Genetic Characterization and Prognostic Relevance of Acquired Uniparental Disomies in Cytogenetically Normal Acute Myeloid Leukemia

Abstract

Purpose:

Uniparental disomy UPD is a way cancer cells duplicate a mutated gene causing loss of heterozygosity (LOH). Patients with cytogenetically normal acute myeloid leukemia (CN-AML) do not have microscopically detectable chromosome abnormalities, but can harbor UPDs. We examined the prognostic significance of UPDs and frequency of LOH in CN-AML patients.

Experimental Design:

We examined the frequency and prognostic significance of UPDs in a set of 425 adult de novo CN-AML patients who were previously sequenced for 81 genes typically mutated in cancer. Associations of UPDs with outcome were analyzed in the 315 CN-AML patients younger than 60 years.

Results:

We detected 127 UPDs in 109 patients. Most UPDs were large and typically encompassed all or most of the affected chromosome arm. The most common UPDs occurred on chromosome arms 13q (7.5% of patients), 6p (2.8%) and 11p (2.8%). Many UPDs significantly co-occurred with mutations in genes they encompassed, including 13q UPD with FLT3-internal tandem duplication (FLT3-ITD) (P<0.001), and 11p UPD with WT1 mutations (P=0.02). Among patients younger than 60 years, UPD of 11p was associated with longer overall survival (OS) and 13q UPD with shorter disease-free survival (DFS) and OS. In multivariable models that accounted for known prognostic markers including FLT3-ITD and WT1 mutations, UPD of 13q maintained association with shorter DFS, and UPD of 11p maintained association with longer OS.

Conclusions:

LOH mediated by UPD is a recurrent feature of CN-AML. Detection of UPDs of 13q and 11p might be useful for genetic risk stratification of CN-AML patients.

Introduction

Acute myeloid leukemia (AML) is a genetically heterogeneous disease. The prognosis of AML patients is strongly influenced by chromosomal aberrations and gene mutations, and patients are prognostically stratified based on the results of gene sequencing and cytogenetic analysis (1–3). Although genetic risk stratification of AML patients has been well-studied, there remains room for improvement, especially for the cytogenetically normal (CN) group which comprises 40–45% of adult patients. For CN-AML patients, the 2017 European LeukemiaNet (ELN) genetic risk stratification guidelines are determined entirely by gene mutations (i.e. mutation of NPM1, RUNX1, ASXL1 and TP53; bi-allelic CEBPA mutations; and FLT3-internal tandem duplication [ITD] allelic ratio).

CN-AML patients by definition do not have chromosomal abnormalities detectable by karyotyping (3), but these patients often harbor acquired uniparental disomies (UPDs, also called copy-neutral loss of heterozygosity) (4–10), which are somatic losses of a chromosome or segment with duplication of the homologous chromosome or segment. In cancer development UPDs have been shown to mediate loss of heterozygosity (LOH) of mutated oncogenes and tumor suppressor genes by duplicating the mutant alleles thus resulting in homozygous mutations.

The prognostic relevance of UPD in CN-AML has not been adequately investigated, although there is evidence that specific UPDs can impact AML patient outcome (11–20). To better define the utility of UPDs for prognostic risk stratification of CN-AML patients, we herein screened a large cohort of adult de novo CN-AML patients for UPDs and investigated their associations with patient survival and relationship with recurrent gene mutations.

Materials and Methods

Patients and cytogenetic analysis

Studies were conducted using samples obtained from a cohort of 425 adults with de novo AML aged 17–79 years (median 52 years), with 315 patients younger than 60 years. Patients with acute promyelocytic leukemia, AML secondary to myelodysplastic syndromes, or therapy-related AML, and patients who received allogeneic hematopoietic stem cell transplantation in first complete remission (CR) were not included in the study. This study was limited to patients with CN-AML. Pretreatment cytogenetic analyses of bone marrow samples of all patients were performed by the Cancer and Leukemia Group B (CALGB)-approved institutional laboratories using short-term (24–48 hours) unstimulated cultures, and all karyotypes were centrally reviewed (21). In each case, ≥20 metaphase cells were analyzed and no clonal abnormality was found. All patients were similarly treated on CALGB trials (21–33) and did not die within 30 days (see Supplementary Methods for details). Study protocols were in accordance with the Declaration of Helsinki and approved by the institutional review boards. All individuals in this study provided written informed consent.

