Protein Engineering vol.15 no.9 pp.707–711, 2002
Modified solvent accessibility free energy prediction analysis of cyclic urea inhibitors binding to the HIV-1 protease
Fredy Sussman1, M.Carmen Villaverde and Luis Martı´nez Departamento de Quı´mica Orga´nica, Facultad de Quı´mica, Universidad de Santiago de Compostela, 15782 Santiago de Compostela, Spain 1To
whom correspondence should be addressed. E-mail: [email protected]
One of the most successful drug targets against AIDS in the last decade has been the HIV-1 protease (HIV-1 PR), an enzyme that processes the polyprotein gene products into active replicative viral proteins. In our quest for a wideranging, binding free energy function we have extended the solvent accessibility free energy predictor (SAFE_p) method, recently developed for peptidic HIV-1 PR inhibitors, to the study of the binding of cyclic urea (CU) HIV1 PR inhibitors. Our results show that there is a need for a specific term depicting polar contacts to be added to the original SAFE_p analytical expression, an outcome not seen in our studies of HIV-1 PR peptidic inhibitors. Nevertheless, despite the higher profile of the electrostatic interactions in the binding of the CU inhibitors, our analysis indicates that CU inhibitor binding is still driven by the hydrophobic entropic contribution, as much as for the peptidic inhibitors. Keywords: binding driving forces/ binding free energy prediction/cyclic urea inhibitors/ electrostatic interactions/HIV-1 protease
Introduction Many research laboratories have developed HIV-1 protease (HIV-1 PR) inhibitors that have become lead compounds for anti-viral drugs against AIDS (Wlodawer and Erickson, 1993). Some of these compounds have been approved for therapeutic use, such as the drugs saquinavir, ritonavir, indinavir and nelfinavir (Wlodawer and Vondrasek, 1998). It has been found that some of the most effective therapies include combinations of HIV-1 PR inhibitors and HIV-1 reverse transcriptase (HIV-1 RT) inhibitors (Perelson et al., 1997). There is still a search for more potent inhibitors that are less sensitive to strains of this enzyme, called resistance mutants, whose selection is due to the evolutionary pressure of the drugs and the high error prone replication of the HIV gene (Erickson and Burt, 1996). The search for these compounds has led to the discovery of a large variety of non-peptidic inhibitors which include the cyclic urea moiety (CU) inhibitors. These compounds were designed to displace the structurally bound water molecule observed in the complexes between HIV-1 PR and peptidic inhibitors (Lam et al., 1994). There is now a large number of CU derivatives known to be highaffinity HIV-1 PR inhibitors (De Lucca and Lam, 1998; Debnath, 1999). The design of new and more potent HIV-1 PR inhibitors, as well as understanding the effect of the resistance mutations on them, requires a free energy function that will afford us a © Oxford University Press
predictive tool for the affinity of these compounds as well as a structure to binding affinity relationship (Horton and Lewis, 1992; Nauchitel et al., 1995; Wallqvist et al., 1995; Debnath, 1999; Weber and Harrison, 1999; Shimada et al., 2000). The most basic kind of function is a simple quantitative structure activity relationship (QSAR) based solely on the chemical structure of the inhibitor, like the one performed recently on CU inhibitors (Debnath, 1999). The next level of complexity is provided by the evaluation of interaction energy between the ligand and the enzyme (Weber and Harrison, 1999). Other groups (including us) have developed free energy functions that include solvation terms, a very important energetic component of the inhibitor binding affinity. We have developed a very simple function which we have named the solvent accessibility free energy of binding predictor (SAFE_p; Nauchitel et al., 1995; Sussman et al., 1998). The basic assumption is that the free energy of an inhibitor transfer from water to the binding pockets of HIV-1 PR is analogous to the process of transfer from a medium of higher polarity to one of lower polarity, since most residues that line the active site and specificity pockets of this enzyme are hydrophobic in nature. The free energy function fitted closely the observed free energies of binding for a series of known inhibitors. The additive nature of this approach enabled us to partition the free energy of binding into the contributions of single fragments. The