Acute Myeloid Leukemia Stem Cells: Origin, Characteristics, and Clinical Implications

Abstract

The stem cells of acute myeloid leukemia (AML) are the malignancy initiating cells whose survival ultimately drives growth of these lethal diseases. Here we review leukemia stem cell (LSC) biology, particularly as it relates to the very heterogeneous nature of AML and to its high disease relapse rate. Leukemia ontogeny is presented, and the defining functional and phenotypic features of LSCs are explored. Surface and metabolic phenotypes of these cells are described, particularly those that allow distinction from features of normal hematopoietic stem cells (HSCs). Opportunities for use of this information for improving therapy for this challenging group of diseases is highlighted, and we explore the clinical needs which may be addressed by emerging LSC data. Finally, we discuss current gaps in the scientific understanding of LSCs.

Myeloid leukemogenesis and stem cells

Human acute myeloid leukemia (AML) is characterized by accumulations of immature, blastic cells of myeloid phenotype in blood, bone marrow, or other tissues. AML in its varied forms can be manageable for younger patients, with remission common in those able to tolerate standard treatment. However, relapse occurs frequently and ultimately the disease is lethal in most patients [1]. This is a heterogeneous group of disorders which may be thought of as the endpoint of clonal neoplastic ontogeny in the myeloid lineage. This endpoint may be arrived at in a variety of ways and thus the disease which presents clinically as AML varies considerably from patient to patient in phenotype, genetic changes, clinical behavior, and prognosis.

A large literature suggests that AML and its leukemic stem cells (LSCs) arise from the hematopoietic stem cell (HSC) or relatively early committed myeloid progenitors [2]. As discussed in detail below, the roots of leukemogenesis vary greatly from case to case, and thus, the cell of origin may also vary. The leukemogenic cells, including cells which are clonal and transformed but not frankly leukemic, reside and proliferate likely in the same marrow stromal niche which is occupied by hematopoietic stem cells and early progenitors [3, 4]. By stepwise acquisition of (in most cases) multiple transforming mutations, leukemic stem cells are transformed into aggressive cells with relative or absolute block in differentiation into morphologically and functionally normal myeloid cells. This is described pictorially in Figure 1. Transforming events confer survival and proliferative advantages to the leukemic cells versus normal hematopoietic cells, resulting in suppression and failure of normal hematopoiesis.

Fig. 1.

In normal hematopoiesis (green box), quiescent hematopoietic stem cells (HSCs) with self-renewal capacity give rise to multipotent progenitors (MPPs), which can differentiate towards lymphoid primed multipotent progenitors (LMPPs), common myeloid progenitors (CMPs), common lymphoid progenitors (CLPs), granulocyte–macrophage progenitors (GMPs) and megakaryocyte erythroid progenitors (MEP). The primary/initial mutations in HSCs and progenitor cells give rise to pre leukemic stem cells, that, over time can be further transformed into leukemic stem cells (LSCs). In many cases, this process does not involve a clinically recognized pre-leukemic process, but generally, more than one mutation seems needed to generate frank acute leukemia. A restricted progenitor can be transformed to LSC by secondary mutations that confer self-renewal. Acute Myeloid Leukemia (AML) originates from the transformation of normal HSCs, MPPs, or more committed progenitors, developing in leukemic stem cells (LSCs) that subsequently can give rise to full blown leukemia. While treatment with standard induction chemotherapy results in complete remission in the majority of AML patients, a population of (chemo)therapy-resistant cells (TRCs) constituting AML cells with leukemia-initiating potential survive the treatment. LSCs with leukemia-initiating potential within minimal residual disease (MRD) could initiate a relapse. Instead of (chemo)therapeutic selection of pre-existing subpopulations of LSCs, AML cells might adaptively obtain a transient leukemia regenerating cell (LRC) phenotype upon exposure to treatment allowing for regeneration of the leukemia and clinical relapse.

The clonal nature of hematopoietic neoplasms including acute myeloid leukemia has been recognized since the studies of Fialkow and others in the early 1970s [5, 6]. While normal and malignant hematopoietic stem cells have been assayed in congenic murine models since that timeframe, the formal assay of human myeloid leukemia stem cells and thus the functional definition of the human LSC awaited the development of xenograft models in the 1990s [2, 7]. Since that time, enormous progress has been made both in the genetic characterization of AML subtypes and also in the identification of details regarding the stem cell progenitors responsible for establishing and propagating this leukemia in patients [8]. This review will address current data regarding the intersection of basic and translational studies of AML genetics with our understanding of the leukemia stem cell. A better fusion of these fields will ultimately be expected to produce a clearer understanding of how to approach leukemia therapy for individual patients.

In the 27 years since the original description of AML stem cells as defined in xenografts, the number of published articles identified in database searches has increased massively and has recently accelerated. This body of work has clear and vital interests for the community of investigators studying therapies for human leukemias. As leukemia stem cells are by definition the in vivo initiators of the leukemia, ultimate control or cure of human leukemias will require their effective inhibition or elimination.

