DEREGULATION OF DUX4 AND ERG IN ACUTE LYMPHOBLASTIC LEUKEMIA

Abstract

Chromosomal rearrangements deregulating hematopoietic transcription factors are common in acute lymphoblastic leukemia (ALL).^1,2 Here, we show that deregulation of the homeobox transcription factor gene DUX4 and the ETS transcription factor gene ERG are hallmarks of a subtype of B-progenitor ALL that comprises up to 7% of B-ALL. DUX4 rearrangement and overexpression was present in all cases, and was accompanied by transcriptional deregulation of ERG, expression of a novel ERG isoform, ERGalt, and frequent ERG deletion. ERGalt utilizes a non-canonical first exon whose transcription was initiated by DUX4 binding. ERGalt retains the DNA-binding and transactivating domains of ERG, but inhibits wild-type ERG transcriptional activity and is transforming. These results illustrate a unique paradigm of transcription factor deregulation in leukemia, in which DUX4 deregulation results in loss-of-function of ERG, either by deletion or induction of expression of an isoform that is a dominant negative inhibitor of wild type ERG function.

INTRODUCTION

B-precursor ALL is the commonest childhood tumor, and is a heterogeneous disease comprising multiple subtypes with distinct constellations of somatic structural DNA rearrangements and sequence mutations that commonly perturb lymphoid development, cytokine receptor and Ras signaling, tumor suppression and chromatin modification.² However, the genetic basis of a substantial proportion of B-ALL cases remains to be defined. Previous reports identified a subset of B-ALL with a distinct gene expression profile,³ and frequent deletion of ERG, encoding the ETS-family transcription factor v-ets avian erythroblastosis virus E26 oncogene.^4–6 ERG has a key role in hematopoietic differentiation^5,7,8, megakaryopoiesis,⁹ and megakaryoblastic leukemia associated with Down syndrome^10–12. ERG is frequently rearranged in carcinoma of the prostate¹³ and rarely in acute leukemia¹⁴, and ERG overexpression is associated with poor outcome in acute myeloid leukemia¹⁵. ERG is temporally regulated during B lymphopoiesis (Supplementary Figure 1), suggesting it may regulate B lymphoid development, however its role in the pathogenesis of ALL is unknown.

RESULTS

A subtype of B-ALL with DUX4 rearrangement and ERG deletion

To understand the genetic basis of this subtype of B-ALL, we studied a cohort of 1913 individuals with B-progenitor ALL, including 1347 children, 395 adolescents (age 16–20) and 171 young adults (age 21–39) with B-progenitor ALL using gene expression profiling and analysis of DNA copy number alterations by single nucleotide polymorphism (SNP) arrays in all cases, and whole genome (N=32), exome (N=44) and/or transcriptome sequencing (N=54) in a subset of cases (Supplementary Tables 1–2).

Microarray and transcriptome sequencing data identified 141 (7.6%) ALL cases with a distinct gene expression profile (Figure 1a–b, Supplementary Table 3 and Supplementary Figure 2). This form of leukemia constituted 5.2% of childhood standard risk, 9.4% of childhood high risk, 10.2% of adolescent and 5.4% of adult ALL cases.

Figure 1. Gene expression profile and ERG deletions in DUX4/ERG ALL.

a, Hierarchical clustering of the top 100 Affymetrix U133A probe sets upregulated in each subtype of 199 B- and T-lineage ALL cases, including DUX4/ERG ALL. b, Principal component analysis of FPKM gene expression data derived from poly-A RNA sequencing identifying DUX4/ERG ALL c, SNP microarray data of cases with ERG deletions. Data for each case is shown as a column, and shown in a log₂ ratio scale where deletions are blue. The position of the ERG locus is shown as an arrow on the left of the panel, and genomic coordinates shown in megabases on the right. Cases are ordered from left to right according to the extent of the deletion, with focal exon 1 deletions on the left, the common exons 3–7 and 3–9 deletions in the middle of the panel, and whole gene deletion on the right. d, Representative DNA copy number data of four cases with ERG deletion also shown in log₂ ratio scale, with probe level data shown, deletion being below the x axis. The extent of deletion in each case is shown by horizontal bars.

Eighty-five (55.6%) of these cases had focal deletions of ERG at chromosome 21q22.3 (Figure 1c–d), which were not observed in other B- or T-ALL cases. The ERG deletions were confirmed by genomic quantitative PCR and breakpoint mapping, and most commonly involved exons 3–7 (N=27) or 3–9 (N=22) of ERG transcript variant 1 (NM_182918.3, the most abundantly expressed transcript in non-B ALL and normal B cells)^5,16. The presence of conserved heptamer recombinase signal sequences at deletion breakpoints and intervening non-consensus nucleotides indicated that the deletions arise from aberrant recombinase activating gene activity (data not shown). Genomic analysis of a panel of leukemia cell lines showed the B-progenitor cell line NALM-6 exhibited a similar gene expression profile and an intragenic ERG deletion (Supplementary Figure 3).

Notably, genes of the double homeobox gene family on chromosomes 4q/10q were among the top upregulated genes (Figure 2a and Supplementary Table 3). Analysis of transcriptome sequencing data showed that all cases sequenced exhibited rearrangement of DUX4 to IGH, placing DUX4 under the control of the immunoglobulin heavy chain enhancer resulting in increased expression of DUX4 (Supplementary Table 2). DUX4 encodes a double homeobox transcription factor located in a macrosatellite repeat in the subtelomeric repeat unit of chromosome 4q.¹⁷ Deletion of part of this repeat unit is causative of facioscapulohumeral dystrophy, and DUX4 rearrangements have been reported in a subset of Ewing-like sarcoma (CIC-DUX4),^18,19 and recently, a subset of ALL.²⁰ In each case of this subtype, DUX4 is inserted adjacent to the IGH enhancer region (see Supplementary Note and Supplementary Figure 4a), with variable truncation of the C-terminus of DUX4 and appending of a variable number of amino acids from read-through into the IGH locus (See Supplementary Note, Figure 2b–d). IGH-DUX4 rearrangement was confirmed in all 6 cases selected for validation by RT-PCR and/or genomic PCR (Supplementary Table 4 and Supplementary Figure 4b–c), and DUX4 overexpression was confirmed by immunoblotting (Figure 2d). Thus, DUX4 rearrangement is a universal feature of this subtype of B-ALL that exhibits a distinct gene expression profile, with deletion of ERG identified in the majority of cases.

