ABSTRACT
There are now many mammalian examples in which single cell assays of transgene activity have revealed variegated patterns of expression. We have previously reported that transgenes in which globin regulatory elements drive the lacZ reporter gene exhibit variegated expression patterns in mouse erythrocytes, with transgene activity detectable in only a sub-population of circulating erythroid cells. In order to elucidate the molecular mechanism responsible for variegated expression in this system, we have compared the chromatin structure and methylation status of the transgene locus in expressing and non-expressing populations of erythrocytes. We find that there is a difference in the chromatin conformation of the transgene locus between the two states. Relative to active transgenes, transgene loci which have been silenced exhibit a reduced sensitivity to general digestion by DNase I, as well as a failure to establish a transgene-specific DNase I hypersensitive site, suggesting that silenced transgenes are situated within less accessible chromatin structures. Surprisingly, the restrictive chromatin structure observed at silenced transgene loci did not correlate with increased methylation, with transgenes from both active and inactive loci appearing largely unmethylated following analysis with methylation-sensitive restriction enzymes and by sequencing PCR products derived from bisulphite-converted genomic DNA.
A gene is said to exhibit a variegated or mosaic pattern of expression when it is active in some but not all cells of a particular cell type. The best characterized examples of variegation occur in Drosophila, where mosaic expression of genes within the eye disk cells results in mottled patterns of eye pigmentation. This variegated expression results from chromosomal rearrangements which juxtapose a normally euchromatic gene with regions of constitutive heterochromatin (1 ). Cytological observations have revealed a correlation between the transcriptional state of a variegating gene and the degree to which it is compacted within the surrounding chromatin. This has led to the hypothesis that the variegated expression pattern is caused by differences between cells in the extent of heterochromatin spreading (2 ). Several mutations which were observed to suppress variegation in this system identified genetic loci encoding structural components of heterochromatin (reviewed in 3 ).
The frequency with which variegated or mosaic patterns of gene expression occur in mammalian systems may have been underestimated. Variegated expression patterns can only be detected when gene activity is assayed in individual cells or in clones derived from individual cells. Since assays of gene expression in mammals are usually performed on lysates obtained from whole cell populations and the observed level of activity is generally assumed to reflect an equivalent rate of expression in all cells, variegated patterns of expression may go unnoticed. However, there are now several studies in which the ability to assay expression in individual cells or clones has revealed variegated expression of either endogenous genes (4 ) or transgenes (5 -13 , reviewed in 14 ) in mammalian systems.
In plant and animal tissues a correlation has frequently been observed between gene inactivity and hypermethylation of cytosine residues (15 ,16 ), leading to the suggestion that methylation may be the cause of transcriptional silencing in these instances. The finding that in vitro methylated DNA adopts a different chromatin structure to untreated DNA following stable integration into a host genome (17 ) suggests that the silencing effect of methylation upon gene activity may be mediated by a change in chromatin structure from an active to a repressive state, although what this means at the molecular level is still largely unknown. Methylation has also been implicated as the mechanism of gene silencing responsible for variegated expression patterns in plants and animals, with several studies reporting a correlation between the extent of variegation and the degree of transgene methylation (6 ,18 -20 ). However, the exact relationship between DNA methylation, chromatin structure and gene expression still remains to be clearly elucidated.
We have previously reported that transgenes in which the Escherichia coli lacZ gene is driven by human globin regulatory elements are expressed in a variegated manner in mouse erythroid cells (21 ,22 ). Because these transgenes express the cell-autonomous lacZ reporter in the erythroid tissue, histochemical staining of circulating erythrocytes with X-gal allows transgene activity to be analysed in individual cells. The proportion of erythroid cells expressing the transgene was consistent within a transgenic line, but varied between different lines bearing the same construct in a manner which was independent of transgene copy number. This copy number independence suggested that the proportion of cells in which the transgene is active is primarily determined by influences specific to the site of integration.
In the work presented here, we have attempted to identify the molecular mechanism by which transgene expression is silenced in some erythroid cells and not others. We have compared chromatin structure and methylation status of the transgene between cells in which the transgene is active and cells in which it has been silenced. We find that there is a clear difference in the chromatin conformation of the transgene between the two states; the transgene is situated within a relatively open chromatin environment in cells in which it is active, while in non-expressing erythrocytes, the chromatin structure of the transgene locus appears to be more restricted. Surprisingly, this difference in the chromatin conformation of the transgene locus between the two states does not reflect a difference in the extent of transgene methylation. These findings suggest that transgene silencing is mediated by locus-specific epigenetic events which do not involve methylation.
