Supporting InformationTape et al. 10.1073/pnas.1017067108SI Materials and MethodsMaterials. Unless stated otherwise, all chemicals were purchasedfrom Sigma-Aldrich. All eukaryotic cells were grown in RPMI1640 (Invitrogen 21875) supplemented with 10% (v∕v) fetal calfserum (HyClone CH20160.03). The antiTACE catalytic domainpolyclonal antibody pAb33 (Ab28233) and the anti-ADAM10catalytic domain polyclonal antibody pAb56 (Ab39153) wereboth purchased from Abcam. The anti-FLAG-HRP antibodywas purchased from Sigma-Aldrich (A8592). Development andcharacterization of anti-TACE scFvs A9 and D3 has been de-scribed previously (1). Recombinant mouse TACE (2978-AD-010) and human ADAM10 (936-AD-020) were purchased fromR&D Systems. Recombinant human ADAM12 was kindly pro-vided by Ulla Wewer (University of Copenhagen). RecombinantN-TIMP-3 was produced as previously described (2). All proteinmodels were visualized using the PyMOL Molecular GraphicsSystem, Version 1.0, Schrödinger, LLC.

Recombinant Human TACE.Mature recombinant TACE ectodomain(Arg215-Arg651) was expressed in baculovirus infected sf9 cells(a kind gift of Dr. D. Becherer, GlaxoSmithKline) and purifiedas described by Milla et al. (3). The mature catalytic domainof TACE (Arg215-Val477-GlySer-His6) was prepared using anidentical baculovirus system and purified by immobilizedmetal affinity chromatography (IMAC) (the generous gift ofDr. M. Taylor and Dr. P. Newham, AstraZeneca).

ScFv D1 VL Exchange. The VH domain of TACE inhibitory scFv D1was cloned into a naïve human light-chain (λ and κ) phage-displaylibrary developed by McCafferty and random colonies from theresulting library (hereafter the D1-VH-neo-VL library) were PCRscreened to assess VH-insert ratio (86% full scFv). Titrated con-centrations (0.01 nM, 0.1 nM, 1 nM, and 10 nM) of 1∶1 biotiny-lated TACE ectodomain (no CT1746) were exposed to the D1-VH-neo-VL library for two rounds of solution-phase selection. Inaddition, identically titrated selections were performed againstbiotinylated TACE ectodomain immobilized on streptavidincoated Immuno-Tubes (Nunc 444202) (solid-phase selections).Following two rounds of both selections, the eluted polyclonalscFv populations were individually cloned into pSANG10-3F andtransformed into the Escherichia coli BL21(DE3). 1,200 indivi-dual scFv clones were isolated from E. coli periplasm and ELISAscreened against immobilized recombinant TACE ectodomainand catalytic domain. From all 10 selections, the top 21 cloneswere individually expressed in 50 mL auto-induction shake flaskcultures and periplasmic fractions were purified by IMAC (Sator-ius VS-MCMINI24). Titrated concentrations of all matured scFvs(including the original D1 scFv) were ELISA screened against100 nM TACE ectodomain and catalytic domain to identify dualbinders.

Quenched-Fluorescent Peptide Cleavage Assay.Recombinant humanTACE catalytic domain and TACE ectodomain were diluted to1nM in 50mMTris-HCl, 10 mMCaCl2, 0.05%Brij35, 1%DMSO,pH 7.4 and preincubated with titrated concentrations of inhibitorfor 4 h at room temperature. Following incubation, each reactionwas separated into 200 μL technical quadruplets in a 96-wellblack Microwell plate (Nunc 237105), and the fluorogenic sub-strate methoxycoumarinyl acetyl—Lys-Pro-Leu-Gly-Leu-dinitro-phenyl diaminopropionyl-Ala-Arg-NH2 (Peptides InternationalSMO-3670-PI) was added to each well (final concentration ¼1 μM). Every 30 sec fluorescence was excited at 320 nm and emis-

sion recorded at 405 nm in a Tecan Infinite-200 (at 37 °C for 2000seconds). Individual readings were normalised against a substrateonly control and compiled to produce a mean trend for eachvariable. A linear regression slope for each reaction was calcu-lated in GraphPad Prism (ΔFU sec−1) and proteolytic activitywas expressed as the slope percentage of an untreated control(%ΔFU sec−1). Final results represent mean values from threeseparate experiments.

