Proc. Natl. Acad. Sci. USAVol. 90, pp. 6444-6448, July 1993Biophysics

"Diabodies": Small bivalent and bispecific antibody fragments(bacterial expression/phage display/dyad/surface plasmon resonance)

PHILIPP HOLLIGER*, TERENCE PROSPERO*, AND GREG WINTER*t*Medical Research Council Centre for Protein Engineering, Hills Road, Cambridge CB2 2QH, United Kingdom; and tMedical Research Council Laboratory ofMolecular Biology, Hills Road, Cambridge CB2 2QH, United Kingdom

Communicated by M. F. Perutz, March IS, 1993 (receivedfor review January 15, 1993)

ABSTRACT Bivalent and bispecific antibodies and theirfragments have immense potential for practical application.Here we describe the design of small antibody fragments withtwo antigen-binding sites. The fragments comprise a heavy-chain variable domain (VH) connected to a light-chain variabledomain (VL) on the same polypeptide chain (VH-VL). By usinga linker that is too short to allow pairing between the twodomains on the same chain, the domains are forced to pair withthe complementary domains of another chain and create twoantigen-binding sites. As indicated by a computer graphicmodel ofthe dimers, the two pairs ofdomains can pack togetherwith the antigen-binding sites pointing in opposite directions.The dimeric antibody fragments, or "diabodies," can bedesigned for bivalent or bispeciflc interactions. Starting fromthe monoclonal antibodies NQ11.7.22 (NQ11) and D1.3 di-rected against the hapten phenyloxazolone and hen egg lyso-zyme, respectively, we built bivalent fragments (VHNQ1l-VLNQ11)2 and (VHD1.3-VLD1.3)2 and bispecific fragmentsVHNQ11-VLD1.3 and VHD1.3-VLNQ11. The fragments wereexpressed by secretion from bacteria and shown to bindspecifically to the hapten and/or antigen. Those with 5- and15-residue linkers had similar binding afmities to the parentantibodies, but a fragment with the VH domain joined directlyto the VL domain was found to have slower dissociation kineticsand an improved affinity for hapten. Diabodies offer a readymeans of constructing small bivalent and bispecific antibodyfragments in bacteria.

Bivalent and bispecific antibodies have many practical ap-plications, including immunodiagnosis and therapy (1). Biva-lency can allow antibodies to bind to multimeric antigen withgreat avidity; bispecificity can allow the cross-linking of twoantigens-for example, in recruiting cytotoxic T cells tomediate killing of a tumor cell (2). Bivalent (IgG) antibodieshave been derived from hybridomas (3), and bispecific anti-bodies by fusion of two hybridomas with two differentspecificities (4). However, fragments are often preferable tocomplete antibodies, as the Fc region of antibodies can leadto illegitimate targeting to cells expressing Fc receptors (5).Antibody fragments are readily produced by gene technol-

ogy: the genes encoding antibody variable domains can bederived from hybridomas (6) or from filamentous bacterio-phage displaying antibody fragments (7) (for reviews, seerefs. 8 and 9). Recombinant Fab and Fv fragments of anti-bodies can be secreted from bacteria (10, 11) by coexpressionof the two chains comprising the heavy- (VH) and light- (VL)chain variable domains. Alternatively the VH and VL do-mains can be linked on the same polypeptide chain with aflexible spacer stretching between the C terminus of onedomain to the N terminus of the other to create single-chainFv (scFv) fragments (12, 13). However, Fv, scFv, and Fabfragments each carry a single antigen-binding site.

The publication costs of this article were defrayed in part by page chargepayment. This article must therefore be hereby marked "advertisement"in accordance with 18 U.S.C. §1734 solely to indicate this fact.

Recombinant fragments with two binding sites have beenmade in several ways-for example, by chemical cross-linking of the hinge cysteine residues (14) or by including aC-terminal peptide that promotes dimerization (15, 16). Re-cently we noticed that scFv fragments secreted from bacteria(17) were often present as both monomers and dimers (18),suggesting that the VH and VL domains ofdifferent chains canpair. In turn, this indicated a simple way of making smallbispecific antibody fragments that assemble in vivo and canbe harvested directly from culture supernatant. By linking theVH and VL of two different antibodies A and B to form twodifferent "cross-over" chains VHA-VLB and VHB-VLA, wewould expect the chains to recreate both antigen-binding siteson association. To explore this we used the variable-domain-encoding genes of the mouse hybridomas NQ11.7.22 (NQ11)[anti-2-phenyloxazol-5-one (phOx)], and D1.3 [anti-hen egglysozyme (HEL)] to make a range of constructs for expres-sion ofbivalent and bispecific antibody fragments in bacteria.

