Main

Chemical modifications to histone proteins and nucleic acids along chromosomes, collectively referred to as the cell’s epigenome, regulate spatiotemporal gene expression patterns in multicellular eukaryotic organisms1. Analysis of the epigenome has been successfully used to reveal mechanisms of gene regulation and dysregulation in development, disease pathogenesis and aging; however, considerable technical barriers exist in profiling the epigenome of complex tissues due to the heterogeneity of cell types and states within biospecimens. Single-cell epigenomic techniques can circumvent this barrier by revealing the landscape of DNA methylation2, chromosome conformation3, chromatin accessibility4, histone modifications5,6,7 and transcription factor binding8 at single-cell resolution. Single-cell multiomic assays that jointly survey multiple molecular modalities, including gene expression9,10,11,12,13 or protein abundance14,15, further help to decipher the complex interplay between the epigenome and transcriptional machinery. In particular, single-cell CUT&Tag (scCUT&Tag) has enabled the characterization of active or silenced chromatin states at candidate cis-regulatory elements (cCREs) within different cell types of primary tissues16,17. However, current single-cell epigenomic assays have been slow for general adoption, owing to factors such as lengthy procedures and lack of general accessibility.

Here, we report Droplet Paired-Tag, a fast and broadly accessible technique for producing high-quality single-cell joint profiles of histone modifications and transcriptomes in parallel, which has the potential to be quickly adopted by the research community for interrogating the dynamic and cell-type-specific epigenomic landscapes in complex tissues. The key modifications compared to the initial combinatorial indexing-based Paired-Tag protocol include the adaptation of a commercially available microfluidic platform (that is, 10x Chromium Single Cell Multiome) to introduce cellular barcodes and a simplified protocol to prepare sequencing libraries. The new procedure, therefore, offers three key advantages. First, changing to a droplet-based platform greatly shortens the hands-on time in both the molecular barcoding and library preparation steps (Extended Data Fig. 1). As a result, Droplet Paired-Tag can be performed in less than 1.5 days from nuclei preparation to sequencing library construction, considerably shorter than the conventional 3-day-long Paired-Tag procedure. Second, this design can also be more easily adapted owing to the wide availability of the commercial 10x Chromium platform and reagent kits. Third, the simplified procedure also brings improved performance in identifying cCREs and correlating chromatin states of distal elements to expression levels of putative target genes.

In Droplet Paired-Tag, nuclei are first permeabilized, followed by targeted tagmentation using primary antibodies to a histone modification; these antibodies are precoupled with protein A–Tn5 transposase fusion proteins (pA–Tn5) in a procedure modified from the reported CUT&Tag18 method (Methods). The resulting nuclei and barcoded beads are then coencapsulated into droplets with a microfluidic device (10x Genomics Chromium X controller). Two types of oligonucleotides with the same set of barcodes are embedded in the beads: (1) a barcoded poly(dT) oligonucleotide to label cDNA and (2) a capturing oligonucleotide to label DNA fragments derived from tagmentation. Reverse transcription is performed in the droplets with the barcoded poly(dT) oligonucleotide. A ligation reaction is simultaneously performed to attach barcoded capturing oligonucleotide to the tagmented chromatin fragments. To facilitate this ligation process with our in-house pA–Tn5 fusion, we designed the pA–Tn5 adaptor with 3′-extended bridge-pMENTs sequence reverse complementary to the capturing oligonucleotide sequence (Fig. 1a, Extended Data Fig. 2a and Supplementary Table 1). After completion of the reactions, the reverse transcription products and tagmented DNA fragments from the same cells are both labeled with the same unique cellular barcodes. The droplets are then dissolved, and the cDNA and chromatin fragments are purified, amplified and split for sequencing library construction following the manufacturer-recommended protocols (10x Chromium Single Cell Multiome; Extended Data Fig. 2b–d).

Fig. 1: Joint profiling of transcriptomes and histone modifications in single cells using Droplet Paired-Tag.
figure 1

