Colorectal cancer (CRC) is one of the leading causes of cancer-related deaths in the United States. In its early stages, colon carcinogenesis is associated with alterations in molecular signaling pathways and then progresses as a result of a sequential accumulation of events that either activate oncogenes or inactivate tumor suppressor genes.1 Some alterations participating in multistep colon carcinogenesis affect APC, K-Ras, p53 and elements of the TGF-β signaling pathway, such as TGFβRII and SMAD4. The TGF-β signaling pathway serves as a major tumor suppressor in a variety of human gastrointestinal tumors, including colon, gastric and pancreatic cancer.2 The downstream transcriptional targets of the TGF-β signaling pathways are key mediators for regulating cellular proliferation, extracellular matrix production and immune surveillance, and also dictate gastrointestinal epithelial development.
The runt-domain related (RUNX) transcription factors are mammalian homologues of the Drosophila genes runt and lozenge, and are heterodimeric proteins composed of 2 highly conserved DNA binding subunits, α and β.3RUNX transcription factors are one of the important targets of TGF-β superfamily signaling and play critical functions in mammalian development. RUNX proteins have been shown to interact with downstream SMAD proteins in mediating the growth suppressive effects of TGF-β and play an important role in development and oncogenesis.3 Three mammalian runt-related genes, RUNX1, RUNX2 and RUNX3 have been described. RUNX1 is an indispensable factor in hematopoiesis and angiogenesis, and anomalies in this gene are involved in about 30% of the cases of human acute leukemia.4RUNX2 collaborates with c-myc and Pim-1, and its over-expression in mice predisposes to the development of T-cell lymphoma.5, 6
RUNX3 is a putative tumor suppressor gene localized to chromosome 1p36, a region that exhibits frequent loss of heterozygosity (LOH) events in colon, gastric, breast and ovarian cancers. More recently, silencing of RUNX3 has been reported in gastrointestinal cancers in mice as well as humans.7, 8, 9, 10 Li and colleagues8 demonstrated that the gastric mucosa of the RUNX3-null mouse exhibited hyperplasia due to stimulated proliferation and suppressed apoptosis of the epithelial cells. These cells were resistant to growth-inhibitory effects of TGF-β, indicating that RUNX3 regulates the growth of gastric epithelial cells.8 Physical interactions between the C-terminus of RUNX3 and SMAD that mediate transmission of TGF-β induced growth inhibitory signals to the nucleus have been already described.11
Because the TGF-β mediated signaling pathway regulates cell growth in the human colon, it is possible that RUNX3 plays an important tumor suppressor role. In our study, we determined the prevalence of RUNX3 expression and its promoter methylation in human colon cancer cell lines and primary cancers, and defined the association of these events with MSI and TGFβRII status. We further demonstrate that hypermethylation of the RUNX3 promoter is a key mechanism of its inactivation, as treatment of cell lines with 5-aza-2′-deoxycytidine (5-AzaDC) was able to demethylate its promoter region and restore RUNX3 expression.
MATERIAL AND METHODS
Cell lines and cell culture
Seventeen human colon cancer cell lines were obtained from the American Type Culture Collection (Manassas, VA). The HCT116+chr3 cell line was created by the stable transfer of chromosome 3 and was grown in the presence of 400 μg/ml G418 (geneticin) in the culture medium.12 All cells lines were cultured in Iscove's modified Dulbecco's medium (IMDM) with 10% fetal bovine serum (Life Technologies, Inc., Grand Island, NY) and maintained at 37°C in 5% CO2. The characteristics of the cell lines are shown in Table I.
Table I. Characteristics of Cell Lines used and its Correlation with the RUNX3 mRNA Expression and Methylation Status1
One hundred primary colon cancers were obtained from patients with sporadic colorectal cancer treated by physicians associated with either the Cancer and Leukemia Group B (CALGB) or the University of California San Diego (UCSD). Institutional Review board approval was granted for this study. The colon cancers investigated in the current study were randomly selected from a larger pool of 267 specimens, and the cohort descriptions have been previously published.13
5-Aza-2′-Deoxycytidine (5-AzaDC) Treatment
SW48 and HCT15 cells were seeded at 1×105 cells/ml in a 100 mm culture dish in IMDM growth medium. Following 24 hr of growth, the cells were treated with 0.1 μM 5-AzaDC for a 24 hr period. Media was replaced at the end of the treatment period and the cells were allowed to grow further. DNA and RNA were extracted from the cells at pretreatment, and 3 and 5 days posttreatment.