PDI ELISA. PDI-treated TACE ELISAs were performed as de-scribed previously (1). In brief, 30 μMpurified PDI was incubatedin the presence of 10-fold molar excess DTT (300 μM) for 15 minat room temperature. Excess DTT was removed by passing thereduced PDI through two desalting columns (Pierce 89882).30 μM reduced PDI was added to 10 μM biotinylated recombi-nant human TACE and incubated at 37 °C for 2 hours. The TACEsolution was then diluted to 50 nM and aliquoted (50 μL per well)onto an ELISA plate precoated with 50 μL 200 nM streptavidin(Sigma S4762). After a 20 min incubation, each well was washedsix times with 200 μL PBS-Tween (0.1% v∕v) to remove nonbio-tinylated PDI. Immobilized biotinylated TACE was subsequentlyprobed with 100 nM antibody.

Paratope Alanine Scanning Mutagenesis. D1(A12) paratope resi-dues were identified by homology modelling (4) and individualalanine mutants were created using site-directed mutagenesis(Stratagene 200521). Purified recombinant scFvs were subjectedto an eight-point fluorometric titration (½TACE� ¼ 1 nM) (asabove) and a 16-point titration ELISA (½TACE� ¼ 500 nM).IC50 and EC50 values for both D1(A12) (WT) and each alaninemutant (Ala) were calculated using GraphPad Prism. Change inGibb’s free energy (ΔΔG) was calculated using the equation:ΔΔG ¼ þRT lnðAla∕WTÞ.

Expression of Recombinant D1(A12) Human FAb. While useful forhigh-throughput screening experiments, the scFv antibody formatsuffers from avidity complications (due to spontaneous dimeriza-tion) that make it unsuitable for truly reliable kinetic analysis. Toensure monovalent kinetic analysis, D1(A12) was reformatted toa human FAb fragment. The VH and VL domains of D1(A12)were cloned into a novel human FAb expression vector basedon pET22b(+) (upstream of human CH1 and CL-κ respectively).Transformed BL21(DE3) RIPL E. coli were cultured to OD600 >40 in a 5L bench-top fermentor, induced with 10 mM IPTGand harvested after a further 4 h. The periplasmic fractionwas isolated by osmotic shock and human FAb was purified byProtein-G affinity chromatography (GE 17-0404-01).

Surface Plasmon Resonance (SPR). Immobilizing TACE on a BiacoreSPR chip using amine, aldehyde, or biotin coupling rapidly dena-tures the protein (only linear epitopes accessible). This mayexplain why there are no reported SPR experiments using TACE.To circumvent this issue, antiTACE probes were amine-coupledto a CM5 chip (GE Healthcare) (∼200–400 response units(RU)) and titrated concentrations of TACE were injected. Allexperiments were performed in 10 mM HEPES, 150 mM NaCl,1 mM CaCl2, 0.05% P-20, pH 7.4. Results represent the meanvalues of blank-subtracted technical triplicates per concentrationvariable. All experiments were performed on a Biacore T100(GE Healthcare) at 37 °C with a flow-rate of 40 μL∕ sec. Bindingconstants were calculated using Biacore T100 Evaluation Soft-

Tape et al. www.pnas.org/cgi/doi/10.1073/pnas.1017067108 1 of 8

ware (1∶1 binding model; Rmax h200 RU; tci > 1 × 108; Chi2 <0.5 RU2).

Expression of Recombinant D1(A12) Human IgG1. Because the aimof this project was to produce a clinical relevant cell-surfaceTACE inhibitor, we reformatted D1(A12) to a fully humanIgG1 for all cell assays. The VH and VL domains of D1(A12) werecloned into a novel pBudCE4.1 (Invitrogen V532-20) humanIgG1 expression vector (κ-variant) and transfected into HEK-293 cells using Fugene 6 (Roche 11988387001). Stably transfectedHEK-293 populations were grown to maximum confluence in10-layer HYPERFlasks (Corning 10030) and human IgG1 waspurified from the conditioned media by Protein-L affinity chro-matography (Pierce 89929). Traces of bovine serum proteinswere removed using Melon Gel technology (Pierce 45206) andthe final D1(A12) human IgG1 was buffer exchanged into sterilePBS.

TNF-α Cleavage Assay. Recombinant human TACE was combinedwith titrated concentrations of D1(A12) FAb (diluted in 50 mM

Tris-HCl, 10 mM CaCl2, 0.05% Brij35, 1% DMSO, pH 7.4) andimmediately added to 5 μM GST-TNF-α (5). Each reaction wasincubated at 37 °C, resolved by 12% SDS/PAGE, coomassiestained and individual bands were quantified by densitometry[ImageQuant TL (GE Healthcare)].

