Cooperative CAR targeting to selectively eliminate AML and minimize escape

Summary

Acute myeloid leukemia (AML) poses a singular challenge for chimeric antigen receptor (CAR) therapy owing to its phenotypic heterogeneity and similarity to normal hematopoietic stem/progenitor cells (HSPCs). Here we expound a CAR strategy intended to efficiently target AML while minimizing HSPC toxicity. Quantification of target expression in relapsed/refractory patient samples and normal HSPCs reveals a therapeutic window for gated co-targeting of ADGRE2 and CLEC12A: We combine an attenuated ADGRE2-CAR with a CLEC12A-chimeric costimulatory receptor (ADCLEC.syn1) to preferentially engage ADGRE2^posCLEC12A^pos leukemic stem cells over ADGRE2^lowCLEC12A^neg normal HSPCs. ADCLEC.syn1 prevents antigen escape in AML xenograft models, outperforms the ADGRE2-CAR alone and eradicates AML despite proximate myelopoiesis in humanized mice. Off-target HSPC toxicity is similar to that of a CD19-CAR and can be mitigated by reducing CAR T cell-derived interferon-γ. Overall, we demonstrate the ability of target density-adapted cooperative CAR targeting to selectively eliminate AML and potentially obviate the need for hematopoietic rescue.

eTOC Blurb

To address efficacy and safety challenges in AML, Haubner et al. quantify antigen distribution and densities in primary AML and normal cells to guide a cooperative CAR targeting strategy. ADCLEC.syn1 operates via CCR-gated sensitivity and thereby overcomes antigen escape in heterogenous AML and reduces on-target toxicity in humanized mice.

Graphical Abstract

Introduction

Acute myeloid leukemia (AML) remains difficult to treat due to relapsed or refractory (r/r) disease, leading to poor outcomes with approximately 30% 5-year overall survival.¹ Allogeneic hematopoietic stem cell transplantation (allo-HSCT) is currently the only therapeutic option with curative potential for r/r AML, but treatment-related mortality and post-transplant relapse rates are high.² There is an immense clinical need for novel therapies for r/r AML. Chimeric antigen receptor (CAR) therapy has been established as highly potent therapeutic option in the setting of r/r B cell malignancies and multiple myeloma, improving long-term remission rates as well as overall survival.^3–12 Current efforts to further enhance CD19- and BCMA-directed CAR therapies for B cell malignancies and multiple myeloma, respectively, aim to maximize tumor targeting and functional CAR T cell persistence. In these settings, the restricted expression profiles of the CAR target antigens limit on-target toxicities to the depletion of normal B cells or plasma cells without incurring the risk of eliminating hematopoietic stem cells. This is not the case in AML where the surface proteome of leukemic cells bears a high degree of similarity to that of normal HSPCs.¹³ Thus, while CD19 and BCMA stand out as clinically validated CAR targets in FDA-approved CAR T cell products, optimal target choices in AML are still uncertain.

One of the key mechanisms of AML resistance owes to disease heterogeneity. AML is indeed driven by multiple leukemic stem cell (LSC) clones with different trajectories of sequentially acquired mutations leading to an overall AML population that is genomically heterogenous.^14,15 The clonal complexity of AML LSCs is associated with differences in surface protein expression of stem/progenitor and myeloid antigens,^16,17 and heterogenous expression levels of common CAR targets.^13,18,19 A second, perhaps greater, challenge is to achieve selective targeting of AML LSCs without depleting the normal hematopoietic stem/progenitor cell (HSPC) pool. At the transcriptional and surface protein level, AML LSCs have been found to share phenotypic similarities to normal HSPCs, raising the concern for on-target hematotoxicity.^13,18,20,21 Not surprisingly, CAR-related myeloablation due to on-target hematotoxicity has been observed both pre-clinically^22–26 and clinically^27–29 with CD33-, CD123- or CLEC12A-targeted CARs. As emerging clinical CAR reports have demonstrated both poor efficacy and concerning on-target toxicities, CAR therapies are being harnessed before or after obligatory allo-HSCT to rescue normal hematopoiesis. HSCT may either be conventional or a gene-edited transplant lacking expression of the CAR target epitope to protect the HSCT from CAR-related on-target hematotoxicity.^30–35

We have not discarded the prospect of a standalone CAR T cell approach in AML and to this end have searched for a combinatorial targeting strategy that can eliminate LSCs with minimal targeting of HSPCs. Our previous work identified a series of candidate surface targets that pass stringent safety criteria based on their expression profiles in normal hematopoietic and non-hematopoietic tissues.¹³ One of the top candidates with abundant expression in AML and low or absent expression in most normal tissues is ADGRE2, an adhesion G protein-coupled receptor. ADGRE2 expression in AML correlates with poor molecular risk profile as well as shorter overall survival independent of age, molecular risk status and transplant status.³⁶ Given the known heterogeneity in AML and shared target expression in normal tissues, we hypothesized that a quantitative target density profiling of ADGRE2 and other established AML targets such as CD33, CD123 and CLEC12A may inform a strategy for a combinatorial CAR design to target AML safely and efficiently. Based on quantitative measurements of candidate target expression in r/r AML and normal hematopoietic cells we defined a gated, combinatorial strategy targeting ADGRE2 and CLEC12A that achieved efficient elimination of representative AML cell lines and patient-derived xenograft (PDX) models, including in humanized hematopoietic chimeras, with HSPC toxicity that did not exceed that of a control CD19-targeted CAR therapy. These results are the foundation for a phase 1 clinical trial utilizing this design, termed ADCLEC.syn1, in patients with r/r AML (NCT05748197).

Results

Antigen distribution and density inform AML target selection

To rationally select AML targets and an adapted targeting strategy for CAR therapy of AML, we quantitatively profiled surface target protein expression in bone marrow and peripheral blood (PB) samples of AML patients and healthy donors. The AML patient cohort included 39 patients with r/r AML of diverse genetically defined subtypes, including AML with recurrent genetic abnormalities, mutated TP53 and myelodysplasia-related abnormalities (Figure S1A).

We analyzed by flow cytometry both target distribution (percentage positivity) and target density (absolute number of surface molecules/cell) of ADGRE2, CLEC12A, CD33 and CD123. Cell surface expression analysis distinguished between bulk AML and the CD34^pos/CD38^neg LSC-enriched fraction. Target antigen distribution, represented as median percentage of target-positive AML cells in the patient cohort, was overall similar for ADGRE2 (bulk 82% / LSC 91%), CD33 (bulk 93% / LSC 87%) and CD123 (bulk 88% / LSC 96%) (Figure 1A). CLEC12A distribution was more heterogenous (bulk 54% / LSC 14%) and overlapping with ADGRE2 expression (CLEC12A^pos bulk 54% / LSC 14%; ADGRE2^pos∩CLEC12A^pos bulk 46% / LSC 11%). (Figures 1A-B, S1B).

Figure 1. Antigen distribution and density inform AML target selection.

(A-F) Flow cytometric quantitative profiling of AML surface target antigens ADGRE2, CLEC12A, CD33 and CD123 on AML bulk or LSCs in a relapsed/refractory AML patient cohort (n=39, Figure S1) and on normal hematological cells in a healthy adult donor cohort (n=8). Each dot represents an individual patient/donor sample.

(A-B) Target antigen distribution profiles on AML cells as measured by percentage positivity gating. Horizontal bars represent median percentage target positivity in total patient population. ADGRE2^pos∪CLEC12A^pos (A∪C) denotes union of ADGRE2 and CLEC12A positivity (positive for either one or both antigens), ADGRE2^pos∩CLEC12A^pos (A∩C) denotes intersection of ADGRE2 and CLEC12A positivity (positive for both antigens).

(C-E) Target antigen density profiles on AML and normal hematological cells as measured by median absolute numbers of surface molecules/cell. Dashed lines indicate 3×10² molecules/cell and 1×10³ molecules/cell.

(F) Heatmap summarizing target protein densities of ADGRE2, CLEC12A, CD33, CD123 and CD19 on normal bone marrow-derived hematological cell populations. Numbers within heatmap indicate median number of surface molecules/cell. Cell frequency of the respective population relative to total live bone marrow cells is represented by horizontal bars and numbers indicating median percentage from n=8 independent healthy donor samples.

(G) Mean target gene expression in different normal cell types of non-hematological origin based on single-cell RNAseq dataset.⁵⁶

(H) Schematic comparing single-targeting and combinatorial CAR designs (OR gate, AND gate, IF-BETTER gate) and their predicted killing (red filling) and sparing (blank filling) characteristics on tumor (T) and normal (N) cells depending on ADGRE2 and CLEC12A target densities. Axes indicate ADGRE2 and CLEC12A target densities from negative (left and bottom, respectively) to low (middle) and high (right and top, respectively).

Also see Figure S1.

Target antigen density was determined in AML and normal hematopoietic cells (represented as median surface molecule count per cell in Figure 1C-F). In AML LSCs, 95% of patients had detectable ADGRE2 (>3.0×10² molecules/cell), with 79% displaying more than 1.0×10³ ADGRE2 molecules per cell. CLEC12A was detectable in AML LSCs of 62% of patients, of whom 96% had detectable ADGRE2 expression. Detectable levels (>3.0×10² molecules/cell) of CD33 and CD123 in LSCs were found in 97% of patients, and CD33/CD123 co-expression of more than 1.0×10³ molecules/cell in LSC was found in 87% of patients (Figure 1C).

In healthy donor bone marrow-derived HSPCs and granulocytes, we observed a different pattern of ADGRE2 and CLEC12A expression compared to AML. Early HSPCs (phenotypically hematopoietic stem cells (HSCs), multipotent progenitors (MPPs) and multilymphoid progenitors (MLPs)) were found to express only 9×10², 8×10² and 1.1×10³ ADGRE2 molecules per cell, respectively, and no detectable levels of CLEC12A. Committed progenitors (CPs) expressed only low ADGRE2 (7×10² molecules) and CLEC12A (8×10² molecules). Granulocytes expressed 6.5×10³ CLEC12A molecules and no detectable ADGRE2. Only monocytes were found to co-express high levels of ADGRE2 and CLEC12A, with a median of 1.6×10³ ADGRE2 molecules and 1.2×10⁴ CLEC12A molecules per cell. On the other hand, CD33 and CD123 were co-expressed in all early HSPC subsets (up to 2.5×10³ CD33 and 1.9×10³ CD123 molecules), granulocytes (3.9×10³ CD33 and 5×10² CD123 molecules) and monocytes (2.4×10⁴ CD33 and 1.5×10³ CD123 molecules). For comparison, CD19 was only detected in normal B cells, averaging 5.4×10³ molecules per cell (Figure 1D-F). ADGRE2 and CD19 thus presented the most restricted pattern, with <4% positivity in total bone marrow cells, while CD33, CD123 and CLEC12A were detected in >60% of total bone marrow cells (Figure 1F).

