MIRA-Assisted Microarray Analysis, a New Technology for The

Research Article

MIRA-Assisted Microarray Analysis, a New Technology for the Determination of DNA Methylation Patterns, Identifies Frequent Methylation of Homeodomain-Containing Genes in Lung Cancer Cells

Tibor Rauch,1 Hongwei Li,1 Xiwei Wu,2 and Gerd P. Pfeifer1

Divisions of 1Biology and 2Biomedical Informatics, Beckman Research Institute of the City of Hope, Duarte, California

Abstract hypermethylation generally leads to inactivation of gene expres- We present a straightforward and comprehensive approach sion, this epigenetic alteration is considered to be a key mechanism for DNA methylation analysis in mammalian genomes. The for long-term silencing of tumor suppressor genes. The importance methylated-CpG island recovery assay (MIRA), which is based of promoter methylation in functional inactivation of lung cancer on the high affinity of the MBD2/MBD3L1 complex for suppressor genes is becoming increasingly recognized. It is methylated DNA, has been used to detect cell type–dependent estimated that between 0.5% and 3% of all genes carrying CpG- differences in DNA methylation on a microarray platform. The rich promoter sequences (so-called CpG islands) may be silenced procedure has been verified and applied to identify a series of by DNA methylation in lung cancer (1, 11). This means that there novel candidate lung tumor suppressor genes and potential are most likely several hundred genes that are incapacitated by this DNA methylation markers that contain methylated CpG pathway. Some of these genes may be bona fide tumor suppressor islands. One gene of particular interest was DLEC1, located genes, but in other cases, the methylation event may be a at a commonly deleted area on chromosome 3p22-p21.3, consequence of gene silencing or may somehow be associated with which was frequently methylated in primary lung cancers and tumor formation rather than being a cause of tumorigenesis. melanomas. Among the identified methylated genes, homeo- Several specific genes are methylated in lung cancer, including CDKN2A, RASSF1A, RARb, MGMT, GSTP1, CDH13, APC, DAPK, domain-containing genes were unusually frequent (11 of the TIMP3 top 50 hits) and were targeted on different chromosomes. , and several others (12–16). The methylation frequency These genes included LHX2, LHX4, PAX7, HOXB13, LBX1, SIX2, (the percentage of tumors analyzed that carry methylated alleles) HOXD3, DLX1, HOXD1, ONECUT2, and PAX9. The data show ranges from <10% to >80% but these numbers differ widely that MIRA-assisted microarray analysis has a low false- depending on the histologic type of tumor, the study population, positive rate and has the capacity to catalogue methylated and/or the methods used to assess methylation. To improve the CpG islands on a genome-wide basis. The results support the sensitivity of screening tools for the detection of early lung cancer, hypothesis that cancer-associated DNA methylation events DNA methylation markers have shown great promise (7, 17, 18). do not occur randomly throughout the genome but at least However, many more markers that could have improved specificity some are targeted by specific mechanisms. (Cancer Res 2006; in discriminating tumor from normal tissue and are methylated 66(16): 7939-47) at a high frequency in lung tumors have likely not yet been discovered. Introduction To analyze DNA methylation patterns on a genome-wide scale, several techniques have been developed, but none of them has yet In mammalian cells, the DNA base 5-methylcytosine occurs at ¶ reached wide acceptance. Most methods currently available are 5 -CpG dinucleotides and provides the basis for a common mode of labor-intensive and use methylation-sensitive restriction endonu- epigenetic inheritance. Changes in DNA methylation patterns cleases, and thus are limited by the occurrence of the respective occur in a developmental stage– and tissue-specific manner and sites within the target sequence. Another way to find methylated often accompany tumor development, most notably in the form of genes is by using expression microarrays to identify genes CpG island hypermethylation (1–10). During tumorigenesis, both reactivated by treatment with DNA methylation inhibitors (e.g., alleles of a tumor suppressor gene need to be inactivated by geno- ‘5-aza-deoxycytidine; refs. 19–21). This approach is effective but mic changes such as chromosomal deletions or loss-of-function can only be used with cell lines. Recently, genomic tiling and BAC mutations in the coding region of a gene. As an alternative microarrays have been introduced to map methylation patterns mechanism, transcriptional silencing by hypermethylation of CpG (22, 23). These methods are also limited, both in terms of their level islands spanning the promoter regions of tumor suppressor genes of resolution and in terms of the requirements for restriction is a common and important process in carcinogenesis. Because endonuclease recognition sites. An antibody against 5-methylcytosine has been used in immunoprecipitation experiments combined with microarrays (6, 23). However, this antibody requires ssDNA for recognition, which is sometimes difficult to achieve in CpG-rich Note: Supplementary data for this article are available at Cancer Research Online DNA regions. Here we describe a new genome-wide DNA (http://cancerres.aacrjournals.org/). Requests for reprints: Gerd P. Pfeifer, Division of Biology, Beckman Research methylation detection method that depends neither on restriction Institute of the City of Hope, 1500 East Duarte Road, Duarte, CA 91010. Phone: 626- endonucleases nor on specific antibodies. This method is based on 301-8853; Fax: 626-358-7703; E-mail: [email protected]. I2006American Association for Cancer Research. the methylated-CpG island recovery assay (MIRA), which we doi:10.1158/0008-5472.CAN-06-1888 previously applied for testing the methylation status of specific www.aacrjournals.org 7939 Cancer Res 2006; 66: (16). August 15, 2006

