Division of Human Cancer Genetics, Department of Molecular Virology, Immunology and Medical Genetics, The Ohio State University, Columbus, Ohio, OH 43210, USA

Correspondence to: D J Smiraglia, E-mail: Smiraglia.1@postbox.acs.ohio-state.edu

Restriction landmark genomic scanning (RLGS) has been used to study DNA methylation in cancer for nearly a decade. The strong bias of RLGS for assessing the methylation state of CpG islands genome wide makes this an attractive technique to study both hypo- and hypermethylation of regions of the genome likely to harbor genes. RLGS has been used successfully to identify regions of hypomethylation, candidate tumor suppressor genes, correlations between hypermethylation events and clinical factors, and quantification of hypermethylation in a multitude of malignancies. This review will examine the major uses of RLGS in the study of aberrant methylation in cancer and discuss the significance of some of the findings.

Oncogene (2002) 21, 5414-5426. doi:10.1038/sj.onc.1205608

DNA methylation; RLGS; cancer; genomic scanning; hypomethylation

Introduction

Restriction landmark genomic scanning (RLGS) is a highly reproducible two-dimensional gel electrophoresis of genomic DNA that allows the assessment of over 2000 loci simultaneously (Hatada et al., 1991; Okazaki et al., 1995). Examples of RLGS profiles are shown in Figure 1. The technique has been used for various purposes including genetic mapping, identification of novel imprinted genes, genomic amplifications, regions of hypomethylation, regions of hypermethylation, candidate tumor suppressor genes, and measuring the degree of CpG island hypermethylation in cancer (Costello et al., 2000a; Hayashizaki et al., 1993; Miwa et al., 1995; Okazaki et al., 1996; Okuizumi et al., 1995; Plass et al., 1996; Shibata et al., 1995). Two primary observations have been made concerning DNA methylation in cancer: (1) genome wide hypomethylation with an overall decrease in the level of 5-methylcytosine, and (2) hypermethylation of CpG islands (Costello and Plass, 2001 and references therein). In other words, regions of the genome that are normally heavily methylated become hypomethylated, while at the same time, regions of the genome normally unmethylated, become hypermethylated. A variety of molecular techniques have been developed to detect methylation changes in cancer through both global and gene-by-gene approaches (Table 1). RLGS can be used to study both aberrant phenomena, although the real power of RLGS lies with the study of hypermethylation. This review will focus on the use of RLGS to study aberrant DNA methylation in cancer.

The RLGS technique

The technical aspects of RLGS are fairly straightforward conceptually and described in detail elsewhere (Hatada et al., 1991; Hirotsune et al., 1992; Okazaki et al., 1995; Smiraglia et al., 1999). The RLGS technique is outlined in Figure 2. High molecular weight DNA is digested with the 'landmark' enzyme. The landmark enzyme determines the sites of the genome that will be labeled by filling in the enzyme half-site with radioactive nucleotides and thus is responsible for the pattern visualized on autoradiography. Using a methylation sensitive landmark enzyme such as NotI or AscI is what gives RLGS its methylation scanning abilities (sometimes referred to as RLGS-M). Only unmethylated sites can be cut and labeled and therefore the methylated sites will not contribute to the two dimensional pattern seen upon autoradiography. RLGS profiles are quantitative, with the intensity of signal in any spot representative of the number of labeled landmark enzyme half-sites migrating to that position of the gel. Repetitive sequences such as rDNA fragments that contain the unmethylated landmark enzyme site appear as enhanced spots, while the majority of RLGS spots, which arise from diploid sequences, appear as 'normal' spot intensity and haploid sequences appear half as intense (Asakawa et al., 1995).

There are some limitations to the RLGS technique. The most critical consideration for using RLGS is DNA quality. High quality DNA must be isolated with a minimum amount of mechanical shearing and must come from a source with minimal degradation. DNA prepared from archival sources of tissue cannot be used for RLGS. This requires that most studies be done prospectively since fresh-frozen tissue is required for high quality DNA. In addition, although RLGS is high-throughput in terms of the number of loci assessed in a single experiment, it is not high-throughput with regard to the number of samples that can be studied in a short period of time.

RLGS analysis in cancer

In order to use RLGS to study DNA methylation in cancer, tumor DNA profiles must be compared to a 'normal' RLGS profile to identify differences between the two. It is important to consider what source of DNA should be used to produce the 'normal' profile for comparison with any one particular tumor profile. Is consideration of tissue type or of the individual donor source more important to ensure that profile differences really represent methylation changes attributable to the cancer? A comparison of brain and liver tissues taken from the same mouse showed that only 10 out of 2600 (0.38%) RLGS fragments differed between the two tissues and these differences were due to DNA methylation (Watanabe et al., 1995). In another study comparing the RLGS profiles from 13 different human tissues with all comparisons between tissues from the same donor, methylation differences were detected in only an average of 0.26% of RLGS fragments (Smiraglia, unpublished observations).

In contrast, RLGS profile comparison between the same tissues from different individuals revealed that 4.4% of the RLGS fragments are different and these differences are due to sequence polymorphisms (Asakawa et al., 1995). It was further estimated that approximately 13% of RLGS fragments are polymorphic, although only 4.4% vary between any two individuals. Thus, these data show that the tissue used as the 'normal' RLGS profile, to which the tumor profile is compared, is a less critical consideration than having the profile come from DNA from the same patient. Therefore, in such a comparison it is highly desirable to compare the tumor profile to normal adjacent tissue or peripheral blood lymphocyte (PBL) profiles to ensure that RLGS pattern differences observed are not due to genetic polymorphism. In cases where normal DNA is not available from the same patient, RLGS profiles are compared to four other normal profiles and differences in common between all four comparisons are considered to be cancer related differences as opposed to genetic polymorphism.

RLGS fragment cloning

In order to identify the genomic targets of hypo- or hypermethylation in cancer, the RLGS fragments that show differences between tumor profiles and normal profiles must be cloned. The amount of DNA present in an RLGS profile diploid spot is estimated to be in the 10^-15 M range. Elution and direct cloning of such low amounts of DNA proved to be very difficult, and therefore, early RLGS fragment cloning was limited to the high copy number RLGS fragments (Costello et al., 1997; Kuick et al., 1996). Technical improvements, such as use of the restriction trapper and PCR mediated cloning methods (Hirotsune et al., 1993; Ohsumi et al., 1995a; Suzuki et al., 1994), have allowed for the cloning of some diploid RLGS fragments, however these methods remain inefficient and have inherent technical problems.

The cloning of RLGS fragments from human profiles was greatly improved by the creation of an arrayed NotI/EcoRV boundary library (Plass et al., 1997; Smiraglia et al., 1999). A portion of this library was mapped to a normal RLGS profile by systematic mixing of arrayed library clones with genomic DNA to create RLGS mixing gel profiles. RLGS fragments present within the pool of clones mixed into the genomic DNA exhibit enhanced spot intensities (Figure 1b,d). This strategy has produced a significant increase in the number of cloned RLGS fragments and greatly reduced (but not completely eliminated) limitations on RLGS fragment identification. One of the limitations of such an approach is that the cloning capability is restricted to RLGS fragments identified using NotI as the landmark enzyme. It is estimated that a NotI RLGS profile covers approximately 10% of the CpG islands in the genome. In order to study a larger portion of the genome it is desirable to use other landmark enzymes as well. AscI is another enzyme that can be used as a landmark enzyme for RLGS (Figure 1c) that is also methylation sensitive. Analysis of the publicly available human genome sequence shows that only 2.8% of CpG islands contain both a NotI and an AscI site. Consequently, by expanding the RLGS analysis of a tumor sample to include AscI as the landmark enzyme, the number of loci sampled nearly doubles. To take full advantage of this, an arrayed AscI/EcoRV library and mixing gel cloning tool has recently been created (Figure 1d) (Dai et al submitted). This cloning method works the same as for the NotI/EcoRV library.