Leukemogenesis and AML heterogeneity:

The highly heterogeneous nature of AML has been recognized since the early application of karyotype analyses to these neoplasms [9, 10]. Since the advent of Next-generation sequencing (NGS) technology however, this heterogeneity has become better defined and multiple authors have identified relevant genotypic classifications of AML. These classifications identify genetic features that correspond to phenotypic features, tumor behavior, and patient prognosis following therapy. The WHO classification of myeloid neoplasms is commonly used for clinical pathological classification of human AML and identifies a significant number of biologically defined AML subsets [11]. Some of these subtypes are defined by chromosome findings which were described some decades ago and have been molecularly characterized, while others are defined by specific gene mutations including nucleophosmin 1 (NPM1) and biallelic CCAAT enhancer binding protein alpha (CEBPA). The WHO classification however leaves a large body of AML in poorly defined and somewhat amorphous subgroups. Post-chemotherapy and Myelodysplastic syndrome (MDS) associated cases of AML, as identified in the WHO classifications, are recognized to be of poor prognosis and therapeutically resistant. These and other cases are clearly heterogeneous in their biology and genetics thus are likely to represent several genetically and biologically different groups of AMLs. Other investigators have identified relevant AML subtypes based primarily on the characterization of AML related gene mutations and karyotype abnormalities. Perhaps the largest and best studied system of this kind was described by Papaemmanuil in 2016 [12]. Based on 1540 patients (mostly younger and intensively treated) identified at diagnosis and with both chromosome and molecular characterization and clinical follow-up, this work identified 14 subsets of human AML (Table 1) with dramatically different clinical behaviors and prognoses. Since then, the European LeukemiaNet (ELN) has published work based substantially on this classification [13] and its recommendations are commonly used to guide therapy for varying AML subtypes. It is however worth noting that this classification includes 3 subgroups which are ill-defined and based mostly on overlaps with other groups or absence of class defining mutations. Thus, these relatively small classes represent arguably multiple other subclasses of diseases or entities as yet poorly recognized and defined. At present however, this is the best accepted and state-of-the-art subgrouping of human AML. This work makes it clear that AML biological heterogeneity extends well beyond that presented in the WHO classification, and that biological classification of this group of diseases is not yet finished and awaits further insights into genetic variability. A variety of investigators have demonstrated the transforming effects of various mutations by creating murine AML models via genetic engineering of murine HSCs with human AML associated mutant sequences [14, 15].

Table 1.

AML Genomic Subgroup Frequency – 14 SubGroups from [12] (N = 1540) Most Frequently Mutated Genes*

no. of patients (%) gene (%)

1) AML with NPM1 mutation 418 (27) NPM1 (100), DNMT3A (54), FLT3ITD (39), NRAS (19), TET2 (16), PTPN11 (15)

2) AML with mutated chromatin, RNA-splicing genes, or both† 275 (18) RUNX1 (39), MLLPTD (25), SRSF2 (22), DNMT3A (20), ASXL1 (17), STAG2 (16), NRAS (16), TET2 (15),FLT3ITD (15)

3) AML with TP53 mutations, chromosomal aneuploidy,or both‡ 199 (13) Complex karyotype (68), −5/5q (47), −7/7q (44), TP53 (44), −17/17p (31), −12/12p (17), +8/8q (16)

4) AML with inv(16)(p13.1q22) or t(16;16)(p13.1;q22); CBFB–MYH11 81 (5) inv(16) (100), NRAS (53), +8/8q (16), +22 (16), KIT (15), FLT3TKD (15)

5) AML with biallelic CEBPA mutations 66 (4) CEBPAbiallelic (100), NRAS (30), WT1 (21), GATA2 (20)

6) AML with t(15;17)(q22;q12); PML–RARA 60 (4) t(15;17) (100), FLT3ITD (35), WT1 (17)

7) AML with t(8;21)(q22;q22); RUNX1–RUNX1T1 60 (4) t(8;21) (100), KIT (38), −Y (33), −9q (18)

8) AML with MLL fusion genes; t(x;11)(x;q23)§ 44 (3) t(x;11q23) (100), NRAS (23)

9) AML with inv(3)(q21q26.2) or t(3;3)(q21;q26.2); GATA2, MECOM(EVI1) 20 (1) inv(3) (100), −7 (85), KRAS (30), NRAS (30), PTPN11 (30), ETV6 (15), PHF6 (15), SF3B1 (15)

10) AML with IDH2R172 mutations and no other class-defining lesions 18 (1) IDH2R172 (100), DNMT3A (67), +8/8q (17)

11) AML with t(6;9)(p23;q34); DEK–NUP214 15 (1) t(6;9) (100), FLT3ITD (80), KRAS (20)

12) AML with driver mutations but no detected class-defining

Lesions 166 (11) FLT3ITD (39), DNMT3A (16)

13) AML with no detected driver mutations 62 (4)

14) AML meeting criteria for ≥2 genomic subgroups 56 (4)

In addition to the many and varied structural and sequence level genomic mutations identified in human AML, many epigenetic changes have also been identified[16, 17]. These have powerful effects on gene expression at both the somatic protein encoding gene and mRNA level. The relationship between these genetic changes and general gene expression changes in human myeloid leukemias and their stem cells is under active investigation in many labs. Increasingly clinicians treat different groups differently, and European LeukemiaNet (ELN) guidelines for type specific therapy are well accepted [13].

Leukemia Ontogeny

Cell of Origin of LSCs:

The cell of origin of LSCs is of great importance to the field as it can help us understand the evolution and development of leukemia. Given initial findings that LSCs reside in the Lin-CD34+CD38- cell fraction, it was hypothesized that LSCs evolved directly, or through a series of driving mutations, from HSCs [18]. However, the question of whether AML LSCs originate directly from HSCs or from their downstream progenitors has been debated. Majeti et al. further elucidated that in normal HSC Lin-CD34+CD38- populations, only CD90+CD45RA- cells were enriched for the ability to generate successful secondary xenograft transplants, however both CD90+CD45RA- and CD90-CD45RA+ cell fractions demonstrated multipotent long-term hematopoiesis, indicating the latter may be a multipotent progenitor (MPP) cell [19]. Prior and subsequent experiments determined that AML LSCs did not express CD90 [19, 20]. Interestingly, Miyamoto et al. also showed that in patient survivors of the Hiroshima bombing with AML with t(8;21), AML LSCs were found in the CD90- cell fraction [21]. In disease-free survivors however, the AML1-ETO translocation was detected in the CD90+ cell fraction of non-leukemic HSCs [21], providing evidence to the idea that pre-leukemic changes may occur in HSCs, but transformation to leukemia likely occurs in direct downstream progenitors that have lost CD90 expression [20]. Given the considerable heterogeneity in LSC biology and phenotypes, it is plausible that the cell of origin and evolution to leukemia may differ as well. In 2014, Shlush et. al established that upon diagnosis of de novo AML, diagnostic samples contained ancestral pre-leukemic HSCs able to regenerate the entire hematopoietic hierarchy while possessing a competitive advantage over non-leukemic HSCs [22]. These findings support the theoretical framework that multiple genetic subclones or ancestral clones exist within the dominant clone.

