PROTEINS: Structure, Function, and Bioinformatics 67:418–434 (2007)Protein–Protein Recognition and Interaction Hot Spots i...
INTERACTION HOT SPOTS IN ANTIGEN–ANTIBODY COMPLEX                                         419However, much remains to be u...
420                                                     LAFONT ET AL.and antibody chain B with lysozyme chain M. As there ...
INTERACTION HOT SPOTS IN ANTIGEN–ANTIBODY COMPLEX                                              421to the free energy. As w...
422                                                     LAFONT ET conformation of the protein–protein complex was u...
INTERACTION HOT SPOTS IN ANTIGEN–ANTIBODY COMPLEX                                     423peptide was computed. The radii o...
424                                                                  LAFONT ET AL.                                        ...
INTERACTION HOT SPOTS IN ANTIGEN–ANTIBODY COMPLEX                                                425                      ...
426                                                              LAFONT ET AL.   TABLE II. Individual Contributions to the...
INTERACTION HOT SPOTS IN ANTIGEN–ANTIBODY COMPLEX                                               427                       ...
428                                                                     LAFONT ET AL.                                     ...
INTERACTION HOT SPOTS IN ANTIGEN–ANTIBODY COMPLEX                                                      429proach. In parti...
430                                                             LAFONT ET AL.                    Fig. 9. Structural compar...
INTERACTION HOT SPOTS IN ANTIGEN–ANTIBODY COMPLEX                                         431decomposition based on MM/GBS...
Lafont proteins 2007
Lafont proteins 2007
Lafont proteins 2007
Upcoming SlideShare
Loading in …5

Lafont proteins 2007


Published on

Paper to read later

Published in: Technology
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Lafont proteins 2007

  1. 1. PROTEINS: Structure, Function, and Bioinformatics 67:418–434 (2007)Protein–Protein Recognition and Interaction Hot Spots inan Antigen–Antibody Complex: Free Energy DecompositionIdentifies ‘‘Efficient Amino Acids’’Virginie Lafont,1,2 Michael Schaefer,3 Roland H. Stote,4 Daniele Altschuh,2 and Annick Dejaegere1* `1 ´ ´ ´ Structural Biology and Genomics Department, UMR 7104, Institut de Genetique et de Biologie Moleculaire et Cellulaire, CNRS/INSERM/ULP, F-67404 Illkirch Cedex, France2 UMR7175-LC1 CNRS/ULP, ESBS, Parc d’Innovation, 67412 Illkirch Cedex, France3 ´ Laboratoire de Chimie Biophysique, Institut le Bel, Universite Louis Pasteur, 67000 Strasbourg, France4 ´ ´ Laboratoire de Biophysicochime Moleculaire, Institut de Chimie, LC3-UMR7177, Universite Louis Pasteur,67070 Strasbourg Cedex, France ABSTRACT The molecular mechanics Poisson– tions is difficult.1 There are several examples of naturalBoltzmann surface area (MM/PBSA) method was molecules that bind at protein interfaces2 and a growingapplied to the study of the protein–protein complex number of synthetic compounds that are aimed at per-between a camelid single chain variable domain turbing protein–protein interactions (for a recent review,(cAb-Lys3) and hen egg white lysozyme (HEL), and see Ref. 3). Progress in developing synthetic compoundsbetween cAb-Lys3 and turkey egg white lysozyme will benefit greatly from improved understanding of the(TEL). The electrostatic energy was estimated by energetics and dynamics of protein–protein recognition.solving the linear Poisson–Boltzmann equation. A Our current understanding of protein–protein interac-free energy decomposition scheme was developed to tions is based on structural analysis of protein–proteindetermine binding energy hot spots of each complex. complexes mostly determined by crystallography, on vari-The calculations identified amino acids of the anti- ous biophysical characterizations of protein–protein com-body that make important contributions to the inter- plexes that give insight into the kinetics and thermody-action with lysozyme. They further showed the influ- namics of protein–protein association,4 and on biochemi-ence of small structural variations on the energetics cal analyses that help identify amino acids important forof binding and they showed that the antibody amino recognition.acids that make up the hot spots are organized in An important concept that has emerged in the lastsuch a way as to mimic the lysozyme substrate. years is the notion of interaction energy ‘‘hot spots,’’ whichThrough further analysis of the results, we define is the idea that a handful of amino acids at the bindingthe concept of ‘‘efficient amino acids,’’ which can pro- interface make a dominant contribution to the binding af-vide an assessment of the binding potential of a par- finity.5,6 Mutation studies, mostly alanine scanning stud-ticular hot spot interaction. This information, in ies, have indeed shown7 that only a few mutations [nativeturn, can be useful in the rational design of small ? Ala] affect the free energy of binding by more thanmolecules that mimic the antibody. The implications 2 kcal/mol. This has led to a distinction between the func-of using free energy decomposition to identify tional epitope (i.e. ‘‘hot spot residues) and the structuralregions of a protein–protein complex that could be epitope (all residues that participate in the interface).8targeted by small molecules inhibitors are discussed.Proteins 2007;67:418–434. V 2007 Wiley-Liss, Inc. C The Supplementary Material referred to in this article can be foundKey words: MM/PBSA; molecular dynamics; macro- at molecular electrostatics; protein engi- Grant sponsor: French Ministry of Defense, Centre National de la neering; drug design; antibodies; ´ Recherche Scientifique, Institut National de la Sante et de la Recher- ´ ´ che Medicale, and Universite Louis Pasteur, Strasbourg, France. camelids Virginie Lafont’s current address is Department of Biology, Johns Hopkins University, 3400 North Charles Street, Baltimore, MD INTRODUCTION 21218-2685. Michael Schafer’s current address is Novartis Pharma AG, Kly- For a wide range of biological processes, including anti- beckstrasse 141, 4057 Basel, Switzerland.body–antigen interactions, hormone–receptor interac- *Correspondence to: Annick Dejaegere. Structural Biology andtions, signal transduction, and regulation, the formation ´ ´ Genomics Department, UMR 7104, Institut de Genetique et de Biolo- ´ gie Moleculaire et Cellulaire, CNRS/INSERM/ULP, BP 10142,of stable protein–protein complexes is of fundamental im- F-67404 Illkirch Cedex, France. E-mail: annick@igbmc.u-strasbg.frportance. Interfering with or modulating the formation of Received 20 May 2006; Revised 6 September 2006; Accepted 15 Sep-these complexes therefore offers attractive opportunities tember 2006for therapeutic intervention. However, the development of Published online 26 January 2007 in Wiley InterScience (www.small molecules that modulate protein–protein interac- DOI: 10.1002/prot.21259V 2007 WILEY-LISS, INC.C
  2. 2. INTERACTION HOT SPOTS IN ANTIGEN–ANTIBODY COMPLEX 419However, much remains to be understood of the role of are minimally sized, stable, and highly soluble, and there-residues surrounding the hot spots in creating a favorable fore offer a number of benefits in medical and biotechno-binding environment, on the balance between electro- logical applications (for a recent review, see Ref. 25). Highstatic, van der Waals, and hydrophobic interactions in the affinity VHH can be obtained after immunization of lla-formation of complexes, in entropic contributions to bind- mas and dromedaries with specific antigens.23,26 Alterna-ing, and in the role of molecular flexibility on the binding tively, phage antibody libraries can be used.27,28 VHHsprocess. that inhibit several enzymes22,23,26,28–31 have been identi- Computational and theoretical methods permit the sys- fied, as well as VHH that target antigens specific oftematic analysis of individual amino acid contributions to human tumors32,33 or specific of infectious agents.34the formation of protein complexes.9–13 They provide a VHHs against a mouse anti-DNA monoclonal antibodywealth of information that are complementary to experi- that are competitive binders of the original DNA antigenments and allow the determination of the structural and have also been recently identified and are shown to beenergetic consequences of mutations, as well as a better useful for molecular mimicry studies.35 VHHs against pro-understanding of the physical mechanism and origin of teins of interest in human disease have potential diagnos-the intermolecular interactions that drive binding. Free tic33 or therapeutic32 applications. Camelidae VHHs areenergy simulations methods14–16 have been successfully not only able to recognize a wide range of target proteins,used to predict the effect of point mutations on binding af- their high stability and small size makes them amenablefinity in macromolecular