Gene Copy Number Change Events at Chromosome 20 and Their Association with Recurrence in Gastric Cancer Patients Free

Even with the dramatic decline in the incidence of gastric cancer in developed countries, it still remains a leading cause of cancer death worldwide (1). Even with the rapid advancement of molecular biological techniques, there has been slow progress in the understanding of the molecular mechanism of gastric carcinogenesis. This is partly due to the multistep nature of the carcinogenesis and the need to probe the expression and interaction of a large number of genes and proteins simultaneously. Genome-wide monitoring of both of the gene expression and gene copy number changes has now become possible due to microarray technology. In gastric cancer genomic research, microarrays spotted with more than 20,000 human genes are routinely being used to monitor the expression pattern of those genes in a single experiment. Oligonucleotide or cDNA microarrays have been used for a comprehensive survey of the gene expression profiles of the gastric carcinoma cell lines (2), to identify the gastric adenocarcinoma metastasis-related genes in the cell lines derived from a primary tumor and its metastasis counterpart (3, 4), and to select the gene sets that can distinguish between the normal and tumor gastric tissues, or the subtypes of gastric cancer based on gene expression profiling (5–9).

Another important tool for understanding the various types of tumorigenesis has been comparative genomic hybridization (CGH), which traditionally employed whole chromosomes (10, 11). Even though these chromosome-scale CGHs have been informative, the low resolution of sometimes megabases-long inserts has led to the development of matrix-based CGH using P1, PAC clones, where the insert sizes range from 75 to 130 kb in length (12). A breakthrough in the field of CGH was made when Pollack et al. (13, 14) applied cDNA microarray technology to probe the gene copy changes in breast cancer.

Many groups have reported genome instabilities related with gastric cancer. A recent report by Weiss et al. (15) specifically determined an amplicon boundary at 20q13.2 region using BAC clone-based microarrays. The amplified region was shown to be within an ∼800-kb-long region of 20q13.2. Two of the well-known oncogene candidates, ZNF217 and CYP24, were located within this region. However, many of these BAC array-based CGH studies have a low resolution and few have searched for an association between gene copy number change and any clinical information.

In order to delineate the individual genes that undergo copy number changes in gastric cancer, a cDNA microarray-based CGH with 30 pairs of normal and tumor gastric tissues using cDNA microarrays containing 17,000 human genes were done. This report highlights two separate but related aspects of the gene copy number change events in chromosome 20 of gastric cancer. The first is a list of amplified genes in human chromosome 20 of gastric cancer compared with normal gastric tissues. We report that 96 genes including ZNF217, CYP24A, and PTPN1 show more than a 1.5-fold increase in the Cy5/Cy3 intensity ratio in the tumor tissues tested.

In many cases of microarray experiments, a correlation between the expression level of selected genes and the clinical parameters was obtained when the gene expression signature comprised of ratios or intensities of several genes. A group of selected genes, when combined, forms a molecular signature that provides a high correlation with the clinical information, even though they do not seem to provide meaningful information individually (16). Therefore, in the second approach, instead of selecting the genes that pass the 1.5-fold cutoff line, all of the genes that passed the microarray spot quality filtering and are located on chromosome 20 were examined using a t test to determine if any of the genes, whose copy changes are related with a recurrence, are selected. For this, the patients were divided into the recurrent and nonrecurrent group, and the genes that can discriminate between the two groups were searched. Using the t test, 5 genes (P < 0.01) and 39 genes (P < 0.05), whose combined score of copy change ratios could distinguish between the recurrent and nonrecurrent groups in a statistically significant manner, were selected.

Tissue Samples and DNA Isolation. All the experiments using the patient tissues were done after gaining the approval of the supervisory committee of the Yonsei University College of Medicine. Only tissue from the patients who underwent surgery at the Yonsei University College of Medicine (1997-1999), and were followed up for at least 5 years after surgery, were used. Tissues with an at least 70% tumor content were saved as the tumor tissues. Thirty pairs of normal and tumor gastric tissues used in these experiments (Table 1) were stored in liquid nitrogen immediately after surgery. The frozen tissues were ground to a powder, and the genomic DNA was prepared using the phenol/chloroform/isoamylalcohol method (17).

