Distal Transgene Insertion Affects CpG Island Maintenance during Differentiation*

About half of all genes have a CpG island surrounding the promoter and transcription start site. Most promoter CpG islands are normally unmethylated in all tissues, irrespective of the expression level of the associated gene. Establishment of the appropriate patterns of DNA methylation in the genome is essential for normal development and patterns of gene expression. Aberrant methylation of CpG islands and silencing of the associated genes is frequently observed in cancer. One gene with a 5′-CpG island is cytoplasmic β-actin, which is an abundantly expressed protein and a major component of microfilaments. Inserting a βgeo cassette into the 3′-untranslated region of β-actin gene led to widespread but not ubiquitous lacZ expression in mice heterozygous for the modified β-actin allele. Surprisingly, embryos homozygous for this insertion died at mid-gestation. The modified β-actin allele was expressed in undifferentiated embryonic stem cells but was turned off as these cells differentiate in vitro and in vivo. We demonstrate that the insertion affects the maintenance of the methylation status of the CpG island of the modified β-actin allele in differentiated but not in undifferentiated embryonic cells. These data suggest that there is a two-step process to defining a CpG island, requiring both embryonic establishment and a signal that maintains the CpG island in differentiated cells. Furthermore, they indicate that features built into the CpG island are not sufficient to direct CpG island maintenance during differentiation.

About half of all genes have a CpG island surrounding the promoter and transcription start site. Most promoter CpG islands are normally unmethylated in all tissues, irrespective of the expression level of the associated gene. Establishment of the appropriate patterns of DNA methylation in the genome is essential for normal development and patterns of gene expression. Aberrant methylation of CpG islands and silencing of the associated genes is frequently observed in cancer. One gene with a 5-CpG island is cytoplasmic ␤-actin, which is an abundantly expressed protein and a major component of microfilaments. Inserting a ␤geo cassette into the 3-untranslated region of ␤-actin gene led to widespread but not ubiquitous lacZ expression in mice heterozygous for the modified ␤-actin allele. Surprisingly, embryos homozygous for this insertion died at mid-gestation. The modified ␤-actin allele was expressed in undifferentiated embryonic stem cells but was turned off as these cells differentiate in vitro and in vivo. We demonstrate that the insertion affects the maintenance of the methylation status of the CpG island of the modified ␤-actin allele in differentiated but not in undifferentiated embryonic cells. These data suggest that there is a two-step process to defining a CpG island, requiring both embryonic establishment and a signal that maintains the CpG island in differentiated cells. Furthermore, they indicate that features built into the CpG island are not sufficient to direct CpG island maintenance during differentiation.
Actins are abundant structural proteins important in maintaining cell shape and motility (1). Cytoplasmic ␤-actin is a major component of microfilaments and is present in virtually all cell types except differentiated skeletal muscle (2). The promoter of the single-copy ␤-actin gene is expressed in a wide range of cell types in culture and is transiently induced at the transcriptional level by exposure of quiescent cells to serum growth factors (3). In common with many other widely expressed genes in mammalian cells, ␤-actin gene is associated with a CpG island positioned over the promoter and first intron of the gene (4).
DNA methylation at the C-5 position of cytosine is a major form of DNA modification in vertebrate animals (5). The majority of methylated cytosines are found in the palindromic dinucleotide, CpG (6). This dinucleotide is underrepresented ϳ5-fold in the genome because of the increased mutation rate of 5-methylcytosine by deamination. The majority, but not all, of the CpG dinucleotides in vertebrate DNA are methylated. CpG islands are clusters of the of CpG dinucleotides that have maintained their CpG content as they are kept free from methylation (7). These islands remain free from methylation in the germ line and in somatic cells (8). A small proportion of these islands do become methylated, including the island associated with genes of the inactive X chromosome and imprinted genes (9). Genome-wide analysis of CpG islands by restriction landmark genomic scanning has demonstrated that tissue specific CpG island methylation is relatively frequent and that embryonic stem (ES) 3 cells have a propensity to methylate CpG islands (10).
