Histone modification distribution along Y loops

In Drosophila spermatocytes, as part of the spermatogenesis transcription program, three extraordinarily large genes on the Y chromosome become activated and expand into the nucleoplasm as visible Y loops. These loops fill the central nucleoplasm and can be clearly distinguished from the autosomal chromatin that is restricted to the nuclear periphery. Each loop comprises a single transcription unit, several megabases in length, and these Y loops provide a useful model for the visualisation of actively transcribed chromatin [41, 42]. Here we are interested in the distribution of histone modifications within these transcription loops to investigate how histone modification is linked to variation in chromatin structure. For this, we first examined the distribution of one of the key modifications of active chromatin, H3K4me3 [43] by stimulated emission depletion (STED) microscopy [44] on isolated primary spermatocytes immuno-labelled with pan-histone and H3K4me3 specific antibodies (Fig. 1A; the pan-histone antibody labels the core histones and H1 and we will refer to it hereafter as simply histone labelling). To aid overall visual inspection, signal to noise was improved by denoising with N2V (see Methods), but quantifications were performed on the raw data. H3K4me3 labelling was most apparent in chains of clusters close to the nucleolus, with more sparse labelling of loops in the central nucleoplasm (Fig. 1). Some long Y loop regions lack apparent H3K4me3. Notably, the H3K4me3 labelling intensity generally does not directly follow the histone labelling density and the peaks of H3K4me3 intensity overlay rather weak histone labelling, suggesting that the H3K4 mark is associated with relatively disperse chromatin (quantification is presented below). Although the H3K4me3 mark has been specifically linked to transcription initiation [45], the Y loop labelling is clearly not tightly restricted to the sites of initiation which are thought to be close to the Y chromosome mass associated with the nucleolus [36]. Instead, the chains of H3K4me3 labelled clusters emanating from the nucleolus indicate that the H3K4me3 mark extends well into the transcription unit. This and the labelling further along the body of the Y loop transcription units would be consistent with the recent finding that H3K4me3 has a distinct role in transcriptional pause-release and elongation rather than transcriptional initiation [46].

We then examined the distribution of H3K36me3, a well characterised marker for active transcription along gene bodies [47]. We expected H3K36me3 to be enriched along Y loops, since they function as single transcription units [41, 42]. However, we only found high signals at a prominent “spike” region protruding from the nucleolus into the nucleoplasm (Fig. 2A) and rather sparsely along fibres in the nucleoplasm (Fig. 2B). Figure 2C shows an example of H3K36me3 labelling peripheral to the major histone density along the fibre, suggesting that H3K36me3 is associated with relatively decondensed chromatin. The punctate H3K36me3 signals along fibres in the nucleoplasm may indicate the dynamics of the H3K36me3 mark. The observation that the accumulation of active chromatin marks, especially H3K4me3, is highly restricted along Y loops prompted us to determine if we could detect a mark that is associated with inactive transcription. Hence, we turned to H3K27me3, which is a well-established marker of the repressed chromatin state [48]. Surprisingly, STED microscopy revealed a prominent clustered labelling of H3K27me3 along many, but not all, Y loop regions (Fig. 3A). Comparison of the H3K27me3 and histone labelling densities (Fig. 3B) suggests that H3K27me3 is associated with both condensed and relatively decondensed chromatin.

Fig. 3

H3K27me3 distribution along Y loops. STED IF microscopy of histones (magenta and middle panels) and H3K27me3 (green and right panels). (A) overview. Ellipses indicate fibres without detectable H3K27me3. Maximum intensity projection of 10 slices, Δz = 0.5 μm. (B) shows a Y loop fibre with H3K27me3 clusters in decondensed chromatin regions. Maximum intensity projection of 4 slices, Δz = 0.5 μm. Scale bars, 1 μm

Direct comparison between H3K27me3 and H3K4me3 labelling

Each Y loop represents a single gene that is transcribed throughout its whole length and the transcriptional processes must be coordinated. Thus, the relative distributions of marks for different transcriptional activities can be expected to be informative about their mechanism and dynamics. Co-labelling of H3K4me3 and H3K27me3 and observation by laser scanning confocal (LSC) and STED microscopy showed that labelling of these two marks is mutually exclusive (Fig. 4). This occurs however in various arrangements, from alternating clusters to long exclusive fibre sections of either mark (Fig. 4B). This indicates that the transcriptional dynamics along the Y loops are highly variable including a high turn-over of histone modifications and chromatin rearrangements. In some instances the opposing histone marks occur along the same section of the fibre, not overlapping however, but with H3K27me3 more central and H3K4me3 more distal to the main fibre axis (Fig. 4E). This suggests a mechanism for transcription off the main fibre, from which low density chromatin might be looping out and potentially spread to form the more extended transcriptionally active chromatin.

