High-resolution spatial multi-omics reveals cell-type specific nuclear compartments

Abstract

The mammalian nucleus is compartmentalized by diverse subnuclear structures. These subnuclear structures, marked by nuclear bodies and histone modifications, are often cell-type specific and affect gene regulation and 3D genome organization^1-3. Understanding nuclear organization requires identifying the molecular constituents of subnuclear structures and mapping their associations with specific genomic loci in individual cells, within complex tissues. Here, we introduce two-layer DNA seqFISH+, which allows simultaneous mapping of 100,049 genomic loci, together with nascent transcriptome for 17,856 genes and a diverse set of immunofluorescently labeled subnuclear structures all in single cells in cell lines and adult mouse cerebellum. Using these multi-omics datasets, we showed that repressive chromatin compartments are more variable by cell type than active compartments. We also discovered a single exception to this rule: an RNA polymerase II (RNAPII)-enriched compartment was associated with long, cell-type specific genes (> 200kb), in a manner distinct from nuclear speckles. Further, our analysis revealed that cell-type specific facultative and constitutive heterochromatin compartments marked by H3K27me3 and H4K20me3 are enriched at specific genes and gene clusters, respectively, and shape radial chromosomal positioning and inter-chromosomal interactions in neurons and glial cells. Together, our results provide a single-cell high-resolution multi-omics view of subnuclear compartments, associated genomic loci, and their impacts on gene regulation, directly within complex tissues.

Extended Data Fig. 1|. Detailed schematic of single-cell spatial multi-omics.

a, Flow chart of the sample preparation and imaging in single-cell spatial multi-omics. b, Schematic of mRNA seqFISH+ with 4 barcoding rounds with 16-pseudocolors including one round of error correction using one fluorescent channel (640 nm). This coding scheme can resolve up to 4,096 barcodes, while a subset of 1,194 barcodes were used to resolve mRNA species in cell culture experiments. c, Schematic of intron seqFISH+ with 5 barcoding rounds with 12-pseudocolors including one round of error correction using one fluorescent channel (561 nm). This coding scheme can resolve up to 20,736 barcodes, while a subset of 17,856 barcodes were used to resolve intronic RNA species in cell culture and tissue experiments. d, Schematic of non-barcoded mRNA seqFISH using one fluorescent channel (488 nm in cell culture experiments; 640 nm in tissue experiments). Unlike the exponential barcoding scheme in b and c whose coding capacity increases exponentially to the number of barcoding rounds, the number of RNA species that can be distinguished increases linearly to the number of hybridization rounds. e, Schematic of sequential immunofluorescence using two fluorescent channels (640 nm and 561 nm) in cell culture and tissue experiments. Similar to d, the number of antibody species that can be multiplexed scales linearly to the number of hybridization rounds.

Extended Data Fig. 2|. Validation of mRNA and intron seqFISH+ in cell culture experiments.

a, Spearman correlation between average mRNA counts by mRNA seqFISH+ (left) or non-barcoded mRNA seqFISH (right) and bulk RNA-seq in mESCs (top) and NMuMG cells (bottom). b, Pearson correlation of average mRNA counts profiled by mRNA seqFISH+ (orange) and non-barcoded mRNA seqFISH (purple) between two biological replicates from mESCs. n = 1,194 and 53 genes for mRNA seqFISH+ and non-barcoded mRNA seqFISH profiling, respectively, in a, b. c, Visual representation of on- and off-target barcode counts and filtered barcodes by comparing mRNA seqFISH+ results from mESCs between seeds 3 and 4 decoding stringency (Methods). Those filtered barcodes (n = 150 barcodes, including both on-and off-target barcodes) were excluded from the downstream analysis. d, Total on- and off-target barcodes detected per cell by mRNA seqFISH+ in mouse ES cells (top) and NMuMG cells (bottom). e, Average counts per each on- and off-target barcode per cell by mRNA seqFISH+ in mouse ES cells (top) and NMuMG cells (bottom). n = 1,163, 2,783, and 150 for on-target, off-target, and filtered barcodes, respectively in c-e. f, Representative visualization of homologous chromosomes (chromosome 4) by DNA seqFISH+ and transcription active sites by intron seqFISH+ near chromosome territories in the nucleus of mESCs. Intron spots detected within 500 nm from chromosome territories of their own chromosome captured by DNA seqFISH+ were considered to be transcription active sites, and other spots were filtered out from the downstream analysis. g, Pearson correlation between average intron counts per cell computed from all intron spots and transcription active sites defined by the criteria in f. h, Spearman correlation between scaled intron counts by intron seqFISH+ and bulk GRO-seq, measuring the amount of transcriptionally active RNA polymerase II (RNAPII), in mESCs. i, Pearson correlation of average intron counts at the transcription active sites by intron seqFISH+ and non-barcoded intron seqFISH for n = 33 genes in mESCs. The slope of 0.29 indicates a relative detection efficiency of 29%. j, Pearson correlation of average barcode counts per cell by intron seqFISH+ between two biological replicates from mouse ES cells. n = 17,856 genes for intron seqFISH+ in g, h, j. k, Total barcode counts per cell by intron seqFISH+ in mESCs (top) and NMuMG cells (bottom). l, Average counts per each on- and off-target barcode per cell by intron seqFISH+ in mESCs (top) and NMuMG cells (bottom). n = 17,856 and 2,880 for on- and off-target barcodes, respectively in k and l. m, Heatmap of gene expression profiles across single cells for top differentially expressed genes between mESCs and NMuMG cells by intron seqFISH+ (n = 10 genes for each cell line). n, The GO terms (top five plus manually selected three) identified from intron seqFISH+ profiles between mESCs and NMuMG cells represent their corresponding cell type identities. n = 1,076 cells from two biological replicates of mESCs and n = 384 cells from one biological replicate of NMuMG cells in a-n.

