Analysis of Colorectal Cancer by Comparative Genomic Hybridization: Evidence for Induction of the Metastatic Phenotype by Loss of Tumor Suppressor Genes1

Current models suggest that colon cancer initiation and progression are secondary to both the activation of oncogenes and the deletion of tumor suppressor genes. The role of each, however, is still poorly understood, particularly with regard to the induction of metastasis. We hypothesized that genetic differences exist between tumors that metastasize distantly and those that do not, and that oncogenes and tumor suppressor genes participate equally in this process. To address this hypothesis, human tumor specimens from localized [tumor-node-metastasis (TNM) stage I-Ill] and primary colon cancers (n = 10) were directly compared with metastatic (TNM stage IV) lesions (n = 10) using comparative genomic hybridization analysis. Although several alterations were shared equally between primary tumors and metastases (+7q, +19q, and +20q), two patterns of distinguishing alterations were observed: (a) alterations that were more extensive in liver metastases than in primary tumors (+8q, +13q, -4p, -8p, -15q, -l’7p, -18q, -21q, and -22q); and (b) alterations that were unique to metastatic lesions (-9q, llq, and 17q). Overall, genetic losses were more common than gains, and, more importantly, the number of losses/tumor was significantly higher for metastases than for primary tumors (9.3 + 1.3 versus 4.1 + 0.7; P = 0.00062, Wilcoxon’s rank-sum test). The distinct predominance of genetic losses in the metastatic lesions when compared with the primary localized tumors provides evidence that the metastatic phenotype is induced by the deletion of tumor suppressor genes and permits the construction of physical maps targeting regions where novel tumor suppressor genes are likely to exist. Received I 1/14/97; revised 1/23/98; accepted 1/26/98. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. I Supported by NIH Grant CA655 12 (to T. J. Y.). 2 To whom requests for reprints should be addressed, at 12902 Magnolia Drive, Tampa, FL 33612. Phone: (813) 979-7292; Fax: (813) 979-3893; E-mail: [email protected]. INTRODUCTION Carcinoma of the large bowel is a common disease that is frequently lethal because of metastatic spread to the liver (1). For many years, it has been suggested that colon cancer tumorigenesis involves multiple steps (2, 3). More recently, Fearon and Vogelstein (4) suggested that these steps involve the mutution or gain of tumor-promoting oncogenes and the mutation or loss of tumor suppressor genes. Although no particular order of genetic events was considered necessary, early genetic alterations were thought to include mutations in the APC and RAS genes, and later alterations were thought to include mutations and deletions of the p53 and DCC genes (5, 6). The majority of genetic studies involving colon cancer have focused on the early changes (7-9) responsible for tumor initiation. Many studies have addressed the differences between normal mucosa and primary cancers, but little is known regarding the genetic steps (10) that promote the later stages of tumor progression and the development of the metastatic phenotype. Current models (1 1) for the induction of metastasis suggest that multiple molecules must be differentially expressed to produce a cancer cell that is capable of producing its own blood supply, invading blood and lymphatic vessels, adhering to end-organ endothelial cells, and growing in distant organ sites. The process likely requires multiple genetic alterations and seems to be quite complex, and its regulation is poorly understood (12). We hypothesized that alterations in oncogenes and tumor suppressor genes equally account for the genetic differences postulated to exist between primary loco-regional (stage I-Ill) tumors and tumors that have spread to distant organ sites (stage IV) such as the liver. This hypothesis was based on the clinical observation that nearly 70% of patients with loco-regional disease are curable, whereas few patients survive once distant spread has occurred (13). To test this hypothesis, we used CGH,3 a relatively new molecular cytogenetic technology designed to assay the