Integrated spatial genomics in tissues reveals invariant and cell type dependent nuclear architecture
Yodai Takei
1
, Shiwei Zheng
2
, Jina Yun
1
, Sheel Shah
1
, Nico Pierson
1
, Jonathan White
1
, Simone Schindler
1
,
Carsten Tischbirek
1
, Guo
-
Cheng Yuan
2
, and
Long Cai
1
*
1. Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, CA, USA.
2. Department of Genetics and Genomic Sciences and Charles Bronfman Institute for Personalized
Medicine, Icahn School of Medicine at Mount S
inai, New York, NY, USA.
*
Correspond
ing author.
Email
: lcai@caltech.edu
Abstract
Nuclear architecture in tissues can arise from cell
-
type specific organization of nuclear bodies, chromatin
states and chromosome structures. However, the lack of genome
-
wide measurements to interrelate
such modalities wit
hin single cells limits our overall understanding of nuclear architecture. Here, we
demonstrate integrated spatial genomics in the mouse brain cortex, imaging thousands of genomic loci
along with RNAs and subnuclear markers simultaneously in individual cel
ls. We revealed chromatin fixed
points, combined with cell
-
type specific organization of nuclear bodies, arrange the interchromosomal
organization and radial positioning of chromosomes in diverse cell types. At the sub
-
megabase level, we
uncovered a collec
tion of single
-
cell chromosome domain structures, including those for the active and
inactive X chromosomes. These results advance our understanding of single
-
cell nuclear architecture in
complex tissues.
Main Text
The three
-
dimensional (3D) organization o
f the genome is critical for many cellular processes, from
regulating gene expression to establishing cellular identity
(
1
–
3
)
. Genome organization
has been
extensively exam
ined using sequencing
-
based genomics and microscopy approaches
(
4
,
5
)
. In particular,
chromosome architectures, such as topologically associating domains (TADs)
(
6
–
8
)
and high
-
order
chromosomal interactions
(
9
,
10
)
, have been reveal
ed by high
-
throughput genomics approaches such as
Hi
-
C
(
11
)
, genome architecture mapping
(GAM)
(
9
)
, and split
-
pool recognition of interactions by tag
extension (SPRITE)
(
10
)
. Moreover, recent progress in chromosome capture methods has enabled the
exploration of ch
romosome structures at the single
-
cell level
(
12
–
19
)
. These studies have characterized
the variabilities of chromosome structures in single cells derived from various biological samples.
Complementary to the gen
omics approaches, imaging
-
based approaches such as DNA fluorescence in
situ hybridization (FISH)
(
20
)
can directly obtain 3D chromosome structures from measured loci in single
cells without computational reconstructions. Recent multiplexed imaging
-
based methods
(
21
–
32
)
, such
as sequential DNA FISH
(
22
)
and in situ genome sequencing
(IGS)
(
31
)
, have directly characterized the
variabilities of chromosome structures in 3D
, even between homologous chromosomes in a cell, and
reported TAD
-
like domain structures in single cells
(
23
)
, which when averaged over populations of cells
are consistent with sequencing
-
based bulk measurements. Furthermore, super
-
resolution imaging
.
CC-BY-NC-ND 4.0 International license
available under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (which
this version posted April 27, 2021.
;
https://doi.org/10.1101/2021.04.26.441547
doi:
bioRxiv preprint
studies
(
23
,
33
,
34
)
have shown that architectural proteins such as C
TCF and cohesin can play important
roles in the single
-
cell domain structures.
To better understand the principles underlying chromosome organization, it is crucial to integrate
chromosome structures with other measurements that capture transcriptional sta
tes
(
35
)
, chromatin
states
(
36
)
, nuclear bodies
(
37
,
38
)
and radial organization of the nucleus
(
39
)
in single cells. Single
-
cell
multimodal genomics technologies
(
40
)
can evaluate chromosome structures together with, for
instance, DNA methylome profiling
(
41
,
42
)
. However
, sequencing
-
based single
-
cell multimodal
measurements of chromosome structure and the transcrip
tome in the same cell remain challenging. On
the other hand, imaging
-
based approaches allow direct integration of multimodal measurements
including chromosome structures
(
22
,
23
,
25
,
27
–
32
)
.
