Identification of a Small Subpopulation of Candidate Leukemia-Initiating Cells in the Side Population of Patients with Acute Myeloid Leukemia



In acute myeloid leukemia (AML), apart from the CD34+CD38 compartment, the side population (SP) compartment contains leukemic stem cells (LSCs). We have previously shown that CD34+CD38 LSCs can be identified using stem cell-associated cell surface markers, including C-type lectin-like molecule-1 (CLL-1), and lineage markers, such as CD7, CD19, and CD56. A similar study was performed for AML SP to further characterize the SP cells with the aim of narrowing down the putatively very low stem cell fraction. Fluorescence-activated cell sorting (FACS) analysis of 48 bone marrow and peripheral blood samples at diagnosis showed SP cells in 41 of 48 cases that were partly or completely positive for the markers, including CD123. SP cells in normal bone marrow (NBM) were completely negative for markers, except CD123. Further analysis revealed that the SP fraction contains different subpopulations: (a) three small lymphoid subpopulations (with T-, B-, or natural killer-cell markers); (b) a differentiated myeloid population with high forward scatter (FSChigh) and high sideward scatter (SSChigh), high CD38 expression, and usually with aberrant marker expression; (c) a more primitive FSClow/SSClow, CD38low, marker-negative myeloid fraction; and (d) a more primitive FSClow/SSClow, CD38low, marker-positive myeloid fraction. NBM contained the first three populations, although the aberrant markers were absent in the second population. Suspension culture assay showed that FSClow/SSClow SP cells were highly enriched for primitive cells. Fluorescence in situ hybridization (FISH) analyses showed that cytogenetically abnormal colonies originated from sorted marker positive cells, whereas the cytogenetically normal colonies originated from sorted marker-negative cells. In conclusion, AML SP cells could be discriminated from normal SP cells at diagnosis on the basis of expression of CLL-1 and lineage markers. This reveals the presence of a low-frequency (median, 0.0016%) SP subfraction as a likely candidate to be enriched for leukemia stem cells.

Disclosure of potential conflicts of interest is found at the end of this article.


Author contributions: B.M.: conception and design, collection and/or assembly of data, data analysis and interpretation, manuscript writing; A.V.R., A.K., M.V.D.P., M.T., C.B., and A.H.W.: collection and/or assembly of data; G.J.O.: provision of study material or patients, final approval of manuscript; S.Z.: provision of study material or patients, manuscript writing, final approval of manuscript; G.J.S.: conception and design, data analysis and interpretation, manuscript writing, final approval of manuscript.

Acute myeloid leukemia (AML) is generally regarded as a disease likely to originate from the hematopoietic stem cell (HSC) [1]. There is an increasing body of evidence pointing toward the importance of the leukemic stem cells (LSCs) for the occurrence of minimal residual disease (MRD) and relapse. This might well be explained by properties LSCs share with normal stem cells, such as relative quiescence and resistance to apoptosis. Accordingly, we have previously described that a high stem cell frequency in AML predicts MRD cell frequencies after chemotherapeutic treatment, resulting in poor prognosis [2]. To both determine the number of remaining LSCs after therapy for prognostic purposes, as well as to apply targeted therapy, immunophenotypic molecular and functional characterization of LSCs is of utmost importance. However, there is a need for new markers, preferably on the cell surface, to discriminate between normal CD34+CD38 cells and malignant CD34+CD38 cells, as there is considerable overlap in expression of currently available markers. Recently, we found that C-type lectin-like molecule-1 (CLL-1) and leukemia-associated lineage markers provide the opportunity to discriminate between HSCs and LSCs, as they were found to be solely expressed on leukemic CD34+CD38 cells [3, 4]. In addition, new definitions of LSCs should be generated, as not all AML cases have one or more of the aberrant markers present, and moreover, not all LSCs may have the CD34+CD38 immunophenotype (e.g., in true CD34-negative AML). Moreover, the real number of stem cells is likely to be much lower than the number present in the so-called stem cell compartment [1, 5], suggesting that the definition of the stem cell compartment should be refined. An alternative stem cell compartment may be the so-called side population (SP). SP cells are defined by their ability to efficiently efflux Hoechst 33342 dye. In normal bone marrow (NBM), the SP was indeed found to be enriched for stem cells [6, 7]. Accordingly, in AML, the SP compartment is able to initiate leukemia in NOD/SCID mice, whereas the non-side population (NSP) compartment is not [8]. It is tempting to speculate that the frequency of real LSCs within the SP compartment is higher than within the CD34+CD38 compartment, as the SP frequency is much lower than that of the CD34+CD38 cells [8]. Since the SP compartment has previously been shown by fluorescence in situ hybridization (FISH) analysis to contain both malignant and normal cells [9], we sought characteristics/markers with the ability to discriminate between these malignant and normal SP cells. These would allow the primitive AML SP cells to be traced at diagnosis and during/after treatment. Moreover, both normal and AML SP cells could then be studied separately for therapeutic target finding. This would provide functional and molecular biological differences between AML and normal stem cells under the most clinically relevant conditions: both types of stem cells present in the same bone marrow. Lastly, it would enable the definition of AML SP stem cells to be fine-tuned.