Downloaded from cancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. Cancer Research genes (24). It makes use of the high affinity of the MBD2b/MBD3L1 log 2 ratios between the dye swap pairs. Based on our experience, the complex for methylated DNA (24, 25). Here we show that MIRA can combined Lowess and dye swap normalization approach can best reduce be used to analyze the DNA methylation status of a large number of variability. CpG island methylation profiles were determined by ratios genes simultaneously using a microarray approach. between MIRA-enriched and unenriched samples (enrichment factor) for both tumor and normal tissues. The ratios of the enrichment factors between cancer and normal DNA samples will measure the methylation Materials and Methods difference between cancer and normal tissue. To identify the CpG islands MIRA and microarray analysis. DNA obtained from normal human that are differentially methylated between normal and tumor cell DNA, bronchial epithelial (NHBE) cells and from the lung cancer cell line A549 methylation profiles were compared using statistical linear model in was digested with MseI(5¶-TTAA), which produces small (f200-300 bp) LIMMA. For target gene selection, unadjusted P values were set at a level fragments and generally cuts outside of CpG islands. Linkers (upper of 0.05, and the fold change between cancer MIRA/Input versus normal strand sequence, 5¶-TAGAATTCAGATCTCCCG-3¶; lower strand sequence, MIRA/Input (difference factor) was set at >2. Direct comparison of MIRA- 3¶-CTTAAGTCTAGAGGGCCCAGTGGCG-5¶) were ligated to the MseI- enriched fractions from tumor and normal tissue DNA provided digested DNA and enrichment of the methylated fraction was done by independent confirmation for the methylation differences observed MIRA as previously described (24). Briefly, 1 Ag of purified GST-tagged although the latter analysis may be affected by differences in gene copy MBD2b protein and 1 Ag of purified His-tagged MBD3L1 protein were numbers between normal and tumor tissue. preincubated and bound to a glutathione sepharose CL-4B matrix DNA methylation analysis using combined bisulfite restriction (Amersham Biosciences, Piscataway, NJ). The plasmids used to produce analysis and bisulfite sequencing. DNA was isolated from cell lines or these proteins are available on request. This matrix was incubated with 500 frozen tumors and matched normal tissue by standard phenol-chloroform ng of MseI-cut and linker-ligated genomic DNA in 400 AL of a binding extraction and ethanol precipitation. Non–small-cell lung carcinoma tumor reaction mixture [10 mmol/L Tris-HCl (pH 7.5), 50 mmol/L NaCl, 1 mmol/L tissue samples and matching normal tissues removed with surgery were

EDTA, 1 mmol/L DTT, 3 mmol/L MgCl2, 0.1% Triton-X100, 5% glycerol, obtained from the frozen tumor bank of the City of Hope National Medical 25 Ag/mL bovine serum albumin, and 1.25 Ag/mL sonicated JM110 (dcm Center. The combined bisulfite restriction analysis (COBRA) assays were minus) bacterial DNA] for 240 minutes at 4jC on a rocking platform. After done using the method of Xiong and Laird (28). DNA was treated with washing the pelletted sepharose beads thrice with binding buffer containing sodium bisulfite and purified as described (24). PCR primers for 700 mmol/L NaCl, the methylated DNA–enriched genomic DNA fraction amplification of specific targets in bisulfite-treated DNA are listed in was eluted by addition of guanidinium hydrochloride–containing buffer and Supplementary Table S1. For sequence analysis, the PCR products obtained purified using Qiaquick PCR purification kits according to the instructions after bisulfite conversion were cloned into the pDrive PCR cloning vector of the manufacturer (Qiagen, Valencia, CA). The MIRA-captured DNA was (Qiagen) and 6to 10 individual clones were sequenced. then PCR amplified using 4.5 Amol/L linker primers in reaction buffer 5-Aza-deoxycytidine treatment. A549 cells were treated with 5-aza-2¶- (50 AL) containing deoxynucleotide triphosphates (250 Amol/L each) and deoxycytidine (Sigma) at 1, 5, and 10 Amol/L concentrations. Cells, 2 Â 106 Titanium Taq polymerase (Clontech, Palo Alto, CA). Twenty-two cycles of each, were grown for 4 days in the presence of different concentrations of amplification at 94jC for 1 minute, 60jC for 1 minute, and 68jCfor 5-aza-deoxycytidine. The medium was changed daily and 5-aza-deoxycyti- 1 minute were done. The amplified fragments were labeled with either Cy3- dine was replenished each day. RNA was isolated and reverse transcription- dCTP or Cy5-dCTP (Amersham) by random priming. In parallel, the MIRA- PCR (RT-PCR) was done using forward primer 5¶-GATATCTCGCACTTGCT- input DNA from normal and tumor DNA was labeled with either Cy3-dCTP CACC and reverse primer 5¶-ATCCAGCCGCTGCTTATAGA. or Cy5-dCTP by random priming (see scheme in Fig. 2). Then the dye- Gel mobility shift assays. To test the affinity of the MBD2b/MBD3L1 labeled DNA samples (500 ng each) were mixed and hybridized to human complex for CpG-methylated DNA, a double-stranded 55-mer oligo- CpG island microarrays (UHN Microarray Centre, University of Toronto, nucleotide (sequence 5¶-GATCCGCACCGAGGCGGGC-CGCTCCGTAGC- Toronto, Canada). The sequences present on this array are derived from a GCTACGGGACGCCCCGGGCCGCAGG-3¶) containing varying numbers of CpG island library in which 68% of the unique sequences map to the 5¶ end symmetrically methylated CpG dinucleotides was labeled at the 5¶-end and of known or putative genes (26, 27). The dye-labeled DNA samples were incubated with recombinant GST-tagged MBD2b alone (100 ng of protein) ethanol precipitated overnight at À70jC in the presence of human Cot-1 or with the GST-MBD2b/His-MBD3L1 complex (100 ng of each protein). DNA (7.2 Ag Cot-1 DNA/Ag sample). After ethanol precipitation, the samples The protein DNA complexes were incubated for 5 minutes at room were resuspended in 22 AL of hybridization buffer 1 (2.2Â SSC, 0.22% SDS). temperature in a buffer containing 10 mmol/L Tris-HCl (pH 7.5), 50 mmol/L Then, 20 AL of hybridization buffer 2 (70% formamide/3Â SSC, 14.3% NaCl, 1 mmol/L EDTA, 1 mmol/L DTT, 3 mmol/L MgCl2, 0.1% Triton- dextran sulfate) were added to the mixture, and the tube was first heated at X100, 5% glycerol, 25 Ag/mL BSA, and 1.25 Ag/mL sonicated JM110 (dcm 95jC for 5 minutes to denature the DNA and then incubated at 42jCfor minus) bacterial DNA. Mobility shift assays were done using 5% poly- 2 minutes. At this step, 4 Ag of yeast tRNA (Sigma) and 3 AL of 2% BSA were acrylamide gels. added to the mixture, which was then spotted onto the DNA microarray slides. A 25 Â 60 mm coverslip was then gently placed on top of the samples Results and the hybridization was carried out in a hybridization chamber (Corning, Acton, MA) at 60jC overnight in a water bath. After the hybridization, Among all methyl-CpG binding proteins known, MBD2b has the the microarray slides were washed and dried by a spinning in a tabletop highest affinity for methylated DNA and displays the greatest centrifuge. Two independent MIRA assays were done for tumor and normal capacity to differentiate between methylated and unmethylated tissue, and the input and MIRA-enriched fractions were labeled using DNA (29). It recognizes a wide range of methylated CpG sequences either Cy3 or Cy5. A total of eight arrays were labeled as dye-swap pairs with little sequence specificity, unlike MeCP2, which prefers AT- and hybridized. After washing, the slides were scanned using an Axon rich neighboring sequence contexts (30). MBD3L1, a protein with GenePix 4000b scanner and images were quantified by GenePix pro v.6 homology to MBD2b, lacks the NH2-terminal methyl-CpG binding software. Preprocessing of raw data and statistical analysis were done domain but directly interacts with MBD2b (25, 31). Using a using Bioconductor packages in R programming environment. The spots randomly designed 55-mer oligonucleotide with different numbers marked as ‘‘bad’’ or ‘‘not found’’ by the GenePix software were excluded. Background correction was done using the ‘‘normexp’’ method imple- of methylated CpGs, we determined that the MBD2b/MBD3L1 mented in the Bioconductor LIMMA package to adjust the local median complex has a much higher affinity for methylated DNA than background estimates. The background-corrected intensity data were MBD2b alone (Fig. 1A). Depending on the number of methylated normalized using the Print-tip group Lowess method to remove the bias CpGs, the presence of MBD3L1 leads to enhanced binding of the within each array. The dye bias was then further corrected by averaging the complex and reduced mobility, suggesting the formation of

