Model of the �L�2 Integrin I-Domain/ICAM-1 DI InterfaceSuggests That Subtle Changes in Loop OrientationDetermine Ligand SpecificityGlen B. Legge,1* Garrett M. Morris,2 Michel F. Sanner,2 Yoshikazu Takada,3 Arthur J. Olson,2 andFlavio Grynszpan2

1Department of Biology and Biochemistry, University of Houston, Houston, Texas2Department of Molecular Biology, The Scripps Research Institute, La Jolla, California3Department of Cell Biology, The Scripps Research Institute, La Jolla, California

ABSTRACT The interaction of the �L�2 inte-grin with its cellular ligand the intercellular adhe-sion molecule-1 (ICAM-1) is critical for the tightbinding interaction between most leukocytes andthe vascular endothelium before transendothelialmigration to the sites of inflammation. In this articlewe have modeled the �L subunit I-domain in itsactive form, which was computationally dockedwith the D1 domain of the ICAM-1 to probe potentialprotein-protein interactions. The experimentally ob-served key interaction between the carboxylate ofGlu 34 in the ICAM-1 D1 domain and the metalion-dependent adhesion site (MIDAS) in the open �LI-domain was consistently reproduced by our calcu-lations. The calculations reveal the nature of the�L�2/ICAM-1 interaction and suggest an explana-tion for the increased ligand-binding affinity in the“open” versus the “closed” conformation of the �LI-domain. A mechanism for substrate selectivityamong �L, �M, and �2 I-domains is suggestedwhereby the orientation of the loops within theI-domain is critical in mediating the interaction ofthe Glu 34 carboxylate of ICAM-1 D1 with the MIDAS.Proteins 2002;48:151–160. © 2002 Wiley-Liss, Inc.

Key words: �L I-domain; ICAM-1; modeling; dock-ing; integrin, activation

INTRODUCTION

Leukocyte adhesion plays a pivotal role in modulatingthe immune response. Both enhanced and deficient leuko-cyte adhesion results in pathological states; hence, theactivation of leukocyte adhesion is strictly regulated.Small molecule therapeutics that alter such adhesionevents may lead to numerous potential clinical applica-tions, including treatments for cancer and graft rejection,as well as chronic and acute inflammation.1,2 The interac-tion of �L�2 (LFA-1, CD11a/CD18) and its cellular li-gands, namely, the intercellular adhesion molecules(ICAMs), is a key event in the adhesion of most activatedleukocytes to the endothelial wall before transendothelialmigration.1,2

�L�2 is a member of the integrin family and is an ��heterodimer with an �-subunit of 180 kDa and a �-subunitof 95 kDa. Within the integrin family, nine �-subunits (�1,

�2, �10, �11 �D, �E, �L, �M, and �X) (Refs. 3 and 4 andreferences therein) contain an additional domain (so-called “A” or “I”) of about 190 amino acids that adopts theRossmann fold.5 The I-domain is located toward the N-terminus of the �-subunit and through its interaction withthe D1 domain in ICAM-1 plays a central role in ligandbinding.6 The interaction between Glu 34 within theICAM-1 D1 domain and the divalent metal cation at themetal ion-dependent adhesion site (MIDAS) in the �LI-domain is critical for high affinity binding.7–10

There is a significant body of evidence from both anti-body-mapping studies and mutagenesis data indicatingthat �L�2 undergoes a conformational change on activa-tion and subsequent high affinity binding to ICAM-1.11,12

In addition, a structural change in the �L I-domain onbinding the ICAM-1 D1/D2 domains is evident from NMRtitration data.13 This study shows that changes in thechemical shift in the �L I-domain are not limited toresidues near the MIDAS, but also include residues lo-cated within the C-terminal �7 helix.13

The crystal structures of the related �-subunit I-domains from �M (Mac-1) and �2 revealed that theseproteins could adopt two conformations: (a) a ligand (orligand-mimetic) bound, “open” conformation and (b) anunbound, “closed” conformation.5 This change in the ter-tiary structure from an “open” to a “closed” conformation isthought to reflect the active and inactive forms, respec-tively, of the I-domain and may represent a commonmechanism of activation for all I-domain containing inte-grins.14 The �2 I-domain crystal structure adopts the openconformation when it is complexed with a triple helix

Abbreviations: Intercellular Adhesion Molecule-1 (ICAM-1): ProteinData Bank (PDB); metal ion dependent adhesion site (MIDAS).

The minimized docked complex has been submitted to the PDBdatabase (accession number 1IJ4).

Grant sponsor: National Institutes of Health; Grant numbersGM48870, RR08065, and GM49899; Grant sponsor: National ScienceFoundation; Grant number: CA ACI-9619020.

*Correspondence to: Glen B. Legge, Department of Biology andBiochemistry, University of Houston, 353 SR2, Houston, TX 77204-5001. E-mail: glegge@uh.edu

Received 21 November 2001; Accepted 11 February 2002

Published online 00 Month 2002 in Wiley InterScience(www.interscience.wiley.com). DOI: 10.1002/prot.10134

PROTEINS: Structure, Function, and Genetics 48:151–160 (2002)

© 2002 WILEY-LISS, INC.

collagen peptide.14 This crystal structure illustrates thedirect ionic interaction between the divalent cation at the�2 MIDAS and the carboxylate side-chain of a glutamateresidue in the ligand collagen peptide.

