Oncogene (2008) 27, 2412–2421 & 2008 Nature Publishing Group All rights reserved 0950-9232/08 $30.00 www.nature.com/onc ORIGINAL ARTICLE Differentially expressed are marked by 3 9 trimethylation in human cancer cells

JK Wiencke1, S Zheng1, Z Morrison1 and R-F Yeh2

1Laboratory of Neuro and Molecular Epidemiology, Ambulatory Care AC-34, Department of Neurological Surgery, University of California San Francisco, San Francisco, CA, USA and 2Department of Epidemiology and Biostatistics, Center for Bioinformatics and Molecular Biostatistics, University of California San Francisco, San Francisco, CA, USA

Histone H3 lysine 9trimethylation (H3K9Me3) has been H3K9Me3 in three studied genes (HOXA9, TMS1, associated with transcriptional repression, but recent RASSF1A). The localization of H3K9Me3 downstream findings implicate this modification in transcrip- of the TSSs of expressed genes and not within promoter tional activation and mRNA elongation by RNA polymer- regions of hypermethylated and silenced genes is consistent ase II. Here, we applied immunoprecipitation (IP) with a with the proposed coupling of H3K9Me3 with RNA custom DNA tiling microarray containing many transcrip- polymerase II. Our results indicate a need for revising tion factors important in development and cancer (for aspects of the involving H3 lysine . example homeotic genes; N ¼ 683 total genes) to explore Awareness of H3K9Me3 as a mark of activity, not the relationship between H3K9Me3 and other histone repression, is especially important for the classification of modifications with the differential expression of genes. human cancer using chromatin and histone profiles. Cancer cell lines derived from different tissues (2 leukemia, Oncogene (2008) 27, 2412–2421;doi:10.1038/sj.onc.1210895; 2 medulloblastoma) were characterized with IP antibodies published online 29 October 2007 to H3K9Me3, H3K4 dimethylation (H3K4Me2) and H3K9acetylation (). MV4-11 is known to over- Keywords: chromatin immunoprecipitation;histone express the HOXA9 and MEIS1 genes, whereas D283 methylation;histone acetylation;tiling array;; overexpresses the OTX2 homeobox gene. Gene expression histone code was assessed by Affymetrix U133 array. Mapping the number and size of histone markings demonstrated significant colocalization of H3K9Ac and H3K4Me2 with H3K9Me3, indicating a pattern of putative ‘activating’ and ‘repressive’ markings. The median site size was 600–821 bp Introduction and 72–95% or 53–80% of chromatin signal sites were located within 1 kb or 500 bp of transcription start sites Aberrant gene expression in human cancer is thought to (TSS), respectively. A relatively small number of genes be mediated in part through post-translational modifica- displayed additional H3K9Me3 sites in the 50-region distant tions of histone and other DNA binding , which from the TSS. Comparing genes with modification sites to alter the functional properties of chromatin (Gregory those without sites in their promoters confirmed the positive and Shiekhattar, 2004). Covalent modifications of associations of H3K9Ac and H3K4Me2 with gene expres- have been widely studied as pivotal in the sion and revealed that H3K9Me3 is associated with active regulation of gene transcription. A ‘histone code’ genes rather than being a repressive marking as previously hypothesis has evolved to explain how combinations of thought. The positive regulatory effect of all three types of modifications affect transcriptional regulation (Strahl modifications were quantitatively correlated with site size, and Allis, 2000;Turner, 2000). In mammals, histone H3 and applied to absolute gene expression within a single cell lysine 9 acetylation (H3K9Ac) and H3 lysine 4, 36 and 79 line as well as relative expression among pairs of cell lines. methylation are linked with active transcription, whereas Extended patterns of H3K9Me3 upstream of some genes methylation of histone H3 lysine 9, 27 and histone H4 (for example HOXA9 and OTX2) may result from the lysine 20 are associated with transcriptionally silent action of multiple promoter elements. We found an inverse . The division of chromatin patterns relationship between promoter DNA hypermethylation and into exclusive active and repressive domains has been questioned by recent studies. For example, human T cells display both H3K9 acetylation and H3K27 trimethyla- Correspondence: Dr JK Wiencke, Department of Neurological tion among many transcribed genes (Roh et al., 2006). Surgery, Laboratory of Neuro and Molecular Epidemiology, Expressed genes in human (Azuara et al., 2006) and Ambulatory Care AC-34, University of California San Francisco, mouse (Bernstein et al., 2006) embryonic stem cells were 1 Irving St., Box no 441, San Francisco, CA, USA. E-mail: [email protected] identified that displayed both activating and repressive Received 13 February 2007;revised 9 September 2007;accepted 11 histone markings. The latter study termed these see- September 2007;published online 29 October 2007 mingly conflicting histone patterns ‘bivalent’ chromatin Histone 3 lysine 9 trimethylation marks expressed genes JK Wiencke et al 2413 structures (Bernstein et al., 2006). Recently, Vakoc et al. and are associated with a different age at onset, tumor (2005) presented a surprising finding that H3K9Me3 was location and survival times compared with classic or present in the transcribed region of eight genes that they anaplastic types (de Haas et al., 2006). Our results show examined. This observation seems at variance with the that HOXA9, MEIS1, OTX2 and many other genes well described association of H3K9Me3 with both expressed in cancer cells contain overlapping histone constitutive (Peters et al., 2003) and facultative hetero- modifications that include H3K9Ac, H3K4Me2 and chromatin (Heard et al., 2001). Vakoc et al. (2005) H3K9Me3. further reported that heterochromatin 1g (HP1g), which contains a chromo-domain predicted to recognize H3K9 methylation, is also present in the transcribed region and may form a complex with RNA Results polymerase II. Importantly, they demonstrated the H3K9Me3, H3K4Me2 and H3K9Ac modifications specificity of the antibodies they used in their chIP overlap the TSS of genes experiments and ruled out cross-reactivity of H3K9Me3 A description of the 683 genes on the tiling array is with other markings that could have produced spurious provided in Supplementary Table 1. The single gene raw results. These and other observations indicate that RNA data applicable to Figures 1–4 is provided in Supple- polymerase II associates with chromatin modifications mentary Table 2. Using ChIP-on-chip analyses that that include histone H3K9 methylation, a mark most interrogated promoter regions covering upstream 5 kb commonly associated with gene silencing (Eissenberg and downstream 1 kb of 683 genes, we found that all and Shilatifard, 2006). A subsequent report showed that three histone modifications overlapped significantly with H4K20 monomethylation, another modification linked the TSS of genes on our platform, with 72–95 or 53–80% with repression, may also be a mark of transcript of chromatin signal sites were located within 1 kb or elongation (Vakoc et al., 2006). Here, we tested whether H3K9Me3 is associated with 500 bp of TSSs, respectively. The three modifications have similar size distributions with medians ranging differentially expressed genes in human cancer cells and from 600 to 821 bp. Results illustrating these findings for specifically genes that control the action of many transcription factors responsible for maintaining the the MV4-11 cell line are depicted in Figure 1. Note that the site size can be regarded as a surrogate measure for differentiated state of cells. We examined the relationship the strength, or more precisely, the extensiveness of of H3K9Me3 with H3K4Me2 and H3K9Ac modifica- the chromatin modification, since they were defined tions in human cancer cell lines using chromatin by neighboring, statistically significant probes in IP- immunoprecipitation and a custom-designed DNA tiling enriched DNA compared to the input controls (see microarray (ChIP-on-chip). The array interrogates 5 kb Materials and methods for the exact definition). For each upstream and 1 kb downstream of the TSS of 683 genes including all known homeodomain transcription factors of the histone markings there was an apparent decreased signal intensity immediately upstream of the TSS. and many other cancer association genes (for example Similar results were obtained with Rch, D283 and Daoy MAGE genes). The 50-location of probes allows us to examine whether H3K9Me3 is restricted to the regions (data not shown). The smoothed curve in panel d of around and downstream of TSSs as predicted (Vakoc Figure 1 shows a subtle difference in the distribution et al., 2005, 2006). Gene expression was assessed by the of the histone marks with slightly more H3K9Ac and 0 to TSSs compared Affymetrix HGU133 Plus 2.0 expression microarray. We H3K4Me2 signals being located 3 with H3K9Me3. examined two cell lines in depth: MV4-11 and D283, both of which overexpress homeobox genes. MV4-11 is a biphenotypic leukemic cell that contains a t(4;11) H3K9Me3, H3K4Me2 and H3K9Ac modifications translocation and resultant fusion protein that activates overlap each other the mixed lineage leukemia (MLL)gene(Canaaniet al., We calculated the relative location of histone markings 2004). MLL is homologous to the Drosophila Trithorax with respect to each other in order to test whether two chromatin modifying genes that proteins with markings occupied the same genomic location or histone H3K4 methyltransferase activity. MLL rearran- whether one marking occurs upstream (50) or down- gements led to many fold overexpression of the HOXA9 stream (30) another marking. Results are presented for and MEIS1 genes (Rozovskaia et al., 2001;Armstrong the MV4-11 cell line in Figure 2. A symmetrical et al., 2002;Yeoh et al., 2002). The medulloblastoma cell distribution indicates no bias with respect to the location line D283 represents the classic histopathological subtype of one marker versus another. Panel a shows the of these tumors that contain an amplified OTX2 homeo- distribution of differences in site location for double- box gene (Boon et al., 2003, 2005;Acampora et al., 2005; positive sites (measured at the midpoint of the site) for all de Haas et al., 2006). OTX2 is an essential transcriptional three possible combinations of bivalent markings. The factor in the development of the central nervous system relative site distributions indicated that most sites are co- (Kimura et al., 2005;Vernay et al., 2005;Rath et al., localized although some differences are suggested. The 2006). Results from D283 were also contrasted with distribution of relative locations of H3K4Me2 and Daoy, a cell line isolated from the nodular/desmoplastic H3K9Ac appeared the most symmetrical in these subtype of medulloblastoma. Nodular medulloblastoma analyses, suggesting no systematic difference in the rarely displays OTX2 amplification and overexpression position of one mark relative to the other. Some shifts

