The 5'-end transitional CpGs between the CpG islands and retroelements are hypomethylated in association with loss of heterozygosity in gastric cancers

Information on the coding and non-coding sequences as well as CpG islands was obtained from the human genome database in the May, 2004 version [21]. A 1-kb sized non-overlapping window was used to analyze the sequence characteristics of a 10-kb segment upstream and downstream from each transcription start site. This window analysis allowed an examination of the major regulatory regions, which are usually rich in protein binding motifs, and at the same time, produced a profile of the 5'-transitional CpGs between the CpG islands and their nearest retroelements. The sequence data delimited from the promoter regions were input into a local program to calculate the coverage of CpG islands. A CpG island was defined as a DNA segment with a G+C content ≥ 50%, longer than 200 bp nucleotides, and an Observation/Expectation CpG ratio > 0.6 [22]. A similar procedure was used for the genomic position and annotation of retroelements using the RepeatMasker program [23].

The gastric cancer tissue and non-cancerous mucosa were obtained from 50 patients who had undergone surgery between March and December 2003 at St. Paul's Hospital, The Catholic University of Korea. These fresh archives provided the genomic DNA that had been amplified efficiently by methylation-specific PCR (MSP). The histological classification of the gastric cancers was carried out according to the Lauren's classification [24, 25], and the degree of differentiation was graded according to the recommendations of the World Health Organization for the histological typing of gastric cancer [26]. The Tumor-Node-Metastasis (TNM) criteria were used to determine the tumor stage based on the pathological and clinical findings observed from radiography, ultrasonography, computed tomography, and abdominal exploration during laparotomy. The Institutional Review Board approved this study, and written informed consent was obtained from each patient prior to the surgical resection.

A single 5–7 mm diameter tumor site containing a homogeneous cell content was selected from each representative tissue section. Seven-μm-thick hematoxylin-eosin-stained sections were microdissected under a 40-x stereomicroscope using a surgical scalpel. All the microdissected tumor sites were checked for a tumor cell content ≥ 70% prior to DNA extraction. A major fraction (33 sites, 66%) of the microdissected tissues contained a tumor cell content of 80–89%, followed by ≥ 90% (12 sites, 24%) and 70–79% (5 sites, 10%). Approximately 50 microdissected cells were digested in 1 μl of a Tween 20 – Proteinase K lysis buffer. An average of 100 μl of each microdissected tissue lysate was used for microsatellite analysis, and an average of 500 μl of the same lysate was subjected to bisulfite modification for methylation analysis.

The MSI (microsatellite instability) and LOH status of each gastric cancer was determined using a panel of 40 microsatellite markers on eight cancer-associated chromosomes, 3p, 4p, 5q, 8p, 9p, 13q, 17p, and 18q, as reported elsewhere (figure 1A) [5, 6, 27]. The MSI status of the microsatellite sequences was scored if there were any frameshift mutations in > 40% of the homozygous markers. The unilateral reduction of the heterozygous allelic bands was interpreted as LOH. The level of LOHs representing a significant reduction in the genomic dose was determined based on the number of chromosomes showing more than one LOH. According to the genetic classification criterion of gastric cancers [6, 27], the level of LOHs was divided into low (LOH-L, less than four) and high (LOH-H, four or more losses) levels for the intestinal-type cancers, and baseline (LOH-B, zero or one loss), low (LOH-L, two or three losses), and high (LOH-H, four or more losses) levels for the diffuse-type cancers.

Because the quality of the formalin-fixed tissue DNA varied, the amount of template DNA was determined based on a PCR band intensity of 20 ng/μl that had been amplified by the microsatellite primer set, D19S226 (forward, 5'-CCA GCA GAT TTT GGT GTT GTC TA-3'; reverse, 5'-ACA GAG CCA GAG CCA GTA GGA GT-3'; amplicon size, 164 bp). Ninety microliters of the genomic DNA was denatured with 10 μl of 3 M NaOH for 15 min at 37°C before sodium bisulfite modification. This procedure involved modifying 100 μl of the denatured DNA with 1,040 μl of 2.3 M sodium bisulfite and 60 μl of 10 mM hydroquinone for 12 hr at 50°C, as described elsewhere [20]. The modified DNA was then purified using a genomic DNA purification kit (Promega, Madison, WI, USA), precipitated with ethanol, and dissolved in 35 μl of 5 mM Tris buffer (pH 8.0). One μl aliquot of the modified DNA solution was placed in the PCR tube and stored at -20°C.

