TriTACs, a Novel Class of T-Cell–Engaging Protein Constructs Designed for the Treatment of Solid Tumors

T-cell–engaging antibodies are an emerging modality for treating malignancies (1–3). They are engineered to simultaneously bind to a surface protein on cancer cells and to an invariant subunit of the T-cell receptor (TCR) on T cells, resulting in cytolytic synapse formation and subsequent tumor cell lysis. The therapeutic potential of T-cell engagers in patients was first demonstrated by blinatumomab, an anti–CD19/CD3-bispecific antibody construct approved for the treatment of relapsed or refractory adult B-cell acute lymphoblastic leukemia (4–6). Although CD19-, CD20-, and BCMA-specific T-cell–engaging antibodies have shown promising clinical activity in patients with leukemia, lymphoma, and multiple myeloma, respectively (7–14), the activity of T-cell engagers in solid tumor indications appears more limited. For example, CEA- and EPCAM-specific T-cell engagers showed only modest clinical activity in patients with colorectal cancer (15–17). More promisingly, early-stage data for a PSMA/CD3-bispecific T-cell engager showed two long-term responses in patients with prostate cancer (18). Although most of these CEA-, EPCAM-, and PSMA-targeting molecules are small and globular, they are often not half-life extended and must be administered by repeated continuous i.v. infusion over several months. Continuous i.v. infusion imposes convenience issues and increases in port infections (18).

The activity of T-cell–engaging antibodies in solid tumor indications is limited by multiple factors. First, in contrast to hematologic malignancies, few surface lineage markers are known in solid tumors where ablation of healthy cells expressing these markers is tolerated. In solid tumors, tumor-selective targets are often intracellular, such as testis-specific antigens, or secreted, such as mammaglobin or PSA. Lacking tumor-selective targets, antigens are often chosen based on overexpression relative to normal tissues, which limits their therapeutic index because of normal tissue expression. Examples include HER-2, EGFR, or B7-H3. Second, cancer cells within solid tumors are typically embedded in a microenvironment that can reduce the activity of resident T cells or deter T-cell influx. Numerous immune-regulatory proteins hijacked by cancer cells contribute to this inhibitory milieu (19). Third, the microenvironment of solid tumors presents barriers limiting the ability of protein-based drugs to enter the tumor compartment, including extracellular matrix, stromal cells, and a high interstitial pressure. In conclusion, factors critical for developing T-cell engagers for solid tumor therapy are (i) selection of the best possible target antigens, (ii) the smallest possible size to facilitate diffusion into tumor tissue, and (iii) high drug stability to improve safety by avoiding aggregation and potential T-cell activation outside of the tumor compartment.

Here, we describe the development of “Trispecific T-cell Activating Constructs” (TriTACs) as a next generation of T-cell–engaging protein constructs. TriTACs have three domains and bind specifically to a tumor antigen, human serum albumin (HSA), and the invariant TCR subunit CD3ϵ. The domains binding tumor antigen and albumin typically are humanized single-domain antibodies (sdAbs, or VHH domains) derived from the variable regions of camelid heavy-chain–only antibodies (20). The CD3ϵ-binding domain is a humanized scFv antibody fragment. Transient binding to HSA provides TriTACs with extended serum half-life (21–24). TriTAC molecules have a molecular weight of approximately 53 kDa, making them smaller than most other half-life extended T-cell engager formats, which often have an IgG-based design and leverage the dimeric 50-kDa Fc gamma domain for half-life extension. We generated TriTACs targeting EGFR and prostate-specific membrane antigen (PSMA/FOLH1) in multiple domain arrangements and TriTACs targeting PSMA and mesothelin (MSLN) with varying affinities for CD3ϵ. These TriTACs were analyzed for their binding affinities and potency of redirected tumor cell lysis.

Here, we introduce and characterize HPN424, a clinical stage TriTAC targeting PSMA, a cell surface protein that is expressed on >90% of malignant lesions in metastatic castration-resistant prostate cancer (mCRPC). In normal tissues, PSMA expression is mostly restricted to the brain and to the prostate and salivary glands (25). With single-digit picomolar potency, HPN424 directed T cells to specifically kill PSMA-expressing prostate cancer cells in cocultures and led to tumor eradication in a mouse xenograft model. HPN424 had pharmacokinetic properties supporting a once-weekly dosing regimen in humans, making it an attractive option for treating patients with mCRPC.

