THE de novo METHYLATION OF INTEGRATED FOREIGN DNA
Choice of experimental systems. There are many excellent examples to document that the study of viral systems has led to the discovery of fundamental mechanisms in prokaryotic and eukaryotic molecular genetics. Since in most instances, virus replication has to rely on the utilization of cellular mechanisms, it cannot be surprising that viruses have been efficiently exploited as “Trojan horses” inside the cellular milieus. In the 1970s, our laboratory was involved in detailed analyses on the mode of adenoviral DNA integration into the host genome in adenovirus-transformed cells and in adenovirus type 12 (Ad12)-induced hamster tumor cells. In the course of these studies, it became feasible to prove that integrated foreign (adenoviral) DNA was de novo methylated [14]. When we subsequently were able to document the first inverse relationship between genetic activity of integrated viral genes and the extent of their methylation [15], it was obvious that this experimental system could be applied to fundamental studies on the regulation of genetic activity and on the biological function of DNA methylation [2].
A second seminal observation, which emphasized the biological importance of DNA methylation, came from detailed investigations on patterns of DNA methylation in the 5´-upstream regions of two randomly chosen human genes, TNFalpha and TNFbeta. The genomic sequencing method originally developed by Church and Gilbert [16], though difficult to use at the time, helped considerably in these studies. In the promoter and 5´-upstream regions of the TNFalpha and TNFbeta genes, we found cell type-specific and interindividually highly conserved patterns in the distribution of m5C residues, which agreed to the nucleotide site among individuals from different ethnic origins [17, 18].
Similar, though less precise, evidence came from large, randomly chosen segments of the human genome [19], many of them repetitive DNA sequences. These results implied that highly specific patterns of DNA methylation existed and most likely had to have a fundamentally important function. It was not easy in those days to convince others that a systematic endeavor to determine patterns of DNA methylation was not just a descriptive exercise but had to be initiated to learn about the wider gamut of possibilities with functional implications. Hopefully, the human epigenome project will help provide more ample evidence than the study of a single, pioneering, laboratory could possibly have adduced with limited means 15 years ago.
At this time, we also sought the collaboration of clinical researchers in order to extend the basic concepts derived from simpler experimental systems to more complex biomedical problems. In the course of these studies, it became even more obvious that the model developed with the adenovirus system could reliably guide all our efforts. In collaboration with several groups, we determined patterns of DNA methylation in the Prader-Willi/Angelman regions of the human genome [20, 21], in the promoter regions of the RET protooncogene [22], of the FMR1 gene [23, 24], and of several genes of the erythrocyte membrane [25, 26].
The state of methylation in DNA viral genomes. Many DNA virion genomes are unmethylated, others are methylated. Modulation as a bidirectionally active parameter in DNA-protein interactions can be exemplified by the activity of the restriction endonucleases DpnI and DpnII. DpnI cleaves the nucleotide sequences G6mATC only when the A residue in the recognition sequence is methylated, whereas DpnII is inhibited by a 6mA residue in this sequence [27]. Hence a methylated nucleotide can both obstruct or facilitate and be required for the activity of a restriction endonuclease, i.e. for the interaction between a given nucleotide sequence and the protein which specifically recognizes this sequence. A similarly instructive example is not available for a m5C containing recognition sequence of a restriction endonuclease.
Among the DNA containing viral genomes, examples of completely unmethylated as well as completely 5´-CG-3´ methylated virion DNA molecules exist. The encapsidated virion DNAs of the human adenoviruses [28] are unmethylated. In striking contrast, the double-stranded genome of frog virus 3 (FV3), an iridovirus, is completely methylated in all 5´-CG-3´ dinucleotides [29, 30]. As will be discussed later, the intracellular FV3 virion DNA becomes quickly remethylated after replication. Possibly, due to the specific nucleotide sequence of the FV3 genome, the viral and/or cellular proteins, which have to interact with this viral genome in the course of viral transcription and replication, are not inhibited by FV3 DNA methylation. Some of them may even require a methylated genome for full activity.