DNA methylation can induce transcriptional repression in a sequence-independent manner by promoting condensation of chromatin (11). Methyl-DNA-binding proteins such as MeCP2 recruit histone deacetylases through interaction with mSin3A (12). Inappropriate methylation of a CpG island is likely to be associated with a change in the chromatin structure of the locus and resultant silencing of the gene. Tissue culture and cancer cells frequently show a genome-wide increase in CpG island methylation (13). Aberrant de novo methylation of growth regulatory genes has been associated with tumorigenesis in humans; however, very little is known about the mechanisms governing the increases of de novo methylation of CpG islands in cancer cells (14).
Although much is known about CpG islands, the relationship between transcriptional activity and methylation status remains to be established. Studies of both the mouse and hamster Aprt genes indicate that the transcription factor Sp1 may play an important role in preventing methylation of CpG islands (15,16). Mutation of the Sp1 sites in the promoters of these genes leads to methylation of the CpG island in vivo. It has been proposed that Sp1 itself may exert a localized demethylating effect on the surrounding DNA, and therefore binding of Sp1 alone would be sufficient to protect the DNA (17). Alternatively, active transcription induced by binding of Sp1 to the promoter protects the CpG island from methylation. Subsequently, however, it was demonstrated that when a mutation was made in the Sp1 gene in the mouse germ line, CpG islands were still protected and remained methylation-free in homozygous mutant embryos (18). Furthermore, the mechanism underlying the maintenance of methylation status of CpG islands is unclear. For example, the CpG island spanning the second exon of the mouse H2-Ab1 gene, and located some distance away from the promoter, is associated with a separate intronic promoter active only in embryonic and germ cells (19). Taken together these studies suggest that embryonic transcription is required to define a CpG island and that this state is copied to differentiated cells by maintenance methylation.
We now demonstrate that an insertion in the 3Ј-untranslated region (UTR) of the ␤-actin gene interferes with the maintenance of the CpG island associated with this gene in differentiated cells without preventing its transcription in embryonic stem cells. This suggests that maintenance of a CpG island requires a signal that may be dependent on embryonic transcription but must be maintained in differentiated cells to prevent CpG island methylation. Hence the cellular memory of early embryonic transcription may require an epigenetic mark to ensure that it is maintained throughout development and in adult cells.

Plasmids and Construction of Targeting Vectors for the ␤-Ac-
tin Locus-The ␤-actin genomic locus was cloned from a 129Sv/J mouse bacterial artificial chromosome (BAC) library (20). A 7.5-kb fragment, corresponding to the entire coding region of the ␤-actin gene, was subcloned to use as homology arms for the targeting vector. The Actb m1 targeting vector was generated by subcloning an IRES-␤geo cassette (20) into the actin 3Ј-UTR, 55 bp downstream of the translational stop codon. The Actb m2 targeting vector was cloned by inserting an IRES-loxP-Neo-Poly(A)-loxP-lacZ cassette into the same site. The Actb hyg and Actb pur targeting vectors were generated by cloning an IRES-hygromycin (Clontech) or IRES-puromycin (Clontech) cassette, respectively, into the same site in the 3Ј-UTR. Cre recombinase was expressed by transfecting cells with the plasmid pSP-PGK-Cre (21). Further details of the cloning strategy are described in the supplemental "Materials and Methods." ES Cell Culture and Transfection-ES cells were cultured as described previously (22). Details of the electroporation and generation of the series of targeted actin alleles (Actb m1 , Actb m2 , Actb m3 , Actb hyg , and Actb pur ) are given in the supplemental "Materials and Methods." Generation of Targeted Mouse Lines-ES cell clones heterozygous for the targeted insertions were microinjected into C57BL/6 blastocysts to generate chimeras carrying modified actin alleles. Further details of breeding and genotyping of targeted mouse lines are given in the supplemental "Materials and Methods." Differentiation of ES Cells in Vitro-ES cells were lightly trypsinized to give clumps of cells and were induced to differentiate by culturing them as aggregates in suspension in the absence of leukemia inhibitory factor. After 6 days, the aggre-gates were outgrown on gelatinized tissue culture plates in the absence of leukemia inhibitory factor for a further 6 days until the appearance of morphologically differentiated cells. Differentiation was confirmed by staining cells with alkaline phosphatase as described previously (23). Cultures used for analysis contained Ͼ95% alkaline phosphatase-negative cells.
lacZ Staining-␤-Galactosidase activity was determined in situ in cells or embryos by staining with X-gal as described previously (22).
