Prognostic impact, concurrent genetic mutations, and gene expression features of AML with CEBPA mutations in a cohort of 1182 cytogenetically normal AML patients: further evidence for CEBPA double mutant AML as a distinctive disease entity

In the current World Health Organization classification of acute myeloid leukemia (AML), AML with mutated CCAAT/enhancer binding protein alpha (AML with mutated CEBPA), has been designated as a provisional disease entity in the category “AML with recurrent genetic abnormalities.”^1,2 

CEBPA encodes a transcription factor that is essential for neutrophil development. Targeted disruption of Cebpa in mice results in a selective block in early granulocyte development, which is a hallmark of AML.^3,4  Two proteins may be translated from the CEBPA transcripts, such as a 42-kDa (p42) and a shorter 30-kDa (p30) protein both translated from the same mRNA transcript. The p42 isoform contains 2 regulatory transactivation domains (TAD) in the N-terminus (TAD1 and TAD2), whereas the shorter p30 isoform only carries the TAD2 domain. Both isoforms contain the C-terminal basic DNA-binding domain and the leucine zipper (bZIP), involved in DNA binding and protein dimerization. In AML, CEBPA mutations mainly occur in cytogenetically normal AML (CN-AML) with an incidence of 5% to 14%.^5-14  The 2 main types of mutations can be distinguished: N-terminal frame-shift mutations resulting in the translation of a 30-kDa protein only and the C-terminal in-frame mutations in the basic zipper region affecting DNA binding and homodimerization and heterodimerization.^8,15  As a consequence, these mutations create an imbalance between proliferation and differentiation of hematopoietic progenitors.^10,16 

Patients who have AML with CEBPA mutations can be separated into 2 subgroups, namely, those with a single mutation CEBPA (CEBPA^sm) and those with a double mutation CEBPA (CEBPA^dm).^17-21  In the majority of CEBPA^dm AML, both alleles are mutated.¹⁹  These biallelic mutations frequently consist of an N-terminal mutation on one allele and a C-terminal bZIP mutation on the other. In CEBPA^sm AML, mutations occur either in the N terminus or in the C terminus of the gene. In previous studies in which these 2 subgroups were not considered, AML patients with mutated CEBPA had a relatively good outcome.^5,7,12,13  More recent data suggest that this favorable outcome is mainly observed in AML patients with CEBPA^dm and not CEBPA^sm.^17-21  Moreover, it has been suggested that concurrent mutations may occur more frequently in CEBPA^sm than in CEBPA^dm AML. The impact of coexisting mutations remains elusive and needs to be validated in large cohorts.

By applying gene expression profiling (GEP), it was demonstrated that CEBPA^dm AML can be distinguished from CEBPA^sm and the majority of CEBPA^wt AML based on a signature.¹⁸  However, a CEBPA^dm GEP signature did not predict CEBPA^dm AML with maximum accuracy because AML in which CEBPA was silenced (CEBPA^silenced) by promoter hypermethylation carried a highly similar signature.^22,23 

The objectives of this study were to evaluate the impact of CEBPA^dm versus CEBPA^sm on clinical outcome of CN-AML and to investigate the impact of concurrent mutations in nucleophosmin (NPM1^mutant) and/or fms-like tyrosine kinase receptor-3 (FLT3) internal tandem duplication mutations (FLT3^ITD). In addition, we searched for CEBPA-associated gene expression signatures and determined the frequency of predisposing CEBPA germline mutations. For these purposes, we combined datasets from the Dutch-Belgian Hemato-Oncology Cooperative Group (HOVON)-Swiss Group for Clinical Cancer Research (SAKK) and the German and Austrian AML Study Group (AMLSG).

