Overview of studies in adult AML with cut-off values used for analyzing relapse free and overall survival.
Acute myeloid leukemia (AML) is the most common type of acute leukemia in adults. With current treatment strategies, almost 80% of AML patients (18-60 years) will achieve complete remission (CR). However, approximately 50% of these patients will experience a relapse, resulting in a five-year survival rate of only 35%-40% 
2. Minimal residual disease and acute myeloid leukemia
In AML patients, morphologic assessment is performed to evaluate chemotherapy response and to define remission status. By definition, patients are in CR when less than 5% blast cells are present in the bone marrow (BM) concurrent with evidence of normal erythropoiesis, granulopoiesis and megakaryopoiesis. In addition, neutrophils and platelets in peripheral blood should be at least 1.0 x 109/l and 100 x 109/l, respectively 
2.1. Immunophenotypic MRD detection
2.1.1. Principles of immunophenotypic MRD detection
One of the most frequently used techniques to assess MRD in leukemia is based on assessment of immunophenotypic aberrant antigen expression using flow cytometry. For practical purposes, in most cases, this approach is restricted to cell surface antigen expression. At diagnosis, so-called leukemia associated (immuno)phenotypes (LA[I]Ps, further referred to as LAPs) are determined. Such a LAP consists of (an) aberrantly expressed cell surface marker(s), usually combined with a myeloid marker (CD13/CD33) and with a normal progenitor antigen, i.e. CD34, CD117 or CD133. LAPs are grouped into (1) cross-lineage antigen expression (e.g. expression of lymphoid markers in myeloid blasts), (2) asynchronous antigen expression (co-expression of antigens that are not concomitantly present during normal differentiation), (3) lack of antigen expression and (4) antigen overexpression . Such aberrancies can subsequently be used to detect MRD (Figure 2).
Due to large heterogeneity of immunophenotypes in AML, determination of LAPs has to be performed for each individual patient. These LAPs are not, or only in very low frequencies, present on normal BM cells in remission BM. Sensitivities have been reported to be in a range of 10-3 down to 10-5 (1 leukemic cell in 1,000 to 100,000 normal cells) [4-9]. Besides these relatively high sensitivities, it is also a very rapid technique. Main advantage of flow cytometric MRD assessment is its broad applicability: in 80%-95% of all AML patients one or more LAPs can be defined. [4,5,9-11]. There are, however, potential pitfalls/disadvantages that should be taken into account.
2.1.2. Prognostic value of immunophenotypic MRD in bone marrow
The likelihood of achieving CR after therapy and the duration of CR depend on different factors. Important prognostic risk factors available at diagnosis are: history of previous leukemia or myelodysplastic syndrome, age, white blood cell (WBC) count, percentage of BM blast cells and the presence of particular cytogenetic and/or molecular aberrancies . Besides these pre-treatment prognostic factors, it is suggested that MRD detection in BM shortly after treatment offers an important post-treatment prognostic factor. To evaluate the impact of MRD frequencies on relapse rate and overall survival (OS), MRD was related to outcome parameters using survival analyses such as Kaplan Meier curves. For these analyses, most studies set a threshold to define MRD negative (or low) and MRD positive (or high) patients. Different laboratories use different optimal cut-off values after both induction and consolidation therapy (Table 1). However, it should be emphasized that usually, it is not a single cut-off point, but a range of cut-off values that significantly predict clinical outcome.
|San Miguel ||126||<0.01%,|
|Kern ||62||Log difference 1.70||Log difference|
|MRD level||Patients (n)||Relapse rate ± standard error|
|< 0.1 %||16||9% ± 7%|
|0.1% - 1%||45||56% ± 9%|
|"/1%||21||83% ± 10%|
2.2. Molecular MRD detection
Although flow cytometry is an attractive technique for MRD detection, the limitations, including background staining, immunophenotypic switches, complexity of analysis and LAP expression on only part of the leukemic cells, give rise to alternative approaches for MRD detection, including molecular MRD monitoring using the Polymerase Chain Reaction (PCR) technique. This approach allows for the detection of mutations, translocations, inversions, deletions and polymorphisms. Real-time-(qRT-) PCR is the most sensitive technique for MRD detection: it allows detecting MRD with sensitivities that have been reported in a range of 10-4 to 10-6 [18-21]. QRT-PCR is now extensively being studied as approach for MRD detection. Common targets for molecular MRD monitoring, including fusion genes, overexpressed genes and gene mutations, will be reviewed in this section.
