Discovery of Epigenetically Silenced Genes in Acute Myeloid Leukemias

Discovery of epigenetically silenced genes in acute myeloid leukemias

JC Desmond1, S Raynaud2, E Tung1, W-K Hofmann3, T Haferlach4 and HP Koefﬂer1

1Department of Medicine, Cedars-Sinai Medical Center, Los Angeles, CA, USA; 2Hopital de l’Archet, Unite de cytogenetique des hemopathies malignes, Nice, France; 3Department of Hematology and Oncology, University Hospital Benjamin Franklin, Berlin, Germany and 4Department of Internal Medicine III, Ludwig-Maximilians-University, Munich, Germany

The demethylating 5-aza-20deoxycytidine (DAC) and the histone syndromes (MDSs); and many clinical trials are underway using deacetylase inhibitor (HDACi) suberoyl anilide bishydroxamide either DAC or HDACis to treat AML.9–13 However, no reports (SAHA) possess potent antitumorigenic properties in myeloid have examined their effects on global transcription in myeloid disorders. However, the transcriptome alterations mediated by these drugs are poorly understood. We analyzed the transcrip- disorders. Understanding transcriptome alterations mediated by tional effects of DAC and SAHA in the AML cell line KG-1. these drugs should prove vital in the elucidation of the Microarray analyses revealed 76 genes expressed in normal mechanism(s) through which they mediate cell death. Further- CD34 þ cells, absent in KG-1 cells but whose expression was more, it will facilitate the uncovering of potential tumor induced after drug treatment. A total of 39 of these genes suppressor genes, which have become epigenetically silenced harbored CpG islands in their promoters. We examined the in myeloid disorders, albeit either directly or indirectly. expression level of these genes in 120 AML patient samples representing diverse karyotpyes. Gas2l1, tfIIs, ehd3, enolase 2, To this end, we have examined the relative expression levels mx1, dral, astml and pxdn were diminished across all AML of over 22 000 genes in the AML cell line KG-1. Cells exposed to karyotypes examined. Ehd3 was methylated in 63% of AML DAC and SAHA, as well as untreated populations, were patients examined. This methylation was lost upon complete subjected to a high-throughput microarray analysis of their remission, and not observed in normal CD34 þ cells. CD34 þ transcriptome profiles in an attempt to determine the genetic cells expressed ehd3 at approximately 10-fold higher levels reprogramming responsible for the antitumor properties of than AML samples. Another highlighted gene, a-catenin,is located at q31 of chromosome 5. Analyses of 29 5q- AML/ these drugs. Of the genes identified, we report one in particular myelodysplastic syndrome (MDS) samples revealed marked whose expression was reduced across all subtypes of AML, the decreases in expression of a-catenin, compared to non-5q- MDS eps15 homology (EH) domain containing protein 3 (ehd3). samples (6.679-fold). However, no methylation was detected, The 50 regulatory region of ehd3 contains a CpG island that is suggesting indirect effects of these drugs on the expression of aberrantly methylated in these cancers. Furthermore, we have a-catenin. demonstrated an association of this methylated CpG island Leukemia (2007) 21, 1026–1034. doi:10.1038/sj.leu.2404611; published online 1 March 2007 with MBDs, the effectors of DNA methylation-mediated gene Keywords: AML; Epigenetics; MDS; tumor suppressor genes silencing. Our study has also revealed a gene of potential importance in a specific subtype of myeloid disorders. Loss or deletion of the long arm of chromosome 5 (5q- or del(5q)) are common genetic abnormalities in myeloid malignancies.14 Approximately 10% Introduction of de novo AML and MDS patients present with a 5q deletion.15,16 Here, we identify a gene located within common Promoter regions of many genes contain clusters of CpG deleted regions (CDRs) of 5q31, a-catenin (CTNNA1), whose dinucleotides, referred to as CpG islands. These islands are expression is induced by DAC and SAHA treatment in myeloid generally protected from cytosine methylation.1 Aberrant cell lines. Further analysis of the expression level of a-catenin in methylation of CpG residues in a promoter can lead to the a panel of patients with either a 5q- AML or MDS showed that recruitment of methyl-binding proteins (MBDs). These proteins this gene was significantly decreased in expression when recruit and complex with histone-modifying enzymes, resulting compared to non-5q- AML/MDS samples. in the loss of expression of the associated gene product.2 Many tumor suppressor genes can be silenced through aberrant DNA methylation in acute myeloid leukemias (AMLs). Among the most characterized are the cell-cycle inhibitors p15INK4b and Materials and methods p16INK4a, RARB2, SOCS-1 and CRBP1.3,4 The demethylating 0 agent 5-aza-2 deoxycytidine (DAC) and the histone deacetylase Cells lines and patient samples inhibitor (HDACi) suberoyl anilide bishydroxamide (SAHA) 17 KG-1 cells were grown in Roswell Park Memorial Institute have been shown to possess potent antitumorigenic properties media 1640 with 10% fetal calf serum. Other myeloid leukemia against hematopoietic disorders.5–8 Recently, 5-aza-cytidine, cell lines (HL-60, NB-4, U937, THP1) were obtained from DAC’s parental compound, was approved by the Food and Drug American Type Culture Collection (ATCC, Rockville, MD, USA) Administration (FDA) for the treatment of myelodysplastic and maintained according to their recommendations. Samples of 5q- AML/MDS and non-5q- AML/MDS were obtained Correspondence: Dr HP Koeffler, Department of Medicine, Cedars- from patients with informed consent. Diagnoses were based Sinai Medical Center, Los Angeles, CA 90048, USA. E-mail: [email protected] on the FAB classification (See Supplementary Information on the Received 9 March 2006; revised 13 January 2007; accepted 23 Leukemia website (http://www.nature.com/leu). DNA and RNA January 2007; published online 1 March 2007 were extracted from formalin-fixed pellets basically as described Discovery of epigenetically silenced genes JC Desmond et al 1027 by Coura et al.18 with phenol–cholorform extraction of nucleic Results acid following proteinase K treatment. Determination of genes induced by DAC and SAHA treatment in myeloid leukemia cells Treatment with 5-Aza-CdR and SAHA Using the HG-U133A microarray platform, we identified genes Cell lines were treated with DAC (Sigma, St Louis, MO, USA) upregulated in the AML cell line KG-1 after treatment with the and SAHA (Merck Inc., NJ, USA generous gift from Victoria DNA methylation inhibitor DAC (1 mM, 72 h) and the HDACi Richon). Cells were cultured in the presence of DAC (1 mM) for SAHA (1 mM, 24 h). In an attempt to increase the relevance of the 3 days; and on the second day, SAHA (1 mM) was added. Cells results of the array experiment, we modified a gene selection were harvested after a further 24 h of culture. Mock-treated cells method first described by Yamashita et al.20 Figure 1 depicts the were cultured similarly. selection criteria used to bias the final gene list towards genes whose expression was likely altered through epigenetic mechanisms in leukemia. Of the probe sets examined, 3837 were Microarray analyses differentially expressed between the treated and untreated Total RNA (10 mg) was used as starting material for the cDNA 19 populations by at least twofold, and with a significance of preparation, which was performed as described previously. Po0.05. In keeping with the hypothesis that epigenetically cRNA (15 mg) was fragmented, hybridized to Affymetrix HG- silenced (directly or indirectly) tumor suppressor genes are U133A microarray (Santa Clara, CA, USA), scanned and 19 re-expressed in this experimental model, only probe sets called analyzed as described previously. present in the treated samples (2204) and also absent in the untreated samples (492) were examined further. KG-1 cells Real-time RT-PCR, PCR and COBRA analyses are an early myeloid progenitor cell line, which retains the expression of the CD34 antigen. This cell line, therefore, is most RNA was prepared using RNeasy kits (Qiagen, Valencia, CA, suitable for a comparison to the normal CD34 þ hematopoietic USA) and real-time polymerase chain reaction (PCR) performed cell population. In these normal cells, we would reason that any as described previously.19 Amplification followed a three-step tumor suppressor genes relevant to this stage of hematopoiesis PCR with 10 s denaturation (941C), 10 s annealing (601C), would be expressed. We compared the expression level of the elongation at 651C for 20 s and 5 s at melting temperature. aforementioned 492 genes against HG-U133A microarray PCR and combined bisulphite restriction analysis (COBRA) expression data obtained from three independent normal analyses of bisulphite-treated DNA were performed basically as CD34 þ cell populations.21 A total of 95 of the 492 probe sets described previously,19 but using the enzyme BstU1. All primer sequences are available upon request.

