Progressive dosage compensation during Drosophila embryogenesis is reflected by gene arrangement

Abstract

In Drosophila melanogaster males, X‐chromosome monosomy is compensated by chromosome‐wide transcription activation. We found that complete dosage compensation during embryogenesis takes surprisingly long and is incomplete even after 10 h of development. Although the activating dosage compensation complex (DCC) associates with the X‐chromosome and MOF acetylates histone H4 early, many genes are not compensated. Acetylation levels on gene bodies continue to increase for several hours after gastrulation in parallel with progressive compensation. Constitutive genes are compensated earlier than developmental genes. Remarkably, later compensation correlates with longer distances to DCC binding sites. This time–space relationship suggests that DCC action on target genes requires maturation of the active chromosome compartment.

Introduction

Dosage compensation (DC) is a regulatory mechanism that evolved to ensure balanced sex‐chromosomal gene expression in sexually dimorphic species. In Drosophila melanogaster, where males are denoted as XY and females as XX, this adjustment is achieved by an approximately twofold activation of transcription of X‐chromosomal genes in male organisms. In cases where DC fails, male‐specific lethality is observed 1. This coordinated and chromosome‐wide process is carried out by the male‐specific lethal (MSL) dosage compensation complex (DCC), which consists of five protein subunits, MSL1, MSL2, MSL3, maleless (MLE), and males absent on the first (MOF), as well as long non‐coding RNA, RNA on the X (roX1/roX2) 1, 2. This ribonucleoprotein complex decorates the single male X‐chromosome exclusively to enhance chromatin accessibility and transcription 3, 4.

How DCC targets the male X‐chromosome is of great interest and serves as an example for chromosome‐wide regulation. Key to targeting of the X are genetically encoded DCC binding sites, which are termed chromosomal entry sites (CES; 5, 6) or high‐affinity sites (HAS; 7, 8). DCC is targeted to the X through its DNA binding subunit, MSL2 9, 10, which cooperates with the ubiquitous CLAMP protein for tight binding to so‐called MSL response elements (MREs) 11, 12. According to the current model, DCC “spreads” from HAS to transcriptionally active genes in the nuclear vicinity 1, 2. Active genes are marked epigenetically by histone H3 trimethylation at lysine 36 (H3K36me3), which is recognized by MSL3 via its chromo barrel domain 13, 14. Long‐range DCC interactions are guided by the three‐dimensional chromosome topology within the active compartment 15, 16. Upon binding and spreading of the complex on the X‐chromosome, the crucial function of DCC is initiated: H4K16 acetylation by the acetyltransferase MOF. Enrichment of H4K16ac on the X‐chromosome weakens packing of the chromatin fiber 17 and renders chromatin open for more efficient transcription 18, 19.

The establishment of DC during early embryonic development hinges on the expression of the only male‐specific protein of the complex, MSL2, which is repressed in females by the master regulator gene sex‐lethal (Sxl) 1, 2. In the presence of MSL2 and at least one of the two male‐specific roX RNAs 20, 21, DCC assembles. Nuclear localization of MSL proteins has been first observed at blastoderm stage 6 and its enrichment in X‐chromosomal compartment at stage 9 22, 23. The first wave of zygotic transcription starts at least an hour earlier, at stage 4 24, and seems to be partially compensated, presumably by a generic, DCC‐independent mechanism 25, 26.

In our current study, we integrate a transcriptome analysis (RNA‐seq) of single embryos during the 12 h of development representing embryonic stages 4–15, with high‐resolution chromatin immunoprecipitation profiling (ChIP‐seq) of MSL2, MOF, H4K16ac, and H3K36me3 at two time windows that allow monitoring of the progress of dosage compensation. To our surprise, we find only partial dosage compensation of genes between stages 5–8, about 3–4 h into embryonic development, despite robust detection of DCC binding and H4K16ac deposition at this time. However, the extent of H4K16 acetylation on the bodies of active genes increases with time and correlates with full dosage compensation, illustrating that the distribution of the activating acetylation to target genes rate limits dosage compensation. Interestingly, constitutive genes on the X‐chromosome that are compensated earlier tend to be located closer to HAS than developmental genes, which require more time for full compensation. Conceivably, this signifies evolution of HAS closer to genes that require robust dosage compensation early during development.

Results and Discussion

Dosage compensation of the X‐chromosomal transcription progresses gradually during embryogenesis

To track dosage compensation of the X‐chromosome during embryogenesis, we sorted single D. melanogaster embryos to developmental stages based on morphological features 27 and determined their transcriptome by RNA‐seq (Table EV1). Principal component analysis (PCA) on 54 single embryonic transcriptomes confirmed the precision and reproducibility of the sorting (Fig 1A). The first principal component (PC1) categorizes embryos into distinct consecutive clusters based on their developmental stage, whereas PC2 delineates the onset of zygotic transcription. Furthermore, through assessment of sex‐specific gene expression (i.e., Sxl, msl‐2, roX1, and roX2), we unequivocally assigned sex to each embryo and ascertained that both sexes are represented in each developmental stage cluster (Table EV1).

Figure 1. Dosage compensation is progressively established during embryonic development.