Fig. 4

H3K4me3 and H3K27me3 distributions. A)&B) LSCM of H3K4me3 and H3K27me3 labelled spermatocytes. The ImageJ smooth filter was applied. (A) overview, maximum intensity projection of 4 slices (Δ z = 300 nm). Bar, 3 μm. (B) Detail of boxed region in A) showing a fibre emanating from the nucleolus and carrying both histone marks. Single optical section. C)-E): N2V denoised STED images. (C) overview. D)&E) details, 3 optical sections (Δ z = 450 nm). (D) shows fibres in the same nucleus almost exclusively stained either for H3K4me3 or H3K27me3. Images are rotated by 90° relative to C). (E) Detail from another cell showing H3K27me3 clusters along a fibre with locally associated H3K4me3. Scale bars, 1 μm

Quantification of histone modification distributions

To investigate the link between chromatin architecture and active and inactive chromatin marks along Y loops we first measured the distances between histone clusters on long (µm range) fibre sections enriched either with H3K4me3 or H3K27me3 (Fig. 5A-C). Since higher chromatin densities are often thought to impede efficient transcription and also have been shown to stimulate PRC2 activity [49] and lower transcriptional activity corresponds with high H3K27me3 levels, we expected fewer clusters and more extended chromatin in regions displaying the active H3K4me3 mark. Strikingly, there was no difference in the distance between histone clusters, with a mean of 276+-164 nm for H3K4me3 decorated fibres versus 271 +-157 nm for fibres enriched with H3K27me3 when applying a low tolerance factor for peak detection, and 122+-52 nm vs. 126+-58 nm with a low stringency.

This shows that the basic clustered chromatin organization is independent of the studied histone marks, and even regions carrying active marks contain chromatin clusters suggesting that this organization is permissive for transcription.

To obtain a more direct measure of the relationship of H3K4me3 and H3K27me3 with chromatin densities we calculated their respective cross correlations (Fig. 5A, B, D, E). We find a stronger direct correlation of chromatin with H3K27me3, while H3K4me3 is usually found displaced from the chromatin enrichments (median absolute value of shifts (displacement): 50 nm for H3K27me3 and 200 nm for H3K4me3). The t-test for the absolute shift gives a p-value of 0.0049.

Thus, while these two histone marks do not appear to affect chromatin cluster frequencies along Y loops, their differential enrichment with respect to chromatin densities suggest transcription dynamics with inactive transcription or pausing at clusters and transcription progression at their periphery [50]. Pausing could allow e.g. DNA repair or RNA processing, which might form a particular globular chromatin arrangement (Fig. 5F).

Fig. 5

Quantitative analysis of the relationship between histone modifications and chromatin densities. Corresponds to data as shown in Figs. 1 and 3. A)&B) right panels show example intensity profiles along the indicated line in the left (merge) panels. X-axis of the plots gives the distance in µm, normalized intensities are along the y-axis. (A) H3K4me3 vs. histones, example from Fig. 1. (B) H3K27me3 vs. histones. (C) The scheme indicates the expected distribution of chromatin clusters in H3K4me3 vs. H3K27me3 enriched Y loop regions, and the distributions as observed by measuring the distances between histone intensity peaks. Quantifications are given in the box plots; left, for a tolerance factor for peak detection of 0.17, right, for a tolerance factor of 1.0 to avoid a bias caused by false positive or negative detections. Number of measured peak distances are 163, 266, 54 and 102. Total measured Y loop lengths are 22.5 μm (H3K4me3) and 37.76 μm (H3K27me3). D)&E) The cross correlation of fluorescence signal intensities along selected loops was computed for H3K27me3 vs. histones and H3K4me3 vs. histones respectively. (D) shows the overall displacement distributions in box plots. (E) Heat maps indicate the cross correlation values between the lagged signals of respective cluster intensities per nucleus. The x-axis shows the lag distances, the samples are stacked along y. (F) The scheme indicates the shift between histones and the respective histone marks, as well as their mutually exclusive organization (see Fig. 4). The question mark indicates our interpretation of the results that H3K27me3 enrichment at chromatin clusters is correlated with transcriptional pause, which would allow e.g. processing of the transcribed RNA or DNA repair to occur. H3K4me3 enrichment at the periphery of the clusters could indicate pause release into productive elongation

