High-frequency type I/II mutational shifts between diagnosis and relapse are associated with outcome in pediatric AML: implications for personalized medicine

Current treatment of acute myeloid leukemia (AML) is successful in almost 70% of pediatric patients. Although virtually all AML patients achieve a complete remission after initial treatment, a large fraction of patients (30%-40%) will have a relapse, and these patients have a dismal prognosis.^1,2  Factors that are related to prognosis in de novo pediatric AML are clinical parameters such as age and response to treatment. Moreover, biologic characteristics of AML, including cytogenetics,^3,4  minimal residual disease,^5,6  and a number of molecular aberrations,^7-11  also play an important role. According to the proposed “2 hit” model of leukemogenesis,¹²  2 types of molecular aberrations are discernible: (1) activating type I mutations, for example, in RAS or FLT3, thus conferring increased proliferative and survival capabilities on leukemic cells; (2) type II abnormalities, for example, chromosomal aberrations such as t(8;21) or mutations in the CEBPA gene, which cause a differentiation arrest and an increase in self-renewal properties. According to this model, both types of aberrations act in concert to initiate and drive AML progression.

Drugs that directly target aberrant proteins according to their corresponding type I/II mutations may therefore greatly improve prognosis. Kinase inhibitors such as lestaurtinib, midostaurin, or the more recently developed FLT3 kinase inhibitor AC220 were shown to be predominantly effective in mutated FLT3/ITD-positive AML samples in vitro.^13-15  In clinical trials with kinase inhibitors, for example, lestaurtinib¹⁶  or sunitinib,¹⁷  responses were observed particularly in patients harboring mutant FLT3. Ultimately, these targeted approaches may allow personalized chemotherapy of AML that could either eliminate subsets of leukemic cells that are resistant to conventional chemotherapy or eradicate minimal residual disease (MRD), thereby preventing the emergence of relapse. Various novel drugs are now being developed to specifically target leukemic cells according to their specific molecular aberrations.^18,19 

Apart from the use of type I/II aberrations as putative therapeutic targets, they may also be used as molecular markers for the detection of malignant cells. Type I/II aberrations or expression of the genes involved have been proposed in the literature as markers for the molecular assessment of MRD with the use of different techniques (eg, NPM1^20,21 , WT1^22,23 , t(8;21),²⁴  inv(16),²⁵  PML-RAR^26,27 ). For the use of the type I/II aberrations for leukemic cell detection and for subsequent targeted therapy, it will be essential to know the robustness of the mutation status during disease progression, that is, at diagnosis and relapse. Following the initial description²⁸  of cytogenetic changes from diagnosis to relapse, it has been shown by others²⁹  as well as by our own group³⁰  that FLT3/ITD aberrations are not persistently present in initial diagnosis and corresponding relapse AML samples. Such changes in molecular aberrations may theoretically result from outgrowth of otherwise undetectable rare clones present at diagnosis or from acquisition of de novo alterations during or after chemotherapy. These phenomena may lead not only to inaccuracies in malignant cell detection but also to suboptimal drug selection. Therefore, further research on the persistence of these aberrations after initial diagnosis is warranted.

In the current study we determined relevant type I/II molecular aberrations in paired diagnosis and relapse samples of patients with pediatric AML to determine their frequency and persistence during disease progression.

After informed consent was obtained in accordance with the Declaration of Helsinki, bone marrow or peripheral blood samples were collected from 69 paired patients with pediatric AML (n = 138 samples) at initial diagnosis and at relapse. Pediatric samples were obtained from the Dutch Childhood Oncology Group and the AML Berlin-Frankfurt-Münster Study Group. Approval from the Institutional Review Board of VU University Medical Center was obtained. This study is in part an extension of a previous reported study that used paired samples³⁰ : 34 pediatric patients from the previous study were also included in the current study. Mononuclear cells were isolated by standard Ficoll (1.077 g/mL; Amersham Biosciences) gradient centrifugation, and samples were enriched for leukemic blasts by eliminating nonleukemic cells as previously described.³¹  Central review of the diagnosis, classification by morphology or cytogenetics, and clinical follow-up of the patients were performed by both study groups. Percentages of blast cells were assessed by morphologic analysis of May-Grünwald-Giemsa–stained cytospin slides. Leukemic samples were routinely investigated for cytogenetic abnormalities by standard chromosome-banding analysis and screened for recurrent nonrandom genetic abnormalities, typical for AML. Applied screening methods for cytogenetic abnormalities included reverse transcription–polymerase chain reaction or fluorescent in situ hybridization or both. Cytogenetic data were determined for the initial sample only.