Bisulphite sequencing KG1 KG1 + DAC untreated Affymetrix HG-U133A Bisulphite modiﬁcation of DNA was performed using the EZ + SAHA DNA Methylation Kit according to the manufacturer’s recom- Microarray mendations (Zymo Research, Orange, CA, USA). PCR products 22,000 probe sets were subcloned into pGEM-T Easy (Promega, Madison, WI, USA) for sequence analysis. At least six independent clones were sequenced in all samples analyzed. All primer sequences are available upon request. Differentially expressed P<0.05 3,837 Western blot analysis Whole-cell lysates were prepared, resolved by sodium dodecyl sulfate polyacrylamide gel electrophoresis, transferred to poly- Present in treated cells vinylidene membrane (Immobilon, Amersham Corp., Arlington 2,204 Heights, IL, USA) and visualized as described previously.18 Membranes were probed with antibodies against a-catenin, Absent in control cells b-catenin and glyceraldehyde-3-phosphate dehydrogenase 492 (Santa Cruz Biotechnology Inc., Santa Cruz, CA, USA).

Expressed in normal CD34+ cells ChIP assay 95 Chromatin immunoprecipitations (ChIP) experiments were preformed as described previously,19 using the ChIP assay kit (Upstate Inc., Chicago, IL, USA). Prepared lysates were incuba- Genes with promoter containing 1 m ted overnight, at 4 C, with 5 g of antiacetylated lysine anti- a CpG island body (Upstate, Charlottesville, VA, USA), anti-MeCP2 antibody (Upstate), anti-MBD2/3 antibody (Upstate), anti-HA antibody 39 (Santa Cruz, CA, USA) or normal rabbit serum. Immunopreci- pitated DNA was ultimately recovered by phenol/chloroform Figure 1 Schematic representation of experimental design. mRNA extraction and ethanol precipitation and analyzed by PCR and was isolated from triplicate cultures of KG1 cells either treated with DAC (1 mM, 3 days) and SAHA (1 mM, ﬁnal 24 h), as described in gel electrophoresis. The optimal reaction conditions for PCR Materials and methods. Labelled cRNA was prepared and hybridized were determined for each primer pair (sequences available upon to HG-U133A microarrays. The gene selection procedure, and the request). number of genes meeting each criteria, are depicted in the ﬂow chart.

Leukemia Discovery of epigenetically silenced genes JC Desmond et al 1028 were expressed in normal CD34 þ cells, of which 39 were Identification of methylation through bisulphite genes that contained CpG islands in their 50 regulatory regions sequencing and methyl-binding protein association (http://genome.ucsc.edu/). The expression levels of a subset in the 50 regulatory regions of target genes of these 39 genes in the treated and untreated KG-1 cell Bisulphite sequencing analyses of the 50 regulatory regions of a populations were validated by real-time RT-PCR, which con- number of promising genes in Table 1 were performed using firmed the microarray data (data not shown). genomic DNA of KG-1 cells in order to detect the presence of To complete the paradigm, we reasoned that a tumor aberrant methylation. Multiple clones (at least six) were suppressor gene would also not be expressed in primary sequenced for the promoter CpG islands of DRAL, ASTML, Id2 leukemias. We cross-referenced the expression status of and enolase 2. Unexpectedly, no CpG methylation was detected these 39 genes against a panel of primary AML patient in any of these gene promoters, despite the stringent selection samples for which we had access to HG-U133A expression criteria imposed in the original (data not shown) screen. These data. 20 samples of each of the following AML karyotypes findings were not just a KG-1 cell line phenomenon, as were analyzed: t,(8;21) t,(15:17) inv,16 11q23/MLL, complex bisulphite sequencing for these CpG islands was performed on and normal karyotype. The results are described in Table 1. Of three AML patient genomic DNA samples with the same result special note, expression of eight genes was markedly diminished (data not shown). in a subset of patients across all AML karyotypes examined: These data made clear that DAC and SAHA treatment was GAS2L1, TFIIS, EHD3, enolase 2, MXA, DRAL, ASTML and affecting the transcription of many genes irrespective of their PXDN. These represent strong candidates for tumor suppressor methylation status. We developed a ChIP technique to screen genes in AML. for methylated genes in our list of tumor suppressor candidates.

