Analysis of rare driving events in pediatric acute myeloid leukemia

Abstract

Elucidating genetic aberrations in pediatric acute myeloid leukemia (AML) provides insight in biology and may impact on risk-group stratification and clinical outcome. This study aimed to detect such aberrations in a selected series of samples without known (cyto)genetic aberration using molecular profiling. A cohort of 161 patients was selected from various study groups: DCOG, BFM, SJCRH, NOPHO and AEIOP. Samples were analyzed using RNA sequencing (n=152), whole exome (n=135) and/or whole genome sequencing (n=100). In 70 of 156 patients (45%), of whom RNA sequencing or whole genome sequencing was available, rearrangements were detected, 22 of which were novel; five involving ERG rearrangements and four NPM1 rearrangements. ERG rearrangements showed self-renewal capacity in vitro, and a distinct gene expression pattern. Gene set enrichment analysis of this cluster showed upregulation of gene sets derived from Ewing sarcoma, which was confirmed comparing gene expression profiles of AML and Ewing sarcoma. Furthermore, NPM1-rearranged cases showed cytoplasmic NPM1 localization and revealed HOXA/B gene overexpression, as described for NPM1 mutated cases. Single-gene mutations as identified in adult AML were rare. Patients had a median of 24 coding mutations (range, 7-159). Novel recurrent mutations were detected in UBTF (n=10), a regulator of RNA transcription. In 75% of patients an aberration with a prognostic impact could be detected. Therefore, we suggest these techniques need to become standard of care in diagnostics.

Introduction

Acute myeloid leukemia (AML) is a rare disease in children. Over the last decades survival has reached a plateau with current event-free survival (EFS) rates of approximately 50-65% and overall survival (OS) over 70% using contemporary protocols.^1-3 Currently, pediatric AML patients are treated with four or five courses of chemotherapy followed by hematopoietic stem cell transplantation in high risk cases.² Treatment stratification is mainly based on early treatment response and genetic abnormalities.⁴ Further chemotherapy intensification is not feasible as treatment toxicity results in a 5-10% mortality rate.⁵ Therefore, improved treatment stratification or novel, preferably targeted, therapeutic options are necessary.

Pediatric AML is a heterogeneous disease characterized by various type I and type II aberrations.⁶ Type I aberrations generally result in a proliferation advantage, such as mutations in N/KRAS, and FLT3-ITD, which are not mutually exclusive.⁷, ⁸ Type II aberrations generally result in a differentiation block, occur early in leukemic development and are mutually exclusive, such as KMT2A-rearrangements, RUNX1-RUNX1T1 and CBFB-MYH11.⁹ Type II aberrations have been reported to be important predictors of outcome.⁷

In recent years, research focused on refining the risk group stratification and elucidating genetic aberrations in what was previously conceived as ‘(cyto-)genetically normal’ AML).^7,10-12 For instance, Hollink et al. described cytogenetically cryptic NUP98-NSD1 rearrangements as a novel recurrent unfavorable prognostic rearrangement in pediatric AML.¹² Furthermore, De Rooij et al. used next-generation sequencing (NGS) to characterize non-DS acute megakaryoblastic leukemia (AMKL), a subtype of AML which is associated with a poor outcome,¹¹ and identified HOX gene rearrangements with a favorable outcome. Translocations were found in approximately 82% of pediatric AMKL patients.¹¹ In adult AML, genomic profiling revealed recurrent single gene mutations such as NPM1, CEBPA, RUNX1 and DNMT3A, which have distinct prognostic significance.¹³ These single gene mutations are rare in children, further underlining that pediatric AML could be a primarily fusion-driven disease.^14-16 Despite these advances, approximately 25% of pediatric AML patients still present with unknown somatic genetic abnormalities.

In the current study, we aimed to identify novel driving oncogenic events in pediatric AML patients without a known type II aberration by karyotype and/or other molecular analysis. NGS, RNA sequencing (RNAseq), whole exome sequencing (WES) and whole genome sequencing (WGS), was performed on primary pediatric AML samples.

