A novel technique for the identification of CpG islands exhibiting altered methylation patterns (ICEAMP)

Abstract

Aberrant CpG methylation changes occurring during tumour progression include the loss (hypomethylation) and gain (hypermethylation) of methyl groups. Techniques currently available for examining such changes either require selection of a region, then examination of methylation changes, or utilise methylation-sensitive restriction enzymes to identify an alteration. We describe here a novel method that identifies genomic regions as a consequence of altered methylation during tumourigenesis. A methyl-CpG binding domain column isolates methylated GC-rich sequences from both tumours and surrounding normal tissue. Subsequent subtractive hybridisation removes sequences common to both, leaving only methylated sequences unique to the tumour. Libraries of sequences generated using DNA derived from a breast tumour (histological grade; poorly differentiated) as ‘tester’ and from matched normal tissue as ‘driver’ were examined; 26% of clones had the sequence criteria of a CpG island (CGI). Analysis using the bisulfite technique revealed that a number of these sequences were methylated in tumour DNA relative to the normal control. We have therefore demonstrated the ability of this technique, the identification of CGI exhibiting altered methylation patterns (ICEAMP), to isolate tumour-specific methylated GC-rich sequences. This will allow a comprehensive identification of methylation changes during tumourigenesis and will lead to a better understanding of the processes involved.

INTRODUCTION

The aberrant methylation of CpG dinucleotides has been widely reported during tumourigenesis in a variety of cancers (for review see 1). The alterations identified and their consequences include the loss of methyl groups, which is thought to increase chromosomal instability (2). Alternatively, the gain of methyl groups, particularly in CpG islands [CGIs: GC-rich regions of the genome, ~1 kb in length, originally characterised due to their lack of methylation (3)], is linked to the transcriptional repression of the associated genes (for review see 4).

Whether such aberrant methylation patterns are a part of the causative process or a result of secondary effects is unclear. However, once established, aberrant methylation patterns are clonally inherited with high fidelity in all progeny cells. The initial methylation change and subsequent alterations will therefore be represented in the later stages of tumourigenesis. An extensive examination of the differences in methylation (both gains and losses) at different stages of cancer development will lead to a clearer understanding of the mechanisms involved and provide an additional means of tumour classification. In addition, the identification of aberrantly methylated CGIs will allow identification of those genes disrupted during tumour progression. Finally, identification of the earliest changes in methylation will provide useful biomarkers in cancer diagnosis and treatment.

Methods that are currently available for the investigation of methylation changes include restriction landmark genome scanning (5) and genome scanning analysis (6). The method described here sought to identify DNA methylation changes that occur during tumour progression in breast cancers, without prior knowledge of the sequences involved or availability of restriction enzyme recognition sites. We focused specifically on single copy GC-rich regions, such as CGIs, altered by aberrant methylation. Genomic fragments were first isolated from the remainder of the genome by utilizing a methyl-CpG binding domain (MBD) column (7). This comprises the MBD of rat MeCP2 covalently linked to a histidine tag (HMBD), then attached to a Ni-agarose matrix. At low salt concentration the bulk of the genome will bind to the column; however, under conditions of increasing salt concentration only densely methylated sequences will remain bound, allowing their isolation (7). Previous work with an MBD column has revealed that in addition to low and single copy methylated sequences this tightly bound fraction also contained repetitive elements (8), for example SINEs and LINEs (short and long interspersed nuclear elements). In order to isolate only those CGIs with altered methylation patterns, a process of subtractive hybridisation was employed (using DNA extracted from tumour as ‘tester’ and from matched normal tissue as ‘driver’) to remove these repetitive sequences. Using this method, CGIs and other low copy number GC-rich regions with altered methylation patterns between normal and tumour could be extracted, cloned and then analysed.