Table 1 Expression of candidate tumor suppressor genes in AML patient samples

Gene KG1 GenBank ID t(8;21) t(15;17) inv(16) 11q23 Complex Normal % –ve % –ve % –ve % –ve karyotype karyotype % –ve % –ve

CD9 35.8 NM_001769 75 5 10 10 Tubulin, beta 20.9 NM_001069 5 5 Annexin A4 13.8 BC000182 10 5 10 5 10 Agrin 13.5 AF016903 20 10 15 10 HSP27 12.5 NM_001540 5 15 10 5 Cathepsin Z 11.2 AF073890 5 5 10 5 10 SGK 10.5 NM_005627 5 GAR22 8.3 Y07846 25 35 25 H1.0 6.2 BC000145 5 TFIIS 6.1 NM_003195 55 40 60 100 75 70 EHD3 5.8 NM_014600 55 45 75 20 60 35 IFI30 5.7 NM_006332 15 10 Enolase 2 5.4 NM_001975 90 50 100 85 60 75 ATP1B1 5.4 NM_001677 25 30 5 FXYD6 5.1 NM_022003 25 55 25 20 Synaptogyrin 4.7 BC000731 5 Id2 4.6 NM_002166 5 FKH1 4.6 NM_002015 60 25 25 RGC32 3.6 NM_014059 D2S448 3.6 AF200348 25 5 60 45 50 MXA 3.6 NM_002462 40 80 25 65 10 20 DRAL 3.5 NM_001450 95 75 100 95 35 75 RGS2 3.2 NM_002923 PDE4B 3 NM_002600 Cystatin C 2.9 NM_000099 5 5 CTNNA1 2.8 D14705 Cathepsin H 2.5 NM_004390 5 ASTML 2.5 BC002508 50 10 65 60 65 25 ZNF163 2.4 NM_005263 510 Annexin A2 2.2 BC001388 Calpain 2 2.1 M23254 Caspase 7 2.1 NM_001227 20 10 15 10 5 H1.2 2.1 BC002649 Annexin A5 2.1 NM_001154 15 5 15 LPA2 2.1 NM_001401 10 P52 2.1 AF098482 5 Supervillin 2.1 NM_003174 Annexin A6 2 NM_001155 15 5 S100A10 2 NM_002966 5 Gene names and GenBank identiﬁers for the 39 genes selected from the microarray analysis. Microarray reported fold increase seen in KG1 cells after DAC and SAHA treatment is shown in the column headed KG1. Expression analysis of each gene in 20 patients samples for each AML karyotype shown. Results are described as percentage of samples deemed not to be expressing the gene, according to microarray analysis. The lack of a number correlates to all samples being scored positive for expression of that candidate gene.

