Specific Noncovalent Interactions Determine Optimal Structure of a Buried Ligand Moiety: QM/MM and Pure QM Modeling of Complexes of the Small-Molecule CD4 Mimetics and HIV-1 gp120

Abstract

The small-molecule CD4 mimetics (smCD4mcs) are a class of highly potent HIV-1 entry inhibitors characterized by a unique structure–activity relationship (SAR). They share a halogenated phenyl ring (region 1) that deeply inserts into an otherwise water-filled cavity at the CD4 binding site on the gp120 surface, the so-called F43 cavity. Conservative modifications to region 1 away from this halogenated phenyl motif have all led to loss of activity, despite the fact that they are predicted by standard empirical computational approaches to bind equally well, making it difficult to further optimize this region of the compounds to increase binding to gp120. In this study we used quantum mechanical methods to understand the roots of the interactions between region 1 and the F43 cavity. We clearly demonstrate the presence of halogen bond/σ-hole and dispersion interactions between region 1 and the F43 cavity residues F376–N377, which are not captured by standard molecular mechanics approaches and the role played by the smCD4mc in the F43 cavity desolvation. These findings rationalize why the halogenated region 1 has proven so difficult to move beyond in smCD4mc optimization, in agreement with experimental evidence.

1. Introduction

HIV-1 entry into human target cells begins with the interaction of the envelope glycoprotein complex (“Env”) on the surface of HIV-1 and the human T-cell receptor CD4. Env is a trimer of heterodimers of the viral proteins gp120 and gp41. The specific interaction between gp120 and CD4 initiates the entry process, resulting in exposure of the chemokine co-receptor (CCR5 and/or CXCR4) binding site on the gp120 surface. Subsequent co-receptor binding results in the separation of gp41 from gp120 and the insertion of the gp41 N-terminal fusion peptide into the host cell membrane. Inhibition of the entry process, in particular preventing the initial interaction between gp120 and CD4, is one of the most investigated mechanisms for blocking HIV-1 infection at an early stage, particularly preventing the initial interaction between gp120 and CD4.^[1] Small-molecule CD4 mimetic compounds (smCD4mcs) are a potent class of HIV-1 entry inhibitors that block the interaction between gp120 and CD4 by binding to the F43 cavity, that is, the CD4 binding site on the gp120 surface.^[2–5] SmCD4mcs are characterized by a unique SAR and can be divided into three regions: region 1, a p-chlorophenyl (first generation) or a p-chloro-,m-fluorophenyl (second generation); region 2, an oxalamide linker; and region 3, a variously decorated piperidine ring (first generation) or an indane ring (second generation) (Figure 1).

Figure 1.

smCD4mc scaffold, compound JRC-II-191.

Region 1 is deeply inserted into the F43 cavity, whereas region 2 is at the cavity vestibule, and region 3 is solvent exposed.^{[6, 7]} Our efforts at increasing both gp120 binding and the antiviral potency of smCD4mcs have so far been focused on modifying regions 1 and 3 in order to reach conserved residues among the HIV-1 species, to lead to compounds with a higher breadth of activity. Modifications in region 3 have been successful and are still ongoing, whereas most modifications at region 1 deviating from the p-chloro-,m-fluorophenyl ring decrease potency.^[5,6,8–10] Region 2 was never changed, as it combines an optimal length to link regions 1 and 3 and the ability to make hydrogen bonds between the oxalamide nitrogen atom and the backbone oxygen atom of N425 and G472 of gp120.

In this work, we conducted a computational study to investigate how the structure of smCD4mcs affects affinity to inform design of new inhibitors. Because a cluster of three crystallographic water molecules exists in the F43 cavity in the gp120/CD4 complex that is missing in the gp120/smCD4mc complex,^[11] we first analyzed the hydration sites in the gp120/CD4 complex using WaterMap.^{[12, 13]} We then used quantum mechanics/molecular mechanics (QM/MM) and quantum mechanics cluster models (QM-only) using the programs QSite^{[14, 15]} and Jaguar,^[16] respectively. We used these simulations to investigate the nature of the noncovalent interactions (NCIs) between region 1 of the smCD4mcs and the following F43 cavity residues: S256, S375–N377, F382, and W427. These residues form key interactions with region 1 of smCD4mcs, based on X-ray crystallographic data. The smCD4mcs evaluated in this study contain six different phenyl ring systems at region 1: a p-chlorophenyl (1, NBD-556);^[17] a non-substituted phenyl ring (2, JRC-I-300), a p-nitrophenyl (3, JRC-I-236); a p-fluorophenyl (4, JRC-II-191); a p-chloro-,m-chlorophenyl (5, JRC-II-192), and our optimal moiety p-chloro-,m-fluorophenyl (6, JRC-II-191) (Figure 2).^[6] The latest generation of smCD4mcs were not considered in this study because we were only interested in understanding the impact of region 1 modifications on the binding affinity.^[10]

