Preclinical Evaluation of B7-H3–specific Chimeric Antigen Receptor T Cells for the Treatment of Acute Myeloid Leukemia

Acute myeloid leukemia (AML) accounts for approximately 80% of acute leukemias in adults and has an incidence of at least 20,000 cases per year in the United States (1). Despite advances in risk stratification, optimization of chemotherapy regimens, improvement in stem cell transplantation techniques, and the advent of multiple targeted therapies, prognosis remains poor. These clinical findings emphasize the need for novel treatment strategies (1). Adoptive cellular therapy with chimeric antigen receptor (CAR) T cells (CAR-T) has emerged as a highly effective form of immunotherapy for B-cell acute lymphoblastic leukemia (ALL) and certain other B-cell malignancies (2); however, safe and effective CAR-T therapy for AML remains elusive. The success of CAR-T therapy in B-cell malignancies is predicated on the tolerability of the protracted B-cell aplasia that often accompanies CD19-directed CAR-T therapy. In contrast, prolonged disruption of normal myelopoiesis (especially neutropenia) would not be tolerable, and the development of CAR-T therapy for AML has been largely limited by the fact that most AML-associated surface antigens are also expressed on normal myeloid progenitor cells (3).

While it is difficult to envision the identification of cell surface antigens that are absent on all normal myeloid progenitors and yet expressed on all subtypes of AML, it seems plausible that some antigens absent on early myeloid lineage cells may be preferentially overexpressed in certain AML subtypes. We have identified B7-homolog 3 (B7-H3) as one such candidate. B7-H3 (or CD276) is a coreceptor belonging to the B7 family of immune checkpoint molecules. Its structure is that of a 110-kDa type I transmembrane protein with four Ig-like domains, a transmembrane domain, and a short cytoplasmic tail (4). The receptor for B7-H3 has yet to be definitively identified (5, 6), and the nature of its signaling also remains controversial (7). Although B7-H3 mRNA is expressed in many normal tissues, corresponding protein expression is limited (7). In multiple human cancers, however, B7-H3 protein is overexpressed. Examples include various solid tumors (glioblastoma, lung, breast, colon, pancreatic, renal, ovarian, prostate, and melanoma), as well as a subset of patients with AML (7, 8). B7-H3 expression in AML appears to be higher in AML with a monocytic immunophenotype (8), which is often associated with aggressive clinical features (9, 10). To evaluate the potential safety of B7-H3 as a target for CAR-T therapy, we have previously performed a comprehensive evaluation of B7-H3 expression in normal human tissues and demonstrated a lack of major organ toxicity in a syngeneic tumor model (11). We describe here the preclinical assessment of B7-H3–specific CAR T cells (B7-H3.CAR-T) as a possible therapy for a subset of patients with AML and our evaluation of the potential hematopoietic toxicity of this approach.

Deidentified cryopreserved samples of peripheral blood and bone marrow aspirate from patients with newly diagnosed AML were obtained through the University of North Carolina (UNC) Tissue Procurement Facility (UNC Hospital, Chapel Hill, NC). Primary AML blasts for use in patient-derived xenograft (PDX) models were obtained from the Public Repository of Xenografts (PRoXe, Dana-Farber Cancer Institute, Boston, MA). See Supplementary Materials and Methods for additional details, descriptions of other cell lines used, and tissue culture conditions.

T cells were labeled with 1.5 μmol/L carboxyfluorescein diacetate succinimidyl ester (CFSE; Invitrogen) as described previously (12), followed by coculture with irradiated B7-H3⁺ or B7-H3⁻ cell lines. After 5 days, T-cell proliferation was assessed via flow cytometry analysis.

The B7-H3–specific single-chain variable fragment (scFv) from the murine mAb, 376.96 (13), was cloned from the 376.96 mouse hybridoma as described previously (11). Generation of retroviral vectors and activated B7-H3.CAR-Ts has been described previously (11, 14, 15) and is summarized further in the Supplementary Materials and Methods.

Flow cytometry data acquisition was performed using a BD FACSCanto II or BD FACS Fortessa Flow Cytometer and BD Diva Software (BD Biosciences). Data analysis was performed using FlowJo Software versions 9.3 and 10.5 (Tree Star). See Supplementary Materials and Methods for a list of mAbs used.

