Genomic and epigenomic determinants of heat stress-induced transcriptional memory in Arabidopsis

Binding profiles of HSFA2 and HSFA3 are highly similar, but they differ on different sets of genes

To determine the direct targets of memory HSFs and their binding dynamics we performed ChIP-seq analysis of unstressed seedlings and seedlings at different times after a priming HS (termed ACC, with recovery for 4 h, 28 h, or 52 h, Fig. 1a), expressing Flag-HSFA2 or Flag-HSFA3 from the respective endogenous promoters. Both transgenic lines show full complementation of the hsfa2 and hsfa3 phenotypes (Additional File 1: Fig. S1) [22]. We selected the time points 4 h, 28 h, and 52 h after the end of a HS in line with previous analyses. Since both HSFs are strongly induced after HS [7, 22, 24], no-HS (NHS) samples contain only small amounts of tagged protein and were expected to yield low chromatin binding. For each time point, three biological replicates were performed.

Fig. 1figure 1

HSFA2 and HSFA3 targets during HS memory fall into 15 clusters and show differential binding strength and dynamics. a Treatment schematic for ChIP-seq and RNA-seq experiments. Four-day-old seedlings were exposed to a two-step acclimation treatment (ACC) of the indicated temperature and duration. Sampling was at the indicated time points after the end of ACC (arrowheads) or at the corresponding time point for no-HS (NHS) samples. b Median ChIP-seq signal of HSFA2 and HSFA3 in 15 cclusters at the indicated time points. CPM, counts per million; n, number of peaks in each cluster. c Memory cclusters show high binding strength at 4 h after ACC. Empirical cumulative distribution function of HSFA2- and HSFA3-bound peaks at 4 h after ACC. C7, c11, and c12 are highlighted in color. Binding of c12 is significantly higher than for all other clusters (HSFA2) and than for all other clusters except c6 (HSFA3), respectively (KS-test, p < 0.05, Additional File 2: Supplementary Data 1). Cluster c14 is not included as it contains only one peak

The initial peak calling resulted in several thousand peaks that were present in at least one of the samples. Peak width was adjusted to 200 bp, centering on the summit. Compared to the 74 consolidated HSF binding sites in yeast [41], this is a comparatively high number, suggesting that not all of these peaks may be functionally relevant. However, to minimize bias due to filtering with hard thresholds, we used hierarchical clustering to sort the peaks into 15 different ChIP-seq clusters (cclusters) depending on their binding profile over the time course (Fig. 1b, c). Overall, HSFA2 and HSFA3 displayed remarkably similar binding patterns. In terms of median signal strength over all time points, six clusters have a very low signal (c1-5, c8), three clusters (c6, c13, c14) have an exceedingly high signal, and the remaining six clusters (c7, c9-12, c15) have an intermediate signal strength. Because HSFA2 and HSFA3 expression is HS-inducible, we expect true targets to show low (or no) binding in NHS conditions. Clusters c6, c8, c9, c10, c13, and c15 showed the highest signal at NHS, suggesting they do not represent true HSF targets. In contrast, c1, c2, c5, c7, c11, c12, and c14 showed an increase in signal strength after HS, conforming to our expectations. Thus, biologically relevant targets may be concentrated in seven clusters, together comprising 4948 peaks. Interestingly, c12 showed the highest binding signal at 4 h compared to all other clusters (for HSFA2) and all other clusters except c6 (for HSFA3), respectively (KS-test, p < 0.05, Additional File 2: Supplementary Data 1). We next determined the distance of the peaks to the closest transcriptional start site (Fig. 2a–c). Overall, we found a biphasic distribution, with a large fraction of peaks very close to a TSS and another large fraction several hundred to more than 1000 bp away (Fig. 2a). Notably, clusters c7 and c12 contained a much higher proportion of peaks at the TSS than distant peaks; conversely, clusters c6, c8, c9, c10, and c13 were dominated by distal peaks (Fig. 2b). In some cases, one gene was associated with more than one peak, thus there are slightly fewer genes than peaks in the clusters (Fig. 2c).

