The development of drugs against human immunodeficiency virus type 1 (HIV-1) was initiated about 30 years ago [1]. Since then, there have been 30 available antiretroviral drugs approved by the U.S. Food and Drug Administration (FDA). Current therapies against the AIDS pandemic are mainly based on inhibition of the three enzymes expressed by HIV-1 (protease, reverse transcriptase and integrase) [2,3,4,5]. Unfortunately, all of these approaches suffer from drug resistance due to the emergence of viral mutants [6]. In a rapid process of Darwinian evolution, the virus has developed functional enzyme variants not inhibited by current therapies/drugs. Several promising new strategies may, at least in part, counteract such viral hypervariability [7,8]. One of these strategies interferes with the HIV-1 transcriptional activity induced by the HIV-1 trans-activator of transcription (Tat) protein [9,10,11] (Figure 1a). Tat is a small and flexible protein, which acts as a molecular adaptor. Nuclear magnetic resonance (NMR) experiments reveal no secondary structure elements for free Tat in solution [12,13,14,15,16], although dramatic structural changes have been observed (both X-ray and NMR structures) when the protein is in complex with host cell partners [17,18]. The high plasticity of Tat is demonstrated by the fact that it can tolerate up to a 40% of sequence variation without loss of activity [19].

Post-translational modifications—in particular acetylation of Tat residues can play a critical role in HIV replication, enhancing HIV-1 transcriptional activation [20,21,22].

Tat is acetylated at K50 by the p300/CPB co-activator and histone acetyltransferase (HAT) [10,11,21,23,24,25]. The p300/CPB protein contains several functional modules, including a nuclear receptor interaction domain (RID), a CREB and MYB interaction domain (KIX), cysteine/histidine regions (TAZ1/CH1 and TAZ2/CH3), an interferon response binding domain (IBiD), a histone acetyltransferase (PAT/HAT) domain [20,26,27,28,29], and a motif called the bromodomain (BRD). BRDs have recently been discovered to function as acetyl-lysine binding domains [30]. They feature a characteristic left-handed, four-helix bundle (helices αZ, αA, αB, and αC (Figure 1b). A large loop links helices αZ and αA (ZA loop), and a shorter loop links helices αB and αC (BC loop) [23]. Acetylated lysines bind between both loops by inserting the acetyl moiety along the longitudinal axis of the helices and between the four helices bundle [23,25,30]. In all the structures of BRDs in complex with cognate histone peptides so far determined [31], the acetyl moiety forms a hydrogen bond with a very conserved asparagine residue at the end of the cavity, serving as an anchor and providing specificity for acetylated versus non acetylated lysines [32]. Additional contacts with flanking amino acids in the target motif determine specificity/affinity for the large variety of BRD modules so far reported [31]. It is worth noticing that the binding regions belong to post-transcriptional modified histone tails, which are intrinsically disordered [33,34,35]. Therefore, it is not clear whether additional contacts may contribute to the recognition process.

Tat is also acetylated at K28 by the p300/CBP-associated factor (PCAF), thereby enhancing Tat binding to the protein P-TEFb [17]. This may in turn regulate functionally critical steps in transcription [20,21,22,49,50,51]. PCAF is a 90-kDa protein that contains the P300/SRC-1 domain (residues 1 to 352), the HAT domain (residues 518 to 649), which acetylates K28 [18,51] and a C-terminal BRD (residues 719-832) [18]. The BRD binds to the Tat protein, which is acetylated at lysine 50 (AcK50). Despite the fact that the structural determinants of full-length PCAF in complex with Tat have not yet been determined experimentally, the NMR structure of the complex between an acetylated Tat peptide (⁴⁶SYGR(AcK)KRRQRC⁵⁶) (Figure 1c) and the PCAF BRD has been solved by NMR spectroscopy [18].

The structural determinants of the full-length Tat PCAF BRD complex were predicted based on molecular dynamics simulations, combined with biochemical and FRET experiments [52]. This model was built using the structure of the isolated Z2 variant of the HIV-1 Tat [15]. In contrast to the existing structures of BRDs in complex with histone peptides, that model pointed to the presence of contacts between the BRD and non-adjacent regions of the acetylated partner. Furthermore, Tat Regions III and IV made extensive contacts with the ZA loop of the PCAF BRD (Figure 1). Although these contacts were not present in the NMR structure, which only contained a Tat peptide, they were experimentally validated [52], supporting the idea that significant changes in the protein-protein interface may take place in the presence of the entire Tat molecule.

