The search for safe and effective CAR-T targets in AML Authors: Florian Van Oers, Donovan Flumens, Simon Haentjens, Laurens Krekelbergh, Sébastien Anguille Published: 2026 ABSTRACT AbstractAcute myeloid leukemia (AML) remains a highly aggressive malignancy with limited therapeutic options and poor long-term survival. A major barrier to curative treatment is the persistence of leukemic stem cells (LSCs), a chemo-resistant population that drives relapse. Chimeric antigen receptor (CAR)-T-cell therapy has transformed the treatment of B-cell hematological malignancies. However, its application in AML has been met with significant challenges. Among the key challenges are the scarcity of AML-specific antigens and the risk of on-target/off-tumor toxicity due to shared antigen expression on normal hematopoietic stem cells (HSCs) and/or mature blood cells. Early clinical trials of CAR-T-cell therapy in AML —primarily targeting CD123, CD33, or CLL‑1— have demonstrated limited durable complete remissions and/or frequent myeloablation, underscoring the need for more selective targets. While other targets show more restricted expression profiles, they are often expressed only in a small subgroup of AML patients. In this review, we systematically evaluated 63 AML-associated antigens for which CAR constructs have been reported, using five criteria: (1) homogeneous expression across AML patients, (2) uniform expression on AML cells within individual patients; (3) presence on LSCs, (4) absence on normal HSCs, and (5) no or acceptable expression on mature blood cells. Applying a 20-point scoring framework, 13 novel antigens emerged as the most promising candidates for CAR-T-cell therapy in AML: ADGRE2, SIGLEC-6, IL1RAP, MUC1, CCR1, CD155, CD70, LILRB4, GRP78, CD37, ITGB2, TIM-3 and mesothelin. We discuss the advantages and limitations of each target, along with strategies to mitigate associated risks. With no CAR-T-cell therapy currently approved for AML, this comprehensive review provides a prioritized antigen landscape and a framework to guide the rational design of next-generation CARs for this challenging malignancy.Subject terms: Acute myeloid leukaemia, Translational research, Preclinical research, Immunotherapy Introduction Acute myeloid leukemia (AML) is a highly aggressive hematological malignancy arising from the clonal expansion of hematopoietic stem cells (HSCs) and/or myeloid progenitor cells in the bone marrow. Current treatment strategies involve intensive chemotherapy commonly followed by allogeneic HSC transplantation (allo-HSCT) in younger, fit patients, and hypomethylating agents combined with venetoclax in older, non-fit individuals [1, 2]. Despite a substantial number of patients achieving complete remission (CR), relapse remains common due to the persistence of chemotherapy-resistant leukemic stem cells (LSCs). Consequently, the 5-year overall survival rate remains dismal at approximately 30% [1], underscoring the need for more effective and durable treatment strategies. Chimeric antigen receptor (CAR)-T-cell therapy has transformed the treatment of B-cell hematological malignancies, with several products now approved for clinical use [3, 4]. CAR-T-cell therapy involves the use of autologous T-cells genetically engineered to express a CAR, a synthetic receptor typically composed of an extracellular antigen-binding domain derived from an antibody, coupled with one or more intracellular signaling domains. Upon antigen binding, the CAR initiates T-cell activation, proliferation, and targeted cytotoxicity towards the tumor cells [2, 3]. Building on its success in B-cell hematological cancers, CAR-T therapy has also been explored in AML. As shown in Table 1, to date, the largest clinical trials have predominantly focused on C-type lectin-like molecule-1 (CLL-1), CD33 and CD123 as target antigens, reflecting their broad expression in AML. Other, less extensively explored targets include natural killer group 2 member D ligands (NKG2DL), CD7, CD19, CD38, and the Lewis Y (LeY) antigen (Table 1). While efficacy outcomes vary, most studies are single-center and single-arm with relatively small patient cohorts, limiting the generalizability of the findings (Supplementary Table 1). Moreover, the impact of conditioning regimens administered prior to CAR-T-cell infusion should also be considered, as these therapies exert intrinsic antileukemic effects that may confound interpretation of CAR‑T‑cell efficacy (Supplementary Table 1). Table 1Overview of CAR-T clinical trials in AML.Target antigenExpression in AMLClinical data# patients treatedCR/CRi rateSafety concernsAllo-HSCT after CAR-TRef.CLL-1++++ [1, 3, 17]10270.2– 100%Refractory granulocytopenia25–100%[1, 7, 10, 12]CD33++++ [1, 3, 17, 19, 20]740–75%Myeloablation5–75%[1, 5–8]CD123++++ [1, 3, 17]620–66%Neutropenia and thrombocytopenia8.3–33.3%[1, 7]NKG2DL+++ [94]190%/NA[9, 95]CD7++ [17, 19, 20]1770–100%CMV and EBV reactivation30–100%[1, 7, 13, 14]CD19+ [19, 20]1866.7–100%/30–50%[15, 16]CD38+++ [19]2250–70%/50%[1, 7, 10]LeY++ [96]425%/NA[1, 7]Patient expression levels: +(5–25%), ++ (25–50%), +++ (50–75%), ++++ (75–100%).allo-HSCT allogeneic hematopoietic stem cell transplantation, AML acute myeloid leukemia, CAR chimeric antigen receptor, CD cluster of differentiation, CLL-1 C-type lectin-like molecule-1, CMV cytomegalovirus, CR/CRi