Article Tools
ORIGINAL REPORTS
Hematologic Malignancies
Article Tools
ERG Expression Is an Independent Prognostic Factor and Allows Refined Risk Stratification in Cytogenetically Normal Acute Myeloid Leukemia: A Comprehensive Analysis of ERG, MN1, and BAALC Transcript Levels Using Oligonucleotide Microarrays
C.B. and S.K.B. contributed equally to this study.
Abstract
Recently, several novel molecular prognostic markers were identified in cytogenetically normal acute myeloid leukemia (CN-AML). In addition to the well-known influence of FLT3, NPM1, and CEBPA mutations, high transcript levels of the ERG, BAALC, and MN1 genes have been associated with inferior outcomes, but the relative importance of these risk markers remains to be defined.
We analyzed ERG, BAALC, and MN1 expression levels in a cohort of 210 patients with CN-AML who received intensive chemotherapy. Expression levels of ERG, BAALC, and MN1 were determined in bone marrow samples by using oligonucleotide microarrays.
High transcript levels of ERG, BAALC, and MN1 were predictors for inferior overall survival (OS) and a lower rate of complete remissions (CRs). There were significant positive correlations between the expression levels of all three genes. ERG expression levels predicted OS in elderly patients (ie, age 60 years or older) with CN-AML (P = .006) as well as in younger patients (P = .013). In multivariate analyses, high ERG expression was independently associated with a lower CR rate (P = .013), shorter event-free survival (P = .008), and shorter OS (P = .005). Patients who had low ERG levels and absent FLT3 internal tandem duplication (ITD) had a 5-year OS of 44%, and patients who had high ERG expression and FLT3 ITD had a 5-year OS of only 5%.
Approximately 45% of patients with acute myeloid leukemia (AML) have a normal karyotype without chromosomal aberrations on standard cytogenetic analysis,1 and this subgroup is heterogeneous with regard to molecular genetic alterations and therapeutic outcomes.2 Submicroscopic genetic lesions not only are relevant for the pathogenesis of the disease but also influence response to therapy and survival.2,3 The most frequent molecular abnormalities in cytogenetically normal AML (CN-AML) are mutations in the nucleophosmin 1 (NPM1) gene and internal tandem duplications (ITDs) of the fms-like tyrosine kinase 3 gene (FLT3). Mutations in the tyrosine kinase domain (TKD) of FLT3 occur more rarely, and there are conflicting data on their prognostic relevance.4,5 Other molecular aberrations found in smaller proportions of patients with CN-AML include mutations in the CEBPA gene, which convey a favorable prognosis, and partial tandem duplications (PTDs) of the MLL gene.2,6–8
More recently, quantitative differences in the expression levels of several genes (BAALC,9–12 ERG,13,14 MN1,15,16 WT1,17,18 and EVI119), measured by quantitative real-time polymerase chain reaction (qPCR), have been shown to carry prognostic information in patients with CN-AML. However, information on the interactions and relative importance of these risk factors is sparse. In this study, we used data from oligonucleotide microarrays to study the transcript levels of prognostically relevant genes in a well-characterized population of patients with CN-AML. Our approach allowed us to investigate correlations between various molecular markers and to identify those that carry independent prognostic information.
We studied a cohort of 210 patients with previously untreated CN-AML who received intensive induction and consolidation chemotherapy between 1999 and 2004. Two hundred (95.2%) of these patients were enrolled on the AMLCG-1999 multicenter trial of the German AML Cooperative Group,20 and 10 patients were enrolled on the preceding AMLCG-1992 trial21 (details in Data Supplement, online only). The study protocols were approved by the ethics committees of the participating centers, and all patients provided written informed consent. All patients had normal karyotypes on conventional cytogenetic examination of at least 20 metaphases. Patients were characterized at the molecular level with regard to FLT3 ITD, MLL PTD, NPM1, CEBPA, and FLT3 TKD (D835) mutations, as described previously.22–24
Pretreatment bone marrow samples were analyzed by using Affymetrix HG-U133A (n = 154) or HG-U133 Plus 2.0 (n = 56) oligonucleotide microarrays (Affymetrix, Santa Clara, CA). Details regarding sample preparation, hybridization, and image acquisition have been described previously.25,26 For combining individual oligonucleotide probes to probe sets and for the annotation to genes, we used custom chip definition files that were based on the GeneAnnot database.27 Thereby, we ensured that each gene was represented by a single set of oligonucleotide probes and that only one single expression value was obtained per gene (details in Data Supplement, online only).
We studied genes for which an association between transcript levels and prognosis of patients with CN-AML had been reported in the literature (BAALC, ERG, and MN1), and tested their associations with OS and event-free survival (EFS) in univariate Cox models. To delineate distinct patient subgroups on the basis of BAALC and MN1 expression, patients with expression values greater than the median of all samples were classified as having high BAALC or MN1 expression.9–11,15,16 ERG overexpression was defined as a transcript level greater than the 75th percentile of all measurements.13
Survival curves were generated by the Kaplan-Meier method, and P values were calculated by the log-rank test.28 Multivariate Cox proportional hazard regression models were constructed for OS, EFS, and relapse-free survival (RFS), and factors that predicted the chance to reach complete remission (CR) were analyzed in a logistic regression model (details in Data Supplement, online only). All analyses were performed by using the R 2.7.2 software package (available at www.r-project.org).29
Baseline clinical and molecular characteristics of the 210 patients included on this study are detailed in Table 1. Overall, 188 patients (89.5%) were fully characterized for all five mutations considered in this analysis (FLT3 ITD, NPM1, CEBPA, MLL PTD, and FLT3 TKD). The frequencies and combinations of genetic aberrations among these 188 patients are displayed in Appendix Figure A1 (online only). The median OS for the total cohort was 12.7 months, and the median follow-up for survivors was 46 months (range, 2 to 92 months).
