Opposing effects of DNA hypomethylation on intestinal and liver carcinogenesis

Results

Apc^Min/+ Mice with DNA Hypomethylation. To determine the DNA methylation status in colonic tumors and normal mucosa in Apc ^Min/+ mice, we cleaved DNA with the methylation-sensitive enzyme HpaII and analyzed digests by Southern blotting using a centromeric satellite repeat probe (17). Consistent with the observation of genomic hypomethylation in human colon carcinoma (2), we observed hypomethylation of the pericentromeric region in colon tumors from Apc ^Min/+; Dnmt1 ^+/+ mice as compared to wild-type mucosa (Fig. 8, which is published as supporting information on the PNAS web site). We generated isogenic Apc ^Min/+ mice that express three different levels of the maintenance DNA methyltransferase Dnmt1 by using combinations of previously described hypomorphic and null mutations (Dnmt1 ^+/+, Dnmt1 ^chip/+, Dnmt1 ^chip/c) (12) in an attempt to model the degree of hypomethylation observed in tumors. The intestinal mucosa of Dnmt1 ^chip/c mice was significantly hypomethylated at pericentromeric regions (Fig. 8) but had a normal morphology, cell proliferation, and apoptosis levels as assessed by histology, Ki-67 expression, and TUNEL staining, respectively (data not shown). Because DNA hypomethylation is observed in early stage human colorectal adenomas and was proposed to be a result of selection within these tumors (3, 18), we developed an assay to test the effects of DNA hypomethylation on tumor initiation in Apc ^Min/+ mice.

β-Catenin Accumulation as an Early Event in the Colon Tumorigenesis of Apc^Min/+ Mice. Two stages of tumorigenesis can be distinguished in the colonic mucosa of Apc ^Min/+ mice: microadenomas and macroscopic tumors (15). Microadenomas are small intramucosal lesions that can be identified by histological analysis. Most microadenomas consist of a single crypt that displays an aberrant hematoxylin-eosin (HE) staining pattern. We asked whether these lesions might be evidence of early Wnt pathway deregulation by immunostaining with β-catenin. The analysis showed that all histologically visible microadenomas identified by HE staining exhibited increased accumulation of β-catenin relative to normal adjacent crypts (n = 54 of 54; Fig. 1), which persisted in tumors (n = 33 of 33). Therefore, we used β-catenin immunostaining of en face sections to detect colonic microadenomas to assess the number and size of microadenomas and tumors in the same animals of differing Dnmt1 genotypes.

DNA Hypomethylation Suppresses Intestinal Tumorigenesis but Promotes Microadenoma Formation in the Colon. Cohorts of Apc ^Min/+ mice with Dnmt1 ^+/+, Dnmt1 ^chip/+, or Dnmt1 ^chip/c genotypes were aged to 180 days and analyzed for microadenomas in the colon. The number of these lesions as identified by β-catenin immunostaining was significantly increased in the hypomethylated Dnmt1 ^chip/c; Apc ^Min/+ mice when compared with Dnmt1 ^chip/+ ; Apc ^Min/+ or Dnmt1 ^+/+; Apc ^Min/+ mice (Fig. 2A). No microadenomas were observed in Dnmt1 ^chip/c; Apc ^+/+ control mice. The calculated number of colon microadenomas per mouse was 21 in Dnmt1 ^+/+, 20 in Dnmt1 ^chip/+, and 38 in Dnmt1 ^chip/c mice, reflecting a nearly 2-fold increase in colonic microadenomas in hypomethylated Dnmt1 ^chip/c mice when compared to the other experimental groups.

Analysis of tumor incidence in the colon and small intestine from the same animals revealed a block in the formation of macroscopic tumors, consistent with previous findings in Apc ^Min/+ mice (8-10). The total number of intestinal tumors in the hypomethylated Dnmt1 ^chip/c mice was significantly lower when compared to Dnmt1 ^chip/+ or Dnmt1 ^+/+ mice (Fig. 2C;1.9 ± 2.4 in Dnmt1 ^chip/c mice versus 8.0 ± 4.8 in Dnmt1 ^chip/+ mice and 9.6 ± 6.7 in Dnmt1 ^+/+ mice). In particular, no tumors were detectable in the colon of Dnmt1 ^chip/c mice (Fig. 2B). Several intestinal tumors were subjected to histological analysis and all were confirmed as adenomas. There was no difference in tumor histology between the different genotypes.

