IT-scC&T-seq streamlines scalable, parallel profiling of protein–DNA interactions in single cells

Abstract

Single-cell profiling protein-chromatin interactions is often constrained by complex workflows, high cost, or dependence on specialized equipment. We present indexed tagmentation-based single-cell CUT&Tag-sequencing (IT-scC&T-seq), a modular, plate-based strategy using three-round combinatorial barcoding. IT-scC&T-seq robustly profiles histone modifications and transcription factors with high specificity and throughput, supporting simultaneous analysis of multiple samples and epitopes. Notably, it enables sensitive single-cell mapping of lamina-associated domains, low-abundance chromatin features previously difficult to resolve. Applied to adult mouse mammary gland, the method reveals cell-type-specific chromatin landscapes and lineage-regulatory dynamics. Together, IT-scC&T-seq provides a scalable, cost-effective, and broadly accessible approach for high-resolution chromatin profiling.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13059-025-03661-z.

Background

Transcription factors (TFs) and histone modifications bind to specific DNA regions to modulate chromatin structure and accessibility at cis-regulatory elements (CREs), such as enhancers and promoters, in a precise and coordinated manner [1–3]. These interactions form the molecular foundation for establishing cell-type-specific transcriptional programs in response to developmental cues and environmental stimuli [3]. While single-cell RNA sequencing (scRNA-seq) and single-cell ATAC sequencing (scATAC-seq) are widely used to explore gene expression and the regulatory epigenetic landscape [4–7], single-cell profiling of protein–DNA interactions offers an additional layer of precision by directly measuring regulatory proteins and distinguishing active enhancers and promoters (e.g., marked by H3K27ac) from poised or repressive ones (e.g., marked by H3K27me3).

Mapping these regulatory interactions accurately at single-cell resolution remains challenging. Chromatin immunoprecipitation followed by sequencing (ChIP-seq) is widely used to profile protein–DNA interactions. However, its application to single-cell analysis has been limited due to high cell input requirements, low coverage, and poor signal-to-noise ratios [8, 9]. To overcome these constraints, chromatin immunocleavage-based methods—including CUT&RUN [10, 11], CUT&Tag-seq [12, 13], CoBATCH [14], itChIP [15], ChIL-seq [16], ACT-seq [17], and other variants [18]—have been developed, allowing the interrogation of protein–DNA interactions in thousands of individual cells. Among these studies, cell multiplexing is primarily achieved through specialized microfluidics systems or single-cell combinatorial indexing (sci) based method. Microfluidics-based assays via droplet (customized [19, 20] or 10x Chromium-based [21]) or nanowell (iCell8 [12]) rely on expensive, specialized equipment and offer limited throughput (~ 10,000 cells per run). In contrast, sci approach uses a split-and-pool combinatorial indexing strategy [22], in which nuclei are distributed into multiple wells and barcoded for first-round indexing, then pooled, split again into new wells and add another round of barcode. Each cell accumulates a unique combination of barcodes over successive rounds, allowing retrospective cell identification after sequencing [14, 15, 17, 23]. This approach offers high scalability and can be implemented without specialized instrumentation. However, the repeated pooling and redistribution steps increase workflow complexity and raise the risk of barcode collisions and cell/nuclei doublets [22]. Other emerging methods have miniaturized conventional bulk workflows by physically isolating individual cells or nuclei into microwells or tubes, where chromatin profiling is performed separately for each cell [11, 24, 25]. This one-cell-per-well format bypasses the need for complex split-and-pool but requires extensive manual handling or robotic pipetting across hundreds to thousands of wells. As a result, these protocols are labor-intensive, reagent-inefficient, and inherently limited in throughput, making them less suitable for large-scale studies. Taken together, despite notable advancements, current single-cell chromatin profiling methods continue to face limitations in terms of operational complexity, scalability, or data quality. Therefore, a simplified, scalable, and cost-effective solution remains urgently needed.

Here, we present IT-scC&T-seq, an indexed tagmentation-based single-cell CUT&Tag method that enables parallel profiling of protein–DNA interactions in tens of thousands of single cells without reliance on microfluidic instrumentation. The method employs a modular three-round indexing strategy, offering a scalable and cost-efficient workflow compatible with standard laboratory equipment. We benchmarked IT-scC&T-seq across multiple histone modifications and transcription factors, demonstrating high complexity, low background, and concordance with bulk reference datasets. As a proof of concept, we applied IT-scC&T-seq to profile lamina–chromatin interactions, confirming its sensitivity for mapping low abundant chromatin regulatory domains at single-cell resolution. We further demonstrated its multiplexing capacity by simultaneously profiling four chromatin targets across multiple cell types in a single experiment. Finally, we applied IT-scC&T-seq to mouse mammary gland tissue and integrated the resulting profiles with scRNA-seq data to reveal lineage-specific chromatin regulation and pseudotemporal dynamics. These results indicate that IT-scC&T-seq is well suited for characterizing gene regulatory programs across diverse cell types and chromatin contexts at single-cell resolution.

Results

Design of IT-scC&T-seq

To enable high-throughput profiling of chromatin features in single cells using standard lab equipment, we developed indexed tagmentation-based single-cell CUT&Tag-sequencing (IT-scC&T-seq), building upon our previously established IT-scATAC-seq [26]. This method introduces barcodes in three successive steps corresponding to (1) antibody or sample identity, (2) individual cell identity, and (3) plate identity (Fig. 1a, Additional file 1: Figs. S1, S2). In the first step, fixed cells or nuclei are incubated with a primary antibody targeting the chromatin feature of interest, followed by a secondary antibody to amplify the signal. The nuclei are then divided into N bulk transposition reactions, each containing a uniquely barcoded pA-Tn5 complex (assembled with a pair of Q5XX/Q7XX adapters, Additional file 2: Table S1 and Additional file 1: Fig. S2a). Following transposition, nuclei from each of the N indexed tagmentation reactions are sequentially sorted across the 384-well plate, with one nucleus per well per reaction, such that each well ends up with N nuclei (each from a distinct tagmentation reaction) (Additional file 1: Fig. S3). After lysis, a second-round barcoding PCR is performed to assign a unique well-specific index to each nucleus using a 16 (H5XX) × 24 (H7XX) primer scheme for one 384-well plate (Additional file 1: Fig. S2b). The PCR product is then pooled for another round of PCR to add standard Illumina TruSeq adapter pairs (T5XX/T7XX), allowing multiplexing of multiple plates in a single sequencing run (Additional file 1: Fig. S2c–e). Scaling up of throughput can be achieved easily by increasing either the number of pA-Tn5 barcode combinations or the number of indexed plates (second- or third-round indexes). The entire workflow is compatible with basic laboratory infrastructure and no custom microfluidic systems or proprietary hardware are required [21, 27]. This method is also simple and straightforward, circumventing the preparation of > 96 distinct pA-Tn5 transposomes that involves tedious pipetting and handling. The protocol can be completed in 1–2 days and generates high-complexity libraries for over 10,000 cells at a cost of less than $0.01 per cell (Additional file 1: Fig. S1b, c), making IT-scC&T-seq an accessible, scalable, and cost-effective solution for high-throughput single-cell chromatin profiling.

Fig. 1.