Materials and Methods

Leukemic and Normal Bone Marrow Cells

Bone marrow (BM) samples were collected at diagnosis after informed consent from 48 AML patients (20 females, 28 males) with a median age of 49 years (range, 19–75). In four cases, BM was not available at diagnosis, and peripheral blood (PB) was used. NBM was obtained after informed consent from patients undergoing cardiac surgery. The majority of the samples were analyzed immediately. Mononuclear cells were isolated by Ficoll gradient (1.077 g/ml; Amersham Biosciences, Freiburg, Germany, Red blood cells were lysed afterward by 10 minutes of incubation on ice, using 10 ml of a solution containing 155 mM NH4Cl, 10 mM KHCO3, and 0.1 mM Na2 EDTA, pH 7.4, added directly to the cell pellet. After washing, cells were frozen in RPMI (Gibco, Paisley, U.K., with 20% heat-inactivated fetal bovine serum (FBS; Greiner Bio-One, Alphen a/d Rijn, The Netherlands, and 10% dimethyl sulfoxide (Riedel-de Haen, Seelze, Germany, in isopropanol-filled containers and subsequently stored in liquid nitrogen. When needed for analysis, cells were thawed and suspended in prewarmed RPMI with 40% FBS at 37°C. Cells were washed and allowed to recover for 45 minutes in the same medium at 37°C. Cells were washed again and resuspended in phosphate-buffered saline with 0.1% bovine serum albumin (ICN Biomedicals, Aurora, OH,

Flow Cytometry and Cell Sorting

Primary AML cells (1 × 106 cells per milliliter) were stained with 5 μg/ml Hoechst 33342 dye (Molecular Probes, Eugene, OR, with or without breast cancer resistance protein (BCRP) inhibitor KO143 (200 nM; Sigma-Aldrich, Steinheim, Germany, and incubated at 37°C for 2 hours according to Goodell et al. [6]. After Hoechst staining, cells were washed and resuspended into 100 μl of cold (4°C) Hanks' balanced salt solution (HBSS; Cambrex, Verviers, Belgium, + 2% fetal calf serum (FCS) and incubated for 30 minutes on ice with combinations of fluorescein isothiocyanate- (FITC), phycoerythrin- (PE), allophycocyanin (APC)-labeled monoclonal antibodies (MoAbs). Anti-CD45 APC, anti-CD45 PE, anti-CD38 APC, anti-CD34 FITC, anti-CD7 PE, anti-CD19 PE, anti-CD56 PE, anti-CD48 PE [10], and anti-CD123 PE MoAbs were all from BD Biosciences (San Jose, CA, To define the leukemic stem cells, we used CD7, CD19, and CD56, which are frequently used in AML MRD detection using leukemia-associated phenotype; anti-CLL-1 and isotype control were used as previously described [4, 11]. After antibody staining, cells were washed with cold HBSS+ (HBSS + 2% FCS), resuspended in 1 ml of cold HBSS+ and stained for 5 minutes with 2 μg/ml propidium iodide (Sigma-Aldrich), enabling exclusion of dead cells. Cells were kept on ice until fluorescence-activated cell sorting (FACS) analysis. Data acquisition was performed using either a FACSVantage (equipped with red, blue and ultra violet lasers) or a FACSCanto II (with red, blue and violet solid-state lasers), both from BD Biosciences; analysis was performed using CellQuest and FACSDiva software (BD Biosciences). The Hoechst dye was excited with a 350-nm UV (FACSVantage) or 405-nm violet (FACSCanto II) laser and detected with 450/BP20 and 450/BP50 optical filters, respectively. Gates were set to detect the viable SP cells as shown in Figure 1A. Cells were sorted using a FACSAria (with red, blue, and violet solid-state lasers; BD Biosciences). Cells were kept on ice during the whole procedure. For further culturing, cells were sorted directly into cold culture medium. Purity of sorted populations was >98%.

