Loss of Imprinting in Colorectal Cancer Linked to Hypomethylation of H19 and IGF2¹

Epigenetic alterations in human cancer, i.e., alterations in the genome other than the DNA sequence itself, were first described in 1983 by Feinberg and Vogelstein (1), who found widespread hypomethylation of genes in CRCs³ and in premalignant adenomas. Epigenetic abnormalities identified subsequently include global genomic hypomethylation (2), promoter hypermethylation of CpG islands (3, 4), and LOI (5, 6), or loss of the normal parent of origin-dependent gene silencing, affecting at least the genes IGF2, PEG1, p73, and LIT1 (5, 6, 7, 8, 9, 10, 11). LOI of IGF2 causes overexpression of IGF2 (12), an important autocrine growth factor in cancer. LOI was first identified in embryonal tumors in childhood, including Wilms’ tumor, in which it is the most common molecular alteration (5, 6), as well as rhabdomyosarcoma (13) and hepatoblastoma (14). LOI was also later found in common adult malignancies including ovarian (15), colon (16), lung (17), and bladder cancer (18), as well as chronic myelogenous leukemia (19). In CRC, LOI is particularly important because it is found commonly in both the tumor and normal tissue of patients with CRC, at ∼3-fold higher frequency then in patients without colon tumors (16), and, thus, LOI may represent the only common alteration linked to cancer that is found in normal tissue.

In Wilms’ tumors, approximately half of tumors appear to arise by an epigenetic mechanism involving LOI rather than genetic alterations involving, for example, WT1 mutations and LOH, and the tumors with LOI appear in children who develop cancer at a later age, accounting for the bimodal age distribution of Wilms’ tumor (12). LOI was linked to increased methylation, because Wilms’ tumors with LOI of IGF2, i.e., activation of the normally silent maternal allele, show aberrant methylation of the normally unmethylated maternal allele of a DMR upstream of the H19 gene on the same chromosome (20, 21). This result is consistent with the enhancer competition model for regulation of H19 imprinting. By this model, IGF2 and H19 promoters compete on the same chromosome for a shared enhancer, and access of the maternal IGF2 allele to this enhancer is blocked by the H19 DMR when unmethylated, likely because of the insulator activity of CTCF binding to the unmethylated H19 DMR (Refs. 22, 23, 24, 25, 26). Indeed, we observed that in Wilms’ tumor, methylation of the maternal H19 DMR includes CTCF-binding sites (27). These results would suggest that increased or ectopic activity of a DNA methyltransferase might lead to aberrant methylation of the maternal H19 DMR.

Therefore, we were surprised to observe that HCT116, a CRC line with normal imprinting of IGF2, is hypermethylated at H19 and retains normal imprinting after somatic cell knockout of the maintenance DNA methyltransferase DNMT1 but loses imprinting after subsequent somatic cell knockout of DNMT3B (28), a de novo methyltransferase, i.e., that is able to methylate unmethylated sequences and is necessary for normal imprinting (29, 30). This result implies that the loss of methylation, rather than the gain of methylation, causes LOI in CRC. To better determine whether LOI in CRC involves hypomethylation or hypermethylation, we performed genomic sequencing analysis. Our results differ from past studies, and they also suggest a model of IGF2 imprinting in at least the colon that differs from the conventional view of enhancer competition between IGF2 and H19.

H19 CTCF binding site 1 (CBS1) was analyzed as described earlier (27); CBS6 corresponds to GenBank nucleotides 7855–8192 (accession no. AF125183) and was analyzed after bisulfite treatment using primers 5′-GAGTTTGGGGGTTTTTGTATAGTAT-3′ and 5′-CTTAAATCCCAAACCATAACACTA-3′, followed by 5′-GTATATGGGTATTTTTTGGAGGT-3′ and 5′-CCATAACACTAAAACCCTCAA-3′, both annealing at 55°C. The IGF2 DMR sequence analyzed corresponds to GenBank nucleotides 631–859 (accession no. Y13633), and was analyzed after bisulfite treatment using primers 5′-GGGAATGTTTATTTATGTATGAAG-3′ and 5′-TAAAAACCTCCTCCACCTCC-3′, annealing at 55°C, followed by 5′-TAATTTATTTAGGGTGGTGTT-3′ and 5′-TCCAAACACCCCCACCTTAA-3′, annealing at 50°C. Other conditions are as described earlier (27).

