Correct regulation of GM-CSF transgenes in splenic T lineage cells
Our group previously created transgenic mice from a 10.5 kb Xho I - Hind III segment of DNA carrying the human GM-CSF gene and all of the elements required for its correct regulation in vivo (Figure 1A) [19]. Transgene expression was assayed in activated spleen cells where the predominant GM-CSF-expressing cells are T cells. We demonstrated that this transgene is expressed in activated spleen cells in an inducible copy number-dependent fashion at levels equivalent to the endogenous mouse GM-CSF gene in 10 out of 11 independent lines of transgenic mice.
Here we reanalysed GM-CSF transgene activity in a variety of T lineage cell populations in seven of the GM-CSF transgenic mouse lines (C183, A127, J253, M268, G203, F201 and D184) carrying from two to 90 copies of the transgene (Figure 1). Freshly isolated spleen cells were stimulated for 8 h with a combination of the phorbol ester phorbol myristate acetate and the calcium ionophore A23187 (PMA/I) to directly activate T cell receptor signalling pathways. Human and mouse GM-CSF expression levels were then measured by ELISA. After correction for gene copy number, transgene expression in all lines was efficiently induced to a level approximately one to three times that of the endogenous mouse GM-CSF gene (Figure 1B). These data confirm that the 10.5 kb transgene contains sufficient information to correctly regulate GM-CSF gene expression in a position-independent and copy-number-dependent manner.
Transcription-dependent silencing of GM-CSF transgenes in previously activated T cells
In order to investigate transgene regulation in a defined cell type we cultured actively dividing T cells derived from the spleen. In order to induce proliferation, spleen cells were activated in culture for 2 days in the presence of the lectin Concanavalin A (ConA), which activates receptor signalling pathways and induces cytokine gene transcription. We confirmed that the GM-CSF gene was, indeed, induced under these conditions (data not shown). These rapidly dividing cells were further cultured for several cell cycles in the presence of IL-2 and in the absence of any cytokine gene-inducing agent for an additional 2 days. This procedure reliably generates cultures of ~98% pure T cells that have undergone blast cell transformation from inactive non-dividing resting T cells to rapidly proliferating T cells (T lymphoblasts).
After the 2 days of culture in IL-2, the T cells had undergone at least five cell divisions since the cessation of the initial episode of activation. Cells were then restimulated with phorbol ester and calcium ionophore (PMA/I). In lines A127, M268 and F201 the transgenes were efficiently induced to levels roughly equivalent to the endogenous mouse GM-CSF gene (Figure 1C). In contrast, transgene activity was greatly reduced for lines J253 and G203 and almost non-existent for lines C183 and D184. This indicated that a profound degree of gene silencing had taken place subsequent to the initial activation of the spleen cells. Note that splenocytes are comprised primarily of quiescent cells that have had no recent history of activation by agents that induce cytokine gene transcription, whereas the cultured T cells were recently activated and had transiently expressed the transgenes. These observations suggest that a single episode of transcriptional activation is, therefore, sufficient to induce heritable transgene silencing in a specific subset of lines that persists for at least several cell cycles. The results are in marked contrast to our parallel studies carried out on several lines that contain 130 kb GM-CSF transgenes which do not undergo silencing in T blast cells in any instances (Mirabella et al, in press). This change in expression pattern is unlikely to be influenced by any change in the proportions of CD4 and CD8 positive T cells during culture because these two populations show essentially identical levels of GM-CSF expression (Mirabella et al, in press).
In summary, we have defined three transgenic lines where regulation appears to be correct under all conditions and four lines that are correctly regulated until exposed to a cycle of transcriptional activation. In order to further explore the basis of this post-activation-specific silencing we selected two correctly regulated lines (A127 and M268) and two lines that are susceptible to induction of silencing (J253 and D184) for further study. In addition to having similar activities in spleen cells, these four lines are also known to be expressed at equivalent inducible levels in peritoneal myeloid cells [21].
Transgene silencing occurs at the transcriptional level
The above findings were verified by real time polymerase chain reaction (PCR) analysis of human and mouse GM-CSF mRNA levels in T cells stimulated for 4 h with PMA/I. As above, these cells had been first activated for 2 days in the presence of ConA and then cultured for an additional 2 days after the removal of ConA. In lines J253 and D184, human GM-CSF mRNA induction was dramatically reduced compared to line A127, whereas mouse GM-CSF mRNA was expressed at similar levels in all lines (Figure 2A). This analysis also indicated that the homologous mouse GM-CSF gene does not undergo silencing in parallel with the human GM-CSF gene in lines J253 and D184.
