ASXL1 mutations identify a high-risk subgroup of older patients with primary cytogenetically normal AML within the ELN Favorable genetic category

The additional sex combs like-1 (ASXL1) gene, located in chromosome band 20q11, encodes a highly conserved protein that belongs to the enhancer of trithorax and polycomb (ETP) genes, a gene family with dual functions in both epigenetic activation and repression of gene transcription.¹  In mice, Asxl1 is necessary for proper regulation of Hox gene expression during embryogenesis.²  Human ASXL1 is involved in an unbalanced chromosomal rearrangement [dic(9;20)(p13;q11)] recurrent in acute lymphoblastic leukemia.³  More recently, mutations in ASXL1 exon 12 were identified in patients with myelodysplastic syndromes,⁴  myeloproliferative neoplasms,⁵  and acute myeloid leukemia (AML).^6,7  The reported prevalence of ASXL1 mutations in primary (de novo) AML ranges between 5% and 30%, whereas some studies suggest a higher prevalence in secondary AML.^6-10  Cytogenetically normal (CN) AML represents the largest cytogenetic subgroup of adult AML,¹¹  and the one that has been most extensively studied regarding the clinical implications of various gene mutations.^12,13  To date, no study has specifically addressed the association of ASXL1 mutations with clinical and molecular patient characteristics and outcomes in CN-AML.

In a cytogenetically heterogeneous cohort, Chou et al⁸  reported that ASXL1-mutated (mut) AML patients had a lower complete remission (CR) rate and shorter overall survival (OS) than ASXL1–wild-type (wt) patients. ASXL1 mutation status was no longer significantly associated with outcomes in multivariable models adjusting for other prognostic factors. Although more than one-half of all ASXL1 mutations in that study were found in patients 60 years of age or older, 80% of the patients included in the outcome analyses were 60 years of age or younger. It thus remains unclear whether ASXL1 mutation status has independent prognostic relevance in older patients or in specific cytogenetic groups, such as CN-AML.

Therefore, we studied ASXL1 mutations and their associations with baseline patient characteristics, other genetic alterations, and outcomes in relatively large and well-characterized cohorts of younger (< 60 years) and older (≥ 60 years) primary CN-AML patients. We also examined genome-wide gene and microRNA (miR) expression profiles and report the first ASXL1 mutation-associated gene-expression signature in CN-AML, which will help to gain insights into the biology of ASXL1-mut myeloid neoplasms.

Pretreatment BM or peripheral blood (PB) samples containing more than 20% leukemic blasts were obtained from 423 primary CN-AML patients (including 189 patients younger than 60 years and 234 patients 60 years of age or older). The diagnosis of normal cytogenetics was based on 20 or more analyzed metaphase cells in BM specimens subjected to 24- and/or 48-hour culture. Cytogenetic analyses were performed by Cancer and Leukemia Group B (CALGB)-approved institutional laboratories and confirmed by central karyotype review.¹⁴  All patients received cytarabine/daunorubicin-based first-line therapy on CALGB trials (supplemental Appendix, available on the Blood Web site; see the Supplemental Materials link at the top of the online article). As stipulated in the treatment protocols for both younger and older patients, no patient received allogeneic stem cell transplantation in first CR. Study protocols were in accordance with the Declaration of Helsinki and approved by the institutional review boards at each center, and all patients provided written informed consent.

Exon 12 of the ASXL1 gene was amplified from genomic DNA by PCR and analyzed by direct sequencing (supplemental Methods). Patients were also characterized for FLT3-internal tandem duplications (FLT3-ITD),¹⁵ FLT3-tyrosine kinase domain mutations (FLT3-TKD),¹⁶ MLL-partial tandem duplications (MLL-PTD),^17,18  and mutations in NPM1,¹⁹ CEBPA,²⁰ TET2,²¹ IDH1/IDH2,²²  and WT1²³  as previously reported. All mutational analyses were performed at The Ohio State University Comprehensive Cancer Center.