We have used several techniques, total hydrolysis of the virion DNA followed by bidirectional chromatography and electrophoresis [28], which allow the separation of C from m5C residues, as well as genomic sequencing methods [31, 32], to demonstrate that the virion DNA as well as the free, i.e. not host cell genome integrated, intracellular adenovirus DNA in productively or abortively infected cells [33] remains unmethylated. In the latter study, restriction endonucleases were used to document the absence of m5C residues at least in the HpaII recognition sequences 5´-CCGG-3´. The intracellular genomes of the episomally persisting Epstein-Achong-Barr virus (EBV) have become methylated to a certain extent [34]. Similarly, the genome of another persisting virus, herpes virus saimiri, in lymphoid tumor cell lines has been shown to be extensively methylated [35]. The retroviral progenomes also become methylated [36, 37].
SYREC, an Ad12 recombinant genome, which carries unmethylated cellular DNA. When adenovirus type 12 (Ad12) was serially propagated on human cells in culture, a variant Ad12 genome arose which constituted a naturally generated recombinant between the left terminal 2081 nucleotides of Ad12 DNA and a large palindromic fragment of cellular DNA. This viral recombinant could be separated from the authentic Ad12 virions due to its lower buoyant density by equilibrium sedimentation in CsCl density gradients [38]. The existence of this symmetric recombinant (SYREC) proved that recombination could occur between viral and cellular DNAs in human cells, which were productively infected with human Ad12 [39].
The cellular DNA in this huge palindromic genome of some 34 kb with identical left ends of Ad12 on either terminus of the recombinant genome comprised cellular DNA sequences of both the unique and repetitive types. Interestingly, these cellular DNA sequences were completely unmethylated in the virion recombinant, but the same cellular DNA sequences were highly methylated in the human cellular genome from which they had been originally derived [38]. This finding demonstrates that free adenovirion DNA remains devoid of m5C in the same human cell nucleus in which adenovirion DNA replicates and in which the methylation of cellular DNA is maintained in specific patterns. Apparently, the cellular DNMT fail to gain access to the free virion DNA, possibly because adenovirus DNA can avail itself of its own, specific, virion genome-encoded mechanism of DNA replication with the adenovirus terminal protein (TP), its viral DNA polymerase (pol), and the DNA binding protein (DBP). Alternatively, it is conceivable that free intranuclear adenovirus DNA becomes protected from de novo methylation by binding to specific proteins. Adenoviral DNA replication is at least partly independent of the cellular replication machinery, except for the requirement for nuclear factors I, II, and III which might not be linked to any of the cellular DNMT. On the other hand, the intracellularly located, episomal DNA of EBV must be tightly associated with the replication system for cellular DNA with which it replicates in synchrony. Thus, the EBV episomes might be in close contact with cellular DNMT and become methylated.
The adenovirus SYREC molecule and its ability to replicate in human cells in the presence of a helper adenovirus with an intact authentic viral genome has been the model for the construction of the gutless adenovirus vectors of the third generation [40]. These researchers have been able to separate the recombinant virus from its wild type precursor also by equilibrium sedimentation in CsCl density gradients.
Suppression of the frequency of 5´-CG-3´ dinucleotides in the genomes of the small eukaryotic viruses. The dinucleotide 5´-CG-3´ is statistically underrepresented in all but four of the small viruses with a genome size of 41]. In the larger viral genomes, the abundance of this dinucleotide follows statistical expectations. The retrotransposons in eukaryotic genomes are also characterized by low values of 5´-CG-3´ dinucleotides. There are several possible interpretations for these phenomena: methylation effects during the proviral states of some of these genomes which would lead to their silencing, and also dinucleotide stacking energies and mutation mechanisms or selection during evolution.
De novo Studies on integrated Ad12 genomes in transformed or tumor cells. In the course of investigations on the mode of Ad12 DNA integration in Ad12-transformed hamster cells by using restriction endonucleases, the de novo methylation of integrated foreign DNA was discovered. The generated fragments of cellular DNA were separated by electrophoresis on agarose gels and further analyzed by Southern blotting [42] and hybridization to 32P-labeled Ad12 DNA or, more specifically, to the 32P-labeled terminal fragments of Ad12 DNA. In this way, the terminal viral DNA fragments linked to the immediately abutting cellular DNA segments could be identified. methylation of foreign DNA, which was integrated into the mammalian genome.