complexes (for a recent review, to structural studies, and an increasing number of VHH–see Ref. 17). Computational cost, however, puts practical antigen structures are available (for recent examples seelimitations on the number of point mutations that can be Refs. 36–38). The structural information from these com-analyzed by free energy simulations for a given complex. plexes is potentially useful for the development of smallSystematic analysis of the contribution of all amino acids molecule mimicks of the VHH.39 For this to be realized,to the stability of macromolecular complexes have mostly however, it is important to identify the important interac-relied on different empirical free energy functions11,12 as tions in the VHH–antigen complex. From an experimentalwell as on all-atom molecular mechanics methods using analysis alone, this is not necessarily evident. In this pa-implicit representation of the solvent (so called MM/PBSA per, we show, for the VHH cAb-Lys3–lysozyme complex,or MM/GBSA methods18). The MM/PBSA method has that an interaction energy hot spot exploited in the anti-been used to systematically analyze the contribution of all body–lysozyme complex is also used by the N-acetyl gluco-amino acids to the formation of several macromolecular samine (NAG) ring of saccharide substrates in subsite Ccomplexes. For computational efficiency, these analyses of the lysozyme binding pocket. We support the results ofwere performed on a single structure of the complex.10,19 the free energy decomposition approach by the mutationalThe generalized Born model for electrostatics has also analysis of four VHH positions. We further introduce therecently been applied to the free energy decomposition in concept of amino acid efficiency and show the potential ofprotein–protein complexes.13,20,21 The generalized Born free energy decomposition in extracting useful informa-model is faster, but introduces additional approximations tion for drug the electrostatics treatment with respect to numericalsolutions of the Poisson–Boltzmann equations. In the MATERIALS AND METHODSpresent work, we develop a computationally tractable, yetreliable, free energy decomposition protocol that can be We developed a protocol based on the MM/PBSAused to identify amino-acids that have a dominant role in method,13,18,40 in which conformations extracted from mo-binding. In the MM/PBSA approach presented here, we lecular dynamics simulations are processed by a combina-use solutions to the Poisson–Boltzmann equation to tion of methods using a simplified description for the sol-obtain the electrostatic contribution to binding. This is in vent to give an estimate of binding free energy. Individualcontrast to other recent applications of the methodol- contributions of each amino acid to the complex formationogy13,20,21 where the Generalized Born model was are estimated and important amino acid ‘‘hot spots’’ areemployed. In addition, multiple conformations from mo- identified.lecular dynamics trajectories are used in the analysis. Structures The protocol is used to analyze molecular recognition inthe complex between a camel single chain antibody vari- The coordinates of the camel single-domain VH anti-able domain (cAb-Lys3) and its antigen lysozyme.22,23 The body–lysozyme complexes were obtained from the Proteinrecombinant protein cAb-Lys3 is a single variable domain. Data Bank.41 Two structure files containing hen egg whiteSingle domain antigen binding fragments can be gener- lysozyme as the antigen were obtained: 1MEL (solved atated from the variable domain of Camelidae heavy chain ˚ ˚ 2.5 A resolution)22 and 1JTT (solved at 2.1 A resolution).42antibodies. Indeed, the Camelidae possess, in addition to A third structure file containing the coordinates for theclassical IgG molecules, a large fraction of naturally complex of the same antibody with turkey egg white lyso-occurring heavy chain antibodies that are devoid of light zyme as the antigen was also obtained from the Proteinchains.24 These single domains preserve the ability to ˚ Data Bank: 1JTP (solved at 1.9 A resolution).42 For thebind specifically and with high affinity to their antigens. structures 1MEL and 1JTP, there are two complexes perThese small antibodies, also called VHH or nanobodyTM asymmetric unit, antibody chain A with lysozyme chain L PROTEINS: Structure, Function, and Bioinformatics DOI 10.1002/prot
  3. 3. 420 LAFONT ET AL.and antibody chain B with lysozyme chain M. As there are TABLE I. MD Simulations Done forsignificant structural differences between the two com- Conformational Samplingplexes,42 calculations were carried out for each complex HIS Trajectory(AL and BM). Complex PDB ID Chains 111 {ps} The hydrogen atoms were added to the crystal structure cAb-Lys3/HEL MEL AL HSD 180using the HBUILD module43 of the CHARMM44 program cAb-Lys3/HEL MEL AL HSE 180(version 29a1); the all atom parameter set of CHARMM2245 cAb-Lys3/HEL MEL BM HSE 240was used for all calculations. Each structure was sub- cAb-Lys3/HEL JTT AL HSE 250jected to an energy minimization of 5000 steps using the cAb-Lys3/TEL JTP AL HSD 180steepest descent algorithm and 5000 steps using the cAb-Lys3/TEL JTP BM HSE 300adopted basis Newton Raphson algorithm. Harmonic con- Details for four simulations of the cAb-Lys3/Hen egg-white lysozymestraints, added initially to prevent any large perturbation complex, two simulations of the cAb-Lys3/Turkey egg-white lysozymeof structure, were gradually reduced over the course of complex. For each PDB crystallographic structure, both complexes ofthe minimization. The final steps of the minimization chains AþL and of chains BþM were used. The protonation state forwere done without constraints. The minimization was the residue His 111 was determined, see text (HSD/HSE: protonation on Nd/Ne, respectively). For the complex of chains AþL in PDB entrystopped when the gradient of the energy was less than MEL (cAb-Lys3/HEL), trajectories were run for each of the two proto- ˚10À4 kcal molÀ1 AÀ1. A distance dependent dielectric of 4r, nation states of His 111 (HSE and HSD).where r is the distance between two atoms, and a non- ˚bonded cutoff of 13 A were used during minimization. bonds between heavy atoms and hydrogen atoms wereDetermination of Protonation State constrained using SHAKE.50 With the protein–protein complex fixed, the water was equilibrated for 6 ps at 300 The pKa of titratable groups were determined for both K. The constraints were removed and the entire systemthe isolated proteins and for the complexes using contin- was again equilibrated for 6 ps. A time step of 1 fs wasuum electrostatic as described by Schaefer et al.46 The used in the simulation and structures were collected everymethod assumes that electrostatic effects dominate the 10 fs. Van der Waals interactions were truncated at a cut-influence of the surroundings on the pKa of titratable ˚ off distance of 12.5 A using a switch function. Electrostaticgroups. The change in pKa of titratable residues between ˚ interactions were truncated at 12.5 A using a shift func-the solution value (where experimental values for amino tion. All simulations contain between 35,500 and 35,700acid side chains are used) and the value in the protein in- atoms, which includes about 3800 atoms for the proteinterior is then calculated using continuum electrostatics. complex. Six trajectories ranging from 180 to 300 ps inWe used the UHBD program47 to calculate the free energy length were generated using different starting structuresof different states of protonation and a series of c-shell and initial conditions. The different simulations per-scripts46 to obtain the titration curves. The protonation formed are listed in Table I. The purpose of the simula-states of selected residues were determined at physiologi- tions was to generate multiple structures to improve thecal pH of 7.4. Residues located at the interface of the com- statistical sampling of the binding energy calculations.plex are those which are the most susceptible to undergoshifts in pKa when the complex is formed. Therefore we Free Energy Calculations and Free Energydetermined the protonation state of acidic and basic resi- Decompositiondues that are buried upon complexation, i.e. residues thathave a solvent accessible surface area difference between Free energy was estimated according to a MM/PBSA ˚the complex and the free molecules larger than 10 A2. The approach,18 in which the free energy is calculated accord-protonation states of these residues were determined in ing to the following equation:the complex and in the two separated proteins. The proto- Gmolecule ¼ EMM þ GPBSA À TSMM ð1Þnation states of all other titratable groups were put totheir standard values. A dielectric constant of 80 was used where EMM is the total molecular mechanical energy;for the solvent. Two values of the internal dielectric con- GPBSA, the solvation free energy; and TSMM, the entropy.stant for the protein were tested (4 and 20) and both gave The molecular mechanical energy contains an electro-similar results in terms of whether particular groups were static term, a van der Waals term, and an internal energyprotonated or not at a pH of 7.4. term. The solvation free energy is divided into two contri- butions: the electrostatic contribution and the nonpolarMolecular Dynamic Simulations (or hydrophobic) contribution. Molecular dynamic simulations were done using theCHARMM44 program and the all atom parameter set of GPBSA ¼ GPB þ Gnp ð2ÞCHARMM22.45 The minimized protein–protein complexes ˚were solvated in a 93 3 63 3 63 A box of TIP3P water. Mo- For the purpose of the present study, we calculated only alecular dynamic simulations were performed at constant subset of terms entering the MM/PBSA Eq. (1), namelytemperature using the Nose–Hoover method48,49 with the total electrostatic contribution, the van der Waals con-periodic boundary conditions. Bond lengths involving tribution, and the nonpolar, or hydrophobic, contributionPROTEINS: Structure, Function, and Bioinformatics DOI 10.1002/prot