We recently established an imaging
-
based integrated spatial genomics approach that enables the
analysis of nuclear organization beyond chromosome structures
(
32
)
. Briefly, we imaged
thousands of
genomic loci by DNA seqFISH+ along with transcriptional states by RNA seqFISH and subnuclear
localization of histone modifications and nuclear bodies by sequential immunofluorescence (IF) in single
mouse embryonic stem (ES) cells. We discovered that chromosomes consistently associate with specific
nuclear bodies, such as nuclear speckles
(
43
)
and nucleolus
(
44
)
, across many single cells. We found that
individual chromosomes contain unique combin
ations of fixed loci that are consistently associated with
different nuclear bodies and chromatin modifications, suggesting a scaffolding of chromosomes across
multiple nuclear bodies and protein globules. Nevertheless, to what extent these principles for
nuclear
organization extend to diverse cell types, and in a complex and physiologically relevant context such as
mammalian tissues, is largely unknown. Imaging
-
based multimodal measurements of multiple cell types
in tissues will give us a great opportunity
to dissect cell
-
type specific features and invariant principles in
nuclear organization in the native context.
Integrated spatial genomics in the brain
.
To comprehensively investigate the principles of nuclear organization among single cells in a tissue,
we
analyzed sections of the adult mouse cerebral cortex. Specifically, we applied our integrated spatial
genomics approach
(
32
)
to evaluate 3,660 DNA loci, 76 cellular RNAs, and 8 chromatin marks and nuclear
bodies
(
45
)
(Fig. 1 and fig. S1 and table S1 to S4). This technology provides a powerful multimodal tool to
integrate the transcriptional states, chromosome and subnuclear structures, radial nuclear organization,
and chromatin
and morphological features simultaneously in the same cells in tissues. We chose the
mouse brain as a model as it comprises many distinct cell types and has been extensively studied by single
cell RNA sequencing
(
46
–
48
)
as well as by spatial transcriptomics methods
(
49
–
54
)
.
We examined chromosome structures of the entire genome by using DNA seqFISH+ to i
mage 2,460 loci,
at approximately 1
-
Mb resolution, in 20 chromosomes. These data were collected using a 16 pseudocolor
seqFISH+ coding scheme
(
32
,
53
)
in two independent fluorescent channels (Fig. 1C an
d table S1). In
addition, for each of the 20 chromosomes, we examined a local region of at least 1.5 Mb by imaging an
additional 1,200 loci at 25
-
kb resolution
(
32
)
. We collected these data in one fluorescent channel by
.
CC-BY-NC-ND 4.0 International license
available under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (which
this version posted April 27, 2021.
;
https://doi.org/10.1101/2021.04.26.441547
doi:
bioRxiv preprint
imaging 60 consecutive loci on each of the 20 chromosomes (Fig. 1D and table S1). In single cells, individual
chromosomes fo
rmed distinct chromosome territories that have highly variable structures (Fig. 1E). We
detected 2,813.0 ± 1,334.0 (median ± standard deviation) spots per cell in total across three fluorescent
channels in 2,762 cells from three biological replicates (Fig.
1F, 1G and fig. S1A, S1G and S1H
(
55
)
). This
corresponds to an estimated detection efficie
ncy of at least 38.4 ± 18.2% (median ± standard deviation)
in post
-
mitotic cells with the diploid genome. On the other hand, the false positive spots, as determined
by the barcodes unused in the codebook (table S1), were detected at 32.0 ± 27.8 (median ± s
tandard
deviation) per cell (Fig. 1F and 1G).
We validated our DNA seqFISH+ data by comparing it with bulk Hi
-
C data from mouse cortex
(
6
,
56
)
(Fig.
1H and fig. S2A to D). The Hi
-
C normalized read coun
ts correlated with the mean spatial proximity
probability, with a Pearson correlation coefficient of 0.89 and 0.76 at the 1
-
Mb and 25
-
kb resolution,
respectively (fig. S2A to D). Similarly, Hi
-
C normalized read counts showed high concordance with DNA
seqFI
SH+ spatial distance at 1
-
Mb and 25
-
kb resolution (fig. S2A to D). Furthermore, the DNA seqFISH+
data from the three biological replicates were highly reproducible, with a Pearson correlation coefficient
of 0.95
-
0.97 for 1
-
Mb and 25
-
kb resolution data (fig
. S2E and S2F). These results demonstrate the
robustness of DNA seqFISH+ to map 3D chromosome structures in tissue samples.