complete remission (CR) with incomplete count recovery (CRi), EBV Epstein-Barr virus, LeY Lewis Y, NKG2DL natural killer group 2 member D ligand, NA not available. Overview of CAR-T clinical trials in AML. Patient expression levels: +(5–25%), ++ (25–50%), +++ (50–75%), ++++ (75–100%). allo-HSCT allogeneic hematopoietic stem cell transplantation, AML acute myeloid leukemia, CAR chimeric antigen receptor, CD cluster of differentiation, CLL-1 C-type lectin-like molecule-1, CMV cytomegalovirus, CR/CRi complete remission (CR) with incomplete count recovery (CRi), EBV Epstein-Barr virus, LeY Lewis Y, NKG2DL natural killer group 2 member D ligand, NA not available. The first clinical application of CAR-T-cell therapy in AML dates back to 2013, targeting the LeY antigen. As shown in Table 1, only 25% achieved CR, marking a modest start for the field [1]. Subsequent trials targeting CD33 [1, 5–8] and CD123 [1, 7] also demonstrated limited efficacy, with CR rates generally below 50%. These outcomes were further compromised by severe myeloablation, especially in the CD33-directed trials, due to targeting of normal HSCs. Anti-NKGD2L CAR-T-cells showed even less promise, with no objective responses observed in three small, early-phase clinical trials [9]. More encouraging results emerged from trials targeting CLL-1, where response rates ranged from 60% to 100% [1, 7, 10–12]. However, this approach was also associated with substantial myelotoxicity due to CLL-1 expression on normal myeloid-lineage cells, leading to prolonged and sometimes refractory neutropenia, ultimately requiring allo-HSCT to restore hematopoiesis. Similar observations were made in trials targeting CD38 [1, 7, 10], CD7 [1, 7, 13, 14], and CD19 [15, 16]. Importantly, AML is a notoriously heterogenous disease [3, 17, 18] and these trials selectively enrolled patients expressing the respective target antigen. For example, CD7 and CD19 are expressed in only a minority of AML cases [17, 19, 20], limiting their utility as broadly applicable CAR targets. Where CD38 shows an intermediate expression profile, its lack of expression on LSCs may permit relapse [3, 17, 21]. In summary, although CAR-T-cell therapy has shown some degree of clinical activity in AML, no product has yet advanced to late-stage clinical trials. The considerable myelotoxicity and antigenic heterogeneity observed in AML have prevented CAR-T-cell therapy from achieving curative efficacy, limiting its application to a bridge-to-transplant strategy. Central to overcoming these limitations is the identification of the “ideal” CAR-T target antigen. General principles guiding CAR-T target selection include high tumor specificity and broad coverage [22, 23]. In the context of AML, the ideal antigen should be abundantly expressed on the surface of leukemic blasts in the majority of patients, while being absent from healthy cells and tissues. In particular, expression on critical blood cell populations, including but not limited to normal HSCs, neutrophils and T-cells, must be avoided to minimize hematotoxicity [3, 4, 24]. Importantly, both Perna et al. and Daver et al. emphasize that effective targeting of LSCs is critical to prevent disease relapse [3, 24]. Based on these considerations, together with clinical trial observations (Table 1), we defined the following criteria for an ideal AML CAR-T target antigen: (1) homogeneous expression across AML patients, (2) uniform expression on leukemic cells within individual patients; (3) presence on LSCs, (4) absence on normal HSCs, and (5) no or clinically acceptable expression on mature blood cells. In this review, we performed a comprehensive evaluation of 63 AML-related antigens for which CAR constructs have been reported, applying these five predefined criteria to assess their suitability as therapeutic targets. Using a 20-point scoring framework, we prioritized 13 novel antigens that represent the most promising candidates for future CAR-T-cell development in AML. Framework for identifying optimal CAR-T targets in AML To identify optimal CAR-T targets for AML, we systematically evaluated 63 AML-related antigens for which CAR constructs have been reported on PubMed/MEDLINE and Web of Science (cut-off date December 31, 2025). An overview of these 63 antigens can be found in Supplementary Tables 2 and 3. Each antigen was assessed using a structured 20-point scoring framework based on the 5 criteria introduced above (Fig. 1). 20-point scoring framework to select optimal CAR-T targets for AML. Radar plots illustrating the top-ranked targets, each scoring 16/20 or higher based on the following 5 evaluation criteria: Criterion 1: AML patient expression. 0 points for expression <5%, 1 point for 5–25%, 2 points for 25–50%, 3 points for 50–75%, and 4 points for >75%. Criterion 2: AML cell expression. 0 points for <5%, 1 point for 5–25%, 2 points for 25–50%, 3 points for 50–75%, and 4 points for >75%. Criterion 3: LSC expression. 0 points for negative (<5%) and 4 points for positive (>5%). Criterion 4: no HSC expression. 0 points for positive (>5%) and 4 points for negative (<5%). Criterion 5: no mature blood cell expression. 