Immunotherapy-Based Targeting and Elimination of Leukemic Stem Cells in AML and CML

The concept of leukemic stem cells (LSC) has been developed with the idea to explain the clonal hierarchies and architectures in leukemia, and the more or less curative anti-neoplastic effects of various targeted drugs. It is now widely accepted that curative therapies must have the potential to eliminate or completely suppress LSC, as only these cells can restore and propagate the malignancy for unlimited time periods. Since LSC represent a minor cell fraction in the leukemic clone, little is known about their properties and target expression profiles. Over the past few years, several cell-specific immunotherapy concepts have been developed, including new generations of cell-targeting antibodies, antibody–toxin conjugates, bispecific antibodies, and CAR-T cell-based strategies. Whereas such concepts have been translated and may improve outcomes of therapy in certain lymphoid neoplasms and a few other malignancies, only little is known about immunological targets that are clinically relevant and can be employed to establish such therapies in myeloid neoplasms. In the current article, we provide an overview of the immunologically relevant molecular targets expressed on LSC in patients with acute myeloid leukemia (AML) and chronic myeloid leukemia (CML). In addition, we discuss the current status of antibody-based therapies in these malignancies, their mode of action, and successful examples from the field.


Introduction
Acute myeloid leukemia (AML) and chronic myeloid leukemia (CML) are stem cell-derived, life-threatening, hematopoietic neoplasms that are characterized by an uncontrolled expansion of myeloid progenitor cells exhibiting a more or less severe maturation defect. The clinical course, prognosis, and progression patterns vary among patients, depending on the genetic background, age, the sub-type and phase of the disease, cytogenetic and molecular features, and the patient's response to initial anti-leukemic (induction) therapy [1][2][3][4][5][6][7][8].
Whereas in CML the disease is characterized by the pathognomonic BCR-ABL1 oncogene, the molecular landscapes in AML are complex and involve several different oncogenes and a plethora of somatic mutations [4][5][6][7][8][9]. By applying intensive chemotherapy and oncogenic driver-specific drugs in AML and BCR-ABL1-targeting compounds in CML, a majority of all patients achieve remission, and in many cases, long-term disease-free survival is achieved [1][2][3][4][5][10][11][12][13]. However, not all patients have a good response to such therapy, or they relapse after having achieved remission. In patients with multi-resistant disease, hematopoietic stem cell transplantation (HSCT) is usually recommended, but the procedure can only be performed in a limited number of 'young' and fit patients, and carries an inherent mortality risk. As a result, research in AML is currently focusing on new molecular targets and the establishment of more potent drug therapies, including targeted drugs and immunotherapies.
The basic theory of leukemic stem cells (LSC) has been created with the intention to explain cellular and molecular hierarchies and to improve anti-neoplastic treatment through the eradication of disease-initiating and disease-propagating cells [14][15][16][17][18][19][20][21][22]. The concept of LSC is based on the hypothesis that the leukemic clone and sub-clonal evolution are organized in a cellular hierarchy, with (a) more mature leukemic cells that disappear (through apoptosis) after a certain number of cell divisions, and (b) LSC that can augment the bulk population of leukemic cells indefinitely by their unrestricted (unlimited) self-renewing and long-term proliferative abilities [14][15][16][17][18][19][20][21][22]. In the chronic phase of BCR-ABL1 + CML and in some AML variants, LSC were reported to reside in a CD34 + /CD38 − subset of the leukemic clone [14][15][16][17][18][19][20]. However, depending on the molecular background and the phase of the disease, at least some LSC may also be detected within a CD34 + /CD38 + subset of leukemic cells, or sometimes even in a CD34-negative cell population [23][24][25].
Based on their disease-initiating and disease-propagating capacity, LSC are regarded as a major, clinically relevant therapeutic cell target, and numerous studies have been conducted with the goal of identifying new molecular targets in these cells [17][18][19][20][21][22][26][27][28][29]. Of special interest are specific cell surface antigens that can be employed to develop disease-eradicating immunotherapies such as antibody-based or CAR-T cell therapies. However, only a few clinically relevant cell surface targets that are expressed specifically on LSC, but not on normal bone marrow (BM) stem cells, have been identified.
In the current article, we review the cell surface antigens that are expressed preferentially or even specifically on LSC in AML and/or CML, and thus represent potential targets for immunotherapies. In addition, we provide an overview of treatment concepts that have been or are currently being developed based on antigen expression and function in leukemic (stem) cells. Moreover, we discuss the current status of antibody-based therapies in AML. Finally, we discuss future developments in the field, and how LSC-targeting immunotherapies can be translated into clinical application.

