Cross-domain inhibition of TACE ectodomainChristopher J. Tapea, Sofie H. Willemsa,1, Sarah L. Dombernowskya,1, Peter L. Stanleya, Marton Fogarasia,Willem Ouwehandb, John McCaffertyc, and Gillian Murphya,2

aDepartment of Oncology, University of Cambridge, Cancer Research UK Cambridge Research Institute, Li Ka Shing Centre, Cambridge, CB2 0RE, UnitedKingdom; bDepartment of Haematology, University of Cambridge, National Health Service Blood and Transplant, Cambridge, CB2 0PT, United Kingdom;and cDepartment of Biochemistry, University of Cambridge, Tennis Court Road, Cambridge, CB2 1QW, United Kingdom

Edited by James A. Wells, UCSF, San Francisco, CA, and approved February 11, 2011 (received for review November 17, 2010)

Proteolytic release from the cell surface is an essential activationevent for many growth factors and cytokines. TNF-α convertingenzyme (TACE) is a membrane-bound metalloprotease responsiblefor solubilizing many pathologically significant membrane sub-strates and is an attractive therapeutic target for the treatmentof cancer and arthritis. Prior attempts to antagonize cell-surfaceTACE activity have focused on small-molecule inhibition of themetalloprotease active site. Given the highly conserved natureof metalloprotease active sites, this paradigm has failed to producea truly specific TACE inhibitor and continues to obstruct the clinicalinvestigation of TACE activity. We report the bespoke developmentof a specific TACE inhibitory human antibody using “two-step”phage display. This approach combines calculated selection condi-tions with antibody variable-domain exchange to direct indivi-dual antibody variable domains to desired epitopes. The resulting“cross-domain” human antibody is a previously undescribed selec-tive TACE antagonist and provides a unique alternative to small-molecule metalloprotease inhibition.

ADAM17 ∣ antibody phage display ∣ cancer therapeutics

TNF-α converting enzyme (TACE) [a disintegrin and metallo-protease 17 (ADAM17)], is a membrane-bound metal-

loprotease responsible for cleaving a variety of pathologicallysignificant substrates (1). Initially identified as the enzymeresponsible for solubilizing membrane-associated pro-TNF-α(2, 3)—a process subsequently termed “ectodomain shedding,”TACE has since proved capable of cleaving epidermal growthfactor receptor (EGFR) ligands (4), extracellular Notch1 (5),cell-surface receptors (6), and adhesion molecules (7). As proteo-lytic cleavage is an indispensable activation event for manyof these substrates, TACE has emerged as an attractive thera-peutic target for the treatment of cancer (8) and rheumatoidarthritis (9).

Preceding the current clinical interest in TACE, membersof the related matrix metalloprotease (MMP) family were alsoconsidered viable therapeutic targets (10). Despite sound precli-nical rational for antagonizing MMPs, early trials of small-molecule inhibitors (SMIs) failed due to poor inhibitor specificityprofiles (11, 12). Metalloprotease SMIs exclusively target thecatalytic site. This paradigm treats the raw proteolytic capacityof the catalytic site as the only significant target—often withno regard to noncatalytic domains. This catalytic focus forcesthe selectivity of SMIs to rely exclusively on modest biophysicaldifferences surrounding individual metalloprotease catalyticsites and typically yields inhibitors simply with a bias towarda particular metalloprotease. Macromolecular metalloproteaseinhibitors [e.g., tissue inhibitor of metalloproteases (TIMPs) andmetalloprotease prodomains] also focus on binding the catalyticsite and suffer comparable specificity limitations. The unfortu-nate simplification of multidomain extracellular proteases tospatially isolated catalytic sites has ignored the potential fornoncatalytic residues to contribute toward inhibitor specificity. Abroader multidomain TACE antagonist would provide an attrac-tive alternative to the SMI paradigm.

Mature ADAM-family ectodomains contain a globular metal-loprotease catalytic domain, a disulphide-dependent disintegrin-

cysteine rich (Dis-Cys) domain, and, in some cases, an epidermalgrowth factor (EGF)-like domain. Although most ADAM cata-lytic domains appear to share homologous structural topology,significant sequence variation is common throughout the nonca-talytic Dis-Cys domains [especially in the “hyper-variable” region(HVR)] (13). Despite their divergent nomenclature, accumulat-ing evidence suggests the catalytic domain and Dis-Cys domainare spatially associated within the complete ADAM ectodomain(14, 15). Specifically, the crystal structures of ADAM homologsVAP1 (13), VAP2B (16), RVVX (17), and the nonproteolyticectodomain of human ADAM22 (18) support the notion thatADAM proteins are “C-shaped” (Fig. S1). This model impliesthe TACE noncatalytic carboxyl-terminal Dis-Cys domain par-tially obstructs macromolecular access to the amino-terminalcatalytic domain. As a result, large catalytic-cleft inhibitors suchas TIMPs (14) and ADAM prodomains (19) fail to bind completeADAM ectodomains efficiently. To overcome this issue, weproposed that a selective ADAM inhibitor could instead utilizethis spatially connected C-shaped multidomain topology bybroadly antagonizing the catalytic domain while simultaneouslysourcing additional specificity from local noncatalytic Dis-Cysresidues.