Figure 2. Rearrangement of DUX4.

a, deregulation of DUX4 is observed exclusively in the ALL cases with ERG deregulation. Gene expression data shown as fragments per kilobase of mapped reads (FPKM) from RNA-sequencing in box plots, with ALL cases grouped by subtype on the x axis. Horizontal lines show median, and the boxes, interquartile range. b, an example of the commonly complex rearrangements of IGH and DUX4. Each third of the panel shows RNA-seq gene expression data for three loci involved in the rearrangement. The top shows increased expression at DUX4. The dotted line shows the breakpoint of DUX4 that is juxtaposed to a small segment of CDH4 shown in the middle panel, which in turn is rearranged to IGH, shown in the lower panel. c, schematic showing the location of the breakpoints in DUX4, stratified by age group. d, Immunoblotting of cell line and primary leukemic sample lysates showing proteins of variable size corresponding to DUX4 C-terminal truncation and/or appending with amino acid residues encoded by read through into the IGH locus.

The genomic landscape of DUX4-rearranged B-ALL

Analysis of genome, exome and transcriptome data demonstrated that in addition to universal rearrangement of DUX4 and frequent deletion of ERG, this subtype of ALL is characterized by a distinct mutational landscape (Figure 3). We identified a mean of 17.5 non-silent sequence mutations per case (range 2–42) and a paucity of structural genetic alterations (Supplementary Tables 5–9). Alterations of lymphoid transcription factor genes were present in 46.5% of cases (IKZF1 36.7% and PAX5 11.3%). Notably, while IKZF1 alterations are associated with poor outcome in other subtypes of childhood B-ALL^21,22, they were not associated with poor outcome in childhood DUX4/ERG ALL (Supplementary Figure 5). Mutations in transcription factors and transcriptional regulators were observed in 21% of DUX4/ERG cases, including MYC, MYCBP2, MGA and ZEB2, that were uncommon in other subtypes of B-ALL, including 209 representative B-ALL and 16 T-ALL cases subjected to whole genome or exome sequencing (a listing of recurrently mutated and deleted genes across ALL subtypes is provided in Supplementary Table 10). Additional recurrent mutations included those activating Ras signaling (35.2%), cell cycle regulation (22.5%); and epigenetic modifiers (56.3%), most commonly KMT2D, SETD2, ARID2 and NCOR1 (Supplementary Figure 6).

Figure 3. Structural and sequence alterations in DUX4/ERG ALL.

Heatmap showing genomic data for DUX4/ERG ALL cases, each of which is represented in a column and genes are grouped by functional pathway. The colors for each type of genetic alterations are shown at the bottom of the figure. The genomic profiling performed, and presence/absence of ERGalt, ALE and DUX4 rearrangement are shown at the top of the figure, where yellow represents assay performed (or alteration present), white, not performed or absent, and gray, data not available.

ERG deregulation in DUX4-rearranged B-ALL

DUX4-rearranged cases also exhibited profound transcriptional deregulation of ERG. Although the absolute level of expression of ERG was not higher in comparison to other ALL subtypes, this subtype of ALL was characterized by expression of multiple aberrant coding and non-coding ERG isoforms. Using RT-PCR, we detected internally deleted ERG transcripts corresponding to the exon 3–7 and 3–9 deletions (Supplementary Figure 7a). Translation of these internally truncated transcripts was predicted to result in a frame shift and premature truncation. However, corresponding N-terminal truncated ERG proteins were not identified on immunoblotting of leukemic cells harboring ERG deletions (data not shown). In contrast, immunoblotting using C-terminus specific antibodies identified a 28 kDa ERG protein in 50 (63.2%) of 79 DUX4/ERG ALL cases tested, including cases lacking an ERG deletion, suggesting an alternate mechanism of ERG deregulation (Supplementary Figure 7b).

Of the 54 DUX4-rearranged cases subjected to transcriptome sequencing, the majority expressed aberrant ERG transcripts initiated in intron 6 (Supplementary Table 11). The most abundantly expressed isoform was initiated from a novel exon whose 3’ splice site was located 197 nucleotides proximal to exon 7 (exon 6 alternate, or exon 6 alt; Figure 4a). The genomic boundaries of this non-canonical exon were determined by rapid amplification of cDNA ends (RACE), RT-PCR with primer walking, and analysis of the orientation of RNA-seq read pairs mapping to this region. This confirmed splicing of exon 6 alt to exon 7 and downstream exons, without evidence for splicing of upstream exons into exon 6 alt, indicating that exon 6 alt is a non-canonical first exon (Supplementary Figure 8).

Figure 4. Expression of ERGalt in DUX4/ERG ALL.

a, read depth from mRNA-sequencing of an ERG ALL case with expression of exon 6 alt (red) and a case lacking ERG alt expression (gray). b, RT-PCR with PCR primers specific to exon 6 alt and exon 10, showing amplification of a larger amplicon in ALL cases lacking ERG alt, with amplification of the intervening intronic sequence between ERGalt and exon 7, and a smaller amplicon in ERGalt positive cases arising from splicing from exon 6 alt to exon 7, shown in a representative electropherogram. This transcript results in a novel N-terminus of ERG encoded by exon 6 alt comprising 7 residues spliced in frame to exons 7–10. c, structure of the canonical and ERG ALL-associated ERG transcripts and isoforms. d, relative abundance of ERG transcripts in ERG ALL and other B-ALL subtypes. Each column represents a case. The top panel represents expression of wild type (WT) ERG, and the middle panel expression of ERGalt, and the less abundant ERGalt isoforms. The lower panel shows the cumulative abundance of all isoforms as a proportion within each case. Transcripts arising from splicing across ERG deletion are low abundance in most cases with ERG deletion (with the exception of one case with biallelic deletion. Non-coding ERGalt isoforms are more common in ERG ALL cases lacking ERG deletion.

