Functional Features of EVI1 and EVI1Δ324 Isoforms of MECOM Gene in Genome-Wide Transcription Regulation and Oncogenicity

ORIGINAL ARTICLE Functional features of EVI1 and EVI1Δ324 isoforms of MECOM gene in genome-wide transcription regulation and oncogenicity

A Sayadi1, J Jeyakani2, SH Seet1, C-L Wei2, G Bourque2, FA Bard1, NA Jenkins1,3, NG Copeland1,3 and EA Bard-Chapeau1,4

The MDS1 and ecotropic viral integration site 1 (EVI1) complex locus (MECOM) gene encodes several transcription factor variants including MDS1-EVI1, EVI1 and EVI1Δ324. Although MDS1-EVI1 has been associated with tumor-suppressing activity, EVI1 is a known oncogene in various cancers, whose expression is associated with poor patient survival. Although EVI1Δ324 is co-transcribed with EVI1, its activity in cancer cells is not fully understood. Previous reports described that unlike EVI1, EVI1Δ324 protein cannot transform fibroblasts because of its disrupted N-terminal zinc finger (ZNF) domain. To better understand EVI1Δ324 biology and function, we obtained genome-wide binding occupancies and expression data in ovarian cancer cells. We characterized its DNA- binding sites, binding motif and target genes. Comparative analyses with previous study show that EVI1 and EVI1Δ324 share similar transcriptional activities linked to their common C-terminus ZNF domain. They bind to an E-twenty-six family (ETS)-like motif, target to a large extent the same genes and cooperate with AP1 transcription factor. EVI1Δ324-occupied genes were 70.7% similar to EVI1-bound genes. More strikingly, EVI1 and EVI1Δ324 differentially expressed genes were 99.87% identical, indicating comparable transcriptional regulatory functions. Consistently with gene ontologies linked to these target genes, EVI1Δ324 expression in HeLa cells could enhance anchorage-independent growth, such as EVI1, showing that EVI1Δ324 expression also lead to pro-oncogenic effects. The main specific feature of EVI1 variant is its N-terminus ZNF domain that binds DNA through GATA-like motif. We found that most GATA-like EVI1 chromatin immunoprecipitation sequencing peaks are far from genes and are not involved in transcriptional regulation. These genomic regions were enriched in simple sequence repeats and displayed high meiotic recombination rates. Overall, our genomics analyses uncovered common and specific features of two major MECOM isoforms. Their influence on transcription and downstream cell proliferation was comparable. However, EVI1-specific GATA-like binding sites, from its N-terminus ZNF domain, associated with high recombination rates, suggesting possible additional oncogenic potential for EVI1 in modulating genomic stability.

Oncogene (2016) 35, 2311–2321; doi:10.1038/onc.2015.286; published online 3 August 2015

INTRODUCTION the MECOM locus.4,17 The MDS1-EVI1 protein contains an 4,17 The MDS1 an ecotropic viral integration site 1 (EVI1) complex locus extended N-terminus and in contrast to the major EVI1 (MECOM) encodes several alternative transcripts. All but one isoforms has shown tumor-suppressive properties in some – – (MDS1 variant) of these transcripts are isoforms of EVI1, an experimental systems.1 4,17 20 oncogenic transcription factor that is known for its involvement in Although EVI1 overexpression is sufficient to transform fibro- both myeloid disorders1–4 and epithelial tumors, including ovarian blasts, this activity was found missing in the EVI1Δ324 isoform, 15,21 carcinoma.5–10 Recurrent chromosomal rearrangements of the probably due the absence of an intact N-terminal ZNF domain. – MECOM locus at 3q26 have been observed in patients, whereas However, EVI1Δ324 is often found co-expressed with EVI1.22 25 expression of EVI1 is associated with a poor prognosis in various To better understand whether EVI1Δ324 nevertheless display cancer types.9,11,12 The main isoform of EVI1 (hereafter simply oncogenic properties, we characterized here the transcriptional referred to as EVI1) encodes a 1051-amino acid transcription factor properties of EVI1Δ324 in carcinoma cell lines: its DNA-binding with two DNA-binding zinc finger (ZNF) arrays: an N-terminal features, binding sites and target genes. domain with seven ZNFs binding to a GATA-like DNA motif and a In a previous study, we have used chromatin immunoprecipita- domain located more toward the C-terminus with three ZNFs tion sequencing (ChIP-seq) to identify and characterize genomic binding an E-twenty-six family (ETS)-like motif. A major splice binding sites of EVI1 and identify oncogenic transcriptional variant named EVI1Δ324 or EVI1s13–16 lacks ZNFs 6 and 7, which programs.26 To understand the specificities of each variant, compromises the function of the GATA-like binding domain.14–16 we compared both ChIP-seq data sets and studied induced Several other variants have been described, most notably a read- gene expression changes on overexpression of either isoform. through transcript from the upstream MDS1 gene that is part of Comparative analyses of gene expression as well as genome-wide

