Principles of nucleosome organization revealed by single-cell MNase-seq

Summary

Nucleosome positioning is critical to chromatin accessibility, and is associated with gene expression programs in cells^1–3. Previous nucleosome mapping methods assemble profiles from cell populations and reveal a cell-averaged pattern: nucleosomes are positioned and form a phased array surrounding the transcription start sites (TSSs ) of active genes^3–6 and DNase I hypersensitive sites (DHSs)⁷. However, cells exhibit remarkable expression heterogeneity in response to active signaling even in a homogenous population of cells^8,9, which may be related to the heterogeneity in chromatin accessibility^10–12. Here, we report a technique, termed single-cell MNase-seq (scMNase-seq), to measure genome-wide nucleosome positioning and chromatin accessibility simultaneously in single cells. Application of scMNase-seq to NIH3T3, mouse primary naïve CD4 T and embryonic stem cells (mESC) reveals two novel principles of nucleosome organization: (1) nucleosomes surrounding TSSs of silent genes or in heterochromatin regions show large positioning variation across different cells but are highly uniformly spaced along the nucleosome array and, (2) In contrast, nucleosomes surrounding TSSs of active genes and DHSs show small positioning variation across different cells but show relatively low spacing uniformness along the nucleosome array. We found a bimodal distribution of nucleosome spacing at DHSs, which corresponds to inaccessible and accessible states and is associated with nucleosome variation and accessibility variation across cells. Nucleosome variation within single cells is smaller than that across cells and variation within the same cell type is smaller than that across cell types. A large fraction of naïve CD4 T cells and mESCs show depleted nucleosome occupancy at the de novo enhancers detected in their respectively differentiated lineages, revealing the existence of cells primed for differentiation to specific lineages in undifferentiated cell populations.

To understand the principles underlining chromatin heterogeneity related to nucleosome positioning and chromatin accessibility, we developed a technique, called single-cell MNase-seq (scMNase-seq), to simultaneously measure the nucleosome positioning and chromatin accessibility in single cells. We applied scMNase-seq to 48 NIH3T3 single cells, 198 mouse embryonic stem cells (mESCs), and 278 mouse naïve CD4 T cells, obtaining ~3, 0.9, and 0.7 million unique fragments, respectively, on average for each cell type (Fig. 1a, Supplementary Table 1). Sequence reads from sorted human or mouse cells from a mixed population mapped exclusively to the perspective genome, suggesting that there was no DNA contamination across cells (Extended Data Fig. 1a). Pooled single cell reads revealed a size distribution consistent with that obtained by bulk cell MNase-seq (Extended Data Fig. 1b). We considered fragments with a length between 140-180 bp as canonical nucleosomes and fragments with a length ≤80 bp as subnucleosome-sized particles (Extended Data Fig. 1b,c). NIH3T3 libraries have the largest number of non-redundant reads (Extended Data Fig. 1d) and the highest genomic coverage (5%-30%) of nucleosomes compared to CD4 T cells and mESCs (Fig. 1b), probably due to the polyploidy of the cells (Extended Data Fig. 1e). Nevertheless, all three cell types have similar nucleosome density across different genomic regions, suggesting a relatively even representation of the genome (Extended Data Fig. 1f). The nucleosome positioning and the enrichment of subnucleosome-sized particles surrounding DHSs, TSSs of active genes and CTCF binding sites were consistent between pooled scMNase-seq and bulk cell MNase-seq data (Fig. 1c, Extended Data Fig. 2a–h). The density of subnucleosome-sized particles from pooled single cells is correlated with tag density of DHSs and is correlated with gene expression at TSSs, suggesting that subnucleosome-sized particles are predictive of chromatin accessibility (Extended Data Fig. 2i,j). Moreover, the percentage of DHSs detected by scMNase-seq was higher than that by scATAC-seq with the same sequencing redundancy, due to higher complexity and non-redundant read number of scMNase libraries, although the percentage of recovered DHSs per non-redundant read for scMNase-seq subnucleosome-sized particles was relatively lower than scATAC-seq fragments (Extended Data Fig. 2k–n). Nucleosome positions from single cells, aggregated nucleosome density from pooled single cells and tag density from bulk MNase-seq at representative cell-type-specific genes were shown for all three cell types (Fig. 1d). Notably, similarity of aggregated nucleosome profiles between pooled single cells and bulk cells is correlated with nucleosome positioning stringency and nucleosome coverage, and is higher for active promoters than silent promoters (Fig. 1d; Extended Data Fig. 2o). These results demonstrate that scMNase-seq can measure both nucleosome positioning and chromatin accessibility simultaneously in single cells.

