Integrated genetic and epigenetic analysis revealed heterogeneity of acute lymphoblastic leukemia in Down syndrome.

Children with Down syndrome (DS) are at a 20-fold increased risk for acute lymphoblastic leukemia (ALL). Compared to children with ALL and no DS (non-DS-ALL), those with DS and ALL (DS-ALL) harbor uncommon genetic alterations, suggesting DS-ALL could have distinct biological features. Recent studies have implicated several genes on chromosome 21 in DS-ALL, but the precise mechanisms predisposing children with DS to ALL remain unknown. Our integrated genetic/epigenetic analysis revealed that DS-ALL was highly heterogeneous with many subtypes. Although each subtype had genetic/epigenetic profiles similar to those found in non-DS-ALL, the subtype distribution differed significantly between groups. The Philadelphia chromosome-like subtype, a high-risk B-cell lineage variant relatively rare among the entire pediatric ALL population, was the most common form in DS-ALL. Hypermethylation of RUNX1 on chromosome 21 was also found in DS-ALL, but not non-DS-ALL. RUNX1 is essential for differentiation of blood cells, especially B cells; thus, hypermethylation of the RUNX1 promoter in B-cell precursors might be associated with increased incidence of B-cell precursor ALL in DS patients.

acute lymphoblastic leukemia

acute myeloid leukemia

B‐cell precursor

Down syndrome

high hyperdiploid

Japan Association of Childhood Leukemia Study

multiplex ligation‐dependent probe amplification

Philadelphia chromosome

recognition of outliers by sampling ends

single nucleotide polymorphism

INTRODUCTION

Children with DS harbor one extra copy of chromosome 21. Among other congenital deficits and disease susceptibilities, children with DS are at a 20‐fold increased risk for ALL.1 Compared to ALL in children without DS (non‐DS‐ALL), children with DS and ALL (DS‐ALL) show uncommon genetic alterations,2 such as mutations in JAK2, NRAS, and KRAS as well as CRLF2 overexpression, suggesting that certain tumorigenic or other disease‐related processes could be unique to DS‐ALL. Consistent with distinct ALL pathomechanisms in DS, DS‐ALL patients have generally worse prognosis than non‐DS‐ALL patients. Therefore, it is critical to identify factors contributing to enhance ALL propensity and severity in DS. Genes on chromosome 21 are obvious candidates. Indeed, recent studies have implicated HMGN1 3 and DYRK1A 4 overexpression in DS‐ALL.

To better understand DS‐ALL molecular pathogenesis, we carried out an integrated genetic/epigenetic analysis. Expression and methylation analyses revealed high subtype heterogeneity in DS‐ALL. Although each subtype had expression and methylation profiles similar to corresponding non‐DS‐ALL subtypes, subtype frequency distribution differed significantly between patient groups. The hypermethylation of RUNX1 on chromosome 21 was unique to DS‐ALL, which could explain the increased incidence of BCP‐type ALL in DS.

MATERIALS AND METHODS

Patients and materials

Forty‐three patients with DS‐ALL (Table S1), 38 patients with non‐DS‐ALL (Table S2), and 18 DS patients without ALL or AML (Table S3) were enrolled in this retrospective study. The 43 DS‐ALL patients included one large cohort (n = 31) from the JACLS as well as patients from various hospitals in Japan. The 38 non‐DS‐ALL patients were from the University of Tokyo Hospital. Peripheral blood and bone marrow samples were collected after written informed consent was obtained from legal guardians according to protocols approved by the Human Genome, Gene Analysis Research Ethics Committee of the University of Tokyo, and other participating institutions. All protocols conformed to the tenets of the Declaration of Helsinki.

