Characterization of a Novel FLT3 BiTE Molecule for the Treatment of Acute Myeloid Leukemia

Newly approved targeted therapies and cytotoxic agents (1) provide opportunities to improve treatment of acute myeloid leukemia (AML), a disease characterized by low survival rates (2). However, these therapeutics are approved for only certain patient subsets, and treatments to benefit broad patient populations are still needed. To date, the most efficacious treatment consists of intensive chemotherapy followed by allogeneic hematopoietic stem cell transplantation (HSCT; refs. 3, 4). The potent antileukemic effect of HSCT is driven by recognition and elimination of allogeneic antigens on chemoresistant leukemic cells by donor T cells. HSCT, as well as donor lymphocyte infusions, which frequently result in durable complete remissions (3), demonstrate the potential for therapies driven by T-cell cytotoxicity (4). However, this regimen may not be an option for all patients due to comorbidities and the high morbidity and mortality rates associated with graft-versus-host disease, highlighting the urgent need for novel therapies (4).

A promising T-cell–based therapeutic approach is to redirect a patient's own T cells to eliminate leukemic cells. This strategy can be accomplished with bispecific T-cell engaging (BiTE) molecules. BiTE molecules consist of a single chain variable fragment (scFv) against a cell surface-expressed tumor-associated antigen (TAA) linked to an scFv against the T-cell coreceptor CD3. Clinical proof of concept for this modality was demonstrated by the CD19-directed BiTE molecule blinatumomab, which is approved for B-cell precursor acute lymphoblastic leukemia. CD19 is an ideal target for a BiTE molecule because it is broadly expressed on B-cell malignancies, its off-tumor expression is limited to normal B cells, and patients can tolerate prolonged B-cell depletion. The successful translation of BiTE molecules to AML therapy requires identification of a suitable cell surface antigen, one that is broadly and selectively expressed by leukemic cells with limited expression on normal tissues.

fms-Related tyrosine kinase 3 (FLT3, CD135) is a lineage-associated growth factor that was previously reported to be expressed on AML blasts and LSCs (5, 6). Expression of FLT3 on normal hematopoietic cells has been reported to be restricted to a subset of hematopoietic stem and progenitor cells (HSPC) in the bone marrow (BM; ref. 7). These data suggest a favorable expression profile for targeting FLT3 with a BiTE molecule. Mutations in the intracellular portion of FLT3, resulting in constitutive activation, occurring as either internal tandem duplication or point mutations in the tyrosine kinase domain have been identified in approximately 25% or 7% to 10% of patients with AML, respectively. (8–10). Tyrosine kinase inhibitors (TKIs) that target the FLT3 kinase domain were recently approved for patients with mutant FLT3 and others are undergoing clinical evaluation (11–13). FLT3 TKIs are active primarily in the setting of mutant FLT3, whereas BiTE molecules recognize an extracellular protein epitope and bind FLT3 regardless of mutational status.

Here, FLT3 was evaluated as a target for BiTE molecule therapy for the treatment of AML, including expression analysis on disease and normal cells, and two novel FLT3 BiTE molecules were characterized in vitro, ex vivo, and in vivo. Cell surface FLT3 protein expression was observed on most primary AML (pAML) patient bulk and LSC samples, irrespective of FLT3 mutational status. Importantly, comparable FLT3 protein expression was observed on patient samples collected at the time of both initial diagnosis and relapse, suggesting a FLT3 BiTE molecule could provide benefit to patients across multiple lines of therapy. FLT3 transcript and protein expression was rigorously evaluated in a panel of normal human tissues, and cell surface FLT3 protein was detected only on a portion of HSPCs and on rare, scattered cells in the tonsil. FLT3 protein was also detected in some non-hematopoietic tissues, including cerebellum and pancreas; however, extensive characterization revealed that the protein was cytoplasmic. Because FLT3 BiTE molecules selectively bind to cells expressing cell surface FLT3, cells expressing cytoplasmic FLT3 protein would not be expected to be depleted.

Two FLT3 BiTE molecules were generated and evaluated: an experimental FLT3 BiTE molecule comprised an anti-CD3 scFv and an anti-FLT3 scFv, and a FLT3 half-life extended (HLE) BiTE molecule (AMG 427) comprised an anti-CD3 scFv fused to an Fc moiety and a unique anti-FLT3 scFv. Because of the size of the experimental FLT3 BiTE molecule, rapid clearance by glomerular filtration is expected to result in a short serum half-life, requiring continuous intravenous (cIV) infusion to maintain an active concentration in vivo. The larger AMG 427 was designed to have an extended serum half-life relative to the experimental FLT3 BiTE molecule. Both BiTE molecules induced potent and target-specific T-cell-dependent cellular cytotoxicity (TDCC) against AML cell lines in vitro, inhibited tumor growth and provided a survival advantage in vivo in xenograft models and exhibited reproducible pharmacokinetic (PK) and pharmacodynamic (PD) profiles in cynomolgus monkeys. The experimental FLT3 BiTE molecule induced TDCC of pAML samples ex vivo. Increased in vitro TDCC was observed by combining AMG 427 with an anti-PD-1 antibody. These data demonstrate that FLT3 BiTE molecules are capable of inducing TDCC of FLT3-expressing cells in vitro, in vivo, and ex vivo; moreover, although each FLT3 BiTE molecule was efficacious as a single agent against AML cell lines and pAML samples, combination therapy may provide additional benefit for some patients. AMG 427 is being evaluated patients with relapsed or refractory AML.

