siRNA transfections

Kasumi-1 and SKNO-1 cells and AML blasts were transfected with 200nM of siRNA using a Fischer EPI 3500 electroporator (Fischer, Heidelberg, Germany) as described previously.²⁶ For time course experiments, sequential siRNA electroporations were performed at days 0, 4 and 7. The following siRNAs were used: RUNX1/ETO siRNA (sense, 5′-CCUCGAAAUCGUACUGAGAAG-3′ antisense, 5′-UCUCAGUACGAUUUCGAGGUU-3′), mismatch control siRNA (sense, 5′-CCUCGAAUUCGUUCUGAGAAG-3′ antisense, 5′-UCUCAGAACGAAUUCGAGGUU-3′).

RNA and protein isolation, real-time PCR and western blotting

RNA was isolated 2, 4, 7 and 10 days after siRNA electroporation using RNeasy columns (Qiagen, Hilden, Germany). Total protein was precipitated by adding two volumes of acetone to the flow through of the RNeasy columns and then resuspended in urea buffer (9M urea, 4% (w/w) 3-((3-cholamidopropyl)-dimethylammonio)-propanesulphonate, 1% (w/w) dithiothreitol). Reverse transcription, real-time PCR and western blotting were performed as described.²⁷

Chromatin immunoprecipitation

Chromatin immunoprecipitation (ChIP) was performed using the following antibodies: ETO (C-terminus-specific, Santa Cruz Biotechnology, Wembley, UK, sc-9737X), RUNX1 (C-terminus-specific, Abcam, Cambridge, UK, ab23980), RNA-Polymerase II phospho S2 (ab5095) and H3K9Ac (ab4441).

DNase I hypersensitive site mapping

Genome-wide hypersensitive sites were mapped as described in Leddin et al.²⁸

Library generation and sequencing

Libraries of DNA fragments from ChIP or DNase I treatment were prepared from ~10ng of DNA using standard procedures. DNA libraries were subject to massively parallel DNA sequencing on an Illumina Genome Analyzer (Illumina, Little Chesterford, UK).

Luciferase reporter assay

Reporter constructs were generated in the pGL4-Basic plasmid with promoter sequences or in the pGL4-TK plasmid (Promega, Southampton, UK) with distal sequences. Regions of interest were amplified from purified genomic DNA from Kasumi-1 cells. Transient transfections of RAW264.7 cells were conducted using the Lipofectamine 2000 Reagent (Invitrogen, Paisley, UK) according to the manufacturer's protocol. Luciferase assays were performed using a dual-luciferase reporter assay system (Promega).

Expression arrays

Total RNA was labeled and hybridized to the Illumina HT-12 v4 BeadChip (Illumina).

Identification of DNasel- and ChIP-sequencing peaks

The raw sequence data were aligned to the hg18 assembly (NCBI Build 36.1) using BWA,²⁹ and data were displayed using the UCSC Genome Browser.³⁰ Regions of enrichment of DNaseI and ChIP data were identified using MACS software.³¹ De novo motif analysis was performed using HOMER.³²

RUNX1/ETO- and RUNX1-binding sites overlap only partially in t(8;21) cells

Both RUNX1 and RUNX1/ETO showed a similar pattern of distribution of binding sites for the subsets of peaks located within 1.5kb of transcription start sites, and 60% of the RUNX1/ETO peaks overlapped with RUNX1 peaks (Figures 2a and b). The unbiased de novo identification of enriched consensus binding motifs showed that, in contrast to RUNX1-unique peaks, both RUNX1/ETO-associated and RUNX1/ETO-unique peaks were enriched in motifs for E-box-binding proteins (Figure 2c), in agreement with observations that these factors form stable interactions with the NHR1 domain of RUNX1/ETO.¹² The same motifs were found when examining the entire set of RUNX1- or RUNX1/ETO-bound sequences in Kasumi-1 cells (Supplementary Figures 2A and B). Both RUNX1/ETO- and RUNX1-binding regions were enriched for RUNX1 consensus sequences, but only 30% of RUNX1-unique peaks contained such motifs, indicating that at sites not occupied by RUNX1/ETO, RUNX1 may bind via interaction with other transcription factors. These are likely to be ETS family members, which present the top-scoring motif in this peak population (Figure 2c). In contrast, up to 70% of shared and unique RUNX1/ETO-bound sequences contained a RUNX1 motif (Figure 2d), suggesting that the fusion protein needs to be recruited by directly anchoring it to DNA. These results contrast to recently published findings by Maiques-Diaz et al.,⁴⁶ who reported a substantially lower fraction of RUNX1 motifs in RUNX1/ETO peaks. This discrepancy may be due to the fact that in these experiments a ChIP-chip proximal promoter array was used.

