A Monoclonal Antibody to ADAM17 Inhibits Tumor Growth by Inhibiting EGFR and Non–EGFR-Mediated Pathways

ADAM17 is the primary sheddase for HER pathway ligands. We report the discovery of a potent and specific ADAM17 inhibitory antibody, MEDI3622, which induces tumor regression or stasis in many EGFR-dependent tumor models. The inhibitory activity of MEDI3622 correlated with EGFR activity both in a series of tumor models across several indications as well in as a focused set of head and neck patient–derived xenograft models. The antitumor activity of MEDI3622 was superior to that of EGFR/HER pathway inhibitors in the OE21 esophageal model and the COLO205 colorectal model suggesting additional activity outside of the EGFR pathway. Combination of MEDI3622 and cetuximab in the OE21 model was additive and eradicated tumors. Proteomics analysis revealed novel ADAM17 substrates that function outside of the HER pathways and may contribute toward the antitumor activity of the monoclonal antibody. Mol Cancer Ther; 14(7); 1637–49. ©2015 AACR.


Introduction
ADAM17/TACE (TNFa converting enzyme) is a protease originally identified as the major sheddase for TNFa (1, 2). The phenotype of Adam17 knockout mice (3), however, closely aligns with those of EGFR and HBEGF knockout mice implicating ADAM17 in the shedding of ligands that transduce signals through EGFR family receptor tyrosine kinases (4,5) and implying that it is the shed form of these ligands which are responsible for the majority of EGFR-driven signaling. Indeed, knockout of Adam17 in mouse embryo fibroblasts dramatically reduces the amount of shed EGFR ligands (amphiregulin, epiregulin, HBEGF, TGFa) as well as the major ligand for HER3 (heregulin; ref. 6). ADAM17 and its substrates within the HER pathways are important drivers of tumorigenesis and promote resistance to existing therapeutics. High expression of both ADAM17 and TGFa correlate with reduced survival in breast cancer as well as in other indications (7). Amphiregulin and TGFa are both elevated in progressing gefitinib-treated patients relative to those patients with partial clinical responses or stable disease (8). Furthermore, ADAM17 has been implicated in the shedding of HER pathway-independent cytokines, growth factors, receptors, and adhesion molecules (reviewed in ref. 9), many of which have been implicated in cancer development. This HER-independent activity suggests that inhibition of ADAM17 may confer broad antitumor efficacy either as a single agent or in combination with existing HER pathway antagonists targeting HER family receptors. Currently available HER receptor-directed therapeutics have had disappointing clinical response rates (10), and would likely benefit from a more complete shutdown of the HER pathways (11) via combination with a therapeutic targeting the ligand shedding step.
For these reasons, there have been extensive efforts to develop ADAM17 inhibitors (reviewed in ref. 12). Two issues have confounded these efforts and prevented progress beyond phase II clinical trials. First, the structural conservation of the active site across ADAMs and the related ADAMTS, MMP, and SVMP metalloprotease families has made it difficult to develop ADAM17specific inhibitors that do not impact the function of these related proteins (13). Second, compounds with low nanomolar potency in biochemical assays utilizing recombinant ADAM17 were found to be much less potent inhibitors of cellular ADAM17 (14). Although the molecular basis for this discrepancy remains unresolved, active cellular ADAM17 is known to undergo a structural transition allowing accessibility of its active site to small-molecule inhibitors (15). To some extent, these challenges have been overcome in the recent report of a human ADAM17 antibody capable of inhibiting ligand shedding in cells with an IC 50 of approximately 10 nmol/L (16). Nevertheless this antibody does not inhibit mouse ADAM17 and therefore cannot be used to determine the requirement for ADAM17 in either syngeneic mouse tumor models or xenograft models with a significant component of stromal-mediated ligand production. We report here the discovery an ADAM17 antibody, MEDI3622, which has both human and mouse cross reactivity and displays potent antitumor activity in preclinical models as a result of inhibition of both HER-dependent and HER-independent pathways.

Materials and Methods
Identification of ADAM17 substrates in OE21 cells OE21 cells were grown in RPMI media with 10% dialyzed fetal calf serum containing N15 stable heavy isotopes of lysine and arginine to allow mass spectrometric analysis of shed proteins via SILAC. Cells were grown in heavy medium for 3 weeks before experiment to allow replacement of light amino acids with heavy amino acids. 6 T150 flasks of light OE21 cells and 6 flasks of heavy OE21 cells were grown to 70% confluence. Media from all 12 flasks was replaced with identical media but lacking fetal calf serum. Flasks were then pretreated with either 400 nmol/L MEDI3622 or 400 nmol/L nonspecific IgG and ADAM17 was activated by addition of 50 nmol/L phorbol 12myristate 13-acetate (PMA). After incubation at 37 degrees for 6 hours, supernatants from heavy and light cells were mixed, IgG was removed via chromatography over a Protein A column, and flow through was analyzed by mass spectrometry-based proteomics.

