.elsevier.com/locate/bba

Biochimica et Biophysica Acta

Review

Target flexibility in molecular recognition

J. Andrew McCammon *

Howard Hughes Medical Institute, La Jolla, CA 92093-0365, USA

NSF Center for Theoretical Biological Physics, La Jolla, CA 92093-0365, USA

Department of Chemistry and Biochemistry, University of California at San Diego, La Jolla, CA 92093-0365, USA

Department of Pharmacology, University of California at San Diego, La Jolla, CA 92093-0365, USA

Received 2 June 2005; received in revised form 9 July 2005; accepted 10 July 2005

Available online 12 September 2005

Abstract

Induced-fit effects are well known in the binding of small molecules to proteins and other macromolecular targets. Among other targets, protein

kinases are particularly flexible proteins, so that such effects should be considered in attempts at structure-based inhibitor design for kinase targets.

This paper outlines some recent progress in methods for including target flexibility in computational studies of molecular recognition. A focus is

the ‘‘relaxed complex method,’’ in which ligands are docked to an ensemble of conformations of the target, and the best complexes are re-scored to

provide predictions of optimal binding geometries. Early applications of this method have suggested a new approach to the development of

inhibitors of HIV-1 Integrase.

D 2005 Elsevier B.V. All rights reserved.

Keywords: Structure-based drug discovery; Computer-aided drug design; Induced fit; Molecular dynamic; Computer simulation; Free energy

Computer-aided drug discovery has become increasingly

successful in the past 20 years, due to the increasing

availability of experimental structures of molecular targets,

the inexorable increases in the performance of computer

hardware, and the creation of new theory, algorithms and

software. The first clinically useful drugs to emerge from

molecular dynamics simulations (used both in the refinement

of crystallographic structures and in the computational docking

of model compounds to target structures) were the HIV

protease inhibitors [1]. Many useful computational methods

have been introduced for structure-based drug discovery [2].

Here, we focus on the ‘‘Relaxed Complex Method’’, which has

been developed in our group for the docking of potential

inhibitors to intrinsically flexible targets.

1. The Relaxed Complex Method

The Relaxed Complex Method is a computational approach

to discover ligands that may bind even when substantial

‘‘induced fit’’ effects occur in their target molecules [3–5]. The

Relaxed Complex method was inspired by two experimental

1570-9639/$ - see front matter D 2005 Elsevier B.V. All rights reserved.

doi:10.1016/j.bbapap.2005.07.041

* Howard Hughes Medical Institute, La Jolla, CA 92093-0365, USA.

E-mail address: jmccammon@ucsd.edu.

methods for rapid discovery of ligands that bind strongly to a

receptor, namely the ‘‘SAR by NMR’’ method [6] and the

‘‘tether method’’ [7]. These methods recognize that ligands may

bind to conformations that occur only rarely in the dynamics of

the receptor, and that strong binding often reflects multivalent

attachment of the ligand to the receptor. The new computa-

tional approach includes single ligand and double ligand

variants.

The basic element of the new method is the automated

flexible docking of small libraries of compounds to a diverse

selection of target conformations. The first phase of the

approach involves generating the target conformations. This

might make use of a long molecular dynamics simulation of the

unliganded target molecule, an ‘‘accelerated’’ molecular dy-

namics simulation that samples conformational space more

effectively [8], or some other way of generating target

conformations. The second phase involves the rapid docking

of mini-libraries of candidate inhibitors to the selected set of

conformational snapshots of the target. In this phase, a

relatively simple scoring algorithm is used to allow fast

docking. The third phase attempts to improve the scoring of

the best complexes found in the docking calculations by use of

a slower but more accurate algorithm for estimating the

standard free energies of binding.

1754 (2005) 221 – 224

http://www

J.A. McCammon / Biochimica et Biophysica Acta 1754 (2005) 221–224222

The scheme described above represents the single ligand

method. The double ligand variant recognizes that two ligands

with relatively low binding affinities to the target can be linked

to form a high-affinity ligand. Because the binding of the first

ligand could introduce unfavorable interactions for the binding

of the second ligand, the combination of the best-ranked

ligands for respective binding sites does not necessarily

produce the best composite compound. Continuing from the

previous single-ligand studies, the first ligand may therefore be

treated as part of the target, and the docking simulations of the

second ligand may be repeated in a limited search space, based

on the allowable lengths of linkers. Again, the binding of the

second ligand is subsequently re-scored by other more accurate

approaches.