Phenotype of LSC in AML and CML
The classical approach to demonstrate self-renewing and long-term disease-propagating abilities of LSC in vivo is to transplant leukemic cells into immunocompromised mice. Earlier studies employed severe combined immunodeficiency (SCID) mice or non-obese SCID (NOD/SCID) mice for long-term engraftment studies [14][15][16]. In these initial studies, the NOD/SCID mouse (long-term)-repopulating LSC in AML and CML were found to reside preferentially in a CD34 + /CD38 − subset of the leukemic clone [14][15][16]30]. Therefore, most data on LSC refer to these cells. However, it soon turned out that the residual immune system of NOD/SCID mice can eliminate CD38 + AML cells, and that complete immunosuppression enables CD38 + AML LSC to produce long-term engraftment in NOD/SCID mice [23]. Therefore, highly immunocompromised (and thus more permissive) mouse strains were employed and soon accepted as a new standard model of xenotransplantation in AML. One of the most frequently used strains is NSG, a NOD/SCID mouse model lacking a functional interleukin-2 receptor gamma chain. In most AML variants, the NSG mouse-repopulating AML LSC reside in both the CD34 + /CD38 − and CD34 + /CD38 + fraction of the leukemic clone [23]. Similarly, in the blast phase of Ph + CML, LSC appear to express CD38 [31], and the same holds true for Ph + and Ph − acute lymphoblastic leukemia (ALL) [32]. In other words, in acute leukemia models, LSC often reside in a 'progenitor cell-like', CD38 + , fraction of the leukemic clone. Another issue is the molecular and phenotypic complexity and the related sub-clone formation in myeloid leukemias, including AML [22,25,33]. Table 1. Expression of potential therapeutic targets on CD34 + /CD38and CD34 + /CD38 + cells in acute myeloid leukemia (AML) and chronic myeloid leukemia (CML) and comparison to stem cells in normal bone marrow (NBM) *.

Antigen CD
Antigen Expression on Stem/Progenitor Cells in ** n.c.
CD34 + /CD38 − AML LSC express higher levels of CD33 and CD123 compared to normal CD34 + /CD38 − BM stem cells, but the therapeutic window is rather small. In other words, at least some normal stem cells may also be eliminated by intensive treatment with CD33-targeted drugs or CD123-targeted drugs. Other surface targets such as CD44 or CD52 (Campath-1) are expressed abundantly on normal stem cells as well as AML LSC [35,40,46] (Table 1, Figure 1). In most patients with AML, the CD34 + /CD38 − LSC do not display CD26 (dipeptidyl-peptidase-IV = DPPIV) and CD90 (Thy-1) [29,47]. CD34 + /CD38 − BM stem cells, but the therapeutic window is rather small. In other words, at least some normal stem cells may also be eliminated by intensive treatment with CD33-targeted drugs or CD123targeted drugs. Other surface targets such as CD44 or CD52 (Campath-1) are expressed abundantly on normal stem cells as well as AML LSC [35,40,46] (Table 1, Figure 1). In most patients with AML, the CD34 + /CD38 − LSC do not display CD26 (dipeptidyl-peptidase-IV = DPPIV) and CD90 (Thy-1) [29,47]. Examples of expression of cell surface target antigens on CD34 + stem cells in normal bone marrow (NBM) and patients with acute myeloid leukemia (AML). Target expression on aspirated CD34 + /CD38 − cells (left panels, red histograms) and CD34 + /CD38 + BM cells (right panels, blue histograms) was determined by fluorochrome-conjugated antibodies (as depicted) and multi-color flow cytometry. Normal/reactive BM (NBM) was obtained from lymphoma patients without BM involvement or was purchased, and leukemic BM was obtained from three patients with AML. All patients gave written informed consent before BM aspiration was performed. The study was approved by the ethics committee of the Medical University of Vienna. Reactivity of the test antibodies (CD34 + /CD38 − stem cells: red histograms; CD34 + /CD38 + stem/progenitor cells: blue histograms) was assessed on a FACSCantoII (BD Biosciences). Antibody reactivity was controlled by isotype-matched control antibodies (open black histograms). Flow cytometry data were analyzed using FlowJo 8.8.7 software (Flowjo). Examples of expression of cell surface target antigens on CD34 + stem cells in normal bone marrow (NBM) and patients with acute myeloid leukemia (AML). Target expression on aspirated CD34 + /CD38 − cells (left panels, red histograms) and CD34 + /CD38 + BM cells (right panels, blue histograms) was determined by fluorochrome-conjugated antibodies (as depicted) and multi-color flow cytometry. Normal/reactive BM (NBM) was obtained from lymphoma patients without BM involvement or was purchased, and leukemic BM was obtained from three patients with AML. All patients gave written informed consent before BM aspiration was performed. The study was approved by the ethics committee of the Medical University of Vienna. Reactivity of the test antibodies (CD34 + /CD38 − stem cells: red histograms; CD34 + /CD38 + stem/progenitor cells: blue histograms) was assessed on a FACSCantoII (BD Biosciences). Antibody reactivity was controlled by isotype-matched control antibodies (open black histograms). Flow cytometry data were analyzed using FlowJo 8.8.7 software (Flowjo).
Finally, LSC in AML and CML display various immune checkpoint antigens, including the key checkpoint target PD-L1 (CD274) and the 'don't eat me' receptor IAP (CD47) ( Table 1) [61][62][63]. Whereas CD47 is expressed abundantly on LSC in a constitutive manner in most patients with AML and most with CML, the expression of PD-L1 is often weak or absent on LSC in these malignancies. However, the expression of PD-L1 can usually be induced or augmented substantially on LSC by exposure to interferon γ (IFN-G) ( Figure 2, Table 1).
Finally, LSC in AML and CML display various immune checkpoint antigens, including the key checkpoint target PD-L1 (CD274) and the 'don't eat me' receptor IAP (CD47) ( Table 1) [61][62][63]. Whereas CD47 is expressed abundantly on LSC in a constitutive manner in most patients with AML and most with CML, the expression of PD-L1 is often weak or absent on LSC in these malignancies. However, the expression of PD-L1 can usually be induced or augmented substantially on LSC by exposure to interferon γ (IFN-G) ( Figure 2, Table 1).