In vivo antitumor efficacy studies
All cells were grown in monolayer culture, harvested by trypsinization, and implanted subcutaneously into the right flank of 6to 8-week-old female athymic nude mice obtained from Harlan using a 27-gauge needle. Mice were staged and randomized into 10 mice per group when tumors reached approximately 150 to 200 mm 3 . MEDI3622, cetuximab, and a nonspecific IgG1 control (MedImmune) were prepared by diluting with PBS and dosed intraperitoneally twice a week according to body weight. Both tumor and body weight measurements were collected twice weekly and tumor volume calculated using the equation (L Â W 2 )/2, where L and W refer to the length and width dimensions, respectively. AZD8931 was dissolved in 1% (v/v) Tween 80 in sterile water and was dosed per os once a day for 14 days. Antitumor efficacy was measured as percent DTGI. DTGI was calculated as (1ÀdT/dC) Â 100, where dT is the final tumor volume minus the starting tumor volume from the treatment group and dC is the final tumor volume minus the starting tumor volume of the control group. DTGI was calculated using the tumor volume measurement after the last dose. Error bars were calculated as SEM. The general health of mice was monitored daily and all experiments were conducted in accordance to Association for Assessment and Accreditation of Laboratory Animal Care and MedImmune Institutional Animal Care and Use Committee guidelines for humane treatment and care of laboratory animals.

Head and neck PDX cancer models
Female immunocompromised mice (Harlan; nu/nu) between 5 and 8 weeks of age were housed on irradiated papertwistenriched 1/8 00 corncob bedding (Sheperd) in individual HEPAventilated cages (Innocage IVC, Innovive) on a 12-hour lightdark cycle at 68 to 74 F (20-23 C) and 30% to 70% humidity. Animals were fed water (reverse osmosis, 2 ppm Cl 2 ) and an irradiated test rodent diet (Teklad 2919; 19% protein, 9% fat, and 4% fiber) ad libitum. All patient-derived xenografts (PDX) were established by Champions Oncology, Inc. Animals were implanted unilaterally on the left flank with tumor fragments harvested from donor animals. When tumors reached approximately 100 to 300 mm 3 , animals were randomized by tumor volume into treatment and control groups and dosing initiated. MEDI3622 was dosed at 10 mg/kg in these studies under continuous dosing twice per week via intraperitoneal injection. DTGI was calculated using the tumor volume measurement after the last dose.

Pharmacokinetic analysis
To study the pharmacokinetics of MEDI3622, tumor-bearing mice were dosed with MEDI3622 intraperitoneally at 3, 10, and 30 mg/kg on day 0 and day 3, and serum was collected 72 hours after the second dose. Measurement of MEDI3622 serum levels was performed using ELISA. Briefly, 96-well plates were coated overnight at 4 C with goat anti-human IgG Fc (Thermo Scientific; #31125). The next day, mouse serum from the animals was diluted in 1% milk in PBS and incubated in the plate for 2 hours at room temperature. After washing in PBS, goat anti-human lambda was added (Southern Biotech, #20705) for 1 hour at room temperature. Signal was detected using TMB substrate and antibody concentrations determined from a standard curve.

Pharmacodynamic effects of MEDI3622
To study the pharmacokinetics of MEDI3622, tumor-bearing mice were dosed with MEDI3622 intraperitoneally at 3, 10, and 30 mg/kg on day 0 and day 3, and tumor was collected 72 hours after the second dose. Harvested tumor samples were placed in vials containing lysing matrix A (MP Bio) and flash frozen on dry ice at the time of collection. MSD lysis buffer (Meso Scale Discovery) containing protease and phosphatase inhibitors were then added to each vial and samples were homogenized using a FastPrep-24 instrument (MP Bio). Lysates were centrifuged and supernatant quantitated by bicinchoninic acid (BCA) protein assay (Pierce). Equal amounts of protein samples (20 mg) were resolved on a 4% to 12% Bis-Tris gel and transferred onto a polyvinylidene difluoride membrane using an iBlot instrument (Invitrogen). Antibodies used in Western blotting were as follows: phospho-EGFR Y845, (Cell Signaling Technology, #6963), phospho-EGFR Y1068 (Cell Signaling Technology, #3777), phospho-EGFR Y1070 (Abgent, #AJ1252f). phospho-EGFR Y1086 (Invitrogen, #33790G), phospho-EGFR Y1173 (Invitrogen, For the phospho-EGFR array (RayBio #AAH-PER-1-8), 250 mg of tumor lysate was added to each membrane and analyzed accordingly to the manufacturers' protocol. A human phosphokinase array was also used according to the manufacturer's recommendations (R&D Systems, #ARY003B).
For pharmacodynamic effects in the Colo205 model, tumors were collected at the end of the efficacy study 24 hours after the last dose. Tumors were analyzed according to the manufacturer's instructions (MesoScale Discovery).
For densitometry analysis, ImageJ software was used. The baseline percent activation was determined by averaging the phospho/total amounts of the nonspecific IgG control treatment group and this was given a value of 100% activation. The percent activation was determined by dividing the average ratio of phosphor/total of the treatment groups by the average phospho/total ratio of the nonspecific IgG control group.

In vivo LPS induction
Nine-week-old female DBA/1 mice were obtained from Taconic. Mice were injected intraperitoneally with 10 mg/kg nonspecific IgG control or 10 mg/kg MEDI3622, followed by intraperitoneal injection of either 0.3 mg/kg lipopolysachharide (LPS; Sigma #L4130) or PBS 2 hours later. After 90 minutes, whole blood was collected by cardiac puncture and serum was separated for future analysis. Samples were analyzed using the murine TNFa ELISA (R&D Systems, #MTA00B) and quantified with the Softmax Pro 5.4.