MATERIALS AND METHODSVector Construction. Escherichia coli TG1 (19) was used

for propagation of plasmids and expression of antibodyfragments. See Fig. 1 for schematic of most constructs(except for constructs VIII and IX). Constructs II, IV, and Vencoding the scFv (7) and Fv (23) fragments of the mono-clonal antibody D1.3 directed against HEL (26) were pro-vided by T. Simon [Medical Research Council (MRC)-Centrefor Protein Engineering]. Construct III, encoding the Fvfragment of the mouse hybridoma NQ11 directed against thehapten phOx (27) was provided by L. Riechmann (MRC-Laboratory of Molecular Biology). Construct III and thosederived from it (constructs I, VI, VII-IX) include a mutationin the VH domain (Asp-97 -+ Ala) (21).

Further constructs (Fig. 1) were derived by religation ofrestriction fragments (here termed vector fragments and com-prising vector backbone and variable-domain genes or parts ofvariable-domain genes) or their ligation with other restrictionfragments (here termed PCR fragments and comprising vari-able-domain genes, linker, or noncoding sequences) usingstandard methods (28). Vector fragments were provided bydigestion of construct III with BstEII/Sac I, construct II withPst I/Sac I, constructs VI with BstEII or Xho I or Pst I, andthe phagemid pHEN1 (20) with HindIII and Xho I.PCR fragments for the 15-residue linkers were generated

by using primers 1 and 2 with construct IV. Digestion withAsc I and Sac I yielded PCR fragment a-15 comprising thegene 3 leader, the VH segment, and linker. Digestion withBstEII and Asc I yielded PCR fragment b-15 comprisinglinker and the VL segment. PCR fragments for the shorterlinkers were generated by using primer 1 with either primers3 or 4 or 5 (encoding no, 5-, or 10-residue linkers, respec-

Abbreviations: HEL, hen egg lysozyme; phOx, 2-phenyloxazol-5-one; NQ11, hybridoma NQ11.7.22; VH, immunoglobulin heavy-chain variable domain; VL, immunoglobulin light-chain variabledomain; BSA, bovine serum albumin; FPLC, fast protein liquidchromatography; scFv, single-chain Fv.

6444

Proc. Natl. Acad. Sci. USA 90 (1993) 6445

-II

scFv fragments and bivalent fragments

LI VHA L LI P LTn=l, 2,3 rbsl

II p L VHB Lk VLB TII DV 1~~~~~~~~~~~~~~~~n=l, 3 rbs I --

Pstl Sacl

VLA

VHA Lk

bispecific_fragments ..SS(G4S)nDI.. ..SS(G4S.D_..

VI LkLpLI V V L2 VHB LkV T VI VHB VHArbsTIr FfV LB VLA

nf=O, 1, 2,3 HindIl PstI BstEIl SacI Xhol PstIBstEIISacI Pstl XholMcI rbs2

..CGG TAA TAA ggcgcgccacaatttcacact aaggag gtttaactt GTG..stop stop Start

three chain fragments ..VSS(G4)L,D..BLkVG4n I.

VHVII p LI VHA LI VLB L2 V VLA T VHBVLA

n=l, 2NI N

FIG. 1. Vector constructs. Constructs (III-V) used to build the expression constructs (I, II, VI, and VII) are illustrated (constructs VIII andIX encoding VHB-VLA and VHA-VLB, respectively, are not shown but are described in text). Constructs III-V were cloned in fd-tet-DOG1(V2) (20), and the other constructs were cloned in a pUC19-derived expression vector (V1) (21) under control of the lac promoter (P) and witha peptide tag (T) (22). The variable-domain genes of antibodies NQii (VHA, VLA) and D1.3 (VHB, VLB) have internal Pst I, BstEII, Sac I,

