RUNX1 Is a Key Target in t(4;11) Leukemias that Contributes to Gene Activation through an AF4-MLL Complex Interaction

Summary The Mixed Lineage Leukemia (MLL) protein is an important epigenetic regulator required for the maintenance of gene activation during development. MLL chromosomal translocations produce novel fusion proteins that cause aggressive leukemias in humans. Individual MLL fusion proteins have distinct leukemic phenotypes even when expressed in the same cell type, but how this distinction is delineated on a molecular level is poorly understood. Here, we highlight a unique molecular mechanism whereby the RUNX1 gene is directly activated by MLL-AF4 and the RUNX1 protein interacts with the product of the reciprocal AF4-MLL translocation. These results support a mechanism of transformation whereby two oncogenic fusion proteins cooperate by activating a target gene and then modulating the function of its downstream product.


INTRODUCTION
Aberrant epigenetic changes are a driving force in many cancers and are excellent candidates for the development of targeted therapies (Dawson and Kouzarides, 2012). The design of such therapies depends on a clear understanding of the molecular details of disease progression. The Mixed Lineage Leukemia (MLL) protein is an example of an important epigenetic protein that is mutated in a subset of aggressive leukemias (Marschalek, 2010), and thus provides a useful model for studying the link between epigenetic changes and cancer progression.
MLL is important for the epigenetic maintenance of gene activation and is required for normal hematopoietic development (Jude et al., 2007;McMahon et al., 2007). MLL leukemogenic mutations include chromosomal translocations (commonly called MLL rearrangements [MLLr]) that fuse the N terminus of the MLL gene inframe with any one of over 60 different partner genes, producing novel fusion proteins (MLL-FPs). Almost 90% of all MLL-FPs are fusions with AF4, AF9, ENL, ELL, AF10, or AF6 (Meyer et al., 2009). t(4;11)(q21;q23) chromosome translocations (referred to from this point as t(4;11) translocations) fuse MLL in-frame with the AF4 gene and produce both MLL-AF4 and AF4-MLL fusion proteins (Bursen et al., 2004;Bursen et al., 2010). t(4;11) translocations are a major cause of infant acute lymphoblastic leukemia (ALL) and produce an aggressive disease with a poor prognosis. Enforced expression of MLL-AF4 alone is incapable of transforming human CD34 + cord blood (Montes et al., 2011), and mouse models expressing MLL-AF4 alone are not fully representative of the human disease, instead producing B-cell lymphomas (Chen et al., 2006;Metzler et al., 2006) acute myeloid leukemia (AML), or precursor B-ALLs (pre-B-ALLs) (Krivtsov et al., 2008). Conversely, expression of both MLL-AF4 and AF4-MLL together result in either common lymphoid progenitor leukemia or mixed lineage leukemia (MLL), a close recapitulation of the human disease (Bursen et al., 2010). Unlike many acute leukemias, t(4;11) leukemias are associated with very few cooperating mutations (Bardini et al., 2010(Bardini et al., , 2011. This suggests that the t(4;11) translocation by itself may be sufficient for leukemic transformation, perhaps because both MLL-AF4 and AF4-MLL fusion proteins are capable of altering the epigenetic information content of the cell (Benedikt et al., 2011;Krivtsov et al., 2008). Interestingly, knocking down the MLL-AF4 fusion protein alone is sufficient to disrupt t(4;11) leukemic growth in vivo (Thomas et al., 2005), indicating that targeting pathways controlled by the MLL-AF4 protein could be effective in treating t(4;11) leukemias.
Although work with BRD4 inhibitors suggests that multiple MLL-FPs use the same molecular pathway for leukemogenesis (Dawson et al., 2011;Zuber et al., 2011), this cannot explain the fact that MLL-AF4, MLL-AF9, and MLL-ENL produce different leukemias even when expressed in the same cell type (Drynan et al., 2005;Metzler et al., 2006). Furthermore, gene expression analyses in t(4;11), MLL-ENL, and MLL-AF9 patient samples display overlapping as well as distinct gene expression profiles (Stam et al., 2010;Trentin et al., 2009), indicating that individual MLL-FPs could activate unique gene expression pathways.
In this study, we set out to further analyze t(4;11) leukemias on a molecular level. We initially used chromatin immunoprecipitation sequencing (ChIP-seq) and  in patient cell lines to identify key gene targets regulated by MLL-AF4. One direct target of the MLL-AF4 protein is the RUNX1 gene, a key hematopoietic transcription factor that is specifically overexpressed in t(4;11) patient samples. Distinct from other MLL-FPs, RUNX1 expression is important for the growth of t(4;11) leukemia cell lines, in which it plays a role in the activation of specific target genes. Furthermore, we show that RUNX1 interacts with an AF4-MLL complex, providing a new model of how MLL-AF4 and AF4-MLL cooperate on a molecular level. Such a cooperative effect between these two fusion proteins could explain some of the differences between t(4;11) and other MLL-FP leukemias.