Oncogene Histone 3 lysine 9 trimethylation marks expressed genes JK Wiencke et al 2414 K9Ac K9Me3 K4Me2 80 70 0.6 60 30 0.5 60 50 K9Me3 0.4 40 K9Ac 20 40 0.3 K4Me2 30 density

Frequency 0.2 10 20 20 Frequency Frequency 10 0.1 0 0 0 0.0 -5-4 -3 -2 -1 0 1 -5-4 -3 -2 -1 0 1 -5-4 -3 -2 -1 0 1 -5-4 -3 -2 -1 0 1 Distance from TSS (kb) Distance from TSS (kb) Distance from TSS (kb) Distance from TSS (kb)

K9Ac K9Me3 K4Me2 0.0020 30 60 70 25 50 60 0.0015 50 20 40 40 0.0010 30 15 30 density Frequency 10 Frequency 20 0.0005 Frequency 20 10 10 5 0.0000 0 0 0 500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000 500 1000 1500 2000 Site size (bp) Site size (bp) Site size (bp) Site size (bp) Figure 1 Distribution of distance from transcription start site (TSS) and site size for (a) H3K9Ac, (b) H3K9Me3, (c) H3K4Me2 sites, (d) Overlaying the distribution of the three modifications in the MV4-11 leukemia cell line.