Suspension Culture of AML SP Cells

The suspension culture was performed essentially as has been previously described [12]. The sorted SP (sub)populations (3,000–5,000 cells) were mixed with 1 × 105 NSP cells and resuspended in 250 μl per well of CellGro medium (Cellgenix, Vancouver, BC, Canada, containing 20 ng/ml interleukin (IL)-3, 100 ng/ml Flt-3 ligand, and 100 ng/ml stem cell factor (SCF) (all from Peprotech, Basel, Switzerland, prior to plating in 96-well round-bottomed plates (Greiner Bio-One). In addition, 1 × 105 and 1 × 106 NSP cells were plated as control for the mixed SP + NSP. Suspension cultures were incubated at 37°C in 5% CO2 and received weekly half-medium changes. Usually in this assay these weekly half-medium changes are accompanied by demipopulation of cells; however, since the numbers of SP cells were very low, we chose to harvest all cells at one time point only (i.e., 5 weeks). Subsequently, all harvested cells were cultured in a 14-day colony-forming unit (CFU) assay.

CFU Assay

Assays for leukemic CFUs were performed by plating cells in methylcellulose medium (H4434; StemCell Technologies, Vancouver, BC, Canada, Cultures were scored after 14 days for the presence of clusters (4–20 cells) and colonies (more than 20 cells). The number of colonies from the sorted SP + NSP or NSP cells was calculated as previously described [12]. Final clonogenic output was expressed as number of colonies per million input cells (i.e., cells immediately following sorting at the start of the [5 + 2 weeks] experiment).

FISH Analysis of FACS-Sorted SP Cells

For interphase FISH, the FACS-sorted SPs were washed three times with 3 ml of 3:1 methanol/acetic acid fixative and suspended in 100 μl of fixative. Subsequently, one droplet was gently placed on an object slide and air-dried. Dual-color (spectrum green and spectrum orange fluorophores) labeled LSI DNA probes (Vysis, Downers Grove, IL, were applied to the denaturated cells and incubated as previously described [13]. The following probes were used: the LSI AML1/ETO dual color for t(8;21) and the LSI TEL/AML1 ES dual color for del 12(p13). Hybridization and deletion signals were scored in 50 interphase nuclei with an Axioscop 20 (Carl Zeiss, Jena, Germany, fluorescence microscope with three single-band-pass filters and one triple-band-pass filter. Nuclei were scored positive for the fusion gene, when a green spot and an orange spot were less than one spot diameter apart. Nuclei were scored positive for deletion 12, when one green spot was absent. The images were captured with a digital camera using CytoVision 4.1 software (Applied Imaging Corp., Newcastle, U.K.,

Statistical Analysis

Statistical analysis was performed using SPSS 9.0 software package (SPSS, Chicago, The Wilcoxon signed-rank test was used to determine differences between paired samples. Statistical significance was evaluated at p < .05. Average values were expressed as mean ± SEM.


To distinguish AML SP cells from normal SP cells at diagnosis, we attempted to find leukemia stem cell-associated immunophenotypic markers (i.e., those not staining the normal SP stem cells). We used CLL-1 and IL-3 receptor α-chain CD123, previously reported to be leukemic stem cell markers [4, 14], and the lineage markers CD7, CD19, and CD56, used to define blast cell aberrancies suitable for immunophenotypic MRD detection [15] and staining of AML CD34+CD38 LSCs [3].

Relationship Between CD34 and CD38 Expression and the SP Phenotype

To define the immunophenotype of AML SP cells in relation to NBM SP cells, 44 bone marrow samples and 4 PB samples of AML diagnosis patients were investigated. SP cells were detectable in 41 of 48 AML patients (85%), with a median frequency of 0.07% (expressed as percentage of whole blast cells; range, 0.002%–7.6%). In all individual cases with both CD34+CD38 and SP stem cell compartments present (n = 36), the CD34+CD38 compartment had a higher frequency than the SP in this subset of samples: 0.47% (range, 0.01%–26.6%) versus 0.03% (range, 0.002%–7.6%; p = .002). The median frequency of CD34+CD38 cells within the SP compartment was 2.5% (range, 0–49%). SP cells were detected in all 12 NBM samples, with a median frequency of 0.12% (range, 0.008%–4.1%).