Design and benchmarking of IT-scC&T-seq. a Overview of the IT-scC&T-seq workflow. Isolated nuclei are incubated with primary and secondary antibodies, then split into multiple tagmentation reactions (number of reactions = N) using uniquely barcoded pA-Tn5 complexes. Transposed nuclei from each tagmentation reaction are individually and sequentially distributed into 384-well plates via fluorescence-activated nuclei sorting. After lysis, a first round of barcoded PCR is performed to label wells, followed by pooling and a second round of PCR to add Illumina adapters. Libraries are then sequenced. b Species mixing experiment using H3K4me3-targeted IT-scC&T-seq on a 1:1 mixture of HEK293T and mESCs. The scatter plot shows the number of reads per cell aligned to human vs mouse genomes. 99.83% of cells were species-specific, with 1149 humans, 1151 mouse, and 4 unassigned cells. c Genome browser tracks displaying single-cell and aggregated profiles. Pluripotency genes (e.g., Nanog, Klf4, Oct4, Lin28, Rex1, Zscan2) were enriched in mESCs. Housekeeping genes (e.g., GAPDH, MCM2, LMNB1, TP53) were enriched in HEK293T. Per cell quality control metrics of IT-scC&T-seq in K562 cells across histone marks (H3K4me3, H3K27me3) and transcription factors (RNA Pol II, CTCF): mapping rate (d), duplication rate (e), log10 of unique fragments (f), and FRiP scores (g). h Fragment length distributions demonstrating nucleosomal periodicity across histone marks and TFs. Comparisons of library complexity and FRiP between IT-scC&T-seq and other single-cell chromatin profiling methods for H3K4me3 or H3K4me2 (i) and H3K27me3 (j)

Validation and benchmarking of IT-scC&T-seq across chromatin targets

We purified the hyperactive pA-Tn5 and assembled with adapters (Q5XX/Q7XX) to generate the indexed pA-Tn5 transposome complex, followed by quality control assessments to ensure their tagmentation activities (Additional file 1: Fig. S4 and Additional file 2: Table S1). To evaluate the accuracy of nuclei sorting, the precision of barcode assignment, and the overall sensitivity of IT-scC&T-seq, we applied the method to a 1:1 mixture of human HEK293T cells and mouse embryonic stem cells (mESCs) using an H3K4me3 antibody, which marks active promoters (Additional file 3: Table S2). We recovered 2304 single-cell libraries, with 99.86% assigned to a single species based on genome alignment (> 90% reads aligned to either human or mouse and < 10% to the other), leaving only four ambiguous cells classified as potential doublets (Fig. 1b). We aggregated H3K4me3 single-cell profiles to generate pseudobulk tracks for HEK293T and mESCs, respectively. In mESCs, promoter regions of pluripotency-associated genes—including Nanog, Klf4, Oct4, Lin28a, Rex1, and Zscan2—showed strong H3K4me3 signal enrichment, clearly resolved both in pseudobulk and at the single-cell level; in contrast, HEK293T cells lacked signal at pluripotency loci but displayed robust enrichment at housekeeping and fitness-associated genes (GAPDH, MCM2, LMNB1, TP53) (Fig. 1c). These results demonstrate high accuracy and specificity of IT-scC&T-seq.

To further evaluate the performance of IT-scC&T-seq across multiple chromatin binding targets, we profiled K562 cell lines for two histone markers, H3K4me3 (permissive chromatin) and H3K27me3 (repressive chromatin), and two chromatin-binding factors, CTCF and RNA polymerase II (RNA Pol II), each in 1152 cells. All 4608 cells were successfully demultiplexed and retrieved, with > 98.5% reads mapped per cell (Fig. 1d). Median unique fragment counts per cell were 12,806 (H3K4me3), 9304 (H3K27me3), 3344 (RNA Pol II), and 2803 (CTCF) under a median duplication rate of 42.7%–66.8% (Fig. 1e, f). As expected, histone marks yielded 3–6 times more unique fragments than transcription factors. Notably, a substantial proportion of these fragments (56.4% to 85.4%) were located within peak regions, indicating the robustness of IT-scC&T-seq in capturing bona fide CUT&Tag signals rather than background noise (Fig. 1g). The distribution of fragment lengths indicated capture of subnucleosomal features, as mono-, di-, and tri-nucleosomes across all modifications (Fig. 1h). We benchmarked IT-scC&T-seq against ENCODE ChIP-seq and bulk CUT&Tag [12] for the same set of chromatin marks in K562 cells. Pseudobulk and single-cell profiles generated by IT-scC&T-seq showed high internal consistency and strong correlations with both ENCODE ChIP-seq and bulk CUT&Tag datasets (Pearson r = 0.26–0.99, Additional file 1: Fig. S5a, b), recapitulating signal distribution across representative genomic regions (Additional file 1: Fig. S5c, d). For CTCF, IT-scC&T-seq profiles aligned closely with ENCODE ChIP-seq and were strongly enriched at known CTCF peak regions, whereas bulk CUT&Tag profiles showed weaker correlations with both IT-scC&T-seq and ENCODE data (Additional file 1: Fig. S6a–c). MEME [28] motif discovery algorithm confirmed the significantly enriched canonical CTCF binding footprint in both aggregate and single-cell CTCF libraries (Additional file 1: Fig. S6d). RNA Pol II signals were also found to be concordant between the IT profiles (Additional file 1: Fig. S6e, f). Together, IT-scC&T-seq enables robust profiling of chromatin-binding proteins with high coverage, sensitivity, and specificity, for both histone marks and low-abundance transcription factors.

To further evaluate the performance of IT-scC&T-seq, we compared its metrics in targeting active and repressive histone marks in cell lines with those of previously published scCUT&Tag-seq and scChIP-seq technologies, including iCell8 (microfluidics-based nanowell) [12], customized droplet microfluidics [20, 29], 10 × Genomics Chromium system [21, 29], and sci-based protocols [30]. IT-scC&T-seq consistently produced higher library complexity, with more unique fragments per cell, and equal or better FRiP scores compared to these approaches (Fig. 1i, j). These results indicate that IT-scC&T-seq depicts a more comprehensive and specific chromatin binding landscape, improving sensitivity and specificity across both active and repressive histone marks.

IT-scC&T-seq maps lamina–chromatin interactions at single-cell level

The nuclear lamina is a proteinaceous meshwork of intermediate filaments, including A and B-type lamins, located at the nuclear periphery. Lamina-associated domains (LADs) are genomic regions that interact with the nuclear lamina, and are dynamically reorganized during development, differentiation, and disease progression [31]. Traditional methods for LAD profiling, including ChIP-seq and DNA adenine methyltransferase identification (DamID), are limited by low signal-to-noise ratios and insufficient sensitivity or temporal resolution [31–34]. Given these limitations, we sought to test whether CUT&Tag could provide an improved approach for sensitive and high-resolution mapping of LADs.

To validate this approach, we performed immunostaining to confirm LMNB1 antibody specificity (Fig. 2a), followed by bulk CUT&Tag in H1 human ESCs. The resulting libraries exhibited characteristic subnucleosomal fragment patterns and showed high inter-replicate concordance (Pearson r > 0.94; Additional file 1: Fig. S7a, b). LMNB1-enriched regions identified by CUT&Tag closely matched those obtained from bulk ChIP-seq in H9 ESCs [35] and DamID in H1 ESCs (Fig. 2b, c), with over 75% genomic overlap among called LADs across methods (Fig. 2d). These findings establish CUT&Tag as a robust and reliable method for LAD profiling in bulk.

Fig. 2.

IT-scC&T-seq captures lamina–chromatin interactions at single-cell resolution. a Co-immunostaining of LMNB1 (cyan) and LMNB2 (magenta) in H1 ESCs. Nuclei counterstained with DAPI (blue). Scale bar, 10 μm. b Genome browser tracks of LMNB1 LADs profiled by CUT&Tag, ChIP-seq, and DamID in H1 ESCs with three CUT&Tag replicates shown. c Pearson correlation heatmap of LAD signal across CUT&Tag, ChIP-seq, and DamID replicates. d Venn diagram showing > 75% overlap of LAD regions in human embryonic stem cell lines between CUT&Tag, ChIP-seq, and DamID coverage. Violin plots showing log₁₀ unique fragments (e) and FRiP scores (f) per cell across replicates and four human cell lines (H1, HEK293T, K562, MSCs). g UMAP of 3067 single cells clustered by LSI and colored by replicate (top) and barcode-encoded cell type (bottom). h Z-score heatmap of differentially enriched LADs (FDR < 0.05, log₂FC > 0.5) across cell types. i MA plot of differential LADs between MSCs and H1 cells. Red dots represent 1584 upregulated and 2204 downregulated LAD regions. GO enrichment of genes associated with upregulated (j) and downregulated (k) LADs in MSCs vs H1