In order to confirm that these results reflect ongoing transcription in the nucleus, and not just steady state levels of cytoplasmic mRNA, we performed chromatin immunoprecipitation (ChIP) assays to measure levels of the Serine-2-phosphorylated elongating form of RNA polymerase II within the coding region of the human GM-CSF transgenes. Specific ChIP DNA levels were measured before and after stimulation with PMA/I by real time PCR using primer sets located within intron 2 of the GM-CSF gene. No recruitment of the elongating form of RNA polymerase II could be detected in lines D184 and J253, whereas a high level of inducible recruitment was observed in lines A127 (Figure 2B) and M268 (data not shown). This result confirms that silencing takes place at the transcriptional level.
Next, we tried to determine whether transgene silencing was a rapid process, becoming established coincident with the initial transcription initiation, or whether it was a longer process that might even require DNA replication. In order to study short term events we performed a time course of stimulation of splenocytes with PMA/I for up to 11 h (Figure 2C). Over this time period, human GM-CSF mRNA was induced with similar kinetics in both lines A127 and D184 with no major decrease in transgene activity in D184 at the later time points. As GM-CSF mRNA is highly unstable, this suggests that silencing takes more than 11 h to become firmly established. Interestingly, this time course is consistent with the reported kinetics of siRNA-mediated epigenetic gene silencing [18].
Transgene silencing occurs at the level of chromatin structure
The GM-CSF locus contains inducible DNase I hypersensitive sites (DHSs) located at the promoter and enhancer (Figure 3A) [19]. In order to explore epigenetic mechanisms of transgene silencing, DHSs were mapped in cultured T cells within the four chosen lines of mice. In activated T cells from lines A127 and M268 the predicted DHSs formed within the transgenes (Figure 3B). However, the formation of these DHSs was almost abolished in line J253 and completely abolished in line D184. These changes in chromatin structure paralleled the expression data where the transgenes were almost completely silenced in line D184 but incompletely silenced in line J253 (Figures 1C and 2A).
In order to further explore the basis for this chromatin-mediated gene silencing, we performed ChIP assays to measure levels of acetylation and tri-methylation of histone H3 K9 within the enhancer, promoter and coding region of the transgenes. These chromatin modifications have been widely associated with either active (acetylation) or long-term silenced (tri-methylation) regions. The ChIP assays indicated that the enhancer, promoter and coding regions of the silenced J253 transgenes were each heavily modified by trimethylation of histone H3 K9, whereas the active A127 transgenes were not significantly modified (Figure 3C). For an inactive control we also assayed a non-transcribed gene desert region of mouse chromosome 1 (mChrom1) and found that this was equally methylated in both A127 and J253, although not as strongly as the silenced transgenes. ChIP assays also showed that silencing in J253 was accompanied by a decrease in acetylation of K9 in all three regions of the transgenes in comparison with A127 (Figure 3D). Additional ChIP assays were performed on cultured T cells prepared from lines M268 and D184. These assays produced results with values very similar to those shown in Figures 3C and 3D, with D184 transgene silencing being accompanied by decreased acetylation and increased tri-methylation of H3 K9 relative to M268 (data not shown).
