Induction of Tumors in Mice by Genomic Hypomethylation

See allHide authors and affiliations

Science  18 Apr 2003:
Vol. 300, Issue 5618, pp. 489-492
DOI: 10.1126/science.1083558


Genome-wide DNA hypomethylation occurs in many human cancers, but whether this epigenetic change is a cause or consequence of tumorigenesis has been unclear. To explore this phenomenon, we generated mice carrying a hypomorphic DNA methyltransferase 1 (Dnmt1) allele, which reduces Dnmt1 expression to 10% of wild-type levels and results in substantial genome-wide hypomethylation in all tissues. The mutant mice were runted at birth, and at 4 to 8 months of age they developed aggressive T cell lymphomas that displayed a high frequency of chromosome 15 trisomy. These results indicate that DNA hypomethylation plays a causal role in tumor formation, possibly by promoting chromosomal instability.

Human cancer cells often display abnormal patterns of DNA methylation. The role of aberrant hypermethylation in the silencing of tumor suppressor genes is now well documented (1). In contrast, the role of aberrant hypomethylation—which is observed in a wide variety of tumors (25), often together with regional hypermethylation—has remained unclear.

To investigate whether DNA hypomethylation has a causal role in tumor formation, we generated mice with highly reduced levels of Dnmt1, the enzyme that maintains DNA methylation patterns in somatic cells (6). Because mice homozygous for a Dnmt1 null allele (Dnmt1c/c) die during gestation (7, 8), we combined a hypomorphic allele [Dnmt1chip (9)] with a null allele to generate Dnmt1chip/c (referred to here as Dnmt1chip/–) compound heterozygotes with a substantially reduced level of genome-wide DNA methylation. Dnmt1chip/– embryonic stem (ES) cells expressed 10% of wild-type levels (Fig. 1A). To test whether the reduced Dnmt1 expression affected DNA methylation in vivo, we generated mice carrying the different Dnmt1 alleles and determined their global methylation levels with the use of a probe for endogenous retroviral A type particles (IAPs) (Fig. 1, B and D) (10) and centromeric repeats (Fig. 1C). Southern blot analysis of embryonic fibroblasts and adult tissues showed that the DNA from compound heterozygotes was hypomethylated relative to the DNA from Dnmt1chip/chip or Dnmt1+/+ mice, although substantially less so than the DNA from Dnmt1–/– null ES cells. Mice carrying the different Dnmt1 alleles were obtained at the expected Mendelian ratios, indicating that reduction of Dnmt1 expression to 10% was compatible with viability. However, compound heterozygotes (Dnmt1chip/–) were runted and their weight at birth was only 70% that of Dnmt1+/– mice, in contrast to mice homozygous for the hypomorphic allele (Dnmt1chip/chip), which were normal in size (Fig. 2A). Dnmt1chip/– mice, although remaining substantially underweight, were fertile and generated litters of nonrunted pups when bred with wild-type mice.

Fig. 1.

Genomic hypomethylation in Dnmt1 hypomorphic mice. (A) Immunoblot analysis of protein extracts from ES cells, using a C-terminal Dnmt1 chicken antibody (34). The level of Dnmt1 in Dnmt1chip/– ES cells was markedly reduced, demonstrating the hypomorphic nature of this allele. The Dnmt1–/– negative control extracts did not show any bands, as expected [c allele (7)]. A Promega antibody to chicken immunoglobulin Y (IgY), with horseradish peroxidase as secondary antibody, was used; detection was performed with the Amersham Pharmacia Biotech ECL kit. The Coomassie gel shows total protein levels in each sample. (B) Southern blot of methylation status of IAPs in Dnmt1chip/chip and Dnmt1chip/– mice. Genomic DNA was digested with the methylation-sensitive restriction enzyme Hpa II and probed with an IAP cDNA probe (10). Hypomethylation was detected in thymus from 6-week-old Dnmt1chip/– mice and in Dnmt1chip/– embryonic day (E) 14.0 fibroblasts but not in Dnmt1chip/chip homozygotes or wild-type controls, as evidenced by the presence of lower molecular weight DNA fragments. (C) Southern blot analysis of centromeric repeat methylation of E14.0 fibroblasts from mice containing various Dnmt1 alleles. Hypomethylation is evident in the Dnmt1chip/– lanes. (D) Southern blot analysis of IAP methylation status in tissues from a 6-week-old Dnmt1chip/– mouse and in lymphomas. Significant hypomethylation is evident in all Dnmt1chip/– tissues. Controls for methylation (Dnmt1+/+ ES cells) and hypomethylation (Dnmt1–/– ES cells) are also shown (ES, ES cells; W, Dnmt1+/+;M, Dnmt1chip/–).