Extended Data Fig. 3|. Additional schematic and validation of DNA seqFISH+ in cell culture experiments.

a, Implementation of DNA seqFISH+ (left), by leveraging the combinations of region barcoding (top) and chromosome block barcoding (bottom). To implement this, each primary probe (left, middle) contained three identical 15-nt readout binding sites (black) for one of the 60 regions in hybridizations 1–60 as well as three 15-nt readout binding sites (red) for three out of four chromosome paint barcoding rounds (hybridizations 61–96) in each fluorescent channel. Representative images for the nucleus from mESCs across 96 rounds of serial hybridization and imaging of two-layer DNA seqFISH+ (right) for region barcoding (top) and chromosome block barcoding (bottom). A fiducial marker targeting a locally repetitive 3632454L22Rik locus by a single DNA FISH probe appears in all rounds of imaging. Background subtracted images used for an analysis (Methods) are shown from a single z-section for visual clarity. b, Quantification of the median fiducial marker localization relative to the reference image across 96 rounds of two-layer DNA seqFISH+ imaging, representing a high image alignment accuracy with median error of 47.3 nm in 3D across imaging rounds. Shaded areas represent the interquartile range. n = 1,332–2,402 matched fiducial markers in each hybridization round from in three fluorescent channels. c, Quantification of the median reference fiducial marker localization between pairs of fluorescence channels after the chromatic shift correction in 3D, suggesting a minimum chromatic effects across fluorescent channels after the correction. n = 2,301 matched spots. d, Total on- and off-target barcodes detected per cell by two-layer DNA seqFISH+ in NMuMG cells (left). Average counts per each on- and off-target barcode per cell by two-layer DNA seqFISH+ in NMuMG cells (right). n = 100,049 and 31,171 for on- and off-target barcodes, respectively. e, The distribution of mean on-target barcode counts per cell grouped by chromosome identities by two-layer DNA seqFISH+ in mESCs (top) and NMuMG cells (bottom). The differences in detection efficiency between autosomal and X chromosomal loci reflect the identities of male mESCs and female NMuMG cells. We note chromosome 19 in NMuMG cells could be trisomy because of the 43.0% greater average barcode counts per cell than those in the other chromosomes. n = 100,049 loci in total. f, Heatmap comparing average spatial distances of pairs of loci between two-layer DNA seqFISH+ (upper right) and previous DNA seqFISH+⁷ (lower left) at the DNA seqFISH+ 1-Mb resolution loci in Chr4 in mESCs. n = 149 loci. g, Pearson correlation of mean spatial distances of pairs of intra-chromosomal loci between two-layer DNA seqFISH+ and previous DNA seqFISH+⁷ at the DNA seqFISH+ 1-Mb resolution 25-kb loci across the genome in mESCs. h, Pearson correlation of mean spatial distances of pairs of intra-chromosomal loci between two biological replicates of two-layer DNA seqFISH+ using the same loci in g in mESCs. n = 159,397 pairs that were detected in both measurements in g, h. i, Representative visualization of eigenvectors between Hi-C^, (top) and two-layer DNA seqFISH+ (bottom) for mESCs and NMuMG cells, confirming the highly concordant compartment organization between the measurements. j, Spearman correlation of eigenvectors between Hi-C and DNA seqFISH+ across the mouse genome for mESCs (top) and NMuMG cells (bottom). n = 23,547, 23,582 loci that were commonly profiled between the measurements for mESCs and NMuMG cells. 100 kb binning was used in i, j. k, Heatmap comparing median spatial distances of pairs of loci between two-layer DNA seqFISH+ (upper right) and previous DNA seqFISH+⁷ (lower left) at the previously profiled 25-kb resolution loci in Chr3, Chr13, and Chr14 in mESCs. n = 60 loci for each chromosome. Fluorescent channels used for two-layer DNA seqFISH+ loci are shown (top). We note that boundaries between loci in different fluorescent channels may be due to the differences in the efficiency of co-detecting pairs of loci across blocks (Method). l, Pearson correlation of median spatial distances of pairs of intra-chromosomal loci between two-layer DNA seqFISH+ and previous DNA seqFISH+⁷ at previously profiled 25-kb resolution loci at the selected regions in Chr1–19, X in mESCs. m, Pearson correlation of median spatial distances of pairs of intra-chromosomal loci between two biological replicates of two-layer DNA seqFISH+ using the same 25-kb loci in l in mESCs. n = 35,400 pairs that were commonly profiled between the measurements in l, m. n = 1,076 cells from two biological replicates of mESCs and n = 384 cells from one biological replicate of NMuMG cells in b-m.