The production of the DM1 transgenic mouse lines used in this study has been described previously (21 ). Briefly, the DM1 construct contains a 163 bp promoter (comprised of elements from both the human [zeta]- and [gamma]-globin promoters) upstream of a 3.3 kb SDK-lacZpA cassette and a 4.2 kb fragment containing the [alpha]HS-40 DNase I hypersensitive site (Fig. 1 ). The DM1 transgene was excised from vector sequences by digestion with KpnI and NotI and purified by agarose gel electrophoresis. Fertilized eggs of the outbred PO mouse strain were microinjected with the DM1 construct according to standard techniques (23 ). Founder mice were identified by Southern blotting of tail DNA and hemizygous transgenic lines were established. Copy number was determined by Southern blotting of tail DNA from G1 mice. Erythroid expression at 12.5 d.p.c. was determined by bleeding individual embryos into PBS, fixing whole blood cells in glutaraldehyde and staining with X-gal as described previously (21 ). All data presented in this study were obtained from established lines.
Nuclei were prepared as described by Higgs et al. (24 ) from the pooled blood of five individual 12.5 d.p.c. transgenic embryos. For the DNase I hypersensitivity assay, nuclei were digested with DNase I (Boehringer Mannheim) at 6.25 * 10-2 U/ml for 0-50 min at 37oC. At each time point, digestion was stopped by the addition of EDTA to 10 mM and DNA purified by phenol/chloroform extraction following digestion with proteinase K. DNA was then digested with BamHI and resolved on a 1% agarose gel. After Southern blotting, the membrane was probed with the 4.2 kb BamHI fragment of the DM1 transgene containing [alpha]HS-40 (Fig. 1 ). For the general DNase I sensitivity assay, nuclei were digested with DNase I at 0, 0.1, 0.2, 0.5, 1.0 and 2.0 U/ml for 15 min at 37oC. Reactions were stopped and DNA purified as above. After digesting with PstI, Southern analysis was performed using the 3.3 kb SDK-lacZpA cassette as probe. To control for equivalent DNase I digestion, the membrane was rehybridized with a 2.1 kb HinfI-SacI fragment containing the mouse [alpha]-globin gene and 5' flanking sequence. All probes were labelled by random priming using [[alpha]-32P]dCTP and Klenow DNA polymerase.
DNA was recovered from circulating erythroid cells of 12.5 d.p.c. transgenic embryos. Aliquots of 2 [mu]g DNA were digested with either BamHI alone or BamHI together with one of the methylation-sensitive restriction enzymes SmaI, HpaII or HinPI. To monitor complete digestion, in a separate set of reactions an equivalent amount of wild-type mouse DNA spiked with the DM1 plasmid was digested under the same conditions (not shown). The resulting fragments were analysed by Southern blotting after separation in 1.2% agarose gels. Membranes were probed with either the 4.2 kb BamHI fragment of the DM1 transgene containing [alpha]HS-40 or the 3.3 kb SDK-lacZpA cassette. Probes were labelled as described above.
Genomic DNA was purified from circulating erythroid cells obtained from 12.5 d.p.c. embryos of the transgenic lines 27/5A and 16/7B. Aliquots of 15 [mu]g DNA were digested with EcoRI and bisulphite conversion carried out as described previously (25 ), with removal of free bisulphite achieved by passage through PCR Spinclean columns (Progen Industries) according to the manufacturer's instructions. The PCR primers DFG-1 (5'-GGGAGATTTTT
Transgenic mouse lines were generated bearing the DM1 construct shown in Figure 1 . In this construct, a hybrid of the human [zeta]- and [gamma]-globin promoters drives expression of the E.coli lacZ reporter gene in the presence of the major tissue-specific enhancer element of the human [alpha]-like globin locus, [alpha]HS-40 (24 ). Use of the cell-autonomous lacZ reporter in erythrocytes allows transgene activity to be assayed in individual cells by staining whole blood with the chromogenic substrate 5-bromo-4-chloro- 3-indolyl [beta]-D-galactopyranoside (X-gal). We have previously reported that when this cell-by-cell assay was performed on erythrocytes from mice bearing the DM1 construct, in most lines only a sub-population of erythroid cells exhibit [beta]-galactosidase activity (21 ). Since it was theoretically possible that this heterocellular pattern was the result of somatic mutations within the lacZ gene in a sub-population of erythroid cells, PCR was used to