Leukemia Discovery of epigenetically silenced genes JC Desmond et al 1029 (61% cytosines methylated per clone) and patient No. 3 had a low level (4% per clone). eCP2 ntibody The restriction enzyme BstU1 recognizes and cuts DNA at PCR blankNo a Anti HAAnti-acetyllysineAnti MBD2/3Anti M CGCG sites. The CpG island of ehd3 contains four such BstU1 p15 restriction sites (Figure 3a, black boxes), making it possible to screen for methylation in this island using COBRA. COBRA TFIIS2 analysis of this promoter region was validated using these five samples. Figure 3b shows representative restriction digests of EHD3 ehd3 promoter PCR products from bisulphite-treated genomic Id2 DNA. The KG-1 PCR product was completely digested by BstU1, as the sequencing results predicted. The product from MXA normal CD34 þ cells was not digested. Interestingly for the AML Figure 2 ChIP analyses of methyl-binding protein association with patient samples, the amount of digested PCR product appeared the 50 regulatory regions of identified target genes. KG1 cells were to correlate with the level of methylation observed in the subjected to crosslinking with formaldehyde and chromatin fractions sequencing results with no digestion observed in the product were prepared by sonication. DNA was purified from immunopreci- from patient No. 3, slight digestion in the sample from patient pitations with antibodies against acetylated lysine, MBD 2/3 and No. 1 and robust digestion of the PCR product derived from MeCP2, as described in Materials and methods. Anti-HA antibody and no antibody ChIPs were performed as negative controls. DNA eluates patient No. 2’s DNA (Figure 3a and b). To confirm further the were analyzed with primers designed to the CpG islands of the target sensitivity of the COBRA analysis and its dependence on a genes. A total of 35 cycles of PCR were performed in all cases. PCR methylated starting template, the acute promyelocytic leukemia- products were analyzed by agarose gel (2%) electrophoresis. derived cell line NB4 was cultured in the presence and absence of DAC for 72 h, and COBRA analysis was subsequently performed on the ehd3 promoter region. Figure 3c demonstrates ChIPs were performed on KG-1 cells using antibodies raised that this CpG island is also methylated in NB4 cells, and that this against the DNA methyl-binding protein MeCP2, as well as methylation is reduced upon treatment of the cells with DAC. MBD2 and MBD3 (MBD2/3). These proteins are known to COBRA was then used to screen a larger population of bind methylated CpGs in the promoter regions of genes and primary AML samples for ehd3 methylation. Figure 3d shows transduce these repressive marks into silencing of the gene representative BstU1 digests of ehd3 promoter region after PCR through the recruitment of histone-modifying enzymes.22 products were amplified from bisulphite-treated DNA of patient Figure 2 shows the result of a screen on a selection of candidate samples. Eight out of 13 patient samples screened solely by genes. All ChIPs with the control anti-HA antibody were COBRA, were positive for ehd3 CpG island methylation. negative. An anti-acetyl lysine antibody was used as a positive Interestingly, we obtained complete remission (CR) samples control. PCR amplicons corresponded to the CpG island in the from patients No. 4 and No. 5, which showed no evidence promoter regions of the target genes. The p15 tumor suppressor of methylation even though the PCR products from both the gene has been shown to be methylated in 75% of AML samples corresponding disease samples were digested by BstU1 (methy- and to exist as a functionally repressive, variegated promoter lated gene) (Figure 3d). Overall, 10 out of 16 (63%) of the methylation in KG-1 cells. Here, we validate our ChIP approach primary AML samples showed ehd3 promoter methylation, by demonstrating an association of the p15 promoter with both whereas no methylation was detected in a purified normal the MBD2/3 and MeCP2 proteins. Both tfiis2 and id2 were CD34 cell population or in two disease-free remission samples. associated with the transcriptionally permissive chromatin mark and not bound by MBDs. In the case of id2, this was expected as we had already demonstrated lack of methylation in its CpG AML patient samples express low levels of EHD3 as promoter island (data not shown). However, the promoter CpG compared to CD34 þ cells from healthy donors islands of mxa and ehd3 were associated with the MBDs. In the Real-time RT-PCR was performed to confirm decreased expres- case of mxa, the association was weak, but reproducible. Ehd3 sion of ehd3 mRNA in AML samples. cDNA obtained from 15 promoter binding to MBD2/3 and MeCP2 was very robust, and AML patients and five normal CD34 þ cell populations were this gene was chosen for further study. compared and representative data are shown in Figure 4. Relative to KG-1 cells, the mean level of ehd3 expression across the 15 AML samples was 1.971.7 compared to 20.474 in the normal CD34 þ cell fractions (Po0.005). EHD3 is methylated in AML samples. The CpG island in the 50 regulatory region of ehd3 was analyzed in KG-1 cells, normal CD34 þ cells and three AML patient a-Catenin is induced by DAC and SAHA treatment samples by bisulphite sequencing. Genomic DNA was isolated Although not absent in any of the AML samples investigated and modified through sodium bisulphite treatment. Figure 3a (Table 1), a-catenin (CTNNA1) is located on chromosome 5, in depicts the sequencing results of five distinct clones for each the commonly deleted region (CDR) of 5q- MDS and AML. cell type. All CpG dinucleotides analyzed in KG-1 cells were Several groups have delineated a CDR on chromosome 5, but a methylated at the cytosine residue, as represented by black gene contributing to these disorders in this region has yet to circles, confirming the inference of the ChIP experiments been identified.23,24 We decided to characterize the methyla- (Figures 2 and 3a). Furthermore, no methylation was found in tion and expression status of a-catenin in 5q- MDS and AML any of the clones from genomic DNA of normal CD34 þ cells, samples. The increase in a-catenin mRNA expression in as represented by open cirlces. Of the three AML samples, response to DAC and SAHA treatment was confirmed through patient No. 1 showed an intermediate level of methylated real-time RT-PCR analysis. Figure 5a shows an increase in cytosine residues with an average of 34% CpGs methylated per a-catenin transcript levels in all four myeloid cell lines. U937 clone. Patient No. 2 displayed a higher level of methylation cells showed the greatest elevation of a-catenin expression, with

Leukemia Discovery of epigenetically silenced genes JC Desmond et al 1030 abBstU1

KG1 cell line

Normal CD34+(1/4)

AML patient 1

AML patient 2

AML patient 3

c NB4 cells DAC

d Patient 4 Patient 5 D CR DCR Patient 6 Patient 7 Patient 8 CD34+ BstU1 ++ + + + + + +

Patient 9 Patient 10 Patient 11 Patient 12 Patient 13 Patient 14 Patient 15 Patient 16 BstU1 ++ + ++ + + +

Figure 3 Methylation analysis of ehd3.(a) The methylation status of 30-CpG sites in the promoter region of ehd3 was determined in KG-1 cells, normal CD34 þ cells and three AML patient samples. Approximate spacing of the CpG dinucleotides is shown. The analyzed area spans 228 bp located at chr2:31368970-31369198. Methylation was determined by bisulfite sequencing. K, methylated CpG; J, unmethylated CpG. (b) COBRA of the same genomic region. PCR products of bisulfite-treated DNAs of the same five samples were digested with Bstu1. Black boxes in A represent Bstu1 recognition sites (CGCG). (c) PCR products of bisulfite-treated DNA from NB4 cells cultured either with or without DAC for 72 h, were digested with Bstu1. (d) PCR products of bisulfite-treated DNAs of a further 12 AML patient samples were digested with Bstu1 ( þ ). For patients No. 4 and No. 5, CR samples were analyzed alongside their leukemic samples (D). Normal CD34 þ cells were used as a negative control.

30 DAC and SAHA increasing mRNA levels by 4.270.2-fold. Combined drug treated KG-1 cells exhibited a 2.370.2-fold 25 P<0.005 increase over untreated control cells mimicking the 2.8-fold increase observed in the microarray analysis (Figure 5a and Table 1). Importantly, Western blot studies showed that this 20 increase of a-catenin in KG-1 cells was also observed at the protein level (Figure 5b). Bisulphite PCR analyses were 15 performed on the CpG island in the promoter region of a-catenin using genomic DNA isolated from KG-1 and normal CD34 þ 10 cells. No methylated cytosines were found in any of the 100-

Levels of EHD3 mRNA Levels CpG dinucleotides examined in this island (ﬁve clones 5 sequenced for each cell type). Furthermore, no methylation was detected in three 5q- MDS and three 5q- AML patient 0 samples (data not shown). 123456789101112131415 12345 AML normal CD34+ a-Catenin expression is markedly diminished in Figure 4 Expression of ehd3 in AML patient samples. Real-time 5q- MDS and AML quantitative RT-PCR analysis of ehd3 mRNA expression in AML patient samples and CD34 þ cells of normal individuals. Results are Real-time RT-PCR analyses of a-catenin transcript levels were represented as levels of mRNA relative to the mean of the AML patient performed on 29 5q- MDS and AML patient samples as samples (1.0). Unpaired Student’s t-test used for statistical compar- compared to transcript levels in 21 non-5q- MDS and AML isons. Real-time PCR was performed in triplicate. samples. In Figure 6a results are graphed as relative to the mean