Gene-expression profiling was performed using Affymetrix HG-U133 plus Version 2.0 oligonucleotide microarrays; miR-expression profiling was performed using a custom microarray platform (OSU_CCC miR array Version 4.0), as previously reported.^19,21  See supplemental Methods for details on microarray analyses. All microarray data are available on ArrayExpress under accession number E-TABM-1208.

Baseline characteristics were compared between ASXL1-mut and ASXL1-wt patients using Fisher exact test for categorical variables and the Wilcoxon rank-sum test for continuous variables. Definitions of clinical end points (ie, CR, disease-free survival [DFS], OS, and event-free survival [EFS]) and details on outcome analyses are provided in supplemental Methods. Briefly, for time-to-event analyses, we calculated survival estimates using the Kaplan-Meier method, and compared groups by the log-rank test. We constructed multivariable logistic regression models to analyze factors associated with the achievement of CR, and multivariable Cox proportional hazards models for factors associated with survival endpoints. All analyses were performed by the Alliance for Clinical Trials in Oncology Statistics and Data Center.

We studied the coding sequence of ASXL1 exon 12 in 423 primary CN-AML patients 18 to 83 years of age. In previous studies, all ASXL1 mutations were found in this exon, which contains more than 60% of the entire ASXL1 coding sequence.^4,5  Among 189 younger patients, only 6 carried protein-truncating (nonsense or frame shift) ASXL1 mutations (3.2%). In contrast, among 234 older patients, we identified 39 heterozygous nonsense or frame shift mutations affecting 38 patients (16.2%; Figure 1A; P < .001 for the comparison between younger and older patients). In agreement with previous reports,^4-9  duplication of a guanine nucleotide at coding sequence position 1934 (c.1934dupG) accounted for approximately half of the ASXL1 mutations we identified. The remaining patients carried a variety of other nonsense or frame shift mutations. Supplemental Table 1 provides detailed information on all mutations identified in the entire cohort of 423 patients. Besides these mutations, we also detected missense variations leading to exchanges of single amino acids that have not been annotated as single nucleotide polymorphisms (SNPs) in the dbSNP database Version 132 (supplemental Table 2). Such variations were present in 13 younger patients (6.9%) and in 16 older patients (6.8%, 2 also had a nonsense or frame shift mutation). Some of these variants have recently been shown to be germline SNPs.^24,25  Patients who only had missense variations in ASXL1 exon 12 (13 younger and 14 older patients) were excluded because of the uncertain pathogenic relevance of these changes. This is consistent with previous studies.^5-8,26  Baseline clinical characteristics and treatment outcomes of these patients were not significantly different from patients with wild-type ASXL1 (supplemental Figure 1).

Since ASXL1 mutations were rare in younger primary CN-AML patients and since younger and older patients were enrolled on separate CALGB trials that differed in their treatment intensity, we restricted all subsequent analyses of baseline associations and treatment outcomes to older patients (n = 220). Although the small number of ASXL1-mut patients younger than 60 years (n = 6) did not allow a detailed evaluation in this age group, the clinical and molecular characteristics of these patients generally resembled our findings in older ASXL1-mut patients (supplemental Results).

Patients with ASXL1 mutations (n = 38) tended to be more frequently male (P = .08), and had lower pretreatment white blood cell counts (WBC; P = .02), lower PB (P < .001) and BM (P = .04) blast percentages, and a trend toward higher platelet counts (P = .06), compared with ASXL1-wt patients (n = 182; Table 1). ASXL1 mutations rarely occurred concurrently with NPM1 mutations (P < .001) or FLT3-ITD (P = .002; Table 1; Figure 1B). On the other hand, 26% of ASXL1-mut patients also carried CEBPA mutations, compared with only 9% of ASXL1-wt patients (P = .01). Figure 1B visualizes coexisting mutations in NPM1, FLT3, CEBPA and 5 other genes among the 38 ASXL1-mut older CN-AML patients. In 8 of these 38 patients (21%), no further mutations were detected in any of the other 8 genes analyzed.

