Chromatin remodeling complexes regulate genome architecture in Arabidopsis

Abstract

In eukaryotes, three-dimensional (3D) chromatin architecture maintains genome stability and is important in regulating gene transcription. However, little is known about the mechanisms by which diverse ATP-dependent chromatin remodeling complexes regulate the 3D chromatin structure in plants. We examined the 3D chromatin structure within the ATPase subunit of the SWI/SNF, ISWI, INO80, and CHD remodeling complexes in wild-type (WT) and mutant Arabidopsis thaliana plants by combining high-throughput sequencing with in situ Hi-C, the enrichment of histone marks, nucleosome density, and gene expression. We found that compartment regions switched and compartmental strength was significantly weakened in all four enzyme mutants. Chromatin remodeling complexes differentially regulated the nucleosome distribution pattern and density within the switching compartments. Alterations of nucleosome distribution pattern and density were associated with a reduction in H3K27me3 levels in the chromatin remodeling enzyme mutants and led to compartment switching. Our data show that chromatin remodeling complexes regulate the linear nucleosome distribution pattern and density to promote H3K27me3 deposition, which in turn regulates 3D chromatin structure.

Chromatin remodeling complexes regulate the nucleosome distribution pattern and density to maintain a high H3K27me3 level that prevents compartment switching.

Introduction

Chromatin remodeling complexes are conserved in eukaryotic cells. They participate in the regulation of gene expression, DNA replication, and DNA repair by controlling nucleosome dynamics (Clapier and Cairns, 2009; Clapier et al., 2017; Lorch and Kornberg, 2017). There are four distinct families of chromatin remodeling complexes, based on ATPase subunit composition: switch/sucrose non-fermentable (SWI/SNF), imitation switch (ISWI), inositol requiring 80 (INO80), and chromodomain helicase DNA-binding (CHD) complexes (Hargreaves and Crabtree, 2011; Narlikar et al., 2013; Bartholomew, 2014). Chromatin remodeling complexes utilize the energy derived from ATP hydrolysis to mediate nucleosome repositioning, ejection, or exchanging of histone variants (Narlikar et al., 2013; Clapier et al., 2017), thereby controlling the packaging and unpackaging of specific DNA elements into nucleosomes (Chen et al., 2017; Clapier et al., 2017; Gioacchini and Peterson, 2017). Although each chromatin remodeling complex regulates different biological processes, they share basic properties, such as their preference to bind to nucleosomes rather than DNA, and the presence of a single catalytic ATPase (Chen et al., 2017; Clapier et al., 2017).

In mammalian cells, ISWI and SWI/SNF selectively regulate the binding of distinct transcription factors to their target sites (Barisic et al., 2019). Mutations in the catalytic ATPase of ISWI or SWI/SNF alter accessibility to non-overlapping sets of cis-elements bound by different transcription factors (Mathelier et al., 2016; Barisic et al., 2019). For example, binding by the transcription factor CCCTC-binding factor (CTCF) is significantly reduced in SNF2H (catalytic ATPase of ISWI)-knockdown cells (Wiechens et al., 2016). CTCF binds to specific convergent cis-elements and regulates mammalian cell 3D chromatin structure (Lieberman-Aiden et al., 2009; Phillips and Corces, 2009; Rao et al., 2014); this produces a hierarchical pattern including chromosome territories, A/B compartments, topologically associated domains (TADs), long-range interactions, and differences in nucleosome density (Sexton and Cavalli, 2015; Dogan and Liu, 2018). TADs are eliminated by targeted degradation of CTCF protein or mutated cis-elements of CTCF, but these have no effect on A/B compartments (Tang et al., 2015; Nora et al., 2017). SNF2H loss-of-function increases linker DNA length, causing a genome-wide decrease in CTCF binding, affecting TADs and long-range loop formation (Barisic et al., 2019). CTCF and cohesin binding, however, are increased in Chd4 (belongs to CHD family) conditional knockout cells (Goodman et al., 2020).

Cohesin is a ring-like protein complex that regulates 3D chromatin structure (Merkenschlager and Nora, 2016; Zheng and Xie, 2019; Zhu and Wang, 2019). Cohesin binding sites significantly overlap with those of CTCF (Stedman et al., 2008; Merkenschlager and Nora, 2016; Busslinger et al., 2017). Chromatin accessibility increases in mouse brain cells in the absence of CHD4, providing more binding space for CTCF and cohesin (Goodman et al., 2020). Moreover, CHD4 contributes to the regulation of compartmentalization, which is associated with the epigenetic state (Goodman et al., 2020). The 3D chromatin structure, however, is unaffected when Brg1, the catalytic ATPase of SWI/SNF, is knocked out, because CTCF-binding sites are not altered (Barisic et al., 2019). Furthermore, regions bound by the INO80 complex overlap with the enhancer–promoter interaction regions; however, it remains unknown whether or how INO80 regulates 3D chromatin structure (Hsieh et al., 2020).

In general, A or B compartments in both animal and plant cells reflect euchromatin or heterochromatin and are enriched with positive or repressive histone modifications (Lieberman-Aiden et al., 2009; Sexton and Cavalli, 2015; Dogan and Liu, 2018; Zheng and Xie, 2019). Mutations in H1 coding genes cause general chromatin de-compaction and A/B compartment switching across the genome and are associated with changes in levels of H3K36me2 and H3K27me3 (Willcockson et al., 2020; Yusufova et al., 2021). Despite there being no CTCF homolog in plants, A/B compartments and local interaction domains are present (Moissiard et al., 2012; Wang et al., 2015; Liu et al., 2016; Dong et al., 2017; Karaaslan et al., 2020; Ouyang et al., 2020).