1Institute of Molecular and Cell Biology, Singapore, Singapore and 2Genome Institute of Singapore, Singapore, Singapore. Correspondence: Dr EA Bard-Chapeau, Novartis Institutes for Biomedical Research, WKL-125.1.11, Basel CH-4002, Switzerland. E-mail: [email protected] 3Current address: The Methodist Hospital Research Institute, Houston, TX, USA. 4Current address: Novartis Institutes for Biomedical Researtch, WKL-125.1.05A, Basel CH-4002, Switzerland. Received 15 December 2014; revised 9 June 2015; accepted 13 June 2015; published online 3 August 2015 Genomic features of EVI1 and EVI1Δ324 isoforms A Sayadi et al 2312 binding sites permitted us to dissect the contributions of each of Genome-wide identification of EVI1Δ324-binding sites and the two EVI1 DNA-binding domains to oncogenic transformation. genomic occupancy patterns We observed strongly overlapping effects on gene expression We then examined the location of these EVI1Δ324 ChIP-seq peaks linked to the common C-terminal DNA-binding ZNF domains relative to annotated gene structures using the UCSC Genome present in both the investigated MECOM isoforms. However, given Browser Database.29 We obtained a marked enrichment for peaks the fact that EVI1Δ324 lacks transformant capacity and our in the proximity of transcription start sites (TSS) (Figure 1a), observation that EVI1 has strikingly few unique target genes, we consistent with a transcription factor pattern. A large proportion of consider a purely transcription-based model for EVI1-driven peaks (16.5%) fell within the promoters, whereas only 4% were oncogenesis unlikely. Instead, our data suggest that the N-term- located within 5 kb of the 3′-end of genes (Figure 1b). Of the inal ZNF domain of EVI1 is binding to sites of increased 43.5% of peaks that were located at intragenic sites, both intronic recombination and thereby might cause a state of genomic and exonic peaks preferentially occupied the first intron (328 instability. peaks, 47.1% of intronic peaks) or first exon (108 peaks, 62.1% of exonic peaks). A total of 1071 EVI1Δ324-bound genes were fi Δ RESULTS identi ed from 2003 peaks in which the EVI1 324 enrichment peak fell within a region of − 10 kb from the TSS and +5 kb from Δ ChIP-seq of Flag-tagged EVI1 324 in SKOV3 ovarian carcinoma the transcription end site. cells The human MECOM gene encodes multiple expressed alternative EVI1Δ324 binds to a DNA motif similar to EVI1 ETS-like motif splice forms owing to alternative usage of 5′-ends,27 alternative 16 Various experimental analyses have confirmed the oncogenic splicing and intergenic splicing, which results in the formation of 1,4,17 a fusion protein with MDS1.28 The three major isoforms are EVI1, a activity of EVI1. This activity is believed to be mediated by fi well-known oncogene, the EVI1Δ324 truncated isoform and speci c DNA binding of EVI1 through either of its two ZNF MDS1-EVI1 that displays tumor-suppressive functions.1,2,4,17 EVI1 domains, which are able to recognize one of the two previously fi 26 Δ and EVI1Δ324 share the same promoter and are often found de ned EVI1 recognition motifs. We speculated that EVI1 324, fi co-expressed both in human22,25 and mouse23,24 tissues. We which is de cient in the N-terminal ZNF domain, is regulating further confirmed their co-expression by immunoblotting experi- transcription solely by means of its C-terminal ZNF domain that fi fi ments in several cancer cell lines and in E9.5 and E17.5 mouse recognize binding sites with a speci c ETS-like motif. To con rm this hypothesis, we performed a de novo motif search using embryos (Supplementary Figure S1). As ChIP-seq using currently 30 available human EVI1 antibodies would provide a mix of data Weeder for EVI1Δ324 ChIP-seq data. We assessed the quality of representing all of the different isoforms, we decided to express a this novel motif by two complementary tests. The first test Flag-tagged version of the EVI1Δ324 isoform to characterize its consisted in mapping the motif matrix to the 2003 EVI1Δ324 specific target genes and possible involvement in carcinogenesis. genomic peaks obtained by ChIP-seq, to assess its frequency of To transiently express Flag-EVI1Δ324 protein, we used the human occurrence. The second one consisted in testing the power of the ovarian carcinoma cell line SKOV3 known to be causally associated motif matrix, to discriminate between true and false positives. with EVI1, and expressing both EVI1 and EVI1Δ324.26 We could then Several rounds of motif search/quality control were necessary to use a Flag antibody for ChIP analysis (Supplementary Figure S2a) obtain the best motif possible (Supplementary Figure S3a). As 31 and in doing so identify all of the genomic regions that are expected, our search resulted in an ETS-like ‘CCATCTCC’ motif − 10 associated with this particular isoform of EVI1. Subsequent (Figure 2a), identified with a P-value of 3.03 × 10 . 32 sequencing of immunoprecipitated DNA identified 9 995 340 unique The motif-mapping step, performed with Cisgenome , identi- tags that could be mapped to the human genome. The input DNA fied 34% of the 2003 EVI1Δ324 peaks carrying the motif at ± 50 bp, of this ChIP-seq experiment was also sequenced and 11 133 682 against 26% for match control regions, which constituted a unique tags were found and used as a background genomic control. significant enrichment (Figure 2b). This was consistent with Analysis of the ChIP-seq data identified a total of 3263 matrices mapping to ChIP-seq data sets commonly returning a enrichment peaks of overlap count 8 or more. A fold change combination of true- and false-positive results. In the second step, between the ChIP DNA and the input DNA was then calculated as we assessed the discriminative power of the matrices using the described previously.29 Experimental validation studies employing matrix-scan33 and matrix-quality34 software tools implemented in ChIP–quantitative PCR were performed on 111 randomly selected RSAT. EVI1Δ324-binding sequences (empirical) as well as control peaks with various fold changes (Supplementary Figure S2b). Fold sequences (theoretical) were used to search for matrix occurrence. change above 2.5 was selected as threshold of significance for To verify the specificity of the results, the program also used which the estimated false discovery rate was 6.3% (Supplementary column-permuted matrices as a negative control. The theoretical Figure S2c). In total, 2003 Chip-seq peaks of high significance were weight score (Ws) distribution and the empirical Ws distribution found above this cutoff (Supplementary Table S1). were plotted to assess the capacity of the matrix, to predict

Figure 1. Identiﬁcation of EVIΔ324 genome-wide binding sites. (a) Genomic distribution of all 2003 EVI1Δ324 ChIP-seq peaks compared with genes: in promoters (−10 kb from TSS), intragenic sites, gene ends (+5 kb from transcription end site (TES)) or distal sites (outside intra- and peri-genic regions). (b) Frequency of EVI1Δ324 enrichment peaks according to their position relative to the TSS.

Figure 2. Characterization and validation of EVI1Δ324 motif. (a) Sequence logo best representing the EVI1Δ324-binding motif found by de-novo motif search. This motif is similar to an ETS-like motif. (b) EVI1Δ324 motif mapping to Flag-EVI1 and Flag-EVI1Δ324 ChIP-seq data and matched control regions. (c) Distributions of the distance between the ChIP-seq peaks and the genomic location of the EVI1Δ324 motif. (d) Pie charts representing the genomic distribution of EVI1 and EVI1Δ324 enrichment peaks that contain the EVI1Δ324 motif: in promoters (−10 kb from TSS), intragenic sites (exon1, other exons, intron 1 and other introns), gene ends (+5 kb from transcription end site (TES)) or distal sites (outside intra- and peri-genic regions).

EVI1Δ324-binding sites. The empirical distribution reflects the EVI1Δ324 shares similar potential target genes with EVI1 measurement of the enrichment of matrix occurrence at ChIP-seq EVI1 and EVI1Δ324 isoforms share the same C-terminal part that peaks, whereas the theoretical distribution reflects the false includes a DNA-binding ZNF domain.1,4,17 We showed that both prediction rate at each possible Ws. The comparison of the EVI1 and EVI1Δ324 bind to the same ETS-like motif (Figure 2b and empirical and theoretical distributions showed that the matrix Supplementary Figure S3). To understand whether EVI1 and fi Δ could recover a signi cant proportion of EVI1 324-binding sites EVI1Δ324 transcription factors also share the same binding sites (Supplementary Figure S3b). We ran similar enrichment analyses, and target genes, we undertook to compare their occupied this time using an EVI1 ChIP-seq data set, performed under similar 26 genomic regions. When we calculated the distance between the conditions after expression of Flag-EVI1 in SKOV3 cells. Identical Δ Δ centers of EVI1 324 and EVI1 enrichment peaks, we noticed a motif enrichments as for EVI1 324 ChIP-seq were found (Figure 2b marked enrichment of proximate binding sites (Figure 3a). Indeed, and Supplementary Figure S3b). Of note, at high Ws, the false 32% of EVI1Δ324 peaks were located within 100 bp of an EVI1 discovery rate was slightly lower for EVI1 than for EVI1Δ324 ChIP- peak (Figure 3b), showing they bind in the same genomic area. seq data, likely because the total number of significant ChIP-seq When we evaluated the distribution of the distance between peaks is less for EVI1Δ324 (2003) than for EVI1 (12 618). EVI1Δ324 and EVI1 ETS-like motifs mapped to ChIP-seq enrich- Distribution calculations demonstrated high enrichment of Δ occurrence for the ETS-like motif near the 12 618 EVI1 ChIP-seq ment peaks, we observed a perfect overlap for 18% of EVI1 324 Δ peaks (Figure 3c). This indicates that the presence of our peaks and the 2003 EVI1 324 peaks (Figure 2c). Interestingly, the ‘ ’ relative frequencies within EVI1 and EVI1Δ324 ChIP-seq peaks CCATCTCC motif enhances the binding capacity of both EVI1 Δ were comparable, indicating that the motif occurred at EVI1- and and EVI1 324. EVI1Δ324-binding sites in similar proportions. Moreover, the intra- When we considered the corresponding bound genes and intergenic distributions of EVI1 and EVI1Δ324 enrichment rather than peaks, we observed a much higher overlap Δ peaks carrying the motif were nearly identical (Figure 2d), showing between the EVI1 324 and EVI1 data sets, with 70.7% of Δ that this motif is rather associated with peaks near TSS. EVI1 324 target genes also found as EVI1 target gene Collectively, our results indicate that the EVI1Δ324 DNA-binding (Figures 3d and e). A gene was considered bound by an EVI1 motif we identified is representative of not only EVI1Δ324 DNA- isoform when it located in the promoter (−10 kb to the TSS), binding sites but also EVI1-occupied sites. This novel ETS-like motif within the gene or at the transcription end site (+5 kb from gene is more conserved and more specific than the previous ‘TCCTCT’ end). Twenty-two examples of MECOM target genes are shown motif identified26 and shows a good occurrence and enrichment in Supplementary Figures S4a and b. For instance, EVI1 and at both EVI1 (40%) and EVI1Δ324 (34%) ChIP-seq peaks EVI1Δ324 bind the same region of FOSL2 promoter, 123 bp (Figure 2b). It also enriches in EVI1Δ324 enrichment peaks near apart. This result indicates that although the exact binding site TSS, in promoter, exon 1 and intron 1 (Figures 2c and d), indicating may differ, EVI1 and EVI1Δ324 variants, to a very large extent, its functional importance in transcriptional regulation. target the same genes.