Transgene silencing is not mediated by DNA methylation
Changes in gene expression and histone modification patterns are often, but not always, associated with changes in DNA methylation. However, the GM-CSF promoter region contains very few CG sequences that might influence gene expression. Only one CG exists within the -114 to +28 region that constitutes the defined GM-CSF promoter and this is located within the Sp1 site at -70 [22]. In order to investigate whether DNA methylation is involved in the silencing of GM-CSF transgenes we employed methylation-sensitive restriction enzymes and direct DNA hybridisation analysis of genomic DNA to determine the methylation status of the Sp1 site in GM-CSF promoter. In order to measure any changes in the degree of methylation upon silencing, DNA was purified for all four lines from both spleen, which should not be silenced, and ConA-treated cultured T cells (T blasts). Genomic DNA was digested with Hae III in the presence and absence of the methylation-sensitive enzyme Fau I which cleaves the CCCGC sequence at the Sp1 site only if it is not methylated. Hae III alone generates a 175 bp genomic DNA fragment spanning the Sp1 site, whereas Fau I creates a 145 bp sub-fragment. Products were analysed by polyacrylamide gel electrophoresis and filter hybridisation (Figure 4A). Unexpectedly, all samples were equally highly resistant to Fau I digestion of the Sp1 site. This suggested that the Sp1 site was almost fully methylated in both the spleen and the T blast cells, in both the active and the repressed lines, and that there was no change in status upon silencing. We confirmed that the low level of Fau I cleavage was not due to under-digestion because a parallel control analysis of a Fau I site located within a non-methylated CG-island revealed complete cleavage (Figure 4A). However, the significance of methylation of the single CG that exists in the promoter is unclear because DNA methylation does not necessarily interfere with Sp1 binding or function [23]. DNA methylation may not, in fact, play much of a role in the regulation of GM-CSF expression because relatively few CG sequences exist anywhere in the GM-CSF locus. Parallel analyses of 4 Hpa II sites located from 153 to 2091 bp downstream of the transcription start site suggested that similar high levels of DNA methylation existed throughout the GM-CSF gene in T cell DNA prepared from all four transgenic lines and also in primary human T cells (data not shown). Others have similarly shown that the GM-CSF gene is comprised of methylated DNA in T cells [24]. It is also now evident that it is common to find that the bodies of active genes with low CG densities are in fact methylated [25].
Silenced transgenes are associated with inverted convergent repeats
Silencing of GM-CSF transgenes appeared to be transcription-dependent and one potential mechanism of transgene silencing is convergent transcription. In multi-copy transgenes this could lead to the formation of palindromic RNA and siRNAs, which have the potential to direct localised epigenetic silencing [13–16]. Although transgenes typically integrate as head-to-tail copies within multi-copy arrays, silencing could result from convergent transcription of any less commonly encountered tail-to-tail copies of transgenes. In order to determine whether silenced lines do contain convergent gene repeats, we performed a Southern blot hybridization analysis of Afl II-digested DNA from each line, using a probe at the 3' end of the transgene to identify restriction enzyme fragments diagnostic of convergent inverted repeats. Afl II cuts once within the transgene and will generate a 10.5 kb fragment from head-to-tail repeats, a 6 kb band from tail-to-tail repeats and a band of unknown size spanning the site where the 3' end of the transgene array has integrated into the mouse genome (Figure 5). Significantly, the diagnostic 6 kb inverted repeat was present in both of the silenced lines J253 and D184, and absent in A127 and M268 (Figure 5, left hand panel). Each lane also has at least one additional band that most probably represents the Afl II fragment spanning the site of integration. In order to exclude the possibility that the 6 kb Afl II bands represent either fragmented copies of the transgene or site of integration products, we repeated this analysis with three additional restriction enzymes and obtained similar findings, which suggests that the 6 kb Afl II bands are, indeed, inverted repeats (data not shown). Densitometric quantitation of band intensities indicated that each silenced line had just one pair of convergently transcribed transgenes (data not shown).
In order to further confirm that the 6 kb Afl II products are true palindromes, we employed the tactic of denaturing the digested DNA and rapidly renaturing the single stranded products before loading the DNA on a gel for Southern blot hybridisation analysis (Figure 5, right-hand panel). Under these conditions, the hybridization kinetics do not favour the slow reannealing of separate strands of homologous DNA but do favour the rapid formation of hairpin structures from palindromes. This analysis revealed the existence of the expected 3 kb renatured hairpin product of the palindromic 6 kb Afl II fragment only in the silenced lines J253 and D184.
Transgene junctions are transcribed
In order to determine whether there was the potential for convergent transcription to generate palindromic hairpin RNA species, we performed several analyses to determine whether transcription occurred in the vicinity of 3' or 5' junctions between individual copies of transgenes. Transgenic line C42 [21], which does not undergo silencing, is derived from a 130 kb Age I DNA fragment spanning the entire IL-3/GM-CSF locus and including about 35 kb of DNA downstream of the GM-CSF gene. In this line, it is unlikely that the GM-CSF transgene could be subjected to anti-sense transcription arising from adjacent copies. We therefore chose this line as a more appropriate model to search for evidence of sense strand transcription proceeding from the GM-CSF gene and up to and beyond the 3' Hind III site.