Antibody phage display is an established technology forproducing inhibitors against complex extracellular proteins (20).As solution-phase phage display typically produces antibodieswith conformational (i.e., nonlinear) epitopes, this technologyis also theoretically capable of providing the intricate macromo-lecular cross-domain binding desirable in an ADAM inhibitor.Recent technical advances in phage display have produced anti-bodies recognizing multiple disparate antigens (21) and differentconformations of the same antigen (22, 23). These approachesuse a combination of calculated selection conditions and/orantibody variable-domain chain-shuffling to produce antibodiesagainst previously unique epitopes. We hypothesized that similartechniques could be employed to achieve multidomain proteaseinhibition.

We exploited ADAM multidomain topology by first isolatingan inhibitory human antibody (D1) that bound TACE-specificnoncatalytic regions exclusively through its variable heavy (VH)domain. We then used a D1-VH biased scFv phage-display libraryto selectively isolate a new variable light (VL) chain that couldsimultaneously bind to the TACE catalytic domain (Fig. 1).The resulting “cross-domain” human antibody [D1(A12)] is apreviously undescribed biochemically holistic ADAM ectodo-main inhibitor and demonstrates a unique alternative to small-molecule metalloprotease inhibition.

Author contributions: C.J.T., J.M., and G.M. designed research; C.J.T., S.H.W., andS.L.D. performed research; P.L.S., M.F., W.H.O., and J.M. contributed new reagents/analytictools; and C.J.T. and G.M. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1S.H.W. and S.L.D. contributed equally to this work.2To whom correspondence should be addressed. E-mail: gm290@cam.ac.uk.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1017067108/-/DCSupplemental.

5578–5583 ∣ PNAS ∣ April 5, 2011 ∣ vol. 108 ∣ no. 14 www.pnas.org/cgi/doi/10.1073/pnas.1017067108

Dow

nloa

ded

by g

uest

on

Nov

embe

r 2,

202

0

ResultsIsolation of an Anti-TACE Ectodomain Inhibitory Human Antibody.While most TACE drug discovery projects focus on inhibitingthe proteolytic capacity of the isolated catalytic domain, wepurposefully chose to antagonize the complete ectodomain. Inlight of both recent structural advances and prior biochemicalobservations, we hypothesized that selectively targeting noncata-lytic regions of the complete TACE ectodomain would produce

a more specific cell-surface inhibitor. To this end, recombinanthuman TACE ectodomain was biotinylated, checked for wild-type activity, and exposed to a naïve human scFv antibody phage-display library (24) for two rounds of solution-phase selections.As previous attempts to directly target metalloprotease catalyticsites have resulted in unwanted cross-reactivity (e.g., TIMPs,ADAM prodomains, and SMIs), we blocked the TACE catalyticcleft with the broad-spectrum metalloprotease inhibitor CT1746(25) during initial selections (Fig. 1). The resulting TACEectodomain antigen could therefore not select for antibodies withepitopes dependent on residues deep within the catalytic site.Following ELISA screening (Fig. S2A), subsequent positivescFv His-tagged clones were sequenced to remove replicates,expressed in Escherichia coli and affinity-purified [immobilizedmetal affinity chromatography (IMAC)] for functional character-ization. Although several inhibitory antibodies were identified bytheir ability to hinder TACE quenched-fluorescent (QF) peptideproteolysis (Fig. S2C), only scFv D1 retained this inhibitoryprofile when tested against cell-surface shedding of HB-EGF(Fig. S2D). ScFv D1 inhibited TACE ectodomain activity withcomparable potency to the amino-terminal domain of the naturalTACE inhibitor TIMP-3, N-TIMP-3, in both QF-peptide (Fig. 2A)and HB-EGF shedding assays (Table S2). However, unlikeN-TIMP-3, scFv D1 did not bind the isolated catalytic domainof TACE (Fig. 2B). We have previously shown that proteindisulphide isomerase (PDI) can alter the three-dimensionaltopology of the TACE Dis-Cys domain (26). In a similar fashionto the Dis-Cys binding scFv D3, PDI modulation of the TACEectodomain seriously disrupted scFv D1 immunoreactivity(Fig. 3B). When combined with the lack of isolated catalytic do-main binding, this observation suggests scFv D1 primarily boundthe noncatalytic TACE Dis-Cys domain. Alanine-scanning muta-genesis of individual scFv D1 complementarity determiningregion (CDR) loops revealed that residues in the VH chain ofscFv D1 were primarily responsible for TACE ectodomain bind-ing. In contrast, the CDR loops of the VL domain did notappear to significantly contribute toward the active D1 paratope(Fig. 2C). This may explain the modest affinity of scFv D1 (KD ¼26ð�2.2Þ nM) for the TACE ectodomain (Table 1 and Fig. S3).Despite this conservative paratope, scFv D1 appeared to beentirely selective for human TACE (Fig. 2D and Fig. S4). Collec-tively, we conclude scFv D1 is a selective VH-dependent inhibi-tory antibody that primarily binds to the noncatalytic TACEDis-Cys domain.

Fig. 1. Experimental overview. (A) The human TACE ectodomain consists ofan amino-terminal metalloprotease catalytic domain (light red) and a carbox-yl-terminal noncatalytic Dis-Cys domain (light blue) (I-TASSERmodel) (39). Weexploited this multidomain topology to develop a truly specific ADAM inhi-bitor using two-step antibody phage display. (B) (i) First, the catalytic site ofTACE ectodomain was blocked during primary antibody phage-display selec-tions using the small-molecule inhibitor CT1746. This prevented the selectionof antibodies with catalytic-cleft epitopes that could cross-react with nontar-get metalloproteases. (ii) Primary screening revealed the inhibitory scFv anti-body clone D1. This scFv bound specifically to the TACE Dis-Cys domainthrough its variable heavy (VH) domain. (iii) A D1-VH-bias antibody phage-display library was produced to introduce new variable light (neo-VL) chainswhile maintaining the TACE specificity provided by the D1-VH. Secondaryselections were performed in the absence of CT1746 in order to providethe neo-VL chains with uninterrupted access to the TACE catalytic site.(iv) Secondary screening identified several neo-VL scFvs capable of bindingthe isolated TACE catalytic domain. Due to Dis-Cys domain binding throughthe D1-VH these “cross-domain” antibodies maintained their strict specificityfor TACE. D1-VH-neo-VL scFv clone A12 (D1(A12)) exhibited the highestaffinity for the TACE ectodomain and is the most selectively potentcell-surface ADAM inhibitor ever described.

