Transcription rate of noncoding roX1 RNA controls local spreading of the Drosophila MSL chromatin remodeling complex

Altering rates of roX1 transcription controls local MSL spreading

The idea that a high ratio of MSL protein subunits to roX RNA promotes local spreading from sites of roX transcription arose from experiments where the level of MSL subunits or the number of roX genes was altered (Park et al. 2002; Oh et al. 2003). We asked whether the rate of roX transcription might also affect local spreading. We noticed a large difference in the cis spreading behavior of two different roX1 transgenes. GMroX1 carries the complete roX1 gene transcribed from its native promoter and supports robust local spreading when placed on an autosome (Kelley et al. 1999; Park et al. 2002). By contrast, H83roX1 does not support local spreading (Fig. 1A B). It is composed of a nearly full-length 3.4 kb roX1 cDNA transcribed from a strong, constitutive hsp83 promoter (Meller et al. 2000; Stuckenholz et al. 2003). The starkly different spreading behaviors of these two transgenes might be due to sequences at the 5′ end of roX1 RNA, splicing of the tiny intron, a tightly linked spreading initiation site outside the transcription unit, differences between the roX1 and hsp83 promoters, or chance juxtaposition with novel sequences flanking the transgene insertion sites.

We tested these possibilities by moving the roX1 cDNA into a plasmid where it is transcribed from a minimal promoter linked to UAS binding sites for the yeast GAL4 transcription factor (Fig. 1A) (Brand and Perrimon 1993). Males carrying this UASroX1 construct at random autosomal sites were crossed to roX1 roX2 double mutant females (Meller and Rattner 2002) so that the only source of roX RNA for the sons came from the UASroX1 transgenes. All such males had poor viability suggesting that the amount of roX RNA produced was inadequate to support dosage compensation (Table 1). When the polytene chromosomes of these dying males were examined, the X chromosome had a collapsed morphology and very low levels of MSL proteins bound (Fig. 1C) similar to that seen in roX1 roX2 double mutants (Meller and Rattner 2002). However, the autosomal region around the transgene insertion site (within 1-2 Mbp) was heavily coated with MSL complex in all nuclei (Fig. 1C). The same result was obtained for thirteen different insertion sites assayed (data not shown). This demonstrated that no sequence outside the roX1 transcription unit was needed to promote local spreading. We conclude that under conditions of normal MSL protein levels, but exceptionally low roX1 transcription, the small amount of mature MSL complex assembled is predominantly sequestered near the site of roX transcription with little reaching the X chromosome.

We next introduced a source of GAL4 into the UASroX1 stocks. This dramatically increased male viability (Table 1) and roX1 transcription (Fig. 2H). Increased roX1 transcription had a profound effect on the distribution of MSL complex. The X chromosome was painted with MSL complex similar to what is seen in wild type males (Fig. 2B). The most striking observation was that introducing GAL4 abolished local MSL spreading near the autosomal UASroX1 transgene in all eight lines tested (Fig. 2A B and data not shown). This indicates that vigorous roX1 transcription prevents local spreading of the MSL complex. The fact that a single UASroX1 transgene can both support local MSL spreading or not shows that the strength of roX1 transcription is the critical factor controlling MSL spreading. High constitutive transcription from the hsp83 promoter most likely explains why the H83roX1 transgene failed to support local MSL spreading even when assayed at multiple locations.