  4. 4. INTERACTION HOT SPOTS IN ANTIGEN–ANTIBODY COMPLEX 421to the free energy. As we neglect conformational change tions. For example, there is always a destabilizing contri-upon complexation (see later), there are no changes to the bution due to the loss of favorable interactions betweeninternal energy terms. The binding free energy was then charged and polar residues and the solvent moleculesestimated according to the following equation: (desolvation). In addition, there is a change in the num- ber, and the environment, of pair-wise interactions result- DG ¼ DEelec þ DEvdw þ DGelec þ DGnp solv solv ð3Þ ing from complex formation, which counterbalances the unfavorable desolvation term.where DEelec and DEvdw are the electrostatic and van der We used the UHBD program47 to solve the linearizedWaals contributions to the energy of the complex forma- Poisson–Boltzmann equation using a finite-differencetion, respectively; DGelec and DGnp are the electrostatic solv solv method (FDPB). A shell script, which includes interfacesand nonpolar contributions related to solvation, respec- to UHBD, was used to compute the electrostatic bindingtively. free energy of binding of the two molecules. The boundary Equation 3 neglects conformational entropy contribu- between the solute and solvent regions with differenttions to the thermodynamics of binding. We did not esti- dielectric constants is determined by the molecular sur-mate this term because our main goal was to apply this face. van der Waals radii and charges of atoms aremethod to identify residues that play a dominant role in obtained from parameters of the force field CHARMM22.stabilizing the bound complex rather than obtaining a A dielectric constant of 80 was used for the solvent and 1quantitative account of the binding thermodynamics. for the protein (the value of 1 for the protein dielectric con-Although methods have been proposed to analyze individ- stant is used when conformational averaging is per-ual contributions to conformational entropies,51,52 their formed; for the single calculation on the X-ray structure ofgeneralization in the present context would entail a signif- the cAbLys3-HEL, a value of 4 was used for the proteinicant and unnecessary computational cost. Another sim- dielectric constant10). The final mesh grid on whichplification introduced concerns the neglect of conforma- charges are located and on which the electrostatic poten- ˚ tial is calculated has a grid spacing of 0.35 A.tion changes between the bound and unbound proteins. Inthe absence of an experimental structure for the unbound The electrostatic contribution was decomposed by resi-antibody, the structures of the unbound proteins were dues. This decomposition is possible because the lineartaken from that of the complex. There are experimental Poisson–Boltzmann equation is used, which permits aindications that the antibody–antigen association is domi- superposition of the electrostatic potential at point i bynated by enthalpy and occurs without large structural the sum of potentials created by each individual charge.rearrangements.53,54 These simplifications preclude cal- X1culations of absolute values of the binding free energy; DGelec ¼ qi /complex ihowever, they are expected to be satisfactory in the con- i2C 2 8 9text of identifying interaction energy ‘‘hot spots’’ in the >X 1 X1 > > antibody lysozyme > > >protein–protein complex, which is the goal of the present À> : qi /i þ qi /i > ; ð4Þ 2 2study. Similar simplifications have been adopted by other i2A i2Lauthors.10,19,55,56 Each contribution to the binding freeenergy entering Eq. (3) was decomposed as a sum of indi- This equation shows that the total electrostatic energy ofvidual contributions per residue, as described later. a complex C (antibody þ lysozyme) is calculated by sum-van der Waals contributions ming over all charges, the product of the charge qi and the potential /i at the position of the charge. The potential /i The van der Waals energy was calculated by the 6-12 is obtained by solving the linear Poisson–Boltzmann (PB)Lennard-Jones potential of the CHARMM22 force field. equation using the program UHBD. The contribution ofThe individual contributions to the van der Waals interac- one residue j to the electrostatic free energy of complextion energy of each residue of the antibody was estimated formation is estimated by summing over all charges qias one half the interaction with the set of lysozyme resi- using the potentials /i created at the position i by the ˚dues located in a 12.5 A cutoff distance, and vice versa for atomic charges of residue j:lysozyme. The division by two ensures that the sum of 8 9individual contributions gives the total van der Waals X1 >X 1 j2C > j2A X1 j2L > > elec DGj ¼ > qi /i À > qi /i þ qi /i > > ð5Þinteraction energy of the complex. 2 : 2 2 ; i2C i2A i2LElectrostatic contributions The total electrostatic energy of a macromolecular sys- where j belongs either to A (antibody) or to L (lysozyme).tem can be partitioned into contributions from interac- Such decompositions were first used to interpret contin-tions between pairs of atomic partial charges of the macro- uum electrostatic calculations for solvation of phosphor-molecule and from the interactions of individual partial anes in water,57 and a decomposition scheme similar tocharges with the solvent. equation (4) was first applied to proteins by Hendsch and In a process such as protein association, there is always Tidor.10 However, in earlier free energy decompositioninterplay between favorable and unfavorable contribu- work based on the Poisson–Boltzmann method, only a sin- PROTEINS: Structure, Function, and Bioinformatics DOI 10.1002/prot
  5. 5. 422 LAFONT ET conformation of the protein–protein complex was usu- trostatic Coulomb term. We therefore computed the Cou-ally used,10,19,58 which is not the case here. lomb interaction energy in vacuo between the protein cAb-Lys3 and lysozyme for all conformations saved fromNonpolar contributions the MD trajectory (coordinates were collected every 0.1 ps from the MD). The Coulomb electrostatic interaction The nonpolar contributions to the free energy of binding energy in vacuo was calculated using the CHARMM pro-are taken to be proportional to the loss in solvent accessi- gram. A dielectric constant of 1 and a nonbonded cutoff ofble surface area. ˚ 12.5 A was used, with a shift truncation function for elec- trostatics. The conformations were clustered in ten groups DGnp ¼ g SAScomplex À SASantibody þ SASlysozyme ð6Þ unbound unbound that had similar values of the electrostatic interaction energy, and that were distributed evenly between the low- est and highest values recorded for the electrostatic inter-The radii of protein atoms are obtained from the parame- actions over the course of the trajectory. One conformationters of the force field CHARMM22. We used CHARMM for was extracted from each cluster (with vacuum interactioncalculating the accessible surface with a probe radius of energy closest to the cluster average value). Each ˚1.4 A. We determined the accessible surface of each extracted conformation was then postprocessed using theresidue in the free molecules and in the complex and then MM/PBSA procedure. The total free energy and its indi-we calculated the accessible surface area difference vidual components were then obtained for each trajectorybetween the free molecule and the complex for each as a weighted average over the conformations extracted, ˚residue. A value of 5 cal molÀ1 AÀ2 was used13,20,21 for the with higher weight given to conformations extracted fromconstant g. more populated clusters (i.e. the weight of a conformation is proportional to the ratio of the number of conformationsSelection of Structures for the Free with that energy to the total number of conformationsEnergy Decomposition extracted from the MD trajectory). This procedure reduces Past free energy decomposition analysis based on the the computational cost of performing MM/PBSA