In vitro functional analysis was performed using the 293/EBNA1 cell line as described (31) and the pcDNA3Myc vector containing full-length DNMT3B coding sequences, and p220.2 (32) as the assay plasmid. Cotransfected target DNA was digested with the methylation-sensitive restriction endonuclease HpaII, and Southern blot was performed using p220.2 as a probe. All of the transfections were done in duplicate or triplicate for each experiment.

Direct PCR sequencing of genomic DNA was performed to analyze the sequence of DNMT3B. All of the coding exons published including exon-intron junctions were thoroughly examined. LOI of IGF2 was assessed according to hot-stop PCR (33).

HCT116 cells show normal imprinting but undergo LOI of IGF2 after somatic cell knockout of both DNMT1 and DNMT3B (28), suggesting that the loss of methylation rather than the gain of methylation is responsible for LOI in these cells. To test this hypothesis, we examined directly the methylation of two DMRs that distinguish parental alleles in human cells: the H19 DMR 5-kb upstream of H19 and methylated on the paternal allele (H19 active, IGF2 silent); and the IGF2 DMR within intron 2 of IGF2 and methylated on the maternal allele (Ref. 34). Note that the DMRs in humans differ from the mouse, in which there are three rather than one DMR within IGF2 (35). Bisulfite sequencing analysis of HCT116 cells, and HCT116 cells lacking DNMT1, DNMT3B, or both, revealed that in the double-knockout cells, which showed LOI, both the H19 and IGF2 DMRs, were extensively hypomethylated (Fig. 1). This hypomethylation was found in three separate double-knockout lines with LOI and in none of single-knockout or wild-type lines with normal imprinting (Table 1).

To determine whether hypomethylation was also linked to LOI in primary colon cancers, we then analyzed 20 CRC informative for imprinting status of IGF2 (heterozygous for a transcribed polymorphism) by reverse transcription-PCR, 12 with LOI and 8 with normal imprinting. All 8 of the CRC with normal imprinting showed the normal half-methylation pattern at the IGF2 DMR, and all 12 of the CRC with LOI showed marked hypomethylation of the IGF2 DMR (P = 0.000007; Figs. 2 and 3). In tumors with normal imprinting, the fraction of CpG sites that were methylated was 43.6 ± 10.9%, whereas in tumors with LOI the fraction of sites methylated was 10.9 ± 9.4% (P < 0.0001). In addition, for each DMR, 15–20 clones were independently sequenced from the PCR product of each bisulfite-treated sample, and each experiment was repeated at least once. We also observed hypomethylation of the H19 DMR in CRC, although the differences were not absolute as in the case of the IGF2 DMR, but were in marked contrast to Wilms’ tumors with LOI (Table 2). These results also differ markedly from those of Nakagawa et al. (35), who reported hypermethylation of CBS6 in colorectal cancer. Finally, because LOI is found at increased frequency in both tumor and normal tissue of patients with CRC, we also examined the matched normal mucosa of 3 CRC patients whose tumors showed LOI. As we reported earlier (16), the matched normal mucosa also showed LOI of IGF2, although methylation had not been examined in that study. We found the same pattern of hypomethylation in the normal colonic mucosa in each patient as we found in tumors (Table 2), indicating that this epigenetic abnormality was not limited to the cancers.

Because LOI and hypomethylation were present in normal tissue, and DNMT3B appeared to play a role in LOI in HCT116 cells, we examined all 20 of the CRC for germ-line mutations in the DNMT3B gene. Six of 20 patients showed a single variation in the coding sequence leading to amino acid substitutions: G892T (G210W), G1390A (A376T), A1451G (Y396C), G2044A (V594I), G2086A (V608M), and T1436C (L391P). To distinguish between neutral and functional variants, we performed site-directed mutagenesis and transfection into 293/EBNA1 cells, together with an episomal vector, which was the target for de novo methylation. None of the variants disrupted DNMT3B methyltransferase activity (data not shown). Thus, these sequence variations represent neutral polymorphisms.

