HPN328, a Trispecific T Cell–Activating Protein Construct Targeting DLL3-Expressing Solid Tumors

Lung cancer is the leading cause of cancer death worldwide (1). Small cell lung cancer (SCLC) represents 15% to 20% of lung cancer diagnoses and is a highly aggressive and lethal form of lung cancer. There is a strong association with SCLC incidence and tobacco exposure (2). Few effective treatment options exist for patients, and the 5-year survival rate is only about 7% (3). The mainstay of treatment for SCLC has been chemotherapy, with checkpoint inhibitors recently approved as frontline treatment in combination with chemotherapy (4). There remains, however, an urgent unmet clinical need for better therapies to treat this disease.

Novel SCLC therapies have targeted DLL3, a single-pass type I transmembrane protein, identified as being overexpressed in cells isolated from small cell lung cancer (SCLC) and large cell neuroendocrine cancer patient-derived xenografts (5). The prevalence of DLL3 expression in SCLC has been reported to be in the range of 68% to 100% (6, 7). In addition to SCLC, DLL3 is also expressed in other malignancies including glioma (8), neuroendocrine prostate cancer (9), gastrointestinal neuroendocrine malignancies (10), and small cell bladder cancer (11). By contrast, healthy adult tissues have shown limited surface expression of DLL3. In normal tissues, expression of DLL3 was detected at very low levels in the brain, pituitary, and pancreatic islets and showed a cytoplasmic staining pattern by IHC (12–14). Based on its elevated expression in tumors and its restricted and low expression in normal tissues, DLL3 has become an attractive target for antibody- and cell-based therapeutics. Modalities exploring DLL3 as a target in research and clinical development include antibody–drug conjugates, CAR T cells, and CD3-based bispecific T cell–engaging molecules (12–16). Objective responses have been reported for the DLL3-targeting antibody–drug conjugate (ADC) rovalpituzumab tesirine (Rova-T; refs. 12 and 14), although a phase 3 trial failed to demonstrate the survival benefit (17). Dose-limiting toxicity of Rova-T seems to be mainly associated with the tesirine payload and not DLL3. Although the clinical development of Rova-T has been discontinued, its anti-tumor activity in patients has provided clinical validation for DLL3 as an attractive target for other targeting modalities.

Here, we describe HPN328, a trispecific T cell–activating construct (TriTAC) targeting DLL3, which is currently being tested in a phase 1 clinical trial for the treatment of SCLC and other DLL3-expressing tumors (NCT04471727). TriTACs are designed to treat solid tumors by overcoming the challenges seen with previous generations of T cell–engaging molecules. They are engineered to have a long serum half-life, small size, and high stability to improve patient safety and solid tumor activity (18, 19). HPN328 consists of a single-domain antibody (sdAb) specific for DLL3, a sdAb specific for serum albumin for half-life extension, and a single-chain fragment variable specific for the CD3ε subunit of the T-cell receptor complex for T-cell engagement. HPN328 redirects T cells to specifically and potently kill DLL3-expressing SCLC cells in vitro. In an immunocompetent mouse model expressing a human CD3ε epitope (hCD3ε mice), HPN328 induced potent anti-tumor activity and long-term immune memory confirmed by tumor rechallenge studies. HPN328 exhibits a favorable pharmacokinetic and safety profile in cynomolgus monkeys.

Expression constructs were generated for extracellular domains of human and cynomolgus monkey DLL3, as well as human, cynomolgus monkey, and mouse CD3ε, each fused to human Fc using pCDNA 3.4 (Life Technologies, A14697) or pINFUSE-hIgG1-Fc1 (InvivoGen) as DNA vector backbones. Fusion proteins were expressed in Expi293F cells using ExpiFectamine according to kit protocols (Life Technologies). Unless noted otherwise, DLL3-targeting TriTAC expression constructs were made and protein production was performed as described previously (18). HPN328 was expressed by generation of stable CHO-DG44 cell pools containing an expression vector with HPN328 coding sequence using the CHEF-1 expression system (AGC Biologics). Protein was purified from conditioned medium by protein A affinity, desalting, and mixed-mode ion-exchange chromatography or strong cation exchange chromatography and was subsequently formulated in appropriate buffers and excipients. Protein purity and quality were assessed by denaturing SDS-PAGE (Life Technologies) and analytical size-exclusion chromatography (high-performance liquid chromatography) using Yarra SEC-2000, SEC-X150, or SEC-X300 columns or BEH SEC 200 columns resolved on Agilent 1200 or 1290 LC systems and analyzed using ChemStation software. Sequences are provided in Example 5 of reference PCT/US2019/053017, published as WO 2020/069028 (20).

Affinities of HPN328 for albumin, CD3ε, and DLL3 ligands were measured by biolayer interferometry using an Octet RED96 instrument with streptavidin tips (Pall ForteBio). Experiments were performed at 27°C in PBS plus casein in the absence or presence of 15 mg/mL human serum albumin (HSA) as described in “Results” and figure legends (Supplementary Figs. S4–S7). Binding sensorgrams generated from empirically determined ligand loads, appropriate serial dilutions of known analyte concentrations, and association and dissociation times were then fit globally to a one-to-one binding model using Octet Data Analysis 9.0 software.