Methods

Patient selection

Patients (n=161) were selected from five different study groups: St Jude Children’s Research Hospital (SJCRH), the Dutch Childhood Oncology Group (DCOG), the Berlin-Frankfurt-Münster group (BFM), the Nordic Society of Pediatric Hematology and Oncology (NOPHO), and the Associazione Italia di Ematologica e Oncologia Pediatrica (AIEOP) (Figure 1; Online Supplementary Table S1). Each research group selected patients for this study according to their own diagnostic standard work-up (Online Supplementary Appendix). This study was approved by the ethics committee of each research group.

Next-generation sequencing

RNAseq, WGS and WES were performed as previously described.¹¹ Data has been published previously by Fornerod et al.¹⁷ Fusions detected by RNAseq or WGS were validated by reverse transcription polymerase chain reaction (RTPCR). Single nucletide variant (SNV) calls were considered valid if they were detected by at least two NGS techniques. Only Insertions and deletions in a genome (indels) with at least five reads in the AML sample and no reads in the germline sample were considered as valid mutations, in samples where only WES or WGS was performed. If only one NGS technique was available, also SNV calls with at least five reads in the AML sample and no reads in the germline sample were considered as valid mutations.

Gene expression analysis

Gene expression analysis was performed as described previously.¹¹ Briefly, fragments per kilobase of transcript per million mapped reads (FPKM) were used. Genes that were not expressed in any sample group were excluded from the final data matrix for downstream analysis, i.e., FPKM value >=0.5. Only mRNA was used in the analysis and mRNA encoded by sex-specific genes were excluded.¹¹ Downstream analysis was performed using R, version 3.6.1. Differential gene expression was performed using Voom (Online Supplementary Appendix).¹⁸ For gene set enrichment analysis (GSEA), genes were ranked based on t-statistic and analyzed using the Broad Institute GSEA Desktop Application version 2.2.4.^19,20

NPM1 immunofluorescence

NPM1 immunofluorescence was performed on cytospins of pediatric AML samples and one healthy control, using the anti-NPM1 antibody FC-61911 (Invitrogen, Paisley, UK) and DAPI to stain the nuclei (Online Supplementary Appendix).

Serial colony-replating assay

For functional validation of novel aberrations, serial colony replating was performed. Lineage depleted bone marrow cells derived from 6-8 week old BL6 mice were prepared and transduced as previously described¹¹ (Online Supplementary Appendix).

Statistical analysis

Analysis were performed using R, version 3.6.1. All tests were 2-tailed, and a P value <0.05 was considered significant. Complete remission (CR) was defined as <5% blasts in the bone marrow, with regeneration of trilineage hematopoiesis and no leukemic cells in cerebrospinal fluid or elsewhere. If a patient did not reach CR, treatment was considered a failure at day 0. Overall survival (OS) was calculated from the day of diagnosis until the date of last follow-up or death from any cause. Event- free survival (EFS) was measured from the day of diagnosis to the date of the first event or the date of last follow-up. Events considered in this analysis were resistant disease, relapse, occurrence of secondary malignancy and death. Kaplan-Meyer curves for OS and EFS were calculated and plotted using the survival package in R.²¹

Figure 1.Flow diagram of the cohort selection. Patients from different cohorts were selected according to the research groups diagnostic procedures. Patient samples from SJCRH were selected based on karyotype analysis only, patient samples from DCOG and BFM were selected based on karyotype analysis and molecular analysis such as reverse transcription polymerase chain reaction (RT-PCR) and KMT2A split-fluorescence in situ hybridization analysis. NOPHO and AIEOP both provided samples from patients with a normal karyotype, of whom NOPHO excluded NPM1 mutated patients. In total 161 patients had good quality RNA sequencing (RNAseq) (n=152), whole exome sequencing (WES) (n=135) and/or whole genome sequencing (WGS) (n=100). SJCRH: St Jude Children’s Research Hospital; DCOG: Dutch Childhood Oncology Group; BFM: Berlin-Frankfurt-Münster Group; NOPHO: Nordic Society of Pediatric Hematology and Oncology; AIEOP: Associazione Italia di Ematologica e Oncologia Pediatrica. FISH: Fluorescence in situ hybridization; mut: mutated; rearr: rearrangement.