According to the current European LeukemiaNet (ELN) guidelines for the diagnosis and management of AML, CN-AML patients are classified into 2 genetic categories based on their NPM1, FLT3-ITD, and CEBPA mutation status.²⁷  The ELN Favorable category of CN-AML is defined as patients with mutated CEBPA and/or mutated NPM1 without FLT3-ITD. All remaining CN-AML patients (ie, those with wild-type CEBPA, and wild-type NPM1 with or without FLT3-ITD or mutated NPM1 with FLT3-ITD) form the ELN Intermediate-I category. ASXL1-mut patients tended to fall into the ELN Favorable category less frequently than into the ELN Intermediate-I category (P = .08; Table 1). Of the 12 ELN Favorable/ASXL1-mut patients, 10 carried CEBPA mutations, and the other 2 had mutated NPM1 without FLT3-ITD (Figure 1B).

Due to the baseline associations between ASXL1 mutations and established prognostic gene mutations (ie, NPM1, FLT3-ITD, and CEBPA mutations), it was important to study the prognostic relevance of mutated ASXL1 not only in our entire cohort of older CN-AML patients, but also in the genetic subsets defined by the ELN classification (Table 2). To adjust for other, potentially confounding risk factors, we also constructed multivariable models (Table 3). Overall, ASXL1-mut patients had a lower CR rate (53%) than ASXL1-wt patients (71%, P = .04). Even within the ELN Favorable group, only 50% of the ASXL1-mut patients achieved CR, compared with 82% of ASXL1-wt patients (P = .02). In the ELN Intermediate-I genetic category, no significant difference in CR rates was observed between ASXL1-mut and ASXL1-wt patients (Table 2). In a multivariable logistic regression model for achievement of CR, ELN Favorable/ASXL1-mut patients had lower odds of achieving a CR, compared with ELN Favorable/ASXL1-wt patients (P = .03; odds ratio [OR] = 0.24; Table 3), whereas ASXL1 mutations were not associated with CR rate in ELN Intermediate-I patients (P = .47). Other factors associated with a lower CR rate were a higher WBC (P = .01; OR = 0.73), higher platelet count (P = .03; OR = 0.81), and the presence of extramedullary involvement (P = .05; OR = 0.49).

Among patients who achieved CR, those with ASXL1 mutations had shorter DFS than ASXL1-wt patients (P = .03; 3-year DFS, 10% vs 19%; Table 2; Figure 2A). All 6 ELN Favorable/ASXL1-mut patients achieving CR relapsed within 13 months, whereas 27% of the ELN Favorable/ASXL1-wt patients were alive and disease-free at 3 years. ASXL1 mutations were not associated with DFS in ELN Intermediate-I patients (P = .95; Table 2). A multivariable Cox model for DFS showed that ELN Favorable patients had a more than 4-fold increased risk of relapse or death if they carried an ASXL1 mutation (P < .001; hazard ratio [HR] = 4.38; Table 3). In contrast, ASXL1 mutations were not associated with DFS in ELN Intermediate-I patients in this model (P = .65). The P value for interaction between ASXL1 mutation status and ELN category was .01, confirming that the impact of ASXL1 mutations on DFS differed between ELN Favorable and Intermediate-I patients. The only other factor associated with shorter DFS was extramedullary involvement (P = .03; HR = 1.65).

ASXL1-mut patients had shorter OS than ASXL1-wt patients (P = .006; 3-year OS, 5% vs 23%; Table 2; Figure 2B). A negative impact of ASXL1 mutations on OS was also observed within the ELN Favorable subset (P < .001): all ASXL1-mut patients died within 18 months from diagnosis, whereas 34% of ASXL1-wt patients were alive at 3 years (Table 2; Figure 3A). In the ELN Intermediate-I category, ASXL1 mutation status was not significantly associated with OS (P = .73; Table 2, Figure 3B). A multivariable model for OS confirmed that patients in the ELN Favorable category with mutated ASXL1 had a more than 4 times higher risk of death than ASXL1-wt patients (P < .001; HR = 4.43; Table 3), whereas ASXL1 mutations were not associated with OS in the ELN Intermediate-I group (P = .79; P < .001 for interaction between ASXL1 mutations and ELN category). The only other factor associated with OS was higher WBC (P = .01; HR = 1.13).