|
| Variable | Patient Data in Total Cohort (N = 210) | Patient Data by Expression Level | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| BAALC Expression | ERG Expression | MN1 Expression | ||||||||
| Low (n = 105) | High (n = 105) | P | Low (n = 157) | High (n = 53) | P | Low (n = 105) | High (n = 105) | P | ||
| Female sex | .07 | .006 | .017 | |||||||
| No. | 122 | 71 | 51 | 100 | 22 | 70 | 52 | |||
| % | 58 | 65 | 51 | 64 | 42 | 66 | 50 | |||
| Age, years | ||||||||||
| Median | 59 | 57 | 61 | .77 | 57 | 62 | .37 | 56 | 61 | .05 |
| Range | 17-83 | 17-80 | 18-83 | 17-83 | 18-78 | 17-78 | 18-83 | |||
| WHO performance status, No.* | .72 | .58 | .82 | |||||||
| 0 | 46 | 21 | 25 | 37 | 9 | 23 | 23 | |||
| 1 | 89 | 47 | 42 | 69 | 20 | 46 | 43 | |||
| 2, 3, or 4 | 64 | 32 | 32 | 46 | 18 | 31 | 33 | |||
| AML type, No. | .86 | .68 | .50 | |||||||
| De novo AML | 200 | 101 | 99 | 150 | 50 | 100 | 100 | |||
| sAML | 8 | 3 | 5 | 6 | 2 | 3 | 5 | |||
| tAML | 2 | 1 | 1 | 1 | 1 | 2 | 0 | |||
| FAB classification, No. | < .001 | < .001 | .001 | |||||||
| M0 | 6 | 0 | 6 | 2 | 4 | 0 | 6 | |||
| M1 | 57 | 22 | 35 | 33 | 24 | 18 | 39 | |||
| M2 | 65 | 27 | 38 | 53 | 12 | 35 | 30 | |||
| M4 | 46 | 28 | 18 | 35 | 11 | 27 | 19 | |||
| M5 | 25 | 21 | 4 | 25 | 0 | 18 | 7 | |||
| M6 | 9 | 6 | 3 | 7 | 2 | 6 | 3 | |||
| MDS-RAEB | 2 | 1 | 1 | 2 | 0 | 1 | 1 | |||
| Leukocyte count, ×109/L | .17 | .039 | .12 | |||||||
| Median | 31.8 | 33.7 | 26.0 | 26.3 | 41.9 | 33.7 | 24.0 | |||
| Range | 0.85-486 | 0.85-486 | 1.0-440 | 0.90-486 | 0.85-287 | 0.75-486 | 1.0-289 | |||
| Hemoglobin level, g/L | .30 | .80 | .63 | |||||||
| Median | 92 | 90 | 93 | 92 | 91 | 92 | 93 | |||
| Range | 40-147 | 40-142 | 42-147 | 40-147 | 42-131 | 40-142 | 49-147 | |||
| Platelet count, ×109/L | .20 | .14 | .12 | |||||||
| Median | 59 | 60 | 51 | 60 | 48 | 61 | 51 | |||
| Range | 6-471 | 9-280 | 6-471 | 9-471 | 6-367 | 10-471 | 6-367 | |||
| Bone marrow blasts, % | ||||||||||
| Median | 85 | 82 | 85 | .67 | 80 | 86 | .006 | 80 | 90 | .01 |
| Range | 11-100 | 17-100 | 11-100 | 11-100 | 20-98 | 17-100 | 11-100 | |||
| FLT3 ITD status | ||||||||||
| FLT3 ITD negative | .58 | .015 | .58 | |||||||
| No. | 123 | 64 | 59 | 100 | 23 | 64 | 59 | |||
| % | 59 | |||||||||
| FLT3 ITD positive | ||||||||||
| No. | 87 | 41 | 46 | 57 | 30 | 41 | 46 | |||
| % | 41 | |||||||||
| NPM1 status | ||||||||||
| NPM1 wild type | < .001 | .63 | < .001 | |||||||
| No. | 96 | 23 | 73 | 70 | 26 | 31 | 65 | |||
| % | 46 | |||||||||
| NPM1 mutated | ||||||||||
| No. | 114 | 82 | 32 | 87 | 27 | 74 | 40 | |||
| % | 54 | |||||||||
| CEBPA status† | ||||||||||
| CEBPA wild type | .001 | .58 | .005 | |||||||
| No. | 175 | 94 | 81 | 129 | 46 | 91 | 84 | |||
| % | 91 | |||||||||
| CEBPA mutated | 2 | 16 | 12 | 6 | 3 | 15 | ||||
| No. | 18 | |||||||||
| % | 9 | |||||||||
| MLL PTD status‡ | ||||||||||
| MLL PTD negative | .011 | .81 | .40 | |||||||
| No. | 178 | 97 | 81 | 131 | 47 | 92 | 86 | |||
| % | 87 | |||||||||
| MLL PTD positive | ||||||||||
| No. | 26 | 7 | 19 | 20 | 6 | 11 | 15 | |||
| % | 13 | |||||||||
| FLT3 D835 status§ | ||||||||||
| FLT3 D835 wild type | .21 | .25 | .13 | |||||||
| No. | 192 | 93 | 99 | 141 | 51 | 93 | 99 | |||
| % | 92 | |||||||||
| FTL3 D835 mutated | ||||||||||
| No. | 17 | 11 | 6 | 15 | 2 | 12 | 5 | |||
| % | 8 | |||||||||
NOTE. All 210 patients were characterized with regard to FLT3 ITD and NPM1 mutations.
Abbreviations: AML, acute myeloid leukemia; sAML, secondary AML after preceding myelodysplastic syndrome; tAML, therapy-related AML after previous chemotherapy or radiotherapy; FAB, French-American-British classification; MDS-RAEB, myelodysplastic syndrome–refractory anemia with excess of blasts; ITD, internal tandem duplication; PTD, partial tandem duplication.
*Information on performance status was missing for 11 patients.
†The CEBPA status was known for 193 patients.
‡The MLL PTD status was known for 204 patients.
§Information on FLT3 tyrosine kinase domain (D835) status was available for 209 patients. P values were calculated by using Fisher's exact test and the Mann-Whitney U test, as appropriate.
In an initial step, we tested the association of ERG, BAALC, and MN1 transcript levels, determined by oligonucleotide microarray measurements and treated as continuous variables, with patient survival. By using univariate Cox proportional hazards models, we found that higher transcript levels of ERG, BAALC, and MN1 were significantly associated with shorter EFS, and high BAALC and ERG levels also correlated with shorter OS (Appendix Table A1, online only). Importantly, we also found that the transcript levels of ERG, BAALC, and MN1 showed significant positive pairwise correlations. This correlation was particularly strong between BAALC and MN1 (Spearman ρ = 0.77; P < .0001), although it was less marked between BAALC and ERG (ρ = 0.36; P < .0001) and between ERG and MN1 (ρ = 0.27; P = .0001; Appendix Fig A2, online only). Consequently, the dichotomized variables that indicated high versus low BAALC, ERG, and MN1 expression also showed strong pairwise correlations (Appendix Table A2, online only).
On the basis of the microarray-measured transcript levels of BAALC, ERG, and MN1, we defined patient subgroups with high and low expression of each gene. For BAALC and MN1, this dichotomization was performed at the median of all expression values, the threshold used in most previous reports.9–11,15,16 High ERG expression was defined as a transcript level greater than the 75th percentile. Different cut points for ERG have been used in earlier studies,13,14 but the 75th percentile was chosen after inspection of survival curves within quantiles of ERG expression (Appendix Fig A3, online only).
Table 1 lists the clinical characteristics of the entire cohort and of patient subgroups distinguished by high versus low expression of BAALC, ERG, and MN1. Patients with high transcript levels of ERG, BAALC, or MN1 frequently had AML with immature (ie, French-American-British classification [FAB] FAB M0 or M1) cytomorphology, whereas those with low expression of these genes more often showed monocytic differentiation (ie, FAB M5). High ERG expression was also associated with higher leukocyte and bone marrow blast counts, and high expression of all three genes was more frequent among men. High expression of BAALC was strongly correlated with wild-type NPM1 and the presence of MLL PTD. Interestingly, 16 (89%) of 18 patients with CEBPA mutations had high BAALC levels. High MN1 expression was also associated with wild-type NPM1 and mutated CEBPA, whereas high ERG expression correlated with the presence of an FLT3 ITD.
