Computational design of small molecular modulators of protein–protein interactions with a novel thermodynamic cycle: Allosteric inhibitors of HIV‐1 integrase

Abstract

Targeting protein–protein interactions for therapeutic development involves designing small molecules to either disrupt or enhance a known PPI. For this purpose, it is necessary to compute reliably the effect of chemical modifications of small molecules on the protein–protein association free energy. Here we present results obtained using a novel thermodynamic free energy cycle, for the rational design of allosteric inhibitors of HIV‐1 integrase (ALLINI) that act specifically in the early stage of the infection cycle. The new compounds can serve as molecular probes to dissect the multifunctional mechanisms of ALLINIs to inform the discovery of new allosteric inhibitors. The free energy protocol developed here can be more broadly applied to study quantitatively the effects of small molecules on modulating the strengths of protein–protein interactions.

1. INTRODUCTION

Protein–protein interactions mediate many fundamentally important biological processes such as signal transduction and immune response, ¹ while aberrant forms of protein–protein association or aggregation are implicated in several human diseases (e.g., Creutzfeldt–Jakob, Alzheimer's diseases). In recent years, targeting protein–protein interactions (PPI) by small molecular agents has emerged as a viable avenue in modern drug discovery. ² , ³ However, designing small molecules to modulate PPIs is challenging, partly because PPIs typically involve large and more flat interfaces compared with more traditional small molecule drug targets. In the past 10 years molecular dynamics simulation based free energy methods have increasingly been used to better understand the structure, energetics and dynamics of protein–protein interactions. ⁴ , ⁵ , ⁶ , ⁷ , ⁸ Such computational efforts, when combined with experimental assays, could help accelerate the design of more potent and selective small molecules targeting PPIs. We have recently applied umbrella sampling simulations in explicit solvent to determine the free energy landscape and thermodynamics of the protein–protein interaction responsible for the aberrant multimerization of HIV‐1 integrase. ⁹ While that study successfully explained how small molecules modulate PPIs underlying the association of the Catalytic Core (CCD) and C‐terminal (CTD) domains of two HIV‐1 integrase monomers, simulating the reversible process of protein–protein association using umbrella sampling is computationally expensive and not well suited for optimizing the effects of chemical modifications of the small molecules on the strength of the PPI. Here we use a novel thermodynamic cycle to more efficiently estimate the effects of different small molecules on modulating protein–protein interactions involved in the HIV‐1 integrase association of the CCD and CTD domains and design new molecules to probe the complex mechanism of the allosteric inhibitors of HIV‐1 integrase.

HIV‐1 integrase (IN) has been a major antiviral drug target since the early 2000s, when the first integrase strand transfer inhibitor (INSTI) Raltegravir ¹⁰ was approved by the FDA. IN catalyzes 3′‐processing and strand transfer reactions ¹¹ , ¹² , ¹³ to integrate the HIV DNA into the host genome. The protein consists of three independently folded domains connected by linker peptides: the N‐terminal domain (NTD), catalytic core domain (CCD), and the positively charged C‐terminal domain (CTD). Allosteric HIV‐1 integrase (IN) inhibitors (ALLINIs) bind at the LEDGF/p75 (lens epithelium‐derived growth factor/transcriptional co‐activator p75) binding pocket on the CCD dimer interface. LEDGF/p75 is a host factor that interacts with the CCD of IN and mediates the integration of the viral DNA into the host chromatin. ¹⁴ , ¹⁵ ALLINIs have a multifunctional mechanism of action: originally designed to bind CCD to disrupt the LEDGF/p75‐IN interaction and block the viral DNA integration, recent studies have shown that ALLINIs also promote aberrant IN multimerization with a greater potency during the late phase of viral replication. ¹⁶ This late stage activity of ALLINIs is achieved by mediating the protein–protein interactions between CCD and the CTD of an external IN molecule (Figure 1), leading to the formation of aberrant, higher‐order IN multimers, which impairs virus core maturation. ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ To delineate the different mechanisms by which ALLINIs act, derivatives that selectively target the aberrant multimerization have been designed. ²¹ More recently, Bonnard et al. have found that a thiophene‐based ALLINI MUT‐A exhibits reduced activity for aberrant multimerization on an Ala‐125 IN variant while retains the ability to inhibit the IN‐LEDGF/p75 interaction. ²² However, currently, there are no known ALLINIs that inhibit wild‐type HIV‐1 IN only by disrupting the LEDGF/p75‐IN interaction in the early stage of the life cycle, and outstanding questions regarding the specific roles of how ALLINIs function in the early stage versus late stage of the infection cycle remain to be better understood. By dissecting the early stage and late stage inhibition activities separately, it could facilitate the design of new ALLINIs with improved drug resistance profiles.