Results

Fusion detection

In total, 156 of 161 (97%) of analyzed patients had RNAseq and/or WGS data. Of these, fusion genes were identified in 71 patients (44%) (Table 1), confirming the high frequency of translocation events in pediatric AML. Due to the different selection processes of the cohorts, several well-known drivers were detected that were not detected by the research groups using the conventional diagnostic cytogenetic screening methods, like CBFB-MYH11 (n=2), KMT2A rearrangements (n=23), and cryptic NUP98 rearrangements (n=12). Additionally, in 13 patients other rearrangements were identified: DEK-NUP214 in three patients, FUS-ERG (n=4), CREBBP-KAT6A (n=2), KMT2A-PTD (n=2) and enhancer hijacking of MECOM (n=2), leaving ultimately 21 patients (13%) with novel fusions that are either unique or have only been scarcely reported in the literature. These rearrangements involved NPM1 (n=4), ETV6 (n=4), BCL11B (n=2) and GATA1 (n=2) (Table 1). One of the BCL11B rearrangements involved enhancer highjacking detected through WGS, showing a translocation between chromosome 8 and 14. The break point of chromosome 8 was in an intronic region of CCDC26, which is the same location of the t(3;8) translocation involving MECOM enhancer hijacking.²² One of the genes closest to the genomic break point on chromosome 14 was BCL11B and displayed high gene expression, whereas other genes close to the genomic break point were not affected.

Table 1.Characteristics of rearrangements detected in this cohort. This table depicts the characteristics of patients with rearrangements detected in this cohort.

Mutations

WES data was available for 134 of 161 (83%) patients and both WES and WGS data was available for 101 patients. As germline material was used as a reference, only somatic mutations were considered. The median number of mutations with functional consequences in coding regions, i.e., changes in the amino acid composition of the protein, was 23.0 (range, 7-116) per patient. In total, 3,070 unique genes harbored mutations, of which 512 (17%) genes were mutated in at least two patients.

Most of these recurrent mutations had been previously described, such as FLT3 (n=35, 27%), NRAS (n=25, 18%), CEBPA (n=22, 16%), WT1 (n=20, 15%), NPM1 (n=15, 10%), KRAS (n=9, 6%) and RUNX1 (n=8, 6%) (Figure 2A; Table 2). Single gene mutations commonly found in adult AML were rare in this cohort, i.e., IDH1 and IDH2 mutations both occurred in 3.0% of patients (4/134), ASXL1 in 2.2% (3/134), DNMT3A in 1.5% (2/134), TET2 in 1.5% (2/134), and EZH2 in 1.5% (2/134) of patients.

Novel recurrent mutations involved TTN (n=18, 13%), UBTF (n=10), 8%), TCHH (n=9, 7%) and SRA1 (n=9, 7%). As TTN and TCHH are large genes and are often mutated in a variety of cancers without clinical consequences, these mutations were not further analyzed.

Mutations in the UBTF gene had not been previously described in pediatric AML, but rearrangements involving this gene have been described in prostate cancer.²³ All mutations detected in UBTF occurred in exon 13. In total, four patients had a tandem duplication, three patients had in frame deletions, of whom one also had a missense mutation, four patients had missense mutations and one patient had a mutation in the splice region. Patients with a mutation in UBTF had a median variant allele frequency (VAF) of 16.1 (range 10.2-35.6). In five of ten UBTF-mutated patients a mutation in WT1 was detected, and one patient had a concomitant CEBPA DM.

Another novel mutated gene in in nine pediatric AML cases was SRA1. Mutations occurred with a median VAF of 100 (range, 39-100), and all were missense mutations, and resulted in the same amino acid change (V110L). This mutation is registered in the COSMIC database and is predicted to be benign with SIFT and Polyphen scores of 0 (Online Supplementary Table S2).