The patient subgroups with high expression of ERG, BAALC, or MN1 had lower CR rates and shorter EFS and OS than patients with low expression of these genes (Table 2; Fig 1). Furthermore, high transcript levels of ERG and MN1 were associated with shorter RFS among patients who reached CR (Appendix Fig A4, online only). High ERG expression was the marker with the strongest impact on survival. In the entire cohort of patients, with a median age of 59 years, patients with high and low ERG transcript levels had median OS times of 7.7 and 18.4 months, respectively (P < .001) and estimated 5-year OS rates of 17% and 36%, respectively. The effect of high ERG levels was similar in patients age 60 years or older (5-year OS, 11% v 28%; P = .006) and in younger patients (5-year OS, 27% v 43%; P = .013; Fig 2). Similar results were obtained if surviving patients who had been observed for less than 24 months were excluded (Data Supplement, online only).
|
| Outcome | BAALC Expression | ERG Expression | MN1 Expression | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Low (n = 105) | High (n = 105) | P | Low (n = 157) | High (n = 53) | P | Low (n = 105) | High (n = 105) | P | |
| CR | |||||||||
| No. of patients | 75 | 57 | .022 | 108 | 25 | .008 | 75 | 58 | .022 |
| Rate, % | 71 | 55 | 69 | 47 | 71 | 55 | |||
| RFS | |||||||||
| Median, months | 16.8 | 10.8 | .08 | 14.6 | 6.1 | .022 | 14.6 | 8.8 | .038 |
| Rate at 5 years, % | 42 | 24 | 37 | 23 | 42 | 23 | |||
| 95% CI, % | 31 to 55 | 15 to 38 | 28 to 47 | 11 to 48 | 32 to 55 | 14 to 38 | |||
| EFS | |||||||||
| Median, months | 8.4 | 4.1 | .007 | 9.0 | 2.3 | < .001 | 9.5 | 3.9 | .002 |
| Rate at 5 years, % | 28 | 13 | 24 | 11 | 29 | 13 | |||
| 95% CI, % | 21 to 39 | 8 to 22 | 18 to 32 | 5.1 to 24 | 21 to 39 | 8 to 22 | |||
| OS | |||||||||
| Median, months | 15.5 | 9.5 | .049 | 18.4 | 7.7 | < .001 | 17.7 | 9.5 | .047 |
| Rate at 5 years, % | 39 | 24 | 36 | 17 | 39 | 24 | |||
| 95% CI, % | 30 to 50 | 16 to 35 | 29 to 45 | 9.5 to 33 | 30 to 50 | 17 to 35 | |||
Abbreviations: CR, complete remission; RFS, relapse-free survival; EFS, event-free survival; OS, overall survival.
Fig 1. High expression of BAALC, ERG, or MN1 predicts shorter overall survival (OS) in cytogenetically normal acute myeloid leukemia. Kaplan-Meier plots show the OS of patient subgroups with high versus low transcript levels of (A) BAALC, (B) ERG, and (C) MN1.
Fig 2. Prognostic value of ERG expression levels in younger and elderly patients with cytogenetically normal acute myeloid leukemia. High ERG expression levels predicted shorter overall survival in patients (A) younger than 60 years and (B) 60 years and older.
Given the correlations we observed between various molecular risk markers, we performed multivariate analyses to identify those factors that independently predicted prognosis in CN-AML (Table 3). High expression levels of ERG predicted a lower likelihood of reaching CR (P = .013) after analysis adjustment for NPM1 status. Among patients who reached CR, the presence of an NPM1 mutation without concurrent FLT3 ITD (NPM1-positive/FLT3 ITD-negative genotype) was the only independent predictor of RFS. In a multivariate model for EFS, high ERG expression was a significant prognostic factor (P = .008) together with age and the NPM1-positive/FLT3 ITD-negative genotype. Patients with high ERG expression also had shorter OS (P = .005) than those with low expression after analysis adjustment for age, FLT3 ITD, and platelet count at diagnosis. Neither BAALC nor MN1 expression levels exhibited independent prognostic value in any of these multivariate analyses.
|
| Variable by Clinical Outcome | Analysis | |||
|---|---|---|---|---|
| HR | OR | 95% CI | P | |
| CR | ||||
| ERG expression, highest quartile | 0.43 | 0.22 to 0.84 | .013 | |
| NPM1 mutation | 2.16 | 1.17 to 3.96 | .013 | |
| RFS | ||||
| NPM1-positive/FLT3 ITD-negative genotype | 0.30 | 0.17 to 0.54 | < .001 | |
| EFS | ||||
| ERG expression, highest quartile | 1.65 | 1.14 to 2.39 | .008 | |
| NPM1-positive/FLT3 ITD-negative genotype | 0.41 | 0.27 to 0.62 | < .001 | |
| Age per 10-year increase | 1.23 | 1.08 to 1.39 | .002 | |
| OS | ||||
| ERG expression (highest quartile) | 1.82 | 1.20 to 2.75 | .005 | |
| FLT3 ITD* | 2.16 | 1.48 to 3.21 | < .001 | |
| Age per 10-year increase | 1.26 | 1.10 to 1.45 | .001 | |
| Thrombocytes per increase of 50,000/μL | 0.86 | 0.75 to 0.99 | .034 | |
NOTE. Only the 188 patients with complete information on FLT3 ITD and tyrosine kinase domain mutation, MLL partial tandem duplication, and NPM1 and CEBPA mutational status were considered in these analyses. The following variables had a P < .2 in univariate models and were included in the initial multivariate analyses: For overall survival, age, sex, performance status, de novo v secondary AML, pretreatment leukocyte and thrombocyte counts, FLT3 ITD, NPM1, FLT3 tyrosine kinase domain, CEBPA mutations, and BAALC and ERG expression levels; for EFS, age, sex, performance status, pretreatment leukocyte and thrombocyte counts, the FLT3 ITD-negative/NPM1-positive genotype, MLL partial tandem duplication, and BAALC, ERG, and MN1 expression levels; for RFS, age, performance status, pretreatment hemoglobin level and leukocyte count, the FLT3 ITD-negative/NPM1-positive genotype, and BAALC, ERG, and MN1 expression levels; for CR, age, sex, performance status, de novo v secondary AML, pretreatment thrombocyte count, FLT3 ITD, MLL partial tandem duplication, NPM1 mutations, and BAALC, ERG, and MN1 expression levels. A stepwise backward variable selection technique was used, so that variables remaining in the final models were significant at α < .05.
Abbreviations: CN-AML, cytogenetically normal acute myeloid leukemia; HR, hazard ratio; OR, odds ratio; CR, complete remission; RFS, relapse-free survival; ITD, internal tandem duplication; EFS, event-free survival; OS, overall survival.
*In the model for overall survival, FLT3 ITD violated the proportional hazards assumption. Therefore, FLT3 ITD was included in the model together with a time-dependent covariate. The hazard ratio for FLT3 ITD is given for the time point 12 months after diagnosis, and the associated P value refers to a Wald test with two degrees of freedom for FLT3 ITD and the time-dependent covariate.
High ERG expression and FLT3 ITD were both independent predictors of inferior OS, and we observed no statistically significant interaction between the effects of both factors on survival. Thus, we investigated patient stratification by using a combination of ERG transcript level and FLT3 ITD status (Fig 3). Patients with low ERG expression in the absence of a FLT3 ITD constituted a favorable subset of patients with CN-AML who had a median OS of 36 months, whereas the median OS for the three other groups combined was 8.5 months (P < .001). For patients with low ERG levels and absent FLT3 ITD, the estimated 5-year OS was 44% and the RFS at 5 years was 43%. In contrast, the outcomes of patients with both high ERG expression and FLT3 ITD were dismal, as the 5-year OS and RFS rates were only 5% and 9%, respectively. These differences were observed both in patients age younger than 60 years and those who were age 60 years or older (Appendix Fig A5, online only).
Fig 3. Risk stratification of patients with cytogenetically normal acute myeloid leukemia by a combination of FLT3 internal tandem duplication (ITD) and ERG expression. (A) Overall survival and (B) relapse-free survival are shown for the four subgroups defined by FLT3 ITD and ERG status. P values are given for the overall comparison across all four groups.
