Serially expanded T cells from a single donor were used for each individual experiment. Peripheral blood mononuclear cells (PBMCs) were separated from buffy coats (New York Blood Center) by Ficoll. Naïve T cells were purified using Pan T Cell Isolation Kit (Miltenyi Biotec, Cat#130096535) and expanded by CD3/CD28 Dynabeads (Gibco, Cat#11 132D) for 7–14 days in the presence of 30 IU/mL of interleukin-2 (IL-2). Expanded T cells were analyzed for their proportion of CD3(+), CD4(+), and CD8(+) T cells, and the fraction of CD4 or CD8 T cells was allowed between 40% and 60% to maintain consistency. Unless stated otherwise, these activated T cells were used for all T cell experiments.

Naïve T cells were separated from cryopreserved peripheral blood stem cell collections with institutional review board (IRB) approval. These cells were purified using Dynabeads untouched human T cell kit (Invitrogen, Cat# 11 344D) and expanded with CD3/CD28 Dynabeads (Gibco, Cat#11 132D) and 30 IU/mL of IL-2 for 10–14 days.

γδ T cells were expanded in one of two ways: (1) fresh PBMCs separated from buffy coats were cultured with 2 µM of zoledronic acid (Sigma, #Cat 165800066) and 800 IU/mL of IL-2 for 12–14 days according to protocols; and (2) fresh PBMCs were cultured with 2 µM of zoledronic acid and 30 ng/mL of IL15Rα-IL15 for 12–14 days. Cultured PBMCs were tested for their antigen expression using antibodies against human CD3 (BioLegend, Cat# 300308, RRID:AB_314044), CD4 (BioLegend, Cat# 357410, RRID:AB_2565662), CD8 (BioLegend, Cat# 300912, RRID:AB_314116), γδ T cell receptor (TCR) (BioLegend, Cat# 331207, RRID:AB_1575111), and αβ TCR (BioLegend, Cat# 306723, RRID:AB_2563001). The IL15Rα-IL15 complex was prepared as described.17

Representative neuroblastoma cell line IMR-32 (ATCC-CCL-127), osteosarcoma cell line 143B (ATCC-CRL-8303) and U-2 OS (ATCC-HTB-96), primitive neuroectodermal tumor cell line TC-71 (ATCC-CRL-1598), prostate cancer cell line LNCaP-AR (ATCC-CRL-1740), and melanoma cell line M14 (UCLA-SO-M14) were used. All cancer cells were authenticated by short tandem repeats profiling using PowerPlex V.2.1 System (Promega, Cat# DC8942) and periodically tested for mycoplasma infection using a commercial kit (Lonza, Cat# LT07-318). The luciferase-labeled melanoma cell line M14Luc, osteosarcoma cell line 143BLuc, and neuroblastoma cell line IMR32Luc were generated by retroviral infection with an SFG-GF Luc vector.

GD2-BsAb or HER2-BsAb was mainly used for arming T cells. Hu3F8-BsAb specific for GD2 was built on the IgG-[L]-scFv format, in which the anti-CD3 huOKT3 single-chain variable fragment (ScFv) was linked to the carboxyl end of the anti-GD2 hu3F8 IgG1 light chain, where the N297A mutation was introduced to remove glycosylation and the K322A to remove complement activation—a combination to reduce spontaneous cytokine release.18 HER2-BsAb built on the IgG-[L]-scFv format carried a heavy chain variable (VH) identical to that of trastuzumab IgG1, again with both N297A and K322A mutations to silence fragment crystallisable (Fc) functions.19 Hu3F8 × mOKT3 and Herceptin × mOKT3 chemical conjugates were made as previously described by Sen et al.10 20 The other BsAbs were synthesized as previously described (US Patent #62/896415).18 21–23 Anti-CD33/anti-CD3 BsAb or anti-GPA33/anti-CD3 BsAb was used as a control BsAb.22 Biochemistry data of these BsAbs are summarized in online supplemental table 1.

