Identification of a Chemoresistant “Oxidative State-Low” Leukemic Subpopulation in CD34+ Human Acute Myeloid Leukemia

Reactive oxygen species (ROS) are oxygen-containing side products of normal aerobic metabolism. Although it was initially believed that ROS constitute only damaging by-products, recent evidence suggest that ROS actively participate in signal transduction pathways and regulate cell survival and proliferation in a dose-depended manner [1,2]. Endogenous intracellular redox state is tightly controlled by free radical scavenging systems and the appropriate ROS levels differ among cell types and differentiation status [3]. Hematopoietic stem cells (HSCs) in mice reside mainly in the ROSlow subset of hematopoietic progenitors [4], whereas forced increase of ROS in human Lin-CD34+CD38cord blood cells leads to loss of stem cell activity [5-7]. Similarly, though increased ROS levels in tumor cells have been implicated in cancer initiation and progression.


Introduction
Reactive oxygen species (ROS) are oxygen-containing side products of normal aerobic metabolism. Although it was initially believed that ROS constitute only damaging by-products, recent evidence suggest that ROS actively participate in signal transduction pathways and regulate cell survival and proliferation in a dose-depended manner [1,2]. Endogenous intracellular redox state is tightly controlled by free radical scavenging systems and the appropriate ROS levels differ among cell types and differentiation status [3]. Hematopoietic stem cells (HSCs) in mice reside mainly in the ROS low subset of hematopoietic progenitors [4], whereas forced increase of ROS in human Lin -CD34 + CD38cord blood cells leads to loss of stem cell activity [5][6][7]. Similarly, though increased ROS levels in tumor cells have been implicated in cancer initiation and progression.
Cancer stem cells (CSCs) maintain a low redox status, apparently through upregulation of anti-oxidative enzymes [1,2,8]. Elevated ROS levels drive out of quiescence both normal and clonal stem cells, while administration of N-acetylcysteine, an antioxidant, can partially restore the self-renewal capacity [9]. Moreover, ROS production via constitutive activation of the Ras oncogene appears to drive the leukemic progression [10]. In acute myeloid leukemia (AML) free radicals, antioxidant enzymes and oxidative damage products are elevated, but the redox state in leukemia blasts has not been thoroughly characterized and the role of ROS in leukemogenesis is still unclear [11][12][13][14]. Most adults with acute myeloid leukemia (AML) are not cured due to chemoresistance or relapse. AML may be organized in a similar way to normal hematopoiesis and, hence, disease recurrence could be due to a subset of leukemic stem cells (LSC) [15]. Here, we assessed ROS levels in primary AML blasts and we identified a distinct ROS low subpopulation with a quiescent, chemoresistant phenotype and leukemia-initiating cell activity (LIC).

Patients
Bone marrow (BM) or peripheral blood (PB) was obtained from 44 newly diagnosed, CD34 + AML patients (Table 1) after informed consent and Institutional Review Board approval. AML cell lines (U937, THP1) were obtained from DSMZ (Braunschweig, Germany). Mononuclear cells were isolated over Ficoll-Hypaque and used either immediately or after cryopreservation. sensitive fluorescence dye 2'7;-dichlorodihydrofluorescein diacetate (DCF, Molecular Probes Inc) as previously described [16]. Relative ROS levels in AML blasts were determined as the ratio of the Mean Fluorescence Intensity (MFI) values of DCF in AML CD34+ cells to the MFI of total lymphocytes. For cell-cycle analysis sorted subsets were stained with Ki-67 and 7-AAD after fixation and permeabilization with Cytofix/CytoPerm (BD Biosciences).

Fluorescence in situ hybridization (FISH)
FISH was performed in sorted ROS low live cells according to standard protocols, using the Vysis D7S522/CEP7 FISH Probe Kit. In total, approximately 200 nuclei were scored in each of 3 patients bearing the -7 cytogenetic abnormality.