Both NMR and crystal structures of the �L I-domain areavailable in the absence of a ligand or ligand mimetic, witheach structure adopting the “closed” form of the I-domain.8,15,16 The crystal structure of the lovastatin-bound �L I-domain is very similar to each of the published�L Mg2� and Mn2�-bound x-ray structures.17 The bindingof lovastatin to an allosteric regulatory site of the I-domainof �L inhibits the interaction of �L�2 with ICAM-1 bystabilizing the inactive “closed” conformation.17 Both aconstitutively active and inactive form of the �L I-domainhave been obtained by the use of disulfide bonds at theC-terminal helix to lock the domain in the “open” or“closed” conformations respectively.11 Despite all experi-mental efforts, no high-resolution structure of the active“open” form of the �L I-domain complexed with a ligand orligand mimetic is currently available.

Two previous studies have attempted to investigate thespecific interactions in the �L I-domain/ICAM-1 D1 inter-face by a combination of molecular fitting and alaninescanning mutagenesis experiments. However, these mod-els were built by using the “closed” Mg2� �L I-domain,18

which would not reflect the ligand-bound conformation ofthe I-domain. We have modeled the structure of the �LI-domain in the open conformation by using the �MI-domain Mg2� crystal structure (PDB code 1IDO)5 as atemplate. We hypothesize that the conformational changesaccompanying the activation of the �L I-domain are simi-lar to those observed in the �2 and �M I-domains. Theresulting three-dimensional model was used in computa-tional docking studies with the ICAM-1 D1 domain. Thekey interaction of Glu 34 in the ICAM-1 D1 domain withthe open �L I-domain MIDAS was consistently reproduced

by our calculations, suggesting the structural nature ofthis key biological interaction.

RESULTSModel of �L I-Domain in the Open Conformation

To model the �L I-domain in the open conformation(residues Gly 128–Val 308), we used the �M I-domainMg2� crystal structure (PDB code 1IDO)5 as a template.�M was selected on the basis of the high degree ofhomology between the two domains (sequence identity �40%) (Table I). The divalent cation-free �L I-domain (PDBcode 1ZON) was used as the initial structure for modeling,and the �L Mn2� structure (PDB code 1LFA)15 as theclosed conformation [Fig. 1(a)–(c)]. To convert the �LI-domain from the closed to the open form, we moved the�7 helix as a rigid-body toward the bulk of the domain tomatch the orientation seen in the �L divalent cation-freecrystal structure [Fig. 1(b)]. This was followed by a down-ward, four-residue, piston-like movement of the �7 helix (afull helical turn) to arrive at its final orientation [Fig. 1(c)].

Changes in three loops—between �1 and �1 (Ser 139–Gln 143); between �4 and �6 (Gly 262–Thr 267); andbetween �5 and �7 (Leu 289–Phe 292)—are critical inmodeling the �L I-domain in the open conformation. Theorientation of the �1-�1 loop was determined by the �torsion rotation of the �1 helix at residue Ser 139. In theclosed and open �M I-domain structures, this hingelikemovement of the �1 helix is also observed starting atresidue Gly 141.

A swinglike movement of Phe 265 side-chain in �Lresulted from modeling the orientation of the �4-�6 loop inthe open conformation. The open conformation of the�5-�7 loop was generated by the downward translation ofthe �7 helix, which resulted in �10 Å shift in the C�position of the Phe 292. The Ramachandran plot of the

TABLE I. Structure-Based Sequence Alignment of the Open �L Model and �M I-Domain (PDB code 1IDO)†

---b1-� a1 ----b2-� -b2�� a2ss: bbbbbbb hhhhhhhhhhhhhhhh bbbbbbbb bbbb hhhhh�L: 131 DLVFLFDGSMSLQPDEFQKILDFMKDVMKKLSNTSYQFAAVQFSTSYKTEFDFSDYVKWK 190cons: D� FL DGS S� P �F��� �F� VM��L � F� �Q�S �� F F ��

�M: 134 DIAFLIDGSGSIIPHDFRRMKEFVSTVMEQLKKSKTLFSLMQYSEEFRIHFTFKEFQNNP 193a3 a4 ---b3--�

ss: hhhhhh hhhhhhhhhhh bbbbbbbb�L: 191 DPDALLKHVKHMLLLTNTFGAINYVATEVFREELGARPDATKVLIIITDGEATDS----GNI 248cons: �P �L�K � �L T�T I V E�F GAR �A K�L��ITDGE G�M: 210 NPRSLVKPITQLLGRTHTATGIRKVVRELFNITNGARKNAFKILVVITDGEKFGDPLPLGYE 269

a5 --b4--� a6 -b5-� a7ss: hhhh bbbbbbb hhhhhhhhhh bbbb hhhhhhhhhhh�L: 249 DAAKD----------IIRYIIGIGKHFQTKESQETLHKFASKPASEFVKILDTFEKLKDLFTELQ 303cons: D � �IRY�IG�G F����S�� L� ASKP � V �� FE LK � �L�

�M: 270 DVIPEADREGADREGVIRYVIGVGDAFRSEKSRQELNTIASKPPRDHVFQVNNFEALKTIQNQLR 329a7

ss: hhhhh�L: 304 KKIYVIEG 311cons: �KI� IEG�M: 330 EKIFAIEG 337†Conserved residues and residues conserved by type (shown as �) are indicated. Insertions within the �M sequence corresponding to the region ofthe �5 helix are shown in bold. Regions of secondary structure are labeled (b: �-sheet, and h: helix). The alignment was initially generated byBlast28 and edited on the basis of the �L I-domain model, which encompasses residues Gly 128–Val 308.