0 points if expression on both T-cells and neutrophils, 2 points if expression on T-cells or neutrophils and 4 points if no expression on both T-cells and neutrophils. Each axis represents one criterion, with scores ranging from 0 to 4. The larger radar plot at the top serves as a reference, depicting an ideal target meeting all criteria (20/20). ADGRE2 adhesion G protein-coupled receptor E2, AML acute myeloid leukemia, CLL-1 C-type lectin-like molecule-1, CCR1 C-C chemokine receptor type 1, CD cluster of differentiation, GRP78 78 kDa glucose-regulated protein, HSC hematopoietic stem cell, IL1RAP interleukin-1 receptor accessory protein, ITGB2 integrin beta 2, LILRB4 leukocyte immunoglobulin-like receptor subfamily B member 4, LSC leukemic stem cell, MUC1 mucin-1, SIGLEC-6 sialic acid-binding Ig-like lectin 6, TIM-3 T-cell immunoglobulin and mucin-domain containing protein 3. Criterion 1: homogeneous expression across AML patients AML is characterized by significant interpatient variability in antigen expression [1, 3, 4, 17, 25]. As discussed above, some antigens, such as CD19 and CD7, are expressed only in small subgroups [17, 19, 20]. To ensure broad suitability across diverse AML subtypes, we propose as first criterion that an optimal target antigen should exhibit high expression. In our scoring system, this criterion was assigned a maximum of 4 points, based on a semi-quantitative classification, with 4 points reflecting expression in >75% of AML patients (0 points = <5%; 1 point = 5–25%; 2 points = 25–50%; 3 points = 50–75%; 4 points = >75%) (Fig. 1). Criterion 2: uniform expression within individual AML patients In addition to interpatient heterogeneity, AML also exhibits significant intratumoral variability, with subclonal populations potentially differing in antigen expression [1, 17]. This intrapatient heterogeneity poses a major therapeutic challenge, as antigen-negative clones can escape CAR-T-cell-mediated cytotoxicity and drive relapse [3, 26]. Therefore, ideal targets should also display high and uniform expression within individual patients. To account for this, we evaluated the proportion of AML cells expressing the antigen within individual samples, using the same semi-quantitative scale as for criterion 1 (0 points = <5%; 1 point = 5–25%; 2 points = 25–50%; 3 points = 50–75%; 4 points = >75%) (Fig. 1). Criterion 3: expression on LSCs To achieve durable remissions, CAR-T-cells must effectively target the LSC compartment [18]. LSCs are a self-renewing subpopulation of immature leukemic cells whose presence is associated with poor survival [27, 28]. They are thought to be resistant to chemotherapy due to a complex interplay of quiescence, enhanced DNA repair capacity, resistance to apoptosis, and protection within the bone marrow niche in which they reside [29]. They are typically enriched in the CD34⁺/CD38⁻ compartment of the bone marrow, which also harbors their healthy counterparts, HSCs [3, 17, 21]. Consequently, LSCs exhibit considerable antigenic overlap with normal HSCs, posing a risk of myeloablation [1, 2]. Given the binary nature of most available data, this criterion was scored as either 4 points if expression on LSCs was reported, or 0 points if absent or unknown (Fig. 1). Criterion 4: absence on normal HSCs To preserve normal hematopoiesis and avoid myeloablation, CAR-T targets should ideally not be expressed on healthy HSCs. Antigens such as CD33 and CD123, while commonly found on AML cells, are also present on HSCs [3, 17], explaining the myelotoxicity observed in these clinical trials (Table 1). As with LSC expression, this criterion was scored in a binary fashion: 4 points were assigned if expression on HSCs was absent or considered negligible (<5%), and 0 points if present (>5%) or unknown (Fig. 1). Criterion 5: no or acceptable expression on mature blood cells When it comes to the targeting of mature blood cells, preference should be given to antigens that demonstrate an acceptable hematological toxicity profile (Fig. 1). In particular, targeting neutrophils and T-cells should be avoided to prevent persistent neutropenia and T-cell aplasia, which may result in life-threatening infections and viral reactivation of EBV or CMV (Table 1) [30]. Moreover, targeting T-cell antigens introduces a risk of CAR-T-cell fratricide [3, 4, 30]. In contrast, expression on B-cells is considered acceptable since hypogammaglobulinemia can be effectively managed with intravenous immunoglobulins administration [4]. Hence, 2 points were assigned for lack of expression on granulocytes and 2 points for lack of expression on T-cells, for a maximum of 4 points (Fig. 1). Leading candidate target antigens for CAR-T therapy in AML From the list of 63 AML antigens with published CAR constructs (Supplementary Tables 2 and 3), we identified 14 candidates that most strongly met the 5 predefined selection criteria, each scoring ≥16 out of 20 points in our framework (Fig. 1). Although CLL-1 also received a high score (18/20; Fig. 1), it was excluded from the present analysis because our focus was restricted to targets that have not yet undergone clinical evaluation (Table 1), thereby leaving us with 13 antigens for further discussion (Table 2). Table 2Top‑ranked CAR‑T target antigens in AML based on a 20‑point scoring framework.Criterion 1: AML patientsCriterion 2: AML cellsCriterion 3: LSCCriterion 4: HSCCriterion 5: Mature blood cellsScoreRefs.ADGRE2++++++++++++–M20/20[3, 25, 32, 33]SIGLEC-6++++NAYes–Gb, B16-20 /20[34, 35]IL1RAP++++++++Yes–G, M18/20[21, 