Figure 2.

smCD4mcs considered in this work.

2. Results

2.1. Analysis of the hydration sites in the gp120/CD4 complex

WaterMap simulations performed on the gp120/CD4 complex revealed the presence of water molecules in the F43 cavity characterized by high free energy, as shown by Figure 3.

Figure 3.

WaterMap free energy density computed on gp120/CD4 complex: red surface, positive free energy contribution; green surface, negative free energy contribution. Compound 1 (green sticks) is shown only to facilitate recognition of the F43 cavity.

WaterMap is able to recover the clusters of three water molecules found already in the gp120/CD4 co-crystal complex, namely HS10, HS40, and HS44 (Figure 4), which is absent in all gp120/smCD4mc co-crystal structures.

Figure 4.

WaterMap hydration sites (represented red to green from high to low energy). Compound 1 (green sticks) is represented only to facilitate recognition of the F43 cavity.

In particular, the enthalpy and the entropy contributions associated with HS10 and HS40 are greater than 0. HS44 is characterized by an entropy contribution greater than 0 and an enthalpy contribution close to 0 (Table 1).

Table 1.

Hydration sites energy and occupancy in gp120/CD4 complex (PDB ID: 1GC1).^a

Water Number	−TΔS [kcalmor−1]	ΔH [kcalmor−1]	ΔG [kcalmor−1]	Occupancy
10	3.81	−0.78	3.03	0.93
40	2.91	3.92	6.83	0.74
44	2.48	3.89	6.37	0.72

^aOnly the three water molecules in the F43 cavity are reported. Water number 10 is closest to F43_CD4. The occupancy range is 0–1.

Taken together, these data suggest that smCD4mcs with a hydrophobic substituent in region 1 might be better water displacers than those with hydrophilic region 1 moieties, as there is likely no gain of attractive enthalpy via hydrophilic interactions with the mostly hydrophobic cavity.

2.2. Docking and MM-GBSA re-scoring

All compounds showed a good docking pose in the gp120 core monomer (the average RMSD with respect to the reference gp120/1 co-crystal pose is 0.4 Å) and similar Glide SP scores; therefore, minor differences in region 1 were not captured by a scoring function. Similarly, a good correlation between the compounds’ MM-GBSA rank ordering and experimental binding affinity was not observed, especially for 3 characterized by a better MM-GBSA value than 4 (Table 2).

Table 2.

Lack of correlation between MM-based calculations, namely docking and MM-GBSA scores, and experimental binding affinity data.

Compd	Glide SP score	MM-GBSA [kcalmor−1]	Exp. binding affinity [kcalmor−1]a[6]
1b	−9.14	−69.03	−7.4
2	−8.45	−58.04	N.B.
3	−8.40	−64.67	N.B.
4	−9.61	−62.96	−6.5
5	−9.01	−66.68	−8.0
6	−9.08	−67.33	−8.3

^aN.B.: no binding.

^bReference compound.

2.3. QM/MM

Because we could not rank order compounds containing different moieties at region 1 accurately with classical MM, we next considered the possibility of the presence of non-standard interactions between smCD4mc region 1 and the F43 cavity. We used QM-based methods to investigate this hypothesis further.

