Acute myeloid leukemia (AML) is a heterogeneous group of hematopoietic clonal diseases characterized by the accumulation in the bone marrow (BM) and in the peripheral blood (PB) of immature myeloid cells with increased proliferation capacity and survival ability, resulting in bone marrow failure and increased hemorrhagic and infective risk [1]. AML is the most common type of acute leukemia in adults, its incidence increases with age, and despite a consistent proportion of patients initially responding to therapy, the long-term prognosis remains quite poor, as over 60% of patients will eventually relapse and die from the disease [2]. Over the past twenty years, the role of clonal evolution in the leukemogenic process has emerged, identifying many recurrent molecular abnormalities, refining risk across the cytogenetic alterations, and, at the same time, identifying potential therapeutic targets ^[3][4][3,4].

2. Leukemic Cells in Their Bone Marrow Microenvironment: How They Induce Drug Resistance and Escape Immune Response

A hierarchical organization of bone marrow leukemic cells resembling the normal hematopoietic system, with a small subset of leukemic stem cells (LSCs) on the top of the pyramid, continually replenishing the more mature bulky population, was first described 25 years ago by Bonnet and coll. ^[5][8]. As for the normal counterpart, leukemic stem cells are defined by their self-renewal capacity and their ability to initiate leukemia after serial transplantation in SCID mice ^[5][6][8,9]. Interestingly, LSCs seem to originate from the acquisition of driver mutations, not by normal immature hematopoietic stem cells (HSCs) but by committed progenitors. At least two distinct LSC populations have been identified in human AMLs: a more mature population, emerging from granulocyte-macrophage progenitors (GMPs), and a more immature population, from lymphoid-primed multipotent progenitors (LMMPs) ^[7][10]. The recent identification of the clonal hematopoiesis of indeterminate potential (CHIP) ^[8][11] and the recognized ability of more mature LSCs to revert to an immature state ^[9][12] have further complicated the scenario. However, the persistence of pre-leukemic hematopoietic clonal alteration in responding patients confirms the negative role of LSCs ^[10][13]. Although it is well known that LSCs are enriched within the CD34+CD38- fraction, a clear definition of the LSC phenotype does not exist. Compared to normal HSCs, LSCs may have higher expression of CD25, CD32, CD44, CD96, CD123, CD200, GPR56, N-cadherin, Tie2, TIM-3, CLL-1, c-MPL, and HDM2 ^{[11][12][13][14][15][16][17][18][19][20][21]}[14,15,16,17,18,19,20,21,22,23,24]. Unfortunately, the high inter- and intra-individual heterogeneity complicates their enumeration and sequential monitoring. Despite this, the negative prognostic role of LSC frequency at diagnosis or after therapy has been demonstrated by many studies, and different LSC prognostic scores have been obtained by gene expression profiles and whole exome plantation ^[22][23][24][25,26,27]. LSCs escape the effect of anti-leukemic drugs by nesting in the BM microenvironment in a quiescent state, and leukemic progeny gradually occupy the BM niche, converting it into a “leukemic niche”, able to support leukemic cells’ survival and proliferation while decreasing its capacity to maintain normal hematopoiesis ^[25][26][27][28,29,30]. It is interesting to underline that BM failure is not due to a reduction in HSCs but to its inability to produce an adequate number of progenitors as a consequence of the release by mesenchymal cells (MSCs) of hypoxia-associated molecules that increase stemness and prevent differentiation ^[28][31], mimicking the differentiation arrest that is the hallmark of the leukemia phenotype. The modifications of the BM niche by leukemic cells involve all the components of the microenvironment. The endosteal niche is reshaped by the loss of balance between bone formation and resorption as the consequence of the activation of the RANK/RANKL pathway, which promotes osteoclastogenesis and favors osteoclast survival ^[29][32]. Moreover, the secretion by LSCs of a bone morphogenic protein (BMP), of a negative regulator of osteogenesis (DKK), and of a chemokine that decreases osteocalcin and switches MSC differentiation from adipogenic to osteoblastic with the consequent accumulation of progenitor and immature osteoblasts and defects in bone mineralization (CCL3, also known as MIP1α) ^[30][31][33,34], contribute to creating a milieu of facilitating leukemia cell growth and AML progression. Furthermore, in vitro data suggest that MSCs also support leukemia cell survival and promote resistance to chemotherapy, inducing a low cycling rate and anti-apoptotic signals ^[32][33][35,36]. In the AML vascular niche, increased IL-1β and TNF-α levels synergize with increased adhesive receptors E-selectin/CD44 and VCAM-1/VLA-4, thus ensuring the anchorage of LSCs to endothelial cells ^[34][35][37,38]. VEGF/VEGFR and Notch/Delta-like ligand 4 promote neo-angiogenesis, essential for AML progression and extramedullary homing ^[36][37][39,40]. Although less known compared to the other, the reticular niche is considered a transitional niche, able to maintain HSCs in a proliferative state, such as the vascular, but in an undifferentiated state, such as the endosteal one. CXCL12-abundant reticular (CAR) cells and Nestin-expressing cells are the most represented cells in the niche, localized in proximity to sinusoidal endothelial cells ^[38][39][40][41,42,43]. Most HSCs are in contact with CAR cells, but interaction with these cells is established also by B-lymphocytes, plasma cells, plasmacytoid dendritic cells, and NK cells, suggesting that reticular cells might also act as an immune cells niche ^[41][42][43][44,45,46]. CAR cells secrete CXCL12 and SCF, crucial for HSCs trafficking and homing. In AML, the CXCL12/CXCR4 axis regulates the infiltration of leukemic cells in the protective bone marrow niche ^[44][45][47,48]. Nestin-expressing cells are associated with adrenergic nerves and regulate HSC maintenance ^[40][46][43,49]. AML bone marrow is enriched with Nestin-expressing cells. Besides their role in developing AML, Nestin-expressing cells are implicated in the induction of resistance to chemotherapy by enhanced glutathione (GSH)-peroxidase (Gpx) activity ^[47][50].Considered in the past only as passive bystanders, BM adipocytes actually play an important role in regulating normal HSCs and are implicated in hematopoietic recovery after irradiation or 5FU treatment by secreting SCF ^[48][51]. In AML, they promote leukemic cell survival during therapy and drug resistance by enhancing free fatty acid production ^[49][50][52,53] and by sequestering, inactivating, and metabolizing chemotherapeutic drugs ^[51][54]. Among cells from myeloid resident in the bone marrow, myeloid-derived suppressor cells (MDSCs) and leukemia-associated macrophages (LAMs) are also implicated in generating an immunosuppressive microenvironment in AML. MDSCs are a heterogeneous population, able to induce T-cell tolerance by multiple mechanisms, still not completely elucidated. It is well known that MDSCs express high levels of V-domain Ig suppressor of T-cell activation (VISTA) and PD-L1, but also indoleamine 2,3-dyoxigenase (IDO), arginase, ROS, TGFβ, and IL-10 ^{[52][53][54][55]}[55,56,57,58]. 