AML and healthy donor (HD) samples were obtained with written informed consent in accordance with the Declaration of Helsinki and approval by the Institutional Review Board of the Ludwig-Maximilian University (Supplementary Tables S1 and S2). Human tissue specimens for expression analyses were collected under Institutional Review Board approval with appropriate informed consent. In all cases, materials obtained were surplus to standard clinical practice. Patient identity and protected health information/identifying information were redacted from tissue data and clinical data.

Sources of biological samples, all antibodies and other key reagents are listed in Supplementary Table S1.

Cell surface FLT3 protein expression on pAML and HD peripheral blood (PB) or BM samples was assessed by flow cytometry (Navios; Beckman Coulter) using an anti-FLT3 antibody (Supplementary Table S1). Mean fluorescence intensity (MFI) was determined (FlowJo version 10.3) and the MFI ratio (MFI sample/MFI isotype control) was calculated.

FLT3 transcript expression data were retrieved from The Cancer Genome Atlas [TCGA (14), AML patient samples] in February 2018.

FLT3 transcript expression data in normal human tissues were retrieved from the Genotype-Tissue Expression project [GTEx (15), HD samples] in April 2018.

5' rapid amplification of cDNA ends (RACE), digital droplet polymerase chain reaction (ddPCR), reverse transcription PCR, immunohistochemistry (IHC), Western analysis, immunoprecipitation, and RNA-seq were conducted using standard techniques. Details in Supplementary Table S1 and Supplementary Materials and Methods.

Cell lines were initially sourced from DSMZ (MOLM-13, EOL-1, PL-21), ATCC (HL-60, MV4-11 K562, HEL92.1), and ECACC (A2780), and cultured using standard techniques and reagents. In the absence of phenotypic or growth changes, cells were not authenticated or tested for mycoplasma. Cells were used within 2 months of thawing.

Human PBMCs or pan T cells were cultured for 48 hours in the presence or absence of FLT3 expression–positive or FLT3 expression–negative target cells with an effector-to-target (E:T) cell ratio of 10:1 (pan T) or 5:1 (PBMC) and a dose range of FLT3 BiTE molecules. Target cell lysis was measured by loss of luciferase signal (Steady-Glo, Promega; labeled target cell lines express luciferase); or propidium iodide uptake by flow cytometry. T-cell activation markers were assessed by flow cytometry using antibodies against CD4, CD8, CD69, and CD25 labeled with a fluorochrome conjugate (Supplementary Table S1). BiTE-induced cytokine secretion was measured in supernatants using the BD Cytometric Bead Array Human Th1/Th2 Cytokine Kit. Luciferase-based TDCC (pan T, E:T ratio 10:1) was performed with or without 10 ng/mL soluble FLT3 ligand (16) for 48 hours.

Animal experimental procedures were conducted in accordance with the German Animal Welfare Law with permission from the responsible local authorities and within the guidelines of the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC) international standards.

Female nonobese diabetic/severe combined immunodeficient (NOD/SCID) mice at age 7 weeks were sublethally irradiated prior to tumor cell injection. Mice were injected intravenously via lateral tail vein with (10⁷) MOLM-13_Luc or (5 × 10⁶) EOL-1 cells on day 1. After 48 or 72 hours, respectively, mice were injected intraperitoneally with human in vitro–expanded CD3⁺ T cells (2 or 1.2 × 10⁷, respectively) and allocated to treatment groups (n = 10/group). Five mice allocated to the vehicle group did not receive human T cells. Mice were treated with vehicle or AMG 427 (3, 0.6, or 0.12 mg/kg) every 5 days by intravenous bolus injection into the lateral tail vein starting on day 7, then on days 12, 17, 22, 27, and 34 (MOLM-13_Luc). Mice were treated with vehicle or AMG 427 (1, 0.1, or 0.001 mg/kg) every 7 days by intravenous bolus injection into the lateral tail vein starting on day 9 for a total of six administrations (EOL-1). To block binding of AMG 427 to Fc receptors, a mixture of anti-muFcRII (2.4G2) antibody (8 mg/kg) and human normal immunoglobulin (400 mg/kg of Kiovig) was administered once weekly intraperitoneally throughout the treatment period, starting 1 day prior to the first AMG 427 dose. Mice were monitored daily. PK serum concentrations of AMG 427 were determined by electrochemiluminescence immunoassay (Supplementary Materials and Methods).

Cynomolgus monkeys were cared for in accordance to the Guide for the Care and Use of Laboratory Animals, Eighth Edition (17). Animal care is detailed in Supplementary Materials and Methods.

Cell surface FLT3 protein expression on hematopoietic cells from cynomolgus monkeys was assessed as described in the Supplementary Materials and Methods.

The experimental FLT3 BiTE molecule was evaluated in a 16-day cIV study (n = 3) and was administered at step doses increasing every 3 days, intended to achieve C_ss of 0.05, 0.2, 0.5, and 2 nmol/L for 3 days each. Methods for PK and PD assessment are detailed in Supplementary Materials and Methods. AMG 427 was evaluated in an 8-day, repeated-dose study with three dose levels (n = 3/group). AMG 427 was administered intravenously on days 1, 2, and 5.