The relative binding affinities DDE (with respect to 1) between each ligand and the F43 cavity residues W427, F382, S375–N377 were estimated with and without considering the contribution of the solvent upon binding (Table 3), and the NCI strengths were computed.^[18] When the solvation penalty associated with the binding was not considered, there was no correlation between the computed ΔΔE and the experimental binding affinity for 3 and 4. Indeed, 3 and 4 show very similar ΔΔE_QM/MM values despite differences in their binding affinities as previously measured by isothermal titration calorimetry (ITC).^[19] The solvation effect upon binding was then considered by performing a QM single-point energy calculation using the PBF implicit solvent model implemented in Jaguar.^[16] The obtained solvation penalty was then included in the ΔΔE_QM/MM according to Equation (4) (see Experimental Section below). As shown in Table 3, compounds 2 and 3 now show the highest ΔΔE_solv,QM/MM, which is qualitatively consistent with their respective experimental ΔG values.

Table 3.

Correlation between QM-based ΔΔ energies and experimental K_d values.

Compd	ΔΔEQM/MM gas phase [kcalmor−1]a	ΔΔEsolv,QM/MM PBF implicit model [kcalmor−1]b	Relative exp. ΔG [kcalmor−1]c	Exp. Kd [μm]c,d [6]
1e	0	0	0	3.7
2	6.6	5.6	N.B.	N.B.
3	3.7	7.7	N.B.	N.B.
4	3.0	2.8	0.9	17
5	−0.3	−0.1	−0.6	1.3
6	−0.1	0.4	−0.9	0.8

^aCalculated according to Equation (2).

^bCalculated according to Equation (4).

^cN.B.: no binding to gp120.

^dPreviously measured by ITC.

^eReference compound.

The computed NCI map suggests an aryl-CH-halogen hydrogen bond and dispersion interactions between smCD4mc region 1 and the π system of the backbone amide bond between N377 and F376. As shown in Figure 5, the majority of the surface contour is shown between the p-chlorophenyl ring and the amide backbone of F376 and N377 for 1 and between the p-chloro-,m-halophenyl group for 5 and 6.

Figure 5.

Compound 6 NCI map. NCIs are represented by sections of surfaces colored according to the contribution (red, attractive; blue, repulsive). All other compounds show the same binding mode as compound 1 (see Supporting Information).

The hypothesis of an aryl-CH-halogen hydrogen bond/σ-hole is also supported by the geometry of the aryl-chloro system and amide backbone and from their distance that is relatively small (~3 Å) in the crystal structure. In addition, dispersion interactions may be involved given the surface contour around the m-halogen atom in 5 and 6 and S375.

2.4. QM/MM and QM residue scan

To quantify the contribution of the F43 cavity residues to smCD4mc region 1 binding, we performed a QM/MM followed by a QM-based residue scan (Tables 4 and 5). Compound 3 was not considered in this analysis, given the lack of binding due to the associated large solvation penalty. Although 2 does not bind to gp120, it was included to serve as negative control. From the QM/MM-based scan, residues S375–N377 have the highest non-standard effect on the relative binding energy. Indeed, the model where the majority of the interaction is lost is model 3 (Table 4), where S375–N377 is back in the MM region. This supports the importance of residues S375–N377 in the interaction of smCD4mc region 1 with the receptor, as previously shown by the NCI map in Figure 5.

Table 4.

ΔΔE_QM/MM residue effect on small molecules binding to gp120. The complexes are represented by the different models as described in the method section.

Compd	Ref. system	ΔΔQM/MM[kcalmor−1]		S375–N377 Model 3c
		W427 Model 1a	F382 Model 2b
1d	0	0	0	0
2	6.6	6.7	6.3	4.8
4	3.0	2.9	4.3	0.7
5	−0.1	−0.6	−0.1	0.5
6	−0.3	−0.6	−0.5	1.2

^aModel in which W427 is in the MM region.

^bModel in which F382 is in the MM region.

^cModel in which S374–N377 are in the MM region.

^dReference compound.

Table 5.

ΔΔE_QM residue interactions on small molecules binding to gp120.

Compd	Ref. system	−W427 Model 1	ΔΔQM [kcalmol−1]a			+F382 Model 5	+S375– N377 Model 6
			−F382 Model 2	−S375– N377 Model 3	+W427 Model 4
1b	0	0	0	0	0	0	0
2	3.16	3.21	2.82	0.37	0	0.40	2.88
4	0.47	0.43	0.94	−0.42	0.07	−0.45	0.90
5	0.28	0.23	0.35	−0.36	−0.27	−0.40	0.29
6	1.24	1.23	1.38	−0.30	−0.12	−0.30	1.36

^aThe residues effect was computed according to Equation (6), in which the complex is represented by the different models as described in the Experimental Section. “−” represents the reference QM system without the specific residue (QM-S1); “+” represents only the ligand together with the specific residue (QM-S2).