All cancer cell lines were obtained from the ATCC. Cancer cell lines were transduced with firefly luciferase lentivirus with puromycin selection marker (Biosettia GlowCell-14p) and were selected by the addition of puromycin. Cell lines were passaged a maximum of 36 times after being received from the ATCC. Cell line authentication and Mycoplasma testing were not performed.

MC38 cells that stably express hDLL3 or hEGFR were transduced with the lentivirus vector encoding the respective human antigen (i.e., hDLL3, hEGFR) with puromycin selection marker and were selected by the addition of puromycin. Single-cell clones were then generated. Healthy donor T cells were purified from Leukopaks (leukapheresis samples, STEMCELL Technologies) using the EasySep Human T Cell Isolation Kit (STEMCELL Technologies), following the manufacturer’s instructions.

For T cell–dependent cell cytotoxicity (TDCC) assays (21), target cells were engineered to express luciferase, and purified primary resting human T cells were seeded in a 384-well plate at a 10:1 T-cell-to-target-cell ratio. Target cell viability was assessed following incubation for 48 hours at 37°C and 5% CO₂. Because TriTACs contain an albumin-binding domain for half-life extension and are expected to be bound to albumin in serum and tissue when dosed in humans, a TDCC assay was conducted in the presence of 15 mg/mL HSA to simulate the in vivo conditions.

Target cell viability was assessed by incubation with the Steady-Glo reagent (Promega). Luminescence was measured using a PerkinElmer EnVision detection system. The gating strategy for T-cell activation assay was CD3/CD69 and CD3/CD25. The expression of CD3, CD69, and CD25 on T cells was measured by using flow cytometry with anti-CD3 clone SP34-2, anti-CD25 clone M-A251, and anti-CD69 clone FN50 antibodies (BD Biosciences). The samples were analyzed on FACSCelesta (BD Biosciences). Flow cytometry data were processed using FlowJo v10 software (FlowJo, LLC).

Cultured cells or purified human T cells were incubated with HPN328 or anti-GFP TriTAC (control) for 1 hour. Binding was detected by using Alexa Fluor 647 anti-TriTAC antibody in a FACSCelesta flow cytometer (BD Biosciences).

To measure cytokines, a commercially available electrochemiluminescence assay was used per the manufacturer’s instructions (Meso Scale Discovery V-PLEX Proinflammatory Panel 1 Human Kit, K15049D). Plates containing conditioned media from TDCC assays were used for cytokine analyses. Plates were read on a MESO SECTOR S 600 plate reader (Meso Scale Discovery).

hCD3ε mice were developed and bred by genOway. Six amino acids corresponding to the human epitope to which TriTAC molecules bind were added to exon 4 of the mouse Cd3e locus, resulting in human/mouse chimeric CD3ε cDNA under the control of the endogenous mouse Cd3e promoter. Embryonic stem cells carrying the human/mouse chimeric Cd3e locus were validated by PCR and sequencing. Chimeras were generated by the injection of embryonic stem cells into blastocysts and subsequent breeding. Following colony generation, hCD3ε syngeneic mouse lines were maintained at Charles River Laboratories.

All mouse studies were performed in accordance with the policies and approved by the Institutional Animal Care and Use Committee at Harpoon Therapeutics, Inc. For the SHP-77 experiment, NSG (NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ) mice received subcutaneous co-implants of SHP-77 cells (1 × 10⁷) and ex vivo expanded human T cells (5 × 10⁶) in 50% Matrigel (BD Biosciences) on day 0. Mice were dosed on days 1 to 10 via i.p. injection. For NCI-H82 experiments, NSG mice received subcutaneous implants of 5 × 10⁶ cancer cells in 50% Matrigel on day 0. On day 8, when tumors reached approximately 168 mm³, the mice were randomized and injected intraperitoneally with ex vivo expanded human T cells (2 × 10⁷). Treatment was initiated the following day by i.p. injection, from days 9 to 18. The tumor size was measured twice weekly and calculated using the following formula: tumor volume (mm³) = (width² × length)/2. Percent tumor growth inhibition was defined as the difference between the mean tumor volume (MTV) of the control group and the MTV of the treated group, expressed as a percentage of the MTV of the control group.

