Distinct clinical and biological features of de novo acute myeloid leukemia with additional sex comb-like 1 (ASXL1) mutations

Recent advances in the research of acute myeloid leukemia (AML), especially the identification of novel genetic mutations, have enabled us to stratify this heterogeneous disease entity into distinct subtypes beyond the scopes of cytomorphology and cytogenetics. The progress not only brings us insight into the pathogenesis of AML but also helps refine the treatment strategies for this group of patients.¹ 

Recent studies have identified mutations of the additional sex comb-like 1 (ASXL1) gene in various types of myeloid malignances, including myelodysplastic syndrome (MDS), myeloproliferative neoplasia, MDS/myeloproliferative neoplasia, and primary and secondary AML.^2-6 ASXL1 is a human homologue of the additional sex combs (Asx) gene of Drosophila and is mapped to chromosome 20q11, a region commonly involved in cancers.⁷  This gene is highly conserved among multiple species. It consists of an N-terminal ASX homology domain and a C-terminal plant homeodomain (PHD) zinc finger region.^7,8  In Drosophila, Asx is categorized as a member of the enhancer of trithorax and polycomb group because mutations of Asx enhance the homeotic transformation of both trithorax and polycomb gene mutations.^9,10  In humans, the exact functions of ASXL1 remain to be defined, but it can act as a ligand-dependent coactivator for retinoic acid receptor through binding with steroid receptor coactivator-1, a nuclear receptor coactivator.¹¹  Moreover, ASXL1 is involved in regulation of histone methylation by cooperation with heterochromatin protein-1 to modulate the activity of LSD1,¹²  a histone demethylase for H3K4 and H3K9.¹³ 

Given the tremendous effect of aberrant epigenetic modification in the pathogenesis of AML,^14,15  we expect that ASXL1 mutation might play an important role in leukemogenesis. Although ASXL1 mutation has been reported in de novo AML, the number of patients studied was relatively small (< 50 at most), and the characteristics of AML bearing this mutation and its prognostic effect have not been comprehensively studied till now. We thus examined this mutation in 501 patients with de novo AML and found that ASXL1 mutation occurred with similar frequency in patients with a normal karyotype (NK) and patients with abnormal cytogenetics (8.9% vs 12.9%; P = .179), and it was associated with distinct clinical and biological characteristics that have not been reported yet. Moreover, sequential studies of ASXL1-mutated in 269 samples from 127 patients during the clinical follow-ups showed that 2 of the 6 patients with relapsed ASXL1 mutation studied lost the original mutation at relapse or refractory status or both, whereas 2 patients with ASXL1-wild (2 of 109) acquired novel mutations during disease progression, suggesting instability of this gene mutation during disease evolution.

Six hundred seventy-four adults with de novo AML newly diagnosed were seen in the National Taiwan University Hospital from 1995 to 2007. The diagnosis of AML was based on French-American-British (FAB) classification system. Among them, 501 patients (74.3%) had available cryopreserved bone marrow samples and complete clinical data for this study. These patients were representative of the whole patients because their clinical features and treatment outcomes were similar to those of the total 647 patients (data not shown). In this analysis, the patients with antecedent history of MDS or therapy-related AML were excluded because the pathogenesis and survival may differ significantly from de novo AML. In these 501 patients, 360 (71.8%) had received standard intensive chemotherapy. The 326 non-M3 patients received idarubicin 12 mg/m² per day on days 1-3 and cytarabine 100 mg/m² per day on days 1-7 and then 2-4 courses of consolidation chemotherapy with high-dose cytarabine (2000 mg/m² every 12 hours for days 1-4, total 8 doses), with or without one anthracycline idarubicin (12 mg/m²) or mitoxantrone (12 mg/m²) after complete remission (CR) was achieved. The 34 patients with acute promyelocytic leukemia (APL; M3 subtype) received concurrent all-trans retinoic acid and chemotherapy (idarubicin 12 mg/m² per day on days 1-2) as induction therapy. Consolidation chemotherapy with anthracycline-based regimen was given for 3 courses after remission was achieved. The patients were treated on the basis of the consensus of the hematologists in this institute, but not on clinical trials. Mild difference might exist according to the patients' clinical conditions, but we always managed to keep the homogeneity in the treatment of this group of patients. The bone marrow samples were collected at diagnosis and sequentially during follow-ups. Mononuclear cells were isolated by Ficoll-Hypaque gradient and preserved as previously described.¹⁶  All the patients have signed informed consents for sample collection in accordance with the Declaration of Helsinki. This study was approved by the Institutional Review Board of the National Taiwan University Hospital.