cDNA Microarray-Based CGH. The cDNA microarrays containing 17,000 human gene probes (Genomictree, Daejeon, Republic of Korea) were used for the microarray CGH. Six micrograms of the DpnII-digested genomic DNA were purified using a QIAquick PCR purification kit (QIAgen, Düsseldorf, Germany), and was subjected to fluorescent labeling using a Bioprime labeling kit (Invitrogen, Carlsbad, CA). The reaction mixtures were incubated at 37°C for 2 hours in the dark and were quenched by adding 5 μL of 0.5 mol/L EDTA (pH 8.0). The microarray slide was incubated in 3.5× SSC, 0.1% SDS with 10 mg/mL bovine serum albumin for 1 hour at 42°C for blocking. The genomic DNAs from the normal and tumor tissues were labeled with Cy3-dCTP and Cy5-dCTP, respectively. A pair of Cy3- and Cy5-labeled DNAs from a patient was mixed with 30 μg human Cot-1 DNA, 20 μg poly (dA)-poly (dT) oligonucleotides, and 100 μg yeast tRNA. A Microcon-30 filter (Amicon, Bedford, MA) was used to purify and concentrate the hybridization mixture, which was then adjusted to the final concentration of 3.5× SSC and 0.3% SDS in 60 μL. Following denaturation at 100°C for 1.5 minutes and preannealing at 37°C for 30 minutes, the labeled DNA mixture was applied to the cDNA microarray and incubated at 65°C for 20 to 30 hours. The slide was then washed for 2 minutes each in 0.5× SSC/0.01% SDS, 0.06× SSC/0.01% SDS, and 0.05× SSC, consecutively, at room temperature and spun-dried. The microarray slides were scanned using a GenePix 4000B scanner (Axon Instruments, Foster City, CA).

Data Preprocessing. The fluorescence signal was calculated by subtracting the background intensity from the total intensity of a spot in GenePix Pro 4.1 software. Spots with poor signals (F532 ¡−1.5 × B532 < 0 or F635 ¡−1.5 × B635 < 0) were removed from further analysis (spot quality filtering). The discrepancies in the labeling efficiencies were adjusted by a “within print-tip Lowess normalization” (18). Information regarding the chromosomal locations of the genes were obtained from SOURCE http://genome-www5.stanford.edu/cgi-bin/source/sourceSearch).

Statistical Methods I: Selection of Amplified Genes. This paper classified the “copy number alteration” into two categories. The first was “gene amplification”, meaning the conversion of an intact copy of a gene into two copies. For this, the “noise level” of the fluorescence intensity ratio within the normal tissues was first calculated. When the normal gastric tissues from the male (XY) and female (XX) genomic DNA were cohybridized in the control microarray CGH experiments, the genes on the autosomes were varied in log₂ (Cy5/Cy3) ratios within ± 0.2 ranges. After adding 1 SD to this, it was decided that the genes on the normal gastric tissues were varied between the ± 0.3 range of the log₂ (Cy5/Cy3) ratios. The average male to female ratio of the genes on the X chromosomes, XX/X, which could represent a single gene amplification into two genes, was calculated and a log₂ value of 0.58 was obtained. Therefore, the genes with log₂ (Cy5/Cy3) ≥ 0.58 were determined to be those showing gene amplification. The term “frequency” in the text refers to the percentage of copy number alteration incidences for each gene from the 30 experiments. Therefore, after combining the intensity ratios and frequency, the ratio of log₂ (Cy5/Cy3) ≥ 0.58 or ≤ ¡−0.58 was observed in at least 1 of the 30 cases tested (3.3% frequency) for the gene to be qualified as amplified.

Statistical Methods II: Selection of Genes That Distinguish Recurrent and Nonrecurrent Groups. The second aspect of the gene copy number change events in this paper is the effect of combined gene copy change ratios. Without consideration of log₂ (Cy5/Cy3) having to be > 0.58 or < ¡–0.58, all the genes that passed quality filtering and that were located at chromosome 20 were considered. Patients were first divided into a recurrent and nonrecurrent groups. To evaluate the gene copy number change and their possible relation to the distinction of recurrence status of patients, genes that showed differential copy number changes between the recurrent and nonrecurrent groups were selected by a t test. The tissue samples without information for recurrence status were removed from the analysis. From the t test, 5 genes (P < 0.01), and 39 genes (P < 0.05) containing these 5 genes, were selected. A modified version of prognostic scoring system used by Inoue et al. (19) was devised, and each patient was given a score S_i:

where S_i is the score of patient i, and G_ik is the log₂ (Cy5/Cy3) of gene k of patient i.

Logistic regression analysis was done to determine if the score of each patient was correlated with a recurrence. The patients were classified into two or three groups based on the fact that S_i was likely to reflect the genetic alteration level. According to the scores of the 5 or 39 genes, 27 patients were divided into two groups (Table 1; 5-gene grouping of patients) or three groups (39-gene grouping of patients). In the 5-gene grouping of the patients, the patients were divided into a high group with S_i ≥ 30 (14 cases) and a low group with S_i ≤ 30 (13 cases). In the 39-gene grouping, the patients were divided into the high group with S_i ≥ 292 (12 cases), a middle group with 0 ≤ S_i ≤ 292 (7cases), or a low group with S_i ≤ 0 (8 cases).