MATERIALS AND METHODS

The methyl-binding domain column

The MBD column was constructed and operated following a previously described protocol (7). The Fast Pressure Liquid Chromatography HR 10/2 column (Amersham Pharmacia Biotech) used in this study contained ~8 ml of slurry with ~16 mg/ml of bound HMBD. DNA derived from human female blood (100 µg) was digested with 200 U of MseI (New England Biolabs, NEB) in a total volume of 500 µl, for 3–4 h, then loaded directly onto the column with the first 5 ml eluting being reloaded. Fractions were collected across a salt gradient (7) and analysed using PCR primers to amplify regions of known methylation status. These were Sef2 (accession no. U75701), representative of the bulk of the genome, Mer22 (accession no. AC018692), representative of a densely methylated region and MAOA (accession no. M89636), which is X-linked and thus has a methylated copy in DNA derived from female but not male (see Fig. 1). The primer sequences are shown in Table 1, all reactions were in a total volume of 10 µl with each primer at 10 µM concentration, 0.05 µl (5 U) of Biotaq DNA polymerase (Bioline) 1 µl of a 10× buffer containing Tris–HCl (pH 8.8), 4.5 mM (NH₄)₂SO₄, 11 mM MgCl₂, 4.5 mM EDTA, 4.4 µM β-mercaptoethanol and dNTPs at 1.0 mM each (ABgene Surrey, UK). DNA derived from male blood (100 µg) was then treated in an identical manner.

Figure 1.

PCR analysis of bound and unbound fractions from the MBD column. MBD1 refers to the samples collected after one round. The fractions eluting between 0.75–0.85 M NaCl are reloaded and are referred to as MBD2 (see text for details). L, load; FT, flow through; U, unbound 0.45–0.55 M NaCl; B, bound 0.75–0.85 M NaCl; 0, distilled H₂O blank. (A) Primer pair 1 (Sef2) generates a 260 bp product from an MseI fragment of 846 bp, which contains 10 CpGs (representative of the bulk of the genome), that does not bind to the MBD column at higher salt concentrations. (B) Primer pair 2 (Mer22) generates a 539 bp product from an 866 bp MseI fragment that contains 51 CpGs. This fragment is from a tandemly repeated locus, which is predominantly methylated in somatic cells (8) and is in the bound fraction of both male and female DNA after a second round of purification. (C) Primer pair 3 (MAOA) generates a 557 bp product from an 1888 bp MseI fragment, which contains 114 CpGs. This fragment contains the promoter region including the CGI of the X-linked gene Monoamine Oxidase A. It is not present in the bound fraction in DNA derived from male blood, but is present in DNA derived from female blood after a second round of purification.

Table 1. Primer sequences and annealing temperature used in PCR analysis.

		Primers	Temperature/time
Sequences of primers used for testing the MBD columns ability to bind methylated DNA. PCR amplification was performed using the primers listed for 30 cycles with a 95°C/1.5 min denaturing time, the annealing time and temperature shown, then a 72°C/3 min extension followed by a 72°C/10 min elongation step. For bisulfite analysis of cloned inserts, hemi-nested PCR amplification was performed with the first round reaction, using primers (1st) and (2nd) for 30 cycles with a 95°C/1.5 min denaturing time, the annealing time and temperature shown, then a 72°C/3 min extension followed by a 72°C/10 min elongation step. Following this 1 µl of the reaction was diluted 1:250 and 1 µl used in the second round PCR with primers (1st) and (3rd) for 35 cycles. For MS-PCR analysis, bisulfite DNA was conducted in 35 cycles of amplification with a 94°C/1 min denaturing time, the annealing time and temperature shown, then a 72°C/2 min elongation step.
Sef2	1st	5′-AATCCAAACCGCCTTCCAAGTG-3′	67°C/45 s
	2nd	5′-AGGAACGAATGGAGAAAGTGCAAC-3′
Mer22	1st	5′-ACACCAACTGCTGTGGGATTGG-3′	68°C/30 s
	2nd	5′-GCCTGTGTCCTGGGTCCTGTTT-3′
MAOA	1st	5′-GGCACCAGTACCCGCACCA-3′	68°C/1 min
	2nd	5′-GGGGTGCTGAACCCTGAGGA-3′
		Bisulfite primers
XLMP40	1st	XLMP40BisF 5′-AGGGTATTAAGGGTAAGAA-3′	54°C/1 min
	2nd	XLMP40BisR 5′-CTCACTACAACCTCTACCT-3′	54°C/1 min
	3rd	XLMP40BisR2 5′-CCACTTATCAAACCTAAA-3′	54°C/1 min
XLMP107	1st	XLMP107BisF 5′-GATTTTATTTGGAAATGTGA-3′	58°C/1 min
	2nd	XLMP107BisF2 5′-TAAAAGGTTTTTGTTTTGAG-3′	58°C/1 min
	3rd	XLMP107BisR 5′-TTTAACTACCAACTCTATCCA-3′	58°C/1 min
XLMP130	1st	XLMP130BisF 5′-TGTATCGAAGTTTTATTTTTATGTTG-3′	60°C/1 min
	2nd	XLMP130BisF2 5′-TGTGTTTAGTAGGTGGTAGGTG-3′	60°C/1 min
	3rd	XLMP130BisR 5′-AAAAACTCCTAAAATCCCAAAA-3′	60°C/1 min
		MS-PCR primers
S1MP2		M (+) 5′-AAACCGAAAACGACCGCCGCGCG-3′	72°C/1 min
		M (–) 5′-GTTCGCGGTGCGTATCGCGTCGC-3′	72°C/1 min
		UM (+) 5′-GGGTGGATTGAGAGTGATTGTTGTGT-3′	52°C/1 min
		UM (–) 5′-AAAATTCACAATACACACCACACCACAA-3′	52°C/1 min