TACE Cell-Surface Shedding Assays. For all shedding assays, 4×104 cells∕well (in 300 μL media) were plated in 48-well platefor 36 h, washed three times with serum free media, and prein-cubated with either D1(A12) Human IgG1, N-TIMP-3, or controlHuman Plasma IgG (R&D Systems 1-001-A) (diluted in serumfree media) for 1 h. Each well was stimulated with 100 μg∕mLphorbol 12-myristate 13-acetate (PMA) for and supernatantswere harvested after 1 h. Soluble TNF-α, TGF-α, and Amphire-gulin were quantified by sandwich ELISA (R&D System Duoset)and HB-EGF alkaline-phosphatase was measured as describedpreviously (1). All ADAM siRNA knockdown experiments wereperformed as described previously (1).

1. Willems SH, et al. (2010) Thiol isomerases negatively regulate the cellular sheddingactivity of ADAM17. Biochem J.

2. Lee MH, Knauper V, Becherer JD, Murphy G (2001) Full-length and N-TIMP-3display equal inhibitory activities toward TNF-alpha convertase. Biochem BiophysRes Commun 280:945–950.

3. Milla ME, et al. (1999) Specific sequence elements are required for the expression offunctional tumor necrosis factor-alpha-converting enzyme (TACE). J Biol Chem274:30563–30570.

4. Sircar A, Kim ET, Gray JJ (2009) RosettaAntibody: Antibody variable region homology

modeling server. Nucleic Acids Res 37:W474–479.

5. d’Ortho MP, et al. (1997) Membrane-type matrix metalloproteinases 1 and 2 exhibit

broad-spectrum proteolytic capacities comparable to many matrix metalloprotei-

nases. Eur J Biochem 250:751–757.

Fig. S1. Predicted TACE ectodomain topology. (A) A one-dimensional depiction of TACE ectodomain (construct Arg651) illustrating the relative locations of theSignal Peptide (S; white), Pro Domain (green), Furin cleavage site (hollow arrow), Catalytic Domain (red), zinc-coordinating histidines and the Disintegrin-Cysteine Rich (Dis-Cys) Domain (blue). (B) A three-dimensional representation of TACE homologs employing the color system outlined inA. The crystal structureof the ADAM-homolog Russell’s Viper Venom metalloprotease (RVV-X) reported the tight spatial association between the catalytic and Dis-Cys domains. Incombination with earlier vascular apoptosis-inducing protein (VAP) crystallography, this structural evidence supported the notion that ADAMs are “C-shaped.”More recently, the noncatalytic human ADAM22 ectodomain structure (complete with carboxyl-terminal EGF-like domain, yellow) has continued this idea. (C)Unfortunately, given both the difficulty in expressing milligram quantities of TACE ectodomain and the pharmacological focus on developing small moleculeinhibitors, all TACE structural studies have focused exclusively on the isolated catalytic domain. In consolation, structural homology modelling of TACE (1)reveals the full ectodomain likely exists in a comparable C shape. Such models suggest that the Dis-Cys domain obstructs macromolecular access to the TACEcatalytic site in a similar fashion to RVV-X and ADAM22. The close association between the Dis-Cys domain and the catalytic-cleft could explain why thecomplete TACE ectodomain is approximately 10-fold less proteolytic than the isolated catalytic domain. Moreover, this topology might also explain whythe natural protein inhibitor of TACE (and many other metalloproteases), tissue inhibitor of metalloprotease 3 (TIMP-3), is significantly better at inhibitingthe exposed TACE catalytic domain rather than the “guarded” full ectodomain.

1 Roy A, Kucukural A, Zhang Y (2010) I-TASSER: A unified platform for automated protein structure and function prediction. (Translated from eng) Nat Protoc 5:725–738 (in eng).

