AMG 757, a Half-Life Extended, DLL3-Targeted Bispecific T-Cell Engager, Shows High Potency and Sensitivity in Preclinical Models of Small-Cell Lung Cancer

Small-cell lung cancer (SCLC) is an aggressive neuroendocrine tumor prone to early metastasis, accounting for 10%–15% of all lung cancers (1–3) and associated with poor 5-year survival (4). Disease relapse and resistance to therapy are common following an initial response to etoposide and platinum-based chemotherapy with or without radiotherapy (5). Immune checkpoint blockade has increased overall survival (OS) in SCLC despite relatively modest response rates (6–9). The anti-programmed cell death-1 (anti–PD-1) antibodies nivolumab and pembrolizumab received accelerated approval in the United States for the treatment of metastatic SCLC with progression on or after platinum-based chemotherapy and at least one other line of therapy (10, 11); however, subsequent studies did not confirm increased OS (12, 13). The promise of targeted therapies has also not yet been realized in SCLC; a DLL3-targeted antibody–drug conjugate with early evidence of clinical activity demonstrated no benefit in a subsequent phase III trial (14). Therapeutic agents with different mechanisms of action are still urgently needed for patients with SCLC.

Bispecific T-cell engager (BiTE) molecules are a clinically validated therapeutic modality that redirect a patient's T cells to kill tumor cells (15). Blinatumomab, the first BiTE molecule in clinical development, is approved for the treatment of relapsed/refractory B-cell precursor acute lymphoblastic leukemia (16–18) and also demonstrates efficacy in non-Hodgkin lymphoma (19–21). The BiTE molecule AMG 420, which targets BCMA, demonstrated a 70% response rate in multiple myeloma with 50% minimal residual disease–negative complete responses at the maximum tolerated dose in a phase I study (22). In solid tumors, therapeutic index has been a major challenge for the successful development of T-cell bispecific antibodies (23, 24). Development of AMG 110 (MT110), which targets EpCAM, was discontinued due to on-target, dose-limiting toxicity in the liver and gastrointestinal tract (24). The clinical activity of the CEA-targeting T-cell bispecific antibody, cibisatamab, is also associated with dose-limiting on-target toxicity in gastrointestinal tissues (23). These data highlight the need to identify therapeutic targets with tumor-specific or tumor-selective expression profiles (23, 24).

To identify potential BiTE molecule targets in SCLC, we profiled predicted cell surface proteins for differential expression in a panel of SCLC tumors versus an array of normal tissues and identified the Notch ligand, delta-like ligand 3 (DLL3). DLL3 is expressed during embryonic development (25–28), and together with achaete-scute complex homolog 1, a transcription factor that regulates DLL3 expression, is required for neuroendocrine differentiation and SCLC tumorigenesis (29–31). We generated BiTE molecules targeting DLL3 to explore T-cell–redirected lysis against SCLC tumors. We initially evaluated an anti-DLL3 BiTE molecule in the original tandem single-chain variable fragment (scFv) format (32), which requires administration via continuous intravenous (cIV) infusion due to its short half-life (33, 34). We then developed AMG 757, a half-life extended (HLE) BiTE molecule, by fusing an immunoglobulin G (IgG) crystallizable fragment (Fc) domain to the core BiTE molecule structure, which enables longer dosing intervals in patients.

Here, we describe the potent, specific activity of an anti-DLL3 BiTE molecule and AMG 757 against SCLC cell lines in vitro, including high sensitivity for cells with low levels of DLL3 expression (<1,000 DLL3 molecules per cell). We demonstrate that once weekly administration of AMG 757 drives T-cell activation and expansion in established patient-derived xenograft (PDX) and orthotopic SCLC tumors in mouse models, leading to significant tumor regression including complete antitumor responses. We report that AMG 757 is well tolerated in nonhuman primate (NHP) toxicology studies with no AMG 757-related adverse findings up to the highest dose level tested (4.5 mg/kg administered weekly), and that AMG 757 exhibits an extended half-life. The potency and sensitivity of AMG 757 in SCLC models, together with favorable nonclinical safety and pharmacokinetic (PK) profiles, suggests that AMG 757 has potential as a DLL3-targeted immune therapy for SCLC.

DMS 79, NCI-H2171, NCI-H889, SHP-77, NCI-H211, NCI-H460, HEK-293, and Chinese hamster ovary (CHO) cell lines were purchased from the American Type Culture Collection, NCI-H82 from the DSMZ-German Collection of Microorganisms and Cell Cultures, and COR-L279 from Sigma-Aldrich. Cell lines were analyzed for authenticity using DNA-fingerprinting techniques such as short tandem repeat profiling. All cell lines tested negative for mycoplasma contamination. Research Resource Identifiers for cell lines are listed in Supplementary Tables (Supplementary Table S1).