The SP Compartment in AML Contains Normal Lymphocytic Cells

During the immunophenotyping of AML SP cells we detected a small SP with a low forward scatter (FSC) and a slightly lower sideward scatter (SSC); one representative example is shown in Figure 1A3. Some of these cells showed expression of the markers CD7 or CD19 or showed CD56 expression, characteristics shared by blast cells in some AML cases [15]. However, as FSC/SSC was lower than for blast cells and similar to normal lymphocytes, further analysis was performed using antibodies against CD45, which enables discrimination between different types of white blood cells (WBC), and CD48 [10], a glycosylphosphatidylinositol (GPI)-anchored protein that is expressed on mature lymphocytes. Figures 1A5–1A8 identifies three small lymphoid subpopulations (all CD34-negative and CD33-negative) within the SP compartment, which also had low FSC: (a) natural killer-like cells with CD45high/CD56+/CD7low expression, with a median frequency of 4% (percentage of whole SP compartment; range, 2%–8%; n = 8; example in Fig. 1A5); (b) B lymphocytes with CD45high/CD19+ phenotype, and a median frequency of 2% (range, 0%–6%; n = 8; example in Fig. 1A6); and (c) T lymphocytes with CD45high/CD56/CD7high expression, with a median frequency of 7% (range, 2%–14%; n = 8; example in Fig. 1A7). The latter resembled the previously described CD34CD7+ SP cells [16]. All three CD45high populations cluster together in the low FSC area (Fig. 1A8; compare with Fig. 1A3. Similar to the AML samples, NBM samples also showed these three lymphoid SP subpopulations: NK-like, B and T cells, with median frequencies of 5%, 2%, and 8%, respectively (n = 3). Therefore, it is highly likely that the AML SP compartment contains a subset of normal lymphocyte cells. In the malignant samples, all three lymphoid SP subpopulations (CD45high, FSClow) were CD48-positive (Fig. 1B1). In contrast, myeloid SP cells (CD45dim) were CD48-negative (Fig. 1B1, R6). For comparison, the CD45high lymphoid cells within the non-SP (Fig. 1B2, R2) were also CD48-positive (Fig. 1B3), whereas the myeloid cells were not (Fig. 1B4). The NBM samples showed similar results (Fig. 1B5–1B8). Thus, staining with CD48 enables the nonmyeloid cells to be excluded from further analyses of stem cell activity in the myeloid SP subcompartment.

Figure Figure 1..

Side population (SP) cell immunophenotyping: gating strategy and SP subpopulations. Cells were stained for SP, CD45, lineage markers (CD7, CD19, CD56), and CD48, as outlined in Materials and Methods. (A1–A4): One representative example of acute myeloid leukemia (AML) SP immunophenotyping with different marker expression patterns and populations with different FSC and SSC. (A1): FSC versus PI of an AML sample (R1). Gating on PI-negative cells identified viable cells (R1). Gate R1 was used in (A2). (A2): Hoechst red versus Hoechst blue identified the viable side population cells (R2). Gates R1 + R2 were used in (A3). (A3): FSC/SSC was used to select a cluster or clusters of cells (R3). Nonspecific events are excluded here. Note the presence of two clusters differing in SSC but especially in FSC (A3). The large cluster was located in the whole blast region (not shown). Gates R1 + R2 + R3 were used in (A4). (A4): This shows the final population of viable SP cells containing as few as possible nonspecific events (R4). (A5–A8): Representative example of lymphoid SP subpopulations. Three small lymphoid subpopulations were detected within SP cells. (A5): Gate R5 shows CD45high/CD56+ SP cells. (A6): Gate R6 shows CD45high/CD19+ SP cells. (A7): Gate R7 shows CD45high/CD7+ SP cells. The CD45high cells with low expression of CD7 are the CD56+ cells from A5. (A8): Location of cells from gates R5 + R6 + R7 in FSC/SSC. (B1–B4): One representative example of CD48 expression on SP and non-SP subpopulations of an AML patient. (B1): Gate R6 shows CD48/CD45dim myeloid SP. Lymphocytes (CD45high) were CD48+. (B2): CD45high Lymphocytes (R1) and CD45dim myeloid cells (R2) from non-SP were selected and used in (B3, B4). (B3): Lymphocytes (CD45high) from non-SP were essentially CD48-positive. (B4): Myeloid cells (CD45dim) from non-SP were CD48 negative. (B5–B8): One representative example of CD48 expression on SP and non-SP subpopulations of a normal bone marrow. The results were similar to (B1–B4); that is, the lymphoid population was CD48-positive and the myeloid population was CD48-negative. Abbreviations: APC, allophycocyanin; FITC, fluorescein isothiocyanate; FSC, forward scatter; PI, propidium iodide; SSC, sideward scatter.