3. Monoclonal Antibodies

3.1. Potential Targets on LSCs

The main mechanism of action of unconjugated antibodies is antibody-dependent cell-mediated cytotoxicity (ADCC). After the formation of an immunological synapse with the target cell, NK cells trigger a cytolytic response through the exocytosis of granules containing perforin and granzyme into the target cell. Furthermore, they facilitate antibody-dependent phagocytosis (ADCP) by macrophages. Their major limitation is the presence of inhibitory signals counteracting their action and the limited potency of activation signals.

3.1.1. CD33

CD33 is a 67 kDa glycoprotein, a member of the siglec family (siglec-3), expressed in normal hematopoiesis from early myelo-monocytic lineage-committed progenitors to mature cells. It is also expressed in 99% of AML cells and in LSCs ^[56][79]. The use of unconjugated anti-CD33 is challenging due to its low membrane density and the low antibody-induced internalization, such that the first phase 3 trial (NCT0006045) with humanized IgG1 unconjugated antibody anti-CD-33, lintuzumab (HuM195) was stopped due to lack of efficacy ^[57][80]. Despite this, four phase I/II trials are ongoing to test a new anti-CD33 unconjugated IgG1 (BI 836858), alone or in combination with azacytidine (AZA), decitabine (DAC), or F16-IL2, in relapsed/refractory and newly diagnosed AMLs (NCT01690624, NCT03013998, NCT02032721, NCT03207191). The principal advantage of antibody–drug conjugates is the simultaneous selective attack of target cells and the on-site delivery of the potent conjugated cytotoxic agent, not usable as free drugs in conventional chemotherapy schemes for their toxicity. After the spontaneous withdrawal of the manufacturer due to the lack of survival improvement and the high induction mortality, the anti-CD33-calicamycin conjugate gemtuzumab-ozogamicin (GO) has been revived for core binding factor (CBF) AMLs and for elderly patients with AML or high-risk myelodysplastic syndrome (MDS), where it has demonstrated reduced relapse risk and survival advantage in patients with favorable and intermediate cytogenetic risk ^[58][81].