Cancer Res 2006; 66: (16). August 15, 2006 7940 www.aacrjournals.org

CpG island microarrays containing 12,192 CpG islands, of which >60% map to the 5¶ promoter sequences of genes (26, 27). We applied this technique to identify CpG islands methylated in the lung cancer cell line A549 relative to NHBE cells (Fig. 2B and C;

Figure 1. Affinity of the MBD2b/MBD3L1 complex for CpG-methylated DNA and dependence of the MIRA pulldown on the number of methylated CpGs. A, gel mobility shift assay with CpG-methylated oligonucleotides. A 55-mer oligonucleotide, containing varying numbers of symmetrically methylated CpG dinucleotides, was incubated with recombinant MBD2b alone (100 ng of protein) or with the MBD2b/MBD3L1 complex (100 ng of each protein). The protein DNA complexes were incubated for 5 minutes at room temperature and then a mobility shift assay was done using a 5% polyacrylamide gel. B, efficiency of the MIRA pull-down is dependent on the number of methylated CpGs. Restriction-cut human genomic DNA was methylated with different DNA methylases (SssI, HpaII, and/or HhaI) to introduce different numbers of methylated CpGs into the unmethylated CpG island promoter of the TBP gene. After MIRA, the fragments were amplified by quantitative real-time PCRusing TBP-specific primers.

multiple MBD2b/MBD3L1 complexes in the presence of more than one methylated CpG dinculeotide. The MIRA assay, which is based on the strong affinity of the MBD2b/MBD3L1 complex for methylated DNA, has a high specificity for enriching methylated DNA, and unmethylated DNA fragments stay in the supernatant (24). We initially determined that the pulldown of methylated DNA by MIRA is dependent on the number of methylated CpGs present in genomic DNA fragments. Recovery of fragments containing 13 methylated CpGs was more efficient than recovery of fragments containing two methylated CpGs and much more efficient than recovery of fragments with only one or zero methylated CpG sites (Fig. 1B). Thus, the efficiency of the MIRA enrichment depends on CpG density and the approach seems to be ideally suited for pulling down methylated CpG islands. Figure 2. MIRA-assisted CpG island microarray analysis. A, schematic diagram of the MIRA microarray method. Input and MIRA-enriched fractions are Figure 2A outlines the MIRA microarray approach. Genomic labeled with different dyes, mixed, and hybridized to the slides and the relative DNA, isolated from different sources, is cleaved with MseI enrichment factors between different cell types and tissues are determined. ¶ For confirmation, MIRA-enriched DNA from normal and tumor cells can be mixed (5 -TTAA), which produces small 200- to 300-bp fragments and and hybridized directly. B, representative data for MIRA microarrays. Left and cuts outside of CpG islands, and compatible linkers are ligated to middle, red, MIRA-enriched fractions; green, input fractions. Right, green, the ends. Then the MIRA pulldown is done to isolate the MIRA-enriched A549 DNA; red, NHBE cell DNA. C, pairwise comparison. Enrichment factors obtained from NHBE cells (vertical axis) and A549 cells methylated DNA fraction. Input and MIRA-enriched fractions are (horizontal axis) were compared. The dots in the blue circle are the targets labeled with different fluorescent dyes, mixed, and hybridized to selectively methylated in tumor cell DNA. www.aacrjournals.org 7941 Cancer Res 2006; 66: (16). August 15, 2006