DLL3 expression levels on cell lines were quantified in flow cytometry experiments using standard receptor quantitation methods (QIFIKIT assay [Dako] or Quantum Simply Cellular assay [Bangs Laboratories]) and analyzed using BD FACSDiva and GraphPad Prism or QuickCals software (see Supplementary Methods).

Transfected HEK-293 cells were plated at 5,000 cells/well and treated with activated T cells at 50,000 cells/well (10:1 effector to target [E:T] cell ratio) and BiTE molecules at 50 ng/mL final concentration, titrated 4-fold across the plate (12-point titration curve) in duplicate. After incubation for 20 hours, cell viability was measured with the Steady-Glo Luciferase assay (Promega).

Effector cells (2 × 10⁵ human or NHP PBMC or isolated human CD3⁺ T cells) were cocultured with Vybrant (Thermo Fisher Scientific) DiO/Dil–labeled target cell lines at an E:T cell ratio of 2:1 or 5:1 and serial dilutions of AMG 757 in 96-well, flat-bottom plates. Cells were cultured for 48 hours at 37°C in a 5% CO₂ humidified incubator and washed with FACS buffer. PBMCs were stained with a fluorochrome-labeled anti-CD14 antibody (BD Biosciences) to define DiO⁺/CD14⁻ target cells (30 minutes; 4°C), washed with FACS buffer, and resuspended in 100 μL of FACS buffer containing 0.5 μg/mL propidium iodide. All flow cytometry experiments were analyzed using FACSCanto II or LSRFortessa flow cytometers and the BD FACSDiva software (BD Biosciences) and data were fitted to a four-parameter nonlinear fit model sigmoidal dose–response curve (GraphPad Prism). T-cell activation was evaluated by flow cytometry after incubation as described for T-cell–dependent cellular cytotoxicity (TDCC) assays. T cells were washed and stained protected from light for 30 minutes at 4°C for CD4, CD8, and the activation markers CD69, CD71, PD-1, and PD-L1 using directly conjugated antibodies. For evaluation of granzyme B and perforin expression, cells stained for cell surface CD4 and CD8 were fixed and permeabilized (Cytofix/Cytoperm, BD Biosciences) and then stained with FITC-conjugated anti-granzyme B and allophycocyanin-conjugated anti-perforin antibodies. Cells were analyzed by flow cytometry. Cytokine levels were analyzed from supernatants of cytotoxicity assays using human and NHP Th1/Th2 cytometric bead arrays (BD Biosciences; see Supplementary Methods).

All research protocols were approved by the Amgen Institutional Animal Care and Use Committee (IACUC) (Study number: 2009–00046).

For the admixture xenograft studies, mice were inoculated in the subcutaneous flank with a mixture of 5 × 10⁶ luciferase-labeled NCI-H82 or SHP-77 tumor cells and 1 × 10⁶ human CD3⁺ T cells on study day 0. Mice were administered 0.5 mg/kg BiTE molecule by intraperitoneal injection daily from day 1 through day 11. Tumor measurements and body weights were collected twice a week throughout the study.

For the PDX models, female NOD.Cg-Prkdc^scidIl2rg^tm1Sug/JicTac (NOG) mice (Taconic) were implanted subcutaneously with patient-derived SCLC tumor fragments (LXFS 1129 and LXFS 538; Charles River Laboratories). After an induction period of 41 and 29 days, respectively, when the average tumor volume was 94–123 mm³, mice were randomized into two groups and administered activated human CD3⁺ T cells (2 × 10⁷) by intravenous infusion on study day 0. Human CD3⁺ T cells, isolated from PBMCs from two different donors using negative selection, were purchased from AllCells and activated in vitro using IL2 and CD3/CD28/CD2 (STEMCELL Technologies). T-cell expansion was monitored by cell count, and T-cell activation was assessed by measuring CD25 expression by flow cytometry. The PDX models used T cells from two different donors, one of which was used in the admixture xenograft studies. For the PDX efficacy models, mice (n = 9–10/group) were administered 3 mg/kg BiTE molecules by intraperitoneal injection on days 1, 8, and 15. Tumor volumes and body weights were measured twice a week until study end at day 23. For pharmacodynamic analysis, mice were administered a single dose of 3 mg/kg BiTE molecules on day 1 by intraperitoneal injection. At 96 hours postdose, tumors were collected and dissociated for analysis of tumor-infiltrating T cells. Additional details are provided in the Supplementary Methods.

Female NOD-scid IL2Rgamma^null (NSG) mice (The Jackson Laboratory), 6–7 weeks of age, were inoculated intravenously by tail vein injection with 1 × 10⁶ SHP-77 Luc cells or 5 × 10⁴ NCI-H82 Luc cells on day 0. On day 7, mice were imaged ventrally and randomized into treatment groups (n = 10/group) using the deterministic method (Studylog Systems, Inc). Each mouse was injected intravenously with 2 × 10⁷ CD3⁺ human T cells on study day 7, prepared as described above. BiTE molecules were dosed intraperitoneally once a week for a total of two doses (study days 8 and 15). Bioluminescence imaging (BLI) quantifications of the ventral lung area and body weights were collected twice a week throughout the study. In the NCI-H82 Luc model, livers were excised on day 22, and visible metastases were enumerated using a dissecting microscope.