calcula-MM/PBSA or MM/GBSA methods have been performed tions while allowing an estimate of the effect of conforma-using either a single experimental structure9,10 or an en- tional variation on the decomposition analysis. While thesemble of conformations extracted from molecular energy used for selecting the conformations was the vac-dynamic simulations.13,56 It is known that the different uum electrostatic term, which differs from the solventenergy terms used in the MM/PBSA method fluctuate sig- term computed by continuum electrostatics, the two termsnificantly with small changes in structure linked to ther- show a significant statistical correlation, so that confor-mal fluctuations.59 It is therefore necessary to use an en- mations that show large variation in vacuum energy alsosemble of structures rather than one experimental struc- show significant variation in solution energy (data notture in free energy decompositions based on the MM/ shown,60), hence the interest of the simplified protocol.PBSA scheme. Obtaining converged values of binding free A standard deviation of each amino acid contribution isenergies for quantitative assessment of binding thermody- also calculated. The statistical weight of the conformationnamics entails significant computational effort.13 The is taken into account in computing the standard devia-intention of the present work was to reliably identify tion.61 When results from different simulations using dif-those amino acids that play a dominant role in binding, ferent starting structures are averaged, the total numberand to obtain a semi-quantitative estimate of their contri- of conformations in all trajectories is used to normalizebutions to the binding thermodynamics. We, therefore, the weight of each individual conformation in the average.developed a protocol that uses only a limited number ofconformations extracted from molecular dynamics trajec- Amino Acid Efficiencytories for the free energy decomposition, but that allowssufficient sampling to reduce artifacts linked to the use of We define the amino acid efficiency as the free energya single conformation. For the system studied, several contribution of each amino acid divided by its number ofcrystal structures were available (cf. Structures Section). nonhydrogen atoms (i.e. 4 for Gly, 5 for Ala, 14 for TrpThese structures were used as starting points for short etc.). This definition is similar in spirit to the definition ofmolecular dynamics trajectories. Thus, significant confor- ligand efficiency62 which is defined as the binding affinitymational differences would be taken into account by using per heavy atom. When extending the notion to aminothe different initial structures and the MD simulations acids within a macromolecule, we also considered thatwould give small thermal fluctuations around the crystal some amino acids may be partially or even fully buried instructure. Ten conformations were extracted from the MD unbound proteins. We therefore tested an alternative defi-trajectories for analysis. The structures were not nition of the number of atoms as the solvent accessibleextracted at regular time intervals, but selected using a atoms rather than all heavy atoms. The effective numberprotocol that was designed to cover a large, representative of solvent accessible atoms was computed in the followingrange of individual MM/PBSA energy terms. The compo- way: for each of the 20 amino acid, AA, a tripeptide Gly-nent of the binding free energy that shows the largest AA-Gly was constructed in extended conformation andfluctuations with small changes in structures is the elec- the solvent accessible surface of the amino acid in the tri-PROTEINS: Structure, Function, and Bioinformatics DOI 10.1002/prot
  6. 6. INTERACTION HOT SPOTS IN ANTIGEN–ANTIBODY COMPLEX 423peptide was computed. The radii of amino acids atoms cules were injected at different concentrations on a controlwere obtained from the CHARMM force field and surface and on a surface containing low levels of immobi-CHARMM program was used for calculating the accessi- lized lysozyme. The injection and postinjection phases ˚ble surface with a probe radius of 1.4 A. The solvent acces- lasted between 3 and 5 min. The surfaces were regener-sible surface of each amino acid of the unbound protein ated with 50 mM HCl. Data were fitted globally with thewas computed using the same procedure. Then an effec- BIAevaluation 3.1 software (Biacore AB), using the Lang-tive number of atoms (Neffect) for each amino acid within muir binding model.the protein was computed as RESULTS Sprot Neffect ¼ Nheavy Á þ1 ð7Þ Free Energy Decomposition for the Antibody Speptide cAb-Lys3In this equation, Nheavy is the number of heavy atoms in Here, we report the results concerning the contributionthe amino acid; Sprot, the solvent accessible surface of the of the camelid single chain variable domain (cAb-Lys3) toamino acid in the unbound protein; Speptide, the solvent ac- total binding free energy of the complex with hen eggcessible surface of the amino acid in the tripeptide. If the white lysozyme (HEL) and turkey egg white lysozymeamino acid is completely buried in the unbound protein (TEL). The cAb-Lys3 single variable domain fragment(Sprot ¼ 0), the effective number of atoms is set to one. binds to HEL and TEL with nanomolar affinity.65 On av- The amino acid efficiency was calculated in two ways: erage, the solvent accessible surface of the antibody that ˚ ˚ is buried on binding is 998 A2 for the HEL and 945 A2 forfirst by dividing the free energy contribution of eachamino acid by the effective number of atoms (Neffect), and the TEL complex. To identify those amino acids that makesecond, by dividing the free energy contribution by the significant interactions in the bound complex, the freenumber of heavy atom (Nheavy). The results were com- energy decomposition scheme was applied to the cAb-pared and they were not found to be significantly different Lys3/HEL and cAb-Lys3/TEL complexes. In all, 6 simula-between the two procedures. The data using Neffect are tions were run of the different complexes and the MM/presented in the Supplementary Material and the amino PBSA analysis was applied to all trajectories following theacid efficiency calculated by dividing the free energy by procedure outlined in Methods. The results presentedthe number of heavy atoms was used in the analysis of the here in detail are average results over the different anti-results presented. gens, HEL and TEL, respectively. The decomposition results, presented in Figs. 1–3, are for the cAb-Lys3/HELMutations and Binding Affinity Measurements complex. A comparison to the results obtained for the cAb-Production of VHH mutants Lys3/TEL complex is given in Supplementary Material. In cAb-Lys3, the residues that contact lysozyme belong The cAb-Lys3-pHEN6 vector29 containing the gene for to the Complementary Determining Regions (CDRs) (asthe expression of the recombinant cAb-Lys3 with a C-ter- defined in Ref. 67; see also Ref. 22). The results show thatminal His6 was a kind gift from Drs. M. Lauwereys and S. amino acids that constitute the CDRs make the most im-Muyldermans (Laboratorium voor Ultrastructuur, Brus- portant contributions to the binding interactions, whilesels). The mutations T28A, T28D, T28K, D73S, D73R, the framework region of the antibody (defined as the anti-D73K, Y118F, and I29T were introduced by standard site body without the CDRs) makes mostly negligible contribu-directed mutagenesis techniques. All mutant genes were tions (see Fig. 1 and Table II). The residues of the CDRsconfirmed by nucleotide sequencing. The proteins were do not, however, contribute uniformly to the binding inter-expressed and purified as described by Lauwereys et al.29 actions (see Fig. 1). The third Complementary Determin- ing Region (CDR 3), which is particularly long in the cAb-Affinity measurements Lys3 and inserts in the lysozyme active site,22 makes the The interaction of the wild type or mutant cAb-Lys3 dominant contribution to binding (see Fig. 1 and Table II);with either TEL or HEL was analyzed using a Biacore1 in particular, a contiguous stretch of seven residues (Thr2000 instrument (Biacore AB, Uppsala, Sweden) as 101 - Ile 102 - Tyr 103 - Ala 104 - Ser 105 - Tyr 106 - Tyrdescribed previously.63 Briefly, all experiments were car- 107) is shown to contribute 75% of the total binding freeried out at 258C in HBS-EP buffer (10 mM HEPES pH 7.4, energy, while it contributes only 57% of the buried cAb-150 mM NaCl, 3.4 mM EDTA with 0.005% Surfactant Lys3 surface.P20). TEL or HEL were immobilized on a CM5 sensor sur- An analysis of the binding in terms of electrostatic, vanface using the amine coupling chemistry as described.63 der Waals and hydrophobic (based on surface area) contri-The active concentrations of the antibody fragments were butions (see Fig. 2 and Table II) shows that for cAb-Lys3,measured by injecting the samples on a surface containing binding is dominated by the van der Waals interactions,high levels of lysozyme in conditions of total mass trans- while the hydrophobic (surface area) contribution isport.64 The reaction rate was recorded 15 s after injection, smaller. The relative contributions of the three CDRs andand used to calculate the active cAb-Lys3 concentration of the framework region to the hydrophobic term arebased on a previously established calibration curve. For roughly correlated with their respective van der Waalskinetic measurements, the WT and mutant cAb-Lys3 mol- contributions. PROTEINS: Structure, Function, and Bioinformatics DOI 10.1002/prot