The activity of B7-H3.CAR-Ts was evaluated in coculture models with B7-H3⁺ AML cell lines or primary patient-derived monocytic AML blasts. For cell lines, 0.5 × 10⁶ tumor cells were plated in a 24-well plate in 2 mL of media (without any exogenous cytokines added). For primary AML blasts, 1 × 10⁶ cells from cryopreserved bone marrow aspirates were plated in a 24-well plate. Control T cells or B7-H3.CAR-Ts were then added at a 1:5 effector-to-target (E:T) ratio. Twenty-four hours later, 500 μL of coculture supernatant was collected and the concentration of IFNγ and IL2 was measured by ELISA (R&D Systems). On days 4–7 of coculture, cells were collected, and the percentage of T cells and residual tumor cells was assessed via flow cytometry. Monocytic cell lines and primary AML blasts were identified via staining for CD33 and T cells via staining for CD3.

CD34⁺ cells were isolated via Magnetic Microbead (Miltenyi Biotec) selection from umbilical cord blood units obtained from healthy donors. CD34⁺ cells were coincubated with B7-H3.CAR-28-Ts, CD19.CAR-28-Ts, or control T cells [derived from donor peripheral blood mononuclear cells (PBMC)] at an E:T ratio of 10:1 (T-cell concentration of 5 × 10⁵/mL) for 6 hours, after which the coculture was diluted 1:11 (without depletion of T cells) in methylcellulose-containing medium supplemented with recombinant cytokines (MethoCult H4434 Classic; StemCell Technologies) and plated in duplicate as described previously (16). After 2 weeks in culture, granulocyte-macrophage progenitor (GMP) and common myeloid progenitor (CMP) colonies were scored in a blinded fashion using a high-quality inverted microscope.

For mouse xenograft models, monocytic AML cell lines (THP1 and OCI-AML2) were transduced with a retroviral vector encoding the eGFP firefly luciferase (eGFP-FFluc) gene (17). For the THP1 model, a single-cell clone was first selected on the basis of high eGFP expression and high in vitro FFluc activity (18). eGFP-FFluc-THP1 or FFluc-OCI-AML2 cells were injected into the lateral tail vein of 6- to 8-week-old female NOD scid IL2Rγ^−/− (NSG) mice. Seven days later, mice were injected intravenously with CAR-T cells. Further details of each experiment are described below. Tumor growth was monitored via bioluminescence imaging using an IVIS Lumina II In Vivo Imaging System (PerkinElmer; ref. 16). Mice were monitored and euthanized according to UNC (Chapel Hill, NC) Institutional Animal Care and Use Committee (IACUC) standards and/or when the luciferase signal (total flux) reached a predetermined threshold (1 × 10¹⁰ photons/second for eGFP-FFLuc-OCI-AML2 model and 1 × 10¹¹ photons/second for eGFP-FFluc-THP1 model due to increased FFluc intensity compared with eGFP-FFLuc-OCI-AML2).

Patient-derived AML blasts were obtained from PRoXe (samples DFAM-6855 and CBAM-44728) and expanded in vivo in NSG mice. For the first model (DFAM-6855), AML blasts were injected intravenously into 6- to 8-week-old female NSG mice and engraftment was monitored via weekly flow cytometry analysis of peripheral blood CD45⁺CD33⁺ cells. When the majority of mice had >20% peripheral blood AML blasts, they were randomized and injected intravenously with either B7-H3.CAR-28-Ts or control T cells. Eight days later, all mice were sacrificed and flow cytometry was used to quantify AML blasts and T cells in blood, bone marrow, and spleen. In an additional experiment, blasts were retrovirally transduced to express the fusion protein, eGFP-FFLuc, expanded in vivo in NSG mice, sorted by flow cytometry on the basis of eGFP expression, expanded again in vivo, and then similarly injected into female NSG mice. After 4 days, mice were randomized and injected intravenously with either B7-H3.CAR-28-Ts or CD19.CAR-28-Ts. Tumor growth was assessed via bioluminescence imaging using an IVIS Lumina II In Vivo Imaging System (PerkinElmer; ref. 16), unless otherwise indicated. Mice were monitored and euthanized according to UNC IACUC standards. In a second model (CBAM-44728), AML blasts were expanded in vivo and then injected intravenously in 6- to 8-week-old female NSG mice. After 4 days, mice were injected intravenously with either B7-H3.CAR-28-Ts or CD19.CAR-28-Ts. All mice were sacrificed on day 63 after T-cell injection and flow cytometry analysis was used to quantify AML blasts and CAR-Ts in the bone marrow, spleen, and blood.