H3K27me3 deposited by Polycomb group (PcG) proteins shows specific enrichment within long-range and local interacting regions in plants. PcG mutants have significantly fewer Polycomb-associated domains and an almost complete absence of H3K27me3-related interactions (Feng et al., 2014; Zhou et al., 2017; Huang et al., 2021). Moreover, H3K27me3 marks participate in regulating A/B compartment switching (Zhang et al., 2021). In plants, chromatin remodeling complexes have not been shown to regulate 3D chromatin structure. However, many chromatin remodeling enzymes have been reported to interact with or antagonize PcG proteins and influence H3K27me3 enrichment (Cui et al., 2016; Li et al., 2016; Carter et al., 2018), which is reported to play a role in regulating chromatin 3D structure (Feng et al., 2014; Zhou et al., 2017; Huang et al., 2021). PICKLE (PKL), the catalytic ATPase of CHD, for example, works with CURLY LEAF (CLF), a PcG H3 trimethyltransferase, to promote H3K27me3 deposition (Carter et al., 2018); conversely, BRAHMA (BRM), the catalytic ATPase of SWI/SNF, balances H3K27me3 levels by blocking CLF binding at target regions (Li et al., 2015).

Here, we investigated the role of Arabidopsis thaliana chromatin remodeling complexes in regulating 3D genome architecture. B to A compartment switches were found in four different ATPase mutants associated with reduced H3K27me3 levels. We, furthermore, show that the linear nucleosome distribution pattern and density prevent compartment switching by promoting H3K27me3 deposition.

Results

3D chromatin structure changes in chromatin remodeling mutants

Since SWI/SNF, ISWI, INO80, and CHD are conserved in plants, we chose BRM, CHR11/17 (hereafter referred to as CHRs), INO80, and PKL, the catalytic ATPases of these complexes (Kim, 2019), to further investigate their function. To determine the roles of these chromatin remodeling complexes in regulating plant 3D chromatin structure, we performed in situ Hi-C experiments with two biological replicates in wild-type (WT) and brm, chrs, ino80, and pkl mutant seedlings at 10 days after germination (DAG). In total, we identified an average of 50 million valid interaction pairs for each genotype. A high cis-interaction to total interaction ratio indicated the high quality of the Hi-C data (Supplemental Dataset S1; Lajoie et al., 2015). We generated Hi-C maps (Figure 1, A–D and Supplemental Figure S1), which show frequent interactions along the main diagonals; these are consistent with previously reported Hi-C maps of Arabidopsis and other plant species (Dong et al., 2017; Zhu et al., 2017).

Figure 1.

Mutations in chromatin remodeling enzymes cause rearrangement of chromatin high-order structure. A–D, Relative interaction heatmaps showing differences between WT and brm, chrs, ino80, or pkl mutants. The color bars indicate log₂(normalized interaction). E–H, IDEs show the difference in interaction frequency between WT and mutants on chromosome arms.

To determine the effects of the mutations in chromatin remodeling enzymes, we compared the genome-wide chromatin interactions in mutants to those in WT plants at a resolution of 40 kb (Figure 1, A–D and Supplemental Figure S1). There was a general decrease in interactions in all the mutants, particularly in the chromosome arm regions. To examine chromatin changes quantitatively, we calculated the interaction decay exponents (IDEs), which represent trends in interaction frequency with distance on chromosome arms (Lieberman-Aiden et al., 2009). The IDE of WT plants was 0.89, consistent with previously published IDEs in Arabidopsis (Grob et al., 2014); the IDEs of the brm, chrs, ino80, and pkl mutants were 0.779, 0.825, 0.763, and 0.746, respectively (Figure 1, E–H). Consistent with the Hi-C maps, we found that interactions ranging from 100 kb to 3 Mb were reduced in brm, ino80, and pkl mutants, while relatively mild reductions were exhibited in chrs mutants, indicating relatively mild chromatin packaging compared with other three mutants. Taken together, our data show that mutations in chromatin remodeling enzymes caused a global change in 3D chromatin structure.

Chromatin compartmentalization is weakened in chromatin remodeling mutants

The existence of A/B compartments in human and Arabidopsis has been shown through the calculation of bidirectional PC1 (first principal component) values (Lieberman-Aiden et al., 2009; Grob et al., 2014). We found that mutations in any chromatin remodeling enzyme cause A/B compartment switches at chromosome arms (Figure 2A; Supplemental Figure S2 and Supplemental Dataset S2). Continual A compartment regions were observed in all mutants, especially in chrs and pkl mutants. Furthermore, chromatin compartmentalization strength was reduced in all mutants (Figure 2, B–F). Since compartments of the same type have a higher frequency of contacts than compartments of different types (Lieberman-Aiden et al., 2009), saddle plots explained why the general chromatin interactions were reduced in the four mutants (Figures 1 and 2, B–F).

Figure 2.