Fig. 2.

Dnmt1 hypomorphs are runted and develop T cell lymphomas. (A) Average weight of Dnmt1chip/+ (n = 20) and Dnmt1chip/– (n = 20) male littermates at birth (inset). Dnmt1chip/– mice were 66% as large as Dnmt1chip/+ mice. Females showed the same runt phenotype. The error bar indicates ±1 SD, p < 0.0001 (Student t test, StatView 5.0.1 software). Also shown are growth curves of Dnmt1chip/chip, Dnmt1chip/–, and wild-type male mice. Six to 10 mice of each genotype were used for each data point. (B) Cumulative survival of Dnmt1chip/– mice. Most Dnmt1chip/– mice became terminally ill between 4and 8 months of age. Control mice were Dnmt1chip/chip (n = 18); experimental mice were Dnmt1chip/– (n = 33). Mice were autopsied when visibly ill. At autopsy, 23 of 33 Dnmt1chip/– mice had developed tumors (21 lymphomas and 2 fibrosarcomas). Autopsy of Dnmt1chip/chip mice at 6 months (n = 12) and 12 months (n = 6) showed no evidence of tumor formation. (C) Dβ1-to-Jβ1 rearrangement at the TCRβ locus was analyzed by the polymerase chain reaction as described (35), using primers 1 and 4therein. The asterisk denotes the germline configuration [2171 base pairs (bp)]. When rearranged, five different amplified fragments are possible, ranging from 1561 to 381 bp (see wild-type thymus, lanes 1 to 3). Controls also include wild-type tail DNA (lane 14) and thymus DNA from a recombination-deficient RAG1–/– mouse. (D) FACS analysis of wild-type thymus (left) or Dnmt1chip/– tumors (middle and right) stained for CD4and CD8 receptors, T cell–specific markers. Tumors analyzed (n = 16) contained either double-positive CD4high/CD8high cells (9/16, middle panel) or CD4low/CD8high cells (7/16, right panel).

In addition to the runted phenotype, 80% of Dnmt1chip/– mice developed aggressive thymic tumors at 4 to 8 months of age. Cumulative survival of the Dnmt1chip/– mice is shown in Fig. 2B. Histological analysis classified the tumors as T cell lymphomas (11), and fluorescence-activated cell sorting (FACS) analysis revealed that most tumors were CD4/CD8+ or CD4+/CD8+ (Fig. 2D). When tested for D-to-J rearrangements in the T cell receptor β locus, four of 10 tumors showed a predominant Dβ1-to-Jβ1 rearranged band (Fig. 2C, lanes 5, 9, 11, and 13) consistent with monoclonality. Tumors without Dβ1-to-Jβ1 recombination may have rearranged other D and J elements. Monoclonality suggests that hypomethylation induces cancer in a precursor cell, with subsequent events leading to malignant tumor formation. Consistent with frequent activation of the c-myc oncogene in mouse and human lymphoma (12), we found that c-myc was overexpressed in almost all hypomethylated tumors (15/18 Dnmt1chip/–, Fig. 3C).

Fig. 3.