In an attempt to generate small junction fragments, which could be more easily analyzed, frequently cutting restrictases like HpaII were employed. In these experiments, we discovered that the integrated form of Ad12 DNA was not effectively cleaved by HpaII, whereas virion DNA, previously shown to be unmethylated [28], was readily cut. These data implied that the integrated Ad12 DNA had become de novo methylated upon integration into the established hamster genome [2, 14, 43].
This interpretation could be proven when the isoschizomeric restriction endonuclease pair HpaII and MspI became available. Both enzymes recognize the sequence 5´-CCGG-3´. MspI cleaves irrespective of the presence of a m5C residue in the 3´-located C-position in the recognition sequence, but HpaII is capable of cleaving only the unmethylated sequence [44]. Along these lines, the integrated Ad12 DNA, like virion Ad12 DNA studied as a control, was completely cleaved by MspI, whereas HpaII could cleave only the virion DNA to completion. Integrated Ad12 DNA was cut incompletely by HpaII and was thus recognized to be 5´-CCGG-3´ methylated in distinct patterns. Here, we were able to document one of the early examples for the notion that foreign DNA inserted into established mammalian genomes became heavily methylated [14, 15]. This now commonly reproduced finding was later extrapolated to numerous other eukaryotic genomes, including those of plants [45]. The human papilloma viruses 16 and 18 integrated into the genomes of cells from human cervical carcinomas are also methylated in functionally distinct patterns [46].
Site of initiation of de novo methylation - site of foreign DNA integration in the recipient genome. In later studies, we have demonstrated in numerous Ad12-transformed hamster cell lines and particularly in Ad12-induced hamster tumor cells that integrated Ad12 DNA is an excellent substrate for the action of cellular DNMT [47-49]. The patterns generated in different cell lines and tumors exhibited some similarities but did not appear to be identical. Extent and pattern of methylation of integrated foreign DNA were directed rather by the site of integration in the recipient genome than by the nucleotide sequence of the foreign DNA [50], although the latter could have some influence as well. In this context, the observation was of interest that a cloned E1 segment of adenovirus DNA genomically fixed by transfection into hamster cells and integrated at several different loci in the host hamster genome became methylated to different extents or could remain hypo- or unmethylated [50]. Hence, the site of insertion of foreign DNA had to be a strong determinant in its subsequent de novo methylation.
The sites of initiation of de novo methylation were determined by using the cloned HindIII DNA fragments of Ad12 DNA as hybridization probes on HpaII- or MspI-cleaved DNA from different Ad12-induced hamster tumors. These DNA fragments were separated by electrophoresis on agarose gels and subsequently analyzed by Southern blotting and hybridization to the 32P-labeled Ad12 DNA fragments. The results of a large number of experiments [50] demonstrated that de novo methylation was initiated inside the integrated Ad12 DNA molecules in two paracentrally located regions of the Ad12 genomes. De novo methylation did not commence at or close to the termini of Ad12 DNA, which were linked to cellular DNA. The termini, in fact, remained hypomethylated, possibly because continued hypomethylation and expression of the gene products from the Ad12 regions E1 (left terminus) and E4 (right terminus) were likely selected for in Ad12-transformed cells and in Ad12-induced tumors.
We have investigated the site of initiation of de novo methylation in integrated Ad12 genomes also at the nucleotide level by applying the bisulfite genomic sequencing reaction. When CsCl-purified Ad12, 106 to 107 plaque forming units per animal, is injected intramuscularly into newborn hamsters (Mesocricetus auratus), numerous tumors of different sizes develop in the animals' peritoneal cavities. In these intraperitoneal tumors of different sizes, the paracentrally located regions of integrated Ad12 genomes were analyzed by the bisulfite protocol for the state of methylation.
Methylation levels did not exhibit an unequivocal relation to tumor size. Initiation of DNA methylation, moreover, was not emanating from a specific nucleotide or set of nucleotides in the region previously shown to be a site of initiation of de novo methylation. Initiation was rather regional and appeared to emerge from several sites within this region [50, 51]. Hence, de novo methylation seems to commence in a region and not at a single specific nucleotide. Of course, this experimental approach does not allow deriving a definite correlation between the time of foreign DNA integration and that of the initiation of DNA methylation.