Common MLL-AF4 Target Genes Are Overexpressed in Primary B-ALL Patient Samples
To identify potentially important direct target genes of MLL-AF4 in t(4;11) leukemias, we performed ChIP-seq in the RS4;11 cell line and compared our results to a previously published data set from SEM cells (Guenther et al., 2008). RS4;11 and SEM cell lines are both t(4;11) pre-B-ALL patient-derived cell lines (see Extended Experimental Procedures for details on cell lines) that express the MLL-AF4 protein as well as wild-type MLL and wild-type AF4.
No single antibody has been developed to uniquely recognize endogenous MLL-AF4. Instead, using an approach originally taken by Guenther et al. (2008), we performed ChIP-seq experiments using antibodies against the N terminus of MLL (aMLL-N) and the C terminus of AF4 (aAF4-C) ( Figure 1A). To find actively transcribed gene targets bound by MLL-AF4, we identified promoters in RS4;11 cells enriched for both MLL-N and AF4-C as well as H3K79Me2 (an active transcription elongation mark that is highly enriched at important MLL-FP target genes; Krivtsov et al., 2008;Milne et al., 2005) enrichment within the gene body ( Figures 1B-1D). We identified 603 gene targets that met all three criteria ( Figure 1D; Table S1). Two previously identified direct targets of MLL-FPs, the HOXA cluster (Bernt et al., 2011;Guenther et al., 2008;Milne et al., 2005) and CDKN1B (Bernt et al., 2011;Xia et al., 2005), are shown as examples (Figure 1B and C). A similar approach with the SEM cell ChIP-seq data set (Guenther et al., 2008) identified 2,490 putative MLL-AF4 target genes ( Figure S1A; Table S1), which produced a common overlap of 491 genes ( Figure 1E; Table  S1). The 491 target set includes previously identified targets such as JMJD1C, BCL2, FLT3, MYB, MYC, RUNX2, MEIS1, CDKN1B, and HOXA cluster genes, as well as some other potentially interesting gene targets such as EZH2, FOXP1, IZKF1, IZKF2, and SOX4 (Table S1).
In MLLr B-ALL patient samples from three large clinical studies, the average expression of the 491 MLL-AF4 target genes was significantly higher than that of nontarget genes (Figures 1F-1H and S1B-S1D). The 491 target gene set is also significantly overexpressed in MLLr ALL compared to several other B-ALL subtypes (E2A-PBX1, ETV6-RUNX1, and pre-B; Figure S1E-S1G), although not others (e.g., BCR-ABL; Figure S1G). There is no significant difference between t(4;11) and other MLLr patient samples ( Figure 1I), suggesting that this 491 gene target set is generally overexpressed in patients with MLLr ALLs. This correlation between ChIP-seq and gene expression data in patient samples validates ChIP-seq in patient cell lines as a powerful method to identify important target genes, and also suggests that our 491 common MLL-AF4 targets have an in vivo relevance to MLLr leukemia in human patients.