1. Principal component analysis (PCA) of RNA‐seq data derived from 54 single embryos in eight developmental stages. X‐axis shows the first (PC1) and y‐axis the second (PC2) principal component. The variance explained by each component is indicated in parenthesis. Morphological stages are differentially colored and described in Table EV1 27.
2. Ratio of X‐chromosomal over autosomal median transcript abundances (X/A) of preblastoderm (PB), female, and male embryos in seven developmental stages. Each data point is a single embryo. A ratio of one represents balanced transcript levels between the X‐chromosome and autosomes (dotted horizontal line).
3. Log2 fold change (log2FC) of RNA‐seq between female (F) and male (M) embryos of expressed genes on X‐chromosome (X, left, total = 1,515) and on autosomes (A, right, total = 7,451) in seven developmental stages.
4. Log2 fold change (log2FC) of RNA‐seq between female (F) and male (M) embryos of X‐chromosomal genes which are constitutively (left, total = 303) or developmentally expressed (right, total = 303) in seven developmental stages.

Data information: Horizontal lines indicate median value, box ranges denote interquartile range (IQR), and whiskers mark ±1.5x IQR. P‐values (pval) are calculated by one‐sample Wilcoxon signed‐rank test to assess the median log2FC values against zero (balanced expression); ns: non‐significant (P‐value > 0.001).

We determined transcript abundances as log2 transcript per million (TPM) and compared the X‐chromosome (X) to autosomes (A) (Fig EV1A). This shows an increase in median transcript levels between stages 4 and 15 (between 1.5 and about 12 h of development, Table EV1). At any given time, median autosomal transcript levels are similar in both sexes, showing an equivalent rate of genome activation in both sexes (Fig EV1A). Plotting the ratio of X‐chromosomal over autosomal median transcript levels, for which a ratio of one indicates balanced X and A levels, revealed that maternal transcripts are biased for X‐chromosomal expression (Fig 1B). Upon zygotic genome activation (from stage 5 onwards), the X/A ratio drops in both, females and males. In females, the ratios approximate 1, indicating a near‐balanced expression of all chromosomes. Interestingly, the arch‐shaped profile suggests some fine‐tuning of X to A balance even in females. In male embryos, the ratio of X/A expression drops dramatically at stages 5–6, revealing that the first zygotic transcription is not yet fully compensated. During the course of the next 10 h of embryonic development, the X/A ratio in males slowly approximates 1, reflecting the progression of dosage compensation.

Figure EV1. Dosage compensation dynamics of the X‐chromosomal transcriptome during embryonic development (related to Fig 1).

1. Log2 TPM (transcript per million, RNA‐seq) levels of X‐chromosomal (X) and autosomal (A) genes in preblastoderm (PB), female (F), or male (M) embryos in eight developmental stages. Horizontal lines indicate median value, box ranges denote interquartile range (IQR), and whiskers mark ±1.5x IQR. The number of embryos used for each stage analysis can be found on Table EV1.
2. Log2 TPM (RNA‐seq) levels of genes which were used for k‐means clustering to determine the sex of the embryos (top), genes which encode non‐sex‐specific DCC subunits (middle) and additional genes related to dosage compensation (bottom).
3. Western blot showing protein levels of DCC subunits in stages 5–8 and 13–15.
4. Heatmaps of log2 TPM (RNA‐seq) values for X‐chromosomal constitutive (left) and developmental (right) genes. Genes are grouped by the variance across all developmental stages; i.e., constitutive as 20% least and developmental as 20% most variable. Gene Ontology (GO) of gene groups is on Table EV2.
5. Log2 fold change (log2FC) of RNA‐seq between female and male embryos for autosomal genes (A) which are constitutively (top, total = 1,474) or developmentally expressed (bottom, total = 1,474) in seven developmental stages. Horizontal lines indicate median value, box ranges denote interquartile range (IQR), and whiskers mark ±1.5x IQR.

To avoid any possible bias from analyzing different sets of X‐chromosomal and autosomal genes and to measure dosage compensation on a gene‐by‐gene basis, we calculated log2 fold changes of transcript abundance between female (F) and male (M) embryos for each gene. Average value of 0 indicates equal transcript levels originating from the single male X and the two female X‐chromosomes, and thus corresponds to full dosage compensation, whereas a value of 1 implies a total lack of dosage compensation (2‐fold more female than male transcripts). Expectedly, autosomal transcript levels are comparable at all stages (Fig 1C). As previously shown, a log2 fold change of 1 is never observed for X‐chromosomal transcripts, indicating some compensation happens early 25. We were surprised, however, to find high ratios of female over male X expression at stages 6–8, indicating that dosage compensation is incomplete even ~ 4 h into development. In the following hours, male X expression becomes gradually balanced (Fig 1C). Nonetheless, full compensation is not reached by the latest timepoint of our analysis (stage 15, ~ 12 h of development).

The lag in compensation in early stages may be explained by insufficient DCC components; therefore, we compared transcript abundances of the MSL and SXL genes. Noticeable increase in SXL and MSL2 transcripts is seen upon maternal to zygotic transition with the expected sex‐specific bias (Fig EV1B top). Since the expression of both is regulated translationally, it is not surprising to detect transcripts of MSL2 in female embryos, and, conversely, SXL transcripts in males. As reported before, roX1 is transcribed in both sexes up to stage 12, from which point onwards it is male‐specific 20, 21. Furthermore, male‐specific roX2 RNA transcription begins at stage 10 and loosely mirrors the profile of gradual male X compensation 20. Of note, mRNAs of all DCC subunits except for MSL2, namely MSL1, MSL3, MLE, and MOF, were found maternally contributed and stably expressed in both sexes (Fig EV1B). Utilizing a fly line in which sex sorting can be done, protein levels of DCC subunits were confirmed by Western blot (Fig EV1C).