152 G.B. LEGGE ET AL.

model indicates 98.8% of the (�, �) angles of all residuesare in the core and allowed regions, and 1.2% of theresidues are in the generously allowed region.19

The final model structure more closely resembles that ofthe open �M Mg2� I-domain template structure than theinitial starting structure or the closed �L I-domain. Super-position of the backbone heavy atoms of the �L I-domainmodel with the open �M I-domain gave a root-mean-square deviation (RMSD) of 1.5 Å, whereas for the �LI-domain in the closed conformation the RMSD was 2.5 Å.The divalent cation-free �L I-domain crystal structure[Fig. 1(b)], which was used as an initial starting structure,may be interpreted as an intermediate state between theclosed and open conformations.

Model of the �L MIDAS in the Open Conformation

The metal site within the model of the open �L I-domainis based on the differences identified in the open and closedconformations of the �M I-domain5,20 [Fig. 2(a) and (b)]. Inthe model, the ionic bond to Asp 239 is broken and replacedby a water-mediated interaction to the metal ion, which iscritical in allowing a 2 Å shift in the position of the metalatom. The absence of the ionic bond to the Asp 239 residuewould result in enhancement of the electrophilicity of themetal cation in the I-domain, in a similar manner to thatdescribed initially for the �M I-domain.5 The translationof the �1 helix results in a 2 Å shift in the side-chainhydroxyl oxygen of residue Ser 141. This hingelike move-ment is the largest and most important movement of the

Fig. 1. Ribbon diagrams of the “closed” �L I-domain crystal structure (PDB code 1LFA (a), “intermediate” �L I-domain divalent cation-free crystalstructure (PDB code 1ZON); (b:) “open” �L I-domain model. (c) Arrows highlight changes in tertiary structure between each structure from “closed” to“intermediate” to “open.” The two Phe residues (Phe 265 and Phe 292) that undergo a significant change in their orientation between the two states areshown in black wireframe, and the divalent cation in the closed and open structures is represented as a gray sphere. Close-up of the structural changesat the MIDAS motif for the (d) closed �L I-domain crystal structure and the (e) open �L I-domain model with a glutamate sidechain modeled in the freeoctahedral vertex. Coordinating side-chains are shown as ball and sticks, whereas the peptide backbone is represented as a ribbon. Protons are omittedfor clarity, and Wat 1 and Wat 2 are water molecules. Figures were generated by the program AVS (Advanced Visual Systems). [Color figure can beviewed in the online issue, which is available at www.interscience.wiley.com.]

MODEL OF THE �L�2/ICAM-1 INTERACTION 153

Fig

.2.

a:R

ibbo

ndi

agra

mof

the

open

�L

I-do

mai

nsh

own

inbl

ue,

with

the

20lo

wes

ten

ergy

dock

ings

ofth

eIC

AM

-1D

1do

mai

nfr

omth

elo

wes

ten

ergy

clus

ter,

clus

tere

dby

usin

gan

RM

SD

tole

ranc

eof

3Å

repr

esen

ted

asye

llow

tube

s.T

heG

lu34

side

-cha

inis

show

nfo

rea

chdo

ckin

g,an

dth

eca

rbox

ylox

ygen

atom

sar

ere

pres

ente

din

ball

and

stic

kfo

rmat

.b:R

ibbo

ndi

agra

mof

the

open

�L

I-do

mai

nm

odel

and

the

ICA

M-1

stru

ctur

eco

ntai

ning

the

D1

and

D2

dom

ains

,su

perim

pose

don

the

low

est

ener

gydo

ckin

gof

ICA

M-1

D1

dom

ain

(whi

chsh

ows

the

Glu

34ca

rbox

ylat

ew

ithin

bond

ing

dist

ance

ofth

eM

g2�

ion

ofth

eM

IDA

Sin

the

open

�L

I-do

mai

n).T

heM

IDA

Sm

etal

coor

dina

ting

resi

dues

and

the

Glu

34in

ICA

M-1

D1

are

show

nas

ball

and

stic

kre

pres

enta

tions

.T

hem

etal

catio

nis

show

nas

agr

aysp

here

.T

heor

ient

atio

nof

the

Glu

34si

dech

ain

isth

atfo

und

inth

edo

cked

com

plex

.c:

Rib

bon

diag

ram

ofth

eop

en�

LI-

dom

ain

mod

elan

dth

elo

wes

ten

ergy

ICA

M-1

D1

dock

ing

asde

scrib

ed,

whe

reth

eIC

AM

-1D

1is

rota

ted

180°

alon

gth

ey

axis

and

subs

eque

ntly

tran

slat

edal

ong

the

xax

isto

show

an“o

pen

book

”vi

ewof

the

bind

ing

inte

rfac

e.T

heG

lu34

inIC

AM

-1D

1ar

esh

own

asba

llan

dst

ick

repr

esen

tatio

ns,a

ndth

em

etal

catio

nis

show

nas

agr

aysp

here

.A

blac

klin

ebe

twee

nth

etw

odo

mai

nsre

pres

ents

the

“spi

ne”

ofth

ebo

ok.

(a)

and

(c)

wer

ecr

eate

dby

AV

S,

and

(b)

was

crea

ted

byus

ing

Insi

ghtII

(Acc

elry

s).

