Immunophenotype signature as a tool to define prognostic subgroups in childhood acute myeloid leukemia

Acute myeloid leukemia (AML) is a heterogeneous disease group morphologically classified, based on the French–American–British (FAB) classification, into eight main subgroups defined as subtypes M0–M7. Besides morphologic differences, genetic abnormalities have been recognized; cytogenetics and molecular analyses are currently used to identify subgroups of AML with different clinical prognosis. However, in spite of available prognostic factors, accurate prediction of risk for treatment failure or relapse is not completely satisfactory. In order to improve risk assignment and develop new therapeutic strategies, gene expression profiling and proteomic analysis seem to offer important improvements in leukemia classification.

Quantitative multivariate analysis from panels of marker proteins has demonstrated that marker protein expression profiles can distinguish specific ALL subtypes.1, 2 Here, we consider the application of immunophenotyping by flow cytometry (FC) to analyze AML samples using panels of antibodies to characterize specific blast cell populations. We test the applicability of unsupervised cluster analysis of immunophenotypic data in finding potential cryptic relations among AML patients. A total of 96 de novo pediatric AML patients (59 males; 37 females; median age 11, range 1–18 years) were investigated using multiparametric FC and analyzed by clustering methods normally applied in high-throughput gene expression profiling to identify new homogeneous subgroups with clinical relevance. In all cases, AML diagnosis was performed by golden standard protocols for morphology, cytochemistry, cytogenetics, immunophenotype and molecular biology.

The study cohort, a total of 96 patients of which 42 were positive to the screened genetic abnormalities and with a median follow-up of 2 years, is a representative series of Italian pediatric AML patients. We analyzed immunophenotypic data of patients whose diagnosis for AML started from January 2002 to August 2003. Patients were enrolled in the AIEOP AML-2002 protocol. 25 patients were positive for t(15;17) translocation (PML-RARα fusion gene), 11 for t(8;21) (AML1-ETO fusion gene), six for inv(16) (CBFβ-MYH11 fusion gene) and 54 negative for screened aberrations. Indeed, the frequency of genetic abnormalities is similar to a previously reported Italian study.3 Furthermore, the morphologic and genotypic characterizations confirm the correlation between FAB subgroups and the major genetic aberrations. The PML-RARα fusion gene was present in all M3 and M3v cases (acute promyelocytic leukemia, APL); AML1-ETO was present in M2 patients (eight of 12); six of 13 CBFβ-MYH11 cases were classified as M4. Patients with M0, M5, M6 and M7 FAB subtypes resulted negative for the analyzed cytogenetic abnormalities.2, 3

Unsupervised cluster analysis was performed on the entire cohort of 96 AML patients to group samples and antigens according to the similarity of antigen expression profiles. Two-dimensional clustering analysis (Figure 1) for mean antigen expression values separates patients into two main branches (A and B, Pearson correlation value=−0.062729). The dendrogram on the left in Figure 1 shows that patients are clearly grouped in two main clusters: a smaller but homogeneous branch (A), is separated from a larger heterogeneous branch (B) which further divides into six sub-branches (B1–B6).

Figure 1

Clustering on patients (96) and antigens (35) for AML panel (mean values).

Subsequently, Mann–Whitney test has been performed on logarithmic expression values of all markers (mean values) to identify significantly expressed antigens for each cluster group.

Moreover, the overall survival rates were calculated on identified clusters to evaluate the prognostic significance for patient groups with more than 10 samples (i.e. clusters A, B1, B2, B5 and B6) (Figure 2). The results of the Kaplan–Meier plot were significant in distinguishing the clusters considered (P=0.0179).

Figure 2

Kaplan–Meier plot was significant in distinguishing the considered clusters (P=0.0179).

Cluster A shows the highest probability of survival among all groups, as expected. It is mainly composed of M3 and M3v subtypes (APL) with all patients positive for PML-RARα; it has the highest specificity both for morphological and genotypic information. The higher expression of MPO (P<0.001), CD33 (P<0.001) and lower expression of CD11a (P<0.001) and CD11b (P<0.001) matches with APL cases. We kept APL specimens as a proof of concept of our method.

Patients in cluster B1 have relatively poor prognosis: samples are negative for principal aberrations and all patients but one (PML-RARα) are morphologically classified as FAB M0, M1 and M7. Blast cells express CD7 (P<0.001) and CD117 (P=0.004) while are negative for MPO (P<0.001), CD64 (P<0.001) and CD15 (P=0.005).

Cluster B2 predicts a good survival prognosis and includes 10/17 patients with the AML1-ETO translocation; the other seven patients included in cluster B2, but without this specific prognostic translocation, share similar phenotype and relative good prognosis. The most relevant antigen information of cluster B2 was represented by the lower expression levels of CD33 (P<0.001), CD135 (P=0.019) and the higher values of CD34 (P<0.001), CD66b (P=0.001) and CD66c (P=0.004).

The prognostic significance of clusters B3 and B4 cannot be evaluated as both of them comprise only six patients. They can be distinguished by their divergent expression of CD66b, CD66c (both P<0.001 and positive in cluster B3) and CD44 (P=0.017, positive in cluster B4).