3.1.2. CD123

CD123 is the interleukin (IL)-3 receptor alpha chain and is a type I transmembrane glycoprotein ^[59][83]. CD123 is present in 98% of CD34+CD38- LSCs and on blast cells but not in normal hematopoietic progenitors, making it a potential therapeutic target ^[14][17]. The unconjugated antibody (talacotuzumab) failed to demonstrate efficacy in AML and MDS (NCT0299860). The phase I/II trial with anti-CD123-DGN462 conjugate, alone or with AZA± VEN in MRD+ post-induction therapy AMLs, is still recruiting patients (NCT04086264).

3.1.3. CLL-1

C-type lectin-receptor1 (CLL-1) is an ITIM-containing inhibitory transmembrane protein code on chromosome 12p13.31. CLL-1 is not expressed in normal HSCs, but it is present in committed progenitors and in mature peripheral myeloid cells such as monocytes, granulocytes, and dendritic cells ^[60][84].  Being selectively expressed in LSCs, even more than CD123, which is also present in some CD34+CD38- normal progenitors, CLL-1 could be regarded as an ideal target for anti-LSC immune therapies ^[19][61][22,86]. Despite CLL-1 efficiently internalizing after ligand binding, in vitro experiments with an anti-CLL-1 antibody suggested that an un-conjugated antibody cannot have anti-leukemic activity because it does not activate ADCC ^[19][22].

3.1.4. Other Current Clinical Trials of Toxin-Conjugate Antibodies

FLT3 (FMS-like tyrosine kinase 3) is a member of the class III receptor tyrosine kinase family that is highly expressed in the blasts of both AML and ALL patients. FLT3 plays an important but not absolute role in maintaining the survival of normal HSCs, and its recurrent mutations (ITD-FLT3, TDK-FLT3) are expressed in many AML cases ^[62][63][89,90]. Working in conjunction with other growth factors, FLT3 promotes the proliferation and differentiation of myeloid and lymphoid cells. Transplant animal models with non-functioning and wild-type FLT3 showed that hematopoiesis is almost normal in FLT3 knockout animals, while FLT3 mutations give a significant growth advantage. These facts suggest that selective FLT3 inhibition in leukemia cells can block excessive FLT3 leukemia activation with acceptable hematopoietic side effects ^[63][90]. The phase I trial (NCT02864290) testing AGS62P1 anti-FLT3 antibody-amberstatin 269 in relapsed/refractory adult AMLs has been closed for lack of efficacy. NCT03957915, a phase I trial with INA03 drug-conjugate antibody-targeting transferrin receptor (CD71), is active, not recruiting, and NCT01830777, testing brentuximab-vedotin in CD30+ relapsed AMLs, is completed (results not yet available).

3.2. Immune Checkpoint Inhibitors

The immune system activity is closely modulated by the interaction between co-inhibitory molecules and their ligands. These co-inhibitory molecules are involved in the maintenance of immune tolerance, but in the neoplastic setting may represent one of the mechanisms employed by cancer cells to escape immune surveillance ^[64][65][92,93]. Many studies have reported an increased expression of co-inhibitory molecules in solid cancers, and, more recently, similar results have been observed in AML patients ^[66][67][94,95]. The up-regulation of co-inhibitory ligands has been associated with poor clinical outcomes in solid and hematological cancers ^[65][66][93,94], thus identifying a potential new class of therapy targets (Figure 12).

3.2.1. The CTLA-4/B7 Axis

CTLA-4 (CD152) is a member of the immunoglobulin-related receptors interacting with CD80 and CD86 ligands to deliver an inhibitory signal to terminate immune responses. Furthermore, it is involved in the induction of Tregs lymphocytes, playing a pivotal role in regulating tolerance and autoimmunity ^[68][97]. Aberrant expression has been reported in AML, often in association with the expression of other checkpoint inhibitors, and with a negative impact on disease outcome ^[69][98].