2.2.1. Fusion genes
Fusion genes are among the best potential targets for molecular MRD detection. In AML the most common chromosomal rearrangements, producing fusion genes, are t(8;21), t(15;17) and inv(16)/t(16;16). The corresponding fusion genes are
In t(8;21) rearrangement, the AML1 gene on chromosome 21 fuses with the MTG8(ETO) gene on chromosome 8 to produce the fusion gene
More research has been done on the
Mixed-lineage leukemia (
2.2.2. Overexpressed genes
Since in only a small fraction of patients, fusion transcripts are present, overexpressed genes might offer a potential alternative target for molecular MRD monitoring. Such overexpressed genes are either silenced or expressed at very low levels in normal hematopoietic cells. Commonly overexpressed genes are
Another potential marker is
The ecotropic virus integrations-1 (
2.2.3. Gene mutations
Since fusion genes are only present in 15%-45% of AML patients and overexpressed genes seem to be less specific MRD markers, gene mutations may offer another attractive group of targets for MRD monitoring.
A decade ago, fms-like tyrosine kinase 3 (
Mutations in the nucleophosmin (
CCAAT/enhancer binding protein alpha (
2.3. Clinical applications of MRD
As discussed above, MRD frequency assessment using immunophenotypic and molecular parameters in patients with AML in clinical remission has important prognostic value and can predict forthcoming relapses. Therefore, it would be of potential importance to monitor MRD cell frequency for risk stratification. Current AML risk stratification is based on a number of parameters determined at diagnosis, including origin of leukemia (secondary AML, AML after myelodysplastic syndrome), age, WBC count, and presence of certain cytogenetic and/or molecular aberrancies . Novel AML risk stratification should not only be based on risk assessment at diagnosis, but also on MRD cell frequency as a “response to treatment” parameter. Including MRD in AML risk stratification could help identify CR patients after induction therapy with increased MRD levels and therefore high risk of relapse. For instance, good risk patients with high MRD levels after induction therapy may benefit from allogeneic stem cell transplantation, while on the other hand intermediate risk group patients with low MRD levels could be spared from an allogeneic transplantation and the accompanying toxicity. Especially in this intermediate risk group, MRD monitoring would be of great help, since the prognosis of these patients is difficult to estimate. Therefore, MRD based clinical decision making after induction therapy may contribute to better RFS and OS rates.
Also after consolidation therapy, MRD based clinical intervention is promising. Even after an allogeneic transplantation, still a proportion of 20%-40% of the patients will relapse [73-75]. Therapeutic options in the case of post-transplant relapse consist of withdrawal or decrease of dose of immune-suppressive drugs, or immunotherapeutic intervention with a donor lymphocyte infusion. As these approaches intend to boost the graft versus leukemia effect, they are most effective when the leukemic cell load is small. Therefore early detection of impending post-transplant relapses is essential and would allow immunotherapeutic intervention at a low leukemic burden. The current standard to guide post-transplant treatment is the level of donor chimerism. This refers to the percentage of donor cells in PB or BM and it can be determined using short tandem repeat (STR)-PCR. Although mixed chimerism (< 95% of donor cells) has been associated with a higher incidence of relapse [76,77], patients with full chimerism (> 95% donor cells) can still suffer from relapse . Additional monitoring of MRD levels in these transplanted patients could improve successful prediction of relapse, since MRD analysis directly detects the neoplastic part of the patient cell population, while STR analysis reflects total donor and total patient populations. Multiple studies have shown that MRD monitoring after an allogeneic transplantation indeed correlates with clinical outcome and identifies patients who are likely to relapse [78-81]. Therefore, it can be suggested that MRD based pre-emptive immunotherapy after transplantation could reduce relapse and improve survival. Standardization of treatment, based on MRD and chimerism analysis in the post-transplant period, seems therefore warranted. In conclusion, since MRD frequency assessment gives important prognostic information after both induction and consolidation therapy, it seems likely that using MRD for therapeutic intervention in the post-remission phase might reduce relapse rates en prolong OS. To confirm this hypothesis, large prospective studies with MRD based clinical intervention in the post-remission phase are essential.