The dynamic remodeling of LADs during the cell cycle and differentiation contributes to cellular heterogeneity and functional specialization [36–38]. However, existing methods for LAD profiling typically lack single-cell resolution. To determine whether CUT&Tag is applicable for LAD profiling at single-cell resolution, we applied IT-scC&T-seq to four human cell lines, H1, HEK293T, K562, and H1-derived mesenchymal stem cells (MSCs), as a proof-of-concept analysis. Each cell line was profiled across two technical replicates (two 384-well plates per cell line, n = 384 cells per plate, Additional file 3: Table S2). The number of unique fragments per cell for LMNB1 was lower than that for histone marks or transcription factors in K562, with marked variability across cell types (Fig. 2e). The median FRiP scores based on MACS2 broad peaks ranged from 0.375 to 0.443 across cell types (Fig. 2f). We aggregated 3067 single-cell CUT&Tag profiles and applied iterative Latent Semantic Indexing (LSI) for dimension reduction [39]. Uniform Manifold Approximation and Projection (UMAP) projection of these profiles revealed four distinct clusters, with minimal batch effects between two replicates and high accuracy (Additional file 1: Fig. S7c and Fig. 2g). To assess clustering accuracy, we compared these clusters to cell identities inferred from barcode indexes and observed a 98.24% concordance (Additional file 4: Table S3). These findings indicate that IT-scC&T-seq-generated LAD profiles capture cell-type-associated differences, despite the low complexity of LAD-associated chromatin features. Using the pseudobulk profiles of each cell type, we identified broad LAD domains (Methods) and detected 8054 differentially enriched LAD regions (FDR < 0.05 and log2FC > 0.5) among the cell types (Fig. 2h). We further examined the reorganization of differentiation-associated LADs between H1 and H1-derived MSCs, identifying 1584 upregulated LAD regions (FDR ≤ 0.05, log₂FC ≥ 0.5) and 2204 downregulated LAD regions (FDR ≤ 0.05, log₂FC ≤ − 0.5) (Fig. 2i). Functional enrichment analysis of these differential LADs revealed cell-specific signatures. Among significantly enriched Gene Ontology (GO) terms, upregulated LADs in MSCs were associated with embryonic developmental processes, particularly neuronal differentiation and function (Fig. 2j, Additional file 4: Table S3), potentially reflecting a residual neuroectodermal signature acquired during H1 ESCs to MSCs differentiation [40]. In contrast, LADs lost in MSCs are enriched in mesenchymal differentiation, extracellular matrix organization, and Wnt signaling (Fig. 2k, Additional file 4: Table S3), indicating LAD remodeling during differentiation may be linked to lineage-specific transcriptional programs. Together, these findings demonstrate that IT-scC&T-seq enables high-resolution capture of low-abundance protein-chromatin interaction, such as LAD, and resolves the cell-type variant states.

IT-scC&T-seq enables simultaneous multiplexed profiling of diverse chromatin marks across multiple samples

By adopting a modular three-round indexing strategy, IT-scC&T-seq enables scalable multiplexing of samples and profiling of diverse chromatin states within a single experiment. In the first step, indexed pA-Tn5 transposase introduces barcodes to encode sample and/or chromatin mark identity. This design enables simultaneous profiling of multiple chromatin marks across different samples, maximizing data output from limited input and making it particularly well-suited for scarce or low-input material. To validate this capability, we applied IT-scC&T-seq to three human cell lines (H1, HEK293T, and H1-derived MSCs), each profiled for four chromatin marks (H3K4me3, H3K36me3, H3K27me3, and CTCF), using 12 unique indexed pA-Tn5 complexes to distinguish sample–mark combinations. After the first round of tagmentation-based barcoding, the distinctly barcoded samples were distributed across the same plate such that each well received one nucleus from each of the 12 indexes, where subsequent rounds of indexing and amplification were performed collectively, enabling streamlined analysis and robust data generation (Fig. 3a). Two biological replicates were generated: MHT#1 used one plate (n = 4608 cells), and MHT#2 used two plates (n = 9216 cells) (Additional file 3: Table S2). Assignment of single cells based on barcode identity revealed variability in signal abundance and FRiP scores across both cell lines and chromatin marks (Fig. 3b). Notably, the decreased H3K27me3 signal observed in H1 cells compared to the other two populations was consistent with previous findings [27].

Fig. 3.

IT-scC&T-seq enables multiplexed chromatin profiling across cell lines and marks. a Multiplexing strategy for simultaneously profiling H3K4me3, H3K36me3, H3K27me3, and CTCF in three human cell lines (H1, HEK293T, MSCs). b Violin plots showing the distribution of log₁₀ unique fragments (left) and FRiP scores (right) for each cell type and chromatin mark. c UMAPs of H3K4me3, H3K36me3, and H3K27me3 profiles overlaid with MAGIC-smoothed gene scores for lineage marker genes. d UMAPs of single cells for each mark colored by annotated cell type: H1 (H3K4me3 n = 853, H3K36me3 n = 766, H3K27me3 n = 594), HEK293T (n = 971, 854, 1065), and MSCs (n = 554, 551, 748). e Genome browser tracks showing aggregated and 30 single-cell H3K4me3 profiles at representative loci. f Heatmaps of normalized signal for top upregulated H3K4me3 and H3K36me3 peaks and top downregulated H3K27me3 peaks (± 10 kb around peak centers). g Volcano plots showing differentially enriched peaks between MSCs and H1 for H3K4me3 and H3K27me3 (FDR < 0.1, log₂FC ≥ 1 or ≤ − 1). h Gene Ontology terms enriched for genes near upregulated H3K4me3 and downregulated H3K27me3 peaks in MSCs

To determine whether IT-scC&T-seq histone mark profiles can deconvolute mixed cell populations, we combined single-cell fragment counts for each epitope from two replicates and generated tiled matrices at 2 kb, 10 kb, and 50 kb resolutions for H3K4me3, H3K27me3, and H3K36me3, respectively. Dimensionality reduction via iterative LSI and batch effect correction, followed by clustering (see Methods), revealed three major groups per histone mark (Additional file 1: Fig. S8a). For H3K4me3 and H3K36me3, we manually annotated the clusters based on known lineage-associated genes, while for H3K27me3, we annotated clusters using a combination of lineage-specific markers that lacked repressive marks near the respective marker gene regions (Fig. 3c, d). Higher gene activity, indicated by elevated H3K4me3 and H3K36me3 signals across specific loci, was observed in clusters corresponding to distinct cell populations—gene loci of pluripotency markers such as DPPA4, LIN28A, PRMD14, and SALL4, as well as POU5F1 and members of the HOXB and HOXD families, POU4F1 and XIST, which are highly expressed in HEK293T cells, and mesenchymal stem cell markers like CD44, ENG (CD105), NT5E (CD73), and RUNX1 exhibited enriched H3K4me3 and H3K36me3 signals (Fig. 3c, e and Additional file 1: Fig. S8b). In contrast, H3K27me3 signals were depleted at these gene loci in the respective cell populations (Additional file 1: Fig. S8c). Comparing the barcode-encoded identity with annotated identity yields a clustering and annotation accuracy of 100%, 99.8%, and 98.8% for H3K4me3, H3K36me3, and H3K27me3, respectively. Genomic bin width of 50 kb was used for counting per barcode CTCF signal, and the dataset was separated into three clusters after iterative LSI and UMAP visualization, but lineage-specific markers were absent for further annotation (Additional file 1: Fig. S8d).

We next sought to identify cell-type-specific differential peaks for each histone mark. Cells annotated as the same type were aggregated as pseudobulk replicates to ensure sufficient signal, and differential peak analysis was performed for each cluster compared to the remaining cells (see Methods). For H3K4me3 mark, 5552, 2834, and 1570 significantly upregulated loci (FDR ≤ 0.01, log2FC ≥ 1) were identified in H1, HEK293T, and MSCs, respectively; for H3K36me3 mark, the corresponding numbers were 2416, 4556, and 2139 (FDR ≤ 0.1, log2FC ≥ 1); for H3K27me3 mark, 611, 575, and 2210 loci were significantly upregulated (FDR ≤ 0.1, log2FC ≤ − 1) in H1, HEK293T, and MSCs, respectively; these findings reveal distinct cell-type-specific patterns (Fig. 3f, Additional file 1: Fig. S8e, and Additional file 5: Table S4).