One puzzling observation was the complete absence of MSL complex bound to the autosomal UASroX1 gene under conditions of abnormally high transcription (Fig. 2B). The MSL proteins bind a well-defined region within both the roX1 and roX2 genes that form male-specific DNase I hypersensitive sites (DHS) (Kageyama et al. 2001; Park et al. 2003). These act as complex enhancers that stimulate transcription in males (when MSL is bound), and silence the roX genes in females (no MSL bound) (Bai et al. 2004). We tested whether there was any binding defect in the DHS sequence of the UASroX1 transgene by crossing it into a roX1^ex6 roX2⁺ stock that does not allow local MSL spreading from autosomal roX transgenes (Park et al. 2002). A single MSL band was seen at the site of the transgene, demonstrating that the DHS was competent to bind MSL in the absence of GAL4 (Fig. 2C). The important result was that introduction of GAL4 resulted in loss of the MSL signal from the autosomal UASroX1 transgene (Fig. 2D). We asked if the GAL4 protein had some unanticipated ability to interfere with DHS-MSL interaction. We easily detected a band at the roX1 transgene in H83roX1 males expressing of GAL4 (Fig. 2G). In several UASroX1 transgenic lines we noticed that when GAL4 was present, the banded polytene chromosome displays a distinctively puffed morphology surrounding the transgene (Fig. 2E, F) consistent with the very high levels of transcription detected by Northerns (Fig. 2H). We conclude that GAL4 does not interact with either the DHS sequence or the MSL complex, but rather it is unusually high levels of transcription across roX1 that block the DHS-MSL interaction.

We examined how stable the ectopic autosomal MSL pattern was to changes in local roX1 transcription. We constructed flies that carried three transgenes on the same chromosome; UASroX1, actin-GAL4, and actin-GAL80^ts. At 18° the GAL80 binds and inactivates GAL4 bound to the UAS resulting in very low UASroX1 transcription (Fig. 3A lane 6) (Lue et al. 1987; McGuire et al. 2003). Under these conditions, MSL complex spreads extensively from the roX1 transgene again demonstrating that GAL4 bound to the transgene does not affect MSL binding or spreading as long as transcription remains low. At the nonpermissive temperature of 30° the GAL80 releases GAL4 to turn up roX1 transcription (Fig. 3A lanes 7-10). We observed that the newly synthesized roX1 RNA rapidly delivered MSL complex to the X chromosome in trans, at the same time that local MSL spreading around the UASroX1 transgene collapsed (Fig. 3B-E). We conclude that autosomal cis MSL spreading requires continuous replenishment from locally assembled MSL complex. It seemed unlikely that the heat treatment alone could account for the loss of MSL binding around the autosomal roX1 transgene because binding to the X was not hindered. To directly test this alternative, we heat shocked roX1 roX2/Y;[w⁺ GMroX1}/+ males and found that the local spreading around the roX1 transgene was unaltered (Fig. 3F). This confirms that the loss of local MSL spreading seen in the UASroX1 GAL4 GAL80^ts flies was due to increase in roX1 transcription, not the temperature shift.

A 1.2 kb roX1 minigene retains function

Finding that local spreading was an intrinsic property of the MSL complex that did not require any flanking DNA sequences, we set out to test whether any specific region of roX1 RNA might contribute to MSL activity or localization. Although previous analysis showed that the central region of roX1 was not essential for dosage compensation, there were indications that functionally redundant elements might be present that obscured the phenotypes of deletion mutants (Stuckenholz et al. 2003). We constructed a 1.2 kb H83roX1Δ39 minigene (Fig. 4B) modeled after roX1^ex40A (Deng et al. 2005) in an effort to remove potential redundant motifs. This H83roX1Δ39 minigene rescued roX1 roX2 double mutant male viability when transcribed from the Hsp83 promoter and supported strong MSL painting of the X when it was the only source of roX RNA (Fig. 4C and E). No MSL complex was ever seen bound to this transgene, presumably because it lacked the DHS element. More importantly, we never saw any MSL spreading around the transgene insertion sites (Fig.4E and 5B). Vigorous transcription is most likely responsible as described above for the H83roX1 and UASroX1 transgenes. The possibility that the Δ39 minigene is instead missing an RNA domain required for spreading will be addressed below. The 1.2 kb Δ39 RNA accumulated in males but not females (Fig. 4D lanes 4-7) despite being transcribed from a constitutive promoter. The same male-specific accumulation has been reported for full-length roX1 RNA and attributed to rapid turnover of female transcripts that fail to assemble into complexes with the MSL proteins (Meller et al. 2000).