Leukemic Cells with Aberrant Marker Expression Are Present in the SP Compartment

Marker expression on the CD45dim AML SP cells was determined after exclusion of the CD45high lymphocytic cells. In 39 of 41 cases with SP present, the SP cells were partly or completely positive (defined as ≥10% expression) for CLL-1 (in all 41 samples: median, 53%; range, 2%–100%) and in 27 of 41 cases partly or completely positive for CD123 (median, 30%; range, 0%–100%; n = 41). In 25 of 41 cases SP cells were partly or completely positive for one or more of the three lineage markers previously shown to mark AML CD34+CD38 stem cells [3], that is, CD7, CD19, and CD56. Median expression was 35% for CD7 (range, 18%–80%; n = 13), 55% for CD19 (range, 20%–95%; n = 5), and 50% for CD56 (range, 25%–100%; n = 17). As CLL-1 and aberrant lineage marker expression have never been found on normal CD34+CD38 previously by our laboratory [3, 4], this strongly suggest that malignant myeloid cells in the SP cell compartment in AML at diagnosis can be identified using aberrant expression of markers. This was confirmed by comparison with surface marker expression on SP cells in 12 normal bone marrow samples. These SP cells were negative for CLL-1 (median, 0%; range, 0%–4%), CD7 (median, 0%; range, 0%–3%), CD19 (median, 0%; range, 0%–3%), and CD56 (median, 0%; range, 0%–4%), although not for CD123 (median, 27%; range, 1%–82%). Therefore, both CLL-1 and lineage markers, but not CD123, are suitable to discriminate between malignant and normal SP stem cells at diagnosis. FISH analysis for three t(8;21) and one del(12) AML patients (in Table 1, patients 2, 10, 41, and 22, respectively) showed that the majority of cells were indeed malignant (88%, 76%, 78%, and 80%, respectively). However, both marker patterns and FISH analyses indicated that even after correction for the lymphoid compartment (not shown), some of the myeloid SP cells should still be of normal origin.

Table Table 1.. Marker expression on myeloid, scatter-defined SP subpopulations in acute myeloid leukemia patients
original image

The Myeloid SP Compartment Is Heterogeneous in Scatter Properties and CD34 and CD38 Expression

On closer inspection, in 39 of 41 samples heterogeneity for sideward scatter was observed in the myeloid SP compartment (example in Fig. 2A1), showing two different subpopulations: (a) high sideward scatter (HSSC) cells with a median frequency of 56% (percentage of whole SP compartment; range, 4%–91%); (b) low sideward scatter (LSSC) myeloid SP cells with a median frequency of 44% (percentage of whole SP compartment; range, 9%–96%). In 2 of 41 cases only the LSSC SP cells presented as a separate population (an example is shown in Fig. 1A3).

Figure Figure 2..

SSC heterogeneity of SP cells. Cells were stained with Hoechst 33342 and with antibodies for expression CD34, CD38, and lineage marker CD56, as outlined in Materials and Methods. (A1–A4): One representative example of two SP subpopulations with different SSC: LSSC (R4) and HSSC (R2). Dashed line separates HSSC and LSSC. In this case the HSSC population was also characterized by high FSC. The LSSC population was partly CD56+, whereas the HSSC population was largely CD56+. Both the HSSC and LSSC fractions were largely CD34-negative (A3), whereas the HSSC populations, especially, were largely CD38+ (A4), in contrast to the LSSC. (B1–B4): Hoechst staining patterns of HSSC and LSSC SP subpopulations of the same patient shown in (A). HSSC and LSSC showed no significance difference in Hoechst staining: LSSC had mean fluorescence intensities (MFIs) of 70 and 141 for Hoechst red and blue, respectively; and HSSC had MFIs of 75 and 161 for Hoechst red and blue, respectively (three other samples analyzed showed similar results). The SP cells were ablated when BCRP1 inhibitor KO143 (200 nM) was included during the Hoechst incubation (B4). (C1–C4): One representative example of an NBM with two SPs with different SSC. CD34 and CD38 expression ([C3] and [C4], respectively) are typical for NBM. As illustrated in (C2) (and summarized in Table 1), CD56 was absent on both SP subpopulations (C2). Gating on LSSC and HSSC was slightly different from that of AML (A1), since increasing the R1 region to the size of R1 in Figure 1A1 would result in R1 including two populations with clearly different scatter properties. In the absence of further information on other properties, this is dissuaded. Abbreviations: AML, acute myeloid leukemia; FSC, forward scatter; HSSC, high sideward scatter; LSSC, low sideward scatter; NBM, normal bone marrow; PE, phycoerythrin; SP, side population; SSC, sideward scatter.

Backgating of HSSC and LSSC cells in Hoechst bivariate plots showed their presence throughout the SP, with no significant difference in Hoechst staining between LSSC and HSSC SP cells (Fig. 2B1–2B3). HSSC and LSSC populations were also seen in 9 of 12 NBM SP cells (one representative example is shown in Fig. 2C1), with only LSSC SP present in the remaining three cases (not shown).

HSSC AML SP cells had high CD38 expression, with a median frequency of 84% (range, 0%–100%; n = 39; example in Fig. 2A4). LSSC SP cells had significantly (p = .04) lower CD38 expression (example in Fig. 2A4), with a median frequency of 43% (range, 1%–100%; n = 41; p = .04). The median CD34 expression on HSSC and LSSC SP cells was 19% (range, 0%–99%) and 41% (range, 1%–99%), respectively. Apparently, compared with HSSC cells, LSSC cells have characteristics indicative of their primitive nature: higher CD34 expression, lower CD38 expression, and lower SSC.