Table 1. Methylated target genes identified by MIRA microarrays

c Target no. ID FC difference* Genome location Distance Gene symbol Description

1 1_A_12 4.72 chr2:175033605-175034788 5 kb 3¶ CIR1 CBF1 interacting corepressor isoform 1 2 18_E_17 4.34 chr3:38055590-38056657 0 DLEC1 Deleted in lung and esophageal cancer 1 3 1_B_19 3.69 chr1:176933895-176934835 0 LHX4 LIM homeobox 4 4 15_G_63.58 chr1:114111744-114111847 0 PTPN22 Protein tyrosine phosphatase 5 8_E_7 3.54 chr12:14237814-14238207 >30 kb 6 25_H_18 3.53 chr14:23904920-23905609 1650 bp 5¶ NFATC4 Cytoplasmic nuclear factor of activated T-cells 7 22_N_10 3.31 chr6:100542633-100542762 0 GRP145 G protein-coupled receptor 145 8 31_L_22 3.28 chr6:115099013-115099186 >30 kb 9 25_N_11 3.24 chr10:42569409-42570107 0 hmm7851 Hypothetical protein 10 13_P_8 3.16 chr6:99402311-99402598 0 hmm33765 Hypothetical protein 11 31_L_21 3.11 chr6:115099013-115099186 >30 kb 12 26_M_12 3.00 chr13:49163567-49163905 0 EBPL Emopamil binding related protein 13 17_F_9 2.92 chr6:45327693-45327793 0 SUPT3H SUPT3H protein 14 25_O_20 2.78 chrX:48441921-48442544 2.5 kb 5¶ ERAS Small GTPase protein E-Ras 15 17_F_20 2.76 chr9:123856765-123857465 0 LHX2 LIM homeobox 2 16 7_L_22 2.65 chr5:72775997-72776155 3 kb 3¶ FOXD1 Forkhead transcription factor 17 23_B_15 2.63 chr8:91679368-91679403 >30 kb 18 29_A_11 2.63 chr1:18716744-18718019 0 PAX7 Paired box gene 7 19 26_M_11 2.62 chr12:55167740-55168302 0 GLS2 Glutaminase 2 (liver, mitochondrial) 20 7_K_7 2.58 chr17:44187113-44188088 26kb 5 ¶ HOXB13 Homeobox gene 21 29_A_6 2.58 chr4:136284675-136284784 >30 kb 22 2_B_9 2.54 chr10:102972869-102973961 2.5 kb 3¶ LBX1 Homeobox transcription factor 23 21_D_20 2.45 chr2:45139996-45140793 3.6 kb 3¶ SIX2 Sine oculis homeobox homologue 2 24 21_E_12 2.45 chr12:48026625-48026800 0 FLJ13236 Hypothetical protein FLJ13236 25 2_A_10 2.40 chr10:102972869-102973961 0 BRACE2016602 Hypothetical protein 2614_J_18 2.38 chr5:83154024-83154192 0 hmm32907 Hypothetical protein 27 12_B_24 2.35 chr1:16757837-16758127 0 EIF1AP1 Translation initiation factor 1A pseudogene 1 28 22_L_21 2.34 chr20:28164999-28165149 0 hmm117175 Hypothetical protein 29 7_F_8 2.32 chr6:91594928-91595364 >30 kb 30 25_C_14 2.30 chr4:109450308-109450511 3.1 kb 5¶ LEF1 Lymphoid enhancer binding factor-1 31 25_I_22 2.28 chr19:4279686-4280325 0 STAP2 Signal-transducing adaptor protein 2 32 14_C_16 2.26 chr6:144647977-144648462 0 hmm33914 Hypothetical protein 33 32_G_18 2.26chr5:139908213-139908494 0 EIF4EBP3 Eukaryotic translation initiation factor 4E 34 15_M_17 2.25 chr22:27793134-27793763 0 KREMEN1 Inhibitor of Wnt signaling 35 30_E_8 2.24 chr5:54551833-54552394 0 CR626610 Hypothetical protein 36 14_N_16 2.21 chr2:176847561-176847721 6 kb 5¶ HOXD3 Homeobox gene HOXD3 37 18_E_9 2.19 chr10:102972869-102973961 2.5 kb 3¶ LBX1 Homeobox transcription factor 38 20_O_10 2.16 chr2:172770466-172770678 4 kb 5¶ DLX1 Homeobox protein DLX-1 39 15_J_24 2.15 chr4:88700602-88701596 0 NUDT9 Nucleoside diphosphate linked moiety X-type 9 40 1_O_12 2.15 chr2:176802127-176803414 40 kb 3¶ HOXD1 Homeobox gene 41 14_L_15 2.13 chr7:154669328-154670208 0 hmm78131 Hypothetical protein 42 2_D_17 2.10 chr6:134258454-134259172 0.3 kb 3¶ TCF21 Helix-loop-helix transcription factor 43 20_L_5 2.07 chr15:39592659-39593447 0 LTK Leukocyte tyrosine kinase 44 23_H_10 2.07 chr8:145671696-145672153 1.4 kb 5¶ KIFC2 Kinesin motor 45 12_C_14 2.02 chr11:73167747-73168609 18 kb 5¶ RAB6A Small GTPase 4631_G_9 2.00 chr17:70443461-704437800 OTOP3 Otopetrin-3 47 26_N_12 2.00 chr1:148695088-148695652 0 THEM4 Negative regulator of AKT 48 11_C_15 2.00 chr18:53259181-53259461 0 ONECUT2 Homeobox gene 49 26_N_12 2.00 chr1:148695088-148695652 0 KIAA0460 KIAA0460 protein 50 2_D_9 2.00 chr14:36197384-36198099 2.6 kb 5¶ PAX9 Paired box gene 9

NOTE: Target information was verified using the University of California Santa Cruz Genome Browser (May 2004 assembly) and GenBank. *FC difference is the ratio (fold change) of MIRA-enriched over unenriched (input) A549 DNA versus MIRA-enriched over unenriched (input) NHBE DNA. cA distance of ‘‘0’’ indicates that the target falls within a gene including 500 bp upstream of its transcription start site. see Materials and Methods). MIRA enrichment factors indicating in A549 cells relative to NHBE cells (Table 1). Most of the differences in levels of methylation are calculated and compared differentially methylated CpG islands mapped either close to the 5¶ between the two cell lines. Using the data obtained from such ends of known or predicted genes or to the exons/introns within arrays, a list of genes was compiled, which shows hypermethylation genes and may represent regulatory elements.