Patients were all treated with intensive cytarabine/anthracycline-based protocols in the Netherlands and Germany from 1992 until 2004 (protocols Dutch Childhood Oncology Group AML 87, 94, and 97 and AML–Berlin-Frankfurt-Münster 93, 98, and 04).^32-34 

Genomic DNA was isolated from cytospin slides, frozen cell pellets, or liquid nitrogen cryopreserved cells. Mutation analyses were performed in the VU University Medical Center, Amsterdam, and Erasmus Medical Center/Sophia Children's Hospital, Rotterdam, The Netherlands, with the use of sensitive high-throughput methods, which are summarized in Table 1. Mutations in NPM1 exon 12 were analyzed by PCR on genomic DNA in a multiplex reaction with FLT3/ITD analysis. PCR amplification was performed with the following primers: NPM1 forward, 5′-TTAACTCTCTGGT-GGTAGAATGA-3′, and NPM1 reverse. 5′-CTGACCACCGCTACTACTATGT-3′, located in intron 11 and exon 12, respectively. Subsequent fragment analysis was performed with a tetrachlorofluorescein phosphoramidite–labeled (Biolegio) forward primer. Mutations detected with melting curve analysis were confirmed with the use of 2 methods: (1) reamplification of the exon and repeated melting curve analysis or (2) bidirectional DNA sequencing on an ABI 3100 automated sequencer with the use of the BigDye terminator kit (Applied Biosystems Inc).

To compare categorical variables we used χ² analysis and Fisher exact test in the case of small numbers. Nonparametric Mann-Whitney U test, Spearman rank correlation, and analysis of variance were applied to assess differences in the distribution of continuous variables. To assess outcome, the following parameters were used: complete remission (defined as < 5% blasts in a bone marrow aspirate and no evidence of leukemia at any other site and hematologic recovery according to Cancer and Leukemia Group B criteria, relapse (defined by classical morphologic criteria when containing > 5% blasts), overall survival (OS; defined as the period of time in months between the date of diagnosis and of death from any cause or last follow-up), and the time to relapse (TTR; defined as the number of months between the date of initial diagnosis and of relapse). OS was estimated by the Kaplan-Meier method, and different groups were compared with the log-rank test. Prognostic factors were examined by multivariate Cox regression analysis. P values ≤ .05 were considered statistically significant (2-tailed testing). SPSS statistical software Version 17.0 for Windows (SPSS Inc) was used.

We successfully obtained genomic DNA from paired initial diagnosis and corresponding relapse samples from 69 patients with pediatric AML. Not all patients were evaluable for all mutations because of limited amounts of genomic DNA. Patient characteristics are depicted in Table 2. The sex of the majority of the study cohort was male (70%). Patients were equally distributed over different treatment protocols, and no significant difference in OS was found between the protocols (data not shown). Because of the selection of patients with a relatively poor outcome (ie, only for those patients who developed a relapse), the results of patient characteristics and type I/II mutation analysis at diagnosis, might be expected to differ from what has been described in literature for newly diagnosed pediatric AML. Indeed, we found a higher frequency of FLT3/ITD (20.8%), RAS (23.8%), and WT1 (17.8%) mutations compared with previous results.^7,8,35,41  However, the frequency of NPM1 mutations (7.2%) at initial diagnosis was comparable with the incidence described previously.⁹  The coincidence of these mutations with each other also differed; for example, NPM1 mutations coincided less frequently (1 of 67 analyzed cases, 1.5%) with FLT3 mutations compared with the 40% coincidence that was previously described.⁹  Similarly, the numbers of patients with favorable and intermediate cytogenetic aberrations were lower than previously described for unselected patients.⁴²  The selected group is representative for patients with relapsed AML in terms of TTR rates, because approximately one-half of the relapses (57%) were early relapses and occurred within 1 year from diagnosis.¹  The 2-year OS rate was 36%, which is lower compared with all patients with newly diagnosed pediatric AML treated with similar protocols and in line with previous reports on outcome of children with relapsed AML.⁴³ 

Of the 69 patients studied, 42 patients (61%) had a mutation identified either at diagnosis or at relapse. Table 3 depicts the frequency of mutations measured both at initial diagnosis and at relapse. Overall, the frequency of mutations did not differ significantly between diagnosis and relapse, except for WT1 which had a nearly 2-fold increased incidence of mutations at relapse. Mutations were not mutually exclusive in these patients with relapsed pediatric AML. In particular, FLT3/ITDs coincided with NPM1, CEBPA, RAS, and WT1. Overall, 18 of 42 patients (43%) who had a mutation at any stage of the disease harbored more than one different mutation. An overview of all mutation data for each patient is listed in supplemental Table 1 (available on the Blood Web site; see the Supplemental Materials link at the top of the online article).