Colon cancers and matching normal tissue specimens from a site distant from the target lesion were obtained from all patients. The histological type and grade of tumors were classified according to the established criteria. Paraffin-embedded primary tissue and control samples were prepared from H&E-stained 5 mm section slides. Genomic DNA from tumor and matched normal tissues was extracted by microdissection from paraffin-embedded archival tissues. The microdissected tissues were hydrated, digested in Proteinase K, followed by the Wizard DNA Clean-up System (Promega, Madison, WI) using standard protocols described previously.14 For the extraction of DNA from cultured colon cancer cells, we used a DNAeasy System (Qiagen, Inc., Valencia, CA) according to the instructions provided.
MSI determination and TGFβRII mutations
Microsatellite analysis was performed on all matched normal and tumor tissues by PCR amplification using a panel of 5 NCI-workshop recommended markers including 2 mononucleotide (BAT25 and BAT26) and 3 dinucleotide repeat sequences (D2S123, D5S346 and D17S250).15 Polymerase chain reaction (PCR) was performed using 32P-labeled primers and subsequent electrophoresis on 8% polyacrylamide gels to determine the changes in electromobility shifts as described previously.13, 14 Tumors with a shift in at least 2 of the 5 recommended markers were classified as MSI-H, in accordance with the international criteria.15 MSI-L was defined as a shift in only 1 of the 5 markers.
The presence of mutation of the poly (A)10 tract of TGFβRII were investigated by PCR followed by polyacrylamide gel electrophoresis as described above. The presence of bandshifts or an additional band was interpreted as a mutation.13
RNA extraction and RT-PCR
RUNX3 mRNA expression was measured by reverse-transcriptase PCR. Total cellular RNA was extracted using TRIzol (Life Technologies, Inc., Grand Island, NY) according to the manufacturer's instructions. cDNA was reverse transcribed from 1 μg of total cellular RNA in 20 μl reactions containing 25 μg/ml of random hexamers (Roche Molecular Biochemicals, Indianapolis, IN), 10,000 U/ml of Moloney murine leukemia virus reverse transcriptase (Life Technologies, Inc., Grand Island, NY), 50 mM Tris-HCl (pH 8.3), 75 mM KCl, 3 mM MgCl2, 10 mM dithiothreitol, 0.4 mM dNTPs, 2,000 U/ml of RNase Inhibitor (Stratagene, La Jolla, CA) and 8.5 μl of diethyl pyrocarbonate-treated water. Subsequently, we determined the mRNA expression of RUNX3 by PCR.
A PTC200 DNA Engine (MJ Research, Waltham, MA) was used to amplify 2 μl of cDNA products in 50 μl reactions containing 10 mM Tris-HCl, pH 8.3, 10 mM KCl, 0.4 mM each dNTP and 100 units/ml AmpliTaq DNA Polymerase (Life Technologies, Inc., Grand Island, NY) in a semiquantitative RT-PCR using RUNX3 and β-actin specific primers. For human RUNX3, the primers were 5′-GAGTTTCACCCTGACCATCACTGTG-3′ (sense strand), and 5′-GCCCATCACTGGTCTTGAAGGTTGT-3′ (antisense strand), which yielded an 870 bp PCR product.8β-actin mRNA was also amplified in the same PCR reaction as an internal control using the following primers; 5′-TACCACTGGCATCGTGATGGACTC-3′ (sense strand) and 5′-TCCTGCTTGCTGATCCACATCTGC-3′ (antisense strand), resulting in a 642 bp PCR product.16 Hot-start PCR was performed by adding the DNA polymerase at 80°C after initial denaturation. The PCR amplification cycles consisted of denaturation at 95°C for 3 min; 30 cycles of denaturation at 95°C for 30 sec, annealing at 59°C for 30 sec and extension at 72°C for 60 sec, and a final elongation at 72°C for 10 min. PCR products were separated on a 1% agarose gel, stained with 0.5 μg/ml ethidium bromide and visualized by ultraviolet (UV) light.