Apc LOH in Microadenomas from DNA Hypomethylated Mice. It appeared possible that DNA hypomethylation in Dnmt1 ^chip/c; Apc^Min/ mice increased microadenoma formation because of an elevated incidence of Apc LOH, resulting in stablilization and accumulation of β-catenin (Fig. 1; refs. 13 and 14). To determine whether Apc LOH was involved in the development of hypomethylation-induced microadenomas, we performed Apc LOH analysis (16) on microadenomas isolated by using laser capture microdissection. All dissected microadenomas from Dnmt1 ^chip/c mice showed a predominant signal for the Apc ^Min allele (n = 12), indicating loss of the wild-type allele (Fig. 3). Evidence for Apc LOH was further demonstrated in macroscopic tumors in Apc ^Min/+ mice of all Dnmt1 genotypes, in agreement with observations by others (8, 16, 19). These data are consistent with a model that elevated incidence of chromosomal instability in Dnmt1 ^chip/c mice may cause a higher rate of Apc LOH, because previous studies have demonstrated increased genetic instability and elevated rates of mitotic recombination caused by global DNA hypomethylation (11-13). To directly examine rates of Apc LOH in the Apc ^Min/+, Dnmt1 ^chip/c mice, we aged a cohort of animals to 360 days for determining colonic microadenoma and tumor incidence. Importantly, the number of microadenomas increased by 50% between 180 and 360 days in Dnmt1 ^chip/c mice, indicating ongoing initiating events (P < 0.04 by Mann-Whitney U test) (Fig. 4B). Because a majority of Apc ^Min/+, Dnmt1 ^+/+, or Dnmt1 ^chip/+ mice die after 180 days as a result of macroscopic intestinal tumors, we were not able to determine the rate of microadenoma formation in the control mice. Although the incidence of colonic microadenomas increased between 180 and 360 days in Apc ^Min/+, Dnmt1 ^chip/c mice, the size of these lesions did not increase over time (Fig. 4B). Moreover, no colonic tumors developed in the 360-day-old Apc ^Min/+, Dnmt1 ^chip/c mice (n = 6), suggesting that, despite increased initiation events, the progression of microadenomas to macroscopic polyps was inhibited.

DNA Hypomethylation Suppresses Growth of Intestinal Lesions. Similarly, telomere dysfunction, another cause of chromosomal instability, results in an increased microadenoma formation and decreased incidence of macroscopic tumors in Apc ^Min/+ intestinal mucosa (20). Therefore, we examined intestinal tumors for signs of telomere dysfunction by a telomere fluorescence in situ hybridization assay to assess whether DNA hypomethylation caused chromosomal instability through telomere length regulation. No shortening of telomeres was detected in normal tissues (data not shown) or intestinal tumors of Dnmt1 ^chip/c mice when compared to Dnmt1 ^+/+ mice (Fig. 9 and Supporting Text, which are published as supporting information on the PNAS web site).

To further investigate potential mechanisms that might account for the suppression of intestinal adenomas in the hypomethylated animals, we examined the size and proliferative and apoptotic index of the tumors. When we compared the size of microadenomas in each experimental group, the microadenomas in Dnmt1 ^chip/c mice were never larger than 125 μm and were significantly smaller than in Dnmt1 ^chip/+ or Dnmt1 ^+/+ mice (Fig. 4A, P < 0.003 and P < 0.003 by Mann-Whitney U test, n = 182). Consistent with this observation, we found that macroscopic adenomas in the small intestine of Dnmt1 ^chip/c mice were smaller than adenomas in the other experimental groups (Fig. 4A). These findings suggest that, even though DNA hypomethylation increases the formation of microadenomas, it appears to suppress the subsequent growth of the lesions and their progression to tumors as was previously observed (8-10).