RUNX1
Is Overexpressed in Primary ALLs with t(4;11) High expression of HOXA9, HOXA10, and MEIS1 is considered to be a general hallmark of all MLL-FP leukemias, but a detailed analysis of patient expression data show that many t(4;11) leukemias do not express high levels of these genes (Stam et al., 2010;Trentin et al., 2009), indicating that other additional targets are likely to have an important role in t(4;11) leukemogenesis.
Among our list of 491 potential MLL-AF4 target genes, the master hematopoietic transcription factor RUNX1 (AML1) is highly enriched for MLL-N, AF4-C, H3K79Me2, and H3K4Me3 in RS4;11 and also in SEM cells, at both of its two promoters, and at the hematopoietic +23 enhancer element (Nottingham et al., 2007) (Figures 2A, 2B, and S2A). Although Guenther et al. (2008) used different specific antibodies than those used in this study, a direct comparison using our own antibodies in conventional ChIP experiments suggests that RS4;11 and SEM cells have similar levels of MLL-N, AF4-C, H3K4Me3, and H3K79Me2 enrichment across RUNX1 ( Figure S2A).
Mutations in RUNX1 are commonly associated with AML but are also found in B-ALL and T-ALL, and are usually inacti-  (MLL-C) proteins. The t(4;11) breakpoint is marked by a black arrowhead labeled ''bp.'' The translocation fuses part of MLL-N inframe with AF4-C (red box), and also produces a reciprocal AF4-MLL fusing AF4-N (violet box) with the rest of MLL. Antibody positions on the wild-type and fusion proteins are shown. A RUNX1 interaction domain at the C-terminal SET domain  is indicated by blue shading. (B and C) ChIP-seq in RS4;11 cells across the HOXA cluster (B) and CDKN1B (C). The number of reads for peak summits was normalized by the total number of reads per track (set to 1 Gb for each track). Four different primer sets used for real-time PCR ChIP analysis are shown (red boxes) for the following amplicons: A9, A10, CDKN1B-A, and -B. (D) ChIP-seq in RS4;11 cells using antibodies to MLL-N, AF4-C, and H3K79Me2 produced an overlap at 603 target genes. (E) Comparison between the 603 RS4;11 target gene set from (D) and similar ChIP-seq data from SEM cells (Guenther et al., 2008) produced a set of 491 common MLL-AF4 targets (see Table S1).  (Geng et al., 2012). (I) The same data as in (H), split into t(4;11) versus other MLLr patient samples. Boxplots (F-H) represent the median values and error bars represent extreme maximum and minimum whisker values for each plot. Bar plots (I) are the mean and error bars represent SEM. See also Table S1 and Figure S1.
vating, suggesting that RUNX1 normally functions as a tumor suppressor (Blyth et al., 2005;Mangan and Speck, 2011;Zhang et al., 2012). However, overexpression of wild-type RUNX1 can be oncogenic (Blyth et al., 2005). Thus, considering the crucial role of RUNX1 in hematopoiesis and many acute leukemias, we decided to further explore its potential role in t(4;11) leukemias.
We analyzed the expression of RUNX1 and other target genes in specific subsets of primary ALL samples, including t(4;11) and other MLLr samples. Average HOXA9, HOXA10, and CDKN1B expression is significantly higher in MLLr leukemias than in other ALL subtypes (Figures 2C-2E and S2B-S2G; Table S2), but no significant difference in expression levels is seen when comparing MLLr and t(4;11) leukemias (in the Eastern Cooperative  Table S2). Importantly, in the ECOG E2993 patient set, RUNX1 is significantly overexpressed in t(4;11) samples compared to the other MLLr samples ( Figure 2H; Table S2). Interestingly, the non-t(4;11) MLLr samples in the ECOG E2993 data set appear to have a lower than average expression of RUNX1 compared to other leukemia subtypes ( Figure 2H). One possibility is that t(4;11) samples account for the bulk of the high-expressing RUNX1 samples in the St. Jude and Children's Oncology Group (COG) P9906 data sets ( Figures  2F and 2G), but unfortunately, because we do not have t(4;11)specific data on individual MLLr samples in these data sets, we cannot test this directly.
These results are also consistent with a recent analysis that showed RUNX1 is specifically overexpressed in t(4;11) samples compared to several other childhood ALL samples (Montero-Ruíz et al., 2012). It is worth pointing out that the statistically significant increase in RUNX1 expression in the ECOG E2993 data set only represents a 1.3-to 2.3-fold change in microarray expression (Tables S3 and S4). However, a small change in messenger RNA levels for an important master regulatory protein such as RUNX1 could represent a much bigger effect at the protein level. Taken as a whole, these results suggest the possibility that RUNX1 could have a unique role in t(4;11)-mediated leukemogenesis, and we therefore decided to analyze its possible role on a more detailed molecular level.
MLL-AF4 Directly Regulates RUNX1 and Other Target Loci by Stabilizing ENL and AF9 Binding Guenther et al. (2008) previously rejected RUNX1 as a potential MLL-AF4 target gene because it displayed MLL-N, MLL-C, AF4-C binding, and H3K4Me3 and H3K79Me2 in both SEM and the control REH (non-MLLr) cell lines. To determine if MLL-AF4 is a key regulator of RUNX1 expression, MLL-AF4specific siRNA (Thomas et al., 2005) knockdowns were performed in RS4;11 and SEM cell lines. At both the RNA and protein levels, we saw an MLL-AF4-dependent loss of RUNX1 expression ( Figures 3A-3C). Importantly, we also found that wild-type MLL had no effect on HOXA9 or RUNX1 regulation ( Figure 3B), suggesting that MLL-AF4, but not wild-type MLL, is key to maintaining the expression of these target genes.
Because MLL-AF4 is key to maintaining HOXA9 and RUNX1 target gene expression, we wanted to determine if MLL-AF4 was responsible for assembling a specific complex at these target genes in vivo. The AF4-C portion of MLL-AF4 interacts directly with wild-type AF9, ENL, and AFF4, and weakly homodimerizes with wild-type AF4, providing an indirect interaction between MLL-AF4 and the Cyclin T1/CDK9 pTEFb complex (Benedikt et al., 2011;Biswas et al., 2011;Lin et al., 2010;Mueller et (Harvey et al., 2010). (E and H) ECOG E2993 clinical trial (Geng et al., 2012). An asterisk indicates significantly lower average expression for the leukemia subtype relative to MLLr (C, D, F, and G) or relative to t(4;11) (E and H). A two-tailed Wilcoxon test was used to calculate p values, and p values for the different comparisons are in Table S2. See also Tables S3, S4, and Figure S2.
SEM cells, specific siRNA knockdowns of MLL-AF4 reduced binding of MLL-N, AF4-C, and the MLL-AF4 interacting factors ENL and AF9 at RUNX1, HOXA9, HOXA10, and CDKN1B ( Figure 3E), but had no effect on the binding of Cyclin T1 and AFF4. Together, these results indicate that RUNX1, HOXA9, HOXA10, and CDKN1B are direct targets of MLL-AF4, and that MLL-AF4 stabilizes the recruitment of AF9 and ENL, but not Cyclin T1 or AFF4.