Patient samples

Breast tumour samples used in this study, obtained from patients undergoing mastectomies, were classified as infiltrating ductal carcinomas. Adjacent normal parenchyma was obtained to serve as a control. These samples had already been investigated with regard to alterations in methylation patterns using CpG island arrays (9). DNA (5 µg) was extracted from a matched pair of samples (155 were derived from tumour and 156 from the adjacent normal tissue) using standard techniques (9) then digested with MseI, 10 U (NEB) for 3–4 h. The digested DNA from the normal sample was fractionated using an MBD column as described. This was followed by washing with 20 ml of buffer containing 1.0 M NaCl to remove any remaining bound DNA. The column was then equilibrated by washing with 20 ml of buffer containing 0.1 M NaCl followed by the fractionation of samples derived from tumour DNA. Samples eluting from the MBD column between 0.75 and 0.85 M NaCl were precipitated and used in this analysis.

Subtractive hybridisation

The subtractive hybridisation procedure was first optimised by recovery of a plasmid spike using 5 µg of DNA derived from normal female blood as starting material for generation of both ‘tester’ and ‘driver’ (Figs 2 and 3). Linker-adaptors (10 mM) LA2.1 5′-TAGTTAACGCGCTGCATGAGTA-3′ and LA2.11 5′-TACTCATGCAGCGCGTTAAC)-3′ were added to the driver DNA (~150 ng) along with T4 ligase (400 U) and 10× ligase buffer (NEB) in a total volume of 20 µl, then incubated overnight at 4°C. This was followed by incubation for 30 min at 37°C with 1 U of T4 DNA polymerase (NEB) and dNTPs (100 µM). Driver and tester, mixed in a ratio of 3:1, were cleaned using a Qiagen PCR Purification Kit column (Qiagen Ltd, UK) and resuspended in 20 µl of distilled H₂0 with 50 mM HEPES (pH 7.5), 100 mM NaCl, 50 mM EDTA (pH 8) and 0.1% w/v SDS, total volume 50 µl, based on a method described previously (10). Following a 5 min incubation at 95°C the solution was held at 68°C for 48 h, placed immediately on ice for 10 min and cleaned using a Qiagen PCR clean-up column. After resuspending in 20 µl of Tris–HCl pH 8.0, linker adaptors (20 mM) LA2.4 5′-TACTTCTTGCGCCAAGACGTT-3′ and LA2.44 5′-AACGTCTTGGCGCAAGAAG-3′, T4 ligase (800 U) and 10× ligase buffer (NEB) were added, total volume 30 µl, followed by overnight incubation at 4°C. A Qiagen column was used to clean the mixture, and an aliquot (1 µl) used in a PCR reaction with linker LA2.44 as primer (Fig. 3).

Figure 2.

Schematic representation of the subtractive hybridisation procedure. The tester and driver are comprised of MseI-digested DNA that bound to the MDB column at NaCl concentrations of between 0.75 and 0.85 M. The fractions were precipitated, pooled and split in a 3:1 ratio with the smaller aliquot, tester (in black) being spiked with 0.25 pg of an MseI digested plasmid (white box). The ends of the driver aliquot were then filled in (shown as black boxes without ‘overhanging ends’). (1) Both are mixed and denatured at 95°C for 10 min. (2) Following hybridisation for 48 h at 68°C only the plasmid fragments and a background of tester:tester duplexes can re-anneal generating TA overhangs. (3) Linker/adaptors can be attached to these and can then (4) be amplified using a compatible primer.

Figure 3.