Constitutive genes are compensated early, developmental genes later

As a first step toward discovering possible reasons for the observed heterogeneity of dosage compensation during embryogenesis, we defined within our RNA‐seq dataset constitutive genes as the 20% least varied and developmental genes as the 20% most varied transcripts through all developmental stages, regardless of sex (Fig EV1D). To ensure that genes were appropriately classified, we analyzed the gene ontology (GO) terms for each category. Consistent with our definition, those genes characterized as constitutive function in cellular processes such as transcription and translation, and those termed developmental are involved in processes such as anatomical, tissue, and organ development (Table EV2). Interestingly, X‐chromosomal genes defined as constitutive are fully compensated by stage 15 and genes classified as developmental are compensated slower (Fig 1D), whereas developmental and constitutive genes on autosomes are expressed equally in both sexes (Fig EV1E). We proceeded to explore the reasons for delayed dosage compensation of developmental genes.

Chromatin binding of MSL2 precedes complete dosage compensation

Previous immunofluorescence microscopy (IFM) analyses had identified the colocalization of MSL proteins on X‐chromosomal territory at stage 9 22. Since we found dosage compensation incomplete at this stage, we revisited the IFM analysis. In line with earlier report, we failed to detect localized MSLs in embryos at blastoderm stage 5. This is likely due to a combination of low protein levels and the decondensation of chromosomes 28, leading to a distribution of signal below detection limits. Even in a stage 8 embryo, MSL staining in interphase nuclei is difficult to detect, perhaps indicating that coherent territories are not formed yet (Appendix Fig S1A). However, concentrated staining on mitotic chromosomes allows a clear identification of X‐chromosomal staining at stage 8, and by stage 14, focal colocalization of MSL3 and MSL2 was seen in all nuclei (Appendix Fig S1A).

Is the delayed dosage compensation due to incomplete DCC binding to the X or due to a functional deficit, such as inefficient distribution to active genes or low histone acetylation activity? Because IFM is limited in sensitivity and resolution, we resorted to chromatin immunoprecipitation (ChIP), monitoring MSL2 and MOF distribution along the chromosome as a proxy for DCC interactions and H4K16 acetylation on gene bodies as an indicator for DCC activity. We optimized a protocol to generate high‐resolution chromatin interaction profiles combining micrococcal nuclease (MNase) digestion with ChIP followed by sequencing of enriched DNA (ChIP‐seq). We chose MNase digestion for chromatin fragmentation instead of sonication to preserve the integrity of the DCC and allow crosslinking of the complex at both HAS and active gene bodies 8. Two time windows during embryogenesis were chosen based on our RNA‐seq: the first encompassing stages 5–8, where dosage compensation is least measured in male, and the second covering stages 13–15, where compensation of the constitutive genes was achieved. Our embryo staging was verified by monitoring the H3K36me3 profiles of suitable developmental indicator genes through ChIP‐qPCR (Appendix Fig S1B, see Table EV3 for primers). We also monitored the ChIP efficiency of specific antibodies through qPCR of known targets by measuring enrichment over X‐chromosomal HAS for MSL2 and gene body enrichment of active X‐linked genes for MOF and H4K16ac (Appendix Fig S1C).

To characterize the direct DNA binding of MSL2, DNA fragments from paired‐end sequencing runs were computationally subset to sizes between 10 and 130 bp, i.e., sub‐nucleosomal length, since it has been previously shown that HAS are nucleosome‐free 8, 12. For clarity, we refer to those binding events as direct “DNA” binding (Fig 2A). Interestingly, we found robust MSL2 enrichment at PionX sites 10 and HAS 8 during both developmental time windows (Fig 2B). MSL2 binds its target sequences already at times when compensation is least established. Conceivably, the interactions with active genes, which are not genetically encoded but epigenetically determined by H3K36me3 binding, may be weak in stages 5–8. Crosslinking of MSL2 (directly or indirectly via MSL3) to chromatin can be used as a measure for these interactions. We therefore analyzed interaction of MSL2 with nucleosomes defined as DNA fragments of lengths between 130 and 220 bp and referred to them as “chromatin” binding henceforth (Fig 2A). We observed MSL2 interaction with the nucleosomes neighboring HAS, but also with chromatin within a region of up to ~ 4 kb surrounding HAS (Fig 2C). To expand the analysis genome‐wide, we called peaks for each time window (Dataset EV1) and confirmed that ChIP enrichment is concordant across replicates (Fig EV2A).

Figure 2. Binding of MSL2 to high‐affinity sites is established in early embryonic stages.

1. Example genomic region of MSL2 MNase‐ChIP‐seq profiles in early (stages 5–8, orange) and late (stages 13–15, purple) mixed‐sex embryos. DNA binding is defined as sub‐nucleosomal fragments (10–130 bp), while chromatin binding as mono‐nucleosomal fragments (130–220 bp). Called peaks/regions (Dataset EV1) are represented as boxes above the tracks and HAS below. Late‐appearing regions are marked by a plus (+) sign. Graph represents pooled IP/input signal of three biological replicates.
2. Average composite plots of sub‐nucleosomal MSL2 MNase‐ChIP‐seq (DNA) centered at PionX (left) or HAS (right) in early and late mixed‐sex embryos. Total number of PionX = 56 and HAS = 247. Graph represents average IP/input signal of three biological replicates. Shaded area indicates the 95% confidence interval across sites.
3. Average composite plots of mono‐nucleosomal MSL2 ChIP‐seq (Chromatin) centered at PionX (left) or HAS (right) in early and late mixed‐sex embryos.