Extended Data Fig. 4|. Validation of imaging-based chromatin profiling across the mouse genome.

a, b, Representative images by sequential immunofluorescence (n = 65 markers) in a and DNA FISH of repetitive elements (n = 6 markers) in b in mESCs. c, Pearson correlation of imaging-based chromatin profiling for each marker (n = 69 markers in a, b, except E-Cadherin and HA-tag that do not stain nucleus) across the genome between two biological replicates in mESCs (left), showing H3K9me2 as an outlier, and corresponding density plot for individual loci (n = 100,049 loci) for H3K9ac as an example (right). d, Pearson correlation of imaging-based chromatin profiling for each marker (n = 13 markers) between two-layer DNA seqFISH+ and previous DNA seqFISH+⁷ at the DNA seqFISH+ 1-Mb resolution 25-kb loci across the genome in mESCs (left) and corresponding density plot for individual loci (n = 2,460 loci) for H3K9ac as an example (right). e, Pearson correlation of imaging-based chromatin profiling and sequencing-based chromatin profiling (DamID for Lamin B1⁶⁹, CUT&Tag for other markers) across the genome in mESCs. We note that imaging-based chromatin profiling measures spatial proximity between DNA loci and subnuclear structures, while sequencing-based profiling captures molecular interactions between genomic DNA and antibodies of interest. n = 25,110 loci that were commonly profiled between the measurements. f, Representative visualization of Lamin B1 chromatin profiles by DamID and two-layer DNA seqFISH+ in mESCs and NMuMG cells (from top to bottom), representing similar patterns between mESC datasets. g, Comparison of Lamin B1 enrichment among loci with different categories as constitutive lamina-associated domains (cLADs), facultative LADs (fLADs), and constitutive inter-LADs (ciLADs)^, (n = 8,582, 2,908, and 10,239 loci), confirming higher Lamin B1 enrichments at LADs (cLADs and fLADs) than those at ciLADs. In box plots, the center lines for the median, boxes for the interquartile range, whiskers for values within 1.5 times the interquartile range, and points for outliers. h, Pearson correlation of imaging-based chromatin profiling of Lamin B1 and spatial distance from nuclear periphery in mESCs. Lamin B1 enriched DNA loci are physically closer to the nuclear periphery as expected. n = 25,148 loci. 100 kb binning was used in e-h. n = 1,076 cells from two biological replicates of mESCs and n = 384 cells from one biological replicate of NMuMG cells.

Extended Data Fig. 5|. Validation and characterization of spatial multi-omics measurements in the adult mouse brain cerebellum.