Deletion mutagenesis of the extreme 3′ end of roX1

The genomes of twelve Drosophila species covering approximately 40 Myr of evolution have been sequenced (Clark et al. 2007). We found that each contains a homolog of roX1 and roX2 as has been independently reported (Park et al. 2007). A simple BLAST search failed in more distantly related species forcing us to search with adjacent protein coding gene sequences in the hope that the roX genes had remained in the same syntenic arrangement over time. A combination of computational and manual methods produced a tentative alignment (Fig. S1 and S2). Being free of the constraints of a translational reading frame, the roX genes have accumulated a huge number of indels (insertions/deletions) in addition to simple nucleotide substitutions. This, along with simple repeats and extensive runs of polyA (below) partially defeated computational alignment methods necessitating manual inspection to identify meaningful islands of homology separated by variable spans of seemingly unrelated sequence.

We focused on the 600 nucleotides at the 3′ end of roX1 that previous analysis indicated was important (Stuckenholz et al. 2003). Several features were apparent from the species alignment (Fig. 4A). First, the large stem-loop (SL1) analyzed earlier was well conserved in most species except virilis, mojavensis, and grimshawi. A second predicted stem loop (SL2) followed shortly afterwards, which shows better base pairing potential in non-melanogaster species. We noticed that rather than the single copy of the “roXbox” element (RB) previously described (Franke and Baker 1999; Stuckenholz et al. 2003) we found three imperfect copies and a fourth in the inverse orientation in all species (Fig. 4A S1). Any of the three RBs could potentially base pair with the upstream inverted copy to form three alternative dsRNA stems with loops of variable length, although RB1 is closest in both proximity and sequence complemenarity. The genomic sequence suggests that pseudoobscura and persimilis both have four additional copies of the RB (Fig. 4A S1 S3). After these sequences are accounted for, the remainder of the region is remarkably rich in runs of (A)_n although the exact length and locations of these varies widely from one species to another (Fig. S1). We next turned to the much smaller roX2 sequences. Again, we noticed that rather than one copy of the RB noted earlier (Franke and Baker 1999), three copies were present in all species with additional copies in a few species (Fig. S2 and S4). We also found inverted copies lying near some RBs suggesting that these might fold into dsRNA stem-loops (Fig. S4). Again, once the RBs and inverse RB sequences were set aside, the remainder of the sequence was poorly conserved and full of short runs of (A)_n.

We deleted each of these elements as well as one upstream region (Region 1) from the H83roX1Δ39 minigene and assayed their ability to support dosage compensation transgenic male flies. Our initial analysis using standard P element transformation provided suggestive results, but the wide variation from one random insertion site to another made direct comparisons between mutants difficult. Therefore we inserted all of our deletion mutants into the same genomic location using the ϕC31 site-specific integration system (Bischof et al. 2007). After sampling several attB integration sites, we chose VK11 located at 40E at the base of chromosome 2L (Venken et al. 2006) because a nearly full-length 3.4 kb roX1 cDNA at this site gave almost complete male rescue and the 1.2 kb H83roX1Δ39 minigene rescued 47% male viability when the transgenes provided the only source of roX RNA (Fig. 4C). When all the other deletion mutants were integrated at the same site most continued to rescue male viability near 50%. However, removing SL1 nearly destroyed function, and loss of SL2 reduced activity to a lesser degree (Fig. 4C). Loss of the very A rich region also reduced activity. Surprisingly, removing any one copy of the RB element or the complementary IRB had no effect on activity. However, removal of all three RB copies or the entire interval containing them completely destroyed activity. Northern analysis showed that roX1 RNAs lacking any RB element were unstable (Fig. 4D lanes 8-10).

