Chromatin-prebound Crm1 recruits Nup98-HoxA9 fusion to induce aberrant expression of Hox cluster genes

The nucleoporin Nup98 is frequently rearranged to form leukemogenic Nup98-fusion proteins with various partners. However, their function remains largely elusive. Here, we show that Nup98-HoxA9, a fusion between Nup98 and the homeobox transcription factor HoxA9, forms nuclear aggregates that frequently associate with facultative heterochromatin. We demonstrate that stable expression of Nup98-HoxA9 in mouse embryonic stem cells selectively induces the expression of Hox cluster genes. Genome-wide binding site analysis revealed that Nup98-HoxA9 is preferentially targeted and accumulated at Hox cluster regions where the export factor Crm1 is originally prebound. In addition, leptomycin B, an inhibitor of Crm1, disassembled nuclear Nup98-HoxA9 dots, resulting in the loss of chromatin binding of Nup98-HoxA9 and Nup98-HoxA9-mediated activation of Hox genes. Collectively, our results indicate that highly selective targeting of Nup98-fusion proteins to Hox cluster regions via prebound Crm1 induces the formation of higher order chromatin structures that causes aberrant Hox gene regulation. DOI: http://dx.doi.org/10.7554/eLife.09540.001

In leukemia, Nup98 is frequently found in the form of Nup98-fusions, which consist of N-terminal half of Nup98 containing multiple phenylalanine-glycine (FG) repeats and C-terminus of various partner proteins (Gough et al., 2011). More than 30 different proteins with various physiological functions have been reported as Nup98 fusion partners (reviewed in (Gough et al., 2011)). However, the molecular mechanism of Nup98-fusion mediated leukemogenesis is still largely unknown.
With regard to its molecular function, Nup98-HoxA9 was shown to act as a transcriptional regulator (Kasper et al., 1999;Ghannam et al., 2004;Bei et al., 2005;Yassin et al., 2009). In addition, genome-wide gene expression analysis revealed that ectopic expression of Nup98-HoxA9 causes upregulation, rather than downregulation, of numerous genes (Ghannam et al., 2004;Takeda et al., 2006). Mechanistically, the FG repeat of Nup98 is known to associate with CREBbinding protein (CBP)/p300 (Kasper et al., 1999), a histone acetyltransferase that functions as a transcriptional co-activator, and histone deacetylase (HDAC) 1 (Bai et al., 2006). However, the interaction of Nup98 FG repeats with p300 by itself is not sufficient to promote self-renewal of hematopoietic stem cells (Yung et al., 2011). Thus, the exact function of Nup98-HoxA9 still remains unclear.
eLife digest The nucleus of a eukaryotic cell (which includes plant and animal cells) contains most of the cell's genetic material in the form of carefully packaged strands of DNA. Genes are stretches of DNA that contain the instructions needed to produce the proteins and RNA molecules that the cell needs to survive. These molecules move across the membrane that surrounds the nucleus through pores made of proteins. One of these pore-forming proteins is called Nup98. The gene that produces Nup98 is frequently mutated in leukemia, where part of it becomes fused to regions of other unrelated genes. The proteins made from these combined genes are known as "fusion proteins".
The Nup98-HoxA9 fusion protein has been well studied, and appears to cause leukemia by interfering with the process called ("cell differentiation") by which stem cells specialize to form different types of blood cells. During cell differentiation, cells change which sets of genes they activate to become specific types of cells. A family of genes called Hox genes (to which the gene for HoxA9 belongs) is critical in cell differentiation and thus must be fine-tuned. It is also known that the Hox genes form clusters, and its activation is partly controlled by how tightly the DNA is packaged.
Previous studies have shown that the Nup98-HoxA9 fusion protein takes on the form of small dots in the nucleus. Oka et al. have now tracked how these proteins are distributed inside of the nucleus, and examined which part of the DNA they bind to, in more detail. This revealed that the dots of Nup98-HoxA9 tend to associate with tightly packed DNA, especially on Hox cluster genes, and activate these genes.
Oka et al. further found that a protein called Crm1, which is well known as a nuclear export factor that carries molecules out of the nucleus through the pore, is already bound to the Hox cluster genes in the nucleus and recruits the Nup98-HoxA9 protein. This interaction may change how the Hox gene is packaged in the nucleus. A future challenge will be to reveal how the Nup98-HoxA9 fusion protein and Crm1 on Hox cluster genes control gene expression.
Here, we demonstrate that Nup98-HoxA9 is specifically recruited to the vicinity of Hox cluster genes via chromosomally bound Crm1, a nuclear export factor that was originally identified as a protein required for maintaining the chromosomal structure in yeast (Adachi and Yanagida, 1989), to activate Hox cluster regions.