  7. 7. 424 LAFONT ET AL. Fig. 2. Decomposition of the total energetic contribution of antibody residues presented in Figure 1a. The electrostatic (ELEC in blue), the nonpolar (SAS in red) and the van der Waals (VDW in yellow) contribu- tions, as defined in Methods, are displayed separately for selected resi- dues (see Figure 1a for selected residue). Units: kcal/mol. ticularly in the cAb-Lys3/HEL complex. Although more extensive conformational sampling would be required to get a quantitative assessment of the electrostatic contri- bution for these residues, it is striking (cf. Fig. 2) that this sequence of residues emerges as the only region where favorable electrostatic contributions are found. This favor- able region of cAb-Lys3 is shown in Fig. 1b. Free Energy Decomposition for Lysozyme The analysis of the contributions of lysozyme to the binding energetics reveals important qualitative differen- ces with respect to the cAb-Lys3 (cf. Figs. 3 and 4 and Table II). Lysozyme buries on average a solvent accessible ˚ ˚ surface of 900 A2 for HEL and 822 A2 for TEL upon com- Fig. 1. (a) Net binding energy contribution by residue of the antibody plex formation.for the cAb-Lys3/hen egg white lysozyme complex. The values reported The amino acids that make a dominant contribution toare averages of the total contribution DG, as estimated from Eq. (3). The binding are not contiguous, but form two discontinuousresults are presented for residues belonging to complementary determin-ing region (CDR1 from Pro 31 to Gly 35, CDR2 from Ala 50 to Gly 66, and patches around amino acids 58-63 and 106-109 (seeCDR3 from Asp 99 to Asp 121) and for residues of the framework region Fig. 3b). The residues that make the most favorable con-with a contribution different from zero. Averages are calculated over all tributions to binding contribute only 42% of the buried ly-structures extracted from the trajectories (MEL AL HSD, MEL AL HSE,MEL BM, and JTT). Units: kcal/mol. (b) Antibody surface colored accord- sozyme (HEL) surface (50% of TEL). As for cAb-Lys3, theing to the total energetic contribution of each residue. Coloring from green favorable interactions of these residues are dominated by(favorable) to red (unfavorable) with contributions in the range from the van der Waals contribution (cf. Table II and Fig. 4).À7.74 to þ7.13 kcal/mol. Figure generated using the program GRASP.66The residues with the most favorable or unfavorable contributions are These residues line the binding pocket into which thelabeled. CDR3 of cAb-Lys3 inserts (cf. Fig. 3b and discussion given below). Lysozyme (HEL) contains several charged residues at Overall, the electrostatic contribution of the cAb-Lys3 to or near the binding interface that become buried uponbinding is about zero, which results from the compensa- binding. An analysis of the binding in terms of electro-tion between (þ4 kcal/mol) unfavorable electrostatic con- static, van der Waals, and hydrophobic contributions (seetributions of CDR’s 1 and 2 and favorable contributions of Fig. 4 and Table II) shows that several charged residuesa few amino acids in CDR3. The standard deviation on the (D52, R73, and R112) have large unfavorable electrostaticelectrostatic contribution is comparatively large (see terms. Although some of them make important interac-Table II), which is expected, given the magnitude of the tions with cAb-Lys3 amino acids (see later), the interac-Coulomb and the solvation free energy terms that are tion term is not important enough to compensate for theirinvolved. In the small seven residue sequence of CDR 3 desolvation, and as a result, the total electrostatic contri-that dominates the binding energy, some of the residues bution of these amino acids is positive (see Figure 4).have favorable electrostatic contributions to binding, par- While small local conformational differences between thePROTEINS: Structure, Function, and Bioinformatics DOI 10.1002/prot
  8. 8. INTERACTION HOT SPOTS IN ANTIGEN–ANTIBODY COMPLEX 425 relative contribution of lysozyme to the total binding free energy is only about 15% (cf. Table II, À7 kcal/mol from a total of À52 kcal/mol for HEL, and À6 kcal/mol out of a total of À38 kcal/mol for TEL). The observation that the difference in total contribution to binding between the two partners is directly linked to the difference in electrostatic interaction also results from the fact that, by definition (see Methods), the van der Waals interactions are shared equally between the two proteins, and that the difference in buried surface area between two proteins in a complex is very small.68 Electrostatic desolvation, on the other hand, is strongly dependent on the individual structure and charge distribution of each partner, and large differ- ences can be observed for two proteins in a complex as demonstrated here. Although the absolute values of the antibody and lysozyme contributions to binding differ between the HEL and TEL complexes (see Table II), the relative magnitudes of the individual contributions are very similar. Comparison of the Different Structures We studied three experimental structures of the cAb- Lys3 in complex with HEL, and two of the complex with TEL. The three HEL complex structures differ signifi- cantly by a rigid body rotation of one protein with respect to the other42 and both the sequence and the internal structure of the lysozyme monomer are different in the TEL and in the HEL complex structures. The RMS differ- ences between the different protein–protein complexes (in both the crystal structures and structures obtained from the MD trajectory) are given in Table 1 of Supplementary Material. With these differences, it is possible to test the robust- ness of the free energy decomposition scheme with respect Fig. 3. (a) Net binding energy contribution by residue of the HEL lyso- to changes in sequence and in conformation space. Thezyme for the cAb-Lys3/hen egg white lysozyme complex. The valuesreported are averages of the total contribution DG [cf. Eq. (3)]. The results short MD simulations that were run for each structureare presented for residues with a contribution different from zero. The were used to generate ensembles of conformations andaverages were calculated over all structures extracted from the trajecto- thus to reduce statistical error. We are then able to obtainries of the different initial structures (MEL AL HSD, MEL AL HSE, MEL a more general understanding of the recognition processBM, and JTT). Units: kcal/mol. (b) Projection of the total energetic contri-bution of each residue onto the lysozyme surface. The residues are col- in that our results are less dependent on the specificored from green (favorable contributions) to red (unfavorable contribu- three-dimensional structure.tions). The values range from À7.74 to þ7.13 kcal/mol. The picture was A comparison of the results for the two lysozymes (seeprepared using the program GRASP66. The residues with the most favor-able or unfavorable contributions are noted. [Color figure can be viewed Fig. 5) shows that the interaction energy hot spots are thein the online issue, which is available at] same in the two complexes; similar results are obtained for the antibody (see Fig. 1 of Supplementary Material). However, some charged amino acids of the interface, such as Asp 48 and Arg 73 (HEL) or Lys 73 (TEL), show signifi-complexed and uncomplexed lysozyme may modulate the cantly different binding energy contributions for the fivemagnitude of this term, the desolvation effect explains, in structures of the complex (three structures for HEL andlarge part, why the electrostatic contribution is large and two structures for TEL). These differences can bepositive for lysozyme (about þ36 Æ 6 kcal/mol, average of explained either structurally or based on the lysozymeHEL and TEL; see Table II). sequences. For example, Asp 48 of lysozyme can form This is an important difference with the cAb-Lys3, favorable interactions with Tyr 106 of cAb-Lys3, but it iswhere such large positive contributions were not situated in a loop and shows larger structural fluctuationsobserved. As a result, although unfavorable electrostatic than other amino acids. Hence its contribution variesinteractions are compensated by the favorable nonpolar considerably between the different structures. Likewise,ones, the contribution of lysozyme to the binding ener- Arg 73 (HEL) or Lys 73 (TEL) are in a region of the pro-getics is small when compared with that of cAb-Lys3. The tein–protein interface that shows significant structural PROTEINS: Structure, Function, and Bioinformatics DOI 10.1002/prot