The D. pseudoobscura roX1 sequence supports enhanced local MSL spreading

We constructed a hybrid roX1 minigene, H83roX1Δ39pseudo, in which the first 600 nt were derived from the 5′ end of the D. melanogaster roX1 gene and the last 600 nt were derived from the 3′ end of the D. pseudoobscura roX1 gene (Fig. 4C). Despite the overall poor conservation of primary sequence, many indels between the two distantly related species, and a weak SL2-like element, the hybrid minigene rescued roX1 roX2 male viability to the same degree as the all melanogaster construct (Fig. 4C), and produced abundant RNA (Fig. 4D lane 15). The MSL complex containing only the hybrid H83roX1Δ39pseudo RNA painted the male X in a normal pattern demonstrating that the many sequence differences were inconsequential (Fig. 5A). However, we did notice a striking difference between the melanogaster Δ39 minigene and the pseudoobscura hybrid. The latter supported MSL spreading around the autosomal insertion sites tested despite being driven by the strong hsp83 promoter (Fig. 5C-E) while none of the melanogaster constructs did (Fig. 5B). We cannot be certain which of the many sequence differences are responsible for this enhanced spreading behavior, but the most obvious difference is that the pseudoobscura sequence used here carries six copies of RB at its 3′ end rather than three.

MSL binding to DNaseI hypersensitive site

A sequence of approximately 220 bp forms a male-specific DNase I hypersensitive site (DHS) near the middle of the roX1 gene which is the primary site bound by the MSL complex (Kageyama et al. 2001; Park et al. 2003). This site appears to act as a complex transcriptional control element causing silencing in females and activation in males (Bai et al. 2004; Bai et al. 2007). MSL complex clearly binds the DHS sequence in vivo when the gene is not transcribed (Kageyama et al. 2001; Park et al. 2002; Park et al. 2003), during very low transcription (Fig. 1C), and even when transcribed from the hsp83 promoter (Fig. 1B). However, we could not detect the MSL complex bound to the UASroX1 transgene when GAL4 drives transcription much higher than the wild type rate (Fig. 2B and D). This might reveal a negative feedback mechanism that places an upper limit on roX1 expression. In wild type males, as MSL proteins successfully stimulate roX transcription, they might reach a point where they are ejected from the roX gene by a high density of RNA polymerase molecules transiting the DHS. This would ensure that the MSL proteins could drive roX transcription to the correct level, but no higher. This might explain why the MSL-dependent enhancer in located inside the roX1 transcription unit. Although this arrangement results in the DHS sequence being carried along in the mature roX1 RNA, it seems to have no function at the RNA level (Stuckenholz et al. 2003). In our artificial situation roX transcription no longer depends upon MSL proteins, so extraordinarily high roX1 synthesis continues unabated.

Short repeats at the 3′ end of both roX genes

Although an initial comparison between roX1 and roX2 failed to detect any obvious similarities at the primary sequence level (Amrein and Axel 1997), later inspection revealed a single 30 nt element (here called the roXbox, RB) at the 3′ ends of both roX1 and roX2 (Franke and Baker 1999). Deletion of this element did not result in any significant phenotype (Stuckenholz et al. 2003). The availability of 12 sequenced Drosophila species genomes allowed us to determine which segments of roX1 were best conserved and potentially functionally important. The resulting sequence alignment revealed that the RB was actually present in at least three to seven copies in the roX1 gene of different species, and at least three copies were also present at the 3′ end of roX2 (Fig. S1-4). Bolstering the notion that these elements are significant, additional conserved copies were found in the inverse orientation that might form dsRNA secondary structures (this work and Park et al 2007).