Sequences of TriTACs, single-domain antibodies, or extracellular domains of target proteins fused to an Fc domain or 6x His tag were cloned into mammalian expression vector pcDNA 3.4 (Invitrogen) preceded by a leader sequence. Expi293F cells (Life Technologies) were maintained in suspension in Optimum Growth Flasks (Thomson) between 0.2 and 8 × 10⁶ cells/mL in Expi 293 media. Purified plasmid DNA was transfected into Expi293 cells in accordance with Expi293 Expression System Kit (Life Technologies) protocols. Cells were cultured for 4 to 6 days after transfection. Alternatively, sequences of some TriTAC molecules, including HPN424, were cloned into mammalian expression vector pDEF38 transfected into CHO-DG44 DHFR-deficient cells (26), stable pools generated, and cultured in production media for up to 12 days prior to purification. The amount of expressed proteins in conditioned media was quantitated using an Octet RED-96 instrument with Protein A tips (ForteBio/Pall) using appropriate purified control proteins for a standard curve. Conditioned media from either host cell were filtered and purified by protein A affinity and desalted or subjected to preparative size exclusion chromatography using an AKTA Pure chromatography system (GE Healthcare). Protein A–purified TriTAC proteins were further polished by ion exchange and formulated in a buffered solution containing excipients. Final purity was assessed by SDS-PAGE by resolving 2.5 μg/lane on TRIS-Glycine gels and visualized with Simply Blue stain (Life Technologies). Native purity was also assessed by analytical size exclusion chromatography (SEC) using an Acquity BEH SEC 200 1.7 μm 4.6 × 150 mm column (Waters Corporation) resolved in an aqueous/organic mobile phase buffered at neutral pH on a 1290 LC system and peaks integrated with OpenLab Chemstation software (Agilent).

Affinity of TriTAC analytes for albumin, CD3ϵ, and tumor target ligands was measured by biolayer interferometry using an Octet RED96 instrument with anti-human Fc or streptavidin tips (ForteBio/Pall). Experiments were performed at 27°C in PBS plus casein in the absence or presence of 15 mg/mL HSA as described in Results and Figure Legends. 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 DataAnalysis 9.0 software.

Purified HPN424 was aseptically transferred to Type I glass vials at a concentration of 1 mg/mL, sealed, and stressed by five freeze-thaws cycles from −80°C to room temperature with at least 1 hour at each temperature, by orbital shaking at 300 rpm at room temperature for 72 hours, or by incubation at 37°C for 2 weeks. Stressed samples were compared with the same analysis of control nonstressed samples. Each was evaluated for concentration and turbidity by UV spectrometry using UV transparent 96-well plates (Corning 3635) with a SpectraMax M2 and SoftMaxPro Software (Molecular Devices). Samples were further evaluated by SDS-PAGE and analytical SEC as described above. SDS-denaturing capillary electrophoresis of nonreduced or reduced samples was performed with a PA800 Plus instrument and integrated with 32 karat software (Sciex). Samples were subjected to a 0.5°C/min thermal ramp using an UNcle (Unchained Labs), and melting transitions were determined using the barycentric mean of full spectrum fluorescence and aggregation measured by static light scattering at 266 nm and 473 nm using UNcle Software 3.0

All cancer cell lines were obtained from the ATCC except for the OVCAR8 cell line, which was obtained from the NCI. Cell lines were passaged a maximum of 36 times after being received from the ATCC. T cells were purified from leukaphereses using EasySep Human T Cell Isolation Kits (STEMCELL Technologies). T-cell killing assays were performed as described previously (27). Tumor cell viability was measured using CellTiterGlo or by labeling cells with luciferase and measuring luciferase activity. T-cell activation assays were set up using the same conditions as the T-cell killing assays. Cytokines were measured using AlphaLISA kits (Perkin Elmer). CD69 and CD25 expression on T cells was measured by flow cytometry using anti-CD25 clone M-A251 and anti-CD69 clone FN50 antibodies (BD Biosciences).

The care and use of animals were conducted in accordance with the regulations of the Association for Assessment and Accreditation of Laboratory Animal Care. The mouse pharmacokinetic study was performed using NOD CRISPR Prkdc Il2r Gamma (NCG) mice (Charles River). Blood samples for pharmacokinetic studies were collected by orbital bleed under anesthesia or cardiac puncture under deep anesthesia with isoflurane. Details of the xenograft studies were performed as described in the Results or Figure Legends.

Cynomolgus monkey studies were performed at Charles River Laboratories (Reno). TriTACs were administered intravenously by single slow bolus. Serum samples were stored frozen at −80°C until serum TriTAC levels were measured using an electrochemiluminescent ELISA assay. Pharmacokinetic analyses were performed using Phoenix WinNonlin Version 7.0 software (Certara).

We designed TriTACs to be trispecific for a tumor surface protein, HSA, and the invariant CD3ϵ subunit of the TCR. A series of TriTAC molecules targeting PSMA or two different EGFR epitopes (28) were designed to explore the structure/activity relationship of the TriTAC format. These TriTAC molecules contained a humanized scFv specific for CD3ϵ and sdAbs specific for HSA and for EGFR or PSMA. The HSA- and CD3-binding elements, which constitute the core elements of TriTACs, were cross-reactive with respective cynomolgus orthologs. sdAbs were chosen for TriTAC engineering because of their high stability and to limit the overall molecular size to 53 kDa. The three binding domains are joined by GGGGSGGGS linkers to provide flexibility and minimize immunogenicity. The three TriTAC-binding domains can be arranged in six different configurations (Fig. 1A).