Immunophenotypic analysis of leukemic hematopoietic hierarchy
Cells expressing mature lineage markers were depleted from mononuclear cells after staining with an antibody cocktail consisting of anti-CD2, CD3, CD4, CD8, CD19, CD20 and GPA [17] and anti-PE immunomagnetic MicroBeads (Miltenyi Biotec). Purified Lincells were then subjected to 5-color immunophenotypic analysis to define the maturation stage of CD34 + AML and ROS low cells by concomitant staining with DCF and CD34, CD38, CD90, CD123 and CD45RA.

Xenograft assay
The in-vivo engraftment capacity of sorted cells was analyzed as described elsewhere [19]. Briefly, cells (2x10 5 ) ROS low cells or 10 6 -10 7 bulk cells from each patient were intravenously injected into non-obese diabetic severe combined immunodeficient (NOD-SCID) mice and leukemic engraftment was assessed 12 weeks after transplantation.

Single-cell phosphospecific flow cytometry
For the determination of the levels of phosphorylated pSTAT-3 and pSTAT-5, sorted total CD34 + AML and ROS low cells were either left untreated, or were stimulated with Granulocyte-Colony Stimulating Factor (G-CSF) or Granulocyte Macrophage Stimulating Factor (GM-CSF) (Miltenyi Biotec GmbH, Germany) for 15 minutes at 37ºC. Stimulation was halted by fixation with Cytofix Fixation Buffer (BD Biosciences) and then cells were permeabilized with Perm Buffer III (BD Biosciences) and stained with the following antibodies: CD34 (clone 8G12), phospho-STAT3 (clone Y705), and phospho-STAT5 (clone Y694). Fluorescence minus one (FMO) controls were used and basal phosphorylation levels of pSTAT3 and pSTAT5 were expressed as log2

Statistical analysis
All analyses were performed using SPSS 17.0 software (SPSS Science, Chicago, IL). Data are presented as mean ± SEM. The significance of the differences was assessed by unpaired or paired Student's t test and one-way ANOVA as appropriate. Correlations of ROS MFI and the percentage of ROS low subset with age (n=44), karyotype (n=26) and response to initial treatment (n=30) were performed using Pearson's correlation.

ROS content in primary AML cells
Intracellular ROS levels of FACS-sorted CD34 + AML cells exhibited a wide variation among the patient samples (median relative MFI: 3.41, range 0.6-11.52, Figure 1a). In order to test the potential interference of exogenous factors in the fluorocytometric assessment of ROS using the DCF dye, e.g. the immediate exposure of cells to higher oxygen after isolation, or modulation by the activity of P-glycoprotein (P-gp) [20], we performed repeated measurements in a) paired samples before and after the 2-hour sorting procedure, b) in paired fresh and freeze-thawed samples and c) upon treatment with the P-glycoprotein efflux pump inhibitor cyclosporine A (supplementary Figure 1). All measurements showed little variation confirming that the heterogeneous DCF pattern in AML cells reliably reflects the endogenous redox status of AML blasts.
It has been reported that internal tandem duplication of the FLT3 gene (FLT3-ITD) increases ROS production potentially via overactive STAT5 signaling [11], thus the heterogeneity of ROS content may, at least in part, mirror the diversity of oncogenic signaling pathways. Nevertheless, we found no differences in ROS levels between FLT3-ITD positive and negative AML patients (p=0.4, Figure 2). Likewise, there was no association of ROS levels with age, karyotype, AML subtype, i.e. secondary or de novo, and response to induction chemotherapy (Figure 2a-c).