amplify the lacZ coding region from primitive erythrocytes of lines in which the majority (98%) of cells were not expressing the transgene. After cloning the PCR product into a bacterial expression vector and assaying for [beta]-galactosidase activity, it was found that the lacZ gene was fully functional (data not shown). This finding suggests that the inactivity of the transgene locus in a sub-population of cells reflects epigenetic rather than genetic variation. To investigate the epigenetic mechanism responsible for variegated expression of the DM1 transgene, three transgenic lines (27/5A, 22/7A and 16/7B) were selected for analysis, since we had previously found that at 12.5 d.p.c. a very high proportion of erythrocytes express the transgene in 27/5A, while in both 22/7A and 16/7B the transgene is silent in the majority of erythroid cells. Analysis was performed in erythrocytes purified from 12.5 d.p.c. transgenic embryos, since at this stage of development, nucleated primitive cells represent ~95% of circulating erythrocytes (26 ). Consistent with previous data, when peripheral blood from 12.5 d.p.c. embryos of these three lines was stained with X-gal, line 27/5A expressed [beta]-galactosidase in 95% of erythrocytes, while lines 22/7A and 16/7B expressed the transgene in only 8 and 2% of erythroid cells respectively (Table 1 ). Erythrocytes isolated from 12.5 d.p.c. transgenic embryos of these lines therefore approximate pure populations of transgene-active (27/5A) and transgene-inactive (22/7A and 16/7B) erythroid cells. The transgene copy numbers of these three lines are also shown in Table 1 .
DNase I hypersensitivity mapping was performed to assay for the presence of the transgene-specific [alpha]HS-40 hypersensitive site in these three lines. Nuclei were isolated from 12.5 d.p.c. erythrocytes of the lines 27/5A, 22/7A and 16/7B and subjected to DNase I digestion for increasing periods of time. DNA was then purified and digested with BamHI to release a 4.2 kb parent fragment which is reduced by DNase I cleavage at the [alpha]HS-40 hypersensitive site to produce sub-bands of ~3 and 1 kb (27 ). The [alpha]HS-40 hypersensitive site could be detected in erythrocytes from line 27/5A, where the majority of cells express the transgene, but was not present in erythrocytes from 22/7A and 16/7B, where the transgene has been silenced in most cells (Fig. 2 ). Since the [alpha]HS-40 DNase I hypersensitive site is associated with the binding of a variety of transcription factors within a 350 bp core element (27 ), the failure of this site to become established in the transgenes of inactive erythroid cells suggests that access of these transcription factors to their cognate recognition sequences within the transgene has been prevented by a restrictive chromatin structure.
Since the silencing of transgenes and the adoption of an inactive chromatin state has frequently been found to correlate with increased transgene methylation, we compared the methylation status of the DM1 transgene in the expressing and non-expressing erythrocytes. Transgene methylation was initially investigated by digesting with methylation-sensitive restriction enzymes. Chromosomal DNA from 12.5 d.p.c. erythrocytes of lines 27/5A, 22/7A and 16/7B was digested with BamHI to release a 4.2 kb fragment containing [alpha]HS-40 and a 3.5 kb fragment containing the promoter/lacZ. DNA was then digested with one of the methylation-sensitive restriction enzymes SmaI, HpaII or HinPI. The resulting products were analysed by Southern blotting and hybridization with the BamHI parent fragments as probes (Fig. 4 ). Surprisingly, in all three lines these methylation-sensitive restriction enzymes digested the [alpha]HS-40 parent band almost to completion and no difference could be detected between these lines in the pattern of fragments generated (Fig. 4 a). Although some copies of the [alpha]HS-40 fragment had remained uncut following digestion with SmaI, the proportion of digested to undigested fragments in this lane is the same in all three lines. Similarly, a faint 2.0 kb fragment resulting from incomplete HpaII digestion was observed in the DNA of both transgene-expressing (27/5A) and silenced erythrocytes (22/7A and 16/7B). There is therefore no detectable difference in the frequency with which these restriction sites are methylated between active and inactive transgenes.
