`
`9483
`
`Table Ill. Predicted Results for log W Using Neural Networks and
`Regression Analysis
`
`est(RA)b
`est(NN)•
`expt
`ref
`-1.40
`-1.40
`11
`-1.61
`I. 4-heptanol
`-2.72
`-2.35
`18
`-2.03
`2. menthane
`-5.28
`-5.23
`-4.52
`18
`3. 1,1-diphenylethylene
`-0.44
`4. p-cresol
`-0.53
`25
`-0.81
`-4.66
`-4.49
`5. testosterone
`-4.08
`22
`-5.96
`-6.24
`-5.66
`6. 2.4.4'-PCB
`26
`-3.47
`7. dexamethasone
`-3.59
`-3.58
`22
`-1.83
`-2.85
`11
`8. 4-c h Joron i trobenzcne
`-2.66
`27
`-5.45
`9. 2,5-PCB
`-5.75
`-5.06
`27
`-5.35
`10. 2,6-PCB
`-5.52
`-5.21
`27
`-5.88
`11. 2,4,6-PCB
`-6.24
`-6.06
`-4.78
`-4.43
`-4.92
`10
`12. fluorene
`-6.39
`13. pyrene
`-6.04
`-6.17
`10
`-3.27
`-3.10
`-3.04
`10
`14. indan
`11
`15. 3-methylpyridine
`-0.17
`-0.01
`0.04
`-1.11
`-1.24
`16. isoquinoline
`-1.45
`II
`0.59
`17. tetrahydrofuran
`0.48
`0.74
`II
`-3.55
`-2.95
`-3.27
`25
`18. cortisone
`-1.61
`19. 2-naphthol
`-2.08
`-2.25
`25
`• Estimation of log W using neural networks which gives a standard
`deviation 0.43. The standard deviation is 0.37 if we leave out the 4-
`b Estimation of log W using regression analysis
`chloronitrobcnzenc.
`which gives a standard deviation 0.36.
`
`found to be superior to that obtained with the regression analysis
`approach, 0.30. The results clearly demonstrate that the neural
`network has captured the association between the selected prop(cid:173)
`erties of an organic compound and its aqueous solubility.
`The trained neural network was tested on its ability to predict
`the aqueous solubility of an unknown set of organic compounds,
`that is, the compounds were not members of the original training
`set and indeed in some cases were quite unrelated to the original
`members. The test set should therefore provide a severe test of
`
`the neural network's predictive ability. Care should be taken in
`interpreting the results, however, since strictly the neural network
`should only be applied to predicting those compounds containing
`the particular substituents found in the training set. The results
`obtained are shown in Table III together with the values predicted
`by the regression analysis technique. Again the performance of
`the neural network is very satisfactory and compares favorably
`with that given by the regression analysis method. The neural
`network gives a predicted aqueous solubility superior to that
`obtained by regression analysis in 9 of the 19 cases. The poor
`value predicted for 4-chloronitrobenzene is probably due to the
`omission from the training set of any chloronitro compound which
`would reduce the credence attached to the predicted value.
`In conclusion, a neural network model has been applied to the
`prediction of the aqueous solubility of organic compounds and
`the usefulness of the model clearly demonstrated. The predictive
`capability of neural networks has been demonstrated on a number
`of unknown organic compounds. It has been shown in this study
`that neural networks give a superior performance to that given
`by a regression analysis technique. While this work was in progress
`a paper was published 34 describing an application of the neural
`network approach to estimating quantitative structure-activity
`relationships. This work confirms the conclusions derived in this
`study that neural networks can determine such relationships with
`a performance exceeding that of linear multiregression analysis.
`Clearly, the neural network approach would seem to have great
`potential for determining quantitative structure-activity rela(cid:173)
`tionships and as such be a valuable tool for the medicinal chemist.
`
`Supplementary Material Available: Listing of complete ex(cid:173)
`perimental and estimated log W values (6 pages). Ordering
`information is given on any current masthead page.
`
`(34) Aoyoma, T.; Suzuki, Y.; Ichikawa, H. J. Med. Chern. 1990, 33, 2583.
`
`Computational Studies on FK506: Conformational Search and
`Molecular Dynamics Simulation in Water
`
`Julianto Pranata and William L. Jorgensen*
`Contribution from the Department of Chemistry, Yale University, New Haven, Connecticut 06511.
`Received May 6, 1991
`
`Abstract: Computational investigations have been undertaken to elucidate the conformational characteristics and the hydration
`of the immunosuppresant FK506. The calculations made use of the AMBER/OPLS molecular mechanics force field, augmented
`with some newly developed parameters particularly for the a-ketoamide torsion. A conformational search on FK506 using
`an internal coordinate Monte Carlo method found 21 distinct energy minima within 12 kcaljmol of the lowest energy structure.