3.2.2. The PD1/PD-L1 Axis

PD1 (CD279) is a type I transmembrane protein preferentially expressed in almost all activated immune cells ^[64][92]. PD1 binds two different ligands: PD-L1 (CD274) and PD-L2 (CD273). PD-L1 is a member of the B7 family of co-stimulatory/co-inhibitory molecules also present in hematopoietic cells and upregulated or aberrantly expressed in many tumors ^[64][70][92,104]. PD-L1 and PD-L2 over-expression in tumor cells has been associated with poor outcomes ^[66][94]. PD1/PD-L1 engagement produces an inhibitory signal that causes T-cell exhaustion and favors tumor cell immune escape ^[64][92]. Moreover, it induces the apoptosis of tumor-specific cells and favors Tregs differentiation and resistance to CD8+-mediated cytolysis ^[71][105]. In AML, PD1 up-regulation was described in T effector cells and Tregs, explaining the immune suppression observed during AML progression. PD-L1 and PD-L2 were found in blast cells and correlate with poor prognosis ^[72][73][106,107].

3.2.3. Tim-3/Galectin-9 Axis

T-cell immunoglobulin and mucin domain 3 (TIM-3) is a cell-surface molecule first identified on CD4+ TH1 cells and on CD8+ cytotoxic T cells (CTLs) and later on other innate immune cells, such as dendritic cells, monocytes, macrophages, mast cells, NK cells, and in many neoplastic cells ^[74][75][76][111,112,113]. The TIM-3 gene is coded on chromosome 5q33.2, in the same region of IL4 and IL-5 genes. There are four known ligands for TIM-3: the first recognized and most widely studied is galectin-9 (gal-9), which induces the apoptosis of TH1 cells ^[77][114], playing a crucial role in tumor cell immune evasion. Its high expression has been associated with a worse prognosis in solid and hematologic neoplasms ^[64][66][92,94]. In AML, high levels of TIM-3 have been described in immune cells, associated with immune exhaustion, and on LSCs, where it represents a distinctive marker.

3.2.4. LAG-3/MHC Axis

LAG-3 (CD223) is a CD4-like molecule that binds MHC class II with higher affinity compared to CCD4, generating a signal that blocks T-cell activation ^[78][100]. The molecule is expressed on activated T cells, Tregs, NK cells, B cells, and dendritic cells ^[66][94]. The engagement with its ligand reduces T-cell activity and cytokine secretion, blocking T-cell activation and function ^[78][100]. Like other checkpoint molecules, LAG-3 has been identified on Tregs in the cancer microenvironment. The frequent co-expression with PD1 suggests a comparable function to PD1 ^[71][79][101,105].

3.2.5. CD200/CD200R Axis

CD200 is a highly conserved 48kDa type-1 transmembrane glycoprotein, structurally related to the B7 family, coded on chromosome 3q12-q13 near the region coding for CD80/CD86 proteins. Interaction with its ligand CD200R leads to the attenuation of many immune-responsive effects, resulting in the prolonged survival of transplanted allographs but also tolerance for tumor cells. CD200 overexpression has been described in several solid tumors and in AML, where it marks LSCs but not the normal HSC counterpart ^[15][66][18,94]. CD200 overexpression in AML has been associated with poor outcomes, especially in the favorable prognostic groups ^[80][81][117,118].

3.2.6. CD47/SIRPs Axis

Leukocyte surface antigen CD47, encoded on chromosome 3q13.12, is a transmembrane glycoprotein that serves as the ligand for signal regulatory protein alpha (SIRPα), which is expressed on phagocytic cells, including macrophages and dendritic cells. Upon activation, CD47 initiates a signal transduction cascade, resulting in the inhibition of phagocytosis ^[82][123]. CD47 is broadly expressed in a variety of normal tissues, in normal circulating HSCs, as well as leukemic cells ^[83][124]. Moreover, its expression in leukemia cells has an adverse prognostic effect ^[84][125]. Preclinical studies demonstrated that a block of CD47 exerts an anti-neoplastic effect on solid tumors and hematologic malignancies ^[85][86][126,127].

3.3. T Cells Engagers

Given the unsatisfactory results of “naked” and drug-conjugate antibodies in treating AML, over the past years, combination strategies to overcome tumor cell and immune cell resistance have been developed. The purpose was to harness the immune response by targeting specific tumor antigens and engaging the dormant effector cells (T cells, NK-cells, macrophages) against leukemic cells to obtain a fast and potent cytotoxic response and, ideally, also generate a durable immunologic memory ^[87][130]. On this basis, several bispecific T-cell engager (BiTE) antibodies have been developed for hematologic malignancies, and novel structures are constantly emerging ^[88][131]. Bispecific antibodies consist of a single heavy and light chain of the variable region of a tumor-associated antigen and CD3: by combining the two different specificities, they can activate exhausted T cells through sustained tumor antigen exposure, overcoming their possible low-density expression. At the same time, as the T cells are activated only in the presence of target cells, they have limited off-target cytotoxicity. Moreover, BiTEs act in an MHC-1-independent manner, generating a cytolytic synapsis between CD8+ T cells despite the tumor-induced MHC-1 down-modulation, and permits T-cell activation even in the absence of co-stimulatory signals such as CD28 or IL-2 ^[89][132].