DNA Analysis-Genomic DNA was prepared from ES cells and mouse tissues as described previously (24). Selection of probes and conditions for blotting and PCR genotyping are elaborated in the supplemental "Materials and Methods." RNA Analysis-Generation of probes for the RNase protection assay and reverse transcription-PCR analysis is described in detail in the supplemental "Materials and Methods." RNA was extracted from ES cells and mouse tissues using TRIzol (Invitrogen) according to manufacturer's protocol. RNase protection assays were performed using MAXIscript and RPA II kits (Ambion). Digestion products were electrophoresed on 8 M urea, 8% acrylamide gels. The relative abundance of the products was determined by densitometry using a PhosphorImager with ImageQuant software (Amersham Biosciences).

RESULTS
Insertion of ␤geo Cassette in the ␤-Actin 3Ј-UTR-A 12-kb fragment containing the whole ␤-actin locus was subcloned and used to generate a gene targeting vector. An IRES-␤geo cassette was inserted in the 3Ј-UTR of the ␤-actin gene, 55 bp downstream of the stop codon but upstream of the polyadenylation signal (Fig. 1A). HM1 ES cells (25) were electropo- probes used for genotyping are indicated on the diagram. B, Southern blot of targeted ES clones using a 5Ј probe following digestion of genomic DNA with EcoRI. Genomic DNA is from wild-type (ϩ/ϩ), heterozygous (ϩ/Actb m1 ), and homozygous ES cells (Actb m3 /Actb m1 ). The wild-type band is 11.2 kb; following targeting a 6.3-kb band is visible. C, Southern blot of targeted ES clones using a 3Ј probe following digestion of genomic DNA with HindIII. Genomic DNA is from wild-type (ϩ/ϩ), heterozygous (ϩ/Actb m1 ), and homozygous ES cells (Actb m3 /Actb m1 ). The wild-type band is 8.2 kb; following targeting a 7.3-kb band is visible. rated with linearized targeting vector, and G418-resistant clones were analyzed by Southern blotting. Targeted clones were identified using both 5Ј and 3Ј probes for the sequences flanking the insertion (Fig. 1, B and C). During the course of this article, we have described this allele as the modified actin locus (Actb m1 ), as it does not make any mutation in the promoter or coding region of the ␤-actin gene.
Embryos Homozygous for the Actb m1 Insertion Die around Implantation, although Homozygous ES Cells Are Viable-Two ES cell clones carrying the Actb m1 locus gave rise to germ line chimeras. The heterozygous mice showed no obvious abnormality and were indistinguishable from wild-type littermates. Following intercrossing of these heterozygous mice, no liveborn homozygous mice were identified, indicating that the insertion of the lacZ gene cassette into the 3Ј-UTR of the actin gene resulted in prenatal lethality (Table 1). This result was unexpected, as the construct does not disrupt the coding sequence of the ␤-actin gene.
To determine the timing of the embryonic lethality of the homozygous Actb m1 mice, timed matings were established. Genotyping of embryos from day 8.5 to 15.5 of gestation indicated that no homozygote Actb m1 embryos were detected among the 53 live embryos screened, and genotyping of blastocysts revealed that homozygote Actb m1 embryos were present at the expected frequency (Table 1). These results indicate that homozygote Actb m1 embryos died around the time of implantation.