Flow cytometry
T4-2 cells were plated into U bottom plates for flow analysis. Activation utilized 500 nmol/L PMA for 15 minutes at 37 C. Cells were spun and washed with flow buffer (PBS þ 2% FBS) followed by blocking for 15 minutes with 20% FBS. Cells were washed and 1 mg/mL primary antibody was added. Mab9301 (R&D Systems) and MEDI3622 were used to quantitate total and active ADAM17 detection, respectively. These incubation times were 30 minutes at 4 C followed by washing. Secondary antibodies were added at 1:100 dilution for 30 minutes at 4 C then washed. Samples were suspended in 100 mL of FACS buffer and flow cytometry was performed on LSRII followed by FlowJo analysis.
OE21 cells were plated on Costar 3788 plates. Primary antibodies: anti: AREG, HBEGF, EGF, EREG, TGFa, BTC, and HRG1 (R&D Systems 10 mg/mL) were added to cells and incubated for 30 minutes at 4 C followed by washing. Anti-mouse or anti-goat secondary antibodies (Invitrogen, 1:100 dilution) were added to cells and incubated for 30 minutes at 4 C followed by washing. Flow cytometry was performed on LSRII followed by FlowJo analysis.

EGFR ligand shedding
One thousand H358 cells were plated in 96-well plates and allowed to adhere overnight at 37 C. The following day MEDI3622, negative control antibody, or small-molecule control were added to cells at concentrations ranging from 600 to 0.59 nmol/L (1:2 steps) for antibodies or from 10,000 to 0.17 nmol/L (1:3 steps) for PMA. Cells were then incubated for 4 hours at 37 C. Cell supernatant was collected and added to prepared ELISA plates. ELISAs were performed according to ELISA Kit manufacturer specifications (R&D Systems).

TNFa blood shedding
TNFa shedding was performed in Costar 3788 plates with 100 mL of mouse whole blood. LPS (100 ng/mL) was used to stimulate TNFa shedding. MEDI3622, negative control antibody or smallmolecule control were added to cells at concentrations ranging from 600 to 0.59 nmol/L (1:2 steps) for antibodies or from 10,000 to 0.17 nmol/L (1:3 steps) for PMA. Incubations proceeded for 4 hours at 37 C. Plates were spun and supernatants were collected. Supernatants were added to prepared ELISA plates and ran per manufacturer instructions (Invitrogen).

MSD assay
Tumor samples were collected and processed with MSD lysis buffer containing protease and 2Â phosphatase inhibitors (MSD kit) with Lysing Matrix beads (MP) in FastPrep for 30 seconds to generate lysate. Protein concentration was determined via BCA (Pierce). Lysates were added to MSD plates and quantitated according to manufacturer specifications.

Discovery of 80PH3
CAT phage libraries (17) displaying antibody single-chain fragments of vL and vH (scFv) were blocked with 4% milk then incubated with the biotinylated recombinant extracellular domain (ECD) of ADAM17 at room temperature for 1 hour. Four percent milk blocked streptavidin beads (Invitrogen) were added to the phage-antigen mixture and incubated for 15 minutes at room temperature. Streptavidin beads bound with antigen-phage complexes were pulled down with magnetic beads and washed with PBS three times to remove unbound phage. After 3 rounds of panning, high titer phage were produced from approximately 7,000 individual colonies and subjected to large scale highthroughput screening using a laser scanning fluorimeter (Isocyte, MDS) as follows: 293F cells with/without transient transfection of full-length ADAM17 plasmid were harvested 24 hours later and resuspended in blocking buffer (1% BSA and PBS), mixed with anti-M13-Alexa488, and distributed in 384-well plates (Corning) at 2 Â 10 4 cells/well in 20 mL volume each well. Four microliters of phage supernatant were then added to each well. Incubation was carried out at 4 C for 3 to 16 hours and the plates were read via Isocyte. Clones with 1,000-fold higher total intensity on ADAM17-expressing cells over that on nonexpressing cells were considered as positive. Positive clones were converted to IgG and screened in the biochemical ADAM17 assay as described elsewhere in Materials and Methods.

Affinity optimization
Affinity optimization of 80PH3 utilized a high-throughput mammalian expression platform that enabled screening optimized IgG variants in both binding and shedding inhibition assays. Briefly, the IgG expression library was constructed by cloning the VH or VL of 80PH3 into a shuttle vector, followed by parsimoniously randomization of all 6 CDRs using mutagenic primers encoding 18 amino acids at each position via site-directed mutagenesis (Agilent kit). VH/VL were than batch cloned into the pOE IgG expression vector. Library DNA was prepared in 96-well plates using the Qiagen Turbo Kit and used to transfect 293F cells (350 mL/well) in 96-deep well plates. IgG expressed in the supernatant was quantified by Octet in a 384-well plate format and was screened in the shedding and Capture ELISA binding assays. IgG hits from both assays were retransfected in 24-well plates and confirmed by titrated assay. All the amino acid changes from the confirmed positive hits were simultaneously encoded in primers to construct a combinatorial library that was screened for variants encoding multiple amino acids changes that resulted in an additive or synergistic increase in affinity or functional activity.
Shedding assay. H292 non-small cell lung cancer (NSCLC) cells were added to 96-well plates and incubated overnight. The next day, cells were washed and 75 mL of variant protein (crude supernatant or purified IgG) were added to cells along with 25 mL of 2 mmol/L PMA (0.5 mmol/L final) and incubated at 37 C for 4 hours. After incubation, supernatant was removed and added to coated and blocked Nunc maxisorb plates for an amphiregulin ELISA (R&D Systems DuoSet DY262).
ADAM17 capture ELISA. Plates coated with anti-human Fc mAb were used to capture the IgG expressed in the supernatant. Biotinlabeled ADAM17 was added and then the complex was quantitated by neutravidin-HRP.