2.4. Improvement of and alternatives for bone marrow MRD detection
2.4.1. Improvements for immunophenotypic and molecular MRD detection
Although flow cytometric MRD monitoring has many advantages, one of the difficulties is the complexity of MRD analysis. Nowadays, more advanced data analysis programs, that aid to distinguish between normal and malignant cells, are available . This might simplify the analysis and result in more objective results. Notwithstanding the high prognostic value of MRD monitoring, in almost all studies 20%-40% of the patients with immunophenotypic defined low MRD levels still suffer from a relapse [5-7,10,16]. There are several potential reasons for missing these MRD cells. Normal BM cells express LAPs at low frequencies. Counting these cells as leukemic might result in false-positivity. This background expression thus seriously hinders specific identification of leukemic cells. On the other hand, subtracting background levels might under-estimate MRD frequencies and this could result in false-negatives. High specificity and thereby high sensitivity can be achieved when only the most specific immunophenotypic aberrancies are used, i.e. with no expression in normal cells. Inclusion of markers/marker combinations that allow excluding non-specific events in a multi-color approach may increase specificity. This is already shown for the transition of a four to five-color flow cytometric approach . Another explanation for MRD misclassification is low sensitivity of the aberrant immunophenotype. Marker expression may be highly heterogeneous in an AML sample: LAPs may thereby often not be expressed on the total population of blast cells, thereby, at follow up, preventing the identification of all leukemic cells. Improvements can only be expected with the discovery of new aberrancies that cover larger parts of diagnosis blast cells. At present, with the large differences in specificity and sensitivity of LAPs the level of detection of MRD varies between patients: 1:10,000 or even better may be reached in one patient, while in another patient 1:1,000 may be the best attainable. Besides misclassification, immunophenotype shifts can also contribute to false-negative observations. To reduce this, it is recommended to use multiple LAPs for MRD monitoring [6,12,13]. Recently, it has become clear that such shifts may occur through clonal selection: while major molecular clones may disappear upon therapy, minor diagnosis clones may survive chemotherapy treatment, and grow out to relapse . This may be accompanied by immunophenotype changes . More efforts towards recognition of minor clones at diagnosis, that potentially can expand to cause relapse, may identify emerging molecular clones and immunophenotypes instead of disappearing molecular clones and immunophenotypes only. For molecular MRD, in fact most of the pitfalls for immunophenotypic MRD hold here as well. Similar to MFC, multiple molecular MRD studies have reported patients with low molecular MRD levels that still suffer from relapse [25,26,37,38,42]. Underlying causes may include 1) as argued earlier for different LAPs, Q-PCRs for different mutations and fusion genes reach different sensitivities as well; 2) part of the blasts may only be characterized by the molecular aberrancy of interest; and 3) molecular clone shifts occur between diagnosis and relapse. To avoid these false-negative results different molecular markers, if present in the patient, could be quantified for MRD monitoring. There are no real solutions for these problems unless more generally applicable, specific and stable markers are discovered. Until then, combining as many molecular and immunophenotypic targets may contribute to accurate MRD based prediction of relapse. Another possible explanation for finding false-negative MRD results is the fact that it may not only be the number of leukemic blasts that determines the chance for relapse, but more specifically the number of leukemic stem cells (LSCs). These LSCs can cause tumor outgrowth, thereby leading to MRD and finally resulting into overt disease relapse . Although these stem cells are much less frequent than whole blast MRD, LSC frequency assessment may offer an additional specific and biologically relevant determinant of risk on relapse. In section 3 the role of leukemic stem cells in acute myeloid leukemia will be further discussed.
