Single-cell mass cytometry reveals intracellular survival/proliferative signaling in FLT3-ITD-mutated AML stem/progenitor cells

Abstract

Understanding the unique phenotypes and complex signaling pathways of leukemia stem cells (LSCs) will provide insights and druggable targets that can be used to eradicate acute myeloid leukemia (AML). Current work on AML LSCs is limited by the number of parameters that conventional flow cytometry (FCM) can analyze because of cell autofluorescence and fluorescent dye spectral overlap. Single-cell mass cytometry (CyTOF) substitutes rare earth elements for fluorophores to label antibodies, which allows measurements of up to 120 parameters in single cells without correction for spectral overlap. The aim of this study was the evaluation of intracellular signaling in antigen-defined stem/progenitor cell subsets in primary AML. CyTOF and conventional FCM yielded comparable results on LSC phenotypes defined by CD45, CD34, CD38, CD123, and CD99. Intracellular phosphoprotein responses to ex vivo cell signaling inhibitors and cytokine stimulation were assessed in myeloid leukemia cell lines and one primary AML sample. CyTOF and conventional FCM results were confirmed by western blotting. In the primary AML sample, we investigated the cell responses to ex vivo stimulation with stem cell factor (SCF) and BEZ235-induced inhibition of PI3K and identified activation patterns in multiple PI3K downstream signaling pathways including p-4EBP1, p-AKT, and p-S6, particularly in CD34⁺ subsets. We evaluated multiple signaling pathways in antigen-defined subpopulations in primary AML cells with FLT3-ITD mutations. The data demontrated the heterogeneity of cell phenotype distribution and distinct patterns of signaling activation across AML samples and between AML and normal samples. The mTOR targets p-4EBP1 and p-S6 were exclusively found in FLT3-ITD stem/progenitor cells, but not in their normal counterparts, suggesting both as novel targets in FLT3 mutated AML. Our data suggest that CyTOF can identify functional signaling pathways in antigen-defined subpopulations in primary AML, which may provide a rationale for designing therapeutics targeting LSC-enriched cell populations.

Introduction

AML is believed to be propagated from rare cell populations known as leukemia stem cells (LSCs) (1–4). These cell populations are thought to frequently reside within the CD34⁺CD38⁻compartment (5). Several distinct markers have been shown to be expressed preferentially on LSCs, including CD123, CD99, CD33, CD13, CLL-1, CD96, TIM3, CD47, CD44, CD32 and CD25 (6–12). LSC-specific signaling pathways have been identified in myeloid leukemia (13, 14). Recent data from our group using reverse-phase protein array technology suggest a selective protein signature within the LSC/progenitor cell subsets in molecularly defined subtypes of AML(15). However, these datasets were generated from FACS-isolated cell populations, and, as such, represent population averages and do not reflect protein expression at a single-cell level.

The exact phenotype of AML LSCs remains undefined. Initially the LSCs were reported in CD34⁺CD38⁻ cells (5); however, recent studies using more sensitive immunodeficient mice models have demonstrated the CD34⁺CD38⁺ and CD34⁻ cells also possess stem cell capacity (16, 17). A better understanding of the unique phenotypes and complex signaling pathways of LSCs would have implications for the utilization of targeted agents aimed at eradication of these cells. Recent advances in whole-genome sequencing have identified an unexpected clonal heterogeneity of great clinical significance, since minor sub-populations can become dominant and drug resistant (18). While the exact phenotypic attributes of these cells are still largely unknown, high-parametric FCM has contributed substantially to their characterization.

Conventional FCM offers great potential for further dissection of the clonal architecture of this highly heterogeneous malignancy. Multi parametric FCM has been applied to dissect intracellular signaling networks within antigen-defined subsets of AML at the single-cell level, revealing correlations between kinase phosphorylation and patient outcome (19–21). Routine FCM can analyze as many as 12 parameters simultaneously (22), and the feasibility of measuring 17 distinct colors has been demonstrated by polychromatic FCM in the research laboratory (23). However, for polychromatic FCM, the excitation and emission spectra of the fluorescent dyes need to be reconciled by a mathematical algorithm implemented in a process known as compensation. Thus cell autofluorescence and fluorescent dye spectral overlap limit the application of conventional FCM to AML LSCs. Although newer, more powerful lasers, additional fluorochromes, and high-affinity antibodies are available today, FCM has changed little since it was developled by Herzenberg et al. and others 45 years (24).