Figure EV2. Binding of MSL2 to high‐affinity sites and formation of broad MSL2 regions are established in early embryonic stages (related to Fig 2).

1. Pearson correlation heatmaps of MSL2 MNase‐ChIP‐seq profiles at DNA peaks (left) and chromatin regions (right) for three biological replicates at two timepoints.
2. Relative chromosomal distribution of MSL2 DNA peaks as defined in Fig 3A and B. Peaks which do not overlap (top) or overlap MSL2 chromatin regions (mid) and MSL2 chromatin regions which do not overlap MSL2 DNA peaks (bottom) are shown on the left. Relative chromosomal distribution of HAS and chromosome sizes are shown on the right. Number of sites is indicated in parentheses.
3. De novo motif analysis of MSL2 DNA peaks of the same categories in (B). The most significant motif is shown.
4. Size distribution of MSL2 chromatin regions (n = 215 and 231, respectively) present in both early (stages 5–8) and late (stages 13–15) embryos. Horizontal lines indicate median value, box ranges denote interquartile range (IQR), and whiskers mark ±1.5x IQR. P‐value (Wilcoxon rank sum test) is indicated.
5. Log10 distance between the middle of an MSL2 chromatin region and its nearest functional MSL2 DNA peak. Left: early (n = 212) and right: late (n = 87) regions as defined in Fig 3C. Red line indicates median distance to the nearest peak.

Systematic comparison of “DNA” and “chromatin” interactions revealed three classes. First, 139 MSL2 DNA binding events reside within regions of chromatin binding (Figs 2A, and 3A and B). These regions are highly enriched on the X (Fig EV2B) and often contain a DNA sequence motif very similar to the known MRE motif (Fig EV2C). Conceivably, many of these sites represent HAS that are known to often reside in introns 8, 15. Second, we observed 202 isolated MSL2 DNA binding events that are not associated with neighboring chromatin interactions (Figs 2A, and 3A and B). Many of these sites reside on autosomes (Fig EV2B) and do not always contain the typical GAGA‐rich sequences of MRE (Fig EV2C). The lack of nucleosome association implies the absence of the MSL3 “reader”. As these attributes suggest a non‐functional binding of MSL2 to DNA, we omitted them from further analysis. Third, we found 199 chromatin binding regions that do not contain a detected DNA binding site (Figs 2A, and 3A and B). These regions may well reflect the longer‐range interactions of a remote HAS with an active gene facilitated by chromosome conformation 16. To explore the developmental kinetics of MSL2 “chromatin” binding, we classified each region, averaging at 4 kb (Fig EV2D), according to whether MSL2 binds early (stages 5–8) and stays, or whether it associates with the region only late (stages 13–15; Figs 2A and 3C). Interestingly, the delayed chromatin binding of MSL2 correlates with a longer chromosomal distance to the nearest peak of MSL2 DNA binding (Fig EV2E).

Figure 3. MSL2 binds broad chromatin regions in early embryonic stages.

1. Venn diagram showing the overlap of sub‐nucleosomal MSL2 DNA peaks and mono‐nucleosomal MSL2 chromatin regions.
2. Average composite plots of MSL2 [Chromatin] profiles (stages 5–8, orange and stages 13–15, purple) centered at MSL2 DNA peaks which do not overlap (DNA) or overlap MSL2 chromatin regions (DNA + Chromatin), and MSL2 chromatin regions which do not overlap MSL2 DNA peaks (Chromatin). Graphs represent average IP/input signal of three biological replicates. Shaded area indicates the 95% confidence interval across sites.
3. Average composite plots of MSL2 [Chromatin] profiles centered at MSL2 chromatin regions called in both early and late (left) or only in late stages (right). Graphs represent average IP/input signal of three biological replicates. Shaded area indicates the 95% confidence interval across regions. Total number of early = 212 and late = 87.
4. Average composite plots of MOF [Chromatin] profiles centered at MSL2 chromatin regions called in both early and late (left) or only in late stages (right).

MOF‐dependent H4K16ac accumulates on X‐chromosomal target genes during embryogenesis

Contrary to our expectation, we had found robust DNA binding and widespread chromatin contacts of MSL2 already in embryonic stages 5–8, when dosage is least compensated. We wondered whether the developmental delay of compensation may be explained by incomplete DCC assembly and absence of the histone acetyltransferase (HAT) MOF or by a functional restraint of the complex, characterized by reduced HAT activity. To distinguish between those possibilities, we performed high‐resolution MOF and H4K16ac ChIP‐seq on the same soluble chromatin as previously (Appendix Fig S1C; Fig 4A) in three biological replicates (Fig EV3A). We detected robust binding of MOF at MSL2‐bound PionX sites and HAS (Fig EV3B), with interesting differences. First, monitoring small ChIP fragments, we do not recover sharp peaks as for MSL2 (Fig EV3B), in agreement with the earlier finding that MOF does not contact DNA directly, but only via MSL2/MSL3 8. Second, while the breadth of the region of MOF crosslinking aligns nicely with MSL2 interactions and does not change as development proceeds (Figs 3D, and 4A and B, Dataset EV2), crosslinking intensifies both around HAS (Fig EV3B) and on gene bodies (Fig 4C). Increased crosslinking at HAS may indicate the ongoing assembly of functional DC complexes, whereas crosslinking at gene bodies may reflect the refinement of X‐chromosome conformation (see below).

Figure 4. MOF‐dependent H4K16ac accumulates at X‐chromosomal gene bodies during embryonic development.