  9. 9. 426 LAFONT ET AL. TABLE II. Individual Contributions to the Binding Energy for the cAb-Lys3/HEL Complex (top part) and the cAb-Lys3/TEL Complex (lower part) ELEC SAS VDW TOTALCHICKEN CDR1 {P31-G35} 1.3 Æ 1.2 À0.1 Æ 0.04 À2.9 Æ 0.9 À1.8 Æ 2.2 CDR2 {A50-G66} 2.6 Æ 2.6 À0.7 Æ 0.1 À6.4 Æ 1.2 À4.6 Æ 3.9 CDR3 {D99-D121} À3.8 Æ 5.5 À3.4 Æ 0.2 À26.9 Æ 2.0 À34.0 Æ 7.7 HOT SPOT {T101-Y107} À6.4 Æ 4.6 À2.8 Æ 0.2 À23.6 Æ 1.9 À32.8 Æ 6.7 FRAME {all without CDRs} 0.3 Æ 5.3 À0.7 Æ 0.3 À4.0 Æ 1.9 À4.4 Æ 7.4 cAb-Lys3 {all antibody} 0.4 Æ 4.5 À5.0 Æ 0.2 À40.2 Æ 3.6 À44.8 Æ 8.3 Lysozyme: N46, I58, N59, W62, À1.8 Æ 2.1 À1.9 Æ 0.1 À20.6 Æ 1.1 À24.3 Æ 3.3 W63, N103, A107, W108, V109 Lysozyme: D48, D52, R61, R73, R112 29.5 Æ 6.3 À1.5 Æ 0.1 À9.6 Æ 1.7 18.4 Æ 8.1 HEL {all lysozyme} 37.4 Æ 7.0 À4.5 Æ 0.3 À40.2 Æ 3.6 À7.2 Æ 10.9Total 37.8 Æ 8.6 À9.5 Æ 0.5 À80.4 Æ 7.2 À52.0 Æ 16.2TURKEY CDR1 {P31-G35} 2.6 Æ 0.8 À0.2 Æ 0.03 À2.1 Æ 0.5 0.3 Æ 1.3 CDR2 {A50-G66} 3.7 Æ 1.2 À0.7 Æ 0.04 À5.0 Æ 0.4 À2.1 Æ 1.7 CDR3 {D99-D121} 1.8 Æ 3.9 À3.4 Æ 0.2 À26.9 Æ 1.7 À28.5 Æ 5.8 HOT SPOT {T101-Y107} 1.1 Æ 2.9 À2.8 Æ 0.1 À23.8 Æ 1.7 À25.5 Æ 4.7 FRAME {all without CDR’s} 0.4 Æ 0.6 À0.4 Æ 0.1 À2.0 Æ 0.3 À2.0 Æ 1.0 cAb-Lys3 {all antibody} 8.5 Æ 4.7 À4.7 Æ 0.2 À36.0 Æ 2.0 À32.3 Æ 6.8 Lysozyme: N46, I58, N59, W62, W63, 0.02 Æ 1.9 À2.1 Æ 0.1 À20.0 Æ 1.2 À22.08 Æ 3.2 N103, A107, W108, V109 Lysozyme: D48, D52, R61, R73, R112 30.1 Æ 3.3 À1.3 Æ 0.1 À8.8 Æ 0.7 20.0 Æ 4.1 TEL {all lysozyme} 34.3 Æ 3.3 À4.2 Æ 0.2 À36.0 Æ 2.0 À5.8 Æ 5.5Total 42.8 Æ 7.4 À8.9 Æ 0.3 À72.1 Æ 3.9 À38.1 Æ 11.6All contributions are averages over the structures extracted from several molecular dynamics trajectories, as described in the Methods section.The contributions for CDRs (Complementary Determining Regions) 1 to 3, for the hot spot residues (T101 to Y107 for cAb-Lys3), for the frame-work (residues not in the CDR regions of cAb-Lys3, FRAME), for the entire antibody (cAb-Lys3), for lysozyme (HEL or TEL), for the hydrophobicpatches in lysozyme, and for charged residues in lysozyme are obtained by summing over the specified residues. All energies are in kcal/mol. Fig. 5. Comparison of results obtained for the two complexes cAb- Fig. 4. Decomposition of the total energetic contribution of lysozyme Lys3/hen egg white lysozyme (red) and the cAb-Lys3/turkey egg whiteresidues presented in Figure 3a. The electrostatic (ELEC in blue), the lysozyme (green). The contributions plotted are averages of total contri-nonpolar (SAS in red) and the van der Waals (VDW in yellow) contribu- butions of lysozyme residues. The two lysozymes differ by only seventions are displayed separately for selected residues (see Fig. 3a for residues. The sequence used in this plot is the one for the chicken egg-selected residue). Units: kcal/mol. [Color figure can be viewed in the white lysozyme. Amongst the residues shown in the graph, the followingonline issue, which is available at] sequence differences in the turkey egg-white lysozyme are found: Q41H, R73K, V99A, and D101G. The averages were calculated over all trajecto- ries generated for each complex. The standard deviation for each residue is shown. [Color figure can be viewed in the online issue, which is avail- able at]variations. These results underline the need to exercisecaution when interpreting the results from a free energydecomposition for a single crystallographic structure. The For example, in the cAb-Lys3/HEL complex, the confor-conformational sampling applied in this study yields a mational sampling clearly helps discriminating the truemean value with a standard deviation for each contribu- hot spot residues from amino acids that have favorabletion; the standard deviation is a useful first indication of but highly fluctuating contributions (see Fig. 1 of Supple-the dependence of the results on small structural changes. mentary Material). These results suggest that, wheneverPROTEINS: Structure, Function, and Bioinformatics DOI 10.1002/prot
  10. 10. INTERACTION HOT SPOTS IN ANTIGEN–ANTIBODY COMPLEX 427 tion kinetics. From the computational analysis, contribu- tions from Thr 28, Asp 73, and Tyr 118 showed large fluc- tuations with conformational averaging and yielded small binding averages (see Fig. 1a and Fig. S1 in Supplemen- tary material). However, contributions from individual structures can be important, as is shown by the large error bars. DISCUSSION Structural Analysis of the Binding Energy Hot Spots A useful aspect of the free energy decomposition scheme presented here is that it not only identifies the binding energy hot spots, but also gives insight into the nature of the key interactions. The amino acids from the cAb-Lys3 (see Figs. 1 and 2 and Table II) that make the most favorable contributions to binding (between À2 and À8 kcal/mol) form a contigu- ous stretch of residues in the CDR3: Thr 101 - Ile 102 - Tyr 103 - Ala 104 - Ser 105 - Tyr 106 - Tyr 107 (one additional important amino acid is Ile 29 from CDR1), while for lyso- zyme, the most favorable interactions are made by Asn 46- Ile 58 - Asn 59 - Trp 62 - Trp 63 - Asn 103 - Ala 107 - Fig. 6. Comparison of kinetic on-rate (a) and off-rate (b) parameters Trp 108 - Val 109 (see Figs. 3 and 4 and Table II), whichfor the interaction between the WT or mutant cAb-Lys3 with HEL (dark are grouped together spatially.grey) or TEL (light grey). Rate parameters are expressed as percent of the For a more detailed analysis, we determined whetherwild type value. Measurements were repeated 2–3 times for the WT cAb-Lys3 and mutants T28K, I29T, and D73S, for which error bars are shown. the favorable interactions of the hot spot residues are pre- dominantly van der Waals or electrostatic in nature and whether they face each other on the two partners of theavailable, averaging over several experimental structures complex. Concerning cAb-Lys3, the hot spot interactionsis likely to increase the significance of the results obtained are mainly nonpolar. The important amino acids fromfrom the decomposition scheme. CDR3 make favorable contributions to the interaction energy via the solvent accessible surface, or hydrophobic term and via the van der Waals interactions. Some ofMutational Analysis of Four cAb-Lys3 Positions these important van der Waals interactions are made To further investigate the importance of using confor- with amino acids on lysozyme that are also, correspond-mational averaging when calculating a free energy decom- ingly calculated as making important contributions. Forposition, 4 amino acids of the antibody, identified as mak- example, the side chain of Ala 104 of cAb-Lys3 fills aing nonzero contributions to binding free energy by a free hydrophobic pocket lined by the side chains of Ile 58, Alaenergy decomposition calculation on a single crystal struc- 107, and Trp 108 of lysozyme. Particularly favorable vanture, were mutated as specified in Methods. These were der Waals interactions are also made by Trp 62 of lyso-Thr 28, Asp 73, and Tyr 118, which, in the cAbLys3-HEL zyme. This amino acid not only makes an important con-crystal structure (BM complex), have total contributions tact with a hot spot residue from cAb-Lys3 (Ile 102 fromto the binding free energy of À1.1, À0.5 and À2.3 kcal/mol CDR3), but it also makes contacts with several aminorespectively. Thr 28 and Asp 73 were replaced by residues acids from CDR1 (Ile 29 and Tyr 32) and CDR2 (Gly 54),that lead to important modifications of size and/or charge which have smaller van der Waals contributions to the(T28A, T28D, T28K, D73S, D73R, and D73K), while a binding energy. The large contribution of Trp 62 (lyso-more conservative change was made at position 118 zyme) arises from its central position at the protein–pro-(Y118F), to avoid large structural effects of the mutation. tein interface that allows a large number of van der WaalsResidue Ile 29, which was identified as important for bind-, was mutated to Thr as a positive control. The kinetics Thus the size of the amino acid and the shape of the con-of the interaction of the wild type and eight cAbLys3 tact surface plays an important role and in some cases amutants with either HEL or TEL were measured by Bia- van der Waals hot spot is located on one of the proteins,core. Kinetic parameters, expressed as percentage of those whereas there is no dominating van der Waals contribu-measured for the interaction of HEL or TEL with the tion arising from the reciprocal residues on the other pro-unmodified cAb-Lys3, are compared in Figure 6. Except tein. Thus, even though van der Waals interactions arefor the I29T change, which produced a 3- to 4-fold increase shared equally between the two proteins, at the residuein koff, other changes had no measurable effect on interac- level, the hot spots are not necessarily symmetrical. PROTEINS: Structure, Function, and Bioinformatics DOI 10.1002/prot
