A RT I C L E S

The LIR family (also known as the immunoglobulin-like transcripts(ILTs), monocyte/macrophage immunoglobulin-like receptors andCD85) comprises a set of immunoreceptors expressed on the surface oflymphoid and myeloid cells1,2. The LIRs, which are related to naturalkiller cell killer immunoglobulin-like receptors (KIRs) and theimmunoglobulin A (IgA) receptor FcαRI, are highly similar to oneanother, sharing 63–84% amino acid identity in their extracellularregions. All except LIR-4 are type 1 transmembrane proteins, contain-ing either two or four immunoglobulin superfamily domains in theirextracellular regions. One subset of cell surface LIR molecules (LIR-1,LIR-2, LIR-3, LIR-5 and LIR-8) transmits inhibitory signals throughintracellular tyrosine-based inhibitory motifs, whereas another group(LIR-6, LIR-7, ILT7, ILT8, ILT10 and ILT11) transmits activatory sig-nals by associating with signaling adaptor molecules1,2.

After being characterized as ILT2 (ref. 3), LIR-1 (the most broadlyexpressed LIR family member) was identified as the receptor for UL18(ref. 4), a MHC class I homolog encoded by human cytomegalovirus5.UL18 shares 25% amino acid sequence identity with classical MHCclass I molecules in the extracellular region, and associates with host-derived β2-microglobulin (β2M)6 and with short peptides that showcharacteristics of those presented by classical MHC class I proteins7.LIR-1 and LIR-2 are also receptors for a broad range of MHC class Imolecules, including classical (HLA-A, HLA-B and HLA-C) and non-classical (HLA-E, HLA-F and HLA-G) molecules, which they bind withsimilar affinities (in the micromolar range, like several other low affin-ity cell-surface interactions) and kinetics8,9. In contrast, individualKIRs show allele-specific recognition of MHC class I molecules10. Thestructural nature of host MHC class I ligand recognition by LIR-1 andLIR-2 and the molecular basis of LIR-1 targeting by the human

cytomegalovirus UL18 protein remain to be clarified. To address theseissues and to gain information about ligand recognition by the LIRfamily of immunoreceptors, we solved the structure of the ligand-binding D1D2 fragment of LIR-1 in complex with a classical MHC classI ligand, HLA-A*0201.

RESULTSOverall structure of the LIR-1−HLA-A2 complexWe determined the crystal structure of LIR-1 D1D2 bound to HLA-A*0201 presenting a peptide derived from human immunodeficiencyvirus 1 Pol (residues 309–317; ILKEPVHGV) in space group P3121 to3.4-Å resolution (Rcryst = 22.2%, Rfree = 31.0%; Table 1 and Fig. 1a).We determined a 2Fo – Fc annealed omit electron density map relat-ing to a portion of the LIR-1−HLA-A2 molecular interface (Fig. 1b).The D1D2 region of LIR-1 binds to MHC class I and UL18 with thesame binding affinity as the full-length extracellular region(D1–D4)8. The structures of the proteins in the LIR-1−HLA-A2 com-plex resemble previously reported structures determined individu-ally11,12: LIR-1 D1D2 contains two immunoglobulin superfamilydomains arranged at a nearly perpendicular angle, and HLA-A2includes a heavy chain with an α1-α2 domain peptide-binding plat-form and a membrane-proximal α3 domain, which associates nonco-valently with the β2M light chain and bound peptide. Unlike T cellreceptors (TCRs) and KIRs, which interact with the polymorphicpeptide-binding platforms of MHC molecules10,13, LIR-1 D1D2 rec-ognizes the side of the HLA-A2 molecule, forming two contact sur-faces that encompass residues from the HLA-A2 α3 domain, which ismainly nonpolymorphic, and β2M, which is conserved (Fig. 1a andTable 2). The interaction is therefore consistent with recognition by

1Division of Biology 114-96 and 2Howard Hughes Medical Institute, California Institute of Technology, Pasadena, California 91125, USA. 3Present address: CancerResearch UK Institute for Cancer Studies, University of Birmingham, Vincent Drive, Edgbaston, Birmingham B15 2TT, UK. Correspondence should be addressed toP.J.B. (bjorkman@its.caltech.edu).

Published online 3 August 2003; doi:10.1038/ni961

Crystal structure of HLA-A2 bound to LIR-1, a hostand viral major histocompatibility complex receptorBenjamin E Willcox1,3, Leonard M Thomas1,2 & Pamela J Bjorkman1,2

Leukocyte immunoglobulin-like receptor 1 (LIR-1), an inhibitory receptor expressed on monocytes, dendritic cells andlymphocytes, regulates cellular function by binding a broad range of classical and nonclassical major histocompatibility complex(MHC) class I molecules, and the human cytomegalovirus MHC class I homolog UL18. Here we describe the 3.4-Å crystalstructure of a complex between the LIR-1 D1D2 domains and the MHC class I molecule HLA-A2. LIR-1 contacts the mostlyconserved β2-microglobulin and α3 domains of HLA-A2. The LIR-1 binding site comprises residues at the interdomain hinge,and a patch at the D1 tip. The structure shows how LIR-1 recognizes UL18 and diverse MHC class I molecules, and indicatesthat a similar mode of MHC class I recognition is used by other LIR family members.

NATURE IMMUNOLOGY VOLUME 4 NUMBER 9 SEPTEMBER 2003 913

©20

03 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

reim

mu

no

log

y

A RT I C L E S

LIR-1 of a broad range of MHC class I molecules in a peptide-independent way1,2.