We further evaluated target expression profiles in non-hematological cells in normal tissues, defining specific cell types based on single-cell transcriptomic profiles. In this analysis, ADGRE2, CLEC12A and CD33 were all found to be undetectable in n=13 different cell types, while CD123 expression was detected in adipocytes and endothelial cells (Figure 1G).

In summary, our target protein quantification in AML vs normal cells revealed differential expression patterns of ADGRE2 and CLEC12A: ADGRE2 and CLEC12A are co-expressed in AML LSCs, but not in normal HSPCs, while ADGRE2 is absent in granulocytes. In contrast, CD33 and CD123 were similarly expressed in AML but also found in most normal stem/progenitor and mature myeloid cells including granulocytes. Furthermore, the target density of ADGRE2 in normal HSPCs is lower than that of CD33 or CD123. Considering the target distribution and density profiles of these 4 cell surface molecules in AML vs various normal tissues, we proceeded to investigate a combinatorial targeting strategy that takes advantage of a potential therapeutic window to distinguish LSCs and HSPCs based on ADGRE2 and CLEC12A expression.

IF-BETTER gated combinatorial targeting of ADGRE2 and CLEC12A

CAR therapy for AML is constrained by the risk of AML escape on one end and on-target toxicities to normal HSPCs on the other. We hypothesized that suitable co-targeting of ADGRE2 and CLEC12A may enhance efficacy against AML including LSCs while limiting on-target toxicity. An OR-gate is the most common strategy to engage two antigens to provide greater tumor coverage in a heterogenous tumor setting. OR-gated CAR designs rely on co-expression of two independent CARs or one CAR with dual specificity, triggering cytolysis if either one of two targeted antigens is sufficiently expressed^37–43: This approach however increases the cumulative on-target toxicity potential, thus extending the relatively limited toxicity spectrum of ADGRE2 to the larger spectrum of CLEC12A (Figure 1H, second panel). Targeting ADGRE2 and CLEC12A in OR-gate format may limit clinical tolerability by compounding monocyte depletion with agranulocytosis, increasing the risk of life-threatening bacterial infections. AND-gated CAR designs rely on co-expression of two complementary but individually deficient receptors, triggering cytolysis only if two targeted antigens are co-expressed^44,45: This approach may limit on-target toxicity but increase the risk of tumor antigen escape if one of the two targets is absent or scarce (Figure 1H, third panel). We therefore adopted another combinatorial strategy to balance efficacy and safety requirements.

The lower ADGRE2 antigen density found in early HSPCs (<1.0×10³ molecules per cell) relative to its higher density in AML LSCs (>1.0×10³ molecules per cell) indicated that a killing threshold of approximately 1.0×10³ ADGRE2 molecules per cell may be best suited to preferentially target LSCs. As limited sensitivity may allow for antigen escape, we sought to selectively enhance ADGRE2 engagement on LSCs but not HSPCs by co-targeting CLEC12A. We hypothesized that a sensitivity-tuned ADGRE2-CAR co-expressed with a CLEC12A-targeted chimeric costimulatory receptor (CCR)⁴⁶ may display greater activity against ADGRE2^med/highCLEC12A^pos LSCs without increasing on-target/off-tumor toxicity against ADGRE2^lowCLEC12A^neg HSPCs. This combinatorial receptor design to preferentially direct target cell killing based on the presence or absence of a second target is best described as an IF-BETTER gate (Figure 1H, fourth panel).

A CLEC12A-CCR increases sensitivity of ADGRE2-CAR-1XX

To generate IF-BETTER gated chimeric receptor designs combining an ADGRE2-CAR and CLEC12A-CCR, we first screened different scFv candidates for both targets, aiming for an ADGRE2-CAR with a killing threshold sparing ADGRE2^low cells, and a CLEC12A-CCR that efficiently enhances the associated CAR activity (Figure S2A). For the ADGRE2-CAR, a library of n=18 humanized ADGRE2 scFv with different affinities was created by introducing V_H/V_L framework alterations. These ADGRE2 scFv were then screened in the CAR-28z1XX format, a CAR design with only 1 instead of 3 functional ITAMs in the CD3z domain, associated with calibrated activation and increased in vivo functional T cell persistence.⁴⁷ We selected lead scFv candidates in vitro and in vivo based on efficient elimination of ADGRE2^high target cells (1×10⁴ ADGRE2 molecules) and limited activity against ADGRE2^low target cells (1×10³ ADGRE2 molecules) achieving the intended targeting threshold under CAR T cell stress dose conditions (injection of 5.0×10⁵ or fewer CAR T cells). For the CLEC12A-CCR, a library of n=16 fully-human CLEC12A scFv with different epitopes was first screened in the CAR-28z1XX format, selecting a lead candidate with high efficacy in vitro and in vivo under CAR T cell stress dose conditions. The top ADGRE2-scFv and CLEC12A-scFv were chosen to construct a combinatorial design co-expressing an ADGRE2-CAR-28z1XX and a CLEC12A-CCR-BB incorporating a 4–1BB costimulatory endodomain, encoded by an SFG-gammaretroviral bicistronic vector, termed ADCLEC.syn1 (Figures 2A-B, S2B-C).

Figure 2. A CLEC12A-CCR increases sensitivity of ADGRE2-CAR-1XX.

(A) Schematic depicting ADCLEC.syn1 combinatorial receptor design.

(B) SFG-gammaretroviral bicistronic vector map for ADCLEC.syn1 expression.

(C) EL4 murine lymphoma cell line variants expressing either no target (black), ADGRE2 alone (green), CLEC12A alone (violet), or ADGRE2 and CLEC12A together (orange).

(D) 48h in vitro assay to measure cytotoxic activity of ADCLEC.syn1 vs 19-28z1XX CAR T cells at different effector:target (E:T) ratios in co-culture with EL4 cell line variants providing either no target, ADGRE2 alone, CLEC12A alone or ADGRE2 and CLEC12A together. T cell activation is represented by CD25 median fluorescence intensity (MFI) on T cells. Cytotoxicity was determined via flow cytometric enumeration of target cells. Data are represented as mean ± SEM. p value was determined via unpaired t test.

(E) MOLM13 AML cell line variants with low CLEC12A density (5×10² molecules/cell) and varying ADGRE2 antigen density: high/WT (1×10⁴), low (1×10³) and very-low (4×10²).

(F) 18h in vitro assay to measure cytotoxic activity of ADCLEC.syn1 vs its single receptor components (ADGRE2–28z1XX CAR or CLEC12A-BB CCR) vs untransduced T cells (UTD) in MOLM13 variants modeling antigen escape. Data are represented as mean ± SEM. p values were determined via unpaired t test.

(G) M13-A^low-C^high MOLM13 variant with low ADGRE2 (1×10³ molecules/cell) and high CLEC12A (2×10⁵).

(H) AML burden (total flux) and Kaplan-Meier survival analysis of M13-A^low-C^high-bearing mice treated with 5×10⁵ CAR^pos T cells.

(I) M13-A^very-low-C^high MOLM13 variant with very-low ADGRE2 (4×10² molecules/cell) and high CLEC12A (1×10⁵ molecules/cell).

(J) AML burden (total flux) and Kaplan-Meier survival analysis of M13-A^very-low-C^high-bearing mice treated with 5×10⁵ CAR^pos T cells.

Also see Figures S2, S3, S4.

To investigate how ADCLEC.syn1 operates in the context of individual ADGRE2-CAR and/or CLEC12A-CCR engagement, we set up a prototypic in vitro model of target cells expressing none, either single target, or both targets, co-cultured with ADCLEC.syn1 T cells. T cell activation assessed by CD25 upregulation was only observed in conditions where target cells expressed ADGRE2, enabling CAR engagement (Figure 2C-D). In line with T cell activation patterns, target cell lysis was limited to conditions where target cells expressed ADGRE2. In addition, CLEC12A co-expression enhanced target cell lysis (p=0.0061), but CLEC12A expression alone did not trigger T cell activation or target cell lysis (Figure 2D). To model contexts of cellular heterogeneity, we performed in vitro cytotoxicity assays in the presence of mixtures of cells expressing either only ADGRE2 or only CLEC12A. Bystander-mediated CLEC12A-CCR engagement did not alter ADGRE2-CAR-mediated killing of ADGRE2^posCLEC12A^neg AML cells (Figure S3A), and conversely, bystander-mediated ADGRE2-CAR activation did not trigger killing of ADGRE2^negCLEC12A^pos cells (Figure S3B). Overall, these results support that ADCLEC.syn1 T cells efficiently co-engage CAR and CCR to enhance killing of dual antigen-positive cells, while separate CAR and CCR engagement via single antigen-positive cells in close proximity does not result in unintended cytotoxicity.

Having established proof-of-principle evidence that ADCLEC.syn1 operates as IF-BETTER gate, we next tested its application in an AML model of CAR target antigen escape. For this, we created stable MOLM13 cell line variants with unmodified or lowered ADGRE2 antigen densities: M13-A^high (WT, 1×10⁴ ADGRE2 molecules), M13-A^low (1×10³ ADGRE2 molecules) and M13-A^very-low (4×10² ADGRE2 molecules), all endogenously expressing low CLEC12A levels (C^low, 5×10² molecules) (Figures 2E, S4A). In the context of high ADGRE2 levels (1×10⁴ molecules), T cells expressing only the ADGRE2-CAR-28z1XX had high MOLM13 killing efficacy with no difference compared to T cells expressing ADCLEC.syn1. However, in the context of low levels of ADGRE2 (1×10³ molecules) and CLEC12A (5×10² molecules), T cells expressing only the ADGRE2-CAR-28z1XX failed to kill, whereas ADCLEC.syn1 T cells maintained killing capacity, demonstrating in vitro superiority of ADCLEC.syn1 efficacy over a single ADGRE2-targeted CAR (p=0.0011) (Figures 2F).