38–41]MUC1+++++++++–M18/20[42–46]CCR1+++++++++++–G, M17/20[3, 33]CD155+++++++++++–Gn, M, DC17/20[49–52]LILRB4+++++Yes–M, DC17/20[54–56]GRP78+++++Yes–/17/20[58–60]CD70+++++++++–T, B, NK16/20[3, 33, 62–64, 97]CD37++++++++++++–G, M, DC, T, B, NK16/20[66, 67]ITGB2++++++++++–Gn, M, T, NK16/20[18, 20, 63, 68, 69]TIM-3++++++++–M, DC, T, NK16/20[17, 32, 71, 98, 99]Mesothelin+++++++–/16/20[32, 74–76]Expression levels: - (<5%), + (5-25%), ++ (25-50%), +++ (50-75%), ++++ (75-100%) and NA (not available).Mature cells: B, B-lymphocytes; DC, dendritic cells; G, granulocytes; Gb, basophils; Gn, neutrophils; M, monocytes; NK, natural killer cells; T, T-lymphocytes.Other abbreviations used: ADGRE2 adhesion G protein-coupled receptor E2, AML acute myeloid leukemia, CCR1 C-C chemokine receptor type 1, CD cluster of differentiation, GRP78 78 kDa glucose-regulated protein, HSC hematopoietic stem cell, IL1RAP interleukin-1 receptor accessory protein, ITGB2 integrin beta 2, LILRB4 leukocyte immunoglobulin-like receptor subfamily B member 4, LSC leukemic stem cell, MUC1 mucin-1, SIGLEC-6 sialic acid-binding Ig-like lectin 6, TIM-3 T-cell immunoglobulin and mucin-domain containing protein 3. Top‑ranked CAR‑T target antigens in AML based on a 20‑point scoring framework. Expression levels: - (<5%), + (5-25%), ++ (25-50%), +++ (50-75%), ++++ (75-100%) and NA (not available). Mature cells: B, B-lymphocytes; DC, dendritic cells; G, granulocytes; Gb, basophils; Gn, neutrophils; M, monocytes; NK, natural killer cells; T, T-lymphocytes. Other abbreviations used: ADGRE2 adhesion G protein-coupled receptor E2, AML acute myeloid leukemia, CCR1 C-C chemokine receptor type 1, CD cluster of differentiation, GRP78 78 kDa glucose-regulated protein, HSC hematopoietic stem cell, IL1RAP interleukin-1 receptor accessory protein, ITGB2 integrin beta 2, LILRB4 leukocyte immunoglobulin-like receptor subfamily B member 4, LSC leukemic stem cell, MUC1 mucin-1, SIGLEC-6 sialic acid-binding Ig-like lectin 6, TIM-3 T-cell immunoglobulin and mucin-domain containing protein 3. ADGRE2 Adhesion G protein-coupled receptor E2 (ADGRE2) is a member of the adhesion G protein-coupled receptor (GPCR) family and is involved in cell adhesion, migration, and immune responses. In AML patients, its overexpression correlates with poor survival, primarily due to its role in sustaining LSC self-renewal and driving leukemogenesis [31]. Perna et al. [3] reported ADGRE2 expression in all 30 AML patients analyzed, with a mean expression of 93% on blasts and >75% on LSCs. Three independent studies [25, 32, 33] confirmed these findings, showing expression rates of 82%-92% on AML blasts, and 87%-91% on LSCs. Where Perna et al. found minimal expression (<5%) on HS(P)Cs [3], Haubner et al. [25] observed that ADGRE2 is indeed detectable, albeit at substantially lower antigen density levels compared to classical AML targets such as CD33 or CD123. With the exception of monocytes, ADGRE2 is absent on mature blood cells, including T-cells and granulocytes [25], and shows only limited expression on non-hematological cells [3, 25]. ADGRE2 and CLL-1 show co-expression on AML blasts and LSCs but not on normal HSCs and granulocytes, providing a strong rationale for combined targeting. Haubner et al. [25] developed a multispecific CAR construct targeting ADGRE2 in combination with a CLL-1 chimeric costimulatory receptor (CCR). In AML xenograft mouse models, this multi-specific construct outperformed the monospecific ADGRE2-CAR, effectively eradicating AML while sparing normal hematopoietic cells. These promising findings have led to a phase 1 clinical study (NCT05748197). SIGLEC-6 Sialic acid-binding Ig-like lectin 6 (SIGLEC-6) shares structural and functional similarities with CD33 (also known as SIGLEC-3) [34]. Comparable to CD33, SIGLEC-6 demonstrates high expression rates (>75%) on both patient and LSC level, but with a more favorable toxicity profile [34]. SIGLEC-6 is absent on HSCs and its presence on mature blood cells is limited to B-cells and basophils [34, 35]. Among non-hematopoietic tissues, expression has only been observed in placental tissue [34, 35]. Confirmation of robust AML blast surface expression would position SIGLEC-6 as a leading CAR-T target. Two different SIGLEC-6 CAR-T therapies incorporating either CD28 or 4-1BB co-stimulatory domains were evaluated in AML xenograft models, achieving responses in all 16 mice tested [34]. A first-in-human clinical trial of SIGLEC-6 CAR-T-cell therapy in AML is currently ongoing (NCT05488132). IL1RAP Interleukin-1 receptor accessory protein (IL1RAP) acts as a co-receptor for the IL-1 and IL-33 receptors and mediates pro-inflammatory signaling [36, 37]. In AML, it promotes leukemogenesis through activation of FLT3 and KIT signaling pathways [37] and correlates with poor overall survival [36, 37]. Across three studies involving a total of 80 patients, mean IL1RAP positivity ranged from 79% to 100%, with >75% expression on AML cells [38–40]. Importantly, IL1RAP is expressed on LSCs but absent on HSCs [21, 38–40]. The main limitation is its expression on granulocytes, raising concerns for potential neutropenia. Among non-hematopoietic tissues, IL1RAP has been detected on hepatocytes and placental cells [40]. IL1RAP-directed CAR-T-cells showed benefit in AML xenografts [39], achieving a