Next‐generation sequencing

Next‐generation sequencing was carried out using the Illumina HiSeq 2000, 2500, or MiSeq platform (Illumina) with a standard 100‐bp paired‐end read protocol according to the manufacturer's instructions.5

RNA sequencing

High‐quality RNA samples isolated by Agilent TapeStation (Agilent) were available from 25 DS‐ALL patients (RNA integrity number equivalent greater than 5.5). These samples were used to prepare libraries for RNA sequencing using the NEBNext Ultra RNA library prep kit for the Illumina platform (New England BioLabs). Fusion transcripts were detected by Genomon version 2.5.0 and filtered by excluding fusions mapping to repetitive regions. Normalized count data obtained by the variance‐stabilizing transformation function of the R package DESeq2 were used for clustering analysis. Prior to clustering analysis, the data were filtered to remove gene sets deemed unrelated to disease, such as those from sex‐determining regions of X and Y chromosomes and genes (Table S4) from contaminating normal erythrocytes. For clustering analysis, we used Ward's hierarchical clustering method and included 250 of the top 1% of differentially expressed genes (Table S5) extracted using DESeq2. For comparing DS‐ALL and non‐DS‐ALL samples, we also used the open dataset of expression6 (Table S6).

Validation of fusion genes detected by RNA sequencing

Novel in‐frame fusion transcripts were validated by RT‐PCR followed by Sanger sequencing (Table S7).

Detection of the Ph‐like signature

We defined a ROSE gene set (Table S8), including the top 25 genes in clusters R1‐R8.7 Ward's hierarchical clustering method was then used for clustering analysis of the ROSE gene set. As reported,7 samples in cluster R8 showed a signature similar to Philadelphia chromosome‐positive ALL and were thus labeled Ph‐like. To detect other samples with a Ph‐like signature, we undertook hierarchical clustering of 25 DS‐ALL (Table S1) and 118 non‐DS‐ALL samples (Tables S2 and S6) based on genes (Table S9) significantly upregulated (adjusted P < .0001) in samples previously labeled Ph‐like in R8.

DNA methylation analysis

Comprehensive DNA methylation analysis was carried out on 35 DS‐ALL and 24 non‐DS‐ALL samples using the Infinium HumanMethylationEPIC BeadChip (Illumina) for the JACLS samples or Infinium HumanMethylation 450K BeadChip (Illumina) for samples from other sources, all according to the manufacturer's protocols. Beta values were corrected for probe design bias using a beta‐mixture quantile normalization method8 and converted to M values.9 Next, the R package pcaMethods bioconductor was used to impute incomplete M values, which were then converted to beta values. Imputed beta values were later used for further analyses. For the analysis of DNA methylation, the open dataset of DNA methylation profiling in pediatric ALL determined by the Infinium HumanMethylation 450K BeadChip10 was also included. For clustering analysis, we used unsupervised consensus clustering with 8000 probes using Ward's method. Cluster stability was determined by consensus clustering with 1000 iterations using the R package ConsensusClusterPlus. Statistical significance was assessed using Wilcoxon's rank‐sum test, and values were corrected by employing the Benjamini‐Hochberg method.

Bisulfite conversion of RUNX1

To confirm RUNX1 methylation in DS patients without ALL or AML, bisulfite sequencing was carried out using nested primers (forward, 5′‐tcttgaaaagaagaaacagacca‐3′; reverse, 5′‐agtaaattctagcattactccaggga‐3′). Genomic DNA (500 ng) was bisulfite‐modified using the EpiTect Plus DNA bisulfite kit (Qiagen) according to the manufacturer's instructions.

Targeted capture sequencing

We undertook mutation analysis of genes (JAK2, SH2B3, KRAS, NRAS, PTPN11, FLT3, BRAF, NF1, KIT, IKZF1, PAX5, and IKZF1) reported as mutually exclusive oncogenic drivers2 or transcription factors in DS‐ALL (n = 43). The samples from the JACLS (n = 31) were captured and sequenced using SureSelect (Agilent), HaloPlex (Agilent), Nextera (Illumina), or SeqCap (Roche NimbleGen). Polymerase chain reaction‐based targeted deep sequencing was undertaken on other samples (n = 12).

Copy number analysis

DNA samples extracted from peripheral blood or bone marrow were processed using MLPA for the JACLS samples and SNP array analysis for the other samples. The MLPA was carried out using the SALSA MLPA probemix P335‐B1 ALL‐IKZF1 kit according to the manufacturer's instructions (MRCHolland). The SNP array analysis was undertaken using the Affymetrix GeneChip 250K Nsp or CytoScan HD (Affymetrix), according to the manufacturer's protocol.