Fig. 2. ScFv D1 is a VH-dependent TACE ectodomain inhibitory human antibody. (A) TACE ectodomain (Arg215-Arg651) proteolytic activity was measured in aquenched-fluorescent peptide cleavage assay. D1 inhibits TACE ectodomain proteolysis with comparable potency to the catalytic-cleft focused TACE inhibitorN-TIMP-3 (IC50

D1 ¼ 5.4ð�0.4Þ nM; ðIC50N-TIMP-3 ¼ 3.2ð�0.2Þ nMÞ). (B) Immobilized TACE ectodomain and catalytic domain (Arg215-Ser474) were probed with

titrated concentrations of scFv D1 and the anticatalytic domain rabbit polyclonal antibody pAb33. Despite its capacity to inhibit TACE-activity, D1 scFv failedto bind the isolated TACE catalytic domain. (C) Paratope alanine-scanning mutagenesis of scFv D1 CDRs revealed the VH domain was primarily responsible forTACE ectodomain binding. In contrast, the D1 VL domain appeared almost entirely dispensable for TACE ectodomain binding. (D) Multiple TACE homologswere individually ELISA probed with 100 nM D1 scFv, pAb33 and anti-ADAM10 polyclonal pAb56. Despite the close sequence homology between the ADAMantigens, D1 scFv is entirely selective for human TACE ectodomain. All� represent SD.

Tape et al. PNAS ∣ April 5, 2011 ∣ vol. 108 ∣ no. 14 ∣ 5579

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Nov

embe

r 2,

202

0

Introduction of Catalytic Domain Binding by VL-Exchange.While scFvD1 selectively inhibited TACE ectodomain activity with compar-able potency to the broad metalloprotease inhibitor TIMP-3,its limited paratope, poor affinity, and lack of catalytic domainbinding suggested the antagonist could be improved. As scFv D1bound noncatalytic regions through its VH domain, yet was closeenough to the catalytic site to block small peptide hydrolysis,we hypothesized that the currently quiescent VL domain was inclose proximity to the TACE catalytic domain. We subsequentlyhypothesized that the nonfunctional D1 VL-CDRs could beengineered to introduce TACE catalytic domain binding. Toexplore this idea, the D1-VH domain was cloned into a naïvehuman VL phage-display library and the resulting “D1-VH-neo-VL library” was stringently reselected against titrated concentra-tions of biotinylated TACE ectodomain. As the D1-VH wasalready entirely TACE selective through Dis-Cys binding (Fig. 2Dand Fig. S4), we removed CT1746 from all D1-VH-neo-VL selec-tions to provide neo-VL domains with uninterrupted access to theTACE catalytic site. The resulting selection scenario encouragedall D1-VH-neo-VL scFvs to maintain TACE selectivity (by bindingto ectodomain Dis-Cys regions through the D1-VH) while simul-taneously exposing neo-VL domains to a previously inaccessiblecatalytic-cleft epitope (due to the absence of the small-moleculeantagonist) (Fig. 1B).

Following two rounds of both solution and solid-phase selec-tions, 1,200 D1-VH-neo-VL scFvs were rescreened for bothTACE ectodomain and isolated catalytic domain binding. Thebest 30 TACE ectodomain-binding clones were isolated, se-quenced to remove replicates, individually expressed in E. coli,and affinity-purified. D1-VH-neo-VL lead scFvs were then quan-titatively ELISA screened for their capacity to bind both thecomplete TACE ectodomain and the isolated catalytic domain

(Fig. 3A). As predicted, stringent selection of neo-VL chainsagainst TACE in the absence of CT1746 produced multipleD1-VH-neo-VL scFv variants now capable of independentlybinding the isolated TACE catalytic domain and the completeectodomain. Lead scFv “A12” [hereafter D1(A12)] possessedthe highest affinity for both antigens and was advanced forfurther analysis.

Kinetic Characterization of the D1(A12)–TACE Interaction. ScreeningELISAs implied that the D1-VH-neo-VL clone D1(A12) couldindependently bind both the complete TACE ectodomain andthe isolated catalytic domain (Fig. 3A). In addition, D1(A12) islargely resistant to PDI modulation of the TACE Dis-Cys domainwhen compared to the parental scFv D1 (Fig. 3B). Collectively,these results suggest the D1(A12) epitope contains residues fromboth the TACE catalytic and Dis-Cys domains. To characterizethe kinetics of both interactions, D1(A12) was reformatted toa monovalent human FAb, amine-coupled to a CM5 Biacorechip, and titrated concentrations of either the TACE ectodomainor isolated catalytic domain were injected. Surface plasmonresonance (SPR) revealed D1(A12) possessed an affinity con-stant (KD) of 0.46ð�0.7Þ nM for the complete TACE ectodo-main, but only 5.21ð�0.1Þ nM for the isolated catalytic domain(ΔKD ¼ KD

Cat∕KDEcto ¼ 11.3) (Table 1 and Fig. S3). In contrast,

noninhibitory anti-TACE antibodies from the primary selection(catalytic domain antibody A9 and Dis-Cys antibody D3) exhibitdomain-specific profiles. The catalytic-domain focused TACEinhibitor N-TIMP-3 displays a 30-fold affinity preference forthe isolated catalytic domain (ΔKD ¼ 0.03). D1(A12) is thereforeunique: an ADAM inhibitor that shows an affinity preferencefor the complete ectodomain over the isolated catalytic domain.