Next, we asked whether cell populations across the active histone marks, H3K4me3 and H3K36me3, could be aligned and cross-correlated. To achieve this, we integrate the data at gene resolution, treating the H3K36me3 signal as a proxy for gene expression level. The UMAP projection of the integrated data, colored by predicted cell types from H3K4me3-H3K36me3 signal linkage, closely matched the cell types annotated using H3K4me3 signals, indicating high concordance between the datasets (Additional file 1: Fig. S8f). To further explore the relationship between H3K4me3 peaks and gene expression, we identified 33,507 H3K4me3 peaks linked to corresponding gene expression levels marked by H3K36me3 (Additional file 1: Fig. S8g).

Lastly, pairwise comparison between H1 and H1-derived MSCs was performed to identify upregulated and downregulated regions associated with H3K4me3 and H3K27me3 (Fig. 3g). GO analysis revealed shared enrichment in developmental and lineage-commitment pathways, including mesodermal and cardiomyocyte differentiation, skeletal development, and mesenchymal migration, among genes with H3K4me3 gain or H3K27me3 loss in MSCs (Fig. 3h), reflecting the lineage-specification regulation under the histone mark binding. Collectively, these results show that IT-scC&T-seq enables simultaneous profiling of multiple chromatin marks from different samples in a single experiment while maintaining high data quality and allowing accurate cell population identification and functional inference of the gene regulation program.

Profiling mammary gland chromatin states using IT-scC&T-seq

To evaluate the feasibility of IT-scC&T-seq in primary tissue samples, we applied the method to adult mouse mammary gland using antibodies against H3K4me3 and H3K27ac, which respectively mark active promoters and enhancers, simultaneously in a single experiment. Single nuclei were isolated from 10-week-old female mice and processed with 12 pairs of indexed pA-Tn5 complexes (6 indexes per each histone mark, Additional file 3: Table S2). After sequential sorting into three 384-well plates, second- and third-round PCR indexing was performed to generate uniquely barcoded single-cell libraries. We recovered high-quality profiles for 10,821 and 4029 single cells for H3K4me3 and H3K27ac, respectively, across two biological replicates. The median unique fragments per cell ranging between 334 (H3K4me3) and 87 (H3K27ac) (Additional file 1: Fig. S9a). Between 74.4% and 93.2% of H3K4me3 fragments and 34.3%–77.4% of H3K27ac fragments overlapped with MACS2-called broad peaks (Additional file 1: Fig. S9b), indicating high signal specificity. Fragment length distribution revealed expected nucleosomal patterns, and both marks showed robust transcription start site (TSS) and enhancer enrichment, respectively (Additional file 1: Fig. S9c, d).

Integration with a 10-month mouse mammary gland 10x scRNA-seq reference dataset and UMAP visualization identified six major clusters (Fig. 4a), which were reproducible across replicates (Additional file 1: Fig. S9e). By examining H3K4me3 signals near TSS regions and H3K27ac signals near enhancer regions of lineage-specific marker genes, we manually annotated the six major cell populations as fibroblasts (Dcn +, Col3a1 +, Pdgfra +), basal cells (Krt14 +, Trp63 +, Mylk +), mature luminal cells (Krt18 +, Foxa1 +, Prlr +), luminal progenitors (Elf5 +, Aldh1a3 +, Kit +), endothelial cells (Cdh5 +, Flt1 +, Pecam1 +), and immune cells (Cd74 +, Ptprc +, Laptm5 +) (Fig. 4a and Additional file 1: Fig. S9f). We then determined marker genes (adjusted P < 0.05, log2FC > 0) for each of these cell types based on differentially expressed genes from the scRNA-seq dataset (Additional file 1: Fig. S9g and Additional file 6: Table S5).

Fig. 4.

IT-scC&T-seq resolves cell-type-specific regulatory programs and chromatin dynamics in mouse mammary gland development. a UMAP embedding of single-cell profiles generated by IT-scC&T-seq for H3K4me3 and H3K27ac and 10x Genomics scRNA-seq data, jointly visualized and colored by manually annotated cell identity. b UMAPs overlaid with module scores derived from lineage-specific marker genes in scRNA-seq (top), H3K4me3 (middle), and H3K27ac (bottom). The following marker genes were used to calculate module scores for each lineage: Ptprc, Rasgef1c (immune); Trp63, Krt5, Krt17, Krt14, Gatb (basal); Plvap, Gpihbp1, Pecam1, Flt1, Cdh5 (endothelial); Foxa1, Fbxo25, Dpp3, Dynlt1b, Fgfr3 (mature luminal); Aldh1a3, Kit, Tamm41, Epha1, Ecm2 (luminal progenitor); Icam1, En1, Tek, Pdgfra, Col3a1 (fibroblast). c Pseudobulk coverage tracks of H3K4me3 aggregated by cell type, displaying representative lineage-defining marker genes enrichment at promoter regions. d Tile plots showing the presence or absence of signal across the Trp63 locus for H3K4me3 (top) and H3K27ac (bottom) at single-cell resolution. Each row represents a single cell, and each column corresponds to a genomic position for each cell population. e Heatmaps of differentially enriched peaks (log₂FC > 0, adjusted P < 0.05, Wilcoxon test) across major cell types for H3K4me3 (top) and H3K27ac (bottom). f UMAPs of H3K4me3, H3K27ac, and scRNA-seq colored by pseudotime projected from transcriptomic lineage inference to all modalities using KNN transfer. g Smoothed log1p gene activity scores plotted against pseudotime for key luminal differentiation regulators (Kit, Elf5, Aldh1a3, Mcts2), showing coordinated dynamics across modalities

To examine whether transcriptomic signatures are consistently reflected at the chromatin level, we calculated module scores from selected marker gene sets and overlaid them onto the single-cell H3K4me3 and H3K27ac profiles (Additional file 6: Table S5). The resulting activity patterns showed enrichment in the corresponding cell populations (Fig. 4b). We visualized the aggregated coverage tracks of H3K4me3 across lineage-defining genes and observed promoter-specific enrichment at loci such as the immune marker Cd74 and the luminal epithelial marker Foxa1 (Fig. 4c). As a representative example, the basal-cell transcription factor Trp63 (p63) exhibited cell-type-specific enrichment of H3K4me3 and H3K27ac fragments across its gene locus at single-cell level (Fig. 4d). To identify cell-type-specific regulatory regions, we aggregated cells by cluster and performed differential peak calling across cell types. This revealed 5881 and 4132 cell-type-enriched peaks for H3K4me3 and H3K27ac, respectively (adjusted P < 0.05, log2FC > 0) (Fig. 4e). These results demonstrate that in a complex tissue environment, IT-scC&T-seq can resolve lineage-specific chromatin features with cell-type resolution.

To illustrate regulatory coordination between chromatin and transcription, we focused on the luminal lineage. Pseudotemporal trajectory was first inferred from scRNA-seq data using Slingshot, then transferred to the chromatin modalities via k-nearest neighbor (KNN) projection (Methods and Fig. 4f). We next identify genes with dynamic activity patterns across the inferred trajectory as developmental drivers (Additional file 6: Table S5). Key known regulators of luminal differentiation, including Kit [41], Elf5 [42], Mcts2 [43], and Aldh1a3 [44], were among the top-ranked driver genes, exhibited concordant pseudotemporal trends in both gene expression and histone modification levels (Fig. 4g).

Together, these results demonstrate that IT-scC&T-seq enables integrative analysis of transcriptional and epigenomic landscapes in complex tissues at single-cell resolution.

Discussion

Cell-specific gene expression programs rely on the coordinated actions of regulatory proteins that dynamically read, write, and erase histone modifications, reposition nucleosomes, and control DNA accessibility. Single-cell omics technologies have significantly improved our understanding of such regulatory complexities [9]; however, the wider application has been hampered by technical limitations, including operational complexity, dependency on specialized equipment, and challenges in scalability and sensitivity.