Mutations in non‐DS‐ALL samples

Among the 118 samples used for RNA sequencing, DNA was available for 42 paired normal samples, which were evaluated using whole‐exome sequencing. The mutations in the remaining 76 samples, for which paired normal DNA was not available, were called by the RNA sequencing results.

RESULTS

Expression analysis

Initial next‐generation sequencing analysis of 25 DS‐ALL samples (Table S1) identified 19 fusions (Table S10), including 15 frequently reported in pediatric BCP‐ALL, as well as 4 novel fusions. Among common fusions previously identified in ALL, P2RY8‐CRLF2 was detected in 9 samples; IGH‐CRLF2 was not detected in our cohort. Among novel fusions, PDGFA‐TTYH3 was detected in 2 samples. Two DS‐ALL samples with t(14;19)(q32;q13)11 and t(8;14)(q11;q32)12 showed high CEBPA and CEBPD expression levels, respectively (Figure S1). These samples were assumed to harbor IGH‐CEBPA and IGH‐CEBPD fusions, respectively, because both karyotypes involved IGH at 14q32. These fusions are rare in non‐DS‐ALL, but could be more common in DS‐ALL.12, 13 To characterize the unique expression profiles of DS‐ALL, we applied hierarchical clustering analysis (Figure 1 and Table S5) including the 25 DS‐ALL and 118 non‐DS‐ALL samples (expression cohort). These non‐DS‐ALL samples included several subtypes, such as the ETV6‐RUNX1 fusion and HeH (Tables S2 and S6). We also identified 8 samples with PAX5 abnormalities in the absence of other genetic alterations, which we defined as “PAX5‐altered.” Unsupervised hierarchical clustering of the entire ALL sample population yielded 6 clusters (E1‐E6), and DS‐ALL samples fell into 4 clusters, with significantly enrichment in E6 (Table S11). Cluster E3 was heterogeneous and contained several ALL subtypes, including MLL‐rearranged, TCF3‐HLF, IGH‐DUX4, MEF2D‐rearranged, and PAX5‐altered. The PAX5‐altered samples fell into clusters E3 and E6. All DS‐ALL samples with ETV6‐RUNX1 fell into cluster E4, which also included all non‐DS‐ALL samples with the ETV6‐RUNX1 fusion. Additionally, E4 included one DS‐ALL sample without ETV6‐RUNX1. Six DS‐ALL fell into cluster E5, which also included most non‐DS‐ALL samples with HeH. Among these 6 DS‐ALL cases, only 1 had HeH. Cluster E6 was characterized by the presence of BCR‐ABL1 fusions, Ph‐like expression profiles, and CRLF2 fusions.

Figure 1

Gene expression clusters in 143 samples of B‐cell precursor acute lymphoblastic leukemia (BCP‐ALL) (25 samples from children with Down syndrome and ALL [DS‐ALL] and 118 non‐DS‐ALL samples) on hierarchical clustering. Hierarchical clustering reveals BCP‐ALL samples are grouped into 6 clusters (E1‐E6). In those 6 clusters, DS‐ALL samples are clustered into 4 clusters. Expression clusters, DS‐ALL samples, subtypes, and genetic aberrations are shown by colors as indicated. DS‐ALL and non‐DS‐ALL are clustered into the same cluster corresponding to each biological subtype. Philadelphia chromosome (Ph)‐like signature is confirmed by recognition of outliers by sampling ends (ROSE) gene set clustering. HeH, high hyperdiploid

To detect the Ph‐like signature, we clustered the expression cohort by gene sets using the ROSE method7 (Figure S2 and Table S8), which revealed that 7 DS‐ALL samples had the Ph‐like signature. Altogether, expression analysis revealed that DS‐ALL samples were highly heterogeneous, but individual subtypes showed expression patterns similar to corresponding non‐DS‐ALL subtypes. Alternatively, the frequency distribution differed between groups.

Although it is reasonable to speculate that genes on chromosome 21 could be associated with DS‐ALL development, surprisingly no genes on chromosome 21, such as HMGN1 or DYRK1A, showed significantly higher expression in DS‐ALL (Table S12).

