from the Office of Sponsored Research. For computer time, this research used the resources of the Supercomputing Laboratory at King Abdullah University of Science & Technology (KAUST) in Thuwal, Saudi Arabia. This work was supported by the Centre National de la Recherche Scientifique; the Institut National de la Santé et de la Recherche Médicale; the Agence Nationale de Recherche sur le SIDA et les hépatites virales; the Ministry of Science and Technology, China [grant numbers 2007CB914304 and 2006AA02A313]; the National Natural Science Foundation of China [grant numbers 30800181 and 30625011]; the Research Foundation Flanders; the Concerted Action Program of the Katholieke Universiteit Leuven; and the National Institutes of Health (MD Anderson's Cancer Center Support) [grant number CA016672]. S.O. and A.L. were fellows of the ANRS, and X.S. was a fellow of the ‘Ambassade de France en Chine.'

Stefan Arold: Conceptualization, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Writing — original draft, Project administration, Writing — review and editing. Abdullah M. Aldehaiman: Conceptualization, Data curation, Formal analysis, Supervision, Validation, Investigation, Writing — review and editing. Afaque A. Momin: Conceptualization, Data curation, Formal analysis, Validation, Investigation, Writing — original draft, Writing — review and editing. Audrey Restouin: Data curation, Validation, Investigation. Luyao Wang: Formal analysis, Investigation. Xiaoli Shi: Formal analysis, Investigation, Visualization. Safia Aljedani: Investigation. Sandrine Opi: Validation, Investigation, Visualization. Adrien Lugari: Investigation, Visualization, Writing — original draft. Umar F. Shahul Hameed: Supervision, Investigation. Luc Ponchon: Investigation. Xavier Morelli: Conceptualization, Supervision, Funding acquisition, Investigation, Writing — original draft, Writing — review and editing. Mingdong Huang: Conceptualization, Supervision, Funding acquisition, Validation, Writing — review and editing. Christian Dumas: Conceptualization, Supervision, Validation, Investigation, Writing — original draft, Writing — review and editing. Yves Collette: Conceptualization, Supervision, Funding acquisition, Validation, Investigation, Visualization, Writing — original draft, Writing — review and editing.

We acknowledge SOLEIL for the provision of synchrotron radiation facilities. We also thank L. Chavas, P. Legrand, S. Sirigu, and P. Montaville for their assistance with beamline PROXIMA 1 and G. Fox, M. Savko, and B. Shepard with PROXIMA 2A. We also thank M-P. Struband and M-T. Augé-Sénégas for the initial cloning of some of the Fyn_R96I and Fyn_R96W mutants. We thank M. Geyer for providing the expression construct of Nef SF2. We acknowledge the staff at ESRF beamline BM30, Grenoble, France, for their assistance with crystallographic data collection. We also thank the KAUST Supercomputing Laboratory for their assistance with computational resources for molecular dynamics simulations using the IBEX cluster.

HIS3

histidine

ITC

isothermal titration calorimetry

MST

microscale thermophoresis

PDB

Protein Data Bank

PR

proline-rich regions

RMSF

root mean square fluctuations

SAM68

Src-associated in mitosis 68 kDa

URA3

uracil