Open in a separate window

Figure 2

Identification and characterization of RUNX1/ETO and RUNX1 target regions in t(8;21) cells. (a) Positional distribution of RUNX1- (left) or RUNX1/ETO- (right) binding sites relative to the transcription start site (TSS) of their nearest gene. (b) Intersection of RUNX1 and RUNX1/ETO peaks in Kasumi-1 cells. The Venn diagram shows the numbers of high-confidence peaks bound by RUNX1 and RUNX1/ETO. (c) De novo motif discovery performed on the set of regions bound by the RUNX1 and/or RUNX1/ETO in Kasumi-1 cells identified enriched RUNX consensus and different types of ETS factor-binding sites. E-box motifs were significantly enriched in peaks either unique for RUNX1/ETO or common to RUNX1/ETO and RUNX1. (d) RUNX, ETS, E-box and C/EBP consensus sequence were mapped back to all regions bound by RUNX1/ETO and RUNX1. A large proportion of regions bound by RUNX1 did not contain RUNX (TGYGGT) and/or E-box (CANNTG) consensus binding motifs, whereas most regions contain ETS (GGAW) sites.

Differential binding of RUNX1 in AML cells without a CBF complex mutation

The incomplete overlap between RUNX1 and RUNX1/ETO peaks prompted us to investigate whether RUNX1 bound to different sequences in the human genome in RUNX1/ETO-expressing cells compared to AML cells without CBF mutations. To this end we identified genome-wide high-confidence RUNX1-binding sites in AML cells with a normal karyotype (KN-AML) from a patient that expressed a full-length RUNX1 protein (data not shown) but displayed a block at a similar differentiation stage as the t(8;21) patients and had a surface marker expression profile highly similar to the two t(8;21) AML samples (Supplementary Table 2). Similar to Kasumi-1 cells, 90% of all RUNX1 peaks in the KN-AML blast cells colocalized with DHSs (Figure 3a). Direct RUNX1 targets included genes controlling hematopoietic differentiation such as CSF1R, LAT2, and LMO2 (Supplementary Figure 3). The KN-AML RUNX1 peaks showed a similar pattern of distribution around the transcription start sites as the RUNX1 peaks from Kasumi-1 cells (Figure 3b), but their genome-wide distribution and enriched motif composition was strikingly different. When directly compared (Figure 3c), only 26% of the KN-AML RUNX1-binding sites were found in Kasumi-1 cells. GATA3, GCHFR and KLF13 are examples for this notion (Supplementary Figure 3). Most importantly, in contrast to t(8;21) cells, the RUNX1 peaks unique for KN-AML showed a highly significant enrichment for binding sites for inducible factors such as AP1, C/EBP and NF-κB (Figures 3d and e).

Open in a separate window

Figure 3

RUNX1 in t(8;21)-positive and -negative leukemic cells associates with different binding sites motifs. (a) Identification of high-confidence RUNX1 peaks from blasts from a AML patient with a normal karyotype (KN-AML). Venn diagram showing the intersection of RUNX1 peaks with DHS. (b) Positional distribution of RUNX1-binding sites in KN-AML blasts relative to the transcription start site (TSS) of their nearest gene. (c) Intersection of RUNX1 peaks in KN-AML cells and in Kasumi-1 cells. The Venn diagram shows the numbers of high-confidence peaks bound by RUNX1. (d) RUNX1 consensus sequences were mapped back to all regions bound by RUNX1 in t(8;21)-negative AML blasts. A large proportion of regions exclusively bound by RUNX1 contain RUNX, ETS and E-box consensus binding motifs as well as consensus binding motifs for C/EBP and AP1. (e) De novo motif discovery performed on the set of regions bound by RUNX1 in t(8;21)-negative KN-AML blasts identifies enriched RUNX and ETS consensus sequences as well as E-box, AP1- and C/EBP-binding sites.