In this study, we developed IT-scC&T-seq, a modular single-cell CUT&Tag strategy based on indexed tagmentation and PCR barcoding, designed to overcome these limitations. By incorporating a straightforward three-level indexing approach, IT-scC&T-seq allows scalable and high-throughput profiling of protein–DNA interactions in thousands of individual cells without the need for specialized microfluidic platforms. This method simplifies the experimental workflow and reduces costs to less than $0.01 per cell, thereby improving accessibility and practicality within routine laboratory settings.

Benchmarking analyses show that IT-scC&T-seq achieves robust quality control metrics. The species-mixing experiment confirms high accuracy and low doublet rate (> 99.8% purity). Cell line datasets demonstrate library complexity and enrichment specificity, as evidenced by high numbers of unique fragments and fractions of reads within peaks per cell, which are higher or matching current microfluidics- or combinatorial indexing-based methods. Importantly, IT-scC&T-seq consistently and accurately recapitulates known cell identities and regulatory signatures across not only histone marks, but also transcription factors such as RNA Pol II and CTCF, highlighting its sensitivity and robustness.

LADs are characterized by their distinct epigenetic features, such as low gene density, high levels of repressive histone marks (e.g., H3K9me2/3), and DNA methylation, and are linked to various human diseases [31, 32]. The LAD identification initially relied on DamID, which labels lamina-proximal DNA by tethering a bacterial methylase to a nuclear lamina protein [38, 45]. Later studies employed ChIP-seq [33, 34] to identify LADs via lamina-targeted immunoprecipitation. ChIP-seq provides high-resolution binding profiles. However, its application to LAD mapping is limited by the nuclear lamina’s tight chromatin association, which reduces solubilization efficiency and immunoprecipitation yield. This typically requires extensive crosslinking, large starting material, and specialized domain-calling algorithms to distinguish LADs from background noise. In contrast, DamID does not require chromatin fragmentation or antibody enrichment, making it effective for LAD detection, even single-cell level [37, 46]. However, DamID requires exogenous expression of a Dam-fusion protein, which can be infeasible in primary cells or tissues. Additionally, its reliance on adenine methylation motifs (GATC) restricts the resolution (the occurrence of the motif is typically > 1 kb), and the long-term methylation deposition makes real-time LAD tracking unachievable. These limitations highlight the need for an alternative approach that enables. Previously, it remained unknown whether CUT&Tag can be applied for high-resolution, real-time LAD profiling across diverse cell types. Here, we demonstrate for the first time that CUT&Tag can be applied to map LADs, providing a simple and effective approach for profiling lamina–chromatin interactions. Using LMNB1-targeted CUT&Tag, we achieved subnucleosomal resolution and high signal concordance with both ChIP-seq and DamID profiles in bulk H1 cells. Furthermore, IT-scC&T-seq enabled sensitive LAD detection at the single-cell level, capturing cell-type-specific LAD remodeling across four human cell lines. As LAD gain or loss is implicated in gene dysregulation and associated with laminopathies and age-related dysfunction [47–49], IT-scC&T-seq offers a valuable tool to query LAD dynamics in patient samples and disease models, and to elucidate the regulatory roles of nuclear architecture in genome function.

The versatility of IT-scC&T-seq is further shown by its multiplexing capability, which allows simultaneous profiling of multiple histone marks and different biological samples within a single experimental run. Our analyses revealed strong concordance in regulatory profiles between histone modifications such as H3K4me3 and H3K36me3. Moreover, when applied to the complex tissue environment of the mouse mammary gland, IT-scC&T-seq enables integrative analysis with scRNA-seq data to resolve the expected epithelial, stromal, and immune cell populations. This integration provided deeper insights into lineage-specific regulatory elements and dynamic chromatin transitions during luminal differentiation, highlighting the applicability of IT-scC&T-seq in dissecting complex biological systems. Nevertheless, the current study is limited by the lack of validation using human primary cells or patient-derived tissues. Future studies assess the method’s utility in clinically relevant or ex vivo settings.

In addition to its sensitivity and resolution, IT-scC&T-seq offers practical advantages in accessibility and broad utility compared to existing microfluidics-based or sci methods. Microfluidic platforms such as 10x Genomics Chromium and iCell8 require specialized instrumentation, have fixed assay formats, and are limited in scalability or flexibility across protocols. Meanwhile, sci-based strategies rely on multi-round split-pool processes, which add workflow complexity, require more hands-on time, and increase the risk of barcode collisions and doublets. In contrast, IT-scC&T-seq achieves single-cell resolution through indexed pA-Tn5 tagmentation and two rounds of PCR barcoding within a plate-based workflow, requiring only standard laboratory equipment. Although a fluorescence-activated cell (or nuclei) sorter is used for single-cell isolation, it is a widely available instrument routinely used in many biological laboratories for general purposes beyond single-cell omics. All microwell plate-based steps in the IT-scC&T-seq workflow are amenable to automation using benchtop liquid handling platforms (e.g., Labcyte Echo® Acoustic Dispenser used in this study). While not essential, such systems can substantially reduce manual pipetting workload, minimize technical variability, and enhance data consistency. For laboratories without access to automated handlers, these steps can be reliably performed using standard multichannel pipettes. Together, these features support the method’s utility for a wide range of laboratories aiming to explore chromatin regulation without extensive infrastructure or reagent constraints.

Data sparsity remains a major challenge across single-cell epigenomic assays. The stochastic orientation of Tn5-A/B tagmentation results in only half of fragments being successful amplification and sequenced. As IT-scC&T-seq uses on indexed Tn5-A and B adapters, it inherits this limitation. Strategies such as T7 RNA polymerase-based linear amplification [23, 50] or uracil-based adapter switching [51] may help recover additional fragments and improve overall data yield. To preserve nuclear integrity during bulk tagmentation and single-cell processing, we employed light formaldehyde fixation. However, we observed that even low concentrations of crosslinking reagents can negatively impact single-cell data quality. Alternative fixation methods, such as cold methanol treatment, may improve chromatin accessibility and tagmentation efficiency in future implementations.

Like most single-modality omics approaches, IT-scC&T-seq currently profiles one chromatin target per cell, limiting direct inference of combinatorial chromatin states and bivalency analyses. However, its adaptable design is amenable to future development of multi-epitope strategies, potentially employing antibody-pA-Tn5 conjugates for simultaneous multi-target profiling [18, 52–57]. Future integration of IT-scC&T-seq with complementary technologies, including spatial transcriptomics and lineage tracing methodologies, may significantly expand its applications, allowing researchers to reveal dynamic epigenetic changes across spatial and temporal axes with higher resolution and precision.

Conclusions

In summary, IT-scC&T-seq represents a practical, scalable, and cost-effective solution for high-resolution single-cell chromatin profiling. By providing robust sensitivity for challenging chromatin features and compatibility with routine laboratory setups, this method holds the potential to advances the accessibility and applicability of single-cell chromatin analysis, for both fundamental and translational research in epigenetic regulation and cellular heterogeneity.

Methods

Cell culture

The HEK293T cells were routinely maintained in high-glucose Dulbecco’s modified Eagle’s medium (DMEM) containing 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin. The B6 murine ESCs were cultured on gelatin-coated dishes in 2i medium composed of high-glucose DMEM supplemented with 15% stem-cell qualified FBS, 2 mM GlutaMAX, 1 × non-essential amino acids (NEAA), 0.1 mM β-mercaptoethanol, 1000 U/ml recombinant mouse LIF (Merck Millipore), 2i (1 μM PD032591 and 3 μM CHIR99021, Med Chem Express), and 1% penicillin/streptomycin. Human H1 embryonic stem cells (ESCs) were purchased from WiCells and maintained in Essential 8 medium on Matrigel coated plate. The MSCs were reported as before [40] and cultured with MSCs medium composed of αMEM basal medium, 10% FBS, 1 × Glutamax solution, 10 ng/ml bFGF, and 1 × penicillin–streptomycin. All the cells were cultured at 37 °C in 5% CO2 and tested negative for mycoplasma infection using PCR method by the Centre for PanorOmic Sciences, Li Ka Shing Faculty of Medicine.