Subtractive hybridisation to recover a plasmid spike. Tester and driver DNA are derived from the same source and have 0.25 pg of pGEM5zf+ plasmid digested with MseI added to the tester as a ‘spike’ Lane 1 shows ~20 ng of pGEM5zf+ digested with MseI. Lane 2 is the recovered spike; a faint background smear can be seen in the test subtractions, probably due to tester sequences that have re-annealed without forming tester:driver duplexes. It should also be noted that the difference in size is due to the addition of ~40 bp of catch linker to the recovered spike.

To generate libraries, DNA derived from tumour was used as tester and from normal DNA as driver, as illustrated in Figure 2. Analysis of the resulting products using PCR failed to detect representative fragments from Sef2 or MAOA. However, a faint band is detectable using the primers for Mer22 (data not shown).

Cloning and sequence analysis

Amplified fragments cloned using the T-vector system (Promega) and transformed into XL1-Blue (Stratagene) were sequenced using dGTP BigDyes kit (Applied Biosystems). The BLAST (11) analysis results are shown in Table 2. The Grail CpG prediction program (Human Genome Mapping Program) was used to identify cloned fragments from CGIs (http://www.hgmp.mrc.ac.uk/).

Table 2. Blast/NIX analysis of predicted CGIs.

ID	Size (bp)	Comments	Accession no.	Me-CpG.	%GC/CpGO/E
Comments: homology to sequences in the database is shown. (NIX) indicates if region is predicted to be a CGI using the grail CpG prediction program (HGMP). Me-CpG: if methylation analysis of the cloned region was undertaken or if sequence matches to region predicted in alternative study referenced as shown. %GC and CpGO/E of the MseI fragment cloned.
XLMP205	250	Predicted CGI	AF414382	N/T	70/0.96
S1MP64	242	Predicted CGI (NIX)	AF414376	N/T	68/0.88
S1MP79	264	HUMY16EST accession no. L34977	AF414375	N/T	61/0.84
XLMP128	484	Accession no. AB017146	AF414377	ref. 15	66/0.74
XLMP40	517	Predicted CGI (NIX)	AF414387	Fig. 5C	69/0.72
XLMP77	218	Predicted CGI (NIX)	AF414388	N/T	70/0.79
XLMP87	263	Predicted CGI (NIX)	AF414389	N/T	60/0.77
S1MP2	457	Predicted CGI 5′of 3B (3OST3B1) gene	AF414374	Fig. 5A	63/0.84
XLMP130	665	Accession no. AL160175	AF414379	Fig. 5B	63/0.76
XLMP136	537	Predicted CGI mRNA accession no. AB058775	AF414378	N/T	64/0.84
XLMP147	512	Predicted CGI (NIX)	AF414380	N/T	63/0.73
XLMP198	203	Predicted CGI (NIX)	AF414381	N/T	71/0.78
XLMP211	206	Predicted CGI (NIX)	AF414383	N/T	65/0.73
XLMP214	285	Predicted CGI (NIX)	AF414384	N/T	73/0.94
XLMP224	191	Predicted CGI (NIX)	AF414385	N/T	60/0.6
XLMP226	204	93% homology to 5 CGIs in database (7)	AF414386	N/T	53/0.71

Bisulfite analysis

The bisulfite sequencing method employed in this study is a modification of the one used in Frommer et al. (12) with the addition of urea to improve the reaction efficiency (13). Approximately 500–600 bp of genomic DNA inclusive of the cloned fragments from both tumour and normal DNA were sequenced. Based on this sequence information, primers were designed for methylation-specific (MS) PCR (14) for the SIMP2 locus. See Table 1 for the primers used and the methyl-binding domain column section for reaction conditions.

RESULTS

The PCR analysis demonstrated the MBD column’s ability to separate densely methylated fragments from the rest of the genome (Fig. 1). In this study, fractions eluting between 0.75 and 0.85 M were used, estimated (based on previous measurements made when 100 µg of starting material was used) to contain ~250 ng of DNA. The strategy outlined in Figure 2 was developed prior to subtracting the patients DNA. This results in the recovery of a relatively pure sample of the plasmid ‘spike’. Cloning and analysis of the PCR products revealed that ~95% of the inserts contained plasmid fragments (data not shown). However, there appears to be a bias towards the band at ~300 bp (Fig. 3, lane 2) caused either by the subtractive process or the PCR step.