FIGURE 1.

A representative MD snapshot of the complex of CTD and CCD dimer mediated by the ALLINI BI224436 (sphere)

Using protein–protein docking and simulation we had previously successfully predicted the role of ALLINI binding at the LEDGF/p75 binding site in promoting the aberrant IN multimerization. ¹⁷ Our recent free energy simulations study was able to account for the effects of different engineered amino acid mutations on the inter‐subunit interactions between CCD and CTD. Building on this previous work, here we construct a new thermodynamic cycle and use it in free energy calculations to design new analogs of ALLINIs that specifically bind CCD and displace LEDGF/p75, without inducing higher‐order IN multimerization. These new analogs can serve as leads in the search for new ALLINs that act in the early stage of the HIV replication life cycle and avoid the drug‐resistance issues that are hampering the existing ALLINIs. ²³ The new free energy thermodynamic cycle leads to significantly smaller error bars in the estimated relative protein–protein binding free energy compared to those obtained by absolute binding free energy calculations. ⁹ The method employed here is expected to be broadly applicable for computing efficiently the effects of small molecules on the modulation of protein–protein interactions.

2. RESULTS AND DISCUSSION

2.1. Design of ALLINI analogs to probe the complex functionalities of the allosteric inhibitors

Initially designed to disrupt the IN‐LEDGF/p75 interactions in the early stage of the life cycle, ²⁴ , ²⁵ more recent studies have shown that ALLINIs also strongly promote aberrant IN multimerization during the late phase of viral replication. ²⁶ , ²⁷ , ²⁸ , ²⁹ , ³⁰ BI‐224436 (Figures 2 and 3) is one of the most potent ALLINIs (EC₅₀ = 15 nM) and the only ALLINI to date that went into clinical trial. As shown in Figure 2, the ligand contains several functional groups that contribute to the following intermolecular interactions in the multimer interface: (a) the polar carboxyl‐t‐butoxy group interacts with one of the CCD through multiple hydrogen bonds (red dashed box); (b) the nonpolar tricyclic group is buried in the hydrophobic pocket of the other CCD (green dashed box); (c) the nonpolar quinoline group interacts with the CTD residues TYR226, TRP235 and ILE268; and (d) the positively charged LYS266 on the CTD forms a salt bridge with the carboxyl moiety of the BI‐224436, and interacts with the quinoline group via a cation‐π interaction. Most of the known ALLINIs share the same architecture of BI‐224436.

FIGURE 2.

Interaction motifs in the aberrant IN multimer interface involving CTD (upper), BI‐224436 (center) and CCD dimer

FIGURE 3.

Chemical structures of BI‐224436 and GSK1264 and the atom numbering in the quinoline ring

In order to develop molecular probes that can dissect the dual functions of ALLINIs in inhibiting the integration and viral maturation, we designed ALLINI analogs such that (a) they retain the ability to bind CCD and compete with LEDGF/p75 in the early replication cycle, and (b) in the late stage the ALLINI‐bound CCD have much reduced inter‐subunit interactions with an external CTD and hence drastically decreased potency for IN multimerization. Based on the interaction motif of the BI‐224436 in the IN multimer interface between the CCD dimer and the external CTD (Figure 2), the BI‐224436 interacts with the CTD through hydrophobic interactions between the quinoline group and TYR226, TRP235 and ILE268, and a salt bridge between the carboxyl group and LYS266. This observation, together with the fact that the engineered amino acid substitutions from nonpolar to polar residues, for example, Y226D and Y226R, block the aberrant multimerization, lead us to design a series of BI‐224436 analogs by adding polar or charged substituent groups on the quinoline ring centered at or near the 7‐th position (Figure 3), which faces the nonpolar CTD residues TYR226, TRP235 and ILE268, and also adjacent to the positively charge LYS266: see Fig. 2. The polar/charged substituents introduced at the C7 position are expected to cause a large desolvation penalty when buried within the interface between the CCD dimer and the external CTD; adding a positively charged center to the region will also repel electrostatically the like charges on the LYS266 and weakens the protein–protein interaction responsible for the multimerization. In the meantime, in the absence of the CTD, the substitution groups at this position of the quinoline ring are exposed to the solvent and hence will not adversely impact the ALLINI binding with the CCD dimer: see Figure 3.