are indicated for the relative locations of H3K9Me3 with Differential gene expression is marked by H3K9Me3 site respect to either H3K9Ac or H3K4Me2. These distribu- presence and size tions were somewhat skewed in the direction indicating In addition to the above observation that H3K9Me3 is that H3K9Me3 sites were located downstream of associated with active expression in a signal-dependent H3K9Ac or H3K4Me2 sites. Interestingly, even though manner within a cell line, we hypothesize that histone K9Me3 sites are on average shorter than K9Ac and modifications can also be predictive of active gene K4Me2 sites (Figure 1, with median sizes at 680, 742.5, expression when comparing the relative expression of a 722 bp respectively for MV4-11), H3K9Me3 sites gene with different cells. Figure 4 illustrates such a appeared to be larger than H3K9Ac or H3K4Me2 sites positive association by comparing the two medulloblas- when they co-localize (Figure 2). This suggests an toma cell lines Daoy and D283. When stratified by the interaction, likely to be H3K9Me3 dominant, between presence or absence of significant H3K9Me3 sites within H3K9Me3 and other active markings. 1 kb of a gene TSS in Daoy and D283 using a stringent criteria (>4-fold IP/input intensity and Po0.0001) into four groups, it is apparent that genes have relatively Gene expression is strongly associated with H3K9Me3 higher expression in the cell line where H3K9Me3 sites site occurrence, size and distance from the TSS are present (Figure 4a;red and green boxes);and when There were 669 genes on our panel that were also H3K9Me3 sites are present in both cell lines, stronger represented on the expression array. We first tested sites (as measured by site size or number of significantly whether the presence of H3K9Me3 was associated with enriched probes) are correlated with higher relative the log2 intensity for genes on the Affymetrix array. In expression (yellow box in Figures 4a and b) affirming panel a of Figure 3, we show representative results in the the similar positive effect of site strength on gene MV4-11 cell line. The presence of a positive site for expression observed above for absolute expression in a H3K9Me3 (>4-fold IP/input signal and Po0.0001) was single cell line. significantly associated with increased gene expression (Wilcoxon P-value o10À5). To test whether proximity of H3K9Me3 to the TSS was associated with expression Different patterns of H3K9Me3 modification among level, we evaluated the trend between site location and differentially expressed genes log2 intensity using Pearson’s correlation. Panel b Figure 5 provides examples of chromatin scans compar- reveals an increasing trend in expression with proximity ing MV4-11 with D283 (panels a and b) or D283 with to gene TSSs (Po10À5). Correlation analysis shown in Daoy (Panels c and d). As indicated most genes contained panel c also demonstrated a significant association of H3K9Me3 co-localized with other marks located within H3K9Me3 site size with log2 intensity (Pearson’s 1 kb of TSS as single sites. This common result presents a correlation Po10À5). Significant associations were seen relatively simple pattern of modification (panels a and c). in all cell lines tested for each type of analysis shown in Several genes displayed more complex patterns (panels b panels a–c. and d). The MAGEA6 pattern is typical of many

Oncogene Histone 3 lysine 9 trimethylation marks expressed genes JK Wiencke et al 2415 #Genes= 342 327

1.5 12 K9Me3–K9Ac K4Me2–K9Ac K4Me2–K9Me3 10 1.0

8 density

0.5 log2 Intensity 6

4 0.0 No Site within 1kb Sites within 1kb -2 -1 0 1 2 Relative Location (kb)

12

0.0012 10

8 0.0008

Log2 Intensity 6 density 0.0004 4

-2 -1 0 1 0.0000 Distance from TSS (kb)

-1500 -1000 -500 0 500 1000 1500 Site Size (bp) 12 Figure 2 (a) Relative location and (b) relative size of colocalized H3K9Me3, H3K9Ac and H3K9Me2 sites for pairs of modifica- tions in MV4-11. 10 differentially expressed genes (panel a). D283 over- 8 expresses MAGEA6 84-fold compared with MV4-11 and showed large and continuous co-localized acetylation Log2 Intensity and methylation signals around the TSS and extending 6 1 kb downstream. The HOXA9 gene was overexpressed 91-fold in MV4-11 compared with D283 and contained extensive H3K9Ac and three large H3K9Me3 sites 4 extending up to À4 kb. Panel b depicts the consensus TSS;another highly conserved distal promoter is known 500 1000 1500 2000 2500 to exist at about À4.5 kb. High and uniform H3K9Ac Size (bp) 0 signal intensity covered the 5 -region, whereas H3K9Me3 Figure 3 Association of H3K9Me3 sites with gene expression in signals were interrupted by low-intensity regions just MV4-11. (a) Compares the expression level (RMA value, in upstream of the TSS and À1.5 to À2.5 kb. Panel c logarithm base 2 scale) of genes without (N ¼ 342) and with demonstrates another example of co-localized H3K9Me3 (N ¼ 327) H3K9Me3 sites within 1 kb of TSS for a total of 669 genes that are both represented on the tiling and expression arrays. and H3K9Ac around and downstream of the TSS. Here, À5 Wilcoxon P-value o10 (b) The effect of H3K9Me3 site location CD44 is expressed 14-fold greater in Daoy compared and (c) site length on gene expression. The red line is the Lowess with D283. Both marks extend beyond exon 1 of CD44. smoothing curve, which shows a systematic downward trend of GATA-6 in panel d represents a third type of H3K9Me3 expression on the site location as sites go further from the TSS and pattern. The relative expression of GATA-6 is 40-fold in a positive association with site length. The trends are statistically significant (both P-value of Pearson’s correlation 10À5). Daoy compared with D283. Highly significant acetyla- o tion and methylation modifications are evident around

Oncogene Histone 3 lysine 9 trimethylation marks expressed genes JK Wiencke et al 2416 N= 289 30 186 164 extended the regions beyond those on the array and revealed that for HOXA9 and MEIS1 significant 4 H3K9Me3 can be found 1.7 and 2.3 kb downstream each gene, respectively. We assessed in some detail the 2 upstream 50-region of the OTX2 gene. H3K9Me3 extended from the location of the distal promoter 0 (TSSb) through the proximal promoter locations (TSSa -2 À4.8 kb). ChIP-on-chip profiles revealed H3K9Me3 extended over the first exonic regions of OTX2 as well

log2( Daoy / D283 ) log2( Daoy -4 (data not shown). The amplified OTX2 locus contained the largest regions of H3K9Me3 modification we -6 observed in the present study.

