Development of therapeutic anti-JAGGED1 antibodies for cancer therapy

Abstract

The role of Notch signaling and its ligand JAGGED1 (JAG1) in tumor biology have been firmly established, making them appealing therapeutic targets for cancer treatment. Here we report the development and characterization of human/rat-specific JAG1-neutralizing monoclonal antibodies. Epitope mapping identified their binding to the Notch receptor interaction site within the JAG1 Delta/Serrate/Lag2 domain, where E228D substitution prevented effective binding to the murine Jag1 orthologue. These antibodies were able to specifically inhibit JAG1-Notch binding in vitro, downregulate Notch signaling in cancer cells and to block the heterotypic JAG1-mediated Notch signaling between endothelial and vascular smooth muscle cells. Functionally, in vitro treatment impaired 3D growth of breast cancer cell spheroids, in association with a reduction in cancer stem cell number. In vivo testing showed variable effects on human xenograft growth when only tumor-expressed JAG1 was targeted (mouse models) but a more robust effect when stromal expressed Jag1 was also targeted (rat MDA-MB-231 xenograft model). Importantly, treatment of established triple receptor negative breast cancer brain metastasis in rats showed a significant reduction in neoplastic growth. MRI imaging demonstrated that this was associated with a substantial improvement in blood-brain-barrier function and tumor perfusion. Lastly, JAG1-targeting antibody treatment did not cause any detectable toxicity, further supporting its clinical potential for cancer therapy.

Introduction

Notch signaling is an evolutionary conserved pathway that plays an important role in different physiologic processes and its dysregulation is involved in pathologic conditions, including cancer. In mammals, signaling is initiated by the interaction between one of four Notch receptors (Notch1, 2, 3 and 4) with one of five ligands belonging to the Delta-like or Jagged subfamilies (Dll1, 3 and 4, and Jagged1 and 2 respectively). Upon ligand binding, the receptor undergoes a series of proteolytic cleavages that ultimately lead to the release of its intracellular domain into the cytoplasm and subsequent nuclear translocation, where it acts as a transcriptional regulator (1, 2). Alterations of the pathway have been described in many cancer types and affect multiple aspects of tumor biology, these generally confer oncogenic function but, may also have tumor suppressive activity (1, 3).

Among the Notch ligands, human JAGGED1 (hJAG1) has been closely linked to tumor biology, with involvement in metastasis formation, cancer stem cell (CSC) number, angiogenesis, epithelial-to-mesenchymal transition (EMT), cell proliferation, resistance to therapy and immune function regulation (4). Notably both tumoral and stromal Jagged1 have been reported to play a role, with the latter implicating endothelial cells (5, 6), osteoblasts (7, 8) and myeloid derived suppressor cells (9). JAGGED1 is expressed in many normal tissues and Alagille syndrome, caused by inactivating JAG1 mutations, primarily affects the liver, heart, skeleton, eye, face, kidney and vasculature (10). Due to the multi-functional role played by Notch signaling in several different tumor types, it is not surprising that a variety of therapeutic approaches targeting this pathway have been developed, including both small molecules and neutralizing antibodies. Small molecules are predominantly γ-secretase inhibitors (GSIs), a class of compounds that inhibit the last proteolytic cleavage step during Notch receptor activation. These were originally developed for Alzheimer’s treatment (11), but are under extensive clinical testing for a variety of neoplastic conditions. GSIs are characterized by their ability to inhibit Notch signaling mediated by any receptor-ligand combination, but also by their recognition of additional substrates and severe gastrointestinal toxicity, which currently limit their clinical application (12). A more targeted approach involves the use of monoclonal antibodies (mAbs), and this has been employed to neutralize either individual receptors or ligands. Pre-clinical studies have shown therapeutic potential for mAbs targeting Notch1 (13, 14), Notch 2 and 3 (13, 15), as well as ligands such as Dll4 (16–18) and more recently hJAG1 (8). Based on a similar targeted rationale, Notch ectodomain-based decoys have also been developed and these performed positively in pre-clinical models (19, 20). Clinical testing is underway for some of these approaches (12), but further work is needed to identify the optimal Notch pathway target, the effective agents and particularly the right therapeutic setting.

Aggressive triple receptor negative breast cancer (TNBC) represents an important area of unmet clinical need. Patients present with molecular and clinical heterogeneity, have a high likelihood of relapse and as yet there are no systemic approved standard of care therapies beyond classical chemotherapy. Here we show that a therapeutic monoclonal antibody targeting the Notch receptor binding site on the hJAG1 Delta/Serrate/Lag2 (DSL) domain can inhibit Notch signaling, target TNBC cancer stem cells and reduce tumor growth in vivo. Anti-JAG1 immunotherapy offers promise as a future treatment strategy, both in TNBC and other cancer types.