Single-cell time-of-flight mass cytometry (CyTOF) is an entirely new, transformative technology, developed at the University of Toronto, that has been successfully utilized for analysis of immune signaling in normal bone marrow (25–27). CyTOF substitutes transition elements (e.g., lanthanides) for fluorophores as Ab labels, allowing measurement of up to 120 parameters in single cells without correction for spectral overlap. The labeled cells are individually atomized at ~5500K, and the isotope ion cloud that is released is quantified by inductively coupled plasma time-of-flight mass spectrometry. Using this technique, Bendall, et al. identified differential signaling patterns in distinct bone marrow populations in response to various cytokines and kinase inhibitors through simultaneous measurement of 34 parameters (25).

The expression levels of surface lineage markers and intracellular proteins have been demonstrated to be comparable when measured by conventional FCM and CyTOF in normal bone marrow and peripheral blood cells (25, 28). However, the utility of CyTOF in identifying phenotypes and signaling pathways in primary AML LSC populations remains to be determined. For this study, we developed a panel of antibodies that could be used with CyTOF to dissect LSC and signaling pathways in AML cells. Cell-surface phenotypes and validated protein analyses were confirmed by conventional FCM and western blotting. Finally we applied this novel technology for the analysis of primary AML samples harboring FLT3-ITD mutations to investigate activation of intracellular phosphoproteins within distinct leukemia subtypes.

Materials and Methods

Cells

The study protocol was approved by the Institutional Review Board (IRB) of The University of Texas MD Anderson Cancer Center. Peripheral blood samples were obtained from patients with AML after informed consent was obtained in accordance with IRB regulations. Normal bone marrow specimens were collected from healthy donors after informed consent was obtained. Mononuclear cells (MNCs) were separated from the blood samples by Ficoll-Hypaque density gradient centrifugation (Sigma-Aldrich, St Louis, MO). The isolated MNCs were cryopre served in 90% fetal bovine serum (FBS) and 10% DMSO. Myeloid leukemia cell lines TF-1 (American Type Culture Collection, Manassas, VA) and Mo7e (kindly provided by Prof. Z. Estrov, MD Anderson Cancer Center)were routinely propagated in RPMI 1640 medium (Cat. 10-040-CV; Mediatech, Inc. Manassas, VA) with 10% FBS, 1% penicillin/streptomycin, and GM-CSF (2 ng/mL for TF-1, and 20 ng/mL for Mo7e, Cat. NDC 58406-002-01; Immunex Corporation, Seattle, WA).

Antibodies

The antibody clones used in this study, with their manufacturer, and catalogue number, are listed in Table 1.

Table 1.

Antibodies used in CyTOF, FCM, and western blotting analyses

Antigen	Conjugate	Clone	Catalogue Number	Supplier
Mass Cytometry
CD34	148Nd	4H11	14-0349-82	eBioscience
CD38	168Er	HIT2	303502	BioLegend
CD123	151Eu	6H6	3151001B	DVS-Sunnyvale
CD99	165Ho	TU12	555687	BD Biosciences
CD45	154Sm	HI30	3154001B	DVS-Sunnyvale
CD90	143Nd	5E10	328102	BioLegend
CD25	171Yb	BC96	302602	BioLegend
CD7	139La	6B7	343102	BioLegend
CD64	146Nd	10.1	3146006B	DVS-Sunnyvale
CD56	162Dy	NCAM16.2	559043	BD Biosciences
CD2	163Dy	RPA-2.10	300202	BioLegend
CD15	164Dy	W6D3	3164001B	DVS-Sunnyvale
CD41	166Er	HIP8	303702	BioLegend
CD3	170Er	UCHT1	3170001B	DVS-Sunnyvale
CD11b	173Yb	ICF44	301302	BioLegend
p-AKT (Ser473)	159Tb	M89-61	560397	BD Biosciences
p-STAT5 (Tyr694)	175Lu	C71E5	9314BF	Cell Signaling Technology
p-ERK1/2 (Thr202/Tyr204)	167Er	D13.13.4E	3167005A	DVS-Sunnyvale
p-4EBP1 (Thr37/46)	147Sm	236B4	2855BF	Cell Signaling Technology
p-NFκB p65 (Ser536)	149Sm	93H1	NA	DVS-Sunnyvale
p-STAT3 (Tyr705)	152Sm	M9C6	4113BF	Cell Signaling Technology
p-mTOR (Ser2448)	164Dy	Polyclonal	A00643	GenScript
p-S6 (Ser235/236)	172Yb	N7-548	3172008A	DVS-Sunnyvale
p-PI3K	160Gd	Polyclonal	4228BF	Cell Signaling Technology
FCM
CD34	APC	581	555824	BD Biosciences
CD38	PE-Cy7	HB7	335790	BD Biosciences
CD123	Percp-Cy5.5	7G3	558714	BD Biosciences
CD99	FITC	TU12	555687	BD Biosciences
CD45	APC-Cy7	2D1	557833	BD Biosciences
p-AKT (Ser473)	Alexa Fluor 647	193H12	A88881	Beckman Coulter, Inc.
p-Stat5 (Tyr694)	Alexa Fluor 647	47	612599	BD Biosciences
p-ERK1/2 (Thr202/Tyr204)	Alexa Fluor 488	E10	4374	Cell Signaling Technology
Western Blotting
p-AKT (Ser473)	-	N/A	9271	Cell Signaling Technology
p-STAT5 (Tyr694)	-	47	611964	BD Biosciences
p-ERK1/2 (Thr202/Tyr204)	-	E10	9106	Cell Signaling Technology
p-STAT3 (Tyr705)	-	B7	sc-8059	Santa Cruz Biotechnology
Total AKT	-	N/A	9272	Cell Signaling Technology
Total STAT5	-	N/A	AF2168	R&D SYSTEMS
Total ERK2	-	D-2	sc-1647	Santa Cruz Biotechnology
Total STAT3	-	F-2	sc-8019	Santa Cruz Biotechnology
β-Actin		Polyclonal	4967	Cell Signaling Technology