The crystal structure shows a 1:1 LIR-1:HLA-A2 stoichiometry, aspredicted by analytical ultracentrifugation studies8, with no evidencefor LIR-1 or HLA-A2 dimers or oligomers. The LIR-1 N-terminaldomain and N-terminal residues are oriented toward the membrane-proximal portion of the HLA-A2 ectodomain (Fig. 1a), which is mostconsistent with a trans interaction involving recognition of an MHCclass I molecule on a target cell by a LIR-1 protein on an opposing effec-tor cell. A cis interaction between LIR-1 and an MHC class I moleculeon the same cell surface would require LIR-1 domains 3 and 4 and theconnecting region of LIR-1 to nearly reverse directions compared withthe orientation of LIR-1 D1D2, resulting in a horseshoe-like arrange-ment of the four LIR-1 domains that is inconsistent with sedimentationvelocity data8 and conservation of residues involved in interactions atthe interface between D1 and D2 in the D2–D4 region8.

Conformational changes in LIR-1Although the structures of free HLA-A2 and HLA-A2 in a complex arenot very different (r.m.s. deviation on all Cα atoms of 0.91 Å), compar-ison of the structures of free and bound LIR-1 D1D2 showed a changein the D1D2 interdomain angle (Fig. 1c). The angles between the longaxes of the LIR-1 D1 and D2 domains were calculated as 84°, 85° and90° in crystal structures of free LIR-1 D1D2 (ref. 11). When in a com-plex with HLA-A2, the LIR-1 D1D2 angle increases to 100°, indicatingsubstantial flexibility of the D1D2 angle in the unbound state.Although crystal contacts could be involved in this interdomain shift,the interdomain angle in the co-crystal structure allows optimal con-tacts with both the α3 and β2M domains of HLA-A2, and is thereforelikely to reflect stabilization of a particular LIR-1 hinge region confor-mation that facilitates binding. A structurally analogous increase ofsimilar size (∼ 10°) in the interdomain angle of KIR2DL1 noted afterbinding to HLA-Cw4 (Fig. 1c) was also attributed to optimization ofdomain orientation for ligand binding, and was not considered rele-vant to inhibitory signal initiation14.

LIR-1 binding involves two surfaces on HLA-A2There are two distinct contact areas in the LIR-1−HLA-A2 interface(Fig. 2): the A′CC′FG face at the tip of LIR-1 D1 contacts the HLA-A2α3 domain (with the exception of D1 residue Lys42, which contactsβ2M residue Asp96), and the LIR-1 D1-D2 interdomain hinge regioncontacts the β2M domain. The contacting residues on the α3 domainare located at the end of strand A and in the A-B loop (residues193–200) and the E strand (residue 248; Fig. 3a). Contact regions onβ2M are located toward the N-terminus (residues 1–4), in the F-Gstrand loop (residues 86–89) and in the G strand itself (residues91–94, 96 and 99). The size of the LIR-1−HLA-A2 interface (∼ 1,700Å2 total buried solvent-accessible surface area) is marginally largerthan those of KIR-MHC interfaces (∼ 1,500Å2; ref. 10) and similar tothose of TCR-MHC complexes (1,700–1,900 Å2; ref. 13). Of the LIR-1binding surface that is buried after complex formation (835 Å2),about 70% is involved in contacts with 14 β2M residues, with theremainder contacting 6 α3 domain residues (Table 2 and Fig. 2). Thisrelative dominance of β2M is unprecedented among HLA-bindingimmunoreceptors, including KIRs10, TCRs13 and CD8 (ref. 15), andis partly responsible for the broad recognition properties of LIR-1.

Previous studies have suggested that both α3 and β2M interactionsurfaces in the crystal structure contribute energetically to MHC class Iand UL18 recognition. Domain-swapping experiments have shownthat LIR-1 D1-D4 binding was abolished when the α3 domain of HFEwas incorporated into UL18 and MHC class I proteins8, consistentwith the idea that specific LIR-1 D1 contacts with α3 domain residuesin the crystal structure are energetically important. Moreover, a prote-olytic fragment of LIR-1 D1D2, (residues 1–99, referred to as D1, butcontaining residues 97–99 of the interdomain hinge region) boundUL18 and HLA-Cw0702 with only one-quarter to one-third the affin-ity that of D1D2 (ref. 8). The binding of the D1 fragment alone cannow be understood as an interaction with the MHC class I α3 and β2Mdomains, with the reduction in affinity being caused by the absence ofone or more of the D2 residues in the interdomain hinge region(residues 100, 127, 184 and 187), which make contacts to β2M in thecase of D1D2.

A highly conserved LIR-1 binding site on MHC class I moleculesTo further rationalize the broad recognition of MHC class I by LIR-1,we examined the sequences of classical and nonclassical MHC class Imolecules, as well as MHC class I–like molecules that do not bind LIR-1 (FcRn, HFE and ZAG)8 in the α3 domain regions identified

914 VOLUME 4 NUMBER 9 SEPTEMBER 2003 NATURE IMMUNOLOGY

Table 1 Data collection and refinement statistics for the LIR-1–HLA-A2 complex

Parameter Value

Space group P3121

Unit cell dimensions 113.74, 113.74, 89.46

Data collection

Resolution (Å) 50.0–3.4 (3.5–3.4)

Number of observed reflections 61,195

Number of independent reflections 9,493 (928)

Completeness (%)a 99.9 (100)

Rmerge (%)b 18.0 (50.5)

I/σ(I) 11.7 (3.8)

Refinement

Resolution (Å) 50.0–3.4

Reflections in working set 9,000

Reflections in test set 493

Rcryst (%)c 22.2

Rfree (%)d 31.0

Number of protein atoms 4,498 (561 of 582 residues)

Average B factor (Å2) 31.2 (HLA-A2), 37.3 (β2m),

22.5 (peptide), 53.3 (LIR-1)