Considering a lower grade of Pearson correlation, B5 and B6 branches form a single cluster in which mainly M4 and M5 samples are included, respectively. M4 and M5 FAB subtypes share many features as confirmed by the results of the dendrogram; acute monocytic leukemia (AML-M5) differs from acute myelomonocytic leukemia (AML-M4) mainly when monocytes count is greater than 80% without the myeloid component. Blast cells for patients in cluster B5 highly express CD14 (P<0.001) and CD11b (P<0.001) while generally being negative for CD114 (P<0.001), 7.1 (P<0.001), CD80 (P<0.001) and CD61 (P<0.001). Overall prognosis in cluster B5 is poor in spite of the presence of four patients with CBFβ-MYH11 translocation with good prognosis. Cluster B6 includes samples positive for CD64 (P<0.001), CD11a (P<0.001), CD15 (P<0.001), CD86 (P<0.001) while samples are negative for CD83 (P<0.001), MPO (P=0.015), CD34 (P=0.016) and CD114 (P=0.016). Cluster B6 confirms the variable risk related to M5 subtype and includes patients (except one with t(15;17)) without the considered genotype aberrations.

Acute myeloid leukemia is a heterogeneous disease with variable prognosis. Many useful prognostic factors have been studied and validated4 but attempts to accurately predict risk of treatment failure remain unsatisfactory. Recently, comprehensive gene expression profiling analyses have demonstrated to be a new powerful tool for identifying prognostically relevant patient subgroups in AML.4, 5, 6 Patient subgrouping partly coincides with known prognostic categories but new prognostic relevant groups have been also identified. The use of computational analysis performed on gene expression data in medicine strongly suggests that this method can be applied in various diagnostic approaches.

Here, 96 pediatric patients with AML are separated using antigen expression profiles drawn on quantitative FC data. At the same time, patients sharing different morphological phenotypes but homogeneous genotypes are aggregated according to their antigen expression profiles. We demonstrate that these antigenic profiles are relevant in clinical diagnosis and prognosis; additionally, they support morphological and genotypic analyses in defining prognostically relevant AML subgroups. Performing unsupervised analysis on antigen expression values, we distinguished AML patients in seven subgroups that are shown to be prognostically relevant. Notably, our method places AML patients with a normal karyotype into clusters with different prognosis: B2 and B6 (low and variable risk) versus B1 and B5 (high risk). In fact, the clustering approach helps us to associate patients without prognostically relevant features to patients with known survival risk according to immunophenotype information.

Moreover cluster analysis on normal karyotype may support the identification of distinct subgroups involving molecular pathogenesis of AML.7 Indeed, the use of antigen expression profiling like gene expression profiling helps us to separate clinically relevant subgroups.

Previously, a number of studies using FC immunophenotyping have attempted to identify single marker antigens with prognostic relevance for AML patients. In a recent study on adult AML patients, a similar approach was used considering the expression of a limited set of seven antigens. They identified five subsets based on the qualitative differences in antigen expression considering as positive those cases with more than 20% of blast cells, which reacted with the respective monoclonal antibody. The strength of our clustering analysis is ascribed to the quantitative analysis of a large number of antigens that allow us to distinguish clusters of AML patients with clinical impact. Clustering of patients with PML-RARα, AML1-ETO and CBFβ-MYH11 confirmed the potency of antigen expression pattern analysis to distinguish genotypically homogeneous profiles of acute leukemias as reported earlier.2, 8 Thus, the clustering method is a simple, fast and powerful tool for every laboratory in recognizing not only known features but also in finding new correlations among patients.


  1. 1

    De Zen L, Bicciato S, te Kronnie G, Basso G . Computational analysis of flow-cytometry antigen expression profiles in childhood acute lymphoblastic leukemia: an MLL/AF4 identification. Leukemia 2003; 17: 1557–1565.

  2. 2

    De Zen L, Orfao A, Cazzaniga G, Masiero L, Cocito MG, Spinelli M et al. Quantitative multiparametric immunophenotyping in acute lymphoblastic leukemia: correlation with specific genotype. I. ETV6/AML1 ALLs identification. Leukemia 2000; 14: 1225–1231.

  3. 3

    Frascella E, Rondelli R, Pigazzi M, Zampieron C, Fagioli F, Favre C et al. Clinical features of childhood acute myeloid leukaemia with specific gene rearrangements. Leukemia 2004; 18: 1427–1429.

  4. 4

    Yagi T, Morimoto A, Eguchi M, Hibi S, Sako M, Ishii E et al. Identification of a gene expression signature associated with pediatric AML prognosis. Blood 2003; 102: 1849–1856.

  5. 5

    Bullinger L, Dohner K, Bair E, Frohling S, Schlenk RF, Tibshirani R et al. Use of gene-expression profiling to identify prognostic subclasses in adult acute myeloid leukemia. N Engl J Med 2004; 350: 1605–1616.

  6. 6

    Valk PJ, Verhaak RG, Beijen MA, Erpelinck CA, Barjesteh van Waalwijk van Doorn-Khosrovani S, Beverloo HB et al. Prognostically useful gene-expression profiles in acute myeloid leukemia. N Engl J Med 2004; 350: 1617–1628.

  7. 7

    Casasnovas RO, Slimane FK, Garand R, Faure GC, Campos L, Deneys V et al. Immunological classification of acute myeloblastic leukemias: relevance to patient outcome. Leukemia 2003; 17: 515–527.

  8. 8

    Hrusak O, Porwit-MacDonald A . Antigen expression patterns reflecting genotype of acute leukemias. Leukemia 2002; 16: 1233–1258.

Download references

Author information

Correspondence to G Basso.

Rights and permissions

Reprints and Permissions

About this article

Further reading