and Xho I restriction sites as in refs. 6 and 23: in addition, VLA has an internal Pst I site. The polypeptide linker (Lk) or noncoding region (Nior N2) between VH and VL of the same chain is marked. Lk consists of a number of repeats (n = 0-3) of Gly-4 -- Ser (G4S). Ni includes a

ribosome-binding site (rbsl), as described (23), and N2 includes a consensus ribosome-binding site (rbs2), as described (24). The signal sequencesfor secretion of fragments into the bacterial periplasm are the pelB signal sequence (Li) (23) and the phage fd gene 3 signal sequence (L2) (25).Locations ofPCR primers 1-8 are shown in constructs III-V. Schematic depiction of protein products, as deduced from Table 1, is also given.

tively) and primer 2 with either primers 6 or 7 or 8 (alsoencoding no, 5-, or 10-residue linkers, respectively) anddigested as above to create PCR fragments a-0, a-S, a-10, b-0,b-5, and b-10. The PCR fragment (c) for providing thenoncoding sequence (Ni) in construct VII was generated byusing primers 1 and 2 with construct V and digested withBstEII and Asc I.

To make constructs VI, PCR fragments (a-0, a-5, a-10, anda-15) from construct IV were ligated with the correspondingPCR fragment b and the vector fragment III-BstEII/Sac I. Tomake constructs VII, PCR fragments (a-5 and a-10) fromconstruct IV were likewise ligated with fragment c. Con-structs VIII and IX (data not shown) were derived fromconstructs VI by partial digestion with Pst I or Xho I,respectively, and religation of the vector fragment. Con-structs I with 5-, 10-, and 15-residue linkers were likewisegenerated by BstEII digestion of the corresponding con-structs VI. Construct II with 5-residue linker was generatedby ligating fragment II-Pst I/Sac I with fragment a-5 that hadbeen cut with Pst I and Sac I. Construct VI (no linker) wasdigested with Hindlll and partially digested with Xho I andrecloned into the phagemid pHEN1 (20) for phage display.

Ligation mixes and transformation were as described (29).The linker or noncoding regions were sequenced by the dideox-ynucleotide method (30) using VENT(exo-) polymerase (NewEngland Biolabs). The nucleotide sequences of each of theconstructs are available from the authors on request.

Olgonucleoddes. Primer 1, 5'-GAC TCA TTC TCG ACTGAG CTC ACT TGG CGC GCC TTA TTA CCG TTT GATCTC GAG CTT GGT CCC introduces an Asc I site andprimes at the 3' VL; primer 2, 5'-GTC CTC GCA ACT GGCGCG CCA CAA TTT CAC AGTAAG GAG GTT TAA CTTGTG AAA AAA TTA TTA TTC GCA ATT introduces anAsc I site and a synthetic ribosome-binding site and primes atthe 5' phage fd gene 3 signal; primers 3-5, 5'-GAG CCA TCAATC GAT CTG GTC ACC GTC TCC TCA (GGC GGT GGCGGA TCG)n GAC ATT GAG CTC ACC CAG TCT CCAintroduce a linker of 0, 5, and 10 residues, respectively (n =

0,1,2), between VL and VH, include a BstEII site, and primeat 5' VL; primers 6-8, 5'-GAG CCA TCA ATC TCG GAGCTC GAT GTC (CGA TCC GCC ACC GCC) TGA GGAGAC GGT GAC CGT GGT CCC TTG GCC CC introduce alinker of 0, 5, 10 residues, respectively (n = 0,1,2), betweenVH and VL, include a Sac I site, and prime at 3' VH.

Expression of Bispecific Fragments. Fragments were pre-pared from periplasmic lysates (10) or from bacterial super-natants (11). For detection of fragments binding to HEL orphOx-bovine serum albumin (phOx-BSA), ELISA was doneas described (20). For sandwich-ELISA, plates were coatedwith HEL and periplasmic lysate or culture supernatantadded as above. Binding was detected by using 100 01 ofphOx-BSA at 1 pg/ml, followed by 100 M1 of the mouseantibody NQ22.18.7 at 100 ng/ml (31) directed against phOxand peroxidase-conjugated goat anti-mouse immunoglobulin(1:1000; Sigma). All incubations were made in phosphate-buffered saline containing 2% (wt/vol) skimmed milk pow-der. Immunoblots were as described (20). Fragments werepurified from bacterial supernatants by affinity chromatog-raphy on HEL-Sepharose (23) and/or phOx-BSA Sepharoseas described (32).