To compare cell proliferation kinetics in intestinal tumors with different levels of DNA methylation, we performed TUNEL and BrdUrd staining on tumors from each cohort. TUNEL staining revealed no evidence for elevated apoptosis in Dnmt1 ^chip/c intestines or tumors as compared to control intestines and tumors (Fig. 5A). We observed a correlation between BrdUrd incorporation and tumor size (P < 0.0001 by Spearman's rank correlation) in both control and Dnmt1 ^chip/c; Apc ^Min/+ tumors (Fig. 5B). However, there was no significant difference between control and similarly sized Dnmt1 ^chip/c tumors. Because the Dnmt1 ^chip/c; Apc ^Min/+ mice develop few large tumors with elevated BrdUrd incorporation, reduced Dnmt1 levels likely result in reduced cellular proliferation, (21) leading to smaller lesions; however, in large tumors, secondary events may be responsible for increased growth (22). In agreement with this notion, we observed elevated expression of Dnmt1 in ≈80% (n = 10 of 12) of intestinal tumors in Apc ^Min/+, Dnmt1 ^+/+ mice (Fig. 10, which is published as supporting information on the PNAS web site), which likely reflects increased cell division in tumors (23).

Overexpression of Dnmt1 in ES cells can lead to biallelic expression of Igf2 (24), a known mitogen associated with human colon cancer (25) and mouse intestinal carcinogenesis (26). Therefore, we asked whether the smaller lesions in Dnmt1 ^chip/c mice might result from reduced Igf2 expression. Northern blot analysis revealed a consistent but modest 20% reduction in Igf2 expression in the intestine of hypomethylated mice (Supporting Text and Fig. 11, which is published as supporting information on the PNAS web site). However, DNA methylation analysis using bisulfite sequencing did not demonstrate any consistent change in the methylation status of the imprinted differentially methylated region (DMR) adjacent to the H19 gene that determines Igf2 expression (Fig. 11) (27). Whether the modest reduction in Igf2 levels is functionally important in inhibiting tumor progression in this model remains to be tested.

Hepatocellular Carcinoma Development in DNA Hypomethylated Mice. In this study, we did not observe thymic lymphomas, which were reported in a previous study for Dnmt1 ^chip/c mice older than 5 months of age (12), likely because of differences in genetic background of the mice used in the two studies. However, we unexpectedly observed multiple liver tumors in Apc ^Min/+; Dnmt1 ^chip/c mice at 360 days (6.2 ± 2.9 tumors per mouse, n = 6) (Fig. 6A), whereas no liver tumors were detectable in age-matched Apc ^Min/+ (n = 7) or Dnmt1 ^chip/c mice (n = 4) with an identical genetic background. Histological analysis revealed that the tumors were hepatocellular adenomas/carcinomas (Fig. 6 B and C), consistent with recent reports that Wnt pathway activation was involved in liver cancer (28). To assess the activation of the Wnt pathway in Apc ^Min/+; Dnmt1 ^chip/c liver tumors, we first examined whether Apc LOH was detectable (Fig. 6D). More than half of the lesions displayed evidence of LOH (n = 6 of 10) as compared to control normal liver from the same animals. We additionally performed β-catenin immunostaining and found cytoplasmic and/or nuclear accumulations of β-catenin in most tumors (four positive of five analyzed) (Fig. 7 A-C). Consistent with the notion that these tumors rely on activation of the Wnt pathway, glutamine synthetase (GS) was expressed in 23 of 24 tumors analyzed (Fig. 7D). GS is a target of TCF-β-catenin pathway in the liver, and has been used as a marker for hepatocellular carcinoma with activated Wnt pathway (29). No evidence for aberrant GS expression or tumor lesions was observed in any of the age-matched controls (Fig. 7 E and F). These results indicate that DNA hypomethylation promotes liver tumorigenesis in Apc ^Min/+ mice by activation of the Wnt pathway, which in some cases is mediated by loss of the wild-type Apc allele.

Discussion

Here we have shown that DNA hypomethylation promotes early stage tumor formation in the colon and liver tumorigenesis but strongly suppresses overall tumorigenesis in the intestine of Apc ^Min/+ mice. Our results indicate dual effects of DNA hypomethylation on intestinal carcinogenesis, and suggest that DNA hypomethylation can promote early events in tumorigenesis while blocking progression. Two models have been described for the effect of DNA hypomethylation on tumor promotion, including transcriptional derepression of oncogenes and endogenous retroviral elements, or increased genomic instability mediated by hypomethylation of heterochromatic or interspersed repeats (30). Our data indicate that DNA hypomethylation increased the incidence of tumor initiation in both the colon and liver. The increased tumor incidence correlated with an elevated loss of the wild-type Apc allele (Figs. 3 and 6D), which is in agreement with previous studies that demonstrated DNA hypomethylation-dependent LOH events (11-13). Whether DNA hypomethylation also contributes to tumorigenesis via transcriptional derepression is not ruled out, but has not been directly tested in our study.