t(4;11) Cell Lines Support Higher Levels of RUNX1 Expression Than Other MLL-FP Leukemias
To further analyze the potential importance of RUNX1 in t(4;11) leukemias, we compared gene expression patterns and complex assembly at target genes in different MLL-FP cell lines. Typically, both HOXA9 and HOXA10 are highly expressed in MLL-FP cell lines and show almost no expression in non-MLL-FP cell lines ( Figure 4A, top and middle). Although RUNX1 gene expression is complicated by the fact that it appears to be generally higher in ALLs compared to AMLs, consistent with the primary patient data in Figure 2H, RUNX1 expression is upregulated in t(4;11)containing cells compared to other MLL-FPs ( Figure 4A, bottom). In general, although there are some unique isoforms specific to different cell types (perhaps reflecting myeloid versus lymphoid origins), RUNX1 protein levels are higher in t(4;11) cells than in other MLL-FP cells ( Figure 4B).   Figure 4D), in combination with the data in Figure 3, we think the most likely explanation for these results is that MLL-AF4 differs from other MLL-FPs and increases stable AF9 and ENL binding at RUNX1.

RUNX1 Is Required for the Growth of t(4;11) Cells
To determine if RUNX1 expression is important for the leukemic growth of different MLL-FPs, we used colony-forming assays coupled with RUNX1 siRNA knockdowns in SEM (t-4;11), MV4-11 (t-4;11), and THP-1 (MLL-AF9) cells. Cells collected 24 hr after plating contained $50% of RUNX1 mRNA compared to a nontargeting siRNA control ( Figure 5A) and resulted in a large reduction in RUNX1 protein levels ( Figure 5B). In SEM and MV4-11 cells, RUNX1 siRNA treatment inhibited clonogenic ability by $60% after 14 days, while little effect was observed in THP-1 cells ( Figures 5C and 5D). Similar t(4;11) sensitivity to RUNX1 levels was observed in cell growth assays comparing SEM cells and KOPN-8 cells after RUNX1 siRNA treatment ( Figures S4A-S4C). Together, these results suggest that RUNX1 expression specifically contributes to the growth of t(4;11) cells but not other common MLL-FPs.
High RUNX1 Expression Correlates with a Poor Clinical Outcome in ALL Minimal residual disease (MRD) after treatment is generally considered to be an indicator of poor prognosis. In the COG P9906 clinical trial (Harvey et al., 2010), 191 out of 207 ALL patients had MRD data available. As expected, the 67 MRD+ patients had a significantly worse overall survival and relapsefree survival than the 124 MRDÀ cases ( Figures 5E and 5F). We found that the 124 MRDÀ patients had a significantly lower average level of RUNX1 expression compared to the 67 MRD+ patients ( Figure 5G). Among the 191 patients, 17 harbor MLL-FPs (MLLr), among which, the 9 patients that were MRD+ at day 29 had a higher average RUNX1 expression than the 8 patients that did not ( Figure 5H). Interestingly, when these MLLr patients are removed from the data set, the resulting 174  non-MLLr ALL patients showed no significant correlation between RUNX1 expression and MRD status ( Figure 5I). Although we unfortunately do not have specific data for t(4;11) leukemias, the correlation between higher RUNX1 expression levels and worse clinical outcomes in MLLr patients suggests that RUNX1 expression can directly contribute to leukemogenesis in human patients.