To elucidate the functional role of the 4–1BB endodomain of a CCR assisting CAR-1XX-mediated killing in an antigen-low setting, we engineered ADGRE2-CAR-28z1XX-CLEC12A-CCR-del which included a non-signaling truncated CCR but was otherwise identical to ADCLEC.syn1. We compared the efficacy of ADCLEC.syn1 vs ADGRE2-CAR-28z1XX-CLEC12A-CCR-del in a NOD scid gamma (NSG) xenograft model based on a MOLM13 variant expressing low ADGRE2 (1×10³ molecules/cell) and high CLEC12A (2×10⁵ molecules/cell) antigen density to ensure potential CCR effects (Figures 2G, S4 B-C); lower CCR target levels were explored in subsequent experiments. At a CAR T cell stress dose of only 5.0×10⁵ CAR T cells, ADCLEC.syn1 outperformed not only a single ADGRE2-CAR-28z1XX (p=0.0017), but also the ADGRE2-CAR-28z1XX-CLEC12A-CCR-del (p=0.0330), demonstrating superior in vivo efficacy of ADCLEC.syn1 over a 28z1XX-CAR alone, owing to 4–1BB costimulation provided by the CCR (Figures 2H). In a more extreme setting of detectable but very-low ADGRE2 (4×10² molecules/cell) and high CLEC12A (1×10⁵ molecules/cell), ADCLEC.syn1 again outperformed both ADGRE2-CAR-28z1XX (p=0.0023) and ADGRE2-CAR-28z1XX-CLEC12A-CCR-del (p=0.0035), demonstrating that a CLEC12A-CCR-BB selectively enhances ADGRE2–28z1XX sensitivity, even at extremely low CAR target levels (<5×10² molecules/cell) (Figure 2I-J).

CCR engagement regulates cytolysis directed to low antigen density

To test the limits of CCR-gated sensitivity of ADCLEC.syn1, we investigated additional settings where either both CAR and CCR targets are expressed at low density (modeling AML antigen escape), or where the CAR target is expressed at low density while the CCR target is negative (modeling normal tissues). For this, we utilized MOLM13 cell line variants with either high (1×10⁴ molecules) or low (1×10³ molecules) ADGRE2 antigen density and either endogenously low CLEC12A density (5×10² molecules) or CLEC12A-negativity (KO) (Figures 3A,C,E,G, S4D). In the setting of high ADGRE2 expression (1×10⁴ molecules) representing AML, ADCLEC.syn1 led to long-term remission in all mice regardless of CLEC12A co-expression (Figure 3A-D,I), demonstrating that CCR co-engagement is not required to eliminate AML cells with high CAR target density. In the setting of low ADGRE2 expression (1×10³ molecules), CCR co-engagement at low levels of CLEC12A (5×10² molecules) was sufficient for ADCLEC.syn1 to enable durable AML reduction and long-term survival, whereas cells with low ADGRE2 expression (1×10³ molecules) and absence of CLEC12A expression, representing normal cells, were not completely eradicated (p=0.0063, Figure 3E-J). Collectively, these results demonstrate the potential of a CCR to regulate cytolytic activity in settings of low CAR and CCR target antigen densities, resulting in a more selective increase of CAR-1XX antitumor efficacy.

Figure 3. CCR engagement regulates cytolysis directed to low antigen density.

(A,C,E,G) MOLM13 variants with modified ADGRE2 (high vs low) and CLEC12A (low vs KO) levels: M13-A^high-C^low, M13-A^high-C^KO, M13-A^low-C^low, M13-A^low-C^KO. Bivariate plots illustrate how the respective ADGRE2/CLEC12A antigen densities (absolute number of surface molecules/cell) compare to the analyzed AML patient cohort (n=39) and their AML bulk (brown) as well as LSC (red) populations.

(B,D,F,H) AML burden (total flux) and Kaplan-Meier survival analysis of mice bearing MOLM13 variants treated with 5×10⁵ ADCLEC.syn1 T cells.

(I) AML burden (total flux) in week 2 and week 3 post ADCLEC.syn1 T cell injection in mice bearing MOLM13 variants. Data are represented as individual measurements and geometric mean with geometric SD. p values were determined via unpaired t test.

(J) Kaplan-Meier survival of mice bearing either M13-A^low-C^low or M13-A^low-C^KO.

Also see Figure S4.

Low-dose ADCLEC.syn1 efficiently ablates AML with effective recall responses

The ADGRE2 and CLEC12A target antigen quantification in our AML patient cohort allowed us to select AML cell line variants with relevant combinations of high or low ADGRE2 and CLEC12A antigen densities to represent the spectrum of phenotypes observed in the AML patient cohort (Figure 4A,C,E,G). In a xenograft model of ADGRE2^high MOLM13 AML, both a single ADGRE2-CAR-28z1XX and ADCLEC.syn1 induced rapid remissions (Figure 4B). In ADGRE2^low MOLM13 and U937 AML however, only ADCLEC.syn1 maintained its efficacy, whereas the efficacy of a single-targeted ADGRE2-CAR-28z1XX was limited (Figure 4D,F,H). Significantly, the benefit afforded by the CLEC12A-CCR was observed at both CLEC12A^low (5×10² molecules/cell) and CLEC12A^high levels (2×10⁵ molecules/cell), demonstrating functional impact across a range of AML phenotypes. In a stress test model of mice bearing ADGRE2^high(1×10⁴)/CLEC12A^high(2×10⁴) AML, a dose as low as 5×10⁴ CAR T cells was sufficient to induce remissions and long-term survival, further highlighting the in vivo potency of ADCLEC.syn1 T cells (Figure S5A). To assess the functional persistence of ADCLEC.syn1 T cells, we repeatedly challenged mice with MOLM13 injections two months after a single initial injection of only 1–2.5×10⁵ ADCLEC.syn1 T cells, using either MOLM13-ADGRE2^high(1×10⁴)/CLEC12A^low(5×10²) or MOLM13-ADGRE2^high(1×10⁴)/CLEC12A^high(2×10⁴) (Figure 4I-L). All mice re-challenged with ADGRE2^high(1×10⁴) MOLM13 formed effective recall responses, successfully averting relapse. Re-challenge with antigen-escape control ADGRE2^very-low(4×10²)/CLEC12A^low(5×10²) MOLM13 (Figure S5B) led to rapid relapse, thereby ruling out a potential allogeneic graft-versus-leukemia (GVL) effect and re-confirming the threshold antigen density required for reactivating persisting ADCLEC.syn1 T cells.

Figure 4. Low-dose ADCLEC.syn1 efficiently ablates AML with effective recall responses.

(A,C,E,G,I,K) AML cell lines with modified ADGRE2 and CLEC12A densities used for xenograft CAR studies. Bivariate plots illustrate how the respective ADGRE2/CLEC12A antigen densities (absolute numbers of surface molecules/cell) compare to the analyzed AML patient cohort (n=39) and their AML bulk (brown) as well as LSC (red) populations.

(B,D,F,H) AML burden (total flux) and Kaplan-Meier survival analysis of mice bearing MOLM13 variants treated with 5×10⁵ (B,D,F) or 1×10⁶ (H) CAR/CCR^pos T cells, comparing ADCLEC.syn1 vs its single receptor components (ADGRE2–28z1XX CAR or CLEC12A-BB CCR) vs untransduced T cells (UTD).

(J,L) AML burden (total flux) of mice bearing M13-A^high-C^low (J) or M13-A^high-C^high (L) treated with UTD or ADCLEC.syn1 T cells at the indicated dose (1–2.5×10⁵). Arrows indicate repeated MOLM13 re-challenges (dose 5×10⁵, on d66 and d73 post initial CAR T injection) with either the same MOLM13 variant as on d0 (left graphs) or an antigen escape control variant M13-A^very-low-C^low (right graphs) with minimal target levels (Figure S5B) expected to cause AML relapse.

Also see Figure S5.

Prior research has demonstrated that AML cells may induce T cell dysfunction by impairing T cell activation and leading to their senescence.^48,49 We therefore evaluated if an autologous approach to generating ADCLEC.syn1 T cells may be limited by AML-induced T cell dysfunction. To this end, we isolated T cells from PB of an 88-year-old female AML patient with high AML burden (18% blasts in PB manual differential at time of venipuncture for T cell collection). Following T cell isolation and a 7-day T cell transduction and expansion protocol, end-of-production ADCLEC.syn1 T cells were cryopreserved to model the clinical setting. Thawed AML patient-derived ADCLEC.syn1 T cells were then administered to NSG mice engrafted with MOLM13 (WT) AML cells. AML patient-derived ADCLEC.syn1 T cells induced rapid and durable remissions in all mice even at a CAR T cell stress dose of only 2.5×10⁵ CAR T cells per mouse, a dose that is within a 5-fold range of the minimum efficacious dose of fresh, healthy donor-derived ADCLEC.syn1 CAR T cells (Figure S5C). Overall, these results support that T cells derived from AML patients can be a viable source for production of ADCLEC.syn1 CAR T cells with high in vivo potency.

ADCLEC.syn1 eliminates leukemic stem cells in heterogenous AML PDX models

To assess efficacy of ADCLEC.syn1 in the setting of genotypically and phenotypically heterogenous AML, we selected three independent AML PDX models originating from r/r AML patients with prior failure of multiple AML-directed therapies (Figure S6A-D).

PDX#1 originated from an AML patient relapsing after chemotherapy, hypomethylating agent and allo-HSCT. Target antigen density of ADGRE2 and CLEC12A was representative of profiles observed in our studied AML patient cohort and more than 10-fold lower than CD33 (Figure 5A-B). In this PDX model, we evaluated the potential of CAR T cells to eliminate AML LSCs defined by their re-engraftment potential, and chose to compare a reference CD33-targeted CAR-28z1XX (33–28z1XX) against ADCLEC.syn1 (Figure 5C). CD33 was chosen as reference target due to its reported highest percentage of positivity in AML among the most commonly investigated CAR target antigens.⁵⁰ In a primary mouse cohort, PDX#1-engrafted mice were treated with either ADCLEC.syn1 or 33–28z1XX at a stress dose of 5×10⁵ CAR T cells. 33–28z1XX T cells expanded similarly to ADCLEC.syn1 T cells, but failed to completely eliminate AML cells in PB and bone marrow (Figures 5D and S6E). Persistence of AML LSC post 33–28z1XX T cells was evidenced by serial transplantation of relapsing AML cells leading to AML PDX engraftment in all mice in a secondary mouse cohort (Figure 5E, S6F). While 33–28z1XX T cells failed in both cohorts, ADCLEC.syn1 T cells induced remissions, both in the primary cohort and the secondary cohort of mice with relapsed AML post-33–28z1XX therapy (Figures 5D, 5E, S6G-H). In two additional AML PDX models with adverse risk profile and patient-representative target levels of ADGRE2 and CLEC12A, ADCLEC.syn1 induced durable AML remission and long-term survival at a stress dose of 5×10⁵ CAR T cells (Figure 5F-I).