The lethality associated with the modified Actb m1 allele may have been caused by a lower amount of actin protein produced from the modified allele. To allow us to address this possibility, ES cells homozygous for the Actb m1 allele were generated. We pursued two strategies to achieve these homozygous cells. The first strategy involved sequential modification of the ␤-actin locus (Fig. 1A). HM1 cells were targeted with a second targeting vector in which loxP sites flanked the neo gene, an allele termed Actb m2 . After removal of the neo gene by transient transfection of Cre recombinase, the second allele was then targeted by electroporating Actb m3 cells with the Actb m1 construct and selecting on G418 to generate Actb m1 /Actb m3 cells with both ␤-actin alleles modified. Correctly targeted homozygous ES clones were identified by Southern blotting (Fig. 1, B and C). The second strategy involved insertion of IRES-puromycin or IREShygromycin resistance genes into the second allele (data not shown). All of the homozygous cell lines behaved similarly, and in each case the undifferentiated ES cells grew normally and at the same rate as wild-type cells. This indicates that sufficient ␤-actin protein was produced from the modified allele to allow the cells to grow normally. This suggests that, at least in vitro, the reduced ␤-actin levels are compatible with normal cell growth.
When homozygous the ES cells were differentiated by withdrawal of leukemia inhibitory factor, they were able to differentiate as extensively as wild-type cells and give rise a similar range of cell types but grew more slowly upon induction of differentiation. This suggests that the modified allele has an effect on the ES cell growth but only after the onset of differentiation.
Patchy Expression of lacZ Is Seen in Differentiated ES Cells and Adult Mice-Expression of the modified locus was assayed by lacZ staining in both Actb m1 ES cells and mouse embryos. In Actb m1 ES cells the expression of the ␤-galactosidase is widespread in undifferentiated cells (Fig. 2, A and  B). Although there appears to be some variation in the level, all of the undifferentiated cells express the lacZ gene. In contrast, following differentiation of the cells in vitro lacZ expression in the cells becomes very patchy and was detected in only a small fraction of the differentiated cells (Fig. 2C). A similar phenomenon was observed when the cells differentiate in vivo. Widespread lacZ activity was seen in whole mount embryos and tissues examined from newborn mice (Fig. 2, D and E), and lacZ expression subsequently became progressively restricted to certain tissues as the mice aged; even in tissues showing continued expression of the lacZ gene the expression was   (Fig. 2, F and G, and supplemental Table S1). The expression profile of the transgene correlates well with the expression profile described for the endogenous gene (2), and the level of transgene expression does not vary significantly between different Actb m1 lines or between heterozygous littermates within a transgenic line. There is no substantial difference either in the level of lacZ expression in the tissues of a mouse at age 3 months and an 11-month-old mouse (data not shown). This implies that once a tissue has reached maturity, there is no further alteration of expression of the modified allele.

Methylation of Targeted ␤-Actin CpG Island Correlates with Patchy Expression in Mice-
The ␤-actin gene promoter is associated with a CpG island, which normally remains unmethylated in all tissues. In order to test whether aberrant methylation of the CpG island is involved in the silencing of the modified allele, Southern blot analysis of DNA isolated from mouse tissues was performed. Genomic DNA was restricted with EcoRV and either the methylation-sensitive enzyme HpaII or its methylation-insensitive isoschizomer, MspI, and probed with the 1.5-kb region of the actin locus including the promoter, first exon, and first intron of the actin gene (Fig. 3A). This probe spans the entire CpG island at the 5Ј-end of the actin gene. Methylation of the 5Ј-CpG island was apparent in all tissues analyzed from the heterozygous Actb m1 mouse, but no methylation was detected in DNA from tissues of wild-type mice (Fig. 3B). Although both the brain and small intestine still contained a significant percentage of cells expressing lacZ, methylation of the ␤-actin CpG island was still detectable in these tissues (Fig. 3B). In addition there was no firm correlation between the degree of methylation of the ␤-actin CpG island and the expression level of lacZ in a tissue ( Fig. 3B and supplemental Table S1).