entire genome for chromosomal gains and losses (14-16). We attempted to specifically address the hypothesis by directly comparing primary loco-regional cancers with liver-metastatic cancers. By focusing on the interface between cancers that grow locally but do not metastasize to distant organ sites versus those that have already spread to the liver, we have been able to identify and map specific chromosomal loci that are more likely to be altered in the process of conversion to the metastatic phenotype. Both losses that are shared with primary tumors and losses that are unique to metastatic tumors were identified. Our data provide compelling evidence that on May 2, 2017. © 1998 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from 880 Analysis of Colorectal Cancer by CGH Table 1 Clinical and patholo gical data of primary and metastatic col onic carcinomas Case no. Sex Age at diagnosis (yr) Differentiation/grade Size (cm) Location Operative resection Primary carcinomas 1 2 3b 4( 5 6 7 8 9 10” Liver metastases A B C D E F G H’ I J F M F M F M F M M F M M M F F M M M M F 74 73 53 72 43 59 54 48 66 55 66 67 72 56 70 69 75 72 77 65 Moderate1T3N M0 ModeratefF,N( M0 ModeratefF,N ,M0 MucinousIT2N2M1 ModeratefF3N M0 ModerateTF2N ,M0 WellTl’1N ,M() Wellfl’,N2M1 MucinousTF2M ,N0 ModerateTF4N(,M Moderate Moderate Moderate Moderate Moderate Moderate Moderate Mucinous Poor Moderate 4 5.5 4 7 3.5 5.7 2.1 2.8 1.5 4 4 S 8 9 6 8 3.3 8 9.8 8 Rectum Ascending colon Rectosigmoid Ascending colon Sigmoid Rectum Ascending colon Ascending colon Rectosigmoid Rectosigmoid Right lobe Right lobe Left lobe Right lobe Segment 4 Right lobe Right lobe Left lobe Right lobe Left lobe LARU Hemicolectomy LAR Hemicolectomy LAR LAR Hemicolectomy Hemicolectomy Abdominopenneal Abdominopenneal Lobectomy Wedge Lobectomy Lobectomy Segmentectomy Lobectomy Wedge Lobectomy Lobectomy Lobectomy a LAR, low anterior resection. b Primary tumor (tumor 3) and local recurrence (tumor 10) were derived from the same patient. C Primary tumor (tumor 4) and synchronous liver metastasis (tumor H) were derived from the same patient. conversion to the metastatic phenotype may be largely based on the predominance of specific genetic losses, a finding that suggests a primary role for tumor suppressor genes in this complex process. MATERIALS AND METHODS Human Specimens. The material consisted of 20 fresh specimens (at least 2 mm3/specimen), 10 colorectal adenocarcinomas (tumor specimens 1-10) and 10 liver metastases (tumor specimens A-i), that were surgically removed from 18 patients at the Moffitt Cancer Center and Research Institute at the University of South Florida. No patients received preoperative chemotherapy. All specimens were carefully trimmed of all normal adjacent tissues. A frozen section of the specimens was done to confirm the histological diagnosis and the absence of normal adjacent contaminating tissue. The specimens were snap-frozen in liquid nitrogen immediately after resection and stored at 80#{176}C in the Tissue Procurement Laboratory until analysis. High molecular weight tumor DNA was isolated from 200 mg of frozen tumor sections using Trizol reagent (Life Technologies, Inc., Grand Island, NY). Normal DNA was isolated from the peripheral blood lymphocytes of genetically normal males, using the Puregene extraction system (Gentra, Inc., Triangle Park, NC). The samples were checked for purity by nondenaturing agarose gel electrophoresis. Nine of the primary tumors were tumor-node-metastasis (TNM) stage I-Ill. One stage IV tumor (tumor 4) was also included for purposes of comparison with its synchronous liver metastasis (tumor H). Seven primary tumor specimens were derived from the rectosigmoid region, and three specimens were derived from the right colon. Tumor diameters ranged from 1-7 cm. Eight primary tumors were well-differentiated to moderately differentiated, and two were mucinous (tumors 4 and 9). One of the nine primary tumors (tumor 10) represented a local recurrence that was resected for cure. The liver metastasis specimens included seven lobectomies (four of the right lobes and three of the left liver lobes, respectively), one left lateral segmentectomy, and two right wedge resections. Table 