Figure 1. scMNase-seq measures simultaneously the position of both nucleosome and subnucleosome-sized particles in single cells.

a, Schema of scMNase-seq. b, Plot of non-redundant nucleosome read number (x-axis) and genomic coverage of nucleosomes (y-axis) for NIH3T3, CD4 T and mES single cells. c, Average density profiles of nucleosomes (red) and subnucleosome-sized particles (blue) relative to TSS of active genes (left) and CTCF binding sites (right) for pooled CD4 T scMNase-seq data. d, Genome browser view of single-cell nucleosome positions for NIH3T3, CD4 T cell, and mESCs at TSSs of three representative cell type specific gene loci. Single cell libraries that have at least one nucleosome within any of three genomic regions were shown. Tracks for tag density of corresponding bulk cell MNase-seq data (one representative from two repeated experiments was shown) and pooled scMNase-seq data (all single cell libraries were included) were also shown. The nucleosome maps at expressed genes for each cell type were highlighted with pink rectangle. The expression levels of genes were shown in the heat map above the tracks.

While nucleosome positioning¹³ is well-studied^5,14–16 based on a large number of pooled cells, genome-wide nucleosome spacing patterns are poorly understood because current knowledge about nucleosome spacing is limited to the positioned nucleosomes^7,17,18. We profiled the distribution of nucleosome-to-nucleosome distance for different single cells and used relative peak height to measure the uniformness of nucleosome spacing for both positioned and non-positioned nucleosomes (Extended Data Fig. 3a–c, Supplementary Methods). The analysis revealed a high degree in spacing uniformness in single cells regardless of positioning stringency, while decreased spacing uniformness was observed as positioning stringency decreased based on the pooled single cells or bulk cell MNase-seq data¹⁹ (Extended Data Fig. 3d). The bulk cell MNase-seq data failed to reveal the actual spacing pattern due to the mixture of non-positioned nucleosomes from a population of different cells.

Surprisingly, the degree of spacing uniformness in the promoter regions of silent genes is higher than that of active genes (Fig. 2a,b, Extended Data Fig. 4a,b), and the uniformness in non-DHS regions is higher than that in DHS regions (Fig. 2c,d, Extended Data Fig. 4a–c). Notably, the higher spacing uniformness in non-DHS regions than DHS regions was also observed in single haploid mES cells and haploid chromosome X in single mES cells (Extended Data Fig. 4d–g) and was independent of MNase concentration (Extended Data Fig. 4h–m). Further, nucleosome spacing in active chromatin regions associated with H3K4me1, H3K4me3, H3K27ac, H3K9ac, and H2AZ shows a lower degree of uniformness than transcribed regions marked by H3K36me3, and heterochromatic regions marked by H3K27me3 or not marked by any of the studied histone modifications (Extended Data Fig. 4n–o).

Figure 2. Profiling nucleosome positioning in single cells reveals distinct nucleosome organization principles at active and silent chromatin regions.

a, Density plots of nucleosome-to-nucleosome distance within active gene promoters (upper) and silent gene promoters (bottom) for bulk-cell MNase-seq, pooled 48 NIH3T3 single cells, one representative single cell and 48 single-cell scMNase-seq datasets. b, The relative peak heights based on the data from panel (a) reveals higher degree of spacing uniformness within silent gene promoters than active gene promoters. c. Density plots of nucleosome-to-nucleosome distance within DHS regions (upper) and non-DHS regions (bottom) for bulk-cell MNase-seq (blue) and 48 single-cell scMNase-seq (red) datasets. d, The relative peak heights based on the data from panel (c) reveals higher degree of spacing uniformness within non-DHS regions than DHS regions. e, Cumulative density of nucleosome positioning variance of active genes and silent genes within a cell (upper) and across single cells (bottom) at−1 (left) and +1 (right) nucleosomes relative to TSS. Upper left panel, n = 7,574 nucleosome-pairs for active genes; n = 13,107 for silent genes; bottom left, n = 164,512 for active genes, n = 304,847 for silent genes; upper right, n = 11,388 for active genes, n = 17,631 for silent genes; bottom right, n = 237,006 for active genes, n = 416,328 for silent genes. P values were calculated using one-sided Mann-Whitney rank test. f, Cartoon illustrating nucleosome organization patterns in silent (left) and active (right) chromatin states.

Next, we measured variation of nucleosome positioning not only across cells but also within single cells (across different alleles) by calculating the mean value of distances between two overlapping nucleosomes within genomic regions related to a particular feature, e.g. active promoters (Extended Data Fig. 5a). As expected, nucleosome position variation around TSSs of active genes, where nucleosomes are phased relative to TSS, is smaller than that of silent genes (Fig. 2e). In addition, nucleosome positions show smaller variation at the center of DHSs and the center of chromatin regions enriched in active histone modifications (Extended Data Fig. 5b–g).