Hematopoietic stem cells (HSC) were isolated from human fetal liver tissue specimens (obtained from medically indicated or elective pregnancy termination) and were injected intrahepatically into 2- to 6-day-old NSG mice (19). Engraftment of human leukocytes was monitored in the peripheral blood beginning at 6 weeks. Following reconstitution of 40%–60% human leukocytes, 2 mice were sacrificed and Magnetic Microbeads (Miltenyi Biotec) were used to isolate human T cells (CD4⁺ or CD8⁺ cells) from the murine spleens. After activation with CD3 and CD28 antibodies, a subset of T cells was retrovirally transduced to generate B7-H3.CARs-Ts as described above. B7-H3.CAR-28-Ts or control T cells were then injected intravenously into the remaining mice on days 0 and 16. After the initial T-cell injection, mice were monitored regularly for the development of graft-versus-host disease or systemic inflammatory response syndrome. Levels of circulating human CD45⁺CD3⁺, CD45⁺CD19⁺, CD45⁺CD33⁺, and CD45⁺CD14⁺ cells were monitored via flow cytometry on a weekly basis. Mice were sacrificed on day 40 and flow cytometry was used to evaluate these populations in the bone marrow and spleen, as well as levels of human CD34⁺CD38⁺ and CD34⁺CD38⁻ cells in the bone marrow.

Human monocytic AML cell lines (THP1, U937, OCI-AML2, and OCI-AML3) were stained with the B7-H3–specific mAbs, 376.96 and 7-517, demonstrating high expression of B7-H3 (Fig. 1A). In addition to staining with the B7-H3–specific mAb 7-517, the same cell lines, as well as a B7-H3⁻ acute promyelocytic leukemia cell line (HL60), and healthy donor peripheral blood monocytes (CD3⁻CD19⁻CD20⁻CD56⁻HLA-DR⁺CD14⁺) were stained with mAbs specific for other recently identified candidate surface antigens for AML-directed CAR-T therapy (ADGRE2, CCR1, CD70, LILRB2, and CLEC12A; ref. 20). As shown in Fig. 1B, the lack of B7-H3 expression on normal human monocytes and high expression on monocytic AML cell lines were similar to the expression pattern of CD70, whereas ADGRE2, CCR1, LILRB2, and CLEC12A were all found to be expressed on normal monocytes and had variable expression on monocytic AML cell lines.

We obtained deidentified cryopreserved pretreatment bone marrow aspirates (n = 10) from patients with monocytic AML (see Fig. 1E for expression of myeloid and monocytic markers; see Supplementary Table S1 for clinical annotation associated with these samples) and demonstrated that blast populations (identified on the basis of CD45-SSC gating; Fig. 1C) had positive staining with the B7-H3–specific mAbs, 376.96 and 7-517 (Fig. 1D). B7-H3 expression varied across samples, with some samples displaying uniformly high B7-H3 expression.

With the caveat that B7-H3 is likely regulated at the posttranscriptional level (7), we also analyzed a previously published RNA sequencing dataset (21) including 200 patients with untreated, de novo AML. B7-H3 expression was higher in patients with monocytic AML [French-American-British (FAB) M5; P < 0.0001] and patients with acute promyelocytic leukemia (FAB M3 or documented PML-RARA fusion; P < 0.0001 and P = 0.0004, respectively), whereas B7-H3 expression was lower in FAB M2 AML (P < 0.0001; Supplementary Fig. S1A, S1C, and S1D). B7-H3 expression was higher in patients with mutations in NPM1 (P = 0.0224) and lower in patients with mutations in IDH1/IDH2 (P = 0.0460), CEBPA (P < 0.0001), and WT1 (P = 0.0001; Supplementary Fig. S1E). There was also a positive correlation of B7-H3 expression with CD33 (Supplementary Fig. S1F) and CD11b (Supplementary Fig. S1G) expression, and a negative correlation between the expression of B7-H3 and CD34 (Supplementary Fig. S1H). When stratified by B7-H3 z-score of <0 and ≥0, the median overall survival (mOS) was 26.3 and 11.8 months, respectively, although the difference was not significant (P = 0.2005; Supplementary Fig. S1I). When acute promyelocytic leukemia was excluded, mOS was 21.5 and 9.9 months (P = 0.0432; Supplementary Fig. S1J). In a separate dataset including samples from 562 patients with both newly diagnosed and previously treated AML (22), we again found significantly higher B7-H3 expression in patients with FAB M3 (P = 0.0026) and M5 (P = 0.0119) AML, and lower expression in patients with FAB M2 AML (P = 0.0018; Supplementary Fig. S1K). In the same dataset (22), we found higher B7-H3 expression in patients with relapsed versus untreated AML (P = 0.0281; Supplementary Fig. S1L).