Here we construct an updated structural model of the Tat PCAF BRD complex using a more accurate docking procedure. This is the fully flexible protein-protein technique as implemented in the data-driven docking program HADDOCK version 2.1 [53,54] followed by molecular dynamics (MD) simulations. Importantly, we focused on the 86 amino acids Tat BRU variant, PDBID 1JFW [13], a representative of HIV-1 subtype B. This variant is actually the more common HIV subtype in Europe and North America [12]. The Z2 variant of the protein used in our previous study is evolutionally closer to early strains of the virus, but is nowadays much less abundant [12]. Docking studies involving Tat and its cellular partners can be expected to be feasible. Indeed, (i) HADDOCK (High Ambiguity Driven protein-protein DOCKing) [53,54] has been used to predict the protein-protein and protein-nucleic acid complexes based on available experimental data [55,56,57]; (ii) HADDOCK has also been applied to intrinsically disordered proteins for which NMR structural information is available, such as Tat [50,51,52,53]; (iii) HADDOCK has been successfully applied to a variety of NMR structures with a low number of NOE’s per atom [54,55], similar to the ones we found in the available Tat NMR structures. Moreover, HADDOCK complemented by MD also provides reasonable results in those cases where the protein undergoes conformational changes [56,57,58]. Therefore by using the docking procedure combined with the MD simulations, we believe that a reasonable model of Tat can be assured.

The structure of HIV-1 Tat (86 residues) has been solved by NMR (PDBID 1JFW [13]). Eleven conformers have been deposited. His13, His33, His65 are protonated in the Nε [13]. We superimposed residues 47 to 55 in each of these structures with the corresponding ones with the multiple conformers of Tat⁴⁶SYGR(AcK)KRRQRC⁵⁶ peptide in complex with human PCAF BRD, whose structure has been solved by NMR. His717 and His742 were protonated in Nδ as in the NMR structure (1JM4 [18]). The VMD 1.8 program [64] was used. The conformer n.8 of HIV-1 Tat and n.3 of the Tat⁴⁶SYGR(AcK)KRRQRC⁵⁶.PCAF complex exhibits the lowest pairwise root-mean-square deviation (RMSD) in respect Tat⁴⁶SYGR(AcK)KRRQRC⁵⁶ backbone (2.3 Å and 0.7 Å, respectively). All the RMSDs reported in this paper are calculated using only backbone atoms. Thus, these conformers were selected for all subsequent calculations. The full length Tat was fitted to Tat⁴⁶SYGR(AcK)KRRQRC⁵⁶ at the peptide then AcK50 was replaced with K50 in full length Tat using the tleap module in AMBER package [65] with acetylated lysine parameters adapted from Machado et al. [66]. The resulting structure was relaxed for 10 ns molecular dynamics simulation in water (See Supplementary Section S1; RMSF, RMSD of the system plotted as a function of simulated time in Supplementary Figure S3 and Figure S4 for further details). A clustering of the MD trajectory [67] identified seven clusters, 84 structures in total, which represented 98% of the conformations (Supplementary Section S2). These structures were docked on the 25 PCAF BRD conformers present in the NMR family of structures (1JM4 [18]) by a multiple conformations docking procedure using the HADDOCK 2.1 program [53,54,68]. Haddock is an information-driven flexible docking approach for the modeling of biomolecular complexes [40]. The entire protocol consists of four stages: (i) Topology and structure generation; (ii) Randomization of the starting orientation and rigid body energy minimization; (iii) Semi-flexible simulated annealing; (iv) Flexible refinement in explicit solvent (water).

We first validated our docking procedure by redocking the Tat⁴⁶SYGR(AcK)KRRQRC⁵⁶ from PCAF BRD in the NMR structure of the complex (Supplementary Section S3) [18]. Then, the two binding partners were docked one onto the other by rigid body docking with different Ambiguous Interaction Restraints (AIRs) (Supplementary Section S3, Table S2, for the definition of AIRs, see [54]). The adduct exhibited a backbone RMSD of 1.3 Å for PCAF BRD and 1.7 Å for HIV-1 Tat⁴⁶SYGR(AcK)KRRQRC⁵⁶ with respect to the NMR structure (Supplementary Section S3, Figure S4) (Table 3). These results allow us to suggest that HADDOCK performs well for this system (Supplementary Section S3). Next, we performed rigid-body docking between PCAF BRD and Tat AcK50. The contacts F748–, V752–, Y809–, I764–, Y760–, Y802– to AcK50, E756– to R53, V763– to Y47, E756– to Q54, present in the NMR structure [18], were treated as AIRs. The rigid-body docking was followed by a rigid body energy minimization, a semi-flexible simulated annealing in torsion angle space and a refinement by MD at a temperature of 300K in explicit solvent [54,69]. Next we selected the 500 top structures in terms of the highest docking score and the lowest RMSD relative to the NMR structure [18]. In these 500 structures, a short MD simulation was performed to optimize the water positions while proteins were fixed (4,000 steps of MD consisting four times 1,000 steps at a temperature of 600, 500, 400 and 300 K, respectively [69]. This was followed by MD at 300K in water [54]. A clusterization procedure was performed with a cut-off at 3.0 Å from the final 500 structures (94% structures were clusterized). We finally selected the structure that had the lowest RMSD (0.8 Å for PCAF and 4 Å for Tat), which is representative for the most crowded cluster. This is the structure discussed in the Results and Discussion Section. Other representative structures are discussed in Supplementary Table S3.

