An approach of identifying differential nucleosome regions in multiple samples

Abstract

Background: Nucleosome plays a role in transcriptional regulation through occluding the binding of proteins to DNA sites. Nucleosome occupancy varies among different cell types. Identification of such variation will help to understand regulation mechanism. The previous researches focused on the methods for two-sample comparison. However, a multiple-sample comparison (n ≥ 3) is necessary, especially in studying development and cancer. METHODS: Here, we proposed a Chi-squared test-based approach, named as Dimnp, to identify differential nucleosome regions (DNRs) in multiple samples. Dimnp is designed for sequenced reads data and includes the modules of both calling nucleosome occupancy and identifying DNRs.

Results: We validated Dimnp on dataset of the mutant strains in which the modifiable histone residues are mutated into alanine in Saccharomyces cerevisiae. Dimnp shows a good capacity (area under the curve > 0.87) compared with the manually identified DNRs. Just by one time, Dimnp is able to identify all the DNRs identified by two-sample method Danpos. Under a deviation of 40 bp, the matched DNRs are above 60% between Dimnp and Danpos. With Dimnp, we found that promoters and telomeres are highly dynamic upon mutating the modifiable histone residues.

Conclusions: We developed a tool of identifying the DNRs in multiple samples and cell types. The tool can be applied in studying nucleosome variation in gradual change in development and cancer.

Keywords: Chi-squared test; Differential nucleosome regions (DNRs); Multiple cell types; Nucleosome.

Fig. 1

Identification of the differential nucleosome regions (DNRs) in multiple cell types. Shown are samples of the identification in three cell types (a) and four cell types (b). In each subplot, the normalized reads count (nucleosome occupancy) is shown at the top panel. The P-value for each genomic locus is in the middle panel. The P-value cutoff is 10⁻⁵ (dot line). The third panel shows the DNRs (dot) and the DNRs’ center position (triangle). The gene regions are marked at the bottom panel. Subplot A is for wild type (H4WT) and mutants H4R3A and H4K20A. Subplot B is for four mutants H4K5A, H4K20A, H4K91A and H4K16A

Fig. 2

Comparison of local and global background correction methods in Dimnp. Shown is the matching percentage between two correction methods for cell types H4WT, H3R4A and H4K20A (P-value cutoff = 10⁻⁵). The DNRs were identified with the local and global background correction methods, respectively. The matching percentage (y-axis) is the ratio of the matched DNRs number between the two methods (the global and the global) relative to the total number of DNRs with the global method under a certain deviation (from 1 bp to 800 bp (x-axis))

Fig. 3

Validation for Dimnp using the manually identified DNRs as a standard. Shown are receiver operating characteristic (ROC) curves in comparing the DNRs by Dimnp and the manually identified DNRs. The manual DNRs between two cell types, termed as “the manual”, are the regions where the ratio of the normalized reads count between the two cell types is either ≥1/0.6 or ≤ 0.6. The manual DNRs of three cell types (“the manual-3”) are formed by polling the manual two cell types DNRs together. The pooled manual DNRs are used as a standard to test the performance of both Dimnp and literature tool Danpos [7]. The area under the curve (AUC) is 0.87 for Dimnp

Fig. 4

Percentage of the matched DNRs between Dimnp and Danpos. The percentage (vertical axis) represents the ratio of the number of matched DNRs between Dimnp and Danpos relative to the total number of the DNRs by Dimnp under a specified deviation. The deviation is from 1 bp to 100 bp. By pooling together the DNRs that are pairwise identified by Danpos, we generated the DNRs data for multiple cell types for Danpos. The arrows indicated the matching percentage of the multiple cell-type DNRs between Dimnp and Danpos. In calling the DNRs, P-value ≤ 0.01 for Dimnp, and false discovery ratio ≤ 0.15 for Danpos. Two datasets are used. The first is a three cell-type data consisting of wild type (H4WT), mutant H3R4A and H4K20A. The second dataset consists of three mutants, H3K79A, H3S10A and H3K56A

Fig. 5

An application of Dimnp in analyzing the DNRs in histone mutation strains. a, Promoters enrich the DNRs upon mutations at the modifiable histone residues. Shown are the distributions of the DNRs identified by Dimnp in four sets of multiple cell types. Shown at the top panel is the percentages of promoter (−0.5 k bp ~ +0.3 k bp), intragenic region (+0.3 k bp ~ transcription termination site), and intergenic region (other region) to the genome. b, Average P-value (−log10) near the telomeres in the four sets of the multiple cell-type comparisons