We also evaluated the expression of B7-H3 on normal myeloid progenitor populations using cryopreserved normal bone marrow aspirates. Bone marrow aspirates were analyzed by flow cytometry using the gating strategy shown in Fig. 1F to identify the following hematopoietic progenitor populations: megakaryocyte-erythroid progenitors (MEP⁻), common myeloid progenitors (CMP), granulocyte-macrophage progenitors (GMP), HSCs, and multipotent progenitor cells (MPP). We demonstrated no significant staining with the B7-H3–specific mAb 7-517 on any of these populations (Fig. 1G). Similarly, using the gating strategy shown in Fig. 1H, we evaluated the CD45⁺Lineage⁻CD34⁺CD38⁺ and CD45⁺Lineage⁻CD34⁺CD38⁻ populations, and demonstrated no significant staining with the B7-H3–specific mAb 376.96 (Fig. 1I).

Utilizing the scFv from the B7-H3 mAb 376.96 (13, 23), we have previously generated retroviral constructs encoding B7-H3.CAR-28 (incorporating the CD28 endodomain) and B7-H3.CAR-BB (incorporating the 4-1BB endodomain; ref. 11). Following activation with CD3/CD28, donor PBMCs were retrovirally transduced with the constructs mentioned above to generate B7-H3.CAR-Ts (Fig. 2A). CD19.CAR-28-Ts were used as a control. Median (range) transduction efficiencies for the B7-H3.CAR-28, B7-H3.CAR-BB, and CD19.CAR-28 constructs among eight representative donors were 77% (66%–87%), 70% (55%–87%), and 73% (65%–84%), respectively (Fig. 2B). The antileukemia efficacy of B7-H3.CAR-Ts was evaluated in coculture assays. B7-H3.CAR-Ts, but not control T cells, eliminated B7-H3⁺ cell lines (OCI-AML2, OCI-AML3, THP1, and U937; Fig. 2C and D). In the same experiments, we demonstrated significant levels of IFNγ (Fig. 2E) and IL2 (Fig. 2F) in the media harvested, following a 24-hour incubation, from cocultures with B7-H3.CAR-Ts, but not control T cells. To further assess antigen specificity of B7-H3.CAR-Ts, we stained B7-H3.CAR-Ts with CFSE and demonstrated B7-H3–specific proliferation of B7-H3.CAR-Ts (but not control T cells) in coculture with B7-H3⁺ cell lines (THP1 and U937), but not B7-H3⁻ cell lines (Kasumi and HL60; Fig. 2G).

To assess the activity of B7-H3.CAR-Ts against primary AML blasts, we cocultured B7-H3.CAR-Ts derived from healthy donors with primary AML blasts obtained from patients with monocytic AML. When cocultured at a 1:5 E:T ratio for 48 hours, B7-H3.CAR-Ts, but not control T cells, displayed significant cytotoxicity against primary monocytic AML blasts (Fig. 3A and B). We also demonstrated significant release of IFNγ and IL2 by B7-H3.CAR-Ts as measured in the media harvested from cocultures following a 24-hour incubation (Fig. 3C and D). The B7-H3 expression for the same patient samples is shown in Fig. 1D. B7-H3.CAR-Ts appeared to have more cytotoxic activity against AML blasts with higher B7-H3 expression, although this difference was not statistically significant (Supplementary Fig. S3). Given the shorter incubation time for these cocultures as compared with the cocultures with AML cell lines, shown in Fig. 2C–F, we demonstrated that under coculture conditions comparable with those used in Fig. 3A–D and H–K, healthy donor–derived B7-H3.CAR-Ts (n = 3) exhibited similar cytotoxicity (P < 0.0001) against the AML cell line, THP1 (Fig. 3M).

We utilized cryopreserved bone marrow aspirates from patients (n = 4) with AML with variable B7-H3 expression (Fig. 3E) to generate B7-H3.CAR-Ts and activated control T cells. Median (range) transduction efficiencies of B7-H3.CAR-28 and B7-H3.CAR-BB were 56% (49%–68%) and 56% (46%–59%), respectively (Fig. 3F and G). Utilizing a second cryopreserved bone marrow aspirate sample from the same patient, we demonstrated that autologous B7-H3.CAR-28-Ts (P = 0.0085) and B7-H3.CAR-BB-Ts (P = 0.0085) both displayed significant cytotoxicity against primary AML blasts when cocultured at a 1:5 E:T ratio for 48 hours (Fig. 3H and I). On the basis of clinical annotation, the median blast percentage of the bone marrow aspirates used for these cocultures was 86% (range, 78%–92%). For coculture analysis, blasts were identified as CD45⁺/CD33⁺ cells. Release of IFNγ and IL2 by B7-H3.CAR-Ts and control T cells was measured in the media harvested from cocultures following a 24-hour incubation period. IFNγ and IL2 release was detected from B7-H3.CAR-Ts and was mostly undetectable in the media of control T cells (P = 0.0750 and P = 0.1524 for IFNγ and IL2, respectively; Fig. 3J and K). Under identical coculture conditions, we also demonstrated the ability of AML patient–derived (n = 3) B7-H3.CAR-28-Ts (P = 0.0049) and B7-H3.CAR-BB-Ts (P = 0.0049) to efficiently kill the AML cell line, U937 (Fig. 3L).