a, Schematic overview of Droplet Paired-Tag; TSO, template switch oligo. b, Distribution of the unique number of fragments per cell across different single-cell epigenomic assays with the indicated samples and histone targets; n = 13,664 (Droplet Paired-Tag, H3K27ac), 4,501 (Droplet Paired-Tag, H3K27me3), 2,066 (Paired-Tag, H3K27ac), 418 (Paired-Tag, H3K27me3), 3,031 (CoTECH, H3K27me3), 5,111 (scCUT&Tag, H3K27ac), 9,039 (scCUT&Tag-pro, H3K27ac) and 12,585 (scCUT&Tag-pro, H3K27me3); PBMCs, peripheral blood mononuclear cells. c,d, Distribution of unique transcript (c) or gene (d) numbers per cell across different single-cell multiomic assays. For all box plots, hinges were drawn from the 25th to 75th percentiles, with the middle line denoting the median and whiskers denoting a maximum 2× the interquartile range; n = 13,664 (Droplet Paired-Tag, H3K27ac), 4,501 (Droplet Paired-Tag, H3K27me3), 2,066 (Paired-Tag, H3K27ac), 418 (Paired-Tag, H3K27me3), 3,031 (CoTECH, H3K27me3) and 2,711 (10x Multiome, ATAC). e, Genome-wide SCCs between mESC histone modification datasets from Droplet Paired-Tag (indicated with ‘sc’), bulk CUT&Tag and ChIP–seq (indicated with ‘bulk’); rep, replicate. f, Genome browser view showing examples of pseudo-bulk and single-cell histone modification signals of Droplet Paired-Tag in mESCs, along with bulk ChIP–seq and CUT&Tag. The pluripotent gene (Sox2), along with its superenhancer (SCR), and the repressed gene (Gata4) are highlighted in pink; chr, chromosome. g, Number and overlap of peaks called using H3K27ac Droplet Paired-Tag compared to those from ChIP–seq datasets. h, Scatter plot showing the relationships between the total number of H3K27ac peaks called and the number of nuclei after downsampling. The dashed line indicates the cutoff of the number of nuclei that could recover 85% of peaks.

Source data

Full size image

As a proof of concept, we first used Droplet Paired-Tag to analyze histone modifications (H3K27ac and H3K27me3) jointly with the transcriptome in individual mouse embryonic stem cells (mESCs). The complexity of the DNA-dedicated libraries corresponding to histone modifications (median number of 1,448 and 3,224 fragments per nucleus for H3K27ac and H3K27me3, respectively) was comparable to that of the combinatorial indexing-based Paired-Tag. Library complexity was also similar to or higher than library complexities from other published methods for analyzing histone modifications in single cells, such as scCUT&Tag7, CoTECH and scCUT&Tag-pro (Fig. 1b). Compared to H3K27ac, a higher fraction of reads from H3K27me3 Droplet Paired-Tag experiments correspond to mono-, di- and trinucleosome fragments (Extended Data Fig. 3b), consistent with the more compact chromatin structures within Polycomb complex-repressed regions19. For the transcriptomic analysis, Droplet Paired-Tag yielded gene expression-dedicated libraries with comparable complexities as other commonly used single-cell RNA-sequencing (scRNA-seq) platforms (Fig. 1c,d). The gene expression profile detected by Droplet Paired-Tag was in excellent agreement with bulk RNA-seq data from mESCs (Extended Data Fig. 3c). As expected, the detected gene expression levels are, in general, positively correlated with the H3K27ac signal over the transcription start site (TSS) and inversely correlated with H3K27me3 deposition across gene bodies (Extended Data Fig. 3d). These results indicate that Droplet Paired-Tag can reliably capture histone modifications and gene expression simultaneously from the same cell in a high-throughput fashion.

To evaluate the sensitivity and specificity of Droplet Paired-Tag in histone modification profiling, we compared the mESC single-cell data with bulk CUT&Tag and public chromatin immunoprecipitation with sequencing (ChIP–seq) datasets generated from mESCs20,21. For both H3K27ac and H3K27me3, the aggregated single-cell signals faithfully resembled those from bulk CUT&Tag and ChIP–seq experiments (Fig. 1e) and showed high enrichment over peaks identified from ChIP–seq datasets (Extended Data Fig. 3e). As an example, single-cell H3K27ac reads from Droplet Paired-Tag (Fig. 1f) marked the promoter region of the pluripotent gene Sox2 and its downstream superenhancer, while H3K27me3 reads were deposited on genes involved in cellular differentiation and genes that are expressed in specific cell lineages (for example, Gata4). Seventy-two percent of the peaks identified from aggregated single-cell signals from Droplet Paired-Tag overlapped with those from ChIP–seq experiments (Fig. 1g). By downsampling the number of nuclei profiled in Droplet Paired-Tag experiments, we found that the number of H3K27ac peaks detected reached saturation after about 2,500 nuclei, or around 6 million total reads. By comparison, the Paired-Tag required 60% more reads to reach saturation (Fig. 1h and Extended Data Fig. 3f).