a, Large-field DAPI images of the adult mouse brain cerebellum coronal sections from two biological replicates. Yellow boxes represent unique fields of view (FOVs) imaged in each biological replicate. b, Representative images from a single z-section in one field of view for DAPI staining (left) and spatial distribution of cell clusters computed from mRNA seqFISH profiles (right). The cell type annotation for each cell cluster (0–11) was determined by the comparison with the single-nucleus RNA sequencing dataset shown in d. c, Representative images of spatial distribution of cell clusters for Purkinje cell subtypes (top) and corresponding marker gene expression in the nucleus (bottom), reflecting the known patterns of parasagittal stripes between Aldoc positive and negative Purkinje cells in the mouse cerebellum^,,. d, Comparison of cell clusters (0–11) defined by mRNA seqFISH and cell types identified by the single-nucleus RNA sequencing. Based on the degree of Pearson correlation, we annotated our mRNA seqFISH clusters as shown in b. e, Marker gene expression profiles in each cell cluster. Those include previously characterized marker genes^, such as Flt1 in Endothelial cells (cluster 8), Olig1 in oligodendrocyte precursor cells and oligodendrocytes (cluster 10), Aqp4 in Astrocytes (cluster 9), Gdf10 in Bergmann glia (cluster 2), Gabra6 in Granular cells (clusters 0 and 1), Sorcs3 in MLI1 (cluster 4), Nxph1 in MLI2/PLI (cluster 7), and Ppp1r17 in Purkinje cells (clusters 6 and 11, which can be further divided by Aldoc and Plcb4 as shown in c). f, Distribution of nuclear volume for cells in each cell cluster. g, Distribution of total intron counts per cell in each cell cluster by intron seqFISH+. In box plots, the center lines for the median, boxes for the interquartile range, whiskers for values within 1.5 times the interquartile range, and points for outliers in f, g. h, Reproducibility of intron seqFISH+ from two biological replicates with the adult mouse brain cerebellum. n = 17,849 genes that were detected at least once in each replicate. i, Comparison of gene expression profiles by intron seqFISH+ and single-nucleus RNA sequencing in four major cell types showed overall high consistency of cell-type specific gene expression programs. j, Heatmap of nascent gene expression profiles of differentially expressed genes by intron seqFISH+ across single cells grouped by cell clusters defined by mRNA seqFISH. k, Heatmap of sequential immunofluorescence intensity profiles across cell clusters defined by mRNA seqFISH. Note that HDAC1 and SOX2 were expressed in Bergmann glia (cluster 2) and astrocyte (cluster 9), consistent with previous studies^,. l, Representative sequential immunofluorescence images from a single z-section in the adult mouse brain cerebellum. m, UMAP-embedding of cells colored by two biological replicates (left) and each cell cluster defined by intron seqFISH+ profiles (middle) or sequential immunofluorescence intensity profiles (right). n, Comparison of cell clusters defined by mRNA seqFISH and intron seqFISH+ (top) or sequential immunofluorescence (bottom), representing robust identification of similar cell clusters regardless of the measurement modalities in the adult mouse brain cerebellum. n = 4,015 cells (n = 1,504, 832, 518, 357, 263, 164, 113, 88, 76, 56, 29, 15 cells from mRNA seqFISH cluster 0 to 11) from two biological replicates of the adult mouse brain cerebellum.

Extended Data Fig. 6|. Additional validation and characterization of spatial multi-omics measurements in the adult mouse brain cerebellum.

a, Total on- and off-target barcodes detected per cell by DNA seqFISH+ in the mouse cerebellum (left). Average counts per each on- and off-target barcode per cell by DNA seqFISH+ in the mouse cerebellum (right). n = 100,049 and 31,171 for on- and off-target barcodes, respectively. b, Total on-target barcodes detected per cell by DNA seqFISH+ in each transcriptionally-defined cell cluster in the mouse cerebellum. n = 1,504, 832, 518, 357, 263, 164, 113, 88, 76, 56, 29, 15 cells from mRNA seqFISH cluster 0 to 11. c, Pearson correlation of imaging-based chromatin profiling across the genome for each marker (n = 26 markers) in each cell type between two biological replicates of the adult mouse brain cerebellum. In box plots, the center lines for the median, boxes for the interquartile range, whiskers for values within 1.5 times the interquartile range, and points for outliers in b, c. d, Quantification of overlap of top 5% genomic loci associated with each marker between pairs of markers in Purkinje cells, largely separating active and repressive chromatin markers. e, Genomic features and representative imaging-based chromatin profiling with four markers in each cell type in chromosome 16. f, Representative raw immunofluorescence images for various repressive markers from a single z-section for each cell type (left). Purkinje-specific pericentromeric staining by H3K27me2 and mH2A1 is highlighted by green arrows. The intense mH2A1 and H3K27me3 clusters visible in Bergmann glia highlighted by white arrows represent the inactive X chromosome (Xi) territory in the female mouse cerebellum section. Scatter plots for individual loci show the relationship of each repressive marker enrichment in each cell type (right). g, Degree of overlap of top 5% enriched loci between pairs of repressive markers in each cell type. h, Degree of similarity of chromatin profiles between pairs of cell types by Pearson correlation, including cell lines (mESCs and NMuMG cells). 200 kb binning (n = 12,562 loci in total) was used for the analysis in c-h. n = 4,015 cells from two biological replicates of the adult mouse brain cerebellum.