Fig. 2figure 2

Characterization of binding sites of HSFA2/HSFA3 according to their distance from associated genes, chromatin profiles, and chromatin states. a Density plot illustrating global distance of peaks from the nearest transcriptional start site (TSS). Gray line indicates distance of 2000 bp. b Density plot illustrating global distance of peaks from the nearest transcriptional start site (TSS) in ChIP-seq clusters. c Number of peaks and number of associated genes in cclusters. d Enrichment of histone modifications in seedlings grown under control (NHS) conditions in a window of 3 kb around the peak centers. Ccluster c11, c12, and c14 are enriched for H3K4me3 at the flanks of the peaks, cluster c13 is enriched for H3K9me2, and cluster c15 is enriched for H3K27me3. Histone modification data were reanalyzed from [42]. e ChIP-seq peaks are enriched in different chromatin states, with clusters c7, c11, c12, and c14 being enriched in chromatin states 1 and/or 2. The proportion of peaks in cclusters that are associated with the indicated chromatin states is depicted. *p < 0.05 Fisher’s exact test. Chromatin states are taken from [43]

Chromatin states in clusters

To start to address whether different clusters are enriched in certain histone modifications before HS, we analyzed the average profile of 5 frequently profiled histone modifications (H3PanAc (acetylation), H3K4me3, H3K9me2, H3K27me3, H3K36me3) [42]. We observed enrichment of H3K4me3 in clusters c11, c12, c14, of H3K9me2 in c13, of H3K36me3 in c14, and of H3K27me3 in c15; all other clusters did not show any enrichment in any modification (Fig. 2d). Notably, the enrichment in H3K4me3 and H3K27me3 was not observed in the center of the peaks, but at a distance of several hundred base pairs, indicating that there is no specific marking of the HSF binding site, but rather that the modifications occur in the bodies of associated genes.

Chromatin modifications including DNA methylation and histone modifications tend to occur together in certain combinations that determine the overall activity state and accessibility of the chromatin [43, 44]. A recent study in A. thaliana defined nine chromatin states [43]. The promoters of active genes and genes amenable to rapid activation tend to be in the “open” chromatin states 1 and 2, which are enriched in H2A.Z, H3K4me2, and H3K4me3. We investigated whether the clusters differ in their chromatin states. Cclusters c1-5, c7, c11, c12, and c14 are enriched in chromatin states 1 and 2 (Fig. 2e). Conversely, c9, c10, and c13 are enriched in chromatin states 8 and 9, representing heterochromatic chromatin organization (enriched in DNA methylation and H3K9me2), and c6 and c15 in states 4 and 5, representing silenced chromatin enriched in the polycomb mark H3K27me3. Notably, all clusters with a HS-induced increase in HSF binding (clusters c1, c2, c5, c7, c11, c12, c14) are enriched in open chromatin that is typical for promoter regions (states 1 and 2), suggesting that these clusters are enriched in biologically relevant target genes (Figs. 1b, 2e). Together with the overall binding strength, this suggests that clusters c5, c7, c11, c12, and c14 represent biologically relevant targets of HSFA2/HSFA3.

Three clusters contain highly HS-inducible genes with reduced expression in hsf mutants

We next asked whether the increased binding of memory HSFs to genes in clusters c1, c2, c5, c7, c11, c12, and c14 is reflected in induced gene expression after HS. To this end, we integrated the ChIP-seq data with transcript profiling data of Col-0, hsfa2, hsfa3, and hsfa2 hsfa3 double mutants at the corresponding time points (NHS, 4 h, 28 h, 52 h after ACC, Fig. 1a) [22]. Genes that are upregulated at 4 h after ACC are significantly enriched in cclusters c1, c5, c7, c11, and c12 (p < 0.01, Fisher’s exact test, Fig. 3a, Additional File 2: Supplementary Data 1). Genes that are upregulated at 52 h after ACC are enriched in c7 and c12 (p < 0.01, Fisher’s exact test, Fig. 3b, Additional File 2: Supplementary Data 1). The median expression of all of these clusters peaked at 4 h (Fig. 3c, d). Sustained induction at 28 and 52 h was evident in c12 and c14 and weakly in c7 and c11 (Fig. 3d). The induction after ACC depended on functional HSFA2 and HSFA3 in c7, c12, and c14 (Fig. 3c, d). While the genes in clusters c1 and c5 were overall slightly induced by ACC, this was not dependent on memory HSFs (Fig. 3c). This is consistent with the finding that these genes did not show sustained induction. Thus, c12 and c14 (and to a lesser extent c7) are enriched for genes with HSFA2/HSFA3-dependent and overall sustained induction of gene expression during the memory phase. While c12 is associated with 43 genes, c14 has only one peak that is associated with HSFA2 and its neighbor gene. C12 contains many HSP genes and several previously identified and confirmed targets such as HSP22.0, HSP101, and MIPS2 (Additional File 1: Fig. S2, Additional File 3: Supplementary Data 2, [7, 8, 22]). While sustained induction of c7 is less clear than for c12 and c14 based on median gene expression, c7 contains several previously characterized HS-memory genes, including APX2, HSA32, and HSP18.2 [7, 22]. Thus, clusters c7, c12, and c14 are enriched in biologically relevant targets of HSFA2 and HSFA3. We experimentally validated the binding of HSFA2/HSFA3 to five putative target genes from c7 and c12 by ChIP-qPCR (Additional File 1: Fig. S3) and found that the results were fully consistent with previous reports [7, 22] and the time course ChIP-seq analysis reported here.