Identification and characterization of a novel subpopulation of ROS low AML blasts
Unexpectedly, we noticed that in contrast to the AML cell lines U937 and THP-1, both of which expressed ROS homogeneously, primary blasts showed a diversified DCF/DA staining with a discrete fraction of ROS low cells at the left tail of each distribution (median=0.61% of CD34 + cells, range 0.2% to 4.99%, Figure 1b). To confirm that the ROS low subpopulation contained exclusively leukemic cells and did not represent contaminating normal cell populations or debris, we FACS-sorted ROS low events from 3 individuals with AML and subjected them to morphological evaluation after staining with May-Grünwald Giemsa, cytometric analysis of aberrant surface marker expression and fluorescence in situ hybridization (FISH) analysis of specific cytogenetic abnormalities. All analyses confirmed the clonal nature of ROS low cells (Figure 1c-f).
Similar to the intracellular ROS levels, the prevalence of ROS low cells did not differ based on age, gender, karyotype, FLT3-ITD status, and response to initial treatment (Figure 2a-d). By contrast, patients with secondary AML evolving from myelodysplastic syndrome (MDS) displayed a twofold increase of ROS low subpopulation compared with de novo AML patients (1.44% ± 0.32% vs. 0.78 ± 0.12%, respectively, p=0.029, Figure 2d). Notably, the ROS levels and the percentage of ROS low subset were not intercorrelated and were both identical among BM and PB (supplementary Figure 2).
Recent findings challenge the leukemia stem cell (LSC) model and suggest that in approximately 90% of CD34 + AML cases LSCs reside in the granulocyte-macrophage progenitor (GMP) and lymphoidprimed multipotential progenitor (LMPP) compartments (LMPP/ GMP-like AML) [17]. We observed that in five LMPP/GMP-like AML samples, the ROS low fraction contained significantly more CMP-like (p=0.04) and less GMP-like cells (p=0.03) compared to the bulk AML population. Also, LMPP-like cells were lower in the ROS low subset, while megakaryocyte-erythroid progenitor (MEP)-like cells were higher, but statistical significance was not reached for both progenitor types (Figure 3a-c). Of note, CD38 expression was identical among ROS low and total leukemia blasts (supplementary Figure 3). Although these results suggest that the ROS low subset does not reside in the recently proposed LSC compartment of LMPP/GMP-like AML, when we assessed LIC activity in 16 AML samples by transplanting both sorted ROS low and total leukemia blasts from each patient in NOD/ SCID mice we observed identical levels of primary engraftment in 3/16 samples (median values: ROS low 1.4% vs. total blasts 1.5%), indicating that ROS low cells have potentially comparable leukemogenic potential with the bulk leukemic population (Figure 4a).
Next, we evaluated quiescence and chemoresistance, two cardinal properties of leukemia propagating cells, in sorted ROS low cells. Cell cycle analysis revealed that significantly more ROS low cells resided in the G0 phase compared to the bulk leukemic population (70% ± 7.4% vs. 51% ± 9.2%, respectively, p=0.027, Figure 4b-c). Additionally, in in vitro chemosensitivity assays, sorted ROS low cells exhibited significantly increased chemoresistance in daunorubicin and cytarabine compared to the bulk AML population (p=0.02, Figure 4d), whereas in a different set of experiments performed by using total blasts only, the proportion of ROS low viable, non-apoptotic cells increased significantly after culture with either daunorubicin or cytarabine (supplementary Figure 4). G-CSF and GM-CSF-induced overexpression of the phosphorylated form of STAT5 in AML blasts associates with resistance to induction chemotherapy [21,22], whereas the magnitude of GM-CSF-induced activation of pSTAT5 in Juvenile Chronic Myeloid Leukemia, a typically chemoresistant malignancy, correlates with disease status [23].