resulting analysis allowed us to rank the importance of the enzyme’s subsites for binding: although all the enzyme’s pockets contribute to binding, the ones that bind the P2-P2⬘ span of the inhibitor are in general the most critical for high inhibitor potency. Moreover, perusal of the energy contributions of single side chains has shown a broad specificity for some of the inhibitor fragments located in the central portion of the HIV-1 PR inhibitors. These observations are in agreement with experimental data, providing a validation for the physical relevancy of our method (Nauchitel et al., 1995). The earlier applications of the SAFE_p method dealt with predicting inhibitor affinity of peptidic inhibitors for the native (Nauchitel et al., 1995) and some mutant HIV-1 PR strains (Sussman et al., 1998). The aim of this work is to test the applicability of the SAFE_p approach to a set of CU HIV-1 PR inhibitors differing in affinity by three orders of magnitude. We show in this work that the SAFE_p method could be extended to the prediction of the binding of the CU inhibitors, provided a polar contact term is added to take into account the electrostatic interactions. In spite of the need for an intrinsic term for polar interactions, we show that binding of CU-based inhibitors is still driven by entropic contributions, in an analogous way to the peptidic inhibitors studied earlier. Finally, we show that the modified SAFE_p method is able to predict the ranking affinity of a set of peptidic inhibitors as well or better than its predecessor. Materials and methods Methodological approach In what follows we summarize the details of our algorithm for the prediction of the inhibitor affinity to the HIV-1 PR 707
F.Sussman, M.C.Villaverde and L.Martı´nez
presented earlier (Nauchitel et al., 1995), which we have named SAFE_p. Our algorithm bears some resemblance to other contact functions used for binding free energy predictions (Horton and Lewis, 1992). The basic assumption is that the free energy of an inhibitor transfer from water to the binding pockets of HIV-1 PR is analogous to the process of transfer from a medium of higher polarity to a one of lower polarity, since most residues that line the active site and specificity pockets of this enzyme are hydrophobic. To investigate this hypothesis we have generated a simple algorithm containing two terms which fitted closely the observed free energies of binding for a series of known inhibitors. The SAFE_p function can be written as: ∆G/RT ⫽ k1Σ∆gi ⫹ k2Σ[abs(Qi) / (Roi ⫹ Rw)]∆gi
where the first term characterizes the transfer free energy of each atom regardless of its partial charge, and the second term provides a correction to the free energy that depends on the atom charge to radius ratio; ∆gi characterizes the change of accessibility per atom, Qi is the partial charge of atom i, Roi is its van der Waals radius, and Rw is the radius of the water molecule. Equation 1 was derived with the help of the widely used rigid body approximation. In the original work the values of k1 and k2 were fitted to reproduce the experimental binding energies of a series of peptidic inhibitors. The resulting expression was able to reproduce the binding free energies of inhibitors differing in binding by three orders of magnitude. Analysis of the resulting expression allowed us to obtain the critical charges for which ‘burial’ in the enzyme environment is penalized. The results showed that polar atoms have to pay an energy penalty upon binding. Modified SAFE_p function. The hydrogen bond contribution The original SAFE_p function lacks an explicit term for hydrogen bonds and other electrostatic interactions. Since the SAFE_p is basically a knowledge-based contact function, we decided to use a contact function to account for polar interactions (e.g. hydrogen bond). For this purpose we used a hydrogen bond function with a distance and a orientation criteria implemented in the CHARMm suite of programs (Brooks et al., 1983). This kind of function has been used previously together with a van der Waals term in the qualitative analysis of binding of a series of HIV-1 PR CU inhibitors (Ala et al., 1998). The parameters were obtained from the CHARMm force-field in InsightII (Accelrys Inc., San Diego, CA). The modified SAFE_p equation that includes polar terms can be written as follows: ∆G ⫽ ∆GSAFE_p ⫹ k3∆Gpolar ⫹ K
