Molecular stratification model for prognosis in cytogenetically normal acute myeloid leukemia

Patients with acute myeloid leukemia and normal cytogenetics (CN-AML) are usually categorized as an intermediate-risk group, with a 5-year survival rate varying between 24% and 42%. It is probable that differences in outcome reflect the molecular heterogeneity of CN-AML whose prognosis is influenced by several gene mutations or aberrant gene expression.^1,2 FLT3 internal tandem duplications (FLT3-ITD),^3-5 MLL partial tandem duplication (MLL-PTD),⁶  and overexpression of ERG,⁷ WT1,⁸  and MN1⁹  have been associated with a poor prognosis in CN-AML, whereas NPM1 gene mutations are associated with a favorable outcome.^10-12  Furthermore, in the intermediate- and high-risk karyotypic groups, EVI1 overexpression is associated with an adverse prognosis,¹³  whereas a high PRAME expression defines a good prognosis in several AML subtypes, especially those with favorable cytogenetic translocations.^14-16 

Although most studies in CN-AML patients have focused on one or 2 molecular markers, there is increasing evidence suggesting that possible outcomes based on single-gene abnormalities are hard to predict, and a more accurate prediction can be obtained by identifying risk categories based on the information provided by 2 or more parameters.¹  For this reason, we have simultaneously evaluated 9 new molecular markers in 121 CN-AML patients, showing that ERG, EVI1, and PRAME afford independent prognostic information and allow us to establish a simple score system for risk stratification.

We have analyzed pretreatment bone marrow samples from 121 adults diagnosed as novo CN-AML. All patients were treated according to the Spanish Program for the Study and Treatment of Malignant Hemopathies (PETHEMA) LAM-99 protocols.¹⁷  Ten patients (8.3%) died before they had reached complete remission (CR), 91 (75.2%) achieved CR with induction therapy, and 20 (16.5%) were refractory to the standard induction treatment. Nine patients from this latter group achieved CR after salvage therapy. Finally, 43 of the 100 patients who achieved CR eventually relapsed during the evaluation period. The median follow-up for censored patients was 26 months (range, 10-72 months). In addition, 10 bone marrow samples from healthy donors were processed as controls for gene-expression analysis. Informed consent to use biologic samples and clinical data were obtained in all cases in accordance with the Declaration of Helsinki. This study was approved by the Institutional Review Board of the University Hospital of Salamanca and the Scientific and Ethics Committee of the PETHEMA group.

Total RNA from diagnostic bone marrow and subsequent reverse transcription were performed according to the protocols approved by the Europe against Cancer Group program.¹⁸  All samples were analyzed for FLT3-ITD,⁵  mutations in NPM1,¹¹  and relative expression of the following genes: ABL1 (as control gene), ERG, EVI1, MLL-PTD, MN1, PRAME, RHAMM, and WT1, using the TaqMan Gene Expression Assays (Applied Biosystems, Foster City, CA). Relative quantification was calculated using the equation 2^−ΔΔCt, as previously described.¹⁶  The prognostic impact of the gene expression was evaluated using quartiles as cutoff points and selecting the one with the lowest P value.

All tests were carried out using the SPSS 15.0 program (SPSS). For univariate analyses, the Student t test was performed to evaluate refractoriness to treatment and gene-expression levels. The relapse-free survival (RFS) and overall survival (OS) were calculated using the Kaplan-Meier method. The impact of multiple predictor variables on RFS and OS was assessed by multivariate analysis according to the Cox regression model (forward conditional method), as described elsewhere.¹⁶ 

Patients with clinically adverse features, such as white blood cell (WBC) counts more than 50 × 10⁹/L and an age greater than 65 years, were associated with a poorer OS and RFS, whereas patients harboring a FLT3 wild-type (wt) and NPM1-mutated phenotype were associated with a better prognosis (Table 1). In addition, molecular markers with a clinical impact on OS were: ERG (50th percentile, P = .020), PRAME (75th percentile, P = .035), and EVI1 (75th percentile, P = .042). Similarly, the genes that showed significant influence on RFS were: ERG (P = .010), PRAME (P = .017), and EVI1 (P = .051). Interestingly, patients who were refractory to induction therapy showed higher ERG (1.0 ± 0.8 vs 0.6 ± 0.6; P = .01) and lower PRAME (29 ± 53 vs 1641 ± 6102; P = .01) levels compared with patients who achieved CR after the induction therapy.