Figure 3. EVI1Δ324 and EVI1 isoform-binding sites locate in proximity. (a) Distribution of the distances between EVI1Δ324 and EVI1 ChIP-seq peak centers. (b) Overlap between EVI1Δ324 peaks and EVI1 peaks. We considered the overlap true when the distance between the centers of EVI1Δ324 and EVI1 peaks was o100 bp. (c) Distribution of the distances between the ETS-like motifs mapped at EVI1Δ324 and EVI1 ChIP-seq peaks. (d) Overlap between EVI1Δ324 and EVI1 target genes. A target gene is when an enrichment peak occupies the promoter (−10 kb to the TSS), intragenic region or gene end (+5 kb from transcription end site (TES)). The purple insert corresponds to EVI1Δ324 and EVI1 target genes whose peaks are o100 bp apart. (e) Distribution of the distance between EVI1Δ324 and EVI1 peaks that are in same target genes.

Figure 4. Nature of the 757 EVI1- and EVI1Δ324-bound genes. Gene ontology analysis for biological processes (DAVID Bioinformatics: GOTERM) of EVI1Δ324 target genes that are in common with EVI1-bound genes.

Nature of genes occupied by EVI1 and EVI1Δ324 Bioinformatics (https://david.ncifcrf.gov). Interestingly, most As 70.7% of EVI1Δ324 target genes are common with EVI1-bound enriched ontologies regulate several properties frequently genes, we expect that these genes represent the most reliable associated with oncogenesis. Three highly overrepresented genes occupied by EVI1 isoforms through their ETS-like ZNF categories found among the EVI1/EVI1Δ324-bound genes were domain. We classiﬁed them by their gene ontologies using DAVID regulators of gene expression, signal transduction and organ

Figure 5. EVI1 and EVI1Δ324 isoforms trigger similar transcriptional and cellular disruptions in HeLa cells. (a) Intersection of the genes deregulated by Flag-EVI1 and Flag-EVI1Δ324. Microarray experiments were performed from HeLa cells transfected by Flag, Flag-EVI1 and Flag- EVI1Δ324. Analyses of variance (ANOVA) comparing Flag-EVI1 samples versus Flag and Flag-EVI1Δ324 samples versus Flag identified the genes significantly mis-expressed on Flag-EVI1 or Flag-EVI1Δ324 expression. (b) Correlation between EVI1Δ324 and EVI1 fold changes of all microarray probes. (c and d) Cell proliferation (c) and colony-formation (d) assays on HeLa cells transfected with Flag, Flag-EVI1Δ324 and Flag- EVI1 (*Po0.001 versus Flag control, one-way ANOVA, n = 7–9, error bars represent the s.d.). development (Figure 4), consistent with the known function of respectively (false discovery rate = 5%, Po0.0001 and absolute EVI1 as a critical regulator of embryonic development.35,36 We also value of fold change 41.25). Interestingly, there was an overlap found remarkable enrichment for genes controlling processes of 823 genes between these two sets (Figure 5a). Thus, the non- previously associated with EVI1, including apoptosis, cell morpho- overlapping genes constituted only 26.4% and 23.2% of all genesis, angiogenesis and hematopoiesis.9,17,36 More surprisingly, genes regulated by EVI1 and EVI1Δ324, respectively. Moreover, new potential biological functions were associated with EVI1/ these subsets were enriched in genes associated with lower EVI1Δ324-occupied genes, including neurogenesis, response to absolute fold change values (Supplementary Figure S5c), endoplasmic reticulum stress and response to unfolded protein, indicating they may represent variation due to proximity with suggesting a joint action or redundancy of these two isoforms to the threshold limit. Indeed, the correlation coefficient between regulate these processes. EVI1Δ324 and EVI1 fold changes of all microarray probes was 0.9915 (Figure 5b), indicating an almost perfect correlation. To identify the genes differentially regulated when Flag-EVI1 or Gain of expression of EVI1 or EVI1Δ324 triggers identical gene Flag-EVI1Δ324 are expressed, we performed an additional expression regulations analysis of variance comparing Flag-EVI1 and Flag-EVI1Δ324 Δ fi To detect whether EVI1 324 and EVI1 speci cally modulate the samples. Interestingly, only 30 genes were identified above expression of genes, we transiently expressed Flag-tagged cutoff (Supplementary Figures S5d and e), which represented versions in HeLa cells, which display very low levels of only 0.13% of all genes assessed. Among them, three members endogenous EVI1. A DNA plasmid coding for Flag only served of the aldo/keto reductase superfamily (AKR1C1, 2 and 3) were as a control (Supplementary Figure S5a). We then isolated RNA found significantly upregulated in cells expressing Flag-EVI1 from these cells and performed a microarray analysis with four compared with cells producing Flag-EVI1Δ324. AKR1C1 and replicates, using the current generation of cDNA microarray three genes were found as EVI1 targets in the EVI1 ChIP-seq data GeneChips (Supplementary Figure S5b). We first compared the set,26 but not in the ChIP-seq EVI1Δ324 data set, suggesting they Flag-EVI1 samples versus Flag and Flag-EVI1Δ324 samples versus may constitute specific EVI1 direct target genes. DUSP4,oneof Flag separately. We identified 1071 and 1118 genes significantly the most significant genes differentially expressed between deregulated when EVI1Δ324 or EVI1 were expressed, Flag-EVI1 and Flag-EVI1Δ324 treatments, and PHLDB2 were also

Figure 6. Evaluation of the cooperation between EVI1Δ324 and AP1 transcription factors. (a) Immunoprecipitation of FOS protein from SKOV3 nuclear extracts. Western blotting could detect association of FOS AP1 subunit with both EVI1Δ324 and EVI1 proteins. (b) Distribution of the distance between EVI1Δ324 enrichment peaks and AP1 exact TGACTCA motif. (c) In-vitro DNA-binding assays from SKOV3 nuclear lysates to assess EVI1, EVIΔ324, JUN and FOS binding to the motifs found enriched in proximity of the EVI1Δ324 ChIP-seq peaks. The labels at the top represent the DNA probe used. The labels at the side indicate the proteins revealed by immunoblotting. (d) Overlap between EVI1Δ324 peaks and EVI1 peaks carrying an AP1 motif. We considered the overlap true when the distance between the centers of EVI1Δ324 and EVI1 peaks was o100 bp. (e) Distribution of the distances between the AP1 motifs detected in proximity of both EVI1Δ324 and EVI1 ChIP-seq peaks.