The results above reveal different rules of nucleosome organization for different chromatin regions: (1) in silent chromatin states, such as in repressed promoters and heterochromatic regions, nucleosomes are highly uniformly spaced but not positioned relative to the underlying genomic DNA across different arrays and, (2) in active chromatin states, such as transcribed promoters and DHS regions, nucleosomes are positioned but show less uniformness of spacing (Fig. 2f). This model was further supported by the observation that nucleosomes in promoter regions of silent genes, non-DHS regions and heterochromatic regions show higher synchronized shift scores than those in promoters of active genes, DHS regions, and regions marked by active histone modifications (Extended Data Fig. 6a,b). Further, synchronized shift score is dependent on nucleosome spacing, with the highest scores in the spacing range of 180-185 bp, which is dominant throughout genome in all single cells (Extended Data Fig. 6c,d). The nucleosome spacing might indicate a stable structure for packaging nucleosomes²⁰ in silent chromatin states, probably collectively determined by chromatin assembly factors²¹, linker histones^22,23, and the environment surrounding chromatin fibers. In active states, ATP-dependent chromatin remodeling activities^15,24 may reposition nucleosomes^1,25 and consequently change the local nucleosome spacing to facilitate the chromatin accessibility and the gene transcription. Notably, the average nucleosome spacing surrounding the DHSs are shorter than those in non-DHS regions (Extended Data Fig. 6e,f), which may result from repositioning of the nucleosomes to allow accessibility of the DHS regions.

Although nucleosomes are positioned surrounding DHSs to ensure chromatin accessibility⁷, extensive heterogeneity of chromatin accessibility across different single cells^10,12 implies heterogeneity of nucleosome positioning at the same DHS. Profiling nucleosome-to-nucleosome distances over DHSs reveals two distinct peak patterns: one has a summit at 190 bp, and the other has a summit at ~300 bp, presumably corresponding to two different chromatin states (closed or open) (Fig. 3a,b). More than 80% of the DHSs have both spacing types at the same DHS in different single cells (Fig. 3c), and higher DNase I tag density at DHSs measured in bulk cells²⁶ is associated with more wide-spacing DHSs in single cells (Extended Data Fig. 7a). Further, the DHSs with higher fraction of wide space, which is not related to MNase digestion, is associated with higher accessibility measured by bulk cell DNase-seq or by scMNase-seq subnucleosome-sized particles (Extended Data Fig. 7b–d), and with lower variation in DHS accessibility and nucleosome positioning across different single cells (Fig. 3d,e). These results indicate that one DHS may have two types of nucleosome organization (wide or narrow spacing) across different single cells; the degree of accessibility of a DHS and the variation of DHS accessibility and nucleosome positioning across cells are directly linked to the ratio between the two states of nucleosome organization in different single cells.

Figure 3. The bimodal distribution of nucleosome spacing across DHSs is associated with the cell-to-cell variation of nucleosome positioning and chromatin accessibility.

a, Schema of nucleosome spacing across a DHS and two chromatin states inferred by nucleosome spacing. b, Density plot of nucleosome spacing across a DHS within single cells reveals two peaks corresponding to narrow spacing (blue) and wide spacing (red). c, Heat map shows DHS frequency as a function of number of cells with narrow spacing and number of cells with wide spacing. Percentage of DHSs where there are both types of spacing across a DHS in different single cells is shown. d-e, Boxplots showing the cell-to-cell variation of nucleosome position (d) and chromatin accessibility (e) for 5 groups of DHSs defined by fraction of wide space. Data represents 612, 2,088, 3,858, 2,500, and 1,586 DHSs (left to right). f, Scatter plot showing nucleosome variance (y-axis) and DHS variation (x-axis) across cells for 106 bins of DHSs based on DHS variation. Each dot represents the average from 500 DHSs for each bin. Pearson’s correlation was calculated. g, Boxplot showing nucleosome variation at +1 nucleosome relative to TSS for two groups of genes sorted by expression variation (Low, bottom 25% (n = 1,171 genes); high, top 25% (n = 1,174 genes)). d, e, and g, P values were calculated by one-sided Mann-Whitney rank test. Boxplot definition: center line, median; boxes, first and third quartiles; whiskers, 1.5× the interquartile range; notch, 95% confidence interval of the median.

Further, variation in nucleosome positioning around DHSs is positively correlated with variation in accessibility across different single cells (Fig. 3f). The fraction of single cells with nucleosomes positioned around DHSs is correlated with the number of cells detected as DHS (Extended Data Fig. 7e). The variation in nucleosome positioning around TSSs in different single cells is also correlated with variation in gene expression. The TSSs with +1 nucleosomes showing higher variation in nucleosome positioning also show higher variation in expression across different single cells (Fig. 3g). Genes whose expression was detected in a higher fraction of single cells exhibit positioned +1 nucleosomes in a higher fraction of single cells than the genes with a lower fraction of expression (Extended Data Fig. 7f). The top 1000 active genes with smallest nucleosome variance around TSS across cells are enriched in common biological processes such as translation and protein transport (Extended Data Fig. 7g), consistent with the notion that house-keeping genes display less variation in nucleosome positioning. Further, nucleosome position variation around DHSs or at the +1 nucleosome of TSSs of active genes within a cell is smaller than that across different single cells (Extended Data Fig. 7h,i). The variance of nucleosome positioning within the cell type is smaller than that across different cell types (Extended Data Fig. 7j). Clustering based on the nucleosome positioning similarity at all the DHSs across all the single cells from three cell types separated single cells into three clusters corresponding to the perspective cell types and the clustering is independent on experiment time and fragment size ratio (Extended Data Fig. 7k).