We evaluated the in vivo efficacy of B7-H3.CAR-Ts in two murine models of disseminated AML as described above. In the first model, eGFP-FFLuc-OCI-AML2 cells (2 × 10⁶ cells/mouse) were injected intravenously into NSG mice on day −7, followed by injection (1 × 10⁷ cells/mouse) of B7-H3.CAR-28-Ts (n = 10), B7-H3.CAR-BB-Ts (n = 10), or CD19.CAR-28-Ts (n = 7) on days 0 and +20. Leukemia proliferation was assessed via weekly bioluminescence imaging. Tumor proliferation in mice treated with B7-H3.CAR-Ts was significantly delayed, as compared with mice treated with CD19.CAR-28-Ts (P < 0.0001; Fig. 4A–D). We elected to perform a second CAR-T injection in this model because of data from prior experiments in which a higher eGFP-FFLuc-OCI-AML2 dose (5 × 10⁶ cells/mouse) was injected intravenously in NSG mice at day −7, followed by a single injection (1 × 10⁷ cells/mouse) of B7-H3.CAR-28-Ts, B7-H3.CAR-BB-Ts, or CD19.CAR-28-Ts (n = 5 per group) at day 0 (Supplementary Fig. S4A–S4D). In this experiment, we observed a significant prolongation of survival (P = 0.0008) in B7-H3.CAR-T–treated mice, but all mice ultimately relapsed prior to day 40. We hypothesized that this may have been related to the kinetics of the tumor cell growth compared with CAR T-cell expansion in vivo, and that a second CAR T-cell injection could potentially overcome this limitation. In a pilot experiment, we also observed that in mice initially engrafted with eGFP-FFLuc-OCI-AML2 cells on day −7 (5 × 10⁶ cells/mouse), followed by CD19.CAR-28-T injection (1 × 10⁷ cells/mouse) on day 0, tumor growth could be controlled after injection of B7-H3.CAR-28-Ts (1 × 10⁷ cells/mouse) on day 20 (Supplementary Fig. S4E–S4G).

In the second model, eGFP-FFLuc–expressing THP1 cells (5 × 10⁶ cells/mouse) were similarly injected intravenously into NSG mice on day −7, followed by intravenous injection (1 × 10⁷ cells/mouse) of B7-H3.CAR-Ts (n = 10 mice/construct) or CD19.CAR-28-Ts (n = 8 mice). Tumor growth was similarly assessed via weekly bioluminescence imaging. Both B7-H3.CAR-28-Ts and B7-H3.CAR-BB-Ts, but not CD19.CAR-28-Ts, effectively controlled the growth OCI-AML2 cells in vivo (P = 0.0004; Fig. 4E–H). Surface B7-H3 expression was found to be preserved on eGFP-FFluc-THP1 cells isolated from mice that relapsed after treatment with B7-H3.CAR-Ts (Supplementary Fig. S2A and S2B).

To evaluate the activity of B7-H3.CAR-Ts against human primary AML blasts, we utilized PDX mouse models of AML. For these experiments, because of the limited availability of tumor cells, we conducted the experiment using only B7-H3.CAR-Ts carrying the CD28 costimulation. For the first PDX model, we initially confirmed B7-H3 expression on primary AML blasts from sample DFAM-6855 (99.1% B7-H3⁺ using mAb 7-517; Fig. 5A) and confirmed the in vitro activity of B7-H3.CAR-28-Ts against these cells (Fig. 5B). NSG mice were then injected (1 × 10⁶ cells/mouse) on day −21 with primary AML blasts (Fig. 5C). After confirmation of engraftment (Fig. 5D), B7-H3.CAR-28-Ts (n = 7) or control T cells (n = 5) were injected (1 × 10⁷ cells/mouse) on day 0. Mice were sacrificed on day 8 and peripheral blood, bone marrow, and spleens were harvested (Fig. 5C). Compared with mice treated with control T cells, we demonstrated significantly decreased huCD45⁺CD33⁺ AML blasts in the peripheral blood (P = 0.0016) and spleen (P = 0.007; Fig. 5E). Also, compared with control mice, we demonstrated significantly increased CD8⁺ T-cell expansion in the spleens of B7-H3-CAR-28-T–treated mice (Fig. 5F).