`The minima include structures with both cis and trans conformations of the amide bond. A 200-ps molecular dynamics simulation
`in water then provided information on the dynamical behavior of the cis isomer of FK506 as well as its hydration. Two
`conformations of the macrocyclic ring are sampled during the simulation, and some exocyclic groups undergo rapid conformational
`changes. Considerable flexibility is also observed near the amide functionality, which is in the binding region of FK506. The
`hydration of FK506 shows interesting variations owing to differences in the steric environments of potential hydrogen-bonding
`sites. In the critical binding region, there are on average 5 hydrogen bonds between water molecules and FK506.
`
`FK506, rapamycin, and cyclosporin A (CsA) are immuno(cid:173)
`suppresive agents that act by blocking the signal transduction
`pathways that lead toT lymphocyte activation. 1 FK506 and
`rapamycin are structurally similar and appear to bind to the same
`receptor, FKBP,2 while the structurally unrelated CsA, a cyclic
`undecapeptide, binds to a different receptor, cyclophilin. 3 Both
`
`(1) Schreiber, S. L. Science 1991,251, 283.
`(2) Bierer, B. E.; Mattila, P. S.; Standaert, R. F.; Herzenberg, L. A.;
`Burakoff, S. J.; Crabtree, G.; Schreiber, S. L. Proc. Nat/. Acad. Sci. U.S.A.
`1990, 87, 9231. Dumont, F. J.; Melino, M. R.; Staruch, M. J.; Koprak, S.
`L.; Fischer, P. A.; Sigal, N. H. J. lmmunol. 1990, 144, 1418. Fretz, H.;
`Albers, M. W.; Galat, A.; Standaert, R. F.; Lane, W. S.; Burakoff, S. J.;
`Bierer, B. E.; Schreiber, S. L. J. Am. Chern. Soc. 1991, /13, 1409.
`
`receptors have been shown to be peptidyl-prolyl cis-trans isom(cid:173)
`erases (rotamases). 4•5 FK506 and rapamycin inhibit the rotamase
`activity of FKBP, but not of cyclophilin; likewise CsA inhibits
`the rotamase activity of cyclophilin, but not of FKBP.4
`
`(3) Handschumacher, R. E.; Harding, M. W.; Rice, J.; Drugge, R. J.;
`Speicher, D. W. Science 1984, 226, 544. Handschumacher, R. E.; Harding,
`M. W. Transplantation 1988, 46, 29S.
`(4) Harding, M. W.; Galat, A.; Uehling, D. E.; Schreiber, S. L. Nature
`1989, 341, 758. Siekierka, J. J.; Hung, S. H. Y.; Poe, M.; Lin, C. S.; Sigal,
`N. H. Nature 1989, 341, 755.
`(5) Fischer, G.; Wittrnann-Liebold, B.; Lang, K.; Kiefhaber, T.; Schmid,
`F. X. Nature 1989, 226, 544. Takahashi, N.; Hayano, T.; Suzuki, M. Nature
`1989, 226, 473.
`
`ooo2-7863/9111513-9483$02.50/0
`
`© 1991 American Chemical Society
`
`
`
`9484 J. Am. Chern. Soc., Vol. 113, No. 25, 1991
`
`Pranata and Jorgensen
`
`20.0
`
`15.0
`
`1 _I
`/ y""
`
`0
`
`0
`E
`':::::-
`Cl
`()
`0
`'"' 0>
`0:;
`c
`w
`
`<1.> > :g
`o:;
`0::
`
`FK506
`
`rapamycin
`
`~ 6:~·~~~·
`f::t"-r;?"y_~fy
`Me-)y~J~~~JN10
`
`H
`
`- ( o.)._._~.o i
`cyclosporin A
`Crystal structures of all three immunosuppresants have been
`reported.6- 8
`In addition, NMR data have been used to deduce
`the structure of CsA in a variety of solvents.6•9 The solution-phase
`structure for FK506 in CDC1 3 has also been examined. 10
`Schreiber and co-workers have investigated the inhibition of
`FKBP rotamase activity by FK506. 11 The immunosuppresant
`with 13C labels at C8 and C9 was synthesized 12 and used to probe
`the binding process. 13C NMR of the enzyme-inhibitor complex
`shows no evidence for the existence of a tetrahedral adduct at
`either C8 or C9. Thus, the mechanism for rotamase activity does
`not involve the formation of a tetrahedral intermediate; rather,
`a mechanism was proposed which features a twisted amide bond
`in the transition state. The a-ketoamide functionality in FK506
`(and rapamycin) has an orthogonal orientation in the crystal
`structure7•8 and serves as a surrogate for the twisted amide bond.
`Thus, by acting as a transition-state analogue, FK506 potently
`inhibits rotamase activity.
`Interestingly, the NMR of FK506 in solution shows the ex(cid:173)
`istence of two isomers, attributed to the cis and trans conformations
`for the amide bond in a 2: I ratio. 10•11
`It has recently been de(cid:173)
`termined by X-ray crystallography that the trans isomer is bound
`to FK BP, 13 though only the cis isomer is observed in the crystal
`structure of isolated FK506. 7 For CsA, both in the crystal and
`in solution the molecule has a cis peptide bond between residues
`MeLeu 9 and MeLeu 10.6•9 However, there is evidence that CsA
`bound to cyclophilin also adopts a trans conformation for this
`bond. 14
`On the computational side, Lautz et al. have reported molecular
`dynamics simulations of CsA in water, CC14, and the crystalline
`environment. 15 The focus was on comparisons with experimental
`structural data and medium effects on the conformation of CsA.