The representative of the Tat.PCAF model from docking underwent MD simulations in explicit water. Tat.PCAF BRD was solvated in 104 × 70 × 66 ų box, containing 39.363 water molecules, 16 chlorine ions were added to neutralize the total charge of the system. The AMBER force field ff99SB [70] was used for the protein and the chlorine atoms; the TIP3P model [71] was used for water. The acetylated lysine topology parameter was taken from Machado et al. [66].

Periodic boundary conditions were used and a cutoff of 12 Å [72] was adopted for short-range non-bonded interactions. The particle mesh Ewald summation method [73] was used for long-range electrostatic interactions. A dielectric constant of one was assumed. All chemical bonds were constrained by using the SHAKE algorithm [74]. The equations of motion were integrated using a time step of 2 fs. Temperature and pressure were kept constant at 300 K and 1 atm by coupling the system to external baths [75] with a Langevin thermostat (coupling constants st = 0.05) and a Berendsen barostat (coupling constants sp = 0.5 ps), respectively.

After 5,000 steps of minimization (steepest descent algorithm for the first 1,500 steps before switching to the conjugate gradient algorithm for the remaining 3,500 steps), the system was equilibrated for 300 ps. Eighty nanoseconds of MD simulation were performed at 300 K and 2 ps for the time steps. We clusterized [67] the trajectory into five clusters. The backbone RMSD of the Tat.PCAF BRD structures over 80 ns is shown in Supplementary Figure S7. NAMD package was used for MD simulation [76].

Hydrophobic contacts at the protein/protein contact surface were defined as two non-polar carbon atoms at a distance of 4.0 Å or less. Hydrogen bonds were defined as a polar hydrogen at a distance of 2.5 Å or less from a polar atom (nitrogen or oxygen). Salt bridges were considered formed if the distance between any of the oxygen atoms of acidic residues and the nitrogen atoms of basic residues were less than 4.0 Å. Coverage was defined as the percentage of the occurrence of HCs over all frames in MD simulation or all representative structures in docking. The HADDOCK [53,54], LIGPLOT [77] and VMD [64] programs were used to calculate these quantities.

We have presented a model of the BRU Tat.PCAF complex produced by data-driven docking combined with MD simulations, which may add structural information to the intricate mechanism of Tat transcription and trans-activation, up to now not fully understood [10,11]. The structures obtained by docking and those obtained by MD are similar in the binding region. This model was obtained with a more accurate procedure than that used in [52] and it focuses on the more common BRU variant rather than the Z2 strain of HIV-1. The two models show conservation as far as the Tat⁴⁶SYGR(AcK)KRRQRC⁵⁶ and N-terminal contacts with PCAF BRD are concerned. In addition, the model suggests the presence, only for the Z2 strain of Tat, of hydrophobic contacts and hydrogen bonds between Tat and PCAF that are not present in the previous model of Z2 Tat.PCAF [52]. Specifically, the BRU Tat.PCAF model differs from the Z2 Tat.PCAF complex on HBs between R49 and P747, E750, between P758 and E2, P3 and on HCs between S46 and P804, Y809.

The model appears to be consistent with most of the experimental data obtained with Tat⁴⁶SYGR(AcK)KRRQRC⁵⁶.PCAF in vitro [18] (synthesized Tat peptide) and full length Tat.PCAF in vivo [52]. However, this comparison has to be interpreted with caution because of the assumption that most of the intermolecular interactions compared here do not vary when passing from the in vitro to the in vivo situation and/or from the Tat⁴⁶SYGR(AcK)KRRQRC⁵⁶ to the full length protein.

Several BRD amino acids have been identified as crucial for Tat binding (E750, V752, E756, A757, Y802, Y809) and as the targets for development of small molecules blocking Tat.PCAF association [78,79]. Our model allows us to suggest that the hydrophobic interactions seem to be a crucial factor for BRD binding to Tat.

Further, our model allows us to suggest that N798 in the BRD plays a key role for Tat binding, independently of whether this protein is expressed by the BRU or the ZZ variant of HIV-1. Indeed, this residue creates a HB with AcK50 in the X-ray structure of Tat and other cellular partners [41,42,49], in both the Z2 Tat.PCAF model [52] and in our model. Hence, interfering with N798-AcK50 interactions may be a valuable strategy for drug design [63,80]. Because this residue is highly conserved (Supplementary Figure S3 and Table 4), it may form key interactions also with other BRD cellular partners [18,31,32,62,81]. Moreover, some studies showed that specific small molecules binding at the acetylated lysine recognition pocket can antagonize the bromodomains and acetylate histone interactions, especially the HB between asparagine in αB loop and acetylated lysine [30,80,82,83,84,85,86,87,88,89]. Therefore, designing compounds, which bind to N798 may also lead to other potential therapeutic applications.