Fig. 3. Introduction of TACE catalytic domain binding by VL exchange. (A) The VH domain of scFv D1 was cloned into a naïve human VL phage-display libraryand reselected against titrated concentrations of biotinylated TACE ectodomain in the absence of CT1746. Selected scFvs were cloned into a FLAG-taggedexpression vector and screened for both TACE ectodomain and isolated catalytic domain binding (left). The top 30 ectodomain clones were sequencedto remove replicates (21 remained), individually expressed, affinity-purified, and quantitatively rescreened [10 nM] for both TACE ectodomain and catalyticdomain binding (right). For many clones, the neo-VL domain facilitates independent binding to the TACE catalytic domain. D1-VH-neo-VL clone “A12” (*)(hereafter D1(A12)) displayed the highest affinity for both antigens. (B) In contrast to scFv D1, scFv D1(A12) is largely resistant to PDI modulation of the TACEDis-Cys domain (†) (26). This behavior is similar to antibodies with TACE catalytic domain epitopes. All� represent SD.

Table 1. TACE surface plasmon resonance (SPR)

ka (1∕Ms) kd (1∕s) KD (M)

Probe Cat. Ecto. Cat. Ecto. Cat. Ecto. KD Cat:∕KD Ecto. Epitope

N-TIMP-3 1.0E+05 7.2E+03 2.1E−05 5.2E−05 2.1E−10 7.2E−09 0.03 catalytic domainA9 8.2E+04 5.9E+04 6.7E−03 5.3E−03 8.2E−08 9.0E−08 0.91 catalytic domainD3 - 1.9E+05 - 8.1E−02 - 4.2E−07 - Dis-CysD1 - 7.4E+04 - 1.9E−03 - 2.6E−08 - Dis-CysD1(A12) 1.1E+05 5.3E+05 5.8E−04 2.4E−04 5.2E−09 4.6E−10 11.30 cross-domain

Macromolecular anti-TACE probes were amine-coupled to a Biacore CM5 chip and titrated concentrations of either isolated TACE catalytic domain (Cat.) orcomplete ectodomain (Ecto.) were injected. SPR kinetic analysis revealed the catalytic-cleft-focused TACE inhibitor N-TIMP-3 possessed a 30-fold affinitypreference for the TACE catalytic domain over the complete ectodomain. Antibodies produced from primary phage-display selections bind either thecatalytic domain (A9) or the noncatalytic Dis-Cys domain (D1 and D3). In contrast, the D1-VH-neo-VL secondary selection antibody D1(A12) displays a10-fold affinity preference for the complete TACE ectodomain. D1(A12) is therefore the first ADAM inhibitor to show an affinity preference for thecomplete ectodomain over the isolated catalytic domain. SPR plots are illustrated in Fig. S3.

5580 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1017067108 Tape et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 2,

202

0

D1(A12) Paratope Scanning Mutagenesis. As the initial D1scFv did not react with the TACE catalytic domain—yet catalyticdomain binding was effectively introduced through neo-VL-ex-change—we concluded that residues within the original D1-VHinteracted with the TACE Dis-Cys and residues within the neo-A12-VL interacted with the catalytic domain. To comprehensivelycharacterize the D1(A12) paratope, all residues extendingbeyond the β-carbon were individually mutated to alanine, ex-pressed in E. coli, and IMAC-purified. The solution-phaseQF-peptide IC50 of each mutant (IC50

Ala) was calculated forboth the complete TACE ectodomain (IC50

Ecto) and the isolatedcatalytic domain (IC50

Cat). In addition, the wild-type D1(A12)scFv IC50 (IC50

WT) was simultaneously calculated for both theTACE ectodomain (IC50

Ecto∶WT ¼ 0.89ð�0.04Þ nM) and catalyticdomain (IC50

Cat∶WT ¼ 2.3ð�0.1Þ nM) using an identical proce-dure. Subsequent changes in Gibb’s free energy (ΔΔG) werecalculated (ΔΔG ¼ þRT lnðIC50

Ala∕IC50WTÞ) (27) for each mu-

tant and antigen (n ¼ 62) (Fig. 4). In agreement with CDRmutagenesis of scFv D1 (Fig. 2C), many D1(A12) VH residuescontribute to IC50

Ecto∶WT. Interestingly, CDR-H1 residues SH31and YH32, and CDR-H2 residue SH52 (Kabat numbering)appear to exclusively contribute to IC50

Ecto∶WT and are almostentirely dispensable for achieving IC50

Cat∶WT. Conversely,CDR-L1 residues QL27, SL28, and IL29, and CDR-L3 residuesSL91 and FL92, only appear to contribute to IC50

Cat∶WT and arealmost entirely dispensable for achieving IC50

Ecto∶WT. To comple-ment this solution-phase analysis, solid-phase ELISA EC50 ΔΔGvalues were also calculated for all paratope mutants (Fig. S5A).SPR was not used to obtain kinetic data for all 62 interactions

because TACE immobilization led to its denaturation (seeSI Materials and Methods). Despite their disparate methodology,the solution and solid-phase D1(A12) paratope ΔΔG profilesare remarkably similar (Fig. S5B) (ectodomain correlation ¼0.86� 0.05; R2 ¼ 0.91) (catalytic domain correlation ¼ 0.82�0.06; R2 ¼ 0.86). This agreement suggests D1(A12) binding isa direct proxy for TACE inhibition.