Methylation specific PCR (MSP) assay
Methylation specific PCR was performed on bisulfite modified DNA obtained from human colon cancer cell lines as well as clinical materials as described previously.17 Briefly, 0.5–2.0 μg of genomic DNA was denatured with NaOH and was treated with sodium bisulfite and subsequently purified using the Wizard DNA Clean-up System (Promega, Madison, WI). The modified DNA was used as a template for MSP using primers specific for either the methylated or the modified unmethylated RUNX3 promoter sequences. Primer sequences for the unmethylated reaction were 5′-TTATGAGGGGTGGTTGTATGTGGG-3′ (sense) and 5′-AAAACAACCAACACAAACACCTCC-3′ (antisense), and primer sequences for the methylated reaction were 5′-TTACGAGGGGCGGTCGTACGCGGG-3′ (sense) and 5′-AAAACGACCGACGCGAACGCCTCC-3′ (antisense). Step-down PCR reactions were performed in a 25 μl reaction volume containing 1× PCR buffer (Invitrogen Life Technologies, Carlsbad, CA), 2.5 mM MgCl2, 200 μM dNTPs, 0.5 μM of each PCR primer, 0.75 units of AmpliTaq polymerase and approximately 25 ng of bisulfite-modified DNA, as described previously. Reactions were hot-started at 95°C for 5 min. This was followed by 33 cycles at 95°C for 45 sec, 57°C for 30 sec and 72°C for 30 sec, followed by a 10 min extension at 72°C in a PTC 200 DNA Engine™ Thermocyler (MJ Research, Inc., Waltham, MA). The amplification products were separated on a 3% agarose gel and visualized by ethidium bromide staining and UV transillumination.
All statistical analyses were performed at the CALGB Statistical Center, Duke University, Durham, NC. Associations between RUNX3 promoter methylation and protein expression status were tested using the chi-square test or Fisher's exact test, as appropriate. Data were analyzed using SAS and S-Plus statistical software.
Reduced or lack of RUNX3 expression in colon cancer cell lines
Semiquantitative expression of human RUNX3 was determined in 17 colon cancer cell lines. Eleven (65%) of the 17 colon cancer cell lines studied did not express RUNX3 transcripts (Table I and Fig. 1a). Six of the 11 cell lines (55%) that did not express RUNX3 mRNA were microsatellite unstable, and their MSI status was attributable either to the presence of germline defects in the key DNA mismatch repair genes or due to the epigenetic inactivation of the hMLH1 promoter. The remaining 5 cell lines that lacked RUNX3 transcripts were all microsatellite stable (MSS). Among these 5 cell lines, SNU-81 and SNU-503 harbored mutations in hMSH6, while the remaining 3 cell lines (HT29, Caco2 and LIM6) were MMR proficient.
Frequent hypermethylation of RUNX3 promoter in colon cancer cell lines
RUNX3 gene expression is regulated by 2 promoters P1 and P2. However, only P2 promoter region of RUNX3 contains a large (4.2 kb) CpG island (Genebank accession number AL023096).18 To determine whether aberrant CpG island methylation is responsible for transcriptional silencing of RUNX3, we analyzed the promoter region of various colon cancer cell lines using methylation specific PCR, using 2 sets of primers previously described to amplify methylated and unmethylated RUNX3 alleles.8 We found evidence for CpG island methylation in all of the cell lines that exhibited reduced or a complete absence of RUNX3 mRNA expression (Table I and Fig. 1b). In total, 9 (53%) of 17 colon cancer cell lines demonstrated complete methylation of the RUNX3 promoter near its transcription start site. On the other hand, the CpG island of RUNX3 was completely unmethylated in 6 (35%) of the cell lines and 2 (12%) cell lines were hemimethylated.
Methylation of cytosines within CpG islands has been shown to be associated with loss of gene expression due to transcriptional repression and is commonly observed in various human cancers.19. Our data demonstrated a strong correlation between 5′ CpG island methylation and RUNX3 expression, as 9 (82%) of the 11 cell lines that did not exhibit RUNX3 mRNA expression exhibited only methylated alleles for RUNX3 promoter. Additionally, 6 (67%) of 9 cell lines that exhibited complete methylation of RUNX3 promoter demonstrate microsatellite instability as a result of either MMR gene mutations or epigenetic inactivation of hMLH1.
5-AzaDC treatment restores RUNX3 expression
To test the hypothesis that methylation was responsible for transcriptional silencing of RUNX3, we treated colon cancer cell lines with the demethylating agent 5-AzaDC. The SW48 and HCT15 colon cancer cell lines, which were fully methylated at RUNX3, did not express the RUNX3 transcript by RT-PCR. However, following treatment with 5-AzaDC, RUNX3 mRNA was readily detected in SW48 colon cancer cells (Fig. 1c). The ability of 5-AzaDC to restore expression of RUNX3 confirmed the role of methylation in the epigenetic silencing of this gene.