Cancer Res 2006; 66: (16). August 15, 2006 7942 www.aacrjournals.org

We confirmed cancer cell–specific methylation and lack of (e.g., T4, T6, T8, T9, T10, and T15; Fig. 3B) showed clear evidence of methylation in NHBE cells for several of the targets identified by methylation. None of the adjacent, tumor-matched, normal tissues the microarray analysis using a BstUI COBRA assay (Fig. 3A). In had methylation of DLEC1 (Fig. 3B and C). Using sodium bisulfite this assay, a CpG-containing restriction endonuclease cleavage sequencing (33), we verified that the DLEC1-associated CpG island site is conserved after bisulfite treatment when the DNA is meth- was highly methylated in A549 cells and in a primary lung tumor ylated (28). COBRA assays using BstUI (5¶-CGCG) initially con- but was completely unmethylated in NHBE cells or normal lung firmed the methylation of targets ranked number 1, 4, 10, and 20 on tissue (Fig. 3C). Methylation encompassed the entire CpG island of the list of differentially methylated genes, indicating the robustness DLEC1 (Fig. 3C). Methylation of DLEC1 correlated with a lack of and specificity of the MIRA microarray approach. Several target expression in A549 cells and the gene could be reactivated by genes were of particular interest. One methylated gene (target 2 in treatment of the cell line with 5-aza-deoxycytidine (Fig. 3D). In Table 1) is designated as DLEC1 (Deleted in Lung and Esophageal addition, the DLEC1 promoter was methylated in 3 of 15 (20%) Cancers) and maps to chromosome 3p22-p21.3, a common hotspot primary esophageal cancers and in 4 of 10 (40%) primary for loss of heterozygosity and deletions in lung cancer and other melanomas tested (see Supplementary Fig. S1). solid tumors. DLEC1 encodes a protein of 1755 amino acids with Interestingly, 11 of the top 50 hits, with a calculated >2-fold unknown function (32). To investigate if methylation of DLEC1 is enrichment of the methylated fraction in A549 versus NHBE cell present in human primary lung tumors, we analyzed a series of DNA, were mapped to homeobox genes (LHX2, LHX4, PAX7, 30 primary non–small-cell lung cancers by the COBRA assay HOXB13, LBX1, SIX2, HOXD3, DLX1, HOXD1, ONECUT2, and PAX9; (examples are shown in Fig. 3B) and by bisulfite sequencing see Table 1). We confirmed the methylation difference for these (Fig. 3C). Eight of the 30 (27%) undissected tumor samples tested homeobox genes using standard bisulfite-based approaches.

Figure 3. Confirmation of methylation differences and identification of DLEC1 as a target for tumor-specific methylation. A, confirmation of tumor cell–specific methylation of four candidate target genes identified by the MIRA-assisted microarray approach (targets 1, 4, 10, and 20 in Table 1). A549 and NHBE cell DNA was treated with sodium bisulfite and the target CpG island sequences were amplified using specific primers. Methylation was confirmed by a BstUI COBRA assay producing cleavage products (arrows) when methylation at 5¶-CGCG sequences is present. B, methylation of the DLEC1 gene in primary lung cancers. DNA was isolated from primary non–small-cell lung carcinoma tumors (T) and their adjacent normal tissues (N). After sodium bisulfite treatment, the DLEC1 promoter CpG island was PCRamplified. Cutting with BstUI indicates methylation of 5¶-CGCG sequences within this CpG island sequence, which contains three BstUI sites (see samples T4, T6, T8, T9, T10, and T15). C, determination of tumor cell–specific methylation of the CpG island of the DLEC1 gene by bisulfite sequencing. Bisulfite genomic sequencing was done on DNA isolated from A549 cells, NHBE cells, a primary lung tumor, and adjacent normal lung tissue. Primers specific for the DLEC1 CpG island (gray shading) were used. In NHBE and A549 cells, almost the entire CpG island was sequenced (long bar). In the primary tissues, only the area proximal to the transcription start site (arrow) was sequenced (short bar). Sequencing results of several independent clones are shown. Black squares, methylated CpG dinucleotides. D, expression of DLEC1 in NHBE and A549 cells and reactivation by 5-aza-deoxycytidine (5¶-AzadC) treatment.

www.aacrjournals.org 7943 Cancer Res 2006; 66: (16). August 15, 2006

Methylationlevelswerehigherincancercellsforallnine tumors and LHX4 was methylated in 6of 8 (75%) of the tumors homeobox gene targets tested (see Fig. 4A). Methylation was analyzed. The methylation differences were confirmed by sodium either absent in NHBE cells or was partial as seen for targets bisulfite sequence analysis (Fig. 5). Of a total of 210 CpG sequences 3, 15, 38, and 48. The identified methylated sequences at homeo- analyzed for LHX2, 32 were methylated in normal tissue and 95 box genes did not always coincide with the 5¶ ends of the genes, were methylated in the tumor (P < 0.001, m2 test). Of 260 CpG but several targets were localized either to intronic CpG islands sequences analyzed for LHX4, 105 were methylated in normal (LHX4, PAX7, and ONECUT2) or to CpG islands near the 3¶ end of tissue and 187 were methylated in the tumor (P < 0.001, m2 test). the gene (LBX1, SIX2, and HOXD1). For SIX2, we determined that We have also looked for hypomethylation events in the cancer the CpG island spanning the promoter of this gene was also cell line. We had to use less stringent cutoffs (1.5-fold difference methylated, in addition to the CpG island near its 3¶ end (data not factor) to be able to pick up any hypomethylation targets. Of only shown). four such targets identified, two had no BLAST hits on the genome To extend the analysis of homeobox gene methylation to pri- and one did not map near a known gene. The one remaining hit mary lung tumors, we analyzed the methylation status of the CpG mapped to the 3¶ end of the TAF6L gene. Thus, it is clear that hypo- islands associated with the LHX2 and LHX4 genes in primary lung methylation of CpG-rich sequences relative to normal bronchial tumors (Fig. 4B and C). For LHX2, the methylation target spans epithelial cells is not a common event in this cancer cell line. intron 2 of the gene, and for LHX4, the target is a CpG island spanning intron 1. Although low levels of methylation were detected in some normal tissues removed with tumor surgery, Discussion methylation of LHX2 and LHX4 was more pronounced in tumor We present a new method for analysis of DNA methylation samples. LHX2 was methylated in 7 of 12 (58%) of primary lung patterns on a genome-wide scale. The MIRA method is based on

Figure 4. Methylation of homeobox genes in lung cancer. A, methylation differences of homeobox gene targets between NHBE and A549 cells. Several targets contain multiple BstUI sites and the smaller cleavage fragments are not visible on the gel. B, methylation of the LHX2 gene in primary lung tumors (T) and matching normal tissues (N). C, methylation of the LHX4 gene in primary lung tumors (T) and matching normal tissues (N).