Extended Data Fig. 7|. Additional characterization of distinct active subnuclear compartments.

a, Representative classification of three types of active domains (nuclear speckle, RNAPIISer5-P broad and sharp) along with the chromatin profiling of SF3A66 (nuclear speckle marker) and RNAPIISer5-P across cell types. b, Comparison of genomic features across different active domains in each cell type. n = 2,531, 1,789, 458, 4,220 loci (Granule), n = 2,533, 2,060, 1,228, 4,654 loci (MLI1), and n = 2,376 2,015 1,367 4,210 loci (MLI2+PLI) from left to right category. c, Representative raw images of active markers from a single z-section of the adult mouse brain cerebellum, showing distinct compartmentalization between nuclear speckles (SF3A66) and other active markers (e.g. RNAPIISer5-P, H3K4me2, H4K8ac) regardless of the cell types. d, GO term comparison between SF3A66 and RNAPIISer5-P associated genomic loci in each cell type, representing largely distinct enrichment. Similar observation of distinct pathway enrichments between speckle-associating and speckle-non-associating p53 target genes was previously observed in human cell lines. e, Overlap of each active domain across cell types, showing more cell-type specific organization of RNAPIISer5-P sharp domains. f, Comparison of differential association of genomic loci with SF3A66 and RNAPIISer5-P between pairs of cell types. g, Comparison of differential mRNA expression between pairs of cell types at differentially associated loci with either PolII (RNAPIISer5-P) or Speckle (SF3A66) in each cell type. n = 150, 114, 198, 126 loci (MLI1 vs. Purkinje), n = 203, 136, 139, 146 loci (Granule vs. Purkinje), n = 164, 85, 143, 79 loci (Granule vs. MLI1), and n = 147, 96, 117, 87 loci (Granule vs. Bergmann) from left to right category. In box plots, the center lines for the median, boxes for the interquartile range, whiskers for values within 1.5 times the interquartile range, and points for outliers in b, g. h, Representative genomic regions with long genes (>200 kb) (top), corresponding mRNA expression (middle), and chromatin profiles of different active markers (bottom) in each cell type. Nuclear speckles marked by SF3A66 were not enriched at the differentially expressed long genes highlighted with red arrows (top). i, Comparison of Purkinje RNAPIISer5-P enrichment between long genes with the absence or presence of highly open chromatin regions defined by ATAC-seq >20 kb peaks in Purkinje cells. The red dots for the median and error bars for the interquartile range. p values by two-sided Wilcoxon’s signed rank-sum test. 200 kb binning (n = 12,562 loci in total) was used for the analysis in a, b, e-h. n = 2,336, 128, 263, 88, and 518 cells for Granule, Purkinje, MLI1, MLI2+PLI, and Bergmann glia cells from two biological replicates of the adult mouse cerebellum.

Extended Data Fig. 8|. Additional visualization and characterization of cell-type specific repressive subnuclear compartments.

a, The top eleven GO terms identified from H3K27me3 chromatin profiles in each cell type represent the enrichment of similar GO terms across cell types. b, Comparison of genomic features between loci that were differentially associated with H3K27me3 between two cell types. n = 595, 359 loci for Purkinje cells (PC) vs. Bergmann glia (BG) and n = 169, 491 loci for MLI1 vs. BG. c, Additional examples of H3K27me3 profiles across cell types along with Hox cluster in Chr2 and Purkinje-specific H3K27me3-associated long gene loci (top). d, Representative single cell visualization of genomic loci relative to H3K27me3 in Bergmann glia. Scale bars, 500 nm. e, Violin plot of expression changes between P0 and adult Purkinje cells, representing the down-regulaiton of 36 out of 46 differentially expressed genes categorized as Purkinje-specific H3K27me3-associated long genes from the P0 to adult. f, Comparison of spatial distance from nuclear periphery and H3K27me3 or Lamin B1 chromatin profiles in each cell type. g, Comparison of differential association with each subnuclear marker at differentially expressed loci defined by intron seqFISH+ between pairs of major cell types. The increased association with SF3A66 at transcriptionally up-regulated loci was similarly observed in the adult mouse cerebral cortex⁸. n = 114, 129 loci for Bergmann glia vs. Purkinje cells, n = 48, 81 loci for MLI1 vs. Purkinje cells, and n = 151, 114 loci for Bergmann glia vs. MLI1. p values by two-sided Wilcoxon’s signed rank-sum test in b, g. In box plots, the center lines for the median, boxes for the interquartile range, whiskers for values within 1.5 times the interquartile range, and points for outliers in b, g. 200 kb binning (n = 12,562 loci in total) was used for the analysis in a-c, f, g. n = 128, 263, 88, and 518 cells for Purkinje, MLI1, MLI2+PLI, and Bergmann glia cells from two biological replicates of the adult mouse cerebellum.

