Genomic context-dependent histone H3K36 methylation by three Drosophila methyltransferases and implications for dedicated chromatin readers

Abstract

Methylation of histone H3 at lysine 36 (H3K36me3) marks active chromatin. The mark is interpreted by epigenetic readers that assist transcription and safeguard chromatin fiber integrity. In Drosophila, the chromodomain protein MSL3 binds H3K36me3 at X-chromosomal genes to implement dosage compensation. The PWWP-domain protein JASPer recruits the JIL1 kinase to active chromatin on all chromosomes. Because depletion of K36me3 had variable, locus-specific effects on the interactions of those readers, we systematically studied K36 methylation in a defined cellular model. Contrasting prevailing models, we found that K36me1, K36me2, and K36me3 each contribute to distinct chromatin states. Monitoring the changing K36 methylation landscape upon depletion of the three methyltransferases Set2, NSD, and Ash1 revealed local, context-specific methylation signatures. Each methyltransferase governs K36 methylation in dedicated genomic regions, with minor overlaps. Set2 catalyzes K36me3 predominantly at transcriptionally active euchromatin. NSD places K36me2/3 at defined loci within pericentric heterochromatin and on weakly transcribed euchromatic genes. Ash1 deposits K36me1 at putative enhancers. The mapping of MSL3 and JASPer suggested that they bind K36me2 in addition to K36me3, which was confirmed by direct affinity measurement. This dual specificity attracts the readers to a broader range of chromosomal locations and increases the robustness of their actions.

Graphical Abstract

Figure 1.

H3K36 modifications and HMTs mark distinct chromatin states in male Drosophila cells. (A) Genome browser profiles for representative Drosophila chromosomes 2R and X MNase ChIP of K36me1/2/3, the K36me3 reader proteins JASPer and MSL3, and the HMTs NSD and Ash1. “ePol” refers to the signal generated by an antibody against RBP1-S2ph, as a proxy for RNA-polymerase-S2ph-interacting Set2. The Ash1 profile was taken from [41]. The nine-state ChromHMM (modENCODE) is color-coded and explained in panel (C). (B) ChromoMaps representing steady-state enrichment of K36me1/2/3, HMTs, and K36me3 readers in 10-kb genomic bins for chromosomes 2R and X. Scale bars are different for each chromoMap to facilitate visualization of changes. The individual scale values for each state and chromosome are shown in Supplementary Fig. S1. Tracks representing H3K9me2 peaks and annotated genes serve as a reference for mappable PCH domains and transcribed chromatin, respectively. (C) Chromatin state enrichment (nine-state ChromHMM [2]) for K36me1/2/3, HMTs, and K36me3 readers. Published ChIP-seq/CUT&RUN profiles of histone modifications and other chromatin proteins were hierarchically clustered to highlight differences between chromatin states. (D) Genomic features either marked by all K36me1/2/3 domains (top) or filtered for the strongest 10% bottom.

Figure 2.

Effect of HMT depletion on H3K36 methylation bulk levels reveals distinct HMT dependences in male cells. (A) K36me1/2/3 levels in whole cell extracts from S2 cells were determined by quantitative western blotting using specific antibodies. Cells were treated with RNAi against Ash1, NSD, Set2, NSD + Set2 (DKD), or NSD + Set2 + Ash1 (TKD). An irrelevant GST RNAi served as control. Representative blots are shown in Supplementary Fig. S2C. Source data for all blots can be found in Zenodo repository (see the “Data availability” section). Values were normalized to histone H2AV signals on the same membrane and are represented as fraction relative to GST RNAi, which was run on same blot. Each dot represents an independent biological replicate. Calculated ANOVA P-values (null hypothesis: difference between means = 0) are presented for each antibody. (B) Representative IFM images for α-H3K36me1/2/3 in S2 cells treated with RNAi against GST (control), Ash1, NSD, or Set2. The scale bar is 5 μm. (C) Quantification of IFM images [n ∼ 500 nuclei from first biological replicate, shown in panel (B)]. ANOVA followed by post-hoc Tukey HSD was performed to identify groups with significantly different mean relative to GST RNAi (****P < .001).

Figure 3.