2.4.2. Alternative parameters for risk stratification
Perhaps the conceptually simplest method to evaluate treatment response is calculating the decrease rate of peripheral blasts shortly after treatment
3. Leukemic stem cells and acute myeloid leukemia
3.1. Definition of leukemic stem cells
It was hypothesized that a small population of cells, distinct from the bulk of tumor cells, is responsible for tumor initiation and growth in various cancers, including AML [91,92]. These cells are referred to as leukemic stem cells (LSCs) or leukemia-initiating cells (LICs). It is assumed that similar to normal hematopoiesis, leukemia is hierarchically structured. In many respects LSCs resemble normal hematopoietic stem cells (HSCs). Similar to HSCs, LSCs are defined by their ability to undergo self-renewal and the capacity to differentiate to a limited, although highly variable, extent [93,94]. Furthermore, the immunophenotype of LSCs resembles the immunophenotype of normal HSCs. The majority of HSCs are present in the CD34+CD38- immunophenotypic compartment [95,96] and initial AML studies demonstrated leukemia initiating capacity also to be in the CD34+CD38- compartment . This small subpopulation of CD34+CD38- cells was able to engraft and cause leukemia in non-obese diabetic/sever combined immune-deficient (NOD/SCID) mice. These cells were present at a frequency of only 0.2 to 100 cells per 106 mononuclear cells . Nowadays it is known that AML LSCs can also reside within the CD34+CD38+ or the CD34- immunophenotypic compartment [98-102]. There is growing evidence that the transformation of a normal human cell into a LSC not only can occur in a normal HSC, but also in a normal progenitor cell . Mutations in a normal progenitor cell may confer self-renewal properties to progenitors. A recent study demonstrated that CD34+CD38- LSCs, despite the immunophenotypic similarities with normal HSCs, are most related to normal progenitors instead of normal stem cells . In addition, it has been demonstrated that within a patient, the pool of LSCs at diagnosis is often largely heterogeneous. This implies that different subpopulations of LSCs often coexist at diagnosis [84,101] (Figure 1). In CD34 positive patients often both CD34+CD38- cells, CD34+CD38+ and CD34- cells are present and all are able to show leukemic engraftment when infused separately in NOD/SCID mice. However, no information exists on possible competition between these compartments in leukemogenesis. Moreover, the CD34+CD38- compartment has been shown to be less immunogenic compared to the other compartments , which may explain why it was almost exclusively the CD34+CD38- compartment that engrafted in NOD/SCID mice with residual immunity , while in the severely immunocompromised later mouse models, the other compartments engrafted as well. In CD34 negative AML by definition, the CD34- compartment and in particular the CD34-CD38+ compartment contain LSCs . For clinical treatment and patient survival it is important to know which putative LSC will survive therapy. In that respect it is important to realize that the CD34+CD38- compartment has been shown to be most therapy resistant
In the course of time other compartments enriched for LSCs have been identified. These are based on functional properties and include aldehyde dehydrogenase (ALDH) activity and drug efflux (Hoechst) capacity. ALDH is a group of cytosolic enzymes that catalyze the oxidation of aldehydes. It plays an important role in the retinoid metabolism, since it is required for the conversion of retinol (Vitamin A) to retinoic acids. For maturation, loss of quiescence and differentiation of HSCs, these retinoic acids are important [105,106]. Furthermore, ALDH activity is supposed to protect cells from the toxic effects of cyclophosphamide and therefore high ALDH expression in leukemic cells may play a role in chemotherapy resistance [107,108]. Recently it has been shown that leukemic cells and normal hematopoietic cells differ in ALDH activity. Normal stem- and progenitor cells have high ALDH expression [109-112]. It has to be emphasized that it has recently been demonstrated that the population of cells with intermediate ALDH activity appeared to be enriched for leukemic CD34+CD38- cells [113-115]. Several authors have confirmed the leukemia initiating capacity of these cells in NOD/SCID mice [116-118].
Another functional stem cell compartment is the so-called side population (SP). These SP cells are primarily defined by their capability of efficient Hoechst 33342 dye efflux and especially by the way in which fluorescence emission of Hoechst is recorded. In normal BM a population of CD34+CD38- cells was found in the SP [119,120]. In AML, it has been demonstrated that the SP compartment contains a heterogeneous population of cells, containing HSCs, LSCs, LSC progenitors and early lymphocytes . AML SP cells have shown to be able to initiate acute leukemia in NOD/SCID mice [122,123]. All these immunophenotypic and functional findings are important for gaining insight in the process of leukemogenesis and especially for the development of new therapies aiming at eradication of LSCs.