Determination of mutations in various genes, including FLT3-ITD (internal tandem duplication), FLT3-TKD (tyrosine kinase domain), MLL-PTD (partial tandem duplication), CEBPA, NPM1, PTPN11, NRAS, KRAS, JAK2, KIT, AML1 (RUNX1), WT1, and IDH1 was performed as described previously.^17-23  The ASXL1 exon 12 until the stop codon was amplified by 3 pairs of primers and sequenced by another 6 internal primers, as described by Gelsi-Boyer et al,²  with mild modification. Totally, 3075 base pairs were covered. The sequences of these polymerase chain reaction (PCR) and sequencing primers are listed in supplemental Table 1 (available on the Blood Web site; see the Supplemental Materials link at the top of the online article), with the expected length of the PCR products shown. The reference sequence was NM_015338 for mRNA and NP_056153 for protein. In every 35 μL of PCR reaction, there were 50 ng of genomic DNA, 200nM each deoxynucleoside triphosphate, 2mM magnesium sulfate, 200nM each forward and reverse primers, buffer, and 1 U of YEAtaq DNA polymerase (Yeastern Biotech). The PCR reaction included 95°C for 2 minutes, followed by 35 cycles of 95°C for 30 seconds, 61°C for 30 seconds, and 72°C for 1 minute. All the mutations reported in this study were confirmed at least twice. When the mutations were not obvious because of location near the sequencing primers, sequencing from the other direction was done to solve this issue.

A panel of monoclonal antibodies was used for determination of immunophenotype of leukemia cells, including myeloid-associated antigens (CD13, CD33, CD11b, CD15, CD14, and CD41a), lymphoid-associated antigens (CD2, CD5, CD7, CD19, CD10, and CD20), and the lineage nonspecific antigens (human leukocyte antigen [HLA]–DR, CD34 and CD56). The antibodies for CD15, CD16, CD2, and HLA-DR were purchased from BD Biosciences. The other antibodies were purchased from Immunotech. The method was described previously.²¹ 

Bone marrow cells were harvested directly or after 1-3 days of unstimulated culture, and the metaphase chromosomes were banded by the G-banding method as described earlier.²⁴ 

When ASXL1 mutations at diagnosis were no longer seen in relapsed samples by direct sequencing, we cloned the PCR products spanning the original mutation spots by TA cloning (Yeastern Biotech) as described previously,²⁵  followed by sequencing of ≥ 30 individual clones to search for any mutation.

Mann-Whitney tests were used to calculate the significance if the continuous data were not normally distributed while comparing the adult patients with AML with or without ASXL1 mutation. The chi-square test was used to calculate the significance of association between ASXL1 mutation and other parameters, including sex, FAB classification, immunophenotypes, cytogenetics, and mutations of other genes. Fisher exact test was used if any expected value of the contingency table was < 5. Three hundred sixty patients who received standard chemotherapy were included in survival analysis. All the patients in survival analyses were censored at the last follow-up or at the time of allogeneic hematopoietic stem cell transplantation if they received this treatment. Disease-free survival (DFS) was measured from the date when the patients achieved CR to the date of relapse or last follow-up. Multivariate Cox proportional hazard regression analysis was used to investigate independent prognostic factors for overall survival (OS) and DFS. Kaplan-Meier estimation was used to plot survival curves, and log-rank tests were used to calculate the difference of OS and DFS between groups. A P value < .05 was considered statistically significant. All statistical analyses were performed with SPSS 17 software (SPSS Inc).