T cell-mediated cytotoxicity was performed using ⁵¹Cr release,18 and EC₅₀ was calculated using SigmaPlot software. Target cells were labeled with sodium ⁵¹Cr chromate (51CrNa2CrO4; Amersham, Arlington Heights, Illinois) at 100 µCi/10⁶ cells at 37°C for 1 hour. After washing twice, these radiolabeled target cells were plated in 96-well plates. BsAb armed T cells were added to target cells at decreasing effector to target cell (E:T) ratios, at twofold dilutions from 50:1. After incubation at 37°C for 4 hours, the released ⁵¹Cr was measured by a gamma counter (Packed Instrument, Downers Grove, Illinois). The percentage of specific lysis was calculated using the formula where cpm represented counts per minute of ⁵¹Cr released.

100%×(experimentalcpm−backgroundcpm)(totalcpm−backgroundcpm)

Total release of ⁵¹Cr was assessed by lysis with 10% sodium dodecyl sulfate (SDS; Sigma, St Louis, Missouri, Cat# 71736) and background release was measured in the absence of effector cells and antibodies.

Human T helper 1 (Th1) cell-released cytokines including IL-2, IL-6, IL-10, interferon (IFN)-γ and tumor necrosis factor (TNF)-α were measured using LEGENDplex Human Th1 Panel (BioLegend, Cat# 741035) in vitro and in vivo and analyzed by flow cytometry.

Activated T cells were armed with each BsAb for 20 min at room temperature and washed twice with phosphate-buffered saline (PBS). T cell surface BsAb density was measured by mean fluorescence intensity (MFI) using anti-idiotype antibody or anti-human IgG Fc antibody (BioLegend, Cat# 409303, RRID:AB_10900424). MFIs were referenced to antibody binding capacity using Quantum Simply Cellular microspheres (Bio-Rad, Cat# FCSC815A, RRID:AB_10061915).

After arming, EATs were centrifuged at 500 g for 5 min at 4°C, and the cell pellet was suspended in T cell freezing medium (90% of fetal bovine serum (FBS) and 10% dimethyl sulfoxide (DMSO)) to achieve a cell concentration of 5×10⁷ cells/1 mL, chilled to 4°C and aliquoted into 2 mL cryovials. Vials were immediately transferred to freeze at −80°C for 24 hours before transferring to liquid nitrogen. After storage cryovials were thawed in a 37°C water bath with gentle swirling for 1 min. The thawed cells were transferred to F10 media and centrifuged at 500 g for 5 min. They were analyzed for viability, phenotype, antibody binding, and cytotoxicity.

T cells isolated from PBMCs were stimulated with CD3/CD28 Dynabeads (Gibco, Cat#11 132D) for 24 hours. T cells were transduced with retroviral constructs containing tdTomato and click beetle red luciferase in RetroNectin (Takara Bio, #Cat T100A/B)-coated six-well plates in the presence of IL-2 (100 IU/mL) and protamine sulfate (4 µg/mL). Transduced T cells were cultured for 8 days before use in animal experiments.

Tumor cells suspended in Matrigel (Corning, Tewksbury, Massachusetts) were implanted in the flank of male BALB-Rag2^−/−IL-2R-γc-KO (BRG) mice aged 6–10 weeks old (Taconic Biosciences).24 The following tumor cell lines were used: 1×10⁶ of 143BLuc, 5×10⁶ of IMR32Luc, 5×10⁶ of M14Luc, 5×10⁶ of HCC1954, 5×10⁶ of LnCaP-AR, and 5×10⁶ TC-71. Three osteosarcoma, two neuroblastoma, one Ewing sarcoma, and one breast cancer patient-derived tumor xenografts (PDXs) established from fresh surgical specimens with Memorial Sloan Kettering Cancer Center (MSKCC) IRB approval were used for in vivo experiments (online supplemental table 2). Treatment was initiated after tumors were established, with an average tumor volume of 100 mm³ when measured using TM900 scanner (Piera, Brussels, Belgium). Before treatment, mice were randomly assigned to each group. The T cell number administered per dose was 2×10⁷ cells based on previous reports.25 When tumor growth reached 2000mm³or greater, mice were euthanized. CBC analyses, body weight, general activity, physical appearance and graft-versus-host-disease (GVHD) scoring were monitored. All animal experiments were repeated twice more with different donor’s T cells to ensure that our results were reliable.