  11. 11. 428 LAFONT ET AL. In contrast to the unfavorable electrostatic contribu- tions calculated for Arg 112 and Asp 52 of lysozyme, a small favorable contribution is found for Trp 63 of lyso- zyme, which interacts with Tyr 103 main chain from cAb- Lys3. Thus, even though the electrostatic interaction from the main chain of Tyr 103 of cAb-Lys3 is smaller than the interaction made by its side chain (cf. Fig. 7), the main chain H-bond yields a favorable binding energy contribu- tion both on cAb-Lys3 and on lysozyme. Two additional amino acids of lysozyme make favorable (although smaller than those found in cAb-Lys3) electrostatic contributions: Asn 59 (HEL: À1.7 kcal/mol) and Ala 107 (HEL: À0.6 kcal/ mol), see Figure 4. An analysis of the structures shows Fig. 7. Decomposition of the electrostatic contributions into side chain that Asn 59 and Ala 107 are engaged in hydrogen bondsand main chain components for the antibody in the cAb-Lys3/hen eggwhite lysozyme complex. The values represented are averages of the with Ala 104 of cAb-Lys3 (cf. Figure 9), and this buriedelectrostatic contribution of backbone (yellow) and side chain (blue) of network of H-bonds is particularly favorable.antibody residues. (see Fig. 1a for selected residues). The average wascalculated over all the trajectories done with the different initial structures(MEL AL HSD, MEL AL HSE, MEL BM, and JTT). Units: kcal/mol. Comparison of Experimental and[Color figure can be viewed in the online issue, which is available at www. Computational] The conclusions of the calculations were challenged with both existing and new mutational studies of cAb- Even though van der Waals interactions have an impor- Lys3. While the existing studies mainly focus on residuestant contribution to hot spots, some important electro- identified here as ‘‘hot spot,’’ we chose to focus on fourstatic interactions can also be identified. The electrostatic positions with a calculated small contribution to bindingcontribution is favorable for cAb-Lys3 residues Tyr 103 - energy.Ala 104 - Ser 105 - Tyr 106, which are involved in a hydro- Existing experimental studies include replacements ofgen bond network with several residues from lysozyme. residues Thr 101 (by L,S,P) and Ser 105 (by N,H,Q,A,The most favorable electrostatic interactions are made by G,P,M) from CDR3 of cAb-Lys3 for a quantitative struc-the two tyrosine residues; they can be further decomposed ture-activity study,53 and replacements of residues frominto side chain and main chain contributions. As can be CDR1 and CDR2 to revert to the germ line sequence.54seen from Figure 7, the dominant contributions are made The replacements at positions 101 and 105 had variedby the side chain for Tyr 103 (the main chain also makes a effects, the largest losses in binding affinity being 120-foldfavorable, but less important, contribution) and by the for the T100P change, and 100-fold for the S105Qmain chain for Tyr 106. Examination of the structures change.53 In particular, the mutation S105A resulted in aindicates that the Tyr 103 side chain hydroxyl group 10-fold loss in experimental binding affinity (i.e. a changemakes an H-bond with Arg 112 from lysozyme, while the of 1.4 kcal/mol in binding affinity), which is in reasonableTyr 106 main chain NH makes an H-bond with the car- agreement with the computed binding free energy contri-boxyl group of Asp 52 from lysozyme. The favorable contri- bution of À4.8 Æ 2.3 kcal/mol for the serine as a whole (i.e.bution of these H-bonds is directly linked to the strong including main chain contributions that are not lost uponelectrostatic potential created by the charged amino acids mutation to Ala). Multiple mutations were introduced inon lysozyme. Remarkably, the free energy decomposition CDR1 and CDR2 so that the influence of individualfor lysozyme (see Figs. 3 and 4) yields unfavorable electro- changes is not known.54 Thermodynamic data howeverstatic contributions to the binding free energy for Arg 112 show a role of residues from CDR1 and CDR2 forand Asp 52. This is due to the large desolvation penalty enthalpic stabilization of the bound complex during matu-upon binding for the charged side chains of these residues. ration,54 which is in agreement with the results of theThe fact that electrostatic interactions are not always present calculations that show that CDR1 and CDR2favorable to binding has been noted in other free energy make favorable van der Waals interactions with lysozymedecomposition studies.13,19,21,58 In this particular exam- and thus play a role in stabilizing the complex.ple, it can be seen that Arg 112 and Asp 52, although not To investigate whether positions with small, but non-favorable themselves, create a favorable environment for zero, computed contributions can nevertheless influenceTyr 103 and Tyr 106 on the cAb-Lys3, and thus have an binding, we targeted residues Thr 28, Asp 73, and Tyrimportant role in the formation of the complex. The other 118. The residue Ile 29, which is identified as important incharged amino acid on lysozyme that has a large unfavor- CDR1, was taken as a positive control. None of the muta-able desolvation contribution is Arg 73. Arg 73 is not tions at positions 28, 73, and 118 had a measurable effectinvolved in hydrogen bonds with amino acids on cAb- on binding kinetics (see Figure 6), while the mutation atLys3, and its binding energy contribution shows larger position I29 (I29T) increased koff 3- to 4-fold, thus result-fluctuations than those of Arg 112 and Asp 52 (see ing in a loss of binding affinity. These experimentalFig. 3a). results support the validity of our computational ap-PROTEINS: Structure, Function, and Bioinformatics DOI 10.1002/prot
  12. 12. INTERACTION HOT SPOTS IN ANTIGEN–ANTIBODY COMPLEX 429proach. In particular, the mutations data support our face area (hydrophobic solvation) terms as a sum oversimplified conformational averaging scheme, and show individual amino acid contributions (as described in Mate-that the three amino acids tested, which showed large rial and Methods). The decomposition is performed influctuations in the computed contribution, do not influ- such a way that the sum of all individual contributionsence binding, while their contribution was non negligible corresponds to the total free energy (cf. Eq. 1). Decompos-if only one experimental X-ray structure is used. ing into individual amino acid contributions implies that pairwise interactions (such as van der Waals and electro- static) are divided equally between the interacting aminoExperimental vs. Computational Hot Spots acids i.e. if for example the van der Waals interaction The comparison of the experimental data to the compu- energy between amino acids A and B amounts to À4 kcal/tational results raises some important issues. Amino acids mol, a contribution of À2 kcal/mol will be attributed to Athat are found to play an important role in binding are of- and À2 kcal/mol to B, and likewise for electrostatic inter-ten referred to in the computational literature as ‘‘binding actions. It must be noted that in the continuum descrip-energy hot spots,’’ irrespective of whether the quantity tion of the solvent that is used here, desolvation (whethercalculated is the change in binding free energy upon electrostatic or hydrophobic) is a self-energy term and ismutating the residue to alanine, or a related but different thus naturally decomposed in individual contribu-quantity. In this paper, we used the term hot spot for tions.57,71amino acids that make significant individual contribu- The earlier-mentioned differences between computa-tions (À1.5 kcal/mol) to the total binding free energy, as tional free energy decomposition and experimental ala-described in Material and Methods. It is therefore useful nine scanning data should be kept in mind when compar-to clarify the relationship between ‘‘hot spots’’ as dis- ing computational and experimental hot spot residues.cussed in the present computations and ‘‘hot spots’’obtained by alanine scanning experiments. Experimentally, binding energy hot spots are oftendefined as residues that, when mutated to alanine, giverise to a significant drop in the binding affinity (typically,changes in binding free energy larger than 1.5 kcal/mol).7As discussed in Ref. 69, mutations to alanine can affectthe binding affinity by both changing the structure and/orfree energy of the unbound partners as well as those ofthe bound complex. This combination of effects compli-cates the interpretation of alanine scanning experiments,and consequently their exploitation in, for example,designing small molecule inhibitors that mimic importantinteractions. An advantage of computational approachesis that the energetic contributions can be separated in amanner that is not attainable experimentally. For exam-ple, as done in the present study, it is possible to analyzethe energetic contribution of an amino acid withoutresorting to explicit mutation to alanine. This is an important aspect of the present calculations,which is to identify