1. Example genomic region of DNA binding by MSL2 (DNA), and chromatin binding of MSL2 (Chrom.), MOF, and H4K16ac MNase‐ChIP‐seq profiles in early (stages 5–8, orange) and late (stages 13–15, purple) mixed‐sex embryos. Called peaks/regions (Datasets EV1 and EV2) are represented as boxes above the tracks. Graph represents pooled IP/input signal of three biological replicates.
2. Size distribution of X‐chromosomal MOF and H4K16ac regions, which are present in both early and late embryos. Horizontal lines indicate median value, box ranges denote interquartile range (IQR), and whiskers mark ±1.5x IQR. P‐value (Wilcoxon rank sum test) is indicated. Total number of regions is 459, 443, 854, and 400, respectively.
3. Average composite plots of MOF MNase‐ChIP‐seq centered at transcription start site (TSS) or transcription termination site (TTS) in early or late embryos for X‐chromosomal genes (total = 1,515). Graph represents average IP/input signal of three biological replicates. Shaded area indicates the 95% confidence interval across sites.
4. Average composite plots of H4K16ac MNase‐ChIP‐seq centered at TSS or TTS in early or late embryos for X‐chromosomal genes (total = 1,515).

Figure EV3. MOF interaction with chromatin during embryonic development leads to accumulation of H4K16ac on the X‐chromosome (related to Fig 4).

1. Pearson correlation heatmaps of MOF (left) and H4K16ac (right) MNase‐ChIP‐seq profiles at X‐chromosomal genes for three biological replicates at two timepoints.
2. Average composite plots of sub‐nucleosomal (“DNA”, left) and mono‐nucleosomal (Chromatin, right) MOF MNase‐ChIP‐seq centered at PionX or HAS in early (stages 5–8, orange) and late (stages 13–15, purple) mixed‐sex embryos. Total number of PionX = 56 and HAS = 247. Graph represents average IP/input signal of three biological replicates. Shaded area indicates the 95% confidence interval across sites.
3. Average composite plots of H4K16ac MNase‐ChIP‐seq profiles centered at MSL2 chromatin regions called in both early and late (left) or only in late stages (right). Graph represents average IP/input signal of three biological replicates. Shaded area indicates the 95% confidence interval across sites. Total number of early = 212 and late = 87.
4. Average composite plots of MOF and H4K16ac MNase‐ChIP‐seq centered at TSS or TTS in early or late embryos for autosomal genes (total = 7,451). Graph represents average IP/input signal of three biological replicates. Shaded area indicates the 95% confidence interval across sites.

H4K16 acetylation is found in regions containing MOF interactions, but with much broader distribution (Fig 4A). Already at stages 5–8, H4K16ac is enriched on the coding regions of X‐chromosomal genes as far as 20 kb away from the centers of MSL2 chromatin binding regions (Fig EV3C). Remarkably, whereas MOF is found enriched in regions of average size of ~ 5.5 kb that do not change during development, the H4K16ac mark “spreads” with time from an average domain size of 3 kb to 13 kb (Fig 4B). Furthermore, while MOF binding to X‐chromosomal gene bodies increases moderately between early and late developmental time windows (Fig 4C), the H4K16ac mark it places accumulates massively during the same period (Fig 4D). Of note, almost no H4K16ac signal is detected at autosomal gene bodies (Fig EV3D), confirming that the vast majority of H4K16 acetylation is linked to dosage compensation.

To be able to compare our H4K16ac profile to published data, we also generated high‐resolution ChIP‐seq profiles from sonicated soluble chromatin of embryos of the appropriate stages and monitored H3K36me3 as a proxy for gene activity (Fig EV4). ChIP profiles of H4K16ac from sonicated chromatin confirmed the findings seen with MNase‐treated chromatin and further documented that H4K16ac localizes to actively transcribed chromatin marked by H3K36me3 (Fig EV4A, Appendix Fig S2). The similar H3K36me3 profiles at early and late times (Fig EV4A and B, Appendix Fig S2) reveal that the increase in H4K16ac cannot be due to a corresponding increase in H3K36me3.

Figure EV4. Comparison of H4K16ac ChIP‐seq on MNase‐treated and sonicated chromatin (related to Fig 4).

1. Example genomic region of H4K16ac on MNase‐treated (top) or sonicated (mid) and H3K36me3 on sonicated (bottom) ChIP‐seq profiles (stages 5–8, orange and stages 13–15, purple). Called regions (Datasets EV2 and EV3) are represented as boxes above the tracks. H4K16ac (MNase) graph indicates a pool of three biological replicates, while H4K16ac (son.) and H3K36me3 (son.) graphs represent scaled signal (between 0 and 1) of two biological replicates.
2. Size distribution of sonicated X‐chromosomal H4K16ac and H3K36me3 regions, which are present in both early and late embryos. Horizontal lines indicate median value, box ranges denote interquartile range (IQR), and whiskers mark ±1.5x IQR. P‐value (Wilcoxon rank sum test) is indicated. Total number of regions: 772, 558, 678, and 702, respectively.

Compensation of genes is determined by the level of H4K16 acetylation

To explore the relationship between timing of H4K16 acetylation and dosage compensation at the gene level, X‐chromosomal genes were clustered into three groups based on their H4K16ac states (Fig 5A). Comparing acetylation levels between early and late embryos, we see that the vast majority of genes gain acetylation as embryonic development proceeds. Genes with early high acetylation levels (cluster 1) are the first to reach full compensation by stage 15 (Fig 5B). Genes in cluster 2 show moderate levels of H4K16ac; nonetheless, their noticeable gain of acetylation during development leads to appreciable compensation (Fig 5A and B). By contrast, the compensation of genes that show low H4K16ac early (cluster 3) remains incomplete despite additional acetylation (Fig 5A and B). We performed similar clustering analysis for MOF (Appendix Fig S3) as well as for H4K16ac and H3K36me3 from sonicated chromatin (Appendix Fig S4) independently and arrived at a similar pattern. Thus, early or late compensation of genes is reflected by the chromatin features.