median survival of 240 days in treated mice compared to 98 days in controls, without affecting healthy HSCs. Another group recently showed that restoration of microRNA-142 using a synthetic mimic improved T-cell fitness and enhanced IL1RAP CAR-T efficacy in AML xenografts [41]. MUC1 Mucin 1 (MUC1) is a membrane-bound glycoprotein of the mucin family and has been identified as an oncogene in several solid tumors [42–45]. In AML, MUC1 contributes to leukemogenesis through activation of signaling pathways such as NF-κB and β-catenin, as well as upregulation of survivin [43]. Across five studies involving 175 AML patients, mean MUC1 expression was highly variable, ranging from 22.7% to 100% [42–46]. MUC1 is expressed on ±60% of AML blasts [44, 45], predominantly in myelomonocytic and monocytic subtypes [42, 44, 46], with 81.8% positivity reported in FAB M4/M5 cases and minimal expression in non-monocytic AML [46]. While present on healthy monocytes [44, 46], it is not expressed on granulocytes and T-cells [42]. It is absent on HSCs [45], but expressed on LSCs, where it contributes to their leukemia-initiating potential [44, 45]. A clinical trial last updated in 2018 (NCT03222674) incorporated MUC1 as one of several targets in a multi‑CAR‑T-cell strategy for AML patients; however, no results have been made publicly available. CCR1 Like ADGRE2, C-C motif chemokine receptor 1 (CCR1) is a member of the GPCR family and plays a role in immune cell migration [3]. CCR1 signaling activates the MAP kinase pathway, a major oncogenic driver in human cancers [47]. In AML, CCR1 exhibits a highly favorable expression profile, with >75% expression on AML patients, blasts and LSCs [3]. In contrast, Shahswar et al. observed a more modest expression rate of 44% on blasts [33], indicating that CCR1 blast expression is subject to variability. Importantly, CCR1 is minimally expressed (<5%) on HSCs and other non-hematopoietic tissues, indicating a low risk of off-tumor toxicity [3]. However, like CLL-1, CCR1 is expressed on different types of granulocytes, including neutrophils. Chen et al. [48] recently reported the first CCR1-directed CAR-T-cell therapy in a syngeneic AML mouse model. CCR1-specific CAR-T-cells demonstrated potent anti-leukemic activity and significantly prolonged survival, with only mild, transient weight loss and no hematotoxicity. CD155 CD155, a member of the immunoglobulin superfamily, acts as an adhesion molecule driving cell–cell interactions [49]. Upregulation of CD155 often results in immune suppression through interaction with the inhibitory receptor TIGIT [50]. Elevated expression has been associated with poor overall survival, underscoring its role in immune escape and leukemia progression [51]. CD155 is overexpressed in various solid tumors [52] and detected in a high proportion of AML patients, with reported frequencies exceeding 77% [49–52]. At the cellular level, high CD155 expression is observed both on AML blasts and LSCs [50–52]. While absent on T-cells [52], CD155 is present on dendritic cells, monocytes, and granulocytes, raising concerns for neutropenia. CD155 is absent on primitive HSCs but moderately expressed on granulocyte–monocyte and megakaryocyte–erythroid progenitors [52]. On non-hematologic tissues, CD155 shows little to no expression [52]. Functionally, targeting CD155 may offer both direct anti-leukemic activity and restoration of T-cell function. Stamm et al. [51] demonstrated that CD155 targeting with a bispecific T-cell engager led to potent AML cell killing and enhanced cytotoxic T-cell responses. More recently, CD155-directed CAR-T-cells exhibited robust in vitro anti-leukemic activity against primary AML blasts and significantly prolonged survival in a xenograft mouse model [52]. To date, no clinical trials of CD155 CAR-T therapy have been initiated. LILRB4 Leukocyte immunoglobulin-like receptor subfamily B member 4 (LILRB4) functions primarily as an immune checkpoint inhibitor [53]. Its expression on myeloid-derived suppressor cells (MDSCs) and AML cells fosters an immunosuppressive bone marrow microenvironment [53]. In AML, LILRB4 expression is restricted to cases with monocytic differentiation (FAB M4/M5) [54–56], indicating that LILRB4 is not a broadly applicable target in AML. Although limited to monocytic AML, these patients exhibit moderate to high expression on leukemic blasts and LSCs [54], while LILRB4 is absent on HSCs [54] and most mature blood cells except monocytes [54, 57]. This lineage-restricted profile makes LILRB4 an attractive candidate for CAR-T therapy in (myelo)monocytic AML. In a xenograft model, LILRB4-directed CAR-T-cells significantly reduced AML burden and improved survival, although complete remission was not reached [57]. Importantly, no evidence of HSC targeting or depletion of mature blood cells was observed, supporting the safety of this approach. A clinical trial investigating anti-LILRB4 CAR-T therapy in relapsed/refractory monocytic AML is currently ongoing (NCT04803929). GRP78 78 kDa glucose-regulated protein (GRP78) is an intracellular chaperone linked to drug resistance through inhibition of apoptosis [58]. During endoplasmic reticulum stress, GRP78 translocates to the cell surface, making it targetable by CAR-T-cells. This stress response is typically seen in tumor cells, making GRP78 a tumor-specific