Surface and intracellular FCM staining

Cryopreserved peripheral blood MNCs from AML patients and normal bone marrow cells were thawed at 37°C and washed in minimum essential medium (MEM) alpha (Cat. 15-012-CV; Mediatech, Inc.) containing 20% FBS. The cells were incubated in the same medium supplemented with heparin (Cat. 9041-08-1; Alfa Aesar, Ward Hill, MA), DNase I (Cat. 89835; Thermo Scientific, Waltham, MA) and MgSO₄at 37°C for 15 minutes, washed once and filtered to remove debris. The cells were counted by Vi-cell (Beckman Coulter, Brea, CA). The median cell viability after thawing was 89.3% (range 80.3%–97.4%). For surface marker staining, cells were blocked with human Fc receptor binding inhibitor (Cat.14-9161-73; eBioscience, San Diego, CA) for 15 minutes at 4°C and stained with a cocktail of Ab comprising CD34-APC, CD38-PE-Cy7, CD45-APC-Cy7, CD123-PerCP-Cy5.5, and CD99-FITC for 30 minutes at room temperature. The cells were then washed, resuspended in PBS with DAPI, and analyzed on a Gallios Flow Cytometer (Beckman Coulter, Indianapolis, IN).

For intracellular staining, the cells were treated with a kinase inhibitor for 1 hour, including MEK1/2 inhibitor CI-1040 (Cat. 212631-79-3; Selleckchem, Houston, TX) at 10 µM, JAK inhibitor ruxolitinib (Cat. 941678-49-5; Selleckchem) at 100 nM, and PI3K inhibitor BEZ235 (Cat. 915019-65-7; Selleckchem) at 25 nM. Cells were then stimulated with GM-CSF (100 U/mL), stem cell factor (SCF; 100 ng/mL; Cat. PHC6034; Life Technologies, Grand Island, NY) or phorbol 12-myristate 13-acetate (PMA; 100 nM; Cat. P1585; Sigma-Aldrich) for 5 minutes and fixed with paraformaldehyde (PFA; Cat. 15710; Electron Microscopy Sciences, Hatfield, PA) at a final concentration of 1.6% at room temperature for 10 minutes. Cells were spun down, washed twice with wash buffer comprising PBS supplemented with 0.5% bovine serum albumin (Cat. A7906; Sigma-Aldrich), and permeabilized by suspending in 1 mL of 80% cold methanol (Cat. 67-56-1; Fisher Scientific, Waltham, MA) overnight at −20°C. Cells were washed twice with wash buffer and stained with p-STAT5-Alexa Fluor 647 (1:10), p-ERK1/2-Alexa Fluor 488 (3:100), or pAKT-Alexa Fluor 647 (3:100) in 100-µL final reaction volumes for 30 minutes at room temperature in darkness. After being washed twice with wash buffer, cells were analyzed on a Gallios Flow Cytometer.

Mo7e cells were starved overnight at 1 million cells/mL in a 37°C incubator in RPMI 1640 medium with 1% FBS and penicillin/streptomycin and then subjected to inhibition, stimulation, fixation, permeabilization, and staining as described already for patient- and donor-derived cells. TF-1 cells were starved in the same way and stimulated with GM-CSF at 100 U/mL or IL-6 (100 ng/mL; Cat. PHC0066; Life Technologies) for 5 minutes in a 37°C incubator before fixation, permeabilization, and staining with indicated Ab. Data were analyzed by Flowjo software (Tree Star, Ashland, OR).