Discussion
The concept of leukemia stem cells which are able to initiate and propagate the disease is now well established [24]. Accurate identification and targeting of LSCs is of obvious significance for the effective treatment and the complete eradication of the leukemic clones. However, the characterization of LSCs by using surface markers is highly inconsistent among the various research groups, whereas novel data suggest that LSC may even share a progenitor rather than a stem cell phenotype [17,25,26]. Numerous data are now accumulating on the role of energy metabolism and in particular the redox state of cancer stem cells. Recently is has been shown that, similar to the normal HSCs in mice, most LSCs in human AML have low ROS levels [27]. However, no distinct subpopulation with low redox levels was defined. In the present work, we report for the first time the identification of a novel rare subpopulation of CD34 + ROS low AML cells with a chemoresistant, LSC-like phenotype.
In line with recent findings [27], we observed high interpatient variability of ROS content in AML CD34 + cells. The biology under the heterogeneity of ROS staining is obscure. Although leukemogenic pathways via Ras [10] or FLT3 mutations [28] have been associated with increased redox state, we were not able to find a correlation of ROS levels with either the FLT3/ITD mutation, or with clinical parameters of disease aggressiveness. Though the limited size of our cohort weakens the statistical power of the above correlations, ROS homeostasis is characterized of high complexity [3] thus it appears rather unlikely that a single mutation would be responsible for the observed heterogeneity in ROS levels.
The staining pattern of DCF/DA is typically a continuum of intensities, therefore arbitrary cut-offs are often used to define low, "normal" and high levels of ROS [4,27]. By contrast, we identified a small clonal AML subpopulation with very low redox levels, clearly distinguishable from the bulk AML cells. The prevalence of ROS low cells was not correlated to the total ROS content and, though it was not associated with poor risk features, it was significantly increased in patients with sAML. A possible explanation for this finding is that the oxidative stress in the progenitor cell compartment is more intense in MDS compared to AML and potentially selects for resistant, ROS low progenitors [11,29,30]. As mentioned above, the accurate immunophenotypic characterization of LSCs is controversial. In an elegant paper Goardon and colleagues proposed that LSCs in the vast majority of AML patients share progenitor rather than stem cell characteristics, similar to the normal granulocyte-macrophage progenitors (GMP) and the currently poorly characterized lymphoid-primed multipotential progenitor (LMPP) [17]. In 5 patients with LMPP/GMP-like AML we observed that the ROS low fraction contained less LMPP and GMP-like cells compared to the total leukemic population, whereas it was enriched in MEP-like cells implying a more mature phenotype of ROS low cells. Nevertheless, ROS low cells exhibited comparable primary engraftment capacity in xenotransplantation assays with the bulk leukemia cells. Due to the rarity of ROS low cells and the rigorous sorting procedure we could not perform secondary transplantations. However, recent data confirm that AML cells with low ROS levels are enriched in LICs [27], although the authors arbitrarily defined ROS low cells as the 15% brightest dye fluorescence distribution of DCF, thus encompassing a considerably larger population of AML cells compared to the ROS low subset in our study [27].
Quiescence and chemoresistance are considered key characteristics of LSCs [31]. Consistent with previous observations, reporting a strong correlation of ROS levels with the cycling status of normal HSCs [11], we found that, in contrast to the total AML cells, almost three-fourths of ROS low cells were in G0 phase. Additionally, the ROS low subset displayed significantly augmented in vitro chemoresistance to conventional chemotherapeutic agents, which is in line with the reported negative relationship between ROS levels and chemo-and radio resistance of cancer stem cells [2,8,32]. The chemoresistant phenotype of ROS low cells was further corroborated by the over-exuberant response of pSTAT5 after perturbation with GM-CSF, a cell signaling aberration associated with aggressive biologic behavior [21,22].
To conclude, we have identified a novel, biologically distinct ROS low subpopulation in CD34 + AML. Recent data show that mitochondrial priming via BH3-only proteins, which engage a ROS-dependent proapoptotic mechanism [33], correlates with the chemosensitivity of AML blasts [34], thus providing a potential explanation for the chemoresistant phenotype of ROS low cells. Correlational studies of clinical and biological data with the alterations of the ROS low subset in larger patient cohorts will help to determine the exact role of this novel subpopulation in key management issues in AML, such as prognostic assessment, detection of minimal residual disease and cell-specific therapeutic targeting.