Leukemia Discovery of epigenetically silenced genes JC Desmond et al 1031 of the 5q- MDS and AML samples (171.7). Non-5q- MDS/AML a 5 samples showed variable a-catenin expression levels; but as a 4 population, the expression was signiﬁcantly higher than that observed in the 5q- samples. The mean increase in a-catenin 3 transcript levels in the non-5q- MDS/AML samples was 6.5779.3 fold greater than those in the 5q- MDS/AML samples 2 (Po0.01). Figure 6b depicts these results in the form of a dot plot. The expression levels of genes in the immediate vicinity of 1 a-catenin on chromosome 5 were also examined in these same samples. A sample was scored positive for gene expression if CTNNA1mRNA expression 0 DAC + SAHA a PCR product was observed at 25 cycles of RT-PCR. Table 2 KG1 U937 NB4 THP1 summarizes the results of these reactions. Sample concentra- tions were normalized by RT-PCR, and b-actin expression. In general, deletion of 5q- chromosomal material resulted in a 1.5-fold decrease in samples expressing these genes at 25 cycles b of RT-PCR. However, only one of 29 (3.5%) samples was CTNNA1 positive for expression of the a-catenin gene in the 5q- patient group, as compared to six out of 16 (38%) in the non-5q- sample Control DAC + set. This represents a 10.7-fold decrease in expression between SAHA the two groups, which is approximately sevenfold less than any Figure 5 Expression of a-catenin in myeloid leukemia cell lines of the neighboring genes on chromosome 5 (Table 2). These treated with SAHA and DAC. (a) The mRNA expression levels of a- results suggest that a-catenin is silenced in the 5q- samples to catenin were analyzed in four myeloid cell lines by real-time RT-PCR, a level that exceeds what would be expected by haplo- before and after exposure to DAC (1 mM, 72 h) and SAHA (1 mM, 24 h). insufﬁciency alone. Expression levels for each cell line are represented as relative to its untreated control. Real-time PCR was performed in triplicate. (b) Western blot analysis was performed on KG-1 cells before and after treatment with DAC (1 mM, 72 h) and SAHA (1 mM, 24 h). Blots were Discussion probed with an anti-a-catenin antibody. Blot is representative of three independent experiments. Epigenetic alterations play a crucial role in the silencing of genes. Hypermethylation of cytosine residues in the CpG islands of gene promoters has been shown to result in the silencing of

a 34 24 14

2 5q-MDS and AML samples 1 Expression relative to the mean of

0 5q-MDS / AML Non 5q-MDS / AML

b P<0.01 30 20 10 6 5 4 3 2 1 CTNNAI mRNA expression relative to normal CD34+ cells 5q- Non 5q- MDS/AML MDS/AML

Figure 6 Expression of a-catenin in MDS/AML patient samples. (a) Real-time quantitative RT-PCR analysis of a-catenin mRNA expression in 5q- MDS and AML patient samples and in samples of non-5q- MDS and AML patients. Results are represented as levels of mRNA relative to the mean of levels of a-catenin expression observed in the 5q- MDS and AML samples (1.0, dotted line). Real-time PCR was performed in triplicate. (b) Data represented as a dot plot. Unpaired Student’s t-test used for statistical comparisons.

Leukemia Discovery of epigenetically silenced genes JC Desmond et al 1032 Table 2 a-catenin expression is more repressed than its neighboring genes

Gene Expression in non-5q- Expression in 5q- Fold decrease in number of samples MDS/AML (%) MDS/AML (%) expressing gene with loss of 5q

Paip2 75 51.7 1.45 Matr3 87.5 51.7 1.69 Sil 1 56.3 41.4 1.36 Ctnna1 37.5 3.5 10.7 Hspa9b 81.3 58.6 1.39 Etf1 0 0 F Egr1 0 0 F Cdc25c 43.8 58.6 0.75 Percentage of both 16 non-5q- AML/MDS and 29 5q- AML/MDS samples scored positive for expression of each gene is recorded. Fold difference in expression levels are also shown. Positive expression is determined after 25 cycles of PCR analysis. Equal amounts of cDNA were used for each sample. Using real-time RT-PCR, expression of b-actin was determined and samples adjusted accordingly.