2. Simple docking and rescoring to ensembles of protein

conformations

The first applications of the Relaxed Complex methods

focused on an experimentally well-characterized system, FKBP

[3,4]. A long molecular dynamics calculation was used to

sample the FKBP conformations, and the AutoDock software

[9] was used for the initial docking. The re-scoring was done

using the MM/PBSA routines from the AMBER software [10]

and APBS evaluation of the electrostatic energies [11]. The

first paper [3] considered the binding of compounds 2 and 9

from the ‘‘SAR by NMR’’ paper by Shuker et al. [6] to

snapshots obtained from a 2-ns molecular dynamics calcula-

tion. It was shown that the binding of the ligands is quite

sensitive to conformational fluctuations of the target protein

FKBP-12, even though the latter is a relatively rigid protein. In

particular, with the AutoDock 3.0.5 scoring function, the

binding energies of compound 2 covered a range of 3 to 4 kcal/

mol; this corresponds to a 100- to 1000-fold difference in

binding affinities of the same ligand for slightly different

conformations of the target protein (Fig. 1).

In the second paper [4], re-scoring was done using the MM/

PBSA approach [10]. The solutions of the Poisson–Boltzmann

equation were obtained using the APBS software [11]. As in

Fig. 1. Compares experimental findings and our relaxed complex docking results. O

judged by Shuker’s SAR by NMR chemical shifts (from Fig. 3 of [6]). On the rig

scheme. As can be seen, the computed quaternary structure correlates well with th

the first paper, significant ranges of binding energies were

found for the ligands (dimethylsulfoxide, 4-hydroxy-2-buta-

none, and tetrahydrothiophene-1-oxide, in this case). These

variations result in part from steric effects, since the difference

between the largest and smallest solvent accessible molecular

surface of the FKBP-12 binding site is found to be about 187

A2. For these ligands, use of the MM/PBSA re-scoring allowed

the correct prediction of the binding modes, in comparison to

the crystallographic structures, even though these ligands had

weak affinities for the target. The MM/PBSA re-scoring has

proven successful in ranking a number of ligands that bind to

the FK506 binding protein FKBP-12 [4].

With the advent of a new docking algorithm (the Lamarck-

ian genetic algorithm) and a very successful empirical free

energy function, AutoDock [9] is able to perform very efficient

docking of large flexible ligands and so has been used in our

Relaxed Complex scheme. The so-called Lamarckian genetic

algorithm is the hybrid of the original Genetic Algorithm [12]

with the adaptive local search method. The local searcher

modifies the phenotype, which is allowed to update the

genotype. The so-called genome in the genetic algorithm

consists of floating point ‘‘genes’’, each of which encodes one

state variable describing the molecular position, orientation and

conformation. The ligand begins randomly outside the protein,

and explores translations, orientations, and conformations until

an ideal site is found. In order to maintain the consistency of

the free energy function parameters (see below), the restrained

electrostatic potential (RESP) method [13] has been used to

derive the partial charges of the ligands.

3. Rescoring with more accurate free energy calculations

The third phase of the Relaxed Complex method improves

the scoring of the best complexes found in the docking

calculations. This is done by using a type of simulation inspired

by the MM/PBSA (Molecular Mechanics/Poisson–Boltzmann,

Surface Area) Method to calculate more accurate free energies

of binding for a number of best-ranked complexes [10]. In the

MM/PBSA method, protein–ligand complexes that have been

n the left of the figure is the complex of FKBP-12 with compounds 2 and 9 as

ht is the complex generated by our Relaxed Complex computational docking

e observed complex structure.

J.A. McCammon / Biochimica et Biophysica Acta 1754 (2005) 221–224 223

subject to molecular dynamics simulations with an explicit

solvent are post-processed with a continuum solvent model to

estimate the free energy of binding of the ligand to the protein.