`However, the 40-50-ps durations for the simulations severely
`
`(6) Loosli, H. R.; Kessler, H.; Oschkinat, H.; Weber, H. P.; Petcher, T.
`J.; Widmer, A. Helv. Chim. Acta 1985, 68, 682.
`(7) Tanaka, H.; Kuroda, A.; Marusawa, H.; Hatanaka, H.; Kino, T.; Goto,
`T.; Hashimoto, M.; Taga, T. J. Am. Chern. Soc. 1987, /09, 5031. Taga, T.;
`Tanaka, H.; Goto, T.; Tada, S. Acta Crystallogr. 1987, C43, 751.
`(8) Swindells, D. C. N.; White, P. S.; Findlay, J. A. Can. J. Chern. 1978,
`56, 2491.
`(9) Kessler, H.; Oschkinat, H.; Loosli, H. R. Helv. Chim. Acta 1985, 68,
`661. Kessler, H.; Kock, M.; Wein, T.; Gehrke, M. Helv. Chim. Acta 1990,
`73, 1818.
`(10) Karuso, P.; Kessler, H.; Mierke, D. F. J. Am. Chern. Soc. 1990, JJ2,
`9434.
`(II) Rosen, M. K.; Standaert, R. F.; Galat, A.; Nakatsuka, M.; Schreiber,
`S. L. Science 1990, 248, 863.
`( 12) Nakatsuka, M.; Ragan, J. A.; Sammakia, T.; Smith, D. B.; Uehling,
`D. E.; Schreiber, S. L. J. Am. Chern. Soc. 1990, 112, 5583.
`(13) VanDuyne, G. D.; Standaert, R. F.; Karplus, P. A.; Schreiber, S. L.;
`Clardy, J. Science 1991, 252, 839.
`( 14) Fesik, S. W.; Gampe, R. T., Jr.; Holzman, T. F.; Egan, D. A.; Edalji,
`R.; Luly, J. R.; Simmer, R.; Helfrich, R.; Kishore, V.; Rich, D. H. Science
`1990, 250, 1406.
`(15) Lautz, J.; Kessler, H.; van Gunsteren, W. F.; Weber, H. P.; Wenger,
`R. M. Biopolymers 1990, 29, 1669. Lautz, J.; Kessler, A.; Kaptein, R.; van
`Gunsteren, W. F. J. Computer-Aided Mol. Design 1987, /, 219.
`
`10.0
`
`5.0
`
`· · · · AM1
`---· 3-21G
`- - 6-31 G(d)/ /3-21 G
`...... .,. ..... ,
`' ..... ,,
`">t-,,,
`,.,
`'.,...
`'"'+-....
`0.0 L-~--~-..:...:..:L..o.......:::::::::, __ ,£:::::::':~
`0.0
`60.0
`120.0
`180.0
`
`·~.
`
`....
`
`OCCO D;hedral Angle (degrees)
`
`Figure 1. Torsional profiles of N,N-dimethyl-a-ketopropanamide from
`molecular orbital calculations.
`limited the conformational sampling.
`We are interested in investigating the structural and energetic
`aspects of the binding of immunosuppresants and substrates to
`FKBP. The initial efforts, described here, addressed the devel(cid:173)
`opment of needed force-field parameters for FK506, conforma(cid:173)
`tional search for low-energy structures in the absence of solvent,
`and characterization of the hydration and internal motions of
`FK506 through a 200-ps molecular dynamics simulation in water.
`Parameter Development
`A principal difficulty in performing computations on molecules
`like FK506 is the lack of appropriate molecular mechanics pa(cid:173)
`rameters. For proteins and nucleic acids, a variety of standard
`parameter sets are available, e.g., AMBER 16 or CHARMm. 17
`However, FK506 contains functionalities not found in peptides
`or nucleotides. In particular, an accurate description of the a(cid:173)
`ketoamide torsion is required, in view of its importance in the
`binding process. 11
`Where available, parameters and potential functions from the
`AMBER force field were used for bonded interactions (bond
`stretches, bends, and torsions, including improper torsions). 16 The
`OPLS parameters and functions were used for nonbonded in(cid:173)
`teractions. 18 Some parameters were added to the AMBER set
`on the basis of existing parameters for similar functional groups.
`Appropriate parameters were not found for some torsions involving
`esters, ketones, and olefins. Parameters appropriate for an ester
`group were recently developed by Charifson et al. 19 and were used
`in the present work. Other torsional parameters were obtained
`by fitting to ab initio torsional profiles calculated by Wiberg and
`co-workers. 20 A united-atom model has been here for CH. units,
`otherwise all atoms are explicitly represented.