gene transcription. The elucidation of genes epigenetically obtained from 120 AML patients. Six groups of 20 patients were silenced in cancers is dramatically advancing our understanding selected for screening, based upon the overriding karyotype of of these diseases and facilitating the search for new treatment their disease. Eight genes were noted for their low expression regimes. Although the list of known methylated genes asso- levels across the spectrum of AML karyotypes examined: ciated with human leukemias is growing,3 many epigenetically gas2l1, tfIIs, ehd3, enolase 2, mxA (mx1), dral, astml and pxdn. silenced tumor suppressor genes probably remain undiscovered Furthermore, there were genes whose expression was markedly in these disorders. diminished in specific AML subsets. For example, cd9 was Drugs that are able to reverse these epigenetic modifications absent in 75% of cases of t(8;21) AMLs, and fh1 and fxyd6 were are enjoying notable success in the treatment of these disorders. absent in 60 and 55% of samples of t(15;17)s, respectively. Specifically, the demethylating agents 5-aza-cytidine (Vidaza) D2s488 was not expressed in 60% of inv(16)s, 50% of AMLs and DAC (Dacogen) and the HDACi SAHA have been shown to with normal karyotype and 45% of AMLs with MLL transloca- possess potent antitumorigenic properties against hematopoietic tions (11q23). Although beyond the scope of this study, these disorders.5–8 Both 5-aza-cytidine and DAC have been approved genes should be considered for more focused analyses in these by the FDA for use in MDS. Furthermore DAC has undergone specific AML subtypes. successful MDS trials in Europe.10–13,25 The successes of these Surprisingly, bisulphite sequencing analyses did not detect drugs intuitively points to their ability to remove transcription- any methylation in any samples for the promoter regions of dral, ally repressive marks from chromatin in the vicinity of tumor astml, enolase 2 or id2; id2 chosen for further study owing to its suppressor genes. However, these drugs are also likely to elicit a known involvement in myeloid development.26 These findings wide range of effects upon transcription. supported our initial supposition that these treatments would In this study, we attempted to uncover novel tumor suppressor have wide-ranging effects on gene transcription, not limited to genes through their activation by DAC and SAHA by modifying genes directly epigenetically silenced. Similar findings have an approach used by Yamashita et al.20 Importantly, although been reported elsewhere.27 Furthermore, as the drugs were used we utilized several gene selection steps designed to uncover in combination, it is possible that DAC does not play a role in genes directly methylated in this disorders, this did not preclude the activation of these genes. our characterization of any potential tumor suppressor whose Methylated-CpG dinucleotides associate with MBDs, which activation appeared independent of its DNA methylation status. in turn, modulate transcription through the recruitment of Using the Affymetrix HG-U133A microarray platform, the histone-modifying enzymes.2 We reasoned that any of our relative expression levels of over 22 000 genes were compared candidate genes harboring functionally relevant methylation in between untreated KG-1 cell populations and KG-1 cells treated their promoter regions would be associated with one, or more of with the DNA demethylating drug, DAC, and the HDACi SAHA. these proteins. ChIPs were performed on KG-1 (untreated) cells Stringent statistical analyses performed on the resulting data using antibodies raised to the three MBDs most commonly revealed 492 probe sets that represented genes ordinarily associated with methyl-CpG binding in humans, MBD2 and silenced in KG-1 cells, but whose transcription was induced MBD3 (dual specific antibody) and MeCP2. Of the candidate following the reversal of the epigenetic phenotype of the genes examined, the most striking association was found cells. We reasoned that any tumor suppressor genes would, with the promoter region of ehd3. In this assay, ehd3 promoter by definition, be expressed in normal CD34 þ cells. Using DNA was found enriched in both the MBD2/3 and MeCP2 microarray analyses, we determined the expression status of immunoprecipitations, suggesting robust methylation of this these 492 probe sets in three CD34 þ cell populations purified promoter in KG-1 cells. from the bone marrow of three independent healthy donors. A The CpG island of ehd3 was amplified and clonal sequencing total of 95 of the probe sets were expressed in the normal cell was performed. Confirming the ChIP experiment, the ehd3 populations, 39 of which were genes that harbored 50 regulatory promoter of KG-1 cells was completely methylated. Strikingly, CpG islands. It should be noted that the treatment of KG-1 cells the same region was completely unmethylated in the normal with this drug combination also lead to significant number of CD34 þ cell population. Furthermore, methylation of the ehd3 probe sets being downregulated, as compared to untreated promoter was found in all three AML patient samples, to varying control cell populations. This phenomenon, beyond the remit degrees. Using COBRA analysis, 13 more AML patient samples of this study, merits further study. were analyzed for ehd3 methylation. Methylation was observed A thorough analysis of the expression status of these genes in in eight out of 13 of these patients. Moreover, two patients leukemia patient samples was performed using microarray data showed methylation in samples taken during active disease, but