All six possible configurations of anti-EGFR and anti-PSMA TriTACs were produced by transiently transfected 293 cells and purified from conditioned media to >95% purity, as estimated by SDS-PAGE (Supplementary Fig. S1). As analyzed by size exclusion chromatography, all TriTACs were >98% monomeric (Supplementary Table S1). TriTACs were further tested for their ability to bind CD3, HSA, and EGFR or PSMA target antigens, using biolayer interferometry (Table 1). Regardless of configuration, all TriTAC variants bound to recombinant CD3ϵ with K_D values between 1.8 and 4.7 nmol/L, indicating that the CD3-binding affinity of the scFv was relatively insensitive to fusion of sdAbs to its N- or C-terminus. K_D values for binding to HSA were about 3- to 7-fold lower when the albumin-binding domain was placed at the N-terminus of TriTACs. Binding of the EGFR-targeting TriTACs to EGFR was more strongly influenced by configuration. The EGFR TriTACs were engineered with sdAbs recognizing two distinct epitopes (28). When these anti-EGFR sdAbs were placed on the N-terminus, the K_D values were between 56- and 136-fold lower compared with middle or C-terminal positioning. In contrast, binding to PSMA by the PSMA-targeting TriTACs was much less affected by positioning of the anti-PSMA sdAb and varied only 3-fold in K_D values between N-terminal, middle, or C-terminal positioning.

The T-cell–dependent cellular cytotoxicity (TDCC) of TriTACs was determined in a coculture assay (27). Primary, resting human T cells and target-expressing tumor cells were combined with a titration of TriTACs, and after 48 hours, viability of the target cells was determined. TDCC of TriTACs with the anti–EGFR-9G8 domain was determined with EGFR-expressing NCI-H1563 lung adenocarcinoma cells and CAPAN2 pancreatic adenocarcinoma cells. Dose-dependent killing of both target cell lines was observed with all six TriTAC configurations and revealed EC₅₀ values for target cell lysis ranging from 1 to 9 pmol/L (Fig. 1B and D; Table 2). Similar results were observed with EGFR-expressing 22Rv1 prostate adenocarcinoma cells where EC₅₀ values for redirected lysis ranged from 7 to 54 pmol/L (Table 2). With all three target cell lines, the most potent EGFR-9G8 TriTAC configurations were anti-EGFR:anti-CD3:anti-albumin (T:C:A) and anti-EGFR:anti-albumin:anti-CD3 (T:A:C; Fig. 1). These same two configurations bound EGFR with 56- to 136- fold higher affinity compared with the other four configurations (Fig. 1 and Table 1). TriTACs with the anti-EGFR-7D12 domain were also tested in TDCC assays with NCI-H1563 and CAPAN2 cells (Supplementary Fig. S2A and S2C; Table 2). With the EGFR-7D12 TriTACs, cell killing was noticeably more potent in the T:C:A and T:A:C configurations, with EC₅₀ values ranging from 1.5 to 2.1 pmol/L compared with the other four configurations where EC₅₀ values ranged from 314 to 2,363 pmol/L. Similar to EGFR-9G8 TriTACs, the EGFR-7D12 TriTACs in the T:C:A and T:A:C configurations bound with 72- to 134-fold higher affinity compared with the other four configurations. The six PSMA-specific TriTACs were tested in a TDCC assay using 22Rv1 target cells, which express PSMA in addition to EGFR. As seen with the EGFR TriTAC molecules, the T:A:C and T:C:A configurations exhibited the most potent TDCC activity (Fig. 1F and Table 1). Intriguingly, the configurations C:T:A and A:C:T lacked TDCC activity despite binding with similar affinity to recombinant PSMA protein (Table 1) and similar binding to PSMA-expressing cells (Supplementary Fig. S3A), suggesting that binding affinity was not the sole or dominant determinant for potent redirected target cell lysis by T cells and that temporospatial aspects of the binding events are also a determinant of T-cell engager activity.

TDCC assays were performed in the presence of HSA to understand the impact of HSA binding on the activity of TriTAC molecules in serum. To avoid technical issues with the viscosity of highly concentrated albumin solutions, HSA was added to assays at 15 mg/mL, a concentration below the 35–50 mg/mL present in human serum (29), but sufficient to saturate greater than 99.99% of the anti–albumin-binding domain. With TriTACs containing the anti–EGFR-9G8 domain, the addition of HSA reduced the potency of redirected lysis by 5- to 82-fold depending on configuration (Fig. 1C and E; Table 2). In the presence of HSA and testing all three target cell lines, TriTACs in the T:A:C configuration showed the most potent TDCC activity with least impact by HSA, followed by TriTACs in the T:C:A configuration (Table 2). In contrast, the cell killing activity of the EGFR-7D12 TriTACs was barely affected by the addition of HSA (Supplementary Fig. S2B and S2D), with TDCC EC₅₀ values only increasing by 1- to 4- fold in the presence of HSA, with the T:C:A and T:A:C configurations being most potent. Among the six PSMA-specific TriTACs, only the T:A:C configuration showed robust target cell killing in the presence of HSA (Fig. 1G). HSA reduced but did not eliminate the ability of the PSMA TriTACs to bind to PSMA-expressing cells (Supplementary Fig. S2A and S2B). In summary, the T:A:C configuration generally had the most potent TDCC activity in the absence and presence of HSA, with the T:C:A configuration having equivalent or second best activity.