For a NotI/EcoRV/HinfI RLGS profile, the migration of each RLGS fragment in the first dimension is determined by the size of the NotI/EcoRV fragments (and the NotI/NotI fragments without an internal EcoRV site). In the second dimension, these fragments are further digested in gel with HinfI and thus the migration in the second dimension is determined by the NotI/HinfI fragment sizes. Note that only the NotI half-site is labeled and so only the NotI/HinfI fragments are ultimately visualized. The size dependent migration of these fragments in both dimensions can be predicted based upon empirical data. Thus it should be possible to use bioinformatics to predict what an RLGS profile should look like based upon the publicly available human genome sequence. Such an approach could be used as a new RLGS fragment cloning strategy.

Rouillard et al. (2001) have recently developed bioinformatics tools that can create a 'Virtual Genome Scan' or virtual RLGS profile (http://dot.ped.med.umich.edu:2000/vgs/index.html). This is accomplished by identifying all the NotI/EcoRV (and the NotI/NotI fragments without an internal EcoRV site) fragments in the human genome databases, and then all the NotI/HinfI fragments within this subset of the genome and predicting their migration in both dimensions based upon fragment size. This creates a virtual RLGS pattern that resembles experimentally produced patterns. The great benefit of such a virtual profile is that each virtual spot is pre-defined by a specific sequence in the database, and therefore already 'cloned'. This tool was recently reported to have correctly predicted the location of 22 out of 29 previously cloned RLGS fragments within a specified window of probability in the first and second dimensions. In order to clone a single RLGS spot of interest by this method, multiple candidate sequences fall into the window of highest probability depending on the local spot density of the profile and the correct candidate must be determined. In cases where the chromosome from which the spot of interest arises is known (Yoshikawa et al., 1996), then the number of candidates can be greatly reduced. The current utility of this tool is limited by completeness of the human genome sequence, a need for improved fragment migration algorithms, and the ability to warp the predicted pattern of spots to match the experimental patterns. It seems likely that these factors can and will be improved in the near future and this will prove to be an extremely useful tool.

RLGS preferentially scans CpG islands

The choice of landmark enzyme is a critical factor in RLGS, not only for methylation sensitivity but also for their basic sequence characteristics. Both NotI and AscI have GC-rich recognition sequences of eight base pairs with two CpG dinucleotides each (GCGGCCGC and GGCGCGCC, respectively). These GC-rich recognition sequences are preferentially found within CpG islands. Table 2 shows a list of 70 RLGS fragments that have been cloned from the NotI profile in association with various tumor projects (Dai et al., 2001; Fruhwald et al., 2001b; Rush et al., 2001; Smiraglia et al., 2002; and submitted). Sixty-eight out of 70 (97%) of these RLGS fragments have the sequence characteristics of CpG islands (Gardiner-Garden and Frommer, 1987). Thirty-six of these RLGS fragments show identity to known genes, hypothetical proteins, or cDNAs, one shows identity to a pseudogene, two are homologous to mouse genes and 11 show identity to ESTs. Where possible to clearly ascertain the genomic context, the RLGS fragments representing CpG islands were found in the 5' end of the genes in 26 out of 35 (74%) cases, the middle of the genes in six out of 35 cases, and the 3' end of the genes in three out of 35 cases. In addition, in a set of 210 NotI/EcoRV clones, 186 of which were randomly picked from the boundary library, 94% had the sequence characteristics of CpG islands. Similarly, 15 out of 17 (88%) AscI RLGS profile spots have recently been cloned and found to have CpG island sequence characteristics (Dai et al submitted), while 66 out of 69 (96%) of randomly sequenced AscI/EcoRV clones are in CpG islands. These data unequivocally demonstrate that the NotI and AscI human RLGS profiles produce a scan of loci highly enriched for CpG islands.

Hypomethylation studies

The majority of CpG dinucleotides in the genome are found outside of CpG islands in highly and moderately repetitive DNA. The 5-methylcytosine content of unique sequence DNA is very low in comparison to the moderately and highly repetitive DNAs such as the EcoRI family and Alu family repeats (Ehrlich et al., 1982; Gama-Sosa et al., 1983a,c). One of the earliest observations concerning aberrant methylation in cancer was the finding of global hypomethylation represented by both hypomethylation of specific loci and a decrease in the total 5-methylcytosine content of the genome (Feinberg and Vogelstein, 1983; Gama-Sosa et al., 1983b). Specifically, the juxtacentromeric DNA of chromosomes 1, 16 and 9 appear to be extremely susceptible to hypomethylation, which may lead to chromosomal instability (Narayan et al., 1998; Qu et al., 1999).

RLGS fragments that show increased intensity or newly appear in tumor profiles as compared to normal can be the result of hypomethylation of the NotI site (Hayashizaki et al., 1993). In one of the first such applications of RLGS, a 200 copy, 13 kb repeat unit from chromosome 8q21 was cloned from a human RLGS profile (Miwa et al., 1995). This RLGS fragment was not seen in normal tissue profiles but was found at high intensity (representing high copy number) in six out of six melanoma cell lines, two out of five colon carcinoma lines, one out of six pancreatic cancer lines, and one out of one primary melanoma. Cloning of this fragment and subsequent Southern blot analysis demonstrated that in normal tissues the NotI sites in this repeat unit were completely methylated, but in the cancer and cell lines, varying degrees of hypomethylation of the repeat had occurred. In a set of 19 neuroblastomas studied by RLGS, three separate spots showing increased intensity were cloned and found to be from repetitive elements containing NotI sites, which were methylated in normal tissues but became hypomethylated to varying degrees in neuroblastomas (Thoraval et al., 1996; Wimmer et al., 1996). Similar observations were made in prostate cancer and hepatocellular carcinomas (Konishi et al., 1996, 1997; Nagai et al., 1998, 1999a,b).

The finding that many RLGS fragments of stronger than diploid intensity in tumor and cancer cell line profiles results from hypomethylation of repetitive elements containing NotI sites is in agreement with the general observations of global hypomethylation in tumors and cell lines since the majority of 5-methylcytosine is found in highly and moderately repetitive DNA. However, not all RLGS fragments of increased intensity represent hypomethylation of repetitive DNA. In some instances, genomic amplifications of regions that contain NotI sites have been identified, such as CDK6 amplification in human gliomas (Costello et al., 1997), MYCN amplification in neuroblastomas (Wimmer et al., 1997) and medulloblastomas (Smiraglia et al., 1999), and amplification of 22q11 encompassing the BCR gene in the chronic myeloid leukemia cell line K-562 (Rush et al., 2002).

In addition to hypomethylation causing greatly increased intensity of RLGS spots, it can also be responsible for the presence of a diploid intensity spot not seen in the corresponding normal tissue profile. Although less frequently reported, the presence of diploid intensity spots uniquely in tumor profiles may be the result of hypomethylation of single copy sequences. Such events, while perhaps less abundant than hypomethylation of repetitive elements, may have very significant biological effects. One such scenario is at imprinted loci where a CpG rich region normally exhibits allele specific methylation, which is critical for imprinted expression of the affected gene(s). In human bladder cancer the sixth CTCF binding site at the H19/IGF2 imprinted locus is normally methylated on the paternal locus only, but becomes hypomethylated in the cancer (Takai et al., 2001). This hypomethylation event in bladder cancer results in loss of imprinted expression of H19 (biallelic expression of H19); conversely in Wilms tumor and colon cancer, hypermethylation of the maternal allele is seen (Frevel et al., 1999; Nakagawa et al., 2001).