Luc(+) T cell engraftment and trafficking were quantified after intravenous injection of 3 mg D-luciferin (Gold Biotechnology, Cat# LUCK-100) on different days following T cell injection. Bioluminescence images were acquired using IVIS Spectrum CT In Vivo Imaging System (Caliper Life Sciences) and overlaid onto visible light images, to allow Living Image V.2.60 (Xenogen) to quantify bioluminescence in the tumor regions of interest (ROI). The total flux (photon/s) in the ROI was monitored.

Peripheral blood and tumors were collected and analyzed by flow cytometry. Tumor tissue was dissociated into single cell suspensions as previously described.26 Antibodies against human CD3 (BioLegend, Cat# 300308, RRID:AB_314044), CD4 (BioLegend, Cat# 357410, RRID:AB_2565662), CD8 (BioLegend, Cat# 300912, RRID:AB_314116), and CD45 (BioLegend, Cat# 304012, RRID:AB_314400) were used to quantify T cell engraftment and subpopulations. Fluorescence of stained cells was acquired using either a BD FACSCalibur or a BD LSRFortessa (BD Biosciences, Heidelberg, Germany) and analyzed using FlowJo V.10.6.0 software (FlowJo, Ashland, Oregon).

Fresh tumors were embedded in Tissue-Tek OCT (Miles Laboratories, Elkhart, Indiana) and snap-frozen in liquid nitrogen for storage at −80°C. The tumor sections were stained with mouse IgG3 mAb 3F8, as previously described.27 Paraffin-embedded tumor sections were tested for T cell infiltration using immunohistochemical (IHC) staining of human CD3, CD4, and CD8 T cells performed by the Molecular Cytology Core Facility of MSKCC using Discovery XT processor (Ventana Medical Systems), as previously described.26 Histological images were acquired with NIS-Elements V.4.0 imaging software and Nikon ECLIPSE Ni-U microscope.

Statistical analysis of tumor growth or cytokine release was conducted using area under the curve (AUC). Two-tailed Student’s t-test was used to determine statistical difference between two sets of data, while one-way analysis of variance with Tukey’s post-hoc test was used to determine statistical differences among three or more sets of data. All statistical analyses were performed using GraphPad Prism V.8.0 for Windows (GraphPad Software, La Jolla, California; www.graphpad.com). P<0.05 was considered statistically significant. Asterisks indicate that the experimental p value is statistically different from the associated controls at *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001.

We armed T cells with the major formats of GD2-BsAb (figure 1A) and compared the efficacy. The biochemical properties of GD2-BsAbs are summarized in online supplemental table 3. Surface BsAb densities (measured by MFIs) on T cells were analyzed in figure 1B. While chemical conjugate (IgG conjugate) armed T cells had lower MFIs, other GD2-BsAb armed T cells showed comparable MFIs. In vitro cytotoxicity assay showed that IgG-[L]-scFv or monomeric bi-specific T-cell engagers (BiTE) armed T cells had the strongest in vitro potency, while dimeric BiTE or IgG conjugate armed T cells had comparable cytotoxicity (figure 1C). To compare in vivo antitumor potency, a fixed dose of T cells armed with different GD2-BsAb formats (2 µg of each BsAb/2×10⁷ T cells) was intravenously administered into PDX-bearing mice (figure 1D and online supplemental figure 1). All EATs were well tolerated irrespective of BsAb formats, but only IgG-[L]-scFv GD2-BsAb armed T cells (GD2-EATs) exerted significant and durable antitumor responses prolonging survival. The in vivo efficacy of GD2-BsAb armed T cells strongly correlated with the density of tumor infiltrating CD3(+) T cells (TILs) by IHC staining of PDXs (figure 1E); GD2-EATs (T cells armed with GD2-IgG-[L]-scFv) showed substantially more abundant TILs compared with those armed with other BsAb formats.