In a related type of experiment, a plasmid construct, which contained the E2A late promoter of adenovirus type 2 (Ad2) and the prokaryotic gene for the chloramphenicol acetyltransferase (CAT) as a reporter, was transfected into hamster cells. The HpaII-premethylated or unmethylated pAd2E2AL-CAT gene construct was genomically fixed in hamster cells by co-transfection with the unmethylated pSV2-neo plasmid. In this plasmid, the early SV40 promoter controlled the neomycin phosphotransferase gene, which facilitated the selection of transgenic cells. Stability of methylation status and expression of the CAT gene were assessed in a number of clonal transgenic cell lines [52]. The foreign DNA was integrated frequently in multiple tandems of the transfected plasmid. Among 19 clonal cell lines, the unmethylated construct remained in that state; and in 18 of these lines, the CAT gene was continuously expressed. Among 14 cell lines transgenic for the premethylated construct, seven lines failed to express the test transgene, and the three 5´-CCGG-3´ sites in its late E2A promoter remained almost completely methylated. In five cell lines, the promoter remained partly methylated and the CAT gene was only weakly expressed. In two cell lines, the premethylated promoter lost the m5C modification altogether, and the CAT gene was strongly expressed [52].
Factors determining de novo methylation: reinsertion of a mouse gene into its authentic position. We further pursued the general question of which factors would affect the de novo DNA methylation in mammalian genomes. The mouse B lymphocyte tyrosine kinase (BLK) gene was re-integrated by homologous recombination into the genome of mouse embryonal stem (ES) cells. Two different plasmid constructs containing that gene were used for these experiments. One construct also carried the weak E2A late promoter of Ad2 DNA in front of the luciferase gene. In the second, this gene was controlled by the strong early SV40 promoter. Upon homing through homologous recombination to the authentic chromosome 14 or by heterologous recombination to many different loci in the mouse genome, methylation patterns in the integrates were assessed by restriction with the methylation-sensitive endonuclease HpaII or the insensitive MspI. The mouse BLK gene reinserted into the genome by homologous recombination had reestablished the identical methylation pattern characteristic for the authentic, non-manipulated mouse BLK gene [13]. The extent of de novo methylation in the DNA segments adjacent to the BLK gene in the integrated construct depended on the promoter present in the plasmid construct and on the location of the recombined construct in the ES genome. In homologously inserted DNA, which carried the weak Ad2 promoter, de novo methylation was extensive. Presence of the strong SV40 promoter led to hypomethylation or no methylation at all. Even when the enhancer sequence was removed from the SV40 promoter, hypermethylation was still observed. All randomly integrated constructs, independently of the type of promoter or enhancer included, were hypermethylated. We concluded [13] that an authentic mouse gene reinserted into its original genomic site was remethylated to the identical pattern as previously present on the target and on the allelic site; that heterologous recombination to randomly targeted loci did not confer the mouse BLK gene-specific methylation pattern, and that promoter strength in a construct was able to influence the pattern of methylation imposed de novo on the inserted construct after homologous recombination in the mouse genome.
De novo DNA methylation - an ancient cellular defense mechanism? The de novo methylation of integrated foreign DNA is a phenomenon widely documented throughout the phyla of eukaryotic organisms, in mammals as well as in plants [45]. The nowadays, established genomes have evolved over many millennia. We tend to assume that the evolution of genomes is a continuous process which started right after the beginning of organismic life and probably before organisms arose and will continue as long as this biologic system can be maintained. Even under experimental conditions, many organisms have been shown to be capable of accepting and of accommodating foreign DNA, although with unpredictable - advantageous or catastrophic - consequences for the acceptor cell. Since all cells in culture or organisms in environment-challenged life are subject to stringent conditions of selection, even the cell which has been forced by its innate recombination mechanisms to tolerate the genomic insertion of foreign DNA can avail itself of an ancient defense mechanism against the genetic activity of foreign DNA which could carry active genes. Since promoter methylation has been identified as part of a mechanism for the long-term silencing of genes and DNA segments, the de novo methylation of integrated foreign DNA can be contemplated as such a defense mechanism or at least as an integral part of it [53, 54].
A large part of m5C residues is found in the parasitic sequence elements of retrotransposons and endogenous retroviruses which constitute >35% of the human genome. Perhaps intragenomic parasites are recognized by their high copy number. Long-term inactivation by DNA methylation also entails the possibility of m5C residues being deaminated to T's followed by permanent inactivation [55].