154 G.B. LEGGE ET AL.

metal-binding residues between the two conformations.All other residues in the open conformation have positionsthat lie close to that in the closed conformation. A secondwater molecule is modeled to form the water-mediatedinteraction between Asp 137 and the divalent cation. Thewater molecules are in a cis configuration in the closedconformation, whereas in the open form they adopt a transconfiguration.

Docking of the ICAM-1 D1 Domain

We performed protein-protein docking simulations ofthe D1 domain (residues Val 17–Val 82) from the ICAM-1D1/D2 crystal structure (PDB code 1IAM)9,21,22 to themodel of the open �L I-domain, by using the programAutoDock 3.05.23 Because no biologically relevant interac-tions of the �M I-domain with the ICAM-1 D1 domain havebeen observed,24 and the closed conformation of the �LI-domain may represent the low affinity form of thisdomain, ICAM-1 D1 domain dockings with these twoI-domains were performed as controls.

We used a two-step approach. In the first approxima-tion, torsional flexibility was allowed only for the side-chain of the key Glu 34 residue and neighboring surface-exposed side-chains from residues Lys 39, Met 64, and Tyr66. Aside from these defined 11 active torsions, the remain-ing residues in both molecules were fixed. A set of 10docking simulations was performed, with the results clus-tered for all dockings that had an all-atom RMSD within1.0 Å. In the initial calculations, the two lowest energydockings were clustered (RMSD 1 Å), as well as twomore conformers of comparable energy that show the keyGlu 34 residue in approximately the same spatial orienta-tion and close to the metal ion. This group of conformers,which has the ICAM-1 D1 in the same relative orientation,accounts for 40% of all dockings performed. In the lowestenergy cluster, the critical Glu 34 carboxyl group in theICAM-1 ligand is 4.1 Å from the metal atom in the �LI-domain MIDAS and is oriented along the axis of the(previously) unoccupied, sixth coordination site of themetal.

In a second step, the dockings were repeated with a totalof 28 active torsions, selected after inspection of the initialdockings. A total of 100 docking simulations were per-formed. For the open �L model, ICAM-1 D1 dockings wereinitially clustered with a clustering tolerance of 3 Åall-atom RMSD, with the lowest energy cluster containinga total of 50 of the 100 dockings performed. This groupcontains two subgroups that are evident when the dock-ings are clustered with a tolerance of 2.5 Å all-atomRMSD. Figure 2(a) shows the 20 lowest energy dockingsfrom this 3 Å all-atom RMSD cluster. It is immediatelyevident from this diagram that there is only limitedvariation of the Glu 34 side-chain within these dockings,even though there are three active torsions in this side-chain. Indeed the per residue variance within these 20structures is lowest at residue Glu 34 in ICAM-1 D1.

Mutagenesis data indicated that the interaction of theGlu 34 carboxyl on the C-strand of the ICAM-1 D1 domainwith the metal ion at the MIDAS in the I-domain isessential to the �L�2/ICAM-1 interaction.7,10 For any

model to be considered biologically relevant, it must mimicthis interaction. Figure 2(a) illustrates that in most of theopen �L I-domain dockings this residue is placed in anorientation that may potentially result in an ionic bond tothe divalent cation in the MIDAS of the �L I-domain(closest Glu 34 carboxylate-MIDAS metal distance 2.51 Å).In contrast, for none of the dockings of the ICAM-1 D1 tothe “closed” �L I-domain or the “open” �M I-domain wasthe Glu 34 carboxylate of ICAM-1 observed in a favorablecoordination geometry. In addition, the minimum distancebetween the MIDAS metal cation and Glu 34 carboxylatewas greater in each of the control dockings, with theminimum distance being 3.1 Å for the “closed” �L and 2.9Å “open” �M I-domain ICAM-1 D1 domain docking calcula-tions. For the open �L/ICAM-1 D1 docking, lowest energyobtained was 27.10 kcal/mol, which is lower than thevalues obtained for both the closed �L and open �MI-domain/ICAM-1 dockings (24.73 kcal/mol and 26.55kcal/mol respectively).

Description of the “Open” �L I-Domain/ICAM-1 D1Docking Interface

The lowest energy structure that showed the Glu 34carboxyl oxygen within coordination distance from theMIDAS (second lowest energy overall) was selected as arepresentative structure for the dockings and is shown inFigures 2(b) and 2(c). The interface of the complex showsshape complementarity. A shallow groove along the MIDASof the “open” �L I-domain model fits a slight bulge formedby the C-strand in ICAM-1 D1 (see Fig. 3). The interfaceencompasses the C, D, F, and G �-strands in ICAM-1 D1and the loops near the MIDAS in the “open” �L I-domainmodel. Close contacts predicted within the “open” �LI-domain/ICAM-1 D1 interface are summarized in TableII. Figure 2(b) suggests there are no close contacts of theICAM-1 D2 domain with the �L I-domain.

The binding interface between the open �L I-domainmodel and the ICAM-1 D1 also shows electrostatic comple-mentarity (Fig. 3). Apart from the interaction between Glu34 of ICAM-1 D1 and the metal cation in the MIDAS, thereare also several charge-dependent interactions formedwith in the interface (Table II). These electrostatic interac-tions may act in such a way as to steer the binding betweenthe two surfaces. This may also account for the high degreeof clustering within the open �L I-domain/ICAM D1dockings.