Extended Data Fig. 9|. Additional characterization of H4K20me3 subnuclear compartments.

a, Cartoon illustration of a mouse chromosome (top) and representative imaging-based chromatin profiling of repressive markers in Purkinje cells in chromosome 17 (bottom). b, Scatter plots of pairs of chromatin profiles (left), colored by the locus category of MajSat (orange), H4K20me3 (purple), H4K20me3 weak (light purple), or others (gray) and corresponding box plots showing H4K20me3 enrichment for loci in each category (right). c, Comparison of quantile of H4K20me3 and H4K20me3-weak category loci in b for each repressive marker, representing stronger enrichment of repressive markers around pericentromeric heterochromatin (MajSat, H4K20me3, H3K9me3, ATRX, H3K27me2, Fibrillarin, and mH2A1) at H4K20me3 loci (purple) over H4K20me3-weak loci (light purple). In contrast, two repressive markers (H3K27me3 and Lamin B1) are more depleted at H4K20me3 loci compared to H4K20me3-weak loci. d, Barplots comparing the locus characteristics such as mCH desert and SSDRs^, between H4K20me3-weak loci (n = 455) and all 200-kb loci (n = 12,562) in Purkinje cells. e, Gene family characteristics either mCH desert, SSDRs^,, or both (left) and their radial positioning relative to nuclear periphery and H4K20me3 enrichment (middle), as well as their enrichments across chromosomes (right). Only a subset of genomic loci annotated with those categories with corresponding gene family names were included in this analysis. In box plots, the center lines for the median, boxes for the interquartile range, whiskers for values within 1.5 times the interquartile range, and points for outliers in b, c, e. f, H4K20me3 and H4K20me3-weak loci (n = 236, 455, respectively) distribution across chromosomes in Purkinje cells. g, GO term comparison between H4K20me3 and H4K20me3-weak associated genomic loci in Purkinje cells revealed enrichment of distinct gene families at each category. h, Additional visualization of H4K20me3-enriched Vmn and Olfr gene family loci overlaid with H4K20me3 staining with a maximum z-projection of four sections in Purkinje cells. i, Additional examples of ensemble-averaged and single allele chromatin profiles sorted by H4K20me3 enrichment from bottom to top in Purkinje cells. j, Comparison of ensemble-averaged radial positioning of genome-wide DNA loci in the nucleus across cell types separates neurons and glial cells. k, Comparison between RNAPIISer5-P enrichment and spatial distance from nuclear periphery for individual loci. Top 5% H4K20me3-enriched loci (purple, n = 277) in each cell type tend to localize in the nuclear interiors yet excluded from RNAPIISer5-P in neurons (Purkinje, MLI1, and MLI2/PLI) but not in Bergmann glia cells. Those neuron-specific outlier loci found here were not observed in human cell culture. l, Transition of H4K20me3-enriched loci from Purkinje to MLI2/PLI cells showed largely conserved H4K20me3-associated loci between neuronal cell types. n = 252 loci. m, The H4K20me3-enriched regions in Purkinje cells (top) and H4K20me3 and Lamin B1 chromatin profiles across cell types in Chr7 (bottom). 200 kb binning (n = 12,562 loci in total) was used for the analysis in a-g, i-m. n = 4,015 cells from two biological replicates of the adult mouse brain cerebellum.

Extended Data Fig. 10|. Additional visualization and characterization of 3D genome organization with subnuclear structures.

a, Ensemble-averaged spatial distances between pairs of genomic loci along with chromatin profiling from each cell type across chromosomes (Chr1–19, X). b, Additional examples of ensemble-averaged spatial distances between pairs of genomic loci along with chromatin profiling from each cell type at specific chromosomes. c, Cumulative distribution of inter-chromosomal distances between pairs of loci with top 5% association to MajSat or Purkinje-specific H4K20me3 characterized in Fig. 5c (left) or from loci associated with specific gene families (right) compared to random pairs of loci (n = 1,000 trials). d, Representative 3D images of MajSat or H4K20me3 staining and chromosomal loci in each cell type. Identified Purkinje H4K20me3 loci highlighted with colors tend to colocalize at H4K20me3 territories in Purkinje and MLI1 cells, but localize at the nuclear periphery in Bergmann glia cells. 1.5 Mb binning (n = 1,678 loci in total), grouped by the chromosome paint block barcodes, was used for the analysis in a-c. n = 128, 263, and 518 cells for Purkinje, MLI1, and Bergmann glia cells from two biological replicates of the adult mouse cerebellum.