Development from preleukemic states:

While many AML cases are clearly “de novo” diseases developing without clinically identified hematopoietic prodromes, a large proportion of AMLs have long been recognized as arising from myelodysplasia and other clonal myeloid disorders (e.g., myelofibrosis, myeloproliferative neoplasms, chronic myelomonocytic leukemia) (citation). With the availability of NGS studies, clonal myelopoiesis has also been frequently identified in morphologically and (apparently) functionally normal myeloid cells [23–27]. The conditions clonal hematopoiesis of indeterminate potential (CHIP) [24–26] and clonal cytopenias of undetermined significance (CCUS – cytopenias with clonal mutations but no morphological dysplasia) become extremely common with advancing age [23, 24, 27, 28]. It is clear that these conditions may evolve into AML, typically via acquisition of additional mutations. There are now elegant animal models of this sequential process of leukemogenesis. The DNA methyltransferase 3 alpha (DNMT3A) and tet methylcytosine dioxygenase 2 (TET2) genes are frequently found to be mutated in some classes of AML and are frequently clonally mutated in CHIP and CCUS [24, 26, 29]. It follows that these preleukemic clonal states must have their own “stem cells,” and single cell sorting of sequential patient samples has recently identified clonal progression from MDS through AML at the level of the stem cell [30]. The relationships between these entities and normal HSCs vs LSCs are illustrated in Figure 1.

Leukemic Clonal Evolution and Data for multiple leukemia stem cell clones:

For some time, intriguing data has been available in solid tumors suggesting that a malignancy may consist of multiple clones which are separate but functionally interdependent. In other words the clones which form the identified tumor, while distinct from each other, depend on each other for survival and/or growth via extracellular growth factors or other communication [31]. Clonal heterogeneity and independent clones have been recognized in myeloid malignancies, and sub-clonal interdependence has been speculated [32]. Recent data with single cell sequencing reconfirms the existence of independent clones in some cases of AML [33]. More studies are needed in this intriguing area.

Characteristics of Leukemia Stem Cells (LSCs)

Functional Phenotype - Defining the Leukemia Stem Cell:

While the theory that leukemic populations were maintained by rare stem cells existed prior to 1994, these putative LSCs were not characterized in vivo until 1994, when Lapidot et al. of the Dick Lab identified AML initiating cells that could establish human leukemia in immunodeficient (SCID) mice [34]. Initial limiting dilution studies suggested that AML leukemia initiating or stem cells were present in 1 in 1.2×10⁵ to 1 in 5.3×10⁵ cells [7]. In 1997, the Dick lab demonstrated that the cell capable of initiating human AML in non-obese diabetic mice with severe combined immunodeficiency disease (NOD/SCID mice) possessed differentiative and proliferative capacities and the potential for self-renewal expected of a LSC. The authors found that these LSCs (which they termed SL-ICs) were exclusively Lin-CD34+CD38-, aligning them closely with normal hematopoietic progenitor cells [2]. These leukemic clones were able to differentiate in vivo into blasts, shown by the addition of cells that were positive for CD45 and co-expression of CD34 and CD38 [2]. Importantly, this work established a precedent of requiring functional validation of tumor initiating and self-renewal capacity through serial-dilution xenografts. Since then, a number of authors have utilized murine xenografts to assay and characterize leukemia stem cells. Advancements in murine models to produce more immunodeficient mice (NOD/SCID/IL2Rγcnull model, in which residual NK cells are absent) [35] and NSGS mice (NSG mice which express high levels of the human growth factors IL-3, stem cell factor (SCF), and granulocyte macrophage-colony factor) [36], have greatly improved reliability, pace, and quantitative engraftment of human AML [36] and have yielded even more heterogeneity in leukemia stem cell populations, with the observation of LSC activity in more mature Lin-CD34+CD38+ cell populations [37]. Despite heterogeneity, LSCs must retain the property of self-renewal, which can be validated only through clonal serial in vivo repopulation assays [8].

In addition to the defining characteristics of LSCs, functional characteristics have been observed in LSCs that are of vital importance clinically. While progeny of LSCs rapidly proliferate leading to clinical consequences, LSCs themselves are largely quiescent [34, 38]. This quiescent phenotype makes LSCs difficult to eradicate with typical chemotherapy regimens that capitalize on the rapidly dividing nature of cancer cells. For example, the historically most active AML therapies have been based on the cell cycle specific agent cytosine arabinoside (AraC). While quiescence and chemotherapy resistance are generally accepted in the literature and help explain resistance and relapse clinically [39–41], some authors have brought this understanding into question [42–44]. Iwasaki et al. identified a population of cycling non-quiescent LSCs in MLL-arranged AML [45]. Multiple groups have identified LSCs that have shown susceptibility to (chemo)therapy or maturation following therapy exposure [42, 43, 46].

Cell Surface Phenotype:

Various surface antigens have been characterized for their association with LSCs as summarized in Table 2. Like HSCs, AML LSCs generally reside in the Lin⁻CD34⁺CD38^lo/− cell fraction (with important exceptions discussed below) [2, 21]. Much effort has been placed in characterizing and differentiating the cell surface phenotype of AML LSCs from non-leukemic HSCs. Through these efforts, authors have found surface markers that are overexpressed in AML LSC and additional markers that are uniquely expressed.