Binding Affinities. The dissociation kinetics of antibodyfragments to HEL and phOx were measured using plasmonsurface resonance with BlAcore (Pharmacia BiosensorAB),as described (18). For detection of fragments with bindingactivities to both HEL and phOx-BSA, 10 ,1 of periplasmiclysate or affinity-purified fragment was injected onto theHEL-coated sensor chip (see above) with a constant flow ofS ul/min and followed immediately by a pulse of 10 A1 ofphOx-BSA at 1 mg/ml. The binding affinity of fragments forphOx was measured by fluorescence quench titrations (33)with the hapten 4-raminobutyryl-phOx, as described (34).

RESULTS

We first made constructs for expression of bispecific andbivalent fragments oftwo different antibodies with 15-residue

Vectors for constructions

P L1 VHA NI LI VLA TII I' rb- .

i I IHindIll PstI BstEII Sacl PstI XhoI EcoRI

PCR b

Asci BstEII2 ILVHB 3,4,5__ VLB V2

IV .Y.L2 ~T6,7,8 ~T I.

_ _Sacd AscI

PCR a

Ascl2 VHB NP VLB V2

PCR c

Biophysics: HoUiger et aL

Proc. Natl. Acad. Sci. USA 90 (1993)

linkers between the VH and VL domains, as in scFv frag-ments. Fifteen residues are sufficient to stretch from the Cterminus of the VH domain to the N terminus of the VLdomain of the same chain (see Fig. 1 for schematic ofconstructs). However, we found that the chains can also pairwith each other, as the "bispecific" antibody construct VIwith coexpressed cross-over chains (VHA-VLB, VHB-VLA)yielded fragments that bound to both HEL and phOx-BSA(Table 1). By ELISA, the binding was specific: no bindingcould be detected to turkey egg lysozyme or BSA, and thebinding to phOx-BSA could be inhibited by adding solublehapten. By contrast, when the cross-over chains were ex-pressed separately (constructs VIII and IX), as expected,they did not bind to either HEL, turkey egg lysozyme, BSA,or phOx-BSA in ELISA; when mixed together in vitro,however, the two chains did associate as detected by bindingto HEL and to phOx-BSA and produced :10% of the ELISAsignal obtained when the two chains were coexpressed.We also used shorter linker lengths of5 and 10 residues: the

5-residue linker cannot stretch between VH and VL domainsofthe same chain, but it does allow the formation ofbispecificfragments (Table 1). Indeed, prompted by a computer graphicmodel of the dimers (see Discussion), we joined the Cterminus of the VH domain directly to the N terminus of theVL domain. Again the fragment bound to both antigens (Table1) and also when displayed on the surface of phage using thephagemid vector pHEN1 (20) (data not shown).To prove that both antigen-binding sites were located on

the same bispecific fragment, we used sandwich-ELISA(Table 1) and also BlAcore to show that the bispecificfragments that had bound to HEL also bound to phOx-BSA.For example, in Fig. 2B, the fragments were injected andbound to the biosensor chip coated with HEL (segment bc).After refractive index changes (segment cd) and some dis-sociation (segment de), the fragments were seen to bind toinjected phOx-BSA [the difference in resonance units (RU)between e and h reflects the amount ofphOx-BSA bound]. Ascontrol, FabDl.3 was shown to bind to HEL, but the boundfragments did not bind, in turn, to phOx-BSA (Fig. 2A).The sizes of the fragments binding to HEL or hapten were

shown to correspond to dimers by fast protein liquid chro-matography (FPLC). We loaded lysate containing the anti-body fragment on the FPLC column and passed the effluentover a BIAcore sensor chip coated with antigen or hapten.The binding of the fragments was detected (in real time) byan increase in mass at the surface of the chip (18). Thebispecific fragments were found to be similar in size to achimaeric D1.3 Fab fragment and bound to chips coated withphOx-BSA as well as to HEL-coated chips (Table 1). This