Non-random association between aberrations

A non-random association between several type I and type II aberrations was detected (Figure 2A). As previously described, FLT3-ITD was associated with NPM1 mutations (9/35, 26%, P=0.001), and WT1 (8/35, 23%, P=0.047) mutations.¹⁰ Furthermore, patients with mutations in CEBPA had a higher frequency of GATA2 mutations compared to the remaining cohort (27%, P<0.001). Of the patients with both CEBPA and GATA2 mutations, three concerned biallelic CEBPA mutations and three monoallelic mutations (Online Supplementary Figure S1). Another association was found between NPM1 mutations and mutations in RAD21 (3/15, 20%, P<0.001). A novel association was detected between CEBPA mutations and mutations in KMT2C (n=3, 14%, P=0.002).

Copy number alterations

Both WES and WGS data were used to identify copy number alterations, such as chromosome 7 or 7q deletions (n=5), chromosome 8 gain (n=5) and chromosome 5q loss (n=4) (Online Supplementary Figure S2). The most common copy number alteration in this cohort involved a gain of chromosome 1q (n=7). A focal deletion was detected in chromosome 16 involving the CTCF gene in three patients, and one patient had a heterozygous deletion of chromosome 16q. Furthermore, WGS analysis identified chromotrypsis in chromosome 8, 12 and 19 in three individual patients.

Gene expression analysis

Unsupervised clustering, using T-SNE on 152 patients with RNAseq data, revealed six clusters (Figure 2B). As was previously described, KMT2A rearrangements and CREBBP-KAT6A clustered together.²⁴

Furthermore, a large cluster containing 67 patients, was characterized by a high expression of HOXA/B genes. Aberrations such as NPM1 mutations and rearrangements (n=19), NUP98 rearrangements (n=10), DEK-NUP214 (n=3) and KMT2A rearrangements (n=5), were detected in this cluster, as reported before.¹² Also mutations in WT1 (13/19, 68%) and FLT3-ITD (23/35, 66%) were enriched in this cluster (Online Supplementary Figure S3). In total, six patients had both WT1 and FLT3 mutations, of which five had no other rearrangement or mutation in NPM1 or CEBPA. Furthermore, this cluster appeared subdivided into two ‘subclusters’, which might be explained by the morphological subtype (FAB type), as one is enriched in granulocytic leukemias (FAB M1/M2) and the other in leukemias from myelo-monocytic origin (FAB M4/M5) (Online Supplementary Figure S4).

Although only double mutations of CEBPA have been described to have a distinct gene expression pattern, in this cohort, both single (n=12) and double mutations (n=8) clustered together based on gene expression.²⁵ In total, 18 of 20 CEBPA mutated cases with RNAseq data clustered together. The two patients that clustered separately had a single mutation in CEBPA and an additional NPM1 disease, 4 relapses, 1 early death). mutation.

Figure 2.Integrative analysis of mutations and gene expression. (A) Waterfall plot of recurrent mutations occurring in at least 4 patients, mutations occurring in less than 3 patients were excluded from this analysis. The plot was first ordered based on type II aberration. Rearrangements occurring less than 5-times were taken together. From FLT3 downward, genes were ordered based on the number of mutations that occurred. (B) t-distributed stochastic neighbor embedding (T-SNE )analysis using the top 200 most variant genes based on standard deviation of 153 patients with RNA-sequencing data. ITD: internal tandem repeat; rearr: rearrangement.

Table 2.Characteristics of mutations detected in this cohort. This table depicts the characteristics of mutations that occur in at least 8 patients in this cohort.

Furthermore, a small cluster of nine samples could be detected, which was enriched with ERG rearrangements. Of these nine patients that clustered together, four had a FUS-ERG rearrangement, one a FUS-FEV rearrangement and one a EWSR1-ERG rearrangement. The three other patients had an ETV6 rearrangement.

A small ‘’cluster’’ consisted of only three samples of which two had a rearrangement involving GATA1.

