3928
`
`J. Med. Chem. 1998, 41, 3928-3939
`
`Articles
`
`Investigations of Neurotrophic Inhibitors of FK506 Binding Protein via Monte
`Carlo Simulations
`
`Michelle L. Lamb and William L. Jorgensen*
`
`Department of Chemistry, Yale University, New Haven, Connecticut 06520-8107
`
`Received January 27, 1998
`
`The binding and solution-phase properties of six inhibitors of FK506 binding protein (FKBP12)
`were investigated using free energy perturbation techniques in Monte Carlo statistical
`mechanics simulations. These nonimmunosuppressive molecules are of current interest for
`their neurotrophic activity when bound to FKBP12 as well as for their potential as building
`blocks for chemical inducers of protein dimerization. Relative binding affinities were computed
`and analyzed for ligands differing by a phenyl ring, an external phenyl or pyridyl substituent,
`and a pipecolyl or prolyl ring. Such results are, in general, valuable for inhibitor optimization
`and, in the present case, bring into question some of the previously reported binding data.
`
`Introduction
`The R-ketoamide functionality of the immunosup-
`pressant natural product FK506 (Figure 1) is retained
`in many of the highest affinity ligands that have been
`developed to inhibit the rotamase (cis-trans peptidyl-
`prolyl isomerase, or PPIase) activity1 of the FK506
`binding protein (FKBP12, MW ) 12 kDa).2 Originally,
`interpretation of the crystal structure of FK506-
`FKBP12 led to the belief that the R-ketoamide mimics
`a twisted-amide transition state of peptide bond isomer-
`ization, although an endogenous substrate for FKBP12
`had not been discovered. It was thought that blockage
`of the isomerase active site prevented modification of
`downstream proteins necessary for T-cell activation, and
`this was the source of the observed immunosuppression.
`A similar mechanism had been proposed for the activity
`of the undecapeptide cyclosporin A (CsA), which inhibits
`the PPIase cyclophilin, although neither the natural
`products nor the proteins are homologous. However,
`evidence that rotamase inhibition was not sufficient for
`immunosuppression soon began to mount.3 Rapamycin
`(Figure 1), another fungal molecule structurally similar
`to FK506, inhibited FKBP12 but appeared to influence
`a later stage of the T-cell cycle. Schreiber and co-
`workers4 made a significant contribution with the syn-
`thesis of a molecule which retained the FKBP12 binding
`domain of FK506 and rapamycin (pyranose ring, R-ke-
`toamide, pipecolate ester, and cyclohexyleth(en)yl groups),
`but in which the macrocycle was contracted. This
`molecule was a rotamase inhibitor but did not prevent
`T-cell proliferation.
`It later became clear that the formation of an immu-
`nosuppressant-immunophilin complex results in a gain
`of function for the protein. The CsA-cyclophilin and
`FK506-FKBP12 pairs each present a recognition sur-
`face to the calcium-dependent, serine/threonine phos-
`phatase, calcineurin (CN).5 The FK506-FKBP12 com-
`plex binds at least 10 Å from the active site of CN and
`
`Figure 1. Structures and atom numbering for the immuno-
`suppressants FK506 and rapamycin.
`
`S0022-2623(98)00062-4 CCC: $15.00 © 1998 American Chemical Society
`Published on Web 09/19/1998
`
`

`
`Neurotrophic Inhibitors of FK506 Binding Protein
`
`Journal of Medicinal Chemistry, 1998, Vol. 41, No. 21 3929
`
`Table 1. Experimental Activities for Selected FKBP12 Ligands
`
`a Data on neurite outgrowth from chick dorsal root ganglia reported in ref 2. b Data from Guilford Pharmaceuticals, refs 2 and 59.
`c Data from SmithKline Beecham, refs 13 and 16.
`
`must block binding of subsequent phosphorylated pro-
`teins and thus the T-cell signaling pathway.6,7 Reports
`of the association of calcium channels containing -Leu-
`Pro- sequences with both FKBP12 and CN are filling
`in another long-standing piece of the FKBP12 puzzle,
`as these may represent endogenous “ligands” for FKBP12
`mimicked by FK506.8 In contrast, rapamycin-FKBP12
`interrupts a distinct signaling cascade through its
`interaction with another protein, generally termed
`FRAP (FKBP-rapamycin-associated protein).9-11 A
`crystallographic structure of this ternary complex con-
`firms the recognition requirements for rapamycin.12 In
`both FKBP12 ligands, it is the portion of the macrocycle
`opposite the R-ketoamide-pipecolic acid moiety, the
`“effector” region, which contacts calcineurin.