Hypermethylation studies

With regard to identification of fragments, early cancer based RLGS studies were limited in scope to highly represented fragments such as the hypomethylation of repeats and amplifications described above because of difficulty in cloning RLGS fragments. Nevertheless, it was apparent that spot loss, as opposed to newly appearing spots or intensification of spots, was the predominant type of change that was seen in tumor or cell line RLGS profiles. Kawai et al. (1994) studied five mouse cell lines and compared their profiles to those from normal mouse tissues from the same strain. In the 10T1/2 fibroblast line and three SV40 large T antigen transformed lines from E13.5 mouse telencephalon cells, a range of 5-14% of RLGS spots were lost and this was 10-14 times greater than the percentage of newly appearing spots. Subsequent to cloning of seven of these RLGS fragments, Southern blot analysis confirmed that RLGS spot loss was due to hypermethylation of the NotI site (Watanabe et al., 1995). In another study assessing RLGS spot loss in 98 human primary neoplasias, 26 RLGS fragments were cloned and demonstrated by Southern blot analysis to represent hypermethylation events (Costello et al., 2000a). For over 130 loss events examined (the summed number of tumor profiles where loss was detected for each of the 26 RLGS fragments), complete concordance was observed between the detection of loss or decreased intensity on the RLGS profiles and methylation of the NotI site assessed by Southern blotting. These studies demonstrated that in both cell lines and tumors the most prevalent change found in RLGS profiles using NotI as the landmark enzyme is spot loss representing hypermethylation of the NotI sites.

Identification of candidate tumor suppressor genes

Hypermethylation of CpG islands has been identified in the promoters of a number of tumor suppressor genes (TSG) (Baylin et al., 1998, 2001; Herman and Baylin, 2000; Herman et al., 1994; Jones, 2001; Jones and Gonzalgo, 1997). DNA methylation is capable of silencing the promoter of a TSG through mechanisms that may include direct interference with transcription factor binding (Bird and Wolffe, 1999; Robertson and Jones, 2000) or through induction of alterations in chromatin structure resulting in a transcriptionally silent conformation (Bird and Wolffe, 1999; Nan et al., 1997, 1998; Ng and Bird, 1999; Nguyen et al., 2001). In addition, it has recently been demonstrated that in tumors with germline mutations in TSGs, DNA methylation of the normal allele can be the second hit required for tumor development and progression (Esteller et al., 2001; Grady et al., 2000). Thus, given that promoter hypermethylation of TSGs is associated with silencing of the genes, either biallelicly, or in combination with genetic events, it seems reasonable to predict that by scanning the genome for common hypermethylation events, perhaps new TSGs could be identified.

Proof of this principal was demonstrated using a mouse transgenic model for liver tumorigenesis (Akama et al., 1997). The C57BL/6 (B6) inbred mouse strain was used to make a transgenic line carrying the SV40 large T antigen driven by the mouse major urinary protein (MUP) promoter and enhancer (Held et al., 1989; Ohsumi et al., 1995b). This transgenic line spontaneously developed multifocal hyperplasia and hepatocellular carcinoma after SV40-T antigen independent secondary events (Schirmacher et al., 1991). RLGS analysis identified 24 spots demonstrating loss in greater than 75% of the 30 tumors studied and 13 of these RLGS fragments were cloned. Four genes were identified by sequence analysis: p16/INK4a (p16), 4-integrin (Akama et al., 1997), Mac25/insulin like growth factor binding protein-7 (Igfbp-7) (Komatsu et al., 2000), and a novel member of the Snail/Gfi-1 repressor family, Mlt-1 (Tateno et al., 2001). All four genes showed hypermethylation of CpG islands in their 5' ends in greater than 89% of the liver tumors studied. p16 is a well characterized TSG previously reported to undergo transcriptional silencing by hypermethylation (Baylin et al., 1998; Merlo et al., 1995). 4-integrin expression has been shown to be important in inhibiting metastasis in multiple tumor types (Holzmann et al., 1998). Mac25/Igfbp-7 is thought to modulate insulin-like growth factor binding to its receptor leading to growth inhibition and has been demonstrated to be down regulated in association with breast carcinoma progression (Burger et al., 1998). Mlt1 is a newly identified member of the transcriptional repressor protein SNAG family and is thought to be involved in growth suppression (Tateno et al., 2001). Furthermore, for both Mac25/Igfbp-7 and Mlt1 hypermethylation in the liver tumors correlated with lack of expression, and Mlt1 expression could be restored following 5-aza-2'-deoxycytidine treatment in a neuroblastoma cell line in which this gene is normally methylated and not expressed (Komatsu et al., 2000; Tateno et al., 2001). These studies demonstrate the utility of RLGS for the identification of novel targets of hypermethylation and that these targets may represent TSGs or tumor related genes.

In human hepatocellular carcinomas (HCC), 27 RLGS spots were identified that demonstrated significant loss of intensity as compared to normal profiles (Nagai et al., 1994). One of these spots found to be lost in 14 out of 16 (88%) of HCC was cloned and identified as the SOCS1 gene, which is a negative regulator of the JAK/STAT pathway (Nagai et al., 1994; Yoshikawa et al., 2001). The SOCS1 gene had previously been shown to inhibit the biological effects of cytokines but had not been implicated in carcinogenesis. Thus, given its high rate of RLGS fragment loss in HCC and its involvement in negatively regulating the JAK/STAT pathway, dysregulation of which has been implemented in carcinogenesis, SOCS1 made for an attractive candidate tumor suppressor gene inactivated by hypermethylation. Methylation analysis of the CpG island located at the 5' end of this gene demonstrated hypermethylation of this CpG island in primary HCC and HCC cell lines, but not in normal tissues, and this hypermethylation correlated with lack of expression. The functional significance of SOCS1 methylation related transcriptional silencing was explored and strong correlation with constitutive phosphorylation of JAK2 and phosphorylation of STAT3 was found. These observations are consistent with loss of function of SOCS1 and associated with the methylation mediated repression of this gene. In cell lines where SOCS1 is methylated and the JAK/STAT pathway is constitutively active, the chemical inhibitor of JAK2, AG490, significantly reduced colony formation. The reduced colony formation, in both monolayers and soft-agar growth, was also seen upon transient transfection of an expression vector containing SOCS1 (Yoshikawa et al., 2001). These studies provide an excellent example of using RLGS to identify a candidate TSGs followed by further study of the gene whose loss of function is consistent with a tumor suppressor role.

The 5' end CpG island of the human BMP3B gene was found to be methylated by RLGS analysis in three out of 16 primary non-small cell lung cancers (NSCLC) (Dai et al., 2001) and lower level methylation was found in an additional two tumors by Southern blot. RT-PCR analysis revealed that six out of six tumors studied showed no or reduced expression of BMP3B. Interestingly, this included two tumors that did not show methylation of the locus and raises the exciting possibility that in these tumors, expression of BMP3B has been inactivated through other means. Three hypermethylated cell lines did not express, but after 5-aza-2'-deoxycytidine treatment expression was induced. Aside from the observation that this gene is methylated in the tumors and that this methylation appears to be responsible for the lack of expression since it can be restored in cell lines following treatment with an de-methylating agent, two other observations make this gene a good candidate TSG. Although the specific function of this gene is unclear, other genes in the BMP family have been shown to induce apoptosis during organ development (Merino et al., 1999). Treatment of the lung cancer cell line A549 with BMP2 resulted in loss of transformed phenotype and restoration of microfilament organization (Tada et al., 1998). In addition, the BMP3B gene is located on chromosome 10q11.21-11.23 which is a region of loss of heterozygosity (LOH) in 20-30% of NSCLC and 51% small cell lung cancer (Knuutila et al., 1999). Thus, given that two modes of inactivation have now been identified for BMP3B (promoter hypermethylation and LOH) and the predicted gene function is involved with growth regulation and/or apoptosis, this gene is a strong candidate TSG that requires further study.