The patients in the remaining cluster, carried various nonrecurrent rearrangements, including BCL11B rearrangements (n=2), CBFB-MYH11 (n=2), ETV6-MECOM (n=1), KMT2A-MLLT10 (n=1), MECOM (n=1), NUP98-HOXA13 (n=1), SFPQ-ZFP36L2 (n=1), MED12-HOXA9 (n=1), PIM1-BRD1 (n=1), CBFA2T3-GLIS2 (n=1), PICALM-MLLT10 (n=1), RUNX1-USP42 (n=1).

Clinical outcome

Survival data was available of 151 of 161 patients (94%). Surviving patients had a median follow-up time of 6.7 years (range, 0.1-14.5). The entire cohort had a 5-year EFS of 50% (standard error [SE]=4%) and OS of 67% (SE=4%). When patients were stratified according to genetic aberration, trends in differences in outcome were observed (Figure 3). Patients with CEBPA mutations and NPM1 mutations or rearrangements had a 5-year EFS of 71% (SE=10%) and 82% (SE=5%), respectively, the 5-year OS rates were 89% (SE=7%) in both groups. In order to investigate survival of patients with CEBPA mutations, they were split up between single (n=9) and double mutations (n=9) (Online Supplementary Figure S5). Patients with CEBPA DM had no events or deaths, however, patents with CEPBA SM had an EFS of 56% (SE=16%) and OS of 80% (SE=12%). The difference in EFS between CEBPA SM and DM was significant (P=0.03). In contrast, patients with ERG rearrangements had a very poor outcome, i.e, all patients had an event within 2 years after diagnosis (1 refractory disease, 4 relapses, 1 early death).

ERG rearrangements

ERG rearrangements were investigated in more detail as they were associated with a poor outcome. Patient characteristics were verified by combining the current data with the TARGET cohort.¹⁰ The TARGET cohort contained seven pediatric patients with a rearrangement involving ERG or another ETS transcription factor of the ERG subgroup, namely four FUS-ERG, two FUS-FEV and one FUSFLI1. Combining data confirmed the dismal outcome of these patients as in total 11 of 13 patients had an event, and ten of 13 died (Figure 4A). The three surviving patients had a relatively short median follow-up time of 2.6 months (range, 2.0-31) (Online Supplementary Table S3). Furthermore, the date of diagnosis was not significantly different from patients who died (P=0.17). Multivariable analysis was unfeasible due to the limited number of patients. WES and/or WGS data was available in nine of 13 patients. Recurrent mutations occurred in N/KRAS (4/9) and WT1 (2/9). No other genes were recurrently mutated. Differential gene expression analysis comparing ERG-rearranged AML to the remaining AML cohort revealed 1,034 significantly differentially expressed genes, of which 413 genes were upregulated and 621 genes were downregulated (Online Supplementary Table S4). Performing GSEA revealed upregulation of gene sets derived from (embryonic) stem cells and Ewing sarcoma (Figure 4A; Online Supplementary Table 5 and 6). As GSEA of the cluster enriched for ERG rearrangements revealed upregulation of genes involved in Ewing sarcoma, gene expression was compared between AML and a historic Ewing sarcoma RNAseq dataset using the top 50 differentially expressed genes of ERG-rearranged AML versus other AML.²⁶ Using hierarchical clustering, ERG-rearranged AML clustered closely to Ewing sarcoma samples and showed similar expression patterns, whereas other AML samples had opposing gene expression profiles (Figure 4B).

Figure 3.Patient outcome. (A) Event-free survival (EFS) and (B) overall survival (OS) of patients stratified by driving aberration. Aberrations that occur less than 5-times in this cohort have been combined with other rearrangements.

In order to investigate the transforming capacity of ERG rearrangements, FUS-ERG, FUS-FEV and EWSR1-ERG fusions were transduced into murine lineage depleted bone marrow cells and cultured using the colony-forming assay. FUS-ERG and EWSR1-ERG showed self-renewal capacity beyond the cells transduced with an empty vector. However, we could not demonstrate this for FUS-FEV (Figure 4B), suggesting a second hit might be needed for this rearrangement.