RT-PCR demonstrated in-frame splicing from ERG exon 6 alt to exon 7, resulting in expression of an ERG protein with a new N-terminus of 7 amino acids encoded by exon 6 alt followed by exons 7 through 10 of ERG isoform 1 (“ERGalt a”; Figure 4b). The predicted size of this truncated C-terminal ERG protein corresponded to the size of the protein identified by immunoblotting of ERG ALL leukemic cells (Supplementary Figure 7c).

We next systematically analyzed expression of canonical and non-canonical ERG transcripts across a range of leukemias and solid tumors using data from the St Jude Children’s Research Hospital – Washington University Pediatric Cancer Genome Project. In addition to ERGalt a, we identified a second isoform with coding potential also encoded using ERG exon 6 alt, ERGalt b, and two less abundant transcripts that lack coding potential (ERG alt c/d, Figure 4c, Supplementary Table 11). Expression of the coding isoforms, ERG alt a/b, was observed in the majority of ERG ALL cases, and accounted for the majority of transcripts expressed in such cases, but was uncommon and low level in non-ERG cases, such as a minority of BCR-ABL1 ALL cases (Figure 4d). Analysis of other pediatric ALL (N=290) and non-hematopoietic (N=572) tumors showed that expression of non-canonical ERG transcripts were observed in a minority of other B-ALL cases (e.g. Ph+ ALL) but were usually comprised of transcripts without coding potential. In contrast high-level expression of ERG transcripts initiated from this exon was restricted to DUX4-rearranged cases, which was confirmed by quantitative RNA-sequencing and quantitative RT-PCR (Supplementary Table 12).

Additional evidence of ERG deregulation was observed, with intron retention upstream of ERG exon 7 (Supplementary Figure 9). Analysis of total stranded RNA-sequencing data of 12 DUX4/ERG, 40 other B-ALL cases and normal B cell progenitors identified expression of a long non-coding RNA proximal to the first exon of ERG in the majority of DUX4/ERG cases that was restricted to this form of leukemia, and observed in cases with and without ERG deletion (Antisense Long non-coding RNA associated with ERG, or ALE; Supplementary Tables 12–13 and Supplementary Figures 10–11).

DUX4 binds and induces deregulation of ERG

The finding of DUX4 rearrangement as a universal feature of this subtype of ALL, and the unique presence of ERG deregulation and deletion suggests that the two phenomena are related and may contribute to leukemogenesis. DUX4 is known to induce deregulated expression of many genes, including transcripts utilizing alternate exons.²³

To examine the relationship of DUX4 expression and ERG deregulation, we first analysed previously reported chromatin immunoprecipitation and sequencing (ChIP-seq) data for DUX4 expressed in human myoblasts.²³ This identified a peak of DUX4 binding at the first, non-canonical exon of ERGalt with two DUX4 binding motifs within this 372 nucleotide region (Supplementary Figure 12). Moreover, analysis of RNA-seq data showed that expression of DUX4 in myoblasts resulted in expression of ERGalt that was not observed in non-DUX4 expressing cells (Supplementary Figure 13a). Using ChIP-PCR, we observed binding of DUX4 to the first exon of ERGalt in NALM-6 cells (Supplementary Figure 13b). We next performed DUX4 ChIP-seq in NALM-6 and Reh (ETV6-RUNX1 ALL) cells that confirmed DUX4 binding at ERG exon 6 alt (Figure 5a), as well as at additional transcripts deregulated in DUX4/ERG ALL (Supplementary Figure 13c). A comprehensive listing of all peaks of DUX4 binding identified by ChIP-seq annotated with DUX4/ERG gene expression and ATAC-seq data is provided in Supplementary Table 14.

Figure 5. DUX4 induces deregulation of ERG.

a, DUX4 ChIP-seq and ATAC-sequencing data of Reh, NALM6 and DUX4/ERG ALL xenograft showing DUX4 binding at ERG exon 6 alt (arrowed) in NALM6 but not in Reh. ATAC sequencing showing open chromatin at this peak of DUX4 binding at ERG exon 6 alt in the DUX4/ERG xenograft sample and NALM6 but not in Reh. ATAC-seq data is provided in Supplementary Figure 19. b–c, expression of two truncated DUX4 alleles, but not empty vector, results in expression of ERGalt by RT-PCR (b) and immunoblotting (c). b, the top panel shows RT-PCR for ERG using primers specific for exon 6 alt and exon 7. In untransduced or empty vector transduced Reh cells, a larger amplicon is observed that incorporates the intervening intronic transcript (lanes 3 and 4). In DUX4/ERG ALL cells (PARLRH, lane 2) and DUX4-transduced Reh cells (lanes 5 and 6), ERGalt is amplified, with in-frame splicing from exon 6 alt to exon 7 (lower part of panel). c, Immunoblotting showing expression of ERGalt in cells transduced with DUX4. The top part of the panel shows immunoblotting with an N-terminus DUX4-specific antibody; the middle part blotting with a C-terminus specific ERG antibody, and the lower part, actin control. DUX4 alleles induce expression of ERGalt in HEK293T cells (lanes 6 and 7) and Reh cells (lanes 11 and 12). Lane 5 shows 293T cells transfected with ERGalt virus as a positive control for ERGalt (and negative for DUX4); lane 8 shows DUX4/ERG patient sample PARLRH positive for DUX4 and ERGalt.

We systematically analyzed the expression of transcripts utilizing non-canonical first exons in total stranded RNA-seq data, and identified 45 transcripts significantly deregulated in DUX4/ERG ALL, of which three (ERG, NSD1 and RNGTT) were bound by DUX4 (Supplementary Table 15). ERGalt was the second most frequently deregulated transcript utilizing a non-canonical first exon, and the most frequently deregulated of those loci bound by DUX4.

To examine the role of DUX4 overexpression in deregulation of ERG, we transduced human cord blood hematopoietic cells and the ETV6-RUNX1 B-ALL cell line Reh with lentiviral vectors expressing DUX4 alleles corresponding to those identified in patients (DUX4 E415* and Q334*), and empty lentiviral vector as a control. Truncated DUX4 induced expression of ERGalt in both Reh and human CD34 cord blood cells, as shown by both transcript and protein analysis (expression of full length DUX4 was not tolerated in 293T vector producer cells) (Figure 5b–c Supplementary Figure 13d).