Anisotropic B correction B11 = B22 = –11.23 Å2; B33 = 22.46 Å2

r.m.s. deviation from ideality

Bond lengths

Bond lengths (Å) 0.021

Bond angles (deg) 1.85

Ramachandran plot quality

Most favored

Region of Ramachandran plot (%) 74.4%

Additionally allowed

Protein 23.9%

Generously allowed 1.7%

Disallowed 0.0%

aCompleteness = number of independent reflections/total theoreticalnumber. bRmerge (I) = (Σ|I(i) – <I(h)>|/ΣI(i)), where I(i) is the ‘ith’ observationof the intensity of the h, k, l reflection and <I> is the mean intensity frommultiple measurements of the h, k, l reflection. cRcryst (F) = Σh||Fobs(h)| –|Fcalc(h)||/Σh|Fobs(h)| and |Fcalc(h)| are the observed and calculated structurefactor amplitudes for the h, k, l reflection. dRfree is calculated overreflections in a test set not included in atomic refinement. Numbers inparentheses apply to data in the highest-resolution shell.

©20

03 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

reim

mu

no

log

y

A RT I C L E S

from the LIR-1–HLA-A2 structure as making contacts with LIR-1 (α3residues 193–196, 198 and 248). With the exception of Ala193, these allinvolve amino acid side-chain-specific interactions. In all classical andnonclassical MHC class I molecules examined, residues Glu198 andVal248 are conserved, residue 193 is either a proline or an alanine, andresidue 194 is an uncharged hydrophobic amino acid (Fig. 3a). Inaddition, residues 195–197 are conserved in HLA-A, HLA-B, HLA-C,HLA-E and HLA-F. The most divergent sequence is that of HLA-G,which contains two amino acid changes, at positions 195 (Ser to Phe)and 197 (His to Tyr). The positions and nonconservative nature ofthese changes indicate that they could affect interaction with HLA-G,consistent with the slightly higher (three- to fourfold) affinities of LIR-1 and LIR-2 for HLA-G relative to other MHC class I molecules9. Incontrast to most MHC class I molecules, FcRn, HFE and ZAG eachcontain four to six amino acid changes, of which three to four are non-conservative, from the MHC class I consensus sequence in the contactresidues. The comparison described above indicates that in addition toresidues on β2M, LIR-1 recognizes a sequence motif on the α3 domainthat is essentially restricted to classical and nonclassical MHC class Imolecules, and that α3 domain contacts provide essential energeticcontributions to binding energy. Subtle differences in the β2M and α3domain orientation could also affect LIR-1 binding, although the relative positions of the β2M and α3 domains are mostly conserved in

available structures of MHC class I proteins, and of the MHChomologs FcRn and HFE.

Of the six LIR-1 amino acids that contact the HLA α3 domain, four(Arg36, Tyr38, Arg39 and Lys41) are conserved in LIR-2 (Fig. 3b),whereas alternative uncharged polar amino acids are substituted atpositions 43 (Thr to Ser) and 76 (Tyr to Gln). Superposition of the D1domain of each receptor indicates these residues occupy similar posi-tions in LIR-1 and LIR-2. Conservation of interactions involving theseresidues is consistent with the similar affinities and broad specificity ofboth receptors for MHC class I molecules.

DISCUSSIONThe LIR-1−HLA-A2 structure can help in the interpretation of previ-ous studies suggesting that LIR-1 uses a common binding interaction torecognize UL18 and MHC class I molecules8. Domain-swappingexperiments have indicated that LIR-1 D1 and the MHC class I andUL18 α3 domains are interaction sites8, as now verified for LIR-1recognition of HLA-A2. Mutagenesis of LIR-1 identified four D1residues (Tyr38 and one or more of Tyr76, Asp80 and Arg84) thataffected binding to UL18 when substituted11. Tyr38 and Tyr76 areinvolved in specific contacts with HLA-A2 α3 domain residues in theregion of positions 193–196 (Table 2a). Together with the fact that contacts between LIR-1 and β2M are likely to be conserved in the viral

NATURE IMMUNOLOGY VOLUME 4 NUMBER 9 SEPTEMBER 2003 915

Figure 2 Interaction surfaces used by LIR-1 and HLA-A2. Grasp38 representations of theD1D2−HLA-A2 structure (center), with D1D2contact sites (yellow) on HLA-A2 (left), and HLA-A2 contact sites (yellow) on D1D2 (right).

Figure 1 Overall structure of the LIR-1−HLA-A2complex. (a) Ribbon diagram of the LIR1 D1D2−HLA-A2 structure. Cysteines are shown in ball-and-stick form with disulfide bonds in yellow. C, C termini of HLA-A2 heavy chain and LIR-1D1D2. Dashed lines, disordered regions (seeMethods). (b) The D1D2−HLA-A2 model in theregion of the β2M (cyan)−D1D2 (yellow) interfacehinge contact area superimposed on a 3.4-ÅSIGMAA-weighted 2Fo – Fc annealed omitelectron density map contoured at 1.0 σ. (c)Conformational change in LIR-1 and KIR2DL1after MHC class I binding. Comparison of theKIR2DL1−HLA-C14 and LIR-1−HLA-A2 complexstructures. The D1 domains of KIR2DL1 (ref.14) and LIR-1 D1D2 (green Cα representations)not in complexes are superimposed on the D1domains of their bound counterparts (red Cαrepresentations). In both complexes, the D1D2interdomain angle is increased in the boundD1D2 structure compared with that of the freeD1D2 structure.

a b

c

©20

03 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

reim

mu

no

log

y

A RT I C L E S

homolog, and the finding that the presence or absence of bound peptide in UL18 does not affect binding to LIR-1 (ref. 8), these resultsindicate that the LIR-1−MHC class I and LIR-1−UL18 interactioninterfaces are similar. If so, we can use the LIR-1 D1D2−HLA-A2 crystalstructure as a first-order model for a LIR-1−UL18 complex. Whereasboth cis and trans interactions with LIR-1 have been postulated asmodes of UL18 action16, a similar binding mode would favor recogni-tion in trans, such as UL18 on an human cytomegalovirus–infected cellengaging LIR-1 on an effector cell, or alternatively UL18−LIR-1 interac-tion between opposing membranes in intracellular compartments.