The appearance of methylation does correlate with patchy expression of modified allele as judged by lacZ staining in tissues from adult mice. This result was confirmed by analyzing genomic DNA by a PCR assay (Fig. 3C). Additionally the PCR assay was used to demonstrate the presence of methylation at the CpG island in heterozygous embryos. By analysis of DNA from embryos at 9.5 days post-coitus, we detected CpG island methylation indicating a correlation between methylation status and survival of the embryo. Therefore, it is possible that silencing of the modified allele CpG island may underlie embryonic lethality of the Actb m1 homozygotes.
Methylation Is Restricted to the Modified Allele-To determine whether aberrant methylation affected both wild-type and Actb m1 alleles or was confined to the modified allele, a Southern blot was performed using BamHI, which generated a polymorphism between the modified and endogenous allele (Fig. 4A). When this enzyme is used in combination with a CpG methylation-sensitive enzyme such as XhoI or SacII, only the 4-kb band corresponding to the modified allele remains (Fig. 4B). Hence only the Actb m1 allele becomes methylated whereas the endogenous allele remains free of methylation. This result makes it less likely that the Actb m1 allele RNA is involved in promoter silencing. As the modified  allele expresses coding sequences for ␤-actin as well as lacZ, any sequence-specific silencing based on the RNA may target both the modified allele and the endogenous allele simultaneously. There is no apparent effect on either the methylation of the CpG island or the transcription of the endogenous RNA in the heterozygous cell lines. In addition, no methylation could be detected in CpG islands neighboring the Actb locus (supplemental Fig. S1, A and B).
Methylation Is Detectable in ES Cells Only after Differentiation-As undifferentiated embryonic stem cells and early embryos still showed widespread expression of ␤-galactosidase, the methylation status of the CpG island was analyzed in targeted embryonic stem cells. No methylation of the CpG island was detected in wild-type or Actb m1 undifferentiated embryonic stem cells (Fig. 5A), correlating with the widespread lacZ expression seen in these cells (Fig. 2B). The modified ␤-actin allele CpG island in undifferentiated ES cells remained unmethylated despite prolonged culture without selection. Undifferentiated cells were kept growing continuously without selection for 8 weeks before analysis. However, following differentiation of ES cells in vitro, methylation of the CpG island was detectable (Fig. 5B). This correlates with the appearance of morphologically differentiated cells and with the onset of patchy expression (Fig. 2C). In contrast the undifferentiated cells remained unmethylated even after prolonged growth in culture. Both heterozygous and homozygous cells showed similar changes in methylation (Fig. 5B).
Hence the results obtained from analysis of differentiating ES cells in vitro parallels those obtained from mouse tissues in vivo, because the ␤-actin CpG island starts to become methylated upon differentiation and this correlates with the appearance of silencing of expression. Cell lines targeted with hygromycin or puromycin also showed methylation following differentiation of ES cells in vitro (Fig. 5B). This suggests that the silencing was not caused by specific sequences in the lacZ or neo genes. Hence it is likely that it is the location of the insertion rather than the sequence of the inserted DNA that is most significant in contributing to the silencing of the modified allele. This hypothesis is supported by the observation that inserting the same sequences adjacent to CpG islands elsewhere in the genome does not result in methylation. 4 We note that because all our insertions used bacterially derived transgenes, their base composition may have influenced methylation propensity.
RNA Levels for the Targeted Allele Are Reduced-As active transcription has been hypothesized to help maintain CpG islands free from methylation (19), the levels of RNA for the endogenous and targeted alleles were analyzed by a ribonuclease protection assay. This allowed the RNA levels for the two alleles to be distinguished by using allele-specific probes (Fig. 6A).
The RNA corresponding to the modified allele is easily detectable in undifferentiated embryonic stem cells (Fig. 6B). The level of steady state RNA from the Actb m1 allele was 4 D. Strathdee, unpublished data.

FIGURE 5. Methylation is detectable in ES cells only upon differentiation.