Antibodies

Antibodies used include H3K4me3 (1:50, Abcam, Ab8580), H3K27me3 (1:50, Cell Signaling, 9733 T), H3K36me3 (1:50, Abcam, ab9050), RNA Pol II (1:50, Abcam, ab26721), CTCF (1:50, Active Motif, 61,311), Lamin B1 (1:100, Abcam, ab239399), Lamin B2 (1: 100, Santa Cruz, sc-56147), guinea pig anti-rabbit (1:200, Novus Biologicals, NBP1-72,763), and Rabbit anti Mouse IgG (H + L) Secondary Antibody (1:200, Thermo Fisher Scientific, 31188).

Purification of transposase pA-Tn5

The 3xFlag-pA-Tn5 plasmid (Addgene # 124,601) was a kind gift from Dr. Steven Henikoff, and the purification follows Tn5 purification protocol [58]. Briefly, the plasmid was transformed into competent Escherichia coli C3013 cells (NEB, C2527I) and induced with 250 µl 1 M isopropyl β-d-1-thiogalactopyranoside (IPTG) at 23 °C for 5 h. Cell pellet was resuspended in 60 ml HEGX buffer (20 mM HEPES buffer pH 7.2, 1.0 M NaCl, 1 mM EDTA, 10% v/v glycerol, 0.2% v/v triton X-100, and 10 mM PMSF) and sonicated using Covaris sonicator with 10 cycles of 30 s on and 30 s off, 40% duty. The cleared pA-Tn5-CBD protein fraction was enriched with chitin resin (NEB, S6651S) at cold-room for 2 h and further washed with 200 ml of HEGX buffer. The Tn5 protein was released by 100 mM dithiothreitol (DTT) cleavage, concentrated with Pierce™ Protein 30 K MWCO Concentrators and dialyzed twice in 1 L 2X HEPES dialysis buffer (100 mM HEPES pH 7.2, 0.2 M NaCl, 0.2 mM EDTA, 20% w/v glycerol, and 2 mM dithiothreitol (DTT). After dialysis, the pA-Tn5 was equilibrated with pure glycerol to 60% concentration. The final pA-Tn5 was quantified by SDS-PAGE and Coomassie Blue staining based on a standard BSA curve. The pA-Tn5 was quantified as 1.82 µg/µl, approximately 25 µM in this study.

Preparation of indexed pA-Tn5 transposome complex

Dissolve the indexed adapters and pA-Tn5 reverse adapters (ordered from IDT, the detailed sequences of adapters and primers are summarized in Additional file 2: Table S1) with annealing buffer (10 mM Tris–HCl pH 8.0, 50 mM NaCl, 2 mM EDTA) to make 200 µM stock. Prepare 15 µl of individual adapter with 15 µl reverse adapter in 200 µl PCR tube and anneal in a thermocycler as follows: 98 °C for 10 min, and slowly cool down to 23 °C at a rate of − 0.1 °C/s. Mix the annealed adapter with 120 µl 25 µM pA-Tn5 and 50 µl coupling buffer (100 mM HEPES–NaOH, 500 mM NaCl, 50% v/v glycerol, 0.5 mM EDTA, 2 mM DTT), and incubate in thermomixer at 25 °C, 1000 rpm for 1 h. The indexed pA-Tn5 transposome was prepared by mixing 20 µl of the paired two pA-Tn5-adapters with 80 µl coupling buffer and the resulting pA-Tn5 transposome complex was 5 µM and can be stored at − 20 °C without activity loss more than 1 year.

Quality control of assembled transposases

Prepare 1 µl 300 ng/µl genomic DNA, 4 µl 5xTAPS-DMF buffer (50 mM TAPS-NaOH pH 8.2, 25 mM MgCl2, 50% DMF), 13 µl H2O, and 1 µl assembled pA-Tn5. Incubate at 55 °C for 10 min, followed by adding 2 µl 10X STOP buffer (0.2% SDS, 40 mM EDTA) and quench at 37 °C 15 min to dissociate pA-Tn5 from tagmented DNA. Add 5 µl 6 × loading dye and run 1.5% DNA gel. The majority of tagmented DNA sizes are less than 1000 bp, indicating the assembled transposases are quantified for downstream experiments. In this study, we randomly picked 14 indexed pA-Tn5 for quality assessment (Additional file 1: Fig. S4).

Bulk CUT&Tag library preparation

Approximately 100,000 H1 ESCs were harvested and resuspended in 100 µl of antibody buffer (20 mM HEPES (pH 7.5), 150 mM NaCl, 4 mM EDTA, 0.5 mM spermidine, 0.05% digitonin, 0.01% NP-40, 1 × protease inhibitors, and 1% BSA). The suspension was incubated on ice for 3 min to extract nuclei. Following this, the nuclei were centrifuged at 600 g for 3 min and resuspended in 10 µl of antibody buffer, and incubated with 10 µl of activated Con-A beads (Beyotime, P2156) for 10 min. After performing two wash steps with antibody buffer, the nuclei-ConA bead complex was resuspended in 50 µl of antibody buffer and incubated overnight at 4 °C with either 1 µg of Lamin B1 antibody (Abcam, ab239399) or an IgG control (Millipore, 12–370). Following additional washes with antibody buffer, secondary antibodies at a dilution of 1:500 were added to facilitate binding to the primary antibodies. Subsequently, 1 µl of 2 µM pA-Tn5 was introduced into each reaction and incubated at room temperature for 1 h. The cell nuclei were then washed with Dig-300 buffer (20 mM HEPES pH 7.5, 300 mM NaCl, 0.5 mM spermidine, 0.01% digitonin, 1 × protease inhibitors, and 1% BSA) and resuspended in 100 µl of tagmentation buffer (Dig-300 buffer supplemented with 10 mM MgCl₂) before being incubated at 37 °C for 1 h. The DNA was subsequently purified using the Qiagen MinElute Kit and subjected to PCR using Nextera indexed primers and NEBNext High-Fidelity 2 × PCR Master Mix. The PCR products were purified and double-selected using AMPure XP beads before quality control and sequencing on the Illumina NovaSeq PE150 platform (ANOROAD GENOME).

Mammary gland tissue digestion and single-cell preparation

All mice were on a C57BL/6 J background raised in the normal 12/12 light–dark cycle. Mammary gland tissues from 10-week-old female mice were dissected and digested overnight in complete EpiCult-B medium containing EpiCult-B proliferation supplements (STEMCELL Technologies, 05610) with 5% FBS, and 10X Gentle collagenase/hyaluronidase in DMEM medium. Following the lysis of red blood cells with NH4Cl, single-cell suspension was obtained through sequential dissociation of the fragments using 0.25% trypsin for 5 min, followed by 5 mg/ml Dispase (Invitrogen, 07913) and 1 mg/ml DNase I (Invitrogen, 07900) for 3–5 min with gentle pipetting, and subsequently filtered through a 40-μm cell strainer (BD Falcon). The supernatant was removed by centrifuging, and the single-cell pellet was resuspended in 1 ml Hanks’ balanced salt solution enriched with 2% FBS. For single-cell RNA-seq library preparation, cells were washed with DPBS twice to remove fragments and finally resuspended in 50–100 μl of PBS with 0.04% BSA. The single-cell RNA-seq libraries from the suspension were generated using the 10 × Genomics Single Cell 3′ Library Construction Kit according to the manufacturer’s instructions by the Genome Centre, HKU. For the subsequent IT-scC&T-seq assay, cells underwent fixation with 0.2% formaldehyde for 10 min on ice to preserve nuclear integrity. The fixation reaction was terminated by quenching with 125 mM glycine, followed by two washes in DPBS to remove residual reagents.

IT-scC&T-seq library preparation

Two hundred thousand native or 0.1%–0.2% formaldehyde crosslinked cells (conditions and indexing schemes are detailed in Additional file 3: Table S2) were used and centrifuged at 500 g for 5 min, resuspended in 500 μl antibody buffer (20 mM HEPES pH 7.5, 150 mM NaCl, 4 mM EDTA, 0.5 mM spermidine, 0.05% digitonin, 0.01% NP-40, 1 × protease inhibitors, and 1% BSA) and incubated for 3 min on ice to extract nuclei.