Fig. 3figure 3

HSFA2 and HSFA3 target genes display HS-induced sustained gene expression that depends on memory HSFs. a Fraction of genes in ChIP-seq clusters that show differential expression in wild type at 4 h after ACC. Clusters c1, c5, c7, c11, and c12 are significantly enriched in genes that are upregulated at this time point (*p < 0.01, Fisher’s exact test, Additional File 2: Supplementary Data 1). b Fraction of targets genes in ChIP-seq clusters that show differential expression in wild type at 52 h after ACC. Clusters c7 and c12 are significantly enriched in genes that are upregulated at this time point (*p < 0.01, Fisher’s exact test, Additional File 2: Supplementary Data 1). c, d Median expression of genes in ChIP-seq clusters at the indicated time points and genotypes. Gene expression in clusters c7, c12, and c14 is dependent on HSFA2/HSFA3. HS-induced expression in clusters c12, c14, and to a lesser extent c7 shows sustained induction in Col-0 wild type. n, number of genes per cluster

Binding of memory HSFs is largely determined by sequence

To investigate to what extent DNA sequence determines target selection and binding strength of HSFA2/HSFA3, we performed in vitro binding assays (DAP-seq) [45] for HSFA2, HSFA3, HSFA1b, and combinations thereof. During the binding reaction, samples were incubated at either 25 °C or 37 °C to analyze the effect of temperature on binding. In contrast to ChIP-seq, DAP-seq assesses transcription-factor binding to ‘naked’ DNA in the absence of nucleosomes, but with in vivo levels of DNA methylation [45]. The previously published transcription factor-wide DAP-seq experiment [45] did not include HSFA2 or HSFA3. We mapped the DAP-seq reads to the A. thaliana genome and determined normalized read counts for the ChIP-seq clusters defined above. The highest cluster-wide binding intensity was found for clusters c7, c11, c12, and c14 (Fig. 4), in strong agreement with the ChIP-seq results. For c11, binding above the pIX-Halo background was only seen for some of the samples, all of which contained HSFA2. C8 and to some extent c1 and c2 showed binding above the pIX-Halo negative control at least for some combinations. Interestingly, where both temperatures were tested binding was not generally stimulated by incubation at 37 °C. With the exception of HSFA3, individual HSFs bound DNA similarly as their heteromeric combinations, suggesting a minor role of complex formation. Interestingly, for clusters c7 and c12, binding by HSFA2 and/or HSFA3 appeared stronger than binding by HSFA1b. Thus, our DAP-seq results indicate that DNA sequence (and not chromatin structure) is a major factor determining the binding affinities of memory HSFs and confirms c7, c11, c12, and c14 as highly relevant target gene clusters.

Fig. 4figure 4

The binding intensity of memory HSFs is largely determined by sequence, with modulations from composition of heteromeric complexes. The indicated combinations of HSFA2, HSFA3, and HSFA1b were allowed to bind to genomic DNA at the indicated temperatures (37 °C or 22 °C) and purified and analyzed by DAP-seq. Signal intensities at the peaks from the ChIP-seq experiment were determined and averaged for each ccluster. The red line indicates the signal intensity in the pIX-Halo control sample, which only expressed the Halo-tag. Specific binding intensities over background are highest in clusters c7, c12, and c14