Despite the similarities in the binding of LIR-1 to UL18 and to MHCclass I molecules, UL18 achieves an affinity in the nanomolar range,over 1,000-fold higher than that of LIR-1−MHC class I interactions8.When the 13 potential N-linked glycosylation sites of UL18 (ref. 5) aremapped onto the D1D2−HLA-A2 structure, the predicted LIR-1 binding site is one of the few contiguous surfaces that does not contain

a potential glycosylation site (Fig. 4a), consistent with the observationthat the nature of the carbohydrate attached to UL18 (complex orhigh-mannose carbohydrates) does not affect binding to LIR-1 (ref.8). A more likely explanation for increased affinity for the viralhomolog is a more favorable interaction of LIR-1 with amino acids onthe α3 domain of UL18 than the comparable region of MHC class Iproteins. Comparison of the UL18 sequence in the region analogous tothe LIR-1-contacting residues in HLA-A2 (Fig. 3a) shows conservation

916 VOLUME 4 NUMBER 9 SEPTEMBER 2003 NATURE IMMUNOLOGY

Figure 4 Implications of UL18−LIR-1 interaction. (a) View of D1D2−HLA-A2structure with the approximate positions of potential UL18 glycosylation sitesmapped onto the HLA-A2 structure. Potential N-linked glycosylation sites(pink spheres) are distant from the LIR-1 binding site. The closest predictedO-linked glycosylation site (gray sphere) is also distant from the binding site.Three additional O-linked glycosylation sites are predicted in UL18 in aregion corresponding to sequence between the carboxyl terminus of the HLA-A2 ectodomain and the transmembrane region. (b) Closer view of the LIR-1−HLA-A2 interface in the region outlined by the dotted rectangle in a. Thestructures of free LIR-1 (ref. 11) and LIR-2 (ref. 17) are superimposed on the bound LIR-1 structure to show movement of the loop of residues 76–84(mostly disordered in the bound LIR-1 structure (see Methods). The sidechains of LIR-1 Asp80 (which have been linked to the binding of LIR-1UL18; ref. 11) and its counterpart in LIR-2, Arg80, are shown as ball-and-stick representation.

a

b

Figure 3 Amino acid contacts at the LIR-1–MHC class I interface. (a) Comparison of the LIR-1 bindingepitope (residues in color) on the HLA α3 domain across classical and nonclassical MHC class I, UL18and the MHC-like molecules FcRn, HFE and ZAG. Only MHC class I homolog proteins that have beentested for binding to LIR-1 are included. Red, blue and gold indicate nonconservative substitution,conservative substitution and conservation of the consensus sequence (purple), respectively. (b) Conservation of α3- and β2M-contacting residues across the LIR family. Receptor function: i,inhibitory; +, activatory. s, soluble receptor. Colors of amino acid alterations from the LIR-1 referencesequence are as in a, with deletions marked in red.

a

b

Table 2 Amino acid contacts at the LIR-1−MHC class I interface

LIR-1 D1 HLA-A2 α3

Tyr36 Asp196

Tyr38 Val194, Ser195

Arg39 Val194

Lys41 Val194, Glu198, Val248

Thr43 Ala193

Tyr76 Asp196

LIR-1 D1 β2M

Gln18 Gln89

Lys42 Asp96

Trp67 Lys91, Ile92, Val93

Glu68 Lys94

Gly97 Ser88

Ala98 Leu87, Ser88

LIR-1 D2 β2M

Tyr99 Thr4, Thr86, Lys91

Ile100 Val85, Thr86

Gln125 Ile1, Gln2

Val126 Gln2, Arg3, Thr4, Thr86

Ala127 Gln2, Arg3

Asp184 Lys91

Leu187 Ser88

Amino acid contacts (≤4.0 Å) between LIR-1 D1 andHLA-A2 α3 domain (top); LIR-1 D1 and β2M (middle);and LIR-1 D2 and β2M (bottom).

©20

03 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

reim

mu

no

log

y

A RT I C L E S

of Asp196, a residue also conserved throughout classical and nonclas-sical MHC class I molecules, but substituted nonconservatively inFcRn, HFE and ZAG (Fig. 3a). In HLA-A2, this residue makes contactswith LIR-1 Tyr76, a residue that has been linked to the interaction withUL18 (ref. 11). The remaining five residues within the UL18 counter-part of the LIR-1 contact region on the HLA-A2 α3 domain (residues193–196, 198 and 248) differ between the UL18 and MHC class Isequences. Two substitutions are conservative (Ser to Asn at 195; Val toAla at 248), whereas three are nonconservative (Pro to Asn at 193; Ileto Gln at 194; Glu to Arg at 198). These differences, perhaps togetherwith an optimal orientation of the UL18 α3 and β2M domains, maycontribute to the increased affinity of UL18 for LIR-1.