Gemtuzumab ozogamicin (GO) is a humanized anti-CD33 antibody conjugated to a cytotoxic, calicheamicin-type drug through a bifunctional chemical linker [64][65][66]. Previous studies have shown that treatment with GO is effective in chemotherapy-refractory and relapsed AML. Subsequently, the drug was approved by the FDA in 2000. However, in post-marketing studies, the clinical benefit of GO could not be confirmed in relapsed AML [73]. Moreover, veno-occlusive liver disease and prolonged cytopenia (neutropenia and thrombocytopenia) were observed after repeated exposure to GO (as adjunct to chemotherapy) [67][68][69][70][71][72][73]. As a result, Pfizer withdrew GO (voluntarily) from the market in 2010, which may have been a premature decision.
More recent data have shown that therapy with lower and thus less toxic GO doses improves overall survival (OS) in AML patients with favorable cytogenetics, but is less effective in patients with poor cytogenetic features [74][75][76]. In a meta-analysis calculating the outcomes in five controlled randomized trials, GO was found to reduce the risk of relapse and to improved OS in patients with favorable cytogenetics as well as in those with intermediate cytogenetics [76]. Moreover, GO has recently been described to improve outcomes in older AML patients in two Phase III trials [77]. In all studies, when tested, the clinical benefit was preferentially seen in patients in whom AML blasts (CD34 + cells) display CD33 [78]. Correspondingly, GO induces apoptosis and growth arrest in AML LSC [54] (Figure 3). In 2017, the FDA approved GO for the treatment of patients with newly diagnosed AML as adjunct to standard induction polychemotherapy [79]. During the past few years, several attempts have been made to develop additional and more potent CD33-targeting antibody-based drugs. SGN-CD33A (vadastuximab talirine) is a novel CD33targeted agent that contains a humanized anti-CD33 antibody with engineered cysteine residues, conjugated to a synthetic DNA-damaging agent (pyrrolobenzodiazepine dimer) through a proteasesensitive linker [80,81]. The antibody-conjugate is endocytosed by CD33 + AML cells and, once released, the DNA cross-linker delivers major cytotoxic effects [80]. Apparently, SGN-CD33A is a more potent agent against AML blasts compared to GO [80]. The drug exerts growth-inhibitory and apoptosis-inducing effects on chemotherapy-resistant AML blasts [81,82]. SGN-CD33A may be more effective than GO for several reasons. First, SGN-CD33A benefits from a novel linker technology that enables uniform drug loading of the antibody molecules. In addition, unlike GO, the DNA-damaging agent (a pyrrolobenzodiazepine dimer) is considered to deliver cytotoxicity in AML cells independent of multidrug resistance (MDR) antigens, such as MDR-1. Initially, it was also described that vadastuximab talirine is well tolerated in most patients when applied as a single agent [82]. In addition, the drug induced remission in patients with chemotherapy-refractory AML [83]. However, the therapeutic window regarding stem cell suppression was small, and when combined with cytotoxic drugs and polychemotherapy, hematologic toxicity was substantial and led to infectionrelated death, finally leading to the discontinuation of the Phase III trial and stopping of drug development [83].
A number of strategies have been developed to target the IL-3RA (CD123) on AML blasts and AML LSC. One promising strategy is to conjugate cytotoxic agents to IL-3. One such compound is the diphtheria toxin DT388/IL-3 fusion protein. This conjugate consists of the catalytic and translocation domains of diphtheria toxin coupled to human IL-3 [84][85][86]. A variant of this toxin conjugate (DT388 IL-3[K116W]) was obtained by fusing the diphtheria toxin with a modified IL-3 protein, which resulted in a stronger binding of IL-3 to the IL-3 receptor [87]. On a molar basis, DT388 IL-3[K116W] is more active than DT388 IL-3 in destroying leukemic cells [87]. Nevertheless, both DT388 During the past few years, several attempts have been made to develop additional and more potent CD33-targeting antibody-based drugs. SGN-CD33A (vadastuximab talirine) is a novel CD33-targeted agent that contains a humanized anti-CD33 antibody with engineered cysteine residues, conjugated to a synthetic DNA-damaging agent (pyrrolobenzodiazepine dimer) through a protease-sensitive linker [80,81]. The antibody-conjugate is endocytosed by CD33 + AML cells and, once released, the DNA cross-linker delivers major cytotoxic effects [80]. Apparently, SGN-CD33A is a more potent agent against AML blasts compared to GO [80]. The drug exerts growth-inhibitory and apoptosis-inducing effects on chemotherapy-resistant AML blasts [81,82]. SGN-CD33A may be more effective than GO for several reasons. First, SGN-CD33A benefits from a novel linker technology that enables uniform drug loading of the antibody molecules. In addition, unlike GO, the DNA-damaging agent (a pyrrolobenzodiazepine dimer) is considered to deliver cytotoxicity in AML cells independent of multidrug resistance (MDR) antigens, such as MDR-1. Initially, it was also described that vadastuximab talirine is well tolerated in most patients when applied as a single agent [82]. In addition, the drug induced remission in patients with chemotherapy-refractory AML [83]. However, the therapeutic window regarding stem cell suppression was small, and when combined with cytotoxic drugs and polychemotherapy, hematologic toxicity was substantial and led to infection-related death, finally leading to the discontinuation of the Phase III trial and stopping of drug development [83].