When mapped onto D1(A12) Fv RosettaAntibody frameworks(28), residues displaying either antigen bias were shown to clusterat polar ends of the paratope. In addition, CDR-H3 representsa dually important intermediate region within the core of theparatope. Collectively, these data strongly suggest that D1(A12)exclusively interacts with TACEDis-Cys domain through residueson the outskirts of the VH domain and exclusively interacts withthe catalytic domain through select residues in the VL domain.

D1(A12) Potently Inhibits the Complete TACE Ectodomain. Monova-lent D1(A12) FAb proved capable of inhibiting the proteolysisof a macromolecular GST-TNF-α substrate by both the TACEectodomain and the isolated catalytic domain (Fig. 5A). Corre-lating with previous affinity data, D1(A12) FAb inhibited TACEectodomain activity more potently than isolated catalytic domainactivity (IC50

Ecto ¼ 73.9ð�3.2Þ nM; IC50Cat ¼ 124.7ð�6.2Þ nM).

Moreover, the D1(A12) FAb retained this potent inhibitorycapacity in comprehensive QF-peptide analysis (Fig. 5B). D1(A12) FAb inhibited the isolated TACE catalytic domain withsimilar potency to the natural leading TACE inhibitor N-TIMP-3 (ΔIC50 ¼ IC50

Cat∶N-TIMP-3∕IC50Cat∶D1ðA12Þ ¼ 1.35). How-

ever, when identical assays were performed with the completeTACE ectodomain, D1(A12) proved to be over fivefold betterthan N-TIMP-3 (ΔIC50 ¼ IC50

Ecto∶N-TIMP-3∕IC50Ecto∶D1ðA12Þ ¼

5.75).As the rationale for inhibiting the complete TACE ectodomain

was to produce a superior cell-surface TACE inhibitor for clinicalapplication, D1(A12) was reformatted to a human IgG1 formatand compared to N-TIMP-3 in multiple cancer cell-based shed-ding assays (Fig. 5C). The effect of D1(A12) IgG1 on the shed-ding of four separate TACE ligands was investigated across fourhuman cancer cell lines. Interestingly, the QF-peptide TACEectodomain inhibitory profiles of D1(A12) and N-TIMP-3 wererepeated in all four cell-surface shedding assays. Irrespectiveof substrate, cell line, or TACE expression levels, D1(A12)human IgG1 routinely inhibited cell-surface TACE activity five-fold better than N-TIMP-3 (Table S3). Similar inhibitory profileswere obtained with the monovalent D1(A12) FAb (Fig. S6)—sug-gesting only one variable domain per IgG was binding cell-surfaceTACE. ELISA, QF-peptide, and cell-surface shedding assaysalso confirm D1(A12) specifically targets TACE and not closelyrelated proteases (Fig. S7). As only the TACE ectodomain ispresent at the cell surface, this collective data comprehensivelydemonstrates the antagonistic value in specifically targetingthe complete TACE ectodomain.

DiscussionDespite some progress in developing TACE inhibitors over thepast decade (29), attempts to inhibit TACE using a catalytic-site directed approach have largely produced semispecificantagonists. In this study we describe how combining calculatedphage-display selection conditions with subsequent antibodyvariable-domain exchange can produce a selective inhibitor ofthe complete TACE ectodomain. This approach utilizes a two-step selection process to independently direct both heavy andlight antibody variable chains toward desired epitopes. By exploit-ing biochemical knowledge of the complete extracellular antigen,the D1(A12) cross-domain antibody has solved the decade-old problem of how to inhibit an ADAM ectodomain moreefficiently than the isolated catalytic domain (14, 19). Moreover,the selective potency of D1(A12) seriously challenges the exist-

Fig. 4. D1(A12) paratope scanning mutagenesis. (A) All D1(A12) scFv para-tope residues extending beyond the β-carbon were individually mutated toalanine, expressed in E. coli, and affinity-purified. The IC50 for each mutantagainst both the complete TACE ectodomain (IC50

Ecto) and the isolated cat-alytic domain (IC50

Cat) were determined in solution by quenched-fluorescentpeptide assay. The change in Gibb’s free energy (ΔΔG) was calculated(ΔΔG ¼ þRT lnðIC50

Ala∕IC50WTÞ) for each mutant and antigen (n ¼ 62). While

many mutations proved detrimental to the D1(A12) IC50WT for both antigens,

several appeared to specifically alter binding to either the TACE ectodomain(*) or the catalytic domain (†). When mapped onto D1(A12) Fv frameworks(28) (employing colors detailed on the right y-axis), residues displaying anantigen bias cluster at polar ends of the paratope (dashed white lines). Com-parable results were determined by solid-phase titration ELISA (Fig. S5).

Tape et al. PNAS ∣ April 5, 2011 ∣ vol. 108 ∣ no. 14 ∣ 5581

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Nov

embe

r 2,

202

0

ing biochemical paradigm of solely targeting metalloproteasecatalytic sites.