NPM1 rearrangements

In total 21 patients had an event in NPM1, of whom 17 had a mutation and four had an NPM1-rearrangement (3 NPM1-CCDC28A and 1 NPM1-HAUS1). Of the NPM1-rearranged cases, follow-up data was available for three patients, who were all alive at the last follow-up with a median follow-up time of 3.5 years (range, 3.1-7.2).

Differential gene expression analysis comparing NPM1 mutated and rearranged cases to the remaining cohort revealed 1,437 significantly differentially expressed genes, of which 251 genes were upregulated and 1,186 genes were downregulated (Online Supplementary Table S7). When NPM1-mutated and NPM1-rearranged cases were separately compared to NPM1-wild-type patients, this revealed 421 and 113 differentially expressed genes, respectively (Online Supplementary Tables S8 and S9). GSEA revealed upregulation of gene sets derived from NPM1 mutated AML for both NPM1-mutated and NPM1-rearranged gene sets (Figure 5A to C; Online Supplementary Tables S10 and S11). NPM1 mutations cause NPM1 retainment in the cytoplasm.²⁷ In order to investigate whether NPM1 rearrangements also resulted in cytoplasmic NPM1, immunofluorescent staining of NPM1 was performed on cytospins of primary AML patients with or without NPM1 mutations and rearrangements and compared to a sample of peripheral blood from a healthy donor. Cytoplasmic staining could be detected in AML samples with either NPM1 mutations or NPM1 rearrangements (Figure 5D), in contrast to NPM1-wild-type AML samples and healthy peripheral blood, where the staining was mainly localized in the nucleoli.

Discussion

This collaborative study aimed to identify novel genetic aberrations in pediatric AML without known type 2 aberrations using routine clinical diagnostics, which varied among the participating collaborative groups in this study as they all used a slightly different approach to standard diagnostics. In approximately 45% of cases a rearrangement could be detected, emphasizing the fusion driven nature of pediatric AML, versus the relatively high frequency of single-gene mutations in adult AML.²⁸ Some were well-known recurrent events, such as KMT2A rearrangements, others were previously described and known as rare events in pediatric AML, such as DEKNUP214 and FUS-ERG.^29,30 Moreover, several novel or scarcely reported recurrent and some unique rearrangements were detected, such as rearrangements involving BCL11B, GATA1 and ETV6. Of the patients without a known type II event, 30% (25/84) had a novel or unique rearrangement.

Figure 4.Characteristics of ERG-rearranged acute myeloid leukemia. (A) Survival curves combining data from Bolouri et al. and this cohort, comparing survival of ERG-rearranged acute myeloid leukemia (AML) to the remaining cohort. (B) Gene set enrichment analysis plots of gene sets derived from EWSR1-FLI1 targets and embryonic stem cells. (C) Hierarchical clustering of AML and Ewing sarcoma samples using the top 50 differentially expressed genes of ERG-rearranged AML vs. other AML (absolute log-fold change). Ewing sarcoma samples cluster together with ERG-rearranged AML. (D) Colony-forming assay of ERG and FEV rearrangments detected in AML. NUP98-KDM5A was taken along as a positive control. Expression of EWSR1-ERG and FUS-ERG in murine lin- cells resulted in repetitive replating. EFS: event-free survival; OS: overall survival.