Materials and Methods

Generation of anti-JAG1 monoclonal antibodies

All in vivo work, in this and in other experiments, was approved by the local ethical review committee and governed by appropriate UK Home Office establishment, project and personal licenses and complied with the Guidelines for the Welfare and Use of Animals in Cancer Research (21). Antibodies were generated by immunization of MF1 mice with the purified JAG1 DSL-EGF1-3 protein (22) and splenocytes were fused with NS0 cells as described previously (23). Hybridoma supernatants were screened for the presence of antibodies that were reactive with the immunogen by ELISA, and positive hybridoma cell lines were cloned by limiting dilution. For further information on antigen and antibody production, ELISA screening, dot blot and Surface Plasmon Resonance analysis, and antibody humanization see Supplementary Materials and Methods.

Cell lines, culture conditions, treatment of 2D and 3D in vitro models

All cell lines and their growth conditions can be found in the Supplementary table 1. All cell lines were routinely tested for Mycoplasma using the Plasmo Test Mycoplasma Detection kit (Invitrogen) every 3 months. All cell lines were used within 15 passages following thawing (7-8 passages for primary human cells). MDA-MB-231 and MDA-MB-231/BR cells were authenticated using STR profiling by LGC Standards. For 2D cell treatment, cells were plated in 6 well plates and 24h later, when cell density was approximately 70-80%, growth medium was replaced with fresh media containing the treatment. 48h later cells were harvested for further analysis.

For the HUVEC-HUVSMC co-culture experiments, a first layer of cells was plated on day 0 followed by a second one when the first reached approximately 75% confluency (generally on day 3). Three co-culture combinations were prepared as follow: HUVEC + HUVSMC (test sample), HUVEC + HUVEC and HUVSMC + HUVSMC (control samples). HUVEC medium was used for all co-cultures and treatments were added together with the second cell layer. 48h later, cells were harvested, labeled with anti-CD31 MicroBeads and separated using LS columns on a MidiMACS separator according to supplier protocol (all from Miltenyi Biotec).

For JAG1- or vector-transduced U87 cells co-culture with the parental line, cells were co-seeded in a 6 well plate and cultured for 5 weeks. Twice per week cells were split and GFP positive populations (representing transduced cells) were quantified by FACS.

For luciferase reporter assays, clear bottom 96-well-plate (CELLSTAR) wells were coated with recombinant Notch ligands in 0.1% BSA-PBS (see Supplementary table 2) overnight before cell plating (4x10⁴ cells/well; treatments were added at this point). LS174T cells expressing the luciferase gene under the Notch transcription factor RbPJ were used (24). Luciferase activity was quantified 24h later by a luminescence assay (Bright-Glo system, Promega). Each condition was performed in triplicate.

For 3D spheroids treatment, cells were harvested, plated at the density of 5x10³ cells/200μl/well in low-adherence 96 well plates (Corning) and spun at 1800rpm for 10 minutes. Plating medium was normal growth medium (see Supplementary table 1) supplemented with 2.5% Matrigel (BD Bioscience). One day after plating 100μl of medium was replaced with fresh media containing a specific mAb at double the final concentration. Medium was then refreshed every 2 days by replacing half the volume and growth was monitored by taking a picture of each single spheroid. Ten spheroids per condition were analyzed in every experiment. Spheroid volume was then calculated with ImageJ software, assuming perfect sphericity.

All antibodies, including the mouse isotype control IgG1 (R&D Systems), were used at a concentration of 10μg/ml if not specified, DBZ (Calbiochem) was used at a concentration of 100nM and DMSO (Sigma-Aldrich) was used 1:1000.

RNA extraction, reverse transcription and quantitative PCR

Total RNA from cell cultures was isolated using the RNeasy mini kit (Qiagen) according to the manufacturer’s instructions. For ex vivo material, xenograft samples were powdered before RNA extraction. Complementary DNA (cDNA) was synthesized from 0.5-1μg of total RNA using Superscript III first-strand system (Invitrogen). Quantitative PCR (qPCR) analysis was performed in triplicate using the SYBR GreenER qPCR SuperMix Universal (Invitrogen) and Chromo4 fluorescence detector (MJ Research/Bio-Rad). Relative quantification was done using the ΔΔCt method normalizing to housekeeping gene expression (β2-Microglobulin and β-Actin for human and rat samples respectively). For primer sequences see Supplementary table 3.

Mouse xenograft experiments

1x10⁷ tumor cells (U87-vector, U87-JAG1, PC3, MDA-MB-231, and OVCAR3) in 100μl Matrigel (BD Bioscience) were injected subcutaneously into the flank of BALB/c nu/nu female mice (Crl:NU-Foxn1^nu, Charles River Laboratories). JAG1 blocking mAb at the indicated concentration, or an equal volume of PBS were injected on same day intraperitoneally and then twice every week. Tumor volume was calculated as L x W x H x π / 6 (25). For treatment of established OVCAR3 xenografts, tumors were grown to ~100mm³ and grouped into 2 arms of similar size (100mm³) and distribution before twice weekly treatment with J1-65D (20mg/kg).