Compartments switches and compartmentalization strength are decreased in chromatin remodeling enzyme mutants. A, The tracks show PC1 values generated for each 20 kb genomic segment from Hi-C data on the right arm of chromosome 1 (Chr1). Positive PC1 values indicate A compartments (red) and negative PC1 values indicate B compartments (blue). The tracks below the PC1 values represent the B-to-A switching region (orange) and A-to-B switching region (black). Dashed boxes indicate the switched regions. B–F, Saddle plots of Hi-C data. Saddle plots were calculated using the PC1 values obtained from the Hi-C data in Figure 2A. The numbers at the center of the heatmaps indicate compartment strength, calculated as: (A-A + B-B)/(A-B + B-A) using the mean values from the corners indicated by white boxes. G, Bar plot shows the numbers of switched regions in each chromatin remodeling enzyme mutant. The red or green bars indicate the B-to-A or A-to-B switching regions, respectively. H, Bar plots indicate the distribution of B-to-A compartment switching region length. I, Box plots show the average length of B-to-A compartment switched regions. chrs and pkl mutants have longer switching regions than brm and ino80 mutants (two-sided Mann–Whitney U test). J and K, Enrichment and depletion patterns of epigenetic features in no switching (J) or switching (K) B regions compared with the entire genome (represented in red or blue). The relative density ranges from −1.5 to 1.5. Epigenetic features that were neither enriched nor depleted (two-sided Mann–Whitney U test, P > 0.05) are represented in black.

In total, more B compartments switched to A compartments in brm, chrs, ino80, or pkl mutants; however, the total number of A compartment switching to B compartments was fewer in all mutants (Figure 2G). Moreover, 205 B-to-A compartment switching segments overlapped in the four mutants; however, only 24 A-to-B compartment switching segments were shared (Supplemental Figure S3). These results indicate that distinct chromatin remodeling complexes function in a similar way to prevent B-to-A compartment switching. Around three-quarters of B-to-A compartment switching regions in brm and ino80 mutants were between 20 and 50 kb; however, more than 60% of B-to-A compartment switching regions in chrs and pkl mutants were longer than 50 kb (Figure 2H). According to these data, the average lengths of B-to-A compartment switching regions in chrs and pkl mutants were significantly larger than in brm and ino80 mutants (Figure 2I). Collectively, these findings suggest that chromatin remodeling enzymes retain compartmentalization strength and prevent compartment switching.

A or B compartments are associated with positive or repressive histone modifications (Lieberman-Aiden et al., 2009; Sexton and Cavalli, 2015; Wang et al., 2016; Dogan and Liu, 2018; Zheng and Xie, 2019). By analyzing published data (Supplemental Dataset S3; Liu et al., 2018), we found that A compartments in WT plants were enriched with H3K4me2/3, H3K36me3, and H3K27ac, which are typically found in euchromatin (Sexton and Cavalli, 2015; Dogan and Liu, 2018). B compartments, however, were enriched with H3K9me2, H3K27me1, H2A.W, or H3K27me3, which are usually associated with heterochromatin (Supplemental Figure S4; Sexton and Cavalli, 2015; Dogan and Liu, 2018).

To understand the mechanisms underlying certain B compartment region switches, we compared the epigenetic features of switching and non-switching regions in four mutants to those in WT plants. No switching regions in B compartments were enriched with H3K9me2, H3K27me1, and H2A.W (Figure 2J); switching regions in B compartments, however, were specifically enriched with H3K27me3 and exhibited depleted constitutive heterochromatin-related epigenetic features (Figure 2K). Taken together, our results suggest that chromatin remodeling complexes maintain H3K27me3-modified regions in B compartments, which are within repressive euchromatin regions.

To further investigate the consequences of compartment switching, we classified B compartments into three types: those that switch completely, those that switch partially, and those that do not switch (Figure 3A). We found that around 75% of compartment B regions partially or completely switched to A compartments in all enzyme mutants (Figure 3, B–E). Mutations in BRM or INO80 caused 50% of B compartments to partially switch and about 25% to completely switch to A compartments (Figure 3, B and D). Conversely, in chrs or pkl mutants, over 50% of B compartments completely switched, while approximately 25% partially switched or remained as B compartments (Figure 3, C and E). Moreover, scatter plots showed more B compartment switching in chrs and pkl mutants in comparison with brm and ino80 mutants (Figure 3, F–I). These results explain the reason for continual A compartment regions within chromosome arms in all enzyme mutants, especially in chrs and pkl mutants (Figure 2A and Supplemental Figure S2).

Figure 3.

Analysis of compartment switching types in chromatin remodeling enzyme mutants. A, The model shows three types of compartments switching. From top to bottom: complete switch, partial switch, and no switch. The blue lines indicate compartments in WT; the green lines indicate compartments in chromatin remodeling enzyme mutants. B–E, The percentage of B-to-A compartment switching in the enzyme mutants. In the scatter plots, each point represents an independent compartment in WT plants. The x-axis shows the percent of compartment switching in mutants. The y-axis shows the ratio of the switching percentages of different compartments. 0, 1, or 0–1 represent no switch, complete switch, or partial switch, respectively. F–I, Scatter plots showing the correlations between WT plants and brm (F), chrs (G), ino80 (H), and pkl (I) mutants among B compartment regions in WT plants. The x-axis and y-axis indicate the PC1 value of mutants and WT, respectively. The colored points and black points indicate the switch B (B-to-A switched regions) and stable B regions, respectively.

Compartment switching regions are enriched in development-related genes

Almost all developmental stages are regulated by BRM, CHRs, INO80, and PKL (Li et al., 2012, 2015; Zhang et al., 2015; Kim, 2019; Yang et al., 2019). Mutations in any of these enzymes produce pleiotropic phenotypes. Although chromatin remodeling complexes regulate target gene transcription mainly by affecting the binding of corresponding transcription factors (Barisic et al., 2019), we determined how these four enzymes regulate transcription via compartment switching by performing RNA-seq with WT and mutant seedlings (Figure 4A and Supplemental Dataset S4). The number of differentially expressed genes (DEGs) in compartment switching regions was lower than the total number of DEGs across the genome (Supplemental Figure S5A).

Figure 4.