Nup98-HoxA9 dots associate with facultative heterochromatin
Previous reports have demonstrated that Nup98-HoxA9 fusion protein localizes as distinct small dots in the nucleus (Kasper et al., 1999;Bai et al., 2006;Xu and Powers, 2010;Oka et al., 2010). To address the role of Nup98-HoxA9, we first characterized in detail the intranuclear distribution of these dots. When expressed in HeLa cells, the enhanced green fluorescent protein (EGFP)-tagged Nup98-HoxA9 was easily distinguished as small nuclear dots. In contrast, Nup98 FG repeats were detected as large dots mostly associated with the nucleolus, and HoxA9 C-terminus containing the homeodomain, HoxA9-Ct, showed an even nucleopolasmic distribution ( Figure 1A), consistent with earlier reports (Oka et al., 2010;Xu and Powers, 2010). We noticed that Nup98-HoxA9 dots were not randomly localized within the nucleus, but were frequently adjacent to 4',6-diamidino-2phenylindole (DAPI)-dense heterochromatin regions ( Figure 1B, Figure 1-figure supplement 1) as revealed by co-staining with DAPI. Furthermore, immunocytochemical analysis using monoclonal antibodies specific to histone modifications (Kimura et al., 2008) ( Figure 1C) revealed that those dots were frequently associated with H3K9me2 and H3K27me3, facultative heterochromatin markers (Trojer and Reinberg, 2007), but not with the marker for constitutive heterochromatin H3K9me3 (Peters et al., 2003, Rice et al., 2003. These results indicate that Nup98-HoxA9 dots are not randomly positioned in the nucleus but are associated with specific chromatin domains.

Nup98-HoxA9 dots show a cell-type specific localization pattern
We hypothesized that if Nup98-HoxA9 associates with specific chromatin modifications, its intranuclear distribution pattern may differ depending on the cell type. Therefore, we next expressed Nup98-HoxA9 in mouse embryonic stem (ES) cells, whose chromatin modifications are quite different from other cell types (Meshorer and Misteli, 2006). We found that although Nup98-HoxA9 nuclear dots could also be observed in ES cells, they were fewer in number and more heterogeneous in size compared with dots revealed in HeLa cells or NIH 3T3 cells ( Figure 2A). Immunocytochemical analysis in ES cells (Figure 2-figure supplement 1) also showed the partial co-localization of Nup98-HoxA9 dots with H3K9me2 and H3K27me3, but less with H3K9me3. These findings indicated that the intranuclear distribution of Nup98-HoxA9 indeed changes depending on the cell type, although we cannot exclude the possibility that the variation in the transgene expression level caused these differences. Together, these results suggested that Nup98-HoxA9 might be involved in a cell type-specific gene and/or chromatin regulation.