Table 1. Binding activities of monomer and dimer fragments, asdetermined by BIAcore and ELISA

FPLC BlAcore and ELISAConstruct Linkers oligomers phOx HEL Both*

I 5/15 Dt/Mt +/+II 5/15 Dt/M&Dt +/+ -/VI 0/5/10/15 All Dtt All + All + All +VII 5/10 Xtt/Xtt +/+ +/+ +1+VIII 5/15 Dt/Dt -/- -/-IX 5/15 Dt/Mt -/- -I-

M, mainly monomer; D, mainly dimer; M&D, both monomer anddimer clearly detectable; X, species running faster thanM (seen withconstructs VII) presumably due to equilibrium between the threechains.*As shown by sandwich-ELISA and also by BIAcore, binding toHEL, and then binding to phOx-BSA.tAs shown by FPLC and BlAcore of lysates by binding to antigen.*As shown by FPLC and BlAcore of lysates by binding to 9E10antibody.

e

c~~~~~b~~~~

iSLXLI

100 400 700

Time (s)

FIG. 2. BlAcore analysis of antibody fragments. Lysates fromFabD1.3 (A) are compared with lysates from construct VI (B) (with15-residue linker) using a sensorchip coated with HEL and detectingbound mass as resonance units (RU). a, Baseline; segment b-d,injection of fragment; segment d-e, dissociation of bound fragment;segment e-h, injection of phOx-BSA. ARU between e and h reflectsthe amount of phOx-BSA bound.

result indicated that these bispecific fragments must bedimers, formed by association of the two chains.The sizes of the fragments were also checked by binding of

the C-terminal myc tag of the fragments to the monoclonalantibody 9E10 immobilized on the biochip (Table 1). Thisformat detects fragnents irrespective of whether they haveantigen-binding activities and allowed us to show that the singlecross-over chains (constructs VIII and IX) (that do not bindantigen) were dimers with short linkers, but monomers withlong linkers. Again the bispecific fragnents emerged in a majorpeak corresponding to a dimer, irrespective of linker length.We purified bispecific fragments with 5- and 15-residue

linkers and also with no linker by affinity chromatography onHEL and then phOx-BSA-Sepharose columns. The yields ofthe fragments, as detected either by immunoblots usingantibody against the c-myc tag or by binding ELISA wereimproved when cells were transferred to lower temperature(20°C) after induction (35). After overnight fermentation,most ofthe protein was located in the culture supernatant andafter affinity purification yielded fragment at 0.3-1 mg perliter; the yield appears comparable to those reported for scFvor Fab fragments (17, 36).We measured the binding affinity for soluble 4-y-

aminobutyryl-phOx of bispecific fragments purified on HELby fluorescence quench titration. The results (Table 2) showthat the binding affinities of the bispecific fragments with 5-and 15-residue linkers are similar to the parent Fv. However,the fragment with no linker had a 10-fold improved affinity.We also measured the dissociation kinetics of the fragmentsfrom phOx-BSA and HEL by BlAcore. The dissociation ofthe fragment with no linker from phOx-BSA was at least10-fold slower (Table 2).We also constructed three-chain fragments (Fig. 1, con-

struct VII), in which a single-chain VHB-VLA is secretedwith the two complementary domains VHA and VLB. Thefragments were shown to bind both HEL and phOx (Table 1)and to be bispecific, but the chains appeared in fast equilib-rium, as on the FPLC the fragments emerged as a single peakbetween monomer and dimer in size; see ref. 39 for discus-sion and references.