For syngeneic studies, MC38, MC38-hDLL3, and MC38-hEGFR cells were implanted with 2 × 10⁶ cancer cells in 50% Matrigel on day 0. For MC38-hEGFR studies, treatment was initiated the following day by i.p. injection for 10 days. For MC38-hDLL3 studies, when the average tumor size reached approximately 250 mm³, the mice were randomized. Treatment was initiated the following day by i.p. injection for 10 days. For rechallenge studies, when tumors fully receded, MC38 cells were implanted on the opposite flank on study day 30. An additional set of naïve mice that had not been previously treated were implanted on the same flank as that of the rechallenge mice as a control. Tumors from treated mice were minced with razor blades and further homogenized using a Mouse Tumor Dissociation Kit and gentleMACS Octo Dissociator (Miltenyi Biotec). Homogenized tumor cell samples were then run through a 70-µm cell strainer (Miltenyi Biotec). The cell suspensions were washed with FACS buffer (PBS + 2% FBS), blocked with Mouse Fc block (Tonbo), and surface stained with 2.5 µg/mL of antibody in FACS buffer for 30 minutes on ice. Flow data acquisition was performed with a FACSCelesta (BD Biosciences), and the data were analyzed using FlowJo (FlowJo). Markers and antibodies used for immunophenotyping experiments can be found in the Supplementary Tables S1 and S2.

Mouse tumors were harvested at day 6 post dose, fixed overnight in 4% neutral buffered formaldehyde, then stored in 70% ethanol until embedded in paraffin. IHC staining was performed using the Ventana Discovery Ultra autostainer by UCSF Histology and Biomarker Core. NCI-H82 tumors were stained with anti-human CD3 (Ventana 2GV6), anti-human CD4 (Ventana SP35), or anti-human CD8 (Ventana SP57).

The pharmacology, pharmacokinetics, and toxicity of HPN328 were evaluated after repeated intravenous bolus doses once weekly for 4 weeks (study days 1, 8, 15, 22) of the vehicle or 1 or 10 mg/kg HPN328 in 5 male and 5 female cynomolgus monkeys per group followed by either a 1- or 4-week post-last-dose recovery period. The study followed the Institutional Animal Care and Use Committee–approved protocol and standard operating procedures of the testing facility (Charles River Laboratories). Pharmacologic activity based on the hypothesized mechanism of action for HPN328 was evaluated by serum cytokine assessments and flow cytometry. A ligand-binding electrochemiluminescence assay, measuring functional interaction between HPN328 and its targets, was used for measuring HPN328 levels in the serum. HPN328 was captured with biotinylated recombinant CD3ε and detected with a sulfo-tagged recombinant DLL3. Toxicokinetic parameters were estimated using Phoenix WinNonlin software. A ligand-binding electrochemiluminescence assay was used to detect anti-HPN328 antibodies [anti-drug antibodies (ADA)] in the serum following drug exposure. A non-compartmental approach consistent with the intravenous bolus route of administration was used for parameter estimation. Safety and toxicity endpoints included daily morbidity and mortality, daily clinical observations, weekly body weights, daily food consumption, clinical pathology (hematology, clinical chemistry, and coagulation), and anatomic pathology (gross necropsy, organ weights, and histopathology).

Statistical analyses were performed using Prism software (GraphPad). One-way ANOVA followed by Tukey’s post hoc test was applied where noted.

The data generated in this study are available within the article and its supplementary data files.

Built upon the TriTAC platform that has been previously described (18, 19), HPN328 is a DLL3-targeting T-cell engager with extended half-life via serum albumin binding. It is produced as a single-chain polypeptide with a molecular weight of approximately 50 kDa, consisting of a humanized single-chain fragment variable specific for the human CD3ε subunit of the T-cell receptor complex, a humanized llama sdAb specific for HSA, and a humanized llama sdAb selective for human DLL3 (Supplementary Fig. S1). HPN328 protein is pure as demonstrated by SDS-PAGE analysis (Supplementary Fig. S2A). Stability studies subjecting HPN328 to various stress conditions suggest that the protein is stable and stress resistant (Supplementary Fig. S2B).

Because TriTAC molecules consist of three domains, they can be made in six configurations. We previously observed that EGFR-targeting and prostate-specific membrane antigen–targeting TriTACs in the anti-target:anti-albumin:anti-CD3 (T:A:C) configuration were most potent in directed T-cell killing assays (18). In screening anti-DLL3 TriTACs made in different configurations, we observed that a molecule in the anti-CD3:anti-albumin:anti-DLL3 (C:A:T) configuration was more potent than a molecule in the T:A:C configuration. To confirm this result, TDCC assays were performed to confirm this potency difference. In co-cultures of DLL3-expressing NCI-H82, SHP-77, DMS 53, and NCI-H2171 SCLC cells and resting T cells from four different donors, the EC₅₀ values in the C:A:T configured TriTACs (Fig. 1A and B; Supplementary Fig. S3; Supplementary Table S3) were on average 3.1-fold more potent than the EC₅₀ values in the T:A:C configured TriTACs (Supplementary Table S4). Based on these data, HPN328 is configured in the C:A:T configuration.

HPN328 efficiently directed T cells to lyse SCLC target cells with EC₅₀ values ranging from 16 to 286 pmol/L (Fig. 1A and B; Supplementary Fig. S3; Supplementary Table S3) with one donor cell line pair (NCI-H2171 combined with donor 86) failing to reach 50% killing. Specificity of DLL3 targeting by HPN328 was supported by the observation that the control TriTAC targeting GFP was unable to kill these cell lines, and HPN328 was inactive with cell lines that lacked DLL3 expression, HCT116, and NCI-H292 (Fig. 1C; Supplementary Fig. S3; Supplementary Table S3).