target [59, 60]. Importantly, GRP78 is absent on HSCs and healthy hematopoietic cells, indicating an excellent toxicity profile [59]. However, its expression on AML cells is limited: while >75% of patients are GRP78-positive, mean blast expression ranges from 15% to 37%, and LSC expression varies between 2% and 51% [58–60]. Strategies to upregulate GRP78 on AML cells may therefore be required to achieve therapeutic efficacy. Two CAR constructs targeting GRP78 have been reported. A peptide-based CAR prolonged survival in a NOD scid gamma mouse model without any toxicity, although all mice eventually died due to limited CAR-T persistence [59]. A second approach combined GRP78 with CD123 in a multispecific CAR, which outperformed monospecific constructs in vivo and again showed no apparent toxicity [60]. Currently, no clinical trials of GRP78-directed CAR-T therapy are underway. CD70 CD70 is a member of the tumor necrosis factor (TNF) superfamily and plays a role in T-cell activation through interaction with CD27 [61]. Upregulation of CD70 has been associated with tumor proliferation and immune evasion in various solid and hematological cancers [61, 62]. In AML, CD70 is expressed in virtually all patients [3, 62, 63]. At the cellular level, expression on blasts is highly variable, ranging from 29% to 86%, with similar heterogeneity observed on LSCs [3, 62–64]. Hypomethylating agents can upregulate CD70 expression on AML cells and LSCs [61], providing a rationale for combining azacitidine with CD70-targeted immunotherapies. CD70 is absent on HSCs and most non-hematopoietic tissues, with only minimal expression reported in the gut [3]. However, its presence on activated T-cells introduces the risk of fratricide during CAR-T-cell manufacturing and treatment [65]. Sauer et al. [65] developed two CAR constructs targeting CD70: one using a conventional single-chain variable fragment (scFv) and another employing the extracellular domain of CD27 receptor as the antigen-binding moiety. Both constructs effectively eliminated leukemic blasts in mouse models while sparing HSCs, although a reduction in activated T-cells was observed. Several phase 1 clinical trials evaluating CD70-targeted CAR-T-cells in AML are currently ongoing (NCT04662294, NCT06326463, NCT06492304, NCT06345027), but results have not yet been reported. CD37 CD37 is a member of the tetraspanin superfamily and is predominantly expressed in the B-cell lineage [36, 37]. Although its precise function remains unclear [36], high CD37 expression in AML has been associated with poor prognosis [37]. In a study involving 25 AML patients, CD37 was detected in all samples, with an average 80% expression on blasts [66]. In healthy cells, CD37 shows B-cell expression and minimal to no expression on HSCs [67], indicating a safety profile similar to CD19-directed CAR-T therapies [37, 66]. Nevertheless, low-level expression has been reported on T-cells, monocytes, and neutrophils, although findings are inconsistent across studies [66]. In non-hematopoietic tissues, CD37 expression is limited to resident lymphocytes and not observed in parenchymal cells [67]. Originally developed for B-cell and T-cell lymphomas [36], CD37 CAR-T therapy has advanced to early clinical evaluation. In AML, Caulier et al. [66] showed that CD37 CAR-T-cells can effectively target AML cells in vitro and in xenograft models, without toxicity to HSCs or mature blood cells. ITGB2 Integrin beta chain-2 (ITGB2) is a cell adhesion molecule that forms a heterodimer with CD11a to bind ICAM-1 [68]. In AML, ITGB2 is highly expressed across all FAB subtypes, except for M3, with mean expression levels ranging from 72% to 100% [20, 63, 68]. On leukemic blasts, expression rates between 79.1% and 82.1% have been reported [63, 68]. ITGB2 is present on LSCs and absent on HSCs [18]. However, its broad expression on mature blood cells, including monocytes, neutrophils, NK-cells, and T-cells, significantly limits its use, resulting in a 0/4 score for mature blood cell selectivity [69]. To address this challenge, researchers explored whether distinct conformational states of ITGB2 could provide a more tumor-restricted target [69]. Their analysis revealed that the active conformation of ITGB2 is absent on most mature blood cells, except for activated T-cells. Functional validation of ITGB2-specific CAR-T-cells hence required ITGB2 gene knockout. These modified CAR-T-cells improved survival in AML patient-derived xenografts with efficacy comparable to CD33 CAR-T-cells, while preserving mature hematopoietic cells [69]. Building on these results, ITGB2 CAR-T-cells were combined with CD70 CAR-T-cells in a dual-targeting approach [63], which broadened AML cell recognition without increasing off-tumor toxicity. No clinical trials evaluating ITGB2 CAR-T-cells have been registered. TIM-3 T-cell immunoglobulin and mucin domain-containing protein 3 (TIM-3) is expressed on various immune cells, where it plays a role in maintaining immune tolerance [17, 70]. Consequently, TIM-3 is commonly used as a marker of T-cell exhaustion [70]. TIM-3 is highly expressed in AML patients [17, 71], yet its expression on AML blasts appears relatively low [32, 71]. Importantly, TIM-3 is not expressed on HSCs and granulocytes [17, 70], but is present on different T-cell subsets [71]. Regarding