Table 2.

A summary of current surface phenotypic findings for human LSCs. HSC and LSC columns identify relative expression of markers on these respective cell populations.

Cell Surface Marker	Alias	Function of Marker	HSC	LSC	Validated in Xenograft Model*	Ref.
CD9	TSPAN29	Tetraspanin-enriched microdomain	−	+/−	No	[61]
CD25	IL2RA	α- chain of the high affinity IL-2 receptor	−	+/−	Yes	[62]
CD27	T14	Receptor of TNF superfamily, receptor for CD70	−	+	Yes	[63]
CD32	FCGR(2A/2B/2C)	Fc- γ receptor II	−	+/−	Yes	[62]
CD33	SIGLEC3	Myeloid cell surface antigen	+	+	Yes	[52]
CD44	P-glycoprotein 1	Adhesion molecule, intracellular signaling	+	++	Yes	[55]
CD45RA	PTPRC	Isoform of CD45 (typically expressed on naïve T cells)	−	+/−	Yes	[19, 64, 65]
CD47	MER6	Ligand/receptor for SIRPα	+	++	Yes	[49]
CD52	HE5	Campath-1 antigen	+/−	+/−	No	[66]
CD70	LPFS3	Ligand of CD27 (role in immune activation)	−	+	Yes	[63]
CD82	TSPAN27	Tetraspanin-27	+/−	+	No	[67]
CD93	MXRA4	Complement component 1 Q subcomponent receptor 1	−	+/−	Yes	[45]
CD96	TACTILE	T cell surface protein	+	++	Yes	[48]
CD97	ADGRE5	Adhesion G protein-coupled receptor E5	+/−	+	No	[67]
CD99	MIC2	T cell surface glycoprotein E2	+	+	Yes	[68]
CD123	IL3RA	Interleukin 3 receptor subunit α	−/+	+	Yes	[47]
CD200	OX-2	Immunoglobulin superfamily member	−/+	−/+	Yes	[58]
CD366 (TIM3)	HAVCR2	Negative regulator of Th1-T-cell immunity	+/−	+	Yes	[50, 51]
CD371 (CLL-1)	CLEC12A	Transmembrane glycoprotein	−	+/−	Yes	[53]
DRD2	N/A	Dopamine receptor D2	−	+**	Yes	[42]
GPR56	ADGRG1	Adhesion G protein-coupled receptor 56	+	+	Yes	[59]
IL1RAP	IL1R3	Interleukin 1 receptor accessory protein	−	+/−	Yes	[69–71]

^*Xenograft Model: must display the ability to engraft murine model in both primary and secondary transplantations (primary engraftment alone does not differentiate MPP from LSCs) [19, 72].

^**Expressed in post-chemotherapy LSCs, which authors termed LRCs [42].

Jordan et al. found that CD123 (IL-3R-α) was overexpressed in CD34+CD38- AML LSCs but not in normal HSCs which represented an important antigenic marker to identify AML LSCs. While the role of CD123 was not elucidated, given the strong expression of CD123 on these cell populations, the authors proposed the targeting of CD123 and thus AML LSCs [47]. Other cell markers have been found to be overexpressed on AML LSCs, including CD96, CD47, and TIM3 [48–50]. CD47 serves as a ligand for SIRPα, which is expressed on phagocytes and when activated inhibits phagocytosis. Through antibody mediated disruption of the CD47-SIRPα interaction, phagocytosis was enabled in this LSC enriched population and engraftment was inhibited in vivo [49]. Importantly, Majeti et al. found differential expression of CD47 on AML LSC in comparison to HSC and MPP in the same patient sample, providing a method of separating these cell fractions [49]. Increased and differential expression of TIM3, a negative regulator of Th1-Tcell immunity, across cytogenetic AML subgroups compared to normal bone marrow HSCs, provided an additional tool to prospectively separate functionally normal HSC from AML LSC and bulk AML cells. The ability to isolate normal HSC and leukemic populations in patients or patient samples could prove to be a powerful tool for assessing and characterizing preleukemic mutations in HSC populations with applications for leukemia-depleted autologous HSC transplantation therapies [49–51]. Taussig et al. showed that CD33, CD13, and CD123 were expressed in human long-term repopulating cells from cord blood and bone marrow. Further studies revealed that NOD/SCID leukemia initiating cells were restricted to the CD33+ cell fraction, indicating the AML LSCs express this antigen. Of clinical importance, gemtuzumab ozogamicin (GO) was originally designed with the aim of selectively eradicating leukemic blasts that express the myeloid antigen CD33. Taussig et al. revealed that GO in fact may also target AML LSCs, however this targeting is not necessarily specific to the malignant stem cell population [52].

Cell surface markers that are not specific to the AML LSC fraction represent a clinical challenge when developing therapeutics and prognostic tools. In 2007, Rhenen at al. detected C-type lectin-like molecule-1 (CLL-1) expression in AML CD34+CD38- cell samples that confirmed to engraft NOD/SCID mice indicating they were enriched with AML LSCs. Importantly, CLL-1 was absent in both normal CD38+CD34- bone marrow cells, in normal cells following chemotherapy (regenerating bone marrow), and after treatment with G-CSF, making CLL-1 a specific marker of the malignant CD34+CD38- stem cell compartment and a potential marker for minimal residual disease (MRD) following chemotherapy [53]. CLL-1 was also found in the majority of CD34- AML cell fractions and has been shown to be expressed in MDS and chronic myeloid leukemia (CML) cell populations [54].