Binding motifs in the clusters

The observed binding profiles suggest that the promoter DNA sequence is an important determinant of binding intensity. Thus, we asked whether particular sequence motifs are enriched under the ChIP-seq peaks in the clusters. A de novo motif analysis using HOMER2 identified 18 motifs as enriched under the peaks in c7, c11, c12, and c14 compared to the rest of the genome. Of these, only the HSE-like motif (motif 1 in Fig. 5a, Additional File 4: Supplementary Data 3) was present under more than 50% of the peaks. In addition, a TCP targeting motif (TGGGCC, motif 2 in Fig. 5a) was found under more than 50% of the peaks in clusters c11, c12, and c14. By contrast, the other 16 motifs did not reach such a high enrichment in the memory clusters. In a complementary approach, we compared the sequences under the peaks to known transcription-factor binding motifs. This identified 22 motifs, most of which were variations on a tripartite HSE (Fig. 5b, Additional File 4: Supplementary Data 3). In particular, variants of tripartite HSE motif TTCtaGAAnnTTCt (motifs 1–16) were strongly enriched in the main memory cclusters c12 (90%) and c14 (100%), as well as cclusters c7 (67%) and c11 (84%), but at much lower frequencies in the remaining cclusters. Among the known motifs identified in this analysis, there was again the TCP targeting motif (motif 17 in Fig. 5b), as well as the G-box motif CACGTG (motifs 19, 20); however, especially the latter motif was present in less than half of the peaks in the main memory clusters. Thus, while no single motif fully discriminated between peaks in memory vs. non-memory clusters, strong binding of HSFA2 and HSFA3 to the tripartite HSE motif TTCtaGAAnnTTCt contributes to the sustained expression of HS memory genes.

Fig. 5figure 5

Sequence motifs underlying ChIP-seq peaks. a De novo motif analysis was performed with HOMER on clusters c7, c11, c12, and c14. The fraction of peaks with the indicated motif is shown for each ccluster. Motifs are sorted according to increasing p-value (cf. Additional File 4: Supplementary Data 3). The most highly enriched motif contains the core motif AGAAnnTCTT consisting of two conserved HSE elements [46]. b Transcription factor binding motif analysis was performed with HOMER on clusters c7, c11, c12, and c14. The fraction of peaks with the indicated motif is shown for each ccluster. Motifs are sorted according to increasing p-value (cf. Additional File 4: Supplementary Data 3). The first 16 motifs contain variations of HSE elements

Transcriptome clustering provides an orthogonal analysis

HSFA2 and HSFA3 are transcriptional activators that regulate HS memory by maintaining sustained gene expression [7, 22]. Hence, expression of direct target genes is expected to be sustained after ACC and less sustained and possibly less induced in hsfa2 and hsfa3 mutants. In an orthogonal approach to the ChIP-seq, we performed hierarchical clustering on a transcriptome dataset, covering the same treatments and time points, and all mutant combinations (Fig. 1a). Of the 15 RNA-seq clusters (in the following referred to as rclusters, Additional File 5: Supplementary Data 4), three were enriched in strongly HS-inducible genes with sustained expression; all three were dependent on both HSFA2 and HSFA3 (rclusters r11, r14, r15, Fig. 6a). R1, r5, and r8 contained genes that were somewhat induced at 4 h after ACC, and r6, r10, and r13 contained genes that were somewhat repressed at 4 h after ACC, but recovered thereafter (Fig. 6a). Thus, rclusters r11, r14, and r15 are likely enriched in direct target genes of HSFA2 and HSFA3. Indeed, they contain previously characterized targets (HSA32, APX2, MIPS2, HSP101 in r11; HSP22.0 in r14; HSP18.2 in r15).

Fig. 6figure 6

Transcriptomics cluster analysis, baseline expression and overlay with ChIP-seq clusters. a Median RNA-seq expression signal of Col-0, hsfa2, hsfa3, and hsfa2 hsfa3 double mutant in 15 rclusters at the indicated time points. tpm, transcripts per million; n, number of genes per cluster. b Heat map illustrating the overlap of genes in rclusters and cclusters. Colors indicate the fraction of genes in rclusters found in ccluster. Numbers in brackets after cluster number indicate the total number of elements in the respective cluster. *p < 0.05 Fisher’s exact test. c Violin plot representing the basal expression level (at NHS condition) of each rcluster. Rclusters r2, r8, r11, and r14 show low basal expression