Development of biomarkers

Biomarkers can be used for early detection of cancer, prognostic indication of the course of the cancer and/or response to therapy, or to sub-classify tumor types. In a recent study by Lee et al. (2000), 70 patients with oral leukoplakia (a pre-malignant lesion that may develop into head and neck squamous cell carcinoma (HNSCC)) were followed for 10 years following a chemopreventative trial (Lee et al., 2000). In the 22 patients that developed cancer, the investigators identified a panel of three molecular biomarkers (polysomy, high p53 expression, and LOH at either 3p or 9p) that were highly significant in predicting cancer. Importantly, the combination of these three markers was more predictive than any one of them alone (Lee et al., 2000). These data demonstrate a proof-of-principle that sets of biomarkers, not single markers, may have high predictive value in measuring cancer risk and response to therapy. Thus it is clear that ongoing identification and development of molecular biomarkers is crucial.

The use of DNA methylation events as diagnostic markers for certain tumor types or stages has tremendous potential. Belinsky et al. (1998) showed that promoter methylation is an early event in lung cancer and proposed the use of p16 methylation as a biomarker for early detection and monitoring of prevention trials. Fujimoto et al. (2000) found that hypermethylation of the pS2 gene was correlated with intestinal metaplasia, a precursor to intestinal type gastric carcinoma. Similarly, Esteller et al. (2000) reported that methylation of the promoter of the DNA-repair gene, MGMT, that inhibits tumor cell killing by alkylating agents could be a useful prognostic marker for the responsiveness of gliomas to these agents. An interesting study by Herman et al. (1997) identified distinct methylation patterns of p15 and p16 that characterize major types of hematological malignancies.

Methylated DNA was detected in the serum (Sanchez-Cespedes et al., 2000) and the saliva (Rosas et al., 2001) of HNSCC patients by methylation sensitive PCR (MS-PCR) of p16, MGMT, and DAP-kinase. The concordance between methylation of any of these genes in the tumor and also in the serum or saliva was 42% and 65%, respectively. These two studies demonstrate ways in which potential molecular biomarkers based on DNA methylation may be applied.

RLGS analysis has been used to study specific methylation events and to determine if any of them could be correlated with clinical factors. In human HCC, three commonly used prognostic indicators for recurrence are size of the tumor, portal venous invasion, and intrahepatic metastases. However, it is not uncommon for recurrence to occur in patients who are negative for all three of these factors. After RLGS analysis was performed on 31 HCC tumors and corresponding normal tissues, correlations were studied between RLGS profile changes and post-operative recurrence (Itano et al., 2000). Post-operative recurrence correlated significantly with 16 or more RLGS spot changes including both hypo- and hypermethylation events. Thirteen out of 16 patients whose tumors showed 16 or greater changes in their RLGS profiles had recurrence, while only two out of 15 patients whose tumors showed 12 or less changes recurred. Furthermore, the number of RLGS profile changes was more predictive than the three previously used prognostic indicators (Itano et al., 2000).

In human medulloblastomas, up to approximately 1% of RLGS fragments were found to be hypermethylated (Fruhwald et al., 2001b). Seven RLGS fragments showed significant or nominally significant negative correlation with survival upon Cox regression proportional hazard analysis. Methylation of any one of these seven RLGS fragments was a predictor of poor survival time. Hypermethylation of another fragment was actually correlated with improved survival time (Fruhwald et al., 2001b). These eight fragments were methylated at a frequency ranging from 45 to 14% of the 22 tumors studied, with four of the eight fragments methylated in greater than 40%. Thus, these hypermethylated loci have good potential to be developed as biomarkers prognostic for survival time.

The WIT1 gene was identified as a hypermethylated CpG island in a relapsed case of acute myeloid leukemia (AML). WIT1 is located approximately 1 kb upstream of the Wilms tumor suppressor gene (WT1) (Plass et al., 1999). Further analysis of the WIT1 gene by Southern blot analysis demonstrated that 37% of patients who responded to induction chemotherapy had hypermethylation of WIT1 (Plass et al., 1999). For nine cases, methylation was measured at relapse and was detected in five cases. Interestingly, eight refractory AML patients (patients who do not respond to induction chemotherapy) were studied and seven showed hypermethylation. This represents a statistically significant increase in hypermethylation frequency of WIT1 in refractory leukemia compared to chemosensitive AML (Plass et al., 1999).

Quantification of hypermethylation phenotypes

Arguably, one of the most powerful uses of RLGS in cancer is to quantify the degree to which the cancer genome has become hypermethylation at CpG islands. Given that 97% of the RLGS fragments that have been cloned represent CpG islands (Costello et al., 2000a; Dai et al., 2001; Rush et al., 2001; Smiraglia et al., 2002; and submitted) and that in the majority of cases studied thus far RLGS spot loss correlates with hypermethylation of the CpG island (Costello et al., 2000a; Watanabe et al., 1995), RLGS is uniquely suited for this purpose. RLGS assesses the methylation state of well over 1000 CpG islands in a single profile from approximately 1 g of DNA. Importantly, this method assesses essentially the exact same set of CpG islands in each case and therefore the negative results (lack of spot loss) are quantifiable. Consequently, the baseline from which hypermethylation is being measured is clearly set in each case.

RLGS has been used for measuring the amount of CpG island hypermethylation by various groups for multiple tumor types including cell lines (Kawai et al., 1994; Kim et al., 2000; Smiraglia et al., 2001), brain (Costello et al., 2000a,b; Fruhwald et al., 2001a,b; Nakamura et al., 1997; Wimmer et al., 1997), HCC (mouse and human) (Akama et al., 1997; Itano et al., 2000), AML (Costello et al., 2000a; Rush et al., 2001), lung (Dai et al., 2001), prostate (Konishi et al., 1996, 1997), HNSCC (Costello et al., 2000a; Smiraglia et al submitted), testicular germ cell tumors (Costello et al., 2000a; Smiraglia et al., 2002), breast, and colon (Costello et al., 2000a). These studies have clearly shown that hypermethylation of CpG islands on a genome wide scale is a common feature of neoplastic cells.

A study of 98 different tumor RLGS profiles from seven different types of cancer demonstrated the degree and variability of CpG island hypermethylation in cancer (Costello et al., 2000a). In addition, a set of 16 non-small cell lung tumors was also analysed (Dai et al., 2001). Breast, HNSCC, primitive neuroectoderm tumors (PNET) and testicular germ cell tumors showed relatively low levels of hypermethylation while colon, glioma, AML and lung showed relatively high levels of hypermethylation. Statistically significant differences in methylation frequency (defined here as the percentage of analysed RLGS spots lost compared to normal) were demonstrated among the tumor types shown in Table 3. Also, within each individual tumor type there was demonstrated a non-random heterogeneity in methylation frequencies. So even within the set of AML profiles analysed, which overall had a relatively high frequency of methylation; there were patients with 0% methylation, as well as patients with nearly 10% methylation. These data suggest that different tumor types have a different overall likelihood of having a high degree of CpG island hypermethylation; nevertheless individual tumors may have widely variable levels of hypermethylation within the same group of tumors. This leads to the intriguing idea that tumors of the same group could be sub classified into high methylating and low methylating tumors. Such an approach was demonstrated in HCC as described above where high methylating tumors had a significant increase in recurrence rate (Itano et al., 2000).

In a comparison of the above data set with cell lines representing each of the malignancies, a significant increase in the overall level of methylated CpG islands was demonstrated (Smiraglia et al., 2001). The degree of methylation in three cell lines representing each type of cancer was more similar to cell lines of the same type than to cell lines of other types. Furthermore, although the degree of methylation in all cell lines was greatly amplified compared to the primary malignancies, the effect was relative with the lowest methylating primary malignancies represented by the lowest methylating cell lines, and vice versa (Smiraglia et al., 2001). Hence, the cell lines have retained some characteristics of their tumor type of origin in relation to the degree of hypermethylation, despite an overall increase in cell line hypermethylation. These observations suggest that particular cell types and their corresponding malignancies may have predilections or limitations to the degree of aberrant methylation that will arise and that these are relatively maintained in the cancer cell lines. Thus, the provocative question of why leukemia cell lines methylate up to 48% of RLGS fragments while HNSCC cell lines methylate only up to 10% arises. These differences may be due to differing capacities for hypermethylation, differing tolerance of hypermethylation, different accessibility of the genomes to the machinery responsible for hypermethylation, or different underlying genetic defects.