Typically, the ligand and protein are separated and kept in fixed

conformations corresponding to that of the complex. The

solvation energies are then calculated using the PBSA method;

the Poisson–Boltzmann equation provides an estimate of the

electrostatic contributions to solvation, and the Surface Area

method is used to provide a simple estimate of the nonpolar

contributions to solvation. The advantage of replacing the

explicit solvent by the continuum model is that it avoids the

extensive sampling of configurations needed to achieve

converged estimates of the solvation free energy in the explicit

case. Molecular Mechanics (MM) is used to account for the

direct interactions between ligand and protein in the complex.

We have recently improved on the MM/PBSA approach in

several ways [14]. Key to this has been the estimation of the

values of the configuration integrals and corresponding

standard free energies, based on statistical mechanical theory

[15]. The PB (Poisson–Boltzmann) calculations have been

performed using our new APBS software [11].

4. Double-ligand Relaxed Complex method

Two ligands with low binding affinities (e.g., dissociation

constants in the millimolar range) to a target protein can be

linked to form a high-affinity ligand. Therefore, it may be

possible to design a potent drug by combining two or more

ligands with relatively weak affinities. However, the binding of

Fig. 2. This highlights the progression from the single ligand Relaxed Complex m

snapshot from the MD run of FKBP-12 (the single ligand method). The best-ranked

then treated as part of the receptor, and compound 9 was then docked to the sub-re

the first ligand could introduce unfavorable interactions for the

binding of the second ligand; thus, the combination of the best-

ranked ligands for respective binding sites does not necessarily

produce the best composite compound. Here, computational

approaches can help elucidate the complex binding relation-

ships with atomic detail. Continuing from the previous single-

ligand studies, the first ligand can be treated as part of the

receptor, and the docking simulations of the second ligand can

be repeated using a limited search space, based on the

allowable lengths of linkers. Again, the binding of the second

ligand would be subsequently re-scored by MM/PBSA and

other approaches. Fig. 2 depicts preliminary work published by

Lin et al. [4].

5. Perspective and application to drug discovery

The Relaxed Complex Method has been introduced to help

account for the effects of target flexibility in computational

studies of molecular recognition and binding. Because it

involves the docking of full molecules, it is complementary

to other methods such as the Dynamic Pharmacophore Method,

which involves the docking of functional group probes to an

ensemble of target conformations [16,17]. The latter method

seems particularly well suited for somewhat earlier, higher-

throughput stages of a drug discovery program, because it

yields a pharmacophore that represents a consensus among a

number of somewhat different target conformations. Methods

that use soft harmonic modes to sample receptor conformations

have also proven to be very fast and effective [18]. The

ethod to the double ligand method. Compound 2 was docked to every 10th

complex of compound 2 with FKBP-12 is shown on the left. Compound 2 was

gion of FKBP-12 that was within a possible linker distance from compound 2.

J.A. McCammon / Biochimica et Biophysica Acta 1754 (2005) 221–224224

Relaxed Complex Method is more likely to be useful in the

later stages of a drug discovery program, since it is generally

more computationally demanding. But, despite its recent

origin, the Relaxed Complex Method has already proven

valuable in suggesting a new approach to the development of

inhibitors of HIV-1 Integrase [5].

Acknowledgments

The author thanks his former postdoctoral fellow Jung-Hsin

Lin (now Assistant Professor, National Taiwan University) and

graduate students Alex Perryman and Julie Schames Pressman

for their very important contributions to the work that is

reviewed here. This work has been supported in part by the

Howard Hughes Medical Institute, the National Institutes of

Health, the National Science Foundation, the NSF Center for

Theoretical Biological Physics, the W. M. Keck Foundation,

the National Biomedical Computing Resource, the San Diego

Supercomputer Center, and Accelrys Inc.

References

[1] C.N. Hodge, T.P. Straatsma, J.A. McCammon, A. Wlodawer, Rational

design of HIV protease inhibitors, in: W. Chiu, R.M. Burnett, R.

Garcea (Eds.), Structural Biology of Viruses, Oxford Univ. Press,

1997, pp. 451–473.