As the only clinically approved T-cell–engaging molecule is in BiTE format (4–6), the TDCC activities of EGFR- and PSMA-specific TriTACs in the T:A:C configuration were compared with BiTE constructs, which consist of two tandemly arranged scFvs and have no half-life extending moiety (30). An EGFR-specific BiTE based on cetuximab (31) and a PSMA-specific BiTE based on pasotuxizumab (32, 33) were expressed in 293 cells and purified. The EGFR-targeting BiTE and TriTAC molecules had similar TDCC activity against NCI-H1563 cells, with the TriTAC having an EC₅₀ value of 1.1 pmol/L and the BiTE having and EC₅₀ value of 1.3 pmol/L (Fig. 1H). Similarly, the PSMA-specific TriTAC and PSMA BiTE were compared in a TDCC assay using LNCaP as target cells (Fig. 1I). The PSMA-specific TriTAC had an EC₅₀ value of 11 pmol/L, and the PSMA-specific BiTE had an EC₅₀ value of 20 pmol/L. Thus, TriTACs have redirected lysis activity equivalent to BiTEs. The activity of an EGFR TriTAC was also compared with a cetuximab-based EGFR-targeting T-cell–engaging molecule engineered in the 2:1 format (Supplementary Fig. S2E and S2F; ref. 34). The TriTAC mediated T-cell killing with 8- to 9-fold greater potency than the 2:1 formatted molecule.

The in vivo activity of an EGFR-specific TriTAC in the T:A:C configuration was tested in mouse tumor models. To guide dosing, an initial study to determine the half-life of the TriTAC in mice was performed. Serum albumin has a half-life of approximately 1 day in mice, as opposed to 5.5 days in primates and 18 to 21 days in humans (35–37). Mice were given single 50 or 500 μg/kg bolus infusions of the EGFR-7D12 TriTAC, and serum samples were collected at different time points up to 168 hours. TriTAC serum levels were determined using an electrochemiluminescence-based assay and plotted as shown in Fig. 2A. The terminal half-life was calculated to be between 15 and 18 hours in mice (Supplementary Table S2), supporting daily dosing of the TriTAC in subsequent efficacy studies. Irradiated NOD/SCID mice were subcutaneously implanted with HCT116 colon adenocarcinoma cells mixed with resting human peripheral blood mononuclear cells (PBMC). Mice received daily doses of 5, 50, or 500 μg/kg of EGFR-specific TriTAC, or vehicle, during the first 10 days after implantation. Tumor volumes were measured every 2 to 3 days starting at day 10 and ending at day 21. Partial tumor growth inhibition was observed with the 5 μg/kg dose, and no palpable tumors formed with daily 50 or 500 μg/kg doses (Fig. 2B). This in vivo activity was comparable with what has been reported for EGFR-specific BiTE constructs in the HCT116 model (31). To further test the in vivo activity of the EGFR TriTAC, immunocompromised mice with established LoVo tumors were i.p. injected with 20 million expanded human T cells per mouse. The next day, mice were dosed for 14 days with anti–EGFR-7D12 TriTAC or a negative control TriTAC targeting GFP. Significant tumor growth inhibition was observed 7 days after dosing started, and complete tumor regressions were observed by the end of the study (Fig. 2C).

Target-mediated drug disposition through binding to CD3 on T cells has been shown to affect serum exposure of T-cell engagers (38, 39). Although low CD3-binding affinities seem desirable to maximize drug availability, high CD3-binding affinities provide increased activity in vitro. Here, we assessed the impact of CD3-binding affinity on TriTAC activity and exposure. Using a single substitution library method (40), the affinity and stability of hundreds of candidate anti–CD3 scFv fragments were measured, and three representative molecules were elected to explore the impact of CD3-binding affinity on TriTAC in vitro activity and in vivo pharmacokinetics. Three PSMA-specific TriTACs in the preferred T:A:C configuration were engineered with anti–CD3 scFvs having K_D values for CD3ϵ of 4, 18, and 133 nmol/L. In a TDCC assay using LNCaP target cells, these TriTACs showed EC₅₀ values for target cell lysis of 10, 87, and 1,389 pmol/L, respectively (Fig. 3A and Table 3). We also generated a series of MSLN-targeting TriTACs with varying affinities for CD3ϵ. MSLN is also a well-established target for solid tumor therapy (41). MSLN-specific TriTACs in the T:A:C configuration with anti–CD3 scFvs having K_D values of 4, 14, and 40 nmol/L for CD3ϵ were tested in a TDCC assay with SKOV3 ovarian carcinoma cells and showed EC₅₀ values of 2, 7.1, and 20 pmol/L, respectively (Table 3). In conclusion, for both PSMA- and MSLN-specific TriTACs, an increased affinity for CD3ϵ translated into an increased TDCC activity.