For the first RNA assay we performed Northern blot hybridization analysis of total cellular RNA (T), nuclear RNA (N) and cytoplasmic RNA (C) prepared from cultured T cells from line C42 (Figure 6A). We detected the predicted inducible 0.7 kb GM-CSF mRNA transcript in all three fractions of RNA (middle panel, Figure 6A). We also detected an inducible 3' 2.7 kb nuclear RNA species with a 1.0 kb probe encompassing a Bgl II - Hind III fragment comprising the 3' end of the transgene (top panel, Figure 6A). Such a transcript could potentially span the 3' boundary of the transgene and might be expected to exist as a palindrome at sites of inverted repeats. This transcript could potentially represent a read-through transcript from the GM-CSF gene.
We next performed strand-specific reverse transcription of activated C42 T cell RNA, followed by PCR in order to detect transcripts in two adjacent regions just inside the 3' Hind III site (amplicons A and B, Figure 6B). This revealed the presence of sense strand, but not anti-sense strand transcripts, within 40 bp of the Hind III site which probably proceed even further past this point. If the same predicted pattern of transcription occurs in lines J253 and D184, then this would, indeed, generate palindromic hairpin RNA species that could direct gene silencing.
We then examined lines A127, J253 and D184 for the presence of this 3' species of RNA. Curiously, in line A127, in contrast to line C42, we were not able to detect sense but did detect anti-sense strand transcripts downstream of the GM-CSF gene with primer set B (Figure 6C). These transcripts 3' of the gene could potentially arise from anti-sense transcripts originating from the GM-CSF enhancer located within the next downstream copy of the transgene. In support of this view, our group has detected RNA polymerase II in association with the GM-CSF enhancer in activated T cells from transgenic line C42 in ChIP assays (Mirabella et al, in press). We then examined line A127 in order to determine whether there are anti-sense transcripts potentially originating from the GM-CSF enhancer that can traverse the 5' boundary of the transgenes. This could also generate palindromic RNA at inverted repeats. We again used strand-specific reverse transcription and PCR using primers either upstream (B) or downstream (C) of the Xho I site that defines the 5' boundary of each transgene copy (Figure 6C). Note that probe B is, in fact, downstream of the GM-CSF gene and is designed to detect transcription within adjacent copies. We also employed primers that span the junction between transgenes (J). As before, in this analysis we detected anti-sense, but not sense transcripts with both the 5' and the junction primer sets, and also with additional primers closer to the enhancer (data not shown). This suggests that the enhancer does, indeed, direct transcription into neighbouring copies of the transgene. We interpret these observations as an indication that 5' anti-sense transcription from the enhancer can suppress the competing 3' sense strand transcription from the gene in head-to-tail repeats of the transgene. However, in the case of tail-to-tail repeats, there would be no such suppression because there is no adjacent downstream copy of the GM-CSF enhancer. These transcripts were absent prior to the induction of the expression of the gene with PMA and calcium ionophore, indicating that they are inducible (data not shown). We have, however, been unable to detect any of these transcripts in the silenced lines J253 and D184, very likely due to the fact that GM-CSF transcription in these lines is shut down after the act of epigenetic silencing (data not shown).
Taken together, these data suggest that a single inverted copy of a transgene, embedded within a tandem array, does indeed have the potential to generate palindromic RNAs at both boundaries of the transgene, which could be the trigger for epigenetic silencing.
Small interfering (si) RNAs are not detected in silenced transgenes
It has been well documented that double stranded RNAs can lead to production of small RNAs after their processing. In order to search directly for the presence of GM-CSF siRNAs in lines J253 and D184 we have performed an extensive series of Northern blot analyses. One such example is shown in Figure 6D. In this analysis we have assayed either total cellular RNA, or a sample enriched for low molecular weight RNA, from previously activated T cells prepared from lines A127, D184 and J253. Although we can easily detect other specific small RNA species, we were unable to detect any GM-CSF siRNAs (Figure 6D). We have now completed many exhaustive attempts at detecting a GM-CSF siRNA in the silenced lines using various T cell preparations at different stages of blast cell transformation, using a range of probes and have failed to find such an RNA anywhere in the GM-CSF locus (data not shown).
In mammalian cells, siRNAs are thought to be able to direct epigenetic gene silencing by recruitment of Argonaute 1 or 2 [14, 15]. However, we have also been unable to detect any recruitment of either Argonaute 1 or 2 in ChIP assays (data not shown). Hence, although it is clear that GM-CSF transgene silencing is occurring at the level of transcription, and is an epigenetic phenomenon, the precise mechanism of silencing remains unknown.