Different chromatin states have different structural organisation

To extend this analysis and to assess the morphology of chromatin in different activity states we took advantage of the higher resolution of STORM and imaged Y loops single labelled for H3K36me3, H3K4me3 or H3K27me3 (Fig. 6).

Fig. 6

Single-labelled super-resolution STORM images of primary spermatocyte nuclei. (A&B) labelled for H3K36me3, (C&D) labelled for H3K4me3 and (E&F) labelled for H3K27me3. The scale bars for left hand panels are 5 μm. Zoomed in images of yellow box areas are shown on the right. The scale bars in the zoomed in images are 500 nm

The chromatin clusters labelled for H3K36me3 show heterogenous morphology. The clusters appear to be larger than global histone labelled Y loop clusters [36], with more “fuzzy” edges, suggesting the presence of looping DNA emanating from the edges of the clusters. Interestingly, there appear to be clear runs of Y loop chromatin which have continuous enrichment for H3K36me3, as the clusters appear approximately 100 nm apart from one another, the same spacing as the general histone labelled Y loop chromatin. We previously estimated that the average chromatin cluster along the Y loops could contain a median of 54 nucleosomes. In terms of sequence this means that a cluster could be formed of approximately 8.2 kb, and therefore continuous clusters along a fibre modified for the same histone modification implies that very large domains of H3K36me3 modified chromatin are present along the Y loops.

The chromatin clusters of H3K4me3 visually appear much smaller when compared to the Y loops labelled with the histone antibody, and the Y loops labelled for H3K36me3. The H3K4me3 clusters also appear to be more disconnected from each other (Fig. 6D) suggesting that there may be intervening clusters of nucleosomes that do not have the H3K4me3 modification, or that H3K4me3 is enriched in the regions in between the chromatin clusters of the Y loops. This suggests that there are not clear continuous domains of H3K4me3 along the Y loops as was the case for H3K36me3.

H3K27me3 appears to represent mostly large clusters along the Y loops, visually appearing similar to H3K36me3. However, the H3K27me3 labelled chromatin clusters appear to have sharper boundaries suggesting they lack the same extent of decondensed “looping” structure that surrounds the H3K36me3 clusters (Fig. 6D). There are also continuous runs of clusters of chromatin along fibres enriched for H3K27me3, implying the existence of large domains (Fig. 6F). We note that H3K27me3 is not exclusively found labelling larger clusters, but rather can be seen labelling a wide variety of different cluster sizes along the Y loops.

The chromatin clusters labelled by H3K4me3, H3K36me3 and H3K27me3 visualised by STORM therefore showed unique morphologies.

Quantification of structural organisation

In order to better characterise the structural differences of chromatin modified for active and silencing histone modifications, clustering analysis was performed as described previously [36]. This analysis measures the median width (full width half maximum; FWHM) of the clusters identified, enabling the comparison of clusters of chromatin in different states (Fig. 7). Due to the variable numbers of clusters used in each analysis, the frequencies were normalised, and presented as a ‘probability’ value instead of raw frequency counts on the histogram.

Fig. 7

Histograms showing the cluster widths (FWHM) of Y loop clusters labelled with different histone modifications. Total histone labelled compared to (A) H3K27me3, (B) H3K4me3, (C) H3K36me3, and (D) H3K27me3 compared to H3K36me3. For A), B), and C), the red line indicates the median cluster width of total histones, and the black line indicates the median cluster widths of the histone modification. For D), the red line indicates the median cluster widths of H3K27me3 and the black line indicates H3K36me3.