RUNX3 methylation is widespread in colorectal cancers
We next examined primary colorectal cancers for the evidence of RUNX3 CpG island methylation. Informative results were obtained from 91 of 100 cancers. by MSP, RUNX3 methylation was detected in 21% (19 of 91) of colon cancers (Table II). Unlike colon cancer cell lines, methylated primary tumors always displayed evidence of an accompanying unmethylated PCR product. These unmethylated RUNX3 alleles most likely reflect either the unavoidable presence of normal cells in the cancer specimen or heterogeneity of the methylation event within the tumor cell population itself.
Table II. Summary of the Frequency of RUNX3 Promoter Hypermethylation in Sporadic Colorectal Cancers and its Relationship with Microsatellite Instability Status
Significantly different when compared with MSI-L/MSS tumors (chi square, p = 0.012).
To investigate the possibility that RUNX3 promoter methylation is a cancer specific event,17, 20 we studied normal colonic epithelial DNA from 21 individuals with no evidence of tumor. Only 1 of these cases showed evidence for aberrant RUNX3 methylation in the normal mucosa, indicating that the methylation observed is relatively cancer specific. This single case showing RUNX3 methylation in the normal colonic mucosa was found in a patient who also had a MSI-H colon cancer.
In our study, we grouped MSI-L and MSS tumors together for comparison purposes, since both have similar molecular and clinical features and do not differ in clinical outcome.21 Of particular interest, RUNX3 promoter methylation was approximately 3 times more common in MSI-H tumors (33% of MSI-H vs. 12% of MSI-L/MSS tumors; chi-square p=0.012) as depicted in Table II.
TGFβ signaling may be abrogated by inactivation of either TGFβRII or RUNX3 in MSI-H CRCs
RUNX3 is a putative tumor suppressor gene functioning downstream in the TGF-β growth regulatory pathway. Theoretically, the TGF-β signaling pathway could be dysregulated equally well by inactivation of either RUNX3 or TGFβRII. Interestingly we observed that of all the tumors analyzed, 68% (13 of 19) of tumors with evidence for RUNX3 hypermethylation did not exhibit inactivating mutations in TGFβRII, while 32% (7 of 19) tumors also exhibited simultaneous mutations in the TGFβRII (Table IIIa,b).
Table III. Frequency of RUNX3 Methylation and TGFβRII Mutations
chi square, p = 0.92
a) MSI-H sporadic CRCs
b) MSS/MSI-L sporadic CRCs
Furthermore, when this cohort of sporadic CRCs was segregated based on their microsatellite instability status, we observed that 33% (13/40) of MSI-H CRCs were hypermethylated at the RUNX3 promoter, while 45% (18/40) of MSI-H CRCs harbored inactivating frameshift mutations in the TGFβRII gene (Table IIIa). We also found that of the MSI-H tumors with evidence for RUNX3 hypermethylation, more than half (7/13; 54%) did not exhibit mutations at the (A)10 sequence of TGFβRII while 46% (6/13) of MSI-H CRCs demonstrated simultaneous epigenetic inactivation of RUNX3 and mutational inactivation of TGFβRII. RUNX3 methylation was also observed in 12% of MSI-L/MSS CRCs (Table IIIb). TGFβRII mutational inactivation was not found in any of these tumors. These data indicate that the TGF-β growth regulatory pathway may also be deranged in MSI-L/MSS CRCs, and in such cases RUNX3 may be a preferred target.