A number of strategies have been developed to target the IL-3RA (CD123) on AML blasts and AML LSC. One promising strategy is to conjugate cytotoxic agents to IL-3. One such compound is the diphtheria toxin DT 388 /IL-3 fusion protein. This conjugate consists of the catalytic and translocation domains of diphtheria toxin coupled to human IL-3 [84][85][86]. A variant of this toxin conjugate (DT 388 IL-3[K116W]) was obtained by fusing the diphtheria toxin with a modified IL-3 protein, which resulted in a stronger binding of IL-3 to the IL-3 receptor [87]. On a molar basis, DT 388 IL-3[K116W] is more active than DT 388 IL-3 in destroying leukemic cells [87]. Nevertheless, both DT 388 IL-3 and DT 388 IL-3[K116W] induce cytotoxic effects on IL-3R + cells in vitro and in vivo. The extent of cytotoxicity induced by these compounds is directly correlated with the level of IL-3R expressed on the surface of AML blasts [58,85]. Preclinical studies have shown that DT 388 IL-3 doses up to 100 µg/kg are well tolerated [86]. Based on these data, the related IL-3-toxin fusion protein SL-401 (tagraxofusp) was developed and evaluated in a Phase I clinical trial in heavily pre-treated AML patients (NCT 02270463): in a group of 70 AML patients, two complete responses and five partial responses were seen [87]. In a few other patients, stable disease was obtained and observed for more than one year, suggesting that some AML LSC had been eliminated [88]. Other studies have shown that tagraxofusp is active against LSC in patients with CML [89]. However, it remains unknown whether tagraxofusp will be further developed in the context of AML or CML [90]. More recent data suggest that tagraxofusp is an extremely active agent in plasmacytoid dendritic cell neoplasms [90,91], and in 2018, the drug received approval for this indication by the FDA [92].
Another principal way to target the IL-3R is to apply antibodies that bind to IL-3R with high affinity and thereby block the binding of IL-3 to this receptor. The 7G3 antibody binds to the N-terminal domain of the human IL-3R alpha chain (CD123) and acts as potent IL-3R antagonist [93]. 7G3 exerted potent inhibitory effects in vitro and in vivo on the growth of AML cells, including LSC, whereas its inhibitory effects on normal hematopoietic stem cells were reported to be minor if not negligible [93]. The subsequent humanization of 7G3 and its engineering (for optimal antibody-dependent cytotoxicity) resulted in the development of the CSL362 antibody [94]. CSL362 was found to be an effective agent that can inhibit the in vitro growth of CD123 + leukemic AML cells, including LSC [94]. However, when tested in a Phase I study in patients with chemotherapy-refractory AML, CSL362 did not exert sustained anti-leukemic activity in a majority of the patients examined [95].
Finally, several radiolabeled antibodies have been developed and considered for use in AML. These include, among others, 131 I-labeled, 213 Bi-labeled, or 225 Ac-labeled anti-CD33 or anti-CD45 antibodies, and 188 Rhe-labeled anti-CD66 antibodies [103][104][105][106][107]. However, most of these antibodies exert major hematologic toxicity with substantial or even long-term cytopenia (aplasia), which can be explained by the cross-radiation effects on surrounding normal hematopoietic stem cells and by the accumulation of these antibodies in hematopoietic tissues, including the spleen and BM [103][104][105][106][107]. Therefore, these antibodies are mostly applied in combination with an HSCT approach [103][104][105][106][107]. Whether some of these antibody-based therapies when added to conventional conditioning regimens are helpful or even superior in the preparation for HSCT in AML patients remains at present unknown.

Targeting LSC Using Drugs Directed against Immune Checkpoint Molecules
During the past few years, several therapeutic concepts have been developed with the aim to overcome immune checkpoint-mediated, immunologic resistance of neoplastic (stem) cells [107][108][109][110][111][112][113]. Major checkpoint molecules that are detectable on cancer/leukemic cells include CD28, CTLA4, PD-L1, PD-L2, and TIM3. PD-L1 is a checkpoint molecule that has been studied extensively in solid tumors, and the concept of blocking its activity using specific antibodies has recently been translated into clinical application also in myeloid neoplasms [107][108][109][110]112]. In fact, several PD1-targeting or PD-L1-targeting antibody-based drugs are available, and have been shown to block PD1-PD-L1 interaction, and thereby the immune checkpoint-mediated resistance of neoplastic cells in patients with melanoma, Hodgkin disease, and other neoplasms [107][108][109]. More recently, preclinical and clinical studies employing PD1-targeting or PD-L1-targeting antibodies have also been conducted in myeloid neoplasms, including AML [110][111][112][113]. However, responses to these antibodies turned out to be variable and often transient if at all measurable [110][111][112][113]. Therefore, antibodies targeting PD1-PD-L1 interaction have been combined with cytoreductive or hypomethylating agents in myelodysplastic syndromes (MDS) and AML [110][111][112][113][114]. One important aspect here is that LSC in AML, CML, and MDS do not express PD-L1 in a constitutive manner in all patients (Table 1). Rather, in most patients, LSC and more mature leukemic cells only express substantial amounts of PD-L1 in these malignancies under certain conditions. In general, three mechanisms lead to the expression of PD-L1 on LSC. One is the cytokine storm, with IFN-G being the strongest signal ( Figure 2) [61,115,116]. A second mechanism relates to the oncogene-dependent expression of PD-L1: notably, several oncoproteins and related pathways, such as JAK-STAT (a key driver being JAK2 V617F) or MYC-related pathways can upregulate PD-L1 expression on LSC [61,117,118]. Finally, certain drugs, such as the hypomethylating agents can increase the expression of PD-L1 on leukemic cells, including AML blasts and AML LSC [119]. Therefore, combinations of PD-L1 or PD1 inhibitors with hypomethylating agents are currently being tested in clinical trials in patients with AML [110][111][112][113][114]. However, the hypomethylation-induced upregulation of PD-L1 is not seen in AML LSC in all patients (P.V., unpublished observation).