The histone labelled chromatin cluster widths are as described previously [36], with a median of 52 nm, and an interquartile range (IQR) between 44 and 61 nm. H3K27me3 labelled chromatin had a median width of 59 nm, and an IQR between 49 and 72 nm. Compared to the general histone label, the H3K27me3 modified chromatin showed a bias towards larger clusters, with approximately a 14% increase in median width (P = < 0.05). It is important to note that all the modified chromatin cluster sizes were almost exactly within the range of the general histone data, and so the cluster sizes are not larger or smaller than what can be seen in the general histone data. However, the skew of the graphs was different, indicating a bias towards finding different chromatin cluster sizes depending on histone modification. The H3K36me3 labelled chromatin clusters had a median width of 61 nm, and an IQR between 51 and 72 nm. This represents an 18% increase in median width compared to global Y loop histone labelling (P = < 0.05). The H3K4me3 appeared visually much smaller by eye, and indeed the median width was 47 nm, represented a 9% decrease in median size (P = < 0.05). The IQR was between 40 and 55 nm. This suggests that chromatin associated with different histone modifications represent different chromatin structures at the level of nucleosome clusters.

The clusters are in mutually exclusive chromatin state domains

To more closely evaluate the domains of chromatin state along the Y loops, different histone modifications representative of active and inactive state, H3K36me3 and H3K27me3 respectively, were co-labelled, and imaged using STORM (Fig. 8).

Fig. 8

Dual-labelled super-resolution STORM images of the Y loops: H3K27me3 and H3K36me3 different examples are shown from three different nuclei labelled for H3K27me3 (magenta) and H3K36me3 (green). The scale bars are 1 μm

Although both modifications can be seen along the same chromatin fibre nearby to one another, surprisingly, the chromatin clusters themselves appear to be mostly mutually exclusive for either modification. There is very little visual evidence of chromatin clusters that share two of the different types of modification together.

First the apparent exclusivity hypothesis was tested and confirmed via the computation of the mark connection function, which demonstrated a lack of overlap of H3K27me3 and H3K36me3 localisations (Fig. 9C). This analysis also suggests that at distances smaller than 200 nm there is a distinct repulsion.

To further quantify the exclusivity, both H3K27me3 and H3K36me3 were clustered, and the % of localisations within 50 nm radius of the cluster centres (as an approximation of average cluster width) was quantified (Fig. 9). A median of 7% overlap existed between clusters of H3K36me3 and localisations of H3K27me3, demonstrating a clear exclusion of the two along a single Y loop chromatin fibre.

Clustering H3K27me3, a median of 2% of H3K36me3 localisations were within 50 nm of the cluster centres.

Fig. 9

The exclusivity of H3K36me3 and H3K27me3 localisations. (A) clustering the H3K36me3 localisations and quantifying the % of overlap of H3K27me3, and (B) clustering the H3K27me3 localisations and quantifying the overlap of H3K36me3 localisations. (C) mark connection function (magenta) vs. randomly simulated data (cyan) was performed on H3K27me3 and H3K36me3 localisations along the Y loops to give confidence to the results with a more untargeted analysis. Values around zero show the chance of overlap between H3K27me3 and H3K36me3 localisations is random, those above zero show an attraction rule and those below zero show a repulsion rule. A strong repulsion between H3K27me3 and H3K36me3 is shown at distances below 200 nm

This analysis makes some baseline assumptions, therefore in order to provide more confidence in this result, a general exclusivity quantification was undertaken using the mark connection function to further investigate this phenomenon, which also demonstrated a lack of overlap of H3K27me3 and H3K36me3 localisations (Fig. 9C). This analysis shows that at the cluster size level (< 200 nm) there is a distinct repulsion between H3K36me3 and H3K27me3, but above this distance is overlapping with random distribution, matching the data showing that clusters of both chromatin state can be next to one another, but not overlapping in one cluster. This has interesting implications for the mechanisms of how histone modifications spread in domains across chromatin, as this observation implies that individual chromatin clusters made up of many nucleosomes may form a “unit” of chromatin that can have different chromatin states, but that different chromatin states are not contained within the same nucleosome cluster. We considered the possibility that steric hindrance between antibodies might produce artifactual exclusivity but, given the size of the clusters and the minimal overlap, we consider this unlikely.