HPN328 binds with high affinity to human and cynomolgus monkey DLL3, CD3ε, and albumin in the range of 0.13 to 28 nmol/L (Supplementary Figs. S4 and S5) as assessed by biolayer interferometry. The K_D values for binding of HPN328 to mouse DLL3 and ALB were 10.3 and 436 nmol/L, respectively, but did not detectably bind mouse CD3ε (Supplementary Fig. S6). The K_D values for binding to human and cynomolgus orthologs were found to be highly similar for all three target antigens identifying cynomolgus monkeys as a relevant toxicological species. To confirm binding to cynomolgus DLL3 (cyDLL3) with a functional assay, HEK 293 cells genetically engineered to express cyDLL3 were incubated with cynomolgus monkey peripheral blood mononuclear cells in the presence of HPN328 killed with an EC₅₀ value of 26 pmol/L (Fig. 1D). Combined with HPN328 binding (Supplementary Fig. S5), these data further confirm the target specificity of HPN328 and support the use of cynomolgus monkey as a relevant species to evaluate the on-target effects of HPN328 in toxicology studies.

To assess the selectivity of HPN328 for DLL3 over other delta-like ligand family members, biolayer interferometry was used to determine the binding kinetics and affinity of HPN328 to human (hu) DLL1 and DLL4. There was no detectable binding of HPN328 to huDLL1 (Supplementary Fig. S7). HPN328 was observed to bind to huDLL4 with a K_D of 23 ± 1 nmol/L, indicating a 175-fold lower binding affinity to huDLL4 than to huDLL3 (Supplementary Fig. S7). Since cynomolgus monkey is a relevant toxicological species, the ability of HPN328 to bind to both cynomolgus (cy) DLL3 and DLL4 was also compared. HPN328 was observed to bind to cyDLL3 with a K_D of 0.59 ± 0.02 nmol/L and to cyDLL4 with a K_D of 14 ± 0.1 nmol/L (Supplementary Fig. S5), within five-fold and two-fold of the affinity of HPN328 for huDLL3 and huDLL4, respectively (Supplementary Figs. S4 and S7).

Flow cytometry was used to confirm the biolayer interferometry findings. HPN328 exhibited surface binding to human DLL3-expressing NCI-H82 SCLC cells and purified human T cells from two different donors (Supplementary Fig. S8A and S8B), confirming that HPN328 recognized native DLL3 and CD3ε expressed on cells. Furthermore, HPN328 was unable to bind DLL3-negative HCT 116 cells (Supplementary Fig. S8A).

When binding to both T cells and DLL3-expressing cells, HPN328 is expected to activate T cells. To confirm T-cell activation, induction of CD69 and CD25 surface expression on T cells was measured. T cells from four healthy donors were co-cultured with DLL3-positive NCI-H82 or SHP-77 SCLC cell lines (Fig. 2A and B; Supplementary Fig. S9). Depending on the T-cell donor, between 5% and 25% of the T cells in culture upregulated CD69 and CD25 surface expression and were dependent on the amount of HPN328 present. A control TriTAC targeting GFP did not increase CD69 or CD25 expression on T cells. In addition, cells that lacked DLL3 expression, HCT116 (Supplementary Fig. S10), failed to increase CD69 or CD25 expression on T cells following the addition of HPN328. The EC₅₀ values of HPN328 for upregulation of CD69 and CD25 in the presence of DLL3-positive cancer cells ranged from 23 to 1,304 pmol/L (Supplementary Tables S5 and S6).

The release of inflammatory cytokines was also used to assess T-cell activation from T cells co-cultured with DLL3-expressing cancer cells. HPN328 mediated a dose-dependent secretion of TNFα and IFNγ into media from NCI-H82 and SHP-77 co-cultures (Fig. 2C and D; Supplementary Table S7), indicative of T-cell activation. A control TriTAC targeting GFP did not cause cytokine release by T cells (Fig. 2C and D) nor were cytokines released from T cells in the co-culture with DLL3-negative tumor cell lines HCT116 (Supplementary Fig. S11), further confirming that HPN328 is DLL3 specific.

The anti-tumor activity of HPN328 was assessed in two xenograft mouse models established from SCLC that expresses DLL3. In a first study, SHP-77 tumor cells were subcutaneously co-implanted with ex vivo expanded T cells from a normal human donor at an effector-to-target cell ratio of 1:2 into NSG mice (Fig. 3A). A daily dosing regimen was used because previous work with an EGFR-targeting TriTAC molecule was determined to have a serum half-life of 15 to 18 hours (18). Daily doses of HPN328 of 100, 10, or 1 µg/kg were administered over 10 days by i.p. injection starting 1 day after implantation. Doses of 10 and 100 µg/kg HPN328 prevented SHP-77 tumor growth, although 1 µg/kg did not affect tumor growth (Fig. 3A) when compared with a control anti-GFP TriTAC at 100 µg/kg.