RUNX1 Activates Target Genes in t(4;11) Leukemic Cells
To understand the function of the RUNX1 protein in t(4;11) leukemic cells, we performed RUNX1 ChIP-seq in SEM cells and identified 11,013 genes directly bound by the RUNX1 protein (Figures 6A-6D and S5A; Table S5). Interestingly, recent work has shown that RUNX1 can interact with the wild-type MLL protein complex , and we found 3,294 genes that show a specific overlap between MLL-C/H3K4Me3 binding and RUNX1 (Figures 6A-6D and S5A; Table S5). RUNX1 also binds to 617 MLL-AF4 targets (i.e., MLL-N/AF4-C/H3K79Me2 binding sites, Figures 6A and 6C) and 1,664 genes where all the proteins overlap ( Figure 6A; Table S5), including MEF2C ( Figure 6B) and the RUNX1 gene itself ( Figure S5A).
MEF2D and JUNB were both previously identified among a set of 380 genes tightly regulated by RUNX1 in K562 cells , whereas SPI-1 (aka PU.1) is a previously identified important target gene in RUNX1-mediated leukemogenesis . MEF2D, JUNB, and SPI-1 are all bound by RUNX1 in SEM cells ( Figure 6A, S5A, and S5B; Table S5). Interestingly, loss of RUNX1 protein levels appears to have the strongest effect on target genes bound primarily by RUNX1 and MLL-C ( Figures 6E and 6F).
Wild-type MLL knockdowns reduce expression of some MLL-C/RUNX1-bound gene targets, although not to the same degree as knockdowns of RUNX1 ( Figure 6F). Even though MLL-AF4 does not bind directly to SPI-1 or MEF2D ( Figures 6D and  S5A), MLL-AF4-specific knockdowns reduce expression of both of these target genes, likely due to the reduction of RUNX1 protein levels ( Figure S5C). Importantly, RUNX1 knockdowns in THP-1 cells did not reduce target gene expression, and in some cases actually increased expression of RUNX1 target genes ( Figure S5D). Taken as a whole, these data suggest that RUNX1 is functioning as an activator at certain key target genes in t(4;11) SEM cells, and MLL-C:RUNX1-bound target genes are particularly sensitive to the loss of RUNX1.

RUNX1 Activates Gene Targets in t(4;11) Cells by Cooperating with an AF4-MLL Complex
Past work revealed that AF4-MLL is expressed in human patients (Kowarz et al., 2007) and contributes to t(4;11) leukemogenesis (Bursen et al., 2004(Bursen et al., , 2010. AF4-MLL can alter the epigenetic profile of target genes by interacting with components of the SEC and the wild-type MLL-C complex (Benedikt et al., 2011; Figure 6G), but AF4-MLL does not function primarily through the activation of canonical MLL-AF4 target genes such as HOXA9 (Bursen et al., 2010). RUNX1 directly interacts with the C-terminal SET domain of MLL , suggesting that RUNX1 could be a component of a wild-type MLL and an AF4-MLL:MLL-C complex ( Figure 6G, interactions 2 and 3, respectively).
To determine if RUNX1 exists in a complex with AF4-MLL (see Figure 6G), we performed immunoprecipitation (IP) experiments with RS4;11 and SEM nuclear extracts ( Figures 6H and 6I). We found that aRUNX1, aMLL-C, and aAF4-N could coIP a complex containing RUNX1, MLL-C, wild-type AF4 (black arrowhead), and a band that corresponds to the cleaved $194 kDa AF4-MLL protein (white arrowhead; see the legend for Figure S5F for an explanation of the apparent molecular weights of these proteins). AF4-N IPs in CCRF-CEM nuclear extracts failed to detect this 194 KDa AF4-MLL band, and were less enriched for RUNX1 and the MLL-C complex than comparable IPs in SEM or RS4;11 cells ( Figures S5E and S5F). Together, these results support the possibility that AF4-MLL exists in a complex with both MLL-C and RUNX1.   Figure S4. RUNX1 siRNA experiments that reduce expression of SPI-1, MEF2D, JUND, and JUNB ( Figure 6F) disrupt binding of AF4-N, MLL-C, and the MLL-C complex component RBBP5 to these target genes in vivo ( Figures 6J-6M). Further, expression of MEF2D, JUNB, and SPI-1 is higher in SEM and RS4;11 cells than in CCRF-CEM cells ( Figure S5G), and this correlates with an increased binding of AF4-N ( Figures S5H and S5I). Increased AF4-N binding is seen even at the MEF2D target gene, which has approximately equal levels of RUNX1 binding in CCRF-CEM cells compared to RS4;11 and SEM cells ( Figures S5I and  S5J). Unfortunately, AF4-MLL-specific siRNAs failed to reduce AF4-MLL protein levels ( Figures S5K-S5M), and we were not able to directly test whether AF4-MLL regulates RUNX1 target genes. However, taken as whole, these data show that RUNX1 activates certain key target genes in t(4;11) pre-B-ALL cells, and it might accomplish this through recruitment of an RUNX1:MLL-C:AF4-MLL complex (Figure 7).