Rat subcutaneous xenografts

The subcutaneous MDA-MB-231 xenograft experiment was performed by Charles River Discovery Services in compliance with the Guide for Care and Use of Laboratory Animals and accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International. Briefly, 3 days before cell implantation, Taconic female rnu/rnu rats were intraperitoneally administered with 150mg/kg cyclophosphamide to favor tumor engraftment. On day 0, cells were subcutaneously injected (1x10⁷/cells/rat in PBS-50% Matrigel) and J1-65D antibody treatment (20mg/kg) was then administered intravenously twice per week starting from day 2 (15 animals/group). Tumor volume and body weight were measured twice per week. Tumor volume was calculated as L x W² / 2. Blood sampling was performed under anesthesia on day 20 and at the time of culling (when one of the groups reached an average size ≥ 10000mm³).

Rat brain metastasis model

Female nude rats were anaesthetized with 2-3% isoflurane in N₂:O₂ (70:30), placed in a stereotactic frame and focally microinjected in the left striatum (+1.2mm and 2.5mm lateral to Bregma, at a depth of 6.5mm) with 1x10⁴ MDA-MB-231/BR cells in 1μl of sterile PBS using a 75μm-tipped glass microcapillary (Clark Electromedical Instruments).

hJ1-65Dv9 antibody treatment (20mg/kg) was administered intravenously twice weekly starting from day 18 (6 animals/group) until week 7 post cell injection. At this point animals were sacrificed and transcardially perfusion-fixed under terminal anesthesia as previously described (26). The brains were post-fixed, cryoprotected, embedded in tissue-tek (Sakura Finetek Europe) and frozen in isopentane at -40˚C. The isotype control antibody used was human anti-fluorescein IgG1 (Absolute Antibody Ltd). For Magnetic Resonance Imaging analysis and tumor volume reconstruction see Supplementary Material and Methods.

Statistical analysis

Statistical analysis was performed using GraphPad Prism 6 and the tests used are reported in the figure legends. Results are presented as mean ± SD or mean ± SEM. P ≤ 0.05 was considered to be statistically significant.

For information on expression constructs, flow cytometry and histologic analysis see Supplementary Material and Methods (including Supplementary tables 4 and 5 regarding antibodies used for flow cytometry and immunohistochemistry respectively).

Results

JAG1 antibody production and epitope mapping

To generate antibodies that specifically block hJAG1-induced Notch signaling, the hJAG1 DSL domain and the neighboring 3 EGF domains (amino acids 185-335, Fig.1A) (22) was used to immunize mice. Hybridomas were generated using classical techniques and secreted mAbs were tested for reactivity with hJAG1 (Fig.1B). Five mAbs, J1-142B, J1-65D, J1-156A, J1-183D and J1-187B, bound cell surface hJAG1 on overexpressing cells. Specificity testing showed no effective binding to other Notch ligands including human JAGGED2 (hJAG2) and hDLL4 (Fig.1B). Specificity was also confirmed by immunocytochemical labeling of hJAG1 but not hJAG2 transfected HEK293 cells (Supplementary Fig.1A). Only J1-142B showed effective binding to the murine Jagged1 (mJag1) orthologue (Fig.1B).

Figure 1. Anti-JAG1 monoclonal antibody generation, binding specificity and epitope mapping.

(A) Domain organization and structure of the N-terminal region of JAG1. The region comprising the DSL domain and the three neighboring EGF repeats was used as the immunogen for neutralizing antibody production. (B) Binding specificity of human JAG1 mAbs. The mAbs were used to stain HEK293 cells transfected with hJAG1 or hJAG2 and B16F10 cells transfected with hDLL4 or mJag1. The expression of hJAG2 was verified using an anti-JAG2 mAb, and that of hDLL4 and mJag1 by GFP expression from the expression constructs. (C) The hJAG1 DSL domain alone or DSL-EGF1-3 protein was used to coat ELISA plates and binding of the anti-JAG1 mAbs was tested by ELISA. (D) Residues in the hJAG1 DSL domain that are substituted in mJag1 are highlighted in bold. Notably the hJAG1 sequence is also conserved in the cynomolgus monkey (cJAG1), which is widely used in toxicology studies. Note the proximity of these residues (purple; Y190, E228, R231) to the residues shown to be important for binding to Notch (blue; F199, R201, R203, R207). (E) The amino acids at positions 190, 228 and 231 in hJAG1 were each mutated to the mJag1 sequence. These soluble DSL-EGF1-3 recombinant proteins were used in dot blots to identify the amino acids responsible for preferential mAb binding to the hJAG1 protein.

To identify the epitopes recognized by the antibodies, their binding to the original immunogen (hJAG1 DSL-EGF1-3) was compared to the hJAG1 DSL domain alone. The murine cross-reactive antibody J1-142B was shown to require the EGF domains to bind since, unlike the other four mAbs, it was unable to bind to the DSL domain alone (Fig.1C). The DSL domain and adjacent EGF1 sequence of human and cynomolgus monkey JAG1 are identical and are also highly conserved in the mouse where only three amino acids (aa) differ between the orthologues (Fig.1D, aa 190, 228 and 231). To investigate their contribution to human specific binding by the four JAG1 DSL domain targeting mAbs, these three amino acids in the hJAG1 DSL-EGF1-3 recombinant protein were individually mutated to their murine counterpart. Antibody binding to the mutants in a dot blotting assay indicated that mAb binding was unaffected by Y190H and R231K substitutions, while E228D mutation completely abolished the binding of J1-65D, J1-156A, J1-183D and J1-187B (Fig.1E).