In a second SCLC tumor model, NSG mice were injected subcutaneously with 5 × 10⁶ NCI-H82 cells. When tumors reached an average of ∼170 mm³, 2 × 10⁷ex vivo expanded T cells from a normal human donor were implanted by i.p. injection. Mice were treated for 10 days with 1, 10, and 100 µg/kg HPN328 from 1 day since implantation with human T cells. On day 24 post tumor implantation, 10 and 100 µg/kg HPN328 resulted in significant tumor growth inhibition of 61% and 96%, respectively, when compared with control anti-GFP TriTAC–treated mice (Fig. 3B). On day 6 post treatment, three tumors were collected from the 100 µg/kg anti-GFP TriTAC–treated and 100 µg/kg HPN328-treated groups and processed for IHC analysis of human CD3, CD4, and CD8 (Fig. 3C; Supplementary Fig. S12). The HPN328-treated mice showed extensive T-cell recruitment of both CD4- and CD8-positive cells on day 6 post treatment compared with no T-cell recruitment in the anti-GFP TriTAC–treated group. These results demonstrate the activity of HPN328 in recruiting T cells and controlling the growth of established DLL3-expressing SCLC tumors.

To assess the activity of HPN328 in an immunocompetent mouse model, we generated a mouse strain in which six amino acid residues corresponding to the portion of human CD3ε that binds to TriTAC molecules were added to the mouse CD3ε subunit (Supplementary Fig. S13A). Lymphocyte populations in the spleens of homozygous knock-in mice (hCD3ε) were compared with wild-type (WT) mice using flow cytometry. Compared with WT mice, hCD3ε mice demonstrated equal prevalence of T cells, B cells, CD4⁺ and CD8⁺ T cells, NK cells, macrophages, and dendritic cells (Supplementary Fig. S13B–S13D, respectively). As expected, flow cytometry analysis revealed that T cells from homozygous hCD3ε mice express both human and mouse CD3 (Supplementary Fig. S13E) and bind anti-EGFR–targeting TriTAC molecule (Supplementary Fig. S13F). T cells from hCD3ε mice could also be activated ex vivo by incubation with anti-human CD3 and anti-mouse CD28 beads, as demonstrated by proliferation (Supplementary Fig. S14A) and release of IL-2 and IFNγ (Supplementary Fig. S14B and S14C, respectively). Next, we evaluated the in vitro cytotoxic activity of hCD3ε T cells against MC-38 murine tumor cells engineered to express human EGFR (MC38-hEGFR). Following ex vivo activation and expansion, in a TDCC assay, hCD3ε T cells from hCD3ε mice were able to potently kill MC38-hEGFR cells in the presence of an anti-hEGFR–targeting TriTAC molecule (Supplementary Fig. S15A). Together demonstrating that T cells from hCD3ε mice can bind and exert cytotoxic activity in the presence of TriTAC molecules in vitro.

To further validate T-cell activity in the hCD3ε model, we next assessed TriTAC activity in vivo in this model. Homozygous hCD3ε or WT C57BL/6 mice were injected subcutaneously with 2 × 10⁶ MC38-hEGFR cells. Daily doses of anti-hEGFR TriTAC or control anti-GFP TriTAC of 1 mg/kg were administered over 10 days by i.p. injection, starting 1 day after implantation. As expected, no activity was observed in WT mice treated with anti-hEGFR TriTAC (Supplementary Fig. S15B). Anti-hEGFR TriTAC treatment induced a strong anti-tumor response compared with the control anti-GFP TriTAC at 1 mg/kg indicating that the TriTAC molecules can engage and activate T cells in hCD3ε mice.

After confirming that hEGFR TriTAC molecules exhibit cytotoxic activity against murine syngeneic tumor line expressing human EGFR in vivo, we evaluated the activity of HPN328 against MC38 cells engineered to express human DLL3 (MC38-hDLL3) in the hCD3ε mouse model. HPN328 exhibited surface binding to the MC38-hDLL3 cell line (Supplementary Fig. S16). Mice received s.c. injection of 2 × 10⁶ MC38-hDLL3 cells, and tumors grew to ∼215 mm³ before daily doses of 8, 40, 100, and 200 µg/kg of HPN328 or the vehicle were administered over 10 days by i.p. injection. HPN328 exhibited a dose-dependent anti-tumor response with 40 and 100 µg/kg leading to partial tumor growth inhibition and 200 µg/kg completely eradicating tumors (nine of nine mice were cured) when compared with vehicle-treated mice (Fig. 4A). To confirm that HPN328 is indeed activating and engaging intratumoral T cells, a follow-up experiment as shown in Fig. 4A was performed, in which we dissociated tumors following treatment with two doses of 100 µg/kg HPN328 or vehicle and performed FACS analysis. HPN328 treatment led to increases in CD25 and CD69 on CD8⁺ T cells (Fig. 4B).