DISCUSSION
MLL-FPs are thought to promote leukemogenesis through the epigenetic activation and maintenance of master regulatory factors such as HOXA9 and MEIS1, which set up gene expression networks responsible for constitutive activation of cellular growth and proliferation pathways. However, in t(4;11) patient samples, half of the leukemias analyzed do not have elevated levels of HOXA expression (Stam et al., 2010;Trentin et al., 2009), and low-level HOXA expression actually correlates with a worse prognosis (Stam et al., 2010). Furthermore, AF4-MLL is able to induce leukemias in mice without activating HOXA or MEIS1 expression (Bursen et al., 2010). Together, these results suggest that t(4;11) leukemias may activate alternate pathways that are not dependent on HOXA or MEIS1 expression.
In this analysis, we have identified a 491 target gene set that is generally highly expressed among MLLr leukemias. RUNX1 is a unique exception to this in that it is specifically overexpressed in t(4;11) leukemias ( Figure 2H and Montero-Ruíz et al., 2012). RUNX1 siRNA knockdowns inhibited clonogenicity of t(4;11) (SEM and MV4-11) cells but not MLL-AF9 (THP-1) cells, indicating that the oncogenic role for RUNX1 in t(4;11) leukemia appears to be t(4;11) specific but lineage independent, with both B-ALL (SEM) and AML (MV4-11) affected.
Recent analyses have suggested that MLL-AF4 promotes transcription elongation by stabilizing the binding of factors such as pTEFb, DOT1L, ELL, AFF4, AF9, and ENL at target genes in vivo (Lin et al., 2010;Yokoyama et al., 2010). AF4-MLL has been shown to activate gene targets through a similar ability to promote transcription elongation by interacting with a pTEFb-containing complex (Benedikt et al., 2011). Here, we show that the RUNX1 protein can interact with an AF4-MLL complex and stabilize its binding to certain gene targets. Thus, leukemic cells that express AF4-MLL produce an additional coactivator complex (Benedikt et al., 2011) that may push the balance toward RUNX1 functioning as a general activator, and this may have an impact on whether RUNX1 is a tumor suppressor or an oncogene in different cell types ( Figure 7A). Kumar et al. (2011) reported that an AF4-MLL-specific siRNA had no effect on the growth of the SEM t(4;11) leukemia cell line. However, as was pointed out in a rebuttal article (Marschalek, 2011), the specific AF4-MLL siRNA used was not likely to produce a knockdown of the AF4-MLL protein, something we have now confirmed in our results here ( Figures S5K-S5M). Unfortunately, our own attempt to design an AF4-MLL-specific siRNA was also unsuccessful ( Figures S5K-S5M), likely due to can interact with either coactivators or corepressors to cause gene activation or repression. In t(4;11) cells, RUNX1 can also interact with the AF4-MLL complex. (B) In t(4;11) leukemias, MLL-AF4 is expressed from one translocated chromosome, and the MLL-AF4 protein binds to and activates the RUNX1 gene by stabilizing AF9 and ENL binding. AF4-MLL is expressed from the other translocated chromosome, and the RUNX1 protein interacts with the AF4-MLL complex and binds to target genes. the stability and low turnover of the AF4-MLL protein (Marschalek, 2011), so the specific role of AF4-MLL remains to be definitively elucidated.
The data we present are consistent with an interplay between MLL-AF4 and AF4-MLL through the regulation and function of RUNX1, providing a model of how these oncoproteins could cooperate on a molecular level ( Figure 7B). Such a cooperative effect between these two fusion proteins could explain why this particular MLL translocation produces such an aggressive leukemia with relatively few additional mutations (Bardini et al., 2011;Bardini et al., 2010).

EXPERIMENTAL PROCEDURES
Chromatin Immunoprecipitation Assays ChIP (for both real-time PCR and ChIP-seq) experiments were performed as described in Milne et al. (2009), with several modifications, as outlined in Extended Experimental Procedures.

ChIP-Seq Analysis
The RS4;11 MLL-N (Akalin et al., 2012) and RS4;11 MLL-N, AF4-C, and H3K79Me2 (Geng et al., 2012) ChIP-seq data have also been used in separate studies analyzing DNA hypomethylation at target genes. The SEM MLL-N, AF4-C, and H3K79Me2 is from Guenther et al. (2008). Regions of overlap for MLL-N and AF4-C were defined as peaks overlapping in the promoter regions (± 2 kb to transcriptional start site [TSS]), and for H3K79Me2 as the gene body regions (À2 kb to TSS to +1 kb to transcriptional end SITE [TES]). Further details of analysis are included in Extended Experimental Procedures.

Patient Data
Gene expression microarray data from three large cohorts of patients with ALL were analyzed, including the ECOG Clinical Trial E2993, (Geng et al., 2012), the COG Clinical Trial P9906 (Harvey et al., 2010), and the St. Jude Research Hospital pediatric ALL clinical trial (Ross et al., 2003). Further details are provided in Extended Experimental Procedures.