AML patient samples were cultured (Supplementary Materials and Methods) with experimental FLT3 BiTE molecule or control BiTE molecule at 5 ng/mL (92 pmol/L) and replenished at 3-day intervals. Viable CD33⁺/CD2⁻ cells (Supplementary Table S1) were determined by flow cytometry, and total cell count was used to determine AML cell count.

Human pan T cells were stimulated 1:1 with CD3/CD28 Dynabeads (Thermo Fisher Scientific) for 48 hours, then cocultured 1:1 with PD-L1–transfected MOLM-13 cells (MOLM-13_PD-L1) and dose range of AMG 427 in the absence or presence of 10 μg of a PD-1–blocking antibody (Supplementary Table S1). After 24 hours, MOLM-13_PD-L1 cell viability was determined by TO-PRO-3 uptake by flow cytometry.

Leukemic bulk cells from BM or PB from 318 newly diagnosed or relapsed AML patients were evaluated for cell surface FLT3 protein expression. Of the analyzed samples, 78% (248/318) were positive for FLT3 protein expression (MFI ratio >1.5; Fig. 1A, gating strategy, Supplementary Fig. S1A, top; MFI ratio calculation, Supplementary Fig. S1B). Interpatient heterogeneity in FLT3 protein expression was observed (MFI ratio range 0.1–32.7; Fig. 1A), similar to what has been reported for other AML-associated antigens (18). The FLT3 protein expression profile was similar, regardless of FLT3-ITD mutational status (Fig. 1B), time of sample collection (initial diagnosis versus relapse, Fig. 1C), or FLT3-ITD allelic ratio (Fig. 1D). Cell surface FLT3 protein expression was detected on leukemic stem cells (LSC; CD34⁺/CD38⁻) in 79% (122/155) of AML patient samples (Fig. 1E). As observed for bulk cells, FLT3 protein expression on LSCs was similar, regardless of FLT3-ITD mutational status (Fig. 1F) or initial diagnosis versus relapse (Fig. 1G). Higher FLT3 expression was detected on samples with high FLT3-ITD allelic ratio (Fig. 1H, P < 0.0098).

No clear correlation was observed in an analysis of FLT3 protein expression intensity on AML patient bulk cells at initial diagnosis with different disease characteristics, including French American British group, core binding factor abnormalities [i.e., translocation t(8;21) and inversion inv(16)], nucleophosmin 1 (NPM1) and FLT3-ITD mutations, Medical Research Council cytogenetic-based risk classification (19), and 2010 European Leukemia Net classification (Supplementary Fig. S1C–S1F; Supplementary Table S2; ref. 20).

In hematopoietic cell samples derived from HDs, the FLT3 protein MFI ratio was consistently low and less than that of pAML samples. The MFI ratio on HD CD34⁺CD38⁻ cells (n = 18), comprising hematopoietic stem cells and multipotent progenitors, was 0.58 ± 0.26 and the MFI ratio on CD34⁺CD38⁺ cells (n = 36), comprising the oligopotent progenitors, was 1.6 ± 0.5 (Fig. 1I; gating strategy Supplementary Fig. S1A, bottom; MFI ratio calculation, Supplementary Fig. S1B). In comparison, the MFI ratio for pAML samples was significantly higher than either of the HD samples (P < 0.0001 comparison to either HD CD34⁺CD38⁻ cells or HD CD34⁺CD38⁺ cells) at 3.6 ± 3.6 for the bulk samples (n = 318) and 2.9 ± 2.3 for the LSC samples (n = 155). Paired analysis of CD34⁺CD38⁻ and CD34⁺CD38⁺ cells from 13 HDs showed that FLT3 protein expression was statistically lower on CD34⁺CD38⁻ than CD34⁺CD38⁺ cells (Fig. 1J).

FLT3 protein expression was subsequently evaluated on individual stem cell and oligopotent progenitor subsets from two HD. Expression was variable, and no subset was uniformly positive or negative (Fig. 1K, gating strategy; Supplementary Fig. S2A). The rank order of FLT3 protein expression was: granulocyte/macrophage progenitor > common lymphoid progenitor > common myeloid progenitor > HSC and MPP > megakaryocyte/erythrocyte progenitor (MEP), most of the latter population falling below detection.

On mature hematopoietic cells isolated from blood from two HD, there was no detectable cell surface FLT3 protein on T or B lymphocytes, natural killer cells, plasmacytoid or conventional dendritic cells, monocytes, or neutrophils (Supplementary Fig. S2B). Collectively, these data demonstrate that there are differences in expression between disease and normal cells, and within hematopoietic stem and progenitor populations, which may translate to differences in susceptibility to FLT3 BiTE-mediated killing.

The presence of FLT3 transcript in nonhematopoietic tissues was assessed in three different datasets including GTEx RNA-seq database (15), an Amgen-constructed RNA-seq database, and XpressWay Profile Report (Asterand UK Acquisition Limited). Low levels of FLT3 transcript were detected in brain, nerve/ganglia, small intestine, kidney, lung, pancreas, spleen, spinal cord, and testis (Fig. 2A; Supplementary Table S3). Within the brain, FLT3 transcripts localized to the cerebellum (Fig. 2B). Although FLT3 transcript was not consistently detected in all tissues listed above, all tissue types identified as transcript-positive in any dataset were subsequently evaluated for FLT3 protein expression by IHC (spinal cord being the only exception). Of these tissues, the only example of cell surface-localized FLT3 protein was on rare, scattered cells in the tonsil (Fig. 2C). In all other tissues evaluated, including brain stem, cerebrum, cerebellum, kidney, pancreas, pituitary, prostate, skeletal muscle, stomach, testis, and thyroid, FLT3 protein staining was cytoplasmic (Fig. 2D). Within the cerebrum and cerebellum, FLT3 protein staining consisted of cytoplasmic staining of multifocal neurons, and this staining pattern was consistent in multiple sections of brain, with no membranous staining observed in neurons. Diffuse cytoplasmic staining was observed in alveolar macrophages, indicating the likely source of the transcript signal in lung (Supplementary Table S3). Taken together, these data suggest that although FLT3 transcript and protein are present in peripheral tissues, including the brain, FLT3 protein is cytoplasmic and therefore not anticipated to be targeted by an anti-FLT3 BiTE molecule.