Nup98-HoxA9 expressing ES cells show flattened morphology and resistance to spontaneous differentiation
To gain more insights into the function of Nup98-HoxA9, which most likely associates with the impairment of cell differentiation (see Introduction), we tried to obtain stable cell lines expressing the FLAG-Nup98-HoxA9 fusion in ES cells using transposon vectors (Kawakami and Noda, 2004;Urasaki et al., 2006;Oka et al., 2013). Immunocytochemical analysis using stable ES clones demonstrated that FLAG-Nup98-HoxA9 had a similar localization pattern to transiently expressed EGFP-Nup98-HoxA9 in ES cells ( Figure 2B). Immunoblotting using monoclonal antibody that is raised against the N-terminal (A.A. 1-466) of Nup98 (Fukuhara et al., 2005) revealed that FLAG-Nup98-HoxA9 was not overexpressed compared with endogenous Nup98 (Figure 2-figure supplement 2). Unexpectedly, we found that FLAG-Nup98-HoxA9 expressing stable ES cell clones (Nup98-HoxA9 ES) had distinct flat colonies, which was in contrast to tighter colonies of spherical shape detected in parental ES, Nup98FG, or HoxA9-Ct expressing ES cells ( Figure 2C). Further experiments revealed that Nup98-HoxA9 ES cells expressed stem cell marker genes, and had proliferation rates comparable to those of parental ES cells, indicating that they were not differentiated (data not shown). In addition, Nup98-HoxA9 ES cells show more resistance to spontaneous cell differentiation in the absence of leukemia inhibitory factor (LIF), compared with parental ES, Nup98FG, or HoxA9-Ct expressing ES clones ( Figure 2D and    HeLa and NIH3T3 cells. Cells were transfected with the EGFP-Nup98-HoxA9 expressing plasmid for 24 hr, fixed, and stained with DAPI. Samples were analyzed using confocal microscopy. Bar, 10 mm. (B) Subcellular localization of Nup98-HoxA9 and its truncated mutants in ES cells. ES cell clones expressing FLAG-tagged Nup98FG, Nup98-HoxA9, or HoxA9-Ct were fixed and stained with an anti-FLAG (polyclonal) antibody. Nuclei were stained with DAPI. Merged images show FLAG staining (green) and DAPI (red). Bar, 10 mm. (C) Cell morphology of stable ES cell lines expressing FLAG-tagged Nup98-HoxA9, Nup98FG, or HoxA9-Ct. Bar, 100 mm. (D) Differentiation assay of stable ES cell clones. ES cells stably expressing FLAG (control), FLAGtagged Nup98FG, Nup98-HoxA9, HoxA9-Ct, HoxA9, Nup153-HoxA9, or Nup214-HoxA9 were plated at a density of 10 3 cells per well in 12-well plates either in the presence or absence of LIF. After 5 d, the plates were fixed and stained with alkaline phosphatase, a marker for undifferentiated stem cells. (E) Gene expression profiling of ES cell lines stably expressing FLAG-tagged Nup98-HoxA9 (#1 and #9; two independent clones) and HoxA9-Ct is compared with that of parental ES cells. A greater than 10-fold upregulation of gene expression was commonly observed in both Nup98-HoxA9 clone #1 and Nup98-HoxA9 clone #9 cells. (F) Hox gene expression profiling. The log2 fold ratios of the normalized signal value of Hox cluster genes from Nup98-HoxA9 or HoxA9-Ct expressing ES cells relative to signals from parental control ES cells are indicated. (G) Upregulation of Hox cluster genes in FLAG-Nup98-HoxA9 expressing ES cell lines was confirmed using semi-quantitative polymerase chain reaction. DAPI, 4',6-diamidino-2phenylindole; EGFP, enhanced green fluorescent protein; ES, embryonic stem; LIF, leukemia inhibitory factor DOI: 10.7554/eLife.09540.005 The following figure supplements are available for figure 2:

Nup98-HoxA9 expression in ES cells induces expression of Hox cluster genes
Next, to elucidate whether Nup98-HoxA9 affected global gene expression, we performed DNA microarray analysis using the following four clones: two independent Nup98-HoxA9 ES clones (clone #1 and clone #9), parental ES cells, and the HoxA9-Ct ES clone. We found that a small fraction of genes (less than 1%) were strongly induced (more than 10-fold) in Nup98-HoxA9 ES cells compared with their expression levels in parental or HoxA9-Ct ES cells ( Figure 2E). Interestingly, we noticed that Hox cluster genes were highly activated in Nup98-HoxA9 ES cells ( Figure 2F). In particular, more than a half of the top 32 upregulated genes, which were common in both of Nup98-HoxA9 clone #1 and Nup98-HoxA9 clone #9, were Hox cluster genes belonging to Hox-A, -B, and -C clusters (Table 1), which were validated by reverse-transcription polymerase chain reaction (RT-PCR) ( Figure 2G). These results suggest that there is a good correlation between the intranuclear distribution of Nup98-HoxA9 and its function in transcription, since Hox gene cluster regions are known to be organized as facultative heterochromatin (Trojer and Reinberg, 2007).

Influence of Nup98 FG repeats on the activation of Hox genes cannot be mimicked by FG repeat domains of other Nup proteins
To find out how Hox genes could be selectively induced by the Nup98-HoxA9 fusion in ES cells, we first focused on Nup98FG repeats, since all Nup98-fusion proteins, including Nup98-HoxA9, contain dense FG repeats of Nup98 at their N-termini. On the other hand, it is known that other nucleoporins containing such dense FG repeat domains are also implicated in oncogenic transformation (Kasper et al., 1999, Xu andPowers, 2009). Therefore, we created chimeric Nup-fusions by replacing Nup98 FG repeats with those from hNup153FG (A.A.1118-1475) and hNup214FG (A.A.1605-2090), which have been demonstrated to be functionally exchangeable in experiments that addressed the transforming ability in NIH3T3 cells (Kasper et al., 1999). When these chimeric proteins (Nup153-HoxA9 and Nup214-HoxA9) were expressed in HeLa cells, they were not distributed as distinct fine nuclear dots observed in the case of the Nup98-HoxA9 fusion (Figure 3-figure supplement 1), in agreement with a previous report (Xu and Powers, 2013).
We next isolated stable ES cell clones expressing these fusions. As shown in Figure 4-figure supplement 2, we found that, in contrast to the case with Nup98-HoxA9, other Nup-fusions were distributed more evenly in the nucleus (although note that a few dots within the nucleus were seen with the Nup214-HoxA9 fusion). Next, the effect of these Nup-fusions on the expression of Hox genes was monitored in established stable ES cell lines. As shown in Figure 3, the expression of Nup153-HoxA9 or Nup214-HoxA9 did not affect the expression of Hox cluster genes. Of note, we constantly observed substantial numbers of alkaline phosphatase-positive colonies of Nup153-HoxA9-expressing ES cells in the absence of LIF ( Figure 2D and  (Jacinto et al., 2015). We also established stable clones expressing full-length HoxA9. Gene expression analysis revealed that HoxA9 expression did not induce the activation of Hox cluster genes ( Figure 3). These results underscore the physiological significance of Nup98 FG repeats in the dysregulation of Hox cluster gene expression.