For pharmacodynamic analysis of SHP-77 orthotopic tumors, mice were inoculated with SHP-77 cells as described above and then injected intravenously with 2 × 10⁷ CD3⁺ human T cells on study day 32. On study day 36, the mice were imaged ventrally and randomized to treatment groups (n = 4/group) and then treated with a single dose of 3 mg/kg BiTE molecule by intraperitoneal injection. Tumor dissociation and analysis are described in Supplementary Methods.

All research protocols were approved by the Institutional Animal Care and Use Committee (NHP toxicology study number: 121424, NHP PK study number: 121985). PK analysis is described in the Supplementary Methods.

A complete list of reagents and their catalog numbers are listed in Supplementary Tables (Supplementary Table S2).

We used RNA sequencing (RNA-seq) to profile 28 SCLC primary and metastatic tumor samples, and approximately 250 normal tissue samples representing an extensive panel of tissues, to identify potential targets with differential tumor expression using gene-ranking heuristics that prioritized target prevalence in tumors and minimal expression in essential normal tissues. Genes in the preliminary list were then further evaluated on the basis of literature, predicted extracellular domain structure, and feasibility of antibody generation. The Notch ligand DLL3, a type I membrane protein expressed primarily during development (31), was prioritized for further evaluation. We continued to obtain and analyze comparative data on the DLL3 expression profile as additional normal tissue samples and updated gene models became available. The final analysis showed that DLL3 RNA was detected at fragments per kilobase of transcript per million mapped reads (FPKM) ≥1 in 82% of primary and metastatic SCLC tumors (maximum FPKM, 116; median, 11.5), and expressed at low levels in the brain, spinal cord, pituitary gland, and testis (FPKM, 1.06–6.55; Supplementary Fig. S1A).

Saunders and colleagues first reported identification of DLL3 as a therapeutic target based on prevalent mRNA expression in SCLC and large-cell neuroendocrine tumors and limited expression in normal brain, pancreatic, and esophageal tissues (29). Given prior clinical experience with solid tumor BiTE molecule targets, our parallel studies had focused on obtaining a detailed understanding of DLL3 protein expression in normal tissues (Supplementary Table S3). Immunohistochemistry (IHC) analysis confirmed DLL3 expression in neurons in the human brain (cerebrum, cerebellum, hypothalamus, and hippocampus), secretory cells of the human and NHP anterior pituitary and the NHP pars intermedia, and human and NHP pancreatic islets, consistent with the mRNA profile. DLL3 staining in these normal tissues was of weak (1+) to mild (2+) intensity and predominantly cytoplasmic (Supplementary Fig. S2). The low, mainly cytoplasmic localization of DLL3 on normal tissues is not expected to be accessible to BiTE molecule activity.

In SCLC tumors, 30 of 35 (86%) of samples revealed DLL3 expression by immunostaining (Supplementary Fig. S1B), consistent with the RNA-seq data (Supplementary Fig. S1A). The DLL3 staining pattern was homogeneous in both membrane and cytoplasm. Among the DLL3-positive tumors, staining intensity was weak (1+) in 10 (28%), mild (2+) in 14 (40%), and intense (3+) in 6 (17%) tumors, with > 90% of neoplastic cells staining positive for DLL3 in most samples. The intensity of immunostaining did not correlate with tumor stage, grade, or other clinical factors. DLL3 mRNA and protein were also detected in other solid tumor types, including melanoma and glioblastoma (Supplementary Fig. S3). The prevalent membrane-associated staining of DLL3 in SCLC tumors, together with its low, mainly cytoplasmic expression in a few normal tissues, supported the advancement of DLL3 as a target for a BiTE molecule.

In SCLC cell lines, RNA-seq analysis showed a range of DLL3 expression levels (Supplementary Fig. S1A). Flow cytometry analysis of DLL3 cell surface expression showed generally low levels of DLL3, ranging from approximately 800 DLL3 molecules/cell (NCI-H211) to an estimated 3,222 DLL3 molecules/cell (SHP-77) based on quantitative flow cytometry analysis (Supplementary Table S4). These data were concordant with the DLL3 IHC data (Supplementary Fig. S1C), with intense staining (3+) of the SHP-77 cell line and weak staining (1+) of the NCI-H211 cell line. DLL3 mRNA and protein were not detected in the NSCLC cell line NCI-H460, a tumor type where DLL3 expression has not been reported. Collectively, these studies established that DLL3 expression in SCLC cell lines and tumors are similar.