where ∆Gpolar is the contact free energy for polar interactions described above; k3 and K are constants. In some free energy ranking functions a constant is added that represents entropy terms that account for the loss of translational and rotational degrees of freedom that occur upon binding of an inhibitor to the enzyme, as well as a cratic entropy correction (Krystek et al., 1993). In our case, the constant K results from fitting of the above equation to the experimental binding free energies, and may contain other physical components, and as such it can be better described as a scaling factor between experimental and observed affinities. 708
Calculations For our calculations on CU inhibitors we used the structures of these compounds bound to the HIV-1 PR, obtained by X-ray crystallography [Protein Databank entries 1AJX, 1HVR, 1DMP, 1QBS, 1HWR, 1QBU, 1QBR and 1QBT; Ba¨ ckbro et al., 1997; Jadhav et al., 1997; Ala et al., 1998]. The addition of hydrogen atoms to each complex was carried out with the help of the Biopolymer and Builder modules through the InsightII molecular visualization and handling suite of programs (Accelrys Inc.). To calculate electrostatic interactions (e.g. hydrogen bonds) it is necessary to determine the protonation states of the polar groups that interact in the binding between the enzyme and the inhibitor. In our work on binding of peptidic inhibitors, we ascribed a doubly ionized state to the Asp residues that participate in the hydrolysis (Asp25 and Asp25⬘). Recent work on the binding of some CU inhibitors (Trylska et al., 1999) seems to call for a re-assessment of the doubly protonated state for both Asp residues proposed in an earlier work (Yamazaki et al., 1994). Since our present studies aim at comparing a series of CU analogues with the same central CU moiety, we have decided to keep the same protonation state in the studies presented here as the one used in earlier work on peptidic inhibitors (Nauchitel et al., 1995). Some of the CU inhibitors have polar groups in the periphery of the inhibitor. For instance the inhibitor DMP450 (see Table I) has an aniline group at the S2-S3 pockets. The protonated aniline in solution is acidic (has a pKa ~5.0). Nevertheless, the amino group of the aniline lies in a polar micro environment making electrostatic contacts with the carboxylate and amino groups of residues Asp30/30⬘. These interactions should rise the pKa of the aniline fragment considerably. We have carried out the calculations both with the protonated and neutral DMP450 inhibitor, and the best binding is provided by the protonated ligand, whose results are listed in Table I. The pKa value of the benzo-imidazolium cation is ~6.0 (Gilchrist, 1992); we have calculated the affinity of inhibitor SD146 both when the benzoimidazole group is protonated and de-protonated, and we found that the protonated species provides a better binding, so these are the results shown in Table I. The thiazolium cation has a pKa of 2.5 (Gilchrist, 1992), so the thiazole groups of inhibitors XV638 and Q8467 are unlikely to be protonated. Hence, we have kept these groups neutral. To assess the predictive value of the extended SAFE_p method presented here, beyond the CU inhibitors used as a training set, we applied it to a subset of peptidic inhibitors used in our initial study (Nauchitel et al., 1995). This set includes the complexes formed by the HIV-1 PR and the inhibitors U-89360E (PDB entry 1GNO; Hong et al., 1996), A-74704 (PDB entry 9HVP; Erickson et al., 1990), Ro-318959 (PDB entry 1HXB; Krohn et al., 1991) and A-78791 (PDB entry 1HVJ; Hosur et al., 1994). The protonation state of the active site Asp dyad was the same as the one used in our previous study. We performed two sets of calculations: with and without the ubiquitous water molecule found in almost all crystallographic structures of HIV-1 PR complexed to peptidic inhibitors. This water molecule connects the inhibitor to the tip of the HIV-1 PR flaps through a set of hydrogen bonds. The hydrogen positions of all systems studied here were optimized by a 1000-step energy minimization with a steepest
Binding analysis of HIV-1 PR cyclic urea inhibitors
Table I. Observed and calculated binding free energies by the SAFE_p and modified SAFE_p methodsa
aAll energy data are in kcal/mol. bHydrophobic contribution: first term cDesolvation penalty: second dHydrogen bond energy.
in Equation 1. term in Equation 1.
free energy calculated with the modified SAFE_p method. inhibitor has the benzyl groups in P1/P1⬘ replaced by phenoxymethyl groups.