Features selected in the multivariate analysis as having an independent prognostic value for a shorter OS were: WBC more than 50 × 10⁹/L (P < .001), age more than 65 years (P = .004), high ERG expression (P = .024), and high EVI1 expression (P = .030). In addition, patients with no FLT3wt/NPM1-mutated phenotype (P = .055) and a low PRAME expression (P = .066) showed a trend toward a poorer OS. Parameters with an independent prognostic value for a shorter RFS were: age more than 65 years (P = .006), high ERG expression (P = .014), low PRAME expression (P = .026), WBC more than 50 × 10⁹/L (P = .031), no FLT3wt/NPM1-mutated phenotype (P = .037), and high EVI1 expression (P = .050). Our data confirm the adverse prognostic influence that has been shown for ERG^7,19  and EVI1¹³  genes. Preliminary studies have suggested that PRAME overexpression is associated with a good prognosis in childhood AML, although this effect might be the result of its correlation with favorable cytogenetics, ie, t(8;21).¹⁵  Here we show, for the first time, that the prognostic value of PRAME up-regulation is independent of other karyotypic abnormalities because PRAME overexpression was associated with a better response to induction therapy and longer survival in our series, in which all patients had a normal cytogenetics.

Based on the results described, we investigated whether the combination of the ERG, EVI1, and PRAME markers could improve their individual prognostic significance. Thus, we drew up a molecular score by assigning a value of 1 point per gene expression associated with an adverse prognosis (high ERG, high EVI1, and low PRAME RNA levels). By contrast, a value of 0 was assigned to a favorable expression profile (low ERG or low EVI1 or high PRAME). This score allowed us to discriminate 4 different risk categories for both OS and RFS analysis, independently of other clinical-biologic features. The 2-year OS for scores 0, 1, 2, and 3 was 79%, 65%, 46%, and 27%, respectively (P = .001; Figure 1A). Moreover, the 2-year RFS for the same subgroups was 92%, 65%, 49%, and 43%, respectively (P = .005; Figure 1B). Similar results were observed when the analysis was restricted to the 97 patients younger than 65 years (Figure 1C-D). Multivariate analysis confirmed the findings in the complete series because the features selected as having an independent prognostic value for either a shorter OS or RFS were: the proposed molecular score (P < .001 and P < .001), WBC counts more than 50 × 10⁹/L (P = .002 and P = .04), and age more than 65 years (P = .007 and P = .005). Furthermore, the FLT3wt/NPM1-mutated phenotype displayed an independent prognostic value in the multivariate analysis for better RFS (P = .05) and a trend toward longer OS (P = .09).

A further benefit of the proposed score was the discrimination between different prognostic categories within those patients considered as having an intermediate-risk/high-risk based on the FLT3/NPM1 classification.^10-12  Thus, patients harboring FLT3wt/NPM1wt (n = 47) or FLT3-ITD/NPM1-mutated (n = 20) or FLT3-ITD/NPM1wt (n = 16) phenotype displayed differentiated OS (P = .001; Figure 1E) and RFS (P = .013; Figure 1F) according to score subgroup. It is worth noting that scores 0 and 1 showed that 43% of patients (36 of 83) had a good prognosis, which could be considered as redefining their risk category.

Our score system integrates 3 prognostic markers that could provide a more accurate stratification than single marker analysis^7,9,13 ; and, unlike wide gene-expression profiling,^14,20  it could be easily implemented in the context of routine clinical laboratories. Nevertheless, because a molecular score based on gene-expression levels could be less objective than mutation assessment,^2,21  this score system needs to be validated in an independent series of patients before it is incorporated into clinical practice.^20,22 

In conclusion, we propose a score based on ERG, EVI1, and PRAME gene expression that allows a greater distinction between CN-AML patients with significantly different outcomes.

The authors thank the Spanish cooperative group PETHEMA for providing the therapeutic protocol, samples, and clinical information.

This work was supported in part by the Spanish Fondo de Investigaciones Sanitarias de la Seguridad Social (grants PI061351 and 00/0023-00), Gerencia Regional de Salud (89/A/06), Junta Castilla y León, and Centro de Investigación del Cáncer, Instituto de Biologia Molecular y Celular del Cancer (Universidad de Salamanca-Consejo Superior de Investigaciones Cientificas).

A complete list of the PETHEMA group participants appears in the supplemental Appendix (available on the Blood website; see the Supplemental Materials link at the top of the online article).

Contribution: C.M.S. and M.C.C. carried out all molecular studies and prepared the database for the final analysis; C.M.S. performed the statistical analysis and prepared the initial version of the paper; R.G.-S. helped in the design of the work, reviewed the database, contributed toward the statistical analysis, and provided the preapproval of the final version; A. Balanzategui, M.E.S., and M.A. participated in the generation of the molecular results; C.P., M.D.C., F.R., A.G.d.C., J.M.A., P.G., T.B., J.A.Q., J.N.R., P.F.-A., A. Bárez, M.J.P., M.B.V., and J.D.-M. were the clinicians responsible for the patients and who ensured the application of protocol, sampling, and the collection of clinical data; J.F.S.M. and M.G. initially promoted the study, were responsible for securing financial support, were responsible for the group, and were responsible for the most important revision of the draft; and M.G. approved the final version to be sent to the editor.

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

Correspondence: Ramón García-Sanz, Department of Hematology, University Hospital of Salamanca, Paseo de San Vicente, 58-182, Salamanca, 37007 Spain; e-mail: rgarcias@usal.es.