Cancer Res 2006; 66: (16). August 15, 2006 7944 www.aacrjournals.org

Figure 5. Determination of the methylation differences for LHX2 and LHX4 by sodium bisulfite sequencing. Sodium bisulfite sequencing was done for the methylation target in the LHX2 gene (T/N pair #9; Fig. 4B) and the LHX4 gene (T/N pair #3; Fig. 4C).

the selective enrichment of methylated DNA fragments by a matrix The remodeling of genome-wide methylation patterns during containing the methylated-CpG binding protein MBD2b in the development and differentiation can be analyzed. Hypermethyla- presence of its binding partner, MBD3L1. MBD3L1 strongly tion of CpG islands is a hallmark of cancer cells. Thus, the MIRA enhances the binding of MBD2b to methylated DNA (Fig. 1A). microarray approach will have important use for a more com- The specificity of the MIRA method for enriching methylated DNA prehensive characterization of cancer-associated methylation forms the basis of this approach (24). The binding of MBD2b to changes. New tumor suppressor genes or new DNA methylation methylated DNA is relatively sequence independent, in contrast to markers for the early detection of cancer may be identified. Using another methylated-CpG binding protein, MeCP2 (29, 30). We have genomic tiling arrays, it should become possible to more precisely confirmed a lack of sequence bias (beyond that for methylated localize genomic areas that undergo hypomethylation in cancer. A CpGs) by sequence analysis of the CpG island targets identified on complete understanding of the range of CpG islands methylated the arrays (Table 1). No specific sequence motif spanning CpG sites, within a tumor and the occurrence of tumor-specific or even tumor either in all targets combined or specifically in the homeobox subtype–specific methylation patterns may aid in our understand- genes, was present in a majority of the targets identified. The ing of pathways leading to tumorigenesis. MIRA microarray specificity of the MIRA microarray approach has been confirmed. technology may allow the classification of tumors based on their Of the 18 targets identified by microarrays and tested for methylation patterns. methylation differences using standard sodium bisulfite-based The MIRA microarray analysis has generated a list of genes techniques, all 18 have been confirmed as differentially methylated, methylated in a human lung cancer cell line. Many of them require indicating that the number of false positives identified by this follow-up investigations to determine the frequency of methylation method is very low. It is likely that fold difference factors <2 will in primary tumors and the potential biological significance of still indicate differential methylation. We have confirmed tumor methylation silencing for tumor progression. Among the methyl- cell–specific methylation of another homeodomain gene, RAX,in ated genes identified were several genes involved in signal A549 cells although the fold difference factor was only 1.6. We transduction pathways and transcriptional regulation (Table 1). expect that a 1.5-fold difference factor will in many cases be Of particular interest was one gene, DLEC1, which is localized sufficient to indicate differential methylation, although this will within a common area of loss of heterozygosity and deletion on the need to be verified for more genes. short arm of chromosome 3 (34). This chromosomal area is The MIRA microarray has several advantages over existing thought to harbor one or several important lung tumor suppressor techniques for analyzing DNA methylation patterns. It does not genes, but the exact identity of these genes has remained unclear. depend on the occurrence of specific methylation-sensitive Search for mutational changes in genes mapped to this location restriction sites within the target sequence. It also does not has not identified any gene that is mutated frequently. However, depend on the use of antibodies against 5-methylcytosine, which several genes at 3p22-p21.3 are silenced epigenetically in lung require ssDNA for recognition. The recombinant proteins, GST- cancers (16, 35). DLEC1, which is not expressed in a fraction of lung tagged MBD2b and His-tagged MBD3L1, are easy to prepare and and esophageal cancers and inhibits the growth of cancer cell lines can be stored in large quantities as a frozen stock. No complicated on overexpression (32), may be an important target for methylation manipulations need to be done. The technique can easily be silencing in several types of malignancies (Fig. 3). adopted using a variety of array platforms including promoter Our search for genes methylated in lung cancer has uncovered arrays and genomic tiling arrays. Differentiation between methyl- an unexpectedly frequent occurrence of homeobox genes as ated and unmethylated DNA is best achieved when two or more methylation targets (11 of the top 50 hits). This was a much larger methylated CpG sites are present in the MseI fragment. The number than expected because the human genome contains only method will be most sensitive for identifying differentially a few hundred homeodomain genes. Simply by chance, <1 of methylated CpG islands due to the high frequency of methylated 50 genes would be expected to be a homeodomain-containing CpGs in these targets. The technique should be suitable, for gene. Many homeobox genes are aberrantly expressed in a variety example, for the identification of new imprinted genes in mam- of hematologic malignancies and solid tumors, and this change malian genomes. Imprinted genes often are associated with CpG includes more commonly up-regulation and less commonly down- islands that carry differential methylation marks in the germ line. regulation of the gene (36, 37). Methylation of regulatory elements www.aacrjournals.org 7945 Cancer Res 2006; 66: (16). August 15, 2006

Downloaded from cancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. Cancer Research in CpG islands associated with homeobox genes may result in silencing function of the EED-EZH2 complex, has identified a large silencing of homeobox gene expression if the gene is expressed in fraction of the targets as homeobox genes (45, 46). These included adult tissues or stem cells. Individual homeobox genes have been most of the homeotic genes found in the HOX clusters and the reported to be methylated in tumors of various histologic origins. majority of the homeodomain genes. Several of these genes were For example, methylation of HOXB13 has been reported to occur in identical to the ones we identified as targets for DNA methylation in 30% of renal cell carcinomas (38) and methylation of genes in the this study. The polycomb component EZH2 associates with DNA HOXA and HOXD clusters was reported in lung cancer (39). The methyltransferases (47, 48) and EZH2 expression is increased in HOXA5 promoter region was methylated in 16of 20 p53-negative tumors of different histologic types including precancerous tissue breast tumor specimens (40) and the homeobox gene NKX3.1 is (49, 50). Thus, although speculative at present, these data suggest a commonly methylated in prostate cancers (41). Our data indicate potential mechanism by which homeobox genes (and perhaps other that simultaneous methylation of a series of homeobox genes genes) may become targets for de novo methylation in cancer cells. located on different chromosomes occurs in lung cancer cells. In summary, our data indicate that the MIRA microarray Most homeobox genes are not expressed in adult tissues and the approach has the ability to identify genes methylated in human lack of their expression may promote de novo methylation during tumors on a genome-wide basis. This technology is expected to be malignant progression. For example, LHX2 and LHX4 are involved widely applicable for comprehensive analysis of DNA methylation in the development of the central nervous system (42, 43) and are patterns using available spotted microarrays and new-generation not known to be expressed in normal lung tissue. These genes were genome-scanning arrays currently under development. Because not expressed in normal human bronchial epithelial cells as this technique is straightforward and does not require complicated determined by RT-PCR analysis (data not shown) although they manipulations, it should easily be applicable in clinical settings and were only partially methylated (Fig. 4A). LHX2 and LHX4 could not should become useful as a diagnostic tool to classify tumors be reactivated by 5-aza-deoxycytidine treatment. Thus, the according to DNA methylation patterns on a genomic scale. In mechanism of their primary repression is likely independent of addition, this technology may aid in the identification of new DNA methylation. Progressive methylation of LHX2 and LHX4 candidate tumor suppressor genes and potential DNA methylation occurred during lung cancer development (Figs. 4B and 5). markers. Although we have not identified specific sequence similarities in the homeobox gene-associated CpG islands that undergo methylation, a common mechanistic pathway may exist in cancer cells, Acknowledgments which promotes de novo methylation of these targets. Polycomb Received 5/23/2006; revised 6/14/2006; accepted 6/21/2006. Grant support: NIH grant CA104967 (G.P. Pfeifer). repressive complexes are involved in silencing of homeobox genes The costs of publication of this article were defrayed in part by the payment of page (44), although it has been difficult to identify specific polycomb charges. This article must therefore be hereby marked advertisement in accordance response elements in mammalian systems. A recent genome-wide with 18 U.S.C. Section 1734 solely to indicate this fact. We thank Stella Tommasi for bisulfite-treated DNA from melanomas, Maricela survey for localization of polycomb components including SUZ12, Covarrubias for hybridization of microarray slides, and Steven Bates for help with cell which is required for the histone methyltransferase activity and culture experiments.

