Fig 1.
Schematic diagrams of the inter-nucleosome propagation models and three common modules in DBN inferred networks.
(a-c) Schematic diagrams of the direction of the modeled signal propagation around TSS (a), TTS (b), and CTCF-binding regions (c). (d) The cross-validation scheme of dynamic Bayesian network (DBN). (e) Three common modules of heterologous interactions between different factors inferred by DBN at TSS, TTS and CTCF sites. The network modules depict the signal propagation from the factors at “before” nucleosomes (pink nodes) to the factors at neighboring “after” nucleosomes (green nodes).
Table 1.
The similarities and differences between the inter-nucleosome consensus network around TSS, CTCF-binding, and TTS regions.
Fig 2.
In silico validations of the inter-nucleosomal interactions.
(a-e) Correlation between/among the factors in each of the three common modules. The correlation between factor A’s level at “before” nucleosomes and factor B’s level at “after” nucleosomes was illustrated by a scatter plot and quantified by Pearson correlation coefficient (PCC, red dots). Each point represents 100 nucleosome pairs grouped as a bin by the factor A levels at “before” nucleosomes. The relationships between “-1” and “-2” nucleosomes in TSS upstream (a), “+1” and “+2” nucleosomes in TSS downstream (b), “1” and “2” nucleosomes to CTCF-binding sites (c), “-2” and “-1” nucleosomes in TTS upstream (d), and between “+1” and “+2” nucleosomes in TTS downstream (e) are shown by scatter plots. Trend lines are fit to linear regression, whose adjusted R2 are shown together with PCC. See S3 Fig for more illustration of other nucleosome pairs in TSS, CTCF, and TTS regions. The on-site correlations (between the two different factors’ level at the same “before” nucleosomes) are shown with grey dots for comparison. Some non-linear correlations are illustrated by logarithmic converted Mark-A signals (x-axis) versus raw Mark-B signals (y-axis). (f) Co-citation analysis of the functional coherence of the inter-nucleosome consensus networks.
Fig 3.
Visualization of H2A.Z and H4K20me1’s gene-wise positive association and nucleosome-wise negative association.
(a) Nucleosome and H4K20me1/H2A.Z profiles around -2000 ~ +2000 bp of the TSS regions. TSSs are decreasingly ranked by transcription levels, and evenly divided into 10 groups. (b) Comparison between normalized H4K20me1 (green) and H2A.Z (red) profiles around TSSs. Raw nucleosome profiles (grey) are shown. (c) H4K20me1/H2A.Z correlation for “+1” nucleosome versus surrounding nucleosomes along TSS downstream. For each group, the cross-TSS PCC between “+1” and “+2”, “+3”, …, “+10” nucleosomes were calculated with each bin of 100 TSSs respectively. See S5 Fig for more illustration on different nucleosomes. (d) Same as (c), but for H2A.Z/H4K20me1 correlation for “-1” nucleosome versus surrounding nucleosomes along TSS upstream.
Fig 4.
Association of the H4K20me1-H2A.Z interaction with nucleosome profile patterns around TSS.
(a) Four clusters of the nucleosome profiles in -2000 ~ +2000 bp regions surrounding TSS. The H4K20me1-H2A.Z interaction was obtained in Cluster 1, 3 and 4. (b) Length, depth and size of nucleosome free regions, and gene transcription levels for each of the four TSS clusters. The mean and standard error of the mean (SEM) are shown. One-way ANOVA was used for overall comparison, and TukeyHSD test was used to calculate the p-values between Cluster 2 and other clusters. (c) Pipeline for calculating Nucleosome Phasing Index. Nucleosome Phasing Index was quantified based on the Jensen-Shannon divergence distance (JSD) between the binary template vector and the binary-converted nucleosome-signal vector of each TSS (see Methods for details). (d) Phasing Index for each of the four TSS clusters. Statistical tests were same as (b). (e-g) Cross-group linear regression for H4K20me1-H2A.Z anti-correlation versus the NFR length (e), depth (f), and size (length × depth) (g) downstream of TSSs, respectively. Adjusted R2 and p-values are labeled on the panels. (h-j) Same as (e-g), but for the H2A.Z-H4K20me1 anti-correlation at TSS upstream. (k) Cross-group linear regression for H4K20me1-H2A.Z anti-correlation versus nucleosome Phasing Index. (l) Mechanistic model of the H2A.Z-H4K20me1 inter-nucleosome interaction. For lowly expressed genes, neither H2A.Z or H4K20me1 is present (Scenario 1), where H2A and H4K20 from neighboring nucleosome can weakly interact to form a weak interaction between nucleosomes and phasing around TSS through the weak acidic patch of H2A; If there exist H4K20me1 (Scenario 2), it would disrupt inter-nucleosome interaction, cause loss of phasing, and expose DNA to spurious transcription, which is an unfavorable situation. For highly expressed genes, the presence of both H2A.Z and H4K20me1 (Scenario 3) will make a flexible and controllable strong interaction between H4K20me1 of the neighboring nucleosome through the strong acidic patch of H2A.Z; Alternatively, if H4K20 is not methylated (Scenario 4), the interaction is too strong to be regulated and has a propensity to form condensed chromatin structure. The strong extended H2A.Z acidic patch is represented by a lock.
Fig 5.
H2A.Z knockdown induced decrease of nucleosome free regions and nucleosome phasing at TSS regions.
(a) Nucleosome profiles around TSSs in control group and H2A.Z knockdown group. TSSs are decreasingly ranked by their H2A.Z levels (the sum of normalized H2A.Z signal within -2000 to +2000 bp) in control group. (b) Nucleosome free regions (NFRs) marked in red color for control and knockdown groups. (c-e) Length (c), depth (d), and size (e) of nucleosome free regions. The median and quartile are shown. P-value was calculated by one-tailed t-test. (f) H2A.Z knockdown induced increase of nucleosome signals around TSS versus control group. The differential nucleosome profiles between H2A.Z knockdown and control group were calculated by DANPOS with the setting of “quantile normalization”. (g) Increase of “on-site” nucleosome occupancy on TSSs. TSSs are increasingly ranked by their H2A.Z levels in control groups, and the increase of “on-site” nucleosomes is plotted to show the mean ± SEM values within each bin of 1000 TSSs. Spearman correlation was calculated between the difference of “on-site” nucleosome occupancy at TSSs (the sum of differential nucleosome signals within -90 to +20 bp around TSS) versus the H2A.Z level (the sum of H2A.Z normalized-signals within -2000 to +2000 bp around TSS) of the corresponding TSSs in control group. (h) Cross-group linear regression for H2A.Z levels versus the nucleosome Phasing Index of control sample. Adjusted R2 and p-values are labeled on the panels. (I) Comparison of nucleosome profiles between control and knockdown samples. Nucleosome profiles are aligned within -2000 to +2000 bp around TSSs in a 10 bp resolution. TSSs (>20000 TSSs having non-zero H2A.Z signals at “+1” nucleosomes) are decreasingly ranked by H2A.Z levels at “+1” nucleosomes, and evenly divided into 10 groups.