occupied by EVI1 but not EVI1Δ324. Only the MUC13 gene was interaction with the AP1 transcription factor. The distribution of found exclusively occupied by EVI1Δ324. Overall, these result the distance between EVI1Δ324 enrichment peaks and AP1 TRE shows that genes regulated by EVI1 or EVI1Δ324 in HeLa cells exact consensus motifs (TGACTCA) showed a marked enrichment are matching at 99.87%, indicating that these two isoforms of the presence of AP1 motifs near EVI1Δ324-binding sites share very similar transcriptional regulatory functions. (Figure 6b). Interestingly, the AP1 motif was found near but Δ This suggests that EVI1 and EVI1 324 isoforms modify rather not exactly at the center of EVI1Δ324 peaks, suggesting physiological cell functions in an identical manner. To verify this cooperation between AP1 and EVI1Δ324 rather than a competi- hypothesis, we performed cell-based experiments. Suboptimal tion to bind DNA (Figure 6b). Of the 2003 EVI1Δ324-binding sites, EVI1 expression in HeLa cells was previously found to accelerate 14.8% displayed an AP1 motif within ± 250 bp of their center, cell proliferation and colony formations.26 We now compared the whereas only 3.8% of 20 009 control match regions carried an effect of Flag-EVI1 and Flag-EVI1Δ324 transfections in these assays − 86 Δ exact AP1 motif (enrichment P-value = 1.3 × 10 , binomial test). and found that either EVI1 or EVI1 324 expression could induce Δ cell proliferation and colony formation, with similar efficiency These data strongly suggest that AP1 and EVI1 324 transcription fi (Figures 5c and d). Thus, identical transcriptional regulations by factors co-localize at speci c genomic loci. As the ETS-like motif fi EVI1 and EVI1Δ324 may be the reason for similar effects of these we identi ed displays mild similarity with the conserved isoform expressions on HeLa cell proliferation and anchorage- ‘TGACTCA’ AP1 TRE motif, we examined whether EVI1 and independent growth. EVI1Δ324 could bind to an AP1 motif. An in-vitro binding assay confirmed experimentally the association of endogenous EVI1 and Δ Cooperative binding of EVI1Δ324 and AP1 EVI1 324 proteins to DNA biotin-labeled probes carrying the ETS- like motif. However, these two MECOM variants could not bind to In our former study, EVI1 and AP1 (Efer and Wagner37) transcription factors were described to cooperate and synergize an AP1 DNA probe. Conversely, AP1 subunits FOS and JUN did not in invasive cancer.26 Not only EVI1 and AP1 proteins interacted but occupy the probe carrying the MECOM ETS-like motif (Figure 6c). fi also some of their DNA-binding sites co-localized in conserved Thus, AP1 TRE and MECOM motifs were speci c to their genomic regions. As this cooperation involved the transcriptional respective TFs. Δ regulation by the EVI1 ETS-like ZNF domain, we investigated We next overlapped the double EVI1 324/AP1- and EVI1/AP1- whether EVI1Δ324, which also carries this domain, could binding sites. One hundred and sixty-nine sites were found in contribute to the cooperation with AP1 TRE (TPA DNA-response common, which represented 57.0% of all EVI1Δ324/AP1 loci elements). A co-immunoprecipitation experiment demonstrated (Figure 6d). Interestingly, most of these 169 EVI1 ChIP-seq peaks that FOS, a major AP1 subunit, interacted with both EVI1 and were in proximity to the exact same genomic AP1 motifs EVI1Δ324 (Figure 6a). Thus, EVI1Δ324 also displayed a physical (Figure 6e).

Figure 7. Characterization of binding sites specific to EVI1 isoform that enrich in hotspots of recombination. (a–c) The ChIP-seq data set (38 649 peaks) was subdivided in 5 groups according to the peaks’ fold change (a). Higher fold changes (Fc) represent stronger DNA binding for EVI1. The percentage of peaks located within various categories of repeating elements (b) and of simple repeats (SSR) (c) are shown for each fold change group. Peaks in TCTA/TAGA and TGGA/TCCA SSR are overrepresented when the fold change increases (c). (d) Genomic distribution of the 12 618 significant EVI1 peaks. The peaks within TCTA/TAGA and TGGA/TCCA SSR are enriched in loci far from genes. (e) Average meiotic recombination rates were calculated from three independent data sets: deCODE, Marshfield and Genethon.45–47 The EVI1 peaks in TCTA/TAGA and TGGA/TCCA SSR were associated with higher recombination rates, suggesting they tend to locate near hotspots of meiotic recombination. Error bars represent the s.e.m. P-values are obtained by unpaired t-test. (f) The expression of MECOM gene was found significantly upregulated in Gene Expression Omnibus (GEO) database for microarray data sets from tissues including meiotic cells.

Regions bound by the N-terminal EVI1 ZNF domain are enriched in (Supplementary Figures S6a and b). Moreover, these sites often simple repeats and are associated with a high meiotic fell within repeating elements (Supplementary Figure S6a). To recombination rate further characterize such repeat-associated binding sites, we Two reports have previously demonstrated the inability of studied the whole EVI1 ChIP-seq data set of 38 649 peaks that we EVI1Δ324 or EVI1 protein mutated in its first ZNF domain to divided in 5 subgroups according to their associated fold change transform fibroblasts,15,21 thus indicating that the oncogenic (Figure 7a). We found a marked positive association of simple activity of the ETS-like ZNF domain of MECOM variants is not sequence repeats (SSR) with high fold changes (Figure 7b). We fi sufficient for transformation of normal diploid cells. It was then assessed the different SSR families and found that speci c families were responsible for this effect, TAGA, TCTA and CAGAGA, therefore believed that the transcriptional activity of the and TCTCTG (Figure 7c). These repeats generated reiterations of N-terminal EVI1-binding domain accounted for its transformant GATA or GACA motifs (Supplementary Figure S6a) that constitute and oncogenic action.15,21 However, our study revealed that the Δ the main core of the binding site for the EVI1 N-terminal transcriptional activity of EVI1 and EVI1 324 are barely different in domain.26,36,38,39 When we compared the number of GATA or HelLa cells (Figure 5), demonstrating a minimal contribution of the GACA instances present at the peaks with their fold change, we EVI1 N-terminal ZNF domain to transcriptional regulation. found that EVI1-binding affinity increased when GATA/GACA We thought that a better characterization of the EVI1-binding occurrence increased (Supplementary Figure S6c). An experimen- sites that depend on its N-terminal domain may provide insights tal DNA probe assay consistently showed stronger EVI1 binding regarding its transforming activity. Therefore, we carefully when the copy number of TAGA or CAGAGA SSR sequences was analyzed the EVI1 ChIP-seq peaks with GATAGA motifs.26 We higher (Supplementary Figure S6d). noticed that EVI1 often bound such sites with a very high binding A total of 919 peaks among the 12 618 significant EVI1 affinity, as indicated by high EVI1 ChIP-seq fold changes ChIP-seq peaks contained TAGA/TCTA or CAGAGA/TCTCTG SSR