Nup98-HoxA9 dots are maintained through the association with Crm1
What makes Nup98FG repeats differ from Nup153FG or Nup214FG? We speculated that cohesive properties of Nup98FG (Patel et al., 2007;Xu and Powers, 2013;Yamada et al., 2010;Schmidt and Gorlich, 2015) required to form nuclear aggregates (Griffis et al., 2002;Oka et al., 2010;Patel et al., 2007) containing Crm1 (Takeda et al., 2010, Oka et al., 2010) may be important  for the pathogenic function. Indeed, the ectopic expression of Nup98-HoxA9 in HeLa cells caused sequestration of nucleoplasmic and nuclear envelope-localized Crm1 on Nup98-HoxA9 dots within the nucleus ( Figure 4A, top). When we stained Nup98-HoxA9 ES cells, we also noticed partial recruitment of Crm1 to FLAG-Nup98-HoxA9 dots ( Figure 4A, bottom). We have previously demonstrated that treatment with leptomycin B (LMB), a specific inhibitor of the cargo binding of Crm1 (Kudo et al., 1998), causes a dynamic redistribution of Nup98 nuclear dots (Oka et al., 2010). This finding indicated that Nup98-Crm1 association is dependent on the cargo binding status of Crm1; therefore, we examined whether such properties of Nup98FG were still preserved in Nup98-HoxA9. When Nup98-HoxA9 ES cells were treated with LMB for 2 hr, the dots disappeared ( Figure 4B). The disappearance of dots was concomitant with a modest decrease in the protein level (Figure 4-figure supplement 1). In contrast, LMB treatment did not affect the localization of full-length HoxA9, Nup153-HoxA9, and the majority of Nup214-HoxA9, except for a few nuclear dots of Nup214-HoxA9 that disappeared upon LMB treatment (Figure 4-figure supplement 2). These results indicate that Crm1 plays a critical role in the maintenance of Nup98-HoxA9 dots in ES cells.

Crm1 plays an important role in Nup98-HoxA9 functions
Next, to examine whether the association between Nup98-HoxA9 and Crm1 is required for the Nup98-HoxA9-mediated Hox gene activation, we treated Nup98-HoxA9 ES cells with LMB and monitored its effects on gene expression. As shown in Figure 4C, LMB treatment caused a significant downregulation of Hox cluster genes, while the expression of other genes, such as Oct4 and Nanog, was not much affected. Therefore, these results indicate that the interaction of Nup98-HoxA9 with Crm1 is necessary for the selective Hox gene activation.

Nup98-HoxA9 selectively binds to four Hox cluster regions
To further investigate how Nup98-HoxA9 is involved in the Hox cluster gene regulation, we performed chromatin immunoprecipitation sequencing (ChIP-seq) analysis using Nup98-HoxA9 ES cells. Interestingly, we found a strong accumulation of FLAG-Nup98-HoxA9 on chromosomal regions of four Hox clusters, Hox-A, B, C, and D. Especially, three of these, Hox-A, Hox-B and Hox-C cluster regions showed the highest peaks in the whole genome ( Figure 5A). In particular, the Hox-A cluster region showed a single dominant peak in chromosome 6 ( Figure 5C). Examination of individual Hox cluster regions further revealed a selective accumulation of FLAG-Nup98-HoxA9 in the four Hox cluster regions ( Figure 5D). It is noteworthy that, as most of the genes in the Hox-D cluster were not   upregulated by Nup98-HoxA9, the significance of its binding to this region remains unknown. ChIP-qPCR (quantitative polymerase chain reaction) analysis ( Figure 5E) confirmed that a significantly increased amount of FLAG-Nup98-HoxA9 was bound to the Hox-A cluster, but not to the restricted regions, as represented by the intergenic region (between HoxA2 and HoxA3), promoter (HoxA4), or 5'UTR (HoxA9) regions of Hox-A cluster genes.
Chromosomally prebound Crm1 recruits Nup98-HoxA9 to Hox cluster regions Next, we examined whether Crm1 could bind Hox cluster regions using ChIP-seq analysis. Unexpectedly, we found that in parental ES cells, the endogenous Crm1 bound to all Hox regions including Hox-A, B, C, and D ( Figure 5B-D), which were almost the same sites as those occupied by Nup98-HoxA9. Moreover, the binding of Crm1 to Hox cluster regions was significantly elevated in Nup98-HoxA9 ES cells ( Figure 5B-E). Furthermore, LMB treatment induced a drastic decrease of Nup98-HoxA9 binding to Hox clusters ( Figure 5B-E). These results indicate that Nup98-HoxA9 forms higher order nuclear structures containing Crm1 on Hox cluster regions to induce Hox gene activation.