A, PCR of genomic DNA from wild-type (ϩ/ϩ) and heterozygous (ϩ/Actb m1 ) mouse tissues and undifferentiated wild-type (ϩ/ϩ), heterozygous (ϩ/Actb m1 ), and homozygous (Actb m3 /Actb m1 ) ES cells. The positions of the methylation-specific (M1/M2) or control (C1/C2) primers are shown in Fig. 3A. Genomic DNA was digested with: E, EcoRV; M, MspI; H, HpaII. B, PCR of genomic DNA from wild-type (ϩ/ϩ) heterozygous ES cells (ϩ/Actb m1 and ϩ/Actb m3 ) and two clones of homozygous ES cells (Actb m3 /Actb m1 33 and 57) following differentiation in vitro. Methylation is detectable only in differentiated ES cells and mouse tissues. FIGURE 6. RNA levels for the targeted allele are reduced. RNase protection was used to measure the level of RNA from the endogenous and modified alleles. A, diagram of probes derived from endogenous (en) and targeted Actb m1 allele RNA (tg). Probe sizes following digestion are: 240-bp wild-type or 201-bp targeted allele (en probe) and 379-bp targeted or 264-bp wild-type allele (tg probe). B, transcript levels for the endogenous and targeted allele RNA using a probe specific for the wild-type allele. Shown are undifferentiated and differentiated ES cells and Actb m1 mouse brain RNA. C, transcript levels for the endogenous and targeted allele RNA using a probe specific for the targeted allele. Shown are undifferentiated and differentiated ES cell and Actb m1 mouse brain RNA using modified allele probe. reduced to 17% of the level of the endogenous gene when calculated by densitometry. Differences in the relative RNA levels were identical using a probe corresponding to either the endogenous allele or the transgene (Fig. 6, B and C). Analysis of the transcription rate by run-on analysis showed no differences among wild-type, heterozygote, and homozygote ES cells (supplemental Fig. S2). This implies that the reduction of the RNA is more likely because of an alteration in RNA stability than a change of the transcription rate.
The RNA level was subsequently further reduced upon differentiation of the ES in vitro to ϳ1.5% of the endogenous gene. A similar reduction was also apparent in RNA isolated from tissues from adult mice (Fig. 6B).
In undifferentiated ES cells, even though the RNA level of the modified allele was reduced in comparison with the unmodified allele, it remained robustly expressed. When cells differentiate, the RNA level is further down-regulated, which correlates with the onset of patchy expression and the appearance of methylation at the promoter CpG island. This occurs in both ES cells differentiated in vitro and differentiated cells in the tissues of transgenic mice. In contrast, the mRNA levels for genes surrounding the Actb locus are similar in both wild-type and homozygous targeted embryonic stem cells (supplemental Fig. S1C).
Analysis of the five different modified alleles (Actb m1 , Actb m2 , Actb m3 , Actb hyg , and Actb pur ) demonstrated that although there is a slight variation between the individual clones of cells, they all show a reduction to 15-30% of the level of the endogenous gene (Fig. 6C). When measured by densitometry the increase in the modified allele expression in the homozygote cell lines was approximately double the level of a single copy of the gene. Taken together, these data indicate that all modified ␤-actin alleles have reduced mRNA levels but remain expressed prior to differentiation.

DISCUSSION
␤-Actin is found in virtually all cell types (2,26). Building on this broad expression profile, the ␤-actin locus was identified as a potential permissive site for marker gene expression (27). Unexpectedly, our data indicate that transgene insertion into the 3Ј-UTR of the ␤-actin gene results in embryonic lethality in mice homozygous for the insertion. This phenotype is similar to targeted mutations that disrupt the actin gene (27,28). It is not clear what causes the lethality (possibly a reduction of actin protein or silencing of the actin gene upon differentiation). No effects were observed from genes neighboring the Actb locus (supplemental Fig. S1).