DNA sequence plays an important role in nucleosome positioning^4,14,16. Consistent with previous report¹⁴, we observed high CC/GG/GC frequency at nucleosome-occupied sequence and high AA/TT/AT/TA frequency at flanking regions in single cells, as well as a periodical pattern supporting the rotational positioning of nucleosomes^4,16 (Extended Data Fig. 8a). Smaller nucleosomes variation is associated with lower CC/GG/GC frequency and higher AA/TT/AT/TA frequency in flanking region (Extended Data Fig. 8b–e). Next, we explored the relationship between DNA sequence variance and nucleosome positioning variance. Analysis shows that sequences occupied by nucleosomes have higher fraction of alternative base than those occupied by subnucleosome-sized particles, as well as DNase-seq tags and CTCF ChIP-seq tags (Extended Data Fig. 8f,g), which support the notion that sequence variants influence TF binding²⁷ and nucleosome positioning²⁸. Moreover, we found that single base variance within nucleosome regions is positively correlated with nucleosome variance across cells (Extended Data Fig. 9h). Furthermore, the variation at TF motifs is positively correlated with nucleosome variance at DHSs and is also positively correlated with gene expression variation across different single cells (Extended Data Fig 9i,j).

Enhancers display remarkable cell type specificity. Consistent with previous observation that active enhancers are associated with a nucleosome loss³, the naïve CD4 T cell-specific enhancers displayed decreased nucleosome occupancy in naïve CD4 T cells as revealed by the pooled scMNase-seq data from naïve CD4 T cells; by contrast, the Th1- and Th2-specific enhancers showed only a very minor overall nucleosome loss in naïve CD4 T cells (Extended Data Fig. 9a,b). However, examination of the nucleosome patterns at the Th1- and Th2-specific enhancers across different single cells revealed that 19% and 29% of naïve CD4 T cells showed decreased nucleosome occupancy, which is independent of fragment size ratio, at the de novo enhancers of Th1 and Th2 cells, respectively, while much smaller fractions of mESCs and NIH3T3 cells showed decreased nucleosome occupancy at these enhancers (Fig. 4a, Extended Data Fig. 9c–e). Furthermore, subgroups of T cells showing decreased nucleosome occupancy at the Th1 and Th2 enhancers, respectively, do not have much overlap (Extended Data Fig. 9f), suggesting they are specifically primed for the corresponding lineages. The Th1-specific enhancers with the most nucleosome loss in naïve CD4 T cells are linked to genes encoding Th1 cytokine (Ifng) and key regulators (Tbx21, Stat1 and Stat4) (Extended Data Fig. 9g,h); the Th2-specific enhancers with the most nucleosome loss are linked to genes encoding key regulators for Th2 differentiation (Il4 and Stat6) (Extended Data Fig. 9i,j). Motif analysis revealed that the nucleosome loss at Th1 enhancers is specifically associated with motifs for RELA, which promotes Th1 differentiation; the nucleosome loss at Th2 enhancers is specifically associated with motifs for GATA3 and STAT6, which promote Th2 differentiation (Extended Data Fig. 9k). GO analysis revealed that the higher ranked nucleosome losses at both Th1 and Th2 enhancers are associated with functions in T cell differentiation, immune system process and cytokine production (Extended Data Fig. 9l,m). These results suggest that a large fraction of naïve CD4 T cells have already experienced differentiating signaling events during the developmental history of these cells, which have primed the de novo enhancers of Th1 or Th2 cells by decreased nucleosome occupancy in the undifferentiated naïve CD4 T cells.

Figure 4. A subgroup of undifferentiated cells shows a nucleosome signature primed for differentiation.

a,b A large fraction of naïve CD4 T cells shows decreased nucleosome occupancy at the de novo enhancers that are formed either in Th1 (a, top) or Th2 cells (a, bottom), while only a small fraction of mESCs and NIH3T3 cells shows nucleosome depletion at the same enhancers. In contrast, a large fraction of mESCs shows depleted nucleosomes at the de novo enhancers that are formed in EBs, while only a small fraction of naïve CD4 T cells and NIH3T3 cells shows nucleosome depletion at the same enhancers (b). The fractions of primed cells are shown in red. Data represents 237 naïve T single cells, 143 mES single cells, and 48 3T3 single cells.

Similarly, ESCs displayed a substantial nucleosome loss at the ESC-specific enhancers but only a minor loss at EB-specific enhancers, which are formed de novo after differentiation from mESCs (Extended Data Fig. 10a,b). Analysis of single cells revealed that 40% of ESCs showed decreased nucleosome occupancy at the de novo EB-specific enhancers, while only 1% and 2% of naïve CD4 T cells and NIH3T3 cells, respectively, showed decreased nucleosome occupancy at these enhancers (Fig. 4b, Extended Data Fig. 10c,d). The EB enhancers with the most nucleosome loss are linked to genes including mesoderm markers (Brachyury, Wnt3) and endoderm markers (Gata4, Gata6) (Extended Data Fig. 10e,f), and are associated with stem cell differentiation and development of various lineages such as myeloid cell, neural tube, and placenta (Extended Data Fig. 10g). These results revealed the heterogeneity of cultured ESCs and suggests that some of these cells are already primed for differentiation by reorganizing their nucleosome structure at enhancers formed in the differentiating EBs.