Antibody conjugation for CyTOF analysis

All preconjugated Ab were purchased from DVS Sciences (Sunnyvale, CA). When an Ab was not available, conjugation was performed in house as follows. Purified, carrier-free Ab was conjugated with lanthanide isotopes using MaxPar Antibody Labeling Kit (DVS Sciences) by following the manufacturer’s instructions. Protein concentrations were determined by using NanoDrop 2000 (Thermo Fisher Scientific) whereby an absorbance of 1 at 280 nM equalled 1 mg/mL. Metal contents of the conjugated Ab were determined by CyTOF in solution mode by using Claritas PPT Grade Multi-Element Solution 1 (SPEX CertiPrep, Metuchen, NJ) at 0.5ppb as a standard.

Mass cytometry staining

Following cell preparation as described, cells were treated with cisplatin (Cat. 479306; Sigma-Aldrich) at 25 µM for 1 minute at room temperature, as previously described (29). Cells were subjected to centrifugation in 3 mL RPMI with 10% FBS, stimulated, and fixed as already described and then subjected to blocking of human Fc receptor binding inhibitor for 15 minutes at 4°C. Metal-labeled Ab against surface markers were stained in 50-µL final reaction volumes at room temperature for 30 minutes. Following staining, cells were washed twice with wash buffer and once with PBS and resuspended in 500 µL of 1.6% PFA for 10 minutes at room temperature. Cells were spun down, washed twice with wash buffer, and permeabilized with 80% cold methanol at −20°C overnight. After washing twice to remove methanol, cells were stained with Ab against intracellular markers in 50-µL final reaction volumes at room temperature for 30 minutes. Cells were then washed twice with wash buffer and once with PBS and stained in 500 µL of 1:1000 Iridium intercalator (Cat. 201192A; DVS Sciences, Toronto, ON, Canada) diluted in PBS with 1.6% PFA for 20 minutes at room temperature. Cells were then washed twice with wash buffer and filtered through blue-capped tubes. Each sample pellet was resuspended in 50 µL of deionized water and transferred to a 96-deep well plate containing 50 µL of Eu151/153 calibration beads (Cat. 201073; DVS Sciences) in each well. Samples were analyzed on a CyTOF mass cytometer using an AS5 Autosampler (both, DVS Sciences); 0.4 mL of deionized water was added just prior to injection according to published procedures (25). The bead signature was routinely appplied to normalize the raw CyTOF data before analysis using the method previously reported (30). The data were saved in FCS3.0 format and analyzed by Flowjo software.

The spanning-tree progression analysis of density-normalized events analysis

The original FCS files were input into Flowjo software and exported after gating on viable cells based on DNA labeling with the iridium (Ir191/193), cell length, and cisplatin (24). The exported FCS files were transferred into the spanning-tree progression analysis of density-normalized events (SPADE) software for analysis (20, 25). In particular, SPADE was used to analyze FCS files from six patients together. Five surface markers (CD34, CD123, CD45, CD99, and CD38) were chosen to down sample cells, perform clustering, and construct a tree structure. Other surface markers (CD90 and CD25) were tested but not selected because of low signals. The SPADE tree representing all cell types was defined by the seven selected surface markers that existed in at least one of the patients. For each annotated phenotype, median intensity of the marker expression was computed for each phospho-protein marker for each patient and visualized in heat maps to illustrate the difference in phospho-protein expression of each annotated phenotype across the patients. The scale is the mean marker intensity of arcsinh-transformed values.

Western blot analysis

After fixation as described, cells were subjected to lysis in lysis buffer (150 mM NaCl, 1 mM MgCl₂, 1 mM CaCl₂, 10 mM NaF, 5 mM sodium pyrophosphate, 10 mM β-glycerophosphate, 1% Triton X-100, 10 mM iodoacetamide, 1 mM Na₃VO₄, 0.1% NaN₃, 3 mM phenylmethylsulfonyl fluoride) supplemented with a protease inhibitor cocktail (Roche Diagnostics, Indianapolis, IN). Lysates were separated on a 10% polyacrylamide gel, transferred to PVDF membranes (Bio-Rad, Hercules, CA), probed with appropriate Ab, and analyzed on the Odyssey imaging system from LI-COR Biosciences (Lincoln, NE).

Statistical analyses

The Student t-test was used to analyze the statistical significance of differences in the data. All statistical tests were two-sided, and the results are expressed as the mean ± standard deviation. A p value ≤ 0.05 was considered statistically significant. The statistical analyses were performed by using GraphPad Prism v6.0 software (GraphPad, La Jolla, CA).