Diagnostic bone marrow or peripheral blood samples from 1182 younger adults (16 to 60 years of age) with CN-AML were analyzed; 193 patients were enrolled on HOVON-SAKK protocols -04, -04A, -29, and -42 (available at www.hovon.nl),^24-27  and 989 patients on AMLSG protocols AML HD93 (n = 74),²⁸  AML HD98A (n = 313),²⁹  AMLSG 07-04 (n = 376), registered at www.clinicaltrials.gov as NCT00151242, AML SHG 02-95 (n = 94),³⁰  and AML SHG 01-99 (n = 180), registered at www.clinicaltrials.gov as NCT00209833). All patients provided written informed consent in accordance with the Declaration of Helsinki. All trials were approved by the Institutional Review Board of Erasmus University Medical Center, University of Ulm, and Hannover Medical School.

Mutation analyses for the genes FLT3^ITD and FLT3 tyrosine kinase domain (TKD) mutations (FLT3^TKD) and the NPM1 were performed as described previously.^31-33 CEBPA^dm and CEBPA^sm AML were identified by denaturing high-performance liquid chromatography or direct sequencing as described.¹⁸  Cases that carried an insertion polymorphism^18,21  (www.ncbi.nlm.nih.gov/sites/snp; genome.ucsc.edu/cgi-bin/hgGateway; www.ensembl.org/Homo_sapiens/Gene/Variation_Gene) or variations that did not lead to amino acid changes were considered wild-type (wt). Cases were categorized as CEBPA^dm when 2 different mutations or 1 homozygous mutation were present as determined by sequencing analysis; cases with only a single heterozygous mutation were designated as CEBPA^sm. In 71 of the 151 patients with CEBPA mutations, DNA obtained from buccal swabs (n = 52), peripheral blood (n = 8), or bone marrow (n = 11) samples in complete remission (CR) were studied for the presence of CEBPA germline mutations. Patient demographics and molecular characteristics are summarized in Table 1. All CEBPA-mutated patients, except for patients treated within the AMLSG protocol 07-04, have been previously reported in different studies.^7,13,18 

Data from GEP analysis were available in 674 AML patients (53% CN-AML, HOVON-SAKK and AMLSG cohorts), generated using Affymetrix (supplemental Table 1, available on the Blood Web site; see the Supplemental Materials link at the top of the online article). Sample processing and quality control were carried out as described previously.^23,34  For both cohorts, normalization of raw data was processed with Affymetrix Microarray Suite 5 (MAS5) to target intensity values at 100. Intensity values were log₂ transformed, and mean centered per probe set per cohort. GEP data are available at the Gene Expression Omnibus (National Center for Biotechnology Information; accession number GSE14468 [HOVON-SAKK cohort] and GSE22845 [AMLSG cohort]). There were 42 CEBPA^dm and 18 CEBPA^sm cases for which the GEP was determined (supplemental Table 1).

Statistical analyses were performed using Mathworks (Matlab R2009b) with the statistical, bioinformatics, and pattern recognition toolbox (Prtools). For clinical, molecular, univariate, and multivariate analyses, patients with CN-AML and age ≤ 60 years (supplemental Table 1) were included. Molecular and clinical variables of both patient cohorts (HOVON-SAKK and AMLSG) were comparable. Differences were assessed for CEBPA^sm and CEBPA^dm groups in comparison with CEBPA^wt group (Table 1), using the Mann-Whitney U test for continuous variables and the 2-sided Fisher exact test for categorical variables.

Outcome measures of the HOVON-SAKK and AMLSG cohorts were comparable (log-rank test overall survival [OS], P = .08; event-free survival [EFS], P = .47; supplemental Figure 1A-B, for respective cohort). There were no statistical differences in outcome in patients receiving autologous or allogeneic hematopoietic stem cell transplantation between the HOVON-SAKK and AMLSG cohorts (log-rank test OS, P = .68; EFS, P = .89; supplemental Figure 2A-B, respectively).