LSCs, expression is heterogeneous [70]. TIM-3-directed CAR-T-cells demonstrated potent antileukemic activity in vitro and in murine models, including effective elimination of LSCs [72]. Another approach involved a bispecific CAR targeting TIM-3 and CD13, demonstrating potent AML cell eradication combined with successful sparing of HSCs [73]. A phase 1/2 clinical trial is ongoing, evaluating anti-TIM-3/CD123 CAR-T therapy in AML patients (NCT06125652). Mesothelin Mesothelin is typically expressed in mesothelial tissues (pleura, pericardium, and peritoneum) and is strongly associated with several solid tumors [74, 75]. Although its precise biological function remains unclear, it is thought to play a role in cell adhesion [74, 75]. Its absence on HSCs and mature blood cells supports its potential as a safe therapeutic target [74, 76]. In AML, mesothelin expression has been investigated in pediatric cohorts and appears modest overall. It is detected in approximately 40% of AML patients with variable expression on AML blasts across studies [32, 74, 76]. Notably, mesothelin expression appears higher on LSCs, with a reported mean expression of 67.1% [76]. Mesothelin-directed CAR-T-cells demonstrated potent anti-leukemic activity both in vitro and in vivo [76]. In mouse xenografts, mesothelin CAR-T therapy successfully eradicated LSCs without affecting healthy HSCs or progenitor cells [76]. However, given its restricted expression to a subset of AML patients, accurate patient selection will be critical for successful clinical application. Conclusions and future perspectives The identification of CAR targets that balance broad applicability with acceptable hematopoietic safety remains a central challenge in AML. This review provides the first comprehensive and systematic evaluation of all AML-associated antigen targets reported in anti-AML CAR-T designs to date. By applying a structured 20-point scoring framework across 5 predefined criteria —broad coverage across patients, uniform expression on AML cells, presence on LSCs, absence on HSCs, and minimal or no expression on mature blood cells— we offer an evidence-based roadmap for rational CAR-T target selection in AML. From 63 reported CAR targets, we identified 13 novel antigens that best fulfill these criteria. While this framework provides a robust foundation to guide the development of safer and more effective CAR-T therapies for AML, several limitations should be acknowledged. First, high antigen density on AML cells has been proposed as an additional criterion to ensure adequate CAR-T-cell activation [3, 24]. We chose not to include this parameter in our framework due to the limited availability of quantitative surface expression data and the challenges associated with defining standardized thresholds for scoring. Importantly, low antigen density may be relevant for several targets identified here, such as CLL-1 and CD70 [77]. Advances in CAR construct design may help overcome this limitation. For example, incorporation of a high-affinity, membrane-proximal epitope-targeting antigen-binding domain markedly enhanced the capacity of CD33-targeted CAR-T-cells to eradicate AML cells with low antigen density [78]. Costimulatory domain selection may further modulate antigen sensitivity, with CD28‑based CARs often demonstrating greater responsiveness at lower antigen densities compared with 4‑1BB‑based constructs [79]. Furthermore, CARs generally require substantially higher antigen density than native T‑cell receptors (TCR) to trigger effective signaling, motivating the development of TCR‑inspired designs such as HLA‑independent TCR (HIT) constructs [80]. Second, our analysis was restricted to antigens for which CAR constructs have already been developed and published. As a result, potentially promising targets —such as CD180 and CD206 [81], or the LSC marker CD200 [82]— were not included in this work. In addition, among the 63 analyzed antigens, 15 were assigned a score of 13–15; however, several of these showed incomplete or conflicting information, particularly regarding LSC and HSC expression (see Supplementary Table 2, highlighted in yellow). This may have potentially excluded targets that could have been strong candidates if more comprehensive data (e.g., via cell surfaceomics) were available. Third, even when antigens do not fully align with the optimal selection framework, they may still warrant consideration for therapeutic application. Several strategies are being explored to enhance the clinical relevance of such targets. For antigens with intermediate AML cell surface expression, pharmacological agents can upregulate antigen density on AML cells and LSCs [61], potentially improving their suitability for CAR-T therapy. For example, CD44v6 expression increases on blasts following hypomethylating therapy [83]. Because AML is a notoriously heterogeneous disease, some targets—even some of the highest-scoring antigens in our analysis, such as MUC1 and LILRB4—are not universally expressed but instead restricted to specific subtypes. While this limits their applicability for a pan-AML CAR-T approach, it makes them excellent candidates for subtype-specific CAR-T strategies, particularly given their minimal expression on healthy cells [42, 44, 46, 54, 55]. The same applies to LILRB1, which had a score below 16/20 in our