Figure 5. ADCLEC.syn1 eliminates leukemic stem cells in heterogenous AML PDX models.

Three relapsed/refractory AML PDX models (PDX#1 in A-E, PDX#2 in F-G, PDX#3 in H-I) were utilized to assess CAR T cell efficacy (PDX clinical annotations and target phenotype in Figure S6A-D). T cell expansion and AML PDX burden were serially monitored via flow cytometry of PB and are shown as normalized T cell and AML cell counts per 100ul PB. Survival is shown as Kaplan-Meier analysis.

(A,F,H) ADGRE2/CLEC12A antigen densities (absolute number of surface molecules/cell) of PDX LSCs compared to the analyzed AML patient cohort (n=39) and their AML bulk (brown) as well as LSC (red) populations.

(B) ADGRE2/CLEC12A compared to CD33 antigen densities (absolute number of surface molecules/cell) on PDX#1.

(C) Schematic of PDX#1 experimental setup for results shown in D-E

(D) Primary engraftment of PDX#1 in NSG mice on d-11 was followed by treatment with 5×10⁵ CAR/CCR^pos T cells on d0, comparing UTD vs a reference CD33-CAR (33–28z1XX) vs ADCLEC.syn1. Mice receiving the reference CD33-CAR relapsed, and their bone marrow was harvested on d29 for subsequent secondary engraftment in NSG mice on d-16.

(E) PDX#1 post-33–28z1XX failure was secondarily engrafted in NSG mice on d-16 and was followed by treatment with 2.5×10⁵ CAR/CCR^pos T cells on d0.

(G) Engraftment of PDX#2 in NSG-SGM3 mice on d-9 was followed by treatment with 5×10⁵ CAR/CCR^pos T cells on d0.

(I) Engraftment of PDX#3 in NSG-SGM3 mice on d-18 was followed by treatment with 5×10⁵ CAR/CCR^pos T cells on d0.

Also see Figure S6.

Humanized AML mouse model to assess ADCLEC.syn1 efficacy and HSPC toxicity

We next evaluated the on-target/off-tumor activity of ADCLEC.syn1 on normal hematologic cells in the context of an in vivo anti-leukemic CAR T cell response. For this, we established a humanized AML xenograft mouse model to assess in vivo CAR T cell efficacy and hematotoxicity simultaneously in the same mouse. A CD19-targeted CAR (19–28z1XX) was selected as positive control to benchmark expected on-target activity resulting in the depletion of CD19^pos tumor cells and normal human CD19^pos B cells. Furthermore, 19–28z1XX served as an important negative control for on-target activity against normal HSPCs as CD19 is not expressed in HSPCs. G-CSF-mobilized adult healthy donor-derived CD34^pos HSPCs were used for humanization of NSG mice, followed by engraftment of MOLM13 AML cells that had been modified to express CD19 at levels similar to NALM6 B-ALL cells (2.7×10⁴ molecules per cell⁵¹). Humanized MOLM13-CD19^pos AML-engrafted mice received either control 19–28z1XX or ADCLEC.syn1 CAR T cells at a low dose of 2.5×10⁵ CAR T cells (Figure 6A). Both 19–28z1XX and ADCLEC.syn1 led to rapid AML remission in humanized mice; untreated humanized mice did not control AML (Figure 6B). The similar anti-leukemic efficacy of 19–28z1XX and ADCLEC.syn1 allowed to directly assess and compare their respective impact on normal hematologic cells. On day 7 post CAR T cell injection, bone marrow analysis confirmed complete MOLM13-CD19^pos AML eradication in mice treated with 19–28z1XX or ADCLEC.syn1 (Figure 6C), consistent with overall bioluminescence results. We found complete B cell depletion only in 19–28z1XX-treated mice, and partial monocyte depletion only in ADCLEC.syn1-treated mice (Figure 6D-E), indicating that both CAR T cell populations were active in these humanized bone marrows. In the same samples, normal human CD34^pos cells were detected at similar frequency in mice treated with 19–28z1XX or ADCLEC.syn1, showing the same reduction with both CAR treatments relative to untreated mice (Figure 6F). This experimental setting thus did not reveal on-target/off-tumor activity directed to CD34^pos cells but suggested HSPC reduction owing to a target-independent mechanism. These observations were further supported by absolute cell counts (Figure 6G): At time of complete AML eradication and similar CAR T cell expansion, ADCLEC.syn1-treated mice had persistent CD34^pos cells at similar absolute numbers compared to 19–28z1XX, demonstrating that high anti-leukemic efficacy of ADCLEC.syn1 occurred without any more HSPC toxicity than a clinically validated CD19-targeted CAR.

Figure 6. Humanized AML mouse model to assess ADCLEC.syn1 efficacy and HSPC toxicity.

Humanized AML xenograft mouse model to assess CAR T cell hematotoxicity in the context of an in vivo anti-leukemic CAR T cell response. Anti-leukemic response and impact on normal human hematopoiesis was assessed upon receiving either no treatment or treatment with a reference CD19-CAR (19–28z1XX) vs ADCLEC.syn1.

(A) Schematic of humanized AML xenograft CAR T cell hematotoxicity model. NSG mice were humanized via sublethal irradiation and injection of G-CSF-mobilized healthy donor-derived CD34^pos cells on d-21, followed by MOLM13-CD19^pos cell line injection on d-3, either untreated or treated with 2.5×10⁵ CAR/CCR^pos T cells on d0.

(B) AML burden (total flux) of humanized mice bearing MOLM13-CD19^pos AML.

(C-F) Representative ex vivo bone marrow distribution of AML and normal human hematopoietic cells on d7.

(C) MOLM13-CD19^pos AML cells identified via positivity for CD33 and CD19 within total human CD45^pos cells

(D) CD19^pos normal B cells and CD3^pos adoptively transferred CAR T cells within total human CD45^pos cells (excluding CD33^pos/CD19^pos MOLM13)

(E) CD14^pos/CD16^neg normal classical monocytes within CD3^neg/CD19^neg human CD45^pos cells

(F) CD34^pos normal HSPCs within lineage-negative (CD3^neg/CD19^neg/CD14^neg/CD16^neg) human CD45^pos cells

(G) D7 ex vivo quantification of human bone marrow populations, with n=4–5 mice per group. Data are shown as individual counts and geometric mean with geometric SD. p values were determined via Mann-Whitney test.

Off-target hematotoxicity is mitigated by reducing CAR T cell-derived IFN-γ

To further investigate the potential mechanism of off-target (antigen-independent) HSPC toxicity, we tested in vitro and in vivo whether CAR T cell-derived products may account for HSPC reduction. Since IFN-γ has been reported to mediate differentiation and depletion of myeloid-biased HSCs in murine models of chronic inflammation⁵², we hypothesized that excessive IFN-γ release in the context of CAR T cell anti-tumor activity could trigger HSC loss due to rapid differentiation. In an in vitro model of T cell-derived IFN-γ off-target hematotoxicity, we first co-cultured ADCLEC.syn1 T cells with MOLM13 target cells, leading to CAR activation (Figure 7A, upper panel). Cell-free supernatant (CAR T + AML-conditioned medium) was then added to healthy donor-derived HSPCs (Figure 7A, lower panel). HSPC phenotype and absolute cell counts were determined after 24h. The potential effect of CAR T cell-derived IFN-γ was evaluated by either IFNG CRISPR/Cas9 T cell editing or addition of anti-IFN-γ blocking antibody to the conditioned medium. We found that addition of conditioned medium from ADCLEC.syn1 IFNG-WT + MOLM13 co-culture led to a striking reduction of CD34^pos/CD38^neg immature HSPCs while CD34^pos/CD38^pos progenitors increased, both in absolute and relative numbers (Figure 7B). This was not observed when adding conditioned medium from either MOLM13 alone or ADCLEC.syn1 IFNG-WT alone. Genetic disruption of IFNG in ADCLEC.syn1 T cells or blocking of IFN-γ protein was associated with a significant increase of CD34^pos/CD38^neg immature HSPCs and decrease of CD34^pos/CD38^pos progenitors in absolute and relative numbers, compared to HSPCs cultured with conditioned medium from ADCLEC.syn1 IFNG-WT + MOLM13 co-culture.

Figure 7. Off-target hematotoxicity is mitigated by reducing CAR T cell-derived IFN-γ.

(A) In vitro assay to evaluate off-target HSPC toxicity due to soluble factors released upon CAR T cells engaging target cells. ADCLEC.syn1 CAR T cells with or without IFNG CRISPR/Cas9 editing were co-cultured with MOLM13-WT AML cell line for 10h at E:T ratio 1:1. Subsequently, cell-free supernatant (conditioned medium from CAR-T + AML-co-culture and individual controls) was collected and added to a separate in vitro culture of normal human CD34^pos/CD38^neg HSPC, with or without anti-IFN-γ blocking antibody. After 24h, HSPC phenotype (CD34/CD38 expression) and cell counts were measured via flow cytometry.

(B) Absolute cell count and relative distribution of HSPC subsets upon in vitro culture with different CAR T + AML-conditioned media. Horizontal and error bars represent mean value and SD of technical triplicates. p values were determined via unpaired t test. FACS plots show representative CD34/CD38 HSPC phenotypes at time of assay read-out.

(C) Schematic of in vivo model of CAR T cell IFN-γ-mediated off-target hematotoxicity. NSG mice were humanized via sublethal irradiation and injection of G-CSF-mobilized healthy donor-derived CD34^pos cells on d-21 (dose 6.0×10⁵), followed by injection of 19–28z1XX or ADCLEC.syn1 CAR T cells derived from the same donor (dose 3.0×10⁵ CAR/CCR^pos T cells, ± IFNG editing) on d0.

(D) IFNG editing of 19–28z1XX and ADCLEC.syn1 T cells for in vivo study was demonstrated by intracellular IFN-γ staining 10h after in vitro culture with or without MOLM13-CD19^pos target cells at E:T ratio 1:1.

(E) D7 ex vivo quantification of human bone marrow populations (lineage-negative CD34^pos/CD38^neg HSPCs, CD19^pos B cells and CD3^pos CAR T cells), with n=5 mice per group. Data are shown as individual counts and geometric mean with geometric SD. p values were determined via Mann-Whitney test.

Also see Figure S7.