`As part of an effort to design low molecular weight
`PPIase inhibitors as scaffolds for the immunosuppres-
`sive effector components, the crystal structure of
`1-FKBP12 (Table 1) was solved at SmithKline Beecham
`in 1993.13 Figure 2 shows the binding mode revealed
`for the R-ketoamide and pipecolyl portion of 1. The keto
`carbonyl (O4) contacts aromatic hydrogens of Tyr26,
`Phe36, and Phe99, and the pipecoline ring sits over Trp59.
`The 3-phenylpropyl moiety binds in the solvent-exposed
`FK506-cyclohexyl groove of FKBP12 between Ile56 and
`
`Figure 2. Position of compound 1 (yellow) in the aromatic
`binding pocket of FKBP12 (green).13 Molecular graphics im-
`ages were produced using the MidasPlus software system from
`the Computer Graphics Laboratory, University of California,
`San Francisco.60
`
`Tyr82, and these residues form hydrogen bonds with the
`ester (O2) and amide (O3) carbonyl oxygens of the
`ligand. The 1-phenyl substituent interacts with Phe46
`and the tertiary pentyl group of the inhibitor. A
`
`

`
`3930 Journal of Medicinal Chemistry, 1998, Vol. 41, No. 21
`
`Lamb and Jorgensen
`
`Scheme 1
`
`Figure 3. FKBP12-bound conformation of 1 (yellow) overlaid
`with that of FK506 (red).13,15
`
`comparison of the bound conformation of 1 and FK506
`presented in Figure 3 demonstrates that this mode is
`consistent with that found in crystallographic structures
`of FK506-FKBP12 and rapamycin-FKBP12.14,15 How-
`ever, the ability to form hydrogen bonds to the Glu54
`carbonyl observed in complexes with FK506 (C24-OH)
`and rapamycin (C28-OH) is not present in this ligand.
`Binding patterns similar to those for 1 may be expected
`for compounds 2 and 3 (Table 1) as well.13,16 An ex-
`cellent analysis of FKBP12-ligand interactions, includ-
`ing discussion of previously unpublished atomic struc-
`tures, is included in a review of protein-ligand recog-
`nition motifs by Babine and Bender.17
`An additional activity for rotamase inhibitors of this
`class has expanded interest in these compounds beyond
`their potential in immunosuppressant drug design. As
`reviewed recently by Hamilton and Steiner,2 FK506 has
`been shown to induce the regeneration of damaged
`nerves in animal models of Parkinson’s and Alzheimer’s
`diseases. Furthermore, the enriched concentration of
`FKBP12 in neurons has been associated with nitric
`oxide synthesis, neurotransmitter release, and neurite
`extension. Potent, nonimmunosuppressive FKBP12
`ligands, such as V-10,36718-20 and GPI-1046 (6, Table
`
`1),21-23 are able to promote neuronal growth in vitro and
`in vivo without the addition of exogenous growth factors.
`They have a better therapeutic potential than growth
`factors in that they are orally bioavailable and able to
`cross the blood-brain barrier. The requirement of
`binding to FKBP12 for neuronal activity has also been
`demonstrated, but there is no linear relationship be-
`tween rotamase inhibition and activity in neuronal
`cells.2 FKBP12 binding is apparently necessary but not
`sufficient for stimulation of nerve growth, suggesting
`
`that, as in T-cells, the complex may modify the function
`of an additional target.
`Another use of this class of FKBP12 ligands has also
`emerged. The ability of the immunosuppressants to
`induce protein heterodimerization and the knowledge
`of ligand modifications that prevent this association has
`been exploited for control of cellular signaling pathways,
`protein translocation, and gene activation.24,25 Target
`proteins are first artificially attached to the immuno-
`philins (FKBP12 or cyclophilin), CN, or FRAP. The
`ligands themselves or synthetic homo- or heterodimers
`of FK506, CsA, or rapamycin then bring their protein
`partners together, resulting in the proximity of the tar-
`get proteins and transmission of signal.24-30 Recently,
`dimers of 7 have been used to effect cellular apoptosis
`and to induce transcription, again without the immu-
`nosuppressive effects of further binding to calcineurin.31
`This technique of “chemically induced dimerization”,
`used with small, cell-permeable molecules such as 7, is
`
`designed to have application in cellular gene therapy.