Colony Forming Assays
Twenty-four hours after second transfection, cells were plated at a density of 1, 2, or 2.5 3 10 5 cells/ml, in triplicate, plated in IMDM MethoCult media (H4100; STEMCELL Technologies) supplemented with fetal calf serum and cultured for 14 days (37 C, 5% CO 2 ) before counting. Colony-forming assays were run in triplicate with at least three biological repeats.

Western Blotting
A total of 10 mg nuclear extract was loaded per lane on NuPAGE 4%-12% BisTris gels (Life Technologies) and blotted onto polyvinylidene fluoride membrane (Immobilon) at 100V for 1 hr using a Tris-glycine blotting buffer. Blots were probed with the antibodies indicated.

ACCESSION NUMBERS
The Gene Expression Omnibus accession number for the RUNX1 ChIP-seq data reported in this paper is GSE42075.

SUPPLEMENTAL INFORMATION
Supplemental Information includes Extended Experimental Procedures, five figures, and five tables and can be found with this article online at http://dx. doi.org/10.1016/j.celrep.2012.12.016.

LICENSING INFORMATION
This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Colony and Cell Growth Assays 24 hr post second transfection cells were plated at a density of 1, 2 or 2.5 3 10 5 cells per ml, in triplicate. Cells used for colony forming assays were transferred into IMDM (GIBCO) supplemented with 20% FCS and passaged twice before transfection and plating. 4 3 10 3 MV4-11 cells were plated in 30% FBS with 10 À4 M (0.1mM) 2-mercaptoetanol supplemented IMDM Methocult media for the colony assay. For the cell growth assay, manual viable cell counts were performed using 0.4% Trypan blue (GIBCO, Life technologies) and a Neubauer haemocytometer, at the times indicated.

Chromatin Immunoprecipitation Assays
ChIP (for both Real Time PCR and ChIP-seq) experiments were performed as described in (Milne et al., 2009) with the following modifications: H3K79Me2 and H3K4Me3 ChIP samples were fixed using a 1% formaldehyde (FA) fixation protocol for 10 min, while a 45 min, 2mM DSG and a 30 min 1% FA double fixation protocol was used for all other antibodies. Fixed chromatin samples were fragmented using a Bioruptor sonicator (Diagenode) for 30 min at high in a constantly circulating 4 C water bath to an average size of 200-500bps. AF4-N ChIP signal was improved by reducing the sonication time from 30 min to 20 min. Antibody:chromatin complexes were collected with a mixture of protein A and Protein G Dynabeads (Life Technologies) collected with a magnet, and washed 2X with a solution of 50mM HEPES-KOH, pH 7.6, 500mM LiCl, 1mM EDTA, 1% NP-40, 0.7% Na-Deoxycholate. After a TE wash, samples were eluted, RNase and Proteinase K treated, and purified using a QIAGEN PCR purification kit. ChIP samples were quantified relative to inputs as described in (Milne et al., 2009). Briefly, the amount of genomic DNA coprecipitated with antibody is calculated as a percentage of total input using the following formula DC T = C T (input) -C T (chromatin IP), % total = 2 DCT X 5.0%. A 50 ml aliquot taken from each of 1 ml of sonicated, diluted chromatin before antibody incubation serves as the input, thus the signal from the input samples represents 5% of the total chromatin used in each ChIP. CT values are determined by choosing threshold values in the linear range of each PCR reaction.
Primers for ChIP SYBR green primer sets were used for all ChIP figures. ChIP signal was calculated as a % of input as described above.

Gene Expression Analysis and Primers
In Figure 4A, RT-PCR signals were normalized to two different housekeeping genes (GAPDH and bActin) using the DCT method and then the highest expressing cell line was arbitrarily set to 100 and expression in all other lines was normalized to this value.
The following RUNX1, GAPDH and HOXA9 Taqman primer/probe sets were used for the gene expression data in Figure 3, Figure 4 and Figure 6B: RUNX1 20X Taqman primer/probe set from ABI, cat# Hs00231079_m1 RUNX1 GAPDH 20X Taqman primer/probe set from ABI, cat# Hs03929097_g1 GAPDH HOXA9 Forward primer: AAAACAATGCCGAGAATGAGAGCG, Reverse primer: TGGTGTTTTGTATAGGGGGACC, FAM-TAMRA probe: CCCCATCGATCCCAATAACCCAGC The following SYBR green primer sets were used for the gene expression data in Figure 3A, Figure 4A, Figure 5, Figure 6B and 6G and Figure S4 Gene expression was normalized to GAPDH (either Taqman or SYBR green) by the DCT method.
Genomic DNA-Fragment Library Genomic DNA fragment libraries were prepared using the Illumina ChIPseq Library preparation Kit following the manufacturer's instructions (Illumina, CA). Briefly 10ng of purified ChIP DNA was end repaired by conversion of overhangs into phosphorylated blunt ends with the use of T4 DNA polymerase and E. coli DNA polymerase I Klenow fragment. Illumina single-end adapters were ligated to the ends of the DNA fragments. Ligation products were purified on a 2% agarose gel with a size selection of 200-300bp. Fifteen PCR cycles were performed with Illumina genomic DNA primers that anneal to the ends of the adapters. The purified PCR-amplified fragment libraries were quantified with the use of the PicoGreen dsDNA Quantitation Assay with the Qubit Fluorometer (Life Technologies, CA). The size range of libraries was validated on the Agilent Technologies 2100 Bioanalyzer with the High Sensitivity DNA Kit (Agilent, CA).