UL18 is highly selective for LIR-1, binding over 3,000-fold morestrongly then to LIR-2 (ref.11). The crystal structure of LIR-2 D1D2identified conformational alterations relative to LIR-1 in both estab-lished elements of the UL18 binding site: an alternative rotamer con-formation of Tyr38, and an 11-Å shift in the loop of residues 76–84(ref. 17). When the structures of LIR-1 D1 and LIR-2 D1 not in com-plexes are superimposed onto LIR-1 D1 in the LIR-1−HLA-A2 com-plex structure (Fig. 4b and Methods), the LIR-1 loop of residues 76–84(residues 78–83 are disordered in the LIR-1−HLA-A2 structure) is ori-ented toward the α3 domain. However, the main chain conformation

of the LIR-2 loop (which contains a one-residue deletion and twoglycine-to-nonglycine substitutions relative to LIR-1) is shifted 11 Åaway from the contact interface, preventing interaction of residues79–82 with the MHC class I (here representing UL18; Fig. 4b). In addi-tion, the three residues in the D1 loop of residues 76–84 that have beenlinked to UL18−LIR-1 binding (Tyr76, Asp80 and Arg84) are substi-tuted nonconservatively in LIR-2 (to Gln76, Arg80 and Trp83)11.Consequently, even assuming considerable flexibility in this region ofLIR-2, UL18 may interact preferentially with LIR-1 side chains at thesepositions, as indicated by the demonstration that a LIR-2 D1D2 pro-tein containing LIR-1 residues at these positions bound UL18 ∼ 10-foldmore tightly than did wild-type LIR-2 D1D2 (ref. 11). Differences inthe LIR-2 region of the loop of residues 76–84 may also provide anexplanation for the slightly higher (three- to fourfold) affinities forLIR-1 binding to HLA-A, HLA-B, and HLA-C molecules comparedwith that for LIR-2 (ref. 9).

Comparison of the LIR-1 residues that interact with the HLA-A2β2M domain with other LIR receptors separates the LIR family mem-bers into two groups (Fig. 3b). Group 1 members, which include theinhibitory receptors LIR-1 and LIR-2, the soluble receptor LIR-4, andthe activatory receptors LIR-6a, LIR-6b and LIR-7, show high conser-vation (≥10 of 13) of these residues. Group 2 members, which com-prise LIR-3, LIR-5 and LIR-8 (all of which are likely to transmitinhibitory signals), and ILT7, ILT8 and ILT11 (not represented in theLIR family), show poor conservation (≤4 of 13) in which >85% of thechanges are either nonconservative substitutions or deletions. Similarresults are obtained by a comparison of the HLA-A2 α3 domain–contacting residues conserved or substituted conservatively betweenLIR-1 and LIR-2 (residues 36, 38, 39, 41 and 43). Thus, the ligands ofother group 1 members are likely to include MHC class I or MHC classI-like molecules noncovalently associated with β2M, as alreadydemonstrated for LIR-2 and LIR-6 (refs. 1,2,18). In contrast, group 2members LIR-3, LIR-5 and LIR-8 seem unlikely to engage MHC class Iproteins using a binding mode similar to that of LIR-1, and most prob-ably engage a different set of ligands.

Specificity for a broad range of MHC class I molecules distinguishesLIR-1 and LIR-2 from KIR ligand recognition, and indicates that thestrength of LIR-1 and LIR-2 signals may reflect the overall expressionof MHC class I on the target cell. How LIR-HLA binding leads to signalinitiation, and how such signals integrate with activatory signals prop-agated at the cell surface, are key questions. No evidence of ligand-induced oligomerization is apparent from an inspection of crystalcontacts, and alterations in the D1D2 angle may reflect adoption of anideal orientation for binding rather than a signaling mechanism.

NATURE IMMUNOLOGY VOLUME 4 NUMBER 9 SEPTEMBER 2003 917

Figure 5 LIR-1−MHC class I interactions at the cell surface. Comparison of MHC class Icomplexes with KIR2DL1 (ref. 14), the B7TCR40, LIR-1 (including homology modelsrepresenting the structures of D3 and D4; ref. 8)and CD8 (ref. 15). Horizontal dashed linesindicate the approximate locations of cellsurfaces to which the full-length forms of theproteins are attached. The LIR-1−MHC class Iinteraction is depicted in trans (as explained intext). Dotted lines, mucin-like regions of theCD8αα ectodomains; solid horizontal lines,interchain disulfide bonds. The structure of aMHC class I complex with a KIR that containsthree immunoglobulin-like domains has not been determined, but would be expected to result in a larger intermembrane distance thanthe KIR2DL1−HLA-C interaction indicated.

Figure 6 Comparison of the ligand-binding sites on the structures of KIR2DL1(ref. 14), LIR-1, and FcαRI (ref. 23). The positions of residues within 4 Å ofthe binding partner (an MHC class I molecule for KIR2DL1 and LIR-1, and theFc region of IgA for FcαRI) are indicated as yellow (binding site within theD1D2 interdomain hinge) or blue (binding site within D1) spheres. The LIR-176-84 loop region is indicated in magenta; residues associated with binding toUL18 (Asp80 and Arg84) are also marked as blue spheres.

©20

03 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

reim

mu

no

log

y

A RT I C L E S

Alternatively, signaling may result from the LIR-1−MHC complexes’cosegregating with activatory ligand-receptor interactions in areas ofclose contact during cell-cell interaction, thereby recruiting phos-phatases to an otherwise highly activatory environment19. Indeed,LIR-1 and TCR colocalize at the ‘immunological synapse’ formedbetween T cells and APCs expressing TCR and LIR-1 ligands20. Thelengths of protein–protein interactions determining the spacing of theopposing cell membranes are essential parameters in such signalingmodels19. Notably, the length of the LIR-1−A2 interaction derivedfrom the crystal structure and a homology model of the LIR-1 D3 andD4 domains11 indicate that cosegregation with TCR-MHC or KIR-MHC complexes in such ‘close-contact zones’ at the immunologicalsynapse is likely (Fig. 5).