DNA samples from blood of 1,798 non-leukemic individuals recruited in Columbus, Ohio, USA who had never been diagnosed with any cancer, were used as negative controls for somatic UPD detection. All healthy donors provided written informed consent. The ethnicity of the cases and controls were similar as follows: white European, 90% of cases and 91% of controls; African, 6% of cases and 4% of controls; Asian, 2% of cases and 3% of controls; Hispanic, 2% of cases and 1% of controls; Native American, <1% of cases and controls; Native Hawaiian, <1% of cases and controls; and middle-eastern <1% of cases and controls.

UPD and copy number alteration (CNA) detection

All samples were genotyped with Infinium Omni-1 Quad-bead arrays (Illumina, San Diego, CA) by deCODE Genetics (Reykjavík, Iceland) as previously described (34,35). For quality control, samples with <94% genotyping yield were excluded, and variants were excluded if they had <94% yield, showed significant differences among genotyping batches, or if they significantly (P<10⁻⁶) deviated from Hardy-Weinberg equilibrium. GenomeStudio 2.0 software (Illumina) and the cnvPartition plugin (v3.2.0) were used to detect UPDs and CNAs with minimum probe count set to 10. All calls were manually reviewed by examination of LogR ratio and B-allele frequency plots. Nexus Copy Number v10 (BioDiscovery, El Segundo, CA) was used to visualize UPDs. To differentiate between common inherited regions of homozygosity present in germline DNA and somatically acquired UPDs, 1798 non-leukemic control samples were genotyped. Thirty-three different inherited regions of homozygosity, all of which were smaller than 2Mb in size, were present in ≥1% of samples, and were excluded from the UPD analysis (Supplementary Table S1). These non-leukemic samples did not contain any detectable CNAs or large UPDs.

DNA sequencing

Patient DNA obtained from pretreatment bone marrow or blood samples was sequenced for the following 80 protein coding genes using two separate TruSeq Custom Amplicon panels (Illumina) as described (36): AKT1, ARAF, ASXL1, ATM, AXL, BCL2, BCOR, BCORL1, BRAF, BRD4, BRINP3, BTK, CBL, CCND1, CCND2, CSNK1A1, CTNNB1, DNMT3A, ETV6, EZH2, FBXW7, FLT3, GATA1, GATA2, GSK3B, HIST1H1E, HNRNPK, IDH1, IDH2, IKZF1, IKZF3, IL7R, JAK1, JAK2, JAK3, KIT, KLHL6, KMT2A, KRAS, MAPK1, MAPK3, MED12, MYD88, NF1, NOTCH1, NPM1, NRAS, PHF6, PIK3CD, PIK3CG, PLCG2, PLEKHG5, PRKCB, PRKD3, PTEN, PTPN11, RAD21, RAF1, RUNX1, SAMHD1, SETBP1, SF1, SF3A1, SF3B1, SMARCA2, SMC1A, SMC3, SRSF2, STAG2, SYK, TET2, TGM7, TP53, TYK2, U2AF1, U2AF2, WT1, XPO1, ZMYM3, and ZRSR2. Libraries were prepared according to manufacturer’s instructions, pooled and run on a MiSeq instrument using MiSeq v3 reagent kits (Illumina, San Diego, CA). Sequences were aligned to the hg19 genome build with the Illumina Isis Banded Smith-Waterman aligner. Small indel variants and single nucleotide variants were called using VarScan2 and MuTect, respectively. The Mucor program was used as a baseline for integrative mutation assessment (37). The variant allele fraction (VAF) cut-off was set to 0.1. Variants were considered mutations if they were non-synonymous and not present in the dbSNP v142 database or 1000 Genomes database. All called variants underwent visual inspection of the aligned reads using Integrative Genomics Viewer. We excluded variants sequenced with fewer than 15 reads; variants that only occurred in one read direction (if covered by forward and reverse reads); variants in regions marked by low phred-score bases or low-mapping score reads; variants that occurred in all samples; and samples with generally poor quality sequencing for the entire panel. Samples were considered non-evaluable for a specific gene if ≥85% of the amplicons covering the target regions within the coding sequence of the gene were sequenced to a depth of <15 reads. Detection of FLT3-ITD and determination of the allelic ratio was done with reverse transcriptase polymerase chain reaction as described (38). In addition to the 80 gene sequencing panel, testing for CEBPA mutations was performed with Sanger sequencing as described (39).