ChIP Sequencing
After library preparation, the protocols for the Illumina Single-Read Cluster Generation Kit were used for cluster generation on the cBOT (Illumina). The targeted samples were diluted to 10 nmol and denatured with sodium hydroxide. Seven picomoles of each target-enriched sample and Phix control were loaded into separate lanes of the same flow cell, hybridized, and isothermally amplified. After linearization, blocking, and primer hybridization, sequencing was performed for 36 or 51 cycles on an Illumina GAIIx or HiSeq2000. Raw image data were converted into base calls via the Illumina pipeline CASAVA version 1.7 with default parameters. Rigorous quality control was performed with the use of data from reports generated by the Illumina pipeline.

ChIP-Seq Data Analysis
All 36 or 51 bp-long reads were mapped to the reference human genome sequence, hg18, using Illumina's ELAND or BWA (Li and Durbin, 2009) aligner with the default parameters. Only reads mapping uniquely to the genome with not more than 2 mismatches were retained for further analysis. Clonal reads (i.e., reads mapping at the same genomic position and on the same strand) were collapsed into a single read. Peaks from ChIPseq data were called using the ChIPseeqer program (Giannopoulou and Elemento, 2011) with the following parameters: -t 15 -f 2 -fraglen 170. The peaks were annotated to gene bodies, defined as 2kb upstream of the TSS to 1kb downstream of the TES, and to gene promoters defined as within 2kb upstream and 2kb downstream of TSS, based on hg18 refseq genes downloaded from the UCSC Genome Browser. Regions of overlap for MLL-N and AF4-C were defined as peaks overlapping in the promoter regions (± 2kb to TSS), and for H3K79Me2 as the gene body regions (À2kb to TSS to +1kb to TES).

Patient Data
Gene expression microarray data from three large cohorts of patients with ALL were analyzed.

Patient Gene Expression Microarray Data
The microarray raw data was normalized using the RMA method (Bolstad et al., 2003) with Expression Console TM software (Version 1.1, Affymetrix, Santa Clara, CA) for the Affymetrix arrays HG-U133 plus2 (COG data, n = 207) or HG- U133 A and B (St Jude data,n = 132), or NimbleScan software (version 2.5, Roche NimbleGen, Madison, WI) for the NimbleGen arrays HG18 60-mer expression 385K platform (ECOG data,n = 191). The patients in each clinical trial were grouped into subtypes according to their cytogenetic features: BCR-ABL, E2A-PBX1, MLLr (MLL rearrangement), ETV6-RUNX1, or other ALLs which are negative to the above translocations.
T-ALL samples were excluded from this analysis. MLL fusion partner information was available for the ECOG MLLr ALL, which were therefore further separated into MLL/AF4 (n = 17) or other MLLr (n = 8). No MLL fusion partner information was available for the COG or St Jude clinical trials, so MLLr ALL patients were treated as one group. Expression level of a gene in a sample was determined by the average of expression values from multiple probe sets on the array representing this gene. The p values of differential expression of RUNX1, HOXA9, HOXA10 and CDKN1B between MLLr and other ALL subtypes were determined by two-sided Wilcoxon test. The expression values are log2 transformed so the fold change of RUNX1 expression was calculated as 2^(MLLr or t(4;11) RUNX1 expression -other subtype expression). All downstream microarray analysis was performed using R version 2.14.0 (R Development Core Team. R: A Language and Environment for Statistical Computing. 2009; http://www.R-project.org).

Patient Outcome Data
In the COG P9906 ALL clinical trial (n = 207), the minimal residual disease (MRD) was assessed by flow cytometry at the end of induction therapy (day 29), as previously described (Borowitz et al., 2008), and cases were defined as MRD positive (MRD+) or MRD negative (MRD-) using a threshold of 0.01%. Among the 207 COG ALL patients, 191 patients had the MRD data available, and 17 of them were MLLr ALL. We compared RUNX1 expression in the MRD+ and MRD-patients for all 191 ALL and for the subset of 17 MLLr ALL. P values were calculated by two-sided Wilcoxon test using R (R Development Core Team, 2009).