2.2. A thermodynamic cycle for computing small molecule modulation of protein–protein interaction

To verify computationally that the designed ALLINI analogs retain the ability to bind with high affinity to the CCD dimer and thus disrupt the interaction between LEDGF/p75 with the CCD dimer while minimizing the protein–protein interaction between one CCD dimer and an external CTD, we utilize the thermodynamic cycle shown in Figure 4 to compute the change in the protein–protein association free energy $\mathrm{\Delta \Delta }{G}_{A\to B}^{P1:P2}\equiv \Delta G_B^{P1:P2}-\Delta G_A^{P1:P2}$ caused by “mutating” small molecule A to B. In this application, P1 stands for CTD, P2 stands for CCD dimer, A represents the starting ALLINI compound and B the modified inhibitor. Note that since ALLINIs carry functional groups that mimic the LEDGF/p75 interactions with the CCD, in solution, they bind tightly to CCD with nanomolar affinity, but have negligible binding affinity by itself for the CTD. Therefore, in the thermodynamic cycle of Figure 4, such species as P1:ligA and P1:ligB are not shown as their concentrations are vanishingly small.

FIGURE 4.

The thermodynamic cycle used for computing the change to the protein–protein binding free energy caused by converting the small molecule A to B

While $\Delta G_B^{P1:P2}-\Delta G_A^{P1:P2}$ can be computed by two absolute binding free energy calculations (i.e., potentials of mean force) using the two horizontal legs of the thermodynamic cycle in Figure 4, as we have recently reported for the ALLINI‐induced HIV‐1 integrase multimerization, ⁹ computing absolute protein–protein binding free energy requires sampling reversibly the full binding process of two proteins in aqueous solution, which is computationally demanding. Instead, $\Delta G_B^{P1:P2}-\Delta G_A^{P1:P2}$ can be obtained by the two vertical legs in Figure 4, that is, $\Delta G_B^{P1:P2}-\Delta G_A^{P1:P2}=\mathrm{\Delta }{G}_{A\to B}^{P1:P2}-\mathrm{\Delta }{\mathrm{G}}_{\mathrm{A}\to \mathrm{B}}^{\mathrm{P}2}$. The computation of the latter two terms is achievable by reversibly converting ligand A to ligand B in two FEP calculations, one for the two ligands bound in the protein–protein complex and the other with the CCD dimer (protein 2), since the binding affinities of the two ligands for the CTD (protein 1) are negligible. As shown below, since the FEP calculations corresponding to the two vertical legs $\mathrm{\Delta }{G}_{A\to B}^{P1:P2}$ and $\mathrm{\Delta }{\mathrm{G}}_{\mathrm{A}\to \mathrm{B}}^{\mathrm{P}2}$ do not involve simulating the protein–protein unbinding process, the convergence of their free energy difference is significantly faster than the direct potential of mean force calculations through umbrella sampling simulations. ⁹ While FEP is widely used for computing relative binding free energies between different small molecule ligands to a protein target, here we use FEP for the first time for estimating the relative protein–protein binding free energy caused by different small molecule ligands that bind at the interface.

2.3. Computational validation of the designed ligands

The quinoline‐based compounds BI‐224436 ²⁵ and GSK1264 exhibit the highest potency among the known ALLINIs. While BI‐224436 is the first ALLINI to advance into a phase Ia clinical trial, ²⁵ GSK1264 is the first and only crystallized complex with full‐length HIV‐1 IN. ¹⁸ Here, we have designed a series of derivatives based on BI‐224436 and GSK1264 by attaching polar/charged groups to C7 of the quinoline core or nearby positions: see Figures 5 and 6.

FIGURE 5.

Structures of the analogs of BI‐224436

FIGURE 6.

Structures of the analogs of GSK1264

We have tested this hypothesis by using the thermodynamic cycle in Figure 4 to compute the free energy changes caused by the modifications in the binding of the new compounds with the CCD dimer (denoted as CCD₂), and with the CCD₂‐CTD multimer interface. Table 1 shows the changes in the binding free energy for the CCD dimer, which are computed using the FEP cycle that corresponds to the left half of Figure 4:

$$\begin{gathered} \mathrm{CC}{\mathrm{D}}_2+\text{ALLINI}\stackrel{{\mathrm{\Delta G}}_{\text{ALLINI}}^{\mathrm{CC}{\mathrm{D}}_2}}{\to }\mathrm{CC}{\mathrm{D}}_2\bullet \text{ALLINI} \\ \uparrow -{\mathrm{\Delta G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{SOL}}\downarrow {\mathrm{\Delta G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2} \\ \mathrm{CC}{\mathrm{D}}_2+\text{ALLIN}{\mathrm{I}}_{\mathrm{new}}\stackrel{{\mathrm{\Delta G}}_{\text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2}}{\to }\mathrm{CC}{\mathrm{D}}_2\bullet \text{ALLIN}{\mathrm{I}}_{\mathrm{new}} \\ \mathrm{\Delta \Delta }{\mathrm{G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2}\equiv {\mathrm{\Delta G}}_{\text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2}-{\mathrm{\Delta G}}_{\text{ALLINI}}^{\mathrm{CC}{\mathrm{D}}_2} \\ ={\mathrm{\Delta G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2}-{\mathrm{\Delta G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{SOL}} \end{gathered}$$

TABLE 1.

Relative binding free energy between the starting ALLINIs and new analogs for binding to CCD dimer

BI‐224436 →	${\mathrm{\Delta \Delta G}}_{\text{ALLINI}\to {\text{ALLINI}}_{\mathrm{new}}}^{{\mathrm{CCD}}_2}$ (kcal/mol)	GSK1264 →	${\mathrm{\Delta \Delta G}}_{\text{ALLINI}\to {\text{ALLINI}}_{\mathrm{new}}}^{{\mathrm{CCD}}_2}$ (kcal/mol)
B1	−1.21 ± 0.42	G1	−0.34 ± 0.40
B2	0.83 ± 0.40	G2	0.38 ± 0.40
B3	2.15 ± 0.43	G3	0.99 ± 0.41
B4	0.80 ± 0.43	G4	0.92 ± 0.40
B5	−0.66 ± 0.43	G5	−0.50 ± 0.42
B6	−0.29 ± 0.42	G6	−0.06 ± 0.41
B7	0.65 ± 0.40	G7	−2.81 ± 0.41

As can be seen from Table 1, the changes caused by the additions to C7 or nearby positions to the binding affinity for the CCD dimer are generally benign. In the case of G7, the addition of the positively charged piperazine improves the binding to the CCD dimer by −2.8 kcal/mol, or 100‐fold increase in the binding constant. The only modification that led to a greater than 2 kcal/mol decrease in the binding free energy is in the case of B3, where an amino group is added to the tricyclic group. These results confirm our hypothesis that additions to the region centered around the C7‐position will not substantially affect the ALLINI binding to the CCD dimer.

Based on the results of Table 1, we have selected the four most promising analogs of the BI‐224436 and GSK1264 (Table 2) and compute the effects of the modifications on the thermodynamic stability of the protein–protein interface between CCD dimer and the CTD, using the FEP cycle in the right half of Figure 4:

$$\begin{gathered} \mathrm{CC}{\mathrm{D}}_2\bullet \text{ALLINI}+\mathrm{CTD}\stackrel{{\mathrm{\Delta G}}_{\text{ALLINI}}^{\mathrm{CC}{\mathrm{D}}_2\bullet \mathrm{CTD}}}{\to }\mathrm{CC}{\mathrm{D}}_2\bullet \text{ALLINI}\bullet \mathrm{CTD} \\ \uparrow -{\mathrm{\Delta G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2}\downarrow {\mathrm{\Delta G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2\bullet \mathrm{CTD}} \\ \mathrm{CC}{\mathrm{D}}_2\bullet \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}+\mathrm{CTD}\stackrel{{\mathrm{\Delta G}}_{\text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2\mathrm{CTD}}}{\to }\mathrm{CC}{\mathrm{D}}_2\bullet \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}\bullet \text{CTD} \\ \text{ΔΔ}{\mathrm{G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2\bullet \mathrm{CTD}}\equiv {\mathrm{\Delta G}}_{\text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2\bullet \mathrm{CTD}}-{\mathrm{\Delta G}}_{\text{ALLINI}}^{\mathrm{CC}{\mathrm{D}}_2\bullet \mathrm{CTD}} \\ ={\mathrm{\Delta G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2\bullet \mathrm{CTD}}-{\mathrm{\Delta G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2} \end{gathered}$$

TABLE 2.