References 12. Dammann R, Strunnikova M, Schagdarsurengin U, 21. Yamashita K, Upadhyay S, Osada M, et al. Pharma- et al. CpG island methylation and expression of tumour- cologic unmasking of epigenetically silenced tumor 1. Costello JF, Fruhwald MC, Smiraglia DJ, et al. Aberrant associated genes in lung carcinoma. Eur J Cancer 2005; suppressor genes in esophageal squamous cell carcino- CpG-island methylation has non-random and tumour- 41:1223–36. ma. Cancer Cell 2002;2:485–95. type-specific patterns. Nat Genet 2000;24:132–8. 13. TopalogluO,HoqueMO,TokumaruY,etal. 22. Ching TT, Maunakea AK, Jun P, et al. Epigenome 2. Esteller M, Corn PG, Baylin SB, Herman JG. A gene Detection of promoter hypermethylation of multiple analyses using BAC microarrays identify evolutionary hypermethylation profile of human cancer. Cancer Res genes in the tumor and bronchoalveolar lavage of conservation of tissue-specific methylation of SHANK3. 2001;61:3225–9. patients with lung cancer. Clin Cancer Res 2004;10: Nat Genet 2005;37:645–51. 3. Jones PA, Baylin SB. The fundamental role of epi- 2284–8. 23. Weber M, Davies JJ, Wittig D, et al. Chromosome- genetic events in cancer. Nat Rev Genet 2002;3:415–28. 14. Yanagawa N, Tamura G, Oizumi H, Takahashi N, wide and promoter-specific analyses identify sites of 4. Issa JP. CpG island methylator phenotype in cancer. Shimazaki Y, Motoyama T. Promoter hypermethylation differential DNA methylation in normal and trans- Nat Rev Cancer 2004;4:988–93. of tumor suppressor and tumor-related genes in non- formed human cells. Nat Genet 2005;37:853–62. 5. Baylin S, Bestor TH. Altered methylation patterns in small cell lung cancers. Cancer Sci 2003;94:589–92. 24. Rauch T, Pfeifer GP. Methylated-CpG island recovery cancer cell genomes: cause or consequence? Cancer Cell 15. Zochbauer-Muller S, Fong KM, Virmani AK, Geradts assay: a new technique for the rapid detection of 2002;1:299–305. J, Gazdar AF, Minna JD. Aberrant promoter methylation methylated-CpG islands in cancer. Lab Invest 2005;85: 6. Keshet I, Schlesinger Y, Farkash S, et al. Evidence for of multiple genes in non-small cell lung cancers. Cancer 1172–80. an instructive mechanism of de novo methylation in Res 2001;61:249–55. 25. Jiang CL, Jin SG, Pfeifer GP. MBD3L1 is a transcrip- cancer cells. Nat Genet 2006;38:149–53. 16. Dammann R, Li C, Yoon JH, Chin PL, Bates S, Pfeifer tional repressor that interacts with MBD2 and compo- 7. Ushijima T. Detection and interpretation of altered GP. Epigenetic inactivation of a RAS association domain nents of the NuRD complex. J Biol Chem 2004;279: methylation patterns in cancer cells. Nat Rev Cancer family protein from the lung tumour suppressor locus 52456–64. 2005;5:223–31. 3p21.3. Nat Genet 2000;25:315–9. 26. Cross SH, Charlton JA, Nan X, Bird AP. Purification of 8. Bird A. DNA methylation patterns and epigenetic 17. Belinsky SA. Gene-promoter hypermethylation as a CpG islands using a methylated DNA binding column. memory. Genes Dev 2002;16:6–21. biomarker in lung cancer. Nat Rev Cancer 2004;4:707–17. Nat Genet 1994;6:236–44. 9. Feinberg AP, Ohlsson R, Henikoff S. The epigenetic 18. Laird PW. The power and the promise of DNA 27. Heisler LE, Torti D, Boutros PC, et al. CpG Island progenitor origin of human cancer. Nat Rev Genet 2006; methylation markers. Nat Rev Cancer 2003;3:253–66. microarray probe sequences derived from a physical 7:21–33. 19. Shi H, Wei SH, Leu YW, et al. Triple analysis of the library are representative of CpG Islands annotated on 10. Feltus FA, Lee EK, Costello JF, Plass C, Vertino PM. cancer epigenome: an integrated microarray system for the human genome. Nucleic Acids Res 2005;33:2952–61. Predicting aberrant CpG island methylation. Proc Natl assessing gene expression, DNA methylation, and 28. XiongZ,LairdPW.COBRA:asensitiveand Acad Sci U S A 2003;100:12253–8. histone acetylation. Cancer Res 2003;63:2164–71. quantitative DNA methylation assay. Nucleic Acids Res 11. Shiraishi M, Sekiguchi A, Terry MJ, et al. A 20. Suzuki H, Gabrielson E, Chen W, et al. A genomic 1997;25:2532–4. comprehensive catalog of CpG islands methylated in screen for genes up-regulated by demethylation and 29. Fraga MF, Ballestar E, Montoya G, Taysavang P, human lung adenocarcinomas for the identification of histone deacetylase inhibition in human colorectal Wade PA, Esteller M. The affinity of different MBD tumor suppressor genes. Oncogene 2002;21:3804–13. cancer. Nat Genet 2002;31:141–9. proteins for a specific methylated locus depends on