Patients with isolated NPM1 mutations in the absence of FLT3 ITDs seem to form a distinct, prognostically favorable subgroup of patients with AML; therefore, several study groups classify patients with CN-AML into low molecular risk (NPM1-mutant/FLT3 ITD-negative) and high molecular risk (NPM1 wild-type or FLT3 ITD) categories.7,30 We investigated whether ERG expression levels are useful to refine the risk stratification for patients with CN-AML who were already classified according to NPM1 and FLT3 ITD status. Patients in the high molecular risk category who also had high ERG expression had significantly worse OS than those with low ERG levels (estimated 5-year OS, 13% v 27%; P = .003; Appendix Fig A6B, online only). The majority (48 of 57) of patients in the low molecular risk category also had low ERG levels. However, among the nine patients with high ERG expression, there were five deaths within the first 8 months (Appendix Fig A6A; P = .10).
Currently, the prediction of treatment outcomes and selection of treatment strategies for patients with CN-AML are frequently based on FLT3 ITD and NPM1 status, the most extensively characterized molecular prognostic markers. In recent years, however, the situation has become more complicated because of the description of several novel prognostic factors, including quantitative measurements of ERG, BAALC, and MN1 transcript levels. Although the individual prognostic relevance of each of these risk factors has been well documented, it is unclear how molecular markers are best combined to allow for an improved prognostic stratification of patients with CN-AML. The aim of this study was to jointly evaluate a comprehensive set of risk markers in a relatively large and well-characterized patient cohort to delineate which markers are most informative for predicting a patient's risk for treatment failure, relapse, and death.
In our cohort of 210 patients, we found that transcript levels of the ERG, BAALC, and MN1 genes correlated with treatment outcomes in univariate analyses. Our findings extend the results of earlier studies9–16 in several important aspects: This report, to our knowledge, is the largest study so far that simultaneously evaluated expression levels of all three genes. Therefore, we had the possibility to explore correlations between the expression levels of these genes as well as with other molecular risk factors. We found that the expression levels of MN1, BAALC, and (to a lesser degree) ERG showed significant positive correlations with each other. Consistent with these results, MN1 was among the most strongly upregulated genes in patients with high BAALC expression in a previous microarray-based analysis,11 and an association between high BAALC and high ERG expression was described in two earlier studies.11,14 Taken together, these data indicate that the prognostic information conveyed by BAALC, MN1, and ERG transcript levels may be partially overlapping and redundant. This underlines that multivariate analyses that include all three genes are necessary to determine their relative importance as prognostic markers.
Most of the patients on this study were fully characterized with regard to FLT3 ITD, FLT3 TKD, MLL PTD, NPM1, and CEBPA mutations, which thus allowed consideration of a comprehensive panel of molecular and clinical prognostic factors. In multivariate analyses, ERG expression levels emerged as one of the strongest predictors for treatment outcomes in CN-AML. ERG is a transcription factor expressed in hematopoetic stem cells and is involved in early lymphocyte differentiation31 and angiogenesis.32 Of note, the ERG gene is also involved in a recurring chromosomal translocation found in approximately 50% to 60% of all prostate adenocarcinomas,33 which leads to an aberrantly high expression of ERG. High ERG transcript levels have been associated with worse outcomes, not only in AML but also in prostate cancer and T-cell acute lymphoblastic leukemia.34,35 In this study, patients with CN-AML who had high ERG expression were less likely to respond to induction chemotherapy, as indicated by the lower CR rate, but their early death rate (within 28 days) was similar to patients with low ERG expression (Data Supplement, online only). High ERG levels, thus, primarily seem to predict resistance to induction chemotherapy and early treatment failure. The lower CR rate possibly explains the association of high ERG transcript levels with shorter EFS and OS. When the analysis was adjusted for other risk factors, no influence of high ERG transcript levels on RFS was observed. These results are in accordance with previous studies in younger patients.13,14 Additional research is needed to identify the mechanisms that determine ERG expression levels in CN-AML and the signaling pathways that are responsible for the prognostic differences associated with varying ERG levels.
Both ERG expression and FLT3 ITD were independent predictors of OS in this study. Therefore, we investigated whether improved risk stratification in CN-AML can be achieved by combining these two markers. Indeed, the combination of wild-type FLT3 and low ERG expression identified a subset of patients who had relatively good outcomes. At the other end of the spectrum, patients with both an FLT3 ITD and high ERG transcript levels had extremely poor prognoses, comparable to those with a complex aberrant karyotype. Although detailed information on salvage treatments was not available, the median OS of 7.5 months and a 5-year OS rate of only 5% suggest that most diseases with these traits are refractory to current therapies. Patients with this combination of risk markers apparently do not profit from intensive chemotherapy, with the possible exception of those who are candidates for allogeneic transplantation.
Almost all previous studies on the prognostic relevance of ERG, BAALC, and MN1 expression levels exclusively focused on patients with CN-AML who were younger than age 60 years. Only one study that evaluated the prognostic impact of BAALC expression levels12 also included elderly patients. In our cohort, the median age was considerably higher than in the previous study by Bienz et al,12 and nearly half of all patients were age 60 years or older. We did not observe a significant interaction between age and the prognostic impact of ERG, BAALC, or MN1 expression levels, which indicated that these genes are useful prognostic markers in older patients. For example, for patients who were age 60 years or older and who fell into the FLT3 ITD-negative group with low ERG levels, the 5-year OS rate was 34%, which indicated that some of these elderly patients might benefit from intensive chemotherapy.
ERG expression levels also provided additional prognostic information for patients who were already stratified according to their NPM1 and FLT3 status. On the basis of findings in this study and results from other recent studies,7,14 we propose a refined algorithm for risk stratification in CN-AML (Appendix Fig A7A, online only) that could be evaluated in future clinical trials. According to our results, patients with an FLT3 ITD and high ERG levels should be classified as high risk. Because their outcomes resemble those of patients with a complex aberrant karyotype, they may be candidates for allogeneic transplantation or experimental therapies (Appendix Fig A7B). In contrast, patients with NPM1 mutations, low ERG levels, and absent FLT3 ITD, as well as those with mutant CEBPA (Appendix Fig A8, online only), can be considered as low risk.7,8,14 In this cohort, the majority of these low-risk patients were cured with intensive chemotherapy. All remaining patients fell into an intermediate-risk category, and the optimum treatment strategy for these patients still needs to be defined.
All previous studies on the prognostic impact of BAALC, ERG, and MN1 expression were performed by using qPCR,9–15 whereas we used expression measurements from oligonucleotide microarrays. Although microarray measurements tend to underestimate differences in gene expression compared with qPCR, there is generally a good correlation between expression ratios measured by qPCR and by those obtained from oligonucleotide microarrays.11,36,37 Of note, our analysis relied on relative expression differences between samples and not on absolute expression levels, and our results were remarkably consistent with those of previous, PCR-based studies. Nonetheless, before quantitative measures of ERG expression can be used for clinical decision making, additional standardization of the methods used to determine ERG expression levels in combination with prospective trials is necessary.