`Given the diverse biological applications of these
`R-ketoamide ligands and that only slight differences in
`structure can have profound effects on activity, we have
`used theoretical techniques to probe the binding of
`compounds 1-6 (Table 1) at the atomic level, in both
`structural and energetic terms. Previous simulations
`of FKBP12 have addressed the rotamase mechanism
`applied to peptide substrates32,33 and the importance of
`Tyr82 in binding FK506.34 Our current approach has
`focused on free energy perturbation (FEP) calculations,
`using Monte Carlo (MC) methods rather than molecular
`dynamics (MD) for sampling. Computed relative free
`energies of binding, which are obtained from simula-
`tions of the ligands in solution and bound to the protein,
`may be compared with those obtained from experimen-
`tal binding constants (Scheme 1). Averages of the
`computed structures may then be used to analyze the
`origin of the differences in binding affinities.
`The MC method used here has been validated with a
`study of benzamidine inhibitors of trypsin35 and was
`further applied to the analysis of orthogonal CsA-
`cyclophilin pairs as components of a system for chemi-
`cally induced dimerization.36 The present study is
`aimed at understanding factors that influence the
`binding of 1 and its analogues. In particular, the effects
`of removal of the 1-phenyl group, conversion of the
`3-phenyl to 3-(3-pyridyl), and ring contraction of the
`
`

`
`Neurotrophic Inhibitors of FK506 Binding Protein
`
`Journal of Medicinal Chemistry, 1998, Vol. 41, No. 21 3931
`
`pipecolyl ring to prolyl are examined. There are dis-
`crepancies in the binding data from the two experimen-
`tal sources, as indicated by the results for 2 and 3 in
`Table 1. From the crystal structure for 1 bound (Figure
`2), the pyridine nitrogen of 3 is anticipated to be solvent
`exposed. Thus, it would normally not be expected to
`favor the lower dielectric environment of a protein ((cid:15) (cid:25)
`4) over that of bulk water ((cid:15) (cid:25) 80),37 in contrast to the
`binding results from Guilford Pharmaceuticals. This
`was pursued through computations for the 2, 3 and 5,
`6 pairs. Hamilton and Steiner have also pointed out
`that 5 and 6 are the first examples of prolyl compounds
`that bind better than their pipecolyl analogues, but the
`high affinity is attributed only to “improved design”.2
`To investigate further, differences in free energies of
`binding were computed for two pairs of pipecolyl and
`prolyl ligands. Compounds 2 and 5 represent the
`unusual case with the prolyl ligand (5) as the better
`inhibitor. Compounds 1 and 4 represent the more
`common situation in which the presumably more hy-
`drophobic pipecolyl ligand (1) has higher affinity for
`FKBP12.
`
`Computational Details
`Parametrization and Initial MC Simulations.
`The crystal structure of 1-FKBP12 at 2.0 Å resolution13
`from the Brookhaven Protein Data Bank38 (entry 1fkg)
`was used as the starting point for the simulations. The
`computational protocol for the MC simulations was the
`same as in previous applications.35,36 The good precision
`that is obtainable for free energy changes with this
`methodology was addressed extensively in ref 35. The
`MC sampling included variation of all bond angles and
`dihedrals of the ligand and protein side chains as well
`as overall rotation and translation of the ligand and
`water molecules. The protein backbone atoms were held
`fixed in their crystallographic positions. This makes the
`MC simulations more rapid, and the approximation is
`justified for FKBP12. Restricted backbone motion on
`the picosecond time scale has been noted for native
`FKBP12,39 and ligand binding further rigidifies the
`protein structure, as demonstrated by the close resem-
`blance among atomic structures of FKBP12 in numerous
`FKBP12-ligand complexes.17 To be consistent with
`prior MD calculations on the FK506-FKBP12 system,40
`all 79 residues within 12 Å of FK506 in its cocrystal
`structure with FKBP1214 were sampled. This provided
`a greater number of moving side chains than would be
`found in a 12 Å region around 1.
`The OPLS united-atom force field41 with all-atom
`aromatic groups42 provided most parameters for the
`protein; parameters for the inhibitors also came from
`this source and from a previous MD study of FK506.43
`A listing of parameters for the inhibitors is provided in
`the Supporting Information. The torsional parameters
`for the amino acid residues were derived from fitting to
`torsional energy profiles obtained from ab initio calcula-
`tions with the 6-31G* basis set.44 Any missing param-
`eters were derived by fitting to MM245 energy profiles,
`which were generated using Macromodel.46 A scale
`factor of 1/2 was applied to all 1-4 nonbonded interac-
`tions. Histidines 25, 87, and 94 are known to be un-
`protonated,47 and they were designated as (cid:228)-tautomers
`based on visual inspection. This tautomeric state has
`
`also been chosen in MD simulations of FKBP12-ligand
`complexes in solution.32,34
`The unbound ligands and protein-ligand complexes
`were solvated with 22 Å spheres containing 1477 and
`939 TIP4P water molecules, respectively. A half-
`harmonic potential with a 1.5 kcal/mol Å2 force constant
`was employed to prevent waters from migrating away
`from the cluster. A 9 Å residue-based cutoff was used
`for all nonbonded interactions; if any pair of atoms from
`two residues was within this distance, all nonbonded
`interactions between the residues were included in the
`energy evaluation. The list of nonbonded interactions
`was updated every 2 (cid:2) 105 configurations during the
`simulations.