DISCUSSIONHere we describe the design of bispecific antibody fragmentsby taking advantage of the intermolecular pairing of VH and

6446 Biophysics: HoUiger et aL

Proc. Natl. Acad. Sci. USA 90 (1993) 6447

Table 2. Binding affinities and dissociation kinetics (koff)Kd (nM) koff (s-1) koff (s1)

Construct for phOx for phOx for HELFvNQ11 (VH D97A) 99 ± 21* <0.05tIgG NQ11 110 10lFvD1.3 0.0028FabDl.3 0.0030IgG D1.3 0.0051§Fragment VI (liner 15) 110 ± 12 >0.04t 0.0028Fragment VI (linker 5) 106 + 18 >0.04t 0.0022Fragment VI (no linker) 8 + 1.5 0.004 0.001

Affinity constants (Kd) for phOx were determined by fluorescencequench, and dissociation kinetics (koff) for HEL and phOx weredetermined by BIAcore. kon values are not presented [the activefraction can vary between different preparations (18)]. koff values forFv, Fab, and IgG D1.3 on HEL disagree with BIAcore data pre-sented in ref. 37 for reasons to be discussed elsewhere.*Data are from ref. 21.tkoff was too fast to measure by BIAcore; given values are theslowest possible estimate.*Value of k0ff was measured by stop-flow in ref. 34.§Value of k0ff = kon x Kd was measured by stop-flow (kon) andfluorescence quench (Kd) in ref. 38.

VL domains. This result contrasts with the use of chemicalcrosslinking (14) or fusion to dimerization peptides (15, 16).We linked the VH domain of one antibody to the VL domainof another on the same polypeptide chain to create twochains, VHA-VLB and VHB-VLA, that are coexpressed inthe same cell and associate to form dimers with two antigen-binding sites on the same molecule. Although we wouldexpect heterodimers, homodimers, and monomers to beformed on cosecretion of both chains from the same bacte-rium, only the heterodimers bind to antigen and can beisolated by a single round of affinity chromatography.We chose two antibodies, D1.3 and NQ11, for which the

VH and VL domains are known to associate and form a stableFv fragment (21, 23). From the FPLC analysis of the bispe-cific fragments with 15-residue linkers, it appears that dimerspredominate in the lysates (Table 1). Presumably the favor-able interaction between the complementary domains on thetwo different chains helps drive dimer formation. However,to promote dimerization for antibodies for which the VH andVL domains are more weakly associated (10), for example inthe cross-over chains (constructs VIII and IX), we used shortlinkers to prevent the VH and VL domains on the same chainfrom pairing with each other (Table 1).

Short linkers impose major constraints on the ways bywhich the two chains can associate. We, therefore, attemptedto build a model of a dimer using the x-ray crystallographicstructure of the FabDl.3 fragment in complex with theantigen HEL (40). Using the computer graphics programFRODO (41), two molecules of the Fv portion were broughttogether such that the C terminus of each VH domain wasclosely apposed to the N terminus of each VL domain (witha C-N distance of =7 A) and easily satisfying the constraintsof a 5-residue linker. The surfaces of the two VH domainscould be docked together at the loops distal from the antigen-binding site, without bad contacts as detected by eye andconfirmed by the program PROLSQ (42). The antigen-bindingsites point in opposite directions, and the two sets of paireddomains are related by a dyad axis (Fig. 3). Inspection of thecrystallographic structures of other antibodies of theBrookhaven data base indicates they should also be capableofpacking in a similar manner (data not shown). The packingof immunoglobulin domains across the dyad axis (tail to tail)differs from that in antibody or immunoglobulin superfamilystructures (see ref. 43 for recent comparison).

Inspection ofthe model suggested that it might be possible tojoin the C terminus ofthe VH domain directly to the N terminus

FIG. 3. Modeling. Model of the bivalent D1.3 diabody illustratesthe apposition of the N and C termini of the linked domains (markedwith solid circles and connected by dotted line), the packing of thedomains, and the proximity of the C termini of the VL domains(marked with o). The Ca traces for VH and VL on the samepolypeptide chain (a or b) are marked in the same color (red or green)to show that each antigen-binding site is formed from the VH and VLdomains of the two different chains a and b. Thus VHa is associatedwith VLb, and VHb is associated with VLa. The bound antigen HELis marked in blue.

of the VL domain and dispense with the linker polypeptide.Although this would be expected to result in slight clashesbetween the two VH domains as they pack together, we envis-aged that the strain could be relieved by flexibility of thebackbone in these regions orby kinking the N-terminal /-strandof the VL domain. Indeed, the fragment with no linker provedto be dimeric and bispecific (Table 1), lending support to theproposed model. The binding affinity and dissociation kineticsof the fragment for phOx were altered, suggesting that theforced packing ofthe domains can lead to structural alterationsat the antigen-binding site and that it, indeed, might be possibleto build "diabodies" in which the antigen binding of the twobinding sites is cooperative (or anticooperative).