In summary, this work is, to our knowledge, the largest study so far that simultaneously analyzed ERG, BAALC, and MN1 expression levels in a uniformly treated patient population. Several important conclusions can be drawn from this study: We found that the expression levels of ERG, BAALC, and MN1 are strongly correlated, which suggests that their prognostic significance may be overlapping; this necessitates a comprehensive analysis of molecular risk factors in CN-AML. Our patient cohort was well characterized for the most relevant molecular aberrations and gene expression changes. In multivariate models, high ERG expression emerged as a strong negative prognostic marker that independently predicted a poor response to induction chemotherapy and shorter survival. Unlike most previous studies, our analysis also included patients who were older than 60 years of age, and this study is the first to show that high ERG expression is a negative prognostic factor in elderly patients with CN-AML. We demonstrated that combining novel risk markers, such as high ERG expression, with established prognostic factors results in refined prognostic subclassification of CN-AML.
Written on behalf of the German AML Cooperative Group (AMLCG).
Supported by Grant No. NGFN2 01GS0448 from the German Ministry of Education and Research (Bundesministerium für Bildung und Forschung) via the National Genome Research Network (to S.K.B and C.B.).
Authors' disclosures of potential conflicts of interest and author contributions are found at the end of this article.
The author(s) indicated no potential conflicts of interest.
Conception and design: Klaus H. Metzeler, Christian Buske, Stefan K. Bohlander
Financial support: Ulrich Mansmann, Wolfgang Hiddemann, Christian Buske, Stefan K. Bohlander
Administrative support: Maria-Cristina Sauerland, Achim Heinecke
Provision of study materials or patients: Annika Dufour, Tobias Benthaus, Wolfgang E. Berdel, Thomas Büchner, Bernhard Wörmann, Jan Braess, Karsten Spiekermann, Wolfgang Hiddemann
Collection and assembly of data: Klaus H. Metzeler, Annika Dufour, Tobias Benthaus, Manuela Hummel, Maria-Cristina Sauerland, Achim Heinecke, Ulrich Mansmann
Data analysis and interpretation: Klaus H. Metzeler, Manuela Hummel, Ulrich Mansmann, Jan Braess, Karsten Spiekermann, Christian Buske, Stefan K. Bohlander
Manuscript writing: Klaus H. Metzeler, Christian Buske, Stefan K. Bohlander
Final approval of manuscript: Klaus H. Metzeler, Annika Dufour, Tobias Benthaus, Manuela Hummel, Maria-Cristina Sauerland, Achim Heinecke, Wolfgang E. Berdel, Thomas Büchner, Bernhard Wörmann, Ulrich Mansmann, Jan Braess, Karsten Spiekermann, Wolfgang Hiddemann, Christian Buske, Stefan K. Bohlander
| 1. | K Mrozek, NA Heerema, CD Bloomfield: Cytogenetics in acute leukemia Blood Rev 18: 115– 136,2004 Crossref, Medline, Google Scholar |
| 2. | K Mrozek, G Marcucci, P Paschka , etal: Clinical relevance of mutations and gene-expression changes in adult acute myeloid leukemia with normal cytogenetics: Are we ready for a prognostically prioritized molecular classification? Blood 109: 431– 448,2007 Crossref, Medline, Google Scholar |
| 3. | CD Baldus, K Mrozek, G Marcucci , etal: Clinical outcome of de novo acute myeloid leukaemia patients with normal cytogenetics is affected by molecular genetic alterations: A concise review Br J Haematol 137: 387– 400,2007 Crossref, Medline, Google Scholar |
| 4. | AJ Mead, DC Linch, RK Hills , etal: FLT3 tyrosine kinase domain mutations are biologically distinct from and have a significantly more favorable prognosis than FLT3 internal tandem duplications in patients with acute myeloid leukemia Blood 110: 1262– 1270,2007 Crossref, Medline, Google Scholar |
| 5. | U Bacher, C Haferlach, W Kern , etal: Prognostic relevance of FLT3-TKD mutations in AML: The combination matters—An analysis of 3082 patients Blood 111: 2527– 2537,2008 Crossref, Medline, Google Scholar |
| 6. | SP Whitman, AS Ruppert, MD Radmacher , etal: FLT3 D835/I836 mutations are associated with poor disease-free survival and a distinct gene-expression signature among younger adults with de novo cytogenetically normal acute myeloid leukemia lacking FLT3 internal tandem duplications Blood 111: 1552– 1559,2008 Crossref, Medline, Google Scholar |
| 7. | RF Schlenk, K Dohner, J Krauter , etal: Mutations and treatment outcome in cytogenetically normal acute myeloid leukemia N Engl J Med 358: 1909– 1918,2008 Crossref, Medline, Google Scholar |
| 8. | G Marcucci, K Maharry, MD Radmacher , etal: Prognostic significance of, and gene and microrna expression signatures associated with, CEBPA mutations in cytogenetically normal acute myeloid leukemia with high-risk molecular features: A Cancer and Leukemia Group B study J Clin Oncol 26: 5078– 5087,2008 Link, Google Scholar |
| 9. | CD Baldus, SM Tanner, AS Ruppert , etal: BAALC expression predicts clinical outcome of de novo acute myeloid leukemia patients with normal cytogenetics: A Cancer and Leukemia Group B study Blood 102: 1613– 1618,2003 Crossref, Medline, Google Scholar |
| 10. | CD Baldus, C Thiede, S Soucek , etal: BAALC expression and FLT3 internal tandem duplication mutations in acute myeloid leukemia patients with normal cytogenetics: prognostic implications J Clin Oncol 24: 790– 797,2006 Link, Google Scholar |
| 11. | C Langer, MD Radmacher, AS Ruppert , etal: High BAALC expression associates with other molecular prognostic markers, poor outcome and a distinct gene-expression signature in cytogenetically normal acute myeloid leukemia: A Cancer and Leukemia Group B (CALGB) study Blood 111: 5371– 5379,2008 Crossref, Medline, Google Scholar |
| 12. | M Bienz, M Ludwig, BU Mueller , etal: Risk assessment in patients with acute myeloid leukemia and a normal karyotype Clin Cancer Res 11: 1416– 1424,2005 Crossref, Medline, Google Scholar |
| 13. | G Marcucci, CD Baldus, AS Ruppert , etal: Overexpression of the ETS-related gene, ERG, predicts a worse outcome in acute myeloid leukemia with normal karyotype: A Cancer and Leukemia Group B study J Clin Oncol 23: 9234– 9242,2005 Link, Google Scholar |
| 14. | G Marcucci, K Maharry, SP Whitman , etal: High expression levels of the ETS-related gene, ERG, predict adverse outcome and improve molecular risk-based classification of cytogenetically normal acute myeloid leukemia: A Cancer and Leukemia Group B study J Clin Oncol 25: 3337– 3343,2007 Link, Google Scholar |
| 15. | M Heuser, G Beutel, J Krauter , etal: High meningioma 1 (MN1) expression as a predictor for poor outcome in acute myeloid leukemia with normal cytogenetics Blood 108: 3898– 3905,2006 Crossref, Medline, Google Scholar |
| 16. | C Langer, K Maharry, K Mrozek , etal: Low Meningioma 1 (MN1) gene expression to predict outcome in cytogenetically normal acute myeloid leukemia (CN-AML): A Cancer and Leukemia Group B (CALGB) study J Clin Oncol 26: 374s,2008 suppl abstr 7011 Link, Google Scholar |