`All Monte Carlo simulations were performed with the
`MCPRO program.48 An advantage of using internal
`coordinate MC methods is the ability to focus sampling
`on specific regions and degrees of freedom of interest.
`Consequently, bond lengths were fixed to their crystal
`structure values, and aromatic rings were treated as
`rigid units. To prevent inversion at sp3 centers such
`as R-carbons and to enforce planarity of sp2 centers for
`more efficient sampling, improper dihedral angles were
`not varied except as noted below. Otherwise, all bond
`angles and dihedrals in the moving portion of the system
`were sampled.
`The MC simulations were carried out for 25 °C on
`Silicon Graphics workstations and on a cluster of
`personal computers using Pentium processors. It may
`be noted that the experimental results come from an
`assay for rotamase inhibition.49 This widely used
`procedure for measuring FKBP12 binding affinities is
`usually performed somewhat below room temperature,
`e.g., near 10 °C.13 The solvent was first sampled for 1
`million (M) configurations to remove any highly repul-
`sive initial contacts with the solutes. Then, 8M con-
`figurations were performed to equilibrate the 1-FKBP12
`complex. The same protocol was followed for 1 in
`solution, beginning with the bound conformation taken
`from the 1-FKBP12 structure. During equilibration,
`the conformation of the bound ligand remained sim-
`ilar to the crystal conformation; however, partial in-
`version of the pipecolyl ring occurred in solution to
`switch it from a chair to a half-chair conformation
`(Figure 4). In gas-phase optimizations of ligand 1 with
`the present force field, the adopted ring conformation
`is favored by 1.3 kcal/mol. This is likely an artifact of
`using the AMBER C2-N-CH bending parameters with
`ı0 ) 118°, which was not designed for a piperidine
`ring.50 The difference is expected to have little effect
`on the computed free energy changes since the mutated
`phenyl rings are not in contact with the pipecolyl ring
`or, in the case of the ring contraction, the chair con-
`formation was enforced (vide infra).
`Free energy changes were calculated during the MC
`simulations according to standard procedures of statis-
`tical perturbation theory.51-53 The difference in free
`energy of binding (¢¢Gb) for molecule B relative to
`molecule A (Scheme 1) may be obtained from transfor-
`mations of the ligands in solution and bound to the
`protein according to eq 1:
`¢¢Gb(AfB) ) ¢GB - ¢GA ) ¢GFKBP - ¢Gaq
`FEP Simulation Protocol for 1f2. This perturba-
`
`(1)
`
`

`
`3932 Journal of Medicinal Chemistry, 1998, Vol. 41, No. 21
`
`Lamb and Jorgensen
`
`Figure 4. Stereoviews of unbound ligand 1. The initial geometry from the 1-FKBP12 crystal structure has the pipecolic ester
`substituent in an axial conformation (top). Subsequent equilibration resulted in partial inversion of the ring (bottom).
`
`tion involved the removal of the 1-phenyl ring of 1 to
`obtain 2. The atoms of the phenyl group were converted
`to “dummy” atoms without charge or Lennard-Jones
`parameters, and the length of the bond connecting the
`substituent to the remainder of the ligand was reduced
`to 0.65 Å with all other phenyl ring bonds reduced to
`0.35 Å. The transformation of 1f2 was carried out in
`13 windows with double-wide sampling, which yield 26
`free energy increments.51 A coupling parameter, (cid:236), was
`employed such that (cid:236) ) 0 corresponds to the initial
`state, 1, and (cid:236) ) 1 corresponds to the final state, 2. The
`first six windows used ¢(cid:236) ) (0.025, while the remain-
`ing windows used ¢(cid:236) ) (0.050. All were equilibrated
`with 2-4M configurations of sampling; the last config-
`uration of the previous window was used to start the
`next one. Averaging was done in batches of 2 (cid:2) 105
`configurations, with data collected over a total of 4-7M
`configurations in each window. For subsequent analy-
`ses of hydrogen bonding, an additional 1M configura-
`tions were generated at the endpoints of the simula-
`tions.