We then compared DNA methylation status between DS‐ALL and non‐DS‐ALL samples. This methylation cohort included 59 samples of our study cohort (methylation cohort, 35 DS‐ALL and 24 non‐DS‐ALL samples) and 664 samples (14 DS‐ALL samples and 650 non‐DS‐ALL samples) from the open dataset of pediatric ALL.10 In methylation‐based clustering analysis, DS‐ALL samples fell into 5 clusters, M1‐M5 (Figures 2 and S3, Table S13). Clusters M1 and M2 contained mainly HeH, whereas clusters M3 and M5 were highly heterogeneous. Two‐step clustering further divided M3 into 5 (Figure S4) and M5 into 6 (Figure S5) additional clusters. Conversely, cluster M4 included the majority of samples with ETV6‐RUNX1 as well as the two DS‐ALL samples without ETV6‐RUNX1. In previous DNA methylation profiling,10 the ETV6‐RUNX1 cluster also included ETV6‐RUNX1‐like samples, suggesting that the DS‐ALL samples without ETV6‐RUNX1 in cluster M4 might have an ETV6‐RUNX1‐like signature.

Figure 2

DNA methylation clusters in 723 B‐cell precursor acute lymphoblastic leukemia (BCP‐ALL) samples on consensus unsupervised clustering. Consensus unsupervised clustering suggests 723 BCP‐ALL samples are grouped into 5 clusters (M1‐M5). Clusters, cohorts, Down syndrome (DS)‐ALL samples, and subtypes are shown by colors as indicated. Methylation clusters M3 and M5 are heterogeneous clusters, which are reclustered into 5 and 6 clusters, respectively. As shown in the expression analysis, DS‐ALL and non‐DS‐ALL are clustered into the same cluster corresponding to each biological subtype. HeH, high hyperdiploid; NOPHO, Nordic Society for Pediatric Hematology and Oncology; Ph, Philadelphia chromosome

In accordance with expression analysis, DS‐ALL samples were clustered in subtypes similar to those of the non‐DS‐ALL samples by methylation analysis. However, the methylation levels of several genes, including RUNX1 and KDM2B, were higher in the DS‐ALL samples than non‐DS siblings or mothers.14 Although methylation clustering revealed no distinct DS‐ALL subtypes, the direct comparison of methylation status revealed that the P1 promoter region15 of RUNX1 was hypermethylated in DS‐ALL, but not in non‐DS‐ALL (Figure 3A and Table S14). Methylation of RUNX1 promoters was higher in all DS‐ALL subtypes compared to corresponding non‐DS‐ALL subtypes except for ETV6‐RUNX1, although the sample size of several subtypes, such as Ph‐like or PAX5‐altered, were small in non‐DS‐ALL (Figure S6). Next, we used the most significant probe (cg22698744) to compare ALL samples with CD19+ B cells and CD19‐ B cells from fetal bone marrow samples,16 adult bone marrow samples,17 and bone marrow samples of DS‐ALL patients in remission. The methylation of CD19+ B cells from bone marrow samples of DS‐ALL patients in remission was higher than that of CD19+ B cells from fetal and adult ALL bone marrow samples, although the sample size was small, suggesting that RUNX1 promoter hypermethylation might be a unique characteristic of DS‐ALL (Figure 3B). However, RUNX1 promoters were also hypermethylated in DS patients without ALL (Figure S7). RUNX1 promoter methylation was also higher in adults than fetuses.

Figure 3

Methylation of promoter. A, Volcano plot comparing significant delta beta values between samples from children with Down syndrome and acute lymphoblastic leukemia (DS‐ALL) and non‐DS‐ALL samples. Significant probes showing the delta beta value greater than .2 or less than −.2, and −log10 (q value) greater than 10 are colored red. Among these probes, probes of promoter regions of are colored blue and significantly highly methylated in DS‐ALL. q Values are calculated using the Wilcoxon rank‐sum test adjusted by the Benjamini‐Hochberg correction. B, Box plot comparing beta values of the most significant probe of promoter (cg22698744) in ALL samples (n = 723), CD19+ cells of fetuses (n = 16), CD34+ CD19− cells of fetuses (n = 6) and CD19+ cells of adults (n = 3), and CD19+ cells of DS‐ALL samples in remission (n = 2). In normal samples, CD19+ cells of adults and DS‐ALL samples in remission are highly methylated compared to fetal cells. q Values are calculated using the Wilcoxon rank‐sum test adjusted by the Benjamini‐Hochberg correction. BM, bone marrow