We generated an anti-DLL3 BiTE molecule for proof of concept (Fig. 1A). This anti-DLL3 BiTE molecule had high binding affinity for human and NHP DLL3 and CD3, with equilibrium dissociation constants (K_D) of 3.0 nmol/L and 3.1 nmol/L for human and NHP DLL3, and 4.6 nmol/L and 4.4 nmol/L for human and NHP CD3, respectively. The anti-DLL3 BiTE molecule demonstrated potent activity in TDCC assays against HEK-293 cells transiently transfected to express DLL3, with a half maximal effective concentration (EC₅₀) value of 2.2 pmol/L (Fig. 1B). This activity was DLL3-dependent, as no cell killing activity was observed against mock-transfected HEK-293 cells (EC₅₀ > 1 nmol/L; Fig. 1B). The anti-DLL3 BiTE molecule induced cell killing across SCLC cell lines with EC₅₀ values of 18–203 pmol/L (Fig. 1C). These studies validate the approach of targeting DLL3-expressing target cells with a BiTE molecule.

The in vivo activity of the anti-DLL3 BiTE molecule was initially assessed in admixture xenograft models, where SCLC tumor cells mixed with human T cells were implanted subcutaneously into mice. The anti-DLL3 BiTE molecule was administered daily due to its short half-life. An anti-Mec14 BiTE molecule (35) was used as a negative control. In the SHP-77 xenograft model, which expresses 3,222 DLL3 molecules per cell (Supplementary Table S4), mice treated daily with 0.5 mg/kg of the anti-DLL3 BiTE molecule showed 89% tumor growth inhibition (TGI) after 15 days compared with mice treated with the anti-Mec14 BiTE molecule (P = 0.0001; Fig. 1D). In the NCI-H82 xenograft model, which expresses 1,290 DLL3 molecules per cell, mice treated daily with the anti-DLL3 BiTE molecule at 0.5 mg/kg had 75% TGI compared with mice treated with the control molecule (P < 0.0001; Fig. 1E). Together, these studies indicate that an anti-DLL3 BiTE molecule can effectively engage human T cells in xenograft tumors to inhibit tumor growth.

AMG 757 was designed to retain the potency and sensitivity of the anti-DLL3 BiTE molecule and to have an increased serum half-life to facilitate longer dosing intervals. The extended half-life of AMG 757 results from incorporation of a stable, effector-functionless Fc domain at the carboxy terminus of the molecule (Fig. 2A; ref. 36).

The binding affinity of AMG 757 to human and NHP DLL3 was 0.64 ± 0.05 nmol/L and 0.50 ± 0.01 nmol/L, respectively, and AMG 757 binding affinity to human and NHP CD3 was 14.9 ± 0.4 nmol/L and 12.0 ± 0.3 nmol/L, respectively (Supplementary Table S5). TDCC assays demonstrated that the in vitro cell potency of AMG 757 against DLL3-positive SCLC cell lines was similar to that of the anti-DLL3 BiTE molecule and specific for DLL3-positive cells, with no evidence of cytotoxicity against the NCI-H460 cell line that does not express DLL3 (Fig. 2B). AMG 757 effectively engaged human T cells to kill SCLC cell lines, including those with very low DLL3 expression levels (<1,000 molecules/cell; Fig. 2C; Supplementary Table S4). AMG 757 induced potent cell killing with either human or NHP effector cells, with EC₅₀ values within 1- to 7-fold for each cell line, confirming the cross-reactivity of AMG 757 with NHP CD3 (Fig. 2D; Supplementary Table S5).

AMG 757-dependent cytotoxicity was associated with hallmarks of BiTE molecule target engagement, including T-cell activation, cytokine production, and release of cytotoxic granules (Fig. 3). AMG 757 increased granzyme B levels and cytotoxicity over time, with maximal signal observed at 48 hours (Fig. 3A). Markers of T-cell activation or inflammation, CD69, CD71, PD-1, and PD-L1 (37–39) were upregulated following AMG 757 treatment (Fig. 3B). AMG 757-mediated redirected lysis was accompanied a production of proinflammatory cytokines (Fig. 3C). DLL3 cell surface expression on the corresponding human tumor cell lines is shown in Fig. 3D and in Supplementary Table S4. Together, these data demonstrate the AMG 757 mechanism of action.

The cross-reactivity of the anti-DLL3 BiTE molecule and AMG 757 with NHP DLL3 and CD3 established NHP as a relevant model for evaluation of the tolerability of these molecules in vivo. The anti-DLL3 BiTE molecule was evaluated in an exploratory, step-dose, nonterminal safety study in NHP. Animals received the anti-DLL3 BiTE molecule by cIV infusion at doses up to 100 μg/day for 7 days. All doses evaluated were well tolerated, consistent with the idea that the low level, cytoplasmic DLL3 expression in normal NHP tissues is inaccessible to BiTE molecule binding.