Besides rearrangements, WES and WGS detected a median of 24 coding mutations per patient, which is comparable to the mutational burden reported by Bolouri et al.¹⁰ In detail, we detected 502 recurrently mutated genes. Most of the recurrently mutated genes are well known, such as N/KRAS, WT1, CEBPA, NPM1 and FLT3. Novel recurrent mutations were detected in the UBTF (n=10) and SRA1 genes (n=9). SRA1 or Steroid Receptor RNA Activator 1 encodes both a protein and long non-coding RNA (lncRNA). The lncRNA acts as an RNA coactivator, and is associated with breast cancer, whereas the encoded protein SRAP is involved in splicing and cell cycle regulation.^31-33 As mutations in SRA1 were predicted to be benign, we do not expect these mutations to be important pathogens. The UBTF gene on the other hand, encodes a protein which regulates mRNA transcription by RNA polymerase 1 or 2 and is associated with developmental neuro-regression.^34,35 UBTF has been described as a translocation partner of ETV1 or ETV4 in prostate cancer. Furthermore, depletion of UBTF is associated with DNA damage and genomic instability.³⁶ However, there was no significant difference in the number of mutations between UBTF-mutated and wild-type cases. Patients with a mutation in UBTF clustered in the cluster with high HOXA/B expression (n=9), except for a patient with a CEBPA DM and UBTF mutation. Five of these patients had an additional WT1 mutation, and one patient had a enhancher hijacking of MECOM. It is unlikely that this mutation is a type II aberration as it occurs subclonally with variant allele frequencies between 10.2% and 35%. Both mutations have not been described by Bolouri et al. and their transforming capacity is unknown.¹⁰ Due to the selection criteria of our cohort, patients with more rare aberrations and mutations are enriched in this cohort, which may attribute to the detection of recurrent mutations that were not previously described by studies using the same technique.

Figure 5.Characteristics of NPM1-rearranged acute myeloid leukemia. (A to C) Gene set enrichment analysis plots of gene sets upregulated in samples with an event in NPM1 compared to samples without an event in NPM1 (NPM1 wild-type [wt]), of which (A) depicts all NPM1 events (both rearrangements and mutations) vs. NPM1 wt, (B) depicts NPM1 mutations vs. NPM1 wt, and (C) depicts NPM1 rearrangements (NPM1r) vs. NPM1 wt. (D). Cytospins of primary patient material stained with anti-NPM1. NPM1 is located in the nucleoli in healthy peripheral blood and in an acute myeloid leukemia (AML) sample with NPM1 wt, whereas it is localized in the cytoplasm in the NPM1-mutated and NPM1-rearranged AML samples.

In contrast to previous studies, a similar gene expression pattern of CEBPA DM and SM was detected.²⁵ However, in previous studies, CEBPA SM was associated with other type II aberrations such as NPM1 mutations, which could result in differences in gene expression^.14,25 Furthermore, we did find a significant difference in EFS with a worse outcome in the CEBPA SM group, which was in contrast to a study by Tarlock et al., who found no difference in outcome between CEBPA SM and DM.³⁷ However, our study was small with only nine patients in each group, whereas Tarlock et al. has a much larger study, and is potentially more reliable.

One patient subset was characterized by poor outcome, i.e., those with rearrangements in FUS-ERG (n=4), EWSR1-ERG (n=1) and FUS-FEV (n=1). FUS-ERG-rearranged AML has been described as a group with a poor outcome before,²⁹ which was reflected in this cohort and confirmed in the TARGET cohort by Bolouri et al. (Figure 4A).¹⁰ These rearrangements have also been described in Ewing sarcoma.^10,26,29,38 The fusion genes involved the transcription factor binding site of FUS or EWSR1, and the DNA binding site of ERG or FEV. The fact that FUS and EWSR1 are homologs and ERG and FEV, together with FLI1, belong to the same ETS subfamily, could explain why these rearrangements result in a similar gene expression pattern and cluster together in unsupervised clustering analysis (Figure 2B).^39,40 GSEA revealed upregulation of pathways involved in stem cell development. This was expected as genes involved in these fusions are important for the regulation, development and self-renewal of hematopoietic stem cells.^41-46

