Single-cell chromatin immunocleavage sequencing (scChIC-seq) to profile histone modification

Abstract

Our method for analyzing histone modifications, scChIC-seq (single-cell chromatin immunocleavage sequencing), involves targeting of the micrococcal nuclease (MNase) to a histone mark of choice by tethering to a specific antibody. Cleaved target sites are then selectively PCR amplified. We show that scChIC-seq reliably detects H3K4me3 and H3K27me3 target sites in single human white blood cells. The resulting data are used for clustering of blood cell types.

Extended data Figure 1.

Measurement of H3K4me3 profiles by the scChIC-seq using 100, 300, 1,000 and 3,000 NIH3T3 cells a. H3K4me3 Ab-MNase conjugates are added to the pre-treated cells, such that the conjugates could be bound to the locations with the H3K4me3 histone modification mark. b. A genome browser snapshot showing the H3K4me3 profiles around the locus of Dcaf8 for 100, 300, 1,000 and 3,000 cells. c. A genome browser snapshot showing the H3K4me3 profiles around the locus of Mbtd1 for 100, 300, 1,000 and 3,000 cells

Extended data Figure 2.

scChIC-seq detects H3K4me3 profiles in a small number of cells a. A TSS profile plot for H3K4me3 measured by ChIP-seq using bulk cells (black) and by scChIC-seq using 100 (green), 300 (magenta), 1,000 (blue) and 3,000 (red) cells. b. A scatter plot of the H3K4me3 read density of ChIP-seq (bulk cell) versus that of scChIC-seq (3,000 cells rep1) at the enriched regions identified using the H3K4me3 ChIP-seq (bulk cells). The correlation was computed using the Pearson correlation coefficient. c. Scatter plots for the H3K4me3 scChIC-seq data (3,000 cells rep2) versus H3K4me3 ChIP-seq in NIH-3T3 cells. d. Scatter plots for the H3K4me3 scChIC-seq data (1,000 cells) versus H3K4me3 ChIP-seq in NIH-3T3 cells. e. Scatter plots for the H3K4me3 scChIC-seq data (300 cells) versus H3K4me3 ChIP-seq in NIH-3T3 cells. f. Scatter plots for the H3K4me3 scChIC-seq data (100 cells) versus H3K4me3 ChIP-seq in NIH-3T3 cells. g. A Venn diagram showing the overlap of the enriched regions of H3K4me3 profiles measured by ChIP-seq using bulk cells (red) and by scChIC-seq using 3,000 cells rep1 (blue). h. A histogram showing the fractions of enriched regions identified by scChIC-seq that are overlapped with those identified by bulk cell H3K4me3 ChIP-seq. In each scChIC-seq library (100, 300, 1,000 and 3,000 cells), we computed the precision, which is the fraction of these enriched regions that are overlapped with that identified by bulk cell H3K4me3 ChIP-seq. i. A histogram showing the number of enriched regions identified by scChIC-seq libraries of using 100, 300, 1,000 and 3,000 cells NIH-3T3 cells.

Extended data Figure 3.

TSS profile plots for the H3K4me3 profiles detected by scChIC-seq using H3K4me3 Ab-MNase conjugate (red) and IgG Ab-MNase (blue) with different washing conditions A) 200 mM NaCl, B) 300 mM NaCl, C) 400 mM NaCl, and D) RIPA buffer + 150 mM NaCl.

Extended data Figure 4.

TSS profile plots for the H3K4me3 profiles detected by scChIC-seq using H3K4me3 Ab + PA-MNase conjugate (red) and IgG Ab + PA-MNase (blue) with different washing conditions A) 200 mM NaCl, B) 300 mM NaCl, C) 400 mM NaCl, and D) RIPA buffer + 150 mM NaCl.

Extended data Figure 5.

Comparison between the H3K4me3 profiles obtained by scChIC-seq assays using H3K4me3-MNase or H3K4me3 Ab + PA-MNase. a. A genome browser snapshot showing the H3K4me3 profiles identified by ChIP-seq using bulk cell data (black), scChIC-seq using H3K4me3 Ab-MNase (blue) and scChIC-seq using H3K4me3 Ab + PA-MNase (red). b. A scatter plot of the H3K4me3 read density of ChIP-seq versus that of scChIC-seq using H3K4me3 Ab + PA-MNase. c. A scatter plot of the H3K4me3 read density of scChIC-seq using H3K4me3 Ab + PA-MNase versus that of scChIC-seq using H3K4me3 Ab-MNase. d. A Venn diagram showing the overlap of the enriched regions of H3K4me3 profiles measured by ChIP-seq using bulk cells (black) and by scChIC-seq using H3K4me3 Ab + PA-MNase and 3,000 cells (blue). e. A Venn diagram showing the overlap of the enriched regions of H3K4me3 profiles measured by scChIC-seq using H3K4me3 Ab + PA-MNase and 3,000 cells (blue) and by scChIC-seq using H3K4me3 Ab-MNase and 3,000 cells (red).