Biophysics: Holfger et al.

Proc. Natl. Acad. Sci. USA 90 (1993)

The structure of diabodies is compact and with shortlinkers should be rigid. Indeed, it might be possible tointroduce disulfide bonds or other contacts at the interface tomake more stable fragments. The lack offlexibility is unlikelyto compromise the cross-linking oftwo soluble antigens or ofa cell-surface antigen and a soluble antigen. However, forcross-linking of two cells, some flexibility of the surfaceantigens may be required. Longer linkers should allowgreater flexibility for the diabody heads but would also allowpairing within the same chain and formation of monomerfragments. Breaking one of the linkers entirely, as in thethree-chain fragment (Fig. 1), should give highly flexibleheads, but it may prove more difficult to keep the three chainstightly associated.

Diabodies represent a class of bivalent and bispecificantibody fragments similar in size to a Fab fragment. Theirsize should facilitate penetration of tumors and clearancefrom the serum. Diabodies could be derived from hybridomas(as above) or from variable-domain gene repertoires dis-played on phage (for review, see refs. 8 and 9) and shouldprovide a source of human bivalent and bispecific antibodyfragments for medical and industrial exploitation.

We thank colleagues-in particular, L. Riechmann and T. Simonfor vectors; A. D. Griffiths, H. Hoogenboom, J. D. Marks, and M.Malmqvist for ideas and discussions; M. Hirshberg, A. Lesk, K.Henrick, and C. Chothia for their advice and help on the modelingof the diabodies; and M. F. Perutz for his comments on the draftmanuscript. P.H. was supported by the Eidgenossiche TechnischeHochschule Zurich.

1. Winter, G. & Milstein, C. (1991) Nature (London) 349, 293-299.2. Staerz, U. D., Kanagawa, 0. & Bevan, M. J. (1985) Nature

(London) 314, 628-631.3. Kohler, G. & Milstein, C. (1975) Nature (London) 256, 495-

497.4. Milstein, C. & Cuello, A. C. (1983) Nature (London) 305,

537-540.5. Lanzaveccia, A. & Scheidegger, D. (1987) Eur. J. Immunol. 17,

105-111.6. Orlandi, R., Gussow, D. H., Jones, P. T. & Winter, G. (1989)

Proc. Natl. Acad. Sci. USA 86, 3833-3837.7. McCafferty, J., Griffiths, A. D., Winter, G. & Chiswell, D. J.

(1990) Nature (London) 348, 552-554.8. Marks, J. D., Hoogenboom, H. R., Griffiths, A. D. & Winter,

G. (1992) J. Biol. Chem. 267, 16007-16010.9. Hoogenboom, H. R., Marks, J. D., Griffiths, A. D. & Winter,

G. (1992) Immunol. Rev. 130, 41-68.10. Skerra, A. & Pluckthun, A. (1988) Science 240, 1038-1041.11. Better, M., Chang, C. P., Robinson, R. R. & Horwitz, A. H.

(1988) Science 240, 1041-1043.12. Bird, R. E., Hardman, K. D., Jacobson, J. W., Johnson, S.,

Kaufman, B. M., Lee, S. M., Lee, T., Pope, S. H., Riordan,G. S. & Whitlow, M. (1988) Science 242, 423-426.

13. Huston, J. S., Levinson, D., Mutgett-Hunter, M., Tai, M.-S.,Novotny, J., Margolies, M. N., Ridge, R. J., Bruccoleri, R. E.,Haber, E., Crea, R. & Opperman, H. (1988) Proc. Natl. Acad.Sci. USA 85, 5879-5883.

14. Shalaby, M. R., Shepard, H. M., Presta, L., Rodrigues, M. L.,

Beverley, P. C. L., Feldmann, M. & Carter, P. (1992) J. Exp.Med. 175, 217-225.

15. Kostelny, S. A., Cole, M. S. & Tso, J. Y. (1992) J. Immunol.148, 1547-1553.

16. Pack, P. & Pluckthun, A. (1992) Biochemistry 31, 1579-1584.17. Glockshuber, R., Malia, M., Pfitzinger, I. & Pluckthun, A.