Expression and chromosomal analysis of hypomethylated tumors. (A) RNA slot blot of Dnmt1chip/– lymphomas (lanes b, c, and d) probed with MMLV cDNA. Also shown are a positive control lymphoma from a Mov-1 mouse [slot a1 (15)] and negative control thymuses from wild-type 129/Sv (slot a2) and a wild-type littermate of a tumor-bearing mouse (slot a3). (B) Northern blot of IAPs in Dnmt1chip/– tumors. IAPs can be detected in most tumors, whereas wild-type thymus does not show IAP expression. Positive control (lanes 10 to 12, 1:3 serial dilutions) are Dnmt1–/– hypomethylated fibroblasts that have been shown to activate IAP expression (16). Comparison of IAP and glyceraldehyde-3-phosphate dehydrogenase (GAPDH) levels shows that most clones express much less IAPs than the positive control. (C) Expression levels of c-myc were assessed by Northern blot (top two panels) and by immunoblot (bottom two panels). Lanes 2 to 7 are tumors that showed chromosome 15 trisomy; lanes 8 and 9 are tumors that are diploid for chromosome 15. Probes used were c-myc exon 2 for the Northern analysis and a rabbit polyclonal IgG antibody to c-myc for immunoblots (Santa Cruz Biotechnology). (D) Array comparative genome hybridization (CGH) analyses of three Dnmt1chip/– tumors, showing clear single-copy, whole-chromosome gain of chromosome 15 (x, y, and z), whole-chromosome gains of 14and loss on distal 12 (x), and gains of chromosome 14and proximal 9 (y). The X gain (x) reflects a sex difference between tumor and control. Array CGH was performed as in (26). Fluorescence ratios (average of quadruplicate measurements) for each bacterial artificial chromosome are plotted as a function of genome location based on the February 2002 freeze of the assembled mouse genome sequence ( Vertical lines delimit chromosome boundaries.

Genomic hypomethylation may contribute to lymphomagenesis by an epigenetic or a genetic mechanism. We considered three possible mechanisms.

  1. Hypomethylation may induce endogenous retroviral elements, leading in turn to insertional activation of proto-oncogenes (13). To test this idea, we hybridized RNA from randomly selected tumors with a Moloney murine leukemia virus (MMLV) cDNA probe and an IAP probe to detect endogenous retroviral and IAP expression, respectively. Of nine Dnmt1chip/– tumors, none showed C-type retroviral activation (Fig. 3A) (14) and only one of eight tumors showed a moderate increase in IAP expression (Fig. 3B, lane 7). In contrast, strong C-type retroviral expression was seen in a MMLV-induced lymphoma [Fig. 3A, slot a1 (15)] and IAP expression was highly activated in Dnmt1–/– fibroblasts [Fig. 3B, lanes 10 to 12 (16)]. Because c-myc is a frequent target for insertional activation by retroviral elements (17), we searched for inserted proviral elements in hypomethylated and MMLV-induced tumors. In 3 of 12 MMLV-induced tumors, an insertional rearrangement was seen in the vicinity of the c-myc locus, in agreement with previous observations (17). In contrast, no rearrangements were detected in hypomethylated tumors [0/18 (11)]. We conclude that the extent of hypomethylation in Dnmt1chip/– mice does not effectively activate endogenous retroviral elements and that virus insertions may not be a prevalent mechanism in hypomethylation-induced lymphoma.

  2. Hypomethylation may activate protooncogenes through epigenetic effects (18, 19). Indeed, c-myc was overexpressed in most hypomethylated tumors (Fig. 3C). However, it is unlikely that activation of c-myc is a direct consequence of promoter demethylation because the gene is expressed at normal levels in thymuses from 2- and 4-week-old mice that show a level of hypomethylation identical to that of the tumors [Fig. 1D (11)]. In addition, c-myc was not activated in Dnmt–/– fibroblasts that are almost completely demethylated (16). Finally, if oncogene activation by hypomethylation stimulated T cell proliferation as a first step in transformation, one would expect the lymphomas to be polyclonal rather than monoclonal (Fig. 2C).