Combining Scatter Properties and Marker Expression Defines the Presence of at Least Three Different Myeloid SP Subpopulations

Subsequently, aberrant marker expression was studied separately in HSSC and LSSC (data provided per patient in Table 1). CD123 was not included since, as reported earlier, it is expressed on normal bone marrow SP cells as well. HSSC showed heterogeneous expression patterns for the markers CD7, CD19, CD56, and CLL-1 (e.g., patient 10: 62% CLL-1+, 3% CD7+, 6% CD19+, and 65% CD56+). However, the HSSC population usually showed homogeneous expression (either high or low or absent) for all of these markers. In contrast, within the LSSC SP cells in general, two clearly discernable myeloid subpopulations could be identified: one negative for aberrant markers and the other with aberrant markers present. As a particular example, Figure 2A2 shows CD56 expression on the majority of HSSC SP cells but only on part of the LSSC fraction. In general, both HSSC and LSSC compartments showed heterogeneous expression patterns of CLL-1 and aberrant markers, with often large intrapatient differences in expression between both compartments (e.g., patient 6, 10, 11, 15, 27, 35, 38, and 41 in Table 1). Overall, compared with LSSC, marker expression was higher on HSSC for CLL-1 and CD56 but lower for CD7 and CD19. No specific localization of the marker-positive or marker-negative subpopulation was seen in the Hoechst plot (Fig. 2B1–2B3).

To confirm the putative malignant characteristics of the three identified myeloid SP subpopulations, we sorted these for two cases and performed a FISH analysis. In both patient 10 (Fig. 3) and patient 41 (not shown), sorted CD19-positive LSSC SP cells were largely positive for t(8;21), whereas sorted CD19-negative LSSC SP cells were largely FISH-negative (82% and 86%, respectively). However, the HSSC SP cells, although t(8:21)-positive, were largely CD19-negative, thereby explaining the observed discrepancies between marker expression and FISH analysis when studying the whole myeloid SP compartment. In these particular cases, the use of other markers (CLL-1 and CD56 for patient 10, CLL-1 for patient 41) confirmed the malignancy of these CD19-negative cells. The discrepancies between FISH analysis and marker expression and/or between expression of two or more different markers used are caused mainly by the contribution of HSSC, as there is no statistically significant difference in the expressions of CLL-1, CD7, CD19, and CD56 on the LSSC SP (Table 1 summary in the row just below patient 41).

Figure Figure 3..

Fluorescence in situ hybridization (FISH) analysis of high sideward scatter (HSSC) and LSSC SP cells in a t(8;21) acute myeloid leukemia (AML) patient. (A): FSC/SSC plot of the SP cells from a t(8;21) AML patient (patient 10) with two subpopulations (HSSC and LSSC). (A) is the same as Figure 2A. (B): HSSC SP cells (largely CD19-negative; R3) were sorted and investigated by FISH analysis in (D). (C): CD19+ cells (R4) and CD19 cells (R5) present within the LSSC SP cells were sorted and investigated by FISH analysis in (E) and (F), respectively. (D): HSSC: 78% positive (39/50), (E): CD19+ LSSC: 84% positive (42/50), (F): CD19− LSSC: 82% negative (41/50). Between brackets are the ratios of FISH+ cells and total number of scored cells for the different SP subfractions. Abbreviations: FSC, forward scatter; LSSC, low sideward scatter; PE, phycoerythrin; SP, side population; SSC, sideward scatter.

Apart from the difference in SSC, FSC was also lower in 36 patients in the LSSC population (example in Fig. 3). The differences in SSC, FSC, CD34 expression, and CD38 expression strongly suggest that the LSSC population is likely to be more primitive than the HSSC population and contains both marker-positive AML stem cells and marker-negative normal stem cells. Clonogenic assays were performed to support this hypothesis.

The LSSC SP Subfraction Is Enriched with Primitive Cells

Both LSSC (CD38low: range of expression, 22%–35%) and HSSC (CD38high: range of expression, 62%–85%) SP cells for seven AML patients (five CD34-positive, patients 9, 10, 22, 35, and 41; two CD34-negative, patients 37 and 40) were sorted and subsequently cultured in a suspension culture assay. After 5 weeks, all cells were harvested and were placed into methylcellulose for the detection of day 14 colonies. The HSSC SP cells had a very low clonogenic capacity (median of 400 per million input cells; range, 250–650), whereas the LSSC SP cells had a very high clonogenic capacity (median of 19,375 per million cells; range, 7,253–24,645; Fig. 4). The results show that LSSC SP cells with the low CD38 expression have a more primitive character in functional assay.