To demonstrate the utility of Droplet Paired-Tag for single-cell epigenomic analysis of primary tissues, we used it to analyze histone modifications and gene expression at single-cell resolution in the adult mouse frontal cortex. We performed Droplet Paired-Tag experiments to profile either H3K27ac or H3K27me3 together with RNA expression, each in three biological replicates. After filtering out low-quality cells and potential doublets, we recovered 22,054 nuclei in total, of which 11,874 nuclei were profiled for H3K27ac and 10,180 were profiled for H3K27me3. Clustering of these nuclei based on their transcriptomic profiles identified 20 major cell clusters corresponding to nine glutamatergic neuron types (Snap25+Slc17a7+), six GABAergic neuron types (Snap25+Gad1+) and five non-neuron cell types (Snap25–; Fig. 2a, Extended Data Fig. 4a–d and Supplementary Tables 2 and 3). Most cell types were known to exist in the frontal cortex regions, except for three cell types found in one sample, namely, D12MSN (striatum D1/D2 like medium spiny neurons), OBGA (olfactory bulb GABAergic neurons) and OBGL (anterior olfactory nucleus glutamatergic neurons), which likely originate from anatomical regions proximal to the frontal cortex due to variations in surgical sectioning during sample preparation (Extended Data Fig. 4b and Supplementary Table 4). Interestingly, 17 out of 20 and 18 out of 20 clusters can be independently recovered by clustering using histone modification H3K27ac or H3K27me3 signals (Fig. 2a), respectively, suggesting that cell-type annotations are highly concordant between transcriptome- and epigenome-based clustering (Fig. 2b and Extended Data Fig. 5a–d). Additionally, the cell clusters reported in this study were also consistent with those from the previous Paired-Tag dataset and the BRAIN Initiative Cell Census Network (BICCN) reference 10x single-nucleus RNA-seq (snRNA-seq) dataset generated from the mouse primary motor cortex (Extended Data Fig. 5e–h)22.

Fig. 2: Droplet Paired-Tag effectively resolves multiple cell types in the mouse frontal cortex (FC) and identifies the cCREs within each cell type.
figure 2

a, Uniform manifold approximation and projection (UMAP) embedding visualization of frontal cortex Droplet Paired-Tag data clustered and annotated based on the transcriptome (gene expression) and histone modifications (H3K27ac and H3K27me3). b, Overlap of shared annotations between transcriptome and epigenome clustering. c, Representative genome browser view of gene expression and H3K27ac and H3K27me3 distribution over cell-type-specific marker genes. d, Comparison of the number of unique transcripts and genes detected in each cell between Droplet Paired-Tag and Paired-Tag; n = 22,054 (Droplet Paired-Tag, RNA) and 11,026 (Paired-Tag, RNA). e,f, Comparison of the unique fragments and FRiPs in each cell between Droplet Paired-Tag and other methods measuring the histone modifications H3K27ac (e) and H3K27me3 (f) in single cells. All box plot hinges were drawn from the 25th to 75th percentiles, with the middle line denoting the median and whiskers denoting 2× the interquartile range; n = 11,874 (Droplet Paired-Tag, H3K27ac), 5,049 (scCUT&Tag, H3K27ac), 886 (Paired-Tag, H3K27ac), 10,180 (Droplet Paired-Tag, H3K27me3), 4,019 (scCUT&Tag, H3K27me3) and 60 (Paired-Tag, H3K27me3). g, Signal enrichment over CEMBA cCREs in Droplet Paired-Tag and other methods measuring single-cell histone modifications; kbp, kilobase pairs. h, Comparison of the number of H3K27ac peaks from the Droplet Paired-Tag dataset with the original Paired-Tag dataset, intersected with CEMBA-identified cCREs. i, Heat map showing gene expression values from promoter-proximal cCREs. j, Signals of both histone modifications over promoter-proximal cCREs across different cell types. k, Heat map of known motif enrichment for each cCRE module of promoter-proximal cCREs. Examples of known motifs are shown along with the heat map. P values were calculated by one-sided binomial test. FDRs were then calculated to select enriched motifs; TF, transcription factor. l, Schematics for identifying potential targets for cCREs. m, Frequency density plots showing the distribution of SCCs between distal cCRE histone modification signals and their putative target gene expression level. Cutoffs (FDR = 0.05) used to identify cell-type-specific cCRE–gene pairs are also indicated. n, Heat map showing histone modification signals at distal cCREs with potential active or repressive roles and their putative target gene expression levels. Example overrepresented GO terms for genes in selected cell types are shown. P values were calculated by Fisher’s exact test. Benjamini–Hochberg FDRs were then calculated to select overrepresented GO terms.