Thus, in addition to forming part of the epitope for these DSL domain-targeting mAbs, E228 is also the residue that confers their human specificity, and has been shown to contribute to the JAG1 DSL/Notch1 interface (27). While alanine substitutions of other DSL domain residues also shown to interact with Notch1 (F199, R201, R203, F207) (Fig.1D) had no effect on J1-142B and J1-183D mAb binding, they differentially inhibited the binding of the other three mAbs (Supplementary Fig.1B). The DSL domain-targeting mAbs thus recognize distinct epitopes that include residues that play a key role in forming the DSL-Notch1 ligand/receptor binding interface (22, 27).

Surface Plasmon Resonance was performed to quantify the binding affinity of J1-156A, J1-65D, J1-183D, and J1-187B mAbs towards the immunizing hJAG1 protein. J1-65D and J1-183D exhibited the highest binding affinity, having dissociation constants (Kd) of 9.7nM and 4.9nM, respectively (Supplementary Fig.1C).

JAG1 antibodies specifically block JAG1-activated Notch signaling

The anti-JAG1 mAbs were assessed for their functional ability to disrupt the hJAG1-Notch1 protein-protein interaction using a FACS-based binding assay (Fig.2A). Three antibodies, J1-65D, J1-183D, and J1-156A completely blocked the binding of hJAG1-overexpressing cells to human Notch1 (EGF11-13) coated fluorescent beads, but did not block binding to cells expressing mJag1.

Figure 2. Anti-JAG1 antibodies inhibit Notch receptor binding and signaling.

(A) Anti-JAG1 mAbs block hJAG1 binding to hNotch1. Soluble Notch1 (EGF11-13) coated purple fluorescent avidin beads were used to stain hJAG1 or mJag1 overexpressing cells (HEK293 and B16-F10 respectively) in the presence of JAG1 mAbs. Blocking shifts the bold line towards the shaded grey control peak on the left. (B-D) Human colon cancer cells expressing a luciferase reporter gene under Notch/RBPJ control were stimulated with coated recombinant human JAG1 (rhJAG1), JAG2 (rhJAG2), DLL4 (rhDLL4) or control protein (IgG2b). Cells were contemporarily treated with different J1-mAbs or controls (mIgG1/DMSO are negative controls while DBZ is a positive control) and luciferase activity was then analyzed 24h after plating. Bars show the average of 6 technical replicates of representative experiments. (E) FACS analysis showing surface expressed hJAG1 (J1-65D mAb binding) on a panel of five human tumor cell lines (J1-183D mAb in Suppl. Fig. 2). (F) qPCR analysis showing the in vitro effect of J1-mAbs on Notch-target gene expression in the same panel of cell lines described in (E). Bar graphs show the average ± SD of n = 5 (paired Student’s T-test; * P < 0.05, ** P < 0.01 and *** P < 0.001). mIgG1 is the isotype-matched negative control for all mAbs, DBZ is a pan-Notch inhibitor and DMSO is its corresponding vehicle control.

The same assay was performed with titrated purified mAbs, to compare the concentrations required to neutralize the receptor-ligand interaction. With the exception of the murine cross-reactive mAb J1-142B, the remaining mAbs fully blocked Notch1 binding to hJAG1 at ≤ 5μg/ml, with J1-65D and J1-183D blocking at 1μg/ml (Supplementary Table 6). J1-142B was not evaluated further. Sequences for the variable region of J1-65D, J1-183D, J1-156A and J1-187B mAbs are provided in Supplementary Table 7.

To verify the mAbs were capable of Notch signaling inhibition, LS174T cells expressing luciferase under RbPJ/Notch control were stimulated with individual Notch ligands in the presence of our mAbs versus an IgG control. All four antibodies inhibited Notch signaling induced by recombinant hJAG1 (Fig.2B) at levels comparable to pan Notch inhibition with GSI (DBZ). There was no effect on human JAG2- or DLL4-induced Notch signaling, further confirming ligand specificity of the lead mAbs with greatest functional activity, J1-65D and J1-183D (Fig.2C and D).

We screened a panel of human tumor cell lines for surface expression of JAG ligands and Notch receptors, and prioritized five for further mAb testing. These expressed both JAG1 and at least one Notch receptor, and represented different tumor types in which JAG1 expression has been implicated in neoplastic growth (4): MDA-MB-231, HCC1143 (both TNBC), OVCAR3 (ovarian), PC3 (prostate) and H1993 (lung) (Fig.2E and Supplementary Fig.2A).

In vitro treatment of these cell lines with either the J1-65D or J1-183D mAb or DBZ showed a broad spectrum of Notch target gene regulation that did not strongly correlate with the expression level of any of the individual Notch receptors or ligands (Fig.2F and Supplementary Fig.2B).