Together, these findings suggest that ERG deregulation is caused by DUX4 overexpression, which binds to the alternative transcription initiation site in ERG intron 6, and that ERG deletion is a secondary event occurring in a transcriptionally active locus primed for RAG-mediated deletion. Such deletion of ERG may further impair expression of wild-type ERG. Consistent with this proposed mechanism and sequence of deregulation, transcription of ERGalt is initiated in ERG intron 6, which lies in the common regions of ERG deletion (exons 3–7 and 3–9). Thus, in contrast to other key targets of deletion in ALL such as PAX5 and IKZF1, the deleted ERG allele cannot encode the putative oncogenic isoform, but rather it is encoded by the non-deleted ERG allele.

In addition, some DUX4/ERG cases expressing ERGalt lack a clonal ERG deletion but have evidence of subclonal deletions. Of the 46 of 54 cases with transcriptome data that express ERGalt, 27 had a clonal ERG deletion. However, 7 cases lacking such a deletion had evidence of a subclonal deletion on genomic PCR (Supplementary Table 2), and emergence of clonal ERG deletions was observed in xenografted cases, and in cases of relapsed ALL in which the characteristic gene expression profile was present at both diagnosis and relapse, but in which clonal ERG deletions were only observed at relapse (data not shown). Together, these data support the notion that ERG deletions are secondary genomic events, but if acquired sufficiently early in leukemogenesis may be present as clonal events.

ERGalt inhibits activity of ERG and promotes leukemogenesis

ERGalt lacks the N-terminal pointed domain and central regulatory domains of wild type ERG, but retains the DNA-binding ETS and transactivation domains. Both wild-type ERG and ERGalt exhibited nuclear localization and bound DNA target sequences (Supplementary Figure 14a and data not shown). In a transcriptional reporter assay using an ERG target gene, gpIX²⁴, ERGalt displayed diminished transactivating activity and acted as a competitive inhibitor of wild type ERG (Supplementary Figure 14b–c).

These results suggest that ERGalt may directly contribute to leukemogenesis, in part by inhibiting the function of wild-type ERG. To explore this, lineage-negative bone marrow from Arf^−/− mice was transduced with retroviral supernatants expressing wild type ERG or ERGalt, and/or an activated NRAS allele and/or IK6, the dominant negative form of IKZF1 commonly observed in human ALL, followed by in vitro culture and transplantation of transduced cells into lethally irradiated recipients. ERGalt and NRAS^G12D together promoted serial replating of lymphoid cells (Supplementary Figure 15). Consistent with prior reports, expression of wild type ERG induced a lethal erythromegakaryoblastic leukemia (Figure 6a–d)²⁵. In contrast, mice transplanted with ERGalt expressing cells developed lymphoid precursor, biphenotypic or pre-B cell leukemias with longer latency (Figure 6a–d), indicating ERGalt directly promotes lymphoid leukemogenesis.

Figure 6. Expression of ERGalt induces ALL.

a, Transplantation of lineage negative Arf^−/− progenitors induces erythromegakaryoblastic leukemia, and ERGalt expressing marrow a lymphoid progenitor, biphenotypic or B-lymphoid leukemia. *** P< 0.0001 (Mantel-Cox); n= 12 ERG WT, 19 ERGalt mice from 2 independent experiments. b, immunoblotting with a C-terminus specific ERG antibody for ERG ALL samples and splenocytes of mouse leukemias, showing expression of ERG alt in human ERG ALL and mouse tumors induced by expression of ERG alt. c, proportion of leukemias displaying erythromegakaryoblastic, lymphoid, or mixed immunophenotype (mixed = lymphoid and myeloid subpopulations). d, representative immunophenotyping of ERG-induced tumors showing expression of the erythroid marker Ter119 in an ERG WT induced tumor, and coexpression of B220 and CD19 in ERGalt induced leukemia. Cells were gated on GFP expression.

Discussion

These data provide comprehensive genomic characterization of a subtype of B-progenitor ALL characterized by a distinct gene expression profile and deregulation of two transcription factors, DUX4 and ERG. Our findings indicate that DUX4 rearrangement is present in all cases with this distinct gene expression profile, and is a clonal event acquired early in leukemogenesis. DUX4 is not expressed during normal mouse or human B cell development, and translocation to IGH provides a mechanism for hijacking into the B cell lineage, as we have previously described for other genes rearranged to antigen receptor loci such as CRLF2 and EPOR.^26–29

The striking deregulation and deletion of ERG is unique to this subtype of ALL. While multiple prior studies have reported ERG deletions in B-ALL,^22,30–32 including in DUX4-rearranged cases,³³ several of these studies have identified only cases harboring the common deletions and used these as a surrogate to identify a subset of ALL cases with favorable prognosis.^30,31 In contrast, we show that ERG transcriptional deregulation is a hallmark of this subtype, with diverse ERG deletions in a majority, but not all cases. Multiple mechanisms of ERG genomic alteration were observed. These included expression of aberrant ERG transcripts in all cases, including non-canonical coding transcripts, deregulation with intron retention, and long non-coding RNA expression, with expression of the novel coding ERG transcript, ERGalt, in the majority of cases; and clonal or subclonal ERG deletions also observed in the majority of cases. We have shown that DUX4 directly binds to the ERG locus at the first, non-canonical exon of ERGalt, and in multiple hematopoietic and non-hematopoietic cell types directly induces expression of ERGalt. The deregulation of ERG by DUX4 is reminiscent of deregulation of other ETS family genes in solid tumors,¹⁹ and represents a new mechanism of ETS gene deregulation in leukemia. Moreover, as ERG may represent one of potentially many genes deregulated by DUX4, we performed a systematic, genome-wide, integrated analysis of gene expression, expression of genes utilizing non-canonical first exons, DUX4 ChIP-seq and ATAC sequencing and show that ERGalt is the top deregulated gene in this form of leukemia that is bound by DUX4 and utilizes a non-canonical first exon.