To test the effect of CD3 affinity on pharmacokinetics, two MSLN-specific TriTACs with affinities for cynomolgus CD3ϵ of 4 and 43 nmol/L were administered to cynomolgus monkeys as a single 0.02 mg/kg dose. As expected, the molecule with the lower CD3 affinity showed a 1.4-fold higher exposure (AUC) than the molecule with the higher affinity for CD3 (Table 3 and Fig. 3B). However, the increase in exposure came at the expense of a 10-fold reduction in TDCC activity. Among the tested conditions, the TriTAC with a CD3 affinity of 4 nmol/L gave an optimal combination of in vitro activity and in vivo exposure. We therefore selected this anti–CD3 scFv for generating all subsequent TriTAC constructs.

In the same experiment, we determined the serum half-life of the MSLN-specific TriTACs. These TriTACs bind cynomolgus albumin with a K_D of 8 nmol/L. The serum half-lives ranged between 3.5 and 3.7 days (Fig. 3B and Table 3). Albumin has a serum half-life of 5.5 days in cynomolgus monkeys (37), and given the longer half-life of albumin in humans (42), we expect TriTAC serum half-lives in humans to be compatible with once-weekly dosing.

Based on the in-depth characterization of the TriTAC platform, HPN424 was developed as a PSMA-targeting TriTAC for treatment of mCRPC in the T:A:C configuration (see Fig. 1A; Supplementary Fig. S4E). HPN424 currently is in a clinical phase I dose-escalation study with patients with mCRPC. HPN424 contains optimized anti–PSMA, anti–HSA, and anti–CD3 binding domains compared with the tool PSMA TriTAC constructs used for the TriTAC platform optimization. Biolayer interferometry showed that HPN424 binds to human PSMA with a K_D of 0.55 nmol/L in the presence of HSA, whereas there is only weak, nonquantifiable binding to cynomolgus monkey PSMA (Supplementary Table S3). HPN424 binds in the presence of HSA to human and cynomolgus CD3ϵ with similar affinities of 12 and 10 nmol/L, respectively. Its affinities for human, cynomolgus monkey, and mouse albumin are 8.3, 7.7, and 140 nmol/L, respectively.

HPN424 was characterized using multiple analytical methods to demonstrate its purity and stability. Under denaturing conditions, capillary electrophoresis showed that 99.7% of HPN424 was present in the intact main peak and 0.3% was in the form of lower molecular weight species (Supplementary Fig. S4A; Supplementary Table S4). SDS-PAGE showed a prominent band of HPN424 at its predicted molecular weight of 52.5 kDa (Supplementary Fig. S4B). Analytical size exclusion chromatography confirmed that HPN424 had 98.1% monomer content and revealed 1.9% of a higher molecular weight species consistent with the size of a noncovalent dimer (Supplementary Fig. S4C; Supplementary Table S5). To test HPN424′s stability, the protein was subjected to different stress conditions, including multiple freeze/thaw cycles, shaking for 72 hours at room temperature, and storage at 37°C for 2 weeks. Under all conditions, HPN424 maintained greater than 97% monomer content with no detectable increase in higher molecular weight species as measured by denaturing capillary electrophoresis and native analytical size exclusion chromatography. Full spectrum fluorescence and static light scattering were used to measure the melting and aggregation temperatures of HPN424 before and after exposure to stress conditions (Supplementary Fig. S4D and S4F; Supplementary Table S6). For all samples tested, the primary melting transition was between 55.2°C and 55.5°C, and the exponential aggregation temperature was between 83.3°C and 84.3°C. In summary, these data suggest that HPN424 is a stable and stress-resistant monomer that is not prone to aggregation and nonspecific T-cell activation.

HPN424 was confirmed to target and kill PSMA-expressing cancer cells. Purified, resting human T cells from four different donors were cocultured with LNCaP prostate cancer cells. HPN424 directed highly efficient killing of the LNCaP cells by T cells from all four donors with EC₅₀ values of 0.22 to 1.5 pmol/L, whereas no lysis was observed with a TriTAC molecule targeting GFP (Fig. 4A). Potent directed lysis was also observed with these same donor T cells and three additional prostate cancer cell lines, VCaP, MDAPCa2b, and 22Rv1, with EC₅₀ values between 0.16 and 4.8 pmol/L (Supplementary Table S7). The only exception was T cells from Donor 24 with 22Rv1 target cells, which did not result in efficient cell killing under these assay conditions. HPN424-directed killing of LNCaP cells was also measured in the presence or absence of HSA, and the calculated EC₅₀ values were 0.7 pmol/L in absence of HSA versus 1 pmol/L in the presence of HSA (Fig. 4B).