Overall, these in vitro observations suggested that soluble factors released upon CAR T cell activation may lead to reduction of immature HSPCs potentially owing to rapid differentiation, and that CAR T cell-derived IFN-γ is a direct mediator of this effect. To investigate whether this effect may explain our observed in vivo reduction of HSPCs in the context of CAR T cell activation, we treated humanized mice with 19–28z1XX CAR or ADCLEC.syn1 T cells with or without IFNG editing, and compared human engrafted cell counts in bone marrow (Figure 7C-D). Consistent with our previous findings, CAR T cell expansion was similar between 19–28z1XX and ADCLEC.syn1, and only 19–28z1XX led to B cell depletion (Figure 7E). Of note, IFNG editing was not associated with any changes in CAR T cell expansion or B cell counts. The partial HSPC reduction associated with 19–28z1XX and ADCLEC.syn1 T cells (IFNG-WT) was not observed when reducing IFN-γ expression (IFNG-KO), even though on-target CAR T cell activity was maintained (Figures 7E, S7).

Discussion

We provide here a quantitative analysis of target expression in r/r AML, based on which we devised a tethered combinatorial targeting strategy for a CAR T cell therapy. Specifically, we address the challenge of achieving selective tumor targeting while limiting on-target activity against tumor-related normal tissues. We demonstrate that a gated, combinatorial chimeric receptor design can achieve anti-tumor efficacy with limited on-target/off-tumor toxicity to tumor-related normal tissue by following below principles: (1) selection of targets with frequent expression in malignant cells (including cancer stem cells where applicable) and limited expression in vital normal tissues; (2) accurate quantification of target antigen densities that distinguish tumor cells from critical normal cells; (3) design of cooperative receptors that afford activation thresholds adapted to target expression signatures that part tumor from normal cells.

There is a growing number of available chimeric receptor designs that require different target antigen densities to elicit T cell activation, ranging from a few hundred molecules per cell for HIT receptors⁵¹ to a few thousand molecules per cell for 4–1BB/CD3z CARs.^37,53,54 The rational selection of an adapted CAR design thus warrants precise measurement of target antigen densities in tumor and normal cells. This is particularly important in the context of combinatorial receptor formats that aim to skew T cell engagement towards tumor cells and away from normal cells. Here, we report a novel combinatorial CAR approach to target AML, informed by target quantification in AML and normal cells. To select a suitable target combination, we performed quantitative profiling of the target candidates ADGRE2, CD33, CD123 and CLEC12A, based on flow cytometry in a cohort of r/r AML patients (n=39) with diverse AML disease subtypes including AML with recurrent genetic abnormalities, mutated TP53 and myelodysplasia-related abnormalities,⁵⁵ and healthy donors (n=8). We also evaluated the expression of these targets in normal non-hematopoietic tissues using a cell type-specific RNA expression dataset.⁵⁶ For each target antigen, we quantified the number of surface molecules in AML bulk, LSC and normal hematopoietic cells. This allowed us to identify a target combination with differential expression in AML LSCs and normal HSPCs.

While all four targets were abundantly expressed in most AML samples, their different expression patterns in normal tissues informed our target selection. High CD33 levels in normal HSPCs (up to 3×10³), granulocytes (4×10³) and monocytes (2.4×10⁴), and CD123 expression in normal HSPCs (2×10³) and endothelium⁵⁷ complicated the use of differential antigen expression to selectively target CD33 or CD123 in AML. ADGRE2 expression in normal tissues was found to be more restricted, with less than 9×10² molecules per cell on HSC and negativity in non-hematopoietic cell types, allowing us to devise an ADGRE2-targeted CAR strategy with a threshold of activation of approximately 1×10³ molecules per cell. CLEC12A expression was found to be absent in early HSPCs and non-hematopoietic cell types, but its high antigen density in normal granulocytes (6.5×10³) would make those cells immediate CAR targets. We therefore chose to target ADGRE2 with a CAR and co-engage CLEC12A with a CCR instead of a CAR, to selectively enhance killing of ADGRE2^pos CLEC12A^pos AML cells without killing ADGRE2^neg CLEC12A^pos normal cells.

We utilize here a logic-gated targeting approach that we term IF-BETTER gating: CAR target sensitivity is enhanced by CCR target co-engagement, thereby expanding the range of overall targetable cells. This combinatorial CAR approach can be achieved with a variety of different configurations of combined CAR and CCR expression. Its ultimate effect on tumor and normal cells depends on target choice as well as thresholds of activation for CAR and CCR. Our target density profiling in AML and normal cells established that an ideal IF-BETTER gated CAR design targeting ADGRE2 and CLEC12A would trigger killing of any cells with more than 1×10³ ADGRE2 molecules while cells with at least 5×10² ADGRE2 molecules would still be efficiently eradicated if CLEC12A is co-expressed (Figure 8A-B). To achieve this, we developed ADCLEC.syn1, which consists in an ADGRE2-targeted CAR-28z1XX co-expressed with a CLEC12A-targeted CCR providing 4–1BB costimulation. To mitigate the risk of ablating ADGRE2^low HSPCs, we selected an ADGRE2 scFv with limited target sensitivity in the context of a CD3z ITAM-reduced 1XX format.⁴⁷ In vitro, deletion of the two distal CD3z ITAMs has been associated with reduced cytotoxicity at low CD19 target antigen density of 1×10³ molecules per cell.⁵³ The 1XX format has been previously shown to increase functional CAR T cell persistence while limiting effector functions including IFN-γ production relative to CARs retaining 3 functional ITAMs.^47,58,59 For the CCR, a 4–1BB costimulatory domain was selected, building on previous evidence of synergy of combined CD28 and 4–1BB costimulation in the context of endogenous T cell responses⁶⁰ as well as CAR T cells.^61–64

Figure 8. ADCLEC.syn1 enhances distinction between AML and normal cells based on combined target signatures.

(A) Chimeric receptor architecture of single CAR (ADGRE2-CAR) vs IF-BETTER gated CAR+CCR (ADCLEC.syn1).

(B) Schematic outlining CAR T cell activity depending on target densities on AML and normal cells: ADGRE2-CAR kills only ADGRE2^high AML cells but fails to kill ADGRE2^low AML cells; ADCLEC.syn1 kills both ADGRE2^high and ADGRE2^low/CLEC12A^pos AML cells while sparing ADGRE2^low/CLEC12A^neg normal cells

(C) Summary of in vivo activity of ADGRE2–28z1XX-CAR vs ADCLEC.syn1 against AML cell lines or PDX with target antigen densities as shown. Dashed line delineates in vivo target cell killing vs sparing as observed in experiments shown in Figures 2–5.

(D) Projection of line for ADGRE2/CLEC12A in vivo killing threshold onto primary AML target phenotypes from r/r AML patient cohort.

(E) Projection of line for ADGRE2/CLEC12A in vivo killing threshold onto normal hematopoietic cells

We provide experimental evidence that ADCLEC.syn1 operates with killing thresholds that are tuned to allow target killing based on combinatorial antigen density profiles rather than single-antigen expression (Figure 8A-B). In a series of AML cell line xenograft and PDX experiments we applied in vivo stress tests in terms of targeting ADGRE2 at reduced antigen densities, comparing a single CAR vs IF-BETTER gated ADCLEC.syn1. A conventional ADGRE2-CAR-28z1XX failed to eliminate target cells with only 1×10³ or 5×10² ADGRE2 molecules (MOLM13-ADGRE2^low/CLEC12A^low/high and U937, respectively) while ADCLEC.syn1 enabled efficient in vivo elimination of the same target cells owing to CLEC12A-CCR engagement (Figure 8C). We demonstrated that the ADGRE2 targeting sensitivity of ADCLEC.syn1 is purposefully gated by the coopted CLEC12A-CCR, allowing to spare MOLM13-ADGRE2^low/CLEC12A^neg target cells with 1×10³ ADGRE2 molecules and CLEC12A negativity, which represent normal cells, addressing an important safety requirement. On the other hand, MOLM13-ADGRE2^low/CLEC12A^low target cells with 1×10³ ADGRE2 and 5×10² CLEC12A molecules were efficiently eradicated, highlighting that even a low CCR target density of only 5×10² molecules can determine the susceptibility to cytolysis of target cells.

Overall, we demonstrate an in vivo killing pattern of ADCLEC.syn1 that reflects a two-dimensional antigen density threshold for target cell lysis, where CCR engagement determines the CAR sensitivity and therefore killing fate of CAR target-low cells (Figure 8C). The IF-BETTER gated killing thresholds translate into increased potential for complete tumor eradication at limiting CAR target densities (1×10³ or fewer molecules) (Figure 8D) while sparing models for vital normal cells (Figure 8C) and normal HSPCs (Figure 8E).

Previous studies have reported on different uses of CCRs to enhance second-generation CARs: Muliaditan et al. used a 4–1BB CCR to enhance functional persistence of 28z CAR T cells in a breast cancer model.⁶³ Katsarou et al. compared different CD38-CCR formats to enhance efficacy of BCMA- or CD19-targeted 28z-CARs in multiple myeloma and B cell precursor acute lymphoblastic leukemia (B-ALL), respectively. In a Nalm6 B-ALL model of CD19 antigen escape, a CD19-CAR-28z+CD38-CCR-28BB design with shared CD28 transmembrane domains induced remissions in mice.⁶⁴ Hirabayashi et al. reported that CAR and CCR sharing the same transmembrane domain enable CAR/CCR heterodimerization, resulting in CAR activation by the CCR target thus acting as OR-gated killing in settings where on-target toxicity is not a concern.⁶⁵ Other logic-gated CAR approaches include OR gates based on dual or tandem CARs,^{37–39,41,43} AND gates based on two individually dysfunctional receptors,^44,45 and NOT gates incorporating inhibitory signaling.^66,67 Preference for a logic-gate will depend on the malignancy and expression profiles of target antigens. Further investigation of these alternate gates with clinically relevant ranges of target antigen densities is required to determine their relative antigen sensitivities^68,69, as we have done here. In previously reported studies, the CAR+CCR design was investigated to maximize anti-tumor activity by increasing avidity and combining costimulation to promote T cell persistence.^63,64 The present study investigates a different use of a CCR: to amplify a therapeutic window based on antigen density differences so as to skew CAR activity and cytolysis towards tumor cells and away from closely related normal cells.