Despite the unique antagonistic approach reported, the use ofnoncatalytic residues to increase protease inhibitor specificity isnot new. Existing inhibitory antibodies against the serine proteaseMT-SP1 were isolated by competitively selecting for scFv inthe presence of a macromolecular antigen antagonist (30). Thisapproach yielded a steric hindrance antibody antagonist (E2) thatsourced antigen specificity through MT-SP1 noncatalytic surfaceloops and inhibited by binding the active site (31). However,unlike D1(A12), the epitope of E2 resides within the single cat-alytic domain of the MT-SP1 antigen. Given the “C-shape” ofADAM ectodomains, mimicking this approach with an isolatedADAM catalytic domain antigen would likely produce an inhibi-tor that suffers from similar ectodomain affinity limitations asTIMPs (14) and ADAM prodomains (19). We demonstrate howmultiple selection conditions in combination with antibody vari-able-domain exchange can circumvent these issues to produce aselective antagonist against a complete ADAM ectodomain.

Further studies with serine proteases have shown that antibo-dies can operate as either substrate blocking (steric hindrance) orallosteric inhibitors (32). Given that D1(A12) shares a partiallyoverlapping TACE catalytic-cleft epitope with CT1746 andTIMP-3 (Fig. S8), it is logical to presume that D1(A12) inhibitsTACE proteolysis by simply blocking macromolecular substrateaccess to the catalytic site. However, as the catalytic TACE zincion is thought to undergo dynamic charge transitions mediated bynoncatalytic regions (15), D1(A12) could also allosterically dis-rupt crucial communication between distal protein sites by simul-taneously binding both noncatalytic and catalytic regions ofTACE. Monitoring TACE behavior in the presence of D1(A12)using stopped-flow X-ray spectroscopy could help distinguishbetween steric hindrance and allosteric inhibition models.

Two-step phage display produced an antibody capable of bind-ing to both the TACE ectodomain (step one) and isolated cata-

lytic domain (step two). Since recombinant TACE ectodomainimmunoreactivity can be compromised by disulphide rearrange-ment (26) and TACE mutants often do not mature correctly (33),we focused on comprehensively characterizing the antibody para-tope instead of the TACE epitope. Paratope scanning mutagen-esis of D1(A12) described how D1-VH residues SH31, YH32, andSH52 are involved in binding the TACE ectodomain but appeardispensable for binding the isolated catalytic domain. Thesekey residues may provide D1(A12) with its greater than 10-foldaffinity preference for the TACE ectodomain over the catalyticdomain. In contrast, A12-VL residues QL27, SL28, IL29, SL91,and FL92 primarily support binding to the isolated catalyticdomain. Given that D1(A12) CDR-H3 remains unchanged fromthe original scFv D1 (noncatalytic binding), it is interesting thatthese residues play such an important role in binding the isolatedcatalytic domain. This observation suggests either that CDR-H3(i) always interacted with the catalytic domain and simply lackedsupport from the original D1-VL for high affinity binding, oralternatively (ii) the CDR-H3 loop has been contorted by theneo-A12-VL to incorporate previously nonexistent catalytic do-main binding. One of the key changes from the A12-VL wasL94E > I in CDR-L3 (neutralization of negative charge) andL50T > D in CDR-L2 (gain of negative charge). Topologically,this charge migration flanks the CDR-H3 loop and could permitsubstantial macromolecular movement. CDR loop flexibility iswell documented (34, 35) and RosettaAntibody models (28) sup-port multiple D1 CDR-H3 conformations.

As many antibody paratopes are primarily supported by the VHdomain, VL exchange is often used for affinity maturation. How-ever, given that VL exchange was recently shown to modify a VH-biased Her2 antibody to also bind VEGF (21), this reselectiontechnique clearly has considerable potential beyond monoantigenaffinity maturation. Our work further demonstrates how manip-ulating initial selection conditions (e.g., by blocking undesiredepitopes with SMIs or cofactors) can produce primary antibodies

Fig. 5. D1(A12) is a potent inhibitor of TACE ectodomain proteolysis. (A) D1(A12) FAb inhibited the cleavage of recombinant TNF-α by both 100 nM TACEectodomain (IC50

Ecto ¼ 73.9ð�3.2Þ nM) and, to a lesser potency, the isolated catalytic domain (IC50Cat ¼ 124.7ð�6.2Þ nM). Quantitative results from three

separate experiments are displayed as a percentage of a 50 μM CT1746 (CT) metalloprotease inhibitor control. (B) D1(A12) FAb inhibited the proteolyticcapacity of the TACE catalytic domain with a comparable potency to N-TIMP-3. However, when the full-length recombinant human TACE ectodomainwas preincubated with both inhibitors, the D1(A12) FAb proved fivefold more potent than N-TIMP-3 (†). (C) Cancer cells with known expression of TACEsubstrates (TOV21G: TNF-α, IGROV1: TGF-α, PC3: AREG) and HeLa cells stably over expressing HB-EGF-Alkaline Phosphatase were used to assay cell-surfaceTACE activity. Each cell line was stimulated with PMA following a 1 h pretreatment with various concentrations of either D1(A12) human IgG1, N-TIMP-3, orcontrol human plasma IgG. Soluble TACE products were quantified from conditionedmedium by sandwich ELISA or alkaline phosphatase activity. D1(A12) IgG1consistently inhibited cell-surface TACE activity around fivefold more potently than N-TIMP-3 (*). Cell-surface TACE inhibition directly correlates with theenzymatic data in B (i.e., �≈†). Full IC50 data can be found in Tables S1, S2, and S3. All� represent SD.

5582 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1017067108 Tape et al.

