scTAM-seq enables targeted high-confidence analysis of DNA methylation in single cells

Abstract

Single-cell DNA methylation profiling currently suffers from excessive noise and/or limited cellular throughput. We developed scTAM-seq, a targeted bisulfite-free method for profiling up to 650 CpGs in up to 10,000 cells per experiment, with a dropout rate as low as 7%. We demonstrate that scTAM-seq can resolve DNA methylation dynamics across B-cell differentiation in blood and bone marrow, identifying intermediate differentiation states that were previously masked. scTAM-seq additionally queries surface-protein expression, thus enabling integration of single-cell DNA methylation information with cell atlas data. In summary, scTAM-seq is a high-throughput, high-confidence method for analyzing DNA methylation at single-CpG resolution across thousands of single cells.

Keywords: DNA methylation; Epigenetics; Hematopoiesis; Multi-omic analysis; Single-cell profiling.

Fig. 1

ScTAM-seq identifies cellular subtypes in B cells from peripheral blood. a Overview of the scTAM-seq workflow. b Per amplicon comparison of fraction of cells with at least one sequencing read in the undigested and digested bone marrow sample. c False-negative rate (estimated on B-cell differentiation amplicons, n=424, undigested bone marrow control) and false-positive rate (estimated on the constitutively unmethylated amplicons, n=32, digested bone marrow sample) across all amplicons of the respective class. d Heatmap (binary distance, Ward’s method) of binarized, single-cell DNAm values of 9583 cells across 313 high-performance amplicons in the peripheral blood sample. e Comparison of pseudo-bulk and bulk DNAm for the 424 B-cell differentiation-related amplicons. r: Pearson’s correlation coefficient. The colors correspond to the DNAm clusters in d

Fig. 2

scTAM-seq identifies cellular states associated with proliferation. a Heatmap showing the binarized, single-cell DNAm matrix for 4100 ns-memory B cells in the 313 high-performance amplicons. Five clusters of ns-memory B cells and six CpG clusters were defined based on a hierarchical clustering (binary distance, Ward’s method). b Low-dimensional representation of the binarized data matrix for all cells (naive, cs- and ns-memory B cells) using UMAP. The pseudotime was inferred with Monocle3. c Surface-protein expression within the UMAP-space. The surface-protein expression data was binarized using a cutoff of 1 for the CLR-normalized counts. d Surface-protein expression for the different clusters (ordered by increasing pseudotime) identified in a as barplots. Shown is the mean and two times the standard error within each of the clusters. e Average DNAm value per CpG- and cell cluster were estimated by computing the fraction of all methylated amplicons in a given CpG and cell cluster. The error bar indicates two times the standard error across all cells of a cell cluster. The pie chart indicates the genomic distribution of the CpGs within each CpG cluster according to chromatin states of naive, germinal center, ns-, and cs-memory B cells defined in Beekman et al.

Fig. 3

scTAM-seq captures the B-cell differentiation process in bone marrow. a Heatmap of single-cell DNAm values for 5340 cells and 313 amplicons in the digested bone marrow sample. b–d Visualization of DNAm data in low-dimensional space (UMAP), with labels inferred from bulk DNAm (b), across differentiation pseudotime (inferred with Monocle3, c), or with labels transferred from CITE-seq data (d). e Visualization of surface-protein expression for cells embedded in the DNAm UMAP. The value shows the CLR-normalized expression values (see “Methods”). f, g Relationship between DNAm and log-normalized gene expression across B-cell differentiation showing a positive (f, CXCR5, CpG located in promoter region) and negative correlation (g, SMARCA4, CpG located in intronic region), respectively