We demonstrate high in vivo antileukemic efficacy of ADCLEC.syn1 in several AML cell line xenograft models using stress test CAR T cell doses as low as 5×10⁴, matching clinically feasible dosing if normalized to body weight.⁷⁰ Furthermore, we demonstrate efficacy of ADCLEC.syn1 in AML of different phenotypes and genomic background. ADCLEC.syn1 induced durable remissions in xenograft models based on MOLM13 variants representing a wide spectrum of ADGRE2 and CLEC12A antigen densities as well as U937, which endogenously represents ADGRE2^lowCLEC12A^high AML. Of note, ADCLEC.syn1 induced durable remissions in three molecularly distinct AML PDX models. PDX#1 is derived from a patient with KMT2A-rearranged r/r AML and has been reported as highly aggressive AML model that only partially responded to targeted therapy.^71–73 In mice engrafted with PDX#1, ADCLEC.syn1 induced durable remissions and outperformed a reference CD33-CAR-28z1XX. We chose CD33 as reference target due to its known abundant expression in AML including PDX#1, and its wide use as immunotarget in AML, e.g. with the FDA-approved antibody-drug-conjugate Gemtuzumab ozogamicin.⁷⁴

On-target hematotoxicity is one of the key concerns for CAR therapy in AML. Pre-clinical models for CAR hematotoxicity are technically challenging but invaluable to guide safe CAR development, using humanized mouse models to study CAR hematotoxicity in the setting of an in vivo antitumor CAR T cell response. Kim et al. demonstrated the myeloablative potential of CD33-targeted CAR therapy leading to complete depletion of the CD34^pos HSPC compartment, which required genetic removal of CD33 from normal human hematopoiesis to avoid on-target hematotoxicity.³⁰ Sugita et al. evaluated hematotoxicity of CD123-targeted CAR therapy and reported ablation of normal myeloid cells in mice treated with CD123-CAR T cells.⁷⁵ Both pre-clinical studies and the emerging clinical evidence of myeloablation in patients responding to CD33-CAR and CD123-CAR therapies²⁷ underline the potential predictive value of pre-clinical in vivo hematotoxicity models. Other studies reported on alternative CAR approaches targeting GRP78⁷⁶ or SIGLEC6,⁷⁷ demonstrating promising results with limited in vitro hematotoxicity. The in vivo hematotoxicity in the context of GRP78- or SIGLEC6targeted CAR T cell responses remains to be investigated.

In our work presented here, we used normal CD34^pos cells to engraft human hematopoietic cells in NSG mice, which were then co-engrafted with AML cells to assess CAR T cell efficacy and toxicity in the context of normal human bystander cells and an anti-leukemic CAR T cell response. In this model, we compared ADCLEC.syn1 to CD19-CAR-28z1XX as control for on-target toxicity. CD19 is clinically validated as CAR target with a B cell-restricted and overall acceptable on-target hematologic toxicity profile in the context of 28z, BBz or 28z1XX CAR formats.^5,78–81 Therefore, comparing 19–28z1XX vs ADCLEC.syn1 allows to benchmark the on-target HSPC toxicity profile in the context of a potent in vivo anti-leukemic CAR T cell response. We found that ADCLEC.syn1 led to efficient eradication of AML cells without reducing normal HSPCs any more than 19–28z1XX. Notably, efficacy of ADCLEC.syn1 was maintained in a bone marrow environment with proximate normal myelopoiesis which may limit efficacy of CARs targeting antigens of higher density on more frequent normal cells, i.e. due to potential competition for CAR engagement. Consistent with our observed target profiles and independent reports of ADGRE2 and CLEC12A antigen expression,^13,18,82 on-target hematotoxicity of ADCLEC.syn1 was limited to a reduction of normal monocytes. Of note, the clinical benchmark, 19–28z1XX, was associated with a reduction of HSPC compared to untreated mice, suggesting antigen-independent (off-target) HSPC toxicity. Clinically, off-target hematotoxicity is increasingly recognized in patients treated with CD19- or BCMA-targeted FDA-approved CAR T cell therapies. A correlation with CRS severity and pro-inflammatory cytokine levels has been established while the pathophysiology remains to be elucidated.^83–87

We describe here a humanized mouse model in which both on-target and off-target toxicities could be studied, and demonstrate in vivo for both 19–28z1XX and ADCLEC.syn1 that CAR T cell-derived IFN-γ is an important mediator of off-target hematotoxicity. By diminishing IFN-γ expression in CAR T cells, we were able to minimize in vivo off-target hematotoxicity of T cells expressing either a CD19-CAR or ADCLEC.syn1. We thereby provide functional evidence for mitigating off-target CAR hematotoxicity through T cell engineering for limited release of IFN-γ. Off-target hematotoxicity or Immune Effector Cell-Associated Hematotoxicity (ICAHT) has been recognized as one of the most common adverse immune effector cell-associated toxicities, and is distinct from other common toxicities such as cytokine release syndrome (CRS) and immune effector cell-associated neurotoxicity syndrome (ICANS).⁸⁸ We identify here IFN-γ as a directly actionable mediator of ICAHT, warranting further investigation of reducing IFN-γ to alleviate ICAHT and long-term cytopenias in addition to CRS.^89,90

Our study highlights that cooperative CAR design can be adapted to target expression profiles, requiring quantification of surface target molecules in tumor and normal tissues to identify therapeutic windows. We rationally co-target ADGRE2 and CLEC12A with a CAR design termed ADCLEC.syn1, and demonstrate its potential for selective anti-leukemic activity in stringent and controlled pre-clinical models of CAR T cell efficacy and hematotoxicity. We further demonstrate the potential of controlling IFN-γ production to mitigate off-target immune effector cell-associated hematotoxicity (ICAHT) in CAR therapies including CD19-CARs. ADCLEC.syn1 may thus preempt the need for post-CAR T cell allo-HSCT rescue and allow for potential use as a standalone therapy including in consolidation or for treatment of post-transplant AML relapse. A phase 1 clinical trial evaluating ADCLEC.syn1 T cells in patients with r/r AML (NCT05748197) is slated to open in 2023.

STAR METHODS

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Michel Sadelain (m-sadelain@ski.mskcc.org).

Materials Availability

Plasmid requests can be directed to the lead contact.

Data and Code Availability

This paper does not report standardized datatypes or original code.

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

AML patient and normal donor samples

AML patient samples were collected as peripheral blood or bone marrow aspirates from r/r AML patients at MSKCC, after receiving informed consent under IRB-approved protocols (#06–107 and #14–091). Unlike in previous studies^13,18, the present analysis focused exclusively on r/r AML to represent a potential phase 1 CAR T cell trial patient population. Normal donor whole bone marrow aspirates were purchased from AllCells. AML and normal donor samples were received and stored in fresh and EDTA-anticoagulated condition on a tube rocker at room temperature and were processed for flow cytometric analysis within 24h. Samples were filtered through a 70μm cell strainer and red blood cell lysis was performed with a subsequent washing step and Fc receptor blocking before antibody staining per standard flow cytometry protocol. Clinically available patient information on AML burden, cytogenetic and molecular aberrations was used to identify relapsed/refractory cases and their AML subtype according to the ELN 2022 classification.⁵⁵

Cell lines

Unmodified MOLM13 AML cell line was received as gift from Dr. Marion Subklewe (LMU Munich). Unmodified U937 AML cell line was obtained from ATCC. Both AML cell lines were then transduced with FFLuc-GFP-expressing retroviral vector and subsequently sorted for GFP^pos fraction. MOLM13 and U937 cells were cultured in RPMI-1640 medium supplemented with 10% FBS (HyClone FetalClone I), 2mM L-glutamine (Invitrogen), 10U/ml penicillin and 10μg/ml streptomycin (Gibco). MOLM13 cells were CRISPR/Cas9-edited to disrupt ADGRE2 or CLEC12A, and transduced with a retroviral vector expressing human CLEC12A to generate variants with high CLEC12A levels. Limiting dilution was used to identify clones with low and very-low ADGRE2 levels (~1,000 and ~500 molecules per cell, respectively), absent CLEC12A expression (CLEC12A-KO) or high CLEC12A expression (5×10⁴ to 2×10⁵ molecules per cell). ADGRE2 and CLEC12A surface protein densities were determined via flow cytometry as described below. For use in the humanized AML xenograft model, MOLM13 with endogenous ADGRE2 and CLEC12A levels was transduced with a retroviral vector expressing human CD19 and a clone was isolated with CD19 levels matching those found in NALM6 cells. Unmodified EL4 murine lymphoma cell line was obtained from ATCC. EL4 cells were transduced with retroviral vectors expressing human ADGRE2 or CLEC12A. EL4 cells were cultured in DMEM medium supplemented with 10% FBS (HyClone FetalClone I), 2mM L-glutamine (Invitrogen), 10U/ml penicillin and 10μg/ml streptomycin (Gibco). All cell lines were split every 2–3 days, plated at 0.5×10⁶ to 2×10⁶ cells/ml and cultured at 37°C with 5% carbon dioxide (CO2).

AML cell line xenograft models

We used male 8 to 12-week-old NOD/SCID/IL-2Rγ-null (NSG) mice (Jackson Laboratory), under a protocol approved by the MSKCC Institutional Animal Care and Use Committee. In the MOLM13 AML model, a total of 1×10⁶ FFLuc-GFP MOLM13 (WT or target-variant) cells per mouse were administered via tail vein injection (d-5). Five days later (d0), untransduced (UTD) or CAR-transduced T cells were administered via tail vein injection, at a given dose based on transgene-expressing T cells. In the U937 AML model, a total of 0.5×10⁶ FFLuc-GFP U937 cells per mouse were administered via tail vein injection (d-4). Four days later (d0), untransduced (UTD) or CAR-transduced T cells were administered via tail vein injection, at a given dose based on transgene-expressing T cells. AML rechallenge experiments were performed by tail vein injection of 1×10⁶ MOLM13 (WT or CLEC12A-high) cells per mouse at the indicated time points. AML burden was measured by bioluminescence imaging using the Xenogen IVIS Imaging System (Xenogen). Living Image software (Xenogen) was used to analyze acquired bioluminescence data. Survival was monitored for a minimum of 100 days post T cell injection. Mice were euthanized when showing clear clinical signs of distress or when reaching maximum AML burden as defined by hindlimb paralysis. There were no instances at which this maximum was exceeded.

AML patient-derived xenograft models

Peripheral blood samples were collected from AML patients after receiving informed consent under an IRB-approved protocol (#14–091). Samples were processed via isolation of mononuclear cells by Ficoll density centrifugation followed by red blood cell lysis, and engrafted via tail vein injection into immunodeficient mice, as previously described and under a protocol approved by the MSKCC Institutional Animal Care and Use Committee.⁹¹ For PDX#1, female 6 to 8-week-old NSG (Jackson Laboratory) mice were used. For PDX#2,3, female 6 to 8-week-old NOD-scid IL2Rgnull-3/GM/SF (NSG-SGM3, Jackson Laboratory) mice were used. AML PDX models were serially transplanted three times before being deemed established. Clinically annotated cytogenetic and molecular aberrations were confirmed on PDX samples. AML PDX-engrafted mice were treated with TRAC-KO untransduced or CAR T cells to avoid any potential allogeneic graft-versus-leukemia effects. Serial flow cytometry of peripheral blood and bone marrow was performed to monitor AML burden (live/hCD45dim/CD4^neg/CD8^neg) and TRAC-KO CAR T cell expansion (live/hCD45^pos/CD3^neg/CD4^pos or CD8^pos). For serial AML transplantation, bone marrow was harvested, followed by red blood cell lysis and re-engraftment into new mice of the same strain, age and sex. Survival was monitored for a minimum of 100 days post T cell injection. Mice were euthanized when showing clear clinical signs of distress or when reaching maximum AML burden as defined by hindlimb paralysis. There were no instances at which this maximum was exceeded.