We next determined the overlap of genes in the cclusters and rclusters (Fig. 6b). Rcluster r15 only contains HSP18.2, and this is in cluster c7. The 14 genes from r14 were all in either c12 (12 genes including HSP22.0) or c7 (2 genes). Of the 45 genes from r11, more than half (24 genes) were in c7 (18), c12 (5), or c14 (1), respectively. Thus, there is a striking and significant overlap between the cclusters with the strongest HS-induced binding (c7, c12, c14) and the rclusters with the strongest, most sustained, and HSF-dependent HS induction (r11, r14, r15) (Fig. 6b, p < 0.05, Fisher’s exact test). R11, r14, and r15 showed the highest fold-induction at 4 h after HS and the highest sustained induction at 28 h and 52 h. Notably, they also had low baseline expression (Fig. 6c). Besides, several HS-induced genes with lesser fold-induction at 4 h and no or little sustained induction were present in r1, r5, and r8. However, their expression did not depend on HSFA2/HSFA3 at the cluster level. Nevertheless, a substantial fraction of genes in these rclusters were still bound by memory HSFs, mostly in c1, c2, and c7. While memory HSFs may contribute to their transcriptional activation after ACC, they are not required for the HS-induction of these genes probably because other HSFs can substitute. Thus, HS-induced sustained and HSFA2/HSFA3-dependent expression is highly correlated with strong binding of HSFA2/HSFA3, low baseline expression (NHS), and an overall high fold-induction. In summary, by using orthogonal approaches, we identified a highly overlapping set of targets that are enriched in the biologically meaningful features HSFA2/3-dependent expression and HS-induced HSFA2/3 binding.

Overlap with established definitions of memory genes

Previous studies defined type I HS memory genes as genes that show upregulation at 4 h after ACC and sustained induction until 52 h into the recovery phase (1–1-1 genes) [21, 22]. In contrast to the above cluster analysis, this was based on differential gene expression with a hard threshold. We were curious to determine the overlap between the clusters identified in this study and the previously defined 1–1-1 memory genes, early heat inducible genes (1–0-0), and genes that are upregulated at 4 h and 28 h, but not at 52 h (1–1-0), respectively. 1–1-1 genes were significantly enriched in cclusters c7 and c12 (p < 0.01, Fisher’s exact test, Fig. 7a, b, Additional File 2: Supplementary Data 1) and in rclusters r8, r11, r14, and r15 (p < 0.01, Fisher’s exact test, Fig. 7c, d, Additional File 2: Supplementary Data 1). While r8 contained some 1–1-1 genes, it was dominated by 1–0-0 genes. Thus, 1–1-1 genes strongly align with the ChIP-seq and RNA-seq clusters that are enriched in functional HSFA2/HSFA3 targets and 22 of the 168 1–1-1 genes were shared with c7, c12, c14 and r11, 14, 15 (Fig. 7h, Additional File 6: Supplementary Data 5). The promoters of these 22 genes showed further enrichment of HSEs, compared to the rest of the 1–1-1 genes (Additional File 6: Supplementary Data 5).

Fig. 7figure 7

Integration of –omics data sets with a previous definition of HS memory genes. As described previously, 1–1-1 genes are upregulated at 4 h, 28 h, and 52 h after ACC, 1–1-0 genes are upregulated at 4 h and 28 h after ACC, and 1–0-0 genes are upregulated only at 4 h [22]. a, b 1–1-1 genes are enriched in cclusters c7 and c12 (*p < 0.01, Fisher’s exact test, Additional File 2: Supplementary Data 1). The fraction of memory genes across cclusters (a) and the number of 1–1-1 genes per cluster (b) is indicated. c, d 1–1-1 genes are enriched in rclusters r8, r11, r14, and r15 (*p < 0.01, Fisher’s exact test, Additional File 2: Supplementary Data 1). Fraction of memory genes across rclusters (c) and the number of 1–1-1 genes per rcluster (d) is indicated. e 1–1-1 genes have a low basal expression level. Density plot representing the basal expression of 1–1-1, 1–1-0 and 1–0-0 genes. f 1–1-1 genes show increased binding (CPM) of HSFA2 and HSFA3 after ACC. Empirical cumulative distribution function of HSFA2- and HSFA3-binding at the indicated time points after ACC. The distribution of values for 1–1-1 genes was significantly different to that of the other three groups for HSFA2 at 4 h and for HSFA3 at 4 h and 28 h (KS-test, p < 0.05). g 1–1-1 genes are not enriched in any histone modification. Enrichment of histone modifications at memory genes of different categories. Gene bodies ± 500 bp are shown; gene models were scaled between transcriptional start site (TSS) and transcriptional termination site (TTS). n, number of genes per group. Histone modification ChIP-seq data were reanalyzed from [42]. h Venn diagrams illustrate the overlap between HSFA2/HSFA3 target genes according to binding (c7, c12, c14) and expression (r11, r14, 15) and 1–1-1, 1–1-0, 1–0-0 genes, respectively