RESULTS

Assessing LSC phenotypes by CyTOF and FCM

We compared CyTOF with traditional multiparametric FCM for the assessment of LSC phenotype in samples of primary AML and normal bone marrow; LSC were isolated using five surface antigens (i.e., CD45, CD34, CD38, CD123, and CD99). No difference was detected between intact cells and fixed cells in terms of all surface markers (data not shown). The clinical characteristics of the patient samples are shown in Table 2. In the traditional FCM analysis, FSC and SSC were used to detect cells. In the CyTOF analysis, nucleated cells were gated by metal-tagged DNA intercalator-Ir191/193 (Fig. 1A). The CD45^dim blast population was gated to yield the CD34⁺CD38⁻ fraction, from which the CD123⁺CD99⁺-enriched LSCs were obtained. Fluorescence minus one (FMO) controls were applied to identify gating boundaries in FCM analysis (Suppl. Fig. S1). In CyTOF, correction of spectral overlap is not required and cellular autofluorescence is absent because the elemental isotopes do not usually exist in biological systems (25). Signals measured by CyTOF over 10¹ were considered as positive.

Table 2.

Demographic and disease characteristics of AML patients who provided samples

ID#	Source	Status	Age (year)/Sex	Blasts	Cytogenetics	FLT3 Status
1	PB	Relapsed	29/F	96%	Complex	ITD
2	PB	Newly diagnosed	37/F	36%	46,XX,t(1;14)(p10q10)	ITD
3	PB	Newly diagnosed	69/M	20%	46,XY, del(16)(q22)	ITD and D835
4	PB	Newly diagnosed	43/M	48%	Complex	ITD
5	PB	Newly diagnosed	83/M	81%	48,XY,+13	ITD
6	PB	Newly diagnosed	59/M	21%	Complex	ITD
7	PB	Newly diagnosed	32/M	40%	46,XY	ITD and D835
8	PB	Relapsed/Refractory	66/M	39%	46,XY	ITD
9	PB	Refractory	67/M	93%	Complex	WT

PB, peripheral blood; WT, wild type

Figure 1.

LSC phenotypes were defined by CyTOF and FCM in primary AML samples. (A) Surface Ag staining and definitions of cell populations for CyTOF (upper panel) and FCM (Flow; lower panel) for one representative AML sample are shown. Replicate analysis of two primary AML samples and one normal bone marrow sample is shown in Supplemental Figures 1 and 2. (B) Pearson correlation of frequencies of each defined population across samples as determined by FCM and CyTOF is shown.

Representative figures from an AML sample showed that the contour plots were comparable between CyTOF and FCM (Fig. 1A). The CD45^dim blast population accounted for 98.4% and 98.1% of total cells by CyTOF and FCM, respectively. CD45^high lymphocytes were detected by both techniques. Frequencies of CD34⁺CD38⁻ cells within blasts were 5.1% and 2.97% by CyTOF and FCM, respectively, and frequencies of the CD123⁺CD99⁺ population within CD34⁺CD38⁻ cells were 74.8% and 94.8% by CyTOF and FCM, respectively, indicating that CD123 CyTOF Ab is dimmer than flow Ab. Replicate analysis of a second primary AML sample and a normal bone marrow sample is shown in Supplementary Figure S2. Correlation analysis of the three populations (CD45^dim, CD34⁺CD38⁻, and CD34⁺CD38⁻CD123⁺CD99⁺ cells) manually gated in two AML samples and one normal bone marrow sample revealed similar frequencies of AML progenitor subsets between the two platforms (r=0.99, p<0.0001) (Fig. 1B and Table S1).

Intracellular signaling pathways analyzed by CyTOF, FCM, and western blotting in cell lines and primary samples

Next we measured intracellular proteins in myeloid leukemia cell lines and a primary AML sample by means of CyTOF, FCM, and western blotting. First we stimulated cytokine-dependent TF-1 leukemia cells that had been deprived of serum overnight with GM-CSF and SCF. A robust p-STAT5 response was observed following GM-CSF treatment in all platforms (Fig. 2A), whereas SCF consistently failed to stimulate p-STAT5 activation (Fig. 2A). Interleukin 6–induced p-STAT3 activation was completely suppressed in TF-1 cells by pretreatment with JAK inhibitor ruxolitinib for 1 hour (Fig. 2B).

Figure 2.