The disease-free survival curves were constructed according to the Kaplan-Meier method and compared with the G-rho family of tests including the log-rank and Gehan-Wilcoxon test. Cox regression analysis was used for multivariate survival analysis with a backward stepwise selection procedure, and the following were entered as the categorical covariates: the tumor stage (I, II versus III, IV), the age at diagnosis (< 65 years, ≥ 65 years), gender, lymph node metastasis (N_1-2 versus N₃), and genetic group based on the selected genes (5-gene grouping, 39-gene grouping; Table 1). The association between the genetic groups and the stage groups were analyzed using Fisher's exact test.

Global View of Gene Copy Number Changes on Chromosome 20 in Gastric Cancers. From the cDNA microarray-based CGH of the 30 pairs of gastric samples, ∼13,000 genes passed the spot quality filtering. Of these, 7,064 genes had chromosomal location information according to SOURCE (Data Preprocessing in MATERIALS AND METHODS), and the mean of log₂ (Cy5/Cy3) for 7,064 genes in 30 samples were plotted according to their chromosomal locations (Fig. 1A): 296 of the 7,064 genes were located on chromosome 20. Of these, 84 (∼30%) were located on the p arm, 8 had map locations spanning the centromere, and the remaining 204 were at the q arm of chromosome 20. In order to search for the amplified genes in chromosome 20, 296 genes were screened for the log₂ (Cy5/Cy3) ratios of either ≥ 0.58 in at least 1 (3.3%) out of the 30 samples tested. One hundred and twenty-three genes were selected and their mean log₂ (Cy5/Cy3) values (Fig. 1B) and incidences (Fig. 1C) were plotted according to the map locations. Approximately 30% of the copy number-altered genes (39 of 123 genes) were located on the p arm of chromosome 20, whereas the other 70% (84 of 123) had map locations at the q arm.

Amplified Genes at Chromosome 20 in Gastric Cancer. This study surveyed the gene amplification only (not the deletions) at human chromosome 20 [log₂ (Cy5/Cy3) ≥ 0.58, 3.3%]. Of the 123 unique genes, 96 passed these criteria (Table 2); and of the 96 amplified genes, 29 genes (30%) were located at the p arm, 1 at the centromere region, and 66 (70%) at the q arm of chromosome 20. The higher frequency of the gene amplification events at the q arm of chromosome 20 compared with the p arm, does not necessarily guarantee the true biologically active gene copy change distribution at the q arm until proven otherwise, but represents the relative proportions of the copy changes among the genes spotted on the microarray at this point. ZNF217 and CYP24A, which are well-known oncogene candidates that have previously been shown to be within an amplicon boundary in 20q13.2 of gastric cancer (15), were among the genes selected.

Association Between the Gene Copy Number Changes at Chromosome 20 and the Clinical Outcome. Because no correlations could be found between the gene amplification and the clinical parameters when the criteria of ≥ 1.5-fold of Cy5/Cy3 and ≥ 3.3% frequency was used, we examined whether or not the molecular signature of the combined Cy5/Cy3 of multiple genes, albeit with low ratios, could provide any useful information. In order to select the discriminatory genes, we compared the genetic alteration pattern in the recurrent and the nonrecurrent group comprising 27 out of the 30 patients whose recurrence status was known (Table 1). Initially, t tests were used to select a subset of discriminatory genes from all the genes on chromosome 20. t tests with nominal P values of 0.01 and 0.05 produced 5 and 39 genes, respectively, of which the 5 genes, PTPNS1, CST7, PRO0628, PCK1, and COL9A3, with P < 0.01 were also in the 39-gene list with P < 0.05. Two-way hierarchical clustering showed that these 5 and 39 genes, which were visualized via TreeView dendrogram, were able to classify most of the 27 cases into either the recurrent group or nonrecurrent group based on their copy number change patterns (Fig. 2). The next stage was to identify any correlation between the selected groups of 5 or 39 genes and the recurrence status of the patients. In order to classify the patients according to the genetic alteration levels of the selected genes, each patient was given a score using a logistic regression analysis (MATERIALS AND METHODS). The 27 patients were divided into two or three groups based on the patients' scores reflecting the genetic alteration level (Table 1; Fig. 3), and logistic regression analysis was done to determine if the score-based grouping of the patients reflects the recurrence versus nonrecurrence status. The results of logistic regression analysis confirmed that the scores of each patient for both of the 5 and 39 genes were strongly correlated with the recurrence of cancer (P < 0.05 in both cases). Using Fisher's exact test, it was confirmed that the genetic subgroups and the gastric cancer stage were not significantly associated (P = 0.12). Although the use of a traditional staging system for determining the prognosis of gastric cancer represents a statistically significant difference for a nonrecurrence, the Cox regression analysis result and the Kaplan-Meier curves showed that the patients grouped according to the alteration of the specific genes were also highly associated with nonrecurrence (Fig. 4). The group with a higher level of copy number changes in the recurrence-related genes on chromosome 20 showed a shorter survival rate (Fig. 4B and C). Although it was shown that patients with a late stage disease have a shorter survival rate than those patients with an early stage disease (P < 0.05; Fig. 4A), the recurrence difference based on the gene copy number changes were also statistically meaningful (P < 0.05).