To confirm that HPN424 activates T cells in the presence of target cells, T cells and conditioned media were collected from TDCC assays performed with LNCaP as target cells. Using flow cytometry, a HPN424 concentration–dependent increase in expression of the activation markers CD69 and CD25 was observed on T cells from four different donors (Fig. 4C; Supplementary Fig. S5). The EC₅₀ values for induction of CD69 and CD25 expression (Supplementary Tables S8 and S9) were similar to those observed for cell killing (Supplementary Table S7). Comparable results were observed with VCaP, MDAPCa2b, or 22Rv1 as target cells (Supplementary Tables S8 and S9), whereas a TriTAC molecule targeting GFP failed to activate CD69 or CD25 expression (Fig. 4C; Supplementary Fig. S5). A separate study was performed to confirm that HSA binding to HPN424 does not result in target-independent activation of T cells as assessed by measurement of CD69 and CD25 expression. Cocultures of T cells and PSMA-expressing LNCaP cells or PSMA-negative NCI-H1563 cells were incubated with 1 nmol/L of HPN424 or GFP-targeting TriTAC. In the presence of HSA, increased CD69 and CD25 expression was only observed with HPN424 and LNCaP cells but not with the NCI-H1563 cells or with the GFP-targeting TriTAC (Supplementary Fig. S5). Another hallmark of T-cell activation is cytokine production. TNFα and IFNγ levels were measured in conditioned medium collected from TDCC assays with PSMA-expressing LNCaP cells or PSMA-negative HCT116 cells using T cells from two different donors. With T cells from both donors, HPN424-dependent secretion of TNFα and IFNγ was observed with PSMA-expressing LNCaP cells but not with PSMA-negative HCT116 cells (Fig. 4D; Supplementary Fig. S7). EC₅₀ values for cytokine production were slightly higher than the EC₅₀ values for surface expression of CD69 and CD25 (compare Supplementary Tables S8–S11). HPN424 also induces IL2 secretion (Supplementary Fig. S8). Interestingly, although T cells from donor 24 did not efficiently kill 22Rv1 cells under these assay conditions, induction of TNFα and IFNγ production and induction of CD69 and CD25 expression was observed on par with T cells from the other donors and with other PSMA-expressing cell lines (Supplementary Tables S8–S11). These results indicate HPN424 activated T cells from donor 24 in the presence of PSMA-expression 22Rv1 cells, but this activation was insufficient to direct killing of 22Rv1 cells in a 48-hour assay. T-cell activation was dependent on both the PSMA-binding domain in HPN424 and the expression of PSMA on target cells. Even with HPN424 concentrations exceeding EC₅₀ values by 10,000-fold, no T-cell activation was seen with PSMA-negative HCT116 or NCI-H1563 cells (Supplementary Tables S7–S11).

The antitumor activity of HPN424 was assessed in a mouse xenograft model. A 1:1 mixture of PSMA-expressing 22Rv1 prostate cancer cells and human PBMCs was implanted into immunocompromised NCG mice. Five days after implantation and 2 days before palpable tumors were formed, mice were i.v. dosed once per day for 10 days with vehicle or with 2, 10, 50, or 250 μg/kg of HPN424. Significant tumor regression (P < 0.0001) was observed in all dose groups versus vehicle starting at 17 days after implantation (Fig. 5A).

To further characterize the in vivo properties of HPN424, the pharmacokinetics and in vivo stability of HPN424 were investigated in a nonhuman primate model. Cynomolgus monkeys were administered a single dose of 0.03 or 0.1 mg/kg of HPN424. The serum levels of HPN424, as measured using an electrochemiluminescent assay (Fig. 5B), indicated that the TriTAC had a half-life of approximately 3.4 days in the primates (Supplementary Table S12). To test the stability of HPN424 in cynomolgus monkeys, a 168-hour serum sample from the pharmacokinetic study was analyzed in a TDCC assay. The serum sample and a freshly thawed HPN424 sample were serially diluted in cynomolgus serum to control for the impact of serum on in vitro activity. Both samples redirected T cells to lyse LNCaP cells with EC₅₀ values of 2.6 and 4.8 pmol/L, respectively (Fig. 5C), indicating that HPN424 retained its biological activity after circulating in cynomolgus monkeys for 1 week.

Blinatumomab, a CD19/CD3-bispecific BiTE, is thus far the only T-cell–engaging biologic that has obtained global market approval (1, 2). BiTE molecules are composed of two fused scFvs, and with their small size of 50 kDa and flexible linkage between the two binding domains are favorably designed for transiently connecting cancer and T cells for formation of highly efficient cytolytic synapses. Blinatumomab has a very short serum half-life and therefore requires continuous i.v. infusion, which comes with risk of port infection and raises the question of patient compliance due to the low convenience. Next-generation BiTE antibodies were recently engineered with a C-terminal–silenced single-chain Fc gamma domain for half-life extension, giving them a size of approximately 100 kDa (43).

To successfully use T-cell engagers for treatment of solid tumors, we postulated that the therapeutic index, specifically the balance of tumor-specific activity versus off-target activity, must be increased. Although on-target/off-tumor activity can cause toxicities and reduce therapeutic index of a T-cell engager, it can be addressed by targeting tumor surface proteins with limited normal tissue expression. Off-target activity is harder to address and can lead to cytokine release syndrome and toxicity independently of any antitumor activity. Besides nonspecific binding to critical healthy tissues, off-target activity can be driven by target-independent activation of T cells via TCR cross-linking. T-cell engagers that dimerize or aggregate, or have Fc gamma domains with residual Fc receptor–binding activity, have the potential to activate T cells in the absence of target cells, especially if high drug levels are needed in the serum to drive drug penetration into the tumor compartment. This issue is exemplified by the clinical experiences with catumaxomab, a first-generation T-cell engager with intact Fc domain (44). Thus, to improve therapeutic index, T-cell engagers ideally should be physically stable (i.e., not prone to aggregation), have improved tissue penetration, and avoid any residual Fc gamma receptor binding.