During the past few years, the molecular mechanisms underlying the cytokine-induced expression of PD-L1 on LSC have been examined. In these studies, the BRD4-MYC axis and the JAK-STAT pathway have been identified as major drivers of PD-L1 expression [61,117,118]. In line with this concept, the BRD4/MYC-targeting drug JQ1 inhibits the IFN-G-induced expression of PD-L1 on LSC in patients with AML and CML ( Figure 2) [61]. Whether the targeting of PD-L1 expression by BRD4/MYC blockers is relevant clinically remains at present unknown. It is worth noting in this regard that most BRD4/MYC blockers and especially the BRD4 degraders also exhibit strong direct anti-neoplastic effects on AML LSC [120,121], and may overcome multiple forms of LSC resistance in AML and CML (P.V., unpublished observation).
Another interesting checkpoint molecule is CTLA4. Although it remains unknown whether this immune checkpoint molecule plays a role in LSC resistance in AML, it is worth mentioning that CTLA4 blockade may be useful in the post-HSCT setting in hematologic malignancies, including AML [122,123].
Finally, TIM-3 is a promising new checkpoint antigen in the AML context. In fact, TIM-3 and its ligand Galectin-9 have been shown to constitute an autocrine loop that is critical for the survival of AML LSC [124]. Moreover, AML patients have increased Galectin-9 levels, which may augment stem cell signatures and LSC renewal via TIM-3. Interestingly, Galectin-9/TIM-3 expression is upregulated in patients failing chemotherapy [125], and is associated with central memory and memory stem T cell exhaustion in AML patients with disease relapse after HSCT [126]. The clinical value of blocking Galactin-9/TIM-3 interaction alone or in combination with other targeted drugs is currently being explored in clinical trials in high-risk MDS and AML.

Targeting of LSC by Bispecific Antibodies
Bispecific antibodies are engineered drugs recognizing two different epitopes or two different antigens. These agents have several advantages over conventional targeted antibody constructs and monospecific toxin conjugates. First, bispecific antibodies can be engineered to recruit T cells, natural killer (NK) cells, or other relevant cells of the immune system [127][128][129][130]. Moreover, these drugs can be designed to recognize and block multiple surface targets on LSC, including classical molecular targets or key checkpoint molecules, such as PD-L1, PD-1, or CD47. Finally, although receptor internalization is a validated mode of action of some bispecific antibodies, it is not a general prerequisite for the functionality of bispecific antibodies in AML [127][128][129][130].
Among others, there are two types of fragment-based bispecific antibodies used in cancer research: bispecific T cell engagers (BiTEs) are recombinant fusion antibodies designed by utilizing two single-chain variable fragments (scFvs) tandemly arranged on a polypeptide chain. Dual affinity retargeting (DART) antibodies are diabodies consisting of heavy-chain and light-chain variable domains of two antigen-binding specificities linked to two distinct polypeptide chains that can heterodimerize and are connected with a cystine bond.
One of the first bispecific antibodies used in applied hematology was blinatumomab. This BiTE binds CD3 and CD19, and is able to provoke the T cell-mediated killing of B cells [131][132][133]. After first successful clinical trials had been conducted and published, blinatumomab received accelerated approval from the FDA for the treatment of either minimal residual disease-positive or relapsed/refractory B-lineage ALL [134].
Subsequently, several attempts have been made to develop similar agents for the treatment of AML. Indeed, a number of BiTEs, DARTs, and tandem diabodies are in preclinical and clinical development for AML [131][132][133]. The antibodies that are currently being tested are directed against three key target antigens on LSC, namely CD33, CD123, and CLEC12A/CLL-1 (CD371) [131][132][133]. A detailed description of the preclinical and clinical effects of bispecific antibodies in AML is beyond the scope of this article. We refer the reader to the available literature [131][132][133][134][135][136][137][138][139][140][141][142]. In many instances, clinical trials are still ongoing. The final results of these ongoing studies in AML will elucidate the real potential and clinical perspective of these agents in the AML context. A common toxicity associated with T cell redirecting therapy (including bispecific antibodies) is a cytokine release syndrome (CRS) [131][132][133]. Other side effects are cytopenia and liver toxicity. However, in general, BiTEs and DARTs are relatively well tolerated with an acceptable toxicity profile.

