Fig 1.
Diagram of histone modification ratios.
The green bar represents the transcribed region, and the dotted lines represent the 5′ terminal of the core promoter, +1 position (TSS), 3′ terminal of the core promoter, and transcription end site (TES), from left to right. (A) Histone modification ratios at the core promoter region. The pink bar represents the region with H3K4me3, the orange bar represents the region with H3K27ac, and the purple bar represents the region with H3K27me3 within the core promoter region. The ratios of these bars to the area of the core promoter region filled with light blue were calculated. (B) Histone modification ratios at the transcribed region. The red bar represents the region with H3K4me3, the orange bar represents the region with H3K27ac, and the purple bar represents the region with H3K27me3 within the transcribed region. The ratios of these bars to the area of the core promoter region filled with light blue were calculated.
Fig 2.
Histone modification dynamics in each CPE group.
The y-axis represents genes to which the core promoters were assigned, and the x-axis represents histone modifications (H3K4me3, H3K27ac, and H3K27me3 from left to right for each histone modification) from embryos to adults. E0h: embryo at 0–4 h; E4h: embryo at 4–8 h; E8h: embryo at 8–12 h; E12h: embryo at 12–16 h; E16h: embryo at 16–20 h; E20h: embryo at 20–24 h; L1: larval stage 1; L2: larval stage 2; L3: larval stage 3; Pupae; Male: adult male. The order of genes reflects the results of hierarchical clustering using the pheatmap package in R. The warm color indicates that the histone modification ratio was high. (A) Heatmap of histone modification dynamics in Inr group (n = 24), (B) Heatmap of histone modification dynamics in DPE group (n = 25), (C) Heatmap of histone modification dynamics in TATA group (n = 33), (D) Heatmap of histone modification dynamics in TATA-DPE group (n = 14).
Fig 3.
Comparison of the histone modification dependency of RNA expression values by linear regression.
Ten-fold cross validation was performed to check that the regression equations reflected the general relationship between the histone modification ratio and the measured log(FPKM) obtained by RNA-seq. The correlation between the measured and predicted log(FPKM) has been represented by a scatterplot. (A) Scatterplot between the measured and predicted log(FPKM) in the Inr group (n = (24 core promoters) × (11 developmental stages) = 264). The correlation coefficient was r = 0.593. (B) Scatterplot between the measured and predicted log(FPKM) in the DPE group (n = (25 core promoters) × (11 developmental stages) = 275). The correlation coefficient was r = 0.613. (C) Scatterplot between the measured and predicted log(FPKM) in the TATA group (n = (33 core promoters) × (11 developmental stages) = 363). The correlation coefficient was r = 0.465. (D) Scatterplot between the measured and predicted log(FPKM) in the TATA-DPE group (n = (14 core promoters) × (11 developmental stages) = 154). The correlation coefficient was r = 0.357. The TATA and TATA-DPE groups were divided into two types based on the distributions in the scatterplots. (E) Scatterplot between the measured and predicted log(FPKM) in the TATAp group (n = 189, for details see the Methods). The correlation coefficient was r = 0.533. (F) Scatterplot between the measured and predicted log(FPKM) in the TATA-DPEp group (n = 51, for details see the Methods). The correlation coefficient was r = 0.245.
Table 1.
The relative importance of each histone modification in determining RNA expression levels was calculated using the regression equations obtained by linear regression analysis using the LMG method [12].
Values were normalized such that the sum of all values was 1.
Fig 4.
Comparison of the histone modification dependency of RNA expression values by random forest.
Non-linear regression by random forest was performed and the correlation between the measured and predicted log(FPKM) has been represented by a scatterplot. (A) Scatterplot between the measured and predicted log(FPKM) in the Inr group (n = (24 core promoters) × (11 developmental stages) = 264). The correlation coefficient was r = 0.854. (B) Scatterplot between the measured and predicted log(FPKM) in the DPE group (n = (25 core promoters) × (11 developmental stages) = 275). The correlation coefficient was r = 0.857. (C) Scatterplot between the measured and predicted log(FPKM) in the TATA group (n = (33 core promoters) × (11 developmental stages) = 363). The correlation coefficient was r = 0.652. (D) Scatterplot between the measured and predicted log(FPKM) in the TATA-DPE group (n = (14 core promoters) × (11 developmental stages) = 154). The correlation coefficient was r = 0.626. The TATA and TATA-DPE groups were divided into two types based on the distributions in the scatterplots. (E) Scatterplot between the measured and predicted log(FPKM) in the TATAp group (n = 178, for details see the Methods). The correlation coefficient was r = 0.878. (F) Scatterplot between the measured and predicted log(FPKM) in the TATA-DPEp group (n = 46, for details see the Methods). The correlation coefficient was r = 0.837.
Table 2.
The variable importance of each histone modification in determining RNA expression levels was obtained by random forest as mean decrease in node impurity.
The matrices were normalized such that their sum was 1.
Fig 5.
Heatmap of the relative histone modification frequency.
The frequency was obtained by calculating the ratio of core promoters whose histone modification ratio was more than 0.5. The y-axis represents the histone modifications, and the x-axis represents the CPE groups. The orders in the y- and x-axes reflect the results of hierarchical clustering using the pheatmap package in R. The warm color represents the high ratio of core promoters whose histone modification ratio is more than 0.5.
Fig 6.
Hypothetical model of the function of the TATA box.
The green bar represents the transcribed region, and the dotted lines represent the 5′ terminal of the core promoter, +1 position (TSS), and 3′ terminal of the core promoter, from left to right. (A) Model of TATA-less core promoters. TATA-less core promoters exhibit reduced temporal changes during development and could be grouped into three types: core promoters with continuous active marks, those with occasional histone modifications (void), and those with continuous inactive marks. Because core promoters of the first type do not have inactive marks, their RNA expression values can be predicted by the status of the active marks alone. (B) Model of TATA-containing core promoters. TATA-containing core promoters exhibit increased temporal changes during development and could be grouped into two types: core promoters whose RNA expression values are dependent on the status of histone modifications, and core promoters whose RNA expression values are uniform, despite the void state of histone modifications. Since core promoters of the first type have both of active and inactive marks, all of the histone modification statuses are informative for prediction of their RNA expression values.