Accumulating evidence suggests that DNA methylation plays an important role in cancer pathophysiology and that promoter hypermethylation is often an early event in multistep carcinogenesis.22 Aberrant hypermethylation of CpG islands in the promoters of certain tumor suppressor genes leads to loss of gene function in some of them. In colorectal cancers, methylation-associated silencing affects several critical molecular pathways leading to cellular immortalization and transformation, including perturbations in cell cycle regulation (p16 and Rb), cellular adherence (E-cadherin and TIMP-3), metabolic detoxification enzymes (GSTP1) and the DNA damage response pathways (hMLH1, BRCA1, MGMT and p14ARF).23
Several inherited and/or acquired gene and chromosome defects, in parallel with the acquisition of new phenotypic features, have been described for human colorectal preneoplastic and neoplastic lesions during the progression of multistep colorectal tumorigenesis.1 Among the chromosomal aberrations, those affecting chromosomes 5q, 17p and 18q have been demonstrated to be the most frequent in colorectal adenocarcinomas.24, 25 Similarly, classic cytogenetic and molecular genetic studies have revealed that alterations at the telomeric end of the short arm of chromosome 1 are common events in colorectal carcinomas. In particular, microsatellite polymorphism analysis has revealed 1p deletions, the majority encompassing the 1p36 band in about 80% of all colorectal adenocarcinomas examined.26. The minimally deleted region in chromosome 1 from gastrointestinal epithelial cells has been mapped between the genomic markers D1S119 and D1S246, which covers about 45 megabase pairs and contains about 200 genes, including RUNX3, but no other known tumor suppressors.27
Recently, a causal relationship has been identified between the loss of RUNX3 expression and the incidence of gastric cancer in humans, as well as mouse.7, 8 In the current study, we have demonstrated that the majority of human colon cancer cell lines do not express RUNX3, and this lack of expression significantly can be attributed to hypermethylation of the promoter region of the gene. We further demonstrated that inactivation of RUNX3 in CRC cell lines is primarily due to CpG island methylation, as treatment with 5-AzaDC was able to demethylate the promoter and restore RUNX3 expression. Our observation that RUNX3 is frequently silenced in colon cancer cell lines and primary CRCs is consistent with the concept that RUNX3, localized on a strategic location on 1p36, is probably one of the elusive candidate tumor suppressor genes previously not appreciated to play a role in the development of colorectal cancer.
Tumor suppressor genes tend to be preferentially inactivated by specific mechanisms that result in signature mutations that define the type of genetic or epigenetic instability involved. In CRCs, for example, TP53 is most often inactivated by point mutation and hemizygous deletion, p16/CDKN2A is generally inactivated by homozygous deletion and/or DNA methylation, and hMLH1 is inactivated by mutation in familial colon cancer but by DNA methylation in sporadic colon cancers. The implication of such observations is that there might be a group of tumor suppressor genes for which the primary mechanism of inactivation is through promoter hypermethylation.28 Structural mutations and deletions of RUNX3 are rarely observed.8 The high prevalence of RUNX3 promoter methylation observed in our study and in studies of gastric and pancreatic cancers8, 9, 10 further suggest that mechanisms other than methylation are less likely to be involved, as methylation-induced silencing would relieve selective pressure for other mechanisms of tumor suppressor gene inactivation. Our findings underscore the key role of inactivating the growth suppressive function of the TGF-β signaling pathway in colorectal carcinogenesis and indicate that the preferred target in this serial pathway is determined by predominant mode of genetic or epigenetic instability in the tumor. One might speculate there is little selective pressure to inactivate both TGFβRII and RUNX3.
In our study of sporadic CRCs, the association between RUNX3 promoter methylation and MSI-H suggests that these cancers have acquired a “methylator” phenotype, resulting in MMR deficiency by hMLH1 promoter methylation. This mechanism of cancer development has been proposed for sporadic MSI-H CRCs.20 A significant proportion of tumors with RUNX3 promoter methylation did not exhibit frameshift mutations in TGFβRII, and the frequency of tumors demonstrating RUNX3 hypermethylation was quantitatively similar to those with mutations in TGFβRII. Moreover, RUNX3 silencing was also evident in some MSI-L/MSS CRCs that did not harbor TGFβRII mutations. It is generally believed that the TGF-β mediated growth inhibitory pathway may not be involved in carcinogenesis in these tumors.
There is lack of clear understanding on the functional role of RUNX3 gene and its possible involvement in the process of carcinogenesis. However, RUNX3 protein has been shown to bind to Smad2 and Smad3 proteins and these data suggest the possible role of RUNX3 in transducing TGF-β signaling pathway.29 Despite the fact that several recent studies have described the causal relationship of loss of RUNX3 in development of various human cancers,7, 8, 9, 10 we feel that our present study is particularly important owing to the significance of TGFβ mediated signaling in the colon cancer, and in particular in MSI-H CRCs. Since RUNX3 is believed to act downstream of TGF-β, it would be reasonable to hypothesize that mutational inactivation rate of TGFβRII in the colon cancers may be inversely associated with silencing of RUNX3. Surprisingly, in the present study, the mutation rate of TGFβRII in the microsatellite unstable CRCs was not mutually exclusive of RUNX3 methylation, suggesting that RUNX3 may be a redundant component in the TGFβRII pathway and/or may regulate cellular growth by mechanisms not yet completely understood. However, inactivation of RUNX3 by promoter methylation in tumors that did not harbor frameshift mutations in TGFβRII suggests that inactivation of RUNX3 by promoter methylation may be responsible for the loss of TGF-β-mediated cell growth in some proportion of sporadic MSI-H CRCs that lack TGFβRII mutations.