`As mentioned above, the a-ketoamide torsion is particularly
`important, and the parameters for this functionality were developed
`with the aid of quantum mechanical calculations. N,N-Di(cid:173)
`methyl-a-ketopropanamide was chosen as a model system, and
`the torsional profile for the bond between the two carbonyl carbons
`was computed using semiempirical (AM I )21.22 and ab initio
`
`(16) Weiner, S. J.; Kollman, P. A.; Case, D. A.; Singh, U. C.; Ghio, C.;
`Alagona, G.; Profeta, S., Jr.; Weiner, P. J. Am. Chern. Soc. 1984, 106, 765.
`Weiner, S. J.; Kollman, P. A.; Nguyen, D. T.; Case, D. A. J. Comp. Chern.
`1986, 7, 230.
`(17) Brooks, B. R.; Bruccoleri, R. E.; Olafson, B. D.; States, D. J.; Sw(cid:173)
`aminathan, S.; Karplus, M. J. Comp. Chern. 1983, 4, 187. Nilsson, L.;
`Karplus, M. J. Comp. Chern. 1986, 7, 591.
`(18) Jorgensen, W. L.; Tirado-Rives, J. J. Am. Chern. Soc. 1988, JJO,
`1657. Jorgensen, W. L.; Briggs, J. M.; Conteras, M. L. J. Phys. Chern. 1990,
`94, 1683.
`( 19) Charifson, P. S.; His key, R. G.; Pedersen, L. G. J. Comp. Chern.
`1990, /0, 1181.
`(20) Wiberg, K. B.; Martin, E. J. Am. Chern. Soc. 1985, 107, 5035.
`Wiberg, K. B. J. Am. Chern. Soc. 1986, /08, 5817. Wiberg, K. B.; Laidig,
`K. E. J. Am. Chern. Soc. 1987, 109, 5935.
`(21) Dewar, M. J. S.; Zoebisch, E. G.; Healy, E. F.; Stewart, J. J.P. J.
`Am. Chern. Soc. 1985, 107, 3902.
`(22) AM I calculations were performed using the MOPAC program.
`Stewart, J. J.P. MOPAC 5.0; QCPE Program No. 455; Indiana University,
`Bloomington, IN.
`
`
`
`Computational Studies on FK506
`
`Table I. New AMBER Torsion Parameters
`torsion
`
`CH-C(=O)-Q-C•
`o=c-o-c•
`0-C(=O)-CH-X
`
`C-C(=O)-C-X
`C-C(=O)-CH-X
`C-C(=O)-CH2-X
`
`X-C=C-X
`C=CH-CH2-X
`C=CH-CH-X
`C=C-CH2-X
`C=C-CH-X
`C-C(=C)-CH2-X
`C-C(=C)-CH-X
`
`O=C-C=O
`O=C-C-N
`C-C-C=O
`C-C-C-N
`• Reference 19.
`
`V1/2
`
`2.545
`
`0.470
`
`0.203
`0.305
`0.610
`
`0.700
`0.350
`0.350
`0.175
`0.350
`0.175
`
`0.12
`0.12
`0.12
`0.12
`
`J. Am. Chern. Soc., Vol. 113, No. 25, 1991 9485
`
`V3f2
`
`'Y
`
`1.100
`0.550
`0.550
`0.275
`0.550
`0.275
`
`180.0
`180.0
`180.0
`180.0
`0.0
`0.0
`
`'Y
`
`180.0
`180.0
`
`180.0
`180.0
`180.0
`
`180.0
`
`180.0
`180.0
`180.0
`180.0
`
`'Y
`
`0.0
`
`0.0
`
`0.0
`0.0
`0.0
`
`Ester
`
`Ketone
`
`Olefin
`
`V2/2
`
`4.150
`0.500
`
`0.230
`0.345
`0.690
`
`7.500
`
`180.0
`180.0
`180.0
`180.0
`0.0
`0.0
`a-Ketoamide
`0.42
`180.0
`0.42
`0.0
`0.42
`0.0
`0.42
`180.0
`
`10.0 . - - - - - - - - , - - - - - - - - - - - - - - ,
`
`(HF/3-21G and HF/6-31G(d)//3-21G) 23•24 molecular orbital
`calculations. The results are shown in Figure 1. Both the AMI
`and 6-31 G(d) calculations predict a non planar minimum, although
`the barrier at the anti conformation is quite low. For AMI, the
`minimum occurs at 124° and the height for the anti barrier is
`0.79 kcaljmol, while the corresponding values from the 6-31G(d)
`calculations are 135° and 0.65 kcal/mol. However, the 3-21 G
`results appear to overestimate the stability of the anti confor(cid:173)
`mation, making it the minimum. In the crystal structures, the
`a-ketoamide functionality in both FK506 and rapamycin has an
`orthogonal orientation. 7•8 The same trend is observed in other
`molecules when this functionality is doubly substituted at the
`nitrogen; e.g., tetramethyloxamide is twisted by 71 o .25
`AMBER-type parameters were then obtained by fitting to the
`6-31 G(d) torsional profile. The fitting took into account the
`substantial nonbonded interactions in this molecule. In fact, it
`turned out that the nonbonded (i.e., steric) interactions are solely
`responsible for the nonplanarity of the system; there is nothing
`unusual about the torsional parameters that are reported in Table
`I.