Thus, consistent with a previous report,14 the RUNX1 promoter appears congenitally hypermethylated in DS. The RUNX1 isoform transcribed from the P1 promoter is expressed predominantly in hematopoietic cells.18 As RUNX1 is essential for the differentiation of B cells, hypermethylation of the RUNX1 promoter could be associated with increased BCP‐ALL incidence in DS.

Mutation and copy number analysis

To investigate the relationships among expression, methylation, and genetic status, we undertook mutational (Table S15) and copy number analyses (Tables S16 and S17) on 43 DS‐ALL samples previously analyzed for expression (Figure 4A) or DNA methylation (Figure 4B). In addition, we included the mutational analysis results (Tables S18 and S19) of 118 non‐DS‐ALL samples (expression cohort). Six of 25 samples with CRLF2 fusions harbored JAK2 mutations. As all DS‐ALL samples with JAK2 mutations and CRLF2 fusions in the expression analysis had Ph‐like signatures similar to non‐DS‐ALL,19, 20 all 6 DS‐ALL samples in the mutation cohort were labeled as Ph‐like. In cluster E5, one non‐DS‐ALL sample with a JAK2 mutation and CRLF2 fusion was not labeled as Ph‐like.

Figure 4

Relations of gene expression, DNA methylation, and genomic status. A, Mutational and copy number analysis in expression clusters (E1‐E6). Clusters, samples from children with Down syndrome and acute lymphoblastic leukemia (DS‐ALL), subtypes, and genetic aberrations are shown by different colors as indicated. Copy number analysis of and are represented in only DS‐ALL samples. DS‐ALL sample in E3 had amplification. Samples with alteration are clustered into E3 or E6. DS‐ALL sample without has deletion of , implicating ‐like signature. All samples with mutation and fusion, including DS‐ALL samples, reveal Philadelphia chromosome (Ph)‐like signature. In contrast, mutations of pathway genes are detected in several subtypes. B, Results of DNA methylation analysis are combined with results of mutational and copy number analyses. Copy number analyses of and are represented in only DS‐ALL samples. Samples with Ph‐like signature are divided into M3 and M5 clusters. Ph‐like samples in M3 and M5 are clustered in and iAMP21 cluster, respectively. HeH, high hyperdiploid

To detect other samples with Ph‐like signatures, we carried out a hierarchical clustering of expression cohort samples (Tables S1, S2, and S6), based on the significantly highly expressed (adjusted P < .0001) genes (Table S9) in cluster R8 (Figure S8). This analysis revealed 3 additional samples, 2 DS‐ALL and 1 non‐DS‐ALL, with similar gene expression profiles. The 3 B‐other samples, 2 DS‐ALL and 1 non‐DS‐ALL, in this cluster had Ph‐like signatures. In contrast, several subtypes in the non‐DS‐ALL group showed mutations in RAS pathway genes,21 which are common drivers in pediatric BCP‐ALL.

Copy number analysis revealed that one DS‐ALL sample in cluster E3 had a known focal amplification of chromosome 9 involving PAX5 exons 2‐522 (Figure S9). The expression levels of PAX5 exons 2‐5 were higher than other exons (Figure S10), suggesting the expression of structurally aberrant PAX5 protein or a loss of function. In addition, the expression profile of this particular sample (DS‐15) was similar to other PAX5‐altered samples from the expression analysis and so was categorized as PAX5‐altered.

With the exception of PAX5 deletion, PAX5 status has not been evaluated previously in DS‐ALL. Our results raised the possibility that miscellaneous aberrations, such as amplifications or fusions, as well as deletions, might occur in DS‐ALL. Furthermore, several of the PAX5‐altered DS‐ALL samples, including those with PAX5 amplification, were clustered in E3. Although various PAX5 alterations, including fusions, amplifications, or mutations, were detected in BCP‐ALL, the expression profiles of these samples were similar regardless of PAX5 alterations without PAX5 deletions.23 In our cohort, our defined PAX5‐altered samples were also clustered into the single cluster, E3, although our sample size was small and our analysis might not be precise. Because PAX5 deleted samples were not necessarily clustered into E3 and were detected in 4 clusters, PAX5 deletion would have different pathogenicity from other PAX5 alterations. PAX5 deletions decrease the expression of PAX5, however, other PAX5 alterations might express aberrant PAX5 protein.