Targeting of LSC by Chimeric Antigen Receptor (CAR) Cell-Based Therapy
In the past few years, major efforts have been made to develop the adoptive transfer of immune cells expressing genetically engineered CAR (CAR-T or CAR-NK) in hematologic malignancies, including AML [149][150][151][152][153][154][155][156][157][158][159][160]. CARs consist of a polypeptide chain combining the extracellular antigen-binding site of an antibody, in generally a single-chain fragment of its variable region (scFv), to the intracellular CD3ζ chain that links the CAR to the signal cascade of the T cell receptor. CARs can be generated against every identified leukemic-associated antigen for which an antibody exists, and the genetic modification of T cells to express CAR can redirect them against leukemic (stem) cells in a non-HLA-restricted manner. A number of surface molecules expressed on AML LSC and CML LSC may serve as robust CAR-T cell-targeted antigens. These include, among others, CD33, CD44, CD123, CD135 (FLT3), CD371 (CLL-1), and Lewis Y (LeY) [75,[149][150][151][152][153][154][155][156][157][158][159]. However, despite numerous preclinical studies, relatively few CAR-based approaches have been investigated in clinical trials in AML so far [150,159]. A summary of these trials is shown in Table 4. So far, one report of a patient with refractory/resistant AML treated with anti-CD33 CAR-T cells within a phase I trial (NCT01864902) was published [150]. This patient received a total of 1.12 × 10 9 autologous T cells (38% CAR-transduced). Subsequently, the patient suffered from CRS as well as from prolonged pancytopenia (with blast cell clearance), and a relapse of disease was recorded nine weeks after CAR-T cell infusion [150]. In 2013, Ritchie et al. published a phase I study of autologous CAR anti-LeY T cell therapy in AML [159]. The infused CAR-T cells persisted for up to 10 months. Remarkably, grade 3 or 4 toxicity was not observed. Although three patients responded and even one cytogenetic remission was recorded, all patients relapsed after 1-23 months. Most other CAR-T cell therapies are in preclinical development or have just started to enter clinical application in phase I trials in AML (Table 4). Based on the encouraging preclinical data obtained with CAR-T cell therapies directed against CD33, CD123, CLL-1, and FLT3, there is some hope that these approaches will also work in vivo in patients with refractory AML.  However, there are a number of pitfalls and general issues to discuss when considering CAR-T cell approaches in AML. First, apart from logistic and financial obstacles, CAR-T cell therapy requires vast knowledge about the technology, practical issues, and technical details, and can therefore only be offered in specialized centers where AML is also a focus of research. Second, there are several side effects that have to be considered with CAR-T cell treatment in AML, including a tumor lysis syndrome, a CRS, prolonged cytopenia/aplasia, and other 'on-target but off-tumor' toxicities. Finally, current strategies in AML are not addressing the complexity of AML and AML LSC, but rather focus on only one or a few target structures. As a result, relapses are seen quite frequently. Therefore, the future of CAR-T cell therapy in AML may be to direct the CAR-T or CAR-NK approach against multiple LSC-specific targets that cover most AML sub-clones, but still spare normal myeloid stem cells. Another interesting approach is to apply CAR-NK cells instead of CAR-T cells (Table 4). Another question is when to apply CAR-T or CAR-NK cells in AML. Based on the toxicity profile and their ability to eliminate even residual dormant LSC, one proposed strategy is to use CAR cells in patients with minimal residual AML. In CML, no CAR-T or CAR-NK cell therapies have been developed so far. However, preclinical data obtained in mice suggest that CAR-T cells directed against IL-1RAP produce strong anti-leukemic effects on BCR-ABL1+ cells in vivo [160].

Targeting LSC by Employing NK Cells and/or T cells
Independent of therapy or lymph node infiltration, T cell and NK cell production and activity are substantially suppressed in patients with AML and other myeloid neoplasms. Therefore, multiple therapeutic strategies have been considered with the idea to restore or enhance T cell and/or NK cell numbers/activity and to direct effector cell activity against leukemic (stem) cells [161][162][163][164][165][166][167][168][169][170]. These strategies are summarized in Table 5. One approach is to expand allogeneic NK cells in vitro and apply these cells to patients together with HSCT in patients with AML [162][163][164]. Alloreactive NK cells are indeed known to exert cytotoxic effects on AML cells, to improve engraftment, and to boost graft versus leukemia (GVL) activity in AML patients receiving allogeneic HSCT [165][166][167][168]. In a recent phase I study, 21 patients with myeloid malignancies received haploidentical NK cells after conditioning with busulfan and fludarabine prior to HSCT from another donor [163]. No increase in the number of serious acute graft versus host disease (GVHD) events was reported. Although a clear survival benefit was not seen, a trend toward improved survival in NK cell-infused patients was found. Several studies with haploidentical NK cell infusion have been conducted in refractory/relapsed AML. In one study, eight patients with AML or MDS following prior HSCT received lymphodepletion followed by donor NK cell infusion and IL-2 [164]. Although one patient was driven into remission, no overall survival benefit was observed, and no donor NK cells were detected after infusion in these patients [164]. In another study, employing donor NK cell infusions on days +13 and +20 after allogeneic HSCT for refractory/relapsed AML, patients receiving larger numbers of alloreactive NK cells had an improved relapse-free survival [165]. Overall, the anti-AML effect of NK cells is well documented, but there is an ongoing discussion about the optimal way of activating these cells against AML LSC and how and when to apply NK cell infusions. An alternative strategy is to induce or promote antibody-mediated NK cell activity against AML (stem) cells. In fact, NK cells also exhibit antibody-dependent cellular cytotoxicity, and several antibodies have been applied to activate NK cells against AML cells. For example, an Fc-optimized CD133 antibody has successfully been used to activate CD16 + NK cells against AML cells and to provoke the killing of these cells while sparing normal hematopoietic stem cells [171]. Another approach tested recently was to prime NK cells against AML cells with an antibody against NKG2A, with the hope to induce the differentiation and activation of NK cells [172]. Table 5. Overview of strategies aimed at activating NK cells or applying (priming) NK cells or T cells as a therapeutic approach in patients with AML and CML.