RUNX1/ETO knockdown leads to a shift in the pattern of RUNX1 occupancy

Our RUNX1 ChIP data indicate that the genome of hematopoietic precursor cells contains a large number of functional RUNX1-binding sites which are differentially occupied in t(8;21) AML and KN-AML cells and are associated with different binding site motifs. We therefore tested by ChIP sequencing whether depletion of RUNX1/ETO led to an alteration of RUNX1 occupancy in RUNX1/ETO expressing cells (Figure 4). The comparison between RUNX1 binding before and after knockdown showed that the depletion of RUNX1/ETO led not only to an increase in the number of RUNX1 peaks, but also to the formation of a large number of de novo RUNX1-binding sites (Figure 4a), as exemplified by the transcription factor gene NFIL3, which is upregulated by RUNX1/ETO depletion (Figure 4b, right panel). Only 10% of de novo peaks overlapped with RUNX1/ETO-bound sites (Figure 4a, bottom left). In contrast, a much larger proportion (34%) of peaks shared between knockdown and control samples were originally bound by RUNX1/ETO (Figure 4a, bottom right). Interestingly, although at most tested target regions depletion of RUNX1/ETO led to an upregulation of RUNX1 binding and gene expression (an example of this is shown with another upregulated gene, FGR, a member of the SRC kinase family, in Figure 4b, left panel), this was not always the case (Supplementary Figure 4). For instance, PU.1 (SPI1) represents a class of genes, where RUNX1/ETO binding did not or only marginally affect RUNX1 occupancy. The PU.1 upstream regulatory element contains an upstream enhancer element (PU.1 (E)) that is bound by RUNX1 at multiple sites, and this binding is vital for enhancer activity,⁴⁷ but it is also bound by RUNX1/ETO (this study). At this element RUNX1 binding did not respond to RUNX1/ETO knockdown, and gene expression was barely altered, which may indicate a complex binding pattern with multiple occupancies of both RUNX1/ETO and RUNX1.

Open in a separate window

Figure 4

Knockdown of RUNX1/ETO leads to a redistribution of RUNX1 binding. Kasumi-1 cells were electroporated with either mismatch control siRNA (siMM) or RUNX1/ETO siRNA (siRE). Two days after siRNA electroporation, RUNX1 binding was measured by ChIP sequencing in both populations. (a) Top panel: Venn diagram showing the appearance of new RUNX1-binding sites after RUNX1/ETO depletion. Bottom left: Venn diagram showing the overlap between RUNX1/ETO-bound regions and de novo RUNX1 sites, demonstrating that the latter are distinct from sites previously bound by RUNX1/EO. Bottom right: Venn diagram showing the overlap between RUNX1/ETO-bound sequences and sites bound by RUNX1 before and after RUNX1/ETO depletion. (b) Example of alterations in RUNX1 binding. Left panel shows UCSC browser images depicting one gene (NFIL3) at a site not previously bound by RUNX1/ETO and another gene (FGR) showing a de novo RUNX1 with a RUNX1/ETO-bound site showing an increase in RUNX1 binding at this site after knockdown.

RUNX1/ETO depletion can lead to both activation and repression of its direct target genes

In order to examine the effects of RUNX1/ETO depletion on gene expression at the global level and to correlate these changes with identified RUNX1/ETO target genes, we generated expression profiles during a time course of sustained RUNX1/ETO knockdown for 10 days in Kasumi-1 cells (Figure 5, Supplementary Table 6, Supplementary Figure 5A). Hierarchical clustering identified four groups of genes differentially responding to knockdown, consisting of early (Group I) and late (Group III) upregulated genes, downregulated genes (Group II) as well as genes showing a complex pattern of response (Group IV; Figure 5a). These results were validated by manual analysis with a selection of early responding genes both in Kasumi-1 cells and in primary t(8,21) cells obtained from three different AML patients (Figure 5b, Supplementary Figure 5B).