[2] C.F. Wong, J.A. McCammon, Protein flexibility and computer-aided drug

design, Annu. Rev. Pharmacol. Toxicol. 43 (2003) 31–45.

[3] J.H. Lin, A. Perryman, J. Schames, J.A. McCammon, Computational drug

design accommodating receptor flexibility—the relaxed complex scheme,

J. Am. Chem. Soc. 124 (2002) 5632–5633.

[4] J.H. Lin, A. Perryman, J. Schames, J.A. McCammon, The relaxed

complex method: accommodating receptor flexibility for drug design

with an improved scoring scheme, Biopolymers 68 (2003) 47–62.

[5] J. Schames, R.H. Henchman, J.S. Siegel, C.A. Sotriffer, H. Ni, J.A.

McCammon, Discovery of a novel binding trench in HIV integrase, J.

Med. Chem. 47 (2004) 1879–1881.

[6] S.B. Shuker, P.J. Hajduk, R.P. Meadows, S.W. Fesik, Discovering

high-affinity ligands for proteins: SAR by NMR, Science 274 (1996)

1531–1534.

[7] D.A. Erlanson, A.C. Braisted, D.R. Raphael, M. Randal, R.M. Stroud,

E.M. Gordon, J.A. Wells, Site-directed ligand discovery, Proc. Natl. Acad.

Sci. U. S. A. 97 (2000) 9367–9372.

[8] D. Hamelberg, J. Mongan, J.A. McCammon, Accelerated molecular

dynamics: a promising and efficient simulation method for biomolecules,

J. Chem. Phys. 120 (2004) 11919–11929.

[9] G.M. Morris, D.S. Goodsell, R.S. Halliday, R. Huey, W.E. Hart, R.K.

Bellew, A.J. Olson, Automated docking using a Lamarckian genetic

algorithm and an empirical binding free energy function, J. Comp. Chem.

19 (1998) 1639–1662.

[10] W. Wang, O. Donini, C.M. Reyes, P.A. Kollman, Biomolecular simula-

tions: recent developments in force fields, simulations of enzyme

catalysis, protein– ligand, protein–protein, and protein–nucleic acid

noncovalent interactions, Ann. Rev. Biophys. Biomol. Struct. 30 (2001)

211–243.

[11] N.A. Baker, D. Sept, S. Joseph, M.J. Holst, J.A. McCammon, Electro-

statics of nanosystems: application to microtubules and the ribosome,

Proc. Natl. Acad. Sci. U. S. A. 98 (2001) 10037–10041.

[12] J.H. Holland, Adaptation in Natural and Artificial Systems, University of

Michigan Press, Ann Arbor, MI, 1975.

[13] C.I. Bayly, P. Cieplak, W.D. Cornell, P.A. Kollman, A well-behaved

electrostatic potential based method using charge restraints for deriving

atomic charges—the RESP model, J. Am. Chem. Soc. 115 (1993)

9620–9631.

[14] J.M.J. Swanson, R.H. Henchman, J.A. McCammon, Revisiting free

energy calculations: a theoretical connection to MM/PBSA and direct

calculation of the association free energy, Biophys. J. 86 (2004) 67–74.

[15] M.K. Gilson, J.A. Given, B.L. Bush, J.A. McCammon, The statistical–

thermodynamic basis for computation of binding affinities: a critical

review, Biophys. J. 72 (1997) 1047–1069.

[16] H.A. Carlson, J.A. McCammon, Accommodating protein flexibility in

computational drug design, Mol. Pharmacol. 57 (2000) 213–218.

[17] H.A. Carlson, K.M. Masukawa, K. Rubins, F.D. Bushman, W.L.

Jorgensen, R.D. Lins, J.M. Briggs, J.A. McCammon, Developing a

dynamic pharmacophore model for HIV-1 integrase, J. Med. Chem. 43

(2000) 2100–2114.

[18] A. May, M. Zacharias, Accounting for global protein deformability during

proteinprotein and proteinligand docking. Bioch. Biophys. Acta (this

volume). doi:10.1016/j.bbapap.2005.07.045