^bReference compound.

F382 and W427 do not seem to have a high impact on the relative binding affinity of the ligands inside the series investigated. In particular, comparing the non-standard effect of W427 on all compounds, 5 and 6 have the same relative binding energy, and 4 has the same energy as the respective reference system (Table 4).

For the QM-residue scan, two experiments were performed: 1) considering the same systems as in the QM/MM-residue scan (models 1–3 in Table 5); and 2) considering pair interactions (models 4–6 in Table 5).

The agreement between the QM/MM and the QM-residue scan in the first experiment is shown by model 3 in Table 5, as it is again the one where S375–N377 are not considered and where there is the biggest loss of interactions. When comparing the QM/MM- and QM-residue scans, we notice a shift of ~3 kcalmol⁻¹ for 2 and 4 and of ~1 kcalmol⁻¹ for 5 and 6. We think these shifts are probably due to the differences between the QM/MM and the QM model and particularly the smaller size of the system considered in the QM calculation.

The agreement between QM/MM and the QM-residue scans on the second experiment is shown by models 4 and 5 in Table 5, where the effects coming from W427 and F382 are the only one considered. In this case, again the biggest lost in terms of interactions comes from the absence of S375–N377. Notably, we also see W427 has the same impact on 1 and 4 and almost the same impact on 5 and 6. F382 has a higher impact than W427 when changing from a non-substituted phenyl ring to a p-substituted phenyl ring, and the impact on the p,m-disubstituted compounds is similar.

These results do not imply that W427 and F382 are not important for smCD4mc binding. Those two residues mediate important π–π-type interactions with the region 1 phenyl ring. However, these simulations demonstrate that the π–π-like interactions with W427 and F382 and other hydrophobic contacts within the F43 cavity are not enough to stabilize smCD4mcs in their binding site, but that other non-standard interactions such as halogen bond/σ-hole and dispersion interactions are needed.

3. Discussion

Understanding desolvation and hydrophobic effects should be part of a standard structure-based design approach. Specifically, desolvation costs are also likely why some ligands that seem to be good binders according to docking studies are not confirmed experimentally to be inhibitors.^[20] As we demonstrate here, the careful probing of the hydration sites with WaterMap was key to first rationalize why hydrophobic substituents at region 1 of the smCD4mcs were better tolerated than hydrophilic substituents. This result suggests that the probing of hydration sites should be embedded in a standard structure-based design strategy. The subsequent computation of the desolvation cost that a small molecule should pay upon binding by means of the PBF model in Jaguar gave us a clearer picture of the characteristics that a good gp120 CD4 mimic should have to tightly bind the protein. However, if desolvation is the only driving force for a protein–ligand interaction, all binding events involving hydrophobic interactions would have been entropy driven. In this study, we also demonstrate the importance of other types of interactions that classical MM approaches fail to capture, showing that the binding event associated with the smCD4mcs is most likely enthalpy driven. In particular, according to docking followed by MM-GBSA, 3 was predicted to be a strong gp120 binder, and 2 a much weaker gp120 binder. According to WaterMap, 3 was not predicted to bind gp120, and 2 was predicted to bind it. Experimental data show that none of them bind the protein. QM methods are particularly important to capture dispersion forces which play a significant role in gp120/smCD4mc interactions. In this study, QM-based simulations recovered those types of non-standard interactions that were not captured by classical force-fields methods, specifically the halogen bond/σ-hole and dispersion interactions between smCD4mcs region 1 and F43 cavity residues and the solvation penalty paid upon binding. Those are all factors that explain why we could not deviate from the halogenated-phenyl region 1 moiety without losing binding to gp120. To make our case stronger, we performed a search for halogen bonds between chloro- and/or fluorophenyl-containing ligands that obey the Lipinski rule of five and amino acid backbones using the Protein–Ligand Database (PLDB)^[21] to check how often they are observed in known complexes. We found 848 protein–ligand complexes in the PDB halogen bonding with C=O backbone and 676 halogen bonding with N–H backbone. In addition, in some of the complexes those particular interactions were responsible for the acquisition of properties like increasing potency and specificity toward a certain target. An example of increasing potency is the broad-spectrum bactericide triclosan^[20] and an example of increasing specificity is the MAP Erk kinase inhibitor CI-1040.^[22] Triclosan inhibits the Fab1 enoyl reductase of bacterial fatty acid synthesis (PDB ID: 2PD3).^[23] In the solved complex, it is evident that there is a halogen bond between the triclosan p-chlorophenyl ring and the C=O backbone of A95. An extensive synthetic program directed at replacing the Cl atom with other lipophilic or hydrogen bonding substituents did not lead to compounds with higher potency than the original compound.^[20,23] In the case of specificity, compound CI-1040 binds in a hydrophobic pocket adjacent to the Mg-ATP binding site, which has low sequence similarity with the other known protein kinases. One of the key interactions observed in the pocket is a dipolar interaction between the 4-F atom of CI-1040 and the N–H backbone of S212.^[22] Thanks to increasing computing power, QM methods are becoming faster and more popular mostly for estimating relative binding affinities,^[24] thus becoming more feasible to include in the lead optimization stage. Pure QM methods are also valuable to estimate solvation effects, as we demonstrate here. QM/MM methods combine the high accuracy of QM and the efficiency of MM and are valuable to investigate the effects of different substituents on the binding mode or in the assessment of various scaffolds. One of the major limitations of the QM-based methods is still the lack of proper consideration of entropic effects, as the method does not include ensemble averages. However, their advantages are greater accuracy and the ability to recover non-standard interactions between a protein and a ligand that cannot be reproduced in MM force fields and which can be key to understand how to rationally design modifications to a lead compound. Yet, in cases where the receptor is relatively rigid and the ligands are relatively similar, then the entropic contributions to the ligand binding can be neglected and QM-based methods can be successfully applied to get deeper insight into the origin of the observed binding affinity and to better understand the SAR.