Similarly, EFS of ASXL1-mut patients was worse compared with ASXL1-wt patients (P = .002; 3-year EFS, 5% vs 14%; Table 2; Figure 2C). Within the ELN Favorable genetic category, ASXL1-mut patients had worse EFS than ASXL1-wt patients (P < .001; 3-year EFS, 0% vs 22%; Table 2; Figure 3C), whereas no significant difference was observed in ELN Intermediate-I patients (P = .68; Figure 3D). In a multivariable model for EFS, ELN Favorable patients with mutated ASXL1 had a 3.6-fold higher risk of an event than ASXL1-wt patients (P < .001; HR = 3.61; Table 3). In this multivariable model, ASXL1 mutations were not associated with EFS in the ELN Intermediate-I group (P = .44; P = .005 for interaction between ASXL1 mutations and ELN category). Higher WBC (P = .005; HR = 1.18) and extramedullary involvement (P = .02; HR = 1.52) were also associated with shorter EFS. Of note, none of the ASXL1-mut and only 5 ASXL1-wt patients died while in first CR (Table 2), indicating that the observed survival differences reflect differences in the course of the disease rather than treatment toxicity or other unrelated causes.

In summary, ASXL1 mutations were associated with lower CR rate and shorter survival, particularly within the ELN Favorable genetic category and thus identify a previously unrecognized high-risk subgroup among older primary CN-AML patients.

As mentioned previously, ASXL1 mutations were significantly associated with mutated CEBPA in older primary CN-AML. Of 27 CEBPA-mut patients, 10 also were ASXL1-mut. Five CEBPA-mut/ASXL1-mut patients (50%) achieved a CR, compared with 13 of 17 CEBPA-mut/ASXL1-wt patients (76%, P = .21; supplemental Table 3). We could not formally analyze DFS because of inadequate sample size. However, all 5 CEBPA-mut/ASXL1-mut patients who achieved CR relapsed within 13 months, whereas 31% of CEBPA-mut/ASXL1-wt patients were still alive and disease-free at 3 years. The OS (P < .001; supplemental Figure 2A) and EFS (P = .02; supplemental Figure 2C) of CEBPA-mut/ASXL1-mut patients were significantly worse than those of CEBPA-mut/ASXL1-wt patients (supplemental Table 3). Recently, it has been reported that only double, but not single, CEBPA mutations are associated with favorable outcomes.^28,29  We were unable to formally study the prognostic relevance of ASXL1 mutations in patients with single or double CEBPA mutations separately because there were only 5 ASXL1-mut patients with single CEBPA mutations and 5 ASXL1-mut patients with biallelic CEBPA mutations. However, whereas 4 of 5 CEBPA-double-mutated/ASXL1-mutated patients achieved CR, all 4 relapsed within 13 months. All 5 CEBPA-double-mutated/ASXL1-mutated patients died within 18 months from study inclusion. Thus, in our cohort, patients with double CEBPA mutations and mutated ASXL1 appeared to have unfavorable outcomes. Among CEBPA-wt patients, those with ASXL1 mutations tended to have a lower CR rate (54% vs 70%; P = .09), showed no significant difference in DFS (P = .17) or OS (P = .11; supplemental Figure 2B), but had shorter EFS (P = .03; supplemental Figure 2D) compared with ASXL1-wt patients (supplemental Table 3). These exploratory analyses suggest that ASXL1 mutations may have a particularly unfavorable impact in older CN-AML patients harboring CEBPA mutations.