Non-random RLGS fragment methylation

These studies also address the targets of hypermethylation. It is clear that the CpG island methylation observed by RLGS is not random (Costello et al., 2000a; Dai et al., 2001; Fruhwald et al., 2001b; Rush et al., 2001; Smiraglia et al., 2001) as numerous RLGS fragments are hypermethylated at a frequency that rules out random chance. However, this demonstrates only that the output (the tumors) results in a non-random distribution of hypermethylation events and does not necessarily imply that the hypermethylation process itself is non-random. Non-random output arising from a random process could be explained by positive selection of the CpG islands commonly methylated in malignancies. Such a scenario is logical and has recently been supported by the finding that in hereditary forms of colon cancer and breast cancer, the second hit on the non-mutated allele of the MLH1 and BRCA1 genes can be methylation (Esteller et al., 2001), and in hereditary diffuse gastric cancer methylation of the non-mutated allele of CDH1 has been demonstrated (Grady et al., 2000). However, methylation is never observed on the mutated allele when the normal allele is deleted. These observations argue strongly that positive selection for methylation of specific CpG islands does occur. Conversely, it is reasonable to suggest that for many genes, hypermethylation of the CpG island is negatively selected. Hypermethylation of housekeeping genes performing vital cellular function would likely be lethal and so both positive and negative selective pressures probably influence the hypermethylation events found in the cancer cells.

The best understood consequence of CpG island hypermethylation is transcriptional silencing. However, RLGS has shown that up to nearly 10% of CpG islands can be methylated in primary neoplasias. The mean level of CpG island methylation in both colon tumors and gliomas is greater than 3% (Table 3). Assuming 29 000 CpG islands in the genome (Lander et al., 2001; Venter et al., 2001) then this might represent silencing of 870 genes, or as many as 2900 genes. It seems unreasonable to suggest that so many genes, expressed and unmethylated in normal tissues, become methylated and not expressed in neoplastic tissues. How then can this observation be explained? First, not all unmethylated CpG island promoters need be in expressed genes; tissue specific genes often have unmethylated CpG island promoters in non-expressing tissues (Bird, 2002; Warnecke and Clark, 1999). Secondly, not all CpG islands are found in the promoter regions of genes (Larsen et al., 1992) and methylation of these CpG islands generally does not correlate with transcriptional repression (Gonzalgo et al., 1998).

These possibilities argue that a subset of the hypermethylation events detected by RLGS are not correlated with gene silencing either because the gene is already silent in the studied tissue or the CpG island methylation is not in the promoter region and therefore does not affect transcription. This, however, makes it difficult to reconcile the non-random output of methylation solely by proposing positive and/or negative selection. Such events could perhaps be explained by a lack of negative selection; they do no harm to cells' survival chances and therefore are able to come along for the ride. The fact that this distribution is non-random may then simply be due to the fact that in any one particular cell type there are a limited number of CpG islands for which hypermethylation has neither a positive nor negative selective advantage.

Alternatively, different regions of the genome may have differential susceptibility to CpG island hypermethylation. Repetitive DNA sequences have been suggested as triggers for hypermethylation normally in the genome. This is implied by the observation that highly and moderately repetitive elements are typically methylated and contribute the majority of 5-methylcytosine in the genome (Gama-Sosa et al., 1983c; Turker and Bestor, 1997; Walsh and Bestor, 1999). In the mouse Aprt gene, tandem B1 elements are required for de novo methylation (Yates et al., 1999) and at imprinted loci such as the mouse Rasgrf1 gene short tandem repeats are required for differential methylation of the alleles (Yoon et al., 2001). Knockout of these repeats at the Rasgrf1 locus resulted in lack of methylation of the differentially methylated region regardless of maternal or paternal germline transmission (Yoon et al., 2001). These observations demonstrate that cis acting elements are involved in at least some instances of CpG island methylation in normal cells. This raises the possibility that loci with these types of cis acting elements, which are unmethylated in normal cells, are hyper-susceptible to de novo methylation in neoplastic cells. Perhaps then, various regions of the genome have different degrees of susceptibility to de novo methylation based upon cis acting elements or possibly chromatin context. In such a scenario, while the de novo methylation occurring in neoplastic cells may not be specifically directed to certain CpG islands, different regions of the genome may be more likely than others to become de novo methylated. Subsequently, both positive and negative selective pressure may combine with this proposed susceptibility difference to ultimately produce the non-random hypermethylation profiles detected by RLGS. Hypermethylation of CpG islands that do not have an apparent effect on gene expression may be the result of high susceptibility to de novo methylation, but neither positive nor negative effects on growth or survival at these loci.

Concluding remarks

Whether or not aberrant methylation in a cancer cell fits within the classic definition of genomic instability is open to debate. Nevertheless, it is clear now from both studies of specific target genes and genomic scanning approaches that aberrant DNA methylation (both hypo- and hyper-) contributes significantly and meaningfully to the milieu of 'things-that-are-wrong' with the genomes of cancer cells. This has allowed for the creation of new methods to identify novel cancer related genes based on their methylation changes. In addition, it offers up a new class of potential biomarkers with diagnostic and prognostic possibilities that can be assessed purely on their methylation status regardless of whether or not the locus in question is associated with a gene. The unique ability of RLGS to quantify the hypermethylation phenotypes of malignancies presents an opportunity to study the overall contribution that CpG island hypermethylation makes to genomic abnormalities in the cancer cell genome and to carcinogenesis. Thus, in the future we may be able to determine the relative contribution of various types of genomic abnormalities such as microsatellite instability, aneuploidy, LOH, hypomethylation and hypermethylation in the progression, treatment, and outcome cancer.

Acknowledgements

The authors thank Joseph Costello, Laura Rush and Laura Smith for critical reading of the manuscript. This work was supported in part by a grant from the NIDCR, DE13123-02.

Akama TO, Okazaki Y, Ito M, Okuizumi H, Konno H, Muramatsu M, Plass C, Held WA, Hayashizaki Y. (1997). Cancer Res., 57: 3294-3299. MEDLINE

Asakawa J, Kuick R, Neel JV, Kodaira M, Satoh C, Hanash SM. (1995). Electrophoresis, 16: 241-252. MEDLINE

Balaghi M, Wagner C. (1993). Biochem. Biophys. Res. Commun., 193: 1184-1190. Article MEDLINE

Baylin SB, Esteller M, Rountree MR, Bachman KE, Schuebel K, Herman JG. (2001). Hum. Mol. Genet., 10: 687-692. MEDLINE

Baylin SB, Herman JG, Graff JR, Vertino PM, Issa JP. (1998). Adv. Cancer Res., 72: 141-196. MEDLINE

Belinsky SA, Nikula KJ, Palmisano WA, Michels R, Saccomanno G, Gabrielson E, Baylin SB, Herman JG. (1998). Proc. Natl. Acad. Sci. USA, 95: 11891-11896. MEDLINE

Bird A. (2002). Genes Dev., 16: 6-21. MEDLINE

Bird AP, Wolffe AP. (1999). Cell, 99: 451-454. MEDLINE

Burger AM, Zhang X, Li H, Ostrowski JL, Beatty B, Venanzoni M, Papas T, Seth A. (1998). Oncogene, 16: 2459-2467. MEDLINE

Clark SJ, Harrison J, Paul CL, Frommer M. (1994). Nucleic Acids Res., 22: 2990-2997. MEDLINE

Costello JF, Fruhwald MC, Smiraglia DJ, Rush LJ, Robertson GP, Gao X, Wright FA, Feramisco JD, Peltomaki P, Lang JC, Schuller DE, Yu L, Bloomfield CD, Caligiuri MA, Yates A, Nishikawa R, Su Huang H, Petrelli NJ, Zhang X, O'Dorisio MS, Held WA, Cavenee WK, Plass C. (2000a). Nat. Genet., 24: 132-138. Article MEDLINE

Costello JF, Plass C. (2001). J. Med. Genet., 38: 285-303. MEDLINE

Costello JF, Plass C, Arap W, Chapman VM, Held WA, Berger MS, Su Huang HJ, Cavenee WK. (1997). Cancer Res., 57: 1250-1254. MEDLINE

Costello JF, Plass C, Cavenee WK. (2000b). Brain Tumor Pathol., 17: 49-56.