Human chromosome 20 is of special interest due to the presence of a susceptibility locus for hereditary diseases, which include various types of cancers (20). Genomic instabilities at chromosome 20, particularly at the 20q13 region, are frequently examined for the presence of specific oncogenes and gene amplification events in various types of tumors and/or cancer cell lines. These genes include CYP24, ZNF217 (15), and BTAK (21) in gastric cancer; CTSZ, NCOA6, and PTPN1 in pancreatic cancer (22); E2F1, TGIF2, PTPN1 BTAK, and ZNF217 in ovarian cancer (23); and AIB4, ZNF217, CYP24, and aurora 2 kinase in breast (24, 25) and colorectal cancer (26). Almost all these gene amplification events have been examined using the BAC clone-based approaches. In many cases, these BAC clones are several hundred kilobases in length, and approaches of single gene probing, one at a time on a small scale, need to be taken in order to identify the copy-changed genes specifically. These laborious and time-consuming approaches often result in the reconfirmation of a previously known specific gene amplification event. On the other hand, cDNA microarray-based CGH can look at the gene copy change events for all the genes printed on the array in an unbiased manner. The chromosome 20q13.2 region contains differentially amplified genes in primary and metastatic colorectal cancer, and the amplification has been associated with metastasis (27). It has been suggested that there are amplified genes in the 20q13.2 region other than the well-known genes including ZNF217. Many reports repeatedly emphasized that chromosome 20 needs to be examined in more detail because not only are CYP24 and ZNF217 located in this specific region of 20q but other BAC array-based CGH experiments have continuously highlighted this region as being amplified in various types of tumors. The current study of cDNA microarray-based CGH analysis provides a list of candidate amplified genes in chromosome 20 (Table 2). Log₂ (Cy5/Cy3) ≥ 0.58 was used to specifically select genes that are amplified. This is a relatively high level of intensity ratio change for a gene to be qualified as copy number–changed. We specifically used this line to select amplified genes that can be confirmed by independent investigators beyond experimental differences. Several experiments are in the planning stage in order to confirm the gene copy number changes shown in this report. First, some of the well-known amplified genes in gastric cancer including BTAK are located in 20q. Unfortunately, the cDNA microarrays used in this study did not contain such probes. Independent confirmation of the result by fluorescence in situ hybridization or Southern blotting for some of the amplified genes in this report together with the genes that have been reported to be amplified but were not spotted in the cDNA microarrays, notably, BTAK, aurora kinase, and AIB4, will be highly meaningful. Second, we were looking for the gene amplification that could contribute to the linear amplification of mRNA yield by selecting the genes that are more likely to be amplified in this study. We have done RNA expression profiling experiments using the same tissue samples and cDNA microarrays. The genes that are highly correlated in the copy change and RNA expression are strong candidates for novel oncogenes. The initial analysis suggests that a comparison of the gene copy change and RNA expression level might require more than just a fold change–based gene selection in the two databases and direct comparison. Instead, a comparison of Pearson correlation coefficients between gene copy number change profile and RNA expression profile, following the idea of correlation analysis between gene expression profiles and chemosensitivity profiles (28), would provide meaningful information.

This study showed that the minimum five genes of PTPNS1, CST7, PRO0628, PCK1, and COL9A3 comprise a gene copy change signature and that the copy numbers of these genes was higher in the patient group with a recurrent gastric cancer. Furthermore, the patient groups with copy increase for these genes are strongly correlated with a poor survival (Fig. 4). This is probably the first study where the gene copy number changes in chromosome 20 show a direct relationship with the patient recurrence status. Ongoing data analysis for the remaining 17,000 genes in the 30 cases of the microarray CGH will yield an almost complete view of the gene copy number change events that occur in gastric cancer.

We thank G.Y. Lee, C.H. Park, and Y.J. Yang for technical support and helpful comments.