We clustered the RNA seqFISH data and obtained 9 major cell type clusters within the cerebral cortex,
based on the gene expression
of known markers (fig. S1C and S1D), shown in the UMAP representation
(Fig. 1I). These 9 clusters matched well with cell types identified from single
-
cell RNA sequencing
(
47
)
(fig.
S1E). The majority of the cells were excitatory neurons in cluster 9 expressing Slc17a7 (Fig. 1I and fig. S1D).
We also observed four su
bclasses of inhibitory neurons expressing Pvalb, Vip, Ndnf, or Sst, three types of
glial cells (astrocytes expressing Mfge8, microglia expressing Csf1r and oligodendrocyte precursor cells
and oligodendrocytes expressing Olig1), and endothelial cells expres
sing Cldn5 (fig. S1D). We observed a
similar localization accuracy of the FISH spots (fig. S1F) and number of decoded DNA loci (fig. S1H) across
different cell type clusters.
In addition to the genome and RNA imaging, we used sequential IF to detect 6 hist
one modifications or
variants (H3K4me2, H3K27me2, H3K27me3, H3K9me3, H4K20me3, mH2A1), nuclear speckles (SF3a66)
and the methyl CpG binding protein MeCP2 (Fig. 2A and fig. S1K, S1L and S3A and table S3). We incubated
tissue sections with oligonucleotide
-
co
njugated primary antibodies
(
57
,
58
)
prior to the RNA seqFISH steps
and sequentially read out the antibody signals with fluorescently labeled probes
(
32
)
, allowing the
multiplexed detection of primary antibodies from the same single cells in tissues. The protocol previously
used for cell c
ulture experiments
(
32
)
was optimized for tissue sections to preserve the nuclear struct
ure
and accurately align the IF, RNA seqFISH and DNA seqFISH+ data over a total of 125 rounds of
hybridizations and imaging on an automated confocal microscope
(
45
)
(fig. S1I and S1J). Additionally, we
performed sequential RNA and DNA FISH to detect 3 non
-
coding RNA that mark the inactive X
chromosome (Xi, Xist), the nucleolus (ITS1), nuclear sp
eckles (Malat1)
(
43
,
44
,
59
)
as well as 5 repetitive
regions (LINE1, SINEB1, MajSat, MinSat, Telomere) that relate to nuclear organization
(
60
–
62
)
(fig. S1K,
S1L and S3A).
The spatial overlap between individual antibody, RNA and DNA markers of nuclear bodies or subnuclear
compartments was consistent with previously observed subnuclear localization patterns
(
37
,
43
,
44
,
59
–
.
CC-BY-NC-ND 4.0 International license
available under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (which
this version posted April 27, 2021.
;
https://doi.org/10.1101/2021.04.26.441547
doi:
bioRxiv preprint
64
)
. For example, individual cells displayed colocalization of nuclear speckle components (Malat1 and
SF3a66), inactive X chromosome (Xi) components
(Xist, mH2A1, H3K27me3, LINE1) and heterochromatin
components (DAPI, MajSat, H3K9me3, H4K20me3, MeCP2) (fig. S1K and S1L). We also observed minimum
spatial overlap between different nuclear bodies such as nuclear speckles (Malat1 and SF3a66) and the
nucleo
lus (ITS1), as well as between euchromatin
-
(SINEB1) and heterochromatin
-
(LINE1) enriched
chromosomal regions, as expected
(
61
)
(fig. S1K and S1L). Taken together, these results demonstrate that
our integrated spatial genomics approach allows us to explore nuclear organization with unprecedented
detail at the RNA, DNA and prot
ein level across diverse cell types in intact tissues.