Mutation studies of the “closed” �L I-domain and ICAM-1D1 identified three critical residues at the interface withICAM-1 Leu 205, Glu 243, and Thr 243.18 Each of thesecritical residues makes close contacts with ICAM-1 D1 inour docking model (Table II). Five additional residues wereidentified in the �L I-domain as showing a partial reduc-tion in ICAM-1 binding (Glu 146, Thr 175, His 264, andGlu 293).18 They do not show close contacts with ICAM-1D1 in our model and may play an indirect role in mediatingthe �L�2/ICAM-1 interaction.

Residues in ICAM-1 D1 that are known to be importantin �L�2 recognition (Glu 34, Lys 39, Met 64, Tyr 66, Asn68, and Gln 73) all make close contacts with the �LI-domain in our model. It is worth noting that our model

MODEL OF THE �L�2/ICAM-1 INTERACTION 155

Fig. 3. a: Solvent-excluded molecular surface of the open �L I-domain model with the docked ICAM-1 D1domain shown as a tube. b: Solvent-excluded molecular surface of the docked ICAM-1 D1 domain with theopen �L I-domain model shown as a tube. Each surface was colored according to the electrostatic potentialexperienced at one water-radius above the molecular surface with positive (blue), neutral (white), and negativevalues (red) indicated. In the tube representations, the residues well away from the docked surface wereclipped and the N-and C-termini of the tubes were colored blue and red, respectively. Only side-chains andparts of the MIDAS that lie within 4.0 Å closest atom distance from the interfacing molecule are shown. Figureswere generated by using the program MSMS.31

Figure 4.

predicts an interaction of Lys 50 in ICAM-1 D1 with bothThr 267 and Glu 269 in the �L I-domain. Althoughmutation of either of these residues to alanine did notgreatly affect binding to ICAM-1, individual mutation ofeither of these residues may be largely compensated by theinteraction of Lys 50 with the remaining residue.

Source of the ICAM-1 D1-Binding Specificity

A comparison of the docked open �L I-domain/ICAM D1domain the �2 I-domain/collagen peptide crystal struc-ture14 is shown in Figure 4. When the two I-domains aresuperimposed, the bound ligands within the two com-plexes show a similar directionality. In particular, in theregions near the metal-coordinating glutamic acids, theorientation of the middle strand of collagen triple helix andthe C-strand of the ICAM-1 D1 is strikingly similar. Thereis a total of 1292 Å2 of surface area buried within the �LI-domain/ICAM-1 D1 docking interface. This is �30%larger than the surface area within the interface of the �2I-domain/collagen peptide helix (PDB code 1DZI) (980 Å2)calculated by using the same parameters.25 The five-residue insertion (the C-helix) at the �4-�6 loop (residuesLeu 283–Asn 295) in the �2 I-domain plays a role indetermining the selectivity of ligand binding, which ori-ents the collagen ligand to lie to the side of the �2 I-domain[Fig. 4(c)]. The insertion is absent in the �L I-domain, andthe �4-�6 loop (residues Gly 262–Glu 269) is relativelyflattened in our model, which allows the ICAM-1 D1 todock across the MIDAS in the open �L I-domain [Fig. 4(d)].In the docked complex, the ICAM-1 D1 domain makessignificant close contacts with this �4-�6 loop (Table II).

To further investigate which regions of the open �LI-domain are important for the specificity of ICAM-1 D1binding, we superimposed the closed �L I-domain andopen �M I-domain onto the open �L I-domain in thedocked complex [Figs. 5(b) and (c)]. The interaction of�M�2 with ICAM-1 is mediated through the D3 domain24;hence, a specific biologically relevant interaction of theopen �M I-domain with the ICAM-1 D1 domain is notexpected. This is consistent with what was observed in thedockings of the ICAM-1 D1 to the open �M I-domain eventhough it was this structure that was used as a templatefor the open �L I-domain model. Superposition of the open�M I-domain onto the open �L I-domain in the dockedcomplex suggests a single region that may discriminate forICAM-1 D1 binding, the �3-�5 loop (residues Asp 239–Ser245 in �L and Asp 242–Phe 246 in �M) [Fig. 5(b)]. The�3-�5 loop in the open �M structure is different from thatin �L because of an insertion of five residues in the �MI-domain at the beginning of the �5 helix (Table I). Theinsertion in the �3-�5 loop in �M alters the orientation ofthe �3-�5 loop. The superposition of the closed �L I-domain onto the open �L I-domain in the docked complexsuggests two regions that may discriminate betweenICAM-1 D1 binding, the �1-�1 loop and �4-�6 loop [Fig.5(c)]. Both of these loop regions were altered when model-ing the open �L I-domain (Fig. 1).

The rather large, relatively flat, surface of the ICAM-1D1/open �L I-domain docking interface suggests a predomi-nantly preformed complementarity mode of binding ratherthan induced fit. A preformed complementarity mode ofbinding is consistent with the existence of both an “open”and a “closed” I-domain conformations, where the mole-cule is “locked” into an active or inactive conformation bychanges in the quaternary structure of the integrin. Ourdata support the hypothesis that the closed and openconformations of the I-domain represent the active and

Fig. 4. a: Lateral view ribbon diagram of the lowest energy docking ofthe open �L I-domain (shown in blue) and the ICAM-1 D1 domain (shownin yellow). b: Lateral view ribbon diagram of the �2 I-domain (shown inred) complexed with a collagen tripeptide (shown in green) (PDB code1DZI), where the two I-domains were superimposed within the first�-strand and shown in the same orientation. The metal ion is shown as agray sphere, and the coordinating glutamic acid side-chain is shown (Glu34 in ICAM-1 and Glu 11 in strand 1 of the collagen helix). c: Top view of(a). d: Top view of (b). The figure was generated by using program AVS.