Additional analysis of FLT3 transcript and FLT3 protein expression in the cerebellum revealed that the majority of FLT3 transcripts isolated from the cerebellum were shorter than those isolated from a control AML cell line. Transcript sequencing revealed these truncations were due to frequent intron insertion/retention or exon skipping. Quantification of alternatively-spliced FLT3 transcripts using digital droplet PCR (ddPCR) indicated that in this study at least 70% to 85% of cerebellum FLT3 transcripts lacked exonic regions or retained intronic sequences, suggesting that only a small portion of FLT3 transcripts in cerebellum samples analyzed would be intact (Supplementary Fig. S3). Assessment of FLT3 protein from human cerebellum lysate by immunoprecipitation-western analysis identified only FLT3 protein bands that were lower in molecular weight than full-length FLT3 protein from a positive control AML cell line lysate (Fig. 2E). FLT3 protein bands from a cerebellum sample were characterized by mass spectrometry, revealing only peptides from the extracellular domain of FLT3; by contrast, bands from the control AML cell line lysate contained multiple peptides from both the intracellular and extracellular regions of FLT3 (Supplementary Fig. S3D; Supplementary Table S11). In sum, the transcript and peptide data suggest that most transcripts from the cerebellum encode FLT3 peptides that are not full-length and may explain why FLT3 is not detectable on the cell surface of cells in the cerebellum.

Two different FLT3 BiTE molecules (Supplementary Fig. S4A) were evaluated. Each BiTE molecule comprised a distinct anti-FLT3 scFv that bound FLT3 within a 51 amino acid region, associated with an anti-CD3 scFv. The compact size of BiTE molecules (MW ∼55 kDa) has been reported to be important for the generation of a productive immunologic synapse (21); however, proteins this size are generally rapidly eliminated by the kidneys. To increase the serum half-life, an Fc moiety was added to produce AMG 427. To ensure that the presence of the Fc would not impact in vitro or in vivo activity, the two BiTE molecules were evaluated in similar assay panels. Both molecules bound human FLT3 and CD3 protein with sub- or single-digit nanomolar affinities (Supplementary Table S5). A panel of cell lines exhibiting a range of FLT3 protein expression (MFI ratio: 2.6–23.8; Supplementary Fig. S4B) similar to that observed on pAML samples (Fig. 1A) was selected to evaluate FLT3 BiTE molecule in vitro potency. Both molecules similarly induced TDCC against five FLT3 protein-expressing cell lines with single digit picomolar potency (Fig. 3A; Supplementary Table S6). A relationship between FLT3 expression level and potency was not apparent, likely due to the high E:T ratio. TDCC was similar for both BiTE molecules in cell lines homozygous or heterozygous for wild type (wt) or ITD mutant (mut) FLT3, and selectivity was demonstrated as cell lines lacking FLT3 protein expression were not lysed (Fig. 3A; Supplementary Table S6). TDCC was accompanied by upregulation of the T-cell activation markers CD69 and CD25 and secretion of T-cell-derived effector cytokines IFNγ and TNFα in the presence of FLT3 protein-expressing cells, but not in the presence of FLT3 protein-negative cells (Fig. 3B–E; Supplementary Table S6).

Soluble FLT3 (sFLT3) can be detected in AML patient serum at concentrations up to 141 ng/mL (22). In TDCC assays, clinically relevant concentrations of sFLT3 reduced AMG 427 potency 6-44-fold, but maximum killing was still achieved (Fig. 3F). Soluble FLT3 ligand (sFLT3L) can be detected in AML patient serum at concentrations up to 9 ng/mL (23). Although neither the experimental FLT3 BiTE molecule nor AMG 427 binds the ligand-binding domain of FLT3, sFLT3L binding to FLT3 induces internalization of FLT3 (24), and could alter BiTE-mediated TDCC. In the presence of 10 ng/mL sFLT3L, the potency of AMG 427-mediated TDCC was reduced two- to six-fold (Fig. 3F); however, maximum killing was still achieved in all three cell lines tested. These data demonstrate that FLT3 BiTE molecules induce target-specific TDCC equivalently, and that complete killing occurs in the presence of disease-relevant concentrations of sFLT3 and sFLT3L.

Both the experimental FLT3 BiTE molecule and AMG 427 were evaluated in mouse tumor models. As neither BiTE molecule bound mouse Flt3, immunocompromised mice administered with human tumor cells and T cells were used. The experimental FLT3 BiTE molecule was evaluated in an admix model in which athymic nude mice were injected with MOLM-13 AML cells and in vitro-expanded human CD3⁺ T cells in Matrigel. Animals were dosed intraperitoneally with experimental FLT3 BiTE molecule or control BiTE molecule daily for 10 days. Tumor growth was inhibited by 90% in mice treated with the experimental FLT3 BiTE molecule relative to the control BiTE molecule (n = 10, P < 0.0001; Supplementary Fig. S5).