The PK profile of AMG 757 in NHP was characterized in a single-dose study. After a single 12 μg/kg dose of AMG 757, serum concentrations of AMG 757 decreased in a biphasic manner, with a mean clearance of 0.487 mL/hour/kg and a steady-state volume of distribution of 146 mL/kg (Fig. 4). The maximum serum concentration was 0.239 μg/mL and the area under the concentration–time curve was 16.7 hours×μg/mL. The trough exposure level after 336 hours (14 days) was approximately 10-fold higher than the mean in vitro EC₅₀ of AMG 757 across SCLC cell lines. The mean half-life of AMG 757 was 234 hours (9.8 days). These data confirm that AMG 757 has an extended half-life relative to the anti-DLL3 BiTE molecule.

AMG 757 was assessed in a 1-month repeat dose toxicology study in NHP. Weekly doses of AMG 757 up to 4.5 mg/kg were well tolerated and no AMG 757-related adverse findings were observed (40). Taken together, the NHP studies suggest that AMG 757 has the potential to achieve high exposures that may be required for effective antitumor activity in the clinic.

To demonstrate proof of concept of AMG 757 in vivo, we evaluated antitumor activity in an admixture WM266-4 xenograft model, where human T cells and WM266-4 tumor cells were mixed and implanted subcutaneously in immunodeficient mice. Administration of AMG 757 at 0.1, 0.5, or 3 mg/kg every 5 days significantly inhibited tumor growth in this model (Supplementary Fig. S4). We next used PDX models to assess the activity of AMG 757 against established SCLC tumors. PDX tumors replicate the heterogeneity and architecture of human tumors in a mouse model (41), and systemic administration of human T cells to mice bearing PDX tumors enables assessment of T-cell–mediated antitumor activity, as AMG 757 does not cross react with mouse CD3. We evaluated AMG 757 activity in the LXFS 1129 and LXFS 538 PDX models of SCLC, which express DLL3 at levels similar to that of NCI-H82 cells (Supplementary Fig. S5A and S5B). Immunocompromised mice bearing established LXFS 1129 tumors received a single administration of human T cells and were then treated with AMG 757 once weekly for 3 weeks. AMG 757 treatment led to 83% tumor regression, including complete responses in eight of 10 mice by day 23, and an overall significant reduction in tumor volume compared with that in mice which received a control HLE BiTE molecule (P < 0.0001; Fig. 5A). In the LXFS 538 model, AMG 757 treatment induced 98% tumor regression, including complete responses in eight of 10 mice by day 21, and an overall significant reduction in tumor volume compared with that in mice which received the control HLE BiTE molecule (P < 0.0001; Fig. 5B). The mice in the LXFS 538 study treated with AMG 757 were observed through day 32 (17 days after the last AMG 757 dose), at which point nine of 10 mice had complete responses, with one mouse having a residual (<5 mm³) mass that had remained static for 2 weeks. At this last time point, both the human T cells and AMG 757 would have been cleared from circulation.

We developed orthotopic SCLC models to further explore AMG 757 antitumor activity. These models recapitulate the biologic compartment of primary SCLC tumors and, like the PDX model, require that human T cells administered systemically traffic to the tumor site for BiTE molecule target engagement. Consistent with primary SCLC tumors, the SHP-77 orthotopic model is characterized by tumor growth in the lungs but the lesions are not visible macroscopically. In contrast, the NCI-H82 orthotopic model mimics SCLC metastatic disease with discrete, visible tumors primarily in the liver. Both models used cell lines engineered to express luciferase to enable visualization of tumor burden by BLI. Human T cells were administered once, and BiTE molecules were administered weekly by intravenous infusion to immunocompromised mice bearing established orthotopic tumors.

AMG 757 treatment significantly inhibited growth of established orthotopic SHP-77 lung tumors (P < 0.0008; Fig. 5C and D). BLI imaging on day 22 revealed that mice treated with AMG 757 exhibited significant TGI (Fig. 5D). The minimal BLI signal that persisted in the AMG 757-treated mice at the end of the study might reflect residual nodules or cell debris. Tumors in mice treated with vehicle alone or with a control HLE BiTE molecule continued to grow (Fig. 5C and D). In the NCI-H82 orthotopic liver model, AMG 757 treatment significantly inhibited growth of liver metastases compared with vehicle or control HLE BiTE molecule–treated groups at day 22 (P < 0.0001; Fig. 5E). BLI imaging on day 22 revealed a marked decrease in tumor mass after AMG 757 treatment (Fig. 5F). Strikingly, no lesions were visible in the liver at day 23 following AMG 757 treatment, suggesting complete response to treatment (Fig. 5G and H). In contrast, the vehicle and control HLE BiTE molecule–treated cohorts contained multiple liver lesions (mean of 41.5 and 41.8 lesions, respectively; P < 0.0001; Fig. 5G and H).