We provided a powerful method, scMNase-seq to simultaneously measure both chromatin accessibility and nucleosome positioning in single cells, which may be paired with existing approaches such as single-cell RNA-seq⁹, single-cell DNase-seq¹², single-cell ChIP-seq²⁹ for systems analysis to provide more insights into the molecular underpinning of cellular heterogeneity. Application of scMNase-seq to three types of single cells reveals principles in nucleosome organization in different chromatin regions and heterogeneity of nucleosome positioning/spacing at DHSs. Our data suggest that the cellular heterogeneity of undifferentiated cells is related to heterogeneous nucleosome organization at critical regulatory regions, which reflects the differentiation potential of these cells.

Methods

Detailed Methods are described in Supplementary Methods which is available in the online version of this paper.

Code availability

Custom codes for nucleosome spacing uniformness quantification and nucleosome occupancy score calculation are available at https://github.com/binbinlai2012/scMNase.

Data Availability

The scMNase-seq data sets have been deposited in the Gene Expression Omnibus database with accession number GSE96688.

Extended Data

Extended Data Figure 1. Characterizing scMNase-seq datasets.

a, Mapping rates of reads from 100 human cells (left two experiments) or 100 mouse cells (right three experiments) against human genome (blue) and mouse genome (orange) were shown. The cells were sorted from pre-mixed and MNase-digested human and mouse cells. These data show that there is little contamination of DNA of one cell from another cell. b, Fragment length density of pooled scMNase-seq for NIH3T3 cells, pooled scMNase-seq and bulk-cell MNase-seq for T cells and mESCs. c, Box plots of fragment ratio (subnucleosome-sized particle / nucleosome) for NIH3T3, naïve CD4 T cell and mESC scMNase-seq libraries. Single cell libraries were grouped by biologically independent experiments. Refer to Supplementary Table 1 for the library number for each group. The center line, median; boxes, first and third quartiles; whiskers, 1.5× the interquartile range. d, Plot of non-redundant (NR) read number (x-axis) and sequencing redundancy (y-axis) for NIH3T3, CD4 T and mES single cells. e, Plot of NR nucleosome reads (x-axis) and percentage of nucleosomes with overlapping piles ≥3 (y-axis). The plot suggests the polyploid of NIH3T3 cells. f, Nucleosome density at different genomic regions for NIH3T3, CD4 T cell, and mESC scMNase-seq libraries reveal that the nucleosomes in different genomic regions were similarly detected and represented by scMNase-seq.

Extended Data Figure 2. Characterizing pooled scMNase-seq data and subnucleosome-sized particles.

a, Average density profiles of nucleosomes (red) and subnucleosome-sized particles (blue) relative to TSS of active genes (left) and CTCF binding sites (right) for bulk cell naïve CD4 T bulk cell MNase-seq data. b, Average density profiles of nucleosomes (red) and subnucleosome-sized particles (blue) relative to TSS of active genes (left) and CTCF binding sites (right) for pooled mESC scMNase-seq data (top) and bulk cell mESC MNase-seq data (bottom). c, Smoothed scatter plot for the fraction of nucleosome occupied at 8,929 DHS center selected from top 10,000 DHSs (see Methods for criteria) for pooled scMNase-seq (x-axis) versus bulk cell MNase-seq (y-axis) for T cells (left). Pearson correlation coefficient was indicated. As a positive control, the scatter plot for two bulk MNase-seq replicates were also shown (right). d, Pearson correlation coefficient for fraction of nucleosomes occupied at DHS center between pooled sub-sampled CD4 T cell scMNase-seq libraries and bulk cell MNase-seq as a function of sub-sampled cell number (left). Percentage of top 10,000 DHSs represented in the comparison, i.e. the sample size in upper panel, as a function of sub-sampled cell number are also shown (right). e, Smoothed scatter plot for subnucleosome-sized particle density at 83,229 DHSs (h) for pooled scMNase-seq (x-axis) versus bulk cell MNase-seq (y-axis) for T cells (left). Pearson correlation coefficient was indicated. As a positive control, the scatter plot for two bulk MNase-seq replicates were also shown (right). f, Pearson correlation coefficient for subnucleosome-sized particle density between pooled sub-sampled T cell scMNase-seq libraries and bulk cell MNase-seq at 83,229 DHSs as a function of sub-sampled cell number. g-h, Smoothed scatter plot for the fraction of nucleosomes occupied at 8,449 DHS center selected from top 10,000 DHSs (see Methods for criteria) (g) and subnucleosome-sized particle density at 94,250 DHSs (h) for pooled scMNase-seq (x-axis) versus bulk cell MNase-seq (y-axis) for mESCs. Pearson correlation coefficient was indicated. As a positive control, the scatter plot for two bulk MNase-seq replicates were also shown. i-j, Average density profiles of subnucleosome-sized particles around TSSs for gene subgroups with different expression levels (i) and around DHSs for DHS subgroups with different DNase I tag densities (j). k, Table showing the mapping statistics for 198 mESC scMNase-seq libraries and 96 mESC scATAC-seq libraries from publication (Buenrostro et al.). l-m, Scatter plots of number of non-redundant (NR) reads (l, y-axis) and percentage of recovered DHSs (m, y-axis) versus sequencing redundancy (x-axis) for scMNas-seq subnucleosome particles (red, n = 198 single cell libraries) and scATAC-seq reads (grey, n = 96 single cell libraries). Box plots (right) showing the values from left scatter plots for cells with redundancy ranging from 50% to 70% (blue rectangle in the left panel; red, n = 49; grey, n = 58) for two methods. n, Scatter plot showing percentage of recovered DHSs (y-axis) versus number of NR reads for scMNas-seq subnucleosome particles (red, n = 198 single cell libraries) and scATAC-seq reads (grey, n = 87 single cell libraries). o, Aggregated nucleosome profile similarity score at DHSs for different groups of DHSs (left) and two promoter groups (right) for comparison between pooled scMNase-seq and bulk cell MNase-seq (upper) and between two bulk cell MNase-seq replicates (lower). The DHS groups are classified by three positioning stringency levels (low: positioning score (PS) <0.45, moderate: 0.45 < PS <0.65, high: PS > 0.65) and three nucleosome coverage levels (high: ≥15; moderate: 10-15; low: 5-9). The DHS numbers for each group are: low PS and high coverage, n = 803; low PS and moderate coverage, n = 531; low PS and low coverage, n = 450; moderate PS and high coverage, n = 701; moderate PS and moderate coverage, n = 592; moderate PS and low coverage, n = 588; high PS and high coverage, n = 162; high PS and moderate coverage, n = 230; high PS and low coverage, n = 395. The number of promoters for each group: active, n = 6,777; silent, n = 418. Boxplot definition in panels i,m,o: center line, median; boxes, first and third quartiles; whiskers, 1.5× the interquartile range; notch, 95% confidence interval of the median.