© 2016 Macmillan Publishers Limited Oncogene (2016) 2311 – 2321 Genomic features of EVI1 and EVI1Δ324 isoforms A Sayadi et al 2318 (Supplementary Table S2) and they tended to display larger fold of the MECOM gene. EVI1 variant was previously shown as change (Supplementary Figure S6e). In order to investigate expressed and functional in ovarian cancer and ovarian cancer cell whether these peaks were associated with any particular process lines.7,9,10,55 To better understand the function of EVI1Δ324 and of transcriptional regulation, we obtained their distribution the specific differences with the EVI1 isoform, we determined the according to their location in promoters and genes. Interestingly, binding sites of EVI1Δ324 by ChIP-seq in SKOV3 epithelial ovarian most of these peaks were far from genes and therefore likely not adenocarcinoma cells and compared these results with our involved in transcriptional regulation (Figure 7d). The association previous study.26 We confirmed that EVI1Δ324 displayed tran- of Chip-seq peaks to the mentioned SSR was specifically observed scriptional activity through its C-terminus ZNF domain and we for EVI1-occupied sites, and not for EVI1Δ324 or SIRT6, which do identified an ETS-like DNA-binding motif, CCATCTCC, highly not bind to DNA with a GA[T/C]A-like motif (Supplementary Figure prominent at EVI1Δ324-occupied sites. Further comparative S6f). The GATA-1 transcription factor-binding motif ‘AGATAA,’40 on analyses with our previous report26 revealed that EVI1 and the other hand, resembles the EVI1 N-terminus DNA-binding EVI1Δ324 share common transcriptional activities linked to their motif26,36,38,39 and we could thus detect a mild enrichment identical C-terminal ZNF domain. For instance, we found that EVI1 of GATA-1 ChIP-seq peaks in these TAGA/TCTA or CAGAGA/ transcription factor used the same ETS-like motif to bind to DNA. TCTCTG SSR. Indeed, as EVI1Δ324 has only one DNA ZNF-binding domain, EVI1 has recently been implicated in the induction of genomic unlike EVI1 that has two, our present ChIP-seq study allowed to instability,41,42 a cancer hallmark that permits cell transformation identify a more reliable and consistent ETS-like DNA-binding and drives tumor progression.43 Genomic instability involves faulty motif. In addition, EVI1Δ324-occupied genes were EVI1 target DNA repair and mistakes in recombination.44 As the few identified genes to a large extent (70.7%). These genes were enriched in genes that could possibly be regulated by EVI1 through a GATAGA genes that modulate important tumorigenic functions including motif were not involved in cell cycle regulation, recombination, cell differentiation, angiogenesis and apoptosis. These similarities replication or DNA repair,26 we investigated whether the EVI1 sites in DNA binding were reflected by virtually identical gene associated with SSR were located near hotspots of recombination. expression changes (99.87% similar) on expression of Flag- We extracted meiotic recombination rates assigned to human EVI1Δ324 or Flag-EVI1 in HeLa cells. These results strongly suggest genomic regions. Recombination events from three independent that EVI1Δ324 also participates in neoplastic functions transcrip- studies were used, where marker-based genetic maps were built tionally controlled by the MECOM gene. In support of this idea, we from individuals from large families to identify meiotic events on found that EVI1Δ324 expression could enhance HeLa cell each chromosome. The first study was the deCODE genetic map proliferation and anchorage-independent growth to a similar based on 5136 microsatellite markers for 146 families with 1257 extent as EVI1 expression. meiotic events.45 The second one was the Marshfield genetic map, EVI1 was previously found to cooperate with AP1 transcription created with 8325 short tandem repeat polymorphisms for 8 factor.37,56 EVI1 proteins and AP1 proteins interact and bind to families consisting of 134 individuals with 186 meioses46 and the proximal genomic sites. EVI1 also regulate transactivation of AP1 last one was the Genethon genetic map, based on 5264 subunits.26,57 Finally, synergistic cooperative interaction between microsatellites for 8 families consisting of 188 meioses.47 We EVI1 and the AP1 family member FOS in the regulation of cell determined the recombination rates of genomic windows in adhesion and proliferation was described.26 Here we find that which EVI1 enrichment peaks with and without repeat-associated EVI1Δ324 also interacts with FOS AP1 subunit and co-localize at binding sites in TAGA/TCTA or CAGAGA/TCTCTG SSR were located. similar genomic loci. Not only EVI1 and EVI1Δ324 isoforms target Interestingly, the 919 EVI1 peaks in these SSR displayed an the same genes but they also cooperate with the AP1 transcrip- increased average recombination rate compared with other peaks tion factor, indicating that AP1 may help to recruit EVI1 and (Figure 7e). This indicates that EVI1 binding to these specific repeat- EVI1Δ324 proteins to particular AP1-bound sites. associated binding sites may have a function in the regulation of The structural difference of EVI1Δ324 compared with EVI1 is the DNA recombination. Consistently, we observed high mRNA levels of loss of two ZNF motifs that leads to the disruption of the GATA- the murine Mecom gene in four independent microarray experiments like N-terminal DNA-binding domain (Supplementary Figure S7). performed on germ cells or reproductive organs in which meiotic EVI1 variant transforming activity was attributed to its N-terminal DNA recombination occurs (Figure 7f). Mecom expression was ZNF domain.15,21 However, EVI1 targets surprisingly few genes by found high just before meiosis in E13,48,49 murine oocytes50 and means of this particular ZNF domain,26 which may not constitute a spermatogonia.51 Mecom was also found highly expressed in normal sufficient explanation for the differential ability to transform adult testis52 and in normal testis from 1- to 11-day-old mice.53 fibroblasts. Interestingly, GATA-like EVI1-binding sites displayed a Overall, our results demonstrate that EVI1 occupies genomic higher binding affinity and were rather located far from genes. We regions enriched in TAGA/TCTA or CAGAGA/TCTCTG SSR through found that they enrich in specific repeating elements, TAGA/TCTA its N-terminal domain with high binding affinity. These EVI1- and CAGAGA/TCTCTG SSR, which contain repetitions of GATA or binding sites are associated with greater meiotic recombination GACA motifs constituting the main core of the EVI1 N-terminus rates, suggesting they may be involved in the regulation of DNA recognition motif.26,36,38,39 As EVI1 was recently found to trigger recombination. genomic instability in cancer,41,42 and because the maintenance of genomic stability is one of the most important defenses against neoplastic transformation, we aimed to identify whether EVI1 DISCUSSION binding through its N-terminal domain could be responsible for The MECOM complex gene encodes several transcription factor such activity. Insertional activation of EVI1 in patients, who variants including MDS1-EVI1, EVI1 and EVI1Δ324. Although eventually developed leukemia in consequence of gene therapy, MDS1-EVI1 expression has been associated with tumor suppressor was previously found to increase phospho-Histone H2AX signals in activity and good prognosis in cancer,9,54 the EVI1 isoform is a peripheral bone marrow cells,42 indicating the appearance of known aggressive oncogene in various cancer types whose double-stranded DNA breaks and subsequent repair by homo- expression is associated with poor prognosis.1,2,4,17,19 Although logous recombination.58 However, this finding could not be EVI1Δ324 is found co-expressed with EVI1, its activity in cancer, or reproduced in another study on epithelial ovarian cancer.55 function in cells, has not been previously characterized. This is Nevertheless, several microarray experiments showed that murine likely because the EVI1 oncoprotein was found to transform Mecom transcript was highly expressed in germ cells in which fibroblasts, whereas EVI1Δ324 could not,15,21 suggesting that the meiotic DNA recombination occurs.48–51 Indeed, chromosome EVI1 variant may solely be responsible for the transformant action breaks and homologous recombination also take place during