The LIR-1 D1D2−HLA-A2 structure shows no contacts with or con-formational changes in the HLA-A2 α1-α2 peptide-binding platform;thus, simultaneous interaction of LIR-1 D1D2 and either a KIR or aTCR with a single MHC class I molecule should also be possible inprinciple. Either might be physiologically relevant, as LIR-1 is coex-pressed at the cell surface with TCRs on MHC class I–restricted CD8+

T cells and with KIRs on subsets of natural killer cells21. In contrast,simultaneous binding of LIR-1 and the T-cell coreceptors CD8αα orCD8 αβ, which contact MHC class I α3 and β2M domains15, wouldnot be possible. Although the CD8αα and LIR-1 binding sites onMHC class I are mainly nonoverlapping, steric effects would excludebinding of both LIR-1 and CD8 to the same MHC class I molecule(Fig. 5), as confirmed by binding studies9. Thus, LIR-1 may inhibitactivation signals on MHC class I–restricted T cells by competing withCD8 for binding to MHC class I complexes engaged by the TCR, withthe dual effect of preventing stimulatory signals transmitted by CD8and of recruiting SHP-1 phosphatase to the vicinity of the TCR,thereby decreasing the half-lives of phosphorylated signaling compo-nents. In this way, LIR-1 could potentially act as a potent ‘negativecoreceptor’ on MHC class I–restricted T cells or natural killer cellsexpressing LIR-1.

LIR proteins are encoded within the leukocyte receptor cluster, aregion of human chromosome 19 that also includes the genes for KIRsand the IgA receptor FcαRI (ref. 22). These genes have probablydiverged from a common ancestor, resulting in receptor families thatinteract in different ways with structurally diverse ligands. LIR-1 andKIRs use the D1-D2 interdomain hinge to interact with MHC class Imolecules, but only LIR-1 uses an additional binding surface locatedtoward the membrane-distal tip of D1 (Fig. 6). FcαRI also uses the tipof its D1 domain to bind to the Fc portion of IgA, but does not use theD1-D2 interdomain hinge region used by LIR-1, KIRs and other Fcreceptors encoded outside of the leukocyte receptor cluster, such as theFcγR and FcεRI proteins23 (Fig. 6).

The use of two binding surfaces on LIR-1 establishes a molecular linkto the recognition surfaces used by KIRs and FcαRI in ligand binding,and raises questions as to the nature of the evolutionary relationshipsbetween these immunoreceptors. Based on gene structure and the exis-tence of orthologous murine receptors (the paired immunoglobulin-like receptors24), it seems likely LIRs evolved before KIRs. Thehomology of chicken immunoglobulin-like receptors to LIRs andpaired immunoglobulin-like receptors supports the idea of the exis-tence of a common ancestor pre-dating the separation of bird andmammalian lineages25. In contrast, several features of the KIRs, includ-ing the lack of rodent orthologs, high similarity between different KIRloci and differences between chimpanzee and human KIR sequences,indicate a more rapid and recent evolution, restricted to primates25–27.In common with KIRs, FcαRI lacks a murine ortholog, favoring theidea that it originated more recently than the LIRs. These arguments, in

combination with the current structural data, are consistent with therecent proposal that KIR and FcαRI genes evolved from recombina-tions, duplications and shuffling events involving ancestral LIR genes25.The presence of two (rather than one) binding surfaces on LIR-1 maytherefore be characteristic of a more ancient receptor, which subse-quently diverged into KIRs, preserving the use of the D1-D2 interfaceregion, FcαRI, which uses a more extensive D1 membrane-distal bind-ing site, and LIR-1, preserving both interaction sites.

METHODSRecombinant protein production. The HLA-A2 complex (comprising residues1–276 of the mature A2 heavy chain, noncovalently associated with β2M and anonamer peptide (ILKEPVHGV) derived from human immunodeficiencyvirus 1 Pol) and LIR-1 D1D2 (residues 1–198 of the mature protein) were pro-duced using existing methods involving expression in Escherichia coli and dilu-tion refolding28. Renatured LIR-1 and HLA-A2 complex were concentratedseparately, and were purified by size-exclusion chromatography using aSuperdex 75 column. The purified proteins were concentrated and quantifiedby absorption measurements at 280 nm. Extinction coefficients (280 nm) of66,150 M–1cm–1 (HLA-A2 complex) and 48,275 M–1cm–1 (LIR-1) were calcu-lated using amino acid analysis and as described29.

Crystallization, data collection and processing. Hanging-drop crystallizationtrials were done using a 1:1 stoichiometric mixture of purified LIR-1 and HLA-A2 complex (14.5 mg/ml total protein concentration). Microcrystals were ini-tially obtained over a period of 3 weeks in 0.2 M sodium acetate, 0.1 M Tris, pH 8.5, and 30% weight/volume polyethylene glycol 4000. Subsequent additiveand detergent screens resulted in growth of optimized crystals in 0.2 M sodiumacetate, 0.1 M Tris, pH 8.5, 30% weight/volume polyethylene glycol 4000, 20 mM L-cysteine and 1.8 mM Triton X-100. Crystals were transferred to a col-lection buffer consisting of 0.2 M sodium acetate, 0.1 M Tris, pH 8.5, and 31%weight/volume polyethylene glycol 4000 supplemented with increasing con-centrations of ispropanol to a final concentration of 7.5%. Data were collectedfrom cryopreserved crystals at a temperature of 100 K at 0.992 Å at beamline0.3 at the Advanced Light Source, Berkeley National Laboratory (Berkeley,California). Crystals belonged to the space group P3121, with unit cell dimen-sions a = b = 113.74 Å, and c = 89.46 Å, and contain one HLA-A2−LIR-1 com-plex per asymmetric unit. Data were auto-indexed and integrated using theprogram DENZO, and scaled using the program SCALEPACK30.