Dai Z, Lakshmanan RR, Zhu WG, Smiraglia DJ, Rush LJ, Fruhwald MC, Brena RM, Li B, Wright FA, Ross P, Otterson GA, Plass C. (2001). Neoplasia, 3: 314-323. MEDLINE

Ehrlich M, Gama-Sosa MA, Huang LH, Midgett RM, Kuo KC, McCune RA, Gehrke C. (1982). Nucleic Acids Res., 10: 2709-2721. MEDLINE

Esteller M, Fraga MF, Guo M, Garcia-Foncillas J, Hedenfalk I, Godwin AK, Trojan J, Vaurs-Barriere C, Bignon YJ, Ramus S, Benitez J, Caldes T, Akiyama Y, Yuasa Y, Launonen V, Canal MJ, Rodriguez R, Capella G, Peinado MA, Borg A, Aaltonen LA, Ponder BA, Baylin SB, Herman JG. (2001). Hum. Mol. Genet., 10: 3001-3007. MEDLINE

Esteller M, Garcia-Foncillas J, Andion E, Goodman SN, Hidalgo OF, Vanaclocha V, Baylin SB, Herman JG. (2000). N. Engl. J. Med., 343: 1350-1354. MEDLINE

Feinberg AP, Vogelstein B. (1983). Nature, 301: 89-92. MEDLINE

Frevel MA, Sowerby SJ, Petersen GB, Reeve AE. (1999). J. Biol. Chem., 274: 29331-29340. Article MEDLINE

Fruhwald MC, O'Dorisio MS, Dai Z, Rush LJ, Krahe R, Smiraglia DJ, Pietsch T, Elsea SH, Plass C. (2001a). Genes Chrom. Cancer, 30: 38-47.

Fruhwald MC, O'Dorisio MS, Dai Z, Tanner SM, Balster DA, Gao X, Wright FA, Plass C. (2001b). Oncogene, 20: 5033-5042. MEDLINE

Fujimoto J, Yasui W, Tahara H, Tahara E, Kudo Y, Yokozaki H, Tahara E. (2000). Cancer Lett., 149: 125-134. MEDLINE

Gama-Sosa MA, Midgett RM, Slagel VA, Githens S, Kuo KC, Gehrke CW, Ehrlich M. (1983a). Biochim. Biophys. Acta, 740: 212-219.

Gama-Sosa MA, Slagel VA, Trewyn RW, Oxenhandler R, Kuo KC, Gehrke CW, Ehrlich M. (1983b). Nucleic Acids Res., 11: 6883-6894. MEDLINE

Gama-Sosa MA, Wang RY, Kuo KC, Gehrke CW, Ehrlich M. (1983c). Nucleic Acids Res., 11: 3087-3095. MEDLINE

Gardiner-Garden M, Frommer M. (1987). J. Mol. Biol., 196: 261-282. MEDLINE

Gonzalgo ML, Hayashida T, Bender CM, Pao MM, Tsai YC, Gonzales FA, Nguyen HD, Nguyen TT, Jones PA. (1998). Cancer Res., 58: 1245-1252. MEDLINE

Gonzalgo ML, Jones PA. (1997). Nucleic Acids Res., 25: 2529-2531. Article MEDLINE

Grady WM, Willis J, Guilford PJ, Dunbier AK, Toro TT, Lynch H, Wiesner G, Ferguson K, Eng C, Park JG, Kim SJ, Markowitz S. (2000). Nat. Genet., 26: 16-17. Article MEDLINE

Hatada I, Hayashizaki Y, Hirotsune S, Komatsubara H, Mukai T. (1991). Proc. Natl. Acad. Sci. USA, 88: 9523-9527. MEDLINE

Hayashizaki Y, Hirotsune S, Okazaki Y, Hatada I, Shibata H, Kawai J, Hirose K, Watanabe S, Fushiki S, Wada S. (1993). Electrophoresis, 14: 251-258. MEDLINE

Held WA, Mullins JJ, Kuhn NJ, Gallagher JF, Gu GD, Gross KW. (1989). EMBO J., 8: 183-191. MEDLINE

Herman JG, Baylin SB. (2000). Curr. Opin. Microbiol. Immunol., 249: 35-54.

Herman JG, Civin CI, Issa JP, Collector MI, Sharkis SJ, Baylin SB. (1997). Cancer Res., 57: 837-841. MEDLINE

Herman JG, Graff JR, Myohanen S, Nelkin BD, Baylin SB. (1996). Proc. Natl. Acad. Sci. USA, 93: 9821-9826. Article MEDLINE

Herman JG, Latif F, Weng Y, Lerman MI, Zbar B, Liu S, Samid D, Duan DS, Gnarra JR, Linehan WM et al. (1994). Proc. Natl. Acad. Sci. USA, 91: 9700-9704. MEDLINE

Hirotsune S, Hatada I, Komatsubara H, Nagai H, Kuma K, Kobayakawa K, Kawara T, Nakagawara A, Fujii K, Mukai T. (1992). Cancer Res., 52: 3642-3647. MEDLINE

Hirotsune S, Shibata H, Okazaki Y, Sugino H, Imoto H, Sasaki N, Hirose K, Okuizumi H, Muramatsu M, Plass C. (1993). Biochem. Biophys. Res. Commun., 194: 1406-1412. Article MEDLINE

Holzmann B, Gosslar U, Bittner M. (1998). Curr. Opin. Microbiol. Immunol., 231: 125-141.

Huang TH, Perry MR, Laux DE. (1999). Hum. Mol. Genet., 8: 459-470. Article MEDLINE

Itano O, Ueda M, Kikuchi K, Simazu M, Kitagawa Y, Aiura K, Kitajima M. (2000). Oncogene, 19: 1676-1683. MEDLINE

Jones PA. (2001). Nature, 409: 141, 143-144. Article MEDLINE

Jones PA, Gonzalgo ML. (1997). Proc. Natl. Acad. Sci. USA, 94: 2103-2105. Article MEDLINE

Kawai J, Hirose K, Fushiki S, Hirotsune S, Ozawa N, Hara A, Hayashizaki Y, Watanabe S. (1994). Mol. Cell. Biol., 14: 7421-7427. MEDLINE

Kim YS, Yoo HS, Lee KT, Goh SH, Jung JS, Oh SW, Baba M, Yasuda T, Matsubara K, Nagai H. (2000). Int. J. Oncol., 17: 297-308. MEDLINE

Knuutila S, Aalto Y, Autio K, Bjorkqvist AM, El-Rifai W, Hemmer S, Huhta T, Kettunen E, Kiuru-Kuhlefelt S, Larramendy ML, Lushnikova T, Monni O, Pere H, Tapper J, Tarkkanen M, Varis A, Wasenius VM, Wolf M, Zhu Y. (1999). Am. J. Pathol., 155: 683-694. MEDLINE

Komatsu S, Okazaki Y, Tateno M, Kawai J, Konno H, Kusakabe M, Yoshiki A, Muramatsu M, Held WA, Hayashizaki Y. (2000). Biochem. Biophys. Res. Commun., 267: 109-117. Article MEDLINE

Konishi N, Cho M, Yamamoto K, Hiasa Y. (1997). Pathol. Int., 47: 735-747. MEDLINE

Konishi N, Tao M, Nakamura M, Kitahaori Y, Hiasa Y, Nagai H. (1996). Cell. Mol. Biol. (Noisy-le-grand) 42: 1129-1135.