The semiquantitative methylation analysis using a radioisotope was performed in a minimum of amplification rounds for sub-plateau DNA amplification, as described elsewhere [20]. A 1 μl aliquot of the bisulfite-modified DNA was amplified and labeled in 10 μl of a hot-start PCR solution containing an α-³²P dTTP (PerkinElmer, Boston, MA, USA) and dNTP mixture through 32 PCR cycles. Thirty-two MSP primer sets were selected at the 5'-end regions of 20 genes using the MethPrimer software [28] (table 1). The sequences, PCR condition, and genomic position of these primer set are listed in table 2 and [see Additional file 1]. The coverage of CpG islands, the distribution of retrolements, and the MSP primer position are shown in the figure 2. Each MSP primer set was designed to generate a small fragment of ≤ 150 bp, and yield a specific PCR intensity using the genomic DNA obtained from the formalin-fixed paraffin-embedded tissues.

The specificity of each MSP primer set was validated using a standard curve for the universal methylated and unmethylated DNA (figure 1B). Based on the standard curve of the control MSP bands, the density of the methylated CpGs was classified into 5 levels; level 1 (0–20% methylation), level 2 (21–40% methylation), level 3 (41–60% methylation), level 4 (61–80% methylation), and level 5 (81–100% methylation). The intensity of the MSP bands (figure 3A) was compared with the methylated CpG content that was estimated by sequencing the common PCR DNA (figure 3B). The methylation levels estimated by the MSP intensities were consistent with the methylation densities measured from the common PCR DNAs in 23 normal (96%) and 20 tumor (83%) DNAs of 24 normal and tumor DNA pairs (table 3).

In previous studies [29, 30], the MSP bands visualized by the ethidium bromide staining were classified into three methylation levels (no, weak, and strong methylation). Forty eight DNA specimens were amplified through 37 PCR cycles, of which six (13%) failed to generate any visible PCR bands by ethidium bromide staining (table 3). Eight of the 42 visible MSP bands (19%) were scored differently from the common PCR DNA. Twelve of the 14 DNA specimens (85%) showing no visible PCR bands or inaccurate methylation levels were similarly scored by both 32-cycle amplification using a radioisotope and common PCR DNA. This high accuracy of the 32-cycle amplification protocol allowed the MSP bands to be classified into five methylation levels.

When 80 normal and tumor DNA pairs showing a one-level methylation difference were analyzed, 71 pairs (89%) demonstrated same-level methylation differences in the MSP band intensities and the common PCR DNA [see Additional file 2].

The Pearson's correlation coefficient was used to determine the correlation between (i) the MSP band intensity and the proportion of methylated and unmethylated control DNA, (ii) the coverage of CpG islands and the density of retroelements, and (iii) the frequency of methylation alterations and the level of LOHs using SPSS ver. 11.0 software. A χ ² test or Fisher's exact test and an independent t test were used to compare the clinicopathological variables or the frequency of methylation alterations between the different microsatellite genotypes. Two-sided p values < 0.05 were considered significant.

Of a total of 22,297 genes examined, 10,515 and 11,782 were found to have a single CpG island and no CpG islands at the transcription start site, respectively. The densities of Alus, L1s, and LTRs in a 1-kb window were plotted separately a distance 10 kb upstream and downstream from the transcription start sites. Figure 4A revealed the following: (i) the densities of the three retroelement types were commonly higher in the extragenic 10-kb regions than in the 10-kb intragenic regions. (ii) the three retroelement types were commonly depressed in both the intragenic and extragenic first 1-kb regions, (iii) the density of depressed retroelements was lower in the first 1-kb regions containing CpG islands than in the regions containing no CpG islands, (iv) the density of Alu copies rapidly reached a plateau at a distance of 3 kb, which was higher in the 5'-end regions containing CpG islands than in the regions containing no CpG islands.

A transitional area between the CpG islands and their nearest retroelements was demarcated by a distance from the 5'(3')-end of the CpG island to the position of the nearest retroelement in the extragenic (intragenic) region (figure 4B). Figure 4C shows that the CpG islands frequently neighbored the nearest Alu elements (76% and 83%) and infrequently the nearest L1 (13% and 11%) and LTR (11% and 6%) elements in both the extragenic and intragenic regions. Figure 4D shows that the mean extent of the transitional area between the CpG island and the nearest retroelement was significantly shorter in the extragenic regions (989 bp) than in the intragenic regions (1,700 bp). The mean extent of the extragenic and intragenic transitional area bordered by the nearest Alu elements (945 bp and 1,645 bp) was significantly shorter than that by the nearest L1 (1,063 bp and 1,853 bp) and LTR (1,104 bp and 1,967 bp) elements.