DEGs within compartment switching regions are associated with plant development. A, PCA of three RNA-seq replicates in WT, brm, chrs, ino80, or pkl mutants. B, Pie charts indicate the ratio of up- or down-regulated genes among all DEGs in B-to-A switched regions between WT plants and mutants. Red and blue represent up- and down-regulated genes, respectively (|log₂FoldChange| > 1.5; q < 0.01). C–F, GO analysis of DEGs within compartment switching regions (q < 0.05).

In order to investigate the effect of compartment switching on gene transcription regulation, we focused on DEGs within compartment switching regions. In general, B-to-A compartment switching was more dominant than A-to-B switching (Figure 2); this was consistent with more up-regulation within B-to-A compartment switching regions in comparison with A-to-B compartment switching regions in the four enzyme mutants (Figure 4B and Supplemental Figure S5B). DEGs in brm or chrs were enriched in Gene Ontology (GO) terms related to chloroplast development and photosynthesis (Figure 4, C and D). GO analysis of DEGs in ino80 switching regions showed enrichment only for cell wall-related terms (Figure 4E); DEGs within pkl switching regions were associated with isoprenoid metabolism (Figure 4F). Isoprenoids are a highly diverse group of plant metabolites with important roles in plant growth and development (Nazemof et al., 2016). In summary, these data show that chromatin remodeling enzymes prevent compartment switching, which is required to maintain normal developmental gene expression.

H3K27me3 reduction leads to compartment switching

Senescent cells lose their compartmentalization in H3K27me3-enriched regions (Zhang et al., 2021). Since H3K27me3-modified B compartments also tend to switch (Figure 2K), we wondered whether distinct chromatin remodeling complexes affect H3K27me3 deposition, particularly within switched regions. Using ChIP-seq, we first compared H3K27me3 levels in B-to-A compartment switching regions in WT plants to four different enzyme mutants (Supplemental Figure S6A). Our data were consistent with previous studies showing that H3K27m3 levels are elevated in brm mutants at specific target regions, but were decreased genome-wide compared with WT plants (Supplemental Figure S6, B and D); H3K27me3 levels were found to be decreased in pkl mutants at specific target regions, but were unchanged genome-wide (Supplemental Figure S6, C and D; Li et al., 2015; Carter et al., 2018).

When we examined H3K27me3 levels in B-to-A switching regions, we found a significant and specific reduction in H3K27me3 levels in all enzyme mutants compared with WT plants (Figure 5, A–D and Supplemental Figure S7, A–D); however, the random regions show similar H3K27me3 levels in WT plants and four enzyme mutants (Figure 5, A–D). H3K27me3 levels in switching regions decreased more significantly in chrs and pkl mutants in comparison with brm and ino80 mutants, suggesting that there were more complete switching regions in chrs and pkl mutants (Figure 3). Taken together, our results indicate that distinct chromatin remodeling complexes prevent compartment switching through a similar mechanism that affects deposition and/or maintenance of a high level of H3K27me3, although the extent of regulation varies among different chromatin remodeling complexes.

Figure 5.

Chromatin remodeling complexes function to promote H3K27me3 and prevent compartment switching. A–D, Metagene plots showing significant reductions of H3K27me3 levels in B-to-A compartment switching regions in every enzyme mutant in comparison with WT plants (two-sided Mann–Whitney U test). The average H3K27me3 signal was obtained by calculating the log₂ ratio to IPs using inputs. WT, blue; brm, red; chrs, gray; ino80, green; pkl, orange. The dark lines represent H3K27me3 on switching regions, whereas the light lines represent H3K27me3 on random regions. E and F, Magnified views of Marneral (E) and Thalianol (F) gene clusters in enzyme mutants show compartment switches and reduced H3K27me3 levels. H3K27me3 was calculated by subtracting the values of the H3K27me3 levels of mutants from those of WT plants. Red and blue on the PC1 value tracks indicate A and B compartments, respectively. Red and blue on the delta H3K27me3 tracks show increased and decreased levels of H3K27me3, respectively.

To determine whether H3K27me3 reduction leads to compartment switching, we compared the changed compartment regions in WT plants, four enzyme mutants, and clf swn mutants. CLF and SWN are histone methyltransferases that add H3K27me3 to histones (Xiao and Wagner, 2015). In comparison with WT plants, brm and ino80 mutants had fewer changed compartment regions, but chrs and pkl mutants had a greater number of changed compartment regions (Supplemental Figure S8, A–D). Furthermore, we found that clf swn mutants, in which H3K27me3 is completely eliminated (Yin et al., 2021), showed the most changed compartment regions compared with WT plants (Supplemental Figure S8E).

At the chromosome scale, continual A compartment regions could also be observed in clf swn mutants, which indicates that a loss of H3K27me3 leads to B-to-A compartment switching (Figure 2A and Supplemental Figure S2). To further confirm that reduced H3K27me3 levels lead to compartment switching, we examined two biosynthetic nonhomologous gene clusters (Marneral and Thalianol), which are modified via H3K27me3 and therefore isolated from the surrounding chromosome environment (Nutzmann et al., 2020). We found that these two clusters belonged to B compartments in WT plants but switched to A compartments or showed reduced compartmentalization in all enzyme mutants (Figure 5, E and F).

We then calculated the difference in H3K27m3 levels between WT plants and brm, chrs, ino80, and pkl mutants. Consistent with the general decrease in H3K27me3 levels in switching regions in all enzyme mutants (Figure 5, A–D), we found reduced H3K27m3 levels in these cluster regions. Regions that lost H3K27me3 switched to A compartments in the four enzyme mutants (Figure 5, E and F). Moreover, B compartment regions in these two cluster regions also switched to A compartments in clf swn mutants, which lacked H3K27me3 modification (Figure 5, E and F).