3.3.1. Anti-CD33

The CD33 was integrated with an AMG330 bispecific antibody. Pre-clinical studies demonstrated potent T-cell activation and cytokine release, high blast cell clearance, and a reduction in BM monocytic MDSCs ^[90][91][135,136]. Moreover, AMG330 lytic activity was not affected by CD33 polymorphisms nor by the over-expression of drug transporter proteins on AML cells, as happens for gemtuzumab ozogamicin, and it does not modify CD33 membrane density, as conventional anti-CD33 antibody does ^[92][137].

3.3.2. Anti-CD123

Vibecotamba (XmAb14045), is a potent CD3-CD123 bispecific antibody, showing in preclinical models high anti-leukemic activity on PB and BM blasts and a long half-life ^[93][141]. The initial results of a phase I dose-escalation study in R/R AML (NCT 02730312) enrolling 104 patients showed an ORR of 14%, a CR rate of 4%, and stable disease in 71% of patients. CRS was observed in 59% of patients. CD123 expression intensity on AML blasts seemed not to affect response ^[94][142]. APVO436, another CD3-CD123 bispecific antibody, is currently in a clinical trial on AML and MDS (NCT03647800). One of its peculiar features seems to be a low induction of cytokine release while maintaining the ability to reactivate T cells ^[95][143]. However, preliminary results from the ongoing trial suggest a limited efficacy ^[96][144].

3.3.3. Other Targets

MCLA-117 is a CD3-CLEC12A (CLL-1) bispecific antibody currently under investigation in a phase I trial of R/R AML and newly diagnosed elderly AML patients (NCT03038230). In vitro studies demonstrated its ability to lyse cells at a low E:T ratio, even blast cells with low CLL-1 expression ^[97][150]. Its extended half-life permits weekly or bi-weekly infusions ^[98][151]. CLEC12a TriKEs have been developed to target LSCs lacking CD33 expression, but expressing CLL-1, demonstrating lower off-target toxicity and similar lytic capacity compared to the CD33 target, preserving normal HSCs ^[99][152]. AMG427 is an anti-CD3-FLT3 (CD135) bispecific antibody constructed to target the same receptor expressed from leukemic cells. It has demonstrated in vitro T-cell-dependent cytotoxicity but also the increased production of PD1 expression, which can act as an immune escape mechanism without the concurrent administration of anti-PD1 therapy ^[100][153].

4. T- and NK-Cell Therapies

CAR-T cells are engineered autologous peripheral T cells equipped with a synthetic target-antigen receptor (CAR), able to expand after transfusion in a target-antigen-dependent matter and designed to persist after infusion, eventually generating a long-term anti-leukemic memory. Binding CAR to the target antigen initiates intracellular signaling, leading to TCR-independent T-cell activation. CAR’s structure consists of four main components, each of which can affect CAR-T functionality: (1) an extracellular target-antigen-binding domain, (2) a hinge region, (3) a transmembrane domain, and (4) one or more intracellular signaling domains. Extracellular domain structure is crucial not only for antigen recognition but also for the affinity and specificity of binding. Too low or too high affinity may result in the activation-induced death of CAR-T cells and trigger toxicities. The hinge region confers flexibility, allowing the antigen-binding domain to access the target epitope and form immunological synapses. The most-used hinge regions are derived from aminoacidic sequences from CD8, CD28, IgG1, and IgG4. The transmembrane domain influences CAR expression level and stability. Most transmembrane domains are derived from natural proteins including CD3ζ, CD4, CD8α, and CD28. A CD3ζ transmembrane domain probably facilitates T-cell activation, mediates CAR dimerization, and incorporates endogenous TCRs ^[101][155]. However, a CD3ζ domain is less stable than the CD28 one. The hinge region and transmembrane domain influence CAR-T cell cytokine production and activation-induced cell death (AICD), while intracellular domains influence the activation magnitude and persistence of CAR-T cells. The first generation of CAR-T cells has a single signal domain (CD3ζ), while the newer generations have additional costimulatory structures, such as CD28 or CD137 and 4–1BB for 2nd generation CAR-T and CD28plus, CD137, or CD134 (OX40) for the 3rd generation CAR-T ^[102][156].