Changes in the CCD₂‐CTD protein–protein binding free energy caused by the ALLINI analogs relative to the unmodified ALLINIs. As comparison, the changes in the binding free energy between ALLINIs and new analogs for the binding to CCD dimer are also shown $\mathrm{\Delta \Delta }{\mathrm{G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2\bullet \mathrm{CTD}}\equiv \mathrm{\Delta }{\mathrm{G}}_{\text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2\bullet \mathrm{CTD}}-\mathrm{\Delta }{\mathrm{G}}_{\text{ALLINI}}^{\mathrm{CC}{\mathrm{D}}_2\bullet \mathrm{CTD}}=\mathrm{\Delta }{\mathrm{G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2\bullet \mathrm{CTD}}-\mathrm{\Delta }{\mathrm{G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2}$

	$\mathrm{\Delta \Delta }{\mathrm{G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2}$ (kcal/mol)	$\mathrm{\Delta \Delta }{\mathrm{G}}_{\text{ALLINI}\to \text{ALLIN}{\mathrm{I}}_{\mathrm{new}}}^{\mathrm{CC}{\mathrm{D}}_2\bullet \mathrm{CTD}}$ (kcal/mol)
BI‐224436 → B5	−0.66 ± 0.43	5.09 ± 0.42
BI‐224436 → B7	0.65 ± 0.40	12.12 ± 0.43
GSK1264 → G5	−0.50 ± 0.42	1.83 ± 0.40
GSK1264 → G7	−2.81 ± 0.41	2.82 ± 0.42

Table 2 shows that compounds B5, B7, and G7 result in the largest decrease in the protein–protein interaction between the CCD dimer and CTD. For example, B7 which has an added positively charged piperazine group resulted in a destabilization of the aberrant multimer interface by 12.12 kcal/mol, which translates into an 8–9 orders of magnitude decrease in the equilibrium constant for the CCD‐CTD protein–protein interaction. Therefore, the results in Table 2 suggest that B5, B7, and G7 are promising ALLINI analogs that can serve as inhibitors that act specifically in the early stage of the virus life cycle, as they show the largest decrease in the stability of the CCD₂‐CTD multimer interface, and the smallest decrease or even a sizeable increase in the case of G7, in their binding affinity for the CCD dimer.

The free energy simulations also revealed small conformational rearrangements caused by the binding of the new analogs at the CTD‐CCD₂ surface. For example, in the structures of the CCD₂‐CTD complex containing the charged piperazine substituent on the B7 (Figure 7) the interfacial residues Tyr226 and Trp235 of CTD is pushed away from the CCD dimer by about 1.6 Å, compared with the corresponding structures containing the unmodified BI‐224436. This can be seen by comparing the location of Trp235 in the CCD₂‐B7‐CTD complex and that in the CCD₂‐BI‐224436‐CTD complex when CCD dimers in the two structures are aligned. Furthermore, the quinoline ring in the CCD₂‐B7‐CTD complex moves away from the binding pocket and is shifted towards the solvent by 1.8 Å compared with that in the CCD₂‐BI‐224436‐CTD complex. These observations reflect the destabilization to the CCD₂‐CTD inter‐subunit interface caused by the introduction of polar/charged centers to the interface. These conformational changes at the interface correlate with the destabilization of the CCD₂‐ALLINI‐CTD interactions seen from Table 2. Taken together, the FEP results are consistent with experimental observations that nonpolar to polar amino acid substitutions in the region of the CCD₂‐CTD inter‐subunit interface near the ALLINI binding pocket tend to diminish the drug‐induced multimerization. For example, Y226D, Y226R, and W235A mutations which reduce the hydrophobicity of the surface of the CTD that are in contact of the CCD, are all found to result in almost complete abrogation of the protein aggregation. ¹⁸

FIGURE 7.

Overlaid structures of CCD dimers bound to BI‐224436 (blue stick) and B7 (green stick). The CCD dimers are shown in light gray ribbon. The CTD bound to BI‐224436 is shown in blue ribbon, and the CTD bound to B7 is shown in green ribbon. The CTD residues Tyr226 and Trp235 are shown in green and blue lines

Lastly, we discuss the significant improvement in the computational efficiency which resulted from using the new thermodynamic cycle of Figure 4 when computing the changes to the protein–protein binding free energy. In our previous study of the relative binding free energy of protein–protein binding using umbrella sampling, a total of 4 × 720 ns = 2.88 μs of MD simulation time is required to converge the relative free energy of the CCD₂‐ALLINI‐CTD complexation to an uncertainty of 1.4–2.8 kcal/mol. In this work, as shown in Table 2, the calculated relative free energies are converged to within ~±0.4 kcal/mol using just 465 ns simulation time per protein–protein complex.