Our results indicate that the RUNX3 promoter is methylated in colorectal cancers but not in the adjacent nonmalignant mucosal epithelium, suggesting that methylation is a cancer-specific event. Only 1 of 21 cases also showed methylation of the RUNX3 promoter in the nonmalignant colonic tissue, and this case was obtained from a patient with an MSI-H CRC. There are several possible explanations for the presence of methylated alleles in the nonmalignant colonic tissue specimens; one is that they represent premalignant changes. Methylation of tumor suppressor genes in normal tissues has been observed in lung, colon and breast,20 but we did not find this to be the case for RUNX3, as the frequency of such events was conspicuously absent from the normal tissue. Methylation of a variety of genes has been attributed to the aging process, and this may be a possible explanation for detecting methylation in normal colonic tissues. Larger studies of normal specimens should be conducted to address this issue.
In conclusion, our data indicates that inactivation of RUNX3 by promoter methylation constitutes an important epigenetic mechanism in sporadic colon cancer. Although future studies will reveal whether epigenetic inactivation of RUNX3 is a common occurrence in other human cancers, there is now strong evidence that RUNX3 is a candidate tumor suppressor in gastrointestinal carcinogenesis.
The research for CALGB 9865 was supported, in part, by grants from the National Cancer Institute (CA31946) to the Cancer and Leukemia Group B (R.L. Schilsky, Chairman). This work was also sponsored by a grant from the NIH (RO1-CA 72851) to C.R.B and in part by an American Cancer Society-IRG award to A.G. C.N.A. was funded by a grant of the Dr. Mildred-Scheel-Stiftung, Germany. The contents of this manuscript are solely the responsibility of the authors and do not necessarily represent the official views of the National Cancer Institute. The following institutions participated in the study: CALGB Statistical Office, Durham, NC, S. George, Ph.D., supported by CA33601; Dana Farber Cancer Institute, Boston, M.A., G.P. Canellos, M.D., supported by CA32291; Dartmouth Medical School-Norris Cotton Cancer Center, Lebanon, NH, M. Ernstoff M.D., supported by CA04326; Massachusetts General Hospital, Boston, M.A., M.L. Grossbard, M.D., supported by CA12449; Mount Sinai School of Medicine, New York, NY, L. Silverman, M.D., supported by CA04457; Rhode Island Hospital, Providence, RI, W. Sikov, M.D., supported by CA08025; Roswell Park Cancer Institute, Buffalo, NY, E. Levine, M.D., supported by CA02599; Southeast Cancer Control Consortium, Inc., CCOP, Goldsboro, NC, J.N. Atkins, M.D., supported by CA45808; SUNY Upstate Medical University, Syracuse, NY, S.L. Graziano, M.D., supported by CA21060; The Ohio State University, Columbus, OH, C.D. Bloomfield, M.D., supported by CA77658; University of California at San Diego, San Diego, CA, S. Seagren, M.D., supported by CA11789; University of California at San Francisco, San Francisco, CA, A. Venook, M.D., supported by CA60138; University of Chicago Medical Center, Chicago, IL, G. Fleming, M.D., supported by CA41287; University of Illinois at Chicago, Chicago, IL, D. Gustin, M.D., supported by CA74811; University of Iowa, Iowa City, IA, G. Clamon, M.D., supported by CA47642; University of Maryland Cancer Center, Baltimore, MD, D. Van Echo, M.D., supported by CA31983; University of Massachusetts Medical Center, Worcester, MA, M.E. Taplin, M.D., supported by CA37135; University of Minnesota, Minneapolis, MN, B.A. Peterson, M.D., supported by CA16450; University of Missouri/Ellis Fischel Cancer Center, Columbia, MO, M.C. Perry, M.D., supported by CA12046; University of North Carolina at Chapel Hill, Chapel Hill, NC, T.C. Shea, M.D., supported by CA47559; University of Tennessee Memphis, Memphis, TN, H.B. Niell, M.D., supported by CA47555; Wake Forest University School of Medicine, Winston-Salem, NC, D.D. Hurd, M.D., supported by CA03927; Walter Reed Army Medical Center, Washington, DC, J.C. Byrd, M.D., supported by CA26806.