Daoy --++ Inverse relationship between promoter DNA D283 --+ + hypermethylation and H3K9Me3 Sites within 1kb We compared our ChIP-on-chip results for H3K9Me3 4 and other marks with DNA hypermethylation status to see if H3K9Me3 was present in promoter regions of 2 methylated and silenced genes. The results for three genes (HOXA9, TMS1 and RASSF1A) are shown in Supple- 0 mentary Table 4. Using quantitative methylation-specific -2 PCR, we found each gene to be hypermethylated in D283 but not methylated in MV4-11. Gene-expression data

log2( Daoy / D283 ) log2( Daoy -4 confirmed that the non-methylated genes were many fold overexpressed in MV4-11 compared with those that were -6 methylated in D283. Also, H3K9Me3 was accompanied by strong H3K9Ac and H3K4Me2 signals in MV4-11. These results are consistent with H3K9Me3 being a mark -1000 0 1000 2000 of active gene expression and not a mark of gene Site Size (bp): Daoy–D283 silencing, at least for these three genes. Figure 4 Association of H3K9Me3 sites with relative gene expres- sion in D283 versus Daoy. (a) compares the relative gene expression (log2 ratio) in Daoy versus D283 stratified by presence of H3K9Me3 sites within 1 kb of TSS in either cell lines. Genes with sites only Discussion in Daoy (N ¼ 30) and genes with sites only in D283 (N ¼ 186) are significantly differentially expressed (Wilcoxon P-value ¼ 0.0003). Our DNA tiling array was designed to interrogate 5 kb of (b) For genes with H3K9Me3 sites within 1 kb of TSS in both cell lines 0 (N ¼ 164), there is a statistically significant association of relative the 5 -region of our candidate genes and 1 kb downstream gene expression with relative site length (Pearson’s correlation ¼ 0.35, of each gene’s TSS. The rationale for this design was that P-value o10À5). The red line is the Lowess smoothing curve. if H3K9Me3 were a suppressive chromatin marking, then it would be detected in the 50-regions of transcriptionally silent genes. In contrast to this situation our major finding is that the H3K9Me3 marking is detectable by the GATA-6 TSS and downstream in the initial 1 kb ChIP-on-chip analysis among many genes expressed in transcribed region, but also evident are upstream sites cancer cells within genomic regions overlapping known that overlap an antisense transcribed gene AK123079. histone activating signals. Using a sliding window The upstream 50 H3K9Me3 sites show local peaks within approach, we found that H3K9Ac and H3K4Me2 sites exons 1 and 2 of AK123079. In addition to CD44, other associated with gene transcription overlapped with sites genes marked by H3K9Me3 and overexpressed in Daoy containing significant H3K9Me3 enrichment in about compared with D283 are several potential therapeutic 80% of cases of positive sites. For genes containing all targets: MAGEA11 (40-fold),MAGEB2(37-fold), CAV1 three markings, we found that modification sites over- and 2 (26-fold) and EGFR (12-fold). lapped within 1 kb of the TSS in about 90% of genes. These results are consistent with the previous assignment QPCR confirms ChIP-on-chip H3K9Me3 presence in of H3K9Ac and H3K4Me2 to 0.5 kb of either side of the HOXA9, MEIS1 and OTX2 TSS of expressed genes (Liang et al., 2004). Our single We confirmed our tiling array experiments by carrying gene ChIP confirmed the tiling array data in revealing out single gene ChIP studies. Figure 6 summarizes our robust H3K9Me3 signals among highly expressed genes comparisons between MV4-11 and D283 for the such as HOXA9, MEIS1 and OTX2 in cells over- HOXA9 (panel a) MEIS1 (panel b) and OTX2 genes expressing these genes. In their first study Vakoc et al. (panel c). We confirmed that H3K9Me3 was increased (2005) reported the strongest H3K9Me3 signals to be around the TSS and downstream region for each gene as 0.5–1.5 kb downstream of the TSS of several genes. A predicted by the ChIP-on-chip results. These studies more recent study emphasized the 30 location of

Oncogene Histone 3 lysine 9 trimethylation marks expressed genes JK Wiencke et al 2417 -1kb TSS 1

MAGEA6 4 D283.K9Me3 ||||||||||||||| | | ||||||| 3 MV.K9Me3 ||| |||||||||||| ||| ||| |||||||||||| || 2 D283.K9Ac | ||| ||||||||||| || | MV.K9Ac 1 0 Site Enrichment log2(IP/Control) 151536.5 151537.0 151537.5 151538.0 151538.5 151539.0 X position (kb)

1 TSS -1kb -2 -3 -4 -5

<<<<< HOXA9

||||||| || | ||||||| ||||||||||||||||||||| ||||||||||||||||||||||||||||||||||||||||||| 3 || |||||| |||| || ||||||||| ||||| ||| ||||||| || |||| ||||| |||||||||| |||||||||||||||||||| ||||||||||||| || ||||| ||||| ||||||||||||| ||||| 2 ||||||| |||||||||| || ||||||| || |||||||||||||||| || || | || || | | ||| |||| 1

Site Enrichment log2(IP/Control) -1 26978 26979 26980 26981 26982 26983 Chromosome 7 position (kb)

-5 -4 -3 -2 -1kb TSS 1

CD44 CD44 D283.K9Me3 | ||| |||||| || Daoy.K9Me3 |||||||||||||||||||| ||||||||||||||| 3 |||| | ||||||||||||||||||||||| D283.K9Ac | ||||||| || Daoy.K9Ac 1

-1 Site Enrichment log2(IP/Control) 35112 35113 35114 35115 35116 35117 35118 Chromosome 11 position (kb)

-5 -4 -3 -2 -1kb TSS 1

<<<<<<<<<<<<<< <<< AK123079 GATA6 ||||||||| 3 | |||||||||| | |||||||||||||||||||||||||| |||||||| || || | ||| 2 |||||| | | 1 0 log2(IP/Control) Site Enrichment -1 17999 18000 18001 18002 18003 18004 Chromosome 18 position (kb)

Figure 5 Examples of patterns of H3K9Me3 and H3K9Ac modifications associated with gene expression. (a): MAGEA6 gene D283 and MV4-11 cell lines;( b) HOXA9 in D283 and MV4-11 cell lines;( c) CD44 in D283 and Daoy cell lines;( d) GATA-6 gene in D283 and Daoy cell lines. Genomic positions shown relative to TSS (position ¼ 0), negative values are upstream (50) and positive values downstream (30) TSS. Shown above each composite tracing as solid bars are CpG island locations. Beginning of the blue line indicates the initial transcribed portion of each gene. Along the abscissa the tick marks show the chromosomal location of tiling array probes; note gaps indicate regions without probes due to repetitive sequence censoring. Histone signal intensity on the ordinate scale is in units of log2-fold enrichment. Within histone signal tracings tick marks show statistically significant sites as defined by >4-fold IP/input ratio and Po0.0001.