Distinct nuclear features across cell types in the brain
We first examined the differences in global histone modifications and variants across cell types in the
mouse cortex to understand cell
-
type spe
cific global chromatin states. The overall intensities of IF markers
were highly variable across single cells (Fig. 2A and fig. S3A). The 9 major cell types displayed clear
differences in the global levels of both repressive marks (MeCP2, H3K27me3 and mH2A
1) and an active
mark (H3K4me2) (Fig. 2B). Clustering of single cells using the multiplexed IF data was able to distinguish
the same 9 cell types as identified by RNA seqFISH (Fig. 2C), supporting a strong correlation between
global chromatin states and tr
anscriptional states. In particular, we observed a relative enrichment of
mH2A1 in inhibitory neurons and astrocytes, H3K4me2 in inhibitory neurons, oligodendrocyte precursor
cells and oligodendrocytes, and H3K27me3 in excitatory neurons (Fig. 2A and 2B an
d fig. S3A and S3B). In
addition, MeCP2 was enriched in inhibitory and excitatory neurons while lower and almost undetectable
levels of intensities were observed in astrocytes and microglia, respectively (Fig. 2A and 2B and fig. S3A).
This observation agre
es with the previously reported MeCP2 immunostaining intensity in neurons,
astrocytes and microglia in the mouse cortex
(
65
)
.
Interestingly, even the DAPI features alone were sufficient to separate the major cell types in the cortical
areas of the mouse brain using both UMAP and hierarchical clustering (Fig. 2A and 2D to F). We used a
subset
of 701 cells for nuclei that were fully covered in the brain section. Compared to glial cells, neurons
typically had larger nuclear volumes (Fig. 2D and 2F) and lower DAPI intensities per voxel (e.g. mean and
median) (Fig. 2F), consistent with the same DN
A content in both cell types. In addition, among neuronal
cell types, we also observed a larger nuclear volume in excitatory neurons compared to those in inhibitory
neuron subtypes (Fig. 2D and 2F). These results are consistent with the observation that nu
clear
morphological features are often sufficiently distinct in mammalian tissues
(
66
)
t
o determine major cell
types by visual inspection of the images.
Lastly, we examined the spatial distance between pairs of intra
-
chromosomal loci to understand cell
-
type
specific chromosomal scaling in the nucleus. Although previous imaging studies in cel
l culture and
embryos showed differences in the chromosomal scaling of spatial distance as a function of genomic
distance
(
22
,
31
,
32
)
, it remains unclear how the scaling principles operate across different cell types
within the same tissues. We found that the relationship between spatial ve
rsus genomic distance scaling
was distinct in different cell types, in both 1
-
Mb and 25
-
kb resolution data (Fig. 2G and 2H and fig. S3C to
G). For example, loci tens of megabases apart typically displayed a larger spatial separation in inhibitory
neurons c
ompared to excitatory neurons and glial cells (Fig. 2G and fig. S3C), which cannot be simply
explained by nuclear size differences as inhibitory neurons typically had smaller nuclear sizes than
.
CC-BY-NC-ND 4.0 International license
available under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (which
this version posted April 27, 2021.
;
https://doi.org/10.1101/2021.04.26.441547
doi:
bioRxiv preprint
excitatory neurons (Fig. 2D). In contrast, in the 25
-
kb resolu
tion data at the targeted Mb regions, the
scaling relationship differed depending on the chromosomal regions and cell types (Fig. 2G and fig. S3D).
For example, the targeted regions in chromosome 7 and chromosome 17 are more dispersed in neurons
compared t
o glial cells, whereas other targeted regions in chromosome 5 and chromosome 18 have
almost identical scaling relationships among neurons and glial cells. Overall, regions with higher gene
density tend to have less compact spatial organization (fig. S3E),
possibly owing to different underlying
epigenetic states
(
32
,
67
)
. To gain a more integrative picture of nuclear organization, we ne
ed to examine
the interactions between DNA and nuclear bodies beyond characterizing individual components.
DNA loci display unique and shared chromatin profiles across different cell types
To characterize the spatial association between DNA loci, chromati
n marks and nuclear bodies, we
calculated the fraction of time that each DNA locus is associated with each chromatin mark. Because many
chromatin marks form discrete regions within the nucleus and IF images are diffraction limited, we
determine the fractio
n of time each DNA locus is within 300 nm from the exterior of each mark
(
32
,
55
)
(Fig. 3A and 3B and fig. S4A). This imaging
-
based approach of computing “chromatin profiles”
demonstrated a high correlation
with sequencing
-
based bulk measurements such as ChIP
-
seq and SPRITE
at 1
-
Mb resolution in mouse ES cells
(
32
)
.