  3. Hypomethylation may induce genomic instability. In fact, a significantly increased frequency of chromosomal rearrangements such as loss of heterozygosity (LOH) was observed in Dnmt1 mutant ES cells, suggesting that normal levels of methylation are important for genomic stability (20). Defects in DNA methylation have been linked to genome instability in studies of colorectal tumor cell lines (21), mouse tumor models (22, 23), and patients with immunodeficiency–centromeric instability–facial anomalies (ICF) syndrome (24, 25).

To test whether DNA hypomethylation increases genomic instability in Dnmt1chip/– tumors, we performed array-based comparative genome hybridization [array CGH (26)] using thymic tumor genomic DNA prepared from Dnmt1chip/– and Mov-1 (15) and Mov-14 (27) MMLV transgenic mice (Fig. 3D). There was a statistically significant difference in chromosome gains between these tumor classes (Table 1). Ten of 12 hypomethylated tumors exhibited a gain of chromosome 15, whereas only 2 of 12 MMLV-induced tumors showed this change (P = 0.004). Relative to MMLV-induced tumors, hypomethylated tumors also displayed a gain of chromosome 14 (4/12 versus 0/12, P = 0.05) and a higher degree of duplicated and deleted chromosome regions (Table 1) (Fig. 3D).

Table 1.

Gains or losses of chromosomes in Dnmt1chip/- and MMLV-induced tumors. The numbers indicate the number of times a particular event occurred in the Dnmt1chip/- or Moloney tumors. These events were not mutually exclusive; many tumors exhibited multiple chromosomal events.

Chromosomal changes Dnmt1chip/- tumors (n = 12) MMLV-induced tumors (n = 12)
Chr 15 gain 10 2
Chr 14 gain 4 0
Chr 10 gain 0 1
Partial Chr 9 gain 2 0
Partial Chr 4 gain 1 0
Partial Chr 16 loss 1 0
Partial Chr 12 loss 1 0

Together with the centromeric hypomethylation we observed (Fig. 1C), these results suggest a causal link between DNA hypomethylation and chromosomal instability as one mechanism leading to tumorigenesis. The increased fluorescence ratios observed for chromosomes 14 and 15 are consistent with singlecopy whole-chromosome gains throughout the tumor (Fig. 3D), which suggests that they are early events in the development of these monoclonal T cell lymphomas. Chromosome 15 is frequently duplicated in mouse T cell tumors (28, 29) and contains the oncogene c-myc, which when overexpressed causes T cell lymphomas (17). The fact that c-myc is overexpressed (RNA and protein) in most hypomethylated tumors (Fig. 3C) is consistent with a mechanism in which a gain of chromosome 15 contributes, at least in part, to the elevated expression of c-myc. Moreover, c-myc expression was lower in the two tumors that did not show trisomy 15 than in the other tumors (Fig. 3C).

Our results show that genomic hypomethylation causes tumorigenesis in mice and is associated with the acquisition of additional genomic changes. Consistent with this, genomic hypomethylation was found to promote tumorigenesis in a different mouse tumor model and to increase the rate of LOH in cultured fibroblasts (23). However, it remains possible that DNA hypomethylation contributes to tumorigenesis through other mechanisms unrelated to chromosomal instability. The phenotype of hypomethylated mice is also consistent with that of Suv39h histone methyltransferase mutant mice; hence, DNA and histone methylation, pericentric chromatin structure, and the maintenance of chromosomal stability may be linked (30).

DNA methyltransferase inhibitors such as 5-aza-2′-deoxycytidine have been used successfully to treat cancer in humans (19, 31) and mice (32, 33). The efficacy of these drugs is presumably due to their ability to reverse the epigenetic silencing of tumor suppressor genes. In light of our results, however, this therapeutic strategy should perhaps be considered a double-edged sword: Genomic demethylation may protect against some cancers such as intestinal tumors in the ApcMin mouse model (32) but may promote genomic instability and LOH (20, 23) and increase the risk of cancer in other tissues, as seen in hypomethylated mutant mice.

References and Notes

Stay Connected to Science

Navigate This Article