Cancer Res 2006; 66: (16). August 15, 2006 7946 www.aacrjournals.org

their intrinsic binding properties. Nucleic Acids Res sor gene SEMA3B. Proc Natl Acad Sci U S A 2001;98: 43. Porter FD, Drago J, Xu Y, et al. Lhx2, a LIM homeobox 2003;31:1765–74. 13954–9. gene, is required for eye, forebrain, and definitive eryth- 30. Klose RJ, Sarraf SA, Schmiedeberg L, McDermott SM, 36. Grier DG, Thompson A, Kwasniewska A, McGonigle rocyte development. Development 1997;124:2935–44. Stancheva I, Bird AP. DNA binding selectivity of MeCP2 GJ, Halliday HL, Lappin TR. The pathophysiology of 44. Cao R, Zhang Y. SUZ12 is required for both the histone due to a requirement for A/T sequences adjacent to HOX genes and their role in cancer. J Pathol 2005;205: methyltransferase activity and the silencing function of methyl-CpG. Mol Cell 2005;19:667–78. 154–71. the EED-EZH2 complex. Mol Cell 2004;15:57–67. 31. Jiang CL, Jin SG, Lee DH, et al. MBD3L1 and MBD3L2, 37. Samuel S, Naora H. Homeobox gene expression in 45. Boyer LA, Plath K, Zeitlinger J, et al. Polycomb two new proteins homologous to the methyl-CpG- cancer: insights from developmental regulation and complexes repress developmental regulators in murine binding proteins MBD2 and MBD3: characterization of deregulation. Eur J Cancer 2005;41:2428–37. embryonic stem cells. Nature 2006;441:349–53. MBD3L1 as a testis-specific transcriptional repressor. 38. Okuda H, Toyota M, Ishida W, et al. Epigenetic 46. Lee TI, Jenner RG, Boyer LA, et al. Control of Genomics 2002;80:621–9. inactivation of the candidate tumor suppressor gene developmental regulators by polycomb in human 32. Daigo Y, Nishiwaki T, Kawasoe T, Tamari M, Tsuchiya HOXB13 in human renal cell carcinoma. Oncogene 2006; embryonic stem cells. Cell 2006;125:301–13. E, Nakamura Y. Molecular cloning of a candidate tumor 25:1733–42. 47. Vire E, Brenner C, Deplus R, et al. The Polycomb suppressor gene, DLC1, from chromosome 3p21.3. 39. Shiraishi M, Sekiguchi A, Oates AJ, Terry MJ, group protein EZH2 directly controls DNA methylation. Cancer Res 1999;59:1966–72. Miyamoto Y. HOX gene clusters are hotspots of Nature 2006;439:871–4. 33. Clark SJ, Harrison J, Paul CL, Frommer M. High de novo methylation in CpG islands of human lung 48. Hernandez-Munoz I, Taghavi P, Kuijl C, Neefjes J, van sensitivity mapping of methylated cytosines. Nucleic adenocarcinomas. Oncogene 2002;21:3659–62. Lohuizen M. Association of BMI1 with polycomb bodies Acids Res 1994;22:2990–7. 40. Raman V, Martensen SA, Reisman D, et al. Compro- is dynamic and requires PRC2/EZH2 and the mainte- 34. Wistuba II, Behrens C, Virmani AK, et al. High mised HOXA5 function can limit p53 expression in nance DNA methyltransferase DNMT1. Mol Cell Biol resolution chromosome 3p allelotyping of human lung human breast tumours. Nature 2000;405:974–8. 2005;25:11047–58. cancer and preneoplastic/preinvasive bronchial epithe- 41. Asatiani E, Huang WX, Wang A, et al. Deletion, 49. Ding L, Erdmann C, Chinnaiyan AM, Merajver SD, lium reveals multiple, discontinuous sites of 3p allele methylation, and expression of the NKX3.1 suppressor Kleer CG. Identification of EZH2 as a molecular marker loss and three regions of frequent breakpoints. Cancer gene in primary human prostate cancer. Cancer Res for a precancerous state in morphologically normal Res 2000;60:1949–60. 2005;65:1164–73. breast tissues. Cancer Res 2006;66:4095–9. 35. Tomizawa Y, Sekido Y, Kondo M, et al. Inhibition 42. Sharma K, Sheng HZ, Lettieri K, et al. LIM 50. Varambally S, Dhanasekaran SM, Zhou M, et al. The of lung cancer cell growth and induction of apoptosis homeodomain factors Lhx3 and Lhx4 assign subtype polycomb group protein EZH2 is involved in progres- after reexpression of 3p21.3 candidate tumor suppres- identities for motor neurons. Cell 1998;95:817–28. sion of prostate cancer. Nature 2002;419:624–9.

www.aacrjournals.org 7947 Cancer Res 2006; 66: (16). August 15, 2006

Downloaded from cancerres.aacrjournals.org on September 28, 2021. © 2006 American Association for Cancer Research. MIRA-Assisted Microarray Analysis, a New Technology for the Determination of DNA Methylation Patterns, Identifies Frequent Methylation of Homeodomain-Containing Genes in Lung Cancer Cells

Tibor Rauch, Hongwei Li, Xiwei Wu, et al.

Cancer Res 2006;66:7939-7947.

Updated version Access the most recent version of this article at: http://cancerres.aacrjournals.org/content/66/16/7939

Supplementary Access the most recent supplemental material at: Material http://cancerres.aacrjournals.org/content/suppl/2006/08/16/66.16.7939.DC1

Cited articles This article cites 50 articles, 14 of which you can access for free at: http://cancerres.aacrjournals.org/content/66/16/7939.full#ref-list-1

Citing articles This article has been cited by 31 HighWire-hosted articles. Access the articles at: http://cancerres.aacrjournals.org/content/66/16/7939.full#related-urls

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Subscriptions Department at [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://cancerres.aacrjournals.org/content/66/16/7939. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.