In an analysis including both younger and older CN-AML patients, we have recently demonstrated that TET2 mutations also adversely affect patient outcomes in the ELN Favorable genetic subset.²¹  Therefore, we analyzed the outcomes of older ELN Favorable CN-AML patients when taking both ASXL1 and TET2 mutation status into consideration. ASXL1-mut patients (irrespective of their TET2 status) had a significantly worse OS compared with either ASXL1-wt/TET2-wt patients (P < .001) or ASXL1-wt/TET2-mut patients (P < .001; supplemental Figure 3A). Likewise, EFS of ELN Favorable/ASXL1-mut patients was shorter than EFS of ASXL1-wt/TET2-wt (P < .001) or ASXL1-wt/TET2-mut patients (P = .004; supplemental Figure 3B). ELN Favorable patients with ASXL1-wt and TET2-wt had the best outcomes in these analyses, with a 3-year OS of 39% and a 3-year EFS of 26%. Because of small patient numbers, we could not analyze ASXL1-mut/TET2-wt and ASXL1-mut/TET2-mut patients separately.

To gain insight into the biologic consequences of ASXL1 mutations in CN-AML, we studied genome-wide gene-expression signatures associated with these mutations. In our cohort, ASXL1 mutations were almost mutually exclusive with NPM1 mutations and FLT3-ITD, which themselves have strong, characteristic gene-expression signatures.^15,19  To avoid confounding due to the effects of NPM1 mutations and FLT3-ITD on gene expression, we confined our analyses to NPM1-wt/FLT3-ITD-negative patients. Within this subset, we compared 26 ASXL1-mut to 39 ASXL1-wt patients and identified a signature of 92 differentially expressed probe sets, corresponding to 67 named genes (supplemental Table 4). Among the genes most strongly up-regulated in ASXL1-mut patients were LRP6 (a WNT signaling pathway coreceptor³⁰ ), cytochrome P450 CYP1B1 (associated with chemotherapy resistance³¹ ), and gap junction protein α 1 (GJA1, connexin 43). GJA1 couples early hematopoietic and stromal cells in the BM, is important for regeneration of hematopoiesis after chemotherapy,³²  and mediates secretion of stroma-derived factor 1 (CXCL12), a chemokine essential for hematopoietic stem cell homing and function.³³  On the other hand, KISS1R (a G-protein-coupled receptor suppressing CXCL12-CXCR4 signaling³⁴ ) was down-regulated in ASXL1-mut CN-AML. Also down-regulated was MLLT10 (AF10, a fusion partner in recurrent chromosomal translocations in AML³⁵ ). Figure 4 visualizes the expression of the 92 differentially expressed probe sets in all 185 older CN-AML patients with available microarray data.

Since about half of all ASXL1-mut patients had one specific type of mutation (c.1934dupG), whereas the other half had a variety of other mutations, we looked for potential biologic differences between these 2 groups by comparing the gene-expression profiles of patients with c.1934dupG versus other ASXL1 mutations. No significant expression signature was identified, suggesting that c.1934dupG and other ASXL1 mutations share a similar gene-expression profile.

Gene-set analysis was performed to identify sets of genes, representing canonical biologic pathways, which were deregulated in patients with mutated ASXL1. Ten gene sets were found to be significantly deregulated (supplemental Table 5), 7 of which were up-regulated in ASXL1-mut patients compared with ASXL1-wt. Up-regulated biologic pathways include the cytochrome P450 system and gene sets related to glycoprotein and glycolipid biosynthesis and transmembrane transport of metabolites. Three gene sets were down-regulated, including one representing pyrimidine nucleotide metabolism.

In an approach similar to the analyses of gene expression, we also studied genome-wide miR-expression profiles. We compared miR expression in the subgroup of NPM1-wt/FLT3–ITD-negative patients (24 ASXL1-mut and 38 ASXL1-wt patients) but did not identify a signature of significantly deregulated miRs associated with ASXL1 mutations.