Extended data Figure 6.

TSS profile plots of the H3K4me3 profiles around TSS for a) 3T3 cells, b) mouse ESC cells and c) mouse naïve CD4 T cells. In each cell type, the H3K4me3 TSS profiles (blue) are compared to the control IgG (red). d) Two heatmaps showing the clusters of the H3K4me3 enriched regions measured by ChIP-seq using bulk cells (right panel) and scChIC using 3,000 cells (left panel).

Extended data Figure 7.

Application of scChIC-seq to detect the profiles of H3K27me3 profile. a. A Venn diagram showing the overlap of the enriched regions of H3K27me3 profiles measured by ChIP-Seq using bulk cells (red) and by scChIC-seq using 3,000 cells (blue). b. A genome browser snapshot showing profiles of H3K27me3 and Brd4 detected by scChIC and ChIP-seq. Genome browser snapshots showing the H3K27me3 profiles detected by ChIP-seq using bulk cells (top panel in red), by scChIC-seq using 3,000 cells (second panel in blue).

Extended Data Figure 8.

Application of the scChIC-seq to profiling H3K4me3 in single human white blood cells a. A TSS profile plot showing the H3K4me3 profile around TSS for a single cell (red) and for the aggregation of 281 pooled single cells. b. A Venn diagram for comparison between the identified enriched regions from the data by bulk cell ChIP-seq and the pooled 281 single cells by scChIC-seq. c. A scatter plot showing the correlation between the ChIP-seq and pooled top 40 single cell ChIC-seq data on the 52,798 combined H3K4me3 peaks for human white blood cells. The top 40 single cells are selected based on precision. The Pearson correlation between the ChIP-seq and pooled 281 single cell ChIC-seq data is 0.66. d. A boxplot showing the precision from the top 10% single cells (about 48%) and all 242 single cells (48%). They are also compared to the random case, in which reads are randomly positioned in each cell. Precision is defined by the fraction of reads in single cells that are within the enriched regions identified using bulk cell ChIP-seq data. On each box, the central mark indicated the median. The bottom and top edges of the box indicated the 25th and 75th percentiles, respectively. A more detailed explanations of the boxplot could be found in Supplemental Methods. e. A boxplot showing the sensitivity from the top 10% single cells (about 18%) and all 242 single cells (10%). They are also compared to the random case, in which reads are randomly positioned in each cell. Sensitivity is defined by the fraction of enriched regions identified using bulk cell ChIP-seq data that have single cell reads. On each box, the central mark indicated the median. The bottom and top edges of the box indicated the 25th and 75th percentiles, respectively. A more detailed explanations of the boxplot could be found in Supplemental Methods.

Extended data Figure 9.

Application of scChIC-seq to profiling H3K27me3 in single cells a. Genome browser snapshots showing the H3K27me3 profiles from bulk cell H3K27me3 ChIP-Seq data, from the pooled 106 single-cell ChIC-seq data and from 50 individual cells. b. A Venn diagram showing the overlap between the identified enriched regions from the bulk cell H3K27me3 ChIP-seq data and the pooled 106 single cell scChIC-seq data. c. A scatter plot showing the correlation between the bulk cell H3K27me3 ChIP-seq and pooled 84 single cell ChIC-seq data. d. A boxplot showing the precision from all single cells (47%). They are also compared to the random case, in which reads are randomly positioned in each cell. Precision is defined by the fraction of reads in single cells that are within the enriched regions identified using bulk cell ChIP-seq data. e. A boxplot showing the sensitivity for all single cells (9.5%). They are also compared to the random case, in which reads are randomly positioned in each cell. Sensitivity is defined by the fraction of enriched regions identified using bulk cell ChIP-seq data that have single cell reads.

Extended data Figure 10.