To evaluate whether mice treated with HPN328 developed anti-tumor immunity, hCD3ε mice implanted with MC38-hDLL3 tumors were treated with 100, 200, or 1,000 µg/kg HPN328 to elicit an anti-tumor response (Fig. 4C and D), although study-to-study variability in the anti-tumor activity of HPN328 in the hCD3ε syngeneic model has been observed (Fig. 4A and D). On study day 30, mice with no residual disease (7 mice from 200 µg/kg and 10 mice from 1,000 µg/kg groups) were rechallenged by implantation of parental MC38 cells on the opposite flank. Nine age-matched naïve mice were also implanted with parental MC38 cells as a control. On study day 57, 27 days post rechallenge with parental MC38 tumors, mice previously treated and cured with 200 µg/kg HPN328 exhibited slower tumor regrowth than naïve animals (Fig. 4E). By contrast, significant anti-tumor activity upon rechallenge was observed in mice previously cured from the 1,000 µg/kg HPN328, suggesting the development of immune memory. The spleens from mice were collected on study day 60 and processed for memory T cells. A dose-responsive increase in the CD8⁺ memory T-cell population was observed in the 200 and 1,000 µg/kg HPN328-treated mice when compared with naïve control animals (Fig. 4F). HPN328-cured mice successfully rejected tumor rechallenge with hDLL3-negative parental MC38 tumor cells with concomitant increases in memory T-cell populations, demonstrating that HPN328 can promote immunological memory and long-term anti-tumor immunity.

The pharmacokinetics of HPN328 was evaluated in cynomolgus monkeys based on species cross-reactivity of HPN328 to cynomolgus DLL3, CD3, and albumin (see Supplementary Figs. S4 and S5), with functional cross-reactivity of HPN328 validated by the targeted lysis of cyDLL3-expressing HEK 293 cells (see Fig. 1D). A repeat dose of 1 or 10 mg/kg HPN328 was administered on study days 1, 8, 15, and 22, and a functional assay (recombinant CD3 and DLL3 proteins used for capture and detection, respectively) was used for detection of HPN328. Following the first dose of HPN328, concentrations of HPN328 declined biphasically (Fig. 5). C_max and AUC values for HPN328 increased dose proportionally from 1 to 10 mg/kg (Supplementary Table S8). Following four weekly i.v. bolus injections, HPN328 C_max increased dose proportionally, whereas AUC increased slightly more than dose proportionally with limited HPN328 accumulation at 1 mg/kg/dose and approximately 2.5-fold accumulation at 10 mg/kg/dose. The volume of distribution was independent of the dose. The mean half-life ranged from 78 to 187 hours (Supplementary Table S8), confirming half-life extension due to the presence of the anti–albumin-binding domain. Of the 10 animals dosed per group, ADAs to HPN328 were detected in 7 of 10 animals in the 1 mg/kg/dose group and in 3 of 10 animals in the 10 mg/kg/dose group. The presence of ADAs impacted HPN328 systemic exposure on day 22 for 3 ADA-positive animals at 1 mg/kg/dose and 1 ADA-positive animal at 10 mg/kg/dose. Although ADA was observed in both the 1 mg/kg and 10 mg/kg treatment groups, the mean systemic exposure (AUC_0–168hr) on day 22 was similar when all data from all animals were compared with the systemic exposure from ADA-negative animals. Immunogenicity of humanized antibodies is expected in non-human primates (NHPs) and is not predictive of ADA in humans (22). As for any therapeutic, the potential occurrence or impact of ADAs would be closely monitored in patients during clinical development.

Multiple doses of HPN328 were well tolerated in cynomolgus monkeys up to 10 mg/kg. Consistent with T-cell activation and CD3 engagement by HPN328, transient reductions in T-cell subsets were also observed in response to HPN328 treatment (Supplementary Fig. S17A and S17B). Compared with vehicle-treated animals, there were also transient changes in counts of total lymphocytes and neutrophils associated with the first dose of HPN328 to cynomolgus monkeys that returned to normal levels 24 to 72 hours post dose, indicating that HPN328 was well tolerated (Supplementary Fig. S18A and S18B).

The number of stimulated T cells and the potency of T-cell stimulation by T-cell engagers can induce cytokine release syndrome (CRS; ref. 23). Often the most severe cytokine release is experienced following the first dose and dampens following subsequent doses often referred to as the “first-dose effect” (24). On days 1, 8, and/or 22, both sexes at ≥1 mg/kg/dose had increases in IL-6, IL-10, IFNγ, and MCP-1 concentrations at 4 and/or 8 hours post dose, which typically, partially to fully resolved by 24 hours post dose. Increases in cytokines tended to be most pronounced on day 1, with a lessening of the post-dose responses on days 8 and 22. These changes were considered test article–related and indicative of a transient post-dose immune/inflammatory response (Fig. 6; Supplementary Fig. S19). The administration of subsequent doses of HPN328 (on study days 8, 15, 22) did not elevate cytokine levels, suggesting a “first-dose effect”. Together, these data suggest that HPN328 is well tolerated and, as expected, leads to transient T-cell activation in NHPs.