In an additional experiment, the same patient-derived AML blasts were transduced to express eGFP-FFLuc and were injected intravenously (1 × 10⁶ cells/mouse) into NSG mice on day −4, followed by injection (1 × 10⁷ cells/mouse) of B7-H3.CAR-28-Ts (n = 7 mice) or CD19.CAR-28-Ts (n = 3 mice) on day 0 (Fig. 5G). Leukemia proliferation was assessed via weekly bioluminescence imaging. Compared with CD19.CAR-28-T–treated mice, B7-H3.CAR-28-T–treated mice experienced significantly delayed tumor growth (P < 0.0001; Fig. 5H and I) and significantly prolonged survival (P = 0.0027; Fig. 5J). B7-H3 expression was also found to be preserved on patient-derived AML blasts recovered from the spleen after relapse following treatment with B7-H3.CAR-28-Ts (Supplementary Fig. S2E).

In a second PDX model, we used patient-derived AML blasts from sample CBAM-44728 as described above. After confirming B7-H3 expression (Supplementary Fig. S5B), blasts were injected intravenously (8 × 10⁵ cells/mouse) into NSG mice on day −4, followed by intravenous injection (1 × 10⁷ cells/mouse) of B7-H3.CAR-28-Ts (n = 6 mice) or CD19.CAR-28-Ts (n = 5 mice) on day 0 (Supplementary Fig. S5C). All mice were sacrificed on day 63, followed by analysis of patient-derived AML blasts and CAR-Ts in the bone marrow, spleen, and/or blood. AML blasts were significantly decreased (undetectable in 4/6 mice; 2 mice with ≤100 blasts/femur) in the bone marrow of B7-H3.CAR-28-T- compared with CD19.CAR-28-T–treated mice (P = 0.0043) and undetectable in the spleens of B7-H3.CAR-28-T–treated mice (P = 0.0022; Supplementary Fig. S5D). Both B7-H3.CAR-28-Ts and CD19.CAR-28-Ts were detectable in the blood and/or spleens of all mice (Supplementary Fig. S5E).

To evaluate the potential toxicity of B7-H3.CAR-Ts against normal myeloid progenitor populations, we first utilized an in vitro CFU assay. Umbilical cord blood–derived CD34⁺ cells were coincubated with B7-H3.CAR-28-Ts, CD19.CAR-28-Ts, or control T cells derived from donor PMBCs (n = 4 cord blood donors, each cocultured with T cells from two PBMC donors), followed by culture in semisolid methylcellulose medium. No significant differences were observed in the numbers of granulocyte-macrophage progenitor (GMP), common myeloid progenitor (CMP), or total colonies (P = 0.75, P = 0.73, and P = 0.96, respectively; Fig. 6A).

To assess the effects of B7-H3.CAR-Ts on hematopoietic progenitors, we utilized a murine model of normal human hematopoiesis. Human fetal liver–derived CD34⁺ cells were engrafted in NSG mice (n = 16) as described above. When the median human CD45⁺ cell percentage was 47%, 2 mice were sacrificed and human T cells were isolated and used to generate B7-H3.CAR-28-Ts and CD19-CAR-28-Ts (transduction efficiencies 69% and 79%, respectively; Fig. 6B and C). B7-H3.CAR-28-T function was confirmed via in vitro coculture assay with the THP1 (B7-H3⁺) and BV173 (B7-H3⁻) cell lines (Fig. 6D). The remaining mice (n = 14) were injected intravenously with either B7-H3.CAR-28-Ts (n = 5), CD19.CAR-28-Ts (n = 5), or control T cells (n = 4) at days 0 and 16 (5 × 10⁶ cells/mouse at each timepoint). Human CD3⁺, CD33⁺, and CD14⁺ cells in the peripheral blood were assessed at baseline and every 7–9 days thereafter. Compared with mice treated with CD19.CAR-28-Ts and control T cells, mice treated with B7-H3.CAR-28-Ts had no significant decreases in the levels of circulating CD14⁺ or CD33⁺ cells (Fig. 6E). Mice were sacrificed at the conclusion of the experiment and spleen and bone marrow were evaluated via flow cytometry. Compared with mice treated with control T cells, absolute numbers of CD14⁺ and CD33⁺ cells were moderately decreased, but not eliminated, in the spleens of B7-H3.CAR-28-T–treated mice (P = 0.0196 and 0.0359, respectively; Fig. 6F). No significant differences in the level of CD14⁺ and CD33⁺ cells were detected in the bone marrow (Fig. 6G).