Despite a role in vascular biology (28–30) and both primary endothelial and vascular smooth muscle cells (HUVEC and HUVSMC respectively) expressing hJAG1, neither substantially responded to anti-JAG1 mAb treatment (Supplementary Fig.3A-D). As hJAG1 might play a more important role in heterotypic vascular cell interactions, HUVEC and HUVSMC were co-cultured in the presence of J1-65D or isotype control antibody and then the two cell types were separated based on the exclusive HUVEC CD31 positivity and compared to homotypic cultures (Supplementary Fig.4A-D). Co-culture induced Notch-target gene expression (HEY2 and HEYL) in both cell types, while also up-regulating maturation markers in smooth muscle cells (ACTA2 and SMMHC). Importantly, several changes were significantly inhibited by J1-65D treatment, confirming signaling between endothelial-smooth muscle cells as a potential target (Supplementary Fig.4E).

JAG1 antibodies inhibit MDA-MB-231 3D cell growth in vitro

Neither the JAG1 mAbs nor DBZ affected tumor cell line growth/viability in normal 2-dimensional (2D) culture conditions (Supplementary Fig.5A and B). MDA-MB-231 TNBC tumor cells, which exhibited high-level JAG1 expression and significant Notch target gene modulation in 2D culture were grown as 3D spheroids in suspension culture to better mimic tumor biology. Both JAG1 mAbs and DBZ treatment significantly inhibited MDA-MB-231 3D spheroid growth (Fig. 3A). This was accompanied by a dramatic inhibition of expression of the Notch-target HES1 and a reduction in other genes with an important role in breast cancer growth such as IL6 (31) and CA9 (32) (Fig.3B and C) that was dose-dependent (Supplementary Fig.5C and D). Interestingly, JAG1 inhibition was as effective as DBZ at significantly reducing two independent cancer stem cell populations in MDA-MB-231, defined by CD44⁺/CD24^- and Aldefluor⁺, respectively (33) (Fig. 3D).

Figure 3. In vitro effects of JAG1 inhibition on MDA-MB-231 3D growth, gene expression and cancer stem cells.

(A) Effect of J1-65D and J1-183D on MDA-MB-231 spheroid growth (average ± SD of 10 spheroids/group; nonlinear fit test; one representative experiment shown). (B) qPCR analysis on the RNA extracted from treated spheroids (average ± SD of n ≥ 5; paired Student’s T-test). (C) Representative immunohistochemistry images showing reduction in HES1 protein expression in J1-65D-treated versus isotype control-treated spheroids. (D) FACS analysis showing reduction of two distinct cancer stem cell subpopulations (CD44high/CD24low and Aldefluor+) in treated spheroids (average of n ≥ 3; paired Student’s T-test). mIgG1 is the isotype-matched negative control for both mAbs, DBZ is a pan-Notch inhibitor and DMSO is the corresponding negative control (vehicle). * P < 0.05, ** P < 0.01, and *** P < 0.001.

J1-65D inhibits in vivo tumor growth in some mouse xenograft models

Since our lead antibodies were hJAG1 specific, we initially used a cell line model (U87 glioma, which lacked endogenous JAG1) where ectopic hJAG1 expression alone caused accelerated tumor growth (Fig.4A). This provided a model system to identify the antibody dose needed to fully neutralize hJAG1 activity in a subcutaneous tumor, identified as the dose that reverted growth to that of the JAG1-negative parental cell line. In an in vitro co-culture system, JAG1 overexpressing but not vector-transduced U87 cells, were able to out-grow parental cells over time (Fig.4B), and also showed a significant growth advantage in vivo (Fig.4C).

Figure 4. J1-65D mAb prevents hJAG1-induced growth in mouse U87 xenograft tumors.

(A) FACS analysis showing JAG1 surface expression in U87-vector and U87-JAG1 cells (control and hJAG1-overexpressing cells respectively). Staining was performed using the J1-65D mAb. (B) U87 cells overexpressing hJAG1, but not control U87-vector cells, show a growth advantage over parental cells in an in vitro co-culture model. FACS analysis was used to quantify U87-vector and U87-JAG1 cell number (both GFP+) over parental cells (GFP-). Dotted lines show the ratio of 50%. (C) U87 cells overexpressing hJAG1 display accelerated tumor growth in vivo (top) and reduced animal survival (bottom) compared to control cells (n = 6/group). (D) JAG1-neutralizing mAb J1-65D partially prevents hJAG1-induced growth acceleration (top) and animal reduced survival (bottom) in U87-JAG1 xenografts at 10 mg/kg (U87-Vector: n = 7, other groups: n = 6). (E) JAG1-neutralizing mAb J1-65D completely prevents hJAG1-induced growth acceleration (top) and normalizes animal survival (bottom) in U87-JAG1 xenografts at 20 mg/kg (U87-Vector: n = 7, other groups: n = 10). All tumor growth graphs show average ± SEM (Student’s T-test). * P < 0.05, ** P < 0.01. NS, no significance.