binding energy hot spots and, thus,free energy decomposition is used rather than explicitmutation to alanine. Individual amino acid contributionsto binding presented in Figs. 1, 2, 3, 4, 7 are therefore notthe same quantity as the change in free energy uponmutating the residue to alanine. If a direct comparison toAla-scanning experiments was desired, it would be possi-ble to calculate free energy changes upon mutating a resi-due to alanine by constructing the mutant and reevaluat-ing the free energy of binding.11,70 A full estimate of thefree energy change should also consider the structuralconsequences of mutating a specific amino acid to alanine, Fig. 8. Bar graph of the amino acid efficiency. Each free energy contri-both in the complex and in the unbound partners. bution (Electrostatic (ELE), nonpolar solvation (SAS), van der Waals Another difference with experimental alanine scanning energy (VDW), and total DG (TOT)) of amino acids is divided by the num-data is that single amino acid free energy contributions to ber of nonhydrogen atoms. The results are presented for the residuesbinding, as computed here, include contributions to the having the most efficient amino-acids in the interaction for the antibody (top graph) and for the lysozyme (bottom graph). See Methods for details.binding energy from main chain atoms. They are obtained [Color figure can be viewed in the online issue, which is available atby writing the total electrostatic, van der Waals and sur-] PROTEINS: Structure, Function, and Bioinformatics DOI 10.1002/prot
  13. 13. 430 LAFONT ET AL. Fig. 9. Structural comparison of the lysozyme–cAb-Lys 3 complex and lysozyme-N-acetyl glucosamine (NAG) ring complex. The similarity between the hydrogen bond network formed between A107 and N59 of ly- sozyme and A104 of cAb-Lys3 (upper picture) and the network formed between A107 and N59 of lysozyme and NAG is shown. This network involves efficient amino acids of the lysozyme-cAb-Lys3 complex, and shows the energetic mimicry of the substrate (see text for details). [Color figure can be viewed in the online issue, which is available at]The two approaches nevertheless give convergent infor- nine will be detrimental to the binding affinity (unless themation i.e. when the free energy decomposition identifies contribution is made by main chain atoms). Significanta residue as making a large favorable contribution to correlations between experimental free energy changesbinding, one can expect that mutating this residue to ala- upon mutating a residue to alanine and free energyPROTEINS: Structure, Function, and Bioinformatics DOI 10.1002/prot
  14. 14. INTERACTION HOT SPOTS IN ANTIGEN–ANTIBODY COMPLEX 431decomposition based on MM/GBSA have indeed been network of hydrogen bonds and hydrophobic contacts (seereported.13 In contrast, a computed unfavorable contribu- Figure 9) that make them particularly efficient. Thesetion to binding does not necessarily mean that mutating interactions closely resemble65 those made by lysozymethe residue to alanine will increase the binding affinity. substrates, in particular the N-acetyl glucosamine ringAs discussed earlier, charged amino acids such as Asp 52 (NAG) in subsite C of the binding pocket. The acetamidoand Arg 112 from lysozyme have a positive (unfavorable) substituent of NAG in site C exhibits low temperature fac-contribution, yet most likely they have a structural role in tor and a well defined electron density74 and the saccha-stabilizing important residues from cAb-Lys3. ride in site C has been shown to make energetic contribu- tions to binding that are more favorable than the saccha- rides bound to other sites within the lysozyme bindingEfficient Amino Acids and Ligand Binding pocket.75 Thus the cAb-Lys3 not only is a structural mimic A different, instructive way to identify amino acids that of the lysozyme substrate, but it also exploits bindinghave an important role in binding is to compute the energy hot spots in a similar way as the substrate. This‘‘amino acid efficiency,’’ a quantity that we define here in further shows that identifying binding hot spots or con-analogy to ligand efficiency. Ligand efficiency is defined as versely efficient amino acids can pinpoint regions of thethe ratio of affinity to molecular size, a common measure protein–protein interface that can be targeted by syn-of size being the number of nonhydrogen atoms. The thetic ligands.notion of ligand efficiency has recently attracted a consid-erable interest in drug design studies,62,72 particularly in CONCLUSIONSthe context of fragment based drug design. In fragmentbased drug design, the goal is to find small molecular In this work, we analyzed the binding interface betweenbuilding blocks that bind to a medicinal target. Subse- the single chain variable antibody domain (VHH) cAb-quently, several building blocks can be linked or itera- Lys3 and its antigen lysozyme, using a free energy decom-tively increased in size while keeping the molecular position based on Poisson–Boltzmann electrostatics, vanweight in the ‘‘druggable’’ range.73 Small ligands with der Waals interactions, and a hydrophobic surface areahigh binding efficiency are thus sought as a good starting term. We considered conformational flexibility of the com-point for lead identification and drug design. plex explicitly by using several experimental structures By analogy with this concept, we identified ‘‘efficient as starting points and performing short molecular dynam-amino acids’’ of the binding interface between cAb-Lys3 ics simulations on each structure. We selected ten confor-and lysozyme by dividing the free energy contribution of mations from each molecular dynamics simulation thateach amino acid by its number of nonhydrogen atoms (i.e. sample representative values of the Coulomb interaction4 for Gly, 5 for Ala, 14 for Trp etc.). An alternative defini- energies between the proteins and performed the freetion of amino acid efficiency, based on the solvent accessi- energy decomposition on each conformation; the resultsble surface of the amino acids in the unbound proteins represent averages over each individual calculation. Thiswas also tested. The results are not significantly different protocol was shown to be able to discriminate betweenfrom the simple definition mentioned earlier. (See Meth- consistently favorable interactions that give rise to bind-ods and Supplementary Material for details). ing hot spots on one side, and interactions that vary sig- Amino acid efficiency can provide an assessment of the nificantly with small changes in structures on the otherbinding potential of a particular hot spot interaction. In side. The reliability of the free energy decomposition withdrug design strategies aimed at protein–protein interfa- respect to using a single crystal structure is improved,ces, the identification of efficient amino acids could help while avoiding the large computational cost of a freeidentify key interactions and prioritize the region of the energy simulation method. The predictive ability of theprotein–protein interface that could be targeted by small procedure was subsequently tested by point mutations ofmolecules. To gain physical insight in the origin of the amino acids of the interface; and it was shown that aminoamino acid efficiency, we divided each individual compo- acids that have a contribution to binding when only onenent of the binding free energy, as well as the total per res- crystal structure is used, but that show large fluctuationsidue free energy of binding, by the corresponding number when using the proposed conformational sampling proto-of heavy atoms (see Fig. 8). The most efficient amino acids col, can be mutated without influencing binding affinity.are Ala 104 and Ser 105 on cAb-Lys3 and Asn 59, Trp 62 The analysis of the complex has shown that electro-and Ala 107 on lysozyme. In general, the efficiency of the static interactions do not systematically favor bindingamino acids on the antibody is better than on lysozyme, because of the compensation between protein desolvationand the van der Waals contribution is the dominant and protein–protein interactions when forming macro-energy term for efficient binding. A notable exception is molecular complexes. van der Waals interactions, as ex-Asn 59 from lysozyme, where electrostatic and van der pected, systematically favor complex formation. This is inWaals interactions contribute equally. Structurally, the agreement with studies using similar methods on differ-ligand efficiency of Trp 62 from lysozyme is linked to its ent complexes.10,13,19,21,58 To identify amino acids thatposition as a central residue anchoring amino acids from play a particularly important role in the protein–proteinCDR 1, 2, and 3 of cAb-Lys3. The three amino acids Ala interface, we defined and computed the amino acid effi-104 (cAb-Lys3), Asn 59, and Ala 107 (lysozyme) make a ciency, i.e., the free energy contribution divided by the PROTEINS: Structure, Function, and Bioinformatics DOI 10.1002/prot