TABLE II. Intramolecular Residue-Residue Contacts inthe ICAM-1/“open” �L I Domain Complex†

ICAM-1 D1 �L I domain29 LYS 245 SER30 LEU 241 GLU

242 ALA31 LEU 269 GLU33 ILE 141 SER34 GLU 139 SER

140 MET141 SER204 LEU205 LEU206 THRMg2� (magnesium)Wat I

35 THR 140 MET141 SER

36 PRO 140 MET142 LEU

38 PRO 143 GLN39 LYS 141 SER

239 ASP263 LYS

41 GLU 263 LYS266 GLN267 THR270 SER

50 LYS 267 THR269 GLU

64 MET 205 LEU66 TYR 241 GLU

243 THRWat 1

68 ASN 242 ALA243 THR244 ASP245 SER

73 GLN 207 ASN243 THR244 ASP

†Residues in bold were found to be important in intermolecularinteractions by previously reported alanine scanning data.18,29,30 Wat1 is the equatorial bound water molecule to the Mg2� in the MIDAS.

MODEL OF THE �L�2/ICAM-1 INTERACTION 157

inactive conformations of the protein.20 In particular, theabsence of any docked conformation of the ICAM-1 D1domain with the closed �L I-domain, where the key Glu 34carboxyl group can form a coordination bond with the

divalent cation at the MIDAS, supports the idea that thisclosed conformation represents the inactive form. Ourmodel suggests that high-affinity binding of the �L�2integrin with ICAM-1 will result only after a change in the

Fig. 5. a: Solvent-excluded molecular surface of the open �L I-domain model and the ICAM-1 D1. b:Solvent-excluded molecular surface of the closed �L I-domain crystal structure (PDB code 1LFA) and theICAM-1 D1, where the closed �L I-domain was superimposed onto the open �L I-domain the in the dockedcomplex. c: Solvent-excluded molecular surface of the open �M I-domain crystal structure (PDB code 1IDO)and the ICAM-1 D1, where the open �M I-domain was superimposed onto the open �L I-domain the in thedocked complex. Each figure is in the “open book” view [Fig. 2(c)] and is colored according to the intermoleculardistance to closest atom. The color key is shown (0–2 Å: red; 2–4 Å: blue; and 4–6 Å: green). Figures weregenerated by using the program MSMS.31

158 G.B. LEGGE ET AL.

structure of the �L I-domain. This structural changewould be linked with activation of the leukocyte by chemo-kines or activation specific antibodies.

Our data indicate the alternate insertions of the loops inopen �M and �2 I-domains and the alternate orientation ofthe loops in the closed �L I-domain near the MIDAS arecritical for ligand-binding specificity. Structural variationwithin these loops form an obstacle that prevents theformation of the critical interaction of the Glu 34 carboxy-late of ICAM-1 D1 with the metal cation of these domains.The orientation of these loops are essential in determiningboth the shape and charge complementarity within thecomplex that would allow the formation of the key metal–glutamate coordination bond resulting in high-affinitybinding.

MATERIALS AND METHODSModeling and Energy Minimization the �LI-Domain in the Open Conformation

The open �L I-domain was modeled following the C�trace of the �M I-domain Mg2� template structure. Fiveregions of the initial �L I-domain starting structure weremodeled consecutively to the orientations seen in thetemplate structure. These regions include the C-terminal�6 and �7 helices and the �5 strand, which encompassesall residues from Arg 256 to Val 308; the �1 helix and loop(residues Ser 139–Ser 165); and the �3 helix and loop(residues Val 187–His 201) in �L. In all cases, the struc-tures were aligned with the template structure by superpo-sition of the backbone C� atoms of the core �1, �2, �3, and�4 strands, which are structurally highly similar in theopen and closed �M I-domain structures. The �1 helix wasremodeled by disconnecting the peptide bond betweenresidues Ser 165 and Tyr 166, followed by a single �torsion rotation of the Ser 139 N-C� bond; this wassufficient to overlay this helix with the template structure.Similarly, the �3 helix and loop was remodeled by firstdisconnecting the peptide bond between residues Val 187and Lys 188, followed by various rotations starting withHis 201 C�. The peptide bond was reconnected in eachsection after remodeling. All other regions of the �L modelremained untouched, apart from the final minimization ofthe model. Each reconstructed region was separatelytethered to the C� atoms of the template structure andminimized. The three-dimensional model was energy mini-mized in 1000 steps by using the CVFF force field withinthe “Discover” program as implemented by the InsightIIpackage (Accelrys, San Diego, CA). A Morse potential wasused for bond energies until a maximum derivative of 1was accomplished and cross-term energies were included.Calculations were performed in vacuo by using a dielectricconstant of 1. Overall energy minimization was appliedafter the modeling and minimization of all individualsections using fixed C� coordinates for residues Gly 128–Ile 306. The C-terminal residues Ile 307 and Val 308 hadno fixed atoms. After minimization, the final energy param-eters for the model are as follows: bond energy was 408kcal; theta 929 kcal; phi 387 kcal; nonbond 1070 kcal;coulomb 5705; and total energy 2789 kcal.