`FEP Simulation Protocol for 2f3, 5f6. The next
`transformation addressed was the conversion of a phen-
`yl moiety to a 3-pyridyl ring. This perturbation is
`straightforward; the analogous perturbation of benzene
`to pyridine had been performed in the development of
`OPLS all-atom (OPLS-AA) parameters for pyridine.54
`As before, the standard phenyl ring structure was
`transformed to a pyridine geometry determined from
`microwave experiments.54 A model of 5 was required
`prior to the conversion of 5f6 and was obtained by
`mapping a prolyl ring onto the final structure of 2 from
`the 1f2 FEP calculation. In each simulation, the prolyl
`or pipecolyl ring was flexible. The perturbation protocol
`for these calculations was slightly modified from that
`
`used for 1f2 to take advantage of the acquisition of a
`new parallel computing system within our laboratory.
`Seven double-wide windows were run in parallel, with
`4-8M configurations sampled during the equilibration
`phase and with data collected over 4-12M configura-
`tions. A gas-phase FEP calculation was also performed
`for 5f6 to allow estimation of the relative free energies
`of hydration of the two ligands.
`FEP Simulation Protocol for 2f5, 1f4. In our
`experience, perturbations between different cyclic sys-
`tems require much care to implement and can be
`particularly slow to converge. The necessity of account-
`ing for both changes in bonded and nonbonded interac-
`tions within the ring as one atom disappears makes this
`a technically difficult perturbation. One way to simplify
`the present calculations is to drive the ring from one
`fixed six-membered ring conformation to a fixed five-
`membered ring conformation, “disappearing” the re-
`maining atom and simultaneously reeling it in toward
`the others. For this rigid perturbation, changes in
`energy within the ring need not be monitored, as these
`intraligand differences should be very similar in each
`environment (bound and unbound). However, other
`possible conformations for the rings would not be taken
`into account, and the results could be sensitive to the
`path chosen.
`The simulations for the unbound and bound trans-
`formations were started from the final bound conforma-
`tions of 2-FKBP12 or 1-FKBP12 above with the chair
`conformation for the pipecolyl ring. The final prolyl ring
`geometry was obtained from a gas-phase optimization
`of the bound conformation of 2 with one ring atom
`converted to a dummy atom, as illustrated in Figure 5.
`Other than for the internal structures of the pipecolyl
`and prolyl rings, the sampling for the ligands included
`
`

`
`Neurotrophic Inhibitors of FK506 Binding Protein
`
`Journal of Medicinal Chemistry, 1998, Vol. 41, No. 21 3933
`
`Figure 5. The initial and final ring conformations of the 2f5 perturbation. The dummy atom in 5 is noted Du.
`
`all bond angles and dihedral angles, as before. Prior to
`the FEP calculation, the unbound ligands, 2 and 1, were
`each resolvated and relaxed with 4.4M configurations
`of solvent-only sampling, followed by 8M configurations
`of full equilibration. Twenty-one windows were used
`to perform the ring contraction in small steps. Fortu-
`nately, convergence was rapidly achieved within the 4M
`configurations performed for both equilibration and
`averaging.
`
`Results and Discussion
`Effect of the 1-Phenyl Group, 1f2. Removal of
`the 1-phenyl moiety from 1 is a large perturbation but
`a computationally attractive choice given the available
`structural information and the sizable difference in
`binding affinity (Table 1). The free energy change as a
`function of (cid:236) from the FEP calculations for the unbound
`and bound ligands proceeded smoothly (Figure 6a), with
`¢Gaq ) 10.25 ( 0.31 kcal/mol and ¢GFKBP ) 11.69 (
`0.31 kcal/mol (Table 2). The resultant relative free
`energy of binding (¢¢Gb) of 1.4 kcal/mol obtained
`according to Scheme 1 is then in excellent agreement
`with the experimental observations of 1.4-1.6 kcal/mol
`(Table 2).
`A comparison of the averaged structures from the
`simulations with the crystal structure yields several
`interesting observations. First, the average root-mean-
`squared (rms) deviations for non-hydrogen atoms be-
`tween the structures sampled for the complex of 1 and
`the crystal structure13 were computed. The average rms
`for the atoms in the side chains that were varied is 0.7
`Å, and the average rms for the ligand 1 is 1.1 Å. The
`corresponding maximum rms values for individual
`structures did not exceed 0.8 and 1.4 Å, and the values
`at the end of the MC run were 0.7 and 0.9 Å. Thus, the
`MC sampled structures did not drift far from the crystal
`structure, though some differences emerged.