The pharmacodynamic effects of AMG 757 treatment on T cells infiltrating SCLC tumors was evaluated. In the LXFS 538 PDX model, we used flow cytometry to assess T-cell numbers and markers of activation from T cells isolated from disaggregated tumor tissue. A single dose of AMG 757 led to a statistically significant increase in the absolute number of human CD4⁺ and CD8⁺ T cells in the tumors at 4 days posttreatment compared with treatment with the control HLE BiTE molecule (CD4⁺, mean 77,098 vs. 1,601 cells; P = 0.004; CD8⁺, mean 263,289 vs. 2,686 cells; P = 0.012; Fig. 6A). The T-cell activation markers CD25, CD69, and PD-1 were significantly upregulated (P < 0.0001) in CD4⁺ and CD8⁺ T cells in the AMG 757-treated group compared with the control HLE BiTE molecule–treated group in LXFS 538 PDX tumors (Fig. 6B and C).

Similarly, in the SHP-77 orthotopic model, a significant increase in human CD4⁺ and CD8⁺ T-cell counts in the disaggregated lung tissue occurred 7 days posttreatment with AMG 757 compared with those from the control HLE BiTE molecule–treated group (CD4⁺, mean 60 vs. 7.9 cells/mg of lung tissue; P = 0.013; CD8⁺, mean 117.8 vs. 16.7 cells/mg of lung tissue; P = 0.018; Fig. 6D), and 72 hours after administration of AMG 757 in metastatic liver lesions from the NCI-H82 xenograft SCLC model (CD4⁺, mean of 1,080 vs. 78.2 cells/mg of tumor tissue; P < 0.0001; CD8⁺, mean of 1,597 vs. 89.6 cells/mg of tumor tissue; P < 0.0001; Fig. 6E) when compared with the number of cells from groups receiving HLE BiTE molecule.

While a significant upregulation of the T-cell activation markers CD25, CD69, PD-1, and 4-1BB was observed on T cells from metastatic liver lesions from the AMG 757-treated group in the NCI-H82 model, this increase was not observed in T cells from the disaggregated lung tissue from the SHP-77 orthotopic model (Fig. 6F and G; Supplementary Fig. S6). This may be due to the fact that the total population of lung-resident T cells was analyzed in the SHP-77 orthotopic model rather than T cells specific to the lung tumors.

NCI-H82 tumors from mice treated with AMG 757 and the control HLE BiTE molecule were compared to evaluate tumor morphology and the spatial relationship between infiltrating T cells and tumor cells. In hematoxylin- and eosin–stained sections, NCI-H82 tumors treated with AMG 757 had a qualitative increase in infiltrating lymphocytes dispersed throughout the tumor and in the tumor periphery. While the control HLE BiTE molecule-treated tumors were well-demarcated and composed primarily of viable neoplastic cells, AMG 757-treated tumors exhibited viable and degenerate inflammatory cells, necrotic cellular debris, and collapsed stroma in the periphery (Fig. 6H). IHC demonstrated that the infiltrating lymphocytes were CD4⁺ and CD8⁺ T cells (Fig. 6I). Collectively, the orthotopic models demonstrate that AMG 757 can promote T-cell trafficking to tumors in different compartments and engage T cells to DLL3-expressing SCLC cells, enabling complete antitumor responses against established SCLC tumors.

AMG 757 is a first-in-class HLE BiTE immuno-oncology therapy targeting DLL3 for the treatment of SCLC. Here, we demonstrate that AMG 757 engages T cells to induce potent redirected lysis of SCLC cells in vitro and significant antitumor activity in mouse models of primary and metastatic SCLC. In vitro, AMG 757 had potent cell killing activity against SCLC cells with as low as <1,000 DLL3 receptors per cell. In vivo, AMG 757 engaged human T cells administered systemically and DLL3-expressing tumor cells to promote tumor regression. In our models, AMG 757 monotherapy cleared established tumors including visible metastases, and only residual nodules were detected even in the LXFS 538 model, where the mice remained on study 17 days after the last treatment of AMG 757. However, as our tumor models use immunocompromised mice, and the human T cells administered once at the start of our studies do not persist beyond a few weeks, these models are not suitable for the longer-term evaluation necessary to uncover mechanisms of resistance to AMG 757. The impact of DLL3 expression levels and the tumor microenvironment on AMG 757 activity and potential resistance mechanisms will be further assessed in patients with SCLC.

A 1-month repeat dose NHP toxicology study showed that AMG 757 was well tolerated at 4.5 mg/kg, a dose with exposure levels that far exceed the mean in vitro cell EC₅₀ values, with no AMG 757-related adverse findings. Given the high affinity binding of AMG 757 to NHP DLL3 and CD3, and the potent activity of NHP effector cells in redirected lysis of SCLC tumor cells, the in vivo safety profile strongly suggests that DLL3 expressed on normal tissues is largely inaccessible to AMG 757 target engagement. BiTE molecules may access targets expressed in the brain, as blinatumomab has been detected in the brain and can be associated with neurologic events, and an EGFRvIII-targeting BiTE molecule, AMG 596, is in development for glioblastoma (42–44).