Removing the central 70% of the 3.7 kb roX1 sequence only reduces in vivo activity by half, and is fully consistent with prior work (Stuckenholz et al. 2003; Deng et al. 2005). Deletion analysis of the mammalian Xist gene came to a similar conclusion (Wutz et al. 2002). Small deletions in the 1.2 kb H83roX1Δ39 minigene display more drastic phenotypes than reported for larger deletions in the full length gene (Stuckenholz et al. 2003). This is most likely due to the removal of partially redundant domains in the middle of roX1 and our use of a single chromosomal integration site to assay all mutants. Using this system showed that removing any one copy of RB had little effect. However, removing the interval containing all three copies produced an unstable RNA that could no longer support male viability. A similar loss of activity was found when just the three RB elements were deleted precisely. The predicted base pairing between IRB and any copy of RB is not essential for dosage compensation. However, its conservation over time suggests IRB plays some role not measured in this assay or another unidentified sequence can substitute for it.

One might have imagined that the extensive divergence of roX1 primary sequence has been accompanied by equally dramatic coevolution of the MSL subunits. Indeed, while MOF, MSL3 and MLE are very similar across fly species, MSL1 and MSL2 have diverged extensively outside small domains known to contact other protein subunits (unpublished observations). This might result in incompatible protein:RNA combinations if components from different species were mixed. We constructed a hybrid roX1 minigene whose first 600 nt came from the 5′ end of the D. melanogaster homolog and the last 600 nt from highly diverged D. pseudoobscura gene. This supported dosage compensation in transgenic D. melanogaster males whose endogenous roX1 and roX2 genes had been deleted. This argues that roX1 RNAs from pseudoobscura and melanogaster share a core structure despite the divergence in sequence. However, the much more striking result was that the hybrid RNA promoted local MSL spreading around the sites of the autosomal transgenes (Fig. 5). In contrast, the fully melanogaster minigene did not support any MSL binding at or around the transgene. The latter result is easily understood because loss of the DHS sequence from the Δ39 minigene removes the primary MSL binding site within the roX1 gene, and transcription from the strong hsp83 promoter inhibits local spreading of even a full-length roX1 cDNA for reasons discussed above. We speculate that increasing the number of RBs from three to six somehow tips the balance towards local spreading in spite of vigorous transcription from the hsp83 promoter. The fact that the much smaller H83roX1 Δ39pseudo minigene is still capable of supporting local spreading argues that the 2.4 kb of internal sequence removed was not essential for this aspect of roX1 function, but might somehow enhance it.

If complex assembly does begin cotranscriptionally, then the functionally important sequences studied here at the very end of the transcript might link MSL subunit binding to 3′ processing and release the transcript from RNA polymerase. This might then influence how the newly formed complex is distributed locally or to more distant targets (Fig. 6). Removing all RBs may abort assembly leading to destruction of the defective RNA. Conversely, expanding the number of RB copies may enhance some aspect of assembly required for local spreading.

Fly genetics

All fly mutants have been previously described (Meller et al. 2000; Meller and Rattner 2002). The double mutant, y w roX1^ex6 Df(1) roX2⁵² [w⁺ cos4Δ], is referred to as simply roX1 roX2 throughout. Transgenic flies were created by P element transformation using pCaSpeR based vectors injected into y w host flies (Rubin and Spradling 1982; Pirrotta 1988; Robertson et al. 1988). Site specific integration was carried out using the ϕC31 system (Bischof et al. 2007) at attP sites on the second chromosome (Venken et al. 2006). Male viability measurements were made by crossing roX1 roX2 virgins to the appropriate transgenic father and comparing the ratio of transgenic sons to daughters. The UASroX1 transgene carries the 3.4 kb roX1 cDNA20 inserted into pUAST (Brand and Perrimon 1993). The sources of GAL4 came from either an hsp70GAL4 transgene on the second chromosome that produced adequate GAL4 without heat shock, or tubulinGAL4 or actin5CGAL4 transgenes each located on the third chromosome. In every case the UASroX1 was recombined onto the same chromosome as the GAL4 source. Although both transgenes carried w⁺ as the visible marker, the recombinants often had darker orange eye color. In each case the recombinant was scored for high rescue of male viability of roX1 roX2 double mutant males to confirm the desired recombinant had been recovered. For the time course using GAL80^ts, we started with a stock that carried both UASroX1-66B and tubulinGAL4, and recombined it with a tubulinGAL80^ts transgene that also mapped to the third chromosome (McGuire et al. 2003). Rare recombinants with very dark red eyes were tested for the presence of all three transgenes by genomic Southerns hybridized to a white probe and finding the unique junction fragment specific to each of the three starting transgenes.