Extended Data Figure 3. Measuring nucleosome spacing uniformness in single cells.

a, Cartoon illustrates that nucleosome spacing uniformness can be measured by nucleosome-to-nucleosome distance density. Uniformly spaced nucleosomes in a single array result in sharp and high peaks while non-uniformly spaced nucleosomes result in flat peaks or no peaks. Nucleosomes from mixed arrays also result in flat peaks even they are uniformly spaced. b, The nucleosome phasing and relative peak height gradually decreased as the number of mixed cells increases, which indicates cellular heterogeneity of nucleosome organization across different cells. c, Nucleosome space phasing and relative peak height do not change when reducing the library size (number of sequence reads) to 1/2, 1/3, and 1/4. d, Density plots of nucleosome-to-nucleosome distance (upper) and relative peak height in density plot (bottom) for nucleosomes with different positioning stringency for bulk-cell MNase-seq, pooled 48 single cells, one representative single cell and 48 single-cell scMNase-seq datasets.

Extended Data Figure 4. Nucleosome spacing uniformness is higher in silent heterochromatin region than in active chromatin region.

a-b, Density plots of nucleosome-to-nucleosome distance (top) and relative peak height (bottom) for nucleosomes at active or silent promoters and DHS or non-DHS regions for T cells (a) and mESCs (b). c, Relative peak height of nucleosome-to-nucleosome distance density plots for nucleosomes in DHS (red) and non-DHS regions (blue) for low coverage cells (top) and high coverage cells (bottom). d, Density plots of nucleosome-nucleosome distance (top) and relative peak height (bottom) for diploid (black) and haploid (red) mESCs. e, Density plots of nucleosome-nucleosome distance (top) and relative peak height (bottom) at DHS and non-DHS regions for haploid mESCs. f, Mapped nucleosome count normalized by chromosome length at chromosome 1, X, and Y for mESCs and CD4 T cells suggest that mESCs are derived from male mouse. g, Density plots of nucleosome-nucleosome distance at DHS and non-DHS regions at chromosome X for mESCs. h, Violin plots of library size (total NR reads) for NIH3T3 scMNase-seq libraries treated with three MNase concentrations (0.1 unite (0.1U), 0.6 unit (0.6U), and 2.4 unit (2.4U) MNase per million cells). Each condition has 10 single cell libraries. Violin plot: center dot, mean; inner layer, the interquartile range. i-j, Fragment length density of pooled scMNase-seq data with three MNase concentrations. j, Average density profiles of all reads (left) and nucleosome reads with length between 140 and 180 bp (right) around TSSs of active genes (i) and CTCF binding sites (j) for pooled scMNase-seq with three MNase concentrations. l-m, Density plots of nucleosome-nucleosome distance (l) and relative peak height (m) at DHS (red) and non-DHS (blue) regions for scMNase-seq treated by 0.6U (left) and 2.4U (right) MNase concentrations. n-o, Density plots of nucleosome-to-nucleosome distance (n) and relative peak height (o) for nucleosomes within genomic regions marked by different histone modifications.