Like r11, r14, and r15, 1–1-1 genes show low baseline expression when compared to 1–0-0 and 1–1-0 genes (Fig. 7e). Clusters c7 and c12 showed the highest proportion of HS-induced genes at 4 h (Fig. 7a), and rclusters r8, r11, r14, and r15 contained exclusively HS-responsive genes (Fig. 7c). Overall, binding of HSFA2 and HSFA3 at 4 h was stronger in 1–1-1 genes compared to the other groups and the genome-wide average; for HSFA3, this was true also at 28 h (Fig. 7f, KS-test, p < 0.05, Additional File 2: Supplementary Data 1). Under no-HS conditions, we did not find a notable enrichment of any histone modification among 1–1-1 genes (Fig. 7g). Thus, memory gene sets defined previously by differential expression or expression cluster analysis (this work) show a strong overlap with memory gene sets based on HSFA2/HSFA3 binding.

H3K4me3 enrichment after HS is a prominent feature of memory genes

HS-induced transcriptional memory is correlated with the enrichment of H3K4me3 at known memory loci [7, 8, 22]. We thus asked whether this accumulation is observed globally at HSFA2/3-dependent memory genes. To this end, we analyzed the H3K4me3 enrichment after a HS treatment and 72 h of recovery from a published data set [47]. Overall, H3K4me3 enrichment peaked shortly after the TSS, irrespective of the condition (Fig. 8). C7 and c12 (and to some extent c11 and c14) showed hyper-accumulation of H3K4me3 at 72 h of recovery (Fig. 8a). The differential signal of H3K4me3 at 72 h of recovery in the TSS + 0.7 kb interval was significantly higher in c7 and c12 compared to the five large non-memory clusters c1 to c5 (Fig. 8b, Additional File 2: Supplementary Data 1, KS-test, p < 0.05). Similarly, transcriptome clusters r11, r14, and r15 showed pronounced HS-induced H3K4me3 hyper-methylation with values for r11 and r14 higher than for all other clusters (Fig. 8c, d, Additional File 2: Supplementary Data 1, KS-test, p < 0.05). This was not observed in any of the other clusters including r8, which contains HS-induced genes that lack sustained induction and are independent of HSFA2/3. Thus, these analyses provide strong genome-wide support for H3K4me3 hyper-methylation as a hallmark of transcriptional memory after HS. To further test the overall relevance of H3K4me3 hyper-methylation, we calculated the average modification levels at 1–1-1 genes. 1–1-1 genes display hyper-methylation of H3K4me3 after HS and long recovery with significantly higher values than for the other groups (Fig. 8e, f, Additional File 2: Supplementary Data 1, KS-test, p < 0.05). Taken together, sustained H3K4me3 enrichment after HS is a global hallmark of HS memory genes.

Fig. 8figure 8

H3K4me3 is globally enriched at memory genes after 3 days of recovery. a, b Cclusters c7, c11, and c12 show increased H3K4me3 3 days after ACC. Enrichment of histone H3K4me3 at NHS and 3 days after ACC treatment in ccluster-associated genes (a). Gene bodies ± 500 bp are shown; gene models were scaled between transcriptional start site (TSS) and transcriptional termination site (TTS). Empirical cumulative distribution function of differential H3K4me3 accumulation (TSS to TSS + 700 bp) after ACC with c7, c11, and c12 highlighted in color (b). The distribution of values for memory clusters c7 and c12 was significantly different from that of the five large non-memory clusters c1 to c5 (KS-test, p < 0.05). c, d Rclusters r11, r14, and r15 show increased H3K4me3 3 days after ACC. Enrichment of histone H3K4me3 at NHS and 3 days after ACC treatment in rcluster-associated genes (c). Empirical cumulative distribution function of differential H3K4me3 accumulation (TSS to TSS + 700 bp) after ACC with r8, r11, and r14 highlighted in color (d). The distribution of values for rclusters r11 and r14 was significantly different from that of all other clusters (KS-test, p < 0.05). e, f 1–1-1 genes show increased H3K4me

留言 (0)

沒有登入
gif