Therapeutic Approach Indication/Application
Standard therapies: As mentioned before, several efforts have been made to engineer NK cells using the CAR technology, with the aim to augment their affinity and cytotoxicity against AML (stem) cells. Whether CAR-NK cells are better 'CAR drivers' than CAR-T cells is currently under discussion [173]. It also remains unknown whether a combined approach of boosting NK and T cells together in parallel to augment their anti-leukemic effects can elicit synergistic effects on AML LSC in patients.
Finally, a number of attempts have been made to prime NK cells with diverse cytokines to transform them into optimal killers that are capable of eradicating AML LSC. Cytokine-induced killer cells (CIK) are usually generated by exposing blood lymphocytes to IL-1, IL-2, or IL-15. In each case, CIK represent a heterogeneous population of killer cells expressing both T cell (CD3) and NK (CD56) cell markers. These cells have a broad range of anti-neoplastic activities, and are capable of exerting cytotoxicity against leukemic cells in MCH-restricted and MHC-unrestricted fashions. Several preclinical studies and phase I trials employing CIK cells in AML have been conducted and shown the feasibility and safety of this approach [174][175][176][177][178]. In some of these patients, the efficacy of CIK cells could be demonstrated [174][175][176][177]. However, CIK cell infusions are not effective in all patients, and several questions remain concerning the optimal preparation of CIK in AML, the optimal indication of CIK cell therapy, and the ability of CIK cells to eradicate AML LSC. In one recent study, IL-15 was successfully applied as a CIK activator in patients with AML [177]. Another strategy, tested recently, is to combine CIK cells with a specific CAR-T cell approach [178]. In this study, CIK cells directed against AML cells were engineered to express a CD123 CAR via retroviral transduction. These CD123 CIK CAR-T cells exerted major anti-leukemic effects in vitro on AML cells exhibiting CD123 [178]. Whether these combined CIK-CAR-T therapies are able to attack or even eliminate AML LSC is currently being tested in clinical trials.
A quite old strategy is to boost anti-leukemic NK cell and T cell effects by inducing NK cell and T cell expansion and activation with in vivo administrated IL-2. In initial studies, no substantial effects were observed [179][180][181]. Later, it was described that responses of NK cells to IL-2 are blocked by the generation of radical oxygen species (ROS) [182]. In order to overcome the ROS blockage of the immune system, combinations of IL-2 and histamine (blocks production of ROS) were applied [183][184][185]. These studies were mostly performed in AML patients in complete remission (CR), and showed convincing results with a survival benefit for CR patients treated with IL-2+histamine [184,185]. Subsequently, the drug combination was approved as maintenance immunotherapy in AML by the European Medicines Agency (EMA) in 2008. More recent data suggest that IL-2+histamine therapy may be particularly efficacious in AML patients with normal karyotypes [186]. It has also been described that IL-2 and histamine can indeed mobilize the NK cell and T cell system in these patients [187,188]. Despite this effect, side effects are usually mild and tolerable.
However, several issues remain with IL-2+histamine as maintenance therapy. First, the patients and doctors need to be trained in detail in order to avoid immediate adverse reactions. Notably, both compounds, when injected too quickly, may provoke serious adverse events, including hypotension, tachycardia, and flush, or even anaphylaxis. Another problem is that it remains unknown whether IL-2+histamine needs to be administered life-long or only for a certain time period. Moreover, the immunological effects of IL-2+histamine remain insufficiently understood. For example, the drug also induces regulatory T cells, which are assumed to counteract AML-suppressing immune responses. Thus, biomarkers confirming the efficacy and the quality of responses to IL-2+histamine therapy are lacking, and it remains uncertain what cohorts of patients may indeed benefit clinically from this type of immunotherapy.

Targeting LSC by Suppressing or Promoting LSC Homing
Depending on the type of leukemia, LSC are either fixed to the stem cell niche and enjoy niche-mediated protection against various toxic agents including therapeutic drugs (such as normal stem cells) or are mobilized cells that are not only fixed to the niche but are also capable of redistributing easily from the BM into other organ sites, and thus into extramedullary niches [189][190][191]. In most AML variants, some or even most LSC may be protected by the BM niche, although some LSC may be mobilized cells and also infiltrate into other organs sites, especially in monoblastic leukemias. However, during progression, AML (stem) cells become more and more independent of niche protection, and many LSC may infiltrate into extramedullary organs.