Statistical analyses

Associations between UPDs, clinical characteristics and gene mutations were tested using Fisher’s exact tests for categorical variables and the Kruskal-Wallis test for continuous variables. UPDs and gene mutations present in at least eight patients were included in outcome analyses. We used the log-rank test to test for significant associations between categorical variables and survival, with Kaplan-Meier curves for illustration.

We constructed multivariable logistic regression models to analyze the probability of CR attainment and multivariable Cox proportional hazards models for disease-free survival (DFS) and overall survival (OS) using a limited backwards selection procedure. Variables that were significant at the likelihood ratio test P-value <0.20 from univariable models were considered in the multivariable analysis (MVA) (40). Variables considered in univariable models were: age, extramedullary involvement, hemoglobin levels, percentage of blood and bone marrow blasts, platelet count, race, sex, white blood cell (WBC) count, UPD of 11p, UPD of 13q, FLT3-ITD status, FLT3-tyrosine kinase domain mutation status (FLT3-TKD), biallelic mutation of CEBPA status, and mutation status of ASXL1, BCOR, DNMT3A, GATA2, IDH1, IDH2, NPM1, NRAS, PTPN11, RAD21, RUNX1, SMC1A, SMC3, TET2, WT1 and ZRSR2. Data collection and statistical analyses were performed by the Alliance Statistics and Data Center. Analyses were performed using SAS 9.4 and TIBCO Spotfire S+ 8.2, with the database locked on July 5, 2018.

Results

We assessed a cohort of 425 adult patients with de novo CN-AML for UPDs in the autosomes using genotyping arrays. There were 171 UPDs detected in 116 patients (Fig. 1 and Supplementary Table S2). Many UPDs encompassed entire chromosome arms, and the chromosome arms that most frequently contained UPDs were 13q (present in 32 patients), 11p (12 patients), and 6p (12 patients; Fig. 1). To begin to assess the clinical relevance of these UPDs, we grouped UPDs by chromosome arm and determined the associations between the most common UPDs and patient baseline clinical characteristics. UPD of 13q was associated with higher WBC counts (median, 52.5 vs 29.6×10⁹/L; P=0.02), blood blasts (77% vs 58%; P=0.004) and bone marrow blasts (80% vs 69%; P=0.005); UPD of 11p was associated with lower platelet counts (median, 41 vs 60×10⁹/L; P=0.03) and higher blood blasts (85% vs 60%; P=0.002); and UPD of 6p was associated with higher bone marrow blasts (median, 86% vs 70%; P=0.02) (Supplementary Table S3).

Figure 1.

Uniparental disomies (UPDs) in 425 adult patients with de novo cytogenetically normal acute myeloid leukemia. Each orange line represents a single UPD in a single patient. The locations of recurrently mutated genes (mutated in at least 2% of patients) are shown.

UPDs are associated with mutations in genes they encompass

We were able to assess associations between UPDs and gene mutations as these CN-AML patients were previously sequenced for mutations in 81 cancer and/or leukemia-associated genes (Supplementary Table S4) (36). Many UPDs were found to co-occur in the same patients with mutations in genes they encompassed. Specifically, patients with FLT3-ITD (located at 13q12.2) more frequently harbored UPDs of 13q compared to patients without FLT3-ITD (P<0.001; Table 1). Similarly, patients with mutations in the RUNX1 gene (located at 21q22.12) had UPD of 21q significantly more often than patients with wild-type RUNX1 (P<0.001), and patients with EZH2 (located at 7q36.1) mutations had UPD of 7q significantly more often than those with wild-type EZH2 (P<0.001; Table 1). Likewise, there were significant co-associations between mutation of CBL (located at 11q23.3) with UPD of 11q (P=0.01); mutation of WT1 (located at 11p13) with UPD of 11p (P=0.02); bi-allelic mutation of CEBPA (located at 19q13.11) with UPD of 19q (P=0.02); mutation of SF3B1 (located at 2q33.1) with UPD of 2q (P=0.03); and mutation of TET2 (located at 4q24) with UPD of 4q (P=0.04, Table 1). Expectedly, the vast majority of gene mutations that co-occurred with UPDs had VAFs >0.5 (Supplementary Table S5).