To evaluate the effect of hJAG1 blockade in this model, J1-65D (which performed best among the mAbs across our in vitro assays) was administered i.p. twice weekly into tumor-bearing mice from the time of cell injection. At 10mg/kg J1-65D only partially delayed JAG1-induced U87 growth (Fig.4D), while 20mg/kg completely abolished the growth promoting activity of hJAG1 overexpression (Fig.4E).

Having demonstrated effective in vivo hJAG1 blockade, three cell lines with similar hJAG1 levels (Fig.2E) where in vitro experiments showed mAb-mediated modulation of Notch signaling were tested for antibody activity in vivo. In a preventative setting, where twice weekly mAb treatment was initiated on the day of tumor inoculation, J1-65D treatment showed no effect on PC3 xenograft growth, a modest reduction for MDA-MB-231, and significant inhibition of OVCAR3 growth (Fig.5). J1-65D did not inhibit the growth of established OVCAR3 tumors (Supplementary Fig.6A and B).

Figure 5. Differential effects of JAG1 mAb J1-65D treatment on the growth of tumor xenografts in mice.

J1-65D treatment effect on s.c. tumor growth for: (A) prostate cancer cell line PC3 (n = 5/group), (B) breast cancer cell line MDA-MB-231 (n = 10/group), and (C) ovarian cancer cell line OVCAR3 (n = 5/group). J1-65D mAb (20 mg/kg) or PBS (control) were administered twice a week starting from the day of tumor inoculation. For each cell line, average tumor volume ± SEM. is shown in the top panel, and individual tumor growth curves are shown in the bottom 2 panels. Student’s T-test. * P < 0.05.

J1-65D reduces MDA-MB-231 tumor growth in rat xenografts

To evaluate both the potential safety and additional benefits of stromal targeting to therapeutic efficacy we repeated the MDA-MB-231 xenograft experiment in nude rats. Rat Jag1 (rJag1) lacks the E228D alteration that prevents effective binding of our mAbs to mJag1 (Supplementary Fig.7A), and these effectively inhibit rJag1-induced Notch signaling (Supplementary Fig.7B and C), offering the opportunity to test both tumor and host stromal JAG1 inhibition.

J1-65D treatment (20 mg/kg twice weekly i.v. from day 2 post cell injection) significantly delayed the growth of MDA-MB-231 subcutaneous xenografts in immunocompromised rats showing superior efficacy to that observed with only tumor hJAG1 targeting in mice (Fig.6A, B and Fig.5B). We observed significant reductions in the expression of Notch-target gene HES1 and the cancer stem cell marker ALH1A1 (34) by tumor cells (Fig.6C). Histological analysis showed that J1-65D-treated tumors were less necrotic (Fig.6D), perhaps due to their reduced growth.

Figure 6. J1-65D treatment reduces MDA-MB-231 subcutaneous tumor growth in rats without any discernable toxicity.

(A) MDA-MB-231 tumor growth is reduced by J1-65D treatment in a subcutaneous rat xenograft model (average ± SEM of n = 15/group; nonlinear fit test). Individual tumor growth is shown in (B) for both groups. (C) qPCR analysis of tumor RNA showed strong downregulation of the human Notch-target gene HES1 and the stem cell marker ALDH1A1 (n = 4 and 5 for PBS and J1-65D groups respectively; unpaired Student’s T-test). (D) Hematoxylin and eosin staining of the tumors showed important necrosis reduction in J1-65D-treated animals (n = 5/group). (E) No effect on animal weight was observed during the course of the treatment (n = 15/group). (F) Histologic analysis of rat intestines showed no effects of J1-65D on goblet and proliferative cells number (Alcian blue and Ki67 staining respectively; n = 5/group). (G) All bar graphs represent averages ± SD (unpaired Student’s T-test). * P < 0.05, ** P < 0.01, *** P < 0.001.

Importantly, aware of the toxicity reported for several anti-Notch therapeutics (12), we evaluated animal health by monitoring body weight, histology of the intestine and performing extensive blood testing during and at the end of the experiment. Treated animals showed no visible signs of toxicity, no altered body weight (Fig.6E) and no effect on intestinal goblet cell numbers or proliferation (Fig.6F and G). Blood analysis did show some statistically significant differences between the 2 groups (Supplementary Table 8). These were generally small and showed opposite trends at the two time points analyzed, probably primarily reflecting differences in tumor size and disease progression.