Genome-wide characterization of Nup98-HoxA9 and Crm1 binding sites
Next, we characterized the chromatin binding sites of FLAG-Nup98-HoxA9 and their overlap with Crm1-binding sites on a genome-wide scale. Aggregation plot revealed that the peak of FLAG-Nup98-HoxA9 binding site was severely diminished but not shifted by LMB treatment ( Figure 6A, left top). Furthermore, FLAG-Nup98-HoxA9 was preferentially targeted to the chromatin sites that originally bound Crm1 (the Crm1 binding site in control ES cells) ( Figure 6A, right top). Aggregation plots of the Crm1 binding signal revealed that the peak of Crm1 binding was not shifted, but rather significantly enhanced by the expression of FLAG-Nup98-HoxA9 ( Figure 6A, left bottom). In addition, the Crm1 signal also accumulated at the FLAG-Nup98-HoxA9 binding site ( Figure 6A, right bottom). Statistical analyses confirmed these results ( Figure 6B). We also performed an analysis of the target genes that bound FLAG-Nup98-HoxA9 or Crm1 ( Figure 6C, Figure 6-figure supplement 1). As expected, the number of FLAG-Nup98-HoxA9 target genes was severely decreased by LMB treatment (Figure 6-figure supplement 1, top left). On the other hand, the number of Crm1 target genes were significantly increased by the expression of FLAG-Nup98-HoxA9; however, the majority of these genes were not overlapping ( Figure 6C, Figure 6-figure supplement 1, top right), examples of which were observed in the vicinity of Hox-cluster regions ( Figure 5D, see green arrows near Hox-A (Skap2) or Hox-C cluster). In these regions, it is likely that FLAG-Nup98-HoxA9 first binds to chromatin loci where Crm1 is not present, and then, Crm1 is recruited to FLAG-Nup98-HoxA9-bound sites, which results in the increase of Crm1-target genes on a whole genome level ( Figure 6C, Figure 6-figure supplement 1, compare bottom two panels). Therefore, these results demonstrate that both chromatin-bound Crm1 and FLAG-Nup98-HoxA9 can recruit each other.

Binding of Nup98-HoxA9 and Crm1 to chromatin is sensitive to TSA treatment
Our results so far suggest that Nup98-HoxA9 is targeted to Crm1 that associates with specific chromatin modifications. Therefore, we examined the effects of various epigenetic inhibitors on nuclear Nup98-HoxA9 dots. While most of these agents only showed a minor effect on the localization pattern of Nup98-HoxA9 ( Figure 7A), trichostatin A (TSA) treatment caused a drastic redistribution of FLAG-Nup98-HoxA9, a significant increase in the number and a significant decrease in the size of dots, which mimicked that observed in HeLa cells ( Figure 7B), concomitant with the redistribution of endogenous Crm1 (Figure 7-figure supplement 1). We also confirmed that TSA treatment caused  Figure 7. continued on next page which occurred presumably due to the reactivation of partially silenced transgenes cloned in the Tol2 transposon vector. A similar phenomenon was also observed in the case of transgenes created by the Sleeping Beauty transposon (Garrison et al., 2007). Next, we examined the effects of TSA treatment on the genome-wide chromatin binding of Crm1 or FLAG-Nup98-HoxA9. Unexpectedly, despite the increase in the FLAG-Nup98-HoxA9 protein level, TSA treatment caused a drastic decrease in the binding of Nup98-HoxA9 and Crm1 to chromatin ( Figure 6A,B), which was also obvious in Hox cluster regions ( Figure 5B-D). Furthermore, our ChIP-qPCR analysis showed that the binding of endogenous Crm1 to Hox-A cluster in control ES cells was also sensitive to TSA treatment (Figure 7-figure supplement 3).