Results
Screening of a Dyax phage display library for Fabs capable of inhibiting recombinant ADAM17 in a biochemical assay yielded a prominent hit that was converted to the IgG1 antibody 80PH3 which has an IC 50 of 0.46 nmol/L ( Supplementary Fig. S1). In cellbased shedding assays, however, 80PH3 could only inhibit 50% of shedding of ADAM17 substrates at 10 mmol/L ( Supplementary  Fig. S2). The drop in cell-based potency with respect to antibody concentration is consistent with previous reports (14). 80PH3 was therefore optimized using a cell-based shedding assay as the selection criteria leading to the generation of MEDI3622 whose properties are summarized in Table 1 and Fig. 1. The exquisite selectivity of MEDI3622 for ADAM17 is shown by the lack of inhibition of ADAM10, the ADAM family member most closely related to ADAM17 (Supplementary Fig. S3A). Likewise, MEDI3622 shows no inhibitory capacity for MMP12 which is a common off-target activity for ADAM17 small-molecule inhibitors ( Supplementary Fig. S3B).
MEDI3622 potently inhibited amphiregulin shedding in H358 NSCLC cells with an IC 50 of 923 pmol/L (Fig. 1A). This is a 4 log improvement in activity relative to 80PH3. The maximum inhibition of shedding of amphiregulin in H358 was 82%. The 18% of amphiregulin shedding not inhibited by MEDI3622 is most likely due to ADAM17-independent shedding or proteolysis which occurs intracellularly and is not accessible to antibody.
Cellular regulation of ADAM17 shedding activity is a complex process modulated by the ADAM17 redox state (18), protein processing (19), signal transduction pathways downstream of PKC (20), and binding to the naturally occurring ADAM17 antagonist protein TIMP3 (21). ADAM17 sheddase activity is stimulated by the PKC mimetic PMA. To assess whether the structure or accessibility of the active site of ADAM17 is modulated during the activation process, the binding of MEDI3622 to T4-2 breast cancer cells which were either mock or PMA treated was assessed by flow cytometry. ADAM17 activation via PMA treatment resulted in a 3-fold increase in binding by MEDI3622, indicating a significant change in the structure/accessibility of the sheddase during the activation process (Fig. 1B). An increase in ADAM17 cell surface levels as a possible explanation for the increase in binding was ruled out as binding of Mab9301, which binds a nonoverlapping epitope of ADAM17 ( Supplementary Fig.  S4) was unaffected by PMA treatment. These results are consistent with previous observations that accessibility of the active site of ADAM17 to the small-molecule inhibitor DPC333 is modulated by ADAM17 activation status (15).
The potent inhibition of amphiregulin shedding by MEDI3622 suggested that this antibody should be able to inhibit proliferation of EGFR-dependent cell lines which are dependent on this ligand. H292 NSCLC cells express high levels of amphiregulin (22) and are inhibited in xenograft models by cetuximab (23) which competes with ligands for binding to EGFR. MEDI3622 inhibited proliferation of H292 cells with an IC 50 of 25.9 nmol/L (maximum inhibition was 63%; Fig. 1C). The mechanism of action of MEDI3622 in these proliferation assays was confirmed to be inhibition of shedding of EGFR ligands as addition of amphiregulin or TGFa to MEDI3622-treated cells completely restored levels of proliferation in H292 cells to that observed without MEDI3622 treatment ( Fig. 1D and E). Consistent with a mechanism of action involving more broad EGFR pathway inhibition, screening of 33 cell lines from a variety of indications showed a correlation between MEDI3622 sensitivity and cetuximab sensitivity ( Fig. 2 and Supplementary Table S1, R 2 ¼ 0.66). It is noteworthy that these are two-dimensional tissue culture proliferation assays and that ADAM17 likely also sheds and thereby activates EGFR-independent ligands which are drivers of cell growth under different assay conditions and/or in tumors. For example, the known ADAM17 substrate heregulin drives cell proliferation in tumors via HER3 signaling yet has minimal effects on cellular proliferation in tissue culture studies (24). To establish the mouse and human cross-reactivity of MEDI3622, the binding affinity of MEDI3622 to the recombinant ADAM17 proteins was assessed using Biacore analysis. For human ADAM17, the K d for MEDI3622 binding is 39 pmol/L (Table 1), while MEDI3622 binds to mouse ADAM17 with a K d of 132 pmol/L (only 3.4-fold lower affinity). Robust inhibition of cellular mouse Adam17 by MEDI3622 was confirmed by monitoring shedding of TNFa in mouse blood stimulated ex vivo with LPS in the presence of added MEDI3622 or nonspecific IgG (Supplementary Fig. S5A). This was further confirmed in vivo through inhibition of TNFa shedding by MEDI3622 but not control IgG following intraperitoneal injection of LPS (Supplementary Fig.  S5B). Thus MEDI3622 can inhibit shedding of ADAM17-dependent ligands such as TNFa deriving from the stromal component of tumors allowing an assessment of dependency of tumor growth on paracrine signaling mediated by ADAM17 in mouse models.
We next evaluated the in vivo therapeutic activity of MEDI3622 in the OE21 esophageal xenograft model. This model was chosen based on the hypothesis that MEDI3622-mediated reduction of shedding of ADAM17-dependent growth stimulating ligands would inhibit tumor growth driven by these ligands. Relative to other cell lines, OE21 cells display high levels of cell surface amphiregulin, epiregulin, TGFa, and betacellulin as assessed by flow cytometry (Supplementary Fig. S6). Consistent with this hypothesis, MEDI3622 demonstrated a dose-dependent increase in antitumor activity, with 30 mg/kg dose yielding a DTGI of 102% (Fig. 3). The half-life of MEDI3622 is approximately 8 days in rats ( Supplementary Fig. S7). To elucidate the potential mechanism of action of MEDI3622 in this model, EGFR signaling was evaluated in OE21 tumor lysates from mice treated with MEDI3622 using an EGFR kinase array ( Supplementary Fig. S8). These data showed that EGFR phosphorylation was inhibited specifically by MEDI3622 at several sites, including Y1086, Y845, S1070, and Y1173. Inhibition of phospho-tyrosine 1112 of Her-2 was also observed. OE21 tumor lysates were also subjected to a broad kinase array ( Supplementary Fig. S9), where the inhibition of phospho-SRC at Y419 and phospho-YES at Y426 was observed. These kinases have been implicated in the EGFR pathway; indeed EGFR Y845 is phosphorylated by Src (25,26). Despite this clear establishment of EGFR-dependent ligand-driven growth signaling in this tumor model, a role for HER-independent pathways in the OE21 model cannot be ruled out.