Figure 5. H4K16ac levels distinguish early from late compensated genes.

1. Scatterplot comparing X‐chromosomal genic H4K16ac MNase‐ChIP‐seq levels in early (stage 5–8) and late (stage 13–15) mixed‐sex embryos. Genes are grouped by k‐means clustering (k = 3). The numbers of genes are cluster 1 = 979, cluster 2 = 364, and cluster 3 = 172.
2. Log2 fold change (log2FC) of RNA‐seq between female (F) and male (M) embryos at genes grouped as in (A) (clusters 1–3). Horizontal lines indicate median value, box ranges denote interquartile range (IQR), and whiskers mark ±1.5x IQR. The number of embryos used for each stage analysis can be found on Table EV1.
3. Log10 distance between the middle of a gene and its nearest functional MSL2 DNA peak grouped as in (A) (clusters 1–3). Red line indicates median distance to the nearest peak.
4. Comparison of cluster 1 (left) or cluster 3 (right) genes defined by H4K16ac MNase‐ChIP‐seq levels with constitutive and developmental genes as defined in Fig 1C.

Development in time and space

In search for an explanation for the differential timing of DC in development, we found a relationship between genes and genomic distance, where highly acetylated and fast compensated genes are also closest to peaks of MSL2 DNA binding, and conversely, lowly acetylated and least compensated genes are farthest away (Fig 5C). We previously concluded that constitutive genes are compensated faster than developmental genes (Fig 1D). As expected, constitutive genes are mostly represented in cluster 1 (Fig EV5), which are closest to MSL2 DNA binding sites and reach full compensation by stage 15 (Fig 5D). On the other hand, cluster 3 genes, which are situated farther away from MSL2 DNA binding sites and are delayed in reaching full compensation, show a better overlap with developmental genes (Fig 5D). Venn diagrams relating the gene clusters by expression type (constitutive vs. developmental, as defined in Fig 1D), and by H3K36me3 or H4K16ac marks (Fig EV5) reveals a small fraction of developmental genes that acquire H4K16 acetylation and are therefore subject to MSL‐mediated dosage compensation.

Figure EV5. H4K16ac and H3K36me3 levels distinguish early from late compensated genes (related to Fig 5).

Overlap of X‐chromosomal constitutive and developmental genes as defined in Fig 1D with clusters 1, 2, and 3 as defined by independent clustering of sonicated H4K16ac or H3K36me3 in Appendix Fig S4.

Concluding Remarks

According to the prevalent model, targeting the X‐chromosome for dosage compensation in Drosophila requires at least two distinct steps: first, the DCC must associate with genetically encoded high‐affinity DNA binding sites. Second, the complex must spread to epigenetically marked active transcription units to disseminate the activating H4 acetylation 1, 2. We and others suggested that this dissemination may involve dynamic interactions between chromosomal regions of primary binding and target genes in the active nuclear compartment 15, 16, 29. Our current study reveals that the DCC assembles at HAS already at the earliest developmental window we could interrogate by ChIP, but this binding does not immediately lead to maximal H4K16 acetylation. Rather, acetylation gradually increases with time and coincides with better compensation. We consider three non‐exclusive mechanisms that may well explain this finding. First, we observe a modest increase in MOF recruitment to HAS as development proceeds, suggesting continuous assembly of DCC with time. Second, we found that roX2 expression is well correlated with the refinement of dosage compensation. It will be of interest to explore whether and how roX2‐containing DCC differs from roX1‐containing DCC, which dominates earlier developmental stages. Finally, the correlation of late compensation with greater distance to primary MSL2 binding site suggests a role for chromosome conformation. We speculate that the active chromosome compartment is continuously refined during embryonic development and that this “maturation” promotes an increasingly efficient dissemination of H4 acetylation from primary DCC binding sites. This agrees nicely with the refinement of chromosome organization between stages 5–8 and stage 16 that was recently detected in a high‐resolution Hi‐C study 30. From an evolutionary perspective, it must have been vital to evolve HAS closer to genes that required efficient dosage compensation early during embryogenesis, when long‐range contacts were less reliable. Indeed, HAS are strongly enriched within genes, i.e., in introns 8, 15. The dosage‐compensated X‐chromosome thus provides a beautiful illustration of how a temporal order of events is reflected by the spatial arrangement of chromosomal elements.

Materials and Methods

Single‐embryo RNA‐seq protocol

RNA‐seq on individual embryos was described previously in Ref. 31. Briefly, embryos were hand‐picked from collection plates and soaked in a drop of Voltalef 10 S halocarbon oil (Lehman and Voss Co.). Stages were determined by morphological features under a stereomicroscope. Embryos were smashed with a needle into 200 μl lysis buffer (Agencourt RNAdvance Tissue Kit, Beckman Coulter, Cat. No. A32645), and 10 μl 1:100 ERCC Spike‐in RNA Mix was added (Ambion, Cat. No. 4456740). RNA was extracted from the lysate following the manufacturer's protocol with volume adjustment. Ribosomal RNA was depleted using rRNA depletion kit (human/mouse/rat; New England Biolab, Cat. No. E6310). Directional (except eight embryos non‐directional) libraries were prepared using NEBnext Ultra RNA Library Prep Kit for Illumina (New England Biolab, Cat. No. E7530S) following the recommended protocol. In total, 54 embryos from eight stages were sequenced on an Illumina HiSeq1500 instrument in paired‐end mode.