Normal HSPC collection and in vitro culture

Leukapheresis material from G-CSF-mobilized adult healthy donors (Miltenyi Biotec) was used to isolate CD34^pos cells via CliniMACS CD34 GMP MicroBeads (Miltenyi Biotec) and a CliniMACS device (Miltenyi Biotec). T cells from the CD34^neg fraction were isolated using a Pan T Cell Isolation Kit (Miltenyi Biotec). Matched donor CD34^pos HSPCs and CD34^neg pan T cells were separately cryopreserved in CryoStor CS10 (Stemcell Technologies) freezing medium. HSPCs were thawed using thawing buffer (X-vivo 15, 1% human serum albumin, 10U/ml heparin) and subsequently cultured in HSPC cytokine-rich medium (StemSpan SFEM II (Stemcell Technologies), SCF 100ng/ml (R&D Systems), TPO 100ng/ml (Stemcell Technologies), FLT3L 100ng/ml (R&D Systems), IL6 100ng/ml (Peprotech), UM729 0.5μM (Stemcell Technologies), SR1 0.75μM (Cellagen Technology), streptomycin 20mg/ml, penicillin 20U/ml) under hypoxic conditions (2% O2), at 37°C and 5% CO2, maintaining a cell concentration of 0.5×10⁶/ml.^92,93 For off-target hematotoxicity in vitro assays, conditioned media were added at 1:1 volume ratio to the above HSPC culture conditions, and anti-IFN-γ blocking antibody (clone B133.5, Bioxcell) was used at a final concentration of 20 μg/ml.

Humanized AML xenograft model

Humanized AML-engrafted mice were used under a protocol approved by the MSKCC Institutional Animal Care and Use Committee, to evaluate the impact of in vivo activated CAR T cells on normal human hematopoietic cells (model outline described in Figure 6A). CAR T cells were generated from cryopreserved T cells originating from the same donor as HSPCs, using the γ-retroviral T cell transduction protocol as described above. 5 days after HSPC thawing (d-21), in vitro expanded HSPCs were injected via tail vein into sublethally (2Gy) irradiated female 4-week-old NSG mice (7×10⁵ per mouse). 18 days after HSPC injection (d-3), FFLuc-GFP MOLM13-CD19^pos AML cells (5×10⁵ per mouse) were injected via tail vein. 21 days after HSPC injection (d0), CAR T cells (19–28z1XX or ADCLEC.syn1, 2.5×10⁵ per mouse) were injected via tail vein. Antileukemic efficacy was monitored via bioluminescence imaging. On d7, bone marrow aspirates from untreated, 19–28z1XX-treated and ADCLEC.syn1-treated mice were collected and analyzed via flow cytometry. MOLM13 AML cells and normal human hematopoietic cell populations were defined as follows: MOLM13-CD19^pos (hCD45^pos CD33^pos CD19^pos), HSPCs (hCD45^pos CD3^neg CD19^neg CD14^neg CD16^neg CD34^pos), B cells (hCD45^pos CD33^neg CD19^pos), monocytes (hCD45^pos CD14^pos CD16^neg), CAR T cells (hCD45^pos CD3^pos).

METHOD DETAILS

Flow cytometry

All flow cytometry experiments were performed using standard sample processing and staining protocols, and data was acquired on a spectral flow cytometer (Cytek Aurora, 5-laser configuration). Transgene-positive T cells were identified by detection of ADGRE2-CAR, CLEC12A-CCR or EGFRt reporter, wherever applicable, using the following reagents: ADGRE2-CAR anti-idiotype (custom clone 23D5.G5.D4.C10, Rockland Immunochemicals, Inc.), CLEC12A-CCR anti-idiotype (custom clone 18F11.E11.C6, Rockland Immunochemicals, Inc.), EGFRt (clone AY13, Biolegend). Live cells were identified as cells staining negative for DAPI (BD Biosciences). Human hematopoietic cells were characterized using the following reagents: CD3 (UCHT1, BD Biosciences), CD4 (L200, BD Biosciences), CD8 (SK1, BD Biosciences), CD14 (63D3, Biolegend), CD16 (CB16, ThermoFisher), CD19 (SJ25C1, BD Biosciences), CD25 (2A3, BD Biosciences), CD34 (581, Biolegend), CD38 (HB-7, Biolegend), hCD45 (HI30, Invitrogen), mCD45 (30-F11, Biolegend), CD45RA (HI100, BD Biosciences), CD71 (M-A712, BD Biosciences), CD90 (5E10, Biolegend), IFN-γ (B27, Biolegend). AML target expression was detected using the following reagents: CD33 (P67.6, Biolegend), CD123 (6H6, Biolegend), ADGRE2 (2A1, Bio-Rad), CLEC12A (50C1, BD Biosciences).

Flow cytometric gating and quantification of surface target antigen densities

For AML patient and normal donor samples, the following gating strategies were performed to identify relevant cell populations: AML bulk CD45^dim SSC^low CD3^neg CD19^neg. AML LSC CD45^dim SSC^low CD3^neg CD19^neg CD14^neg CD16^neg CD71^neg CD34^pos CD38^neg. HSC (=HSC-enriched MPP fraction) CD45^dim SSC^low CD3^neg CD19^neg CD14^neg CD16^neg CD71^neg CD34^pos CD38^neg CD45RA^neg CD90^pos. MPP CD45^dim SSC^low CD3^neg CD19^neg CD14^neg CD16^neg CD71^neg CD34^pos CD38^neg CD45RA^neg CD90^neg. MLP CD45^dim SSC^low CD3^neg CD19^neg CD14^neg CD16^neg CD71^neg CD34^pos CD38^neg CD45RA^pos CD90^neg. CP CD45^dim SSC^low CD3^neg CD19^neg CD14^neg CD16^neg CD71^neg CD34^pos CD38^pos. Granulocytes CD45^med SSC^high. Monocytes CD45^high SSC^med CD14^pos CD16^neg. B cells CD45^high SSC^low CD19^pos CD3^neg. T cells CD45^high SSC^low CD19^neg CD3^pos. Antigen distribution analysis was defined as negative/positive percentages (wherein positive indicates ≥1,000 molecules per cell), as limiting resolution at antigen densities <1,000 molecules per cell did not allow for reliable gating of negative vs low subpopulations. Absolute surface antigen densities of CAR target candidates were determined by reading the median fluorescence intensity of relevant samples against a calibration curve based on external standard microsphere beads saturated with the relevant staining antibody at known antibody binding capacity. Quantum^™ Simply Cellular^® anti-Mouse IgG (Bangs Laboratories, Inc.) beads were stained and acquired in parallel with the sample at identical flow cytometry instrument settings. Quantitation bead staining, acquisition and subsequent calculations were performed according to the manufacturer’s protocol and the QuickCal^® (Bangs Laboratories, Inc.) analysis template.

Single-cell target RNA expression profile in defined cell types of normal tissues

The data for single-cell target RNA expression in normal tissues and their associated cell types were obtained from the Human Protein Atlas (version 21.1).⁵⁶ Single-cell data and their annotated cell types are available for download through the Human Protein Atlas (https://www.proteinatlas.org/download/rna_single_cell_type_tissue.tsv.zip).

T cell isolation, activation and culture

Buffy coats from healthy volunteer donors were obtained from the New York Blood Center. Peripheral blood mononuclear cells were isolated by density gradient centrifugation. T cells were purified using a Pan T Cell Isolation Kit (Miltenyi Biotec) and activated with Human T-Activator CD3/CD28 Dynabeads (Thermo Fisher) at 1:1 bead:cell ratio in X-vivo 15 medium (Lonza) supplemented with 5% human serum (Gemini Bioproducts) with 5ng/ml human recombinant interleukin-7 (Miltenyi Biotec) and 5ng/ml human recombinant interleukin-15 (Miltenyi Biotec) at a density of 1×10⁶ cells/ml. The medium was changed every 2 days, and cells were replated at 1–1.5×10⁶ cells/ml. T cells were incubated at 37°C, at 5% carbon dioxide (CO2). For AML patient-derived CAR T cells, clinical information and peripheral blood of an MSKCC AML patient were obtained under IRB-approved protocols (protocols #14–091 and #06–107).

gRNA, Cas9 protein and RNP formation

TRAC gRNA target sequence: 5′-CAGGGTTCTGGATATCTGT as described in Eyquem, Mansilla-Soto et al.⁹⁴ IFNG gRNA target sequence: 5’-CCAGAGCATCCAAAAGAGTG as described in Bailey et al.⁸⁹ TRAC and IFNG gRNAs were ordered from Synthego with 2′-O-methyl 3′-phosphorothioate modifications in the first and last three nucleotides.⁹⁵ Guide RNA was resuspended with TE buffer at 40μM. Cas9 protein (40μM) was obtained from QB3-Berkeley Macrolab core facility. TRAC ribonucleoprotein (RNP) was prepared by mixing Cas9 protein and TRAC gRNA at 1:1 molar ratio, incubating at 37°C for 15 min and immediately using it for T cell editing experiments.

Gene editing

CRISPR/Cas9 editing was used to disrupt the TCRα subunit constant gene (TRAC) in T cells for PDX experiments and IFNG in T cells for off-target hematotoxicity experiments. 48h after initiating T cell activation, the CD3/CD28 beads were magnetically removed, and T cells were transfected by electrotransfer of TRAC or IFNG RNP using a 4D-Nucleofector device (Lonza). Then, 2×10⁶ or 10×10⁶ cells were resuspended in P3 buffer (Lonza) and mixed with 60 or 300pmol TRAC or IFNG RNP in a total volume of 20 or 100μl, respectively. Following electroporation and assuming 66.7% recovery, cells were diluted and incubated in culture medium at 1×10⁶ cells/ml. 12 to 24h post electroporation, T cells were transduced via SFG-γ-retroviral vector following the transduction protocol as described for non-edited T cells. Subsequently, TRAC or IFNG-edited and SFG γ-retrovirally-transduced T cells were cultured using T cell growth medium, replenished at least every two days and as needed to maintain a density of 1–1.5×10⁶ cells/ml.