Nuclei were centrifuged at 800 g for 3 min, resuspended in 200 μl antibody buffer pre-mixed with 1:50 diluted primary antibodies, and incubated at 4 °C on a roller. After overnight incubation, the nuclei were centrifuged at 800 g for 3 min, washed twice with antibody buffer, and resuspended in 200 μl antibody buffer. Secondary antibody (1:200) was added and incubated for 30 min at room temperature. The nuclei were washed twice with 500 μl antibody buffer and twice with Dig-300 buffer (20 mM HEPES pH 7.5, 300 mM NaCl, 0.5 mM spermidine, 0.05% digitonin, 0.02% NP-40, 1% BSA, and 1 × protease inhibitors) supplemented with 4 mM EDTA.

Nuclei were then resuspended in 800 μl Dig-300 buffer supplemented with 4 mM EDTA, aliquoted (200 μl per reaction) into tubes pre-washed with 0.5% BSA-PBS, and incubated with 1.5 μl indexed pA-Tn5 complex (0.02 μM final) at room temperature for 1 h with rotation. The barcodes of the indexed pA-Tn5 were recorded. Nuclei were washed twice with Dig-300 buffer containing 4 mM EDTA and once with Dig-300 buffer. Tagmentation was performed by resuspending nuclei in 100 μl tagmentation buffer (20 mM HEPES pH 7.5, 300 mM NaCl, 0.5 mM spermidine, 0.05% digitonin, 0.02% NP-40, 1% BSA, 1 × protease inhibitors, and 10 mM MgCl₂) and incubating at 37 °C with shaking (850 rpm) for 1 h. Tagmentation was quenched by adding 500 μl 0.5% BSA-PBS with 20 mM EDTA and 2 mg/ml DAPI before single-nucleus FACS sorting into 384-well plates.

The plates were centrifuged at 2000 rpm for 3 min, nuclei were lysed at 55 °C for 10 min, and 100 nl of 10% Triton X-100 was added to quench SDS. Then, 25 nl of 20 μM indexed forward and reverse primers (H5XX and H7XX) and 0.5 μl NEBNext High-Fidelity 2 × PCR Master Mix were dispensed per well. The first-round PCR was carried out with the following program: 72 °C for 5 min, 98 °C for 30 s, 15–17 cycles of 98 °C for 20 s, 63 °C for 30 s, 72 °C for 1 min, final extension at 72 °C for 5 min, hold at 4 °C. Steps involving reagent dispensing into microwell plates, including addition of lysis buffer, PCR primers, and master mix, were automated using the Labcyte Echo® 550 acoustic liquid handler.

PCR products were pooled and purified using a MinElute PCR Purification Kit and eluted in 50 μl nuclease-free water. Undesired fragments were removed via Exo I digestion, followed by 1.0 × AMPure XP bead cleanup, and eluted in 25 μl nuclease-free water. Truseq P5/P7 adapters were added in a second PCR (3 cycles), followed by double-sided size selection (0.55 ×/0.4 ×) with AMPure XP beads. Final libraries were sent for QC and next-generation sequencing by ANOROAD GENOME.

Data processing and alignment of bulk CUT&Tag and IT-scC&T-seq libraries

Adapter and barcode trimming were performed using Cutadapt v4.5 [59] in four sequential steps to accommodate the IT-scC&T-seq library structure. In the first round, H5XX and H7XX (second-round PCR indexes) were trimmed using the -g and -G options with corresponding 8 bp barcode sequences at each end. Subsequent rounds removed linker sequences and extracted Q5XX and Q7XX (first-round pA-Tn5 indexes). All barcodes were concatenated in the order H5XXH7XXQ5XXQ7XX and appended to the read headers using the –rename = CB:Z:(r1.adapter_name)(r2.adapter_name) option and parameters -e 0.1, –no-indels, and –action = trim.

CUT&TAG reads alignment was performed following the previously published protocol [13]. Briefly, reads were aligned to the human hg38 genome for human cells, mouse mm10 genome for mammary gland datasets, or to a hybrid human–mouse reference (hg38 + mm10) for species-mixing experiments using Bowtie2 [60] v2.5.3 with the following parameters: –end-to-end, –very-sensitive, –no-unal, –no-mixed, –no-discordant, –phred33, -I 10, and -X 700. The resulting BAM files were sorted by the cell barcode (CB) tag and split into single-cell BAM files using SAMtools [61] v1.17. Duplicate reads were marked and removed using Picard Tools v3.1.0.

Species mixing experiment data analysis

For each single-cell BAM file, SAMtools idxstats was used to determine the proportion of reads mapped to the human (hg38) and mouse (mm10) genomes. Cells with > 90% of reads mapped to the human genome and < 10% to the mouse genome were classified as human, while cells with > 90% of reads mapped to the mouse genome and < 10% to the human genome were classified as mouse. Cells that did not meet either criterion were labeled as unidentified.

Peak calling and motif discovery

Transcription factor (TF) peaks were identified from the deduplicated single-cell aggregate BAM files using MACS2 [62], with the following parameters: -f BAMPE, –keep-dup all, -q 0.1, and -g hs (for human genome hg38) or -g mm (for mouse genome mm10). For histone modifications, peak calling was performed using the same settings but with the –broad option to capture the broader regions associated with histone marks. MEME [28] ChIP was used for de novo motif discovery within the called peak sequences of CTCF IT-scC&T-seq single-cell aggregates or individual single-cell mapped sequences.

IT-scC&T-seq library quality control

Duplication rate was estimated from the metric files generated by Picard MarkDuplicates during deduplication. SAMtools idxstats and flagstats were used to compute the number of unique fragments and the mapping rate. The number of reads overlapping with peaks, defined by the MACS2 narrowPeak or broadPeak file, was determined using BEDtools intersect -u, and FRiP values were calculated by dividing the number of reads in peaks by the total number of reads counted by SAMtools. CollectInsertSizeMetrics of Picard Tools 3.1.0 were used to calculate the fragment size of single-cell aggregates’ libraries.

Comparison and correlation with bulk datasets and data visualization

K562 bulk CUT&Tag datasets (GSE124557) were downloaded from Gene Expression Omnibus (GEO) and processed according to the original publication [12, 63], and the ENCODE (https://www.encodeproject.org/) [64, 65] K562 ChIP-seq datasets (Experiment CTCF: ENCSR000DWE, ENCSR000AKO; H3K27me3: ENCSR000AKQ, ENCSR000EWB; H3K4me3: ENCSR000AKU, ENCSR668LDD; IgG: ENCSR000EHI; Input: ENCSR000AKY) were retrieved in BAM format. H1 LMNB1 DamID and DamOnly data were downloaded from the 4DN data portal (accession: 4DNESXKBPZKQ). Raw data LMNB1 ChIP-seq data for H9 hESCs were retrieved from GEO accession GSE155244 [66] (GSM5669226 and GSM5669227 for LMNB1 ChIP-seq replicates 1 and 2, and GSM5669228 and GSM5669231 for corresponding input controls) in FASTQ format and processed according to the methods described in the original publication [35]. Deeptools bamCoverage was used to compute the normalized coverage of single-cell aggregates, and the signal tracks were displayed in IGV v2.14.1. For correlation analysis, multiBigwigSummary with the –outRawCounts parameter was used to calculate raw metrics for determining the Pearson correlation coefficient (r) across replicates of the single-cell libraries, sampled single-cell profiles, and with bulk and ENCODE datasets. ENCODE peaks were downloaded and incorporated into multiBigwigSummary BED-file using the –BED parameter to visualize signals around ENCODE peak regions in heatmaps.

Dimensionality reduction and clustering for LMNB1 and MHT datasets

For each replicate, BAM files were first converted to fragment files using Sinto (https://timoast.github.io/sinto). These fragment files were then used as input for downstream analysis with the ArchR package [39]. For histone mark datasets (H3K4me3, H3K27me3, and H3K36me3), iterative Latent Semantic Indexing (LSI) was applied to the binned count matrices at 2 kb, 20 kb, and 50 kb resolutions, respectively. Batch correction across biological replicates was performed using the Harmony algorithm [67] with default parameters (for MHT datasets only). UMAP was used for dimensionality reduction and visualization, with single cells colored by cell-barcode encoded identity and replicate. Clustering was performed using Seurat’s FindClusters via ArchR’s addClusters function at a resolution of 0.1. Marker genes were identified using the getMarkerFeatures function, and GeneScoreMatrix was calculated in ArchR with default parameters. To annotate clusters, gene activity scores were imputed using the MAGIC algorithm implemented in addImputedWeights. Lineage-specific markers were visualized and used for manual annotation based on H3K4me3 and H3K36me3 promoter enrichment and absence or depletion of H3K27me3 signal around the same loci.