4. Conclusions

We performed a QM-based study that helped us to understand the roots of the interaction between smCD4mc region 1 and F43 cavity residues that have been difficult to rationalize through standard drug design approaches using empirical force fields. In addition, this study helped us to understand the large role played by small molecules in F43 cavity desolvation and the key role of halogen bond/σ-hole and dispersion interactions in stabilizing the smCD4mc in their binding site. Using QM/MM and QM approaches, we rationalized why our efforts in reaching S375 by decorating region 1 with hydrogen bond donor/acceptor groups have been unsuccessful. We can conclude that non-standard interactions are the driving forces for the binding/stabilization of the smCD4mc to the F43 cavity and that the binding to gp120 is enthalpy rather than entropy driven, as shown by Liu et al.^[19] In a prospective view, to optimize the smCD4mc it appears to be difficult to deviate from the aryl-halogen moieties in region 1 unless some protein conformational changes of the pocket are possible. Nevertheless, in the view of de novo design, F376 and N377 are the F43 cavity residues that might be key to target to reach potency and high breadth of activity.

Experimental Section

5.1. WaterMap

The gp120/CD4 complex (PDB ID: 1GC1)^[11] was prepared using the Protein Preparation Wizard.^[25] Missing side chains were added, and all crystallographic water molecules were retained. WaterMap combines 2 ns of MD simulation and trajectory analysis to map the locations and the thermodynamic properties of the hydration sites (water molecules) of a protein. WaterMap is based on the inhomogeneous solvation theory of Lazaridis and Karplus, where enthalpy is taken directly from non-bonded interactions and entropy is computed from a local expansion of spatial and orientational correlation functions.^{[12, 13]} Here, WaterMap was used to map the thermodynamic properties of the hydration sites of the prepared complex gp120/CD4 complex with default settings.^{[12, 13]}

5.2. Docking and MM-GBSA

The gp120/1 complex (PDB ID: 3TGS)^[26] was prepared using the Protein Preparation Wizard at default settings, adding the missing side chains. The gp120 core monomer extracted from 3TGS coordinates was used as receptor for the subsequent docking calculation of 2 to 6. Docking was performed using Glide with the SP scoring function;^[27–29] the oxalamide torsional angle was restrained to 1808 and the number of output poses was increased to 10. The best-ranked docking pose for each compound was then re-scored using MM-GBSA with default settings.^[18,30]