While differential and unique expression can be utilized clinically, these phenotypes can also elucidate aspects of LCS biology. Jin et al. showed that CD44 ligation (mAb-mediated activation) could selectively eradicate LSCs in vivo by blocking LSC trafficking to supportive microenvironments and by altering their stem cell fate. CD44 is an adhesion molecule with the primary ligand of hyaluronan. Upregulation of CD44 on LSC surfaces may allow these cells to more effectively migrate to a niche highly abundant in CD44 receptors. This paper elucidated the importance of the niche (which will be discussed in detail below) in LSC viability and function and reinforced the idea that focusing on stem cell properties can be an advantageous method of developing future treatments that target LSCs [55]. Further analysis of LSC biology may also guide the discovery of additional cell surface markers. In total, these findings have elucidated potential therapeutic targets, uncovered new biological understanding of LSCs, and provide new laboratory and clinical tools to differentiate or prognosticate.

Many cases of AML conform well to the Lin⁻CD34⁺CD38^lo/− phenotype for LSCs, however more immunodeficient murine models have led to LSC populations being found in some CD34+CD38+ cell fractions[56]. Additionally, some AML subtypes do not express CD34, particularly the relatively frequent and favorable prognosis cases with mutation of the NPM1 gene [57]. In these cases, substantial proportions of leukemia stem cells are found in CD34- cell populations [56]. This heterogeneity has made it difficult to ensure all LSC clones are being represented in xenograft transplantations and makes characterizing additional stem cell markers more challenging. In a recent article by Ho et al., the authors have proposed that utilizing CD45^dim blast populations rather than CD34+ populations to fractionate cells for xenografts can provide a more complete representation of the LSC population. In addition, they have identified CD200 as a marker expressed in both CD34+ and CD34- LSC fractions that can assist in differentiation from CD45^high blasts [58]. G protein-coupled receptor 56 (GPR56) is an additional surface maker expressed on xenograft validated LSCs that was found in both CD34+ and CD34- cell fractions. GPR56 expression varied based on AML subtype, with minimal expression in MLL-rearranged, inv(16), and t(8;21) AML subtypes, again highlighting LSC heterogeneity [59].

Much of the genetic and phenotypic understanding of AML LSCs has formed from treatment naïve populations or in steady-state environments. Recent work has explored AML cell populations in dynamic states and following significant stressors. In 2018, Boyd et al. described an elegant set of experiments to characterize LSCs following cytoreductive therapy [42]. The authors utilize serial bone marrow aspirate sampling in a murine model to characterize AML cell populations prior to and following AraC treatment, with particular emphasis on the point of maximal chemotherapy response. Importantly, the authors’ identified cells that displayed self-renewal potential yet were molecularly distinct from treatment naïve LSCs [60], which they termed leukemia regenerating cells (LRCs). Transcriptional analysis of post-treatment LRCs and targeted antagonism (of dopamine receptor D2) of their unique gene signatures resulted in profound inhibitory effects on LRC progenitor activity in vitro and serial transplantation ability in vivo compared to treatment naïve LSCs. In addition, AraC + LRC specific antagonism blocked disease regenerative potential in comparison to AraC alone in vivo [42]. This article importantly identified a transient cellular state that could be specifically targeted to prevent relapse following standard chemotherapy.

Self-renewal in stem cells, as defined by the ability to give rise to progeny identical to the parent cell, is best evaluated by long-term hematopoiesis and engraftment and ability to generate successful serial transplants [8]. The ability to generate secondary xenografts can help distinguish HSCs from multipotent progenitors (MPP) [19]. Experiments evaluating engraftment potential do not necessarily distinguish between HSC and MPP cell populations unless completing these secondary transplant experiments, a consideration when evaluating cell surface phenotypes on AML LSCs. This differentiation, although difficult and time-intensive, is essential to better understand cell surface markers in the context of their accurate cellular fraction. Laboratory techniques themselves, such as the use of monoclonal antibodies to target cell surface markers, may affect the ability of cells to engraft the murine host [37, 52]. To successfully repopulate mice in serial transplantations, cell populations must not only contain stem cells, but must be able to survive laboratory manipulation. We must consider that the very nature of these experiments bias towards more robust and durable phenotypes [8]. Finally, much of our understanding of LSC biology and phenotypes has come from therapy naïve LSCs. Better understanding of LSCs following chemotherapy will allow us to more accurately characterize the nature of relapsed disease and develop therapies to target this process [42].

Metabolism and Gene Expression of Leukemia Stem Cells

Energy Metabolism of LSCs:

Metabolic characteristics of cancer have been explored for many decades. In 1956 cancer cells were shown by Warburg et al to generate cellular energy primarily through aerobic glycolysis, despite the decreased efficiency of energy generation [73]. Aerobic glycolysis in cancer cells has been proposed to function to provide the cell with essential biomass (nucleotides, amino acids, and lipids) rather than maximal ATP, with biomass the limiting factor for the proliferating cell [74]. Leukemia cells are highly glycolytic, yet can be found in the blood, an environment with an abundance of oxygen. They utilize up to 20 times more glucose than normal hematopoietic cells and manipulate host tissues to ensure high bone marrow glucose concentrations [75–77]. Targeting metabolic characteristics of cancer is not new, as most chemotherapy employs mechanisms involving induction of oxidative stress, a strategy that can be effective due to an elevated oxidative state commonly observed in cancer cells. However, relative redox levels in tumors can be heterogeneous, suggesting intratumoral differences in underlying metabolism and effectiveness of therapy [78].

Analyses of metabolic characteristics in bulk AML cells are only of limited value as LSCs are likely responsible for maintenance of disease and relapse following treatment yet make up only a small fraction of AML cells. Targeting LSCs requires an understanding of their underlying cell biology, which has been shown to diverge from our current understanding of AML metabolism in bulk cell populations as discussed below.