`
`As a test of the new parameters, they were incorporated into
`AMBER and used to compute the torsional profile of a non(cid:173)
`macrocyclic fragment of the FK506 structure. This is compared
`to the profile computed using AM I in Figure 2. Although the
`AMBER profile has a shallower minimum, the difference is not
`great, and the two profiles have the same overall shape. The
`minimum in the results with AMBER occurs at -108°, compared
`to -96° with AM 1, while in the crystal structure of FK506 this
`torsional angle is -89°. 7
`All of the torsional parameters added to the AMBER set for
`the purpose of this work are presented in Table I. In addition,
`a complete listing of the parameters for FK506 is given in the
`Supplementary Material. The same parameters were used for
`the conformational search and molecular dynamics.
`Conformation Search
`Procedure. A fairly extensive search for the conformational
`minima of FK506 in the ideal gas phase was performed using an
`
`(23) Hehre, W. J.; Radom, L.; Schleyer, P. v. R.; Pople, J. A. Ab Initio
`Molecular Orbital Theory; Wiley: New York, 1986.
`(24) Ab initio calculations were performed using the GAUSSIAN 90
`programs. Frisch, M. J.; Head-Gordon, M.; Trucks, G. W.; Foresman, J. B.;
`Schlegel, H. B.; Raghavachari, K.; Robb, M.A.; Binkley, J. S.; Gonzalez, C.;
`Defrees, D. J.; Fox, D. J.; Whiteside, R. A.; Seeger, R.; Melius, R.; Baker,
`J.; Martin, R. L.; Kahn, L. R.; Stewart, J. J. P.; Topiol, S.; Pople, J. A.
`GAUSSIAN 90 Revision F; Gaussian Inc.: Pittsburgh, PA, 1990.
`(25) Adiwidjaja, G.; Voss, J. Chern. Ber. 1977, 110, 1159.
`
`I
`I
`
`I
`
`I
`
`I
`
`I
`I.
`I.
`I.
`
`/
`
`I
`
`I
`
`I
`
`I
`
`5.0
`
`--AM1
`----AMBER
`
`I
`0.0 k::-~-,-,---------~--;,.7'1~:..__ _ _ ___,
`...... ,
`
`...................
`
`............. _____ .........
`
`~///
`...
`
`-5.0 L_ ____ _.__.:::::=:::::::::.._ _ _._ _ _ ~--
`0.0
`-60.0
`-120.0
`-180.0
`
`OCCO Dihedral Angle (degrees)
`Figure 2. Torsional profiles of a fragment of FK506.
`
`internal coordinate Monte Carlo method. 26 The focus was on
`the conformation of the 21-membered macrocycle. No search
`was conducted which involved variations of the exocyclic torsions.
`Also excluded were torsions within six-membered rings, namely
`those for the C2-N7, Clo-GS, and 05-cl4 bonds, as well as the
`double bond (Cl9-c20). The remaining 17 dihedral angles were
`randomly varied; however, 01-c1 was defined as the ring-closure
`bond, so its torsion and the torsions around the two adjacent bonds
`(Cl-C2 and C26-01) were not explicitly varied. Of course, no
`constraints were applied in the subsequent energy minimizations.
`The starting structure for the search was obtained from the
`X-ray crystal structure.7 Initially, all hydrogens bound to carbons
`were removed, resulting in representation of FK506 as 60 explicit
`atoms. This structure was energy minimized, and the resulting
`structure was used to start the conformational search. The RMS
`deviation between the actual X-ray structure and the energy(cid:173)
`minimized form is only 0.48 A. The only significant change is
`the formation of a hydrogen bond between 06-H 1 and 04 that
`is not in the crystal structure (vide infra). A total of 8499
`structures were generated using a random walk procedure; these
`were initially minimized to a root mean square gradient of I
`kJ/(mol·A) (=0.239 kcalj(mol·A)). Nonduplicate structures
`whose energies were within 50 kJ /mol ( = 12 kcal/mol) above the
`lowest energy minimum were saved. This resulted in 28 structures
`which were further minimized to a root mean square gradient of
`
`(26) Chang, G.; Guida, W. C.; Still, W. C. J. Am. Chern. Soc. 1989, Ill,
`4379.