Metal Coordination Site

After minimization of the model, the Mg2� cation sitewas built by using the program InsightII (Accelrys). Anoctahedral coordination sphere was assumed, with five ofthese sites of the magnesium ion being bonded to twomodeled waters and residues Ser 139, Ser 141, and Thr206 of the I-domain. The divalent cation (formal charge�2) in the model of the open �L I-domain is translated by1.9 Å, between the “closed” and the “open” conformations,which is similar to the metal translation observed for the�M I-domain crystal structures.5 The translation of themetal ions are in the opposite direction of the movementsof �1 and �7 helices and results in a �30° rotation of thesixth, unoccupied octahedral site of the cation.

Molecular Dockings

In the AUTODOCK 3.05 program,23 the ligand is pres-ently compiled to a maximum size of 1024 atoms; hence, atruncated form of the D1 of ICAM-1 containing all theatoms between residues Val 17 and Val 82 was used. Thenonpolar hydrogens were merged to give a total of 1010atoms within the ligand. The truncated version of ICAM-1D1 fully includes the molecular surface that makes inter-molecular contacts with �L.18 The solvent-exposed Mg2�

octahedral vertex was left empty in the model duringdocking calculations.

Atomic solvation parameters and fractional volumeswere assigned to the protein atoms by using the AddSolutility, and grid maps were calculated by using AutoGridutility in AutoDock 3.05.23 A grid map with 127 � 127 �127 points and a grid point spacing of 0.375 Å included thewhole MIDAS-containing face of the I-domain as well as aspace large enough to accommodate the truncated ICAM-1D1 structure with ease. In the first round of dockings, theCVFF force field from Discover (Accelrys) was used toassign partial charges to the proteins, whereas in thesecond round, Kollman “united-atom” charges26 were used.AutoDock 3.0523 uses a Lamarckian Genetic Algorithm(LGA) that couples a typical Darwinian genetic algorithmfor global searching with the Solis and Wets algorithm forlocal searching. The LGA parameters were defined asfollows: the initial population of random individuals had asize of 50 individuals; each docking was terminated with amaximum number of 5 � 105 energy evaluations or amaximum number of 27,000 generations, whichever camefirst; mutation and crossover rates were set at 0.02 and0.80, respectively. An elitism value of 1 was applied, whichensured that the top ranked individual in the populationalways survived into the next generation. A maximum of300 iterations per local search was used. The probability ofperforming a local search on an individual was 0.06,whereas the maximum number of consecutive successes orfailures before doubling or halving the search step size was4. This set of parameters was used for all dockings.

Nonpolar hydrogens were added to the docked complexfor the “open” �L: I-domain/ICAM-1 D1 domain, which wasthen energy minimized by using the Discover 3 module ofInsightII 2000 (Accelrys). The CVFF force field was used,and no fixed atoms were constrained. There were sixdistance constraints applied to the Mg2�-oxygen bonds:

MODEL OF THE �L�2/ICAM-1 INTERACTION 159

the allowed distance range was from 2.0 to 2.1 Å.27 Theconjugate gradients algorithm was used for a total of 800iterations. No significant changes in the structure of thecomplex occurred: the energy-minimized structure had anall-atom RMSD of 0.91 Å from the docked complex struc-ture.

ACKNOWLEDGMENTS

We thank R. A. Lerner, P. E. Wright and H. J. Dyson forsupport and discussions. G. B. L. thanks Novartis forproject funding, and F. G. thanks the J. S. GuggenheimMemorial Foundation for a Fellowship. This work wassupported in part by grants from the National Institutes ofHealth GM48870 (G.M.M and A.J.O.), RR08065 (M.S.F.and A.J.O) and GM49899 (Y.T.) and from the NationalScience Foundation CA ACI-9619020 (M.S.F. and A.J.O.).

REFERENCE

1. Springer TA. Traffic signals for lymphocyte recirculation andleukocyte emigration: the multistep paradigm. Cell 1994;76:301–314.

2. Gahmberg CG. Leukocyte adhesion: CD11/CD18 integrins andintercellular adhesion molecules. Curr Opin Cell Biol 1997;9:643–650.

3. Kamata T, Liddington RC, Takada Y. Interaction between colla-gen and the �2 I-domain of integrin �2�1: critical role of conservedresidues in the MIDAS region. J Biol Chem 1999;274:32108–32111.

4. Velling T, Kusche-Gullberg M, Sejersen T, Gullberg D. cDNAcloning and chromosomal localization of human alpha(11) inte-grin: a collagen-binding, I domain-containing, beta(1)-associatedintegrin alpha-chain present in muscle tissues. J Biol Chem1999;274:25735–25742.

5. Lee JO, Rieu P, Arnaout MA, Liddington R. Crystal structure ofthe A domain from the alpha subunit of integrin CR3 (CD11b/CD18). Cell 1995;80:631–638.

6. Takada Y, Kamata T, Irie A, Puzon-McLaughlin W, Zhang X.-P.Structural basis of integrin-mediated signal transduction. MatrixBiol 1997;16:143–151.

7. Staunton D, Dustin M, Erickson H, Springer T. The arrangementof the immunoglobulin-like domains of ICAM-1 and the bindingsites for LFA-1 and rhinovirus [published errata appear in Cell1990 Jun 15;61(2):1157 and 1991 Sep 20;66(6):following 1311].Cell 1990;61:243–254.

8. Qu A, Leahy DJ. The role of the divalent cation in the structure ofthe I domain from the CD11a/CD18 integrin. Structure 1996;4:931–942.