For univariate analysis, significance was assessed using the stratified log-rank test and Kaplan-Meier estimates for OS, EFS and relapse-free survival (RFS). The recommended consensus criteria³⁵  were used for the definition of CR and survival end points such as OS, EFS, and RFS. Multivariate analysis was performed using stratified Cox proportional hazard model. For all analyses, a value for P ≤ .05 was considered statistically significant and for survival analyses, values for P were computed using the full time span. Note that the closed testing procedure³⁶  was applied, and a correction for multiple testing³⁷  was only performed if the global log-rank test resulted in P > .05.

For gene expression–based classification of CEBPA^dm cases, GEP of the HOVON-SAKK cohort was used to derive the 25-probe set predictive signature and the AMLSG cohort as validation set. To summarize, a logistic regression model with Lasso regularization (a continuous feature selection procedure) was used because it takes the correlation structure between the probe sets into account (see supplemental “creation and evaluation of the CEBPA^dm predictive signature”).

CEBPA mutations were detected in 151 of the 1182 (12.8%) CN-AML patients; 91 (60%) had CEBPA^dm among which the combination of N- and C-terminal mutations was the predominant genotype (82 of the 91). CEBPA^dm cases with only N- or C-terminal mutations were less frequently observed (4 of the 91 or 5 of the 91, respectively). A total of 60 of 151 (40%) CEBPA-mutated cases had CEBPA^sm, which occurred most frequently in the N terminus (47 of the 60). Only 13 of the 60 CEBPA^sm cases had in-frame insertion or deletion mutations affecting the bZIP domain (Figure 1).

Five of 71 (7%) CEBPA mutant AML patients analyzed carried CEBPA germline mutations: in 2 of the 5 patients, the germline mutation was localized in the N-terminus, and both acquired a C-terminal mutation. Both patients had a family history of AML and were diagnosed at a young age. In the remaining 3 patients, the germline mutation was in the C terminus; 1 of these patients gained an additional N-terminal mutation and the second patient an additional C-terminal mutation at the time of AML-diagnosis. None of these 3 patients had a family history of AML. Alignment to distinct Single Nucleotide Polymorphism databases (www.ncbi.nlm.nih.gov/sites/snp; genome.ucsc.edu; or www.ensembl.org/Homo_sapiens/Gene/Variation_Gene) did not identify one of these germline sequence variations as a polymorphism. Using the PolyPhen database (http://genetics.bwh.harvard.edu/pph/), all 3 C-terminal mutations were predicted to be damaging to the function and structure of the protein (Table 2).

Concurrent mutations were seen less frequently in CEBPA^dm than in CEBPA^sm AML (22% vs 60%; P < .0001, Figure 1); frequency for NPM1^mutant in CEBPA^dm versus CEBPA^sm AML was 3.3% and 35% (P < .0001), and for FLT3^ITD was 7.7% and 30% (P < .001), respectively (Table 1). Comparing CEBPA^sm and CEBPA^wt AML, NPM1^mutant were slightly less frequent in CEBPA^sm AML (35% vs 54.3%; P = .018); the frequency for FLT3^ITD was comparable between the 2 groups (30% vs 33.7%).

Regarding presenting clinical characteristics, CEBPA^dm mutations were associated with younger age (median age 44 vs 48 years; P = .04) and lower platelet counts (median number 38 × 10⁹/L vs 65 × 10⁹/L; P < .0001) compared with CEBPA^wt patients (Table 1).

For correlation with clinical outcome, 1182 CN-AML were considered. CEBPA^dm was associated with a higher CR rate compared with CEBPA^sm (92% vs 78%, P = .02) and with CEBPA^wt (92% vs 79%, P = .002). There was no difference to CR probability between CEBPA^sm and CEBPA^wt patients (78% vs 79%, P = .86).

The median follow-up time for survival in the 1182 CN-AML patients was 33 months (95% confidence interval [CI], 25.6-0.4); the estimated 5-year OS and RFS were 42% (95% CI, 39%-45%) and 34% (95% CI, 31%-38%), respectively.