HMTs act on largely distinct genomic regions. (A) Schematic depicting Z-score analysis workflow. Pairwise Z-score scatter plots representing correlations among changes in ChIP signal upon RNAi of HMTs with respect to control RNAi in genome-wide 5-kb bins. Only bins overlapping at least one of K36me1/2/3 peaks in any RNAi condition were included. A negative Z-score indicates a reduction relative to control, while a positive Z-score indicates an increase. Schematic was generated with BioRender.com. (B–D) Pairwise Z-score scatter plots for K36me1/2/3 upon RNAi of indicated factor (right). The color overlaid on the scatter indicates the local density of points. Pearson correlations are provided for each pair. Corresponding chromoMaps representing regions of significant difference in K36me3/2/1 signal for indicated RNAi conditions as derived from csaw analysis for chromosome 3L at 2-kb resolution (left). The color of the regions (as indicated by the common scale in Fig. 3B) represents log₂-transformed value of number of normalized reads in RNAi condition relative to control condition. Control K36me3/2/1 signal overlaid above chromoMaps aids interpretation of relative changes. K9me2 peaks and gene annotations are provided for reference.

Figure 4.

A gene-centric view of the K36 methylation landscape. (A) Clustered heatmaps of gene body-averaged ChIP signal for K36me1/2/3 indicated on the right and RNAi condition indicated on the left. Only genes overlapping at least one of K36me1/2/3 peaks in any RNAi condition (n = 10 477) were used for clustering. Clusters are numbered 1–12 as indicated above the heatmap. The track below the heatmap represents the manual grouping of clusters to define superclusters that display similar patterns, resulting in SC-I (n = 6619), SC-II (n = 1710), and SC-III (n = 1873). (B) Genome browser profiles of representative genes from each supercluster along with the number of grouped genes. Supercluster II was further classified into euchromatic (IIA) or heterochromatic (IIB) based on overlap with H3K9me2 peaks. RNAi condition and immunoprecipitation target indicated on the left; nine-state chromatin states (as in Fig. 1C) serve as a reference. Representative genes shown for SC-I, SC-IIA, SC-IIB, and SC-III lie within clusters 1, 8, 7, and 9, respectively. (C) Density plot of Z-scores representing change in K36me2 signal upon Ash1 RNAi [41] for superclusters defined in Fig. 4A. (D) Genome browser profiles of supercluster 3 representative gene highlighting effect of Ash1 RNAi on K36me2. RNAi condition and immunoprecipitation target indicated on the left; chromatin states (as in Fig. 1C) shown for reference. (E) Cumulative plots for each cluster representing gene body relative distributions of K36me1/2/3. 1-kb regions centered around TSS and TTS are unscaled, while the rest of the gene body was scaled to 500 bins. SC-0 genes lack any detectable H3K36 methylation and serve as a reference for zero signal/baseline. This is also represented by the horizontal dotted line in all individual plots. Only genes of minimum length of 1500 bp were included in the analysis.

Figure 5.

Gene clusters defined by HMT methylation patterns are correlated to different genic features. (A) Violin plots representing average ChIP signal for 2-kb windows around HMT gene body peaks for NSD, ePol, and Ash1 within superclusters defined in Fig. 4A. Supercluster 0 represents randomly sampled genes (n = 3000), which lack any detectable K36 methylation, and serves as reference for zero signal/baseline. (B) Boxplots showing the proportion of introns (sum of length of introns/total gene length), transcriptional activity (denoted by log₁₀-transformed RNA-seq TPM values), and log₁₀(gene length) for gene superclusters. (C) Proportion of tissue-specific/invariant genes for gene superclusters based on FlyAtlas expression data for 25 tissues.

Figure 6.

Robust binding of K36me3 readers above threshold K36me3 density. (A) Heatmap of gene body average ChIP signal for JASPer in indicated RNAi condition ordered according to heatmap from Fig. 4A. Note that supercluster III was excluded as it has very few genes bound by JASPer. (B) Genome browser profiles for three distinct genomic loci highlighting different responses of JASPer to Set2 RNAi and NSD RNAi within indicated clusters. RNAi condition and immunoprecipitation target are indicated to the left and right, respectively. The dotted line in K36me3 ChIP in Set2/NSD RNAi represents an arbitrary threshold below which strong reader binding reduction is observed, based on empirical observation of genome browser tracks. (C) Proportional change density plots representing direction and magnitude of change in gene body average ChIP signal for K36me2/3 along with reader JASPer for indicated representative clusters. Dashed line indicates the median value of each curve. (D) Equilibrium binding between recombinant JASPer and unmodified, dimethylated, or trimethylated nucleosomes, respectively, determined using MST. Error bars represent the standard deviation from the mean values obtained from n = 2 experiments. Calculated dissociation constants are indicated within parentheses.