Cetuximab is an inhibitor of EGFR in clinical use for head and neck cancer (27) and in KRAS wild-type colorectal cancer (28). Cetuximab and other approved HER pathway antagonists all target the HER pathways at the receptor level and likely suffer from incomplete pathway shutdown (11). It was hypothesized that a combination of MEDI3622 and cetuximab would be additive/synergistic due to pathway inhibition at two distinct points: receptor activity as well as ligand shedding. As shown in Fig. 4A, OE21 tumors partially responded to cetuximab. MEDI3622 treatment resulted in tumor stasis during the dosing phase consistent with previous results, but tumors began to regrow 2 weeks after termination of MEDI3622 dosing, supporting the notion that MEDI3622 causes tumor stasis in this model. Remarkably, the combination of MEDI3622 with cetuximab led to complete tumor regression in all mice. To elucidate the signaling pathways that were modulated, Western blot analyses were performed on tumor extracts from mice that were dosed with MEDI3622, cetuximab, and the MEDI3622 þ cetuximab combination. MEDI3622 was found to modestly inhibit EGFR phosphorylation at all sites evaluated, whereas cetuximab was found to inhibit mainly Y1086, and Y1173 of EGFR (Fig. 4B). Interestingly, the combination of MEDI3622 and cetuximab appeared to not decrease EGFR phosphorylation to a greater extent than that seen with either agent alone. MEDI3622 was found to inhibit p-Src, p-Fak, and p-Akt to a greater extent than cetuximab, however, the combination of MEDI3622 with cetuximab was required to generate the most robust inhibition of p-Akt. Reduction in p-Akt levels is the biomarker which most closely correlated with antitumor efficacy in these studies.
To determine the pharmacokinetic nature of the MEDI3622 exposure that was correlated with inhibition of these pharmacodynamic markers, OE21-tumor bearing mice were dosed on days 0 and 3 with MEDI3622 at 3, 10, and 30 mg/kg, and tumors and serum were analyzed 72 hours after the second dose for Akt and Src inhibition as well as MEDI3622 antibody concentration. MEDI3622 inhibited phospho-Akt and phospho-Src in a dosedependent manner (Fig. 5A). Densitometric analysis demonstrated that at a dose level of 3 mg/kg MEDI3622, Akt, and Src activity was inhibited approximately 67%. The activation was further inhibited to approximately 85% to 90% at a 10 mg/kg dose. The dose of 30 mg/kg did not further decrease Akt or Src activation significantly, which correlated well with the dose response of antitumor activity (Fig. 3). The decrease of Akt and Src activity correlated with an increase in MEDI3622 antibody concentration at each dose level (Fig. 5B). The pharmacokinetic analysis demonstrated that a serum trough MEDI3622 concentration of 46.2 mg/mL (achieved at the 10 mg/kg dosing regimen) led to maximal inhibition of Akt, while MEDI3622 serum trough levels of 82.16 mg/mL (achieved under the 30 mg/kg dosing regimen) led to maximal inhibition of Src.
In addition to the OE21 model, the antitumor activity of MEDI3622 was examined in a variety of xenograft models. These models varied in response to MEDI3622 and included strong responders (Cal27, H292, and OE21) and poor responders (Fig. 6). On the basis of the prior discovery that MEDI3622 inhibited the EGFR pathway in OE21 tumors, tumor extracts from these models were analyzed for phospho-and total EGFR expression by Western blot analysis. Remarkably, the three most   responsive models to MEDI3622 had the strongest levels of phospho-EGFR at Y845, Y1086, and Y1173. OE21 tumors in particular were hyperphosphorylated at all of these sites compared with any other tumor model. These three models also appeared to have the highest total EGFR levels.
On the basis of the observation that esophageal and head and neck xenograft models, OE21 and Cal27 had the best response to MEDI3622, 11 head and neck cancer PDX models were evaluated for antitumor response to MEDI3622 (Table 2). There was a diverse response to MEDI3622 that included strong responders (models 840 and 786) and weak responders (models 432 and 776). Model 840, the best responder, was additionally analyzed for inhibition of HER pathway receptors and kinases through a Western blot evaluation of both total and phosphorylated protein levels (Fig. 7A). In contrast to the Akt inhibition that was observed in the OE21 model upon MEDI3622 treatment, MEDI3622 inhibited primarily total and phospho-MEK in addition to phospho-EGFR while leaving levels of total and phosphorylated Akt relatively unchanged. In all of the head and neck PDX models, total EGFR (H1T) and active EGFR/EGFR homodimer (H11D) levels were assessed by VeraTag analysis (Monogram Biosciences) and percent activated EGFR was calculated as the H11D/H1T ratio ( Table 2). This ratio strongly correlated with MEDI3622 antitumor activity (Fig. 7B, R 2 ¼ 0.69 with outlier model 953 excluded) thereby establishing the H11D/H1T ratio as a pre-dictive biomarker in these models. This correlation exceeded that seen with cetuximab ( Supplementary Fig. S10, R 2 ¼ 0.49). Furthermore, MEDI3622 activity trended with that of cetuximab across most (but not all) models, consistent with EGFRdriven growth signaling as the primary driver of tumor growth in these models and consistent with the findings from xenograft studies where the most responsive models were found to have hyperphosphorylated EGFR.