1 shows a summary of the clinical and pathological data. All liver lesions were resected for cure, but only 30% were alive without disease at 2 years median follow-up. CGH. CGH was performed using directly labeled fluorescent dUTP in extracted DNA according to specific protocols (Vysis, Inc., Downers Grove, IL). DNA samples from tumors were labeled with Spectrum Red dUTP (Vysis, Inc.), and reference DNA was labeled with Spectrum Green dUTP (Vysis, Inc.) with the nick translation technique. The nick translation reaction was monitored to obtain a probe size of 300-3000 bp. The probe size was checked through nondenaturing agarose gel electrophoresis. In Situ Hybridization. Spectrum Red-labeled tumor DNA and Spectrum Green-labeled reference DNA (200-400 ng) and 10 i.g of human Cot-i DNA were combined and precipitated in the presence of 3 M sodium acetate and 2.5 volumes of iOO% ethanol. The hybridization mix was then centrifuged, and the supernatant was removed, resuspended in CGH buffer, allowed to preanneal at 37#{176}C for 30 mm, and denatured at 73#{176}Cfor S mm. The hybridization probe was immediately applied to previously denatured normal chromosome metaphase slides (Vysis, Inc.), coverslipped, sealed with rubber cement, and placed in a sealed moist hybridization chamber in a 37#{176}C incubator for 72 h. After hybridization, the slides were washed two times in 0.4X SSC and 0.3% NP4O wash on May 2, 2017. © 1998 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from Clinical Cancer Research 881 solution at 74#{176}C ± i#{176}C for 2 mm and once in 2X SSC and 0.i% NP4O at ambient temperature for S s to 1 mm. After air drying, the slides were counterstained with 4’,6-diamidino-2-phenylindole (Vysis, Inc.) in antifade solution. Digital Image Analysis. Three successive images/ metaphase were captured with a cooled charge-coupled device camera (Photometrics, Tucson, AZ) mounted on a Leitz Orthopan 2 epifluorescence microscope interfaced to a Power Macintosh 8i00/i00. Ten to 15 three-colored digital images of metaphase chromosomes were collected per hybridized tumor. Of these, the best 8-10 smooth high-quality images were selected for further analysis. Nearly 10,000 individual chromosomes were analyzed. Relative DNA copy number changes were found by analyzing the fluorescent ratio of tumor and reference DNA along the axis ofeach chromosome, as described previously (14, 17). After background correction and normalization of the red: green ratios to 1 .0, red:green fluorescence intensity ratio profiles were calculated along the length of each chromosome using IPLabs and QUIPS software (Vysis, Inc.). Hybridizations of Spectrum Red-labeled and Spectrum Green-labeled normal male DNAs disclosed a mean ratio and corresponding SD between 0.9 and ii. Positive control experiments using the MPE600 breast cancer cell line were performed, and the results were directly compared with described chromosomal alterations (18). Two negative control CGH experiments in which aiiquots of normal DNA were labeled differentially with green and red fluorochromes, respectively, were performed to identify chromosomal regions with artifactually abnormal fluorescent profiles. Regions immediately adjacent to and including heterochromatic segments were excluded from analysis, because they typically showed very low red and green fluorescence intensities, due to partial hybridization suppression with Cot-i DNA. A threshold of 1 .2 and 0.8 for the red:green ratio values was set for gains and losses of test DNA material. Statistical Analysis. The statistical significance between the number of gains and losses of primary and metastatic tumors and the frequency of alterations in selected chromosome arms was calculated using Wilcoxon’s rank-sum test and the twotailed Fisher’s exact test, respectively. RESULTS Controls. To assess the sensitivity and validity of the CGH methodology, two normal negative control male DNAs (Vysis, Inc.) and a positive breast cancer DNA specimen (MP600; Vysis, Inc.) were studied (data not shown). The CGH analysis of the normal DNA specimens (negative controls) identified no genetic alterations using the parameters outlined in “Materials and Methods,” but analysis