The TriTAC platform was engineered to produce small monovalent half-life–extended T-cell–engaging protein constructs with high physical stability. By using single-domain antibody fragments for tumor targeting and transient binding to serum albumin for half-life extension, the overall size of the half-life–extended TriTAC format is similar to blinatumomab and is much smaller than T-cell engaging antibody constructs that use an Fc domain for half-life extension (45). In addition, TriTACs likely retain the globular shape of BiTE molecules as opposed to the nonglobular shape of antibodies. Their small size and globular shape are expected to improve tumor penetration (46), specifically in solid human tumors. Biomolecules are distributed in normal tissues by convection, a passive process largely independent of the shape and size of protein drugs. This process is impaired in solid tumors, due to the high interstitial pressure and the lack of a functional lymphatic system, thus the primary way biologics distribute inside tumor tissue is by diffusion, a process governed by a molecule's shape and size (46). Further promoting efficacy, TriTACs are among the most potent half-life–extended T-cell engagers with in vitro EC₅₀ values for redirected lysis in some cases below 1 pmol/L. We surmise that ultralow EC₅₀ values reflect formation of the most efficient cytolytic synapse with the fewest number of T-cell engager molecules. Lastly, TriTACs have favorable biophysical properties as they were engineered with sdAbs, known for high stability (47, 48), and contain a notably stable scFv fragment for CD3 binding. TriTACs could be produced with high monomer content in all six configurations independent of the target-specific sdAb used. As exemplified with HPN424, TriTACs have low aggregation propensity and high resilience to physical stress.

We consider the robust stability of TriTACs in combination with their high potency to be important for both antitumor activity and safety. T-cell engager formats that tend to aggregate can cause multivalent CD3 binding leading to TCR stimulation in the absence of target cells and, eventually, internalization of TCRs. Both are undesirable and may limit the therapeutic window and potency, respectively, of the T-cell–engaging molecule. Herein, we show that a TriTAC with specificity for green fluorescent protein or target-negative cell lines in coculture assays do not elicit any signs of T-cell activation or mediate directed T-cell–mediated killing. These assays are highly sensitive measures of TriTAC specificity and suggest that TriTACs will not have off-target toxicities even at high doses.

As determined with TriTACs targeting EGFR and PSMA made in six different configurations, the arrangement of the binding domains in the TriTAC format significantly affected the biological properties of TriTACs, including binding affinities and TDCC activity, particularly in the presence of albumin. Previous reports indicated that the anti-target and anti-CD3 domains needed to be proximal to achieve robust directed T-cell killing (49). Thus, we were surprised that the T:A:C configuration, where the anti-target and anti-CD3 domains were separated by the anti-HSA domain, was generally the most potent, especially in the presence of albumin. Based on our data, we speculate that with certain TriTAC configurations that albumin, a 65-kDa molecule, can interfere with the formation of a functional synapse between a T cell and a target cell. We believe that the magnitude of this interference is dependent on the three-dimensional presentation of the target epitope relative to the cell surface and the configuration of the binding domains within a TriTAC molecule. We speculate that with the T:A:C configuration, flexibility afforded by the linkers, enables the anti-CD3 and anti-target domains to position in a manner that generally most favors stable synapse formation by an albumin-bound TriTAC, although with certain epitopes, such as the epitope recognized by the EGFR-7D12 TriTACs, other configurations may be equally optimal. In a mouse xenograft tumor study, where HPN424 was bound to mouse albumin, HPN424 caused tumor regression with a dose as low as 2 μg/kg. These results suggest under in vivo conditions that albumin bound transiently to HPN424 also did not interfere with the formation of a cytolytic synapse between T cell and target cell.

We also explored and optimized the binding affinity of TriTACs for CD3ϵ on T cells. As expected, with higher affinity for CD3, more potent TDCC activity was observed. On the other hand, a concern with high-affinity CD3 binding is sequestration of TriTACs by the T-cell compartment with a negative impact on PK properties. In a cynomolgus monkey study with two MSLN-specific TriTACs with differing affinities for CD3ϵ, we observed the molecule with a reduced affinity for CD3ϵ had a higher AUC compared with the higher affinity molecule. The increased AUC, however, came at the expense of reduced TDCC activity. Empirically, we concluded the optimal balance between TDCC activity and AUC is conferred with a CD3-binding affinity of 5 to 10 nmol/L, which was then selected for generation of HPN424.