There remains the possibility that ADAM17-mediated shedding of EGFR-independent growth driving ligands may confer MEDI3622-mediated antitumor efficacy in models insensitive to EGFR-dependent inhibitors. In support of this idea, model 505 had a very low level of EGFR activation (H11D/H1T index of 0.83) yet was inhibited in tumor growth by 56% by MEDI3622. Cetuximab was unable to inhibit growth of this model, and for reasons which are unclear, actually stimulated growth. Taken together, the data indicate that ADAM17 can drive tumor proliferation through both EGFR-dependent and independent mechanisms, both of which can be mitigated by MEDI3622 treatment.
An EGFR-independent mechanism of action could also be involved in the OE21 model in which MEDI3622 antitumor activity is both superior to and additive with cetuximab. The additive pharmacodynamic effect of MEDI3622 and cetuximab on pAKT levels could be a possible explanation for the tumor regressions which occurred when the two antibodies were combined in the OE21 tumor model. However, a lack of additivity of the combination in the inhibition of pEGFR (Fig. 4B), pHER2 (data not shown), or pHER3 (Fig. 4B) suggests that a second pathway which is independent of the HER family members may contribute toward Akt activation in an ADAM17-dependent manner. To address this possibility, shed ligands in cell supernatants from OE21 cells treated with either MEDI3622 or a nonspecific IgG control antibody were analyzed by proteomics in an attempt to identify hitherto unidentified substrates for  ADAM17 which might contribute toward tumor growth in the OE21 model. Hits from this analysis are presented in Table 3 as fold reduction in protein shedding in MEDI3622 treated versus control treated cells. The top two hits were epiregulin and amphiregulin which decreased 14.9-and 10-fold, respectively, thus validating the capacity of the assay to identify known ADAM17 ligands. Other hits previously identified in the literature as ADAM17 substrates included MICA (reduced 4.3-fold), CD166 (reduced 3.4-fold), heregulin (reduced 2.5-fold), ULBP2 (reduced 2.9-fold), and JAG1 (reduced 2.6-fold). Noteworthy among novel hits were several single pass type one plasma membrane proteins, which are members of the protein tyrosine phosphatase family: PTPRS (reduced 6.3-fold), PTPRF (reduced 5.3-fold), PTPRG (reduced 4.1-fold), and PTPRU (reduced 2.9-fold). PTPR family members have been implicated as important tumor suppressors (29). In support of this, PTPRS levels were rescued in MEDI3622-treated tumors in a dose-dependent manner (Figs. 4B and 5A). This suggests that MEDI3622 exposure shifts the equilibrium of PTPRS from inactivated shed protein to active full-length tumor suppressor protein. Therefore, exposure to MEDI3622 reduces the amount of shed PTPRS (as described in the proteomics results in Table 3) and increases the amount of intact cell surface PTPRS (as seen in Figs. 4B and 5A).The increased activity of MEDI3622 versus cetuximab could be a consequence of these events. Unexpectedly, combination treatment with cetuximab further increased the levels of this putative tumor suppressor. This increased rescue of PTPRS correlates with the increased antitumor activity seen with the MEDI3622-cetuximab combination.
Proteomic analysis of OE21 cell supernatants identified numerous potential HER-independent ADAM17 substrates involved in tumorigenesis (Table 3). However the OE21 model is largely EGFR driven, and distinguishing how much antitumor activity of MEDI3622 derives from inhibition of HER-dependent versus HER-independent pathways is problematic. In an effort to understand the mechanistic contribution of ADAM17-dependent HER   Figure 7. A, MEDI3622 activity trends with EGFR homodimerization index in head and neck PDX models. A, tumor lysates from sensitive model CTG-840 were probed by Western blot analysis using samples taken 72 hours after the second of two 10 mg/kg doses on days 0 and 3 using the indicated antibodies. B, DTGI trends with the EGFR dimerization index (H11D/H1T) as further described in Table 2.
family signaling versus HER-independent signaling, several cell lines which express ADAM17 as assessed by flow cytometry (data not shown) were screened for lack of ADAM17-dependent shedding of known HER pathway ligands via proteomic analysis of cell supernatants. Approximately one third of the identified proteins dependent on ADAM17 for shedding in Colo205 were in common with the OE21 analysis (Supplementary Table S2). This confirmed the reliability of the assay. No HER pathway ligands were identified by the analysis of this cell line. The lack of HER pathway dependence of Colo205 was further confirmed by lack of growth inhibition when cells in culture were exposed to either cetuximab or the small-molecule pan-HER inhibitor AZD8931 (ref. 30; Supplementary Fig. S11). To explore the potential for HER-independent MEDI3622 antitumor activity in greater detail, we treated mice bearing Colo205 tumors with MEDI3622 or AZD8931. MEDI3622 treatment resulted in 50% tumor growth inhibition compared with untreated mice (Fig. 8). In contrast, the pan-Her inhibitor AZD8931 had no significant antitumor activity in this model, suggesting that the activity of MEDI3622 in Colo205 tumors is largely HER pathway independent. Consistent with this conclusion, analysis of Colo205 tumor lysates of MEDI3622-treated animals showed no modulation of the activation status of EGFR, HER2, or HER3 ( Supplementary Fig. S12). Elucidation of the mechanism of action of MEDI3622 in this model will require further experimentation.

