Determination of antigen expression patterns is, in addition to morphologic analysis, essential to the diagnosis of acute myeloid leukemia (AML). The present study was performed to determine (a) the degree of changes in immunophenotype and their consequences on the monitoring of minimal residual disease (MRD) in childhood AML and (b) whether certain clusters of changes in antigen expression patterns exist between diagnosis and relapse.
Bone marrow specimens of 48 children enrolled in the German AML-BFM-93/98 (Acute Myeloid Leukemia-Berlin-Frankfurt-Münster) studies were analyzed immunologically, morphologically, and genetically at diagnosis and at first relapse.
The immunophenotypes by flow cytometry differed by at least one antigen between samples at presentation and relapse in 42 of 48 children (88%). More children displayed an immature phenotype at relapse (43 of 47, 91.5%, vs. 37 of 48, 77%; P = 0.05) with expression of CD34 and/or CD117. This was reflected by a gain of markers that are associated with lineage immaturity in 18 of 25 (72%) of cases, whereas the loss of such antigens was observed in 6 of 25 (24%) patients. We did not observe significant changes for lineage specific markers, with comparable occurrences of loss or gain of myeloid and lymphoid antigens in the sample pairs. Only minimal changes were seen for morphologic and genetic features.
For the diagnosis of acute myeloid leukemia (AML), the determination of antigen expression pattern, in addition to morphologic analysis, is essential. The detection of leukemia-associated immunophenotypes using multiparametric flow cytometry allows the reliable diagnosis of AML in up to 90% of patients. Recently Kern et al. reported that at least one aberrant/infrequent immunophenotype was identified in 68 adults with AML, suggesting that flow cytometry could be used for minimal residual disease (MRD) follow-up in most patients (1). Immunophenotypes in AML include (a) asynchronous expression of progenitor-associated antigens and myeloid lineage markers, (b) aberrant expression of lymphoid cell lineage antigens on myeloid blast cells, and (c) over- or underexpression of antigens (2–4).
The relevance of MRD diagnostics in AML for risk stratification based on treatment response has not been evaluated, mainly because molecular markers (e.g. AML1/ETO, PML/RARα, CBFβ/MYH11) are infrequent in AML (5, 6). The use of multiparametric flow cytometry for the monitoring MRD during follow up permits the early detection of relapse and is therefore considered to provide a universal tool for MRD follow-up (7–9). However, the main obstacle to the application and standardization in MRD diagnosis is the occurrence of antigen shift during treatment.
Until now, immunophenotypic changes in antigen expression of patients with AML have been reported only in adults. The ratio of patients with antigen shift varied from 50% to 91%, depending on the antibody panel and treatment used (10–14). However, several reports concerning the high incidence of antigen shift in childhood and adult acute lymphoblastic leukemia patients have compared immunophenotypes at diagnosis and relapse (11, 13–17)
In this study we compared flow cytometric profiles of 48 pediatric AML cases at the time of initial diagnosis and first relapse to (a) establish the frequency and type of antigenic shift, (b) correlate clusters of immunophenotypic changes with morphologic and cytogenetic classifications, and (c) clarify the requirements for an optimal antibody panel for MRD monitoring in childhood AML.
MATERIALS AND METHODS
Bone marrow specimens of children with de novo AML (n = 48, including one child with Down's syndrome) were obtained after informed consent from a parent or guardian of the patient. All children were treated according to the German AML-BFM-93/98 (Acute Myeloid Leukemia-Berlin-Frankfurt-Munster 1993 and 1998) studies. Treatment schedules have been reported elsewhere (18, 19). Patient characteristics are listed in Table 1. All investigations were performed after approval by the local human investigations committee and in accord with an assurance filed with and approved by the Department of Health and Human Services.
Four-color flow-cytometry was performed at the immunology laboratories at the University Children′s Hospital Münster and at the University Hospital Göttingen at initial diagnosis and at first relapse. Anticoagulated bone marrow samples had been sent to one of the laboratories by overnight mail.