Oncogene (2016) 2311 – 2321 © 2016 Macmillan Publishers Limited Genomic features of EVI1 and EVI1Δ324 isoforms A Sayadi et al 2319 meiosis to produce new combinations of DNA sequences. It is transfection, chromatin extraction, ChIP, quantitative PCR and sequencing known that meiotic recombination sites are not randomly located were performed as previously described.26 Peaks were called and the fold on the chromosomes and breaks often occur at recombination change of ChIP DNA versus input DNA was calculated for each peak as 29 hotspots, regions in chromosomes that have high rates of described previously. recombination.45–47 We found that the EVI1 sites within TAGA/ TCTA and CAGAGA/TCTCTG SSR are genomic regions associated Co-immunoprecipitations and biotin-labeled DNA probe binding with significantly higher rates of meiotic recombination. This assays suggests that EVI1 binding through its N-terminal domain may be Protein extraction, immunoprecipitations, DNA probe pull down and involved in the control of genomic maintenance. Thus, the western-blottings were performed as previously described.22,26 suspected transforming activity of the EVI1 N-terminus ZNF Antibodies against the C-terminus of murine EVI1 protein was kindly domain could be linked to a role in the regulation of homologous provided by Archibald Perkins. recombination rather than specific transcriptional changes. In line with this, a mass spectrometry analysis, further confirmed with Microarray experiments validation by co-immunoprecipitations, reported interactions of HeLa cells were used and confirmed mycoplasma free (MycoDtect kit from EVI1 with proteins involved in DNA damage repair, DNA Greiner BioOne). Cell transfection, RNA extraction and microarray recombination and meiosis in SKOV3 and HeLa cells.22 Overall, experiments were performed as described previously with Human Gene these findings hint a direct involvement of EVI1 in regulating 1.0ST arrays (Affymetrix, Santa Clara, CA, USA).26 The statistical analyses to genomic instability through the protein recruitment at specific identify genes that were significantly differentially expressed were genomic sites. performed by analysis of variance using Partek Genomics Suite (version As reported by several studies, a number of transcription factors 6.4, St Louis, MO, USA). also regulate genomic stability.59 For instance, the BRCA1 tumor suppressor gene, known to maintain genomic stability,60 also has Cell-based functional assays a role in the regulation of transcription.61 Other transcription The proliferation assay was carried on with mycoplasma-free HeLa cells factors were found to associate with complexes regulating DNA transfected with indicated plasmids in 24-well plates. The cells were recombination. For example, the YY1 oncogene activates or counted 3 days after transfection with a Vi-CELL Cell Viability Analyzer represses promoters and is essential in homologous (Beckman Coulter, Vienna, Austria). The soft-agar colony-formation assay recombination-based DNA repair in meiosis and cancer.62,63 was performed using Cytoselect 96-well Cell Transformation Assay Kit (Cell Biolabs, San Diego, CA, USA) as previously described.26 Unpaired t-test Similarly, the tumor suppressor p53 was found to downregulate was applied, as results showed normal distribution (Shapiro–Wilk homologous recombination independently of its transcriptional normality test). transactivation function.64 Previous data41,42 and our present data suggest that the EVI1 isoform of MECOM displays dual roles in transcriptional regulation and maintenance of genome stability. De novo motif search However, the exact mechanism by which EVI1 expression triggers The quality of the obtained motif depends essentially on the input data Δ phospho-H2AX,42 reflecting DNA breaks, or centrosomal used. We used sequences of 100 bp length at EVI1 324 ChIP-seq peaks, 41 located in promoters and with a fold change 410. Motif prediction was amplification is still unclear. Our data suggest that EVI1 may 30 fi performed using Weeder program and the CCATCTCC motif was found locate to some speci c hotspots of recombination through its to be statistically enriched in promoter’s sequences. To assess the proximal ZNF domain binding to specific simple repeats. As EVI1 enrichment of the motif in all peaks of the ChIP-seq data, we mapped has no enzymatic activity, the protein may serve to recruit factors the motif matrix to the peaks and to a match control data set using that mark DNA, catalyze double-stranded DNA break formation65 CisGenome software.32 Enrichment P-values were then calculated by a or prevent the repair of these breaks.66 binomial test. The MECOM product, previously called PRDM3, belongs to the 67,68 PRDM family of SET domain-containing proteins. PRDM factors Matrix quality control and mapping to EVI1 and EVI1Δ324 ChIP- either function as direct histone methyltransferases or recruit seq data histone-modifying enzymes to target genomic sites. The PRDM9 Matrices obtained by de-novo search was submitted to a matrix-quality family member has been shown to locate to specific recombina- program, to assess its occurrence in EVI1 (Bard-Chapeau et al.26) and tion hotspots with its ZNF domain and to label them with its EVI1Δ324 ChIP-seq data. We generated corresponding permuted control histone methylase domain, while it recruits other proteins to matrices. A sequence of 100-pb around each ChIP-seq peak was used, initiate the double-stranded breaks needed for meiosis.69–71 giving 12 618 or 2003 sequences representing EVI1 or EVI1Δ324 genome- Although the EVI1 isoform of the PRDM3/MECOM gene lacks the wide binding sites (hg18), respectively. The RSAT tool33,34 generated a PR domain and therefore cannot modify histones, partial structural second data set of upstream-noorf sequences of the human genome to be similarities between EVI1 and PRDM9 could serve to recruit used as a control, to determine the background matrix occurrence. The matrix was mapped to the 12 618 and 2003 sequences and the control proteins to initiate the double-stranded breaks. ‘ ’ fi sequences using the program matrix-scan implemented in the RSAT Overall, our genomic analyses uncovered common and speci c tool,33 using defaults parameter. To identify whether matrices can functions of two major isoforms of the MECOM gene, EVI1 and differentiate true EVI1/ EVI1Δ324-binding sites to control sequences, the EVI1Δ324. Unexpectedly, their impact on transcription was highly software tool ‘matrix-quality’ implemented in RSAT tool measured the Ws concordant and thus appears insufficient to explain the difference and the P-value assigned to each sequence.34 Positive Ws indicate that the in their oncogenicity. EVI1 displayed additional binding sites sequence is more likely to be an instance of the motif than an instance of linked to its N-terminal ZNF domain. These genomic regions were the background. P-value for dCDF (decreasing cumulative distribution associated with higher recombination rates, which may explain function) is the probability to obtain a given Ws by chance in the fl why this specific isoform presents additional transforming activity. background sequences; the P-value thus re ects the risk of false positive. For example, a P-value threshold of 0.001 correspond to one false-positive prediction every kilobase.34 Distribution of the Ws values for EVI1- or MATERIALS AND METHODS EVI1Δ324-binding sequences and control sequences fit at low Ws values and separate at high Ws values. The separation of the distribution curves ChIP and ChIP-seq assays corresponds to high-scoring functional EVI1/EVI1Δ324-bindings sites. Flag-EVI1Δ324 expression plasmid was made similarly as Flag-EVI1,26 from We mapped the matrices to 100 bp sequences at each of the 12 618 a cDNA library generated from HEL cells. SKOV3 epithelial ovarian EVI1- or 2003 EVI1Δ324-binding sites using CisGenome.32 The control adenocarcinoma cells were used and confirmed mycoplasma free match regions were generated with CisGenome. Enrichment P-values were (MycoDtect kit from Greiner BioOne, Kremsmunster, Austria). SKOV3 cell calculated using a binomial test.