When DNA derived from a tumour source was subtracted with that derived from the adjacent normal tissue, the average size of inserts analysed was ~350 bp. Analysis of 173 cloned inserts revealed that both the GC content and CpG Observed/Expected (CpG_O/E) are much higher than that previously reported when an MBD column was used to fractionate the genome (8). The average GC content of 59% and CpG_O/E of 0.70 are in fact closer to that reported when the MBD column was used to generate libraries of CGIs (7). Following BLAST searches (15) analysis of the cloned sequences revealed that 19% were derived from repetitive elements, for example Alus, with 10% matching to exons or pseudogene sequences. An additional 14% had homology to multiple BAC clones in the high throughput genome sequence (HTGS) database or fragments derived from the rDNA non-transcribed spacer (16), considered ‘methylated low copy number repeats’. Those inserts with no significant homology to sequences in the database (10%) may represent difficult to sequence clones (GC 62% and a CpG_O/E of 0.88). The final category of sequence (19%) had the GC content and CpG_O/E of the bulk of the genome.

The remainder (~26%) were MseI fragments from predicted CGIs (17). Following removal of duplicated sequences 16 unique cloned inserts remained, as shown in Table 2, with a further three adjacent to, but not within a region predicted to be a CGI (GC 65% and a CpG_O/E of 0.79).

Analysis of methylation at three of these predicted CGIs involved bisulfite modification and amplification of 500–600 bp followed by digestion with restriction enzymes as shown (Fig. 4A). A more comprehensive analysis involves cloning the amplified fragments then sequencing. In Figure 4B–D a graphic representation of the methylation changes detected at CpGs in the regions amplified, cloned and sequenced is shown. Clone S1MP2 (Fig. 4B) is part of an ~5 kb CGI co-localised with the 5′ end of the heparan sulfate d-glucosaminyl 3-O-sulfotransferase-3B (HGNC) gene at chromosome 17p12. Clone XLMP130 (Fig. 4C) is part of an ~1 kb CGI co-localised with the 3′ end of a gene encoding a novel protein similar to interferon regulatory factor (bA243J16.5) on chromosome 20. Finally clone XLMP40 (Fig. 4D) is part of an ~1 kb CGI, which does not co-localise with any predicted genes using the programs available through the HGMP. A further two clones were analysed (following bisulfite treatment and amplification) using restriction enzymes as shown in Figure 4A. This indicated that both these loci occurred in a methylated and unmethylated form in both the normal and the tumour DNA at the sites tested (data not shown).

Figure 4.

Restriction enzyme digests and sequencing of PCR products generated with primers specific for bisulfite-modified DNA. (A) Amplified fragments are digested with BstUI, MseI, MspI or HhaI. Following bisulfite modification, methylated CpG will be retained, digestion with BstUI or HhaI indicates methylation. Digestion with MseI indicates conversion of unmethylated cytosines (CCAA to TTAA); failure to digest with MspI indicates successful conversion (CCGG-TCGG or TTGG depending on methylation at internal CpG). (B–D) Several inserts were analysed for each region amplified from both tumour and normal DNA, representative examples are shown. The methylation status of all CpGs in the amplified region including the cloned MseI fragment is graphically represented. Sequence amplified from normal DNA was analysed as shown in the top line, amplified DNA from tumour tissue is shown below. Methylated CpGs are shown as black circles, unmethylated CpGs are represented as white circles. The fragment cloned in the library is represented by a white box with a dashed line indicating the cloned inserts extended beyond the region analysed.

One of the cloned inserts present in the library has previously been reported to show altered methylation patterns during tumour progression (XLMP 128) using alternative methods (18). A second clone matched a CGI in the database generated using the method of Cross et al. (7). As this technique was used to generate libraries of unmethylated CGIs, this may indicate a methylation change in this region.

A number of inserts (10%) have no homology to sequences currently represented in the database. Many of these sequences required several sequencing attempts and contain inserts that are both very GC rich and have a high CpG_O/E.. Their sequences may not yet be included in the database.

We extended the methylation analysis of SIMP2 to a panel of 37 primary tumours and seven normal breast tissue samples. Bisulfite primers (positions 89–112 and 290–318 nt, respectively) were designed for methylation-specific PCR (MS-PCR) (19) in the area devoid of DNA methylation in a normal CGI (Fig. 4B). Hypermethylation was seen in 30% (11 of 37) of the breast tumours examined, but no methylation was detected in the seven normal controls (Fig. 5A). Statistical analysis revealed that this hypermethylation was frequently seen in patients of younger age (<50 years old) and more advanced clinical stages (P <0.005) (Fig. 5B).