There is evidence of regular switching between an active to an inactive chromatin state along one single chromatin fibre. In Fig. 8A, from left to right there is a continuous run of approximately 6 chromatin clusters labelled with H3K36me3, then a domain of a few clusters of H3K27me3, which switches back to a domain of H3K36me3 again. As previously described, this supports the findings that there are whole domains of different chromatin states along one fibre. However, at STORM resolution individual clusters along the Y loop are able to be visualised, and the extent of the proximity of these domains, and their apparent sharp borders at chromatin cluster edges is striking.

This interspersion of different chromatin states can also be seen for H3K27me3 and H3K4me3 along Y loop chromatin (Fig. 10).

Fig. 10

Dual-labelled super-resolution STORM images of the Y loops: H3K27me3 and H3K4me3. (A-C) different examples are shown from three different nuclei labelled for H3K27me3 (magenta) and H3K4me3 (green). C) shows a dense region of chromatin on the periphery of the nucleolus thought to be near the origin point of the Y loops, and is strongly enriched for both H3K27me3 and H3K4me3. The scale bars are 1 μm

H3K36me3 is associated with elongating polymerase

To assess the functional association of chromatin state with active transcription, RNA Polymerase II with the phospho-Ser2 modification on the C-terminal domain repeats (RPol-Pser2) that is associated with active elongation [51, 52] was immunolabelled alongside both H3K27me3 and H3K36me3 (Fig. 11).

Fig. 11

Dual-labelled super-resolution STORM images of primary spermatocyte nuclei labelled for histone marks with RPol-PSer2. (A) H3K36me3 (magenta) and RPol-Pser2 (green) and (B) H3K27me3 (magenta) and RPol-Pser2 (green). Scale bars are 2 μm

RPol-Pser2 is still seen in association with chromatin labelled with H3K27me3, however, RPol-Pser2 appeared to be more closely associated with chromatin labelled with H3K36me3. To quantify this, the nuclear interior region of several cells labelled with H3K27me3 or H3K36me3 and RPol-Pser2 were cropped out. We computed the k-nearest neighbour distance for H3K27me3 and H3K36me3 localisations to RPol-Pser2 for k = 21, and the respective empirical cumulative distribution functions (Fig. 12A). Furthermore, we compared the median distance and the 90% quantiles for samples in the two cases (Fig. 12B). On average it was seen that the RPol-PSer2 signal was significantly closer to the H3K36me3 signal than to H3K27me3 along the Y loops (P = < 0.05). A mark connection function confirmed this (Fig. 12C). A robustness test using a lower knn [5], and 50% quantiles confirmed the same trend of H3K36me3 having a closer association with RPol-Pser2 (see Supplementary Information). This observation, on the one hand, provides image-based support linking H3K36me3 with active transcription, with H3K27me3 acting as a comparator. On the other hand, this data shows that H3K36me3 is not simply associated with individual nucleosomes where transcription is occurring as H3K36me3 is found up to several hundred nanometers distant from the RPol-Pser2 signal. We interpret this as perdurance of the H3K36me3 mark.

Fig. 12

Quantification of the average overall distance between RPol-Pser2 and H3K36me3, or H3K27me3. (A) individual ROIs containing the histone modifications and RPol-Pser2 along sections of the Y loop are plotted as individual lines using the empirical distribution function (ECDF); H3K36me3 in cyan, H3K27me3 in magenta. (B) the ECDF values at 90% were pooled for both H3K27me3 and H3K36me3 and plotted on a box and whiskers plot showing that H3K36me3 was closer on average to RPol-Pser2 signal. (C) the mark connection function was used as an alternative method in a more unbiased fashion, and also showed that RPol-Pser2 is closer to H3K36me3.

In specific examples of the association between H3K36me3 labelled Y loop chromatin and RPol-Pser2 it can often be seen that RPol-Pser2 is associated on the edge of a domain of H3K36me3 (Fig. 13). This would align with a function of the enzyme SET2 that is responsible for catalysing the H3K36me3 modification, which may modify a region of chromatin as polymerase transcribes it, and the modification remains for a period of time behind the elongating polymerase. This is schematically represented in Fig. 13D.

Fig. 13

Dual-labelled super-resolution STORM images of the Y loops labelled for H3K36me3 (magenta) and RPol-Pser2 (green). (A-C) different examples are shown from three different nuclei. (D) schematic of selected area from (C), indicated with a yellow box. The scale bars are 1 μm in (A) and (B), and 500 nm in (C)