We also investigated the relationship between H3K27me3 levels and compartment status in roots and leaves. It was reported that the two biosynthetic nonhomologous gene clusters mentioned above have different H3K27me3 modifications in roots and leaves (Nutzmann et al., 2020). We found that a high level of H3K27me3 is linked to B compartmentalization in leaves, whereas a low level of H3K27me3 is linked to A compartmentalization in roots (Supplemental Figure S9). In summary, our data provide evidence that a high level of H3K27me3 functions to maintain B compartmentalization.

Linear nucleosome arrangement affects H3K27me3 deposition and compartment switching

The main functions of chromatin remodeling complexes are involved in regulation of the nucleosome distribution pattern and density, including eviction, sliding, insertion, or replacement of nucleosomes (Clapier and Cairns, 2009; Clapier et al., 2017; Lorch and Kornberg, 2017). To investigate how chromatin remodeling complexes regulate the nucleosome distribution pattern and density to affect H3K27me3 deposition in switching regions, we performed micrococcal nuclease sequencing (MNase-seq) in WT and four mutant seedlings at 10 DAG (Supplemental Figure S10). Analysis of the distribution of nucleosomes surrounding the transcription start site (TSS) of genes via genome-wide nucleosome mapping in different organisms reveals a nucleosome-rich region containing highly occupied and well-phased nucleosomes downstream of the TSS (Yague-Sanz et al., 2017; Barbier et al., 2021). Since B compartment switching occurred in all of the chromatin remodeling enzyme mutants (Figures 1–3), we focused on the nucleosome coverage of +1 kb regions downstream of the TSS within B-to-A compartment switching regions in WT plants and four enzyme mutants.

Globally, metagene plots of nucleosome coverage in the B-to-A switching regions showed different patterns and distributions in the four enzyme mutants compared with WT plants. brm and ino80 mutants showed lower or higher distribution patterns in the proximal (+1 nucleosome) or distal regions downstream of the TSS (Figure 6, A and C). The general nucleosome distribution pattern did not change much in brm and ino80 mutants, which is consistent with weak B-to-A compartment switching (Figure 3). However, chrs mutants lost the evenly spaced pattern of nucleosomes downstream of the TSS (Figure 6B); pkl mutants had lower nucleosome density downstream of the TSS (Figure 6D). A severe altered nucleosome distribution pattern reflected more B-to-A compartment switching (Figure 3).

Figure 6.

Nucleosome pattern and distribution within compartment switching regions differ from those in random regions. A–D, Metagene plots indicate the nucleosome coverage of genes in +1 kb regions downstream of the TSS within B-to-A compartment switching regions in WT or brm, chrs, ino80, or pkl mutants. WT, blue; brm, red; chrs, gray; ino80, green; pkl, orange. The dark lines represent the nucleosome coverage within switching regions, while the light lines represent the nucleosome coverage within random regions. E, Box plots indicate the relative nucleosome density in switched compartment regions compared with that in randomly selected regions (two-sided Mann–Whitney U test). The relative difference was calculated by subtracting the nucleosome coverage of mutants from that of WT plants. Red, gray, green, or orange boxes indicate brm, chrs, ino80, and pkl mutants, respectively. F, Working model: chromatin remodeling complexes regulate the nucleosome distribution pattern and density to maintain H3K27me3 modification, which prevents compartment switching.

To further investigate the regulation of nucleosome density regulation by chromatin remodeling complexes within B-to-A compartment switching regions, we compared the nucleosome density in these regions to randomly selected regions; the nucleosome density was significantly reduced in brm, ino80, and pkl mutants (Figure 6E). Moreover, when we focused on protein coding genes, we found that nucleosome density was altered in both gene body regions and promoter regions in four enzyme mutants (Supplemental Figure S11). The lower nucleosome density in mutants reflected switching from a repressive B compartment to an active A compartment; this further explained the reduced H3K27me3 levels observed in the enzyme mutants (Figure 5). Meanwhile, CHRs, which function as a molecular ruler (Barisic et al., 2019), promote H3K27me3 mainly by regulating the nucleosome distribution pattern (Figure 6B). In summary, our data suggest that the mechanisms underlying promotion of H3K27me3 deposition differ between CHRs and BRM, INO80, and PKL.

Discussion

In this study, we show that four conserved chromatin remodeling complexes contribute to 3D chromatin genome regulation. We found reduced compartment strength and more switched compartments in the chromatin remodeling enzyme mutants. The combined Hi-C and MNase-seq data showed that the linear nucleosome distribution pattern and density affected compartment stability. Our findings suggest that chromatin remodeling complexes prevent compartment switching by maintaining a high level of H3K27me3.

In mammalian cells, the loss of CTCF promotes compartmentalization and the formation of finer compartments, which are further associated with histone modifications (Nora et al., 2017). Our data confirmed that A or B compartments are associated with positive or repressive modifications in plants. Compared with A-to-B compartment switching, B-to-A compartment switching was prevalent in all mutants, and switching regions showed significant overlap. Moreover, we found that switched B compartments were enriched with H3K27me3 modifications. In mammalian cells, H1 proteins compact chromatin and regulate H3K27me3 deposition (Willcockson et al., 2020; Yusufova et al., 2021). H1 coding gene knock-down reduces H3K27me3 levels and leads to the predominance of B-to-A compartment switching. To demonstrate that H3K27me3 reduction has a causal effect on B-to-A switching, we showed that (1) B-to-A switching is the dominant event at the chromatin level (Figure 2, A and G) and in H3K27me3-modified regions in clf swn mutants (Figure 5, E and F), in which H3K27me3 is completely eliminated and (2) a high level of H3K27me3 is important for B compartment maintenance in different tissues (Supplemental Figure S9). The present study provides evidence showing that distinct chromatin remodeling complexes function in similar ways that maintain a high level of H3K27me3 to prevent compartment switching (Figure 6F).