J1-65D treatment strongly inhibits breast cancer brain metastasis growth

Metastatic TNBC is generally considered to be an incurable disease, and shows a predilection for the brain and lung as metastasis sites compared to other breast cancer sub-types (35). To address this unmet clinical need we moved away from the earlier preventative s.c. experimental settings and focused on a therapeutic brain metastasis model, with characterized JAG1 expression (Supplementary Fig.8A) where involvement of the Notch/JAG1 pathway had been reported (36, 37). MDA-MB-231/BR cells (variant having a brain tropism) were injected into the striatum of immunocompromised rats. Established tumors were then treated twice weekly i.v. from day 17/18 with control antibody or a humanized IgG1 variant of J1-65D (hJ1-65Dv9) having enhanced affinity (13.9pM) for JAG1 (sequence for both heavy and light chains in Supplementary Table 7). At this time point the tumor is detectable by T₂-weighted MRI, and blood-brain-barrier (BBB) breakdown is imminent (already detectable at day 21, Fig.7D). Tumor growth and vascular function were assessed weekly by MRI and histologically at the end of the experiment (Fig.7A and B). Assessment of hyperintense regions evident on T₂-weighted MR images suggested a reduced neoplastic growth (Fig.7B) and histological 3D reconstructions subsequently confirmed a significant reduction in tumor volume in animals treated with hJ1-65Dv9 compared to control group (Fig.7C). Gadolinium-enhanced MRI, to assess BBB integrity, showed that the volume of compromised barrier in control animals progressively increased over time, as expected. Interestingly, JAG1 inhibition drastically reduced BBB breakdown, with animals showing no sign of increased signal despite similar volumes of T₂ hyperintensity to controls at week 6 by MRI (Fig.7D). Cerebral blood flow (CBF) was also assessed and showed a progressive reduction in control group tumors over time, whilst stable flow was observed in hJ1-65Dv9-treated animals and in contralateral normal brain for both groups (Fig.7E and Supplementary Fig. 8B and C). Unsurprisingly, we found significant correlations between tumor size and both BBB breakdown (positive correlation) and tumor perfusion (negative correlation) in control animals. Interestingly, no significant correlations were observed for tumors treated with hJ1-65Dv9 (Supplementary Fig. 8D), indicating that the differences observed for BBB breakdown and blood flow cannot be explained by the reduced tumor growth caused by Jagged1 inhibition.

Figure 7. Anti-Jagged1 treatment reduces MDA-MB-231-BR tumor growth in a rat brain metastasis xenograft model.

(A) Scheme of an experiment mimicking the therapeutic treatment of an established breast cancer brain metastasis with the hJ1-65Dv9 humanized/de-immunized antibody. (B) Representative T₂-weighted MR images of hJ1-65Dv9 and control-treated MDA-MB-231-BR xenograft tumors (red arrows) at 3- and 7-weeks post cell implantation. (C) Analysis of hyperintense regions on T2-weighted MR images suggests reduced growth of hJ1-65Dv9-treated tumors (top panel; average size ± SEM; n = 6 and 5 for hJ1-65Dv9 and control respectively; nonlinear fit test). Metastasis volume at endpoint (week 7) was assessed histologically and confirmed reduced growth upon Jagged1 inhibition (bottom panel; average size ± SD; n = 6/group; unpaired Student’s T-test). (D) BBB leakage/permeability assessed by gadolinium-based T1-weighted MRI shows reduced BBB breakdown in hJ1-65Dv9-treated animals (average ± SEM; n = 6 and 5 for hJ1-65Dv9 and control respectively; nonlinear fit test). (E) MRI-based perfusion analysis shows that Jagged1 inhibition stabilized tumor perfusion whilst control group tumors showed a progressive reduction in perfusion over time (average ± SD of tumor values normalized to normal contralateral brain; n = variable between 2 and 6 per data point; 2way ANOVA). * = P < 0.05; ** = P < 0.01; *** = P < 0.001.

Discussion

Notch signaling has a well-established role in tumor biology (1), and JAG1 is the ligand with the broadest demonstrated involvement, with reported roles in several aspects of cancer growth across a variety of different tumor types (4). Our aim was to generate neutralizing mAbs to target the receptor-binding region of hJAG1 (22), in order to inhibit this specific Notch signaling axis, while sparing the others and therefore avoid toxicities often associated with broader pathway inhibition (12).

Here we report the successful generation and characterization of specific anti-human/rat JAG1 mAbs, which target the receptor binding site within the DSL domain of the ligand (22). Selected antibodies were able to inhibit Notch signaling in vitro in a variety of tumor cell lines proving that JAG1 activates homotypic cell interactions between cancer cells, contrary to what has been suggested by others (8, 38). Interestingly, the level of inhibition varied, in terms of the magnitude, and the identity of the genes involved, depending on the cell line. Generally, pan Notch inhibition with DBZ showed a greater magnitude of gene expression changes and the number of affected genes per cell line. JAG1-specific blockade also significantly inhibited Notch target gene expression in four out of five cell lines proving that both J1-65D and J1-183D mAbs inhibit endogenous JAG1-mediated Notch signaling. Interestingly, no individual gene was consistently affected in all lines indicating that JAG1 triggers cell-type specific programs. Intriguingly, JAG1 blockade also up-regulated at least one Notch target gene in four out of five cell lines, indicating that this ligand may work contemporarily as both an agonist and antagonist in the same cell type, as previously reported for endothelial cells (30). This was not predictable based on the cell surface expression of Notch pathway components, including hJAG1 itself.