Although the number of microadenomas in the colon of Apc ^Min/+; Dnmt1 ^chip/c mice was increased, the size of the lesions was significantly smaller than those in Apc ^Min/+; Dnmt1 ^chip/+ or Apc ^Min/+; Dnmt1 ^+/+ mice (Fig. 4A). Macroscopic tumors in Apc ^Min/+ mice were recently shown to originate in some cases from two or more crypts, suggesting a non-cell autonomous mechanism for promoting tumor formation (31). Whether the rate of crypt fusion is suppressed by DNA hypomethylation, and accounts for the elevated numbers of smaller microadenomas, is difficult to rule out. However, because increased LOH events and subsequent Wnt pathway activation were observed in both the colon and liver, we do not think a difference in crypt fusion best explains these results.

Suppression of tumor formation in Apc ^Min/+ mice by DNA hypomethylation has been observed by several groups (8-10), yet the molecular basis for this block to tumor formation is not understood. The prevailing notion is that DNA hypomethylation blocks the epigenetic silencing of candidate tumor suppressor loci; however, identification of the critical target genes is lacking. We tested three nonexclusive models to explain the tumor suppression phenotype in the intestine. First, we tested whether DNA hypomethylation may exert an effect on telomere shortening, because mice with telomere dysfunction similarly increase microadenomas and block progression marked by elevated apoptosis (20). However, no affect on telomere length or apoptosis was observed in the hypomethylated mice (Figs. 5 A and 9). Second, we asked whether the Dnmt1 hypomorphic mice showed reduced proliferation in microadenomas and tumors. Consistent with analysis of Apc ^Min/+; Dnmt1 heterozygous mice (9), Apc ^Min/+; Dnmt1 ^chip/c mice showed fewer and smaller tumors than control animals. Dnmt1 expression in Apc ^Min/+; Dnmt1 ^+/+ tumors was elevated, perhaps because of increased proliferation in tumors, suggesting that reduced levels of Dnmt1 in Dnmt1 ^chip/c mice limit cell proliferation accounting for their fewer smaller tumors. However, BrdUrd incorporation correlated with tumor size rather than Dnmt1 genotype, suggestive of secondary events that in rare cases allow larger tumors to form in Dnmt1 hypomorphic animals. Finally, we asked whether down-regulation of Igf2 in Dnmt1 ^chip/c mice provided a molecular basis for reduced tumor incidence. Although a subtle effect on Igf2 mRNA levels was seen, no clear effect of DNA hypomethylation of the imprinted regulatory sequences was observed.

The observation that Apc ^Min/+ mice with DNA hypomethylation developed multiple liver tumors in contrast to the strong inhibition of intestinal tumorigenesis clearly indicates opposing effects of DNA hypomethylation in different cell types. Reduced DNA methylation is frequently observed in human hepatocellular carcinomas (32). In addition, diets deficient in methyl group donors have been consistently observed to induce DNA hypomethylation and lead to liver cancers in rodents (33). Our findings provide a functional link between DNA hypomethylation and liver tumorigenesis. It has been shown that conditional inactivation of Apc leads to liver tumorigenesis in mice (28), suggesting that the Apc LOH observed in Dnmt1 ^chip/c ; Apc ^Min/+ mice is the cause of liver tumorigenesis in this model. Because Apc LOH was not observed in all liver tumors, it is possible that promotion of liver tumors also occurs through other mechanisms. However, as Apc is a large gene, it is possible that our assay is insensitive to LOH events caused by intrachromosomal deletions at a distance from the Apc ^Min point mutation.

Our findings further indicate that DNA hypomethylation has opposing effects on early and late stage events in intestinal neoplasia through promotion of genetic instability and inhibition of growth, respectively. Two known risk factors for human colon cancer are dietary insufficiency for folates and common polymorphic mutations of the MTHFR gene, both of which can lead to decreased genomic methylation levels (34). Although the significance of DNA methylation in epigenetic silencing in human tumors is well established, whether DNA hypomethylation drives chromosomal instability in human cancers remains an interesting and open question. Further analyses of genetic alterations, together with systematic tests for gene expression affected by DNA hypomethylation, will be required to describe the molecular mechanisms that underlie the opposing effects of DNA hypomethylation in different tumor stages and different cell types.