ISWI, not SWI/SNF, affects chromosome conformation by regulating CTCF binding, particularly the reduction in insulation at TAD boundaries in mammalian cells (Barisic et al., 2019). However, compartments are retained in ISWI knock-out cells (Nora et al., 2017). In Drosophila, CTCF is a DNA binding protein that regulates chromatin 3D structure (Merkenschlager and Nora, 2016). Distinct architectural proteins regulate the Drosophila chromatin 3D structure at different developmental stages and in different cell types (Rowley and Corces, 2016). While plants do not have a CTCF homolog, distinct proteins and mechanisms may mediate different levels of chromatin high-order structure in plants. Our Hi-C results showed that all four plant chromatin remodeling complexes regulate A/B compartment switching and compartmentalization strength, demonstrating that the molecular mechanism of 3D chromatin genome regulation differs between mammalian and plant cells. Although plant insulator-like cis-elements, which could be bound by TCP proteins, were identified through Hi-C analysis in rice, tcp mutants did not show any change in their chromatin 3D structures (Karaaslan et al., 2020). Factors that are involved in plant 3D chromatin structure regulation, therefore, still need to be identified. In this study, we found that chromatin remodeling complexes affect the 3D chromatin structure by regulating nucleosomes (Figure 6F): (1) BRM, INO80, and PKL mainly regulate nucleosome density; (2) CHRS mainly regulates the nucleosome distribution pattern. Changes in the nucleosome distribution pattern and density are associated with H3K27me3 deposition. In summary, we examined chromatin 3D structure in chromatin remodeling enzyme mutants at high resolution and provided clear evidence of how chromatin remodeling complexes regulate the nucleosome distribution pattern and density to promote H3K27me3 deposition, which in turn prevents compartment switching.

Materials and methods

Plant materials and growing conditions

Arabidopsis thaliana plants were grown in a Percival growth chamber under cool white light (Philips, ∼100 μmol·m⁻²·s⁻¹) at 22°C in long day (16-h light/8-h dark) conditions. Seeds were incubated at 4°C in the dark for two days to stratify germination. For RNA-seq, ChIP-seq, MNase-seq, and Hi-C experiments, seeds of Col-0, brm-1 (Li et al., 2015), chrs (Li et al., 2014), ino80-5 (Zhang et al., 2015), pkl-1 (Joe Ogas, 1997), and clf swn (Yin et al., 2021) mutants were sterilized in 75% ethanol and sown on Murashige and Skoog medium. Materials from 10-day-old seedlings were collected.

RNA extraction and library preparation

For RNA-seq, total RNA was extracted from 10-day seedlings using the E.Z.N.A. Plant RNA Kit (Omega, R6827-01). RNA-seq libraries were prepared with a VAHTS Universal V6 RNA-Seq Library Prep Kit for Illumina (Vazyme, NR604). Briefly, messenger RNA was purified with mRNA Capture Beads and fragmented with Frag/Prime Buffer at 85°C for 6 min. Double-stranded cDNA was then synthesized, followed by ligation of adapters and PCR amplification for 13 cycles. Libraries were sequenced based on Novaseq 6000 guidelines (Illumina). Three biological replicates were performed for RNA-seq.

ChIP and ChIP-seq library preparation

ChIP experiments were performed as previously described (Yin et al., 2021). Briefly, 10-day-old seedlings were fixed in 1% formaldehyde in phosphate-buffered saline (PBS, pH 7.4) under vacuum for 2 × 10 min; the fixed seedlings were then homogenized in liquid nitrogen. Chromatin was extracted and sonicated with a Bioruptor Pico (Diagenode) to produce DNA fragments of 200–500 bp. The sheared chromatin was immunoprecipitated overnight at 4°C using the following antibodies: Anti-H3K27ac (Abcam, ab4729), Anti-H3K4me3 (Abcam, ab4729), and anti-H3K27me3 (Millipore, 07-449). Protein A Sepharose beads CL-4B (GE Healthcare) were washed and used to capture H3K27me3-, H3K4me3-, or H3K27ac-associated DNA. The chromatin was eluted and de-crosslinked overnight at 65 °C. DNA from the immunoprecipitated chromatin was recovered by phenol–chloroform extraction, followed by ethanol precipitation. For ChIP-seq, two independent biological replicates were performed for next-generation sequencing library preparation. All libraries were prepared with Ovation Ultralow Library Systems (NuGEN) according to the manufacturer’s instructions. After end repair of sheared DNA, adapters were ligated. The ligated product was then amplified with 18 cycles of PCR. To confirm amplification, an aliquot of the library was tested by quantitative PCR before and after PCR. DNA (200–600 bp) was purified with VAHTS DNA Clean Beads (Vazyme, N411-01). Library sequencing was performed on an Illumina Hiseq-Xten PE150 or a NovaSeq 6000 by generating 2 × 150 bp paired-end reads.