Fig. 1|. Development and validation of single-cell spatial multi-omics technology.

a, Schematic of spatial multi-omics, imaging mature and nascent transcriptome as well as chromosome and subnuclear structures within the same single cells. b, Detailed schematic of DNA seqFISH+. Chromosomes are splitted into 1.5-Mb blocks, in which each 25-kb locus is imaged as a diffraction limited spot during the first 60 rounds of imaging with three orthogonal fluorescent channels. Then each 1.5-Mb block is uniquely barcoded across 4 rounds with 9-pseudocolors during the subsequent 36 rounds of imaging. Finally, by decoding the region and block identities, 100,049 of 25-kb loci can be uniquely resolved across 20 chromosomes. c, Visualization of decoded RNA (left) and DNA spots (middle), immunofluorescence raw images (top right), and DNA spots colored by z-score normalized immunofluorescence intensity (bottom right) in the nucleus from mESCs. d, Total on- and off-target barcode counts per cell (top). Average counts of individual barcode per cell (bottom). n = 100,049 and 31,171 DNA loci for on- and off-target barcodes. e, f, Comparison of our imaging-based chromatin profiles with those from previous DNA seqFISH+⁷ and sequencing-based technologies for the selected genomic region (e) and across the genome (f) with 100 kb binning (n = 25,110 loci). n = 1,076 cells from two biological replicates of mESCs and n = 384 cells from one biological replicate of NMuMG cells in c-f.

Fig. 2|. Single-cell spatial multi-omics in the adult mouse brain cerebellum.

a, Spatial distribution of cell type clusters from a single z-section of the adult mouse cerebellum. b, Cell type clusters determined by mRNA seqFISH projected onto UMAP-embedding of mRNA seqFISH (left), intron seqFISH+ (middle), sequential immunofluorescence (right). c, Decoded DNA spots (top) from a single z-section of cells in the black box in a. Zoomed-in 3D views of decoded intron spots (middle left) and DNA spots (middle right, bottom) in the nucleus from a yellow-boxed Purkinje cell (top). d, Immunofluorescence raw images (left) and DNA spots colored by z-score normalized immunofluorescence intensity (right) in the nucleus from a cell highlighted in c. e, Genomic features and representative imaging-based chromatin profiling in Purkinje cells in chromosome 15. f, Representative raw immunofluorescence images from a single z-section for each cell type (top). Illustration showing cell-type specific organization of repressive markers (bottom). The intense H3K27me3 cluster visible in Bergmann glia, representing the inactive X chromosome territory, is not depicted. g, Degree of similarity of chromatin profiles between pairs of cell types (left) and corresponding examples (right). n = 12,562 loci. 200 kb binning was used for the visualization and analysis (e, g). n = 4,015 cells from two biological replicates of the adult mouse cerebellum in a-g.

Fig. 3|. Distinct subnuclear compartments organize transcriptionally-active genomic loci.

a, Representative classification of three types of active domains (nuclear speckle, RNAPIISer5-P broad and sharp) in Purkinje cells and Bergmann glia along with the chromatin profiling of SF3A66 (nuclear speckle marker) and RNAPIISer5-P. b, Visualization of genomic loci colored by the ensemble-average active domain classification in a (left) and corresponding raw immunofluorescence image (right) from a single z-section for each cell type. c, Comparison of genomic features across different active domains in each cell type. n = 1,405, 4,102, 679, 4,892 loci (Purkinje) and 3,161, 3,115, 411, 4,080 loci (Bergmann) from left to right category. d, Comparison of differential association of genomic loci with SF3A66 and RNAPIISer5-P between pairs of cell types. e, Comparison of differential mRNA expression between pairs of cell types at differentially associated loci for either PolII (RNAPIISer5-P) or Speckle (SF3A66) in each cell type, colored in d. n = 110, 55, 121, 94 loci (top) and 134, 94, 109, 67 loci (bottom) from left to right category. In box plots, the center lines for the median, boxes for the interquartile range, whiskers for values within 1.5 times the interquartile range, and points for outliers (c, e). f, Representative genomic regions with long genes (>200 kb) (top), corresponding mRNA expression (middle), and RNAPIISer5-P chromatin profiles (bottom) in each cell type. g, Heatmaps for RNAPIISer5-P chromatin profiles (left), nascent RNA expression (middle), and mRNA expression (right) for highly correlated long genes (n = 132 genes, Methods). h, Similarity of RNAPIISer5-P peaks with other markers on the long genes in g in each cell type. i, Representative single cell visualization of long gene loci with cell-type specific gene expression (Dpp10 in Purkinje cells and Adgrl3 in MLI1), relative to nuclear speckles (SF3A66) and RNAPIISer5-P with a maximum z-projection of two z-sections. Scale bars, 500 nm. j, Illustration showing nuclear speckle and RNAPIISer5-P subnuclear compartments associated with distinct genomic loci in a cell-type specific fashion. 200 kb binning (n = 12,562 loci in total) was used for the analysis and visualization (a, c-h). n = 2,336, 128, 263, 88, and 518 cells for Granule, Purkinje, MLI1, MLI2+PLI, and Bergmann glia cells from two biological replicates of the adult mouse cerebellum in a-i.