Whereas some chromatin profiles were highly concordant across cell types, others showed specific
patterns that varied between cell types (Fig. 3A and 3B and fig. S4A). The DNA loci associated with nuclear
speckle markers (SF3a66 and
Malat1) were highly correlated among different cell types at 1
-
Mb
resolution, with a Pearson correlation coefficient of >=0.83 even including mouse ES cells (Fig. 3B). This
highly conserved DNA
-
nuclear speckle spatial association has been reported recently
with various cell
lines using TSA
-
seq
(
68
)
. Interestingly, the chromatin profiles for n
uclear speckles were highly correlated
with gene densities at 1
-
Mb resolution (Fig. 3A and 3B and fig. S4A), suggesting a robust relationship
between spatial genome organization around nuclear speckles and underlying genomic sequences.
On the other hand,
the associations between DNA loci and DAPI
-
rich constitutive heterochromatin
showed lower correlation between neurons and astrocytes (Fig. 3A and 3B), even though the associations
were typically enriched in the centromere proximal genomic loci in all cell
types (Fig. 3A and fig. S4A and
S4B). Furthermore, the relationships between DNA loci and H3K27me3 were more distinct across cell
types (Fig. 3A) and showed lower correlation across cell types in the brain as well as with mouse ES cells
(Fig. 3B), reflecti
ng the underlying differences in DNA loci
-
histone modification globule associations across
cell types.
We observed similar cell
-
type
-
dependent association of chromosomes with the nucleolus in the mouse
brain cortex (Fig. 3A). The spatial
proximities between ITS1 non
-
coding RNA and DNA loci in Pvalb
inhibitory neurons and astrocytes displayed a relatively low correlation, with a Pearson correlation
coefficient of 0.37 (Fig. 3B). We observed some cells in which the 45S ribosomal DNA (rDNA)
-
c
ontaining
chromosomes 15 and 19
(
69
)
were not in physical proximity to the nucleoli but
were close to DAPI
-
rich
heterochromatin regions (fig. S4C), possibly due to rDNA silencing
(
70
)
. This rDNA silencing could lead to
the cell
-
type specific nucleolar association of genomic loci (Fig. 3A and 3B and fig. S4A). To confirm this,
we performed imaging of rDNA loci by DNA FISH and found that different fractions of rDNA loci were
.
CC-BY-NC-ND 4.0 International license
available under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (which
this version posted April 27, 2021.
;
https://doi.org/10.1101/2021.04.26.441547
doi:
bioRxiv preprint
associated with the nucleolus and DAPI
-
rich heterochromatin regions among neurons, astrocytes, mouse
ES cells and cultured fibroblasts (fig. S4D to F), lending support to the notion that nucleolar organizer
regions can be stably silenced
(
70
)
in a tissue
-
specific fashion.
Cell
-
type specific fixed loci anchor chromosomes to nuclear bodies in
single cells
To further gain insight of single
-
cell nuclear architecture across cell types, we defined DNA loci that were
consistently associated with a particular chromatin mark or nuclear body in each cell type to be “fixed
loci”
(
45
)
(Fig. 3C). We previously observed that these fixed loci for each IF marker consistently appear on
the exterior
of the respective marker in single mouse ES cells
(
32
)
. In the mouse brain, we observed
similar
associations of the fixed loci with the exterior of nuclear bodies and chromatin marks in single cells (Fig.
3C), despite the differences in the morphological features and arrangement of nuclear bodies in individual
neuronal and glial cells. As ex
amples, fixed loci for SF3a66, DAPI, and H3K27me3 are consistently observed
on the exterior of nuclear speckles, heterochromatin bodies and H3K27me3 globules in single neurons
and glial cells (Fig. 3C).
Fixed loci for different chromatin marks are distribu
ted across the genome such that each of the
chromosomes has distinct patterns of fixed loci in each cell type (Fig. 4A to D and fig. S5A). These fixed
loci can constrain the nuclear organization of chromosomes by their association to the nuclear bodies or
chromatin marks in individual nuclei (Fig. 4C and 4D and fig. S5A). For example, chromosomes 7 and 17
have fixed loci for nuclear speckles (SF3a66) and heterochromatic bodies (DAPI) in excitatory neurons,
inhibitory neurons and astrocytes (Fig. 4D), and bo
th chromosomes straddled these nuclear bodies in
single cells of all three cell types. Similarly, chromosome 8 has nuclear speckles and H3K27me3 fixed
points in all three cell types, and spanned those nuclear globules in single cells. A small number of loc
i
were associated with two nuclear bodies (orange dots) and were observed near both nuclear bodies in
single cells (Fig. 4C and 4D). These features were observed across different cell types for other
chromosomes (fig. S5A). Thus, despite the differences in
the arrangement of nuclear bodies in individual
cells and the different fixed point patterns on the chromosomes in each cell type (Fig. 4D and fig. S5A),
the association between DNA loci and nuclear bodies are consistent across single cells of each cell t
ype.