One DS‐ALL sample (DS‐13), without the ETV6‐RUNX1 fusion in cluster E4, had homozygous deletions of ETV6 (Figure S11), suggesting an ETV6‐RUNX1‐like signature.24 Cluster M4 included 2 DS‐ALL samples (DS‐10 and DS‐13) without the ETV6‐RUNX1 fusion. Copy number analysis revealed one more sample with an ETV6‐RUNX1‐like signature (Figure S12). Although DS‐10 was not available due to poor RNA quality, copy number analysis showed that DS‐10 harbored deletions of several genes, including ETV6 and ARPP21, 25 suggesting that DS‐10 was also the ETV6‐RUNX1‐like signature.

DISCUSSION

Although incidence of a Ph‐like signature was high in DS‐ALL, especially in children younger than 10 years old, the genetic profile of DS‐ALL was highly heterogeneous (Figure 5). Although the expression and DNA methylation analyses revealed many similarities between DS‐ALL and non‐DS‐ALL, the subtype profile of DS‐ALL was distinct, with lower frequencies of ETV6‐RUNX1 and HeH subtypes26 and a high incidence of a Ph‐like signature, especially in children younger than 10 years, compared to non‐DS‐ALL. There were also discrepancies in cluster designation based on differential expression and DNA methylation. Differences in gene expression could stem from mutations, the presence of chimeric genes, and (or) DNA methylation. Previous studies reported that the frequencies of MLL‐rearranged, BCR‐ABL1, ETV6‐RUNX1, and HeH subtypes were low in DS‐ALL; however, considering that ALL risk is 20‐fold higher in DS,1 the incidence rates of these subtypes were predicted to be 5‐ to 8‐fold higher in DS compared to healthy children. Alternatively, the incidence of the Ph‐like subtype was predicted to be 70‐fold higher. Indeed, Fisher's exact test revealed a significantly higher Ph‐like subtype frequency in DS‐ALL (P = .0056) compared to non‐DS‐ALL. Generally, the Ph‐like samples were divided into 3 classes: (i) ABL‐class rearrangements involving ABL1, ABL2, CSF1R, and PDGFRB; (ii) JAK/STAT pathway alterations of CRLF2, JAK2, EPOR, IL7R, and SH2B3; and (iii) other rare kinase fusions involving NTRK3, DGKH, or FGFR1. The Ph‐like signatures with JAK/STAT pathway alterations accounted for approximately 50% of all Ph‐like signatures in the non‐DS‐ALL group. However, most DS‐ALL samples with the Ph‐like subtype (all except 2) in our cohort had JAK2 mutations and CRLF2 fusions, whereas none with Ph‐like signatures had fusions of JAK2, EPOR, or ABL‐class kinases. No studies to date have reported a Ph‐like signature in DS‐ALL cases harboring fusions without CRLF2 fusions. The high incidence of a Ph‐like signature in DS‐ALL could be due to the high frequency of CRLF2 fusions in DS patients. Indeed, CRLF2 fusion preceded JAK2 mutation frequency in ALL with a Ph‐like signature.

Figure 5

Genetic landscape of Down syndrome and acute lymphoblastic leukemia (DS‐ALL). A, Genetic landscape of DS‐ALL combined with clinical information. The NCI criteria (standard risk [SR] or high risk [HR]), sex, and outcomes (alive or dead) together with subtypes, the affected genes, and the types of genomic aberrations are shown by colors as indicated. plus was defined as deletions co‐occurring with deletions in ,,, or in the absence of deletion. We detected Ph‐like, ‐like, alteration, fusion, and fusion in addition to known subtypes such as , high hyperdiploid, dic(9;20), , and . Samples are ordered by biological subtypes. B, The relation between mutational status and biological subtypes in DS‐ALL samples. The upper half of this figure shows already known genetic alterations in DS‐ALL; the lower half shows subtypes of ALL detected by our analysis. All samples with mutations and fusions have the Philadelphia chromosome (Ph)‐like signature. In contrast, several subtypes of DS‐ALL have mutations of pathway genes