`
`
`
`9486 J. Am. Chern. Soc., Vol. JJ3, No. 25, 1991
`
`Pranata and Jorgensen
`
`Table II. Values of Macrocyclic Dihedral Angles (deg) in the Conformations Found in the Monte Carlo Search and in the X-ray Structure
`conformation
`8
`5
`7
`9
`3
`4
`6
`2
`dihedral angle
`-165
`-178
`-179
`87
`179
`91
`-168
`01C1C2N7b
`86
`43
`-99
`-100
`-95
`-101
`-100
`-100
`-93
`-100
`-95
`ClC2N7C8<
`-4
`-3
`2
`-5
`-2
`1
`0
`0
`C2N7C8C9
`165
`-87
`-104
`-99
`-96
`-115
`-68
`-63
`-105
`167
`N7C8C9C10
`81
`68
`68
`72
`55
`91
`65
`90
`75
`C8C9Cl005
`175
`177
`178
`179
`178
`169
`173
`179
`177
`C9C1005Cl4'
`-177
`-175
`-177
`-176
`-177
`-176
`-179
`-176
`-177
`C1005Cl4ClY
`92
`75
`76
`78
`49
`85
`62
`86
`75
`05Cl4CI5C16
`-78
`89
`60
`68
`52
`68
`65
`65
`57
`C14Cl5Cl6Cl7
`-170
`-159
`-179
`-170
`-169
`-159
`-172
`-151
`-168
`Cl5Cl6Cl7Cl8
`-56
`167
`70
`60
`74
`-63
`155
`66
`60
`C16C17Cl8Cl9
`-67
`-126
`-119
`16
`-123
`-121
`130
`71
`-107
`Cl7Cl8Cl9C20
`-177
`171
`-176
`179
`-176
`175
`-179
`175
`176
`C18C19C20C2F
`-118
`-129
`-124
`-124
`-115
`-110
`-142
`-123
`-115
`C 19C20C21 C22
`-48
`-39
`101
`138
`122
`101
`83
`106
`73
`C20C21C22C23
`-148
`-157
`174
`-120
`148
`-99
`122
`178
`-152
`C21 C22C23C24
`-64
`169
`169
`175
`172
`-65
`172
`-58
`146
`C22C23C24C25
`-67
`-65
`-69
`-84
`-69
`-163
`-61
`-145
`-71
`C23C24C25C26
`-24
`-37
`-23
`-42
`-55
`45
`-23
`-56
`-40
`C24C25C2601
`-155
`-149
`-147
`-147
`-146
`-153
`-147
`-154
`-154
`C25C2601C1b
`-171
`-176
`-165
`-160
`-177
`-169
`-162
`-169
`-175
`C2601C1C2b
`25.9
`25.7
`26.4
`24.5
`24.4
`23.9
`energyd
`20.6
`23.3
`17.6
`a After minimization. bQJCI is defined as the ring closure bond, thus these dihedral angles are not explicitly included in the search. <These
`
`180.0
`
`~
`
`~
`
`!
`
`1
`
`I
`
`! I
`
`,..-...,
`UJ
`Q)
`Q)
`I...
`O'l
`Q)
`.....__
`"0
`
`Q)
`::::l
`
`0 >
`
`90.0
`
`0.0
`
`-90.0
`
`I
`
`(
`
`I
`
`I
`
`-180.0
`
`m ,
`
`• p
`
`6
`
`~
`I
`
`11
`
`il
`
`i
`
`t
`p
`
`16
`
`~
`
`I
`
`21
`
`Dihedral Angle
`Figure 3. Distribution of macrocyclic dihedral angles in the 21 structures found in the Monte Carlo search. Open squares are for trans amide isomers
`and filled squares for cis. Also shown are the dihedral angles of the X-ray structure after minimization (X). The numbers of the dihedral angles on
`the abscissa correspond to the list in Table II.
`
`0.1 kJ /(mol·A). After elimination of duplicates and high-energy
`structures, 21 distinct minima were found.
`The energy minimizations were performed using a dielectric
`constant of 1.0 and a cutoff distance of 9.0 A for both van der
`Waals and electrostatic interactions. The calculations were
`performed with the BATCHMIN program, Version 2.7, on a DEC
`VaxStation 3500 minicomputer. 27
`Results. The dihedral angles for the macrocyclic ring in the
`21 conformational minima found during the search are listed in
`Table II along with relative potential energies. A distribution of
`the dihedral angles is represented in Figure 3. Both Table II and
`Figure 3 also contain data for the conformation from the X-ray
`
`(27) BATCHMIN is the noninteractive part of the MACROMODEL
`molecular modeling program. Mohamadi, F.; Richards, N. G. J.; Guida, W.
`C.; Liskamp, R.; Lipton, M.; Caufield, C.; Chang, G.; Hendrickson, T.; Still,
`W. C. J. Comp. Chern. 1990, II, 440.
`
`structure after minimization. The 21 energy minima include both
`trans and cis amide isomers. Stereopictures of the lowest energy
`trans (1) and cis (2) forms are shown in Figure 4, along with the
`energy minimized X-ray structure; corresponding pictures and
`coordinates for all 22 structures are available in the Supplementary
`Material.
`It is emphasized that we did not attempt to locate all the
`conformational minima of FK506; to do so would have necessitated
`a much longer search as well as the inclusion of variations of the
`exocyclic and six-membered-ring torsions. However, the 21
`structures may be expected to be representative of the low-energy
`conformations of FK506.