Dow

nloa

ded

by g

uest

on

Nov

embe

r 2,

202

0

against directed epitopes. Secondary neo-variable domain selec-tions using a modified antigen (e.g., no inhibitor/cofactor or adifferent antigen entirely) can then reexpose the initial antibodyto new epitopes while retaining paratope elements from the pri-mary antigen. This technique could be routinely used to produceantibodies against multidomain epitopes. For example, two-stepphage display could theoretically produce antibodies againsttertiary protein complexes (e.g., ligand-bound receptors), multi-ple local loops in membrane proteins (e.g., tetraspanins), andprotein-nucleic acid complexes (transcription-factor DNA com-plexes). It is important to note that while our primary selectionconditions promoted D1 scFv TACE non-catalytic-cleft binding(by blocking the TACE active site), the isolation of a purelyVH-bias antibody was not encouraged by this method. Althoughmany antibodies possess a VH-dominant paratope (particularlyCDR-H3), our process relied on chance to provide an antibodysuitable for VL exchange. Future efforts to develop a more pre-dictable “VH then VL” selection approach could consider usinga VH-only phage-display library for primary selection—followedby the cloning of selected VH genes into a naïve neo-VL library(complete scFv) for secondary selection. However, such anapproach would suffer from instability issues associated withexpressing human antibody variable domains in isolation (36). Amore stable alternative could be to employ a scFv library withmultiple VH domains and a single constant VL domain for theprimary selection. A subsequently selected VH domain could thenbe cloned into a naïve neo-VL library for secondary selectionagainst a modified antigen.

In summary, we describe the bespoke development of a cross-domain inhibitory TACE antibody using two-step phage display.

By combining calculated selection conditions with antibodyvariable-domain exchange, we have produced a specificallypotent ADAM ectodomain inhibitor. The resulting cross-domainhuman antibody D1(A12) provides a unique opportunity tostudy the individual pathological impact of TACE activity. Ourapproach is extendable to other multidomain macromoleculesand provides a previously undescribed alternative to small-molecule metalloprotease inhibition.

Materials and MethodsRecombinant human TACE ectodomain (Arg215-Arg651) was biotinylated ata 1∶1 ratio using N-succinimidyl biotin (Invitrogen AL-01), checked for wild-type activity in a quenched-fluorescent peptide cleavage assay, and exposedto the naïve human scFv phage-display library of McCafferty (24) in thepresence of 50 μM CT1746 (25). Following two rounds of solution-phaseselection, the eluted polyclonal scFv population was cloned into pSANG10-3F (37) and transformed into BL21(DE3) RIPL E. coli (Stratagene 230280).Individual scFv clones were isolated from E. coli periplasm and ELISA screenedagainst immobilized recombinant TACE ectodomain in the absence ofCT1746. Comprehensive screening details have been outlined previously(24, 26). Following initial screening, selected anti-TACE scFv clones wereexpressed in 500 mL auto-induction (38) shake flask cultures and periplasmicfractions were purified by IMAC. Purified scFvs were screened for recombi-nant TACE inhibition in a quenched-fluorescent peptide assay and forcell-surface TACE inhibition in a PMA stimulated HB-EGF-alkaline phospha-tase assay.

Detailed methods are described in SI Materials and Methods.

ACKNOWLEDGMENTS. We thank Dr. David Becherer (GlaxoSmithKline) forTACE ectodomain expressing baculovirus and Dr. Mark Taylor and Dr. PeterNewham (AstraZeneca) for TACE catalytic domain expressing baculovirus.This work was funded by Cancer Research UK.

1. Edwards DR, Handsley MM, Pennington CJ (2008) The ADAMmetalloproteinases.MolAspects Med 29:258–289.

2. Black RA, et al. (1997) A metalloproteinase disintegrin that releases tumour-necrosisfactor-alpha from cells. Nature 385:729–733.

3. Moss ML, et al. (1997) Cloning of a disintegrin metalloproteinase that processesprecursor tumour-necrosis factor-alpha. Nature 385:733–736.

4. Sunnarborg SW, et al. (2002) Tumor necrosis factor-alpha converting enzyme(TACE) regulates epidermal growth factor receptor ligand availability. J Biol Chem277:12838–12845.

5. Brou C, et al. (2000) A novel proteolytic cleavage involved in Notch signaling: The roleof the disintegrin-metalloprotease TACE. Mol Cell 5:207–216.

6. Marin V, et al. (2002) Chemotactic agents induce IL-6Ralpha shedding from polymor-phonuclear cells: Involvement of a metalloproteinase of the TNF-alpha-convertingenzyme (TACE) type. Eur J Immunol 32(10):2965–2970.

7. Condon TP, et al. (2001) ADAM17 but not ADAM10 mediates tumor necrosisfactor-alpha and L-selectin shedding from leukocyte membranes. Antisense NucleicA 11:107–116.

8. Kenny PA (2007) TACE: A new target in epidermal growth factor receptor dependenttumors. Differentiation 75:800–808.

9. Moss ML, Sklair-Tavron L, Nudelman R (2008) Drug insight: Tumor necrosis factor-converting enzyme as a pharmaceutical target for rheumatoid arthritis. Nat Clin PractRheum 4:300–309.

10. DeClerck YA, Imren S (1994) Protease inhibitors: Role and potential therapeutic use inhuman cancer. Eur J Cancer 30A:2170–2180.

11. Egeblad M, Werb Z (2002) New functions for the matrix metalloproteinases in cancerprogression. Nat Rev Cancer 2:161–174.