Discussion
We report here the discovery of a highly specific and potent ADAM17 monoclonal antibody that efficiently inhibits shedding of ADAM17 substrates and proliferation of ADAM17-dependent cells in vitro and tumor growth in vivo. The initial lead antibody was discovered by phage panning using recombinant ADAM17 and functional screening via inhibition of recombinant protein in biochemical assays. Previous efforts to develop ADAM17 inhibi-tors have noted large drops in cell-based potency versus biochemical assays leading to speculation that the structure/accessibility of ADAM17 is modified in the cellular context (12). Because of this, optimization of the initial lead antibody utilized a cell-based functional screen. The PMA-regulated accessibility of the epitope recognized by MEDI3622 is consistent with previous reports demonstrating that phosphorylation of the intracellular domain of ADAM17 orchestrates a transition from inactive dimer to active monomer and the release of the extracellular inhibitory molecule TIMP3 (21).
Much of the inhibitory activity of MEDI3622 in these studies is due to inhibition of the EGFR pathway. In tissue culture, the antiproliferative activity of MEDI3622 was rescued by both recombinant amphiregulin and TGFa. Furthermore, screening of a 33 cell line panel across several indications demonstrated a strong correlation between sensitivity to cetuximab and MEDI3622. Cetuximab sensitivity is predicted by high EGFR ligand levels (31,32), strongly suggesting that high ligand levels predict MEDI3622 sensitivity as well. In support of this, MEDI3622 sensitive cell lines expressed high mRNA levels for at least one EGFR ligand, and no correlation was found between MEDI3622 sensitivity and mRNA levels of ADAM17, EGFR, HER2 or HER3 ( Supplementary Fig. S13). The EGFR pathway was also critical for MEDI3622 activity in vivo. In a panel of head and neck PDX models, MEDI3622 activity correlated with the percent of EGFR which was in the activated dimeric form. Percent activated EGFR therefore likely serves as an indirect readout of high EGFR ligand levels in predicting MEDI3622 sensitivity. Similarly to the 33 cell line panel described above, there was no correlation between sensitivity to MEDI3622 and mRNA levels of ADAM17, HER2 or HER3 in the PDX models ( Supplementary Fig. S14). In the OE21 model, pEGFR was significantly reduced by MEDI3622 exposure. Finally, in a panel of xenografts across several indications, the most sensitive models had higher levels of pEGFR. The involvement of the EGFR pathway in the mechanism of action of MEDI3622 is consistent with genetic data indicating that the Adam17 knockout mouse phenotype bears resemblance to that of the EGFR and HBEGF knockout mice (3)(4)(5). Also noteworthy, however, was the clear potential of MEDI3622 to inhibit EGFR-independent pathways as evidenced by the 56% DTGI in the cetuximab resistant 505 PDX model.
In the OE21 esophageal tumor model, antitumor activity of MEDI3622 was superior to cetuximab. Combination of MEDI3622 with cetuximab led to complete regression of tumors in this model. Although cetuximab is approved for use in metastatic colon cancer and head and neck cancer, the response rates are low (12.6% in the case of head and neck; ref. 27). Incomplete pathway inhibition is thought to contribute to the lack of efficacy of HER pathway inhibitors as the pathways are designed to reactivate following partial inhibition due to the presence of downstream negative feedback loops (11). Complete pathway inhibition blunts pathway reactivation. Inhibition of the EGFR pathway at two distinct points, receptor inhibition and ligand shedding, would be expected to provide a more complete pathway inhibition. Superiority to and additivity with cetuximab could also be explained by inhibition of HER-independent ligands. Although MEDI3622 is additive with cetuximab in the OE21 model, it is not possible to conclude whether the mechanism involves additivity within the EGFR pathway or the inhibition of shedding of HER-independent ligands. The  ambiguity in the mechanistic basis for the additivity is evident when considering the effects of cetuximab and MEDI3622 on p-Akt which is inhibited by each single agent and reduced further by the combination. The lack of additive effect of cetuximab and MEDI3622 on upstream HER pathway pharmacodynamic responses suggests an HER-independent mechanism of additivity. The pharmacodynamic response to MEDI3622 varied in different models, inhibiting p-Akt in the OE21 model and p/t-ERK in the PDX model 840. It is noteworthy that this is consistent with previous EGFR literature and is likely due to model-specific association of EGFR with the Ras and/or PI3K pathways (33). HER-independent ligands that might contribute to the mechanism of inhibition of MEDI3622 were identified via proteomics. PTPRS was a noteworthy hit for four reasons. First, PTPR proteins are mutated with a 30% frequency across 15 types of solid tumors (29). Second, PTPRS is specifically microdeleted in 26% of head and neck cancer patients (34). Third, PTPRS function is modulated by shedding (35) suggesting that its tumor suppressor activity could be abrogated by ADAM17-mediated shedding. Finally, MEDI3622-mediated inhibition of ADAM17 rescues PTPRS in the OE21 model. Identification of substrates modulated by PTPRS and an assessment of the relative contribution of PTPRS rescue to overall MEDI3622 activity in the OE21 model will require additional experiments.