AMG 427 was evaluated in two orthotopic mouse xenograft models in which either EOL-1 or MOLM-13 AML cells were injected on day 1 and after 72 or 48 hours (EOL-1 and MOLM-13, respectively), mice were injected with in vitro-expanded human CD3⁺ T cells. Mice were treated with vehicle or AMG 427 every 7 days starting on day 9 (EOL-1) or every 5 days starting on day 7 (MOLM-13). In the EOL-1 model, all animals from the control groups developed leukemic disease and were euthanized between days 27 and 52 following AML cell injection with median survival of 36 and 37 days (Fig. 4A). Weekly treatment with AMG 427 prolonged survival at all doses tested, with 17 of 30 animals surviving until study end on day 108. As ≥50% of animals were alive at study end, the median survival could not be calculated; however, compared with vehicle, AMG 427 significantly extended survival (n = 10, P < 0.001; Fig. 4A). In the more aggressive MOLM-13 model, all mice in the control groups died within 20 days after injection of AML cells, with median survival of 18 days. Compared with vehicle, treatment with AMG 427 significantly extended survival at all doses tested (n = 10, P ≤ 0.0015, Fig. 4B). No significant difference in overall survival was observed between different dose levels. Comparable PK profiles were observed within each cohort for all dose levels, and serum concentrations remained above the TDCC assay-determined EC₅₀ for at least 9 days following the final administration (days 34–43, Fig. 4C). Serum half-life of AMG 427 ranged from 33 to 47 hours. These data demonstrate that both FLT3 BiTE molecules were active in vivo in mouse tumor models.

The experimental FLT3 BiTE molecule and AMG 427 bound human and cynomolgus monkey FLT3 and CD3 protein with comparable affinity (Supplementary Table S5). BiTE-induced TDCC was similar for both constructs using either cynomolgus monkey or human effector cells (Fig. 3; Supplementary Fig. S6). To assess the PK/PD relationship, both FLT3 BiTE molecules were evaluated in vivo in cynomolgus monkeys. Both molecules were well tolerated. PD endpoints included FLT3 transcript levels (primer and probes in Supplementary Table S9) in BM and blood and circulating sFLT3L levels. Reduction of FLT3 transcript levels in BM was likely due to direct killing of FLT3 transcript-expressing hematopoietic progenitor cells, and reduction in the blood was likely due to lack of replenishment of FLT3 transcript-expressing cells from the BM. This hypothesis is supported by data showing that there are cells in the BM that express both FLT3 transcript (Fig. 5B and F) and surface-localized FLT3 protein (Supplementary Fig. S7B), making them recognizable by FLT3 BiTE molecules, whereas none of the FLT3 transcript-expressing cells in blood express detectable surface-localized FLT3 protein (Supplementary Fig. S7A), and are therefore not recognizable by FLT3 BiTE molecules.

The experimental FLT3 BiTE molecule was evaluated in a 16-day study (Fig. 5A–D) in cynomolgus monkeys, with intra-animal (n = 3) dose escalations every 3 days intended to achieve steady-state concentrations (C_ss) of 0.05, 0.2, 0.5, and 2 nmol/L (Fig. 5A; Supplementary Table S8). FLT3 transcript levels were reduced in BM at day 17 (the only time point evaluated) relative to non-treated animals (Fig. 5B) and in blood on days 4, 7, 10, and 17, by an average of 85% to 92%, relative to levels measured before treatment (Fig. 5C). Soluble FLT3L levels increased dose-dependently over the course of the study, reaching maximum levels of 13,000 to 15,500 pg/mL at the end of the study (Fig. 5D). Ligand accumulation is likely due to depletion of FLT3 protein-expressing cells. The fold-over-EC₅₀ (in vitro TDCC data; Supplementary Fig. S6) for each of the four dose levels (C_ss ∼0.05, 0.2, 0.5, and 2 nmol/L) was 25-, 64-, 165-, and 780-fold. The percent reduction in FLT3 transcript level in blood did not deepen once drug concentration was above C_ss 0.2 nmol/L (64-fold-over-EC₅₀), suggesting that the concentration required to achieve maximal target cell elimination from blood was somewhere between C_ss 0.05 and 0.2 nmol/L (25- and 64-fold-over-EC₅₀, TDCC data; Supplementary Fig. S6; Supplementary Table S7).

