Single-cell CUT&Tag analysis of chromatin modifications in differentiation and tumor progression

Abstract

Methods for quantifying gene expression¹ and chromatin accessibility² in single cells are well established, but single-cell analysis of chromatin regions with specific histone modifications has been technically challenging. In this study, we adapted the CUT&Tag method³ to scalable nanowell and droplet-based single-cell platforms to profile chromatin landscapes in single cells (scCUT&Tag) from complex tissues and during the differentiation of human embryonic stem cells. We focused on profiling polycomb group (PcG) silenced regions marked by histone H3 Lys27 trimethylation (H3K27me3) in single cells as an orthogonal approach to chromatin accessibility for identifying cell states. We show that scCUT&Tag profiling of H3K27me3 distinguishes cell types in human blood and allows the generation of cell-type-specific PcG landscapes from heterogeneous tissues. Furthermore, we used scCUT&Tag to profile H3K27me3 in a patient with a brain tumor before and after treatment, identifying cell types in the tumor microenvironment and heterogeneity in PcG activity in the primary sample and after treatment.

Fig. 1 |. scCUT&Tag resolves distinct cell types and maps repressive chromatin domains in early hESC development.

a, Schematic of scCUT&Tag applied to nuclei isolated from cell culture, a model endoderm differentiaton system, blood cells and a human brain tumor. Single cells are then partitioned using either the 10x Genomics or ICELL8 microfluidic systems. b, UMAP embedding of scCUT&Tag for a repressive histone modification, H3K27me3, in K562 (n = 908) and hESC (n = 804) single cells. c, UMAP embedding of scCUT&Tag for a repressive histone modification, H3K27me3, in a 5-d differentiation time course from hESC to definitive endoderm (total n = 1,830). Cell types are colored according to the day along the time course in which they were harvested. d, Top, bar plot representing the percent of single cells (n = 350, 171, 474, 274 and 561 from days 1–5, respectively) that are repressed at each specific gene, where the upper axis corresponds to scCUT&Tag (percent of single cells repressed). Bottom, jitter plot depicting scRNA-seq for similar time points (n = 92, 66, 172, 138,and 188), where the lower axis corresponds to scRNA-seq (normalized messenger RNA counts from GSE75748). From left to right, well-known TF markers for pluripotent, mesendoderm and definitive endoderm cells. mRNA, messenger RNA.

Fig. 2 |. scCUT&Tag for H3K27me3 readily identifies major subtypes in PBMCs.

a, Left, UMAP embedding of single-cell data from PBMCs. Unsupervised clustering revealed five clusters. Right, UMAP projection of downsampled ChIPseq bulk data from primary sorted bulk datasets for major PBMC cell types (see Supplementary Methods for GSE citations) on single-cell CUT&Tag data on left. b, Heat map of genes with significantly low (top) or high (bottom) H3K27me3 signal (CSS) in each cluster (row). Fold change < −2 (top) or > 2 (bottom); q < 0.05 (both). Cell-type-specific genes are highlighted. c, Sparse mixture model clustering (using souporcell) of genotype variant calls from the PBMC data colored by genotype assignment (before multiplet removal). NK, natural killer.

Fig. 3 |. scCUT&Tag data for H3K27me3 for a human glioblastoma primary and relapse sample demonstrate heterogeneity in PcG distribution within tumor cell clusters and cluster enrichment after treatment.

a, UMAP embedding of single cells from a primary human glioblastoma based on H3K27me3 signal. b, Cluster annotation using CSS for key marker genes identifies microglia (PTPRC), neurons (RBFOX3), oligodendrocytes (MOBP) and tumor cells (SOX2). c, UMAP transform and projection of bulk ChIP-seq (monocytes and astrocytes) or bulk CUT&RUN (UW7gsc) onto patient sample. d, Left, UMAP co-embedding of tumor cells from primary and relapse sample. Inset highlights locations of cells from relapse sample. Right, bar plot demonstrating fraction of cells in each sample (Primary and Relapse) that belong to each cluster. e, Left, two pseudotime trajectories starting with cluster T1 (presumed stem-like cluster) and ending in either Cluster T4 (Trajectory 1) or Cluster T2 (Trajectory 2). Right, heat map of 132 significant motif deviations based on H3K27me3 activity within peaks from aggregated tumor cell ATAC-seq data. Motif deviations are ordered by pseudotime. f, UMAP plots for tumor cells colored by deviation scores for selected motifs. Left column shows early motifs in pseudotime that are commonly silenced, including NEUROD1, SNAI2 and TCF12. Middle column shows silenced programs that diverge according to trajectory (NR1D2 in Trajectory 1 and ETV5 in Trajectory 2) or are common across trajectories (RFX4). Right column shows silenced programs specific to terminal pseudotime for Trajectory 1 (HES5), Trajectory 2 (GATA6) or both (DNMT1).