Tape et al. www.pnas.org/cgi/doi/10.1073/pnas.1017067108 2 of 8

Fig. S2. Isolation of anti-TACE ectodomain human scFv antibodies. (A) Following two rounds of solution-phase phage-display selections against biotinylatedTACE ectodomain (with CT1746), the eluted phagemid population was transferred into the pSANG10-3F vector and individual scFvs were expressed in E. coli.Each antibody was ELISA screened against TACE ectodomain (without CT1746) and BSA. Results are expressed as the fold-difference between the two antigens(ΔA450 ¼ A450TACE∕A450BSA). All clones ΔA450 ≥ 10 (dashed line) were isolated, sequenced to remove replicates, individually expressed, and purified byaffinity chromatography. In addition, a negative control clone (B9) was also advanced. (B) Titration ELISAs against TACE ectodomain revealed EC50s rangingbetween 11 and 420 nM. (C) Quenched-fluorescent (QF) peptide activity of 1 nM recombinant TACE ectodomain following preincubation with 500 nM purifiedlead antibody revealed several scFvs possessed some inhibitory capacity (*). (D) To investigate the translation of QF-peptide inhibition to cell-surface TACEinhibition, HeLa cells overexpressing alkaline-phosphatase (AP) tagged HB-EGF were preincubated with 500 nM of each scFv and stimulated with PMA. PBS andthe amino-terminal domain of the endogenous TACE inhibitor TIMP-3, N-TIMP-3 (NT3), were used as positive and negative and respectively. TACE-mediatedshedding of HB-EGF was measured by assaying AP activity from the conditioned media (displayed as A405). Despite the identification of several recombinantTACE ectodomain inhibitors in (C), only scFv D1 (**) appeared to retain this inhibitory potency at the cell-surface. All� represent SD.

Tape et al. www.pnas.org/cgi/doi/10.1073/pnas.1017067108 3 of 8

Fig. S3. Surface plasmon resonance plots for Table 1. AntiTACE probes were amine-coupled to a CM5 chip [approximately 200 response units (RU)] and titratedconcentrations of TACE were injected. Results represent the mean values of blank-subtracted technical triplicates per concentration variable. All experimentswere performed on a Biacore T100 (GE Healthcare) at 37 °C with a flow-rate of 40 μL∕ sec. Binding constants were calculated using Biacore T100 EvaluationSoftware (1∶1 binding model; Rmax h200 RU; tci1 × 108; Chi2 < 0.5 RU2).

Fig. S4. D1 scFv does not inhibit human ADAM10. Recombinant human ADAM10 ectodomain was preincubated with titrated concentrations of either humanD1 scFv or the small molecule metalloprotease inhibitor, CT1746. Subsequent ADAM10 proteolytic activity was measured in a quench fluorescent peptidecleavage assay. Results are expressed as the percentage of an untreated control. While CT1746 rapidly inhibits ADAM10 proteolysis, scFv D1 has no effect.All� represent SD.

Tape et al. www.pnas.org/cgi/doi/10.1073/pnas.1017067108 4 of 8

Fig. S5. Solid-phase paratope mutagenesis. (A) The EC50 for each paratope mutant was calculated by solid-phase ELISA and used to calculateΔΔGðΔΔG ¼ þRT lnðEC50

Ala∕EC50WTÞ. Echoing the pattern described in Fig. 4, residues SH31, YH32, and SH52 (*) support EC50

Ecto∶WT, and residues QL27,SL28, IL29, SL91, and FL92 (†) support EC50

Cat∶WT. (B) Collated ΔΔG values from solution-phase inhibition (QF-peptide) and solid-phase binding (ELISA). Despitetheir independent approaches, D1(A12) paratope ΔΔG profiles are remarkably similar (ectodomain correlation ¼ 0.86� 0.05; R2 ¼ 0.91)(catalytic domain correlation ¼ 0.82� 0.06; R2 ¼ 0.86).

Fig. S6. D1(A12) IgG1 and FAb inhibition of cell-surface TACE. To investigate whether the bivalent nature of D1(A12) IgG1 influenced its TACE inhibitorypotency, titrations of IgG1 and monovalent FAb were compared in a PMA stimulated HeLa alkaline-phosphatase (AP) tagged HB-EGF assay. The D1(A12) FAbdisplayed a comparable inhibitory profile to the IgG1. This suggested only one variable domain of each IgG1 was responsible for inhibiting cell-surface TACE.All� represent SD.