12. Coussens LM, Fingleton B, Matrisian LM (2002) Matrix metalloproteinase inhibitorsand cancer: Trials and tribulations. Science 295:2387–2392.

13. Takeda S, Igarashi T, Mori H, Araki S (2006) Crystal structures of VAP1 reveal ADAMs’MDC domain architecture and its unique C-shaped scaffold. EMBO J 25:2388–2396.

14. Lee MH, et al. (2002) The C-terminal domains of TACE weaken the inhibitory action ofN-TIMP-3. FEBS Lett 520:102–106.

15. Solomon A, Akabayov B, Frenkel A, Milla ME, Sagi I (2007) Key feature of the catalyticcycle of TNF-alpha converting enzyme involves communication between distal proteinsites and the enzyme catalytic core. Proc Natl Acad Sci USA 104:4931–4936.

16. Igarashi T, Araki S, Mori H, Takeda S (2007) Crystal structures of catrocollastatin/VAP2B reveal a dynamic, modular architecture of ADAM/adamalysin/reprolysin familyproteins. FEBS Lett 581:2416–2422.

17. Takeda S, Igarashi T, Mori H (2007) Crystal structure of RVV-X: An example of evolu-tionary gain of specificity by ADAM proteinases. FEBS Lett 581:5859–5864.

18. Liu H, Shim AH, He X (2009) Structural characterization of the ectodomain of adisintegrin and metalloproteinase-22 (ADAM22), a neural adhesion receptor insteadof metalloproteinase: insights on ADAM function. J Biol Chem 284:29077–29086.

19. Gonzales PE, et al. (2004) Inhibition of the tumor necrosis factor-alpha-convertingenzyme by its pro domain. J Biol Chem 279:31638–31645.

20. Carter PJ (2006) Potent antibody therapeutics by design. Nat Rev Immunol 6:343–357.21. Bostrom J, et al. (2009) Variants of the antibody herceptin that interact with HER2 and

VEGF at the antigen binding site. Science 323:1610–1614.22. Gao J, Sidhu SS, Wells JA (2009) Two-state selection of conformation-specific antibo-

dies. Proc Natl Acad Sci USA 106:3071–3076.23. Hu X, Kang S, Lefort C, Kim M, Jin MM (2010) Combinatorial libraries against libraries

for selecting neoepitope activation-specific antibodies. Proc Natl Acad Sci USA107:6252–6257.

24. Schofield DJ, et al. (2007) Application of phage display to high throughput antibodygeneration and characterization. Genome Biol 8:R254.

25. An Z, et al. (1997) Conversion of highly malignant colon cancer from an aggressive to acontrolled disease by oral administration of a metalloproteinase inhibitor. Clin ExpMetastasis 15:184–195.

26. Willems SH, et al. (2010) Thiol isomerases negatively regulate the cellular sheddingactivity of ADAM17. Biochem J 428:439–450.

27. Cunningham BC, Wells JA (1993) Comparison of a structural and a functional epitope.J Mol Biol 234:554–563.

28. Sircar A, Kim ET, Gray JJ (2009) RosettaAntibody: Antibody variable region homologymodeling server. Nucleic Acids Res 37:W474–479.

29. DasGupta S, Murumkar PR, Giridhar R, Yadav MR (2009) Current perspective ofTACE inhibitors: A review. Bioorg Med Chem 17:444–459.

30. Sun J, Pons J, Craik CS (2003) Potent and selective inhibition of membrane-type serineprotease 1 by human single-chain antibodies. Biochemistry 42:892–900.

31. Farady CJ, Egea PF, Schneider EL, Darragh MR, Craik CS (2008) Structure of anFab-protease complex reveals a highly specific non-canonical mechanism of inhibition.J Mol Biol 380:351–360.

32. Ganesan R, Eigenbrot C, Kirchhofer D (2010) Structural and mechanistic insight intohow antibodies inhibit serine proteases. Biochem J 430:179–189.

33. Li X, Fan H (2004) Loss of ectodomain shedding due to mutations in the metallopro-tease and cysteine-rich/disintegrin domains of the tumor necrosis factor-alphaconverting enzyme (TACE). J Biol Chem 279:27365–27375.

34. Mylvaganam SE, Paterson Y, Getzoff ED (1998) Structural basis for the binding of ananti-cytochrome c antibody to its antigen: Crystal structures of FabE8-cytochrome ccomplex to 1.8 A resolution and FabE8 to 2.26 A resolution. J Mol Biol 281:301–322.

35. Jimenez R, Salazar G, Baldridge KK, Romesberg FE (2003) Flexibility and molecularrecognition in the immune system. Proc Natl Acad Sci USA 100:92–97.

36. Jager M, Pluckthun A (1999) Domain interactions in antibody Fv and scFv fragments:Effects on unfolding kinetics and equilibria. FEBS Lett 462:307–312.

37. Martin CD, et al. (2006) A simple vector system to improve performance and utilisationof recombinant antibodies. BMC Biotechnol 6:46.

38. Studier FW (2005) Protein production by auto-induction in high density shakingcultures. Protein Expres Purif 41:207–234.

39. Roy A, Kucukural A, Zhang Y (2010) I-TASSER: A unified platform for automatedprotein structure and function prediction. Nat Protoc 5:725–738.

Tape et al. PNAS ∣ April 5, 2011 ∣ vol. 108 ∣ no. 14 ∣ 5583

BIOCH

EMISTR

Y

Dow

nloa

ded

by g

uest

on

Nov

embe

r 2,

202

0