A total of 501patients with de novo AML were included in this study. There were 216 females and 285 males, with a median age of 52 years (range, 15-90 years). The basic demographic data are listed in supplemental Table 2. ASXL1 mutations, including frame-shift and non–sense mutations, at exon 12, occurred in 54 patients (10.8%) at diagnosis. All these mutations resulted in disruption of the PHD domain of ASXL1 (Figure 1). Other single nucleotide variations, including c.3671G>C (p.R1224T), c.2305A>G (p.T769A), c.3019G>C (p.G1007R), c.2063C>T (p.T688M), c.3221A>T (p.Q1074L), c.2077C>G (p.R693G), c.3416C>A (p.T1139K), c.1954G>A (p.G652S), and c.3215T>A (p.V1072D), detected in 1, 1, 1, 1, 1, 1, 1, 37, and 48 patients, respectively, were not regarded as true mutations because these variations were possible single nucleotide polymorphism. Moreover, the amino acids at these sites are not well conserved among humans, mice, zebrafish, and fugu fish (data not shown). The basic clinical and biological data of the patients with ASXL1 mutations are listed in Table 1.

No significant difference was seen in white cell counts, hemoglobin levels, platelet counts, and serum lactate dehydrogenase levels between patients with and without ASXL1 mutation (data not shown). However, ASXL1 mutations occurred more frequently in older (median age, 66 vs 49 years; P < .001) and male patients (13.7% vs 6.9%; P = .012) and were closely associated with FAB M0 subtype (P = .016), isolated trisomy 8 (P = .011; Table 2), and expression of HLA-DR (P < .001) and CD34 (P = .020; supplemental Table 3). A trend of positive correlation was observed between ASXL1 mutation and t(8;21) (P = .067). The incidence of ASXL1 mutations in patients of ≥ 60 years was 18.0% (34 of 189), compared with 5.4% (20 of 312) in patients< 60 years. In contrast, ASXL1 mutations were inversely related to FAB M1 subtype (P = .003), t(15;17) (P = .025), complex cytogenetics (P = .047; Table 2), and expression of CD33 (P = .036) and CD15 (P = .024; supplemental Table 3). None of the 38 patients with APL and the 19 patients with inv(16) had ASXL1 mutation at diagnosis (Table 2). ASXL1 mutations occurred with similar frequency in patients with a NK and patients with abnormal cytogenetics (8.9% vs 12.9%; P = .179; Table 2).

We also examined the association of ASXL1 mutations with 13 other gene mutations (Table 1; Table 3). Thirty-nine of the 54 patients with ASXL1 mutation had concurrent other gene mutations at diagnosis; 31patients had 1, 7 had 2, and 1 had 3 additional mutations (Table 1). The patients with ASXL1 mutations had significantly lower incidences of NPM1 mutation, WT1 mutation, and FLT3-ITD (3.7% vs 23.0%, 0% vs 7.4%, and 7.4% vs 25.0%, respectively; Table 3). On the contrary, ASXL1 mutation cooperated frequently with RUNX1 mutation (16 of 54, 29.6%, vs 45 of 447, 10.1%; P < .001).

Survival analyses were performed in 360 patients with AML who received standard intensive chemotherapy after diagnosis, as described previously.^26,27  Sixty-nine patients were > 60 years; 17 of them had ASXL1 mutation. The 360 patients included in survival analyses were younger than the total cohort (supplemental Table 2) and had a higher incidence of t(15;17) but lower incidence of complex chromosome abnormalities. No differences were seen in other characteristics between these 2 groups. The median follow-up time of our patients was 53 months (range, 1.0-160 months). Patients with ASXL1 mutations had a lower CR rate than others (15 of 26, 57.7%, vs 267 of 334, 79.9%; P = .013), but the relapse rate was similar between the 2 groups once they obtained a CR (data not shown). Patients with ASXL1 mutation had worse OS (median OS, 14 months vs 58 months; P = .009; Figure 2A; Table 4), compared with patients without ASXL1 mutation, even when patients with APL were excluded (median OS, 14 months vs 34 months; P = .022; Figure 2B; Table 4), but DFS was similar between these 2 groups (data not shown). The same finding was also seen in the subgroup of patients with intermediate-risk cytogenetics (median OS, 10.0 months vs 25 months; P = .009; Figure 2C; Table 4). When the outcome was analyzed separately for the younger and older patients, the OS was not different between the patients with and without ASXL1 mutation in the younger group of patients with age of < 60 years (n = 291; median, 22 ± 9.3 months vs not reached; P = .221), nor in the group > 60 years (n = 69; median, 8 ± 7.5 months vs 11 ± 2.5 months; P = .093). The lack of difference of survival after this age division may be related to the smaller number of patients in each group, compared with the total cohort. In multivariate analysis using covariables that were significantly associated with OS in univariate analysis such as age, karyotypes, NPM1mutation without FLT3-ITD (NPM1⁺/FLT3-ITD⁻), RUNX1 mutation, WT1 mutation, and biallelic CEBPA mutations (Table 4), we could not establish ASXL1 mutation as an independent poor prognostic factor for OS, even when only patients without APL or patients with intermediate cytogenetics were analyzed (Table 5).