Northerns

Either adults or third instar larvae were sorted by sex and homogenized in Trizol reagent (Invitrogen) according to the manufacturer's instructions. In most cases 10μg of total RNA was run on 1.0% agarose formaldehyde gels, transferred to Hybond N+ membranes in 50 mM NaOH. For the time course shown in Fig. 3, larvae were grown at 18° until reaching third instar (about 10 days) then shifted to 30°. Male larvae were harvested at 2, 4, 8, or 12 hours after the shift and homogenized in Trizol.

Polytene chromosome staining

Transgenic male larvae were generated by mating either y w roX1^ex6 virgins or roX1 roX2 virgins to males that were homozygous for the roX1 transgene of interest. The resulting sons were harvested as third instar and their polytene chromosomes were stained with anti MSL1 antibodies as described (Lyman et al. 1997).

Mutations of roX1

The roX1Δ39 minigene was assembled from the 5′ half of roX1Δ3 ligated to the 3′ portion of roX1Δ9 using the NheI site placed at the deletion point (Stuckenholz et al. 2003). All of the small deletions were generated by PCR mediated cloning using primers that spanned the desired deletion. The final sequence flanking each deletion is given below with <nt> indicating the number of nucleotides removed. A MluI restriction site placed at each junction (except –inverse roXbox) is underlined.

• -Region 1: CTAATTAACTTCCACGCGT<87>CCATACTATTCCTATA

• -SL1: TATTCCTATATAAACGCGT<63>ACATTTAAACAATAT

• -IRB: TTAAACAATATTTT<15>AATAAATGGAACGG

• -SL2: GTAAAACGAATAAATACGCGT<38>AGAAGTATAACTA

• -A bulge: CATTACGGTTCACGCGT<83>CGTTCTACGCAGT

• -RB1: TAAAAACTTGCTGATCAAACGCGT<11>GTTCTTAAAAAGA

• -RB2: CAAAGCAAGTAAAAATGTACGCGT<22>CACCAAGTGCCCAAA

• -RB3: CAAACTATAAAAAGAAAATACGCGT<14>GCCTGAGCCGATT

• -3RB: AACTTGCTGATCAAACGCGT <126>GCCTGAGCCGAT

• -RB123: RB2 and RB3 were deleted as above, RB1 was changed to TCAACAAACGAATCGGT where the underlined nucleotides are mutant.

The pseudoobscura hybrid minigene was made by PCR amplification of the appropriate genomic DNA. The primers were 5′ CACGCTAGCGCGCCACACCACAGTCCCCA3′ and 5′ CACGCGGCCGCGCGCCAGTAAGTACGTATAACGTTACCAAATC 3′ where the underlines sequences are artificial NheI and NotI sites incorporated to allow ligation into the hybrid gene vector. All of the small deletions and the pseudoobscura hybrid genes were sequenced after construction to check for PCR induced mutations.

Sequence Analysis

The roX genes could be readily obtained for species up to pseudoobscura by simple BLAST searches with the melanogaster starting sequences. Beyond that we relied on recovering better conserved protein coding genes adjacent to roX and then intensively searching the surrounding region for diverged roX genes. The closely related sequences were aligned using ClustalW (http://www.ebi.ac.uk/Tools/clustalw/index.html), but we resorted to manual alignment for the more distant species in order to deal with the many indels. These were guided by local BLAST searches of short regions and Dot Plot analysis to search for forward and inverted repeats (http://www.vivo.colostate.edu/molkit/dnadot/).