Kuick R, Asakawa J, Neel JV, Kodaira M, Satoh C, Thoraval D, Gonzalez IL, Hanash SM. (1996). Genetics, 144: 307-316. MEDLINE

Kuo KC, McCune RA, Gehrke CW, Midgett R, Ehrlich M. (1980). Nucleic Acids Res., 8: 4763-4776. MEDLINE

Lander ES, Linton LM, Birren B, Nusbaum C, Zody MC, Baldwin J, Devon K, Dewar K, Doyle M, FitzHugh W, Funke R, Gage D, Harris K, Heaford A, Howland J, Kann L, Lehoczky J, LeVine R, McEwan P, McKernan K, Meldrim J, Mesirov JP, Miranda C, Morris W, Naylor J, Raymond C, Rosetti M, Santos R, Sheridan A, Sougnez C, Stange-Thomann N, Stojanovic N, Subramanian A, Wyman D, Rogers J, Sulston J, Ainscough R, Beck S, Bentley D, Burton J, Clee C, Carter N, Coulson A, Deadman R, Deloukas P, Dunham A, Dunham I, Durbin R, French L, Grafham D, Gregory S, Hubbard T, Humphray S, Hunt A, Jones M, Lloyd C, McMurray A, Matthews L, Mercer S, Milne S, Mullikin JC, Mungall A, Plumb R, Ross M, Shownkeen R, Sims S, Waterston RH, Wilson RK, Hillier LW, McPherson JD, Marra MA, Mardis ER, Fulton LA, Chinwalla AT, Pepin KH, Gish WR, Chissoe SL, Wendl MC, Delehaunty KD, Miner TL, Delehaunty A, Kramer JB, Cook LL, Fulton RS, Johnson DL, Minx PJ, Clifton SW, Hawkins T, Branscomb E, Predki P, Richardson P, Wenning S, Slezak T, Doggett N, Cheng JF, Olsen A, Lucas S, Elkin C, Uberbacher E, Frazier M et al. (2001). Nature, 409: 860-921. Article MEDLINE

Larsen F, Gundersen G, Lopez R, Prydz H. (1992). Genomics, 13: 1095-1107. MEDLINE

Lee JJ, Hong WK, Hittelman WN, Mao L, Lotan R, Shin DM, Benner SE, Xu XC, Lee JS, Papadimitrakopoulou VM, Geyer C, Perez C, Martin JW, El-Naggar AK, Lippman SM. (2000). Clin. Cancer Res., 6: 1702-1710. MEDLINE

Liang G, Salem CE, Yu MC, Nguyen HD, Gonzales FA, Nguyen TT, Nichols PW, Jones PA. (1998). Genomics, 53: 260-268. Article MEDLINE

Lindsay S, Bird AP. (1987). Nature, 327: 336-338. MEDLINE

Maekawa M, Sugano K, Kashiwabara H, Ushiama M, Fujita S, Yoshimori M, Kakizoe T. (1999). Biochem. Biophys. Res. Commun., 262: 671-676. Article MEDLINE

Merino R, Ganan Y, Macias D, Rodriguez-Leon J, Hurle JM. (1999). Ann. NY Acad. Sci., 887: 120-132. MEDLINE

Merlo A, Herman JG, Mao L, Lee DJ, Gabrielson E, Burger PC, Baylin SB, Sidransky D. (1995). Nat. Med., 1: 686-692. MEDLINE

Miwa W, Yashima K, Sekine T, Sekiya T. (1995). Electrophoresis, 16: 227-232. MEDLINE

Nagai H, Baba M, Konishi N, Kim YS, Nogami M, Okumura K, Emi M, Matsubara K. (1999a). DNA Res., 6: 219-225. MEDLINE

Nagai H, Kim YS, Yasuda T, Ohmachi Y, Yokouchi H, Monden M, Emi M, Konishi N, Nogami M, Okumura K, Matsubara K. (1999b). Gene, 237: 15-20. Article MEDLINE

Nagai H, Ponglikitmongkol M, Fujimoto J, Yamamoto H, Kim YS, Konishi N, Matsubara K. (1998). Cancer, 82: 454-461. Article MEDLINE

Nagai H, Ponglikitmongkol M, Mita E, Ohmachi Y, Yoshikawa H, Saeki R, Yumoto Y, Nakanishi T, Matsubara K. (1994). Cancer Res., 54: 1545-1550. MEDLINE

Nakagawa H, Chadwick RB, Peltomaki P, Plass C, Nakamura Y, de La Chapelle A. (2001). Proc. Natl. Acad. Sci. USA, 98: 591-596. MEDLINE

Nakamura M, Konishi N, Tsunoda S, Hiasa Y, Tsuzuki T, Aoki H, Kobitsu K, Nagai H, Sakaki T. (1997). J. Neurooncol., 35: 113-120. MEDLINE

Nan X, Campoy FJ, Bird A. (1997). Cell, 88: 471-481. MEDLINE

Nan X, Ng HH, Johnson CA, Laherty CD, Turner BM, Eisenman RN, Bird A. (1998). Nature, 393: 386-389. Article MEDLINE

Narayan A, Ji W, Zhang XY, Marrogi A, Graff JR, Baylin SB, Ehrlich M. (1998). Int. J. Cancer, 77: 833-838. Article MEDLINE

Ng HH, Bird A. (1999). Curr. Opin. Genet. Dev., 9: 158-163. Article MEDLINE

Nguyen CT, Gonzales FA, Jones PA. (2001). Nucleic Acids Res., 29: 4598-4606. MEDLINE

Ohsumi T, Okazaki Y, Hirotsune S, Shibata H, Muramatsu M, Suzuki H, Taga C, Watanabe S, Hayashizaki Y. (1995a). Electrophoresis, 16: 203-209.

Ohsumi T, Okazaki Y, Okuizumi H, Shibata K, Hanami T, Mizuno Y, Takahara T, Sasaki N, Ueda M, Muramatsu M et al. (1995b). Biochem. Biophys. Res. Commun., 212: 632-639.

Okazaki Y, Okuizumi H, Ohsumi T, Nomura O, Takada S, Kamiya M, Sasaki N, Matsuda Y, Nishimura M, Tagaya O, Muramatsu M, Hayashizaki Y. (1996). Nat. Genet., 13: 87-90. MEDLINE

Okazaki Y, Okuizumi H, Sasaki N, Ohsumi T, Kuromitsu J, Hirota N, Muramatsu M, Hayashizaki Y. (1995). Electrophoresis, 16: 197-202. MEDLINE

Okuizumi H, Okazaki Y, Ohsumi T, Hanami T, Mizuno Y, Muramatsu M, Hayashizaki Y, Plass C, Chapman VM. (1995). Electrophoresis, 16: 253-260. MEDLINE

Plass C, Shibata H, Kalcheva I, Mullins L, Kotelevtseva N, Mullins J, Kato R, Sasaki H, Hirotsune S, Okazaki Y, Held WA, Hayashizaki Y, Chapman VM. (1996). Nat. Genet., 14: 106-109. MEDLINE

Plass C, Weichenhan D, Catanese J, Costello JF, Yu F, Yu L, Smiraglia D, Cavenee WK, Caligiuri MA, deJong P, Held WA. (1997). DNA Res., 4: 253-255. MEDLINE

Plass C, Yu F, Yu L, Strout MP, El-Rifai W, Elonen E, Knuutila S, Marcucci G, Young DC, Held WA, Bloomfield CD, Caligiuri MA. (1999). Oncogene, 18: 3159-3165. MEDLINE

Qu G, Dubeau L, Narayan A, Yu MC, Ehrlich M. (1999). Mutat. Res., 423: 91-101. MEDLINE