Effect of TSA treatment on the chromatin structure of Hox cluster regions
It is known that TSA treatment causes differential effects on the epigenetic status of chromatin depending on the dosage, duration, cell-type, and target genes. Interestingly, it is known that treatment of ES cells with TSA at a concentration of 50 nM drastically affects bivalent histone modifications H3K4me3 (upregulation) and H3K27me3 (downregulation), in addition to the acetylation (Karantzali et al., 2008). Since bivalent histone marks are critical for the regulation of Hox cluster genes in ES cells , Pan et al., 2007, we speculate that TSA treatment may dramatically modulate the epigenetic status of Hox cluster regions. Indeed, as shown in Figure 7-figure supplement 4, TSA treatment caused a moderate upregulation of H3K4me3 and a robust decrease of H3K27me3 on the promoter region of Hox-A cluster genes, which is in line with a previous report (Karantzali et al., 2008). These results suggest that inactive histone marks, including H3K27me3, are important for the recruitment of Crm1 and FLAG-Nup98-HoxA9 to the Hox-A cluster region.

Nup98-HoxA9 binding causes epigenetic modifications and recruitment of RNA polymerase II onto Hox-A cluster
We then compared the epigenetic status of the Hox-A region in Nup98-HoxA9 and parental ES cells. We revealed that Nup98-HoxA9 ES cells exhibited significantly higher levels of H3K4me3, an active histone modification, in all tested Hox regions ( Figure 7C). On the other hand, enhanced levels of H3K27me3 were significant only on the HoxA9 5'UTR exon ( Figure 7C). Furthermore, ChIP-qPCR analysis of phospho-CTD isoforms of RNA polymerase II (RNAPII) revealed that the binding of both S5P (initiating) and S2P (elongating) forms of RNAPII to the Hox-A cluster regions increased in the Nup98-HoxA9 ES cells compared with parental ES cells ( Figure 7D). Collectively, our results suggest that Crm1 may associate with Hox genes via inactive histone marks, such as H3K27me3, and subsequently, Nup98-HoxA9 may form a local higher order chromatin structure on Crm1 binding sites to further recruit active histone modifier(s) and RNAPII to induce a selective upregulation of Hox genes.