`In the
`R-ketoamide region of the ligand, the crystallographic
`interaction of O4 with an (cid:15)-hydrogen of Phe99 is not
`maintained, but there is a frequent interaction of the
`œ-hydrogen with the amide carbonyl oxygen (O3). The
`hydrogen bonds to Ile56 and Tyr82 are unaffected by the
`perturbation (Figure 2). Within compound 1, the 1-phen-
`yl and isopentyl groups remain in contact. However,
`the 1-phenyl group moves away some from Phe46; the
`shortest contact between aromatic carbons increases
`from 4.5 Å in the crystal structure to 5.6 Å in the
`simulation. In both the crystal structure and from the
`
`Figure 6.
`(a) Free energy profile for the transformation of
`1f2, with error bars for each window shown. (b) Profiles for
`2f3 and 5f6 (squares).
`(c) Profiles for 2f5 and 1f4
`(squares). Solid lines represent the unbound simulations, and
`the dashed lines result from simulations of FKBP12-ligand
`complexes.
`
`MC simulations, the 1-phenyl and isopentyl groups pack
`well into the hydrophobic pocket that is outlined by
`Phe46, Phe36, Ile90, Ile91, Tyr82, and His87 (Figure 2). Loss
`of hyrdrophobic contacts upon removal of the 1-phenyl
`group is unfavorable for binding.
`The simulations also suggest that some specific
`contacts with the 1-phenyl group may be relevant. As
`highlighted in Figure 7, the 1-phenyl substituent makes
`aryl CH(cid:226)(cid:226)(cid:226)O contacts with the backbone oxygen of Glu54
`and the Tyr82 hydroxyl oxygen, and it has an amino-
`aromatic interaction with the (cid:15)-nitrogen of His87. While
`the interactions of the 1-phenyl group in 1 with Tyr82
`and Glu54 are found in most of the structures analyzed,
`the interaction with His87 occurs in only 29% of the
`analyzed structures. Average distances and frequencies
`
`

`
`3934 Journal of Medicinal Chemistry, 1998, Vol. 41, No. 21
`
`Lamb and Jorgensen
`
`Table 2. Experimental Binding Free Energies, Calculated Free Energy Changes, and a Comparison of Experimental and Calculated
`Relative Binding Free Energiesa
`
`exptlb
`
`calcd
`exptlc
`AfB
`¢GA
`¢¢Gb
`¢¢Gb
`¢GFKBP
`¢Gaq
`¢GB
`11.69 ( 0.31
`10.25 ( 0.31
`-9.0, -9.5
`-10.6, -10.9
`1f2
`1.4
`1.6, 1.4
`-0.4, 0.3
`2.32 ( 0.08
`1.52 ( 0.09
`-9.4, -9.2
`-9.0, -9.5
`2f3
`0.8
`-1.0
`1.48 ( 0.07
`0.59 ( 0.08
`-11.1
`-10.1
`5f6
`0.9
`-1.7
`-1.1
`-10.07 ( 0.10
`-8.36 ( 0.09
`-10.1
`-9.0, -9.5
`2f5
`-1.9
`-8.76 ( 0.11
`-6.86 ( 0.11
`-9.9
`-10.6, -10.9
`1f4
`0.7
`a All free energies in kcal/mol. b Absolute binding free energies are derived from rotamase inhibition data given in Table 1, using ¢G
`) RT ln Ki and T ) 25 °C (298 K). c Relative binding free energies are only listed when experimental data from the same source may be
`compared.
`
`calcd
`
`tion, 2f3 and 5f6. Experimental data from Guilford
`Pharmaceuticals suggests that modification to a pyridyl
`substituent improves binding of 3-phenylpropyl com-
`pounds, although in one case this is contradicted by data
`obtained at SmithKline Beecham (Table 2). Considering
`the ca. 4 kcal/mol more favorable free energy of hydra-
`tion of pyridine than of benzene54 and assuming that
`the pyridyl compounds bind in a manner similar to 1
`in the solvent-exposed region of the binding pocket, the
`phenyl to pyridyl conversion would be expected to
`decrease binding affinity. Simply put, a dipole is better
`solvated in a medium with a higher average dielectric
`constant.