Transient binding to serum albumin has been shown to dramatically increase the serum half-life of fused proteins (21, 50). In incorporating an HSA-binding domain into TriTAC molecules, our goal was to achieve a half-life of greater than 3 days in cynomolgus monkeys, which was achieved with the MSLN- and PSMA-specific TriTACs. Half-life extension by albumin binding relies on FcRn and multiple other mechanisms (50). Albumin binding is unlikely to negatively affect the extravasation and tissue distribution of TriTACs because specific transporters allow albumin to leave the vasculature and reach the interstitial space of tumor tissue (51). In addition, because the TriTACs are not covalently bound to albumin, at any point in time, a small portion of TriTACs can diffuse freely. Finally, the long serum half-life of TriTACs and an adequate AUC will keep up a diffusion gradient between the capillary bed and the interstitial space between tumor cells for sufficient drug delivery. Consistent with this hypothesis, we also believe that TriTAC molecules will freely diffuse into the tumor to engage resident tumor lymphocytes rather than being carried into a tumor already bound to a T cell. Supporting this statement, we did observe free TriTAC molecules, not bound to T cells, in pharmacokinetic studies in cynomolgus monkeys (Figs. 3B and 5B).

HPN424 is a PSMA-targeting TriTAC for treatment of mCRPC. Although there are no approved commercial therapeutics targeting PSMA, there is target validation from radioimmunoconjugates (52, 53), antibody–drug conjugates (54), and pasotuxizumab, a PSMA-targeting BiTE that resulted in two long-term clinical responses in a phase I clinical trial (18). HPN424 had a half-life of 3.4 days in a cynomolgus monkey pharmacokinetic study, which may enable a once-weekly dosing regime in humans. HPN424 is currently being tested for safety and therapeutic activity in a phase I clinical trial in mCRPC (ClinicalTrials.gov Identifier: NCT03577028).

R.J. Austin reports a patent for US9708412B2 issued, a patent for US10066016B2 issued, a patent for US009920115B2 issued, a patent for US20180161428A1 pending, a patent for US 20180161428A1 pending, a patent for US20180162949A1 pending, a patent for US20200231672A1 pending, a patent for US20190112381A1 pending, a patent for US10543271B2 issued, a patent for US20180327508A1 pending, and a patent for US20200270362A1 pending. B.D. Lemon reports a patent for US9920115 issued, a patent for US10066016 issued, a patent for US10100106 issued, a patent for US10543271 issued, a patent for US10544221 issued, a patent for US10730954 issued, a patent for US20180161428A1 pending, a patent for US20180162949A1 pending, a patent for US20190031749A1 pending, a patent for US20190112381A1 pending, a patent for US20200270362A1 pending, and a patent for US20200289646A1 pending. P.A. Culp reports being an employee at Harpoon at the time the work was performed and owns Harpoon stock. L.B. Evnin reports personal fees from Harpoon Therapeutics outside the submitted work; in addition, L.B. Evnin has a patent for US20200231672A1 pending, a patent for US009920115B2 issued, a patent for US10066016B2 issued, and a patent for US9708412B2 issued; and he also served as a Board member on, and has a derivative financial interest in, two other companies working on T cell-engaging biologics: Amphivena Therapeutics and Maverick Therapeutics. K.L. Strobel reports employment at Harpoon Therapeutics during the conduct of the study. C.L. Law reports full time employee of Harpoon Therapeutics and own equity in Harpoon Therapeutics. P.A. Baeuerle reports personal fees from MPM Capital during the conduct of the study; in addition, P.A. Baeuerle has a patent for Harpoon issued. H. Wesche reports other from Harpoon Therapeutics Inc. outside the submitted work. No disclosures were reported by the other authors.

R.J. Austin: Conceptualization, resources, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. B.D. Lemon: Conceptualization, resources, data curation, formal analysis, supervision, validation, investigation, visualization, methodology, project administration, writing-review and editing. W.H. Aaron: Resources, data curation, supervision, investigation, methodology. M. Barath: Resources, data curation, investigation, methodology. P.A. Culp: Resources, data curation, supervision, investigation, methodology, project administration, writing-review and editing. R.B. DuBridge: Conceptualization, resources, supervision, investigation, methodology, project administration. L.B. Evnin: Conceptualization. A. Jones: Data curation, investigation, methodology. A. Panchal: Resources, data curation, investigation, methodology. P. Patnaik: Data curation, formal analysis, investigation, methodology. V. Ramakrishnan: Conceptualization, supervision. S.S. Rocha: Data curation, investigation, methodology. P. Seto: Data curation, investigation, methodology. K. Sexton: Resources, investigation. K.L. Strobel: Data curation, investigation, methodology. R. Wall: Resources, data curation, investigation, methodology. S. Yu: Resources, data curation, investigation. T.Z. Yu: Resources, data curation, investigation, methodology. C.-L. Law: Supervision, project administration. P.A. Baeuerle: Conceptualization, writing-original draft, writing-review and editing. H. Wesche: Conceptualization, formal analysis, supervision, project administration, writing-review and editing.

The authors thank Evan Callihan for measuring the IL2 data shown in Supplementary Fig. S8.

All funding was provided by Harpoon Therapeutics, 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.