Our study of 423 patients represents, to our knowledge, the largest cohort of primary CN-AML patients analyzed for ASXL1 mutations so far. We identified 45 ASXL1 exon 12 nonsense or frame shift mutations in 44 patients, for an overall prevalence of 10.4%. ASXL1 mutations were 5 times more common in patients 60 years of age or older (16.2%) than in younger patients (3.2%), the largest difference in the proportion of mutated cases between age groups for any gene mutation that we have studied so far. Our findings are in agreement with the report by Chou et al, who found ASXL1 mutations in 5.4% of cytogenetically heterogeneous younger AML patients, 18.0% of cytogenetically heterogeneous older patients, and in 8.9% (17 of 192) of CN-AML patients across both age groups.⁸  To our knowledge, our study is the first to show that ASXL1 mutations are associated with lower WBC and blast percentages and with mutated CEBPA in CN-AML patients. Our results also confirm previous reports showing that ASXL1 mutations very rarely occur together with NPM1 mutations or FLT3-ITD.^7,8 

We are not aware of any published data regarding the prognostic relevance of ASXL1 mutations in primary CN-AML, and particularly in older patients who account for the majority of AML cases and have the highest prevalence of ASXL1 mutations. Chou et al⁸  reported data on treatment outcomes in a cytogenetically heterogeneous cohort of 360 patients, including 202 with intermediate-risk cytogenetics. In multivariable analyses, ASXL1 mutations had no impact on OS in the entire cohort or in cytogenetic intermediate-risk patients. Of note, more than 80% of patients in the outcome analyses in that report were younger than 60 years, and only two-thirds of the patients with intermediate-risk cytogenetics had CN-AML. CN-AML not only is the most extensively studied AML subgroup on the molecular level, it also constitutes the largest cytogenetic subgroup of adult AML.^{11-13,15-23,29}  Therefore, it is important to understand the prognostic implications of mutated ASXL1 within this group. Novel molecular markers need to be studied in cytogenetically defined and molecularly well-characterized, homogeneously treated patient cohorts because the prognostic relevance of a gene mutation may vary between different cytogenetic groups,³⁶  depend on the context of other gene mutations present in the leukemic clone,^21,37,38  or be modified by treatment regimens.^18,39 

In our cohort of older primary CN-AML patients, univariable analyses revealed that ASXL1 mutations were associated with a lower CR rate and shorter DFS, OS, and EFS. At the same time, ASXL1 mutations showed baseline associations with other gene mutations, which themselves have well-recognized prognostic impact. ASXL1 mutations very rarely occurred together with NPM1 mutations,^7-9  which confer a favorable prognosis among older CN-AML patients.¹⁹  In addition, ASXL1-mut patients rarely carried FLT3-ITD (a marker associated with shorter DFS and OS in older patients¹⁵ ), but more often harbored CEBPA mutations (a favorable prognostic marker studied mainly in younger patients²⁰ ) than ASXL1-wt patients. These complex associations make it difficult to interpret the results of univariable outcome analyses concerning the entire cohort of older CN-AML patients. To better delineate the relevance of ASXL1 mutations in the context of other gene mutations, we explored their prognostic impact within the molecular subgroups defined in the recent ELN guidelines for the diagnosis and management of AML.²⁷  According to these internationally accepted guidelines, CN-AML patients are stratified into the ELN Favorable and ELN Intermediate-I genetic categories based on their NPM1, FLT3-ITD, and CEBPA mutation status, which are the only 3 mutations that showed baseline associations with ASXL1 mutations. ASXL1 mutations were relatively rare among older CN-AML patients belonging to the ELN Favorable group. However, those ELN Favorable patients who carried mutated ASXL1 had unfavorable outcomes, including a significantly lower CR rate and shorter OS and EFS compared with ELN Favorable/ASXL1-wt patients. Multivariable models supported these findings, confirming that ASXL1 mutations remained associated with a lower CR rate and shorter DFS, OS, and EFS among ELN Favorable patients after adjusting for prognostic pretreatment characteristics and gene mutations. Further exploratory analyses suggested that ASXL1 mutations have a strong negative prognostic impact in older CN-AML patients with mutated CEBPA, including those with double CEBPA mutations, but these results require confirmation because of the small patient numbers. Among ELN Intermediate-I older CN-AML patients, ASXL1 mutations did not impact outcomes. The ELN Intermediate-I group consists of 3 genetic subsets (mutated NPM1 and FLT3-ITD, wild-type NPM1 and FLT3-ITD, and wild-type NPM1 without FLT3-ITD).²⁷  All but 3 ELN Intermediate-I, ASXL1-mut patients belonged to the last of these subsets (Figure 1B), reflecting the negative association of ASXL1 mutations with NPM1 mutations and FLT3-ITD and precluding further subgroup analyses.