Our cohort also included DS‐ALL samples with the IKZF1 G158S mutation, the amplification of PAX5 exons 2‐5, and an ETV6‐RUNX1‐like signature, all of which were recently identified in other cohorts. In total, the B‐other subtype accounted for approximately one‐third of the cohort samples. Half of the B‐other subtype samples in the DS‐ALL cohort harbored mutations in RAS pathway genes. Mutations in RAS pathway genes are frequently detected in non‐DS‐ALL subtypes, especially the HeH subtype,27 and thus might not be specific for DS‐ALL. One other reason for the high proportion of the B‐other subtype in DS‐ALL could be the unique IGH‐CEBPA or IGH‐CEBPD subtypes revealed by our analysis.

Previous reports revealed DS patients have a genetic predisposition to ALL due to the presence of extra copies of genes such as HMGN1 and DYRK1A. In addition, as our analysis revealed, hypermethylation of RUNX1 promoter is also a potential predisposing factor. RUNX1 is the differentiation factor from common lymphoid progenitors to pre‐pro‐B cells and hypermethylation of the RUNX1 promoter is predicted to decrease RUNX1 expression28 in common lymphoid progenitors. Furthermore, chromosomal instabilities29 caused by aneuploidy in DS patients might be associated with ALL development. These genetic and epigenetic factors could act synergistically to promote DS‐ALL. Several genes, including RUNX1, were also shown to be hypermethylated in DS patients with myeloid leukemia or transient myeloproliferative disorder compared to DS noncancerous fetal liver mononuclear cells.30 However, our comparison of DS‐ALL and non‐DS‐ALL samples revealed similar expression levels of RUNX1, HMGN1, and DYRK1A as well as other chromosome 21 genes between the 2 groups, suggesting that the roles of these genes could be limited to the development of ALL and not afterwards. In addition, hypermethylation of the RUNX1 promoter was not detected in several subtypes, including ETV6‐RUNX1, some Ph‐like, and some B‐other samples, suggesting that these subtypes have unique molecular pathomechanisms independent of RUNX1.

In our cohort, DS‐ALL prognosis was worse than non‐DS‐ALL prognosis (Figure S13A,B), possibly due to the greater sensitivity of DS‐ALL patients to chemotherapy side‐effects and higher treatment‐related mortality rate. Additionally, unique high‐risk subtypes and common high‐risk subtypes with higher frequencies influence the general risk classification of DS‐ALL. However, it is difficult to accurately estimate outcomes of these various subtypes due to the small individual sample sizes. Based on previous clinical investigations, the outcomes of ETV6‐RUNX1, ETV6‐RUNX1‐like, and hyperdiploid may be excellent. Compared to Ph‐like non‐DS‐ALL, the outcome of the Ph‐like DS‐ALL was fair (Figure S13C,D). Because outcomes of B‐other DS‐ALL cases are also still worse (Figure S13E,F), further identification of B‐other DS‐ALL subtypes and associated risk classifications are needed. In addition to personalized supportive care for DS‐ALL patients, precise risk classification based on each subtype identified by molecular profiling is a promising treatment approach.

The major limitation of this study is the small DS‐ALL sample size, which precluded direct comparisons of many subtypes between groups and limited whole‐genome sequencing. Thus, our findings must be verified by a prospective study involving a large cohort.

In conclusion, we confirmed that DS‐ALL is highly heterogeneous with many subtypes. Although subtype expression and methylation profiles were similar to non‐DS‐ALL, subtype frequencies differed significantly between groups. The Ph‐like subtype was the most common in DS‐ALL, accounting for approximately one‐third of all samples. Hypermethylation of the RUNX1 promoter was also a unique characteristic of DS‐ALL, and so could explain the increased ALL incidence in DS.

DISCLOSURE

The authors declare no competing financial interests.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.

Click here for additional data file.