During early embryogenesis, global methylation patterns are erased, and a wave of de novo methylation follows around implantation (29). The timing of the inactivation of the modified ␤-actin CpG island appears to correlate with this wave of de novo methylation. This raises the possibility that silencing of the modified allele rather than the modification itself is responsible for lethality. This conclusion is also supported by the fact that homozygously modified ␤-actin ES cells are viable and grow normally. It remains possible, however, that ES cells and early embryonic tissues have different requirements for ␤-actin protein and that the reduction in levels of actin is responsible for the lethality or that lethality results from a combination of low actin levels and silencing. Silencing of the modified ␤-actin allele is associated with methylation of the CpG island encompassing the ␤-actin promoter. However triggered, methylation of the promoter occurred even though the transgene insertion was in the 3Ј-UTR and hence distal to the promoter. This indicates that maintenance of CpG island methylation status in differentiated cells is not governed autonomously by features built into the island itself.
It is widely accepted that CpG islands are normally kept free from methylation in differentiated cells in the absence of transcription (30,31). It has also been shown that CpG islands are often associated with promoters that are active in early embryonic cells (19). The ␤-actin gene normally follows these rules, but the modified allele does not. The modified ␤-actin allele is actively expressed and remains unmethylated in embryonic cells, although mouse ES cells have high levels of de novo methylation (32). It is switched off upon differentiation, resulting in patchy expression in differentiated ES cells and mouse tissues, which correlates with the onset of methylation of the modified ␤-actin locus. This implies that the CpG island of the modified ␤-actin allele is recognized as normal in the undifferentiated cells, but its methylation-free status is not maintained in differentiated cells.
These observations are compatible with a two-phase definition of a CpG island, an initial establishment phase in early embryonic cells followed by a maintenance phase in differentiated cells. From our modified ␤-actin data, we propose that these two stages are separable and independent and that there must be a persistent signal, which normally maintains CpG islands free from methylation in differentiated cells. In this scenario, embryonic transcription of the wild-type ␤-actin allele results in the production of a signal that identifies the promoter as a CpG island and prevents methylation from occurring in differentiated cells. Following modification of the gene, the promoter continues to function appropriately in ES cells, but the signal defining the CpG island is disrupted. Following differentiation the CpG island is now susceptible to mechanisms inducing gene silencing, as the protective signal has been lost. This hypothesis is supported by the observation that CpG islands become specifically demethylated in embryonal carcinoma cells, whereas differentiated cells such as fibroblasts do not demethylate these sequences or methylate CpG islands with inactive promoters (33). In addition, inactive CpG island promoters remain hypomethylated in human primary fibroblasts and have higher levels of H3K4 dimethylation even in the absence of transcription, implying that chromatin structure has a role in maintaining CpG islands methylation-free (34).
We have not identified the mechanism of methylation in this study, but the trigger must be the presence of the transgene within the 3Ј-UTR. Although transgene silencing is a well described phenomenon, effects of gene targeting on endogenous gene promoters are less well characterized. Because we found methylation regardless of whether we inserted ␤geo, puromycin, or hygromycin gene, the sequence inserted was not the stimulus for methylation; rather, it was where the transgene was inserted into the ␤-actin gene that triggered methylation. Inserting the same construct into other genes, such as Hras1, does not result in methylation of the endogenous allele CpG island. 4 Once triggered, methylation of the promoter occurred even through the transgene insertion was in the 3Ј-UTR. Furthermore methylation occurred only after differentiation of ES cells in vitro and the embryo in vivo. We cannot exclude the possibility of RNA-mediated events; however, we consider this unlikely, as methylation did not occur in undifferentiated ES cells and only the modified allele was silenced with the endogenous allele remaining unaffected. Instead we favor a spreading methylation process (18), presumably nucleated from the transgene inserted into the 3Ј-UTR. It has been demonstrated that methylation can spread from B1 repetitive sequences into a CpG island of the Aprt gene (35,36).
These results indicate that the targeted insertion of a transgene into a gene, even though distant from the gene promoter, may affect the activity of the promoter. The lacZ gene is used widely as a marker for gene expression following gene targeting and gene trapping (37). These results suggest that even when the lacZ gene is targeted to a locus as a single copy, it can disrupt the endogenous promoter. Our data indicate that the same is true for puromycin and hygromycin genes. It follows that care must be taken when interpreting data obtained from inserted marker genes.