Besides the ability of LSCs to initiate and sustain the initial AML, there is increasing evidence pointing towards the importance of LSCs in the occurrence of MRD and the emergence of a relapse. LSCs are thought to be more resistant to standard chemotherapy compared to the total population of malignant blast cells and therefore these LSCs are able to escape apoptosis. Other essential LSC features are their acquired capacity for self-renewal and proliferation. Such properties allow LSCs to survive chemotherapy treatment, to divide and to grow out and cause a relapse (Figure 1). Consequently, identification and characterization of LSCs is fundamental to gain insight in the mechanisms that underlie relapse and how to evade relapse.
3.2. Identification of leukemic stem cells
Since the assumed role of LSC in the emergence of an AML relapse, identification of these probably most malignant cells becomes imperative. The hypothesis would thus be that quantitation of LSCs in AML patients would give important information about treatment response and risk of relapse. Similar to MRD identified by flow cytometry, LSCs in BM can be identified using cell surface antigen expression. As mentioned before, LSCs can reside in different immunophenotypic compartments, but, as argued before, the CD34+CD38- defined LSCs may be most malignant/resistant [84,104]. Since both HSCs and LSCs reside within this compartment, discrimination between CD34+CD38- HSCs and LSCs is challenging. Immunophenotypic LSC detection is often possible making use of the fact that the lineage marker combinations used for MRD detection, are frequently aberrantly expressed on CD34+CD38- cells too . These lineage markers include CD2, CD7, CD11b, CD13, CD15, CD19, CD22 CD33, CD56 and HLA-DR. Combinations of lineage markers could also be used, like CD33+CD13- and CD15+HLA-DR-. Besides these lineage markers, a growing number of other markers is now available to discriminate between LSCs and HSCs. These include CLL-1 CD25, CD32, CD33, CD44, CD47, CD96, CD123 and TIM-3 (Figure 3). An overview of LSC markers is given in Table 3.
It is important to realize that there is a large heterogeneity in marker expression. This implies that marker expression differs between AML patients and even within an individual patient different stem cell markers are often differentially expressed (Figure 4). Thus, none of the individual markers are expressed in all AML cases. For accurate LSC detection, high specificity of stem cell markers is essential. Both CLL-1 and lineage markers have proven to be highly specific, since these are present on leukemic CD34+CD38- cells in a substantial part of the AML patient population, but absent on normal CD34+CD38- cells, also after chemotherapy [124,125]. For the other stem cell markers high specificity and stability during treatment/disease still have to be confirmed. The established differences in ALDH activity between CD34+CD38- LSCs and CD34+CD38- HSCs were confirmed using this aberrant marker approach [114,115], thereby strengthening that the functional ALDH assay offers an alternative tool for CD34+CD38- LSC identification, which importantly, could be applied in absence of aberrant antigen expression. In contrast, the SP phenotype does not discriminate between HSCs and LSCs since both may be present in the SP compartment. Here the immunophenotypic marker approach is necessary to discriminate between LSCs and HSCs . Both ALDH and SP assays not only identify leukemia initiating cells with the CD34+CD38- immunophenotype, but also other cell types, like CD34+CD38+ progenitors or CD34- cells [114,115,117].
Although functional assays, like ALDH and SP, are complex and time-consuming compared to standard immunophenotypic LSC detection, they may offer promising alternatives for CD34+ AML patients without detectable CD34+CD38- cells, as well as for AML patients who are defined as CD34 negative. The latter patients usually have less than 1% expression of CD34 on the leukemic blast cells which all are of non-neoplastic origin 
|CLL-1||C-type lectin-like molecule-1|||
|Lineage markers||Lymphoid lineage and myeloid lineage markers|||
|CD25||Interleukin-2 receptor α-chain|||
|CD32||Fc fragement of IgG, low affinity IIa receptor|||
|CD44||Receptor for hyaluronan|||
|CD47||Integrin associated protein|||
|CD96||T cell-activated increased late expression protein|||
|CD123||Interleukin 3 receptor alpha chain|||
|TIM-3||T-cell Ig mucin-3|||
Seen the large clonal heterogeneity at diagnosis [84,101], and the possibility that not just the major clone at diagnosis, but often low-frequency CD34+CD38- clones may grow out 
CD34+CD38- population was analyzed for the expression of six aberrant markers: CD2 (A), CLL-1 (B), CD22 (C), CD96 (D), CD123 (E), CD11b (F). Expressions percentages for marker positive and marker negative CD34+CD38- cells are shown for each marker.