Figure Figure 4..

Long-term clonogenic capacity of sorted side population (SP) subpopulations of seven acute myeloid leukemia patients. HSSC and LSSC fractions were sorted and put in liquid culture. After 5 weeks of suspension culture, sorted LSSC SP cells showed high clonogenic capacity in a 2-week colony-forming unit assay. The sorted HSSC SP cells showed no or very low clonogenic capacity. Abbreviations: HSSC, high sideward scatter; LSSC, low sideward scatter.

To further discriminate within the LSSC population between the clonogenic ability of presumed normal and AML cells, both marker-positive and marker-negative SP LSSC cells from patient 37 (CD34-negative) and patient 41 (CD34-positive) were used in the functional assay. All populations formed high numbers of colonies. For patient 37, the data were 14,645 colonies per 106 cells for marker-positive and 8,010 colonies per 106 cells for marker-negative populations. For patient 41, these figures were 6,460 and 5,270 colonies per 106 cells, respectively (not shown). FISH analysis performed on week 5 of the colonies originating from marker-positive or marker-negative LSSC cells of two patients (patients 22 and 41 with del(12) and t(8;21), respectively), showed that the majority of the colonies originating from marker-positive LSSC cells were indeed malignant (80% and 78%, respectively; counted on 45 and 50 cells, respectively). The colonies originating from marker-negative LSSC cells of both patients were predominantly of normal origin (both 86% FISH-negative; counted on 35 and 38 cells, respectively). No colonies were formed in the controls starting either with 100,000 or 1 million NSP cells (not shown).

To control for the possible negative effect of Hoechst exposure on the function of NSP, sorted SPs for four patients (patients 22, 37, 40, and 41) were reincubated with Hoechst (2 hours at 37°C) in the absence or presence of the inhibitor and subsequently put into CFU assay. After 14 days, both populations showed similar numbers of colonies for all four patients: medians of 52 (range, 40–58) and 47 (range, 37–51) colonies per 1 million cells in the absence and presence of the inhibitor, respectively. These results show that increasing Hoechst binding in SP cells by allowing these to become non-SP cells had no impact on the clonogenic capacity at concentrations used for this study. When these data are taken together, the median frequency of total SP cells was 0.03% (n = 25), using the markers with highest expression, whereas the frequency of marker-positive LSSC SP cells was only 0.0016% of that of WBC (range, 0.0002%–0.0056%).


AML is regarded to originate, at least in some cases, in the hematopoietic stem cell compartment [1]. The lack of durable response in a high percentage of AML patients suggests that current treatments do not effectively target LSCs. Therefore, LSCs have to be identified to prove their persistence after current treatments, whereas their characterization might pave the way to identify new therapeutic targets.

In CD34-positive AML the stem cell has been recognized as CD38-negative [3]. However, not all stem cells can be defined as CD34+CD38. At diagnosis, 5% of AML patients with a CD34+ phenotype lack a clearly detectable CD34+CD38 compartment (< 0.01%), and approximately 20% are CD34-negative and thereby, by definition, CD34+CD38-negative [2]. An alternative stem cell compartment for these cases is offered by the side population. This population is highly enriched for HSCs [6] in NBM. SP cells are also present in bone marrow of AML patients and are capable of initiating leukemia after transplantation into NOD/SCID mice, suggesting that these cells contain candidate LSCs [8].

In the present study we reported that the frequency of SP cells is far lower (factor, approximately 16) than that of CD34+CD38 cells. However, using flow cytometry, we have found that the SP fraction was still highly heterogeneous and contained different subpopulations: (a) three small normal lymphoid subpopulations (T, B, and NK-like cells) with low FSC/SSC properties, high CD45 expression, and CD48 expression; (b) a differentiated (high FSC/SSC, high CD38, low CD34) myeloid population with or without aberrant leukemic marker expression; (c) a primitive low-frequency myeloid fraction (low FSC/SSC, low CD38, high CD34), negative for aberrant leukemic markers, and likely enriched for primitive normal cells; (d) a primitive low-frequency myeloid fraction (low FSC/SSC, low CD38, high CD34), with aberrant leukemic markers present and likely enriched for primitive AML stem cells. NBM showed the first three populations; however, these were always, with marker expression, absent in the second population. These results strongly suggest that the majority of SP cells are malignant, which was indeed confirmed by FISH analysis in four of our patients and has also been shown by others [8, 9]. Suspension culture followed by CFU assays showed that low FSC and SSC SP cells with low CD38 expression have a more primitive character in both CD34-positive and CD34-negative AML patients.