Leukemia Discovery of epigenetically silenced genes JC Desmond et al 1033 none was observed in samples taken from the same individuals augmented. Bisulphite PCR analyses were performed on the after they had entered CR. This implied that the methylation of CpG island located in the promoter region of a-catenin. Using ehd3 was restricted to the cancerous cells in these patients and genomic DNA isolated from KG-1 and normal CD34 þ cells, not present in their normal cells. In summary, across the none of the cytosines were found to be methylated in any of the different detection methods, 16 primary AML samples were 100 CpG dinucleotides examined (five clones sequenced for analyzed and methylation of the ehd3 promoter was observed in each cell type). Furthermore, no methylation was detected in 10 patients (63%). All normal samples, either purified normal three 5q- MDS and three 5q- AML patient samples (data not CD34 þ cells or normal patient blood samples, were negative shown). Real-time RT-PCR analyses of a-catenin transcript levels for ehd3 methylation. Furthermore, methylation of this promoter were performed on 29 5q- MDS and AML patient samples. in NB4 cells could be reversed by 72 h DAC treatment, as These results were compared to observed transcript levels in 21 analyzed by COBRA (Figure 3c). non-5q- MDS and AML samples. Non-5q- MDS / AML samples Expression levels of ehd3 mRNA were examined in a separate showed variable a-catenin expression levels; but as a popula- panel of AML patient samples. Real-time RT-PCR revealed tion, their expression was significantly higher (6.5779.3 fold) expression levels of ehd3 in AML blast cells to be dramatically than that observed in the 5q- MDS/AML samples (Po0.01). lower than those observed in normal CD34 þ cells. Normal Furthermore, expression of genes in the immediate vicinity of a- CD34 þ cells expressed ehd3 transcripts at approximately catenin was not diminished to the same degree in the 5q- MDS 10-fold higher levels than those observed in the AML samples versus non-5q- MDS samples, suggesting that a-catenin is (Po0.005). silenced to a level that exceeds what would be expected by These data clearly show that ehd3, a gene expressed in haploinsufficiency alone in the 5q- samples. Taken together, normal CD34 þ cells, is methylated and repressed, in a signi- these results imply an indirect effect of DAC and SAHA on the ficant proportion of AML patient’s blast cells. Ehd3 thus expression of the a-catenin gene. represents a prime candidate as a novel tumor suppressor gene a-Catenin is part of the adheren junction protein complex in in this disease. Biologically, EHD3 (EH domain containing mammalian cells. It binds and links b-catenin to the cellular protein 3) is involved in the shuttling of endocytic vesicles. The cytoskeleton.36 Its role in stabilizing this complex may point to a protein appears to be a feature of the endocytic recycling mechanism whereby loss of a-catenin protein could promote compartment (ERC), as it has been shown to colocalize with either cellular growth or transformation. By destabilizing transferrin containing recycling vesicles.28,29 A recent study has b-catenin at the cell surface, loss of a-catenin could promote demonstrated that RNAi-mediated knockdown of EHD3 in HeLa its nuclear translocation and the subsequent activation of cells results in the failure of internalized transferrin to localize to oncogenic b-catenin target genes.37 However, our experiments the ERC, instead remaining contained in large, peripheral showed no changes in b-catenin localization in myeloid cell organelles.30 In the same study, the authors note recent lines upon treatment with DAC or SAHA (data not shown). evidence implying that blocking vesicle trafficking to the ERC Furthermore, no changes in b-catenin-mediated transcription may, in fact, increase the rate of recycling. This is thought to were observed in U937 cells transfected with the TOP and FOP occur through a ‘fast recycling’ pathway.31 It is unclear as to the FLASH TCF/LEF b-catenin reporter constructs (data not shown). effects of loss of EHD3 in myeloid cells, but it is interesting to Recently, a-catenin has been reported to play a role in note various studies that show an involvement of transferrin symmetry maintenance in stem cells.38 Loss of a-catenin in and/or the transferrin receptor levels in the pathogenesis of primitive CD34 þ myeloid progenitor cells could possibly lead AMLs.32,33 Furthermore, the elucidation of other cargo con- to asymmetrical cell division, resulting in a dysplastic myeloid tained within EHD3-associated endosomes may provide crucial cell population. insights into the pathogenesis of various AMLs. In summary, the use of drugs engineered to reverse epigenetic Loss or deletions of the long arm of chromosome 5 (-5/5q-) are phenotype in myeloid cells results in the activation of genes common genetic abnormalities in myeloid malignancies.14 The independent of their CpG-island methylation status. We have most frequent occurence of 5q- is in secondary, therapy-related uncovered a group of genes expressed in normal CD34 þ cells, leukemias and MDSs (tAML, tMDS), where deletion is observed and whose transcription is reinduced upon DAC and SAHA in over 43% of cases.16 5q- leukemia is typified by poor treatment in KG1 cells. Our results identify ehd3, as a gene response to chemotherapy and CR rates of under 30%.34 whose expression is markedly dimished in a host of AMLs, a-Catenin is located on chromosome 5, in the CDR of 5q- seemingly through the hypermethylation of the CpG island in its MDS and AML. Several groups have delineated a CDR on promoter region. In addition, we also identify a-catenin as a chromosome 5, but a gene contributing to these disorders in this gene deregulated through an epigenetic event(s) in 5q- related region has yet to be identified.15,24,35 As a-catenin was on our myeloid disorders. list of 39 genes, we decided to characterize the methylation and expression status of a-catenin in 5q- MDS and AML samples. Given the lack of a good 5q- cell model system, myeloid Acknowledgements leukemia cell models systems were used. Both a-catenin mRNA and protein expression levels increased in response to DAC and This work was supported in part by NIH grants, Inger Fund, Parker SAHA treatment. a-Catenin transcript levels increased in all Hughes Trust and C&H Koeffler Foundation. Dedicated to the four myeloid cell lines examined. mRNA levels in U937 cells memory of David Golde, a friend and mentor. HPK holds the increased by 4.270.2-fold. Combined drug-treated KG-1 cells Mark Goodson Chair of Oncology Research at Cedars-Sinai exhibited a 2.370.2-fold increase over untreated control cells Medical Center and is a member of the Jonsson Comprehensive mimicking the 2.8-fold increase observed in the microarray Cancer Center and the Molecular Biology Institute, UCLA. analysis. Western blot analyses demonstrated that the increase of a-catenin seen in KG-1 cells was also observed at the protein References level. Moreover, a-catenin protein was undetectable in the untreated KG-1 cells, implying that gene transcription is turned 1 Jones PA. The DNA methylation paradox. Trends Genet 1999; 15: on by DAC and SAHA at this locus, rather than merely 34–37.