| 17. | E Barragan, J Cervera, P Bolufer , etal: Prognostic implications of Wilms' tumor gene (WT1) expression in patients with de novo acute myeloid leukemia Haematologica 89: 926– 933,2004 Medline, Google Scholar |
| 18. | M Weisser, W Kern, S Rauhut , etal: Prognostic impact of RT-PCR-based quantification of WT1 gene expression during MRD monitoring of acute myeloid leukemia Leukemia 19: 1416– 1423,2005 Crossref, Medline, Google Scholar |
| 19. | S Barjesteh van Waalwijk van Doorn-Khosrovani, C Erpelinck, WL van Putten , etal: High EVI1 expression predicts poor survival in acute myeloid leukemia: A study of 319 de novo AML patients Blood 101: 837– 845,2003 Crossref, Medline, Google Scholar |
| 20. | T Buchner, WE Berdel, C Schoch , etal: Double induction containing either two courses or one course of high-dose cytarabine plus mitoxantrone and postremission therapy by either autologous stem-cell transplantation or by prolonged maintenance for acute myeloid leukemia J Clin Oncol 24: 2480– 2489,2006 Link, Google Scholar |
| 21. | T Büchner, W Hiddemann, WE Berdel , etal: 6-thioguanine, cytarabine, and daunorubicin (TAD) and high-dose cytarabine and mitoxantrone (HAM) for induction, TAD for consolidation, and either prolonged maintenance by reduced monthly TAD or TAD-HAM-TAD and one course of intensive consolidation by sequential HAM in adult patients at all ages with de novo acute myeloid leukemia (AML): A randomized trial of the German AML Cooperative Group J Clin Oncol 21: 4496– 4504,2003 Link, Google Scholar |
| 22. | S Schnittger, C Schoch, W Kern , etal: Nucleophosmin gene mutations are predictors of favorable prognosis in acute myelogenous leukemia with a normal karyotype Blood 106: 3733– 3739,2005 Crossref, Medline, Google Scholar |
| 23. | S Schnittger, U Kinkelin, C Schoch , etal: Screening for MLL tandem duplication in 387 unselected patients with AML identify a prognostically unfavorable subset of AML Leukemia 14: 796– 804,2000 Crossref, Medline, Google Scholar |
| 24. | T Benthaus, F Schneider, G Mellert , etal: Rapid and sensitive screening for CEBPA mutations in acute myeloid leukaemia Br J Haematol 143: 230– 239,2008 Crossref, Medline, Google Scholar |
| 25. | T Haferlach, A Kohlmann, S Schnittger , etal: Global approach to the diagnosis of leukemia using gene expression profiling Blood 106: 1189– 1198,2005 Crossref, Medline, Google Scholar |
| 26. | C Schoch, A Kohlmann, S Schnittger , etal: Acute myeloid leukemias with reciprocal rearrangements can be distinguished by specific gene expression profiles Proc Natl Acad Sci U S A 99: 10008– 10013,2002 Crossref, Medline, Google Scholar |
| 27. | F Ferrari, S Bortoluzzi, A Coppe , etal: Novel definition files for human GeneChips based on GeneAnnot BMC Bioinformatics 8: 446,2007 Crossref, Medline, Google Scholar |
| 28. | BD Cheson, PA Cassileth, DR Head , etal: Report of the National Cancer Institute–sponsored workshop on definitions of diagnosis and response in acute myeloid leukemia J Clin Oncol 8: 813– 819,1990 Link, Google Scholar |
| 29. | R: A Language and Environment for Statistical Computing 2006 R Development Core Team Vienna, Austria R Foundation for Statistical Computing Google Scholar |
| 30. | G Marcucci, MD Radmacher, K Maharry , etal: MicroRNA expression in cytogenetically normal acute myeloid leukemia N Engl J Med 358: 1919– 1928,2008 Crossref, Medline, Google Scholar |
| 31. | M Anderson, G Hernandez-Hoyos, R Diamond , etal: Precise developmental regulation of ETS family transcription factors during specification and commitment to the T cell lineage Development 126: 3131– 3148,1999 Medline, Google Scholar |
| 32. | GM Birdsey, NH Dryden, V Amsellem , etal: Transcription factor ERG regulates angiogenesis and endothelial apoptosis through VE-cadherin Blood 111: 3498– 3506,2008 Crossref, Medline, Google Scholar |
| 33. | SA Tomlins, DR Rhodes, S Perner , etal: Recurrent fusion of TMPRSS2 and ETS transcription factor genes in prostate cancer Science 310: 644– 648,2005 Crossref, Medline, Google Scholar |
| 34. | CD Baldus, P Martus, T Burmeister , etal: Low ERG and BAALC expression identifies a new subgroup of adult acute t-lymphoblastic leukemia with a highly favorable outcome J Clin Oncol 25: 3739– 3745,2007 Link, Google Scholar |
| 35. | J Wang, Y Cai, C Ren , etal: Expression of Variant TMPRSS2/ERG fusion messenger RNAs is associated with aggressive prostate cancer Cancer Res 66: 8347– 8351,2006 Crossref, Medline, Google Scholar |
| 36. | S Draghici, P Khatri, AC Eklund , etal: Reliability and reproducibility issues in DNA microarray measurements Trends Genet 22: 101– 109,2006 Crossref, Medline, Google Scholar |
| 37. | PB Dallas, NG Gottardo, MJ Firth , etal: Gene expression levels assessed by oligonucleotide microarray analysis and quantitative real-time RT-PCR: How well do they correlate? BMC Genomics 6: 59,2005 Crossref, Medline, Google Scholar |
Before treatment started, patients in the AMLCG-1999 trial (n = 200) were randomly assigned upfront to one of two double-induction therapies (thioguanine, cytarabine, and daunorubicin [TAD] –high-dose cytarabine and mitoxantrone [HAM] or HAM-HAM). The TAD regimen for induction consisted of cytarabine 100 mg/m2/d by continuous intravenous (IV) infusion on days 1 and 2 and via 30-minute IV infusions every 12 hours on days 3 to 8; daunorubicin 60 mg/m2 via 60-minute IV infusions on days 3, 4, and 5; and thioguanine 100 mg/m2 orally every 12 hours on days 3 to 9. The HAM induction regimen combined cytarabine 3 g/m2 (in patients younger than 60 years) or 1 g/m2 (in patients ≥ 60 years) via 3-hour IV infusions every 12 hours on days 1 to 3, with mitoxantrone 10 mg/m2 via 60-minute IV infusions on days 3 to 5. One week after the first induction course, a bone marrow aspirate was examined. A second induction course was administered to all patients younger than 60 years, and to patients 60 years or older if 5% or more residual blasts appeared in their bone marrow. After achieving complete remission by bone marrow and peripheral blood criteria (Cheson BD et al, J Clin Oncol 21:4642-4649, 2003), all patients received consolidation with one course identical to the TAD induction regimen. After TAD consolidation, patients younger than 60 years received postremission therapy by either prolonged maintenance or by autologous stem-cell transplantation, whereas older patients received maintenance chemotherapy. Details on maintenance treatment have been published before.20
Six patients from the AMLCG-1999 trial were randomly assigned to the common intergroup arm of the German AML study groups and received induction treatment according to a standard 7 + 3 regimen of cytarabine and daunorubicin.
Ten patients who were treated in the AMCG-1992 trial received induction treatment according to the TAD-HAM regimen and TAD consolidation, as described above (Buchner T et al, J Clin Oncol 21:4496-4504, 2003). Overall, 106 patients were assigned to TAD-HAM induction, 97 patients were randomly assigned to the HAM-HAM regimen, and six patients received 7 + 3 induction therapy.
