Multimodal chromatin profiling using nanobody-based single-cell CUT&Tag

Abstract

Probing histone modifications at a single-cell level in thousands of cells has been enabled by technologies such as single-cell CUT&Tag. Here we describe nano-CUT&Tag (nano-CT), which allows simultaneous mapping of up to three epigenomic modalities at single-cell resolution using nanobody-Tn5 fusion proteins. Multimodal nano-CT is compatible with starting materials as low as 25,000-200,000 cells and has significantly higher sensitivity and number of fragments per cell than single-cell CUT&Tag. We use nano-CT to simultaneously profile chromatin accessibility, H3K27ac, and H3K27me3 in juvenile mouse brain, allowing for discrimination of more cell types and states than unimodal single-cell CUT&Tag. We also infer chromatin velocity between assay for transposase-accessible chromatin (ATAC) and H3K27ac in the oligodendrocyte lineage and deconvolute H3K27me3 repressive states, finding two sequential waves of H3K27me3 repression at distinct gene modules during oligodendrocyte lineage progression. Given its high resolution, versatility, and multimodal features, nano-CT allows unique insights in epigenetic landscapes in complex biological systems at the single-cell level.

Fig. 1. nano-CUT&Tag (nano-CT).

a, Schematic image of the Tn5 fusion proteins used in the experiments. b, Bar plot depicting number of cells used as input for nano-CT and number of cells recovered. c, Comparison of the antibody- and Tn5-binding strategy between scCUT&Tag and nano-CT. d, Cartoon depiction of the tagmentation and library preparation strategy. The nano-Tn5 is loaded with MeA/Me-Rev oligonucleotides, tagmented genomic DNA is used as template for linear amplification, which is then tagmented in a second round with standard Tn5 loaded with MeB/Me-Rev oligonucleotides. The resulting library is amplified by PCR and sequenced. e, Violin plot depicting number of unique reads per cell obtained by scCUT&Tag and nano-CT targeting H3K27me3 per replicate. Violin plots 1–4 from left show multimodal nano-CT performed without ATAC (1 and 3 from left) or with ATAC-seq (2 and 4 from left), and violin plot 5 depicts unimodal nano-CT experiment. f, Individual UMAP embeddings of the single-modality scCUT&Tag (left) and nano-CT (right) data depicting the identified clusters, (scCUT&Tag: 13,932 cells in 4 biological replicates; nano-CT: 6,798 cells in 1 biological replicate; 200,000 cells used as input) g, UMAP co-embedding of the scCUT&Tag data (13,932 cells in 4 biological replicates) together with nano-CT data (6,798 cells in 1 biological replicate; 200,000 cells used as input). Raw matrices obtained by scCUT&Tag and nano-CT were merged together and analyzed without integration. VASC, vascular; AST, astrocytes; RGCs, radial glial cells; OECs, olfactory ensheathing cells; OPCs, oligodendrocyte progenitor cells; MOLs, mature oligodendrocytes; BG, bergman glia; EXC, excitatory neurons; INH, inhibitory neurons; MGL, microglia.

Fig. 2. Multimodal nano-CT.

a, Cartoon depicting the strategy used to profile multiple epigenomic modalities. Individual Tn5 and nano-Tn5 are loaded with barcoded oligonucleotides that are used in the analysis to identify the source of tagmentation and demultiplex the modalities. b, Violin plots depicting the number of unique fragments per cell per replicate and modality. c, Violin plots depicting FrIP per cell per replicate and modality. d, UMAP embeddings of the multimodal nano-CT data for ATAC-seq, H3K27ac, and H3K27me3. The lines connect representations of the same cells in the individual modalities (4,434 cells in two biological replicates, which passed quality control for all three modalities individually and originate from the three-modal datasets; 200,000 cells used as input for all replicates). e, UMAP embedding of the individual modalities with cluster labels. n = 2 biological replicates—each biological replicate was profiled both by nano-CT with ATAC (3-modal) and nano-CT without ATAC (2-modal): 4,960 cells ATAC-seq, 12,464 cells H3K27ac, 12,763 cells H3K27me3; 200,000 cells were used as input for all replicates. Cell is shown in modality UMAP if it passes quality control in its respective modality regardless of the other modalities. AST_NT, astrocytes non-telencephalon; AST-TE, astrocytes telencephalon; AST_3, astrocytes 3; AST_4, astrocytes 4; INH1–4, inhibitory neurons; EXC1–4, excitatory neurons; MGL1–3, microglia 1–3; MAC, macrophages; VEC, vascular endothelial cells; PER, pericytes; CHP, choroid plexus epithelial cells; EPE, ependymal cells; CHP-EPE, choroid plexus + ependymal cells; BG, Bergmann glia; VSMC, vascular smooth muscle cells; ABC, arachnoid barrier cells.