Structure solution, refinement and analysis. The structure was determined bymolecular replacement using the program AmoRe31 and the coordinates of LIR-1 D1D2 and HLA-A2. Unambiguous solutions were found in the cross-rotation and translation functions for HLA-A2 and LIR-1 (Rcryst = 41%; Rfree =43%) for data between 20 Å and 4.0 Å. After four-domain rigid-body refinementwith REFMAC5, as implemented in the CCP4 program suite32, rebuilding wasaccomplished with the program O (ref. 33) using 2Fo – Fc annealed omit maps(Fig. 1b), alternating with reciprocal space refinement in the crystallographyand nuclear magnetic resonance system (CNS)34. Final rounds of simulatedannealing refinement and subsequently B factor refinement using grouped tem-perature factors in CNS34 resulted in a final Rcryst of 22.2% (Rfree = 31.0%) for alldata between 20 Å and 3.4 Å (Table 1 and Supplementary Fig. 1 online). TheHLA-A2−LIR-1 complex model consists of residues 1–276 of the HLA-A2 heavychain; peptide residues 1–9; β2M residues 1–16, 21–73 and 76–99; and LIR-1residues 4–27, 32–77, 84–138 and 141–198. The side chains of residues 17, 82and 268 of HLA-A2; 41 and 94 of β2M; and 33, 34, 52, 53, 56, 57, 84, 86 and 87 ofLIR-1 were disordered and modeled as alanine residues. Disulfide bonds arefound between LIR-1 residues 26 and 75, 122 and 174, and 134 and 144; HLA-A2 residues 101 and 164, and 203 and 259; and β2M residues 25 and 80. Foranalysis of interdomain angles, contacts and buried surface areas, D1 is definedas residues 1–98 and D2 is defined as residues 99–198. Interdomain contactresidues were identified using the program CONTACT32, and were defined asresidues containing an atom of ≤4.0 Å of the partner domain. Buried surfaceareas were calculated using SURFACE32 with a 1.4-Å probe radius. Interdomainangles were calculated using the program Dom_angle35, which determines theangle between the long axes of adjacent domains that are approximated by

918 VOLUME 4 NUMBER 9 SEPTEMBER 2003 NATURE IMMUNOLOGY

©20

03 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

reim

mu

no

log

y

A RT I C L E S

ellipsoids calculated from the coordinates. Molscript36, Raster3D37, Grasp38 andPyMOL39 were used to prepare Figures 1–5.

Accession numbers. Coordinates of the structure have been deposited with theProtein Data Bank under accession code 1P7Q. Protein Data Bank accessioncodes: LIR-1 D1D2, 1G0X; HLA-A2, 1AKJ.

Note: Supplementary information is available on the Nature Immunology website.

ACKNOWLEDGMENTSWe thank members of the Bjorkman laboratory for technical assistance, and C. O’Callaghan and A. van der Merwe for critical reading of the manuscript.B.E.W. was supported by a Wellcome Trust Travelling Fellowship and is nowfunded by a Medical Research Council Career Development Award.

COMPETING INTERESTS STATEMENTThe authors declare that they have no competing financial interests.

Received 22 April; accepted 9 July 2003Published online at http://www.nature.com/natureimmunology/

1. Borges, L. & Cosman, D. LIRs/ILTs/MIRs, inhibitory and stimulatory Ig-superfamilyreceptors expressed in myeloid and lymphoid cells. Cytokine Growth Factor Rev. 11,209–217 (2000).

2. Cella, M., Nakajima, H., Facchetti, F., Hoffmann, T. & Colonna, M. ILT receptors atthe interface between lymphoid and myeloid cells. Curr. Top. Microbiol. Immunol.251, 161–166 (2000).

3. Colonna, M. et al. A common inhibitory receptor for major histocompatibility complexclass I molecules on human lymphoid and myelomonocytic cells. J. Exp. Med. 186,1809–1818 (1997).

4. Cosman, D. et al. A novel immunoglobulin superfamily receptor for cellular and viralMHC class I molecules. Immunity 7, 273–282 (1997).

5. Beck, S. & Barrell, B.G. Human cytomegalovirus encodes a glycoprotein homologousto MHC class-I antigens. Nature 331, 269–272 (1988).

6. Browne, H., Smith, G., Beck, S. & Minson, T. A complex between the MHC class Ihomologue encoded by human cytomegalovirus and β2 microglobulin. Nature 347,770–772 (1990).

7. Fahnestock, M.L. et al. The MHC class I homolog encoded by human cytomegalovirusbinds endogenous peptides. Immunity 3, 583–590 (1995).

8. Chapman, T.L., Heikema, A.P. & Bjorkman, P.J. The inhibitory receptor LIR-1 uses acommon binding interaction to recognize class I MHC molecules and the viral MHChomolog UL18. Immunity 11, 603–611 (1999).

9. Shiroishi, M. et al. Human inhibitory receptors ILT2 and ILT4 compete with CD8 forMHC class I binding and bind preferentially to HLA-G. Proc. Natl. Acad. Sci. USA100, 8856–8861 (2003).

10. Natarajan, K., Dimasi, N., Wang, J., Mariuzza, R.A. & Margulies, D.H. Structure andfunction of natural killer cell receptors: multiple molecular solutions to self, nonselfdiscrimination. Annu. Rev. Immunol. 20, 853–885 (2002).

11. Chapman, T.L., Heikema, A.P., West, A.P., Jr. & Bjorkman, P.J. Crystal structure andligand binding properties of the D1D2 region of the inhibitory receptor LIR-1 (ILT2).Immunity 13, 727–736 (2000).

12. Madden, D.R., Garboczi, D.N. & Wiley, D.C. The antigenic identity of peptide-MHCcomplexes: a comparison of the conformations of five viral peptides presented byHLA-A2. Cell 75, 693–708 (1993).

13. Rudolph, M.G. & Wilson, I.A. The specificity of TCR/pMHC interaction. Curr. Opin.Immunol. 14, 52–65 (2002).

14. Fan, Q.R., Long, E.O. & Wiley, D.C. Crystal structure of the human natural killer cellinhibitory receptor KIR2DL1-HLA-Cw4 complex. Nat. Immunol. 2, 452–460 (2001).