We also determined the stability of ASXL1 mutations during clinical follow-ups in 269 samples from 127 patients, 18 with ASXL1 mutations and 109 without (Table 6). Among the 18 patients with ASXL1-mutated, the ASXL1 mutations disappeared in 15 patients (supplemental Figure 1A) but were retained in UPNs (unique patient numbers) 7, 12, and 33 in clinical CR (Table 6). The mutant signals in CR samples from all these 3 patients were less intense (supplemental Figure 1B), compared with those at diagnosis or relapse, suggesting there were still leukemia cells but in a lower amount in the samples, despite the morphologic remission. In consistency with the findings, the NRAS mutation and RUNX1 mutation were also detectable in the CR samples of UPNs 7 and 33, respectively (Table 6). For UPN 12, his ASXL1 mutation was G646WfsX12, which was the most common mutation in this study and was suggested to be acquired because it disappeared in CR in UPNs 16, 17, 21, 23, 24, 30, 40, and 44, so the detection of this mutation in CR samples indicated the presence of leukemia cells. All these 3 patients relapsed shortly, 4.2 months, 4.0 months, and 7.4 months, respectively, after CR was achieved. Among the 6 patients with ASXL1-mutated who had ever experienced disease relapse and had cryopreserved samples for study, 4 patients (UPNs 7, 12, 16, and 33) retained the original mutation at relapse. For the remaining 2 patients (UPNs 30 and 53), the original ASXL1 mutation could not be detected at relapse by direct sequencing (supplemental Figure 1C-D). In UPN 30, the mutation could be shown in 1 of 20 clones by TA cloning, but it was obviously lost as the disease became refractory. The c-KIT mutation site in this patient also changed from exon 8 at diagnosis to exon 17 at disease relapse and refractoriness. In UPN 53, the original ASXL1 mutation could not be detected at relapse by both direct sequencing and TA cloning, and it was replaced by a new non–sense mutation (supplemental Figure 1D). FLT3-TKD mutation in this patient also disappeared at the same time (Table 6). For 109 patients without ASXL1 mutation at diagnosis, 2 (UPNs 55 and 56) acquired a novel ASXL1 mutation (c.1934dupG) at relapse (supplemental Figure 1E; Table 6). Sequencing for 36 and 30 clones, respectively, using the DNA from diagnostic samples of these 2 patients, did not yield any mutant clone, suggesting that either the ASXL1 mutation was newly formed at relapse or was present in low level < 3% at the time of diagnosis. These results indicate that ASXL1 mutation status can change during disease evolution.

Although ASXL1 mutations have been reported in various myeloid malignancies, several important questions remain unsolved.^2-6  First, the incidence of this mutation in de novo AML is still not clearly defined, ranging from 6.5% (3 of 46) to 29.6% (8 of 27) in 2 studies with relatively small number of patients.^5,6  Second, the association of this mutation with karyotypes has also not been well addressed yet. For example, 40% (34 of 86) of patients with MDS/AML with NK were found to have ASXL1 mutation in one study,⁶  but only 13.0% (6 of 46) of such patients had this mutation in another study.⁵  Third, ASXL1 mutation was shown to be mutually exclusive with NPM1 mutation,⁵  but the association of this mutation with other genetic mutations as well as other biological features in ASXL-mutated AML remains unknown. Finally, no studies have investigated the effect of this mutation on clinical outcome.