In situ Hi-C and library preparation

In situ Hi-C was performed as previously described. Seedlings were fixed in 1% formaldehyde solution in MS buffer (10 mM potassium phosphate, pH 7.0; 50 mM NaCl; 0.1 M sucrose) at room temperature for 2 × 10 min under vacuum. After fixation, the seedlings were subjected to 5 min of vacuum treatment with 0.1 M glycine. The fixed tissue was homogenized with liquid nitrogen and resuspended in nuclei isolation buffer (20 mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), pH 8.0, 250 mM sucrose, 1 mM MgCl₂, 5 mM KCl, 40% [v/v] glycerol, 0.25% Triton X-100, 0.1 mM phenylmethanesulfonyl fluoride (PMSF), 0.1% [v/v] 2-mercaptoethanol) and then filtered through two layers of Miracloth (Merck Millipore). The nuclei were resuspended in 0.5% SDS, denatured at 62°C for 5 min, and then digested with 50 units of DpnII overnight at 37°C. The next day, the digested DNA was blunt-ended by the Klenow enzyme (Thermo Scientific), during which biotin-14-dCTP (Invitrogen) was incorporated. After ligation with T4 DNA ligase, DNA isolation was performed via phenol–chloroform extraction, followed by ethanol precipitation. DNA was then sheared by sonication with a Bioruptor Pico (Diagenode). Sheared DNA (200–600 bp) was selected with VAHTS DNA Clean Beads (N411-01). The DNA was then purified with Dynabeads MyOne Streptavidin C1 beads (Invitrogen). Following biotin enrichment, on-bead end repair and adapter ligation were performed. After washing, the beads were resuspended in 15 μL of 10 mM Tris–HCl buffer (pH 8.0). The DNA was detached from Dynabeads MyOne Streptavidin C1 beads by incubation at 98 °C for 10 min. Amplification of library molecules was performed with 11 cycles of PCR. PCR products were then purified with VAHTS DNA Clean Beads (Vazyme, N411-01) and libraries were sequenced on an Illumina Hiseq-Xten PE150 platform by generating 2 × 150 bp paired-end reads.

MNase experiment and library preparation

Seedlings (0.3 g) were collected and immediately fixed in 1% formaldehyde under vacuum for 2 × 10 min at room temperature; this was followed with a 5-min neutralization by adding 0.1 M glycine. The fixed seedling tissues were ground into a fine powder in liquid nitrogen and resuspended in Honda buffer (20 mM HEPES–KOH, pH 7.4, 0.44 M sucrose, 1.25% Ficoll, 2.5% Dextran T40, 10 mM MgCl₂, 0.5% Triton X‐100, 5 mM DTT, 1 mM PMSF, and protease inhibitor cocktail; Roche). After filtering the nuclei through two layers of Miracloth (Merck Millipore), the sample was centrifuged and the pellet was carefully collected, washed twice with Honda buffer, and resuspended in MNase buffer (20 mM Tris–HCl, pH 8.0, 5 mM CaCl₂, and 1 mM PMSF) (Roche). Approximately 500,000 nuclei were used per reaction. To digest genomic DNA into individual nucleosomes, the aliquoted nuclei were cleaved at 37 °C for 10 min with 40 units of micrococcal nuclease (Takara, 2910A). Digestion was stopped by adding 10 μL of 0.5 M EDTA, 20 μL of 1 M Tris–HCl (pH 6.8), and 1.5 μL of 14 mg/mL proteinase K, followed by incubation at 45°C for 1 h. DNA was extracted by phenol–chloroform extraction, followed by ethanol precipitation and treatment with RNase A at room temperature for 30 min. The recovered small-sized DNA fragments were used for the construction of an Illumina library according to the procedures of the NEBNext Ultra II DNA Library Prep Kit for Illumina (NEB, E7645S). Briefly, the proper size of DNA was selected by VAHTS DNA Clean Beads (Vazyme, N411-01) after end repair and adapter ligation. The selected product was then amplified with three cycles of PCR. Library sequencing was performed on an Illumina Nova Seq6000 by generating 2 × 150 bp paired-end reads.

RNA-seq and ChIP-seq data analysis

For RNA-seq, quality control of the data was processed by fastp (Chen et al., 2018) with flag -l 70. Clean reads were then conveyed to HISAT2 (Kim et al., 2015) with default parameters and mapped to the Arabidopsis reference genome (TAIR10, https://www.arabidopsis.org) guided by Araport11 annotations, after which SAMtools (Danecek et al., 2021) was used to sort, index, and compress mapped reads. To estimate gene expression levels with RNA-seq, we first used StringTie (Pertea et al., 2015) to estimate transcript abundance. DESeq2 (Love et al., 2014) was then applied to find DEGs (q < 0.05 and fold change >1, unless otherwise noted); the top 1,000 DEGs were used to reveal reproducibility by principal component analysis (PCA). Pie charts were generated by ggplot2 and clusterProfiler (Yu et al., 2012) was used to process the GO analysis.

For ChIP-seq, clean reads were mapped to the Arabidopsis reference genome (TAIR10, https://www.arabidopsis.org) using Bowtie2 (Langmead and Salzberg, 2012) with default parameters. To remove PCR duplication by PICARD (http://broadinstitute.github.io/picard/), we generated Bigwig files by using bamCoverage from deepTools (Ramirez et al., 2016) with parameter –normalize RPKM. We then processed the IP log₂ratio by input using bigwigCompare from deepTools (Ramirez et al., 2016) with –operation log2. SICER (Zang et al., 2009) was used to call peaks with parameters -s tair10 -rt 1 -w 200 -f 150 -egf 0.9 -g 200 -fdr 0.05. Integrative Genomics Viewer (IGV) (Robinson, 2011) was used to check data quality by comparing typical loci with published data. Metagene plots were generated by SeqPlots (Stempor and Ahringer, 2016). The ChIP-seq results were evaluated to find relative differences among groups using pairwise Mann–Whitney U tests. Reproducibility was determined by peak overlap via Venn diagram (Hanbo Chen, 2011) and Pearson correlation.