Alongside GD2-BsAb armed T cells, we also compared the efficacy between HER2-EATs (armed with HER2-IgG-[L]-scFv) and HER2-IgG conjugate (Herceptin × OKT3) armed T cells proven safe in clinical trials.9 28 HER2-EATs showed higher BsAb densities (MFIs) (figure 1F) and more potent tumor cell killing than HER2-IgG conjugate armed T cells (figure 1G). HER2-EATs were much more effective in vivo for tumor response (p<0.01) and for survival (p=0.0020) (figure 1H).

With this IgG-[L]-scFv BsAb platform we conducted the rest of our experiments. Given the finite CD3 TCR density on human T cell,29 we set out to identify the optimal BsAb density on T cell. The in vitro cytotoxicity of EAT over a range of E:T ratios and BsAb arming dosages was studied (figure 2A). GD2-EATs and HER2-EATs both showed maximal cytotoxicity between 0.05 µg and 5 µg of BsAb/1×10⁶ T cells where quantitated between 500 and 20,000 BsAb molecules per T cell (figure 2B). When we compared the in vitro cytotoxicity of EATs with soluble BsAb under identical E:T ratios (figure 2C), the potencies of GD2-EATs and HER2-EATs were ~10-fold lower (EC₅₀) than soluble BsAbs; however, the maximal killing efficacy was comparable.

Next, we compared cytokine release between EATs and T cells in the continuous presence of BsAb (BsAb plus T cells). Cytokine release was evaluated throughout each step of T cell arming (figure 3A). First, Th1 cell cytokines (IL-2, IL-6, IL-10, IFN-γ, and TNF-α) were measured in the supernatants after 20 min incubation (prewash) and then after the washing step (postwash) (figure 3B). Although the released cytokine levels were generally low, IFN-γ and TNF-α did increase with the increment of GD2-BsAb (online supplemental figure 2A), which were removed after the washing steps (online supplemental figure 2B). Next, after co-culture with target cells, T cell cytokine levels were measured again and compared among groups. The cytokines surged after exposure to target cells. Both GD2-EATs and T cells coincubated with GD2-BsAb (GD2-BsAb plus T cells) released more cytokines in proportion to BsAb dose (online supplemental figure 2C,D). However, GD2-EATs produced significantly less amount of IL-2 (p=0.0071), IL-10 (p=0.0203), IFN-γ (p=0.0043), and TNF-α (p=0.0162) compared with GD2-BsAb plus T cells. Finally, we analyzed in vivo cytokine levels at different time points after GD2-EAT or GD2-BsAb plus T cell injection (figure 3D and online supplemental figure 3). GD2-EATs showed an early IL-2 and IFN-γ release, while GD2-BsAb plus T cells showed a surge of TNF-α at 4 hours and IL-6 at 48 hours. The AUC for TNF-α was significantly higher in GD2-BsAb plus T cells compared with GD2-EATs (p=0.0033).

T cell trafficking and persistence in tumors play an essential role in tumor immunity. To quantitate how efficiently EATs traffic into tumors and persist in vivo, we generated luciferase transduced T cells and armed with GD2-BsAb. Luc(+) GD2-EATs or Luc(+) unarmed T cells were intravenously administered into neuroblastoma PDX-bearing mice. Without GD2-BsAb arming, Luc(+) T cells did not localize to tumors and dissipated. In contrast, Luc(+) GD2-EATs rapidly trafficked into GD2(+) tumors following a transient sequestration in the lungs on day 1. The luciferase signals (total flux) of GD2-EATs in tumors increased over time to peak on day 4 (figure 4A,C). On the other hand, Luc (+) unarmed T cells with intravenous GD2-BsAb showed delayed egress from the lungs and the liver, and the bioluminescence signals in tumors peaked around day 7 (figure 4B,C). As tumors regressed, the total bioluminescence of Luc(+) GD2-EATs also diminished (figure 4C,D). In a second set of EAT trafficking study using osteosarcoma PDX model, EATs persisted longer with residual tumors (figure 4E,F). The bioluminescence of Luc(+) GD2-EATs or Luc(+) HER2-EATs in tumors peaked around day 5 and diminished slowly with still detectable signals over 1 month (figure 4G).