3. CONCLUSION

We have presented a thermodynamic free energy cycle to efficiently estimate the effects of small molecule modulators of protein–protein interactions and applied it to design a series of quinoline‐based ALLINI analogs that retain their strong affinity to the LEDGF/p75 binding pocket in the CCD dimer of HIV‐1 integrase, without inducing the IN‐IN interactions that leads to aberrant IN multimerization. The newly designed compounds will be useful to act selectively in the early‐stage of the HIV‐1 life cycle, as opposed to the existing ALLINIs that also promote multimerization and primarily act in the late stage of the infection cycle. These compounds can serve as molecular probes to delineate the complex functions of ALLINIs and aid the rational design of next generation ALLINIs to inhibit HIV‐1 integrase with improved drug resistance profiles. The novel thermodynamic cycle presented here can be applied broadly for estimating the effects of small molecules on the modulation of protein–protein interactions.

4. METHODS AND MATERIALS

4.1. System setup and simulation details

The starting structure of GSK1264‐CCD₂ is taken from IN^F185K(CCD₂)·GSK1264 crystal structure (PDB 4OJR) ³¹ and the starting structure CTD‐GSK1264‐CCD₂ is from the crystal structure of full‐length HIV‐1 IN/ GSK1264 complex with PDB ID 5HOT. ¹⁸ The starting structure of BI‐224436‐CCD₂ is taken from the crystal structure with PDB ID 6NUJ and the starting structure of CTD‐BI‐224436‐CCD₂ is taken from the previous study. ²⁰

Molecular dynamics simulations were performed with GROMACS version 2016.3. ³² The HIV‐1 integrase (CCD dimer or CCD₂‐CTD multimer) were modeled with the ff14SB force field, ³³ ligands with the General Amber force field (GAFF) ³⁴ and the AM1‐bcc charge models. ³⁵ Integrase was placed at the center of a rhombic dodecahedral box and the distance between the solute and the edge of the box was set to be ≥1.0 nm. For ligand perturbations in the CCD dimer, a 10.1 nm × 9.2 nm × 8.4 nm rectangular box containing 20,595 TIP3P water molecules and three Na⁺ was used. The total number of atoms in the simulation box is ~66,000. For ligand perturbations in the CCD₂‐CTD complex, a bigger box with the dimension 8.6 nm × 11.1 nm × 9.9 nm containing 25,095 TIP3P water molecules and five Cl⁻ was used; and the entire system contains ~81,000 atoms. For ligand perturbations in water, the ligand was solvated in a 5.6 nm × 6.2 nm × 6.5 nm box containing 6,105 water molecules and one Na⁺. The number of alchemical λ windows for ligand perturbations was set to be 31, including 21 van der Waals (VdW) λ and 10 coulombic λ. When an ALLINI is alchemically mutated to its charged analog (compounds B1/G1 and compounds B7/G7), one Na⁺ ion is counter‐mutated to water to keep the simulation system charge neutral throughout the alchemical free energy simulation. ³⁶

For each individual λ state, an energy minimization was performed to relax the initial system, followed by 100 ps equilibration of NVT ensemble and 1 ns equilibration of NPT ensemble. The production run in the NPT ensemble was performed for 15 ns at a temperature of 300 K and pressure of 1 atm. The temperature was scaled by the modified Berendsen thermostat. ³⁷ In an NPT ensemble, the pressure was scaled to 1 bar with a compressibility of 4.5 × 10⁻⁵ bar by the Parrinello‐Rahman barostat. ³⁸ The electrostatic interactions were treated with the particle mesh Ewald (PME) ³⁹ method with a real‐space cutoff of 1.0 nm. MD simulations were performed with a time step of 2 fs and trajectory files were saved every 2 ps.

AUTHOR CONTRIBUTIONS

Qinfang Sun: Conceptualization; data curation; formal analysis; funding acquisition; investigation; methodology; project administration; resources; software; validation; visualization; writing‐original draft; writing‐review and editing. Vijayan Ramaswamy: Conceptualization; investigation. Ronald Levy: Conceptualization; investigation; methodology; project administration; resources; software; supervision; writing‐review and editing. Nanjie Deng: Conceptualization; data curation; formal analysis; funding acquisition; investigation; methodology; project administration; resources; software; supervision; validation; visualization; writing‐original draft; writing‐review and editing.

Sun Q, Ramaswamy VSK, Levy R, Deng N. Computational design of small molecular modulators of protein–protein interactions with a novel thermodynamic cycle: Allosteric inhibitors of HIV‐1 integrase. Protein Science. 2021;30:438–447. 10.1002/pro.4004