15. Gao, G.F. et al. Crystal structure of the complex between human CD8α(α) and HLA-A2. Nature 387, 630–634 (1997).

16. Cosman, D., Fanger, N. & Borges, L. Human cytomegalovirus, MHC class I andinhibitory signalling receptors: more questions than answers. Immunol. Rev. 168,177–185 (1999).

17. Willcox, B.E. et al. Crystal structure of LIR-2 (ILT4) at 1.8 Å: differences from LIR-1(ILT2) in regions implicated in the binding of the human cytomegalovirus class I MHChomolog UL18. BMC. Struct. Biol. 2, 6 (2002).

18. Allen, R.L., Raine, T., Haude, A., Trowsdale, J. & Wilson, M.J. Leukocyte receptorcomplex-encoded immunomodulatory receptors show differing specificity for alterna-tive HLA-B27 structures. J. Immunol.167, 5543–5547 (2001).

19. van der Merwe, P.A., Davis, S.J., Shaw, A.S. & Dustin, M.L. Cytoskeletal polarizationand redistribution of cell-surface molecules during T cell antigen recognition. Semin.Immunol. 12, 5–21 (2000).

20. Dietrich, J., Cella, M. & Colonna, M. Ig-like transcript 2 (ILT2)/leukocyte Ig-likereceptor 1 (LIR1) inhibits TCR signaling and actin cytoskeleton reorganization. J. Immunol. 166, 2514–2521 (2001).

21. Young, N.T., Uhrberg, M., Phillips, J.H., Lanier, L.L. & Parham, P. Differential expres-sion of leukocyte receptor complex-encoded Ig-like receptors correlates with the tran-sition from effector to memory CTL. J. Immunol. 166, 3933–3941 (2001).

22. Wende, H., Colonna, M., Ziegler, A. & Volz, A. Organization of the leukocyte receptorcluster (LRC) on human chromosome 19q13.4. Mamm. Genome 10, 154–160 (1999).

23. Herr, A.B., Ballister, E.R. & Bjorkman, P.J. Insights into mucosal immunity fromthe structures of human FcaRI and its complex with IgA1-Fc. Nature 423,614–620 (2003).

24. Kubagawa, H., Burrows, P.D. & Cooper, M.D. A novel pair of immunoglobulin-likereceptors expressed by B cells and myeloid cells. Proc. Natl. Acad. Sci. USA 94,5261–5266 (1997).

25. Martin, A.M., Kulski, J.K., Witt, C., Pontarotti, P. & Christiansen, F.T. Leukocyte Ig-like receptor complex (LRC) in mice and men. Trends Immunol. 23, 81–88 (2002).

26. Barten, R., Torkar, M., Haude, A., Trowsdale, J. & Wilson, M.J. Divergent and conver-gent evolution of NK-cell receptors. Trends Immunol. 22, 52–57 (2001).

27. Trowsdale, J. Genetic and functional relationships between MHC and NK receptorgenes. Immunity 15, 363–374 (2001).

28. Garboczi, D.N., Hung, D.T. & Wiley, D.C. HLA-A2-peptide complexes: refolding andcrystallization of molecules expressed in Escherichia coli and complexed with singleantigenic peptides. Proc. Natl. Acad. Sci. USA 89, 3429–3433 (1992).

29. Pace, C.N., Vajdos, F., Fee, L., Grimsley, G. & Gray, T. How to measure and predict themolar absorption coefficient of a protein. Protein Sci. 4, 2411–2423 (1995).

30. Otwinowski, Z. & Minor, W. Processing of X-ray diffraction data collected in oscillationmode. Methods Enzymol. 276, 307–326 (1997).

31. Navaza, J. AMORE—an automated package for molecular replacement. ActaCrystallogr. A 50, 157–163 (1994).

32. CCP4. The CCP4 suite: programs for protein crystallography. Acta Crystallogr. D 50,760–763 (1994).

33. Jones, T.A., Zou, J.Y., Cowan, S.W. & Kjeldgaard. Improved methods for building pro-tein models in electron density maps and the location of errors in these models. ActaCrystallogr. A 47, 110–119 (1991).

34. Brunger, A.T. et al. Crystallography & NMR system: A new software suite for macro-molecular structure determination. Acta Crystallogr. D. Biol. Crystallogr. 54,905–921 (1998).

35. Su, X.D. et al. Crystal structure of hemolin: a horseshoe shape with implications forhomophilic adhesion. Science 281, 991–995 (1998).

36. Kraulis, P.J. MOLSCRIPT: a program to produce both detailed and schematic plots ofprotein structures. J. Appl. Crystallogr. 24, 946–950 (1991).

37. Merritt, E.A. & Murphy, M.E.P. Raster3D Version 2.0, a program for photorealisticmolecular graphics. Acta Crystallogr. D 50, 869–873 (1994).

38. Nicholls, A., Sharp, K.A. & Honig, B. Protein folding and association: insights fromthe interfacial and thermodynamic properties of hydrocarbons. Proteins 11,281–296 (1991).

39. DeLano, W.L. The PyMOL Molecular Graphics System. (DeLano Scientific, SanCarlos, California, 2002).

40. Ding, Y.H. et al. Two human T cell receptors bind in a similar diagonal mode to theHLA-A2/Tax peptide complex using different TCR amino acids. Immunity 8,403–411 (1998).

NATURE IMMUNOLOGY VOLUME 4 NUMBER 9 SEPTEMBER 2003 919

©20

03 N

atu

re P

ub

lish

ing

Gro

up

h

ttp

://w

ww

.nat

ure

.co

m/n

atu

reim

mu

no

log

y