Heterogeneity in terms of malignancy of SP cells has been previously described [9], suggesting that normal stem cells are present and probably constitute the CD34+CD38 SP compartment. Although the relationship between the SP compartment and the CD34+CD38 compartment was not the focus of the present paper, we were able to confirm that observation, although only in CD34-negative AML, as defined by CD34 expression lower than 1% [13]. In contrast, in CD34-positive AML, the CD34+CD38 population consisted of both a malignant and a normal compartment (unpublished data).

Identification of malignant stem cell subpopulations among normal stem cells has thus become possible using aberrant marker expression. Both the earlier defined CLL-1 antigen [4] and lineage markers [3] were now found to characterize the AML SP compartment. There were, however, large differences between patients in expression of the markers used. Moreover, even for a given patient, the expression pattern of the markers showed large differences on many occasions. Evidently, one or more particular markers may not cover the whole AML part of the SP compartment, which in turn may well result in underestimation of the AML component. However, this problem may be circumvented due to the following peculiar characteristics: whereas CLL-1 expression was significantly higher in the HSSC fraction, CD7 and CD19 were higher in the LSSC fraction (Table 1). As a result, even in the absence of CD7 and/CD19 expression, the HSSC population may still be malignant as illustrated for patient 10 in Figure 3 and Table 1. Most importantly, the median expression of the four markers did not differ significantly in the most relevant compartment: the LSSC myeloid compartment. On the other hand, there were still differences in marker expression in the LSSC fraction within individual patients: for example, in patient 6 (lower CLL-1 expression compared with CD7 and CD56) or patient 39 (higher CLL-1 expression compared with CD7). Careful examination of such cases shows that this may reflect an old dilemma in flow cytometry with regard to the definition of positivity: a small shift of a whole population is likely to indicate that all cells are positive, but with low intensity of expression. However, using expression as the percentage positivity compared with isotype controls results in percentages far lower than 100%. It is therefore quite likely that if there is good congruence between the markers used, the marker of choice should then be the one that allows the best discrimination between putative normal and putative AML cells.

When the marker with the best discriminatory power on LSSC SP cells was used for each individual case of AML, the range of AML LSSC SP cells found was 0.0002%–0.0056% of WBC (median, 0.0016). This is now closer to the real AML stem cell frequency, which may be as low as 0.0001%, as estimated using limiting dilution experiments [1, 17]. The future challenges will be to study the interrelationship between the CD34+CD38 and SP stem cell compartment in CD34-positive AML and to identify the AML stem cell compartment in truly CD34-negative AML.

Interestingly, we could distinguish between the lymphoid and myeloid compartments within the SP and non-SPs by using CD48. As the lymphocytes have low FSC and SSC and may thereby interfere with a correct analysis of malignant and, especially, normal LSSC SP cells, it is important to be able to distinguish these from myeloid stem cells. This would facilitate the study of the stem cell compartment: CD48 fulfils this demand. CD48 is a GPI-anchored protein belonging to the CD2 subfamily [18]. It is a low-affinity ligand for CD2 and is implicated as an important costimulatory molecule in lymphocyte activation. CD48 has also been used in other studies, together with other signaling lymphocytic activation molecule (SLAM) family markers (CD150 and CD244), to identify stem cells and progenitors in mice [19].

In addition to hematopoietic malignancies, SP cells have been identified in various tumors [20]. It will be interesting to establish whether the SP compartment in these tumors has heterogeneity similar to that detected in this study for AML. Such information might be a valuable tool with which to study and target these cells. In this respect we have already been able to identify LSSC and HSSC SP cells in glioblastoma and chronic myeloid leukemia (CML) SP cells (unpublished data).

Similar to the characterization of the leukemic blasts by using aberrant lineage markers [15, 21], the identification of immunophenotypical characteristics specific for the malignant primitive SP cells at diagnosis would offer opportunities to study primitive AML SP cells under conditions of minimal residual disease after chemotherapy. This might not only identify patients at risk for relapse, thereby improving the clinical significance of MRD cell studies [14], but also enable characterization of cells that probably represent the most relevant target cell population to design new therapies [3, 4]. This very low frequency SP subpopulation is a likely candidate to be enriched for leukemia-initiating cells, although the leukemia-initiating ability of the identified primitive SP compartment will need proof in animal models.


The SP compartment in AML is highly heterogeneous in terms of different cell lines and differentiation status. Using expression of CLL-1 and lineage markers, AML SP cells can now be discriminated from normal SP cells. Using functional assays, this has revealed the presence of a low frequency immature SP subfraction as a likely candidate to be enriched for leukemia initiating cells.

Disclosure of Potential Conflicts of Interest

The authors indicate no potential conflicts of interest.


The Department of Cardiac Surgery of the VU University Medical Center and its patients are gratefully acknowledged for providing normal bone marrow samples.