`All 21 structures are reasonable in that no bonds or angles are
`unduly strained. Somewhat surprisingly, the X-ray crystal
`structure is not among these 21 structures. Its energy, or rather,
`the energy of the minimum closest to it, is 14.7 kcaljmol above
`the energy of the lowest minimum found. A minimization was
`
`
`
`Computational Studies on FK506
`
`J. Am. Chern. Soc., Vol. 113, No. 25, 1991 9487
`
`conformation
`13
`15
`16
`14
`12
`11
`10
`167
`176
`18
`45
`35
`19
`51
`-97
`-98
`-91
`-103
`-106
`-98
`-95
`-172
`5
`167
`-167
`167
`180
`2
`-80
`-136
`-120
`-110
`-90
`-172
`152
`75
`84
`67
`69
`69
`72
`78
`170
`175
`172
`173
`174
`172
`175
`-177
`-174
`-170
`-178
`-169
`-168
`-175
`-171
`-169
`76
`159
`66
`66
`72
`-43
`79
`173
`177
`57
`60
`60
`-162
`-178
`-141
`-137
`-105
`-170
`-166
`-57
`67
`77
`62
`63
`84
`81
`-3
`-104
`-84
`-98
`-127
`-105
`-113
`-177
`-175
`176
`-179
`178
`178
`180
`-112
`-97
`-122
`-118
`-101
`-106
`-120
`-57
`-22
`-72
`-64
`139
`68
`52
`-154
`107
`-133
`175
`-169
`154
`125
`-70
`78
`-176
`173
`-58
`70
`146
`-93
`-157
`-73
`-153
`-76
`-68
`46
`-47
`-40
`-55
`-55
`-45
`-41
`43
`-154
`-158
`-151
`-152
`-149
`-!51
`69
`-170
`-178
`-174
`177
`178
`179
`173
`27.4
`27.4
`27.0
`26.8
`26.7
`26.7
`26.4
`dihedral angles are not included in the search. d Energies in kcaljmol.
`
`17
`-172
`-99
`-2
`-79
`89
`176
`-169
`78
`123
`64
`72
`-108
`180
`-115
`-46
`-142
`-68
`-63
`-23
`-155
`-167
`27.5
`
`18
`173
`-99
`-2
`-115
`72
`176
`-177
`86
`-57
`71
`171
`-18
`177
`-122
`105
`-142
`173
`-68
`-38
`-150
`177
`27.7
`
`19
`19
`-92
`171
`-126
`64
`170
`-173
`58
`54
`-170
`55
`-113
`174
`-134
`98
`-130
`177
`-61
`63
`65
`160
`28.2
`
`20
`89
`-98
`-3
`-88
`81
`177
`-174
`82
`65
`-177
`65
`-122
`-177
`-117
`127
`-147
`172
`-162
`51
`-!55
`-165
`28.3
`
`21
`15
`-94
`167
`-177
`69
`173
`-177
`74
`62
`-174
`68
`-117
`-179
`-117
`101
`-108
`-64
`-91
`-45
`-155
`175
`28.3
`
`X-ray•
`124
`-93
`0
`-78
`87
`176
`-176
`80
`63
`177
`63
`-128
`-174
`-122
`!35
`-132
`63
`-169
`180
`!58
`173
`32.3
`
`also carried out starting with the geometry of bound FK506
`obtained from the X-ray crystal structure of its complex with
`FKBP. 13 The resultant structure had an energy 21.5 kcal/mol
`above the lowest energy minimum.
`Interestingly, the cis isomers appear to favor a perpendicular
`orientation of the adjacent carbonyl groups (C8-C9 dihedral
`angle), while the trans isomers are more tolerant of an anti
`orientation in this position. Another noticeable difference is that
`the conformation about the CI-C2 bond is always gauche-like
`in the trans isomers; it is usually anti in the cis. These trends,
`of course, only reflect the results for the 21 structures. However,
`these torsions are in the binding region of FK506, and the tend(cid:173)
`encies they show may be important in view of the observation that
`only the trans isomer is bound to FKBP. 13 No other trends are
`strikingly apparent which differentiate the cis and trans isomers.
`Not surprisingly, the four torsions excluded from the search
`(those around the C2-N7, CI0-05, 05-CI4, and CI9-C20
`bonds) remain in their initial conformation (Figure 3). Three other
`torsions also show no tendency for variations, namely those around
`the C9-C I 0, C2G-C21, and 0 1-C I bonds. The anti orientation
`for the last torsion means that the ester is always in the Z form.
`Most of the rest of the torsions are clustered into easily identifiable
`gauche and anti conformations. Of course, the torsions flanking
`the double bond (CI8-CI9 and C2G-C21) are expected to cluster
`around skew and syn conformations instead, and they do, except
`that C2G-C21 is a torsion that remains skew in all 21 structures.
`The torsions flanking the isolated keto group (C21-C22 and
`C22-C23) show a rather large spread of values, and do not appear
`to be easily categorized into gauche or anti conformations.
`A major reason for the absence of the X-ray crystal structure
`from the set is the presence of intramolecular hydrogen bonds.
`The actual X-ray structure does not have any intramolecular
`hydrogen bonds, but one (06-HI .. ·04) was formed upon energy
`minimization (Figure 4). This hydrogen bond is present in all
`the other structures; furthermore, 16 of these structures have
`additional hydrogen bonds involving 010-H2 as a donor (Table
`III).