HPN328 is a novel TriTAC molecule being developed for the treatment of patients with advanced cancers that express DLL3. Here, we demonstrate that HPN328 is highly potent and specific in redirecting T cells to kill DLL3-expressing tumor cells in vitro and in multiple SCLC mouse tumor models. Unlike other TriTACs, which are configured in the anti-target:anti-albumin:anti-CD3 (T:A:C) format (18, 19), we found the activity of HPN328 was enhanced in the anti-CD3:anti-albumin:anti-DLL3 (C:A:T) configuration. This difference in configuration seems to be unique to this DLL3-targeting TriTAC and may possibly reflect how the anti-DLL3 domain in HPN328 interacts with DLL3. In vitro, HPN328 treatment at picomolar concentrations leads to T-cell activation, cytokine release, and T-cell killing of SCLC cells. In vivo, HPN328 recruited systemically administered T cells to the tumor site in mice, leading to potent anti-tumor activity. HPN328 administration to cynomolgus monkeys was well tolerated with a serum half-life of 78 to 187 hours. The safety and efficacy of HPN328 is currently being evaluated in SCLC patients and in patients with other DLL3-expressing tumors (NCT04471727).

The clinical success seen with blinatumomab in the treatment of acute lymphoblastic leukemia (ALL; ref. 25), TECVAYLI for the treatment of multiple myeloma (26), and tebentafusp for the treatment of melanoma (27) has led to further interest in the development of T-cell engagers to treat various solid tumor types. High differential expression of DLL3 on tumor versus normal tissue makes it an attractive target for T-cell engagers. Tumors from SCLC patients showed DLL3 positivity in 1,040 of 1,362 or 76.4% of tumor specimens (6, 7). In addition to SCLC, DLL3 expression was also reported in various neuroendocrine tumors, including glioma (8), gastrointestinal neuroendocrine malignancies (10), small cell bladder cancer (11), Merkel cell carcinoma (28), and neuroendocrine prostate cancer (9).

A previous drug targeting DLL3-expressing tumors was unsuccessful in clinical trials. This drug, an anti-DLL3 antibody drug conjugate (ADC) called rovalpituzumab tesirine (Rova T), initially showed early clinical activity, but eventually, clinical development was halted due to dose-limiting toxicities attributed to the payload of the ADC (12). The mechanism through which ADC technologies function, such as Rova-T, differ greatly from that of T-cell engagers. The relatively low abundance of DLL3 on tumors may not be sufficient to deliver the payload required for ADCs to be effective at tolerated doses. On the other hand, only a very small number of sites likely need to be occupied by a T-cell engager such as HPN328 to kill cancer cells. The EC₅₀ values for redirected lysis by HPN328 are a log lower than the K_D values measured for HPN328 binding to DLL3 and CD3ε, suggesting that only a very few TriTAC molecules are needed for cytolytic synapse formation. Cytolytic synapses have been described for an epithelial cell adhesion molecule/CD3 bispecific T-cell engager molecule (29), and these serve as sites for T-cell activation followed by perforin and granzyme release and ultimately target cell lysis through caspase 3- and 7-mediated apoptosis (30). The mechanism through which T-cell engagers target tumor cells suggests that they may be more effective in treating DLL3-expressing cancers.

In a multiple-dose Good Laboratory Practice (GLP) study on cynomolgus monkeys, HPN328 was well tolerated at 10 mg/kg. This dose leads to circulating HPN328 levels that are 1,000- to 10,000-fold higher than its EC₅₀ values determined by TDCC assays, suggesting a potentially large therapeutic window for HPN328. Both the 1 and 10 mg/kg doses led to transient reductions in circulating T cells consistent with the mechanism of action of TriTAC molecules.

Mitigating or managing the impact of cytokine release syndrome is an important aspect in the clinical management of T-cell engagers. In a repeat-dose study, cytokines in the serum of high dose–treated NHPs peaked between 6 and 8 hours and recovered to baseline values within 24 hours in most cases except for IL-6, which returned to baseline by day 8. The observation that cytokines did not spike again following the second, third, or fourth dose suggests that the transient increase in cytokines are related to the “first-dose effect” in which the most severe release of cytokines is observed after the first administration of the drug (24). Limiting the peak levels of proinflammatory cytokines by utilizing step-fractionated dosing in which a portion of the dose is given in a stepwise manner can be effective (23).

An additional challenge faced by T-cell engagers targeting solid tumors compared with liquid cancers is the relative lack of T cells within the tumor microenvironment. Here, by IHC, we show that HPN328 was able to recruit T cells to “cold” tumors that lead to an immunogenically active tumor microenvironment and anti-tumor activity. The ability of HPN328 to recruit T cells to the tumor may lead to wider immunological changes in the tumor microenvironment and be advantageous for patient outcomes.