Energy Metabolism and Apoptosis:

Stem cells from multiple tumor types have been shown to rely on oxidative phosphorylation (OXPHOS) [79, 80]. Similar to HSCs, the majority of self-renewing and chemotherapy-resistant LSCs are quiescent [7, 41]. Consistent with this phenotype, AML LSCs have relatively low production of reactive oxygen species (ROS-low) compared to AML blasts [80]. This ROS-low phenotype also extends to cancer stem cells more broadly [81]. Targeting mitochondrial pathways and oxidative phosphorylation has revealed a vulnerability in LSCs not seen in bulk AML blasts or normal HSCs. Further investigation of these mechanisms has begun to provide clarity on the specific metabolic characteristics of LSCs.

ROS-low AML cells, confirmed to be enriched with LSC populations by xenograft assay, serial transplantation, and overexpress B-cell lymphoma 2 (BCL2) oncogene. [80]. In addition to the well-established role of BCL2 as an inhibitor of the mitochondrial-initiated apoptotic pathway, BCL2 has also been shown to regulate oxidative state and mitochondrial metabolism [82]. When BCL2 is inhibited, oxidative phosphorylation is decreased and LSCs are selectively eradicated [80]. Upon further investigation, during inhibition of OXPHOS, HSCs upregulated glycolysis, an expected maneuver during energy deficit. However, in ROS-low LSCs, this glycolytic response was minimal leading to reduced cell viability, thus indicating that LSCs are “metabolically inflexible”. This inability to flex to other forms of energy generation when necessary, represents a vulnerability of LSCs not present in HSCs, a vulnerability that could prove valuable when developing therapeutics [80].

In 2018, treatment with the BCL2 inhibitor venetoclax in combination with the hypomethylating agent azacitidine (ven/aza) was shown to produce a robust and durable response in elderly patients who could not tolerate standard chemotherapy induction treatment [83]. Molecular analysis of patient samples pre- and post-treatment from this trial revealed that ven/aza disrupted the tricarboxylic acid (TCA) cycle with specific markers suggesting inhibition of electron transport chain complex II and suppression of OXPHOS [84]. Importantly, this study revealed that this treatment was effective across a variety of AML phenotypes, irrespective of traditional biological risk factors [84].

Despite a robust initial response, patients with AML generally relapse following ven/aza treatment with further ven/aza treatment less efficacious in this patient population [85]. With ven/aza shown to target energy production through inhibition of OXPHOS, relapse may indicate an adaptation of LSCs to additional or alternative forms of energy production. Specific understanding of cellular metabolism in both de novo AML LSC and relapsed and refractory (R/R) AML LSC may therefore reveal insights underlying treatment failure.

De novo AML LSC Metabolism:

A recent paper by Jones et al. revealed that LSCs isolated from de novo primary AML samples are uniquely reliant on amino acid (AA) metabolism for oxidative phosphorylation and survival [86]. Metabolomic analysis showed significantly increased amino acid metabolism in ROS-low LSCs compared to ROS-high AML blasts, which were separated using flow cytometry. In addition, AA depletion in vitro significantly reduced OXPHOS, viability, and function in LSCs compared to AML blasts and HSCs, indicating preferential importance of AA metabolism in the LSC population. In contrast, glucose and lipid depletion did not affect LSC viability or function, yet glucose depletion significantly reduced viability in ROS-high AML, highlighting the metabolic differences between the cell types. Importantly, treatment with ven/aza led to a reduction in AA uptake in LSCs, which contributed to a reduction in OXPHOS, thus targeting LSCs [86].

Relapsed and Refractory (R/R) AML LSC Metabolism:

In comparison to de novo LSCs, relapsed LSCs were more resistant to ven/aza treatment. AA depletion in this cell population did not result in decreased viability or cause a reduction in OXPHOS in relapsed LSCs, indicating the relapsed cells had likely adapted through upregulation of other metabolic pathways. Further testing revealed that relapsed LSCs compensated for AA depletion through an increase in fatty acid metabolism to the TCA cycle to maintain OXPHOS [86]. In a follow-up study by Jones et al., the authors showed that R/R LSCs have increased AA metabolism and fatty acid oxidation (FAO) contributing to OXPHOS compared to de novo LSCs, which was driven at least in part by an upregulation of nicotinamide metabolism. Targeting nicotinamide metabolism decreased OXPHOS and the viability of R/R LSCs, providing an example of understanding and targeting metabolic adaptations in LSCs [87].

Aldehyde dehydrogenase (ALDH):

ALDH activity is relatively specific for proliferative and progenitor cell populations and appears important for detoxification of reactive oxygen species (ROS) and alkylating agents [88, 89]. ALDH expression has been identified as marking normal HSCs and the stem cells of various neoplasms [90–93]. In AML, its expression also seems restricted to putative stem cells including those engrafting as xenografts [94]. The expression is however highly variable from case to case, and authors have suggested that for many AML LSCs, expression may be lower than that of normal HSCs [95, 96]. This has suggested a potential to successfully treat AML via targeting or inhibition of ALDH either via the use of alkylating chemotherapeutics or ALDH specific inhibition [97, 98]. We have studied a novel ALDH inhibitor KS99 which has shown powerful anti-human AML toxicity (vs normal human progenitors) and in vivo activity [99].

Telomerase:

Maintenance of telomere length and prevention of its natural attrition if all the vision is an evolutionary early conserved mechanism. Nutrition of telomeres blood cell division appears protective event which results in prevention of many human neoplasms prior genomic instability[100]. Telomerase is expressed in stem cells of various types including HSCs. Human cancers in general and AML in particular have shorter telomeres than do age-matched control tissues[101]. In certain genetic subtypes of AML and in younger patients, telomeres are shorter than in other types[102]. This appears particularly to be associated with TP53 mutated AML’s in those with substantial aneuploidy[103]. Some transforming mutations in AML however are associated with increased telomerase activity. Telomerase is particularly active in AML progenitor and leukemia stem cell fractions[103].