Vascular cells proved resistant to mAb treatment in homotypic cell cultures but showed significant response when treated in co-culture, confirming that hJAG1 is an active ligand that mediates endothelial-vascular smooth muscle cell interactions (29). As previously reported by others (14), cell line treatment in 2D did not affect cell viability (including pan Notch inhibition by GSI treatment) despite the significant effect on gene expression. Growth was however reduced when cells were grown in 3D, indicating that Notch signaling plays a more important role in this condition that more resembles in vivo tumor growth. Notch signaling, including aspects mediated by JAG1, has been reported to regulate the stem cell-like phenotype in breast cancer (33, 39, 40). Treatment with our mAbs confirmed this by reducing CSC numbers in MDA-MB-231 cells.

In vivo testing showed that in a U87 model with ectopic hJAG1 expression, our lead antibodies were extremely effective, being able to completely inhibit hJAG1-induced enhancement of tumor growth. However, inhibition of endogenous tumor-expressed JAG1 alone (mouse xenograft models in which our mAb cannot inhibit host Jagged1) exhibited highly variable efficacy, reminiscent of the variation within in vitro gene expression changes caused by anti-JAG1 antibody treatment. This and the ability to prevent the engraftment of the ovarian OVCAR3 cell line, but not impair the growth of established OVCAR3 tumors, suggested that the antibodies should be evaluated in models where the host stroma could also be targeted, in order to fully assess their therapeutic potential.

JAG1 antibody treatment of MDA-MB-231 xenografts (partially responsive in mice) implanted into nude rats subsequently demonstrated a strong growth inhibitory effect. Confirming the important roles played by Jag1 in stromal cells (5–8) and indicating that inhibition of stromal Jag1 is necessary to achieve significant therapeutic benefits in this model. Importantly, targeting rJag1 in normal tissues did not show any evidence of toxicity, both at the level of general health and in subsequent histologic/cytological studies. The lack of gastrointestinal toxicity is consistent with inducible genetic deletion of mJag1 being dispensable for intestinal stem cells homeostasis (41).

The role of JAG1 in breast cancer has been broadly established (4, 42, 43), and despite important progress in therapeutic management of the disease, subtypes still show poor patient outcome due to treatment resistance and metastatic spread (44). TNBC in particular, shows higher metastatic affinity for the brain, making treatment harder and reducing patient life expectancy (35). The clinical potential of targeting JAG1 in TNBC bone metastasis has been demonstrated (8), here we evaluated whether this might also be true for brain metastasis, a process in which Jagged1-Notch signaling has already been implicated (36, 37, 44). In a model of established brain metastasis, treatment with the hJ1-65Dv9 mAb significantly reduced neoplastic growth, preserved blood-brain-barrier function and tumor perfusion. These findings indicate that Jag1 neutralization has a protective effect on the tumor-associated brain vasculature. Interestingly, this does not seem to be solely a consequence of reduced tumor burden, since hJ1-65Dv9-treated tumors maintained better vascular function than size-matched tumors from the control IgG group at earlier time points. Inhibition of Notch signaling would be expected to negatively affect barrier establishment and function (45), but the latter could be preserved indirectly by inhibiting Jag1-induced aberrant angiogenesis (46, 47). Overall, this improvement in tumor vascular function might indicate a form of vascular normalization, and as such could be exploited to improve the efficacy of other therapeutic approaches that are normally impaired by poor tumor perfusion, such as radiotherapy, chemotherapy and immunotherapy (48).

The different therapeutic outcomes in experiments performed in mice and rats are most likely to derive from the targeting of both tumoral and stromal Jagged1 in nude rats. However, a number of experimental variables, including the route of antibody administration, cyclophosphamide pre-conditioning and use of the humanized Jagged1 antibody could contribute to maximizing the therapeutic efficacy of JAG1 neutralizing antibodies. The ongoing development of a mouse in which the DSL domain has been humanized will enable this to be tested during future preclinical development.

Further studies are warranted to evaluate the potential for combination therapies involving our JAG1 antibodies. Tumor vasculature that is immature and poorly covered by pericytes is sensitive to VEGF-targeting antiangiogenic therapy (49). The discoveries that endothelial specific genetic Jag1 depletion was associated with poor vessel coverage by VSMC (29) and that our JAG1 antibodies target heterotypic Notch signaling between endothelial and vascular smooth muscle cells suggests that anti-JAG1 mAb therapy may benefit patients who are refractory/relapsed to Bevacizumab. It will be important to identify the most relevant patient groups and to identify biomarkers of response that encompass the heterogeneity we have observed in Notch target gene regulation. The recent definition of at least four molecular subtypes within TNBC (BL1, BL2, M and LAR) has demonstrated that the BL1 subtype responds most effectively to chemotherapy, while BL2 and LAR are thought to be more likely to present with residual chemo-resistant disease (50). Thus, identifying patients with particular TNBC subtypes may further stratify those that would benefit most from an additional therapeutic approach, such as inhibition of Jagged1 signaling. Importantly the data shown here, in association with supportive findings from other groups (8), clearly demonstrate the clinical potential of JAG1 neutralizing antibodies for cancer therapy, including the treatment of metastatic breast cancer.