Discussion
Our findings reveal that Nup98-HoxA9 is selectively recruited to Hox cluster regions via its interaction with Crm1 to induce gene expression in ES cells. Although activation of several Hox genes by Nup98-HoxA9 has been previously reported in other types of cells (Calvo et al., 2002;Takeda et al., 2006;Ghannam et al., 2004), those effects were relatively weak and not highly specific, which is in contrast to the pronounced changes observed in our present study. We believe that such variable outcomes could stem from two possible reasons. First, the chromatin structure, epigenetic status or chromatin-bound protein complexes of Hox cluster regions may be dissimilar in different cell types, causing varying susceptibility to Nup98-HoxA9-mediated chromatin modifications. Second, the association status of Crm1 with Hox cluster gene loci, which is a prerequisite for the selective deregulation of Hox cluster genes, may be differentially modulated in different cell types, causing cell type-specific variability in the Nup98-HoxA9-mediated gene expression.
Notably, although both FG repeats of Nup153 and Nup214 are known to interact with Crm1 (Roloff et al., 2013;Nakielny et al., 1999;Fornerod et al., 1997), we found that these fusions, Nup153-HoxA9 and Nup214-HoxA9, failed to activate Hox cluster genes. Thus, our results suggest that the ability of Nup98 FG repeats to form nuclear dot-like structures is critical in Hox gene activation. We propose a model whereby the interaction of Nup98-HoxA9 with prebound Crm1 induces the alteration of chromatin structures specifically on Hox cluster regions and this modification, in turn, causes robust activation of gene expression.
It is known that Hox clusters in ES cells harbor a distinctive long-range histone modification signature called the bivalent motif, which possesses both transcriptionally active H3K4me3 and repressive H3K27me3 marks (Bernstein et al., 2006;Mikkelsen et al., 2007). This modification is maintained through the activity of polycomb complexes (Schwartz and Pirrotta, 2007). Our results demonstrate that Nup98-HoxA9/Crm1 binding to Hox region is sensitive to TSA treatment, which results in a robust decrease of the H3K27me3 mark on Hox cluster genes. Moreover, the expression of Nup98-HoxA9 causes a significant increase of the active histone mark H3K4me3 on Hox cluster region genes. Previously, it was reported that the FG repeat region of Nup98 interacts with histone acetyltransferases CBP and p300 (Kasper et al., 1999). In addition, MBD-R2/NSL and Trx histone modifying complexes have been demonstrated to interact with full-length Nup98 in fly (Pascual-Garcia et al., 2014). Thus, we speculate that Nup98-HoxA9 first binds to facultative heterochromatin and then recruits histone and/or chromatin modifying enzymes to induce active histone modifications and a subsequent activation of Hox cluster genes. Importantly, in CD133+ hematopoietic stem cells, histone modifications of Hox-A and Hox-B loci are also bivalent (Cui et al., 2009), suggesting that Hox cluster genes are similarly regulated both in ES and hematopoietic progenitor cells.
Selective activation of Hox genes has also been reported in other Nup98-fusions, Nup98-NSD1 and Nup98-Jarid1a, in myeloid progenitor cells (Wang et al., 2007, Wang et al., 2009. Mechanistically, Nup98-NSD1 binds to the promoter region of Hox-A cluster genes and modifies histones via its H3K36 methyltransferase activity (Wang et al., 2007). Similarly, Nup98-Jarid1a binds to di-or trimethylated H3K4 of the Hox-A cluster region through its plant homeodomain (PHD) finger domain and causes deregulation of Hox genes activation (Wang et al., 2009). However, it remained unclear how the expression of these Nup98-fusions actually causes selective activation of Hox genes. Our results suggest that chromosome prebound Crm1 may recruit Nup98-NSD1, Nup98-Jarid1a, and possibly other Nup98-fusions to specific Hox cluster genes loci through its interaction with the Nup98FG repeat, which commonly exists in all Nup98-fusions. Unexpectedly, our data showed that Nup98FG could only weakly bind to Hox-A cluster region ( Figure 5-figure supplement 1). However, since Nup98FG dots are frequently localized within the nucleolus (data not shown), and Nup98FG is more prone to form nuclear aggregates than Nup98-HoxA9 with only a weak nucleoplasmic diffuse staining (see Figure 2B), we speculate that Nup98FG is hardly accessible to the genomic DNA, including the Hox cluster region, when expressed by itself.
Notably, a recent report (Conway et al., 2014) also demonstrated that Crm1 binds to the HoxA9 and HoxA10 gene regions, both in human leukemia cell line and immortalized mouse embryonic fibroblast cell line. Conway et al. (2014) further showed that Crm1 recruits CALM-AF10 fusion through its interaction with the nuclear export signal on CALM to activate Hox gene expression. Thus, the association of Crm1 with Hox loci could be a common molecular basis for aberrant Hox gene dysregulation mediated by numerous leukemic fusions.

Transfection and generation of stable clones
For transient transfection assays, cells were plated into 35-mm dishes and transfected with plasmid DNA using lipofectamine 2000 (Life Technologies). Generation of stable ES cell lines expressing a transgene was performed as described previously (Oka et al., 2013). Briefly, ES cells were co-transfected with pCAGGS-m2TP, which is an expression vector for Tol2 transposase, and the Tol2 transposon-based pT2A-CMH expression vector containing various inserted cDNA sequences. After 2 d, cells were re-plated into the ES-LIF medium containing hygromycin B (200 mg/ml). The medium was changed every other day until the appearance of colonies. Colonies were picked and the expression product of the transgenes was confirmed by western blotting and immunofluorescence staining.

Immunofluorescence staining and confocal microscopy
Cells were grown on coverslips and fixed with 3.7% formaldehyde in phosphate-buffered saline (PBS) for 15 min at room temperature. After treatment with 0.5% Triton X-100 in PBS for 5 min, the cells were incubated in blocking buffer (3% skim milk in PBS) for 30 min and again incubated with primary antibodies overnight at 4˚C. After washing with PBS, cells were incubated with Alexa Flour 488-and/or 594-conjugated secondary antibodies (Life Technologies) for 30 min. The cells were then stained with DAPI and the coverslips were mounted with Vectashield (Vector Laboratories, Burlingame, CA). Images were acquired using FV1000 (Olympus) or SP8 (Leica) confocal microscopes.