`First, the solvation of the ligands was addressed. The
`aqueous transformation of 2f3 resulted in a free energy
`difference of 1.52 ( 0.09 kcal/mol, somewhat larger than
`that for 5f6 (0.59 ( 0.08 kcal/mol). The gas-phase
`transformation of 5 to 6 yielded a free energy change of
`3.15 ( 0.04 kcal/mol, which combines for a net relative
`free energy of hydration of -2.6 kcal/mol favoring 6.
`This is reasonable given the previous benzene to pyri-
`dine results,54 and the larger, flexible ligand. Hydrogen
`bonds to water were similar for both sets of ligands; as
`expected, the pyridyl nitrogen provides an additional
`acceptor site in 3 and 6 (Table 4).
`The free energy profiles for both pairs of unbound and
`bound perturbations are displayed in Figure 6b. They
`are again notably smooth. As seen from Table 2, the
`transformations in the protein are ca. 1 kcal/mol less
`favorable than those in solution, and the net result is a
`consistent preference for the phenylpropyl ligands. No
`direct protein contacts are made by the pyridyl nitrogen
`atom; its interaction with water is maintained (Table
`4). For this hydrogen bond, the optimal interaction
`energy is -6.2 kcal/mol,54 much stronger than an aryl
`CH(cid:226)(cid:226)(cid:226)O- or N- interaction. However, the pyridyl
`nitrogen does participate in a hydrogen-bonding net-
`work with water molecules and carbonyl oxygens in the
`protein backbone. As reflected in Figure 9, a two-water
`bridge between the oxygen of Val55 and the nitrogen is
`found consistently in 6-FKBP12. In 3-FKBP12 the
`interaction is often mediated by three water molecules,
`with a few structures in which one molecule also bridges
`to the Gln53 carbonyl oxygen (Figure 9). The Val55
`oxygen has been noted previously as a consensus
`hydration site in FKBP12-ligand complexes.57 Crystal-
`lographic waters from the 1-FKBP12 structure13 were
`not explicitly included in the calculations, so it is
`gratifying that this interaction is established during the
`MC simulations. Water surrounding the ligand in the
`binding pocket can also bridge longer distances; for
`example, in 6-FKBP12, four molecules link the hy-
`
`Figure 7. Intermolecular aryl CH(cid:226)(cid:226)(cid:226)N, O contacts with His87,
`Tyr82, and Glu54 made in 1-FKBP12 that are lost on trans-
`formation to 2-FKBP12. One representative configuration
`from the Monte Carlo simulation is illustrated.
`
`of occurrence of these and other key interactions be-
`tween the ligands and FKBP12 are summarized in
`Table 3. Such aryl interactions are commonly observed
`in protein crystal structures, and their orientational
`distributions have been analyzed.55,56 To provide a
`sense of the strength of the aryl CH(cid:226)(cid:226)(cid:226)X interactions
`with the OPLS-AA force field, gas-phase optimizations
`for complexes of benzene with phenol, imidazole, N-
`methylacetamide, and water yielded the interaction
`energies shown in Figure 8. With the hydrogen bonds
`constrained to be linear, the intrinsic interaction ener-
`gies are in the 1-2 kcal/mol range.
`In both complexes, the 3-phenylpropyl moiety remains
`in the FK506-cyclohexyl region of FKBP12, and there
`is usually a water molecule well-positioned on the face
`of one or both phenyl rings in 1 (Figure 7). For
`reference, in the gas phase, the optimal (cid:240) hydrogen bond
`between benzene and a water molecule has an interac-
`tion energy of ca. -3.5 kcal/mol.42 An aromatic hydro-
`gen of the 3-phenylpropyl group often interacts with the
`carbonyl oxygen of Val55 in 1-FKBP12; this contact is
`shorter and more frequent in 2-FKBP12 (Table 3). In
`addition, while the 1-phenyl ring no longer interacts
`with the Glu54 backbone carbonyl oxygen when the
`transformation to 2 is complete, the 3-phenyl ring shifts
`to pick up this contact. Considering these observations,
`it is likely that the major contribution to the weaker
`binding for 2 than 1 is the overall reduction in hydro-
`phobic interactions and, possibly, the specific loss of the
`aryl CH(cid:226)(cid:226)(cid:226)O, N contacts between the 1-phenyl group and
`Tyr82 and His87.