Are integrated foreign DNA sequences stabilized by hypermethylation? The Ad12-transformed cell line T637 was obtained by the in vitro transformation of BHK21 hamster cells with Ad12 [56] and carries about 15 copies of Ad12 DNA integrated at a single chromosomal locus [57-59]. Unintegrated, free Ad12 DNA could never be detected in any of the adenovirus-transformed cell lines or in Ad12-induced tumor cells. The integrated Ad12 DNA is methylated in functionally significant patterns [15, 51]. Upon continuous propagation in cell culture, a small number of morphological revertants arose from the cell line T637 [60]. These revertants exhibit a fibroblastic phenotype in contrast to the more rounded, epithelioid appearance of the T637 cells. The revertants can be selected due to their resilience to spent, acidic medium in which T637 cells detach.
We have studied several of these revertants with respect to their content of residual viral DNA. In one of these revertants, TR12, only one copy of Ad12 DNA and an approximately 3.9 kb long fragment from the right terminus of Ad12 DNA persist in the integrated state (N. Hochstein, I. Muiznieks, and W. Doerfler, manuscript in preparation). Studies using methylation-sensitive restriction endonucleases revealed that the integrated Ad12 DNA in the revertant cell line TR12 was even more extensively methylated than in the cell line T637 from which TR12 originated [61]. Preliminary results adduced from experiments, in which the bisulfite genomic sequencing method was applied, confirm these data. We propose that hyper- or nearly completely 5´-CG-3´-methylated foreign DNA sequences would be more stably integrated than less completely methylated foreign genomes. Repetitive sequences, like endogenous retroviral retrotransposons, appear to be significantly, but not completely, methylated [62].
De novo methylation of Ad12 and of cellular DNA in hamster tumors. The insertion of foreign DNA into established mammalian genomes has consequences for the inserted foreign DNA and for the recipient host genome. We have chosen to study these events in Ad12-induced hamster tumor cells, a model in basic research as well as for its significance in viral oncology. When Ad12 is injected subcutaneously into newborn hamsters within 24 h after birth, undifferentiated tumors develop at the site of topical application of the virus within a few weeks in 70 to 90% of the animals, which survive injection. As mentioned above, upon the intramuscular injection of Ad12 into the gluteal region, numerous tumors are found intraperitoneally. Histologically, these tumors exhibit Homer-Wright rosette-like structures indicative of primitive neuroectodermal tumors (PNET).
The tumor cells express proteins, which are characteristic both for neuroepithelial as well as for mesenchymal cells [51, 63]. Each tumor cell carries multiple copies of integrated Ad12 DNA. Free viral DNA has never been found in any of these tumor cells.
The cells from an individual tumor carry the viral integrates, with few exceptions, all at a single chromosomal location which is identical in all cells of a given tumor. Among 60 different tumors investigated only one showed two different chromosomal loci to be occupied by Ad12 DNA when investigated by the fluorescent in situ hybridization (FISH) technique. When we compared the sites of Ad12 DNA integration in more than 100 different hamster tumors, the sites of viral DNA integration in the host cell genome were different from tumor to tumor [43, 47, 64]. We, therefore, favor a model of clonal origin of these Ad12-induced tumors. Aside from the de novo methylation of the viral integrates in all of the Ad12-induced tumors, there can be changes in the extent of DNA methylation also in the host genome [62]. Moreover, the transcription patterns of cellular genes in different tumors can be very similar but there are also differences. The patterns of transcription of the integrated Ad12 genes are strikingly similar in all Ad12-induced hamster tumors analyzed and resemble those patterns in the Ad12-transformed hamster cell line T637 [51, 63].
Loss of Ad12 genomes is compatible with maintenance of the oncogenic phenotype. Upon continuous cell culture, some of the Ad12-induced hamster tumor cells, which carried multiple copies of integrated and de novo methylated Ad12 genomes, lost the viral DNA sequences completely or almost completely. Surprisingly, these revertants devoid of Ad12 DNA and notably lacking its left terminus with the transforming E1 region of the viral genome retained their oncogenic phenotypes when re-injected into hamsters [48, 65]. We consider this result to be highly significant in that it demonstrates that Ad12 is capable of inducing tumors, which keep their oncogenic properties in animals, even when the E1 region is completely lost from these tumor cells.