Myeloid progenitor populations were subsequently evaluated in the bone marrow by gating on live, huCD45⁺, lineage⁻ (CD3⁻CD14⁻CD16⁻CD19⁻CD20⁻CD56⁻), CD34⁺CD38⁺, and CD34⁺CD38⁻ populations. Compared with mice treated with control T cells, there was a significant decrease in the percentage of CD34⁺CD38⁺ cells in mice treated with B7-H3.CAR-28-Ts (P = 0.0169). However, this effect did not appear to be specific to B7-H3.CAR-Ts as a similar decrease in this population was seen in mice treated with CD19.CAR-28-Ts (P = 0.0206); as discussed below, we hypothesize that this is related to nonspecific CAR-T activation (Fig. 6H). Apparent decreases in the percentage of CD34⁺CD38⁻ cells in B7-H3.CAR-28-T- and CD19.CAR-28-T–treated mice were not statistically significant (P = 0.0705; Fig. 6H).

We have previously generated B7-H3–specific CAR-T cells, demonstrated antitumor efficacy in several solid tumor models, confirmed the absence of significant B7-H3 expression in most normal human tissues, and, by taking advantage of the cross-reactivity of B7-H3.CAR-Ts with the murine form of B7-H3, demonstrated a lack of evident toxicities in a syngeneic tumor model, further supporting the potential clinical translation of B7-H3.CAR-Ts (11). In this study, we investigated the potential of B7-H3 as a clinically relevant target antigen for AML-directed CAR-T therapy. We have demonstrated that B7-H3.CAR-Ts could represent a safe and effective therapy for a subset of patients with AML, and that such therapy is unlikely to cause unacceptable hematopoietic toxicity.

In addition to direct antitumor effects, it is also possible that B7-H3.CAR-Ts could augment adaptive immune responses in AML (7, 24, 25). B7-H3 expression on tumor cells, and on myeloid-derived suppressor cells in the tumor microenvironment, is likely to play an immunosuppressive role, and may drive immune escape in multiple cancer types (7, 24–27). Patients with solid tumor treated with an anti-B7-H3 mAb were shown to have significant posttreatment increases in T-cell receptor clonality (26). B7-H3 and other costimulatory ligands (e.g., PD-L1 and CD80) are also upregulated in patients with AML who relapse following allogeneic stem cell transplantation, suggesting that B7-H3 may contribute to impaired allorecognition by donor T cells in this setting (28).

B7-H3 is thus a promising therapeutic target for human cancers, and multiple clinical trials are evaluating anti-B7-H3 mAbs, including a bispecific antibody targeting B7-H3 and CD3 (7, 27–34). In this study, we have provided efficacy and additional safety data to support the clinical assessment B7-H3.CAR-Ts as a treatment for AML. Our data suggest that B7-H3.CAR-Ts could potentially overcome the prolonged hematopoietic toxicity that has limited the development of CAR-Ts targeting certain other AML-associated antigens (35, 36). In the humanized mouse model described above, compared with mice treated with control T cells, we observed a similar decrease in myeloid progenitor populations in mice treated with both B7-H3.CAR-28-Ts and CD19.CAR-28-Ts. We suspect that this is explained by nonspecific activation of the CAR-Ts, rather than antigen-specific toxicity against myeloid progenitors. Although plasma cytokine levels were not measured in this experiment, in a similar humanized mouse model, our group previously demonstrated significantly increased levels of IFNγ, TNFα, and IL6 in the plasma of mice treated with CD19.CAR-Ts as compared with control T cells (37). Our data, together with the lack of reported myeloid toxicity in patients infused with a B7-H3–specific mAb (33), suggest that targeting B7-H3 will be associated with a better safety profile compared with other pan-myeloid markers. Nonetheless, for patients enrolled in a phase I clinical study of B7-H3.CAR-Ts, it would be recommended to have the option of HSC rescue, if needed.

Although we did not observe B7-H3 antigen loss in cells isolated from relapsed mice in either our xenograft mouse models using AML cell lines or in the PDX models, it remains to be seen whether antigen escape will limit the clinical efficacy of B7-H3.CAR-Ts in patients with AML. It is also possible that increased adaptive immune responses could overcome this potential limitation. If antigen escape does occur, B7-H3 could still represent an attractive potential target antigen for use in a combinatorial approach. Finally, we recognize that this study also does not fully characterize the expression of B7-H3 on AML leukemic stem cell (LSC) populations. AML LSCs are most well defined in the approximately 75% of AML samples that are positive for CD34, and typically reside in the CD34⁺CD38⁻ population of AML blasts. In patients with CD34⁻ AML, the majority of LSCs are thought to reside in the CD34⁻ population; however, LSCs in this subset of AML are less well studied (30). Indeed, recent studies evaluating potential LSC targets for CAR-T therapy in AML have focused only on the CD34⁺CD38⁻ blast population (32). We have found that most B7-H3⁺ AMLs are negative for CD34 (which is consistent with a monocytic immunophenotype), and it is thus more challenging to reliably evaluate the LSC expression of B7-H3. It is notable, however, that B7-H3.CAR-Ts demonstrated in vitro cytotoxicity against patient-derived AML samples that were also shown to have the ability to reliably engraft in NSG mice. Nonetheless, in the absence of serial transfer experiments in vivo, it is not possible to definitively conclude whether there is differential activity of B7-H3.CAR-Ts against LSCs versus healthy hematopoietic progenitor cells.