Open in a separate window

Figure 5

Analysis of RUNX1/ETO-dependent gene expression patterns. (a) Hierarchical clustering of genes responding by an at least twofold change in expression levels to RUNX1/ETO knockdown in Kasumi-1 cells. The heat map shows early upregulated (Group I), downregulated (Group II) and late upregulated genes (Group III) over a time course of 10 days. At the bottom of the heat map note non-clustered genes that are either upregulated more than threefold or show a more complex response pattern (Group IV). Expression levels were compared between RUNX1/ETO siRNA and mismatch siRNA-treated cells. Time points are indicated on the top of the heat map. Dark red indicates highly upregulated genes and black indicates highly downregulated genes. (b) Effect of RUNX1/ETO depletion on gene expression in primary t(8;21) AML blasts. The graph shows real-time PCR-based validation of early responding genes. The columns represent the means for three t(8;21) AML patients and the error bars the s.e.m. Inset: immunoblot showing siRNA-mediated depletion of RUNX1/ETO in blasts from a t(8;21) AML patient. (c) GSEA ranked according to the correlation of genes with a metagene (F1) summarizing the gene expression profile of Kasumi-1 cells after RUNX1/ETO knockdown. From left to right: significant enrichment of gene sets downregulated in human hematopoietic stem cells upon RUNX1/ETO overexpression⁶ and enrichment of genes determined in this study to have a high-confidence RUNX1/ETO-binding site with a corresponding DHS in the region 2kb upstream of the start of transcription in Kasumi-1 cells. P, nominal P-value; q, false discovery rate. (d) The numbers of upregulated and downregulated genes upon RUNX1/ETO depletion in Kasumi-1 cells. The bottom row indicates genes with RUNX1/ETO peaks. (e) Classification of RUNX1/ETO target genes. The columns show the percentage of genes with high-confidence RUNX1/ETO peaks of all genes with changed gene expression upon RUNX1/ETO knockdown. siMM, mismatch siRNA; siRE, RUNX1/ETO siRNA.

To compare global gene expression profiles, we produced two coordinately regulated metagenes F1 and F2 in an unsupervised manner that summarized the expression of RUNX1/ETO-dependent genes in a single score using non-negative matrix factorization.^{48, 49} The F1 and F2 metagenes were highly expressed in Kasumi-1 cells with and without RUNX1/ETO knockdown, respectively. We also tested the relevance of the RUNX1/ETO-dependent global gene expression changes for another t(8;21)-positive cell line, SKNO-1, and for primary t(8;21)-positive AML. Kasumi-1-derived RUNX1/ETO-dependent metagene expression was mirrored and validated in the SKNO-1 and primary AML data sets, demonstrating an excellent concordance between the RUNX1/ETO-associated global gene expression changes in these three different t(8;21)-positive cell types (Supplementary Figure 5C). Furthermore, RUNX/ETO-associated gene expression changes inversely correlated with gene expression changes observed in previous knockdown studies and in overexpression studies with primary human CD34+ cells (Figure 5c, Supplementary Figures 5D and E).^{6, 27, 39} These combined analyses confirmed that RUNX1/ETO-associated shifts in gene expression in Kasumi-1 faithfully reflect gene expression features in the human t(8;21) AML.

The metagene was also used to rank genes for the purpose of GSEA according to their correlation with the metagene score (Supplementary Table 7). Genes with high-confidence RUNX1/ETO peaks were highly enriched in the RUNX1/ETO knockdown signature (Figure 5c). However, not all genes responding to RUNX1/ETO knockdown were associated with RUNX1/ETO peaks. Genes responding early at day 2, such as CST7, IGFBP7 and MTSS1, consisted of 50% direct RUNX1/ETO target genes, whereas this was true for only 20% of the late responding genes at day 10 (Figures 5d and e). Moreover, almost 80% of all early responding target genes were upregulated upon RUNX1/ETO knockdown, in contrast to 60% of all late responding target genes (Figures 5d and e). This indicates that downstream effects of the removal of RUNX1/ETO contribute rapidly and progressively to the reorganization of the transcriptional network.

The depletion of RUNX1/ETO leads to complex changes in the histone acetylation pattern and RNA Polymerase II occupancy

As RUNX1/ETO recruits histone deacetylases to its target promoters^{13, 14} and alters gene transcription, we examined the immediate consequences of RUNX1/ETO depletion on histone H3 lysine 9 acetylation (H3K9Ac) and occupancy of the elongating form of RNA Polymerase II (RNA Pol II). We therefore performed ChIP experiments with RUNX1/ETO-depleted and -control cells at day 2 after knockdown, and integrated RNA Pol II and H3K9Ac data in Kasumi-1 cells on RUNX1/ETO target loci with the gene expression data (Figures 6a and b, Supplementary Table 6). After 2 days of knockdown, half of the early upregulated genes, including IGFBP7, NFE2 and PRG2, exhibited more than twofold increased RNA Pol II occupancy, whereas most downregulated genes such as CD34 and 75% of the late upregulated genes generally showed no or little change in RNA Pol II binding at this stage (Supplementary Figure 6A).