This study has two major results. First, we report that hypomethylation, rather than hypermethylation, is linked to LOI of IGF2 in human CRC based on two lines of evidence. In CRC lines in which hypomethylation is induced artificially by DNMT1/DNMT3B double knockout, LOI is found only in the hypomethylated lines. Indeed, unmodified HCT116 cells with hypermethylation of the H19 DMR exhibit normal imprinting, even though Wilms’ tumors with hypermethylation of the same sites show LOI (27). Furthermore, we find that in primary human CRC, as well, LOI is linked to hypomethylation rather than hypermethylation. The latter result is in contrast to the findings of Nakagawa et al. (35), who reported hypermethylation of the H19 in CRC with LOI of IGF2. It should be remembered that the first epigenetic alterations found in human cancer was hypomethylation of DNA (1) and that CRC show global hypomethylation even in the presence of specific sites of increased DNA methylation (2). Furthermore, the assumption that CpG islands are universally hypomethylated is incorrect, as imprinted genes show normal methylation, and we have also identified recently many normally methylated CpG islands in normal cells (36). Therefore, a more correct and inclusive view is that cancers show epigenetic instability, including global hypomethylation, and sites of both aberrantly increased and decreased methylation, that lead to altered gene regulation.

The second major result of this study is that normal imprinting in the colon and LOI in CRC is specifically linked to the methylation status of a DMR within IGF2 and not H19. Thus, all 8 of the cancers with normal imprinting showed normal half-methylation of the IGF2 DMR and all 11 of the cancers showed hypomethylation of this DMR, as well as 3 matched normal mucosal specimens that also showed LOI. Takai et al. (37) recently described partial or complete hypomethylation of the H19 ICR in two of four bladder cancers, but no relationship to H19 imprinting; IGF2 was not examined in that study. We did not find any alteration of H19 imprinting in the CRC examined here. We would argue that it is the IGF2 DMR, not the H19 DMR, that is important in maintaining imprinting in CRC. We had reported earlier that cancers with LOI also show LOI in the matched normal mucosa (16), so we would expect that this methylation abnormality is generally present in the colon of these cancer patients.

An important implication of this result is that it suggests a mechanism for regulation of IGF2 imprinting independent of enhancer competition. By the enhancer competition model, IGF2 and H19 promoters compete on the same chromosome for a shared enhancer, and access of the maternal IGF2 allele to this enhancer is blocked by the H19 DMR when unmethylated, likely because of the insulator activity of CTCF binding to the unmethylated H19 DMR (22, 23, 24, 25, 26). However, in CRC with LOI, the H19 DMR is hypomethylated on both alleles, and hypomethylation of the IGF2 DMR is specifically linked to LOI of IGF2 in both primary CRC and in HCT116 cells in which methyltransferases have been disrupted experimentally.

Some clues to function are available from mouse studies, although it is difficult to relate mouse experiments precisely to the human, as the DMR sequences themselves differ between species. Nevertheless, the region corresponding to the human DMR studied here is in same physical relationship to human IGF2 exons 2 and 3, as is mouse “DMR0” to mouse Igf2 pseudoexons 1 and 2 (34). To date, no mouse knockout of DMR0 by itself has been reported, although deletion of DMR1, or of DMR0 and DMR1 together, lead to activation of the normally silent maternal allele of Igf2 (38, 39). The mouse knockout experiments suggest the existence of a transcriptional repressor within Igf2 (38, 39). We would agree with that hypothesis and additionally state that our results suggest that methylation of this human IGF2 DMR recruits transcriptional repressors to the maternal allele. By this model, hypomethylation would lead to LOI by loss of association of these repressors to the IGF2 DMR. Our results also suggest two potentially valuable lines of experimentation: knockout of DMR0 in mouse and biochemical studies aimed at identifying factors of which the binding to the human IGF2 DMR is lost in tumors with LOI.

We thank Ina Rhee and Bert Vogelstein for the HCT116 knockout cell lines, and Mindy Graber for preparing the manuscript.