In this study, we tried to solve these issues by focusing on a large cohort of patients with de novo AML. We found that ∼ 10% of patients with primary AML had ASXL1 mutations. Unlike other common genetic mutations in AML such as FLT3-ITD and mutations of NPM1, CEBPA, and IDH1, which are significantly more common in patients with a NK,^{17,18,21,25,28,29} ASXL1 mutation does not have such preference (17 of 192 or 8.9% in patients with NK vs 33 of 256 or 12.9% in patients with abnormal cytogenetics; P = .179; Table 2). Interestingly, none of the patients of newly diagnosed AML with t(15;17) and inv(16) had ASXL1 mutations, in contrast to the finding that 31.6% of patients with trisomy 8 and 19.5% of patients with t(8;21) had this mutation. Moreover, ASXL1 mutation frequently cooperated with RUNX1 mutation in the same patients but was mutually exclusive with NPM1 mutation (P < .001), similar to the previous report.⁵  It was also inversely associated with WT1 mutation (P = .038) and FLT3-ITD (P = .003), which have not been reported before. The close association of ASXL1 mutation with RUNX1 mutation and a trend to be positively correlated with t(8;21), a translocation resulting in RUNX1-ETO fusion, suggest this mutation may closely cooperate with mutations of RUNX1, which has been implicated in chromatin modification of hematopoietic system, in leukemogenesis.^30,31 

We found that ASXL1 mutation was associated with a worse OS (median, 14 vs 58 months, P = .009 for all patients; 14 vs 34 months, P = .022, for patients with APL; and 10.0 vs 25.0 months, P = .009, for patients with intermediate cytogenetics). In a multivariate analysis that combined age, NPM1⁺/FLT3-ITD⁻, biallelic mutations of CEBAP, WT1 mutation, RUNX1 mutation, and karyotypes as covariables, which were all significantly associated with OS in both univariate and multivariate analyses,^26,28,32-39  we could not establish ASXL1 mutation as an independent prognostic factor whether patients with APL were included or not or when only patients with intermediate cytogenetics were analyzed (Tables 4-5). This may be related to the fact that ASXL1 mutation was highly associated with old age and RUNX1 mutation, 2 poor prognostic factors,^26,32  but other possible reasons await exploration. Although ASXL1 mutations in our analysis did not represent a significant independent factor for prognosis, it is highly associated with a worse survival and lower CR rate. Thus, we can predict a worse prognosis for the patients bearing this mutation. At present, determination of ASXL1 mutations in routine clinical practice may not be necessary, but, as more and more knowledge about this mutation comes out, the situation may change.

Of the 109 patients with ASXL1-wild, 2 acquired a G646WfsX14 mutation of ASXL1 at disease relapse (Table 6; supplemental Figure 1E). This finding implies that ASXL1 mutation can be acquired as diseases evolve into a late stage, similar to the situation in MDS.⁶  In addition, in 1 of the 6 patients with ASXL1-mutated who had relapse and were studied, the ASXL1 mutation detected at diagnosis diminished at relapse and disappeared at refractory status (UPN 30 in supplemental Figure 1C and Table 6), and in the other patient (UPN 53) the mutation was even replaced by a novel non–sense mutation as the disease further evolved (supplemental Figure 1D; Table 6). Thus, this mutation may not be as good as NPM1 and IDH1 mutations^16,25  to serve as a marker for monitoring of minimal residual disease.

The functions of ASXL1 in humans have not been well delineated, but several pieces of evidence indicate a role of this gene in chromatin modification. First, Asx in Drosophila, the homolog of human ASXL1, is a member of enhancer of trithorax and polycomb group genes, both of which are associated with histone modification.^9,10  Second, the PHD domain at the C terminal of ASXL1 is well conserved among different species and can recognize methylated H3K4.^3,7,40-42  Finally, ASXL1 was shown to modulate LSD1, a H3K4 and H3K9 demethylase, in vitro.^12,13  Given the important role of histone methylation in leukemogenesis, we expect that ASXL1 mutation may play an important role in the pathogenesis of AML.