Oncogene Histone 3 lysine 9 trimethylation marks expressed genes JK Wiencke et al 2418 3.0 MV-4-11 that H3K9Me3 is abundant within the initial transcribed D283 regions of differentially expressed genes. Interestingly, the 2.5 distribution of histone markings immediately upstream of 2.0 the TSS was suggestive of a local decrease in signal 1.5 intensity for all markings (Figure 1). Such a decrease may result from depletion among gene promoters 1.0

Fold difference Fold undergoing activation (Boeger et al., 2003;Bernstein 0.5 et al., 2004). Our results also indicated that for some highly 0.0 expressed cancer-associated genes such as HOXA9, -1.1 TSS 0.6 1.7 OTX2 and GATA-6 there are additional sites of Distance from TSS (kb) H3K9Me3 within very broad regions of 50 modification. 5.0 These upstream modifications associated with transcrip- MV4-11 tion may represent upstream multiple start sites for gene 4.0 D283 transcription, the presence of different genes in close proximity or other unidentified RNA products (for 3.0 example regulatory RNA). In the case of HOXA9 and 2.0 OTX2 it seems most likely that alternative TSSs could

Fold difference Fold explain the presence of these distal modifications. For 1.0 example, three major HOXA9 transcripts (a9a, a9b, a9T), which are highly conserved between species have 0.0 been identified and the long a9a transcript is regulated -0.8 TSS 0.5 0.7 2.3 28 76 112 136 through a start site approximately 4.5 kb upstream of Distance from TSS (kb) the consensus TSS. Our finding of extended 50 markings 5.0 in HOXA9 is consistent with previous ChIP experiments showing that the MLL fusion protein leads to broadly MV4-11 4.0 distributed histone activation marks (that is , D283 H3K9Ac) across the entire HOXA9 locus encompassing 3.0 all three transcripts (Milne et al., 2005). Our results add H3K9Me3 as being abundant over the HOXA9 locus in 2.0 MLL-positive leukemia cells. The OTX2 locus encodes Fold difference Fold two major isoforms that are regulated by a proximal 1.0 promoter approximately 4.8 kb downstream of a second upstream promoter (Fossat et al., 2005). In mice it has 0.0 been shown that as embryonic neurodevelopment -5.6 TSSa -4.5 -4.1 -3.8 -2.2 -0.3 TSSb 1.4 3.3 proceeds, there is a time-dependent shift toward the Distance from TSS (kb) utilization of the distal promoter (Fossat et al., 2005). Figure 6 Single-gene ChIP results in MV4-11 and D283 cell lines. Two additional TSSs have also been detected with one Quantitative real-time PCR was performed on H3K9Me3 ChIP being upstream about 2.2 kb and an additional proximal and input DNA samples;solid bars (D283), open bars (MV4-11). site (Acampora et al., 2005). Our ChIP results indicated The graphs show fold difference (IP/In) þ standard error for three separate experiments, with each PCR run done in triplicate. Genes strong H3K9Me3 modification across a broad area assayed are: (a) HOXA9;(b) MEIS1;and ( c) OTX2, including two encompassing the predicted initial transcribed portions isoforms noted as TSSa and TSSb. Numbering (x ¼ 0) for OTX2 is associated with both the distal and proximal OTX2 relative to the upstream promoter location TSSb (Fossat et al., promoters;the strongest signals were located in the 2005). region between the two in an area corresponding to the second distal TSS. This may indicate that the amplifica- tion of OTX2 and many fold overexpression of the gene H3K9Me3 markings in the polyA binding protein gene involves the action of multiple OTX2 promoters. ChIP- (Vakoc et al., 2006) and that regions of hyperacetylated on-chip scans demonstrated H3K9Me3 deposited histone H3 were relatively devoid of H3K9Me3. There throughout the exonic regions as well. The GATA-6 was a suggestion from our results in MV4-11 that gene was overexpressed 40-fold in Daoy compared with H3K9Me3 signals may be more 30 compared with D283 and contained strong H3K9Me3 signals around H3K9Ac or H3K4me2 although the significant overlap the TSS as well as 1–2 kb 50-upstream. These upstream of H3K9Me3 with these modifications is the most markings do not correspond to known GATA-6 TSSs prominent feature of our data. There are limitations in but are near the TSS of a newly described antisense localizing these signals with all ChIP methods given the expressed sequence AK123079. Multiple promoter shearing of DNA to about 200–1000 bp. In addition, for usage and closely spaced or antisense transcribed gene ChIP-on-chip studies the tiling array probe spacing and sequences should be considered in the interpretation of the sliding window scoring method also limit resolution. ChIP-on-chip chromatin profiles. Nonetheless, our finding of H3K9Me3 among regions at Previous studies indicated a tight linkage between or up to 1 kb downstream of the TSS strongly indicates CpG island methylation, histone H3 lysine 9 methylation