The clearest evidence for HER-independent antitumor activity of MEDI3622 was provided by the Colo205 model. This model was chosen due to the presence of ADAM17 expression, absence of ADAM17-dependent shedding of HER pathway ligands, and resistance to cetuximab and the pan-HER inhibitor AZD8931. MEDI3622 inhibited tumor growth of Colo205 50% in sharp contrast to the pan-HER inhibitor AZD8931 which had no significant activity. Genomic analysis of the response to MEDI3622 will be required to understand the mechanism of inhibition in this model. JAG1, JAG2, DLL1, and FZD6 were identified as ADAM17 substrates in the proteomics studies suggesting a role for the sheddase in the maintenance of stem cells. This concept is bolstered by recent data indicating that ADAM17 colocalizes with LGR5-expressing stem cells of gastric tumors (36), stimulates Notch1 signaling in gastric cancer and glioblastoma stem cells (37), and promotes the cancer stem cell phenotype in colon cancer through shedding of JAG1 (38). The lack of regrowth of the OE21 model when treated with the combination MEDI3622 and cetuximab suggests that cancer stem cells are being killed. If the mechanism were to include one or more of the four ligands identified in the proteomics studies, it is not clear why cetuximab is required to see the effect and this will be an area of future investigation. Future experiments will be necessary to test the requirement for ADAM17 in cancer stem cell maintenance. For example, does the treatment of tumorbearing mice with MEDI3622 eliminate the ability of tumor cells to initiate new tumors? What are the effects of MEDI3622 on growth of three-dimensional sphere cultures which enrich for cancer stem cells? Finally, what is the prevalence of the requirement for ADAM17 for cancer stem cell maintenance beyond the OE21 model?
Overall, the results presented here provide a framework for the exploration of HER-independent mechanisms for ADAM17 requirement in tumor proliferation. Identification of additional models insensitive to pan-HER inhibitors such as AZD8931, yet sensitive to MEDI3622, should prove to be fertile grounds for these efforts in conjunction with proteomic studies to identify novel ADAM17 substrates in these models. It is noteworthy that ligands which are immunosuppressive upon shedding, such as ULBP2 (39) and MICA (40), were among the proteomics hits in the OE21 model (Table 3) and H358 lung cancer cells (Supplementary Table S3). MEDI3622-mediated inhibition of circulating levels of these ligands has the potential to restore CD8 þ T-cell and NK cell function. Future studies utilizing mouse syngeneic tumor models will query an immunostimulatory role for MEDI3622.
The identification of novel ADAM17 substrates involved in critical aspects of tumor proliferation such as stem cell maintenance and immune suppression suggests that translational strategies for ADAM17 inhibitors should focus on the identification of patient tumors reliant on ADAM17 for the shedding of two or more critical ligands. An advantage to such an approach is that patients could be selected on the basis of high circulating levels of these ligands obviating the need for tumor biopsies. Suitably quantitative detection of ADAM17-dependent circulating ligands would similarly enable straightforward clinical dose/schedule optimization of an ADAM17 inhibitor. While determination of which combinations of ligands to prioritize for patient stratification will require additional experimentation, it is possible that some combinations may provide synergies. For example, inhibition of shedding EGFR ligands may promote an immunogenic cell death (41) which is recognized by an immune system no longer suppressed by high levels of shed ULBP2.
In summary, we have discovered an inhibitory ADAM17 antibody with potent antitumor activity which recognizes a novel epitope which is modulated during the ADAM17 activation process. The antibody is likely to have a significant clinical impact as it can contribute additional tumor growth inhibition to existing HER pathway antagonists through two mechanisms. First, targeting ligand shedding is likely to be additive with currently available antagonists which all target HER receptors. Second, inhibition of shedding of HER-independent tumor promoting ligands will likely also enhance the efficacy of existing HER therapeutics as well as offer the opportunity to use MEDI3622 as a single agent.