A wide antibody panel was applied, including fluorescence-conjugated myeloid markers CD13 plus phycoerythrin (PE; clone SJ1D1; Immunotech, Marsalles, France), CD15 plus fluorescein isothiocyanate (FITC; clone MMA; Becton Dickinson, San Jose, CA), CD33-PC5 (clone D3HL60.251; Immunotech), and HLA-DR-FITC (clone L243; Becton Dickinson); lymphoid markers CD7-PE (clone 8H8.1; Immunotech), CD10-FITC (clone ALB2; Immunotech), CD19-FITC (clone J4.119; Immunotech), and CD56-PE (clone NCAM 16.2; Becton Dickinson), CD79a-PE (clone HM47; Immunotech); and progenitor-associated markers CD34 plus allophycocyanin (clone 8G12; Becton Dickinson), and CD117-FITC (clone 95C3; Immunotech).
After incubating bone marrow samples with monoclonal antibodies for 15 min, erythrocytes were lysed for 7 min with FACS Lysing Solution (Becton Dickinson). Afterward specimens were washed twice with 2 ml phosphate buffered saline (pH 7.4) and centrifuged (5 min, 20°C, 600g) to remove excess antibodies and lysed red blood cells. Specimens were measured with a FACSCalibur (Becton Dickinson) or EPICS XL (Beckman Coulter) analyzing at least 30,000 events. Extensive interinstrumental comparisons were performed to ensure that either site could detect similar percentages of positivity.
Specimens were evaluated with Paint-a-gate PRO software (Becton Dickinson). To identify a malignant cell population within residual normal hematopoietic cells, cluster analysis using six parameters (forward and side scatter properties and four antigens simultaneously) was performed. Antigens were considered positive when fluorescence intensity could be clearly separated from negative controls (isotypic controls or negative cell populations for the examined antigen). All bone marrow samples of patients were analyzed by four-color immunophenotyping according to the consensus panel of the AML-BFM MRD study group as described elsewhere (20). In AML, expression of antigens is frequently heterogeneous, e.g., the expression of one or more antigens within the blast cell population may range from negative to positive (Fig. 1). According to the consensus guidelines for immunologic diagnosis of acute leukemias, an antigen is supposed to be negative if it expressed by fewer than 20% of cells (21). In our analysis, further differentiations, low positive (expressed by ≥20% and <40%) and positive (expressed by ≥40%), were made. In general, blast cells were roughly gated by forward and side scatter properties followed by precise gating according to CD33/CD34 expression. In cases of CD34 negativity, gating was based on CD117 or an aberrant expression of a lymphoid antigen. To avoid bias, all samples were evaluated by at least two investigators.
Diagnosis and classification of AML were performed according to the standard morphologic and cytochemical criteria of the French-American-British group by the reference laboratory of the AML-BFM studies in Münster and reviewed by an expert group of extern hematologists.
Cytogenetics and Molecular Genetics
Cytogenetic analyses were performed in the reference laboratory of the AML-BFM study. Chromosome preparation from bone marrow and/or peripheral blood was done according to standard techniques after culturing for 24 or 48 h (22). Metaphase spreads were prepared according to routine methods including Colcemid treatment, hypotonic shock, and 3:1 methanol:acetic acid fixation, and chromosome analysis was carried out on G-banded metaphases. In general at least four metaphases were karyotyped and described according to the 1995 International System for Human Cytogenetic Nomenclature (23). Additional molecular genetic studies were executed to confirm or exclude the presence of specific leukemia-associated gene rearrangements using fluorescence in situ hybridization with the LSI Dual Color Break Apart Rearrangement Probe for mixed lineage leukemia (MLL) or reverse transcriptase polymerase chain reaction (RT-PCR) for the detection of AML specific gene rearrangements (24).
McNemar's test for dependent proportions was applied to compare antigen expression of the same children at diagnosis and at relapse. Chi-square test was used to compare frequencies of antigen expression at diagnosis between the children who relapsed and those who did not.