Table 1.

Number of adult patients with cyotgenetically normal acute myeloid leukemia who had co-occuring mutations and uniparental disomies.

Gene (location)	Mutation Status	UPD
		Yes	No	P-valuea
FLT3-ITD	present	30	129	< 0.001
(13q12.2)	absent	2	263
RUNX1	mutated	4	33	< 0.001
(21q22.12)	wild-type	2	386
EZH2	mutated	3	10	< 0.001
(7q36.1)	wild-type	4	408
CBLb	mutated	2	6	0.01
(11q23.3)	wild-type	6	411
WT1	mutated	4	38	0.02
(11p13)	wild-type	8	375
CEBPAc	mutated	2	56	0.02
(19q13.11)	wild-type	0	349
SF3B1	mutated	1	10	0.03
(2q33.1)	wild-type	0	414
TET2d	mutated	3	59	0.04
(4q24)	wild-type	3	360
IDH2	mutated	2	56	0.09
(15q26.1)	wild-type	2	365
DNMT3A	mutated	3	164	0.68
(2p23.3)	wild-type	3	255

Abbreviations: UPD, uniparental disomy; FLT3-ITD, internal tandem duplication of the FLT3 gene.

Associations between UPDs and gene mutations are shown for all genes in which at least one patient had a co-occurring mutation and UPD. For each gene, the numbers of patients with and without a UPD that encompasses the gene are listed, stratified by mutation status.

^aFisher’s exact tests were used to calculate P-values.

^b One patient with a UPD on 11q that did not encompass CBL is counted as not having an 11q UPD.

^c Only patients with bi-allelic CEBPA mutations are included.

^d Two patients with UPDs on 4q that did not encompass TET2 are counted as not having 4q UPDs.

UPD of 11p is associated with improved outcome and UPD of 13q with poor outcome

Because of differences in treatment intensity between older and younger adult AML patients enrolled onto CALGB/Alliance treatment protocols, outcome studies are typically performed separately in younger (<60 years of age) and older patients (≥ 60 years of age). Between the 315 younger patients and the 110 older patients in our study, the sample size for examining associations with CR status, DFS and OS for 11p UPD and 13q UPD was adequate only for the younger patients.

We found that UPD of 11p was associated with longer OS (P=0.02; Fig. 2A), and UPD of 13q was associated with both shorter DFS (P<0.001) and shorter OS (P<0.001; Fig. 2B and C) in younger patients. MVA was used to examine the effect of UPDs on outcome in the context of known prognostic markers. For OS, the risk of death was lower for patients with 11p UPD (P=0.04) after adjusting for age (P=0.005), FLT3-ITD (P<0.001), and mutations in FLT3-TKD (P=0.04), DNMT3A (P=0.003), RUNX1 (P<0.001) and WT1 (P<0.001; Table 2). For DFS, patients with 13q UPD had a higher risk of relapse or death (P=0.009) after adjusting for hemoglobin levels (P=0.02), FLT3-ITD status (P<0.001), and mutations in the DNMT3A (P<0.001), RUNX1 (P=0.002) and WT1 (P=0.03) genes (Table 2).

Figure 2.

Associations between uniparental disomies (UPDs) and outcome in younger patients (aged <60 years) with cytogenetically normal acute myeloid leukemia. A, Overall survival of patients with and without 11p UPD. B, Overall survival and C, disease-free survival of patients with and without 13q UPD. D, Disease-free survival of patients who have FLT3 internal tandem duplications (FLT3-ITD) and either harbor 13q UPD or do not.

Table 2.