Open in a separate window

Open in a separate window

Figure 6

RUNX1/ETO silencing leads to changes in RNA-Polymerase II occupancy and the histone H3K9 acetylation pattern at RUNX1/ETO target genes. (a) Comparison of RNA Pol II occupation with transcriptional profiling reveals a substantial correlation between changes in gene expression and changes in RNA Pol II association upon RUNX1/ETO knockdown. RNA Pol II occupation was analyzed 48h after siRNA treatment, ranked according to fold change in occupancy and compared with changes in gene expression during a time course of 10 days with siRNA treatment. Dark red indicates high occupancy or upregulation, black low occupancy or downregulation, respectively. (b) The H3K9Ac pattern correlates partially with changes in gene expression associated with RUNX1/ETO depletion. H3K9Ac occupation was analyzed and compared analogously to (a). (c, d) UCSC Genome Browser image of the human ZBTB16 and WT1 loci depicting ChIP-Seq tags for RUNX1/ETO, RNA-Pol II and H3K9Ac in mock-treated (control), mismatch siRNA (siMM) and RUNX1/ETO siRNA (siRE)-treated Kasumi-1 cells. (e) Heat map resulting from an unsupervised clustering of H3K9Ac- and RUNX1/ETO-binding sites with and without RUNX1/ETO knockdown (siMM and siRE, respectively) as well as without transfection (control) showing two groups of sequences and their genomic location. In each lane, the ChIP enrichment score is shown 10kb upstream and downstream from the peak center. (f) Integration of the sequence enrichment of each group into a density plot showing the location of RUNX1/ETO-binding sites with and without knockdown in relation to H3K9Ac. Dark blue: H3K9Ac (siMM); pink: H3K9Ac (siRE); green: RUNX1/ETO (control); yellow RUNX1/ETO (siMM); light blue: RUNX1/ETO (siRE).

A more complex pattern of changes at RUNX1/ETO peaks was seen when measuring H3K9Ac. Although after 2 days of knockdown most upregulated genes displayed a rapid increase in H3K9Ac, we also found an increase at 50% of the downregulated genes (Figure 6b). This group of down-regulated genes partially overlapped with genes displaying an increase in RNA Pol II occupancy (Figure 6a, Supplementary Figures 6B and C). Such peaks were often localized in introns (for example, ZBTB16 or WT1, Figures 6c and d) indicating that not all RUNX1/ETO peaks that are associated with alterations of gene expression colocalize with classical promoter or enhancer elements. This diversity was also reflected in reporter gene assays of the promoters of target genes (Supplementary Figure 6D): Many, but not all, of these promoter elements directly responded to transactivation by RUNX1 and repression by RUNX1/ETO.

We next used our histone acetylation data to examine the chromatin architecture and the genomic distribution of RUNX1/ETO-binding sites (Figures 6e and f). Unsupervised clustering showed that RUNX1/ETO-binding sites could be clustered into two groups (Figure 6e). The largest group contained intergenic and intragenic sites that are most likely to be enhancers, as these elements are also DNase I hypersensitive. The analysis of the histone H3K9 acetylation profile before and after knockdown (Figure 6f) showed that in both groups the RUNX1/ETO complex was flanked by acetylated histones, demonstrating that histone acetylation was not absent in such elements. This is consistent with findings that described the presence of histone acetyl transferases at these sites.^{7, 20} However, the depletion of RUNX1/ETO led to a strong increase in H3K9Ac, while maintaining a protein complex between nucleosomes, demonstrating that RUNX1/ETO does not lead to the formation of inactive chromatin, but shifts the balance between active and inactive chromatin.

This work was funded by a specialist program grant from Leukaemia and Lymphoma Research (CB, PC, RT), grants from Yorkshire Cancer Research (PC and CB), Cancer Research UK (BDY), the NIH (NIH R01 CA118316, DGT) and Kay Kendall Leukaemia Fund (OH). We thank Berthold Gottgens (Cambridge) for helpful comments on the manuscript.