Gammaretroviral vector construction, production and transduction

Plasmids encoding the SFG γ-retroviral vector⁹⁶ were used to clone bicistronic constructs for screening of candidate CAR and CAR+CCR designs, control CARs and ADCLEC.syn1, as outlined in Figure S2A-B. The CAR hinge, transmembrane and endodomain have previously been described and incorporate a modified CD3ζ domain with a single proximal immunoreceptor tyrosine-based activation motif (ITAM).⁴⁷ CARs were either co-expressed with a truncated EGFR (EGFRt) reporter, a chimeric costimulatory receptor (CCR) or a CCR-del control, as previously described.^44,46 SFG γ-retrovirus encoding CAR or CCR constructs was produced via transfection of amphotropic Phoenix-AMPHO cell line (ATCC) with SFG vector plasmids, pCMV-gag-pol plasmid (Cell Biolabs) and pCMV-VSV-G envelope plasmid (Cell Biolabs), using FuGene HD transfection reagent (Promega), followed by virus collection 24h post transfection. T cells were transduced with retroviral supernatants by centrifugation on Retronectin-coated plates (Takara), as described previously.⁹⁷

CAR scFv origin and selection

ADGRE2 scFv candidates were generated based on humanization of V_H and V_L sequences of a murine ADGRE2 antibody (clone 2A1), using Xoma and Molecular Operating Environment (MOE) humanization methods, leading to amino acid changes in the V_H and V_L framework and associated changes in affinities (parental clone 2A1 scFv: K_D 2.66×10⁻¹¹M, lead humanized scFv: K_D 2.10×10⁻¹⁰M). The parental murine antibody clone 2A1 has previously been reported.⁹⁸ CLEC12A scFv candidates were generated based on V_H and V_L sequences from human anti-CLEC12A antibodies generated in Trianni mice immunized with recombinant human CLEC12A protein. Candidate scFv targeting ADGRE2 (n=18 scFv) and CLEC12A (n=16 scFv) were screened for in vitro and in vivo efficacy in the 28z1XX CAR format. The lead ADGRE2 scFv was chosen based on preferential killing of ADGRE2^high and relative sparing of ADGRE2^low AML cell lines. The lead CLEC12A scFv was chosen based on highest killing of CLEC12A^high and CLEC12A^low AML cell lines. ADGRE2 and CLEC12A scFv sequences are listed in patent WO2022232016A2. The CD33 scFv was generated using heavy (V_H)- and light (V_L)-chain variable regions from a published anti-CD33 antibody sequence (clone 280–31-01, as described in patent WO2012045752A1). The CD19 scFv is based on clone SJ25C1, as previously reported.⁹⁹

Cytotoxicity assays

In vitro anti-leukemic cytotoxicity of CAR T cells was determined via standard FFLuc-based assay. FFLuc-expressing MOLM13 cells served as target cells. The effector (E) and target (T) cells were co-cultured in triplicates at the indicated E:T ratio using black-walled flat-bottom 96-well plates with 5×10⁴ target cells in a total volume of 100μl per well in T cell medium. 18h later, 50μl D-luciferin (Goldbio) at 1.5μg/μl was directly added to each well. Immediately after, emitted light was detected in a luminescence plate reader, and % cytotoxicity was calculated using the formula 100x(1−(RLU_sample)/(RLU_target alone)). For studies using EL4 cells, in vitro cytotoxicity was determined via flow cytometric quantification of target cell counts. EL4 cells (T) were co-cultured with CAR T cells (E) at the indicated E:T ratio using flat-bottom 96-well plates with 5×10⁴ target cells in a total volume of 100μl per well in T cell medium. 48h later, flow cytometry was used to determine the residual total target cell counts (live/CD3^neg cells) as well as T cell activation (CD25 MFI). % cytotoxicity was calculated using the formula 100x(1−(RLU_sample)/(RLU_target alone)).

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistics

All experimental data are presented as described in the figure legends. In vitro experiments were performed in technical triplicates and replicated with at least two independent donors for CAR T cell production. For in vivo experiments, n=5 mice per group were used unless indicated otherwise. Across all in vivo experiments, CAR T cells from a total of n=13 different adult healthy T cell donors were used. For all Kaplan-Meier survival analyses, log-rank (Mantel-Cox) testing was performed. Statistical tests are described in the figure legends. Statistical analysis was performed on GraphPad Prism v10.0.1 software. Significance was set at p<0.05.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
ADGRE2-CAR (anti-idiotype, 23D5.G5.D4.C10, Dylight405)	This paper	N/A
CLEC12A-CCR (anti-idiotype, 18F11.E11.C6, PE)	This paper	N/A
human EGFRt (AY13, BV711)	Biolegend	Cat# 352920; RRID: AB_2687123
human CD3 (UCHT1, BUV615)	BD Biosciences	Cat# 612992; RRID: AB_2870263
human CD4 (L200, BUV395)	BD Biosciences	Cat# 564107; RRID: AB_2738596
human CD8 (SK1, BUV805)	BD Biosciences	Cat# 612889; RRID: AB_2833078
human CD14 (63D3, SparkBlue550)	Biolegend	Cat# 367148; RRID: AB_2832724
human CD16 (CB16, SuperBright600)	ThermoFisher	Cat# 63-0168-42; RRID: AB_2662479
human CD19 (SJ25C1, BUV563)	BD Biosciences	Cat# 612916; RRID: AB_2870201
human CD25 (BC96, PE)	Biolegend	Cat# 302606; RRID: AB_314276
human CD34 (581, APC/Fire750)	Biolegend	Cat# 343536; RRID: AB_2650736
human CD38 (HB-7, BV650)	Biolegend	Cat# 356620; RRID: AB_2566233
human CD45 (HI30, PacificOrange)	ThermoFisher	Cat# MHCD4530; RRID: AB_10376143
murine CD45 (30-F11, AF488)	Biolegend	Cat# 103122; RRID: AB_493531
human CD45RA (HI100, BV480)	BD Biosciences	Cat# 566114; RRID: AB_2739516
human CD71 (M-A712, BV711)	BD Biosciences	Cat# 563767; RRID: AB_2738413
human CD90 (5E10, PE/Dazzle594)	Biolegend	Cat# 328134; RRID: AB_2566343
human CD33 (P67.6, PE/Cy7)	Biolegend	Cat# 366618; RRID: AB_2566420
human CD123 (6H6, BV785)	Biolegend	Cat# 306032; RRID: AB_2566448
human ADGRE2 (2A1, AF647)	Bio-Rad	Cat# MCA2330A647T; RRID: AB_2231142
human CLEC12A (50C1, BB515)	BD Biosciences	Cat# 565325; RRID: AB_2739182
human IFNg (B27, PE)	Biolegend	Cat# 506507; RRID: AB_315440
human IFNg (blocking)	Bio X Cell	Cat# BE0235; RRID: AB_2687717
Biological samples
Mobilized Leukopaks®, G-CSF x 5 Days - Apheresis Day 6, Fresh	Miltenyi Biotec	Cat# 150-000-461
Normal donor whole bone marrow, fresh	Allcells	N/A
Buffy coats	New York Blood Center	N/A
AML patient samples	MSKCC	IRB# 06-107 and IRB# 14-091
AML patient-derived xenograft models	MSKCC, Antitumor Assessment Core Facility	IRB# 14-091
Chemicals, peptides, and recombinant proteins
Cas9-NLS purified protein	QB3-Berkeley Macrolab	N/A
P3 Primary Cell 4D-Nucleofector® X Kit	Lonza	Cat# V4XP-3024
Human IL-7, premium grade	Miltenyi Biotec	Cat# 130-095-363
Human IL-15, premium grade	Miltenyi Biotec	Cat# 130-095-765
Dynabeads™ CD3	ThermoFisher	Cat# 11151D
Critical commercial assays
Quantum™ Simply Cellular® antiMouse IgG	Bangs Laboratories	Cat# 815
Pan T Cell Isolation Kit, human	Miltenyi Biotec	Cat# 130-096-535
Deposited data
Normal tissue scRNAseq data	Human Protein Atlas (v21.1)	RRID: SCR_006710; Download:https://www.proteinatlas.org/download/rna_single_cell_type_tissue.tsv.zip
Experimental models: Cell lines
Phoenix-AMPHO	ATCC	Cat# CRL-3213; RRID: CVCL_H716
EL4	ATCC	Cat# TIB-39; RRID: CVCL_0255
MOLM-13	DSMZ	Cat# ACC 554
U-937	ATCC	Cat# CRL-1593.2; RRID: CVCL_0007
Experimental models: Organisms/strains
NSG mouse: NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ	Jackson Laboratory	RRID: IMSR_JAX:005557
NSG-SGM3 mouse: NOD.Cg-Prkdcscid Il2rgtm1Wjl Tg(CMV-IL3,CSF2,KITLG)1 Eav/MloySzJ	Jackson Laboratory	RRID: IMSR_JAX:013062
Oligonucleotides
TRAC sgRNA target sequence: 5′-CAGGGTTCTGGATATCTGT	Synthego	N/A
IFNG sgRNA target sequence: 5′-CCAGAGCATCCAAAAGAGTG	Synthego	N/A
Recombinant DNA
pCMV-gag-pol	Cell Biolabs	Cat# RV-111
pCMV-VSV-G	Cell Biolabs	Cat# RV-110
pSFG-ADCLEC.syn1	This paper	N/A
pSFG-ADGRE2-28z1XX-P2A-EGFRt	This paper	N/A
pSFG-CLEC12A-BB-P2A-EGFRt	This paper	N/A
pSFG-ADGRE2-28z1XX-P2A-CLEC12A-del	This paper	N/A
pSFG-CD33-28z 1XX-P2A-EGFRt	This paper	N/A
pSFG-CD19-28z1XX-P2A-EGFRt	This paper	N/A
Software and algorithms
GraphPad Prism v10.0.1	GraphPad Software	N/A
FlowJo v10.8.1	BD Biosciences	N/A
Living Image	Perkin Elmer	N/A
Adobe Illustrator v27.8.1	Adobe	N/A

Highlights.

• Antigen co-expression signatures guide cooperative CAR T cell design

• CAR sensitivity is controlled by combining a calibrated CAR and a CCR (IF-BETTER gate)

• ADCLEC.syn1 overcomes AML antigen escape and HSPC on-target toxicity

• Reducing CAR T cell interferon-γ release mitigates off-target hematotoxicity