The observed frequency of mutations was distinct in the cytogenetic subgroups. For instance, FLT3/ITD, WT1, RAS, and NPM1 mutations were predominantly found in the group of cytogenetically normal (CN) patients with AML and the subgroup of “other cytogenetics.” Occasional cases of RAS mutations were observed in mixed-lineage leukemia (MLL), AML-ETO1, and core binding factor AML. Supplemental Figure 1 summarizes the observed incidence of mutations per cytogenetic subgroup at diagnosis.

To assess the retention of mutations during disease progression, we compared the mutation data obtained at diagnosis and their matched relapse pairs. We observed mutational shifts between diagnosis and relapse for the currently analyzed genes in 26 of 69 patients (38%), which constitutes 61% of the 42 samples harboring a mutation at some stage of disease (Table 4; supplemental Table 1). This analysis shows frequent mutational shifts in FLT3/ITD, RAS, and WT1. In FLT3 and RAS, both gains and losses of mutations occurred between diagnosis and relapse. Both types of mutations were more frequently lost than gained. In WT1, however, mutations accumulated at relapse because these mutations were only gained and none were lost. Occasional changes were also found in other targets such as CEBPA and FLT3 D835. The frequencies of mutational shifts per gene of interest are given in Table 4. When stratified according to cytogenetic subgroups, mutational shifts occurred more frequently in patients wit AML with CN-AML or MLL rearrangements (54% and 80% of analyzed cases, respectively), compared with other cytogenetic subgroups (supplemental Table 2).

The mutational shifts (Table 4) strongly influenced the associations of mutational status with TTR. The patients who were positive for FLT3/ITD or RAS or WT1 mutations at initial diagnosis showed shorter TTR; however, this did not reach statistical significance (Table 5). The lack of a clear effect of these mutations on TTR could be anticipated because, by definition, our patient subgroup lacks the nonrelapsing good prognosis group. However, when TTR was determined after stratification of groups on the basis of mutation status at relapse, the presence of FLT3/ITD, RAS, or WT1 mutations was associated with a significant or a strong trend (P = .05, P = .12, and P = .06, respectively) for a shorter TTR: 16.5 and 7.7 months for FLT3 wild-type and ITD-positive patients, respectively, as well as 17.4 and 9.4 months for wild-type RAS and patients harboring mutant RAS, respectively, and 17.6 and 8.9 months for wild-type WT1 and patients with WT1 mutations (Table 5). Of this patient group, clinical data were available to perform OS analysis, and we hence stratified the patients according to mutation status either at diagnosis or at relapse. In this patient group, the OS at 2 years was 36%, hence reflecting again the selection of a patient subgroup with poor prognosis. Our results show that the presence of a FLT3 or WT1 mutation at relapse was a more significant predictive factor for an adverse outcome of patients with AML compared with the presence of these mutations at diagnosis (Figure 1). The persistence or shifts of mutational status for individual genes was too low to assess the effect on OS of the mutational shifts in these genes. However, when data were combined for genes with frequent mutational shifts, a significant correlation between shifts or persistence of mutations and OS was found (P = .039; supplemental Figure 2). Multivariate OS analysis, including FLT3, RAS, or WT1 mutations, either at diagnosis or relapse, and age, white blood cell count, French-American-British type and sex showed that only the negative association of FLT3/ITD mutations at relapse with OS remained significant (P = .001).

In the current study on a large cohort of paired diagnosis and relapse pediatric AML samples (n = 69), we examined the incidence and retention of established type I/II molecular aberrations at diagnosis and at relapse. We focused our study to 7 genes and found substantial differences in the mutation status within the predominant leukemic populations at initial diagnosis and at relapse in 38% of the patients. The actual number of patients who experience such differences may even be higher, considering that not all mutations could be determined in all paired samples, because of limited availability of genomic DNA. Moreover, the observed frequency of mutational shifts may be an underestimate, because many other mutated genes that are potentially involved in leukemogenesis or disease progression were not studied here. Therefore, the extent of the alterations between initial diagnosis and relapse remains to be elucidated. The observed mutational shifts occurred most frequently in CN-AML and MLL rearranged AML (54% and 80% of cases, respectively), which suggests that these AML types are genetically less stable at these gene loci. Mutational shifts were more frequently observed for those genes for which mutations at diagnosis are associated with a dismal prognosis (FLT3, WT1) or which are presumed to be involved in leukemogenesis (eg, RAS). The instability of FLT3 and WT1 mutations between diagnosis and relapse is in concordance with previous publications in which FLT3/ITD mutations were shown to be gained or lost during disease progression, whereas WT1 mutations were only gained.⁸  Given the effects on patient outcome, these changes in FLT3/ITD mutation status and particularly WT1 mutations may resemble the leukemia-driving capacity of these genes (ie, their type I properties). KIT or CEBPA genes that are also implicated in leukemogenesis showed less mutational shifts, which can be explained by the low incidence of mutations in these genes. In the current study we observed a loss of NPM1 mutation in one case; although NPM1 is often regarded as a robust and stable molecular marker,^9,21,44  similar losses of NPM1 mutations between diagnosis and relapse were observed in adults by other groups.^45,46  To the best of our knowledge, other mutational shifts, including in RAS, have not previously been described.