Fig. 3. ATAC, H3K27ac and H3K27me3 at loci harboring microglia and mature oligodendrocyte marker genes.

Genome browser tracks of the multimodal data for several clusters showing marker peak regions. Markers: Mag for ATAC/H3K27ac in mature oligodendrocytes and Dmkn for H3K27me3 in microglia.

Fig. 4. Quality control and benchmarking of nano-CT.

a, Genome browser nano-CT pseudo-bulk view of the HoxA region on chromosome 6 for all three modalities. b, Principal component analysis (PCA) of pseudo-bulk tracks for each cluster identified from the respective modalities by scCUT&Tag and nano-CT. Top 50 marker regions were selected from each nano-CT cluster and modality, and all peaks were merged and flattened before running PCA. c, Metagene plots showing the signal distribution of H3K27ac and H3K27me3 in astrocyte populations obtained by nano-CT and scCUT&Tag around specific H3K27ac and H3K27me3 peaks. The peaks were defined and selected on the basis of reference scCUT&Tag data. d, Scatter plots matrix showing correlation of H3K27ac and H3K27me3 signal in astrocyte populations defined by scCUT&Tag and nano-CT. r, Pearson correlation coefficient. Cell labels as in Fig. 2. e, Venn diagram showing the genomic overlap of significant H3K27ac and H3K27me3 peaks in cluster AST-TE.

Fig. 5. Multimodal analysis and visualization of the nano-CT data.

a, UMAP embedding of the individual modalities with cluster labels identified through WNN analysis. Embedding is based on individual modalities, whereas cluster identities are assigned from WNN dimensionality reduction. b, Venn diagram showing the overlap of peaks identified from the individual modalities. c,d, UMAP projection and visualization of ATAC, H3K27ac and H3K27me3 signal intensity in single cells at the Foxg1 (c) and Irx2 loci (d). Gray lines connect the cells with same the single-cell barcodes across the different modalities. Clusters for telencephalon astrocytes (AST_TE) and non-telencephalon astrocytes (AST_NT) were selected for the visualization. Aggregated pseudo-bulk tracks for all modalities together with genomic annotations are shown to the right.

Fig. 6. nano-CT reveals sequential H3K27me3 waves during oligodendrocyte differentiation.

a, UMAP embedding showing pseudo-time calculated by slingshot on the basis of WNN dimensionality reduction and cluster identities. b, Scatter plot depicting meta-region score for all modalities (y-axis) and pseudo-time (x-axis). The score was calculated as a sum of normalized score across all regions. The regions were selected on the basis of P value (P < 0.05, Wilcoxon test) and log fold change > 0 at the marker regions of the ATAC modality, and top 200 regions were used. The line depicts local polynomial regression fit (loess) of the data and shaded regions depict 95% confidence intervals. c, Heat map representation of the H3K27me3 signal intensity at the regions the marker regions that are gaining H3K27me3 during oligodendrocytes differentiation (P < 0.05, Wilcoxon test, log fold change > 0, top 200 regions). Each column depicts one single cell and row single genomic region (peak). Cells are ordered by pseudo-time calculated as shown in a. The order of the regions is based on k-means clustering of the matrix with k = 2. d, Scatter plots depicting meta-region score for all modalities (y-axis) and pseudo-time (x-axis). The score was calculated as a sum of normalized score across all regions. The regions were selected on the basis of P value (P < 0.05, Wilcoxon test) and log fold change > 0 at the marker regions of the H3K27ac modality, and top 200 regions were used. The regions were further stratified to wave 1 and wave 2 regions on the basis of k-means clustering as shown in c. The line depicts local polynomial regression fit (loess) of the data and shaded regions depict 95% confidence intervals.

Fig. 7. nano-CT-based chromatin velocity analysis.

a, UMAP projection and chromatin velocity visualization. The chromatin velocity was calculated by using ATAC-seq gene-by-cell matrix as input into the unspliced layer and H3K27ac gene-by-cell matrix into the spliced layer and then running scvelo algorithm using default parameters. b, Phase plots of ATAC-seq and H3K27ac signal for key genes associated with oligodendrocyte differentiation (Mal, Mag). c, UMAP projection of the latent time calculated by the scvelo algorithm. d, Heat map showing H3K27ac signal normalized with sctransform. Rows depict individual top velocity driver genes, sorted by time of value with maximum intensity and columns depict individual cells sorted by latent time. e, Heat map representing gene expression profiles measured by scRNA-seq in the oligodendrocyte lineage. Rows depicts individual genes, clustered by similarity and columns depict single cells ordered in pseudo-time. f, Violin plot showing normalized expression of set of marker genes identified in scRNA-seq dataset, and normalized expression of a set of genes identified as the key driver genes by scvelo. g, UMAP projection and velocity vectors projection of chromatin velocity calculated using H3K27ac gene-by-cell matrix used as input into unspliced layer and H3K27me3 gene-by-cell matrix used as input into the spliced layer and then running the scvelo algorithm using default parameters.