Extended Data Figure 5. Nucleosome positioning variation within a cell or across different single cells around the center of DHS or histone modification peaks.

a, Cartoon illustrating the definition of nucleosome variance within a cell or across different single cells. b-c, Heat maps showing the distribution of nucleosome variance at the position relative to DHS center within cells (b) or across different single cells (c). d, Nucleosome variance within a cell (red) and across single cells (blue) becomes smaller when getting closer to DHS center. e, Calculations of mean value of nucleosome variance from two ranges (3-82 bp and 0-82 bp) reveal the same trend of increase when nucleosomes become farther away from DHS center. f-g, Average profiles of nucleosome variance at the position relative to the center of histone modification peaks within cells (e) or across different cells (f).

Extended Data Figure 6. Nucleosomes show synchronized shift in silent gene promoters and heterochromatin regions and show compressed spaceing flanking DHS centers.

a, Cartoon illustrates synchronized shift of adjacent nucleosomes within single nucleosome arrays. b, Bar plot showing synchronized shift score for different genomic regions. Silent promoter: silent gene promoter; active promoter: active gene promoter; not marked: regions not marked by any histone modifications as shown; DHS: ±2000 bp region surrounding DHS center; non-DHS: intervals of DHS regions. c, Synchronized shift score for nucleosome pairs with different nucleosome space distances. d, Density plot of nucleosome-to-nucleosome distance in single cells reveals dominant nucleosome space at ~182 bp. e, Density plot of nucleosome spacing in the regions flanking strong and weak DHSs as well as non-DHSs. f, Distances between each pair of nucleosomes in the chromatin regions flanking strong DHS, weak DHS or non-DHS described in (e).

Extended Data Figure 7. Heterogeneity of nucleosome spacing and positioning around DHS across different single cells.

a, Heat maps showing DHS frequency as a function of number of cells with the narrow spacing (x-axis) and number of cells with the wide spacing (y-axis) for four DHS subgroups with different tag densities. Numbers indicate percentage of DHSs that have more wide space than narrow space. b-c, Boxplots showing the accessibility level from cell population measured by DNase-seq tag density (b) and pooled scMNase-seq subnucleosome-sized particle density (c) for 5 groups of DHSs defined by fraction of wide space. Data represents values on 612, 2,088, 3,858, 2,500, and 1,586 DHSs (left to right). d, Scatter plot of the ratio of wide to narrow space at DHS in a single cell (x-axis) and fragment size ratio of subnucleosome-sized particles to nucleosomes (y-axis) on 48 NIH3T3 sc MNase-seq libraries. Pearson correlation coefficient and P-value were indicated. P-value is the probability that one would have found the current result if the correlation coefficient were in fact zero (null hypothesis), and was calculated using R package. e, Boxplot showing fraction of cells with positioned nucleosomes around a DHS for different groups of DHSs. DHSs were grouped based on the number of cells detected as DHS in scDNase-seq experiment. Number of DHSs for each group was 44,040, 15,622, 11,056, 8,009, 4,063, and 1,180 (from left to right). f, Boxplot showing fraction of cells with a positioned +1 nucleosome for two groups of genes sorted by expression variation (Low, n = 1,171; High, n = 1,174). g, GO analysis of top 1000 active genes with smallest nucleosome variance across cells. Significant GO terms with P-value were reported by David Bioinformatics Resources (v6.7). h, Density plot showing nucleosome variance around DHSs for within a cell (n = 73,274 nucleosome pairs) and across different cells (n = 752,398 nucleosome pairs). i, Cumulative density plot for nucleosome variation at +1 nucleosome relative to TSS of active genes for within cell (red, n = 11,388 nucleosome pairs) and across cells (blue, n = 237,006). j, Boxplot showing nucleosome variance around DHSs across cells for within a cell type (3T3-3T3, n = 1,128 nucleosome pairs; T-T, n = 23,936; ES-ES, n = 5,775) and across different cell types (3T3-T, n = 11,856; 3T3-ES, n = 6,962; T-ES, n = 20,442). k, Heat map reveals clustering of NIH3T3 cells, T cells and mESCs based on cell-to-cell nucleosome dissimilarity score around DHSs. Color bar on the right indicates cell types and color bars on the bottom indicates experiment time and fragment size ratio. P values in panels b, c, e, f, h, and i were calculated using one-sided Mann-Whitney rank test. Boxplot definition: center line, median; boxes, first and third quartiles; whiskers, 1.5× the interquartile range; notch, 95% confidence interval of the median.