5.3. QM/MM

The prepared 3TGS complex was first subject to QM/MM optimization using M06-2X functional and the LACVP* basis set as implemented in QSite.^[14,15] The QM region included regions 1 and 2 of 1 (hydrogen cap between regions 2 and 3), the side chains of F382 and W427 (hydrogen caps between C_α and C_β) and the full chain from S375 to N377 (hydrogen caps between the C_α and the backbone carbonyl groups). The shell of atoms beyond 10 Å of 1 was frozen and tight convergence criteria were used for this first optimization (energy change: 10⁻⁵, maximum number of cycles: 9999, long-range forces cutoff set to 1000 Å and updated every cycle). To maximize error cancelation in the calculations, all the other complexes were built using this first optimized complex as a template by modifying the ligand accordingly. All subsequent QM/MM calculations were carried out using the same QM region and keeping the same frozen region for all the ligands considered. QM/MM calculations were also performed on free ligands in gas phase, using the same settings as for the protein–ligand complex. The ΔE_QM/MM is defined as the difference in energy between the complex and the sum of the energy of the free target and free ligand [Eq. (1)]:

$$\mathrm{\Delta }{E}_{\text{QM}/\text{MM}}=E_{\text{QM}/\text{MM}}(\text{complex})-E_{\text{QM}/\text{MM}}(\text{ligand})$$ (1)

The relative binding energy ΔΔE_QM/MM is defined with respect to the ΔE_QM/MM of 1 [Eq. (2)]:

$$\mathrm{\Delta }\mathrm{\Delta }{E}_{\text{QM}/\text{MM}}=\mathrm{\Delta }{E}_{\text{QM}/\text{MM}}-E_{\text{QM}/\text{MM}}(1)$$ (2)

To include solvation effects on the binding of the small molecules to gp120, in Jaguar we calculated single-point energies for the unbound compounds using PBF as solvation model.^[16] The relative solvation penalty ΔE_solv is defined with respect to 1 [Eq. (3)]:

$$\mathrm{\Delta }{E}_{\text{solv}}=E_{\text{solv}}(\text{ligand})-E_{\text{solv}}(1)$$ (3)

And the relative binding energy with solvation effects included ΔΔE_solv,QM/MM is defined in Equation (4):

$$\mathrm{\Delta }\mathrm{\Delta }{E}_{\text{solv},\text{QM}/\text{MM}}=\mathrm{\Delta }\mathrm{\Delta }{E}_{\text{QM}/\text{MM}}-\mathrm{\Delta }{E}_{\text{solv}}$$ (4)

5.4. QM/MM-residue scan

Starting from the QM/MM system described above, three protein–ligand models were generated modifying the QM region: 1) model 1 including back in the MM region W427; 2) model 2 including back in the MM region F382; and 3) model 3 including back in the MM region S375–N377. The QM/MM-residue scan was performed in gas phase using the same settings used for the QM/MM calculation, as described above. The ΔΔE_QM/MM contribution was computed according to Equation (2).

5.5. QM-residue scan

The QM-residue scans were divided into two series. In the first series, the models considered were directly comparable to the models studied with the QM/MM-residue scan. Hence, the four models generated were: 1) the full ligand and residues S375–N377, W427 and F382, that was considered as the reference model; 2) model 1 comprising the full ligand, residues S375–N377, and F382; 3) model 2 comprising the full ligand, residues S375–N377 and W427; 4) model 3 comprising the full ligand, W427 and F382. In the second series, pair interactions were computed on three models: 1) model 4 comprising the full ligand and W427; 2) model 5 comprising the full ligand and F382; and 3) model 6 comprising the full ligand and residues S375–N377. Both calculations were performed doing a single-point calculation using Jaguar^[16] with the M06-2X density functional in gas phase and LACVP* basis set taking as reference the models with 1.

The relative binding affinity with respect to 1 in the QM-residue scan was computed analogously to that computed in the QM/MM calculations [Equations (5) and (6)]:

$$\mathrm{\Delta }{E}_{\text{QM}}=E_{\text{QM}}(\text{complex})-E_{\text{QM}}(\text{ligand})$$ (5)

$$\mathrm{\Delta }\mathrm{\Delta }{E}_{\text{QM}}=\mathrm{\Delta }{E}_{\text{QM}}-\mathrm{\Delta }{E}_{\text{QM}}(1)$$ (6)