Source data

Full size image

To jointly analyze Droplet Paired-Tag data corresponding to different histone modifications, we used transcriptome-based clustering and cell-type annotation in all subsequent analyses. For histone modality, pseudo-bulk-level signals showed high concordance with both bulk CUT&Tag and ChIP–seq experiments (Extended Data Fig. 6a,b). Pseudo-bulk single-cell histone signals from cells within each cluster showed that the H3K27ac signal is abundant at TSSs of cell-type-specific genes, whereas these regions are generally silenced by H3K27me3 in other cell types (Fig. 2c and Extended Data Figs. 6c and 7a–d). Compared to scCUT&Tag6, Droplet Paired-Tag yielded a comparable fraction of reads in peaks (FRiP) but higher numbers of unique fragments per nucleus for both histone modifications. Compared to combinatorial indexing-based Paired-Tag12, Droplet Paired-Tag recovered fewer unique reads per nucleus but showed higher FRiP and thus captured a higher number of peak-associated reads. The improvements in both signal sensitivity and specificity likely contributed to the higher resolution in separating cell types (Fig. 2d-f and Extended Data Fig. 6f–i). Compared to the list of open chromatin regions identified from the same brain cell types from a recent single-nucleus ATAC-seq atlas (BICCN)23, Droplet Paired-Tag yielded the lowest level of H3K27me3 signals at the open chromatin regions, indicating minimal off-target Tn5 transposase activities in our procedure (Fig. 2g). To evaluate the sensitivity of Droplet Paired-Tag, we identified the peaks of H3K27ac signals in each cell cluster and retained those that appeared in two or more replicates. The resulting union set of 63,799 peaks was two times more than that detected with the previous combinatorial indexing-based Paired-Tag method (27,522) from a similar number of nuclei (11,874 versus 11,749; Extended Data Fig. 6d). A higher fraction of H3K27ac peaks detected in Droplet Paired-Tag overlapped with the open chromatin regions from the same brain cell types than the previous Paired-Tag dataset (90.8% versus 80.5%; Fig. 2h). Taken together, these results suggest that Droplet Paired-Tag can generate high-quality transcriptomic and epigenomic profiles at single-cell resolution from complex tissues.

To further demonstrate the utility of Droplet Paired-Tag in characterizing cCRE activity states, we examined variations of chromatin states (H3K27ac or H3K27me3) at identified cCREs in different brain cell types23. We classified cCREs as distal or proximal based on their distance to promoter regions (Methods) and performed non-negative matrix factorization to group all the cCREs with sufficient levels of H3K27ac (reads per kilobase (kb) per million (RPKM) > 1) and H3K27me3 (RPKM > 1) signals in at least one brain cell type into 20 cCRE modules, each showing a distinct pattern of cell-type specificity (Fig. 2i,j and Extended Data Fig. 8a). For proximal cCREs, H3K27ac signal in the promoter region showed a strong positive correlation with transcription, while H3K27me3 signal showed an overall inverse correlation (Fig. 2i). Transcription factor binding motif analysis of each CRE module revealed known transcription factors involved. For example, a cCRE module corresponding to the ITL23GL cluster (excitatory neurons from cortex layers 2 and 3) was enriched for NEUROD1 and MEF2C motifs, and the OGC (mature oligodendrocytes) cCRE module was enriched for motifs of oligodendrocyte-critical transcriptional factors, such as SOX10 (Fig. 2k, Extended Data Fig. 8b and Supplementary Table 5).

Droplet Paired-Tag data enable the prediction of putative target genes of distal cCREs due to the joint profiling of gene expression levels and chromatin states from the same cells. We calculated the pairwise Spearman’s correlation coefficients (SCCs) between histone modification signals at distal cCREs and promoter regions of potential target genes within a 500-kb window (Fig. 2l). This analysis identified 20,241 significant cCRE–gene pairs with a positive correlation between H3K27ac signals and gene expression and 4,738 pairs with a negative correlation between H3K27me3 and gene expression (false discovery rate (FDR) of <0.05; Fig. 2m and Supplementary Table 6). Interestingly, Droplet Paired-Tag data captured stronger linkages between cCREs and genes over background than combinatorial indexing Paired-Tag (Extended Data Fig. 8c). Gene ontology (GO) analysis showed that H3K27ac-associated genes in the OGC population were enriched for terms related to the myelination process, consistent with the likely enhancer function of the distal cCREs. However, H3K27me3-associated genes in the same cell type were enriched for terms related to neuronal function (Fig. 2n and Supplementary Table 7). Through integrative analysis with a recently published mouse brain single-cell chromosome contacts dataset24, we found that cell-type-specific cCRE–gene pairs with long-range chromatin contacts were overall more positively (for H3K27ac) correlated than cCRE–gene pairs with no detectable chromatin contacts (Extended Data Fig. 8d,e).

Discussion

In summary, Droplet Paired-Tag is a fast and robust method for joint profiling of histone modifications and gene expression in single cells. We demonstrated the utility and superior performance of this method for analyzing cell-type-specific gene regulatory programs in complex tissues. By using a widely available microfluidic device (that is, 10x Genomics Chromium), this shortened, more easily adaptable procedure will likely facilitate the quick adaptation of this method in the field of epigenetics. Droplet Paired-Tag adds a new tool kit for investigation of the gene regulatory mechanisms in disease and life span.