© 2016 Macmillan Publishers Limited Oncogene (2016) 2311 – 2321 Genomic features of EVI1 and EVI1Δ324 isoforms A Sayadi et al 2320 Recombination rates 11 Lugthart S, van Drunen E, van Norden Y, van Hoven A, Erpelinck CA, Valk PJ et al. We extracted meiotic recombination rates associated to human genomic High EVI1 levels predict adverse outcome in acute myeloid leukemia: prevalence regions that were identified through marker genetic analyses of of EVI1 overexpression and chromosome 3q26 abnormalities underestimated. families.45–47 In these data sets available on UCSC, recombination rates Blood 2008; 111: 4329–4337. are assigned to 1-Mb windows representing the average recombination 12 Ogawa S, Kurokawa M, Tanaka T, Tanaka K, Hangaishi A, Mitani K et al. Increased rate of the bases contained within the window. We averaged the Evi-1 expression is frequently observed in blastic crisis of chronic myelocytic recombination rates of genomic windows in which EVI1 peaks fell. We leukemia. Leukemia 1996; 10:788–794. assessed various subgroups of EVI1 ChIP-seq peaks, with or without SSR. 13 Bartholomew C, Clark AM. Induction of two alternatively spliced evi-1 proto- 9 – The P-value was calculated using an unpaired two-tailed t-test using oncogene transcripts by cAMP in kidney cells. Oncogene 1994; :939 942. recombination rates assigned to each peak. 14 Bordereaux D, Fichelson S, Tambourin P, Gisselbrecht S. Alternative splicing of the Evi-1 zinc finger gene generates mRNAs which differ by the number of zinc finger motifs. Oncogene 1990; 5: 925–927. Computational analyses 15 Kilbey A, Bartholomew C. Evi-1 ZF1 DNA binding activity and a second distinct The other ChIP-seq data sets were obtained on Gene Expression Omnibus transcriptional repressor region are both required for optimal transformation of database. The accession numbers are GSE25213 for Flag-EVI1 ChIP-seq, Rat1 fibroblasts. Oncogene 1998; 16: 2287–2291. GSE31477 for GATA-1 ChIP-seq40 and GSE31477 for SIRT6 data set. The 16 Morishita K, Parganas E, Douglass EC, Ihle JN. Unique expression of the human exact AP1 motif consensus was mapped to EVI1- and EVI1Δ324-binding Evi-1 gene in an endometrial carcinoma cell line: sequence of cDNAs and struc- sites and to a control match sequences created with Cisgenome tool.32 ture of alternatively spliced transcripts. Oncogene 1990; 5: 963–971. Enrichment P-values were calculated using an exact Fisher test. Biological 17 Wieser R. The oncogene and developmental regulator EVI1: expression, processes clustering and classification was performed using DAVID biochemical properties, and biological functions. Gene 2007; 396:346–357. bioinformatics (GOTERM, biological processes). 18 Hirai H, Izutsu K, Kurokawa M, Mitani K. Oncogenic mechanisms of Evi-1 protein. MECOM expression levels in tissues containing meiotic cells subpopula- Cancer Chemother Pharmacol 2001; 48:S35–S40. tions were obtained from the Gene Expression Omnibus database; 19 Mitani K. Molecular mechanisms of leukemogenesis by AML1/EVI-1. Oncogene accession numbers of these microarray data sets are indicated in 2004; 23: 4263–4269. Figure 7f. The data analysis tool of Gene Expression Omnibus was used 20 Yoshimi A, Kurokawa M. Evi1 forms a bridge between the epigenetic machinery to extract the results for microarray probes corresponding to the and signaling pathways. Oncotarget 2011; 2:575–586. MECOM gene. 21 Palmer S, Brouillet JP, Kilbey A, Fulton R, Walker M, Crossley M et al. Evi-1 transforming and repressor activities are mediated by CtBP co-repressor proteins. J Biol Chem 2001; 276: 25834–25840. Accession numbers 22 Bard-Chapeau EA, Gunaratne J, Kumar P, Chua BQ, Muller J, Bard FA et al. The ChIP-seq and the microarray data have been deposited in the Gene EVI1 oncoprotein interacts with a large and complex network of proteins and Expression Omnibus database under accession numbers GSE42354 and integrates signals through protein phosphorylation. Proc Natl Acad Sci USA 2013; GSE42251, respectively. 110: E2885–E2894. 23 Bard-Chapeau EA, Szumska D, Jacob B, Chua BQ, Chatterjee GC, Zhang Y et al. Mice carrying a hypomorphic Evi1 allele are embryonic viable but exhibit severe CONFLICT OF INTEREST congenital heart defects. PLoS One 2014; 9: e89397. 24 Goyama S, Yamamoto G, Shimabe M, Sato T, Ichikawa M, Ogawa S et al. Evi-1 is a The authors declare no conflict of interest. critical regulator for hematopoietic stem cells and transformed leukemic cells. Cell Stem Cell 2008; 3: 207–220. ACKNOWLEDGEMENTS 25 Zhou LY, Chen FY, Shen LJ, Wan HX, Zhong JH. Arsenic trioxide induces apoptosis in the THP1 cell line by downregulating EVI-1. Exp Ther Med 2014; 8:85–90. This work was supported by the Agency for Science, Technology and Research 26 Bard-Chapeau EA, Jeyakani J, Kok CH, Muller J, Chua BQ, Gunaratne J et al. (Singapore). We thank Archibald Perkins, University of Rochester, for sharing his Ecotopic viral integration site 1 (EVI1) regulates multiple cellular processes mouse Evi1 antibodies. important for cancer and is a synergistic partner for FOS protein in invasive tumors. Proc Natl Acad Sci USA 2012; 109: 2168–2173. 27 Aytekin M, Vinatzer U, Musteanu M, Raynaud S, Wieser R. Regulation of the REFERENCES expression of the oncogene EVI1 through the use of alternative mRNA 5'-ends. 1 Goyama S, Kurokawa M. Pathogenetic significance of ecotropic viral integration Gene 2005; 356:160–168. site-1 in hematological malignancies. Cancer Sci 2009; 100:990–995. 28 Fears S, Mathieu C, Zeleznik-Le N, Huang S, Rowley JD, Nucifora G. Intergenic 2 Goyama S, Kurokawa M. Evi-1 as a critical regulator of leukemic cells. Int J Hematol splicing of MDS1 and EVI1 occurs in normal tissues as well as in myeloid leukemia 2010; 91: 753–757. and produces a new member of the PR domain family. Proc Natl Acad Sci USA 3 Metais JY, Dunbar CE. The MDS1-EVI1 gene complex as a retrovirus integration 1996; 93: 1642–1647. site: impact on behavior of hematopoietic cells and implications for gene therapy. 29 Chen X, Xu H, Yuan P, Fang F, Huss M, Vega VB et al. Integration of external Mol Ther 2008; 16: 439–449. signaling pathways with the core transcriptional network in embryonic stem cells. 4 Nucifora G, Laricchia-Robbio L, Senyuk V. EVI1 and hematopoietic disorders: Cell 2008; 133: 1106–1117. history and perspectives. Gene 2006; 368:1–11. 30 Pavesi G, Mereghetti P, Zambelli F, Stefani M, Mauri G, Pesole G. MoD Tools: 5 Bonome T, Levine DA, Shih J, Randonovich M, Pise-Masison CA, Bogomolniy F regulatory motif discovery in nucleotide sequences from co-regulated or et al. A gene signature predicting for survival in suboptimally debulked patients homologous genes. Nucleic Acids Res 2006; 34: W566–W570. with ovarian cancer. Cancer Res 2008; 68: 5478–5486. 31 Wei GH, Badis G, Berger MF, Kivioja T, Palin K, Enge M et al. Genome-wide analysis 6 Brooks DJ, Woodward S, Thompson FH, Dos Santos B, Russell M, Yang JM et al. of ETS-family DNA-binding in vitro and in vivo. EMBO J 2010; 29: 2147–2160. Expression of the zinc finger gene EVI-1 in ovarian and other cancers. Br J Cancer 32 Ji H, Vokes SA, Wong WH. A comparative analysis of genome-wide chromatin 1996; 74: 1518–1525. immunoprecipitation data for mammalian transcription factors. Nucleic Acids Res 7 Dutta P, Bui T, Bauckman KA, Keyomarsi K, Mills GB, Nanjundan M. EVI1 splice 2006; 34:e146. variants modulate functional responses in ovarian cancer cells. Mol Oncol 2013; 7: 33 Turatsinze JV, Thomas-Chollier M, Defrance M, van Helden J. Using RSAT to scan 647–668. genome sequences for transcription factor binding sites and cis-regulatory 8 Mok SC, Bonome T, Vathipadiekal V, Bell A, Johnson ME, Wong KK et al. A gene modules. Nat Protoc 2008; 3: 1578–1588. signature predictive for outcome in advanced ovarian cancer identifies a survival 34 Medina-Rivera A, Abreu-Goodger C, Thomas-Chollier M, Salgado H, Collado-Vides J, factor: microfibril-associated glycoprotein 2. Cancer Cell 2009; 16: 521–532. van Helden J. Theoretical and empirical quality assessment of transcription factor- 9 Nanjundan M, Nakayama Y, Cheng KW, Lahad J, Liu J, Lu K et al. Amplification of binding motifs. Nucleic Acids Res 2011; 39:808–824. MDS1/EVI1 and EVI1, located in the 3q26.2 amplicon, is associated with favorable 35 Hoyt PR, Bartholomew C, Davis AJ, Yutzey K, Gamer LW, Potter SS et al. The Evi1 patient prognosis in ovarian cancer. Cancer Res 2007; 67: 3074–3084. proto-oncogene is required at midgestation for neural, heart, and paraxial 10 Sunde JS, Donninger H, Wu K, Johnson ME, Pestell RG, Rose GS et al. Expression mesenchyme development. Mech Dev 1997; 65:55–70. profiling identifies altered expression of genes that contribute to the inhibition of 36 Yuasa H, Oike Y, Iwama A, Nishikata I, Sugiyama D, Perkins A et al. Oncogenic transforming growth factor-beta signaling in ovarian cancer. Cancer Res 2006; 66: transcription factor Evi1 regulates hematopoietic stem cell proliferation through 8404–8412. GATA-2 expression. EMBO J 2005; 24: 1976–1987.