In 42 of 48 children (88%), changes in at least one antigen between diagnosis and relapse could be observed. Table 2 presents morphologic, immunologic, and genetic data of each child at diagnosis and relapse and the kind of change concerning immunophenotype and karyotype. Classified by cell lineage, changes in myeloid antigens (CD13, CD15, CD33, HLA-DR) were the most frequent (34 of 48, 71%), followed by changes in lymphatic antigens (CD7, CD10, CD19, CD56; 30 of 46, 65%) and changes in progenitor associated antigens (CD34, CD117; 24 of 47, 51%). An example of one child's bone marrow samples at diagnosis and at relapse with changes of immunophenotype is shown in Figure 1. Initially, the blast cells (95%) expressed CD13 and CD33 and were negative for CD34, CD117, and CD19. However, at relapse, the blast cells showed a more immature immunophenotype with expression of CD33, CD34, CD117, and CD19 (negative: CD13). Typically, at relapse the progenitor-associated antigens CD34 and CD117 are not expressed by the whole blast cell population homogenously, but the cells seem to be in different stages of maturation/differentiation. In most cases, changes in progenitor-associated antigens included gain of antigen expression only (17 of 24, 71%), whereas changes in the myeloid and lymphatic antigens were almost balanced for loss and gain (Fig. 2).
Table 2. Immunophenotypes and Karyotypes of 48 Children With AML at Initial Diagnosis and Relapse
***Fluorescence in situ hybridization analysis; −, negative; +, positive; ++, high positive; bi, biphenotypic; cl. evol., clonal evolution; D, diagnosis; G, gain; L, loss; nd, not done; PCR, aberrant clone proofed by RT-PCR; R, relapse.
There were no significant differences in the frequency of single antigen expression between diagnosis and relapse. Equally no significant differences in initial antigen expression were determined in comparison with a control group of children enrolled in the AML-BFM-93/98 studies without relapse (n = 154; Table 3). Regarding the expression of progenitor-associated antigens CD34 and CD117 at diagnosis, 37 of 48 (77%) of children showed an immature immunophenotype with at least one of the two analyzed markers, 24 of these (65%) were positive for both antigens (CD34+/CD117+). Blast cells of 10 children expressed one of the progenitor-associated antigens (CD34+/CD117−, n = 3; CD34−/CD117+, n = 7) in three children who were positive for CD34, CD117 was not analyzed. Eight children initially negative for both antigens (CD34, CD117) expressed at least one of the markers at relapse, whereas only one child showed minimal loss (from positive to low positive) of CD34 and CD117 at relapse as compared with (high) positive at initial diagnosis. Blast cells of three children remained negative for both markers. At relapse, a significantly higher percentage of children (43 of 47, 91.5%, vs. 37 of 48, 77%; P = 0.05) showed an immature immunophenotype expressing CD34 and/or CD117.
Table 3. Expression of Antigens on Initial and Relapse Blast Cells in Patients Who Had a Relapse Compared With Those Who Did Not‡
Both groups were diagnosed within the same period. N.S., not significant.
CD34 and/or CD117
Regarding antigen expression of single antigens (Fig. 3), changes in the expression of CD15, CD13, CD117, and CD7 occurred in more than 35%, whereas changes in the expression of CD33 were seen in 2%. Changes within the other antigens analyzed (CD34, HLA-DR, CD56, CD19, and CD10) ranged from 11% to 21%. The ratio of gain and loss of antigen expression within the progenitor-associated antigens was more than 2.5, whereas the ratio within the other antigens (myeloid and lymphatic) was nearly balanced.
In 41 of 42 (98%) children with immunologic changes, the population with the antigen expression pattern shown at relapse could not be detected at diagnosis. Only one child with 70% blast cells initially expressing CD13/CD15/CD33/CD117 showed proportion of 17% of the cells with the antigen expression pattern of the blast cells at relapse (in addition to CD34).