Methods

Cell culture

mESCs used in this study have been described in a previous study20. Specifically, a hybrid mouse embryonic cell line CAST x s129 was engineered to have both alleles of the SOX2 gene tagged (CAST allele with eGFP; s129 allele with mCherry), and 4 copies of the CTCF binding sites inserted between SOX2 and downstream super-enhancer on the CAST allele. mESCs were maintained in feeder-free and serum-free 2i medium at 37 °C with 5% CO2. To isolate nuclei, mESCs were dissociated with Accutase (Innovative Cell Technologies, AT104), collected by centrifugation, washed twice with PBS (Gibco, 10010023) and resuspended in cold nuclei permeabilization buffer (10 mM Tris-HCl (pH 7.4; Sigma, T4661), 10 mM NaCl (Sigma, S7653), 3 mM MgCl2 (Sigma, 63069), 1× protease inhibitor (Roche, 05056489001), 0.5 U µl−1 RNaseOUT (Invitrogen, 10777-019), 0.5 U µl−1 SUPERaseIn inhibitor (Invitrogen, AM2694), 0.1% IGEPAL CA630 (Sigma, I8896) and 0.02% digitonin (Sigma, D141)) for 1 min. Nuclei were counted using a Bio-Rad TC20 cell counter with 0.4% Trypan Blue (Gibco, 15250061) staining. For each Droplet Paired-Tag experiment, 0.5 million nuclei were used.

Mouse brain dissection and nuclei extraction

All animal work described in this manuscript has been approved and conducted under the oversight of the University of California, San Diego, Institutional Animal Care and Use Committee. Male C57BL/6J mice were purchased from the Jackson Laboratory (000664) at 12 weeks of age and were housed in the barrier facility at University of California, San Diego, under a 12-h light/12-h dark cycle in a temperature-controlled room with ad libitum access to water and food until euthanasia and tissue collection at 16 weeks of age. The temperature in the animal facility was maintained within the range of 20 to 22.2 °C, while the humidity levels varied between 35 and 60%. The frontal cortex was dissected from 16-week-old male mice, snap-frozen in liquid nitrogen and stored at −80 °C before proceeding to nuclei extraction.

Single-cell suspensions were prepared from frozen tissues by Dounce homogenization in douncing buffer (0.25 M sucrose (Sigma, S7903), 25 mM KCl (Sigma, P9333), 5 mM MgCl2, 10 mM Tris-HCl (pH 7.4), 1 mM DTT (Sigma, D9779), 1× protease inhibitor, 0.5 U µl−1 RNaseOUT and 0.5 U µl−1 SUPERaseIn inhibitor). The cell suspension was then filtered through a 30-µm Cell-Tric filter (Sysmex) for debris removal and centrifuged for 10 min at 300g at 4 °C. Cell pellets was washed once with douncing buffer, centrifuged again and resuspended in cold nuclei permeabilization buffer for 10 min. Permeabilized nuclei were pelleted by centrifugation for 10 min at 1,000g and 4 °C and washed with sort buffer (1× PBS (Gibco, 10010023), 1× protease inhibitor (Roche, 05056489001), 0.5 U µl−1 RNaseOUT (Invitrogen, 10777-019), 0.5 U µl−1 SUPERaseIn inhibitor (Invitrogen, AM2694), 1 mM EDTA (Invitrogen, 15575020) and 1% bovine serum albumin (BSA; Sigma, A1595)) once. After resuspension in sort buffer, nuclei were stained with 2 μM 7-AAD (Invitrogen, A1310) for 10 min on ice and were sorted by fluorescence-activated nuclei sorting with an SH800 cell sorter (Sony) for the isolation of single nuclei (Extended Data Fig. 9). Nuclei were collected in collection buffer (1× PBS (Gibco, 10010023), 5× protease inhibitor (Roche, 05056489001), 2.5 U µl−1 RNaseOUT (Invitrogen, 10777-019), 2.5 U µl−1 SUPERaseIn inhibitor (Invitrogen, AM2694), 1 mM EDTA (Invitrogen, 15575020) and 5% BSA (Sigma, A1595)) at 5 °C and immediately centrifuged for 10 min at 750g and 4 °C. Nuclei were washed twice with sort buffer and counted on an RWD C100-Pro fluorescence cell counter with DAPI staining to better estimate the number of nuclei. For each histone modification, around 0.5 million sorted nuclei were aliquoted and used for Paired-Tag experiments.