The optimal choice of costimulatory domain for CAR-T therapy remains unclear, and likely differs depending on factors related to the tumor growth kinetics, target antigen characteristics, and the scFv used. CD28-containing CARs are known to activate faster, to produce larger magnitude changes in protein phosphorylation, and to lead to a more effector T-cell–like phenotype as compared with 4-1BB–containing CARs (31). Prior studies suggest that 4-1BB–containing CARs may be better able to maintain effector functions in the setting of chronic antigen stimulation (31). Reduced levels of cytokine production with 4-1BB–containing CARs may also correlate with a lower potential for cytokine release syndrome or neurotoxicity (31), and the only FDA-approved CAR-T therapy for ALL incorporates the 4-1BB signaling domain. Prior work by our group in a pancreatic adenocarcinoma model has shown that 4-1BB–containing B7-H3.CAR-Ts expressed lower levels of PD-1 and showed greater antitumor efficacy against pancreatic ductal adenocarcinoma (PDAC) cell lines transduced to constitutively express PD-L1 (11). However, the kinetics of PDAC tumor growth are clearly different than that of AML, and these advantages may not necessarily translate to the treatment of AML. In this study, most experiments were designed to evaluate B7-H3.CAR-28-Ts and B7-H3.CAR-BB-Ts in parallel. Although there was a trend toward greater efficacy of B7-H3.CAR-28-Ts, our data generally suggest similar efficacy between the two constructs, both in vitro and in mouse xenograft models. Any potential advantage of B7-H3.CAR-28-Ts is likely explained by the rapid kinetics of AML cell line expansion in the model systems used. While the optimal costimulatory domain remains unclear and may depend, in part, on the burden of disease at the time of CAR-T infusion, we have elected to pursue B7-H3.CAR-28-Ts for initial clinical evaluation in AML.

In conclusion, our data support further evaluation of B7-H3 as a target for AML-directed CAR-T therapy, demonstrate that B7-H3.CAR-Ts are able to control the proliferation of B7-H3⁺ AML both in vitro and in xenograft models, and with the limitations discussed above, suggest that B7-H3.CAR-Ts are unlikely to cause significant hematopoietic toxicity.

E.I. Lichtman reports grants from ASCO/Conquer Cancer Foundation, NIH/NCI, and North Carolina University Cancer Research Fund and other from NIH/NCI during the conduct of the study; E.I. Lichtman served (without compensation) on an advisory board for Seattle Genetics. H. Du reports a patent for B7-H3–specific CAR issued, licensed, and with royalties paid from bluebird Bio. B. Savoldo reports personal fees and other from Tessa Therapeutics and grants from bluebird bio, NIH, and Cell Medica outside the submitted work. G. Dotti reports grants from Cell Medica, Bellicum Pharmaceutical, and bluebird bio and personal fees from Catamaran, Bellicum Pharmaceutical, and Tessa Therapeutics outside the submitted work, and has been issued a patent on the B7-H3.CAR. No disclosures were reported by the other authors.

E.I. Lichtman: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. H. Du: Conceptualization, investigation, methodology, writing–review and editing. P. Shou: Conceptualization, investigation, methodology, writing–review and editing. F. Song: Investigation, methodology. K. Suzuki: Methodology. S. Ahn: Investigation, methodology, writing–review and editing. G. Li: Investigation, methodology, writing–review and editing. S. Ferrone: Resources, methodology, writing–review and editing. L. Su: Resources, methodology, writing–review and editing. B. Savoldo: Conceptualization, resources, supervision, investigation, methodology, writing–review and editing. G. Dotti: Conceptualization, resources, data curation, supervision, funding acquisition, methodology, writing–review and editing.

This work was partially supported by the University Cancer Research Fund at the University of North Carolina (to G. Dotti). E.I. Lichtman was supported by a Conquer Cancer Foundation of the American Society of Clinical Oncology Young Investigator Award and NCI grant 1T32CA211056-01A1 (to principal investigator, Jonathan Serody). S. Ferrone was supported by W81XWH-16-1-0500DOD, R01DE028172, and R03CA223886. The small-animal imaging core and flow cytometry core facilities were supported, in part, by an NCI Cancer Core grant (P30-CA016086-40). Cryopreserved human AML samples were provided by the Tissue Procurement Facility at UNC-LCCC.

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.