Leukemia Discovery of epigenetically silenced genes JC Desmond et al 1034 2 Newell-Price J, Clark AJ, King P. DNA methylation and silencing of 21 Komor M, Guller S, Baldus CD, de Vos S, Hoelzer D, Ottmann OG gene expression. Trends Endocrinol Metab 2000; 11: 142–148. et al. Transcriptional profiling of human hematopoiesis during 3 Esteller M. Profiling aberrant DNA methylation in hematologic in vitro lineage-specific differentiation. Stem Cells 2005; 23: neoplasms: a view from the tip of the iceberg. Clin Immunol 2003; 1154–1169. 109: 80–88. 22 Hendrich B, Tweedie S. The methyl-CpG binding domain and the 4 Teofili L, Martini M, Luongo M, Diverio D, Capelli G, Breccia M evolving role of DNA methylation in animals. Trends Genet 2003; et al. Hypermethylation of GpG islands in the promoter region of 19: 269–277. p15(INK4b) in acute promyelocytic leukemia represses 23 Dodge JE, Munson C, List AF. KG-1 and KG-1a model the p15 CpG p15(INK4b) expression and correlates with poor prognosis. island methylation observed in acute myeloid leukemia patients. Leukemia 2003; 17: 919–924. Leuk Res 2001; 25: 917–925. 5 Guo Y, Engelhardt M, Wider D, Abdelkarim M, Lubbert M. Effects 24 Zhao N, Stoffel A, Wang PW, Eisenbart JD, Espinosa 3rd R, Larson of 5-aza-20-deoxycytidine on proliferation, differentiation and p15/ RA et al. Molecular delineation of the smallest commonly deleted INK4b regulation of human hematopoietic progenitor cells. region of chromosome 5 in malignant myeloid diseases to Leukemia 2006; 20: 115–121. 1-1.5 Mb and preparation of a PAC-based physical map. Proc 6 Leone G, Voso MT, Teofili L, Lubbert M. Inhibitors of DNA Natl Acad Sci USA 1997; 94: 6948–6953. methylation in the treatment of hematological malignancies and 25 Wijermans PW, Lubbert M, Verhoef G, Klimek V, Bosly A. MDS. Clin Immunol 2003; 109: 89–102. An epigenetic approach to the treatment of advanced MDS; 7 Sakajiri S, Kumagai T, Kawamata N, Saitoh T, Said JW, Koeffler the experience with the DNA demethylating agent 5-aza-20- HP. Histone deacetylase inhibitors profoundly decrease prolifera- deoxycytidine (decitabine) in 177 patients. Ann Hematol 2005, tion of human lymphoid cancer cell lines. Exp Hematol 2005; 33: 1–9. 53–61. 26 Ishiguro A, Spirin KS, Shiohara M, Tobler A, Gombart AF, Israel 8 Xu Y, Voelter-Mahlknecht S, Mahlknecht U. The histone deace- MA et al. Id2 expression increases with differentiation of human tylase inhibitor suberoylanilide hydroxamic acid down-regulates myeloid cells. Blood 1996; 87: 5225–5231. expression levels of Bcr-abl, c-Myc and HDAC3 in chronic 27 Schmelz K, Sattler N, Wagner M, Lubbert M, Dorken B, Tamm I. myeloid leukemia cell lines. Int J Mol Med 2005; 15: 169–172. Induction of gene expression by 5-Aza-20-deoxycytidine in acute 9 Silverman LR, Demakos EP, Peterson BL, Kornblith AB, Holland JC, myeloid leukemia (AML) and myelodysplastic syndrome (MDS) Odchimar-Reissig R et al. Randomized controlled trial of but not epithelial cells by DNA-methylation-dependent and azacitidine in patients with the myelodysplastic syndrome: a study -independent mechanisms. Leukemia 2005; 19: 103–111. of the cancer and leukemia group B. J Clin Oncol 2002; 20: 28 Galperin E, Benjamin S, Rapaport D, Rotem-Yehudar R, Tolchins- 2429–2440. ky S, Horowitz M. EHD3: a protein that resides in recycling tubular 10 Wijermans PW, Krulder JW, Huijgens PC, Neve P. Continuous and vesicular membrane structures and interacts with EHD1. infusion of low-dose 5-aza-20-deoxycytidine in elderly patients Traffic 2002; 3: 575–589. with high-risk myelodysplastic syndrome. Leukemia 1997; 11: 29 Naslavsky N, Caplan S. C-terminal EH-domain-containing pro- 1–5. teins: consensus for a role in endocytic trafficking, EH? J Cell Sci 11 Wijermans P, Lubbert M, Verhoef G, Bosly A, Ravoet C, Andre M 2005; 118: 4093–4101. et al. Low-dose 5-aza-20-deoxycytidine, a DNA hypomethylating 30 Naslavsky N, Rahajeng J, Sharma M, Jovic M, Caplan S. agent, for the treatment of high-risk myelodysplastic syndrome: a Interactions between EHD proteins and Rab11-FIP2; A role multicenter phase II study in elderly patients. J Clin Oncol 2000; for EHD3 in early endosomal transport. Mol Biol Cell 2006; 17: 18: 956–962. 163–177. 12 Lubbert M. DNA methylation inhibitors in the treatment of 31 Hirst J, Borner GH, Harbour M, Robinson MS. The aftiphilin/p200/ leukemias, myelodysplastic syndromes and hemoglobinopathies: \{gamma\}-synergin complex. Mol Biol Cell 2005; 16: 2554– clinical results and possible mechanisms of action. Curr Top 2565. Microbiol Immunol 2000; 249: 135–164. 32 Nakamaki T, Kawabata H, Saito B, Matsunawa M, Suzuki J, Adachi 13 Issa JP. Optimizing therapy with methylation inhibitors in D et al. Elevated levels of transferrin receptor 2 mRNA, not myelodysplastic syndromes: dose, duration, and patient selection. transferrin receptor 1 mRNA, are associated with increased Nat Clin Pract Oncol 2005; 2: S24–S29. survival in acute myeloid leukaemia. Br J Haematol 2004; 125: 14 Look AT. Molecular Pathogenesis of MDS. Hematology (Am Soc 42–49. Hematol Educ Program) 2005, 156–160. 33 Staber PB, Linkesch W, Zauner D, Beham-Schmid C, Guelly C, 15 Lai F, Godley LA, Joslin J, Fernald AA, Liu J, Espinosa 3rd R et al. Schauer S et al. Common alterations in gene expression and Transcript map and comparative analysis of the 1.5-Mb commonly increased proliferation in recurrent acute myeloid leukemia. deleted segment of human 5q31 in malignant myeloid diseases Oncogene 2004; 23: 894–904. with a del(5q). Genomics 2001; 71: 235–245. 34 Schiffer CA, Lee EJ, Tomiyasu T, Wiernik PH, Testa JR. Prognostic 16 Dann EJ, Rowe JM. Biology and therapy of secondary leukaemias. impact of cytogenetic abnormalities in patients with de novo acute Best Pract Res Clin Haematol 2001; 14: 119–137. nonlymphocytic leukemia. Blood 1989; 73: 263–270. 17 Koeffler HP, Golde DW. Acute myelogenous leukemia: a human 35 Horrigan SK, Arbieva ZH, Xie HY, Kravarusic J, Fulton NC, Naik H cell line responsive to colony-stimulating activity. Science 1978; et al. Delineation of a minimal interval and identification of 9 200: 1153–1154. candidates for a tumor suppressor gene in malignant myeloid 18 Coura R, Prolla JC, Meurer L, Ashton-Prolla P. An alternative disorders on 5q31. Blood 2000; 95: 2372–2377. protocol for DNA extraction from formalin fixed and paraffin wax 36 Nagafuchi A. Molecular architecture of adherens junctions. Curr embedded tissue. J Clin Pathol 2005; 58: 894–895. Opin Cell Biol 2001; 13: 600–603. 19 Takai N, Kawamata N, Walsh CS, Gery S, Desmond JC, Whittaker 37 Harris TJ, Peifer M. Decisions, decisions: beta-catenin chooses S et al. Discovery of epigenetically masked tumor suppressor genes between adhesion and transcription. Trends Cell Biol 2005; 15: in endometrial cancer. Mol Cancer Res 2005; 3: 261–269. 234–237. 20 Yamashita K, Upadhyay S, Osada M, Hoque MO, Xiao Y, Mori M 38 Nemade RV, Bierie B, Nozawa M, Bry C, Smith GH, Vasioukhin V et al. Pharmacologic unmasking of epigenetically silenced tumor et al. Biogenesis and function of mouse mammary epithelium suppressor genes in esophageal squamous cell carcinoma. Cancer depends on the presence of functional alpha-catenin. Mech Dev Cell 2002; 2: 485–495. 2004; 121: 91–99.

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

Leukemia