Assembly of the active transposon complex

Sequences for all custom oligonucleotides used in this study are provided in Supplementary Table 1. To assemble pA–Tn5 complexes suitable for the 10x Single Cell Multiome ATAC + Gene Expression platform, we mixed transposome oligonucleotides (100 µM) in two separate PCR tubes (USA Scientific, 1402-2300) with 25 µl of AdapterA + 25 µl of bridge-pMENTs or 25 µl of AdapterB + 25 µl of pMENTs. Oligonucleotides were annealed in a thermal cycler with the following program: 95 °C for 5 min and slowly cool to 12 °C at a speed of 0.1 °C s–1. Each 1 µl of the annealed transposome DNA was mixed separately with 6 µl of unloaded pA–Tn5 (0.5 mg ml–1, MacroLab) and quickly spun down. The mixtures were incubated at room temperature for 30 min then at 4 °C for an additional 10 min, and equal volumes of assembled pA–Tn5–AdapterA and pA–Tn5–AdapterB were mixed to form functional pA–Tn5 complexes. Assembled transposon complexes can be stored at −20 °C for up to 6 months, and transposon activity was validated with a bulk CUT&Tag assay before use.

Antibodies

Antibodies used in this study include H3K27ac (Abcam, ab177178, recombinant; Abcam, ab4729, polyclonal) and H3K27me3 (Abcam, ab192985, recombinant). We found that antibody specificity is critical for high-quality signals of single-cell histone data. For H3K27ac, although the recombinant antibody yielded a higher fragment number per cell than the polyclonal antibody, its enrichment at TSSs or ChIP–seq peaks was lower (Extended Data Fig. 6e). Therefore, except for replicate 1 of the mouse frontal cortex dataset, all other experiments targeting H3K27ac were performed with the polyclonal antibody. One microgram of antibody was used per Droplet Paired-Tag reaction.

Experimental protocol for Droplet Paired-Tag

A brief description of the Droplet Paired-Tag experimental procedure is provided below. A more detailed, step-by-step protocol is provided in Supplementary Data 1 and on the Protocol Exchange25.

Antibody-guided tagmentation

pA–Tn5 and primary antibody were preconjugated during nuclei extraction. One microgram of primary antibody and 1 µl of assembled pA–Tn5 were premixed in 35 µl of MED1 buffer (20 mM HEPES (pH 7.5), 300 mM NaCl, 0.5 mM spermidine, 1× protease inhibitor cocktail, 0.5 U µl−1 SUPERaseIn RNase inhibitor, 0.5 U µl−1 RNaseOUT, 0.01% IGEPAL CA630, 0.01% digitonin, 2 mM EDTA and 1% BSA) and rotated at room temperature for 1 h, as a previous study showed that high-salt conditions are critical to reducing undesired open chromatin background9. Nuclei extracted with the above-described protocol were also resuspended in MED1 buffer, and 0.15 million–0.50 million nuclei were distributed into the premixed antibody and pA–Tn5 to a final volume of 75 µl. The mixtures were rotated at 4 °C overnight for epitope targeting along with pA–Tn5 tethering.

After overnight incubation, nuclei were isolated by centrifugation for 10 min at 300g and 4 °C and washed with MED2 buffer (20 mM HEPES (pH 7.5), 300 mM NaCl, 0.5 mM spermidine, 1× protease inhibitor cocktail, 0.5 U µl−1 SUPERaseIn inhibitor, 0.5 U µl−1 RNaseOUT, 0.01% IGEPAL CA630, 0.01% digitonin and 1% BSA) three times to remove excess antibody and pA–Tn5. Tagmentation was next performed in 50 µl of MED2 buffer supplemented with 10 mM MgCl2 (Sigma, M1028) at 550 r.p.m. and 37 °C for 60 min in a ThermoMixer (Eppendorf). The tagmentation reaction was terminated by adding an equal volume of stop solution (10 mM Tris-HCl (pH 8.0), 20 mM EDTA, 2% BSA, 2× protease inhibitor cocktail, 1 U µl−1 SUPERaseIn inhibitor and 1 U µl−1 RNaseOUT). Nuclei were spun down for 10 min at 500g and 4 °C and washed once with 1× nuclei buffer (10x Genomics, PN-2000207; supplemented with 1 mM DTT, 0.5 U µl−1 SUPERaseIn inhibitor and 0.5 U µl−1 RNaseOUT). Finally, nuclei were resuspended in 10 µl of 1× nuclei buffer, and 10,000–16,000 nuclei were aliquoted into PCR tubes with a total volume of 8 µl. Normally, the nuclei recovery rate is around 30–60% depending on the starting input materials. To assess CUT&Tag performance, around 10,000–50,000 nuclei were used for bulk analysis. Fragmented DNA in these nuclei was purified with MiniElute (Qiagen, 28004) columns and amplified for quality control assessment. Seven microliters of ATAC Buffer B (10x Genomics, PN-2000193) was added to 8 µl of nuclei mixture to reach a final reaction volume of 15 µl, the same as specified in the user manual of the Chromium Next GEM Single Cell Multiome kit (CG000338, RevF), except that we substituted 3 µl of ATAC Enzyme B (10x Genomics, PN-2000265/PN-2000272) with 1× nuclei buffer.