It is, therefore, tempting to speculate on the origin of these mutational shifts. The latter are consistent with the immunophenotypic shifts^47-49  or cytogenetic changes²⁸  that are also observed in AML between diagnosis and relapse. Given the sensitivity of the methods used for the detection of mutations and considering the high blast percentages in the analyzed samples, it is highly unlikely that we have failed to detect mutations that were present in the bulk of leukemic blasts at either stage of the disease. One explanation for the observed shifts may be the presence of small numbers of leukemic clones with a distinct molecular make-up that differs from the bulk of leukemic blasts at initial diagnosis (ie, oligoclonality). These chemotherapy drug–resistant clones appear to initiate the relapse in a posttherapy bone marrow on the basis of their superior tumor-initiating properties. This plausible explanation is substantiated by findings of Pollard et al⁵⁰  that at diagnosis the immature AML cells are predominantly heterogeneous in their FLT3/ITD mutation status. In particular, the presence of the mutation in this immature subpopulation was associated with a drug-resistant disease. In patients, the selective pressure of induction chemotherapy may hence result in the survival of these leukemia-initiating cells, which expand and give rise to the relapsed blast population. This proposed model is in line with the “leukemic stem cell” model and explains both gains and losses of mutations. The other possible explanation could be that AML cells are genetically unstable and acquire appropriate mutations or chromosomal aberrations necessary for survival during chemotherapy, thereby resulting in clonal expansion and relapse. In this model, complex DNA recombination events, in which loss of heterozygosity occurs, are required to explain the loss of mutations and reversion to the wild-type genotype; clearly, both possibilities remain to be explored experimentally.

In this study we confirm the mutational shifts in FLT3/ITDs in paired samples and provide similar results for other molecular aberrations. In addition, within this extended patient group with relevant clinical data, we could show an effect on outcome for patients with mutational shifts in WT1 or RAS. The clinical relevance of our observations is evident and may further have direct implications for the patients involved, because genetic imbalances may affect the use of molecular markers for MRD assessment as well as the nature and timing of future targeted therapies. Currently, molecular diagnostics are often performed only at the time of definitive diagnosis, and patients are stratified into risk groups accordingly. If subsequent changes in mutational status occur between diagnosis and relapse, the strategy to treat such patients at relapse may not be properly tailored to their actual risks, and patients might be overtreated or undertreated. This will become particularly important when the molecular aberrations will also be used for selecting targeted drugs against the aberrant proteins of these mutated genes. Moreover, other aberrations, apart from mutations, may represent future targets for treatment, and these may also vary between initial diagnosis and relapse. Studies at multiple biologic levels in larger series of consecutive patient samples could elucidate the extent of mutational shifts or other genetic alterations, including aberrant gene expression or DNA methylation status.

The results of this study clearly indicate that mutation analysis at diagnosis alone is not sufficient for proper risk stratification, MRD detection, and optimal selection of personalized treatment. We hence propose that risk stratification, malignant cell detection, and selection of personalized treatment should be based also on type I/II mutation status at both initial diagnosis and relapse.

This work was supported by grants from the Dutch Cancer Society (VU 2005-3666) (J.C.), Children Cancer-free (Y.G.A.), the Netherlands Organization for Scientific Research (Y.G.A.), and The Netherlands Academy of Arts and Sciences (Y.G.A.).

Contribution: C.B. performed experiments, analyzed data, and wrote the paper; G.J.S. and J.C. supervised the research and edited the paper; I.H.I.M.H. provided experimental data; Z.J.K. performed experiments and analyzed data; B.F.G. and G.J.L.K. designed the research and edited the paper; C.M.Z. and M.M.v.d.H.-E. provided clinical and experimental data and edited the paper; E.S.J.M.d.B. provided experimental data; V.d.H., D.R., and U.C. provided samples and clinical data; and Y.G.A. discussed the research and edited the paper.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Jacqueline Cloos, Department of Pediatric Oncology/Hematology, VU University Medical Center, PO Box 7057, 1007 MB Amsterdam, The Netherlands; e-mail: j.cloos@vumc.nl.