`Effect of the 3-Phenyl to 3-(3-Pyridyl) Substitu-
`
`

`
`Neurotrophic Inhibitors of FK506 Binding Protein
`
`Journal of Medicinal Chemistry, 1998, Vol. 41, No. 21 3935
`
`Table 3. Key Intermolecular Hydrogen Bond Distances (Frequencies %) Less than 3.2 Åa
`
`1f2b
`
`2f3c
`
`ligand
`FKBP12
`O4
`Tyr26H(cid:15)
`Tyr26HŁ
`O4
`Phe36H(cid:15)
`O3
`Phe36H(cid:15)
`O4
`Gln53O
`H22
`Glu54O
`H23
`Glu54O
`H25
`Val55O
`H21
`Val55O
`H22
`Val55O
`H23
`Ile56H
`O2
`H19
`Tyr82C(cid:15)
`Tyr82H(cid:15)
`C19
`Tyr82H(cid:15)
`O3
`Tyr82HŁ
`O3
`Tyr82HŁ
`O1
`Tyr82OŁ
`H29
`His87N(cid:15)
`H28
`Phe99Hœ
`2.9 (100)
`O3
`Phe99Hœ
`2.9 (22)
`O4
`Phe99H(cid:15)
`3.0 (46)
`2.9 (72)
`2.9 (22)
`O4
`a Only interactions found in >10% of saved structures are reported. The frequency in parentheses records the percentage of the analyzed
`structures that had the feature. b 1f2 for 45 structures saved every 2 (cid:2) 105 configurations. c 2f3 and 5f6 for 50 structures saved every
`2 (cid:2) 105 configurations.
`
`5f6c
`
`2.8 (92)
`3.1 (16)
`
`2.6 (100)
`2.8 (96)
`2.8 (100)
`
`3.1 (26)
`3.0 (40)
`
`2.0 (100)
`3.0 (20)
`
`3.0 (60)
`1.8 (100)
`
`2.8 (96)
`3.0 (42)
`
`2.6 (98)
`
`2.9 (68)
`
`2.9 (40)
`2.0 (100)
`3.0 (76)
`3.0 (28)
`3.1 (52)
`1.8 (100)
`3.0 (74)
`
`2.7 (100)
`
`3.0 (82)
`2.8 (96)
`3.0 (53)
`2.8 (93)
`
`2.8 (91)
`
`2.9 (60)
`2.0 (100)
`3.0 (69)
`3.1 (42)
`2.9 (93)
`1.8 (100)
`3.1 (31)
`2.8 (93)
`2.9 (29)
`2.9 (78)
`
`2.8 (96)
`2.9 (62)
`3.0 (16)
`2.7 (96)
`
`2.8 (84)
`
`2.7 (96)
`
`2.0 (100)
`3.0 (49)
`3.0 (69)
`3.0 (69)
`1.8 (100)
`3.1 (44)
`
`2.7 (98)
`3.0 (62)
`3.0 (56)
`2.7 (96)
`
`2.8 (88)
`
`2.8 (84)
`3.0 (34)
`2.0 (100)
`3.1 (16)
`3.1 (38)
`3.0 (70)
`1.8 (100)
`3.0 (60)
`
`2.9 (94)
`2.9 (44)
`3.0 (22)
`2.6 (98)
`
`2.8 (86)
`
`3.1 (10)
`2.1 (100)
`3.0 (52)
`3.1 (40)
`3.0 (92)
`1.8 (100)
`3.0 (98)
`
`2.7 (96)
`
`2.6 (100)
`
`2.8 (92)
`
`droxyl oxygen of Tyr26 to the backbone carbonyl of Glu54
`(not shown).
`Finally, to ensure that the phenylpropyl 3-position
`selected for transformation did not influence possible
`protein-ligand contacts or the computed free energy
`difference, the final FEP window of 5f6 bound to
`FKBP12 was also started with the 3-pyridyl ring flipped
`by 180°. The free energy change calculated in this
`window was unaffected, and no new contacts with the
`protein were observed. In all, the simulations support
`the intuitive idea that binding for 3-(3-pyridyl)propyl
`compounds should be less favorable than that for
`3-phenylpropyl ligands. This finding agrees qualita-
`tively with the SmithKline observations for 3 and 2, but
`not with either pair of binding affinities (3 vs 2 or 6 vs
`5) reported by Guilford Pharmaceuticals, as sum-
`marized in Table 2.
`Effect of Ring Contraction on Binding, 2f5 and
`1f4. With the relatively hydrophobic binding pocket
`of FKBP12, one might expect the larger pipecolyl
`compounds to bind with higher affinity than prolyl
`homologues. Though this is true for most published
`FKBP12 inhibitors,2 the opposite pattern is reported for
`the 3-phenylpropyl and 3-(3-pyridyl)propyl compounds
`in Table 1.
`With this in mind, the 2f5 transforma