Since CD3 is present on diverse subpopulations of T cells, all kinds of effector T cells should be able to arm T-BsAb. We tested the potential applicability of EATs using γδ T cells known to have a reduced alloreactivity with potential as an allogeneic ‘off-the-shelf’ T cell source.32 After expansion from fresh PBMCs with IL-2 and zoledronate for 12 days, most of CD3(+) T cells were γδ TCR(+) (online supplemental figure 14A). Majority of the γδ T cells were CD3 positive and CD4 and CD8 double negative, contrasting with T cells expanded using CD3/CD28 beads, where majority were αβ TCR positive and CD4 or CD8 positive (online supplemental figure 14B). After arming γδ T cells with GD2-BsAb (γδ-GD2-EATs) or HER2-BsAb (γδ-HER2-EATs), surface BsAb density was measured by flow cytometry, and their MFIs were comparable with those of αβ-EATs (figure 7A). In the presence of GD2-BsAb or HER2-BsAb, γδ T cells mediated potent tumoricidal activity in vitro, exceeding the efficacy of αβ-EATs (figure 7B). The maximal cytotoxicity of γδ-GD2-EATs and γδ-HER2-EATs was achieved at arming doses between 0.05 µg and 5 µg of BsAb/1×10⁶ T cells (figure 7C), where antigen-specific cytotoxicity was maintained after freeze and thaw (figure 7D).

We next tested the in vivo efficacy of γδ-EATs and compared with corresponding αβ-EATs (online supplemental figure 15A). With supplementary IL-2, γδ-GD2-EATs and γδ-HER2-EATs did not produce significant antitumor response irrespective of additional zoledronate. Flow cytometry analyses of T cells in the blood and tumors after γδ-EAT treatment presented substantially fewer human CD45(+) T cells in the circulation and in the tumors compared with αβ-EATs (p=0.0006 and p=0.0007, respectively), suggesting poor in vivo survival of γδ-EATs (online supplemental figure 15B).

However, when exogenous IL-15 (as IL15Rα-IL15 complex) instead of IL-2 was used as T cell survival cytokine, the in vivo frequencies of γδ-EATs significantly increased (figure 7E). γδ T cells expanded from fresh PBMCs using 2 µM of zoledronate plus 30 ng/mL of IL-15 for 12–14 days were armed with GD2-BsAb or HER2-BsAb and administered intravenously into osteosarcoma PDX-bearing mice, with 5 µg of subcutaneous IL-15 or 1000 IU of IL-2 (figure 7F and online supplemental figure 16A). While γδ-GD2-EATs and γδ-HER2-EATs sustained with IL-2 failed to suppress tumor, same γδ-EATs sustained with IL-15 exerted significant antitumor effects without toxicities or weight loss during follow-up period. The PDXs treated with γδ-GD2-EATs plus IL-15 showed a diffuse T cell infiltration into tumors, in contrast to tumors treated by γδ-GD2-EATs plus IL-2 or unarmed γδ-EATs plus IL-15 (online supplemental figure 16B). Furthermore, cryopreserved γδ-EATs showed strong in vivo antitumor effect when supplemented with IL-15 and zoledronate (online supplemental figure 17A). Thawed γδ-STEAP1-EATs exerted a significant antitumor response against STEAP1(+) Ewing sarcoma PDXs in vivo. These findings implicate the potential applicability of IL-15 and allogeneic γδ T cells for IgG-[L]-scFv-platformed BsAb arming for the treatment of human cancers.