The observations that ERG is deleted in only a subset of cases, and that expression of ERGalt is directly induced by DUX4 suggests that ERG deregulation is an important, but secondary event in leukemogenesis. In addition, we have shown that for the majority of cases harboring an ERG deletion, deregulation of the ERG locus involves both alleles, and cannot be explained by perturbation of a single copy of the gene. The common ERG deletions remove intron 6 which harbors ERG exon 6 alt, the first exon encoding ERGalt, and the region bound by DUX4. Thus, in a case with a clonal ERG deletion, ERGalt must be expressed from the non-deleted ERG allele. Thus, our data suggest a sequence of events in which rearrangement of DUX4 is an early, leukemia-initiating event that results in binding to the ERG locus and deregulated expression of coding and non-coding transcripts. The resulting increased chromatin accessibility evidenced by ATAC-sequencing renders the locus susceptible to RAG-mediated deletion of ERG. Should this occur early in leukemogenesis, ERG deletion is observed as a clonal event, but if later, a subclonal event that may become clonal during disease progression (e.g. at relapse, or as a surrogate, passage in an immunocompromised mouse).

The interplay of DUX4 and ERG deregulation in leukemogenesis will require detailed future examination in appropriately engineered mouse models that must account for the fact that ERG exon 6 alt is incompletely conserved in the mouse, and primary mouse hematopoietic cells were thus not suited to examining induction of ERG alt by DUX4. That notwithstanding, the findings of inhibition of the transcriptional activity of wild-type ERG by ERGalt, and documentation of aberrant intron retention and/or deletion of ERG in all cases in this form of leukemia, but rarely in any other tumor indicate that inhibition or loss of ERG activity is required in the pathogenesis of human DUX4/ERG ALL. The notion that ERGalt sustains lymphoid colony replating, and induces leukemia with longer latency than wild type ERG, but with a shift to a biphenotypic or lymphoid lineage, further supports this notion.

These findings have important clinical implications, as DUX4/ERG ALL is associated with favorable outcome, irrespective of the presence of concomitant genetic alterations otherwise associated with poor outcome in other contexts, such as deletion of IKZF1. DUX4 rearrangement is not evident on karyotypic analysis, and is challenging to identify on analysis of genome and RNA-sequencing data due to the repetitive nature of both breakpoints and the nature of DUX4 insertion into the IGH locus. Thus, in contrast to prior studies that have used identification of only the common (and clonal) ERG deletions as a surrogate for identification of a distinct type of leukemia,^30,31 future studies must move beyond identification of ERG deletions alone to identify this form of leukemia. All cases harbor rearrangement of DUX4, detection of which requires transcriptome and/or genome sequencing. Quantitation of DUX4 may also identify cases with rearrangements, but must be performed carefully in view of the highly paralogous nature of the DUX family of genes. Conventional cytogenetic approaches such as fluorescence in situ hybridization are challenging due to the repetitive nature of DUX4 locus. Identification of such cases, however, is important and to accurately assign risk and guide therapy.

ONLINE METHODS

Patients and samples

Diagnosis and remission samples were obtained from St Jude Children’s Research Hospital, the Children’s Oncology Group, the Alliance – Cancer and Leukemia Group B, the Eastern Cooperative Oncology Group and the MD Anderson Cancer Center. We examined leukemia samples obtained at diagnosis from 1913 children (age up to age 16), adolescents (age 16–21) and young adults (age 21–39) with B-progenitor ALL (Supplementary Table 1) using microarray gene expression profiling and single nucleotide polymorphism array analysis. The study was approved by the St Jude Institutional Review Board and informed consent was obtained from patients or legal guardians.

Genomic analysis

Whole genome, whole exome, and transcriptome sequencing, and gene expression and SNP array analysis were performed as previously described.^29,30 Additional details of bioinformatic analysis are provided in the Supplementary Note. The genomic landscape was summarized using ProteinPaint³⁴.

Whole genome and/or exome sequencing was performed for ERG-altered B-ALL cases (N=72 cases), ETV6-RUNX1 ALL (54), ALL with high hyperdiploidy (39), BCR-ABL1 ALL (39), ALL with low hypodiploidy or near haploid karyotypes (N=19), TCF3-PBX1 ALL (N=17), BCR-ABL1-like ALL (17), miscellaneous B-ALL (N=17), T-lineage ALL (N=17) and B-ALL with rearrangement of CRLF2 (7). Paired end whole genome sequencing of tumor and normal DNA was performed using HiSeq 2000 genome sequencers (Illumina) as previously described to at least 30 fold haploid coverage. Exome sequencing was performed using Truseq exome capture baits (Illumina) and GAIIx or HiSeq 2000 sequencers as previously described³⁵. Sequence mapping and variant calling was performed as described^35,36. Data were visualized in ProteinPaint.³⁷

Transcriptome sequencing was performed for 175 B-ALL cases including ERG-altered cases (54 cases), ETV6-RUNX1 (N=54), BCR-ABL1-positive and BCR-ABL1-like ALL (N=27 each), hypodiploid ALL (N=8), high hyperdiploid and miscellaneous ALL (N=5). Total RNA was extracted from leukemia cells using TRIzol (Life Technologies, NY). Total RNA quality and quantity were assessed on Agilent RNA6000 Chip (Agilent, CA) and Qubit (Life). RNA-Seq was prepared from 1 µg total RNA following Illumina RNA-Seq protocols including DNase treatment and Phenol purification, PolyA+ RNA selection by using Oligo-dT beads, cDNA conversion, fragmentation by Covaris Ultrasonicator (Covaris, MA), end repairing, deoxyadenosine tailing, adaptor ligation and PCR amplification (10 cycles). The library with 10 pM was clustered on Illumina cBot and the flowcell was loaded on HiSeq for sequencing using Illumina 2×100bp sequencing kit (Illumina, CA).

Transcriptome sequencing data were mapped to the human genome (hg19) using in-house software, and the resulting alignments were analyzed to identify evidence of both known and novel splicing events. Predicted splice junction sites were post-processed with a software package called RNApeg to correct mapping ambiguities and apply minimum quality control requirements to novel junction calls. RNApeg evaluates junctions for microhomologous mapping ambiguity versus reference junctions in the refGene, ENSEMBL, AceView, and UCSC known gene databases, also correcting against novel exon skips and single-edge matches in those isoforms. Novel junctions without any available reference anchoring are compared with each other and across samples, facilitating standardized comparison. Where ambiguity was identified, coordinates are adjusted and supporting read evidence is combined, producing a more compact and consistent set of junction calls.