These findings in the mouse brain extend our previous work in mouse ES cells
(
32
)
t
o show that chromatin
fixed loci serve as organizational invariants in the nuclei of single cells across cell types, despite the highly
variable appearance of individual chromosomes and nuclear bodies in individual cells.
Cell
-
type specific nuclear bodies
determine inter
-
chromosomal proximity and radial positioning
The cell
-
type specific inter
-
chromosomal interaction and radial positioning of chromosomes in the nucleus
were previously characterized by chromosome paint
(
71
–
73
)
and single
-
cell chromosome conformation
capture
(
18
,
19
)
. H
owever, it remains unclear what determines the distinct chromosomal features across
cell types. Similarly, although the arrangement of nuclear bodies, such as heterochromatin bodies and
nucleoli, has been shown to be cell
-
type specific as well as dynamic e
ven within a cell type during
development
(
61
,
74
)
, it remains unclear how those differences in nuclear body arrangements relate to
3D chromosome organization at the single
-
cell resolution. Thus, we exa
mined our integrated spatial multi
-
modal datasets to test the hypothesis that nuclear body organization and the fixed point association in
.
CC-BY-NC-ND 4.0 International license
available under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (which
this version posted April 27, 2021.
;
https://doi.org/10.1101/2021.04.26.441547
doi:
bioRxiv preprint
different cell types accounts for the cell
-
type specific inter
-
chromosomal interaction and radial
chromosomal positio
ning.
We first characterized the nuclear bodies in each cell type (Fig. 5A to C and fig. S5B). Compared to
excitatory neurons and astrocytes, Pvalb inhibitory neurons had fewer but larger heterochromatic bodies
and nucleoli (Fig. 5B and fig. S5B). Heteroc
hromatic bodies were often localized close to the center of the
nucleus in Pvalb inhibitory neurons, but more distributed in other cell types (Fig. 5C). Nuclear speckles
appeared to be more dispersed in cells with preferred localization in the nuclear inte
rior (Fig. 5A and 5C).
Furthermore, H3K27me3 globules tend to localize more at the nuclear interior in astrocytes as compared
to neurons (Fig. 5A and 5C).
Next, we characterized radial positioning of individual chromosomes and individual loci from nuclear
interior to exterior, and we observed their changes across cell types (Fig. 5D and fig. S5C and S5D).
Consistent with previous reports
(
71
–
73
)
, we observed the correlation
between radial positioning and
gene density as well as chromosome size across cell types (fig. S5E and S5F). Interestingly, gene
-
dense
chromosomes (e.g. Chr7, 11, 17 and 19) tend to have different radial positions across different cell types
(Fig. 5D and f
ig. S5C and S5F), which was also observed during post
-
natal brain development
(
19
)
. For
example, gene dense chromosome 7 tends to localize close to the nuclear center in excitatory neurons,
but not in glial cells (Fig. 5D and fig. S5C). At the same time, we observed a cell
-
type dependence in the
average inter
-
chromosome spatial distances (Fig
. 5E, top). Notably, those pairwise inter
-
chromosome
distance maps agree with averaged radial positioning of pairs of chromosomes, such that chromosomes
in the nuclear interior tend to be spatially close to each other (Fig. 5E bottom, and 5F and fig. S5G).
These
features were observed across cell types, suggesting a common principle in chromosome organization.