Extended Data Figure 8. Cell-to-cell single base variation is associated with nucleosome positioning variation and gene expression variation across different single cells.

a, CC/GG/GC frequency is higher in nucleosome occupied region than in flanking region while AA/TT/AT/TA frequency shows an opposite pattern. b, CC/GG/GC frequency in flanking regions increases as nucleosome variance within a cell (left) or across different single cells (right) increases. c, AA/TT/AT/TA frequency in flanking regions decreases as nucleosome variance within a cell (left) or across different single cells (right) increases. d, Nucleosome variances within a cell and across different single cells are reversely correlated with AA/TT/AT/TA percentage in flanking regions. e, Weblogos for sequence preferences across MNase cleavage sites are shown for subgroups of nucleosomes with different positioning variance across cells. f, An example showing a CTCF motif with the reference base (green) in some cells and alternative base (red) in other cells. scMNase-data show that the reference base is associated with subnucleosome-sized particles while the alternative base is associated with the nucleosome structure. Fragments from DNase-seq and CTCF ChIP-seq datasets within the window are also shown with the bases at SNP location highlighted. Tracks for tag densities of CTCF ChIP-seq, DNase-seq, and nucleosomes and subnucleosome-sized particles from pooled single cells are shown in a zoom-out window. g, The number of CTCF motif matches containing alternative/reference base at the SNP locus occupied by nucleosomes, subnucleosome-sized particles, sequence reads obtained by DNase-seq and by CTCF ChIP-seq. P-value was calculated using one-sided Fisher’s exact test. The ratio between alternative to reference base was also shown (bottom panel). h, SNP frequency is correlated with nucleosome variation across different single cells. Variant frequencies at each position relative to nucleosome midpoint for four nucleosome subgroups with different levels of nucleosome variance across cells are shown. i, SNP frequency within TF motifs at DHSs for 4 DHS subgroups sorted by nucleosome variance around DHS across different single cells (each subgroup has 22,139 DHSs that contains at least one TF motif match). j, SNP frequency within TF motifs at DHSs in promoters for gene subgroups sorted by expression variation across different single cells (each subgroup has 2,136 genes). P value in panels i,j is defined as the probability of observing a larger difference than current result between two groups by random. P value calculation was described in Methods.

Extended Data Figure 9. Characterization of primed enhancers in undifferentiated naïve CD4 T cells.

a, Heat maps show H3K27ac in naïve T cells and p300 in Th1 and Th2 cells around naïve T-specific, Th1-specific and Th2-specific enhancers. b, Profile of nucleosome occupancy from pooled naïve T scMNase-seq around T-specific, Th1-specific, and Th2-specific enhancers. c, Normalized nucleosome occupancy within ±200 bp from the center of de novo Th1 enhancers (left) or de novo Th2 enhancers (right) for subgroups of T cells primed for Th1 cells (green), Th2 cells (blue) or none (black). d-e, Plots of fragment size ratio of subnucleoosme-sized particles to nucleosomes versus nucleosome occupancy score at de novo Th1 (c) and Th2 (d) enhancers for 237 naïve CD4 T cells reveal that nucleosome occupancy score is not correlated with fragment size ratio. Pearson correlation coefficient and P value are indicated. P-value is the probability that one would have found the current result if the correlation coefficient were in fact zero (null hypothesis), and was calculated using R package. f, Subgroups of naïve CD4 T cells primed for Th1 and Th2 do not have much overlap. g, Plots of de novo Th1 enhancers ranked based on nucleosome occupancy difference between pooled primed cells and the non-primed cells (y-axis, Methods). Enhancers associated with key genes for Th1 were labeled by genes along with ranks. h, Nucleosome positions in pooled or single primed (red) and non-primed (blue) cells at de novo Th1-specific enhancers for Ifng gene. i, Plots of de novo Th2 enhancers ranked based on nucleosome occupancy difference between pooled primed cells and the non-primed cells (y-axis, Methods). Enhancers associated with key genes for Th2 were labeled by genes along with ranks. j, Nucleosome positions in pooled or single primed (red) and non-primed (blue) cells at de novo Th2-specific enhancers for Il4 gene. k, Motifs enriched in top 1000 Th1/Th2-primed enhancers were shown. l-m, Gene ontology analysis for top 1000 Th1-primed (l) and Th2-primed (m) enhancers. Significant GO terms with P values were reported by GREAT v3.0.0.

Extended Data Figure 10. Characterization of primed enhancers in undifferentiated mESCs.

a, Heat maps show H3K27ac in mESCs and EB cells and p300 in EB cells around mESC-specific and EB-specific enhancers. b, Profile of nucleosome occupancy from pooled mES scMNase-seq around mESC-specific and EB-specific enhancers. c, Normalized nucleosome occupancy within ±200 bp from the center of de novo EB enhancers for subgroups of mESCs that are primed for EB (red) or not primed for EB (black). d, Plots of fragment size ratio of subnucleosome-sized particles to nucleosomes versus nucleosome occupancy score at de novo EB enhancers for 144 mESCs. Pearson correlation coefficient and P value are indicated. P-value is the probability that one would have found the current result if the correlation coefficient were in fact zero (null hypothesis), and was calculated using R package. e, Plots of de novo EB enhancers ranked based on nucleosome occupancy difference between pooled primed cells and the non-primed cells. Enhancers associated with key genes for EB cells were labeled by genes along with ranks. f, Nucleosome positions in pooled or single primed (red) and non-primed (blue) cells at de novo EB-specific enhancers for Brachyury gene. g, GO analysis for top 1000 EB-primed enhancers. Significant GO terms with P values were reported by GREAT v3.0.0.