9. Bella J, Kolatkar PR, Marlor CW, Greve JM, Rossmann MG. Thestructure of the two amino-terminal domains of human ICAM-1suggests how it functions as a rhinovirus receptor and as an LFA-1integrin ligand. Proc Natl Acad Sci USA 1998;95:4140–4145.

10. Stanley P, Hogg N. The I domain of integrin LFA-1 interacts withICAM-1 domain 1 at residue Glu-34 but not Gln-73. J Biol Chem1998;273:3358–3362.

11. Lu C, Shimaoka M, Ferzly M, Oxvig C, Takagi J, Springer TA. Anisolated, surface-expressed I domain of the integrin alpha Lbeta 2is sufficient for strong adhesive function when locked in the openconformation with a disulfide bond. Proc Natl Acad Sci USA2001;98:2387–2392.

12. Lu C, Shimaoka M, Zang Q, Takagi J, Springer TA. Locking inalternate conformations of the integrin alpha Lbeta 2 I domain

with disulfide bonds reveals functional relationships among inte-grin domains. Proc Natl Acad Sci USA 2001;98:2393–2398.

13. Huth JR, Olejniczak ET, Mendoza R, Liang H, Harris EA, LupherML Jr, Wilson AE, Fesik SW, Staunton DE. NMR and mutagen-esis evidence for an I domain allosteric site that regulateslymphocyte function-associated antigen 1 ligand binding. ProcNatl Acad Sci USA 2000;97:5231–5236.

14. Emsley J, Knight CG, Farndale RW, Barnes MJ, Liddington RC.Structural basis of collagen recognition by integrin alpha2beta1.Cell 2000;101:47–56.

15. Qu A, Leahy D. Crystal structure of the I-domain of the CD11a/CD18 (LFA-1, �L�2) integrin. Proc Natl Acad Sci USA 1995;92:10277–10281.

16. Legge GB, Kriwacki RW, Chung J, Hommel U, Ramage P, CaseDA, Dyson HJ, Wright PE. NMR solution structure of the inserteddomain of human leukocyte function associated antigen-1. J MolBiol 2000;295:1251–1264.

17. Kallen J, Welzenbach K, Ramage P, Geyl D, Kriwacki R, Legge G,Cottens S, Weitz-Schmidt G, Hommel U. Structural basis forLFA-1 inhibition upon lovastatin binding to the CD11a I-domain.J Mol Biol 1999;292:1–9.

18. Edwards CP, Fisher KL, Presta LG, Bodary SC. Mapping theintercellular adhesion molecule-1 and -2 binding site on theinserted domain of leukocyte function-associated antigen-1. J BiolChem 1998;273:28937–28944.

19. Laskowski RA, MacArthur MW, Moss DS, Thornton JM. PRO-CHECK: a program to check the stereochemical quality of proteinstructures. J Appl Crystallogr 1993;26:283–291.

20. Lee JO, Bankston LA, Arnaout MA, Liddington RC. Two conforma-tions of the integrin A-domain (I-domain): a pathway for activa-tion? Structure 1995;3:1333–1340.

21. Greve JM, Davis G, Meyer AM, Forte CP, Yost SC, Marlor CW,Kamarck ME, McClelland A. The major human rhinovirus recep-tor is ICAM-1. Cell 1989;56:839–847.

22. Olson NH, Kolatkar PR, Oliveira MA, Cheng RH, Greve JM,McClelland A, Baker TS, Rossmann MG. Structure of a humanrhinovirus complexed with its receptor molecule. Proc Natl AcadSci USA 1993;90:507–511.

23. Morris GM, Goodsell DS, Halliday RS, Fig Huey R, Hart WE,Belew RK, Olson AJ. Automated docking using a Lamarckiangenetic algorithm and an empirical binding free energy function.J Comp Chem 1998;19:1639–1662.

24. Diamond M, Staunton D, Marlin S, Springe RT. Binding of theintegrin Mac-1 (CD11b/CD18) to the third immunoglobulin-likedomain of ICAM-1 (CD54) and its regulation by glycosylation. Cell1991;65:961–971.

25. Sanner MF. Python: a programming language for software integra-tion and development. J Mol Graphics Mod 1999;17:57–61.

26. Cornell W, Cieplak P, Bayly C, Gould I, Merz KJ, Ferguson D,Spellmeyer D, Fox T, Caldwell J, Kollman P. A second generationforce field for the simulation of proteins and nucleic acids. J AmChem Soc 1995;117:5179–5197.

27. Frausto da Silva JJR, Williams RJP. The biological chemistry ofthe elements, Oxford (UK): Oxford University Press, 1991. p244–245.

28. Tatusova TA, MTL. Blast 2 sequences—a new tool for comparingprotein and nucleotide sequences. FEMS Microbiol Lett 1999;174:247–250.

29. Huang C, Springer TA. A binding interface on the I domain oflymphocyte function-associated antigen-1 (LFA-1) required forspecific interaction with intercellular adhesion molecule 1 (ICAM-1). J Biol Chem 1995;270:19008–19016.

30. Kamata T, Wright R, Takada Y. Critical Thr and Asp residueswithin the I domains of �2 integrins for interactions with ICAM-1and C3bi. J Biol Chem 1995;270:12531–12535.

31. Sanner M, Olson AJ, Spehner JC. Reduced surface: an efficientway to compute molecular surface. Biopolymers 1996;38:305–320.

160 G.B. LEGGE ET AL.