Our results exemplify how novel genetic markers can be useful to refine existing, clinically applicable classification schemes. In this respect, our findings add to our previous report on TET2 mutations, which included both younger and older CN-AML patients, and also identified a subset of the ELN Favorable group with inferior outcomes.²¹  When we combined our data on ASXL1 and TET2 mutations with the established ELN classification system, we identified a genetically defined subgroup of older primary CN-AML patients (ELN Favorable, ASXL1-wt, TET2-wt) that achieved a 3-year OS of 39% after cytarabine/daunorubicin-based chemotherapy regimens as used in our trials (supplemental Figure 3A). In contrast, all ELN Favorable/ASXL1-mut patients died of their disease within 18 months from study inclusion. Thus, ASXL1 mutations identify a previously unrecognized high-risk subgroup among older CN-AML patients who would be categorized as low-risk based on currently used classification systems. ELN Favorable/ASXL1-mut patients might therefore be considered candidates for experimental treatment approaches.

Recently, our group reported that, in older CN-AML patients, high expression of BAALC (measured in PB) is associated with a lower CR rate and shorter DFS and OS and that high ERG expression is associated with shorter OS.⁴⁰  Information on BAALC and ERG expression in PB was unavailable for one-third of patients in the present study because of lack of adequate material. This, along with the limited number of ASXL1-mutated patients, precluded inclusion of BAALC and ERG expression levels in our multivariable analyses. It is therefore unclear whether BAALC or ERG expression offers additional prognostic information in patients characterized for ASXL1 mutation status, or vice versa. However, prospective clinical use of quantitative measurements of gene expression requires extensive standardization and calibration efforts,⁴¹  whereas analyses of gene mutations can be more easily transferred into routine use. Therefore, ASXL1 mutations may be a clinically valuable prognostic marker in older patients.

Similar to previous studies in patients with various myeloid neoplasias,^4-9  one specific variant (c.1934dupG) accounted for more than half of the ASXL1 mutations we found. A variety of other frame shift and nonsense mutations were detected in the remaining patients. One previous report raised doubts as to the pathogenic relevance of the c.1934dupG mutation by suggesting it was a false-positive finding, possibly representing an artifact introduced during PCR amplification.⁴²  We believe that c.1934dupG is indeed a true somatic mutation based on 3 lines of evidence. First, using a DNA polymerase enzyme that is resistant to this particular type of artifacts for PCR,⁴³  we reproducibly observed c.1934dupG in patients' BM or PB samples, but not in germline DNA from matched buccal swabs (supplemental Figure 4). Second, c.1934dupG, like other ASXL1 mutations, was strongly associated with older age, wild-type NPM1, and absence of FLT3-ITD. These associations, which were statistically highly significant and consistent between our study and other reports,^7-9  would be unlikely to occur if c.1934dupG was a technical artifact occurring ex vivo. Moreover, outcomes of older CN-AML patients with c.1934dupG were not significantly different from those with other ASXL1 mutations (data not shown). Third, ASXL1 mutations were characterized by a common gene-expression signature, which was shared between patients with c.1934dupG and other ASXL1 mutations, but was distinct from ASXL1-wt.