`In the actual crystal structure, OIO-H2 acts as an inter(cid:173)
`molecular hydrogen bond donor to 09 of a neighboring molecule.7
`In addition, one water molecule was located in the crystal which
`forms the hydrogen bonds 06-H J ... Q(W) and 0(W)-H .. ·04,
`preventing the formation of a direct hydrogen bond between
`06-H I and 04. The water molecule also forms a hydrogen bond
`to 03 of a neighboring FK506 molecule. Another intriguing
`difference in the crystal structure is the anti orientation about the
`C25-C26 bond. The orientation is invariably gauche in the 21
`
`Table III. Intramolecular Hydrogen Bonds in the Structures Found
`in the Monte Carlo Search•
`conformation
`1
`2
`3
`4
`5
`6
`7
`8
`9
`10
`11
`12
`13
`14
`15
`16
`17
`18
`19
`20
`21
`X-ray
`(minimized)
`• A hydrogen bond is deemed to exist if the distance between the
`hydrogen and the acceptor is less than 2.5 A and the donor-hydrogen(cid:173)
`acceptor angle is greater than 120°.
`
`hydrogen bonds
`06-Hl .. ·04; 010-H2 .. ·08
`06-Hl .. ·04; 010-H2 .. ·05; 010-H2 .. ·01
`06-Hl···04
`06-H1· .. 04; 010-H2 .. ·09
`06-H1· .. 04
`06-H\· .. 04; 010-H2 .. ·05; 010-H2 .. ·01
`06-H1· .. 04; 010-H2 .. ·09
`06-Hl .. ·04
`06-Hl .. ·04; 010-H2 .. ·09
`06-Hl .. ·04
`06-Hl .. ·04; 010-H2 .. ·09
`06-Hl· .. 04; 010-H2· .. 08
`06-Hl· .. 04; 010-H2· .. 09
`06-Hl· .. 04; OIO-H2 .. ·03; 010-H2 .. ·01
`06-HJ .. ·04; 010-H2· .. 01
`06-Hl .. ·04; 010-H2· .. 09
`06-Hl .. ·04
`06-Hl .. ·04; 01Q-H2 .. ·09
`06-Hl· .. 04; OIO-H2· .. 09
`06-Hl .. ·04; OIO-H2 .. ·09
`06-Hl .. ·04; OIO-H2· .. 05
`06-Hl· .. 04
`
`other structures, which facilitates the intramolecular hydrogen
`bonding with OIO-H2.
`Molecular Dynamics
`Procedure. In order to investigate the dynamical behavior and
`solvation of FK506, a molecular dynamics (MD) simulation of
`the molecule in water was performed. The energy-minimized
`X-ray structure was taken as the starting point for the MD sim(cid:173)
`ulation. This structure was immersed in a box of TIP3P water
`molecules. 28 Any water molecule with its oxygen atom closer
`than 1.5 A or with a hydrogen closer than 0.5 A to any FK506
`atom was removed. Any water molecule with its oxygen farther
`away than 7.0 A from any solute atom in the x, y, or z directions
`was also removed, resulting in a system consisting of the solute
`
`(28) Jorgensen, W. L.; Chandrasel<har, J.; Madura, J. D.; Impey, R. W.;
`Klein, M. L. J. Chern. Phys. 1983, 79, 926.
`
`
`
`9488 J. Am. Chern. Soc., Vol. /13, No. 25, 1991
`
`Pranata and Jorgensen
`
`figure 4. Stcreorcpresentations of the lowest energy trans (top) and cis (middle) isomers found in the conformational search. The conformation obtained
`from energy minimintion of the crystal structure is shown at the bottom.
`
`and 558 water molecules ( 1734 total atoms) in a box whose initial
`dimensions were 28.6 X 28.2 X 21.4 A. From this point on,
`periodic boundary conditions were imposed on the system.
`Equilibration of the system was achieved in several phases. To
`begin, the water molecules were energy minimized for 100 steps
`using a ~teepc~t de~cent algorithm, while keeping the solute ge(cid:173)
`ometry fi>.ed. The solvent was then subjected to 5 ps of constant
`volume d) namics, during which the temperature of the system
`was raised from 100 to 298 K in a stepwise fashion. An additional
`4 ps of simulation followed. keeping the temperature at 298 K,
`with the solute stilllixcd. Finally, I ps of dynamics was performed
`in which the solute was allowed to move.
`An M 0 simulation was then performed on this equilibrated
`system at constant temperature (298 K) and pressure (I bar =
`0.987 atm) for 200 ps. Data analysis was carried out on the entire
`trajectory: there was no significant transient behavior during the
`early part of the simulation that might have resulted from im(cid:173)
`perfect equilibration.
`
`All bond lengths and the H- H distance in water molecules were
`constrained to their equilibrium values, using the SHAKE al(cid:173)
`gorithm with a tolerance of0.0004 A.29 Nonbonded interactions
`were calculated using a spherical residue-based cutoff, with FK506
`defined as a single residue. The cutoff distance was 9.0 A. To
`accelerate the calculations, a nonbonded pair list was used