The allelic profiles of the 40 microsatellite sequences were initially analyzed for a MSI at the homozygous markers that showed a few shadow or stutter bands in a pair of normal and tumor DNAs. Five of 50 gastric cancers examined had a high-frequency MSI in more than 40% of the homozygous alleles examined. A LOH was analyzed at the stable heterozygous alleles in the remaining 45 MSI-negative gastric cancers. The number of chromosomes showing more than one LOH event was counted in each MSI-negative case. Using the number of chromosomal losses, the extents of LOH was categorized into high level involving four or more chromosomes (LOH-H) and low level involving less than four chromosomes (LOH-L) for intestinal-type cancers. One and zero chromosomal loss were scored as baseline-level loss (LOH-B) for diffuse-type cancers. Overall, 20 LOH-H (40%), 19 LOH-L (38%), 6 LOH-B (12%), and 5 MSI (10%) cases were identified in the 50 surgical specimens (table 4). A comparison of the clinicopathological variables in the LOH-H and LOH-L gastric cancers revealed a lymphatic invasion (p = 0.001) and advanced stage (p = 0.015) to be significantly associated with the LOH-H cases. Intestinal-type cancers (p = 0.032) as well as well and moderate differentiation (p = 0.046) were more frequent in the LOH-L cases.

Fifty pairs of normal and tumor tissues were examined using the MSP primer sets for a total of 32 CpG amplicon sites (figure 5). The CpG islands of the CDH1, RABGEF1, and STAG1 genes, which neighbored a high density of Alu elements at a distance < 2 kb, were found to be hypomethylated in the normal tissues at a level < 2. These CpG islands showed few methylation alterations in the tumor tissues. The transitional CpGs of the MYBPC2 gene bordered by the high-density Alus were hypermethylated at a level ≥ 3 in the normal tissues but were unchanged in the tumor tissues. The long CpG islands of the PAX5 and RUNX2 genes neighboring a small number of retroelements at a long distance, > 6 kb, showed insignificant methylation alterations in the tumor tissues.

The CpG island boundaries neighboring a low density of Alu elements or L1 and LTR elements at a distance of 2–6 kb became increasingly methylated at a level of ≥ 2. The CpG sites in close proximity to the transcription start sites with no accompanying CpG islands were densely methylated at a level of ≥ 3. The CpG island boundaries and the proximal sites of the non-CpG-island genes were categorized as transitional CpGs, because these CpG sites could be distinguished from the proximal sites of the CpG islands that are consistently unmethylated. The methylation level of these 5'-transitional CpGs in the normal and tumor DNA was compared (table 5). The methylation level of the LOH-H cancers was significantly lower at the 5'-transitional CpG sites of the VDR, FLJ43855, CDKN2A, MUC8, MAGEA2, and MSLN genes. The LOH-L cancers had a significantly higher methylation level at the 5'-transitional CpG sites of the MLH1 and CDKN2A genes.

The frequency of methylation alterations occurring in a total of 14 transitional CpG amplicons was scored according to the number of the amplicons showing a difference in the level of methylation between the normal and tumor DNA. The frequency of hypomethylation and hypermethylation in the transitional CpGs were linearly (r = 0.738, p < 0.0001) and inversely (r = 0.673, p < 0.0001) proportional to the number of chromosomal losses, respectively, (figure 6A and 6B). There were significant differences in the mean frequency of hypomethylation (44% versus 15%) and hypermethylation (5% versus 26%) between the LOH-H and LOH-L cancers (p < 0.0001) (figure 6C and 6D). The frequency of hypermethylation in the transitional CpGs was highest in the MSI cases.

Although the proximal portion of the CpG islands was largely unmethylated in both the normal and tumor DNAs, the CpG islands of the MLH1 and RUNX3 genes was hypermethylated in most tumor tissues with a MSI (MLH1) and LOH-B (RUNX3), respectively (figure 5). The distal portion of the RUNX3 CpG island was frequently hypermethylated in both LOH-H and LOH-L cases.

The relationship between the frequency of methylation alterations and the clinicopathological features was analyzed in 39 gastric cancers with high-level and low-level chromosomal losses (table 6). The frequency of hypomethylation in the transitional area increased with increasing lymphatic invasion (p = 0.006) and advanced tumor stage (p = 0.015). The frequency of hypermethylation decreased with increasing venous invasion (p = 0.041), lymphatic invasion (p = 0.003), and advanced tumor stage (p = 0.023).

Of a total of 598 methylation alterations, 472 (79%) showed a one-level difference in methylation between the paired normal and tumor DNA and 126 (21%) showed a two- or higher-level difference with an inverse methylation status between the normal and tumor DNA [see Additional file 2]. The normal tissues of the LOH-H and LOH-L cases showed a similar level of methylation in the transitional area.