Multivariable analyses for disease-free survival and overall survival performed in younger (aged <60 years) patients with cytogenetically normal acute myeloid leukemia

Variables		Disease-free survival (n=266)			Overall survival (n=315)
		P-value	HR (95% CI)		P-value	HR (95% CI)
Age, continuous					0.005	1.20 (1.06–1.37)
DNMT3A, mut vs wt		<0.001	1.94 (1.41–2.67)		0.003	1.58 (1.17–2.14)
FLT3-ITD, yes vs no		<0.001	2.31 (1.67–3.19)		<0.001	2.08 (1.55–2.80)
FLT3-TKD, mut vs wt					0.04	0.52 (0.28–0.97)
HG, continuous		0.02	0.91 (0.84–0.98)
RUNX1, mut vs wt		0.002	2.82 (1.44–5.52)		<0.001	3.82 (2.17–6.75)
UPD of 11p, yes vs no					0.04	0.35 (0.13–0.98)
UPD of 13q, yes vs no		0.009	2.10 (1.20–3.67)
WT1, mut vs wt		0.03	1.70 (1.04–2.77)		<0.001	2.62 (1.70–4.03)

Abbreviations: UPD, uniparental disomy; CI, confidence interval; HG, hemoglobin levels; HR, hazard ratio; FLT3-ITD, internal tandem duplication of the FLT3 gene; mut, mutated; n, number; FLT3-TKD, tyrosine kinase domain mutation in the FLT3 gene; wt, wild-type

A HR >1 (< 1) corresponds to a higher (lower) risk for first category listed of a dichotomous variable or higher values of a continuous variable. A limited backward selection technique was used to build the final models with variables that were significant at the likelihood ratio test P-value <0.20 from univariable models for each outcome. For DFS those variables were: HG, platelet count, white blood cell count, UPD of 11p, UPD of 13q, FLT3-ITD status, biallelic mutation of CEBPA status, and mutation status of BCOR, DNMT3A, FLT3-TKD, GATA2, RUNX1, SMC1A, and WT1. For OS those variables were: age, HG, white blood cell count, UPD of 11p, UPD of 13q, FLT3-ITD status, biallelic mutation of CEBPA status, and mutation status of DNMT3A, FLT3-TKD, GATA2, RUNX1, SMC1A, WT1 and ZRSR2.

UPD of 13q is associated with shorter DFS in patients with FLT3-ITD

Since both FLT3-ITD and 13q UPD were associated with shorter DFS, and co-occurred in the same patients, we sought to determine if 13q UPD has any additional utility as a prognostic marker independent of its association with FLT3-ITD by performing outcome analyses for 13q UPD in only the 112 younger CN-AML patients who harbored FLT3-ITD. Patients with 13q UPD still had significantly shorter DFS (P=0.004) in the FLT3-ITD-positive group, demonstrating that 13q UPD status is useful for prognostic stratification even when considering the known prognostic marker FLT3-ITD (Fig. 2D). Expectedly, all of the patients with both FLT3-ITD and 13q UPD had a FLT3-ITD allelic ratio >.5 (i.e. FLT3-ITD^high), supporting the view that the UPD caused LOH for FLT3-ITD (Supplementary Table S6).

Small copy number alterations are present in CN-AML patients

In addition to UPD analysis, the genotyping data from these patients allowed us to screen for microdeletions too small to be detected microscopically by karyotyping. Among the 425 CN-AML patients there were 28 deletions and 3 amplifications that ranged in size from ~100kb to ~19Mb (Supplementary Table S7). Most of the CNAs were observed in only one sample each, with six occurring in two patients each (Supplementary Table S7). Because of the infrequency with which these CNAs were observed, we were unable to assess their associations with clinical or molecular characteristics.

Discussion

The use of genotyping arrays to assess UPD in AML was first described over a decade ago, and subsequent studies have established that UPDs occur in ~20% of CN-AML patients (5,13). In the time since these initial studies assessing the frequency and prognostic relevance of UPDs in AML, the focus of genetic risk stratification has largely shifted beyond cytogenetics to gene mutations, with some additional studies exploring the prognostic relevance of expression signatures and epigenetic changes (41–48). Our work validates earlier efforts that reported associations between 11p and 13q UPDs with AML patient outcome (including CN-AML), (5,13,14,16,49) and again highlights the importance of UPDs in CN-AML. We believe our sample set is the largest series of CN-AML patients screened for UPDs to date, and the finding that both 11p and 13q UPDs were separately associated with patient outcome in multivariable models suggests that these UPDs are clinically relevant prognostic markers for younger CN-AML patients treated with standard “7+3” induction therapy.