AMG 427 was evaluated in an 8-day multiple dose study (Fig. 5E–H) in cynomolgus monkeys. All animals were treated on days 1, 2, and 5 with doses intended to achieve a maximal serum concentration (C_max) of 1 nmol/L (Group 1), 5 nmol/L (Group 2), and 10 nmol/L (Group 3; n = 3/group). The study duration was limited to 8 days to minimize loss of exposure due to antidrug antibody formation, and multiple doses were administered to ensure target coverage for the entire study. Exposures of AMG 427 over 7 days were reproducible within each of three dose groups (Fig. 5E), and exposure, C_max, and C_min all increased in an approximately dose-proportional manner (Supplementary Table S8). The terminal half-life ranged from 33 to 50 hours (Supplementary Table S9). Hallmarks of BiTE molecule activity including upregulation of CD69 on T cells and cytokine secretion were observed (Fig. 8A and B). Because FLT3 is not expressed on the surface of cells in the blood (Supplementary Fig. S7), AMG 427-mediated upregulation of CD69 on T cells likely resulted from cells expressing FLT3 surface protein in the BM. Increases in serum concentrations of IFNγ, IL6, MCP-1, and TNFα were observed in response to the first dose but were attenuated in response to subsequent doses (Supplementary Fig. S8B). The fold-over-EC₅₀ levels at C_min (in vitro TDCC data; Supplementary Fig. S6; Supplementary Table S9) for the respective groups was 45-, 158-, and 396-fold. Within the BM, FLT3 transcript levels were reduced by 85% to 95% on day 4 and by 93% to 97% on day 8 (Fig. 5F). Within the blood, the FLT3 transcript levels were reduced to a nearly undetectable level (≥97%) at the lowest dose level and earliest time point, and a similar level of depletion was maintained across all higher exposures and time points (Fig. 5G). Monocytes were reduced at the end of the study (Supplementary Fig. S8C), which may reflect lack of replenishment due to direct killing of BM progenitors. Minor decreases in plasmacytoid dendritic cells (pDC; Supplementary Fig. S8C) are challenging to interpret as the number of circulating dendritic cells was low and enumeration of rare cells is prone to error. Soluble FLT3L levels increased dose-dependently over the course of the study, reaching maximum levels of 12,000 to 23,000 pg/mL in each of the three groups (Fig. 5H). Time-dependent improvements in PD were observed for those endpoints which had not already reached maximal levels when first analyzed, as demonstrated by the increase in FLT3 transcript reduction in BM from group 1 on day 4 (85%) to day 8 (93%, Fig. 5F) and changes in sFLT3L (Fig. 5H; Supplementary Table S9). This demonstrates that greater efficacy at a given dose level may be observed with longer duration of exposure.

The in vivo activity of the experimental FLT3 BiTE molecule and AMG 427 was most directly comparable using the PD endpoint of FLT3 transcript in blood. For the experimental FLT3 BiTE molecule, the greatest activity occurred between days 4 and 7 at a C_ss of 0.05 to 0.2 (25- to 64-fold-over-EC₅₀; Supplementary Table S8). For AMG 427, the greatest activity was observed on day 3 at a C_min of 0.24 nmol/L in Group 1 (≤45-fold-over-EC₅₀; Supplementary Table S9). Although the time points of data collection differed, these results suggest that both FLT3 BiTE molecules are active at similar fold-over-EC₅₀ values in vivo.

A long-term culture system (25) was used to evaluate experimental FLT3 BiTE molecule-mediated cytotoxicity in 14 pAML samples (Supplementary Table S10) over 9 days. The autologous E:T ratio was calculated from the number of T cells and pAML cells in each sample at the beginning of the experiment and ranged from 1:2.5 to 1:74. Three patterns of cytotoxicity were observed: (1) continuously-increasing cytotoxicity (Fig. 6A, left, representative sample Supplementary Fig. S9A); (2) initial cytotoxicity followed by sustained or decreased killing (Fig. 6A, middle); (3) transient or no cytotoxicity over the 9 days (Fig. 6A, right). Analysis of FLT3 surface protein expression of the pAML cells and E:T ratio revealed that most of the samples in groups 1 and 2 contained FLT3 protein-positive pAML cells (MFI Ratio > 1.5) and a higher E:T ratio (>1:38, 75^th percentile; Fig. 6B), whereas most of the samples in group 3 expressed low levels of FLT3 protein (MFI ratio <2) and/or had a low E:T ratio (<1:38). These data demonstrate that both target expression and T-cell abundance are important factors for FLT3 BiTE-mediated target cell killing.

T-cell activation induces PD-1 expression, and reports show that PD-1 engagement by ligands PD-L1 or PD-L2 decreases T-cell activity (26). Co-culture of pAML specimens with a CD33-targeting BiTE molecule induces PD-1 expression on T cells and PD-L1 expression on AML blasts (27). Similarly, AMG 427-mediated T-cell activation induced a dose-dependent increase in PD-1 expression (Supplementary Fig. S9B), and potency in TDCC assays was reduced five-fold if the target cells expressed PD-L1 (relative to target cells lacking PD-L1; Fig. 6C). Combination of a PD-1–blocking antibody with AMG 427 restored TDCC potency, decreasing the EC₅₀ by an average of 2.5-fold (n = 3 T-cell donors, P = 0.02) and increasing maximum killing by 12% (Fig. 6D and E). These data demonstrate that AMG 427-mediated target cell killing may be enhanced by combination with PD-1 blockade, as has been demonstrated for other BiTE molecules (27, 28).