of the positive control found all of the previously reported (18) alterations (+ lq, -9p, +liq, -liqter, +i3qter, -i6q, +l7q, +X, and -Y). Overview of Genetic Changes. Figs. i and 2 summarize the chromosomal abnormalities (gains and losses, respectively) found in the primary and metastatic colorectal carcinomas. All evaluated specimens showed DNA copy number changes by CGH. The tumors harbored at least 3 chromosomal alterations, with a range of 3-16 and 7-20 chromosomal alterations in the primary and metastatic tumors, respectively. Four of 10 primary tumors (40%) showed more relative DNA sequence gains than losses, 5 of 10 tumors (50%) had more losses than gains, and 1 of 10 tumors (10%) had equal numbers of gains and losses. On average, there were 9.6 alterations/primary tumor: 5.3 gains (range, 2-12 gains) and 4.3 losses (range, 1-8 losses; Fig. 3). Metastatic tumors harbored more chromosomal abnormalities than primary tumors, and all had more relative DNA sequence losses than gains, with an average of 12.6 alterations/tumor [9.3 losses (range, 5-17 losses) and 3.3 gains (range, 0-6 gains)]. A representative three-color CGH image demonstrating visually detectable gains and losses in a liver metastasis is shown in Fig. 4. The mean number of losses/tumor was greater in the metastatic neoplasms than in the primary neoplasms (Fig. 3), with a highly significant P of 0.00062 (Wilcoxon’s rank-sum test). The total number of gains and losses was also greater in the metastatic neoplasms, but the difference was not statistically significant. The values for chromosomes X and Y were consistent with the gender of the patients. Primary Tumors. All patients bearing primary localized tumors that were resected for cure were alive without evidence of disease at follow-up (median follow-up, >2 years). All of the tumors showed gains (Fig. i) and losses (Fig. 2) collectively affecting all of the chromosomes with the exception of chromosome 2. Four tumors showed sequence copy number gains, and seven tumors showed gains at one or more chromosomal region. Frequent gains involved 20q (four tumors), 13q (four tumors), and 19 (three tumors). The minimal overlapping regions of gain were 9q21.3, 2oqli.23-qi3.l, l3qi4-q2i, and 7cen-qil.2. Frequently lost chromosomal regions were l8q, l7p, i8q, 8p, and Sq. The minimal common regions of loss were 18q2i-qter, l7pter-p12, i8pter-pl 1.3, 8pter-p23, and Sq2l.2-q2i.3. There was no relationship between the tumor size and the number of chromosomai alterations. Metastatic Tumors. Consistent with national statistics for hepatic resection of colorectal liver metastases, 30% of all patients who underwent resection of liver metastases were alive at the time of follow-up (median follow-up, >2 years). Eight of the 10 metastatic tumors showed both regional gains (Fig. i) and losses (Fig. 2), and two tumors had only losses. Losses were more common than gains, and the total number of losses/tumor for metastatic lesions was significantly greater than the total number of losses/tumor for primary lesions. Two tumors (tumors F and G) showed copy number losses of chromosome i8. Common regional losses included regions involving l8q (90%), l7p (90%), 8p (80%), and 22q (70%), and 40% of the tumors had a regional loss at 4p, 9q, iSq, and 17q. The minimal overlapping regions of loss for chromosome 18 were 1 8q22qter in 100%, l8q2l-qter in 80%, and 18qi2-qter in 70% of the tumors. Chromosome 17 had overlapping losses involving l7p23-cen (four tumors), l7pter-p12 (four tumors), and l7pter-pi3 in six tumors. The remaining chromosomes had the following common overlapping regions of loss: (a) 8pter-p22; (b) 22ql1.2-qi2; (c) 22ql2.3-qi3.l; (d) 22ql3.3; (e) 4pl6; (f) 9q2l; (g) 9q34.3; and (h) i5q26. Two distinct patterns of common chromosomal losses were observed: (a) losses in metastases that involved the same chromosome arms as the primary tumors but were more frequent or extensive (4p, 8p, lSq, l’7p, and i8q); and (b) regional losses unique to the metastatic tumors (9q, 1 lq, and 17q). Regional gains per tumor most commonly affected l3q (50%), 8q (40%), on May 2, 2017. © 1998 American Association for Cancer Research. clincancerres.aacrjournals.org Downloaded from