RNA‐seq data analysis

Paired‐end RNA‐seq reads were aligned to the reference genome (dmel release 6) using STAR (version 2.5.3a) with the parameter –outFilterMultimapNmax 1. Genic reads were counted using STAR –quantMode GeneCounts and the Flybase GTF annotation (dmel‐all‐r6.17). Expressed genes were filtered by a minimum threshold of 21 counts present in at least five embryos. TPM (transcript per million) values were calculated by first dividing each genic count by the effective exonic gene length (in kilobases), second dividing by the sum of length corrected counts, and finally, multiplying by a million. Principal component analysis (PCA) was performed on log2 TPM values including genes with a variance lower than the 90^th percentile. X/A ratios were calculated by dividing X‐chromosomal by autosomal median TPM values in each individual embryo. The sex of the embryos was determined by k‐means clustering on established sex‐specific marker genes (i.e., Sxl, msl‐2, roX1, and roX2) with k = 3 for preblastoderm, female, and male groups. For each stage and sex group, the mean of log2 TPM values was calculated. The difference in mean log2 TPM values between females and males was visualized as boxplots for each stage. Developmental genes were defined as the most variable 20% genes (variance of group mean log2 TPM), whereas constitutive genes as the least variable 20% genes. Gene Ontology analysis on developmental or constitutive genes was performed using the topGO BioConductor package (version 2.30.1) with classic algorithm and Fisher statistic.

Antibodies

αMSL1, αMSL2, αMSL3, αMOF, and αMLE antibodies were previously described in Refs 12, 32, 33, 34. αSXL antibody was kindly provided by F. Gebauer and previously described in Ref. 35. Commercially available antibodies used were αGFP (ChromoTek, Cat. No. 029762), αH3K36me3 (Abcam, Cat. No. ab9050), and αH4K16ac (EMD Millipore, Cat. No. 07‐329). Secondary αRat‐Alexa Fluor 488 and αRabbit‐Alexa Fluor 647 were obtained from Jackson Immunoresearch.

Large‐scale embryo collection

Timed embryo collection was dechorionated in 25% commercial bleach and fixed in PBS/3.7% formaldehyde/heptane at room temperature for 20 min, agitated. After formaldehyde removal, embryos were shaken in heptane/methanol (1:1) solution. Devitellinized embryos were rinsed in methanol and stored at −20°C for later use.

Immunostaining, image acquirement, and analysis

For immunofluorescence staining, embryos were rehydrated and washed in PBSTx‐0.1% (PBS/0.1% Triton X‐100). Thereafter, blocking was done twice for 20 min each in block solution (PBSTx‐0.1%/5% normal donkey serum). Primary antibodies were added at the working dilution in block solution, and samples were incubated overnight at 4°C with agitation. After rinsing and washing four times in PBSTx‐0.1% for 15 min each, secondary antibodies diluted in block solution were added and samples were incubated in dark for 2 h at room temperature with agitation. Samples were rinsed and washed four times in PBSTx‐0.1% for 15 min each. DAPI was added during the last wash to stain DNA before a final wash in PBS for 10 min. Thereafter, embryo was mounted onto a 3‐well glass slide (VWR, Cat. No. 631‐0453) in VECTASHIELD mounting medium (Vector Laboratories, Cat. No. H‐1000), and coverslip was applied and sealed with nail polish. Images were acquired on a Leica TCS SP5 confocal microscope using the same settings. Images were then processed and analyzed using Fiji 36.

Western blot

For the sorting of male vs. female embryo, a fly line expressing GFP under the control of the early promoter of SXL (SXL‐Pe‐EGFP) was utilized 37. Dechorionated embryos were collected in a cell strainer/basket and blotted dry. A droplet of Voltalef Oil 10 S was placed onto a glass slide, and embryos were moved into the oil with a fine paintbrush. Under a fluorescent dissection microscope, appropriately staged embryos were collected, and male/female embryos were sorted based on the GFP expression. Embryos were resuspended in 1 μl/embryo urea lysis buffer, smashed with a micropestle (LLG Labware, Cat. No. 9.314 501), and heated up at 65°C for 15 min before electrophoresis.