Blinatumomab demonstrates that a BiTE molecule can engage patient T cells to eliminate CD19-expressing disease cells, and this activity can provide clinical benefit for patients with acute lymphoblastic leukemia and non-Hodgkin lymphoma (29–31). Here FLT3 BiTE molecules for the treatment of patients with AML are characterized. FLT3 meets the requirements of a BiTE molecule target as cell surface protein is broadly expressed on disease samples, with limited expression on normal tissues. In disease samples, cell surface FLT3 protein was detected on the majority of 318 pAML samples. The level of expression was comparable between bulk samples and LSCs, suggesting that both subsets could be targeted at similar therapeutic exposures. In addition, mean FLT3 protein expression on pAML samples was comparable, regardless of FLT3 mutational status, FLT3-ITD allelic ratio, or initial diagnosis versus relapse, suggesting that a FLT3 BiTE molecule would benefit a broad patient population. In normal hematopoietic cells, cell surface FLT3 protein was detected on subsets of BM stem and progenitor cells (excluding MEPs). Within each subpopulation, a portion of cells were cell surface FLT3-positive (MFI ratio ≥ 1.5), so as to suggest their elimination by a FLT3-targeting BiTE molecule. However, a portion of these cells were FLT3 protein-negative, consistent with literature reports of FLT3 protein- and transcript-negative cells within HD hematopoietic progenitor populations and HSCs (GEO accession code GSE75478; refs. 7, 32).These results suggest that although some HSPCs would be eliminated by a FLT3 BiTE molecule, there is a FLT3-negative population that could potentially repopulate the BM following cessation of treatment. FLT3 transcript and protein expression were also evaluated in normal non-hematopoietic tissues. To ensure a thorough assessment, FLT3 transcript expression was evaluated in several databases and further characterized by qPCR-based analysis of a panel of tissues. Protein expression was subsequently evaluated in tissues shown to contain FLT3 transcript in any dataset. Although FLT3 transcript and protein were detected in some solid tissues, no membranous protein staining was observed, indicating that these cells would not be targeted by a FLT3 BiTE molecule. Additional analysis of FLT3 transcript and protein in the cerebellum demonstrated that most transcripts were not full-length due to alternative splicing, and similarly, the FLT3 protein was also not full-length. By evaluating both FLT3 transcript and protein expression using multiple sources and orthogonal methods, it was possible to build a detailed understanding of the normal tissue expression and based on these results, FLT3 BiTE treatment is not anticipated to target normal non-hematopoietic tissues.

Given the favorable expression profile of FLT3 as an AML target, two potent and specific BiTE molecules were generated: one experimental FLT3 BiTE molecule that has a short serum half-life and the other, AMG 427, which contains an Fc-moiety to extend serum half-life. In vitro, these molecules demonstrated TDCC against human FLT3-positive cancer cell lines with similar picomolar potency (EC₅₀) and this TDCC was associated with T-cell activation and cytokine secretion and was not affected by the presence of sFLT3 or sFLT3L at concentrations found in patients with AML. These data demonstrate it is possible to generate a BiTE molecule capable of eliminating FLT3-expressing cells, and that despite incorporation of the Fc moiety, the larger size does not impact the ability of AMG 427 to effectively form an immunologic synapse and induce T-cell–mediated target cell killing.

In cynomolgus monkeys, both FLT3 BiTE molecules mediated depletion of cell surface FLT3-expressing target cells as demonstrated by decreases in FLT3 transcript in the blood and BM. Although cell surface FLT3 protein expression was not detected on human or cynomolgus monkey peripheral immune cells, FLT3 transcript can be detected in pDCs and monocytes. These cell types have short half-lives in vivo (33, 34), and administration of a FLT3 BiTE molecule is expected to eliminate a portion of the precursor cells that give rise to them, which may explain the decreases in FLT3 transcript observed in the blood of cynomolgus monkeys treated with a FLT3 BiTE molecule. Within the BM, FLT3 transcript was reduced by ≥85% at all doses by the first timepoint tested (day 4), demonstrating that BiTE-mediated target cell killing can occur rapidly. At this same time point, the degree of depletion increased as the dose increased [85% (low dose) vs. 95% (medium dose), and 93% (high dose), respectively], suggesting that increased exposure can lead to deeper responses. Within the low-dose group, the reduction in FLT3 transcript levels increased from 85% to 93% between days 4 and 8, suggesting that deeper responses may also be achieved by maintaining the same exposure for longer. This hypothesis is supported by the sFLT3L endpoint, which improved with either higher exposure or increased time of exposure.

BiTE molecule-mediated lysis of AML blasts within patient samples was evaluated in long-term culture assays using autologous T cells. The degree of anti-AML activity was associated with FLT3 expression on the target cells and the E:T ratio, with improved activity seen in AML samples with a higher FLT3 protein expression and an E:T ratio >1:38. The impact of the E:T ratio highlights the importance of T-cell fitness to enable successful responsiveness to BiTE molecule therapy. One well-established mechanism of reducing T-cell activity is induction of PD-1 expression. BiTE molecule–mediated T-cell activation is accompanied by expression of PD-1 on corresponding T cells and this expression has been associated with resistance to blinatumomab treatment (28, 35). PD-1 is expressed on 20% to 30% of AML patient T cells (36, 37) and has been shown to increase to 50% to 60% at relapse (38). PD-L1 mRNA expression is upregulated in patients with AML (39) and correlates with cell surface protein expression (40). Although not usually detected at diagnosis (41), PD-L1 protein is upregulated on AML blasts during therapy, after HSCT, and at relapse (39, 42). Upregulation of PD-L1 on AML blasts is reported to be induced by cytokines such as IFNγ (43, 44), which may be the mechanism of PD-L1 upregulation on pAML blasts treated ex vivo with a CD33-targeting BiTE molecule (27). In a mouse model engineered to express human CD3, combination studies of BiTE molecules with checkpoint inhibitors exhibit additive effect (45). Herein, AMG 427-mediated activation of T cells was associated with rapid induction of PD-1 expression and subsequent reduced killing of PD-L1–expressing target cells, suggesting that combination with PD-1 blockade may improve BiTE-mediated activity. Indeed, the combination of BiTE molecule and a PD-1–blocking antibody in a TDCC assay resulted in decreased EC₅₀ and increased maximum killing in all donors tested. As expression of checkpoint molecules, including PD-1, has been observed in patients with AML (39), and may be increased following chemotherapy (43), this combination therapy warrants clinical evaluation.