The transcript expression levels in transcriptome sequencing data were estimated as Fragments Per Kilobase of transcript per Million mapped reads (FPKM). Briefly, the read counts in the GENCODE annotated gene model were obtained with the HTseq-count program.³⁸ The FPKM values were computed by normalizing the obtained read counts of the transcripts or genes with the length of the transcripts or genes and the total mapped reads.

A gene was called “expressed” in a given sample if it had a FPKM value ≥0.35 based on the distribution of FPKM gene expression levels and filtered out genes that were not expressed in any sample from the final gene expression data matrix for downstream analysis. Similar to expression arrays, limma³⁹ with estimation of false-discovery rate (FDR)⁴⁰ was also performed between ERG and other B-ALL samples (FDR<0.05).

Total stranded RNA sequencing was performed for 6 B lymphoid progenitor samples flow sorted from human bone marrow, 12 ERG ALL and 40 non ERG B-ALL cases. Expression levels of ERGalt and ALE were examined in 922 hematopoietic, brain and solid tumor cases sequenced by the St Jude Washington University Pediatric Cancer Genome Project, and adult tumor cases sequenced by The Cancer Genome Atlas (non-small cell adenocarcinoma, N=308, squamous non-small cell lung cancer, N=279, low grade glioma, N=467, glioblastoma multiforme, GBM N=167 and ovarian cancer, N=15).

ATAC sequencing

ATAC (assay for transposase-accessible chromatin) sequencing was performed using cell lines (NALM-6, Reh) and xenografts of DUX4/ERG ALL as previously described with minor modifications ⁴¹. NALM-6 and Reh were obtained from the Leibniz Institute DSMZ – German Collection of Microorganisms and Cell Cultures and were subjected to SNP array and transcriptome sequencing, and tested for Mycoplasma spp. contamination prior to use. Briefly, 50,000 cells were lysed and isolated nuclei were resuspended in 2× TD buffer. Transposase reactions were purified using MinElute columns (Qiagen) and library fragments amplified using NEB Next HiFi 2× MM (New England Biosciences), with quantitative PCR cycle optimization performed using Kapa SybrFast 2× MM (Kapa Biosystems). Post amplification products were purified using two rounds of incubation with the Agencourt Ampure XP SPRI beads at a 1.4× ratio (Beckman Coulter). The libraries were quantified via Qubit HS DNA kit (Life Technologies), and evaluated for library size distribution on a Bioanalyzer 2100 (Agilent Technologies). Sequencing was performed on a HiSeq 2000 (Illumina) generating paired 100 nucleotide reads.

Reads were aligned to hg19 with BWA with default parameters after trimming illumina adapter sequences. Observed fragment size showed similar enrichment of nucleosome free, mononucleosome, dinucleosome, trinucleosome as described.⁴¹ Reads were adjusted by transposon inserts offset, and the unique aligned nucleosome free fragments of less than 100 nucleotides in size were extraacted. Wig files were generated by extending these nucleosome free fragments to 80bp from the center and uploaded to UCSC Genome Browser for visualization. For annotation of DUX4 binding sites, we called nucleosome free regions considering every nucleosome free fragments as two single-end reads and called nucleosome free regions with MACS2⁴² (version 2.1.0.20150603; default parameters with “--extsize 200 --nomodel”).

Establishment of DUX4/ERG xenografts

Eight xenografts of human DUX4/ERG ALL were established using NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NOD-SCID gamma-null, or NSG) mice.⁴³ Primary leukemia cells were intravenously injected in NSG mice and engraftment monitored by flow cytometric analysis of peripheral blood using antibodies against human CD19 and CD45. Engraftment was monitored by retro orbital bleeding after 4 weeks, and then every 2–4 weeks as necessary and xenograft cells were harvested when levels of peripheral blood engraftment exceeded 50%.

Cloning and expression of DUX4 in human cord blood samples

All cord blood samples were obtained with informed consent according to procedures approved by the institutional review boards of the University Health Network, Trillium Health Centre, Brampton Civic Hospital, and Credit Valley Hospital, Ontario, Canada. Mononuclear cells were obtained by centrifugation on Lymphoprep medium (Stem Cell Technologies) and were depleted of Lin⁺ cells (lineage depletion) by negative selection with the StemSep Human Progenitor Cell Enrichment Kit according to the manufacturer’s protocol (Stem Cell Technologies). Lin⁻ CB cells were stored at −150°C. Lin⁻ cells were thawed by drop-wise addition of X-VIVO 10 (Lonza), 50% HyClone Cosmic Calf Serum, 100 µg/ml DNase (Roche) and were cultured at a cell density of 1.6 – 2.5×10⁶/mL in X-VIVO 10 medium supplemented with 1% BSA (Roche), L-glutamine (GIBCO), Pen/Strep (GIBCO), and the following cytokines (all from Miltenyi): SCF (100 ng/ml), Flt3L (100 ng/ml), TPO (50 ng/ml) and IL7 (IL-7; 10 ng/ml).

For lentiviral overexpression studies a pRRL based and Gateway® (ThermoFisher) adapted lentiviral vector was used in which transgene expression is driven by the SFFV promoter and tagBFP by a chimeric EF1α/SV40 promoter (J.E.D., K.K., ST., E.W., manuscript in preparation). The cDNA of wild type and truncated DUX4 were PCR amplified from leukemic cell RNA and were inserted via Gateway® LR clonase reaction according to manufacturer’s instructions (ThermoFisher). VSV-G pseudotyped lentiviral vector particles were produced by polyethyleneimine (PEI) based co-transfection of 10.5 µg pMD.G2, 20.5 µg pCMVR8.74 (both Addgene) and 38 µg transfer vector into HEK-293T cells⁴⁴ and titrated on MOLM-13 cells. For transduction, pre-stimulated (15 hrs) CB Lin- cells (4×10⁵/mL) were exposed to virus for 32 hrs at an MOI of 4.5 resulting in 22–78% BFP+ cells at day 4 post transduction when cells were sorted on FACSAria II (Becton Dickinson) to obtain BFP+ and BFP− fractions for RNA (5×10⁴) and protein lysates (0.9-5.6×10⁶).