The safety profile of AMG 757 in NHP is very different from that of another DLL3-targeted therapy, the antibody-drug conjugate (ADC) rovalpituzumab tesirine. Safety-related findings in preclinical toxicology studies with rovalpituzumab tesirine included myelosuppression, kidney degeneration, and thickening and hyperpigmentation of the skin (29). These adverse effects were attributed to the cytotoxic payload of the molecule, pyrrolobenzodiazepine (PBD), and were not related to DLL3 expression (45). Despite early clinical activity of rovalpituzumab teserine (12%–18% objective response rates; refs. 45, 46), treatment was associated with dose-limiting adverse events, including pleural and pericardial effusion and skin reactions, which limited treatment to two cycles on average (45). These findings were also attributed to PBD, and development of rovalpituzumab tesirine was halted after a phase III study showed no survival benefit (14). The toxicity profile of rovalpituzumab tesirine in preclinical and clinical studies is, therefore, not associated with DLL3 expression (46). While recent advances in linker technology, antibody design, and payload optimization may improve the therapeutic index of an ADC targeting DLL3, the relatively low abundance of DLL3 suggests that it could be challenging to deliver sufficient amounts of a toxic payload to DLL3-positive SCLC tumors without dose-limiting toxicity caused by unconjugated payload. The combination of activity in preclinical tumor models and preliminary safety data in cynomolgus monkeys highlight the promise of targeting DLL3 with a BiTE molecule. In addition to AMG 757, chimeric antigen receptor (CAR)-T cells targeting DLL3 in SCLC and other neuroendocrine tumors are in development, and DLL3-targeted radioimmunoconjugates are also being explored (31, 47).

The treatment paradigm in SCLC has recently changed with the approval of PD-1/PD-L1 immune checkpoint inhibitors in multiple settings, including as part of first-line therapy in combination with chemotherapy (8, 48). Although these approvals represent a significant advancement in SCLC treatment, the overall response rates remain low. AMG 757 provides an option for therapeutic targeting of DLL3-expressing SCLC cells by a mechanism that is completely distinct from that of immune checkpoint inhibitors. AMG 757 demonstrates engagement of T cells and direct T-cell–mediated killing of established tumors, and also the potential to promote T-cell infiltration into tumors. AMG 757 therefore may also be considered for combination therapy with anti–PD-1/PD-L1 inhibitors in patients where it may enhance the activity of immune checkpoint inhibitors (15, 49); blinatumomab and other BiTE molecules have been shown to upregulate the PD-1/PD-L1 axis and patients may benefit from combination therapy with immune checkpoint inhibitors (50). AMG 757 provides a new option for SCLC therapy with a compelling nonclinical safety:efficacy profile, and clinical evaluation in patients with SCLC is ongoing (NCT03319940).