Many researchers in the field of adenovirus tumor biology consider the E1 region of the adenoviral genome akin to an oncogene and its maintenance paramount in preserving the oncogenic potential of the adenovirus-induced tumor cells. However, this notion has been experimentally derived mainly by using rat embryo fibroblasts, which were transformed in culture through the transfection of adenovirus type 5 (Ad5) DNA fragments [66]. Obviously, tumor induction by Ad12 in animals can be a quite different process and can probably not appropriately be mimicked by transfection experiments in cell culture. Below I propose an alternate way of looking at the mechanism of viral oncogenesis. In this view, global changes in the cellular genome as a consequence of viral DNA integration are invoked as a mechanism of viral oncogenesis in addition to the expression of a single “oncogene” [4, 5, 7].
DNA methylation in non-CpG dinucleotides; hemimethylated DNA. The establishment of de novo patterns of DNA methylation in mammalian genomes is characterized by the gradual spreading of methylation, which has been documented to occur across multiple copies of integrated adenovirus genomes as well as, at the nucleotide level, in the integrated E2A promoter of Ad2 DNA [52, 67]. A few 5´-CG-3´ sequences can remain hemimethylated for several cell generations before they become totally methylated. Hemimethylation may be a transient phenomenon but could also persist for a certain period in specific segments of a transgene or in the genome in general. In the Ad2-transformed cell line HE2, the E2A promoter in the integrated Ad2 genome is heavily methylated not only in all 5´-CG-3´ dinucleotides but also in some 5´-CA-3´ and 5´-CT-3´ dinucleotides [68]. Evidence for the occurrence of m5C in non-5´-CG-3´ dinucleotides has also been presented by Woodcock et al. [69]. There is at present no plausible explanation whether and how this type of methylation can be maintained following DNA replication.
Initiation and spreading of de novo methylation. The mechanism of de novo methylation is not well understood. From the results of experiments performed with the adenovirus system, the following conclusions appear well founded:
- in a transgene the size of Ad12 DNA, 34,125 bp [70], de novo methylation starts in internal regions of the genome [49-51] and spreads from there across the transgene [67, 68]. In transformed or tumor cells, the early regions, particularly E1, can be partly spared from de novo methylation, because their gene products are required during the selection for the oncogenic phenotype;
- initiation of de novo methylation is regional and not confined to one or a few contiguous 5´-CG-3´ dinucleotides [51]. In Ad12-induced tumors, the extent of Ad12 transgene methylation is not only dependent on tumor size. Transcribed regions of the transgene are hypomethylated, inactive segments hypermethylated [15, 71, 72];
- spreading of DNA methylation is not entirely contiguous. For unknown reasons, certain 5´-CG-3´ dinucleotides can remain unmethylated (I. Hochstein, I. Muiznieks, and W. Doerfler, manuscript in preparation);
- the extent of de novo methylation and the speed of its spreading seem to be determined by the chromosomal site of transgene localization, to a lesser extent also by the nucleotide sequence of the transgene [13, 50];
- the reestablishment of the authentic pattern of DNA methylation in the correctly reinserted BLK gene in mouse embryonic stem cells [13] implies that each segment of the mammalian genome is capable of exerting a memory function which could be directly related to the mechanism of maintenance methylation. Specific chromatin structures at different sites of the genome may be decisive in directing the methylation reaction;
- in the same set of experiments [13], we noted that the same reintegrated BLK gene, when it was inserted at randomly selected sites in the genome, was methylated in completely different patterns. Moreover, the extent of the de novo methylation was influenced by the strength of the viral promoter in the construct integrated. The weaker E2A late promoter of Ad2 DNA led to a more extensive de novo methylation than the presence of the stronger early promoter of SV40 DNA;
- when Ad12-transformed cells or Ad12-induced hamster tumor cells were continuously passaged in culture, revertants arose which had lost all or a part of the multiple copies of integrated Ad12 DNA. The then persisting Ad12 DNA genomes in the revertants seemed to be more completely methylated than the lost copies. Hence, the idea was put forward that the levels of DNA methylation of transgenes might be related to their stability of fixation in the host genome [61]. Moreover, continuous culture of different Ad12-induced tumor cells led sometimes to the selection of cell lines with very similar integration patterns [73]. It is conceivable that the integration sites in the thus selected cells were most compatible with the survival of these cells in culture.
Answers to all your Biology Questions