CEBPA^dm AML was associated with a significantly superior outcome compared with CEBPA^wt AML (5-year OS, 63% vs 39%, P < .0001; EFS, 45% vs 28%, P < .0001; RFS, 44% vs 32%, P = .05), as shown (Figure 2A and supplemental Figure 3A,D). A somewhat better outcome was also found for CEBPA^sm AML compared with CEBPA^wt AML (5-year OS, 55% vs 39%, P = .05; RFS, 49% vs 32%, P = .02; but not EFS, 37% vs 28%, P = .22). No significant difference was evident between CEBPA^dm and CEBPA^sm AML (5-year OS, P = .06; EFS, P = .16; RFS, P = .48). Of note, no differences in outcome were observed among CEBPA^sm patients with either C-terminal (n = 13) or N-terminal (n = 47) mutations (5-year OS, 54% vs 56%, P = .58; supplemental Figure 4).

In multivariate analyses considering other prognostic indicators (Table 3), the presence of CEBPA^dm was an independent prognostic factor for favorable OS (HR 0.36, P < .0001), EFS (HR 0.41, P < .0001), and RFS (HR 0.55, P = .001), whereas CEBPA^sm did not impact these 3 end points (Table 3).

Finally, we performed explorative subgroup analyses in CEBPA^sm and CEBPA^wt AML to evaluate the impact of 4 FLT3/NPM1 genotype subgroups: FLT3^ITD/NPM1^mutant (n = 10); FLT3^ITD/NPM1^wt (n = 8); FLT3^wt/NPM1^mutant (n = 11); and FLT3^wt/NPM1^wt (n = 21). There were 10 cases from the CEBPA^sm group excluded for which the genotypes were unknown.

Among patients with CEBPA^sm AML, the FLT3^ITD/NPM1^wt genotype had an inferior OS compared with patients with the FLT3^wt/NPM1^wt genotype (5-year OS, 25% vs 49%, P = .05; Figure 2B); for EFS and RFS, there was a trend toward an inferior outcome (supplemental Figure 3B,E); in contrast, the FLT3^wt/NPM1^mutant genotype associated in trend with a favorable outcome compared with the FLT3^wt/NPM1^wt genotype (5-year OS, 78% vs 49%, P = .2; EFS, 59% vs 32%, P = .08; RFS, 66% vs 40%, P = .38; Figure 2B and supplemental Figure 3B,E). In analogy, in the CEBPA^wt group, the FLT3^ITD/NPM1^wt genotype had a significantly inferior survival compared with the FLT3^wt/NPM1^wt genotype (5-year OS, 17% vs 34%, P = .001; EFS, 11% vs 14%, P = .04; RFS, 15% vs 24%, P = .002; Figure 2C and supplemental Figure 3C,F), whereas the FLT3^wt/NPM1^mutant genotype was associated with a favorable outcome (5-year OS, 57% vs 34%, P < .0001; EFS, 47% vs 14%, P < .0001; RFS, 50% vs 24%, P < .0001; Figure 2C and supplemental Figure 3C,F). Thus, we observed comparable trends for favorable (FLT3^wt/NPM1^mutant) and inferior (FLT3^ITD/NPM1^wt) outcome in the CEBPA^sm and CEBPA^wt subgroups. The outcome for all CEBPA^smFLT3/NPM1 genotypes was higher (not significantly, P > .05), compared with the CEBPA^wt genotypes; however, the distinct groups were relatively small. For CEBPA^dm AML, sample sizes of the composite genotypic subgroups were too small for analysis.