descendent algorithm and then the SAFE_p and hydrogen bond energies were calculated. Results and discussion Choice of binding free energy algorithm Table I lists the observed binding free energies, the SAFE_p energies and the contact hydrogen bond energies as well as the sum of these two contributions, for all CU inhibitors studied. The values for the constants k1 and k2 (see Equation 1) used in these calculations were those obtained for peptidic inhibitors (Nauchitel et al., 1995). As seen in Table I, the binding free energies calculated by SAFE_p do not correlate with the observed binding free energies. The inhibitors that generate the strongest differences are those that have polar groups in P2: replacement of the second ring in the naphthalene moiety in P2 by a polar group generates an inhibitor with an
equal or better binding energy, a trend that is not seen in the SAFE_p calculated binding energies. For instance, the SAFE_p approach predicts a higher affinity of binding for XK263, an inhibitor with a bulky two-fused rings hydrophobic group, over those CU inhibitors that have the second ring replaced by polar groups like NH2 or OH (see inhibitors DMP450 and DMP323 in Table I). Experimentally, all three above-mentioned compounds have the same binding affinities. Analogously, the inhibitors that have a thiazolylbenzamide group, replacing one or both naphthalene groups in inhibitor XK263, are inhibitors with a binding constant one order of magnitude lower than the parent inhibitor, a trend that is not observed in the SAFE_p calculations. Including a term for the enzyme–inhibitor polar contacts produces energies that correlate better with the observed binding free energies. The least squares (LSQ) fit of all three 709
F.Sussman, M.C.Villaverde and L.Martı´nez
Table II. Calculated binding free energies with new LSQ weightsa Inhibitor
∆GDP ⫹ ∆GHB
SD146 XV638 Q8467 XK263 DMP323 DMP450 XK216 AH1
24.9 21.5 18.1 21.0 20.6 20.6 15.2 20.2
–14.9 –14.7 –13.7 –12.2 –12.6 –12.6 –11.8 –11.8
–14.8 –14.5 –14.1 –13.1 –13.1 –13.1 –11.5 –10.9
values for k1, k2 and k3 for Equation 2 were obtained by LSQ fitting of the observed binding free energies to that equation. The values for these constants are k1 ⫽ –0.0055, k2 ⫽ 0.068, k3 ⫽ 0.44 and K ⫽ –9.6.
terms in Equation 2 (hydrophobic contribution, desolvation energy and polar contacts) to the observed free energies of binding produces a high correlation factor (R ⫽ 92%), indicating a very high agreement between the observed free energies and the calculated ones. Table II contains the predicted binding free energies with the new set of parameters derived from the LSQ fit. The only inhibitor whose binding ranking is not well predicted by this protocol is AH1, the single compound that has a different side chain in P1/P1⬘. A possible explanation is that the binding constant for this inhibitor was determined in a different laboratory, using different experimental conditions than for the other seven compounds. Entropic and enthalpic contributions to binding In our original work, the prediction of the free energy of binding of peptidic inhibitors did not require the inclusion of an explicit term to take into account the electrostatic interactions. The rationale behind this finding was based on the idea that the electrostatic interactions contributed only indirectly to free energy of binding, through the binding specificity generated by these interactions, allowing for the optimal placement of the side chain of the ligands into their pockets. This was the result of producing a hydrogen bond or other electrostatic interactions in a very hydrophobic environment, where the desolvation penalty due to the burial of polar groups is not compensated by the formation of the electrostatic interactions (Nauchitel et al., 1995). In our analysis of the binding of peptidic inhibitors, the hydrophobic burial (given by the first term of Equation 1) is the driving force for affinity. The same conclusion has been arrived at by other authors, using the analysis of the thermodynamics of binding of peptidic inhibitors to the HIV-1 PR (Velazquez-Campoy et al., 2000). Table II lists the predicted binding free energies using the weighting constants obtained from the LSQ fit. This table also lists the sum of the weighted desolvation term and the polar contacts (∆GDP ⫹ ∆GHB). This is a very important term, since its sign should predict whether polar groups contribute to drive the binding of CU inhibitors to the enzyme. As seen from Table II, the contribution of polar contacts is not enough to overcome the penalty of desolvation of charged atoms, as shown by the positive sign of the sum of the two terms. Hence, as in the case of the peptidic inhibitors, the electrostatic interactions contribute to the affinity only indirectly by determining the specificity of binding. The binding energy is driven by entropically based hydrophobic forces as shown by the large negative contribution of the first term of the SAFE_p equation (see Table I). Nevertheless, the free energy of binding of CU analogues is modulated by the electrostatic interactions 710
Fig. 1. Top, the superimposition of one of the longest peptidic inhibitors (U-85548E) with the CU inhibitor SD146. The overlap was obtained by superimposing the peptidic chains of both complexes. Bottom, the chemical structures of both superimposed inhibitors as well as the fragment partition for the peptidic inhibitor following the definition for substrate molecules (Schetcher and Berger, 1967).