Chromatin immunoprecipitation assays

ChIP assays were carried out as described previously.⁴⁵ Chromatin shearing was performed according to the truChIP shearing protocol (Covaris; http://covarisinc.com/wp-content/uploads/pn_010179.pdf) with minor modification. Briefly, 25×10⁶ Reh and NALM6 cells were incubated for 10 min in 1% formaldehyde in phosphate buffered saline at room temperature, quenched by adding 1/10 volume of 2.5 M glycine. The cells were then washed three times by cold phosphate buffered saline containing proteinase inhibitors and lysed on ice for 10 minutes in lysis buffer (50 mM HEPES PH7.9, 140 mM NaCl, 1 mM EDTA, 10% glycerol, 0.5% NP-40, 0.25% Triton-X100). Chromatin was washed twice in washing buffer (10 mM TrisCl, pH 8, 200 mM NaCl,1 mM EDTA, 0.5 mM EGTA), then twice in shearing buffer (0.1% SDS, 10 mM TrisCl, pH8, 1 mM EDTA) before resuspension in 1 mL shearing buffer. Chromatin was sonicated in 1 ml AFA millitubes using a Covaris E210 instrument for 20 mintues at 5% duty cycle, intensity 4, 200 cycles per burst at 7°C. The sheared chromatin was spun down 10 minutes at 13,000 rpm, 4°C and the supernatant was mixed with equal amount of ChIP-dilution buffer (0.1% SDS, 30 mM TrisCl, pH 8, 1 mM EDTA, 300 mM NaCl) before ChIP experiments. Immunoprecipitation was performed with a DUX4 antibody (ab124699) and a normal rabbit IgG control (Santa Cruz).

Real-time PCR (ΔCt method) was employed to determine the enrichment of human ERG locus by antibodies for DUX4. The signals were normalized to input as percentile. The primers used are shown in Supplementary Table 16.

To prepare ChIP-seq libraries, 10 ng of ChIP DNA was end-repaired and adapter ligation performed using the NEBNext® ChIP-Seq Library Prep Reagent Set for Illumina® (New England Biosciences). Libraries were purified after 14 rounds of PCR amplification with Q5 DNA Hot Start polymerase (New England Biosciences). Each ChIP-seq library underwent 50-cycle single-end sequencing using TruSeq SBS kit V3 on the Illumina Hiseq 2000.

Sequence reads were aligned to human genome hg19 (GRCh37) by BWA⁴⁶ (version 0.5.9-r26-dev using default parameter), and duplicated reads were then marked with Picard (version 1.65(1160)) and only non-duplicated reads retained for analysis (using SAMtools⁴⁷ (parameter “-q 1 -F 1024” version 0.1.18 (r982:295)). Cross-correlation plots were generated by a non-duplicated version of SPP⁴⁸ (version 1.11) for quality control (QC) and fragment sizes estimated using caTools (version 1.17) and bitops (version 1.0–6) implemented in R (version 2.14.0). The best fragment size estimated by the cross-correlation plot were used to extend each read to generate bigwig files for viewing in Integrative Genomics Viewer⁴⁹ (version 2.3.40), with scaling of the height of bigwig tracks by normalization to 15 million non-duplicated reads. All results of quality control analysis and visualization indicated our data passed ENCODE criteria except for Reh DUX4-ChIP seq, where results were similar to input DNA used as a conrol (relative strand correlation < 1) which was as expected due to the lack of DUX4 expression in this cell line. For annotation, MACS2 was used for peak calling (again using input DNA as a control) and only peaks overlapping in the two replicates for NALM6 were retained and merged as finalized peaks. MAST⁵⁰ in the MEME suite⁵¹ (version 4.10.2) was used for scanning DUX4 motifs (JASPAR⁵² accession MA0468.1) within finalized NALM6 DUX4 ChIP-seq peaks.

Fluorescent in situ hybridization

Fluorescence in situ hybridization for disruption of the ERG locus was performed for 10 novel BALL cases in the training cohort using diagnostic bone marrow or peripheral blood leukemic cells in Carnoy’s fixative as previously described³². BAC clones RP11-50G3 (5’ of ERG) and RP11-720N21 (3’ of ERG) were labelled with rhodamine and fluorescein isothiocyanate, respectively. At least 100 interphase nuclei were scored per case.

RT-PCR and cloning

Wild type ERG, ERGalt and ALE were amplified using Advantage 2 DNA polymerase (Clontech, Mountain View, CA) as previously described³². PCR primers are listed in (Supplementary Table 16). PCR products were purified, and sequenced directly and after cloning into pGEM-T-Easy (Promega) and subcloned into MSCV-IRES-GFP retroviral vectors.

Luciferase assays

Twenty four hours after plating 20,000 293T cells per well in a 96-well plate, cells were transfected with equimolar amounts of ERG wild type (MIG-ERG isotype 1), mutant (MIG-ERG-e6alt) or MIG empty vector along with 250 ng of pGL3-gpIX luciferase reporter plasmid and 50 ng of pRL-TK Renilla luciferase plasmid DNA (Promega) using Fugene 6 Transfection Reagent (Roche Diagnostics, Alameda, CA). Competition assays were performed by transfecting increasing amounts of mutant ERG plasmid together with a fixed amount of wild type ERG plasmid or vice versa complemented with empty vector. Forty-eight hours post-transfection, cell lysis and measurement of firefly and Renilla luciferase activity was performed using the Dual-Glo® Luciferase Assay System (Promega) according to the manufacturer’s instructions. Transfections were performed in triplicate in at least two independent experiments. The firefly luciferase - Renilla luciferase ratio was reported as mean ± s.e.m.