The development of a mouse strain containing the epitope of hCD3ε on mouse T cells allows for investigation of TriTAC activity in an immunocompetent system. We demonstrate that T cells in this mouse model are functional and demonstrate cytotoxic anti-tumor activity in the presence of an anti-hEGFR TriTAC or HPN328. Rechallenge studies further demonstrate that HPN328 treatment has led to long-term anti-tumor immunity in the hCD3ε model. T-cell activation and cytokine release induced by TriTACs or other T-cell engagers have the potential to recruit and activate other immune cell types to the tumor microenvironment leading to epitope spreading and sustained anti-tumor immunity. Dao and colleagues (31) have provided evidence that T-cell engagers can induce epitope spreading in vitro. Epitope spread and this novel mechanism of anti-tumor activity induced by T-cell engagers can potentially contribute to efficacy seen in patients where intra- and inter-tumor heterogeneity may contribute to disease progression.

The stability and half-life of HPN328 in cynomolgus monkeys support a once weekly dosing regimen. The use of two sdAbs allow HPN328 to be highly stable while retaining its small size. The intrinsic stability of sdAbs compared with antibody fragments make them ideal for use in a T-cell engaging molecule (32, 33). For half-life extension, HPN328 relies on an sdAb that binds to HSA. In NHPs, the sdAb confers a prolonged half-life of 78 to 187 hours for HPN328 when compared with only 2 hours for the bispecific T-cell engager antibody blinatumomab, which requires continuous intravenous infusion (25, 34). The half-life of HPN328 observed in NHPs is consistent with other TriTAC molecules (18, 19) and supports a weekly or less frequent dosing schedule in patients.

The present study provides a clear rationale to assess the safety and tolerability of HPN328 in patients with advanced DLL3-expressing cancers. The purported mechanism of action, preclinical activity, and safety profile observed in cynomolgus monkeys following the administration of HPN328 all support the potential for a safe and effective therapy for patients whose tumors express DLL3. A phase 1/2 study of HPN328 as monotherapy to assess the safety, tolerability, and pharmacokinetics in patients with advanced cancers associated with DLL3 expression, including SCLC and other neuroendocrine malignancies is currently ongoing (NTC NCT04471727).

M.E. Molloy reports former employment at Harpoon Therapeutics, Inc. and owns shares of Harpoon Therapeutics. M. Barath reports employment at Harpoon Therapeutics, Inc. from December 2016 to November 2022 and, until recently, owning shares of Harpoon Therapeutics, Inc. M. Cremin reports employment at Harpoon Therapeutics, Inc. B.D. Lemon reports a patent for US 11,453,716 issued, a patent for US 11,623,958 issued, a patent for US 10,730,954 issued, a patent for US 10,844,134 issued, a patent for US 11,136,403 issued, a patent for WO-2021-163364-A1 pending, a patent for WO-2022-192225-A1 pending, and a patent for WO-2022-212732-A1 pending. L. Santiago reports employment at Harpoon Therapeutics, Inc. K.L. Strobel reports personal fees from Harpoon Therapeutics, Inc. and Engage Bio outside the submitted work. C.-L. Law reports personal fees from Harpoon Therapeutics, Inc. outside the submitted work and having a patent for combination therapy with immune cell–engaging proteins and immunomodulators pending. H. Wesche reports being an inventor of T cell Engager related patents and employment at Harpoon Therapeutics, Inc. R.J. Austin reports personal fees and other support from Harpoon Therapeutics, Inc., outside the submitted work and having a patent for US10815311B2 issued. No disclosures were reported by the other authors.

M.E. Molloy: Conceptualization, data curation, formal analysis, methodology, writing–original draft. W.H. Aaron: Conceptualization, data curation, formal analysis, supervision, investigation, methodology. M. Barath: Investigation. M.C. Bush: Investigation. E.C. Callihan: Investigation. K. Carlin: Investigation. M. Cremin: Investigation. T. Evans: Investigation. M.G. Guerrero: Investigation. G. Hemmati: Investigation. A.S. Hundal: Investigation. L. Lao: Investigation. P. Laurie: Investigation. B.D. Lemon: Conceptualization, formal analysis, supervision. S.J. Lin: Conceptualization, investigation, methodology. J. O’Rear: Investigation. P. Patnaik: Investigation. S. Sotelo Rocha: Formal analysis, supervision, investigation, methodology. L. Santiago: Formal analysis, writing–review and editing. K.L. Strobel: Formal analysis, investigation. L.B. Valenzuela: Investigation. C.-H. Wu: Investigation. S. Yu: Investigation. T.Z. Yu: Formal analysis, investigation. B.S. Anand: Writing–review and editing. C.-L. Law: Conceptualization, formal analysis, supervision, writing–review and editing. L.L. Sun: Conceptualization, formal analysis, writing–original draft. H. Wesche: Conceptualization, resources, supervision, methodology. R.J. Austin: Conceptualization, formal analysis, supervision, methodology, writing–review and editing.

All funding was provided by Harpoon Therapeutics, Inc.