Morphologically, there were no differences in French-American-British (FAB) subtypes between diagnosis and relapse. Concerning cytochemistry, there were only minimal changes between diagnosis and relapse (periodic acid-Schiff, 5 of 26, 19%; phosphatase, 6 of 20, 30%; peroxidase, 8 of 30, 27%; a-napthyl-acetate-esterase (ANAE), 9 of 31, 29%).
Thirty-three of the 48 children were analyzed by cytogenetic and/or molecular genetic methods at diagnosis and at relapse. Initially, six children (18%) showed a normal karyotype including one child with trisomy 21, whereas most children (n = 27) showed at least one aberration. In most patients (28 of 33, 85%), at least one of the aberrations was detected at both points (n = 26) or the karyotype remained normal (n = 2). Complete changes were observed in five children (15%): four patients showed a normal karyotype at diagnosis and aberrations at relapse, whereas one child had an initial abnormality that disappeared at relapse. No differences were found in 15 children, whereas a clonal evolution was obvious in nine. In four children the initial abnormality was not seen at relapse by cytogenetics (in two of them, blast percentage was very low; in the other two, no analysis was performed at relapse), but RT-PCR proofed the aberrant clone.
Comparing immunologic and cytogenetic data, an association between karyotype and antigen expression within these children was the same as that found in children without relapse (25). Moreover, there was no correlation between changes in antigen expression pattern and secondary aberrations.
Determination of antigen expression pattern is essential to the diagnosis of AML. To evaluate changes of immunophenotype from initial diagnosis to first relapse, antigen expression patterns of 48 children enrolled in the German AML-BFM-93/98 studies were analyzed with four-color flow cytometry and a wide antibody panel including progenitor-associated, myeloid, and lymphoid markers. Compared with three-color flow cytometry, a noticeable improvement in terms of sensitivity and specificity leading to a higher significance can be achieved by the addition of a fourth color, resulting in six different properties of each cell achieved simultaneously. Because the vast majority of AML blasts at diagnosis and mainly at relapse coexpress the myeloid antigen CD33 and the progenitor-associated antigen CD34, we integrated a CD33/CD34 backbone to each tube in the antibody panel. In cases of CD34 negativity, the gating strategy was based on CD117 or the aberrant expression of lymphoid antigens, respectively, and CD45. Additional antibodies were combined according to the typical immunophenotypes of AML blasts. By using this panel, we were able to detect the blast cell population in all children who had been analyzed at diagnosis and relapse.
Changes within the immunophenotype, including gain and loss of antigens, were determined in nearly all children, indicating that the immunophenotype of AML blasts is unstable during treatment. These findings are in accordance with results presented by Baer et al. (10) who analyzed antigen expression pattern at diagnosis and relapse in adults (n = 136) who had AML. These investigators found changes in 91% of patients. Macedo et al. (12) reported changes in the expression of at least one antigen in 10 of 16 adults with AML; however, with particular attention to aberrant antigens, the initial immunophenotype generally remained unchanged. Oelschlagel et al. (13) described a significant decrease in aberrant antigen expression from diagnosis to relapse in AML patients (n = 52).
These findings provide important implications for MRD monitoring and immunologic diagnosis of a relapse. It is not sufficient to analyze the initial antigen expression pattern; rather, one must carefully evaluate a defined setting of antigens. Although recent studies have identified the MRD level by monitoring a patient's individual initial immunophenotype as the strongest independent prognostic factor, the results must be relativized. San Miguel et al. (26) monitored 41 different immunophenotypes without providing any details concerning sensitivity and specificity; therefore, reproducibility of the analysis seems to be impaired. Sievers et al. (27) included all patients with morphologically less than 15% blasts after first induction therapy. A 4.8-fold increased risk of relapse was obtained for those patients with at least 0.5% blast cells after first course of chemotherapy. In this approach, the high incidence of relapses in the MRD-negative group is problematic in terms of specificity. Further, the prognostic differences between the two groups can be explained by other risk factors. Recently, Coustain-Smith et al. (28) published their results of monitoring MRD in childhood AML by observing the initial immunophenotype. Because the amount of relapses in the MRD-positive and MRD-negative groups was almost similar (6 of 29, 21%, vs. 5 of 17, 29%) and complete remission (CR) rate was equal to relapse rate (both 5 of 17, 29%) in the MRD-positive group, this approach seems not to be sufficient to predict outcome. Furthermore, the results of this study have to be questioned, because the MRD-positive group also included four children morphologically classified as nonresponders. For MRD monitoring, these patients should be excluded, because they are already detectable by conventional methods and MRD diagnosis does not provide further information.