Reverse transcription, cell barcoding and library preparation

Barcoding reaction mixtures were prepared as described in the manual for the Chromium Next GEM Single Cell Multiome kit. Sixty microliters of prepared master mix was added to 15 µl of nuclei mixture before being loaded onto a Chromium Next GEM Chip J and proceeding to droplet generation with a Chromium X microfluidic system (10x Genomics). Reverse transcription and cell barcoding were performed inside 10x GEM. Final DNA and RNA library amplification was performed according to the Chromium Single Cell ATAC Library kit manual except that we used an increased number of amplification cycles (12–13 in total) for histone modality libraries.

Preprocessing of Droplet Paired-Tag data

All sequencing was performed with an Illumina Nextseq2000 sequencer. Droplet Paired-Tag fastq files were demultiplexed using cellranger-arc (v2.0.0) with the command ‘cellranger-arc mkfastq’; however, DNA and RNA data were preprocessed using cellranger-atac (v2.0.0) and cellranger (v6.1.2), respectively, and barcodes were manually paired with a custom script using the matching relationship provided in cellranger-arc. To select high-quality nuclei, we first aggregated histone modification data from the same sample and performed peak calling to find narrow peaks (for H3K27ac) or broad peaks (for H3K27me3) using MACS2 (v2.1.2)26. We then filtered the histone modality based on the per cell fragment number and FRiP. We next selected nuclei by pairing pass-filtered nuclei from both modalities (histone modification and transcriptome), as shown in Extended Data Fig. 3a. Before clustering, nuclei with a high fraction of mitochondrial and ribosomal RNA reads were filtered out. Nuclei with an extremely high number of reads were also filtered out because most of them were doublets. Potential doublets were identified and removed using Scrublet27 for individual RNA datasets, and corresponding DNA barcodes were also removed.

For genome browser track generation, sample- or cell-type-specific bigwig files were generated from bam files with deepTools (v3.5.1)28 and visualized in Integrative Genomics Viewer (v2.15.4)29.

For FRiP calculation, duplicate reads were removed using Samtools (v1.14)30 and Picard MarkDuplicates (v2.25.0)31, taking barcode information into account. Peak calling was then performed using MACS2 with default parameters, except that for H3K27me3, we called broad peaks with ‘–broad’ due to its broad domain enrichment. Preprocessing pipelines and scripts are shared at https://github.com/Xieeeee/Droplet-Paired-Tag.

Analysis of Droplet Paired-Tag data

Signal enrichment calculation

Density plots and heat maps of signal enrichment over ChIP–seq peaks or CEMBA cCREs were generated using deepTools. Peaks that overlapped with ENCODE blacklist (v2) or CUT&RUN blacklist regions were removed during the calculation of enrichment32,33.

Clustering and annotation

Clustering of single-cell transcriptomic data was performed in R using Seurat (v4.1.0)34 and Signac (v1.6.0)35. In short, gene counts were normalized, and the top 2,500 variable genes were selected for dimension reduction by principal-component analysis. For all datasets, the first 35 principal components were used to correct batch effects with Harmony36, followed by UMAP visualization and Louvain clustering. Marker genes for each cluster were identified using Seurat, with the log2 (fold change) threshold set to 0.25. Annotation of cluster identities was done with marker genes characterized in previous studies. For epigenomic data, 10x fragment files were converted to cell-by-bin matrices using 5-kb non-overlapping genomic bins, and clustering was performed using Signac. Briefly, sequencing depth was normalized using the two-step term frequency-inverse document frequency. The top 85% of genomic bins were selected for linear dimension reduction by singular value decomposition, and batch effects were corrected with Harmony, again followed by UMAP visualization and Louvain clustering. Gene activity scores were computed by signal density in promoter and gene body regions.

To compare clustering results from different modalities, we first annotated the epigenome clusters by nominating the most abundant cell type identified with transcriptome clustering. Overlap coefficients (Oi) were calculated according to the proportion of cells sharing the same labels from both the transcriptome (A) and epigenome clusters (B) in the transcriptome clusters:

$${O}_{i}=\max \left(\frac{{A}_{x}\cap {B}_{i}}{{A}_{i}}\right).$$

Integration with public snRNA-seq datasets

Integration of the Droplet Paired-Tag RNA dataset with the original Paired-Tag dataset and 10x snRNA-seq dataset was performed using Seurat. Briefly, gene counts for all datasets were normalized, and the top 2,000 shared variable genes across datasets were identified as integration features. Dimensional reduction (canonical correlation analysis) was performed to project all nuclei into the same embedding, and ‘anchors’ (pairs of cells from different datasets) were identified by mutual nearest neighbors searching. Low-confidence anchors were filtered out, and shared neighbor overlap between anchor and query cells in an overall neighbor graph was computed. Louvain clustering on the overall neighbor graph was used for coembedded cluster identification. To compare clustering results from different datasets, overlap coefficients (Oi,j) were calculated based on the number of cells sharing the same labels from both the query (A) and reference clusters (B) in the coembedding clusters (C; i indicates query cell type, j indicates reference cell type, and k indicates coembedding cluster):

$${O}_{i,j}=\min \left(\left[\frac{{A}_{i}\cap {C}_{k}}{{A}_{i}}\right],\,\max \left[\frac{{B}_{j}\cap {C}_{k}}{{B}_{j}}\right]\right).$$