Extended Data Fig. 1. Comparison of nano-CUT&Tag and scCUT&Tag H3K27me3 profiles.

a. qPCR amplification curve for libraries generated using the same amount of input material using scCUT&Tag method, scCUT&Tag combined with linear pre-amplification and nano-CT protocol with two-step tagmentation. b. Bioanalyser plot showing the size distribution of libraries generated using scCUT&Tag and nano-CT. Typical nucleosome profile can only be seen in the scCUT&Tag library. c. UMAP co-embedding of the scCUT&Tag and nano-CT dataset (single modality). The raw count matrices were merged, and dimensionality reduction was done on both datasets together. d. Violin plot showing fraction of reads in peak regions (FrIP) for data collected by scCUT&Tag and nano-CT.

Extended Data Fig. 2. Benchmarking and QC of marker regions identified by nano-CUT&Tag.

a. Heatmap showing the cell by marker matrix for scCUT&Tag dataset. Top markers were selected based on adjusted p-value calculated using Seurat’s FindAllMarkers function (two-sided Wilcoxon rank sum test and corrected for multiple hypothesis testing by Bonferroni correction). b. Heatmap showing the cell by marker matrix for nano-CT dataset. Top markers were selected based on adjusted p-value calculated using Seurat’s FindAllMarkers function (two-sided Wilcoxon rank sum test and corrected for multiple hypothesis testing by Bonferroni correction). c. Boxplot depicting fraction of cells with any signal (counts >1) in the most significantly enriched marker regions (top 100) for the respective cluster. Markers are selected based on adjusted p-value calculated using Seurat’s FindAllMarkers function (two-sided Wilcoxon rank sum test and corrected for multiple hypothesis testing by Bonferroni correction). Lower and upper bounds of boxplot specify 25th and 75th percentile, respectively, and lower and upper whiskers specify minimum and maximum, respectively, no further than 1.5× interquartile range. Outliers are not displayed. d. Violin jitter plot showing the p-value determined by two-sided Wilcoxon rank sum test (Seurat FindAllMarkers function), distribution for the top 50 marker bins for each cluster of nano-CT and scCUT&Tag dataset. *** p value = 1.3×10⁻³⁴ (Two sided wilcoxon rank sum test) e. Volcano plot showing p-value determined by two sided wilcoxon rank sum test (Seurat FindAllMarkers function) and average log2 fold change for top 50 marker bins in nano-CT and scCUT&Tag. f. Genome browser snapshot showing pseudobulk scCUT&Tag, nano-CT and Encode dataset (forebrain, midbrain and hindbrain) for H3K27me3 histone mark. g. Fingerprint analysis of H3K27me3 pseudobulk scCUT&Tag, nano-CT and encode datasets.

Extended Data Fig. 3. QC and integration of multimodal nano-CUT&Tag.

a. Upset plots showing the overlap between cells that pass QC within the different modalities for combinations of H3K27ac and H3K27me3, and also ATAC, H3K27ac and H3K27me3. b. Violin plot showing the distribution of fraction of reads per cell in nano-CT and scCUT&Tag. c. Violin plot showing the distribution of number of reads per in peak regions (FrIP) in nano-CT and scCUT&Tag. d. Violin plot showing the number of fragments per cell for scCUT&Tag and nano-CT datasets downscaled to 30,000,000 reads for each dataset. e. Violin plot depicting the number of fragments per cell for multi-CUT&Tag and nano-CT and for two modalities (H3K27ac and H3K27me3). f. Fraction of reads not mapped, unique, result of linear amplification duplicates or result of PCR duplicates in nano-CT and multimodal nano-CT experiments. g. UMAP co-embedding of CCA-integrated H3K27ac nano-CT together with scRNA-seq dataset. Gene body and promoters were used as gene activity scores for the integration. h. UMAP co-embedding of CCA-integrated nano-CT ATAC-seq together with scRNA-seq dataset. Gene body and promoters were used as gene activity scores for the integration.

Extended Data Fig. 4. UMAP and violin plot visualization of marker H3K27ac regions.

a. UMAP embedding and visualization of marker peak activities for H3K27ac nano-CT. Exact genomic region together with the closest gene is shown in the plot title. b. Violin plot visualization the of peak scores for the same peak regions as in a).