Oncogene Histone 3 lysine 9 trimethylation marks expressed genes JK Wiencke et al 2419 and transcriptional silencing (Bird, 2002;Richards and regions. Farther upstream signals (that is 2–5 kb) were Elgin, 2002;Kondo et al., 2003, 2004). Reversal of gene detected only rarely, but notably were present among silencing by the demethylating agent 5Aza-dC was very highly expressed genes containing multiple tran- shown to lead to the disappearance of histone H3 lysine scription initiation sites. Second, our results may help to methylation (Nguyen et al., 2002;Kondo et al., 2004; inform efforts to use ChIP-on-chip for tumor classifica- Zhao et al., 2005). Our results could appear to contradict tion. Surveying the chromatin status of important these observations as we found that in D283 the transcriptional regulators such as the homeodomain HOXA9, TMS1 and RASSF1A loci were hypermethy- family of genes could aid in understanding the genesis lated and nonexpressed but we were unable to detect the of specific cancer cell types, in developing new therapies presence of any H3K9Me3. In contrast, these three genes and guiding patient stratification in clinical trials. The are expressed and display strong H3K9Me3 signals in subtype of medulloblastoma characterized by OTX2 the MV4-11 cell line. A possible explanation for these overexpression and extensive chromatin modifications differences is that earlier studies used IP antibodies has recently been shown to be responsive to all trans specific for histone H3K9Me2 in contrast to the present retinoic acid-induced apoptosis (Acampora et al., 2005) work, which focused on H3K9Me3. Thus, while histone whereas nodular/desmoplastic tumor types, which do not H3K9Me2 and H3K9Me3 are dynamic intermediates of overexpress OTX2, are not responsive. In the Daoy cell different methylated states they appear to mark very line, CD44, CAV 1 and 2, EGFR and several MAGE different functional chromatin domains Di- and tri- genes were overexpressed and contained high H3K9Me3 methyl histone lysine 9 were found to reside as stable levels compared with D283;all these genes have been methylation states within different chromatin sites (Wu considered potential therapeutic targets (Chomez et al., et al., 2005). Our study cannot provide further insights 2001;Annabi et al., 2005;Arslan et al., 2006;Jin et al., on this point as we did not carry out ChIP experiments 2006;Tijink et al., 2006). using anti-H3K9Me2-specific antibodies. Future clinical applications of chromatin activity In addition to its physical association with RNA profiling requires a clear definition of the nature of polymerase II, Vakoc et al. (2005) confirmed the histone alterations that signal ‘active’ and ‘repressed’ presence of H3K9Me3 within heterochromatic major chromatin patterns. Our results add new insights into satellite repeats. H3K9Me3 modifications may be unique the complex relationship between histone modifications among modifications in being and gene expression that highlight the difficulty in associated with both heterochromatin and actively assigning chromatin activity to specific markings. The transcribed genes (Eissenberg and Shilatifard, 2006). present work strengthens the link between H3K9Me3 An intriguing aspect of this dual localization is the and the previously established histone lysine marks such association of H3K9Me3 with the heterochromatin as and H3K72Me2 (Miao and Natarajan, binding protein HP1g (Vakoc et al., 2005), which unlike 2005) with actively transcribed genes. Our results HP1a or HP1b is found in euchromatin as well as emphasize the fact that H3K9Me3 is not a suitable heterochromatin (Minc et al., 2000). HP1g is present in marker for classifying genic regions associated with mammals but not in the widely studied fission yeast. transcriptional repression. Recent studies suggest phosphorylation of HP1g at Ser 83 is responsible for targeting the protein to euchromatin and it’s binding to the Ku70 protein (Lomberk et al., Materials and methods 2006). There are no known functional roles for H3K9Me3 and HP1g in transcribing genes although it Cell lines has been suggested that the presence of H3K9Me3 may Cell lines were obtained from the American Type Tissue prevent cryptic promoter elements from becoming Culture Association. D283 and Daoy are derived from aberrantly activated following the passage of RNA medulloblastomas;MV4-11 and RcH are acute leukemic cell polymerase II (Vakoc et al., 2006). Neither the present lines. Cells were cultured in RPMI-1640, 10% fetal calf serum results nor previous reports can establish whether and 5% CO2 at 371C. H3K9Me3 is a ‘normal’ counterpart in the transcription of all genes or whether it may signify an aberrant readout Chromatin immunoprecipitation, amplicon labeling and of the histone code that is more common in cancer cells. hybridization Although nontransformed cells have been found to Cells were harvested, and chromatin was formalin crosslinked exhibit H3K9Me3 marking within actively transcribed and sheared to approximately 200–1000 bp using a Sonic genes, we note here that some differentially expressed Dismembrator model 550 at settings 1.5, for 2–4 cycles 50 s per genes did not contain H3K9Me3 (Figure 4, panel a). cycle on ice. We followed the microarray application protocol Hence, H3K9Me3 may not be universally associated as described in the Chromatin Immunoprecipitation Assay Kit with transcription and further studies are required to (Upstate cat. #17–295) to isolate chromatin DNA complexes. define the generalizability of our observations more We used the EZ-ChIP Kit (Upstate cat no 17–371) for single gene confirmatory ChIP. We used ChIP grade antibodies to globally among normal and malignant cells. H3K9Me3 (Abcam Inc., Cambridge, UK, ab 8898), H3K9Ac Implications of our study are several fold. First, (Abcam, ab4441) and H34Me2 (Upstate, Lake Placid, NY, because most signals were detected within 1 kb of the USA cat no 07–030). The anti-H3K9Me3 antibody was TSS, future tiling array designs that target activating previously shown not to crossreact with H3K4Me2 epitopes histone modifications may be abbreviated to these (Vakoc et al., 2005). Chromatin was also purified from