Robertson KD, Jones PA. (2000). Carcinogenesis, 21: 461-467. Article MEDLINE

Rosas SL, Koch W, da Costa Carvalho MG, Wu L, Califano J, Westra W, Jen J, Sidransky D. (2001). Cancer Res., 61: 939-942. MEDLINE

Rouillard JM, Erson AE, Kuick R, Asakawa J, Wimmer K, Muleris M, Petty EM, Hanash S. (2001). Genome Res., 11: 1453-1459. MEDLINE

Rush LJ, Dai Z, Smiraglia DJ, Gao X, Wright FA, Fruhwald M, Costello JF, Held WA, Yu L, Krahe R, Kolitz JE, Bloomfield CD, Caligiuri MA, Plass C. (2001). Blood, 97: 3226-3233. Article MEDLINE

Rush LJ, Heinonen K, Mrozek K, Wolf BJ, Abdel-Rahman M, Szymanska J, Peltomaki P, Kapadia F, Bloomfield CD, Caliguiri MA, Plass C. (2002). Blood, 99: 1874-1876. Article MEDLINE

Sanchez-Cespedes M, Esteller M, Wu L, Nawroz-Danish H, Yoo GH, Koch WM, Jen J, Herman JG, Sidransky D. (2000). Cancer Res., 60: 892-895. MEDLINE

Schirmacher P, Held WA, Yang D, Biempica L, Rogler CE. (1991). Am. J. Pathol., 139: 231-241. MEDLINE

Shibata H, Yoshino K, Muramatsu M, Plass C, Chapman VM, Hayashizaki Y. (1995). Electrophoresis, 16: 210-217. MEDLINE

Smiraglia DJ, Frühwald MC, Costello JF, McCormick SP, Dai Z, Peltomäki P, MS OD, Cavenee WK, Plass C. (1999). Genomics, 58: 254-262. Article MEDLINE

Smiraglia DJ, Rush LJ, Fruhwald MC, Dai Z, Held WA, Costello JF, Lang JC, Eng C, Li B, Wright FA, Caliguiri MA, Plass C. (2001). Hum. Mol. Genet., 10: 1413-1419. MEDLINE

Smiraglia DJ, Szymanska J, Kraggerud SM, Lothe RA, Peltomäki P, Plass C. (2002). Oncogene, 21: 3909-3916. MEDLINE

Suzuki H, Kawai J, Taga C, Ozawa N, Watanabe S. (1994). DNA Res., 1: 245-250. MEDLINE

Tada A, Nishihara T, Kato H. (1998). Oncol Rep., 5: 1137-1140. MEDLINE

Takai D, Gonzales FA, Tsai YC, Thayer MJ, Jones PA. (2001). Hum. Mol. Genet., 10: 2619-2626. MEDLINE

Tateno M, Fukunishi Y, Komatsu S, Okazaki Y, Kawai J, Shibata K, Itoh M, Muramatsu M, Held WA, Hayashizaki Y. (2001). Cancer Res., 61: 1144-1153. MEDLINE

Thoraval D, Asakawa J, Wimmer K, Kuick R, Lamb B, Richardson B, Ambros P, Glover T, Hanash S. (1996). Genes Chrom. Cancer, 17: 234-244. MEDLINE

Turker MS, Bestor TH. (1997). Mutat. Res., 386: 119-130. MEDLINE

Ushijima T, Morimura K, Hosoya Y, Okonogi H, Tatematsu M, Sugimura T, Nagao M. (1997). Proc. Natl. Acad. Sci. USA, 94: 2284-2289. Article MEDLINE

Venter JC, Adams MD, Myers EW, Li PW, Mural RJ, Sutton GG, Smith HO, Yandell M, Evans CA, Holt RA, Gocayne JD, Amanatides P, Ballew RM, Huson DH, Wortman JR, Zhang Q, Kodira CD, Zheng XH, Chen L, Skupski M, Subramanian G, Thomas PD, Zhang J, Gabor Miklos GL, Nelson C, Broder S, Clark AG, Nadeau J, McKusick VA, Zinder N, Levine AJ, Roberts RJ, Simon M, Slayman C, Hunkapiller M, Bolanos R, Delcher A, Dew I, Fasulo D, Flanigan M, Florea L, Halpern A, Hannenhalli S, Kravitz S, Levy S, Mobarry C, Reinert K, Remington K, Abu-Threideh J, Beasley E, Biddick K, Bonazzi V, Brandon R, Cargill M, Chandramouliswaran I, Charlab R, Chaturvedi K, Deng Z, Di Francesco V, Dunn P, Eilbeck K, Evangelista C, Gabrielian AE, Gan W, Ge W, Gong F, Gu Z, Guan P, Heiman TJ, Higgins ME, Ji RR, Ke Z, Ketchum KA, Lai Z, Lei Y, Li Z, Li J, Liang Y, Lin X, Lu F, Merkulov GV, Milshina N, Moore HM, Naik AK, Narayan VA, Neelam B, Nusskern D, Rusch DB, Salzberg S, Shao W, Shue B, Sun J, Wang Z, Wang A, Wang X, Wang J, Wei M, Wides R, Xiao C, Yan C. (2001). Science, 291: 1304-1351. MEDLINE

Walsh CP, Bestor TH. (1999). Genes Dev., 13: 26-34. MEDLINE

Warnecke PM, Clark SJ. (1999). Mol. Cell. Biol., 19: 164-172. MEDLINE

Watanabe S, Kawai J, Hirotsune S, Suzuki H, Hirose K, Taga C, Ozawa N, Fushiki S, Hayashizaki Y. (1995). Electrophoresis, 16: 218-226. MEDLINE

Wimmer K, Kuick R, Thoraval D, Hanash SM. (1996). Electrophoresis, 17: 1741-1751. MEDLINE

Wimmer K, Thoraval D, Kuick R, Lamb BJ, Hanash SM. (1997). Biochem. Soc. Trans., 25: 262-267. MEDLINE

Xiong Z, Laird PW. (1997). Nucleic Acids Res., 25: 2532-2534. Article MEDLINE

Yates PA, Burman RW, Mummaneni P, Krussel S, Turker MS. (1999). J. Biol. Chem., 274: 36357-36361. MEDLINE

Yoon BJ, Herman H, Sikora A, Smith LT, Plass C, Soloway PD. (2001). Nat. Genet., 30: 92-96.

Yoshikawa H, de la Monte S, Nagai H, Wands JR, Matsubara K, Fujiyama A. (1996). Genomics, 31: 28-35. Article MEDLINE

Yoshikawa H, Matsubara K, Qian GS, Jackson P, Groopman JD, Manning JE, Harris CC, Herman JG. (2001). Nat. Genet., 28: 29-35. Article MEDLINE

Figure 1 RLGS profiles. (a) A NotI RLGS profile from normal DNA. (b) A NotI mixing gel RLGS profile. Normal DNA was mixed with a pool of clones from a single 384 well plate from the NotI/EcoRV boundary library. Intense spots not seen in (a) represent clones within the library plate used. (c) An AscI RLGS profile from normal DNA. (d) An AscI mixing gel RLGS profile. Normal DNA was mixed with a pool of clones from a single 384 well plate from the AscI/EcoRV boundary library. Intense spots not seen in (c) represent clones within the library plate used

Figure 2 Schematic of the RLGS technique. The major steps of the RLGS technique are outlined and diagramed. In the DNA sequence, 'M' indicates 5-methylcytosine, and the bold, italics, and underlined G and C indicate radiolabeled nucleotides. The majority of spots in the cartoon RLGS profile represent diploid sequences while the larger spots represent rDNA sequences

Table 1 Techniques for detecting methylation changes in cancer

Table 2 Characteristics of cloned RLGs fragments

Table 3 RLGS analysis of CpG island hypermethylation of various tumor types