CyTOF, FCM, and western blotting yield comparable information on intracellular signaling pathway markers in leukemia cell lines. Serum-deprived TF-1 cells were stimulated with SCF or GM-CSF (GM) (A) or treated with ruxolitinib (Ruxo; B) for 1 hour before stimulation with IL-6. (C) Serum-starved Mo7e cells were treated with ruxolitinib, CI-1040, or BEZ235 for 1 hour. P-STAT5, pERK, and pAKT were detected by CyTOF, FCM (Flow), and western blotting. Pearson correlation of the magnitude of response of each intracellular protein to stimulation and inhibition across samples as determined by FCM, CyTOF, and western blotting is shown in Supplemental Figure 3. (D) CyTOF, FCM, and western blotting yielded comparable information on intracellular signaling pathways markers in primary AML samples.

Next, we applied three phosphoprotein agonists (i.e., GM-CSF, PMA, and SCF) to the cytokine-dependent cell line Mo7e. As shown in Figure 2C, GM-CSF induced a strong increase in p-STAT5 in all three platforms, which in turn was partially reduced by specific JAK inhibitor ruxolitinib. PMA-evoked p-ERK was eliminated by CI-1040, and BEZ235 effectively blocked the SCF-activated p-AKT signaling pathway. Fold-changes of median signal intensity of stimulated/treated groups compared to the untreated groups showed significant correlation between FCM and CyTOF results (r = 0.92, p = 0.0093) (Table S2 and Fig. S3).

We further assessed phosphoprotein responses in the primary AML MNCs (#9) under conditions similar to those shown in Figure 2C. MNCs were rested in HPGM medium for 2 hours, then were treated with a kinase inhibitor for 1 hour and cytokine stimulation for 5 minutes. No constitutive p-STAT5, p-ERK, or p-AKT was detected in the AML sample (Fig. 2D). CyTOF, FCM, and western blotting demonstrated consistency in measurements of cytokine-induced signaling pathway activation and inhibition. STAT5 phosphorylation was strongly induced by GM-CSF and partially reduced after treatment with ruxolitinib. Activation of p-ERK and p-AKT was detected in response to PMA and SCF, respectively, and CI-1040 and BEZ235 blocked p-ERK and p-AKT completely. Taken together, these data show that CyTOF can provide data on intracellular phosphoprotein expression equivalent to that generated by traditional FCM and western blotting, but without the potential limitations of FCM.

Next we studied modulation of the signaling pathway in antigen-defined cell population in response to SCF and PI3K inhibitor BEZ235 in patient #9 using an Ab panel. MNCs were stained with seven stem/progenitor cell surface markers (i.e., CD34, CD38, CD123, CD99, CD45, CD90 and CD25) and nine intracellular signaling markers (i.e., p-4EBP1, p-NFκB, p-STAT3, p-AKT, p-mTOR, p-ERK, p-S6, p-PI3K and p-STAT5). The SPADE tree plot was built by using CD45, CD34, CD38, CD123, and CD99 on pre-gated viable blasts (sample contained 93% blasts, Table 2). The tree structure colored by individual markers is shown in Figure 3A, and phenotypes for all annotated populations are summarized in Fig. 3B and Supplementary Table S3. Constitutive activation of p-4EBP1 and p-mTOR, but not of p-S6 and p-AKT, was observed, particularly within annotations A1-3, which expressed the CD34⁺CD38⁺ progenitor phenotype (Fig. 3C). SCF induced activation of signaling in p-4EBP1, p-S6, p-AKT, and p-ERK, less in p-mTOR, in all annotated populations but predominantly in CD34⁺ populations (annotations A1-8). BEZ235 completely blocked activation of the p-4EBP1 and p-AKT signaling pathways, and partially blocked the p-S6 signaling pathway, indicating involvement of a different kinase upstream of p-S6. P-ERK and p-mTOR pathways did not show responses to BEZ235. These data provide a rationale for tracking responses of primary AML cells to ex vivo signaling inputs and therapeutics in distinct cell populations.

Figure 3.

CyTOF detected activation of signal transduction pathways in response to ex vivo perturbations. Cells from AML patient #9 were treated with BEZ235 and/or stimulated with SCF. (A) Tree plots were generated from patient 1using five surface markers (CD34, CD123, CD45, CD99, and CD38) with the annotations shown in Table S3. (B) The heatmap shows expression of surface markers in all annotations A1 through A9. (C) For each annotated phenotype, median intensity was computed for each phosphoprotein marker for each annotation, and the results are visualized in heatmaps to illustrate differences in phosphoprotein expression of each annotated phenotype under different conditions.

Activation of signal transduction pathways in dissectedstem/progenitor cell subsets

Taking advantage of the multiparametric potential of CyTOF, we investigated intracellular signaling pathways in antigen-defined subpopulations in seven samples of primary AML with FLT3-ITD mutations (Table 2). MNCs were stained with the Ab panel already described. To assess the heterogeneity across AML samples and variability between AML and normal samples, we included three normal bone marrow samples as reference. The data were analyzed by the SPADE software (31). SPADE clustered the single-cell data by using the k-means algorithm, constructed a minimum spanning tree to connect clusters that exhibited similar marker combination, and displayed the tree structure colored by median intensity of individual protein markers. All surface markers except low CD90 and CD25 were applied to build the SPADE tree.