MNase-seq data analysis

Upstream analysis was performed using the same methods described for ChIP-seq. To normalize data among samples, we applied quantile normalization methods to normalize occupancy with Wiq function in DANPOS (Chen et al., 2013). To identify high-quality nucleosome peaks, the Dpos algorithm was used. To analyze the distributions of chromatin features flanking each group of genomic sites (e.g. TSS), we calculated nucleosome read coverage by bedmap from BEDPOS (Neph et al., 2012) with option -wmean. Nucleosome occupation scores among different groups were then filtered with function boxplot.stats() to remove the outlier values and evaluated by pairwise Mann–Whitney U tests. Metagene plots were generated by SeqPlots. Reproducibility was determined by PCA.

Public genomic data analysis

Various Arabidopsis ChIP-seq datasets were downloaded from the Plant Chromatin State Database (Liu et al., 2018). Bedmap (Neph et al., 2012) with -wmean was used to calculate the average occupation score for each selected modification within target regions. ChIP-seq occupation scores among different groups were then evaluated by pairwise Mann–Whitney U tests and the Bonferroni–Holm method for multiple comparisons (Holm, 1979).

Hi-C data analysis

Fastp (Chen et al., 2018) was used to filter raw Hi-C sequencing reads. HiC-Pro (Servant et al., 2015) was then used to filter read pairs and identify valid ligations. Bowtie2 (Langmead and Salzberg, 2012) was used to map sequencing data in HiC-Pro with default parameters. We subsequently obtained contact matrices and transformed them into different formats to fit the requirements of downstream analysis software (refer to Sun et al. [2020] for Hi-C data analysis methods). To compare Hi-C heatmaps, we first normalized Hi-C contact maps in HiCdatR (Schmid et al., 2015), with correction via the iterative coverage normalization method (Zhang et al., 2012). We then calculated relative differences and used HiCdatR to visualize the relative differences and raw heatmaps. To quantitatively compare differences in chromatin organization between samples, IDEs were calculated with HiCExplorer (Wolff et al., 2018) using a modified version of hicPlotDistVsCounts (Sun et al., 2020) and visualized by ggplot2 (http://ggplot2.tidyverse.org), which applied the function stat_poly_eq() in the ggpmisc R package based on the log₁₀ values of distance and contacts, after which the equation was calculated for the fitted polynomial with formula y ∼ x. To define chromatin compartments in WT and mutant plants, we used HOMER (Heinz et al., 2010) with a window size of 60 kb for each 20 kb genomic segment to perform PCA analyses of Hi-C. H3K4me3 ChIP-Seq data were input to HOMER (Heinz et al., 2010) to determine the orientation of PC1 values. Chromatin compartmentalization saddle plots and chromatin strength were calculated and visualized by FAN-C (Kruse et al., 2020). All Hi-C datasets were normalized to the same sequencing depth, corrected with KR balancing methods, and used to generate observed/expected matrices with obs_exp_lieberman methods by hicNormalize, hicCorrectMatrix, and hicTransform from HiCExplorer, respectively. pyGenomeTracks (https://github.com/deeptools/pyGenomeTracks/) was used to visualize typical loci and their features based on Hi-C data, ChIP-Seq data, and annotations.

Accession numbers

All sequencing data generated from this study have been deposited into the SRA database (https://www.ncbi.nlm.nih.gov/sra/) under the accession number PRJNA780072.

Supplemental data

The following materials are available in the online version of this article.

Supplemental Figure S1. Chromatin-interaction patterns in chromatin remodeling enzyme mutants.

Supplemental Figure S2. The tracks show the PC1 values generated from Hi-C data on the right arm of chromosome 5.

Supplemental Figure S3. Compartment switching regions in four enzyme mutants.

Supplemental Figure S4. Compartments A or B are associated with positive or repressive epigenetic modifications.

Supplemental Figure S5. DEGs in chromatin remodeling enzyme mutants compared with WT plants.

Supplemental Figure S6. H3K27me3 ChIP-seq in WT and four enzyme mutants.

Supplemental Figure S7. Heat map of H3K27me3 ChIP-seq read density.

Supplemental Figure S8. Correlation of PC1 values between WT and mutants.

Supplemental Figure S9. The association between decreased H3K27me3 level and B-to-A compartment switching in roots and leaves.

Supplemental Figure S10. MNase-seq in WT and four enzyme mutants.

Supplemental Figure S11. Nucleosome relative coverage within gene bodies or gene upstream regions in compartment switching regions differs from that in random regions.

Supplemental Dataset S1. Summary statistics for Hi-C data.

Supplemental Dataset S2. Identified compartment switching regions, including A-to-B switching and B-to-A switching, in brm, chrs, ino80, and pkl mutants.

Supplemental Dataset S3. List of public epigenetic data sources.

Supplemental Dataset S4. DEGs within compartment switching regions in brm, chrs, ino80, and pkl mutants.

 

These authors contributed equally (T.Y., D.W., G.T., and L.S.)

T.Y., G.T., and X.Y. performed all the experiments. D.W., L.S., M.Y., and H.H. analyzed high-throughput sequencing data. J.X., Y.S., D.Z., and Y.Z. interpreted the data. Y.Z. planned the experiments and wrote the manuscript. All authors read and approved the final manuscript.

The author responsible for distribution of materials integral to the findings presented in this article in accordance with the policy described in the Instructions for Authors (https://academic.oup.com/plcell/pages/General-Instructions) is: Yue Zhou (yue_zhou@pku.edu.cn).