3.3. Prognostic value of LSC frequency
Since it has been hypothesized that the subpopulation of chemotherapy resistant LSCs is responsible for relapse, LSC frequency, similar to MRD frequency, should have direct prognostic impact.
|Cut-off||Number of patients above cut-off||Relative risk of relapse||95% C.I.|
|First cycle||5 x 10-6||14||5.0||1.8-14.0|
|Second cycle||5 x 10-6||18||4.7||2.2-10.1|
|Consolidation||2 x 10-6||14||8.5||1.8-41.4|
All together, several studies showed CD34+CD38- LSC frequency to be an independent prognostic risk factor. Important to emphasize, however, is that these studies focus on LSC detection and quantification at AML diagnosis. Because LSCs are hypothesized to be chemotherapy resistant and to grow out after treatment and then cause a relapse, it would be of utmost importance to study the frequency of these LSCs during follow-up. For the first time we also demonstrated that the frequencies of LSCs after different courses of therapy significantly correlated with clinical outcome . More effort is needed to identify LSCs and their prognostic value in immunophenotypic compartments other than CD34+CD38-, like the CD34+CD38+ and CD34- compartment using the ALDH and SP assay. Ultimately, when the clinical importance of different stem cell compartments have been prospectively confirmed, this, together with MRD based strategies, should offer new diagnostic tools to guide clinical intervention and to monitor effectiveness of therapy and, moreover, to design new therapies that specifically target LSCs while leaving the normal HSCs intact.
3.4. Leukemic stem cell targeted therapy
Apart from the clinical application of LSCs, characterization of these malignant cells offers the design of new therapies that specifically target LSCs while leaving the normal HSCs intact. The most direct example of such therapy is the application of antibodies that are used to specifically discriminate between LSC and HSC. CD123 and CD33 are examples. It has been reported, using NOD/SCID mice, that treatment with the anti-CD123 antibody 7G3 improved mouse survival . A humanized version of the anti-CD123 antibody (CLS360) has been studied in a phase 1 study in relapsed, refractory and high-risk AML patients. Interim analysis showed no treatment related toxicity, besides two mild infusion reactions and one infection possibly related to the treatment. Of eight patients treated with CLS360, one CR had been observed 
CD33 is expressed on leukemic blasts in 85%-90% of AML patients and therefore, already years ago, it had been suggested as a potential target for anti-AML therapy. The CD33 immunoconjugate gemtuzumab ozogamicin (Mylotarg) has been studied in several trials and, after initial disappointment relating to toxicity, new studies with altered treatment schedules suggest that Mylotarg is beneficial in certain subgroups of AML patients, including patients with favorable cytogenetics 
4. Conclusions and future perspectives
MRD frequency assessments by RQ-PCR and MFC in AML patients are more sensitive methods to define remission status compared to current morphologic assessment. Although RQ-PCR is in general the most sensitive technique, MFC is applicable in almost all AML patients. Since the importance of flow cytometric MRD detection has now been validated in a first prospective study, it is of utmost importance that, when these data are confirmed in other prospective studies, MRD status will be implemented in clinical decision-making. We have described that alternatives for BM MRD may include MRD assessment in peripheral blood and blast reduction, frequency of B-lymphocytes precursors and CD34+ myeloid/lymphoid ratios. It thus seems that development of algorithms including all such parameters may ultimately contribute to improved detection of residual therapy resistant cells and early and accurate prediction of relapses. Also, based on the observation of immunophenotypic and molecular shifts, occurring between diagnosis and relapse, a new issue in MRD research may be that not only disappearing phenotypes, but also emerging “new” phenotypes have to be monitored. An alternative, probably more specific method to predict clinical outcome is LSC frequency assessment. Results so far on the clinical importance of LSCs are limited, but very promising, especially since for the first time the correlation between the presence of LSCs after treatment and clinical outcome has been reported. When the value of LSC assessment is confirmed in other retrospective and eventually prospective studies, it may be hypothesized that in the future, not only MRD, but also LSC frequency assessment may be implemented in clinical decision-making.
Hopefully, using the suggested approaches in this chapter, it will become possible to significantly improve clinical outcome of acute myeloid leukemia patients.