Transgenes in which globin regulatory elements drive expression of lacZ are expressed in a heterocellular or variegated pattern in mouse erythrocytes. In the work presented here, a clear difference was observed between the chromatin structure of active and inactive transgenes. Silenced loci are characterized by an absence of the [alpha]HS-40 hypersensitive site and a reduced general sensitivity to DNase I relative to that of active loci. The finding that variegation of this transgene is associated with a closed chromatin configuration is consistent with other studies which have compared the chromatin structure of variegating transgenes between expressing and non-expressing populations. For human CD2 transgenes which exhibit variegated expression in mouse thymocytes, it was found that a transgene-specific DNase I hypersensitive site was present in expressing thymocytes but could not be detected in those cells in which the transgene had been silenced (11 ,13 ). Similarly, it was found that the variegated silencing of a hsp26/white transgene in Drosophila was accompanied by packaging within a more regular nucleosome array and decreased sensitivity to micrococcal nuclease (30 ).
Despite the observed difference in chromatin structure, we were unable to detect any difference in the extent of transgene methylation between active and silenced transgene loci. Transgenes from both active and inactive erythrocytes were equally susceptible to digestion by a variety of methylation-sensitive restriction enzymes. Further, analysis of methylation by sequencing PCR products obtained from bisulphite-converted genomic templates revealed no consistent difference in either the overall density of 5-methylcytosine within the promoter or in the methylation of any specific cytosine residue between active and inactive transgenes. While we have not analysed the methylation of every cytosine within the transgene and therefore cannot definitively rule out differential methylation at an as yet undetected specific site, these results strongly argue against methylation as the cause of variegated transgene silencing in this system.
While there are many reported examples in which the inactivation of a gene has been found to correlate with hypermethylation of its regulatory elements (15 ,16 ), there is still considerable controversy as to whether gene methylation is the primary inactivating event or whether it is a secondary modification which takes place on genes which have already been silenced by a different mechanism. Where methylation has been proposed as the cause of gene inactivation, the mechanism by which this is achieved remains unclear. The existence of transcription factors whose binding affinity is reduced by the methylation of cytosine residues within their cognate recognition sites (31 -36 ) suggests that methylation can interfere with gene expression directly. However, other transcription factors show no difference in their ability to bind methylated and unmethylated target sequences (37 ) and the number of known methylation-sensitive transcription factors cannot account for all the reported observations of gene silencing associated with methylation. The observation that in vitro methylated DNA adopts a chromatin structure that is different from that of untreated DNA upon genomic integration (17 ) suggests that the repressive influence of methylation on gene expression can also be indirect and involve a change in chromatin structure. Consistent with these findings, it was reported that methylation alone was not sufficient for the transcriptional silencing of a HSV TK transgene but that repression was also dependent on the formation of chromatin within the recipient cell (38 ,39 ).
A number of studies have investigated the role of methylation in establishing variegated patterns of expression. The proportion of neural tube cells expressing a mouse hsp68/lacZ transgene was found to be inversely proportional to the extent of transgene methylation (6 ) and the variegated expression of a maize A1 cDNA transgene in the flower of Petunia hybrida correlated both with high copy number and transgene methylation (18 ). Similarly, two variegating epi-alleles of the maize P-rr gene were found to be hypermethylated relative to the non-variegating parent allele (19 ). Kennedy et al. (40 ) found that methylation did not correlate with variegated silencing of a human [alpha]1 antitrypsin/lacZ transgene in mouse hepatocytes, although analysis was performed with only a single methylation-sensitive restriction enzyme. Our finding that the inactivity of the variegating DM1 transgene correlates with a less accessible chromatin structure but not increased methylation contrasts with these earlier studies and is consistent with the proposal that the activity of a gene is determined primarily by its location within a permissive or restrictive chromatin domain and is not dependent on methylation status per se. For some genes it may be that methylation of one or more cytosine residues is responsible for initiating the repressive chromatin structure. In other cases, different factors, including the proximity of the locus to spreading heterochromatin or the presence within the locus of multiple repeated copies of a monomeric unit (41 ), may trigger the formation of a repressive chromatin structure in a methylation-independent manner. The primary importance of chromatin structure in determining the activity of a variegating locus is consistent with the observation of variegated expression patterns in Drosophila, where methylation of genomic DNA does not occur.
This work was supported by a grant from the National Health and Medical Research Council of Australia to E.W. D.G. and H.S. are both supported by Australian Postgraduate Awards. We thank Hugh Morgan for his contribution to the work, David Martin for the DM1 clone, Susan Clark for advice on bisulphite treatment of genomic DNA and Merlin Crossley for helpful comments on the manuscript.
*To whom correspondence should be addressed. Tel: +61 2 9351 2549; Fax: +61 2 9351 4726; Email: e.whitelaw@biochem.usyd.edu.au
REFERENCES
Return