In addition, expression signatures previously described in Ewing sarcoma were upregulated, and all fusions in this group also occur in Ewing sarcoma. When combining RN seq data from AML and Ewing sarcoma patients, the ERG-rearranged AML patients clustered together with Ewing sarcoma patients, whereas other AML patients clustered separately. This suggests that ERG-rearranged AML shares oncogenic pathways with Ewing sarcoma in contrast to other AML samples. Finding common ground between cancers can be useful for detecting novel therapeutic targets. Current chemotherapy protocols of AML and Ewing sarcoma are rather comparable, and not many additional novel therapeutic agents have proven to be effective.^2,47,48 Another recurrent set of rearrangements included NPM1 rearrangements (n=4). We detected two different types of rearrangements: NPM1-CCDC28A (n=3) and NPM1-HAUS1 (n=1). Bolouri et al. also detected four patients with NPM1 fusions, i.e., NPM1-MLF1.¹⁰ All three NPM1-rearranged patients with survival data in our cohort were in CCR, whereas of the four NPM1-rearranged patients in the TARGET cohort, all patients relapsed and three patients subsequently died. Therefore, the prognostic value of NPM1 rearrangements may be treatment or translocation partner dependent. The breakpoint of NPM1 in the rearrangements in our study was between exon 11 and exon 12. As mutations and rearrangements both result in disruption of exon 12 of NPM1, we hypothesized that both events could result in cytoplasmic retention of NPM1, driving leukemogenesis in these cases.²⁷ Immunofluorescence of NPM1 showed that NPM1-CCDC28A result in cytoplasmic retention of NPM1 (Figure 5B). This has been previously shown for NPM1-HAUS1 as well, which makes it likely that NPM1 rearrangements and NPM1 mutants functionally have a similar mechanism of action.⁴⁹ Furthermore, performing unsupervised clustering, all patients with an NPM1 event clustered in the large cluster which is characterized by upregulation of HOXA and HOXB, further underlining the conceivable similarities in oncogenesis of NPM1 mutations and rearrangements.

Rarer recurrent rearrangements involved BCL11B (n=2). BCL11B mutations and rearrangements are associated with T-ALL.^50,51 In this cohort, one patient had a ZEB2-BCL11B rearrangement and in the other patient, enhancer hijacking of BCL11B was detected. Both rearrangements led to high expression of BCL11B. This rearrangement has been described in adult AML before, but did not lead to self-renewal capacity in CFU-GEM assay.⁵²

One of the drawbacks of this study is that 37% of the cases that were included in the study had a known driving genetic aberration, such as a KMT2A rearrangement, which was dependent on what was considered the diagnostic standard of care in the participating collaborative groups. This shows the relevance of implementing more genome-wide diagnostics, as for example karyotyping may appears not sufficient to detect all relevant fusions. Therefore, we suggest to implement NGS, i.e., RNAseq and WES, as standard of care in pediatric AML to readily detect driving aberrations, and stratify patients according to genetic events to improve outcome.⁵ Moreover, with the availability of targeted agents such as menin inhibitors, it becomes relevant to readily identify all KMT2A and NUP98 rearrangements.

In 24% (n=39) of the patients in this study, despite NGS efforts, still no (potential) type II aberration could be identified. However, in eight of these 39 cases, only RN seq was performed, therefore, potentially driving single gene mutations could have been missed. As the mutational burden in coding regions of the genome in pediatric AML is low, we anticipate that potentially non-coding mutations may play an important role here, as well as epigenetic events.^10,53

In conclusion, this study has characterized several novel and rare genetic aberrations occurring in pediatric AML. Some novel rearrangements, such as ERG rearrangements should be considered high risk, whereas for others, such as NPM1 rearrangements this is not yet clear. Ideally for pediatric AML, diagnostics includes RNAseq when resources allow it, as we demonstrated that this detects rare relevant fusion genes with prognostic implications.

Footnotes

• Received November 29, 2021
• Accepted July 19, 2022

Correspondence

C. M. Zwaan c.m.zwaan@erasmusmc.nl

Disclosures

No conflicts of interest to dislcose.

Contributions

CMZ, TAG, MF and MMvdH-E designed the study. DR, MP, FL, HH, JA, MJ, CK, SP, CMZ contributed materials and clinical data. SN, MF, JM, MW and GS analyzed the data. SN, JvO and EARG performed experiments. SN and MF performed statistical analysis. SN, MF, MMvdH-E and CMZ wrote the paper. CMZ, MF, MMvdH-E and TAG supervized the study and all co-authors performed critical review of the manuscript and gave their final approval.

Data-sharing statement

Data for the sequenced samples in this study have been deposited to the St. Jude Cloud (www.stjude.cloud; ref. 33) and European Genome-phenome Archive (study ID EGAS00001004701).