In CML, the situation is different: here, most LSC may be mobilized cells that can traffic into other organs, such as the spleen and form local pools of LSC, thus promoting extramedullary myeloproliferation. Several different mechanisms may underlie stem cell mobilization in AML and CML. First, the loss of certain adhesion molecules may promote redistribution out of the niche [192][193][194]. Second, enzymes produced by granulocytes or endothelial cells can promote mobilization by degrading surface adhesion receptors or cytokines responsible for stem cell-niche interactions. A good example is CML: here, LSC display dipeptidyl-peptidase IV, DPPIV (CD26). This is an enzyme that degrades stroma cell-derived factor-1 (SDF-1), which is known to be essential for stem cell homing in BM niches [29]. As a result, CML LSC mobilize into the blood [29]. An unresolved question is whether the mobilization of LSC in AML and CML can support or would even block the drug-induced killing of LSC. Initial studies with the CXCR4-inhibitor plerixafor-a drug known to induce the mobilization of AML LSC-did not show convincing beneficial effects in AML patients [195,196]. Another strategy is to block DPPIV in CML with gliptins [29]. However, so far, no clinical trials with DPPIV inhibitors or CD26-blocking antibodies have been conducted. Another strategy is to block the extramedullary spread of AML LSC by applying antibodies directed against homing and invasion receptors such as CD44. Indeed, antibodies against CD44 were found to reduce leukemic expansion in vivo in a xenotransplantation model. However, this effect was not confirmed in clinical studies so far: for example, the effects of the humanized anti-CD44 antibody RG7356 was tested in a phase I trial in patients with refractory or relapsed AML [99]. Only one out of 44 patients achieved a complete response with incomplete platelet recovery; one patient achieved a partial response, and one experienced a stable disease with hematologic improvement [99]. All in all, neither the mobilization concepts proposed nor the invasion-receptor blockers were found to improve therapy in patients with AML.

Limitations of LSC-Targeting Immunotherapy in AML and CML: LSC Resistance
Based on their selective ability to propagate the disease for unlimited time periods, LSC are attractive therapeutic targets in AML and CML, and numerous attempts have been made to selectively kill these cells by applying specific therapies [14][15][16][17][18][19][20][21][22]. However, the LSC pool (in individual patients and overall) consists of heterogeneous populations of cells (sub-clones) with varying molecular expression profiles and different patterns of cell surface antigens [22][23][24][25]31,33,[198][199][200][201]. Moreover, most of the sub-clones are small-sized, and thus are not detectable at first diagnosis, which is a phenomenon that is associated with the low proliferative capacity of pre-leukemic neoplastic stem cells [22,[197][198][199][200][201]. However, the most important issue is that LSC and their pre-leukemic neoplastic stages are highly resistant against various drug therapies [17][18][19][20][21][22][198][199][200][201][202][203]. In particular, a number of different mechanisms underlie LSC resistance in AML and CML (Table 6). In general, LSC resistance can be divided into i) intrinsic resistance (common to all LSC sub-clones and often also shared with normal stem cells), ii) acquired resistance (due to somatic mutations, the loss of tumor suppressors, or the loss of other antigens), iii) niche-mediated resistance, and iv) immunological resistance (often associated with the expression of immune checkpoints on LSC). Intrinsic LSC resistance is often associated with stem cell dormancy and/or with drug-exporter (drug-pumping) molecules. This type of resistance can often be overcome by antibody-based (ADC-based) or immune cell-based therapy, unless the antibody-conjugated, cytostatic drug is exported by an active drug pump once it is released from the antibody complex. Otherwise, antibody-based or cell-based therapies are optimal strategies to overcome intrinsic LSC resistance. Acquired resistance may or may not be overcome by immunotherapy or cell-based therapies (Table 6). For example, many cell-based therapies overcome acquired mutation-induced resistance against tyrosine kinase inhibitors. On the other hand, acquired resistance may also lead to loss of the major molecular surface targets (e.g., the loss of CD33 during treatment with GO). Another form of resistance against immune surveillance is oncogene-induced or cytokine-induced expression of checkpoint molecules, such as PD-L1 [61,[115][116][117][118]. There are also therapy-specific forms of LSC resistance that have to be considered. For example, apart from target loss, a decrease or loss of CAR-T cells and/or the development of autoantibodies are important mechanisms of resistance against CAR-T cell therapies (Table 6) [204,205]. Finally, the niche-mediated resistance of LSC plays an important role in many leukemic disorders, including AML and CML [189][190][191]. In summary, multiple forms of LSC resistance have to be addressed in AML and CML when considering the development of curative (LSC-eradicating) drug therapies, and even immunotherapy-based approaches may not be able to overcome all these resistance mechanisms. Therefore, we believe that immunotherapy concepts have to be combined with other LSC-targeting treatment concepts in order to improve the overall outcome and survival in these patients. The type of drug combination depends on the type and phase of disease, the expression of cell surface and cytoplasmic drug targets, the presence of resistance-inducing mechanisms, and also patient-related factors such as age.

Concluding Remarks and Outlook to the Future
From a historical point of view, the first therapy that combined LSC eradication with immunotherapy in myeloid neoplasms was HSCT, and this approach is still considered standard in eligible patients with relapsed or refractory AML and CML. Later, donor lymphocyte infusion therapy confirmed the critical role of the allogeneic immune system in these patients. However, only a few patients who are young and fit can undergo HSCT. During the past 10 years, more elegant ways to eradicate LSC and to mobilize the immune system against LSC have been developed.
First, the phenotype and target expression profiles of LSC in AML and CML have been established. In addition, novel targeted treatment concepts have been designed, including new antibody-based therapies, checkpoint inhibition, CAR-T and CAR-NK cell strategies and T/NK-mobilizing approaches. More recently, profound attempts have been made to merge LSC-targeting and immunotherapy concepts using novel tools and agents, with the ultimate aim to eliminate most or even all LSC in AML and CML. However, the oncogenic machineries in LSC are complex and associated with multiple mechanisms of resistance. In addition, most of these therapies may provoke side effects, including prolonged cytopenia and CRS requiring special attention and management in highly specialized centers. In order to overcome LSC resistance and to reduce side effects, combination strategies have been considered and are currently being tested in clinical trials. Whether these approaches improve the treatment and the overall outcomes of patients with AML and CML remains to be determined.