Fig. 4|. Cell-type specific organization of H3K27me3 repressive subnuclear compartments.

a, Comparison of genomic features between loci that were differentially associated with H3K27me3 in two cell types. n = 600, 260 loci for Purkinje cells (PC) and MLI1. p values by two-sided Wilcoxon’s signed rank-sum test. The center lines for the median, boxes for the interquartile range, whiskers for values within 1.5 times the interquartile range, and points for outliers. b, Comparison of Lamin B1 and H3K27me3 chromatin profiles in each cell type. c, H3K27me3 profiles across cell types, highlighted with Hox cluster. Identified Purkinje-specific long gene loci associated with H3K27me3 are shown as blue dots (b) and binary heatmap (c). d, Representative single cell visualization of genomic loci highlighted (b, c), relative to H3K27me3. Scale bars, 500 nm. e, Illustration showing H3K27me3 subnuclear compartments associated with common and Purkinje-specific genomic loci at the nuclear interior or periphery. 200 kb binning (n = 12,562 loci in total) was used for the analysis and visualization (a-c). n = 128, 263, 88, and 518 cells for Purkinje, MLI1, MLI2+PLI, and Bergmann glia cells from two biological replicates of the adult mouse cerebellum in a-d.

Fig. 5|. The H4K20me3 compartment confines specific genomic loci in neurons.

a, DNA FISH (MajSat) and sequential immunofluorescence images for markers enriched near pericentromeric repressive heterochromatin in Purkinje cells. b, H4K20me3 enrichment across chromosomes in Purkinje cells. c, Barplots comparing the locus characteristics such as mCH desert and SSDRs^, between MajSat- and H4K20me3-enriched loci in Purkinje cells. d, Ensemble-averaged and single allele chromatin profiles sorted by H4K20me3 enrichment from bottom to top in Purkinje cells. e, Visualization of H4K20me3-enriched Vmn and Olfr gene family loci with H4K20me3 staining. f, Illustration showing the differences of subnuclear localization between Vmn (magenta) and Olfr (cyan) gene family loci. g, Comparison of other genomic features (top) along with radial positioning of chromosomal loci across cell types (bottom). h, H4K20me3 and Lamin B1 chromatin profiles at the H4K20me3-enriched regions, highlighted by triangles (g). i, Visualization of H4K20me3 enriched loci in Chr4 and Chr7 (g) overlaid on the H4K20me3 immunofluorescence image in Purkinje and Bergmann cells. j, Transition of H4K20me3-enriched loci from Purkinje cells to MLI1 or Bergmann glia. n = 252 loci. k, Illustration showing the localization switching of genomic loci between neurons and Bergmann glia. 200 kb binning (n = 12,562 loci in total) was used for the analysis and visualization. n = 128, 263, 88, and 518 cells for Purkinje, MLI1, MLI2+PLI, and Bergmann glia cells from two biological replicates of the adult mouse cerebellum in a-e, g-j.

Fig. 6|. Cell-type specific subnuclear compartmentalization and 3D genome organization.

a, Ensemble-averaged spatial distances between pairs of genomic loci along with chromatin profiling at 1.5 Mb resolution from each cell type. Locations of pairs of H4K20me3-associated loci in Purkinje cells in Fig. 5c are highlighted as black circles. b, Representative 3D images of H4K20me3 staining and chromosomal loci in each cell type. c, Cumulative distribution of inter-chromosomal distances between pairs of loci with top 5% association to a given marker compared to random pairs of loci (n = 1,000 trials). 1.5 Mb binning (n = 1,678 loci in total), grouped by the chromosome paint block barcodes, was used. d, Illustration showing the differences of cell-type specific intra- and inter-chromosomal spatial arrangements around H4K20me3-enriched subnuclear compartment in neurons or at the nuclear periphery in Bergmann glia. n = 128, 263, and 518 cells for Purkinje, MLI1, and Bergmann glia cells from two biological replicates of the adult mouse cerebellum in a-c.