to a greater extent than the binding of the peptidic inhibitors, as proven by the need for a polar contact term in the free energy predictor function in order to reproduce the binding of the former. The increased importance of the polar contributions to binding in the CU analogues indicates that some of these groups, mainly those on the edge of the P2 substituents, form these interactions in a relatively polar environment, since the latter groups are in the periphery of the enzyme and, as such, partially exposed to solvent (like in the case of inhibitor SD146), or are located in a polar micro-environment like the amino group of the aniline fragment in the inhibitor DMP450, which makes polar contacts with the carboxylate groups of residues Asp30/30⬘ (Ala et al., 1998). The forces that drive binding to a protein are not yet fully understood, although recent studies indicate that the desolvation penalty required for the burial of polar and charged groups, upon binding, frequently precludes their involvement in electrostatic interactions that directly lower the free energy of binding (Lee and Tidor, 2001). This result is in agreement with our studies of inhibitor binding to the HIV-1 PR using the SAFE_p method. In order to search for the structural underpinnings of the solvation accessibility differences between peptidic and certain CU-based inhibitors we have superimposed some complexes formed by peptidic inhibitors and the HIV-1 PR with those formed by CU inhibitors. Superimposition of the HIV-1 PR complexes of a CU inhibitor (SD146), with one of the longest peptidic inhibitors U-85548E (PDB entry 8HVP; Jasko´ lski et al., 1991) shows that the periphery fragments high exposure to the solvent of some CU inhibitors is due to the extended conformation of the CU inhibitor, that bypasses the P3 fragment (see Figure 1). As seen from Figure 1, the CU inhibitor SD146 does not have the P3/P3⬘ fragments observed in the peptidic inhibitor. The CU inhibitor fragment next to fragment P2 lies
Binding analysis of HIV-1 PR cyclic urea inhibitors
in the more solvent exposed S4 pocket rather than in the S3 region. Affinity prediction for other inhibitors To assess the predictive power of the modified SAFE_p method we have determined the ranking of a set of peptidic inhibitors that differ in their affinities by 4 kcal/mol (approximately three orders of magnitude). The binding order of the selected inhibitors is: U-89360E ⬍ A-74704 ⬍ Ro-31-8959 ⬍ A-78791 In our previous work (Nauchitel et al., 1995) the calculated ranking obtained for inhibitors A-74704 and Ro-31-8959 was the reverse of the experimental one. The calculated ranking obtained with the modified SAFE_p method (using the same constants as in Table I for the CU inhibitors) gives a ranking order closer to the experimental order:
Sussman,F., Martı´nez,L.A. and Villaverde,M.C. (1998) Adv. Exp. Med. Biol., 436, 91–97. Trylska,J., Antosiewicz,J., Geller,M., Hodge,C.N., Klabe,R.M., Head,M.S. and Gilson,M.K. (1999) Protein Sci., 8, 180–195. Velazquez-Campoy,A., Todd,M.J. and Freire,E. (2000) Biochemistry, 39, 2201–2207. Wallqvist,A., Jernigan,R.L. and Covell,D.G. (1995) Protein Sci., 4,1881–1903. Weber,I.T. and Harrison,R.W. (1999) Protein Eng., 12, 469–474. Wlodawer,A. and Erickson,J.W. (1993) Annu. Rev. Biochem., 62, 543–585. Wlodawer,A. and Vondrasek,J. (1998) Annu. Rev. Biophys. Biomol. Struct., 27, 249–284. Yamazaki,T. et al. (1994) J. Am. Chem. Soc., 116, 10791–10792. Received November 29, 2001; revised June 4, 2002; accepted June 11, 2002
U-89360E ~ A-74704 ⬍ Ro-31-8959 ⬍ A-78791 This result is obtained only when the single ubiquitous water molecule (observed in most crystallographic HIV-1 PR complexes) is included. A full account of the results of the present SAFE_p version for peptidic inhibitors will be presented elsewhere. The next step in this research will be the application of this protocol to other inhibitor–enzyme complexes, starting with other aspartic protease enzymes. Acknowledgements We thank the MEC-DGESIC for financial support and the Xunta de Galicia for financial support and for a grant awarded to F.Sussman.