Microarray analysis
Microarray analysis was performed using total RNA prepared from indicated ES cell clones using the TRIzol reagent (Invitrogen Carlsbad, CA) as follows: cyanine-3 (Cy3) labeled complementary RNA (cRNA) was prepared from 0.1 mg total RNA using the Low Input Quick Amp Labeling Kit (Agilent Technologies, Santa Clara, CA) according to the manufacturer's instructions, followed by RNAeasy column purification (Qiagen, Valencia, CA). Cy3-labeled cRNA (0.6 mg) was fragmented and hybridized to SurePrint G3 Mouse GE 8Â60K Microarray (Agilent Technologies) for 17 hr at 65˚C in a rotating Agilent hybridization oven. After hybridization, microarrays were washed for 1 min at room temperature with GE Wash Buffer 1 (Agilent Technologies), then for 1 min with GE Wash buffer 2 (Agilent Technologies) at 37˚C, and dried immediately by brief centrifugation. Slides were scanned immediately on the Agilent DNA Microarray Scanner (G2505C) using one color scan setting for 8Â 60k array slides (scan area 61 Â 21.6 mm, scan resolution 3 mm, dye channel was set to Green and Green PMT was set to 100%). The scanned images were analyzed with the Feature Extraction Software 10.10.1.1 (Agilent Technologies) using default parameters to obtain background subtracted and spatially detrended processed signal intensities.

RT-PCR and qPCR
Total RNA was extracted from ES cells using the TRIzol reagent (Invitrogen) and used for cDNA synthesis with the Transcriptor First Strand cDNA Synthesis kit (Roche Applied Science, Mannheim, Germany) or PrimeScript 1st strand cDNA Synthesis Kit (Takara Bio, Otsu, Japan). All procedures were conducted according to the manufacturer's recommendations. RT-PCR was performed using KOD plus DNA polymerase (Toyobo, Osaka, Japan). qPCR analysis was performed on a 384-well plate with QuantStudio 6 Flex Real-Time PCR System (Life Technologies) using Power SYBRÒ Green PCR Master Mix (Applied Biosystems, Foster City, CA). As for mRNA expression levels, the relative expression levels were normalized using GAPDH mRNA levels as control. The primer sequences are listed in Supplementary file 1.

ChIP analysis
Cells were fixed in a medium containing 0.5% formaldehyde at room temperature for 5 min. After washing with ice-cold PBS twice, cells were resuspended in the ChIP buffer (10 mM Tris-HCl [pH 8.0], 200 mM KCl, 1 mM CaCl 2 , 0.5% NP40) containing protease inhibitors (aprotinin, leupeptin, pepstatin A at a concentration of 1 mg/ml each) and briefly sonicated (Branson 250D sonifier, Branson Ultrasonics, Danbury, CT). After centrifugation, the supernatants containing chromatin were digested with 3 units/ml micrococcal nuclease (Worthington Biochemical, Lakewood, NJ) for 40 min at 37˚C, and the reaction was stopped with ethylenediaminetetraacetic acid (EDTA; final concentration of 10 mM). Enzyme-treated supernatants were incubated with anti-mouse or anti-rabbit IgG magnetic beads (Dynabeads, Life Technologies) preincubated with anti-FLAG (M2), anti-Crm1, or anti-histone modification antibodies (2 mg) for 6 hr. (For Pol II ChIP, we first incubated the beads with rabbit anti-Rat IgG antibody [Jackson Immuno Research, West Grove, PA] before its incubation with phospho-specific Pol II antibodies.) The beads were washed thoroughly twice with each of the following three buffers, ChIP buffer, ChIP wash buffer (10 mM Tris-HCl [pH 8.0], 500 mM KCl, 1 mM CaCl 2 , 0.5% NP40), and TE buffer (10 mM Tris-HCl [pH 8.0], 1 mM EDTA), and eluted in the elution buffer containing 50 mM Tris-HCl (pH 8.0), 10 mM EDTA, and 1% sodium dodecyl sulfate overnight at 65˚C. DNA was recovered using the DNA gel extraction kit (Promega, Madison, WI) and used for ChIP-qPCR analysis or ChIP-seq analysis.

ChIP sequencing
The ChIP library was prepared according to the Illumina protocol and sequenced on the Illumina HiSeq1500 system. The sequence reads were uniquely mapped to the reference mouse genome (mm9) using Bowtie 2 software (version 2.2.2) with default parameter (Langmead and Salzberg, 2012). PCR duplicates were removed from mapped reads using SAMtools (version 0.1.19).

ChIP-seq data analysis
The mapped reads of ChIP and input DNA control data were counted in non-overlapping 200 base windows on the genome and the counts were normalized as RPM (read per million). We then calculated normalized ChIP-Seq signal intensities on each window as RPM difference between ChIP and input DNA control data (RPM ChIP -RPM input ). We used MACS (version 2.0.10) with the default parameters to detect ChIP-Seq signal enriched regions.

Accession numbers
The microarray data have been deposited in the Gene Expression Omnibus (GEO) database under the series entry GSE67967. The ChIP-Seq data are accessible through GEO Series accession number GSE68783.