R.L. Goldstein is a scientist at Amgen Inc. and reports ownership of Amgen Inc. stock. C.M. Karbowski is an employee/paid consultant at Amgen, Inc. and reports ownership of Amgen Inc. stock. A. Henn is a senior scientist at and has ownership interest (including patents) in Amgen Research Munich GmbH. C.-M. Li is a principal scientist at and reports of receiving a commercial research grant from Amgen Inc. V.L. Bücklein has received speakers bureau honoraria from Pfizer. P. Koppikar has ownership interest (including patents) in and has provided expert testimony for Amgen Inc. S. Haubner has ownership interest (including patents) with royalty payment from MSKCC. J. Wahl is a senior scientist at Amgen Inc. and reports ownership of Amgen Inc. stock. C. Dahlhoff is a senior scientist at Amgen Research Munich GmbH and reports ownership of Amgen Inc. stock. T. Raum is an executive director at and has ownership interest (including patents) in Amgen Research Munich GmbH. M.J. Rardin is a senior scientist at Amgen Inc. and reports ownership in Amgen Inc. stock. B. Frank is a senior research associate at Amgen Inc. and reports ownership in Amgen Inc. stock. K.H. Metzeler reports receiving a commercial research grant from Celgene and has received speakers bureau honoraria from Otsuka, Pfizer, Daiichi Sankyo, and Celgene. R. Case is a principal scientist at Amgen Inc. and reports ownership in Amgen Inc. stock. M. Friedrich is a scientific director at and has ownership interest (including patents) in Amgen Research Munich GmbH. A. Coxon is a vice president of research at Amgen Inc. and reports ownership of Amgen Inc. stock. M. Subklewe is a paid consultant at, reports receiving other commercial research support from, and has received speakers bureau honoraria from Amgen Inc. T. Arvedson is an executive director at and has ownership interest (including patents) in Amgen Inc. No potential conflicts of interest were disclosed by the other authors.

B. Brauchle: Conceptualization, methodology, data curation, formal analysis, writing-original draft, writing-review and editing. R.L. Goldstein: Conceptualization, methodology, data curation, formal analysis, writing-original draft, writing-review and editing. C.M. Karbowski: Conceptualization, methodology, data curation, formal analysis, writing-original draft, writing-review and editing. A. Henn: Conceptualization, methodology, data curation, formal analysis, writing-original draft, writing-review and editing. C.-M. Li: Methodology, data curation, formal analysis, writing-original draft, writing-review and editing, project administration. V.L. Bücklein: Data curation, formal analysis, writing-original draft, writing-review and editing. C. Krupka: Conceptualization. M.C. Boyle: Conceptualization, methodology, data curation, formal analysis, writing-original draft, writing-review and editing. P. Koppikar: Conceptualization, data curation, formal analysis, writing-original draft, writing-review and editing. S. Haubner: Formal analysis. J. Wahl: Conceptualization, formal analysis, writing-original draft, writing-review and editing. C. Dahlhoff: Conceptualization, formal analysis, writing-original draft, writing-review and editing. T. Raum: Conceptualization, formal analysis, writing-original draft, writing-review and editing. M.J. Rardin: Formal analysis, writing-original draft, writing-review and editing, project administration. C. Sastri: Methodology, data curation, formal analysis, writing-original draft, writing-review and editing. D.A. Rock: Conceptualization, formal analysis, writing-original draft, writing-review and editing. M. von Bergwelt-Baildon: Writing-original draft, writing-review and editing. B. Frank: Conceptualization, methodology, data curation, formal analysis, writing-original draft, writing-review and editing. K.H. Metzeler: Data curation, formal analysis, writing-original draft, writing-review and editing. R. Case: Methodology, data curation, formal analysis, writing-original draft, writing-review and editing. M. Friedrich: Conceptualization, formal analysis, writing-original draft, writing-review and editing. M. Balazs: Conceptualization, methodology, data curation, formal analysis, writing-original draft, writing-review and editing. K. Spiekermann: Formal analysis, writing-original draft, writing-review and editing. A. Coxon: Conceptualization, writing-original draft, writing-review and editing, study supervision. M. Subklewe: Conceptualization, methodology, data curation, formal analysis, writing-original draft, writing-review and editing, project administration, study supervision. T. Arvedson: Conceptualization, methodology, formal analysis, writing-original draft, writing-review and editing, study supervision.

Qualified researchers may request data from Amgen clinical studies. Complete details are available at the following: https://wwwext.amgen.com/science/clinical-trials/clinical-data-transparency-practices/.

We thank Kelly Hensley for IHC support, Bradford Gibson for proteomic expression analysis, Ivonne Archibeque and Angus Sinclair for BiTE characterization, Natalia Grinberg for Octet binding affinity studies, Herve Lebrec for discussions, Oliver Homann for gene expression analysis, Sandra Ross and Elizabeth Leight for discussions and excellent medical writing support, and Urszula Domanska for xenograft support. This study was funded by Amgen, Inc.

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.