Aryl Hydrocarbon Receptor Signaling:

The aryl hydrocarbon receptor (AHR), identified as important in maintenance of stem cell characteristics for normal hematopoietic progenitors, has been studied for human AML. Another group from the University of Toronto have identified that diminished signaling through this receptor complex is associated with maintenance of stem cell characteristics in the AML cell population [104]. Inhibitors of signaling through this pathway have also been found to be helpful in the culture of AML progenitors in vitro [105, 106].

Gene Expression:

A leukemia stem cell gene expression signature has been identified by Ng et al. and has been validated as predicting for survivorship and a large series of AML patients. This group studied cells sorted into CD34 positive and CD38 negative LSCs fraction versus those in other fractions and generated a unique 17-gene stemness scoring system which has high prognostic capacity along with accurate prediction of initial treatment resistance in AML based on the differential gene expression profiles [107]. This expression profile was then shown to be more predictive for outcome in AML treatment then other I can identify gene expression or clinical parameters.

The Leukemia Stem Cell Niche – Interactions with Stroma

Normal hematopoietic stem cells have a complex but obligate relationship with the nonhematopoietic cells of the bone marrow stroma. This stem cell stromal relationship has recently been reviewed, and it is noted that there is considerable controversy regarding what the most important cell-cell relationship is for the earliest, quiescent HSC [108]. A variety of non-hematopoietic cells (osteocytes, endothelial cells, adipocytes, and various mesenchymal cells) and hematopoietic derived cells (monocytes, megakaryocytes, Treg cells) seem to participate in this environment. Available evidence suggests that LSCs similarly, at least early in the disease, require interaction with the bone marrow stroma and in fact compete with normal HSCs for the sites in the marrow niche. The relationship to the HSC and LSC to various stromal marrow elements is depicted in Figure 2. Certainly AML, even in relatively low bulk conditions, is frequently suppressive of normal myelopoiesis. Various authors have explored the interaction of AML and its progenitors with the marrow stroma [4, 108]. There is data that AML Stem cells modify the marrow stroma via their growth in a variety of ways (see Figure 2) [4]. A number of growth factors and adhesion molecules participate in the AML stromal interaction [3]. A hypoxic zone in the endosteal area appears to be functionally important for quiescent HSC and LSC. The proximity of the LSC to stroma provides a potential therapeutic avenue via “mobilization” of LSC from stroma to blood, thereby removing survival supports and potentially sensitizing these cells to therapy. Clearly in many AML cases, an interaction of the AML progenitors or LSCs to stroma occurs via C-X-C motif chemokine receptor type 4 (CXCR4) interaction with C-X-C motif chemokine ligand 12 (CXCL12) produced by stromal cells. There are ongoing efforts to sensitize AML to therapy by “mobilizing” these LSCs through blockade of the CXCR4 interaction [109, 110].

Fig.2.

The comparison between stem cell niche organization during normal and malignant hematopoiesis. The HSCs (left) and LSCs (right) reside in bone marrow (BM) niche consisting of hematopoietic cells, cell populations of stromal origin such as mesenchymal stromal cells (MSC) and perivascular endothelial cells (located in blood vessels sinusoids/arterioles), and Extracellular matrix (ECM). In particular, the nestin and/or leptin receptors positive MSC around endothelial sinusoid/arteriole have high expression of chemokine CXCL12 that act as attractant for CXCR4-expressing HSCs. The CD146 positive mesenchymal progenitors provide factors that facilitate homing, proliferation, transendothelial migration, and differentiation of HSCs/ LSCs. The quiescent HSCs are dormant (G0) cells, whereas active HSCs represent cells that have just exited quiescence or actively cycling or migrating. The relevant stromal cell populations were illustrated here along with their interactions and soluble factors such ascytokines that are secreted to support expansion of stem cells in the niche. The arrows represent the direct effects/factors, whereas dotted arrows indicate indirect effects and thick arrows for showing differentiation process. The hypoxic BM niche is regulated by Hypoxiainducible factors (HIF-1/HIF-2), and proliferation of LSCs leads to the expansion of hypoxic niches during malignancy/leukemias. Not all depicted features of the LSC niche will be found to operate in every case of AML. The figure is intend to simplify and represent some of the best candidate niche cells and molecules that were found to be altered during leukemias. OB, Osteoblast; OC, Osteoclast; Mac, Macrophages; Treg, Regulatory T cells; HSC, hematopoietic stem cell; LSC, leukemia stem/initiating cell; MSC, mesenchymal stromal/ stem cells; ECM, extracellular matrix; nmSC, non-myelinating Schwann cells; NA, noradrenaline; CAR cells (CXCL12-abundant reticular cells); CXCL12, C-X-C motif chemokine 12; G-CSF, granulocyte-colony stimulating factor; TGF-β, transforming growth factor-β; VEGF, vascular endothelial growth factor; TPO, thrombopoietin;CCL3, Chemokine (C–C motif) ligand 3; IL-1β/6/7, interleukin 1β/6/7; SCF, stem cell factor; ANGPT1, angiopoietin 1

Surface targeting of leukemia stem cells

Numerous efforts have been made to target human AML via expression of surface molecules. Limiting these approaches for AML however is the lack of leukemia specific markers. Thus, for most markers described above in Table 2, there are normal elements of hematopoiesis (and in some cases other sensitive tissues) which express the surface antigen and are susceptible to targeting via antibody coupled drugs or other approaches. In addition, many cell surface markers are incompletely expressed on LSCs given stem cell heterogeneity [58].