M.J. Giffin reports nonfinancial support from Scott Medical Communications, LLC and Cactus Life Sciences, part of Cactus Communications during the conduct of the study; in addition, M.J. Giffin reports employment and stock ownership in Amgen. K. Cooke is an Amgen employee and owns Amgen stock. E.K. Lobenhofer reports nonfinancial support from Scott Medical Communications, LLC and Cactus Life Sciences, part of Cactus Communications, during the conduct of the study; in addition, E.K. Lobenhofer reports employment and stock ownership in Amgen, Inc. J.C. Estrada is an Amgen employee and owns Amgen stock. J. Zhan is an Amgen employee and owns Amgen stock. P. Deegen reports nonfinancial support from Scott Medical Communications, LLC and Cactus Life Sciences, part of Cactus Communications, during the conduct of the study; in addition, P. Deegen reports employment and stock ownership in Amgen. M. Thomas reports nonfinancial support from Scott Medical Communications, LLC and Cactus Life Sciences, part of Cactus Communications, during the conduct of the study; in addition, M. Thomas reports employment and stock ownership in Amgen. C.M. Murawsky reports nonfinancial support from Scott Medical Communications, LLC and from Cactus Life Sciences, part of Cactus Communications, during the conduct of the study; in addition, C.M. Murawsky reports employment and stock ownership in Amgen. J. Werner reports nonfinancial support from Scott Medical Communications and Cactus Life Sciences during the conduct of the study; in addition, J. Werner reports employment and stock ownership in Amgen. S. Liu reports nonfinancial support from Scott Medical Communications, LLC and Cactus Life Sciences, part of Cactus Communications, during the conduct of the study; in addition, S. Liu reports employment and stock ownership in Amgen. F. Lee reports nonfinancial support from Scott Medical Communications, LLC during the conduct of the study; and reports employment and stock ownership in Amgen. O. Homann reports nonfinancial support from Scott Medical Communications, LLC and Cactus Life Sciences, part of Cactus Communications, during the conduct of the study; in addition, O. Homann reports employment and stock ownership in Amgen. M. Friedrich reports nonfinancial support from Scott Medical Communications and Cactus Life Sciences during the conduct of the study; in addition, M. Friedrich has a patent for US20170349668A1 pending; and reports employment and stock ownership in Amgen. J.T. Pearson reports nonfinancial support from Scott Medical Communications, LLC and Cactus Life Sciences during the conduct of the study; personal fees from Merck & Co outside the submitted work; in addition, J.T. Pearson also reports past employment (ending December 3, 2018) and stock ownership in Amgen. T. Raum reports nonfinancial support from Scott Medical Communications, LLC and Cactus Communications during the conduct of the study; in addition, T. Raum has a patent for US20170037130 pending; in addition, T. Raum is an Amgen employee and has stock ownership in Amgen. S. Caenepeel reports nonfinancial support from Scott Medical Communications, LLC and Cactus Life Sciences, part of Cactus Communications during the conduct of the study; in addition, S. Caenepeel reports employment and stock ownership in Amgen. J. Stevens reports nonfinancial support from Scott Medical Communications, LLC during the conduct of the study; in addition, J. Stevens reports employment and stock ownership in Amgen. P.J. Beltran reports nonfinancial support from Scott Medical Communications and Cactus Communications during the conduct of the study; other from Amgen outside the submitted work. J. Canon reports nonfinancial support from Scott Medical Communications, LLC and Cactus Life Sciences during the conduct of the study; in addition, J. Canon was an employee and stock holder of Amgen. J.M. Bailis reports nonfinancial support from Scott Medical Communications, LLC and Cactus Life Sciences, part of Cactus Communications, during the conduct of the study; in addition, J.M. Bailis is an employee of Amgen, Inc. and holds stock ownership in Amgen, Inc. P. Hughes reports nonfinancial support from Scott Medical Communications, LLC and Cactus Life Sciences, part of Cactus Communications during the conduct of the study; in addition, P. Hughes is a shareholder and employee of Amgen Inc. No disclosures were reported by the other authors.

M.J. Giffin: Conceptualization, formal analysis, supervision, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. K. Cooke: Conceptualization, resources, formal analysis, supervision, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. E.K. Lobenhofer: Conceptualization, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. J. Estrada: Formal analysis, validation, investigation, visualization, methodology, writing-review and editing. J. Zhan: Formal analysis, validation, investigation, visualization, methodology, writing-review and editing. P. Deegen: Formal analysis, supervision, validation, investigation, visualization, methodology, writing-review and editing. M. Thomas: Resources, formal analysis, supervision, validation, investigation, methodology, writing-review and editing. C.M. Murawsky: Resources, data curation, formal analysis, supervision, validation, investigation, methodology, writing-review and editing. J. Werner: Data curation, validation, investigation, visualization, methodology, writing-review and editing. S. Liu: Validation, investigation, visualization, methodology, writing-review and editing. F. Lee: Validation, investigation, visualization, methodology, writing-review and editing. O. Homann: Conceptualization, data curation, software, formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. M. Friedrich: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing-review and editing. J.T. Pearson: Formal analysis, validation, investigation, visualization, methodology, writing-original draft, writing-review and editing. T. Raum: Formal analysis, supervision, validation, visualization, methodology, writing-original draft, writing-review and editing. Y. Yang: Formal analysis, validation, investigation, visualization, methodology, writing-review and editing. S. Caenepeel: Formal analysis, validation, investigation, visualization, methodology, writing-review and editing. J. Stevens: Formal analysis, supervision, validation, visualization, methodology, writing-review and editing. P.J. Beltran: Supervision, writing-review and editing. J. Canon: Supervision, writing-review and editing. A. Coxon: Conceptualization, supervision, validation, writing-review and editing. J.M. Bailis: Conceptualization, formal analysis, supervision, validation, visualization, methodology, writing-original draft, writing-review and editing. P.E. Hughes: Conceptualization, resources, formal analysis, supervision, writing-original draft, writing-review and editing.

The results shown here are based, in part, on data generated by the TCGA Research Network (https://www.cancer.gov/tcga). The authors acknowledge Karen Hettwer, Rodolfo Yabut, Roberto Guzman, David Smith, Joachim Wahl, Kathy Manchulenko, Tao Osgood, Melody Richardson, and Amy Gilbert for technical support. Medical writing support was provided by Ben Scott, PhD (Scott Medical Communications, LLC), Sukanya Raghuraman, PhD (Cactus Life Sciences, part of Cactus Communications), Micah Robinson, PhD (Amgen Inc.), and Jacqueline Sayyah (Amgen, Inc.). This study was funded by Amgen Inc. We thank Mark Salvati, PhD, and Marie-Anne Damiette Smit, MD, MS, for critical review of the manuscript.

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