Figure 5.

Methylation-specific PCR for the S1MP2 locus in 37 primary breast tumours. (A) Representative results. T, Tumour; N, normal; P, positive control; M, methylated allele; U, unmethylated allele. (B) A three-dimensional plot showing the relation between S1MP2 hypermethylation and clinical parameters of patients. Hypermethylation of S1MP2 is depicted as black circles, whereas no methylation is depicted as white circles in this patient group. As shown, this hypermethylation event was associated with younger patients of more advanced clinical stages (P < 0.005, χ² test).

DISCUSSION

We have described a novel technique for the extraction of genomic regions that differ between two samples with regards to methylation. In this example, the tester is derived from a tumour, the driver from the adjacent normal parenchyma. This method both complements and extends existing techniques, such as the one employed by Shiraishi and co-workers, which used an MBD column to isolate densely methylated regions from human lung adenocarcinoma DNA (18). Out of the original examination of 1000 clones, nine were identified as potential CGIs. Further analysis confirmed that one of these CGIs was methylated in the adenocarcinoma cells and unmethylated in normal cells (18). The addition of the subtractive hybridisation step has therefore greatly increased the methods capability of rapidly generating libraries of CGI fragments. In addition, three of the five putative CGIs examined to date exhibit a methylation change between the tumour and normal tissue examined (Fig. 4B–D).

However, the method described may not detect all methylation changes occurring in or near a repetitive element as fragments containing such sequences could be removed during the subtractive process. In addition, it should be noted that the method may be affected by polymerase bias as the ~300 bp band seen (Fig. 3) predominates following subtraction and subsequent amplification, and inserts in this size range appear to be enriched in the library. The generation of sufficient quantities of tester:tester duplexes by using a larger amount of starting material would allow direct cloning and remove the problem of PCR amplification bias. A larger quantity of starting material would also allow a greater driver:tester ratio, which may reduce the number of cloned inserts that are representative of both low and high copy number methylated repeats (33%) in the final library. Attempts to amplify the driver using the attached catch-linkers introduced significant PCR bias, rendering such a strategy untenable.

A number of the cloned inserts (19%) were derived from MseI fragments that are single copy but which did not have the GC content or CpG_O/E of fragments expected to be retained by the MBD column at high salt concentration. Previous studies have shown that significant ‘contamination’ remains after the first fractionation and that a further purification of material bound after one pass through the MBD column is required (Fig. 1) (7,8). Our present study analysed cloned inserts generated by subtracting tester and driver passed once over an MBD column (i.e. MBD1, Fig. 1). Libraries generated using tester and driver subjected to two rounds of purification (i.e. MBD2, Fig. 2) are currently being analysed.

We intend to use this method to generate additional libraries for the development of methylated CGI arrays similar to those described by Yan et al. (9). However, it has previously been reported that methylation changes can be detected in the adjacent tissue before pathological detection (20). Analysis of the inserts generated in this library also revealed some methylation in tissue histologically classified as normal (Fig. 4B–D). While such alterations may be specific to the tissue type or the patient’s DNA used in this study, to ensure a comprehensive representation of methylation changes in any subsequent arrays, it will be necessary to undertake a more thorough analysis of samples histologically classified as normal. This will ensure that major methylation changes have not occurred in the ‘driver’ sample resulting in the removal of sequences indicative of a methylation change during the subtraction process.

It is not possible from this study to say at what stage in the tumourigenic process these changes have occurred or how frequent they may be in the cells that comprise the tumour. However, our observation of SIMP2 shows that hypermethylation of this locus usually occurred in younger breast cancer patients with more advanced disease status. Further investigation will be conducted to ascertain if there is any disruption to the transcriptional regulation of the associated HGNC gene and whether this aberrant event is occurring in later stages of tumour development. Thus, this initial study sets an example for further analysis of the effect that methylation changes at the other loci may have on tumour cell physiology.

The generation of libraries of sequences showing methylation changes during neoplasia will be a valuable tool in breast tumour diagnosis and classification and in assessing the efficacy of any treatments. Additional potential applications of this technique include analysis of methylation changes that occur in other cancer types, identification of hypomethylation changes occurring during tumour progression (by using DNA derived from normal tissue as tester), methylation changes that occur during ageing and the recovery of islands methylated on the inactive X chromosome.

DDBJ/EMBL/GenBank accession nos AF414374–AF414389