In our multicenter MRD study within the AML-BFM study, we used a standardized four-color antibody panel, independent of the initial immunophenotype, at defined times during treatment. We defined four groups of leukemia-associated immunophenotypes with different specificities and determined sensitivity using dilution assays to perspicuously improve quality and reliability of MRD monitoring (20)
Complete karyotype changes occurred in 5 of 33 children (15%) analyzed at diagnosis and at relapse; in all of them, at diagnosis or at relapse, the karyotype was normal. Changes in terms of clonal evolution were detected in 27% (9 of 33) of children, and all other karyotypes remained the same at diagnosis and relapse. These findings conform with results obtained in adults: Changes in karyotype have been described in 38% to 55%, in the majority involving clonal evolution (10, 29, 30)
These minimal changes within genetic features of blast cells between initial diagnosis and relapse are the prerequisite for monitoring MRD using molecular genetic techniques (real-time quantitative RT-PCR) by analyzing the initially determined gene rearrangement. The only impairment is that a specific rearrangement is detectable in only about 30% of children (31, 32). Because the leukemia-associated karyotype remained stable, molecular genetic MRD diagnosis is suitable in those patients with initially detectable gene rearrangements.
The fact that, with the exception of one child, cells with the immunophenotype expressed at relapse could retrospectively not be observed at initial diagnosis may corroborate the hypothesis for the existence of one single blast clone with a less differentiated immunophenotype at relapse than at diagnosis. Otherwise at least a minimum proportion of cells with the antigen expression pattern of relapse would have been detectable at diagnosis. The development of a secondary leukemia could be suggested in those cases with karyotypic and immunologic changes. Nearly all blast cells of children with relapsed AML showed an immature immunophenotype present at initial diagnosis or the antigen expression pattern became less differentiated from diagnosis to relapse. Likewise Thomas et al. (33) ascertained that AML tends to relapse with a less differentiated immunophenotype as determined by the expression of CD34. Possibly, cells expressing the progenitor-associated antigens CD34 and CD117 are more resistant to external influences than other cells characterized by more mature markers. Another possible explanation for the less differentiated immunophenotype at relapse may be the earlier detection recognizable by lower blast cell proportions.
Compared with normal hematopoietic cells, malignant cells appear to have different potentials for differentiation and/or maturation. During clonal evolution, a leukemic stem cell with distinct genetic features and capacities for differentiation and proliferation may arise from normal primitive cells. Therefore, human AML seems to be organized hierarchically with different stages of maturation and/or differentiation characterized by antigen expression pattern (34). During treatment, a small population of leukemic stem cells, detectable as MRD (31), seems to survive and due to a second hit starts to proliferate very quickly with only minimal differentiation, characterized by the expression of progenitor-associated antigens. The hypothesis of a “sleeping leukemic stem cell” as the origin of initial and relapse leukemic populations is supported by the fact, that in most children, at least one aberration is identical at both times. Our results support the model that the residual cell population may have a stem cell phenotype, which gives rise to relapsed leukemia with a more undifferentiated immunophenotype after the acquisition of additional genetic lesions.
We thank all the hospital personnel and clinicians who participated in the German AML-BFM Study Group for providing bone marrow samples and clinical data. We also thank Carolin Augsburg, Jutta Meltzer, Anke Reinkemeier, and Elisabeth Kurzknabe for excellent technical assistance.