Oncogene Histone 3 lysine 9 trimethylation marks expressed genes JK Wiencke et al 2420 crosslinked DNA that had not been immunoprecipitated;this probe sets, we chose the most ‘informative’ probe set, as served as an input DNA control for tiling arrays, which was defined by the intensity range across the four cell lines under co-hybridized to provide a reference (that is, total input study to be the representative probe set. amplicon). To prepare samples for array hybridization, we applied the method developed in Bing Ren’s laboratory to Single gene ChIP using quantitative real-time PCR carry out ligation-mediated PCR amplification of ChIP DNA DNA samples from antibody-bound and input chromatin were (Boeger et al., 2003). At least 4 mg of ligation-mediated PCR- analysed by real-time PCR using SYBR Green detection and amplified DNA was used to carry out the subsequent labeling the 7900HT Real-Time PCR System (Applied Biosystems, and hybridization steps. The IP amplicon was labeled with Foster City, CA, USA). Primers were designed to interrogate Cy5, and an equal amount of the total input amplicons labeled several regions around the TSSs of the HOXA9, MEIS1 and with Cy3;the two are combined to hybridize to the tiling array. OTX2 genes. Primer sequences and amplicon locations are Details of the tiling array can be found at NimbleGen.org. available on request. Reaction conditions were used as ChIP on chip microarray design, Gene-expression studies— recommended by the manufacturer and as determined by RNA isolation on Affymetrix Plus 2.0. ChIP on chip experimental optimization (Power SYBR Green master mix, microarray data analysis: normalization probe annotation site 100 or 200 nM of each primer, 0.5 mL template DNA diluted 1:5, detection and characterization;characterization and summary 10 mL reaction volume) with a two-step PCR cycle (1 Â 95 1Cfor of sites can be found in Supplementary Table 6. Normalized 10 min.;40 Â 95 1C for 15 s, 60 1C for 60 s). We calculated the log2 ratios are shown in Supplementary Table 3. fold difference of a target sequence in the immunoprecipitation Correlation between sites and expression. With the modified (IP) fraction versus a fixed amount of the input DNA (In) using site data and matching mRNA expression data, we can test the C C the following expression: Fold increase ¼ 2Àð T;IPÀ T;In Þ.See association of each site with gene expression;669 of the 683 Supplementary Table 2 for raw data and Table 5 for PCR genes have probe sets on the HGU133 plus 2.0 arrays based on primers and amplicon locations for Figure 6. the UCSC Genome Browser mapping and Affymetrix’s probe set annotation, and all site-expression analyses were done for this set of genes. The basic strategy is to compare the relative Quantitative methylation-specific PCR expression changes with two cell lines, such as MV4-11 vs See Supplementary Table 4 for results and Table 5 for methods. D283, in relation to their histone modification patterns such as the presence/absence of H3K9Ac. If H3K9Ac represents an ‘on’ signal for genes, then we would expect to see positive Abbreviations MMV vs D283 ¼ log2(MV/D283) for genes with H3K9Ac sites in MV4-11 but not D283, negative M for genes with MV vs D283 ChIP, chromatin immunoprecipitation;H3K9Me3, histone 3 sites in D283 but not MV4-11, and no effect for genes with lysine 9 trimethylation;MAGE, melanoma antigen gene;TSS, H3K9Ac sites in neither or both cell lines. There are 14 genes transcription start site. with no Affymetrix probe set: BARHL2, FLJ40321, FOXD4L3, FOXD4b, IRX6, LOC120237, LOC257468, MA- GEB10, TGIF2LX, VAX1, WNT3A, FLJ27255, LOC402176 Acknowledgements and LOC494141. Note that only one Affymetrix probe set per gene was chosen for site-expression analysis even though many Grant sponsors: National Institute of Environmental Health genes map to multiple probe sets. For genes with multiple Science NIEHS, ES06717 and ES000002.

References

Acampora D, Annino A, Tuorto F, Puelles E, Lucchesi W, Bird A. (2002). DNA methylation patterns and epigenetic Papalia A et al. (2005). Otx genes in the evolution of the memory. Genes Dev 16: 6–21. vertebrate brain. Brain Res Bull 66: 410–420. Boeger H, Griesenbeck J, Strattan JS, Kornberg RD. (2003). Annabi B, Bouzeghrane M, Moumdjian R, Moghrabi A, unfold completely at a transcriptionally active Beliveau R. (2005). Probing the infiltrating character of promoter. Mol Cell 11: 1587–1598. brain tumors: inhibition of RhoA/ROK-mediated CD44 cell Boon K, Eberhart CG, Riggins GJ. (2005). Genomic surface shedding from glioma cells by the green tea catechin amplification of orthodenticle homologue 2 in medulloblas- EGCg. J Neurochem 94: 906–916. tomas. Cancer Res 65: 703–707. Armstrong SA, Staunton JE, Silverman LB, Pieters R, den Boon K, Edwards JB, Siu IM, Olschner D, Eberhart CG, Boer ML, Minden MD et al. (2002). MLL translocations Marra MA et al. (2003). Comparison of medulloblastoma specify a distinct gene expression profile that distinguishes a and normal neural transcriptomes identifies a restricted set unique leukemia. Nat Genet 30: 41–47. of activated genes. Oncogene 22: 7687–7694. Arslan MA, Kutuk O, Basaga H. (2006). Protein kinases as Canaani E, Nakamura T, Rozovskaia T, Smith ST, Mori T, drug targets in cancer. Curr Cancer Drug Targets 6: 623–634. Croce CM et al. (2004). ALL-1/MLL1, a homologue Azuara V, Perry P, Sauer S, Spivakov M, Jorgensen HF, John of Drosophila trithorax, modifies chromatin and is directly RM et al. (2006). Chromatin signatures of pluripotent cell involved in infant acute leukaemia. Br J Cancer 90: lines. Nat Cell Biol 8: 532–538. 756–760. Bernstein BE, Liu CL, Humphrey EL, Perlstein EO, Schreiber Chomez P, De Backer O, Bertrand M, De Plaen E, Boon T, SL. (2004). Global nucleosome occupancy in yeast. Genome Lucas S. (2001). An overview of the MAGE gene family with Biol 5: R62. the identification of all human members of the family. Bernstein BE, Mikkelsen TS, Xie X, Kamal M, Huebert DJ, Cancer Res 61: 5544–5551. Cuff J et al. (2006). A bivalent chromatin structure marks de Haas T, Oussoren E, Grajkowska W, Perek-Polnik M, key developmental genes in embryonic stem cells. Cell 125: Popovic M, Zadravec-Zaletel L et al. (2006). OTX1 and 315–326. OTX2 expression correlates with the clinicopathologic