GEP was performed in a subset of the CN-AML patients and includes cytogenetically abnormal patients (n = 674) as shown (supplemental Table 1). Unsupervised analyses, by computing pairwise Pearson correlation coefficients of 674 AML cases, revealed distinct GEP clusters (Figure 3A), including the known clusters of AML with inv(16), t(15;17), or t(8;21), as shown previously.²³  These subtypes revealed high correlation within the GEP cluster (average correlation, 0.42, 0.49, and 0.49, respectively) and differed significantly (P < .0001) among the AML cases without any of these aberrations, (supplemental Figure 5B-C and supplemental Figure 5E). We observed that the CEBPA^dm AML cases were highly similar within the cluster (average correlation, 0.35) and differed significantly from cases without a CEBPA^dm (P < .0001; supplemental Figure 5D). CEBPA^sm AML cases showed reduced similarity (average correlation, 0.15) and did not differ from cases without CEBPA^sm (P = .12; supplemental Figures 3A,5A).

The previously predictive CEBPA^dm signature¹⁸  was hampered by the recently reported CEBPA^silenced AML cases that carry a similar GEP.²²  The 2 independent AML cohorts were used to train and evaluate the predictive value of the CEBPA^dm signature in terms of sensitivity and specificity. A predictive signature was created (Figure 3B and supplemental Table 2), containing 25-probe sets, using a logistic regression model with Lasso regularization^38,39  that selects discriminative probe sets between the classes, CEBPA^dm (n = 26) and all other AML cases, CEBPA^wt and CEBPA^sm (n = 494). Subsequently, a classifier was trained on the entire HOVON-SAKK cohort based on a 2-class approach; with 26 CEBPA^dm versus 494 other (CEBPA^wt and CEBPA^sm) cases. This trained classifier subsequently classified 16 candidate CEBPA^dm cases (supplemental Table 3) in the AMLSG cohort of 154 AML cases (16 CEBPA^dm, 6 CEBPA^sm, and 132 CEBPA^wt; supplemental Table 1). Among the CEBPA^dm cases were 5 with either homozygous N- or C-terminal CEBPA^dm mutations, and a CEBPA^dm patient with a germline C-terminal mutation. This approach showed perfect sensitivity and specificity (both 100%; Figure 3C). In addition, we performed a classification among CEBPA^dm, CEBPA^sm, and CEBPA^wt to infer whether we were able to accurately classify CEBPA^sm cases. We observed that all CEBPA^sm cases were classified as CEBPA^wt, thus CEBPA^sm cases did not have a consistent gene expression pattern and were different from the CEBPA^dm group.

In this study, we established the value of CEBPA^dm mutation in a large cohort of CN-AML patients from AMLSG and HOVON-SAKK treatment trials. Applying denaturing high-performance liquid chromatography and whole gene sequencing, we detected 91 (7.7%) CEBPA^dm and 60 (5.1%) CEBPA^sm mutations among 1182 patients. In multivariate analyses, we demonstrate that the presence of CEBPA^dm but not CEBPA^sm is an independent factor for favorable outcome in AML, which confirms previous findings reported in studies with relatively small cohorts.^17-19,21 

Concurrent mutations were significantly less frequent in CEBPA^dm compared with CEBPA^sm AML. This factor was true for FLT3^ITD and in particular for NPM1^mutant, which were virtually not present among CEBPA^dm cases, a finding that is consistent with previously published data.²⁰ 

Compared with previous studies^17-21  and with the large number of cases, we were able to evaluate the prognostic impact of the CEBPA mutational status in the context of the FLT3/NPM1 genotypes. Among CEBPA^sm AML, the 4 combined genotypes showed similar trend with regard to outcome compared with CEBPA^wt AML (Figure 2B-C). Nevertheless, we observed a higher outcome (not significant) for all CEBPA^smFLT3/NPM1 genotypes compared with the CEBPA^wt genotypes, but these groups are relatively small. These findings, supported by data from multivariable analysis, strongly suggest that not the existence of CEBPA^sm per se but rather the combined effects of CEBPA^sm and FLT3^ITD and/or NPM1^mutant determine outcome in these AML patients.