Extended Data Fig. 5. Genome browser tracks at Hox regions.

a. Genome browser tracks for all modalities profiled by multimodal nano-CT (ATAC, H3K27ac, H3K27me3) around the a. HoxB, b. HoxC and c. HoxD loci.

Extended Data Fig. 6. QC of cross-talk between the profiled modalities.

a. Scatter plot showing the signal of scATAC-seq (single modality profiled. from 10x Genomics) and multimodal ATAC-seq performed with nano-CT in cluster astrocytes and in astrocyte-specific peaks. b. The same scatter plot as in a.,but stratified by H3K27ac signal. H3K27ac low represents peaks with the 20% lowest quantile of H3K27ac signal and H3K27ac high represents peaks with 20% highest quantile of H3K27ac signal. c. Metagene heatmaps showing the genomic distribution of the ATAC-seq signal in the same regions as in b. d. Scatter plot showing the cluster astrocytes H3K27ac signal profiled either together with H3K27me3 (without_ATAC) or together with ATAC-seq and H3K27me3 (with ATAC) within the astrocyte-specific peaks. e. The scatter plot of H3K27ac astrocytic signal within peak regions stratified by ATAC-seq signal. ATAC low represents peaks with the 20% lowest quantile of ATAC-seq signal and ATAC high represents peaks with the 20% highest quantile of ATAC-seq signal. f. Metagene heatmaps showing the genomic distribution of H3K27ac signal in the same regions as in e.

Extended Data Fig. 7. QC of cell identities across clusters in individual modalities.

Confusion matrix of broad cell identities between a. ATAC and H3K27ac b. ATAC and H3K27me3 and c. H3K27ac and H3K27me3. Confusion matrices for fine cluster identities across the different modalities for d. ATAC and H3K27ac e.ATAC and H3K27me3 and f. H3K27ac and H3K27me3.

Extended Data Fig. 8. QC of cell identities and markers across clusters in individual modalities.

a. Alluvial diagram of corresponding cell identities between the ATAC, H3K27ac and H3K27me3 modalities. b,c. UMAP projection and visualization of ATAC, H3K27ac and H3K27me3 signal intensity in single cells at the Lhx2 (b) and Foxb1 (c) locus. Gray lines connect the cells with same single-cell barcodes across the different modalities. Clusters telencephalon astrocytes (AST_TE) and non-telencephalon astrocytes (AST_NT) were selected for the visualization.

Extended Data Fig. 9. Multimodal pseudotime and chromatin velocity.

a. UMAP embedding and visualization of pseudotime determined based on the WNN dimensionality reduction. Pseudotime is projected on UMAPs generated by dimensionality reduction of the individual modalities. b. UMAP representation of co-embedded H3K27ac nano-CT dataset and scRNA-seq dataset. Diffusion pseudotime calculated by scvelo is projected onto the H3K27ac UMAP. c. Histogram of correlation of gene expression and H3K27ac across binned, integrated pseudotime categories. d, e Phase plot of H3K27ac and gene expression across binned, integrated pseudotime for d. Itpr2 and e. Mal. f. Scatter plot depicting meta-region score for all modalities (y-axis) and pseudotime (x-axis). The score was calculated as a sum of normalized score across all regions. The regions were selected based on p-value (p < 0.05, Wilcoxon test) and log fold change > 0 at the marker regions of the H3K27ac modality, and top 200 regions were used. The line depicts local polynomial regression fit (loess) of the data and shaded regions depict 95% confidence intervals g. UMAP embedding of the scRNA-seq data from the mouse brain atlas. 20,000 single cells were sampled randomly from the dataset. Wave 1 and wave2 scores were calculated as a mean of scaled gene expression data for genes identified in Fig. 6c. h. Gene ontology enrichment analysis of genes, which are repressed in the wave 2 of H3K27me3-mediated repression. P value was calculated using R package enrichGO, using one sided Fischer’s test with Benjamini-Hochberg correction for multiple hypothesis testing.

Extended Data Fig. 10. Chromatin velocity inference from nano-CUT&Tag data.

a. Phase plots of H3K27ac and ATAC, velocity and expression signal for top 10 most likely velocity driver genes identified by scvelo. b. Heatmap showing ATAC-seq signal normalized with sctransform. Rows depict individual top velocity driver genes, sorted by time of value with maximum intensity and columns depict individual cells sorted by latent time. c. Heatmap showing the chromatin velocity. Rows depict individual top velocity driver genes, sorted by time of value with maximum intensity and columns depict individual cells sorted by latent time. d. Gene ontology enrichment analysis of the most important velocity driver genes. P value was calculated using R package enrichGO, using one sided Fischer’s test with Benjamini-Hochberg correction for multiple hypothesis testing.