Analyses of gene expression data from Affymetrix (Santa Clara, CA) oligonucleotide microarrays are complicated by the huge progress in genome sequence annotation since the array layout was designed originally. This results in the following problem: when standard Affymetrix probe set definitions are used, one given gene can be represented by multiple microarray probe sets. This often leads to discordant expression signals when differentially expressed genes are analyzed. Moreover, each Affymetrix probe set is a combination of 11 different oligonucleotide probes, and some of these probes may show poor specificity or sensitivity (ie, probes may anneal with > one transcript or match no transcript at all; Harbig J et al, Nucleic Acids Res 33:e31, 2005). We circumvented these difficulties through the use of custom probe annotations that were based on the GeneAnnot database (Chalifa-Caspi V et al, Bioinformatics 20:1457-1458, 2004). This database contains detailed, up-to-date information on sequence alignments between individual 25-mer probes and transcripts, and it provides an assessment of sensitivity and specificity for each probe-transcript pair. On the basis of this information, Ferrari et al27 have recombined the individual oligonucleotide probes into novel probe sets, so that all probes that specifically match one gene were combined into one single probe set. Thereby, they eliminated the problems created by having more than one probe set per gene. Moreover, these GeneAnnot-based custom probe sets include only probes matching transcripts that are linked to a single gene. As such, they preserve a one-to-one correspondence between genes and probe sets, thus avoiding additional noise caused by the use of a probe in multiple probe sets. These novel probe set definitions are contained in custom chip definition files (CDFs), which are publicly available at http://www.xlab.unimo.it/GA_CDF/. In summary, by using GeneAnnot-based CDFs instead of standard Affymetrix CDFs, we were able to obtain one single expression measurement for each gene of interest.
Data normalization was performed separately for HG-U133A and HG-U133 Plus 2.0 microarrays (Affymetrix) by using the variance stabilizing normalization algorithm (Huber W et al, Bioinformatics 18 Suppl 1:S96-S104, 2002), and expression values were calculated by the median polish method. After selection of probe sets common to both microarray types, the data from HG-U133A and HG-U133 Plus 2.0 arrays were merged by using the median rank score (MRS) method to ensure numerically comparable measurements of gene expression (Warnat P et al, BMC Bioinformatics 6:265, 2005). The larger HG-U133A data set served as the reference set. Box plots of the expression values were assessed before and after MRS adjustment to ensure that the procedure resulted in similar value distributions for both types of arrays (Appendix Fig A9, online only).
All clinical end points were defined according to the criteria proposed by Cheson et al (J Clin Oncol 21:4642-4649, 2003). Specifically, overall survival (OS) was measured from the date of entry onto study until death as a result of any cause, and patients who were not known to have died were censored at the time of last follow-up. Event-free survival (EFS) was defined as time from the date of study entry until treatment failure, relapse, or death as a result of any cause, whichever occurred first. Relapse-free survival (RFS) was defined as time from reaching complete remission (CR) until the date of acute myeloid leukemia (AML) relapse or death as a result of any cause, whichever occurred first.
In all multivariate analyses, only patients with complete information on FLT3 internal tandem duplication (ITD) and tyrosine kinase domain (TKD) mutations, NPM1, and CEBPA mutational status and MLL partial tandem duplication (PTD) were included. The following variables were first evaluated in univariate models: Age; sex; leukocyte count; hemoglobin level; thrombocyte count; bone marrow blast percentage; de novo versus secondary AML; presence of FLT3 ITD and TKD (D835) mutations; NPM1; MLL PTD; CEBPA status; and expression levels of BAALC, ERG, and MN1 (dichotomized into high v low expression). Variables with a univariate P < .2 were included in the initial multivariate models, and a stepwise backward variable selection procedure was applied. Variables that remained in the final model were significant at α < .05. The Cox proportional hazards models were stratified according to microarray type (HG-U133A v HG-U133 Plus 2.0). As an alternative to considering NPM1 mutations and FLT3 ITD separately, models were also analyzed with a variable that indicated the presence of the favorable NPM1-positive/FLT3 ITD-negative (low molecular risk) genotype, and the model with the better overall fit was chosen. Because the strong correlation between MN1 and BAALC expression levels might have caused problems as a result of collinearity, we also tested models were either only MN1 or only BAALC were considered. The results were similar to the models with both variables included, and BAALC or MN1 expression never entered the final multivariate model. The proportional hazards assumption was tested for each variable individually by inspection of scaled Schoenfeld residual plots and via the Grambsch-Therneau test (Grambsch and Therneau, Biometrika 81: 515-526, 1984). If a variable violated the proportional hazards assumption, it was included in the multivariate model together with a time-dependent covariate.
The 200 patients from the AMLCG-1999 study that were analyzed for BAALC, ERG, and MN1 expression levels represent a subset of a total of 1,354 patients with a normal karyotype treated on that protocol so far. The clinical outcome of the 200 patients who were included in our analyses was similar to the outcome of the 1,154 patients for whom no data on quantitative gene expression was available: CR rates were 63.3% for patients included in our analyses versus 63.6% for those who were not included (P = .94); the estimated 5-year EFS was 20.5% versus 17.3%, respectively (log-rank P = .70); the estimated 5-year RFS was 32.2% versus 27.6%, respectively (P = .86), and the 5-year OS was 31.1% versus 27.2%, respectively (P = .64). These results indicate that our patient cohort was representative of patients with CN-AML who were included in the AMLCG-1999 study.
Similar to the results reported for the entire AMLCG-1999 study cohort,20 there were no significant differences in CR rates, RFS, EFS, or OS between patients assigned to HAM-HAM or TAD-HAM induction (data not shown). There also was no evidence of an interaction between the type of induction regimen and the dichotomized ERG expression level in the analyses of OS, EFS, or RFS. Likewise, there was no interaction between the type of microarray used for the analysis (HG-U133 Plus 2.0 v HG-U133A) and the effects of ERG expression on OS, EFS, or RFS.
Of the 210 patients included in this analysis, 11 (5.2%) were lost to follow-up after less than 24 months of observation. The frequency of high ERG, BAALC, or MN1 expression levels was not different between patients who were censored in the first 24 months and those with longer follow-up. Patients who were lost to follow-up within 24 months had lower median hemoglobin levels (P = .04) and thrombocyte counts (P = .03) than those with longer follow-up, but all other baseline characteristics (including the frequencies of NPM1, FLT3, MLL, and CEBPA mutations) were similar between both groups.
When the 11 patients with early censoring were excluded from the cohort, the major results of our analyses remained unchanged. For example, in univariate analyses, patients with high ERG expression had a median OS of 15.9 months versus 7.6 months for patients with low ERG levels (P < .001), and the corresponding estimates of 5-year OS were 35% and 16%, respectively. In a multivariate model, the hazard ratio for death for patients with high ERG expression was 1.80 (95% CI, 1.19 to 2.73; P = .005) after adjustment for FLT3 ITD status, thrombocyte count, and age, exclusion of these 11 patients.
Because our analyses showed that high ERG expression independently predicted a lower likelihood of attaining CR, we tested whether high ERG transcript levels were associated with an increased risk of early death (defined as death as a result of any cause within 21 days from study inclusion). The early death rates were 7.5% (four of 53 patients) for patients with high ERG levels and 7.0% (11/157 patients.) for those with low ERG expression (P = 1.0). If early death was defined as death until day 28, the corresponding frequencies were 11.3% (six of 53 patients) and 7.6% (12 of 157 patients) for patients with high and low ERG expression, respectively (P = .40).