analysis, but may nevertheless be particularly suited as a target in monocytic AML [84]. In AML, identifying targets that are exclusively expressed on AML cells and LSCs, while being completely absent from HSCs and normal blood cells, remains inherently challenging [4, 17, 25, 26]. Selectivity can be enhanced by targeting epitopes that appear only in cancer-specific conformations, as illustrated by the activated form of ITGB2 [63, 69]. As shown in Table 1, clinical trials targeting CLL-1, CD33, and CD123 have reported considerable hematotoxicity, often requiring rescue with allo-HSCT. Because allo-HSCT not only restores hematopoiesis but is also an established curative option in AML, these findings suggest that such CAR-T therapies may be optimally deployed as a bridge-to-transplant rather than as a standalone approach. Similar considerations may apply to other antigens with broad hematopoietic expression listed in Supplementary Table 2, including CD84 [85]. Next to administering CAR-T-cells before allo-HSCT, the opposite sequence - performing allo‑HSCT prior to CAR‑T therapy - also emerges as a strategy that could expand the clinical utility of CAR‑T therapy in AML. Koehne et al. successfully performed allo-HSCT with tremtelectogene empogeditemcel, a CRISPR/Cas9-edited CD33-negative allograft, followed by CD33-directed CAR-T-cell therapy, enabling sparing of donor HSCs from on-target toxicity and allowing selective targeting of residual AML cells [86]. Finally, some targets, such as CD7, pose unique challenges due to their expression on T-cells (Table 1), creating a risk of CAR-T-cell fratricide [4, 65]. This can be effectively avoided through genome editing approaches, such as CRISPR/Cas9 or base editing, to selectively delete the target antigen in CAR-T-cells [30, 63, 69, 87]. A major consideration in the development of effective CAR-T therapies for AML is the profoundly immunosuppressive bone marrow microenvironment, which is shaped not only by leukemic cells but also by inhibitory immune populations such as bone marrow–resident MDSCs. These cells can substantially impair CAR-T-cell function, supporting the rationale for combination strategies incorporating immune-modulating agents. For example, stimulation with the poly I:C ligand has been shown to reduce MDSC abundance and enhance CAR-T-cell activity [88], while CD47 blockade promotes macrophage-mediated phagocytosis of AML cells [89]. As discussed above, LILRB4 is expressed on both AML cells and MDSCs and plays a key role in establishing an immunosuppressive bone marrow microenvironment; hence, targeting LILRB4 may combine direct antileukemic effects with the reversal of MDSC-driven immune suppression [53]. A final challenge in CAR-T therapy, which cannot be resolved through antigen selection criteria alone, is the well-recognized phenomenon of target antigen loss or downregulation on tumor cells [25, 26]. To address antigen loss, overcome antigen heterogeneity and enable more selective recognition of LSCs and AML blasts while sparing HS(P)Cs and mature blood cells, combinatorial targeting - including bispecific/multispecific CARs and logic‑gated CARs - is being actively pursued and has progressed to early clinical evaluation in AML [25, 26, 90]. Among the 13 antigen candidates in this review, ADGRE2/CLL-1 [25], GRP78/CD123 [60], CD70/CD33 [91], CD70/ITGB2 [63], TIM-3/CD13 [73, 92], and mesothelin/CD33 [93] have been tested. As illustrated earlier, combining ADGRE2 and CLL-1 in a CAR-CCR construct prevented antigen escape, improved antileukemic activity and markedly reduced “on-target/off-tumor” hematotoxicity [25]. Our analysis provides a structured basis to support rational antigen selection and the design of new multi‑antigen combinations for future CAR-T‑cell products in AML. In conclusion, this review highlights 13 antigen targets that currently represent the most compelling candidates for advancing CAR‑T therapy in AML. These targets warrant rigorous and systematic evaluation to determine whether their promising preclinical profiles, characterized by broad leukemic coverage and hematopoietic safety, can be translated into meaningful clinical benefit. Prioritizing these candidates for early‑phase testing, and particularly evaluating them in multispecific CAR designs, may be key to unlock the full therapeutic potential of CAR-T-cell therapy in AML. Supplementary information Supplementary Table 1 Supplementary Table 2 Supplementary Table 3 Acknowledgements DF was funded by an Emmanuel van der Schueren starters grant from Stand up to Cancer, Belgium, followed by a DOCPRO4 PhD grant of the Special Research Fund of the University of Antwerp, Belgium (BOF; GRANT MULTPLEX). LK is supported by a grant from Flanders Innovation and Entrepreneurship (VLAIO; HBC.2024.0265). SA is a Senior Clinical Investigator of the Research Fund - Flanders, Belgium (FWO; 1806225N). The other authors do not have any funding to declare. Author contributions FVO and SA conceived this review. FVO reviewed the literature, prepared figures and tables, and wrote the original manuscript. DF, LK, SH, and SA edited and critically revised the manuscript, contributing to its final version. SA supervised the work. All authors read and approved of the final manuscript. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41408-026-01544-5. References