Immunofluorescence and immunoblotting

Cytospins of pre-B cells were fixed with 4% paraformaldehyde and washed with phosphate-buffered saline (PBS), followed by a 30 minute incubation in a blocking/permeabilization solution of 10% normal goat serum (NGS)/0.1% Triton-X 100/PBS, and then incubated for one hour with primary ERG antibody (Abcam EPR3864) diluted in 3% NGS/0.1% Triton-X 100/PBS. Slides were washed three times in PBS, and then incubated for 45 minutes in 3% NGS/0.1% Triton-X 100/PBS containing a secondary antibody conjugated to Ig-Alexa Fluor 555 (Invitrogen). Slides were washed three times in PBS and mounted with Vectashield (Vector Labs) containing 4’-6-diamidino-2-phenylindol (DAPI). All steps were carried out at room temperature. Images were captured using a Nikon C2 confocal fluorescence microscope.

Immunoblotting of cell line and leukemic cell whole cell lysates was performed as previously described^32,53 using ERG N-terminus specific antibodies (C20 and H95, Santa Cruz Biotechnologies, Santa Cruz, CA), and a C-terminus specific rabbit monoclonal antibody (Abcam 133264). DUX4 immunoblotting was performed with a C-terminus specific (E5-5, ab124699, Abcam) and N-terminus specific (aa82–131, LS-C205474, Lifespan Biosciences, Inc.) antibodies.

Gene Transduction and Transplantation of Lineage-negative Bone Marrow Cells

Mice were housed in an American Association of Laboratory Animal Care (AALAC)-accredited facility and were treated on Institutional Animal Care and Use Committee (IACUC)-approved protocols in accordance with NIH guidelines. For in vivo lineage-negative cell transplantation experiments, bone marrow from 8-week-old Arf^−/− mice⁵⁴ was extracted from tibiae and femora. Red blood cells were lysed and the remaining bone marrow cells were incubated with a cocktail of biotinylated anti-mouse antibodies (Gr-1, B220, Ly-1, Ter119, Mac-1; BD Biosciences, San Jose, CA) followed by mixing with streptavidin-coated beads (Dynabeads M-280 Streptavidin, Dynal 112.16; Life Technologies, Grand Island, NY). Cells were separated on a magnet, and unbound cells were collected and incubated at 37°C, 5% CO₂ for two days in the presence of IL-3, IL-6, SCF, IL-7, and Flt3 cytokines (Peprotech; Rocky Hill, NJ). Cells were retrovirally-transduced with MIG-ERG WT or MIG-ERGalt on RetroNectin (Takara Bio, Otsu Shiga, Japan) for 48 hours priors to sorting for GFP-positive cells (BD FACSAria, BD Biosciences, San Jose, CA). Recipient 10-week-old wildtype C57Bl/6 mice were lethally irradiated (11 Gy) 24 hours prior to transplantation with 200,000 GFP-positive cells via tail-vein injection. Animals were not randomized prior to transplantation. Animals were monitored, without blinding, for the development of leukemia clinically by measurement of GFP expression in peripheral blood and sacrificed when moribund, or at the end of the study. Post-mortem flow analysis for B220, CD19, Mac-1, Gr-1, Ter119, and Thy1 was performed on the GFP-positive population of bone marrow and spleen samples to determine lineage of disease. GraphPad Prism v.6 software was utilized to generate Kaplan-Meier survival curves and Mantel-Cox p values for pairwise comparisons of cohorts.

Lineage-negative enrichment and colony assays of murine haematopoietic cells

Bone marrow mononuclear cells were harvested from 10-week old Arf^−/− mice and labeled with biotin-conjugated lineage antibodies (Ly-6G and Ly-6C, CD11b, CD45R/B220, CD5, TER-119; BD Biosciences, San Jose, CA), followed by incubation with streptavidin coated magnetic beads (Dynabeads M-280 Streptavidin; Thermo Fisher Scientific Inc., Franklin, MA). Lineage-negative cells were purified by magnetic separation and cultured for 48 hours in Iscove’s Modified Dulbecco’s Medium/20% fetal bovine serum supplemented with penicillin-streptomycin, L-glutamine, recombinant mouse stem cell factor (SCF; 50 ng/ml), FLT-3 ligand (40 ng/ml), IL-6 (30 ng/ml), IL-3 (20 ng/ml), and IL-7 (10 ng/ml) (Peprotech, Rocky Hill, NJ). Cells were infected on RetroNectin-coated plates for 48 hours (Takara Bio Inc., Shiga, Japan) with MSCV-IRES-GFP (MIG) and MSCV-IRES-RFP (MIR) retroviruses expressing the following: wild-type, NRAS^G12D, ERG^wt, ERG^e6alt, tricistronic vector of ERG^wt and NRAS^G12D, and ERG^e6alt and NRAS^G12D for MIG; wild-type and IK6 (isoform derived from IKZF1 deletion of exon4-7) for MIR. These combinations resulted in twelve conditions. Transduced GFP and RFP-positive cells were obtained by fluorescence-activated cell sorting. For clonogenic assays, 10,000 cells were plated in triplicate in Methocult M3231 (Stem Cell Technologies, Inc., Vancouver, BC, Canada) with the appropriate factors (SCF, 100 ng/ml; IL-7, 20 ng/ml; FLT-3 ligand, 10 ng/ml) and colonies were scored 7 days later. For re-plating, 10,000 cells were cultured in identical conditions, with colonies counted on day 10–12. Colony identity was confirmed by morphological analysis through cytospin and Wright-Giemsa staining and by flow analysis for a panel of multi-lineage markers (Gr-1-PerCPCy5.5, Ter119-PE-Cy7, B220-eFluor605, CD3-APC, Mac1-Alexa700 and CD19-APC-Cy7) and specific B-cell lineage markers (CD43-PerCP-Cy5.5, IgM-PE-Cy7, and BP1-Alexa647).

Statistical analysis

Associations between categorical variables were examined using Fisher’s exact test. Associations between DUX4/ERG status with treatment outcome (event free survival and relapse) were performed as previously described^55–59. Differences in survival between mice transplanted with ERG wild type or ERG alt transduced cells were compared using the Mantel Cox test. Analyses were performed using Prism v6.0 (Graphpad, La Jolla, CA), R (www.r-project.org)⁶⁰, SAS (SAS v9.1.2, SAS Institute, Cary, NC) and SPLUS (SPLUS 7.0, Insightful Corp., Palo Alto, CA) and StatXact (v 8.0.0, Cytel Inc, Cambridge, MA).