Oncogene (2016) 2311 – 2321 © 2016 Macmillan Publishers Limited Genomic features of EVI1 and EVI1Δ324 isoforms A Sayadi et al 2321 37 Eferl R, Wagner EF. AP-1: a double-edged sword in tumorigenesis. Nat Rev Cancer 54 Zhang Y, Stehling-Sun S, Lezon-Geyda K, Juneja SC, Coillard L, Chatterjee G et al. 2003; 3: 859–868. PR-domain-containing Mds1-Evi1 is critical for long-term hematopoietic stem cell 38 Delwel R, Funabiki T, Kreider BL, Morishita K, Ihle JN. Four of the seven zinc fingers function. Blood 2011; 118: 3853–3861. of the Evi-1 myeloid-transforming gene are required for sequence-specificbindingto 55 Jazaeri AA, Ferriss JS, Bryant JL, Dalton MS, Dutta A. Evaluation of EVI1 and EVI1s GA(C/T)AAGA(T/C)AAGATAA. Mol Cell Biol 1993; 13: 4291–4300. (Delta324) as potential therapeutic targets in ovarian cancer. Gynecol Oncol 2010; 39 Yatsula B, Lin S, Read AJ, Poholek A, Yates K, Yue D et al. Identification of binding 118:189–195. sites of EVI1 in mammalian cells. J Biol Chem 2005; 280: 30712–30722. 56 Jochum W, Passegue E, Wagner EF. AP-1 in mouse development and tumor- 40 Fujiwara T, O'Geen H, Keles S, Blahnik K, Linnemann AK, Kang YA et al. Discovering igenesis. Oncogene 2001; 20: 2401–2412. hematopoietic mechanisms through genome-wide analysis of GATA factor 57 Tanaka T, Mitani K, Kurokawa M, Ogawa S, Tanaka K, Nishida J et al. Dual functions chromatin occupancy. Mol Cell 2009; 36:667–681. of the AML1/Evi-1 chimeric protein in the mechanism of leukemogenesis in 41 Karakaya K, Herbst F, Ball C, Glimm H, Kramer A, Loffler H. Overexpression of EVI1 t(3;21) leukemias. Mol Cell Biol 1995; 15: 2383–2392. interferes with cytokinesis and leads to accumulation of cells with supernumerary 58 Yuan J, Adamski R, Chen J. Focus on histone variant H2AX: to be or not to be. FEBS centrosomes in G0/1 phase. Cell Cycle 2012; 11: 3492–3503. Lett 2010; 584: 3717–3724. 42 Stein S, Ott MG, Schultze-Strasser S, Jauch A, Burwinkel B, Kinner A et al. Genomic 59 Nicolas A. Relationship between transcription and initiation of meiotic instability and myelodysplasia with monosomy 7 consequent to EVI1 activation recombination: toward chromatin accessibility. Proc Natl Acad Sci USA 1998; 95: after gene therapy for chronic granulomatous disease. Nat Med 2010; 16:198–204. 87–89. 43 Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell 2011; 60 Wood RD, Mitchell M, Sgouros J, Lindahl T. Human DNA repair genes. Science 144: 646–674. 2001; 291: 1284–1289. 44 Heyer WD, Ehmsen KT, Liu J. Regulation of homologous recombination in 61 Mullan PB, Quinn JE, Harkin DP. The role of BRCA1 in transcriptional regulation eukaryotes. Annu Rev Genet 2010; 44:113–139. and cell cycle control. Oncogene 2006; 25: 5854–5863. 45 Kong A, Gudbjartsson DF, Sainz J, Jonsdottir GM, Gudjonsson SA, Richardsson B 62 Wu S, Shi Y, Mulligan P, Gay F, Landry J, Liu H et al. A YY1-INO80 complex et al. A high-resolution recombination map of the human genome. Nat Genet regulates genomic stability through homologous recombination-based repair. 2002; 31: 241–247. Nat Struct Mol Biol 2007; 14: 1165–1172. 46 Broman KW, Murray JC, Sheffield VC, White RL, Weber JL. Comprehensive human 63 Zaprazna K, Atchison ML. YY1 controls immunoglobulin class switch recombina- genetic maps: individual and sex-specific variation in recombination. Am J Hum tion and nuclear activation-induced deaminase levels. Mol Cell Biol 2012; 32: Genet 1998; 63: 861–869. 1542–1554. 47 Dib C, Faure S, Fizames C, Samson D, Drouot N, Vignal A et al. A comprehensive 64 Janz C, Wiesmuller L. Wild-type p53 inhibits replication-associated homologous genetic map of the human genome based on 5,264 microsatellites. Nature 1996; recombination. Oncogene 2002; 21: 5929–5933. 380: 152–154. 65 Yanowitz J. Meiosis: making a break for it. Curr Opin Cell Biol 2010; 22:744–751. 48 Borum K. Oogenesis in the mouse. A study of the meiotic prophase. Exp Cell Res 66 Tan FJ, Hoang ML, Koshland D. DNA resection at chromosome breaks promotes 1961; 24: 495–507. genome stability by constraining non-allelic homologous recombination. PLoS 49 Speed RM. Meiosis in the foetal mouse ovary. I. An analysis at the light micro- Genet 2012; 8: e1002633. scope level using surface-spreading. Chromosoma 1982; 85: 427–437. 67 Fog CK, Galli GG, Lund AH. PRDM proteins: important players in differentiation 50 Bouma GJ, Affourtit JP, Bult CJ, Eicher EM. Transcriptional profile of mouse and disease. Bioessays 2012; 34:50–60. pre-granulosa and Sertoli cells isolated from early-differentiated fetal gonads. 68 Hohenauer T, Moore AW. The Prdm family: expanding roles in stem cells and Gene Expr Patterns 2007; 7:113–123. development. Development 2012; 139: 2267–2282. 51 Namekawa SH, Park PJ, Zhang LF, Shima JE, McCarrey JR, Griswold MD et al. 69 Baudat F, Buard J, Grey C, Fledel-Alon A, Ober C, Przeworski M et al. PRDM9 is a Postmeiotic sex chromatin in the male germline of mice. Curr Biol 2006; 16:660–667. major determinant of meiotic recombination hotspots in humans and mice. 52 Xia W, Mruk DD, Lee WM, Cheng CY. Unraveling the molecular targets pertinent Science 2010; 327:836–840. to junction restructuring events during spermatogenesis using the Adjudin- 70 Parvanov ED, Petkov PM, Paigen K. Prdm9 controls activation of mammalian induced germ cell depletion model. J Endocrinol 2007; 192: 563–583. recombination hotspots. Science 2010; 327: 835. 53 Schultz N, Hamra FK, Garbers DL. A multitude of genes expressed solely in meiotic 71 Segurel L, Leffler EM, Przeworski M. The case of the fickle fingers: how the PRDM9 or postmeiotic spermatogenic cells offers a myriad of contraceptive targets. Proc zinc finger protein specifies meiotic recombination hotspots in humans. PLoS Biol Natl Acad Sci USA 2003; 100: 12201–12206. 2011; 9: e1001211.

Supplementary Information accompanies this paper on the Oncogene website (http://www.nature.com/onc)