In a multivariate model, the only factors that independently predicted early death (ie, death that occurred before day 28) were higher age (odds ratio per 10-year increase, 2.48; 95% CI, 1.39 to 4.45; P = .002) and a higher initial leukocyte count (odds ratio per increase of 10,000/μL, 1.08; 95% CI, 1.02 to 1.15; P = .005), but not ERG expression levels or pretreatment performance status. Moreover, the WHO performance status of patients with high ERG expression was similar to those with low ERG expression (P = .58). However, it should be noted that only 14 patients in the entire cohort had a WHO performance status of 3 or 4.
Two recent, large studies by Marcucci et al8 and Schlenk et al7 have highlighted mutations in the CEBPA gene as an important favorable prognostic factor in CN-AML. In our cohort, patients with CEBPA mutations exhibited a trend towards longer OS (median, 40 months v 10 months for mutant v wild-type CEBPA; estimated 5-year OS, 45% and 28%, respectively). This difference did not reach statistical significance (P = .08), possibly because of the small number of patients with mutant CEBPA (n = 18, with nine deaths). In a multivariate model, after adjustment for FLT3 ITD, ERG expression, thrombocyte count, and age, mutant CEBPA was associated with a hazard ratio for death of 0.61 (95% CI, 0.30 to 1.21; P = .16). When we compared patients who had mutant CEBPA with those who had low ERG expression and absent FLT3 ITD, the survival curves for both groups were similar (Appendix Fig A8). These results, together with the studies cited above, provided the rationale to include mutant CEBPA as a positive prognostic factor in the prognostic algorithm proposed in Appendix Figure 7A.
|
| GeneAnnot Probe Set by Gene | OS | EFS | ||||
|---|---|---|---|---|---|---|
| HR | 95% CI | P | HR | 95% CI | P | |
| ERG | ||||||
| GC21M038675_at | 1.60 | 1.26 to 2.04 | .0001 | 1.64 | 1.32 to 2.04 | < .0001 |
| BAALC | ||||||
| GC08P104222_at | 1.18 | 1.02 to 1.39 | .031 | 1.30 | 1.13 to 1.49 | .0002 |
| MN1 | ||||||
| GC22M026468_at | 1.10 | 0.96 to 1.27 | .16 | 1.24 | 1.10 to 1.39 | .0004 |
NOTE. The associations of the transcript levels of the three genes, analyzed as numeric continuous variables, with patient survival were evaluated by using univariate Cox models. Hazard ratios refer to an expression difference equal to the interquartile range of all samples.
Abbreviations: OS, overall survival; EFS, event-free survival; HR, hazard ratio.
|
| Expression Variable | Expression Variable | |||||
|---|---|---|---|---|---|---|
| BAALC Expression | ERG Expression | |||||
| Low (n = 105) | High (n = 105) | P | Low (n = 157) | High (n = 53) | P | |
| MN1 expression | ||||||
| Low (n = 105) | 82 | 23 | < .001 | 91 | 14 | < .001 |
| High (n = 105) | 23 | 82 | 66 | 39 | ||
| ERG expression | ||||||
| Low (n = 157) | 90 | 67 | < .001 | |||
| High (n = 53) | 15 | 38 | ||||
Fig A1. Frequencies and combinations of molecular genetic aberrations in 188 patients with cytogenetically normal acute myeloid leukemia. In this analysis, only patients for whom complete information on FLT3 internal tandem duplication (ITD) and tyrosine kinase domain mutations, MLL partial tandem duplication (PTD), and NPM1 and CEBPA status was available were included. Each tick mark at the top of the figure indicates one patient; in each line, patients who were positive for the respective mutation are marked in black. Overall, 26 patients (14%) had none of the five aberrations, 90 patients (48%) had one aberration, 66 (35%) had two aberrations, and six patients (3%) had three aberrations. An MLL PTD was never detected in combination with an NPM1 or CEBPA mutation.
Fig A2. Correlations between BAALC, ERG, and MN1 transcript levels. There was a particularly strong correlation between the transcript levels of (A) BAALC and MN1. The correlations between ERG expression and (B) BAALC and (C) MN1 levels were less strong but still significant. P values were calculated by using the Spearman rank correlation method.
Fig A3. Survival according to percentiles of ERG expression. To test whether a cut point at the 75th percentile was appropriate to define high- and low-risk patient subgroups on the basis of ERG expression levels, survival curves were generated for patient cohorts with ERG expression levels less than the median, between the 50th and 75th percentiles, between the 75th and 90th percentiles, and greater than the 90th percentile.
Fig A4. Kaplan-Meier analysis of relapse-free survival (RFS) according to (A) BAALC, (B) ERG, and (C) MN1 expression levels. High ERG or MN1 expression levels, but not high transcript levels of BAALC, predicted shorter RFS among patients who reached complete response.
Fig A5. Risk stratification of patients with cytogenetically normal acute myeloid leukemia by a combination of FLT3 internal tandem duplication (ITD) and ERG expression, stratified according to age. (A) Kaplan-Meier analysis of overall survival in the subgroup of patients who were younger than 60 years (n = 109). Patients were divided into four subgroups on the basis of FLT3 ITD and ERG status. (B) The same analysis was performed in older patients (ie, those ≥ 60 years; n = 101). P values are given for the overall comparisons across all four groups.
Fig A6. ERG expression levels provide additional prognostic information in patients with known NPM1 and FLT3 internal tandem duplication (ITD) status. (A) Patients with mutated NPM1 and absent FLT3 ITD represent a distinct low molecular risk subgroup of patients with cytogenetically normal acute myeloid leukemia. Most of these patients had low ERG transcript levels; however, among the few patients with high ERG levels, there was a trend toward shorter overall survival. (B) Patients who had FLT3 ITD and/or wild-type NPM1 can be additionally stratified by analyzing ERG transcript levels.
Fig A7. Proposed algorithm for risk stratification in cytogenetically normal acute myeloid leukemia (CN-AML). (A) Classification flowchart based on our own data and on results from other recent studies (Schlenk et al7; Marcucci et al8; and Marcucci et al14); the flowchart is intended to serve as a basis for additional discussions and prospective studies. The CEBPA mutation status was included in this scheme in response to the results of Schlenk et al7 and because of a trend toward better overall survival for patients with mutant CEBPA that we observed in our own data (Data Supplement, online only). When the patient cohort described in this manuscript was stratified according to the suggested risk group definitions, (B) three subgroups of CN-AML with remarkably different clinical outcomes could be distinguished. This provides an example of how improved risk stratification in CN-AML can be achieved by a structured combination of old and novel molecular markers. mut, mutation; wt, wild type.
Fig A8. Effect of CEBPA mutations on survival. The Kaplan-Meier plot displays the overall survival of patients with a CEBPA mutation (irrespective of ERG and FLT3 internal tandem duplication [ITD] status), of those with low ERG levels and absent FLT3 ITD, and of all remaining patients. The survival of patients with CEBPA mutations is comparable to those who have low ERG expression levels and no FLT3 ITD.
Fig A9. Microarray-derived expression values before and after cross-platform normalization. These box-and-whisker plots visualize the median expression values; 25th and 75th percentiles; and maximum and minimum values for BAALC, ERG, and MN1 expression levels that were obtained from the HG-U133 Plus 2.0 and HG-U133A arrays (Affymetrix, Santa Clara, CA). The left column shows the expression values before the median rank score (MRS) method was applied, and the right column shows the values after cross-platform normalization. For all three genes, the MRS algorithm resulted in value distributions that were similar for both types of microarrays.