(1990) Biochemistry 29, 1362-1367.18. Griffiths, A. D., Malqvist, M., Marks, J. D., Bye, J. M., Em-

bleton, M. J., McCafferty, J., Baier, M., Holliger, P., Gorick,B., Hughes-Jones, N., Hoogenboom, H. R. & Winter, G.(1993) EMBO J. 12, 725-734.

19. Gibson, T. J. (1984) Ph.D. thesis (Univ. of Cambridge, Cam-bridge, U.K.).

20. Hoogenboom, H. R., Griffiths, A. D., Johnson, K. S.,Chiswell, D. J., Hudson, P. & Winter, G. (1991) Nucleic AcidsRes. 19, 4133-4137.

21. Riechmann, L., Weill, M. & Cavanagh, J. (1992) J. Mol. Biol.224, 913-918.

22. Munro, S. & Pelham, H. R. B. (1986) Cell 46, 291-300.23. Ward, E. S., Gussow, D., Griffiths, A. D., Jones, P. T. &

Winter, G. (1989) Nature (London) 341, 544-546.24. McLaughlin, J., Murray, C. L. & Rabinowitz, J. C. (1981) J.

Biol. Chem. 256, 11283-11288.25. Zacher, A. N., Stock, C. A., Golden, J. W. & Smith, G. P.

(1980) Gene 9, 127-140.26. Amit, A. G., Mariuzza, R. A., Phillips, S. E. & Poljak, R. J.

(1986) Science 233, 747-753.27. Berek, C., Griffiths, G. M. & Milstein, C. (1985) Nature

(London) 316, 412-418.28. Sambrook, J., Fritsch, E. F. & Maniatis, T. (1990) Molecular

Cloning: A Laboratory Manual (Cold Spring Harbor Lab.Press, Plainview, NY).

29. Marks, J. D., Griffiths, A. D., Malmqvist, M., Clackson, T.,Bye, J. M. & Winter, G. (1992) BiolTechnology 10, 779-783.

30. Sanger, F., Nicklen, S. & Coulson, A. R. (1977) Proc. Natl.Acad. Sci. USA 74, 5463-5467.

31. Berek, C., Jarvis, J. M. & Milstein, C. (1987) Eur. J. Immunol.17, 1121-1129.

32. Dreher, M. (1992) Ph.D. thesis (Univ. of Cambridge, Cam-bridge, U.K.).

33. Eisen, H. N. (1964) Methods Med. Res. 10, 115-121.34. Foote, J. & Milstein, C. (1991) Nature (London) 352, 530-532.35. Takagi, H., Morinaga, Y., Tsuchiya, M., Ikemura, H. &

Inouye, M. (1988) BiolTechnology 6, 948-950.36. Skerra, A. & Pliickthun, A. (1991) Protein Eng. 4, 971-979.37. Borrebaeck, C. A. K., Malmborg, A.-C., Furebring, C.,

Michaelson, A., Ward, S., Danielson, L. & Ohlin, M. (1992)BiolTechnology 10, 697-698.

38. Foote, J. & Winter, G. (1992) J. Mol. Biol. 224, 487-499.39. Jones, D. H., McMilian, A. J., Fersht, A. R. & Winter, G.

(1985) Biochemistry 24, 5852-5857.40. Fischmann, T. O., Bentley, G. A., Bhat, T. N., Boulot, G.,

Mariuzza, R. A., Philipps, S. E. V., Tello, D. & Pobak, R. J.(1991) J. Biol. Chem. 266, 12915-12920.

41. Jones, T. A. (1982) in Computational Crystallography, ed.Sayre, D. (Clarendon, Oxford), pp. 303-310.

42. Hendrickson, W. A. & Konnert, J. H. (1981) in Structure,Conformation and Evolution, ed. Srinivasan, R. (Pergamon,Oxford), Vol. 1, pp. 43-57.

43. Jones, E. Y., Davis, J. D., Williams, A. F., Harlos, K. &Stuart, D. I. (1992) Nature (London) 360, 232-239.

6448 Biophysics: HoUiger et aL