Identification of peak set

To identify peaks using H3K27ac data, we adopted a previously described method to perform peak calling and merge peaks across replicates23. Properly paired reads from all pass-filtered nuclei in the same replicate were merged to generate a pseudo-bulk dataset for all biological replicates. After shift correction for the pA–Tn5 cleavage site, peak calling was performed using MACS2 with the following parameters: ‘-q 0.01 –nomodel –shift −75 –extsize 150 –keep-dup all -B –SPMR’. Because we used two different antibodies to reduce variation caused by antibody affinity and specificity, we retained peaks identified in at least two replicates as conserved peaks during merging. Finally, we extended peak summits to a fixed width of 500 base pairs for merging and downstream analysis.

Identification of cCRE modules

To ensure a fair comparison between different techniques, we filtered the CEMBA cCREs list for elements with an arbitrary cutoff (histone modification signal RPKM of >1) in at least one transcriptome-based cluster from both Droplet Paired-Tag and Paired-Tag datasets and retained cCREs with H3K27ac (289,437, 87.9% of cCREs) or H3K27me3 (127,005, 34.9% of cCREs) signal RPKM values of >1 in at least one cluster for downstream analysis. For visualization, shared cCREs (108,319) from H3K27ac and H3K27me3 groups were selected for plotting. cCREs were classified as distal or proximal based on their distance to ±2 kb of the TSS in GENCODE mm10 (vm25). Proximal or distal cCREs were grouped into 20 different modules by non-negative matrix factorization37 based on their histone modification signal intensity. Downstream motif enrichment and GO analysis are based on this classification of cCRE modules. For visualization, 95,799 distal cCREs and 12,520 proximal cCREs that passed both H3K27ac and H3K27me3 signal cutoffs were plotted. Cell types with <100 nuclei were excluded for further analysis (vascular and leptomeningeal cells).

Linking cCREs with putative target genes

We used a previously described method to link cCREs with their putative regulatory genes for both histone modifications38. First, cCREs with co-occupancy of H3K27ac or H3K27me3 within a genomic distance of 500 kb were identified using Cicero39. cCREs with co-occupancy (Cicero score) of >0.1 were retained for further analysis. Next, cCREs were classified as distal or proximal based on their distance to ±2 kb of the TSS in GENCODE mm10 (vm25). In our analysis, only distal-to-proximal pairs were selected for comparison. We calculated SCCs between gene expression and histone modification signal over cCRE across clusters to examine the relationship between coaccessibility pairs. To estimate random background levels, we shuffled the cell identities for each read and calculated the corresponding SCCs. Finally, we fit a normal distribution model and set a cutoff SCC score with an FDR of <0.05 as an empirically defined significance threshold to select significant positively (H3K27ac) or negatively (H3K27me3) correlated cCRE–gene pairs. To compare to the original Paired-Tag dataset on the strength of putative cCRE–gene pairs, we used a Kolmogorov–Smirnov test to calculate the difference between putative cCRE–gene linkages over random background. For comparison to the previously published Paired-Tag dataset, the greatest distance (D) between real and background distributions was calculated for each dataset, respectively.

Motif enrichment and GO analysis

Motif enrichment for each cCRE module was performed using HOMER (v4.11) and the ‘findMotifsGenome.pl’ function40. The displayed motif heat maps in Fig. 2k and Extended Data Fig. 8b were from the results of known motif discovery. GO analysis for each enriched gene set was performed using PANTHER with default parameters, and biological process terms were used for annotation41. To exclude ambiguous terms, we only selected the top enriched terms ranked by fold enrichment × –log10 (adjusted P value).

Integrative analysis of Droplet Paired-Tag and single-nucleus methyl-3C sequencing (snm3C-seq) data

The mouse brain snm3C-seq dataset was downloaded from Gene Expression Omnibus (GEO) with accession number GSE156683 (ref. 24). Contact pairs from individual cells were merged and visualized using pairtools (v1.0.2) at a 5-kb resolution42. To summarize putative cCRE–gene pairs at loop anchors, we first performed loop calling using HiCCUPS (Juicer tools v1.22.01) at resolutions of 5, 10 and 25 kb (ref. 43). Merged loop sets were then intersected with cCRE–gene pairs using bedtools pairtobed (v2.27.1)44.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.