Oncogene Histone 3 lysine 9 trimethylation marks expressed genes JK Wiencke et al 2421 classification of medulloblastomas. J Neuropathol Exp Peters AH, Kubicek S, Mechtler K, O’Sullivan RJ, Derijck AA, Neurol 65: 176–186. Perez-Burgos L et al. (2003). Partitioning and plasticity of Eissenberg JC, Shilatifard A. (2006). Leaving a mark: the repressive histone methylation states in mammalian chroma- many footprints of the elongating RNA polymerase II. tin. Mol Cell 12: 1577–1589. Curr Opin Genet Dev 16: 184–190. Rath MF, Munoz E, Ganguly S, Morin F, Shi Q, Klein DC Fossat N, Courtois V, Chatelain G, Brun G, Lamonerie T. et al. (2006). Expression of the Otx2 homeobox gene in the (2005). Alternative usage of Otx2 promoters during mouse developing mammalian brain: embryonic and adult expres- development. Dev Dyn 233: 154–160. sion in the pineal gland. J Neurochem 97: 556–566. Gregory RI, Shiekhattar R. (2004). Chromatin modifiers and Richards EJ, Elgin SC. (2002). Epigenetic codes for hetero- carcinogenesis. Trends Cell Biol 14: 695–702. chromatin formation and silencing: rounding up the usual Heard E, Rougeulle C, Arnaud D, Avner P, Allis CD, Spector suspects. Cell 108: 489–500. DL. (2001). Methylation of histone H3 at Lys-9 is an early mark Roh TY, Cuddapah S, Cui K, Zhao K.. (2006). The genomic on the during X inactivation. Cell 107: 727–738. landscape of histone modifications in human T cells. Proc Jin L, Hope KJ, Zhai Q, Smadja-Joffe F, Dick JE. (2006). Natl Acad Sci USA 103: 15782–15787. Targeting of CD44 eradicates human acute myeloid Rozovskaia T, Feinstein E, Mor O, Foa R, Blechman J, leukemic stem cells. Nat Med 12: 1167–1174. Nakamura T et al. (2001). Upregulation of Meis1 and Kimura J, Suda Y, Kurokawa D, Hossain ZM, Nakamura M, HoxA9 in acute lymphocytic leukemias with the t(4 : 11) Takahashi M et al. (2005). Emx2 and Pax6 function in abnormality. Oncogene 20: 874–878. cooperation with Otx2 and Otx1 to develop caudal forebrain Strahl BD, Allis CD.. (2000). The language of covalent histone primordium that includes future archipallium. J Neurosci 25: modifications. Nature 403: 41–45. 5097–5108. Tijink BM, Buter J, de Bree R, Giaccone G, Lang MS, Staab A Kondo Y, Shen L, Issa JP. (2003). Critical role of histone et al. (2006). A phase I dose escalation study with anti- methylation in tumor suppressor gene silencing in colorectal CD44v6 bivatuzumab mertansine in patients with incurable cancer. Mol Cell Biol 23: 206–215. squamous cell carcinoma of the head and neck or esophagus. Kondo Y, Shen L, Yan PS, Huang TH, Issa JP. (2004). Clin Cancer Res 12: 6064–6072. Chromatin immunoprecipitation microarrays for identifica- Turner BM.. (2000). Histone acetylation and an epigenetic tion of genes silenced by histone H3 lysine 9 methylation. code. Bioessays 22: 836–845. Proc Natl Acad Sci USA 101: 7398–7403. Vakoc CR, Mandat SA, Olenchock BA, Blobel GA. (2005). Liang G, Lin JC, Wei V, Yoo C, Cheng JC, Nguyen CT et al. Histone H3 lysine 9 methylation and HP1gamma are (2004). Distinct localization of histone H3 acetylation and associated with transcription elongation through mamma- H3-K4 methylation to the transcription start sites in the lian chromatin. Mol Cell 19: 381–391. . Proc Natl Acad Sci USA 101: 7357–7362. Vakoc CR, Sachdeva MM, Wang H, Blobel GA. (2006). Lomberk G, Bensi D, Fernandez-Zapico ME, Urrutia R. Profile of histone lysine methylation across transcribed (2006). Evidence for the existence of an HP1-mediated mammalian chromatin. Mol Cell Biol 26: 9185–9195. subcode within the histone code. Nat Cell Biol 8: 407–415. Vernay B, Koch M, Vaccarino F, Briscoe J, Simeone A, Miao F, Natarajan R. (2005). Mapping global histone Kageyama R et al. (2005). Otx2 regulates subtype specifica- methylation patterns in the coding regions of human genes. tion and neurogenesis in the midbrain. J Neurosci 25: Mol Cell Biol 25: 4650–4661. 4856–4867. Milne TA, Martin ME, Brock HW, Slany RK, Hess JL. (2005). Wu R, Terry AV, Singh PB, Gilbert DM. (2005). Differential Leukemogenic MLL fusion proteins bind across a broad subnuclear localization and replication timing of histone H3 region of the Hox a9 locus, promoting transcription and lysine 9 methylation states. Mol Biol Cell 16: 2872–2881. multiple histone modifications. Cancer Res 65: 11367–11374. Yeoh EJ, Ross ME, Shurtleff SA, Williams WK, Patel D, Minc E, Courvalin JC, Buendia B. (2000). HP1gamma Mahfouz R et al. (2002). Classification, subtype discovery, associates with euchromatin and heterochromatin in mam- and prediction of outcome in pediatric acute lymphoblastic malian nuclei and . Cytogenet Cell Genet 90: leukemia by gene expression profiling. Cancer Cell 1: 279–284. 133–143. Nguyen CT, Weisenberger DJ, Velicescu M, Gonzales FA, Lin JC, Zhao W, Soejima H, Higashimoto K, Nakagawachi T, Urano T, Liang G et al. (2002). Histone H3-lysine 9 methylation Kudo S et al. (2005). The essential role of histone H3 Lys9 is associated with aberrant gene silencing in cancer cells and di-methylation and MeCP2 binding in MGMT silencing is rapidly reversed by 5-aza-2’-deoxycytidine. Cancer Res 62: with poor DNA methylation of the promoter CpG island. 6456–6461. J Biochem (Tokyo) 137: 431–440.

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

Oncogene