We have previously derived gene expression signatures that predict AML with inv(16), t(15;17), and t(8;21) with 100% accuracy. In this study, we generated a refined GEP signature of 25-probe sets that predict CEBPA^dm AML cases (6 genes overlapped with the previous signature,¹⁸  as indicated in “Supplemental Materials”). This signature showed sensitivity and specificity of 100% and has a better predictive power than the CEBPA^dm signature that we defined before.¹⁸  In fact, in contrast to the previous signature, the new signature also discriminates CEBPA^dm from AML with hypermethylation of the proximal promoter region of CEBPA.²²  Classification results were not affected by homozygous N- or C-terminal CEBPA^dm mutations or because of germline mutation. Because this 25-probe set signature was optimized for classification it does not necessarily provide insight into the biologic meaning of CEBPA^dm mutations.

Currently, nucleotide sequencing is used as the gold standard for the identification of CEBPA mutations. Because of the much higher effort required, the GEP technique should not be considered as a primary diagnostic tool in AML. However, GEP can be confirmatory, especially in cases in which the CEBPA gene appears difficult to sequence. More importantly, GEP provides relevant insights in the biology of the disease and the affected signaling pathways and therefore allows further classification/refinement of AML.

Finally, we evaluated the frequency of CEBPA germline mutations in this large cohort of CEBPA-mutated cases. Among 71 mutated patients, 5 revealed germline mutations. Of these cases, 4 developed CEBPA^dm AML, that is, 4 cases acquired a mutation in the second allele. This finding is in line with previous data.^40,41  Interestingly for the first time, we identified 3 C-terminal germline mutations of which 2 cases acquired a second CEBPA mutation at the time of AML diagnosis. In GEP analysis both cases clustered within the CEBPA^dm group and were classified as a CEBPA^dm, providing evidence that these C-terminal sequence variations are mutations rather than polymorphisms. In line with the GEP data, all 3 C-terminal germline mutations were predicted to be damaging for the function and the structure of the protein.

In the current World Health Organization classification of AML, AML with mutated CEBPA has been designated as a provisional disease entity in the category “AML with recurrent genetic abnormalities.” Based on our data obtained from a large patient cohort together with previous findings, we propose that CEBPA^dm AML should be clearly distinguished from CEBPA^sm AML and that only AML with CEBPA^dm should be considered as an independent entity in the classification of the disease.

The authors thank Martin van Vliet and Jelle Goeman for the discussions.

This research was supported by the Center for Translational Molecular Medicine and by Else Kröner-Fresenius-Stiftung grant P38/05//A49/05//F03, the Network of Competence Acute and Chronic Leukemias grant 01GI9981, and the Bundesministerium für Bildung und Forschung, Germany, grant 01KG0605 (“IPD-meta-analysis: a model-based hierarchical prognostic system for adult patients with acute myeloid leukemia [AML]”).

Contribution: E.T. performed research, data analysis, data interpretation, creation of the figures, and manuscript writing; L.B. and A.C. performed research, data analysis, and interpretation; M.A.S. performed data analysis, data interpretation, and manuscript writing; C.A.J.E., B.J.W., and S.C.v.d.P.-v.d.L. performed research; F.D. performed research and data interpretation; J.K. and A.G. provided study material; R.F.S. performed research, data interpretation, and manuscript writing; and B.L., R.D., H.D., P.J.M.V., and K.D. designed the study and performed data interpretation and manuscript writing.

Conflict-of-interest disclosure: B.L., R.D., and P.J.M.V. have declared ownership interests in Skyline, a spinoff company of Erasmus University Medical Center, held in a Special Purpose Foundation of Erasmus University Medical Center. The remaining authors declare no competing financial interests.

Correspondence: Konstanze Döhner, Department of Internal Medicine III, University of Ulm, Albert-Einstein-Allee 23, 89081 Ulm, Germany; e-mail: konstanze.doehner@uniklinik-ulm.de.