Biologically, UPDs are a mechanism by which a cancer cell can duplicate an activating mutation in an oncogene or eliminate the wild-type copy of a tumor suppressor gene to augment the mutation-associated phenotype, such as multiplying an increase in growth advantage caused by a loss of function mutation in a tumor suppressor gene. Many of the recurrent UPDs identified in our study were associated with pathogenic mutations in genes residing on the same chromosome arms including FLT3 (13q), RUNX1 (21q), EZH2 (7q), CBL (11q), WT1 (11p), CEBPA (19q), SF3B1 (2q), and TET2 (4q) (Table 1). Although 12 patients had UPDs that encompassed most of chromosome 6p, we did not identify any recurrently mutated genes on 6p. Notably the only gene located on chromosome 6p on our sequencing panel was HIST1H1E, which was only mutated in two of the 425 CN-AML patients, neither of whom had a UPD of 6p.

Our findings validate previous reports of 7q UPDs causing biallelic inactivation of the EZH2 tumor suppressor gene, although both oncogenic and tumor suppressor roles have been ascribed to EZH2 in AML and its progression from myelodysplastic syndromes (50–51). Likewise UPD-mediatied LOH in RUNX1-mutated AML patients has been recently described and was shown to be associated with poor outcome (52,53). We were unable to assess the relationship between UPD of 21q and outcome in our samples due to insufficient numbers. We found 11p UPD was associated with WT1 mutation status, and also associated with improved DFS. However, WT1 mutation is a known prognostic marker associated with poor outcome (54,55). In our MVA for OS, WT1 mutations and 11p UPD were associated with shorter and longer OS, respectively, indicating it is likely that 11p UPDs mediate an effect on OS independently from their association with WT1 mutations.

We found that patients with FLT3-ITD who also had 13q UPD had a FLT3-ITD^high allelic ratio, which is consistent with 13q UPD as a common mechanism for acquisition of FLT3-ITD^high (56). Our study also validates reports that FLT3-ITD^high is associated with poor outcome, and supports the inclusion of FLT3-ITD^high in the 2017 ELN genetic risk classification system (57). We found that among the younger CN-AML patients who harbor FLT3-ITD, 13q UPD was still associated with poor outcome. Finally, 13q UPD was associated with poor outcome in a MVA that included FLT3-ITD.

Our analysis of CNAs did not identify any frequently occurring CNAs in these cases. We hypothesized that CN-AML patients might harbor micro-CNAs smaller than 5Mb in critical genes, which would not be detectable microscopically by karyotyping. Several recurrent copy number alterations have previously been reported in CN-AML but they are generally rare. These include deletions encompassing TET2 (4q24), ETV6 (12p12.3), TP53 (17p13.1) and NF1 (17q11.2), and amplifications encompassing MYC (8q23.2) and KMT2A (11q23.3) (7,13,58,59). Notably we detected deletion of NF1 in two patients and deletion of TET2 in one patient (Supplementary Table S7). However, our results indicate that micro-CNAs are not a defining feature of CN-AMLs.

Our data provide an overview of recurrent UPDs found in CN-AML and show UPDs are frequently associated with mutations in genes they encompass, leading to LOH. Together our work validates the prognostic importance of 13q UPD and 11p UPD in CN-AML patients, and indicates these UPDs could be clinically relevant for risk stratification of younger de novo CN-AML patients treated with standard induction therapy.

Statement of Translational Relevance.

Acquired uniparental disomy (UPD) is recognized as a common mechanism by which cancer cells can achieve homozygous mutations in oncogenes or tumor suppressor genes. UPDs frequently occur in cytogenetically normal acute myeloid leukemia (CN-AML) patients, and specific UPDs are reportedly associated with patient outcome. Our study reports the largest CN-AML patient set screened for UPDs, to our knowledge. Consistent with previous reports, we found that UPDs often co-occur with mutations in genes they encompass resulting in loss of heterozygosity, including UPD of 13q with FLT3 internal tandem duplication and UPD of 11p with WT1 mutations. In the CN-AML patients younger than 60 years, we found UPD of 13q and UPD of 11p were associated with shorter and longer survival, respectively, even in multivariable models that accounted for known prognostic markers. These results imply that genetic risk stratification of younger CN-AML patients could be improved by the inclusion of UPD testing.