The development of gemtuzumab ozogamicin and its approval for therapeutic use marked an important milestone in AML therapy [111]. This immunotoxin was designed to target the myeloid antigen CD33 expressed on most AML blasts, with further studies revealing it also targeted AML LSCs [52, 112]. Unfortunately, GO targets CD33 expressed on normal HSC and progenitor cells as well as other immature cells including sinusoidal cells in the liver parenchyma, leading to impaired hematopoiesis and significant liver toxicity, which have limited its use [113, 114]. To circumvent the nonspecific targeting of CD33 in normal hematopoietic cells, one group used CRISPR-Cas9 to generate CD33-deficient human HSPCs resistant to CD33-targeted CAR T cells [115], highlighting a unique method that could improve applicability of antigen-specific immunotherapy. CD123 has also been targeted with monoclonal antibodies in pre-clinical [116] and clinical trials. Talacotuzumab, a CD123 mAb, was unsuccessful in phase II (as a monotherapy) [117] and phase II/III (in combination with decitabine) [118] clinical trials due largely to toxicity or limited efficacy. The cell surface interaction of CD47 and SIRPα has been targeted with the development of the CD47-directed monoclonal antibody Magrolimab, leading to leukemic cell engulfment and elimination [119]. CD47-directed antibodies are now in clinical trials of patients with AML, MDS, and other malignancies [120, 121]. Post-chemotherapy LSCs have been studied as well, with recent literature suggesting a transient population of cells that express unique gene signatures and cell surface markers may be targetable with specific, timed therapy [42]. This has been evaluated clinically with the use of thioridazine, a dopamine-receptor D2 antagonist, in combination with cytarabine in a phase 1 trial of relapsed and refractory AML patients [122], which has shown preliminary evidence of efficacy however was limited by thioridazine adverse effects at higher doses [122].

LSC-specific immunophenotype targeting continues to be an area of great potential in AML therapeutics, however heterogeneity in stem cell populations and non-specific toxicity, particularly of HSCs and committed progenitors, has limited its clinical application. Continued efforts are underway to identify and develop which selectively target large proportions of the LSC population.

Understudied Areas

The divide between functionally and phenotypically defined LSC:

The originally described leukemia stem cell was defined based on engraftment of human leukemias in immunodeficient murine hosts. These foundational works validated the cells defining phenotype of self-renewal through serial limiting dilution assays of cells fractionated based on their surface phenotype. These are expensive and labor-intensive experiments and thus the number of labs currently undertaking such work is limited. Many labs have generated xenografts of human AMLs in suitable murine hosts and arguably, particularly when the xenograft surpasses into secondary hosts, this must identify leukemia stem cells as the cell responsible for long-term propagation of the neoplasm in the xenograft. However, as LSC markers and gene signatures become more widely adopted to identify LSCs, the field must continue to ensure LSCs are functionally validated in serial transplantation assays.

LSCs as Specific for AML subtypes:

As discussed above, AML is the final endpoint for clinical evolution of a number of different myeloid diseases. Arguably, these different entities or AML subtypes are biologically very different and thus their leukemia stem cells are also different one from another. The available literature defining these subtype specific LSCs is limited [123]. Efforts to develop subtype specific therapy in the coming decade will depend on better definition and understanding of subtype specific long-term clonogenic cells.

Differentiation therapy for APL, AML and its implications for LSCs:

For several important subsets of AML, small molecule inhibitors are available which target expressed mutant proteins. There is clear data that these inhibitors result in in vivo differentiation of clonal AML elements into normal neutrophils and other myeloid elements without extensive apoptosis or other cell death. Thus, under the influence of these agents, leukemia stem cells are induced to give rise to cellular progeny with relatively normal hematopoietic differentiation properties. Clinically, these drugs provide dramatic therapeutic results. Cell arising from removal of the differentiation block caused by mutations, are functional hematopoietic cells which are still clonal and derived from the mutant leukemia stem cell populations. This therapeutic in-vivo differentiation has been long recognized for acute promyelocytic leukemia, a subset of AML who is discussion is beyond the scope of this review. The Isocitrate dehydrogenase 1 and 2 genes however are also associated with mutations which are similarly targetable and yielding differentiation of blastic clonal elements into mature cells[124, 125]. It follows that leukemia stem cell and progenitor behavior has been altered to allow their differentiation and restrict their self-renewal, but to date the authors are not aware of laboratory studies examining the in-vivo differentiation of LSCs.

Implications for AML Treatment:

AML therapy has recently been in rapid evolution with the integration of venetoclax (targeting BCL-2) with historical therapies and with new agents targeting IDH 1 and 2 and FLT3 mutations. Many patients, even those older and relatively unfit, enter remission – defined as a state of relatively normal hematopoiesis but often with minimal residual leukemia or disease (MRD). Unfortunately, the great majority will relapse from residual MRD driven by LSCs. Thus relapse, and the control or eradication of this LSC population, is perhaps the greatest therapeutic obstacle. Despite the large and growing body of basic research surrounding leukemia stem cells, translation of this work into the clinic has been difficult and minimal. Reasons for this are multiple, but in part because the well-recognized clinical subtypes discussed above have not been successfully integrated with leukemia stem cell genetic and phenotypic findings. Additionally, heterogeneity in LSC populations and similarities to normal hematopoietic cells have made it difficult to identify successful and specific drug targets. The field is also moving towards the intensive use of MRD detection. MRD detection, whether via multiparameter flow cytometry or mutation quantitation, must closely correlate with clonogenic cells to be useful. Detection of LSCs and progenitors in MRD will more accurately assess cells capable of regenerating leukemic colonies, and thus will more accurately predict relapse. In the coming decade, combined efforts of basic, translational, and clinical investigators will be necessary to ensure the advances discussed in this review, as well as future discoveries, translate to meaningful improvements in care for patients with acute myeloid leukemias.

Data Availability

N/A