Differential peak analysis

For histone modifications and transcription factors, pseudobulk replicates were generated using addGroupCoverages. Peaks were called with MACS2 using ArchR’s addReproduciblePeakSet function per annotated cell population, and a unified peak set was appended to the Arrow file via addPeakMatrix. To identify differential chromatin regions between H1 and H1-derived MSCs, the getMarkerFeatures function was applied on the PeakMatrix with the Wilcoxon test, controlling for TSS enrichment and fragment count bias (TSSEnrichment, log10(nFrags)). For H3K4me3 and H3K36me3, regions with log2FC ≥ 1 and FDR ≤ 0.1 were retained. For H3K27me3, repressive regions were defined as log2FC ≤ − 1 and FDR ≤ 0.1. Differential peaks were annotated using ChIPseeker to identify genes located within ± 2.5 kb of the nearest peak, and pathway enrichment was performed using the clusterProfiler package.

LAD-specific domain calling and analysis

For all LMNB1 datasets, including ChIP-seq, DamID, bulk and single-cell CUT&Tag, in addition to peak calling via MACS2, broad LMNB1-enriched domains were identified using EPIC2 [68]. The following parameters were used: –keep-duplicates, –bin 8000, -g 6, and genome index hg38. For IT-scC&T-seq pseudobulk profiles, the resulting LAD domains were used as a custom peak set for ArchR analysis, enabling quantification and differential comparison across cell types using the same “getMarkerFeatures” framework. Differential LAD regions were defined at FDR < 0.05, log2FC > 0.5 and subjected to functional annotation with rGREAT [69].

Single-cell RNA-seq processing and integration with IT-scC&T-seq data

Raw scRNA-seq data from adult mouse mammary gland were processed using Cell Ranger (v8.0.1) with default settings, aligned to the mm10 reference genome (mm10-2020-A). Filtered feature-barcode matrices were imported into Seurat (v5.2.1). Cells with fewer than 500 genes, more than 8000 genes, or > 5% mitochondrial transcripts were excluded. Data were normalized using LogNormalize, followed by PCA and UMAP for dimensionality reduction. Clustering was performed using FindClusters with a resolution of 0.02. Marker genes were identified using FindAllMarkers, and six major cell types were annotated based on canonical markers.

Single-cell chromatin profiles for H3K4me3 and H3K27ac were processed using Signac (v1.14.9002) and Seurat (v5.2.1). Fragment files were loaded via CreateFragmentObject, and broad peaks were called using MACS2, then filtered for downstream analysis. Fragment count matrices were generated using FeatureMatrix over peak regions, followed by construction of ChromatinAssay objects and Seurat objects. Annotations were obtained from the EnsDb.Mmusculus.v79 database and assigned to each object. Cells were retained if they had more than 50 unique fragments and a FRiP score above 0.3, calculated based on total fragment counts. Gene activity matrices were computed using GeneActivity, summarizing chromatin accessibility signals over gene bodies extended ± 10 kb from the transcription start sites (TSSs), based on annotations from the mm10 genome (EnsDb.Mmusculus.v79).

To integrate scRNA-seq with H3K4me3 and H3K27ac IT-scC&T data, we first constructed gene activity matrices from IT-scC&T fragment and peak files using Signac [70] (v1.14.9002) and Seurat (v5.2.1). Dimensionality reduction of IT-scC&T data was performed using PCA, and cross-modality integration was achieved using FindIntegrationAnchors (reduction = “rpca”) and IntegrateEmbeddings.

The integrated dataset was clustered using FindClusters (resolution = 0.1) and visualized with RunUMAP. Cell-type annotation was guided by scRNA-seq-derived marker genes. Module scores for lineage-specific gene sets were computed using AddModuleScore using the corresponding raw gene activity score, and the cell type corresponding to the highest score within each cluster was assigned. Visualization of chromatin features across cell types was performed using FeaturePlot (Seurat), CoveragePlot, and TilePlot (Signac).

Pseudotime inference and trajectory-based analysis

To investigate regulatory coordination between gene expression and chromatin dynamics in the luminal lineage, we first inferred pseudotime from the scRNA-seq modality using Slingshot [71], with luminal progenitor as the root. The inferred trajectory was then projected onto the H3K4me3 and H3K27ac modalities using k-nearest neighbor projection (k = 20) based on the joint UMAP embedding. For each modality, we used tradeSeq [72] to fit negative binomial generalized additive models (GAMs), with pseudotime as a covariate. Genes with significant dynamic patterns (P < 0.05) were identified as candidate developmental regulators. Pseudotemporal patterns of gene expression and chromatin activity were visualized using LOESS-smoothed expression curves (span = 0.4) based on log1p-transformed raw gene activity count.

Authors’ contributions

J.M., W.J. and Z.Z. conceived the project. W.J., J.M. and L.R. designed the experiments. W.J. did the experiments with help from F.L., H.S., Z.M., ZX.Z. and X.L. L.R. did the sorting part. J.M. performed bioinformatic analysis with help from Z.G. G.J. helped with data discussion. W.J., J.M., and Z.Z. interpreted the data. W.J. and J.M. prepared the manuscript with comments and inputs from all authors. Z.Z. applied the fundings for this project. All authors read and approved the final manuscript.

Funding

This work was supported by grants of Theme-based Research Scheme (T13-602/21N), Guangdong High-level Hospital Construction Project (KJ012019517), Science, Technology and Innovation Commission of Shenzhen Municipality (JCYJ20210324 114408024), and Guangdong Basic & Applied Basic Research Foundation (2021B1515130004).

Data availability

The IT-scC&T-seq data generated in this study have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession number GSE299567 [73], accessible at: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE299567.

The source code for the IT-scC&T-seq analysis pipeline is available on GitHub at https://github.com/jiingc/IT_scCT under the MIT License [74]. The version used in this study has been archived on Zenodo (DOI: https://doi.org/10.5281/zenodo.15636718 [75].

Previously published datasets used in this study include K562 bulk CUT&Tag data (GSE124557) [12, 63]; K562 ChIP-seq datasets: CTCF (ENCSR000DWE, ENCSR000AKO), H3K27me3 (ENCSR000AKQ, ENCSR000EWB), H3K4me3 (ENCSR000AKU, ENCSR668LDD), IgG control (ENCSR000EHI), and input (ENCSR000AKY) from ENCODE project (https://www.encodeproject.org/) [64, 65]; H1 LMNB1 DamID and Dam-only data from the 4D Nucleome data portal (https://data.4dnucleome.org/) with accession 4DNESXKBPZKQ; and H9 LMNB1 ChIP-seq data retrieved from GEO under GSE155244 [66], including LMNB1 ChIP (GSM5669226, GSM5669227) and input controls (GSM5669228, GSM5669231) [35].

No custom in-house software was used in this study. All reagents and materials are available from the corresponding authors upon reasonable request.

Declarations

Ethics approval and consent to participate

All experimental procedures involving animals were conducted in accordance with institutional ethical guidelines and regulatory standards under approval from the Committee on the Use of Live Animals for Teaching and Research (CULATR) at the University of Hong Kong (Animal License No. [25–144] in DH/HT&A/8/2/3 Pt.80).

Consent for publication

Not applicable.

Competing interests

W.J., J.M., Z.Z. and L.R. declare that a patent application related to the IT-scC&T-seq technology is in preparation. The intended application will cover the indexed tagmentation and barcoding strategy described in this study. This does not restrict academic or non-commercial use of the method. All experimental protocols and relevant details are fully described in the manuscript and supplementary materials, allowing full reproducibility. Researchers are free to use and implement the method for non-commercial purposes without requiring a license. Any future commercial use may require appropriate licensing arrangements. The remaining authors declare no competing interests.