Tape et al. www.pnas.org/cgi/doi/10.1073/pnas.1017067108 5 of 8

Fig. S7. D1(A12) is a selective TACE inhibitor. (A) 50 nM immobilized human TACE ectodomain, human TACE catalytic domain, mouse TACE ectodomain,human ADAM10 ectodomain, human ADAM12 ectodomain, and BSA were individually ELISA probed with 100 nM D1(A12) FAb, antiTACE catalytic domainpolyclonal (pAb33), and anti-ADAM10 polyclonal (pAb56). Despite the close sequence homology between the ADAM antigens, D1(A12) FAb is entirely selectivefor human TACE. (B) Recombinant human ADAM10 ectodomain was preincubated with titrated concentrations of either human D1(A12) FAb or the smallmolecule metalloprotease inhibitor, CT1746. Subsequent ADAM10 proteolytic activity was measured in a quench fluorescent peptide cleavage assay. Despitepartially binding the TACE catalytic domain, D1(A12) FAb has no effect on ADAM10 activity. (C) MCF7 cells stably transfected with alkaline-phosphatase taggedHB-EGF provide a useful model to distinguish between ADAM10 and TACE cell-surface shedding activity.Western blot analysis ofMCF7 breast cancer cell lysatesfollowing treatment with nontargeting, ADAM10 or TACE siRNA. (D) PMA-stimulation of MCF7 cells results in TACE dependent shedding of HB-EGF-AP.(E) Ionomycin-stimulation of MCF7 cells results in ADAM10-dependent shedding of HB-EGF-AP. (F) D1(A12) IgG1 only inhibits TACE-mediated PMA stimulationof HB-EGF-AP shedding in MCF7 breast cancer cells. For cell assays, one-way ANOVA tests were performed comparing control cells (e.g., nontargeting siRNA) tovariable (e.g., PMA stimulated). P values: * ¼ < 0.05, ** ¼ < 0.01, *** ¼ < 0.001. All� represent SD.

Tape et al. www.pnas.org/cgi/doi/10.1073/pnas.1017067108 6 of 8

Fig. S8. D1(A12) shares a partial TACE catalytic-cleft epitope with CT1746 and N-TIMP-3. (A) AntiTACE probes were amine-coupled to a CM5 chip and titratedconcentrations of TACE ectodomain were injected in the presence of either 100 μM CT1746 TACE catalytic-cleft small molecule inhibitor (SMI) or an equalvolume of DMSO. Surface plasmon resonance (SPR) studies revealed preincubation with CT1746 completely inhibited N-TIMP-3 binding to the TACE catalytic-cleft. In contrast, the affinity of TACE Dis-Cys antibody D1was unaltered by the presence of CT1746. This is unsurprising given that D1 was selected against TACEectodomain in the presence of CT1746. The subsequent selection of D1-VH-neo-VL antibodies in the absence of CT1746 permitted the introduction of TACEcatalytic-cleft residues into the epitope of D1(A12). As such, the association constant (ka) of D1(A12) for the TACE ectodomain is almost 10-fold slower in thepresence of the CT1746. This suggest that D1(A12) shares at least a partially overlapping TACE catalytic-cleft epitope with CT1746. (B) 100 nM TACE ectodomainwas immobilised onto an ELISA plate and preincubated with various concentrations of D1(A12) monovalent FAb. The subsequent accessibility of the TACEcatalytic-cleft was monitored by assaying for N-TIMP-3 binding. D1(A12) blocks access to the TACE ectodomain catalytic-cleft at 1∶1 molar ratio. (C) Reversingthe probe orientation (i.e., TACE preincubated with various concentrations of N-TIMP-3 and probed with D1(A12) FAb) produced comparable results.All� represent SD.

Table S1. QF-peptide IC50 values from Figs. 2A and 5B

IC50 (nM)

Inhibitor Epitope Ectodomain Catalytic Domain Cat/Ecto

N-TIMP-3 catalytic-cleft 3.1 (±0.15) 0.6 (±0.02) 5.2D1(A12) FAb cross-domain 0.45 (±0.01) 0.54 (±0.02) 0.8

Tape et al. www.pnas.org/cgi/doi/10.1073/pnas.1017067108 7 of 8

Table S2. HB-EGF-AP IC50 values from Fig. 2B,Fig. S6, and Fig. 5C

Inhibitor Epitope IC50 (nM)

N-TIMP-3 catalytic-cleft 47.3 (±2.30)D1 scFv Dis-Cys 67.2 (±2.1)D1(A12) FAb cross-domain 8.6 (±1.8)D1(A12) IgG1 cross-domain 7.9 (±1.22)

Table S3. Cell-surface shedding IC50 values from Fig. 5C

IC50 (nM)

Substrate Cell Line D1(A12) IgG1 N-TIMP-3 TIMP-3/D1(A12)

TNF-α TOV21G 11.2 (±0.95) 48.5 (±3.56) 4.3TGF-α IGROV-1 9.4 (±2.34) 44.5 (±4.58) 4.7Amphiregulin PC3 9.3 (±1.65) 53.3 (±1.31) 5.7HB-EGF-AP HeLa 7.9 (±1.22) 47.3 (±2.30) 6.0TNFR1a IGROV-1 10.4 (±0.97) 49.5 (±2.51) 4.7

In addition, results from TNFR1a shedding are described. All� represent SD.

Tape et al. www.pnas.org/cgi/doi/10.1073/pnas.1017067108 8 of 8