Eleven cell populations were annotated manually by coloring the tree via each cell surface marker (Fig. 4A, Suppl. Fig. S4A, and Suppl. Table S4). Representative figures depicting CD45 and CD34 expression are shown. Annotations1 and 2 (A1 and A2) were defined as CD45^dimCD34⁺CD38^−/dim, which phenotypically corresponded to the putative stem cell population, with high CD123 and CD99 co-expression in A1 (6, 7,12, 32). A3 and A4 were the progenitor populations defined as CD34⁺CD38⁺, which lacked CD123 and CD99 expression. A5, A6, and A11 were negative for CD34, which may represent the more mature leukemia blasts. A7 through A10 showed high CD45 expression, putatively suggesting the lymphocyte populations; however, this needs to be confirmed with additional lineage markers. Biaxial plots of CD34 versus CD38 are shown to confirm the immunophenotypes of the annotations (Suppl. Fig. S4B).

Figure 4.

CyTOF detects variability in phenotypes and activation of signal transduction pathways in stem/progenitor cell subsets in primary AML samples harboring FLT3-ITD mutations compared to normal bone marrow samples. (A) The SPADE tree plots were generated using five surface markers (CD34, CD123, CD45, CD99, and CD38) from seven patients and three normal bone marrow (NBM) samples together and was colored by the median intensity of CD45 and CD34. Phosphoprotein markers were not used to construct the tree. Different cell populations were manually annotated as A1 to A11 (Table S4). (B) For each annotated phenotype, median intensity of the marker expression was computed for each phosphoprotein marker and for each sample and results were visualized as heatmaps to illustrate differences in phosphoprotein expression of each annotated phenotype across different patients. (C) Differences between samples were compared using the earth mover distance metric (EMD). For each individual sample, the percentages of cells in that sample belonging to each node of the SPADE tree were computed. The pairwise EMD distances were organized by hierarchical clustering and displayed.

To gain insights into activation of the signaling downstream of FLT3-ITD, we generated heatmaps to illustrate differences in phosphoprotein expression of each annotated phenotype across AML and normal samples (Fig. 4B and Suppl. Fig. S4C). Activation of mTOR target p-4EBP1 was observed in samples from all AML samples but none of the normal samples, with high activities restricted to A1–A4 in the majority of AML samples. It exhibited a wide range of activity in all annotated populations in patient 8. Another mTOR target, p-S6, displayed distinct patterns by showing activities in A1-4 in AML 4–7 and A9-11 populations in AML 3–6. AML 8 lacked p-S6 activity. Both signaling pathways were absent in normal bone marrow samples. Expression of additional phosphoproteins within annotated cell populations is shown in Supplemental Figure S4C.

To assess the overall variability across AML samples and between AML and normal samples, we compared the samples by using the earth mover’s distance metric (EMD) (33). For each individual sample, we computed the percentage of cells in that sample belonging to each node of the SPADE tree. These percentages formed a distribution of cells across the tree, similar to the percentages computed in manual gating analysis. EMD is a distance measure between the cell distributions of two samples with respect to the tree, reflecting the similarity among the samples. The pairwise EMD distances were organized by hierarchical clustering and are displayed in Figure 4C. The three normal bone marrow samples were highly similar to each other. Two AML samples (6 and 7) were similar to the normal samples, whereas the remaining AML samples were not similar to the normal samples or each other. This result demonstrated that the variability among the AML samples was much higher than the variability among the normal samples.

It has to be noted that current antibody panel included only surface markers against progenitor populations that have been defined as lineage negative. We questioned whether the progenitor markers alone or combined with lineage markers could result in similar progenitor populations. To address this, we analyzed AML sample #2 using a CyTOF panel that includes both progenitor markers as above and lineage markers (CD7, CD64, CD56, CD2, CD15, CD41, CD3). The SPADE trees were built using either progenitor markers only or combined with lineage markers (Fig S5). We found that both analyses were able to capture CD34⁺ stem/progenitor phenotypes, which did not show signals for any of the tested lineage markers (Fig S5). However, combination with lineage markers gave better resolution of the mature cell phenotypes (Fig S5B). Therefore, inclusion of lineage markers in CyTOF panels would provide additional information for the high-dimensional data analysis of primary AML samples.