The function of ASXL1 in normal hematopoiesis and the role of ASXL1 mutations during leukemogenesis are not well understood. Asxl1 knockout in mice leads to defects in B- and T-lineage lymphopoiesis but does not cause severe myelodysplastic changes or leukemias.⁴⁴  However, this model may not adequately reflect the situation in human myeloid cancers, where heterozygous ASXL1 mutations are predicted to result in a truncated protein with possible gain-of-function or dominant-negative effects.⁴⁴  The ASXL1 protein modulates gene expression induced by retinoic acid receptors (RARs) and peroxisome proliferator-activated receptors, at least in part through affecting histone methylation.^45,46  Thereby, ASXL1 is involved in pathways controlling cell proliferation and differentiation in the hematopoietic system and other tissues.⁴⁴  Most reported ASXL1 mutations are predicted to truncate the protein before the RAR-binding domain (Figure 1A), but it is largely unknown how this affects gene expression in AML. We thus studied ASXL1 mutation-associated gene- and miR-expression profiles, using precautions to avoid confounding by other gene mutations, and identified an ASXL1 mutation-associated gene-expression signature composing 67 named genes. These include several genes involved in pathways related to hematopoietic stem cell maintenance, such as the WNT and CXCL12-CXCR4 pathways.

Two other groups previously studied gene-expression signatures associated with ASXL1 mutations in myeloid disorders.^6,47  Boultwood et al, in a cohort of 36 patients with myelodysplastic syndromes or chronic myelomonocytic leukemia, identified 30 differentially expressed genes, and Ingenuity Pathway Analysis suggested deregulation of the RAR pathway.⁶  Thol et al applied Gene Set Enrichment Analysis (a method similar to the gene-set analysis algorithm we used) in 30 AML patients, and found activation of the immune response pathway and down-regulation of cell adhesion and phosphatase pathways.⁴⁷  In our larger set of patients, we did not find deregulation of these or closely related pathways. Because both previous reports^6,47  provide no lists of individual differentially expressed genes, we are unable to assess their concordance with our own results in more detail. Finally, we did not detect a significant miR-expression signature associated with ASXL1 mutations, suggesting that deregulation of miRs is not a prominent feature of CN-AML with mutated ASXL1.

In conclusion, we have shown that ASXL1 mutations are rare in younger primary CN-AML patients, but that they affect 16.2% of patients 60 years of age or older. ASXL1 mutations are mainly found in older CN-AML patients without NPM1 mutations or FLT3-ITD. We found that ASXL1 mutations are associated with mutated CEBPA, and we report signatures of differentially expressed genes in ASXL1-mut CN-AML. Importantly, we demonstrated that ASXL1 mutations associate with inferior response rates and survival, particularly in the ELN Favorable genetic category. Thereby, ASXL1 mutations may be useful to refine existing genetic classification systems and identify a group of older patients who may be considered candidates for experimental treatment approaches.

The authors thank Donna Bucci and the Cancer and Leukemia Group B Leukemia Tissue Bank at The Ohio State University Comprehensive Cancer Center, Columbus, OH, for sample processing and storage services, and Lisa J. Sterling and Dr Colin G. Edwards for data management.

This work was supported in part by the National Cancer Institute (grants CA101140, CA114725, CA140158, CA31946, CA33601, CA16058, CA77658, CA129657, and CA41287), the Coleman Leukemia Research Foundation, Deutsche Krebshilfe–Dr Mildred Scheel Cancer Foundation (H.B.), and John B. and Jane T. McCoy Chair in Cancer Research (G.M.).

National Institutes of Health

Contribution: K.H.M., K. Maharry, M.D.R., J.K., K. Mrózek, G.M., and C.D.B. designed the study, analyzed the data, and wrote the manuscript; K.H.M., H.B., S.P.W., Y.-Z.W., and S.S. carried out laboratory-based research; K. Maharry, M.D.R., J.K., and D.N. performed statistical analyses; B.L.P., T.H.C., M.W., J.O.M., J.E.K., M.R.B., A.J.C., R.A.L., M.A.C., G.M., and C.D.B. were involved directly or indirectly in the care of patients and/or sample procurement; and all authors approved the final version of the manuscript.

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

A complete list of the Cancer and Leukemia Group B institutions, principal investigators, and cytogeneticists participating in this study can be found in the online supplemental Appendix.

Correspondence: Clara D. Bloomfield, The Ohio State University, Comprehensive Cancer Center, 1216 James Cancer Hospital, 300 West 10th Ave, Columbus, OH 43210; e-mail: clara.bloomfield@osumc.edu.