References Ala,P.J. et al. (1998) J. Biol. Chem., 273, 12325–12331. ¨ sterlund,K., Atepo,J., Unge,T., Hulte´ n,J., Ba¨ ckbro,K., Lo¨ wgren,S., O Bonham,N.M., Schaal,W., Karle´ n,A. and Hallberg,A. (1997) J. Med. Chem., 40, 898–902. Brooks,B.R., Bruccoleri,R.E., Olafson,B.D., States,D.J., Swaminathan,S. and Karplus,M. (1983) J. Comput. Chem., 4, 187–217. Debnath,A.K. (1999) J. Med. Chem., 42, 249–259. De Lucca,G.V. and Lam,P.Y.S. (1998) Drugs Fut., 23, 987–994. Erickson,J.W. and Burt,S.K. (1996) Annu. Rev. Pharmacol. Toxicol., 36, 545–571. Erickson,J. et al. (1990) Science, 249, 527–533. Gilchrist,T.L. (1992) Heterocyclic Chemistry. Longman Group UK Ltd, p. 286. Hong,L., Treharne,A., Hartsuck,J.A., Foundling,S. and Tang,J. (1996) Biochemistry, 35, 10627–10633. Horton,N. and Lewis,M. (1992) Protein Sci., 1, 169–181. Hosur,M.V., Bhat,T.N., Kempf,D.J., Baldwin,E.T., Liu,B., Gulnik,S., Wideburg,N.E., Norbeck,D.W., Appelt,K. and Erickson,J.W. (1994) J. Am. Chem. Soc., 116, 847–855. Jadhav,P.K., Ala,P., Woerner,F.J., Chang,C.-H., Garber,S.S., Anton,E.D. and Bacheler,L.T. (1997) J. Med. Chem., 40, 181–191. Jasko´ lski,M., Tomasselli,A.G., Sawyer,T.K., Staples,D.G., Heinrikson,R.L., Schneider,J., Kent,S.B.H. and Wlodawer,A. (1991) Biochemistry, 30, 1600–1609. Krohn,A., Redshaw,A., Ritchie,J.C., Graves,B.J. and Hatada,M.H. (1991) J. Med. Chem., 34, 3340–3342. Krystek,S., Stouch,T. and Novotny,J. (1993) J. Mol. Biol., 234, 661–679. Lam,P.Y.S. et al. (1994) Science, 263, 380–384. Lee,L.P. and Tidor,B. (2001) Protein Sci., 10, 362–377. Nauchitel,V., Villaverde,M.C. and Sussman,F. (1995) Protein Sci., 4, 1356– 1364. Perelson,A.S., Essunger,P., Cao,Y., Vesanen,M., Hurley,A., Saksela,K., Markowitz,M. and Ho,D.D. (1997) Nature, 387, 188–191. Schechter,I. and Berger,A. (1967) Biochem. Biophys. Res. Commun., 27, 157–162. Shimada,J., Ishchenko,A.V. and Shakhnovich,E.I. (2000) Protein Sci., 9, 765–775.