In a recent report, ASXL1 c.1934dupG was suspected not as a somatic mutation but rather as an artifact.⁴³  In that report, DNA from normal tissue (buccal mucosa) from patients with ASXL1-mutated showed c.1934dupG. Moreover, the same alteration was seen in > 25% of samples from patients without obvious blood diseases. Because c.1934dupG is the most common ASXL1 mutation, this issue is critical in assessing the clinical and biological significances of ASXL1 mutations in myeloid neoplasms. Our data, however, strongly suggest that ASXL1 c.1934dupG is a somatic alteration, based on the following reasons. (1) The TA cloning of the PCR products from the samples with this mutation did yield mutant clones. In addition, repeated sequencing from both directions showed c.1934dupG (data not shown). The results from these 2 steps of studies indicate that c.1934dupG is not an artifact from sequencing per se. (2) As mentioned in the results of sequential studies, in the 109 patients without ASXL1 mutation at diagnosis, only 2 acquired c.1934dupG at the time of relapse. That is, in the remaining 107 patients who did not have ASXL1 mutation at diagnosis, including 22 with > 1 relapse, no c.1934dupG was detected in a total of 133 samples taken at the time of relapse, a finding in stark contrast to the report that c.1934dupG could be detected in ≤ 25% of control persons without a hematologic malignancy⁴³  and arguing against the possibility that this mutation was an artifact randomly occurring because of PCR errors. Moreover, all the mutations reported in our study were confirmed at least twice. Therefore, we are quite confident in the mutation results in our study. (3) Patients UPNs 16, 17, 21, 23, 24, 30, 40, and 44 had mutation c.1934dupG at diagnosis, but no more mutation was detected at the time of CR (Table 6). In UPN 16, subsequent relapse was accompanied with the reappearance of c.1934dupG. These results indicate that this mutation is not a random instance (please see supplemental Figure 1A for the sequencing patterns). We have also investigated ASXL1 mutations in 38 additional samples taken at the time of CR from the patients without this mutation at diagnosis, and none of them had detectable mutations (data not shown). This result suggests that c.1934dupG is very unlikely an artifact with a frequency as high as 25% in normal control samples. (4) From our analysis, ASXL1 mutations were highly correlated with several distinct parameters such as older age, male sex, isolated trisomy 8, RUNX1 mutation, and expression of HLA-DR and CD34, but inversely associated with t(15;17), complex cytogenetics, FLT3-ITD, NPM1 mutations, WT1 mutations, and expression of CD33 and CD15. From the view of statistics, ASXL1 cannot be an event occurring by chance on the basis of these statistically significant correlations. (5) To further consolidate our results, we serially diluted one patient's sample containing c.1934dupG by wild-type genomic DNA, followed by PCR of ASXL1 across nucleotide 1934, and found that the mutation peaks progressively decreased toward undetectable at a dilution fold of 50 (1 ng of gDNA containing c.1934dupG in 49 ng of wild-type gDNA; data not shown), indicating that c.1934dupG truly existed in the patient's DNA and did not come from an artifact.

In conclusion, we have performed by far the most comprehensive analysis of ASXL1 mutations in de novo AML and found several distinct clinical and biological characteristics of this disease entity. Further investigations are needed to study the mechanistic significance of this mutation and its cooperation with other genetic alterations in the leukemogenesis.

This work was supported by the National Science Council, Taiwan (grants NSC 96-2628-B002-013-MY2 and 98-2314-B-002-033-MY3), the National Health Research Institute (grant NHRI-EX97-9731BI), DOH99-TD-C-111-001, Department of Health (Taiwan), NTUH 98-S1052, NTUH 98-S1383, and YongLin Healthcare Foundation (W.-C.C.), and by grants NSC 97-2314-B002-015-MY3 and NSC-97-2628-B-002-002-MY3 (H.-F.T.).

Contribution: W.-C.C. and H.-F.T. designed the experiment; W.-C.C., H.-H.H., H.-F.T., and H.-A.H. analyzed the data and wrote the paper; C.-Y.C., Y.-N.H., Y.-C. Chang, F.-Y.L., M.-C.L., C.-W.L., M.-H.T., and C.-F.H. performed the experiment; and J.-L.T., M.Y., W.T., B.-S.K., S.-J.W., S.-Y.H., S.-C.H., and Y.-C. Chen provided important materials.

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

Correspondence: Hwei-Fang Tien, Department of Internal Medicine, National Taiwan University Hospital, Taipei, Taiwan 100; e-mail: hftien@ntu.edu.tw; or Wen-Chien Chou, Department of Laboratory Medicine and Internal Medicine, National Taiwan University Hospital, Taipei, Taiwan 100; e-mail: wchou@ntu.edu.tw.