Correlation between H3K4me3 scChIC-seq data and scRNA-seq data a. A violin plot showing the relationship between variability in H3K4me3 and heterogeneity in gene expression for T cell. The p-value is computed using Wilconxon ranksum test. The central mark indicated the median. b. A violin plot showing the relationship between co-methylation and co-expression for T cell. The p-value is computed using Wilconxon ranksum test. The central mark indicated the median. c. A violin plot showing the relationship between variability in H3K4me3 and heterogeneity in gene expression for monocytes. d. A violin plot showing the relationship between co-methylation and co-expression for monocytes. e. Two groups of annotated genes are selected from the highly co-methylated peaks (blue) and the highly variable peaks (red) using the T cells identified from the scChIC-seq data. Four cdf plots are plotted for the two groups of genes using the gene expression in B cells (top left), monocytes (top right), T cells (bottom left), and NK cells (bottom right). The p-values for the difference between the gene expression of the two groups are computed using Wilcoxon ranksum test. f. Two groups of annotated genes are selected from the highly co-methylated peaks (blue) and the highly variable peaks (red) using the monocytes identified from the scChIC-se data. Four cdf plots are plotted for the two groups of genes using the gene expression in B cells (top left), monocytes (top right), T cells (bottom left), and NK cells (bottom right). The p-values for the difference between the gene expression of the two groups are computed using Wilcoxon ranksum test.

Extended data Figure 11.

TFs enriched in the highly co-methylated and highly variable peaks are associated with cell-specific expression. a. A volcano plot of the comparison between the enriched TFs that are specific to the highly co-methylated peaks and highly variable peaks in the T cells identified from the H3K4me3 scChIC-seq data. The y-axis is the negative log of p-value in the differential TFs analysis. X-axis is the difference between the mean value of two TF-bias corrected deviations vectors obtained from chromVAR. b. The enriched TFs, which are specific to highly co-methylated peaks in T cells (Panel a), are preferentially expressed in Th1 cells. The bar plot show the gene expression levels (RPKM) of the enriched TFs in Th1 cells and naïve T cells. c. The enriched TFs, which are specific to highly variable peaks in T cells (Panel a), are preferentially expressed in naïve T cells. The bar plot shows the gene expression levels (RPKM) of the enriched TFs in Th1 cells and naïve T cells. d. A volcano plot of the comparison between the enriched TFs that are specific to the highly co-methylated peaks and highly variable peaks in monocytes identified from the H3K4me3 scChIC-seq data. The y-axis is the negative log of p-value in the differential TFs analysis. X-axis is the difference between the mean value of two TF-bias corrected deviations vectors obtained from chromVAR. e. The enriched TFs, which are specific to highly co-methylated peaks in monocytes (Panel c), are preferentially expressed in monocytes. The bar plot shows the gene expression levels (RPKM) of the enriched TFs in monocytes and macrophages. f. The enriched TFs, which are specific to highly variable peaks in monocytes (Panel c), are preferentially expressed in macrophages. The bar plot shows the gene expression levels (RPKM) of the enriched TFs in monocytes and macrophages.

Figure 1.

scChIC-seq detects H3K4me3 profiles in a small number of cells and single cells a. Experimental procedures of the scChIC-seq protocol. Following pre-treatment of fixed cells with RIPA buffer (with 0.2% SDS) for chromatin de-condensation, the Ab-MNase conjugates are added to allow Ab binding. Following washing of the unbounded and excess Ab-MNase conjugates in the nucleus, the MNase is activated by addition of calcium ion into the cell nucleus. Standard library preparation procedures are applied to the samples for library preparation and sequencing. b. A genome browser snapshot showing panels of H3K4me3 profiles in NIH 3T3 cells obtained by scChIC-seq analysis using the direct conjugate between H3K4me3 Ab and MNase. The top panel in black refers to H3K4me3 profiles measured by ChIP-seq using bulk cells. H3K4me3 profiles measured by scChIC-seq using 100 (green), 300 (magenta), 1,000 (blue) and 3,000 (red) cells. c. Genome browser snapshots showing the H3K4me3 profiles from pooled bulk cells ChIP-seq data (Supplemental Methods), pooled 281 single-cell ChIC-seq data and 50 individual cells. The ChIP-seq data sets are downloaded from ENCODE (top panel in blue). The H3K4me3 data from the pooled 281 single cells are displayed in the bottom panel.

Figure 2.

Identification of sub-cell types in white blood cells based on clusters generated from single-cell H3K4me3 profiles a. A t-SNE visualization of cells by applying the t-sne analysis on the consensus matrix obtained from the software SC3. Cell type annotations of clusters are obtained via the analysis in Fig. 2b. Some of the cell type marker peaks which are also the marker peaks of clusters are explicitly shown next to the annotated clusters. b. A heatmap showing negative log2 of p-values for the overlap comparison between the marker peaks of clusters and the cell type marker peaks. c. Genome browser snapshots showing the H3K4me3 profiles from bulk cells ChIP-Seq data and single-cell scChIC-seq data. The ChIP-Seq data for B cells, monocytes, T cells and, NK cells are downloaded from ENCODE (top panel). The H3K4me3 scChIC-seq data from the 12 randomly selected single cells for each identified cell type are displayed in the bottom panels.