Chromatin preparation

For chromatin preparation, staged mixed‐sex embryos were dechorionated in 25% commercial bleach and weighted. 1 g embryos were washed in 50 ml PBS/0.01% Triton X‐100 (PBSTx‐0.01%) and resuspended in 10 ml fix solution (50 mM HEPES pH 8.0, 100 mM NaCl, 1 mM EDTA, and 0.5 mM EGTA)/3.7% formaldehyde (Merck, Cat. No. 1040031000). After addition of 30 ml n‐heptane, embryos were rigorously shaken for 1 min and rotated at room temperature for 13.5 min. Following a spin at 2,000 g for 1 min, crosslinking was halted by addition of 50 ml PBSTx‐0.01%/125 mM glycine. Washed embryos were flash‐frozen and stored at −80°C until further use. Frozen embryos were resuspended in 10 ml homogenization buffer (15 mM HEPES pH 7.6, 10 mM KCl, 2 mM MgCl₂, 0.1 mM EDTA, 0.5 mM EGTA, 350 mM sucrose, 1 mM DTT, 0.2 mM PMSF, Roche cOmplete Protease inhibitor without EDTA) and dounced 20 times with a loose pestle and 20 times with a tight pestle before being spun down at 170 g for 10 min at 4°C. The nuclei pellet was resuspended in 4 ml RIPA buffer (25 mM HEPES‐NaOH pH 7.6, 150 mM NaCl, 1 mM EDTA, 1% Triton X‐100, 0.1% SDS, 0.1% Na‐deoxycholate, 1 mM PMSF, Roche cOmplete Protease inhibitor without EDTA). Fragmentation of chromatin was done either by MNase treatment (Sigma, Cat. No. N5386) or by sonication shearing using the Covaris S220 instrument. To obtain similar digestion degree, stage 5–8 embryos were digested using 0.9 units MNase/g embryo and stage 13–15 embryos were digested using 2.9 units MNase/g embryo at 37°C for 30 min, shaking. Reaction was stopped by the addition of EDTA to a final concentration of 10 mM. Additional mechanical shearing was done by passing lysate through a 27G needle 15 times. Alternatively, sonication was performed at 100 W Peak Power, 20% Duty Factor, and 200 Cycles/Burst for 20 min. Soluble chromatin was retrieved by centrifugation at 15,000 g for 15 min at 4°C. Chromatin fragment size distribution was evaluated on a Bioanalyzer (Agilent).

ChIP

1–2 μg soluble chromatin was used as input for each ChIP. Pre‐cleared soluble chromatin was incubated with antibody in RIPA buffer overnight and retrieved by incubation with 50% slurry mix of protein A+G (1:1) sepharose beads. Reversal of crosslinking was done by an overnight incubation, shaking at 65°C, followed by an incubation with 1 μg RNase for 30 min at 37°C and with 0.1% SDS/1 μg proteinase K for 1.5 h at 55°C. DNA was then purified by phenol–chloroform using MaXtract tube (Qiagen, Cat. No. 129046). To evaluate efficiency of ChIP, qPCR was performed using primers listed in Table EV3. Libraries for sequencing were prepared using the NEBNext Ultra II DNA Library Prep Kit for Illumina (New England Biolab, E7654).

ChIP‐seq data analysis

Sonicated single‐end ChIP‐seq reads were aligned to the reference genome (dmel release 6) using Bowtie2 (version 2.2.9) and were filtered using SAMtools with parameter ‐q 2. MNase‐based paired‐end ChIP‐seq reads were concordantly aligned using Bowtie2 with distinguishing sub‐ (parameters: ‐I 10 ‐X 130) or mono‐nucleosomal (‐I 130 ‐X 220) fragments and were filtered using SAMtools with parameter ‐q 12. Reads were processed and normalized (to total number of reads and to input) by Homer Software package 38. Peak calling was performed on pooled replicates by Homer findPeaks tool with parameters ‐style histone ‐size 1000 ‐F 2 for sonicated H4K16ac and H3K36me3; ‐size 1000 ‐F 2.5 for MNase MOF and H4K16ac; ‐size 2000 ‐F 2 for MNase MSL2 mono‐nucleosomal fragments; or with ‐style factor ‐size 200 ‐F 6 for MSL2 sub‐nucleosomal fragments. Distance to MSL2 peaks was calculated between middle of the region/gene and the nearest peak.

Average composite plots were generated from input normalized bedgraph coverages, which were centered at sites and finally replicates were averaged. MNase MOF, H4K16ac, and MSL2 sub‐ or mono‐nucleosomal coverages were centered on HAS/PionX sites in a 4 kb window and mean ± 95% confidence intervals across sites were visualized. Sub‐ or mono‐nucleosomal peak/region centered plots were generated in 4 kb or 10 kb windows, respectively. For correlation analysis at MSL2 peaks, ChIP enrichment was averaged in a 400 bp or 4 kb window for MSL2 sub‐ or mono‐nucleosomal profiles, respectively, and Pearson correlation coefficient was calculated across samples and replicates. De novo motifs on MSL2 sub‐nucleosomal peaks were identified using meme (version 5.0.2) with parameters ‐nmotifs 1 ‐dna ‐revcomp ‐mod zoops ‐maxw 18. Sonicated H4K16ac and H3K36me3 coverages were further scaled between the 1^st and 99^th percentile in 10 kb windows surrounding TSS and TTS or in a 40 kb window surrounding MSL2 mono‐nucleosomal regions.

For genic analysis, MNase MOF and H4K16ac, sonicated H4K16ac and H3K36me3 ChIP‐seq, and input reads were counted over exons for each gene. Counts were normalized to TPM (similarly to RNA‐seq). Sonicated H4K16ac and H3K36me3 were also offset normalized for ChIP and input independently using csaw Bioconductor package (version 1.12.0). Finally, normalized counts were log2 transformed, input was subtracted, and replicates were averaged. These genic ChIP enrichments were used for calculating Pearson′s correlation coefficients or for k‐means clustering (k = 3) to define high, moderate, and low enrichment levels.

Author contributions

PBB conceptualized the study. KP and TS performed experiments. TS performed bioinformatics analyses. KP, TS, CR, and PBB analyzed data and prepared article.

Conflict of interest

The authors declare that they have no conflict of interest.

Supporting information

Appendix

Expanded View Figures PDF

Dataset EV1

Dataset EV2

Dataset EV3

Table EV1

Table EV2

Table EV3

Review Process File

EMBO Reports (2019) 20: e48138

Data availability

Datasets are available with the GEO number GSE127177 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE127177).