Discussion

In this study, we employed the novel CyTOF mass cytometry technique for the characterization of LSC surface markers and intracellular phosphoproteins in primary AML samples. Detection of myeloid and lymphoid lineages in normal bone marrow and peripheral blood MNCs by CyTOF and FCM has already been shown to be comparable (25, 29). In this first cross-platform analysis of primary AML samples, optimized CyTOF and FCM demonstrated similar frequencies of the LSC fractions defined by five surface markers. The frequencies of three manually gated populations, including CD45^dim blasts, CD34⁺CD38⁻cells, and CD34⁺CD38⁻CD123⁺CD99⁺ cells from AML samples and normal bone marrow samples significantly correlated across the CyTOF and FCM platforms. The CyTOF technique measuring intracellular proteins was validated in myeloid leukemia cell lines and one primary AML sample. Although CyTOF and FCM showed slightly different fold changes of intracellular protein signals, correlation analysis suggested that the two platforms provided consistent information. These data strongly suggest that CyTOF can be reliably applied for characterizing phenotypes and signaling states in AML samples.

AML comprises a group of highly heterogeneous hematological malignancies. Several surface markers have been reported to be expressed preferentially on AML LSCs (6–11), which initiate and maintain the leukemias. These LSCs possess unique properties, including specific signaling pathways (13, 14) and gene signatures (16). The further characterization of LSCs may assist in defining novel, potentially targetable cell survival/proliferation pathways. CyTOF with its greater dimensionality than FCM, may be an ideal tool for studies of AML LSC biology. Thus, we applied CyTOF analysis to seven AML samples harboring FLT3-ITD mutations to evaluate its capacity for simultaneous assessment of surface markers and phosphoproteins to distinguish AML stem and progenitor cell phenotypes and phenotype-specific cell signaling cascades. SPADE tree plots showed annotations representing putative LSCs (CD45^dimCD34⁺CD38⁻CD123⁺99⁺) and leukemia progenitor cells (LPCs; CD34⁺CD38⁺), demonstrating phenotypic heterogeneities across patients harboring the same genetic make-up. Distinct patterns of signaling activation were observed in different annotated cell populations between samples, demonstrating the high capacity of CyTOF in combination with multi-dimensional data analysis to dissect cell type–specific functional events. Variability in terms of cell distribution and signaling activation was also observed between AML patient samples and normal bone marrow samples.

LSCs in AML were first identified to reside in the CD34⁺CD38⁻ fraction by their capacity to repopulate in NOD/SCD mice (34). This notion was challenged by recent studies showing that CD34⁺CD38⁺(16, 35, 36) and CD34⁻ (17)cell fractions also contained AML LSCs. Gibbs et al. showed that, in the HoxA9-Meis1 model of AML, LSC activity existed in three distinct cell fractions, corresponding to normal progenitors and lymphoid lineages (37). Gene expression profiling studies generated LSC-specific signatures that were not restricted to the CD34⁺CD38⁻ population (16, 38). Moreover, Goardon et al. found, by xenotransplantation into NOD/SCID/IL2Rγ mice, that two dominant populations phenotypically representing normal progenitors possessed LSC capacity (36). In this study, we found that multiple intracellular signaling pathways were also highly activated in the LPCs and may thus represent potentially targetable cell survival/proliferation pathways in AML progenitor cells. Future xenograft assays are required to determine whether these populations contain LSCs.

The mTOR targets p-4EBP1 and p-S6 were expressed in immature leukemia cells but not in normal stem/progenitor cells in the majority of tested AML samples, suggesting both as novel targets in FLT3 mutated AML. Experiments testing this hypothesis are in progress.

Incorporation of specific kinase inhibitors may also aid in understanding the properties of multiple signaling pathways and their role in survival of LSCs. Incorporating ex vivo stimuli and treatment with a kinase inhibitor or apoptosis inducer offered additional insights into AML biology; for example, stroma/cytokine dependencies could be deduced, which may help in selecting agents specifically targeting LSC-enriched populations. Thus CyTOF analysis is providing intriguing data regarding AML biology and warrants future correlations with functional studies. The technology is highly applicable to studying distinct molecular subtypes of AML, which likely differ in LSC composition and oncogene-specific signaling, such as demontrated here for FLT3-ITD AML.

In conclusion, by taking advantage of the CyTOF technique, with its negligible background signal and increased numbers of parameters, we were able to identify multiple functional signaling pathways in antigen-defined subpopulations of AML, especially in LSC-enriched populations. The mTOR targets p-4EBP1 and p-S6 were exclusively found in FLT3-ITD stem/progenitor cells, but not in their normal counterparts. These data demonstrate the potential of this platform to assist with selection and monitoring of rationally designed therapies targeting LSC/progenitor cells in AML.