We further investigated how the cell
-
type dependent changes in nuclear bodies, chromosome radial
positioning and inter
-
chromosome arrangement are con
nected in single cells. We observed that pairs of
inter
-
chromosomal loci assigned as fixed points at nuclear bodies tend to be spatially closer to each other
than the other non
-
fixed point pairs in neurons and astrocytes (Fig. 5G). These fixed points can i
nfluence
the cell type
-
specific arrangement of the chromosomes. For example, chromosomes 11 and 19 have many
H3K27me3 fixed points in astrocytes (fig. S5A) and H3K27me3 globules tend to be in the interior of the
astrocyte nuclei (Fig. 5C). Thus, chromosome
s 11 and 19 tend to be observed near the interior and interact
with other chromosomes in astrocytes (Fig. 5E and F), but less in neurons. Similarly, chromosome 17
contains many heterochromatin fixed points in neurons (Fig. 4D) and heterochromatic bodies te
nd to be
in the nuclear interior in Pvalb inhibitory neurons (Fig. 5C). Consequently, chromosome 17 tends to be
radially positioned near the nuclear interior (Fig. 5D). In addition, because chromosomes 7 and 17 are
gene dense and contain many fixed points
to nuclear speckles which tends to be positioned in the nuclear
interior, both chromosomes are observed in the nuclear interior of neurons in bulk and single cells (Fig.
5D to F and 5H). Thus, the complexity in the global organization of chromosomes in div
erse single cells in
the brain, such as the cell type
-
dependence in radial positioning and inter
-
chromosomal distances, can be
dimensionally reduced and captured in the different nuclear body arrangements and fixed point
chromatin profiles.
Lastly, we comp
ared the population
-
averaged and single
-
cell picture of radial organization of
chromosomes and nuclear bodies. At the population
-
averaged level, the radial positioning of
.
CC-BY-NC-ND 4.0 International license
available under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (which
this version posted April 27, 2021.
;
https://doi.org/10.1101/2021.04.26.441547
doi:
bioRxiv preprint
chromosomes and genomic loci to nuclear interior appear to be correlated with nuclear
speckle
association (Fig. 3D), consistent with other bulk analysis
(
75
)
. However, visualization of the genomic loci
as a function of gene density or expression levels in single cells shows that the radial positioning effect is
highly variable in each nucleus (Fig. 3E). Th
us, while speckle associations for gene dense regions are highly
consistent across cells (Fig. 3A and 3B), because the arrangement of nuclear speckles is heterogeneous in
single cells with a weak propensity for nuclear center (Fig. 5C), the positioning of
the gene
-
dense speckle
associated loci do not show strong radial gradient from nuclear interior to nuclear exterior in single cells,
which directly supports the refined model of radial nuclear organization
(
75
)
.
Domain boundaries are variable in single cells.
The high
-
resolution DNA seqFISH+ data covering the genomic regions at 25
-
kb resolu
tion enables us to
further examine the domain organization of chromosomes in single cells at sub
-
megabase scales.
Analyses based on bulk
-
averaged chromosome conformation capture data at the sub
-
megabase
resolution suggest that chromosomes are organized int
o topologically associating domains (TADs) with
clear insulation boundaries
(
6
–
8
)
. The TADs appear largely unchanged across species
(
6
,
76
)
and their
boundaries preferentially reside at CCCTC
-
binding factor (CTCF)
-
and cohesin
-
binding sites
(
6
,
23
)
.
Single
-
cell chromosome conformation capture measurements
confirmed preferential contacts within
TADs in single cells
(
12
)
. In addition, imaging e
xperiments confirmed that TAD
-
like domain structures
exist in single cells and depletion of cohesin resulted in shifting boundaries of the examined TADs
stochastically across single cells
(
23
)
. However, it is unexplored how single chromosome domain
structures are organized across genomic regions with different bulk TAD characteristics.
To sys
tematically investigate single chromosome domain structures, we first determined whether there
are subpopulations of chromosomes with distinct configurations that differ from the bulk averaged
configurations in excitatory neurons. We clustered the single c
hromosome pairwise spatial distance data
with 25
-
kb resolution using principal component analysis (PCA) and visualized by UMAP (Fig. 6A to C).
The chromosome 3 region (7.7
-
9.3 Mb) displayed three domains in the bulk data (Fig. 6D, top).
However, analysis o
f subpopulation of chromosomes that were clustered together (Fig. 6C) showed
multiple configurations with different pairwise spatial associations and domain boundaries (Fig. 6D,
bottom, and 6E, top, and fig. S6A). We further examined single chromosome stru
ctures in each
structural cluster, and found that instead of all three major domains being present in single cells, in
many cells, only a subset of the domains appeared (Fig. 6E, bottom). In the bulk measurements, domain
boundaries corresponded to CTCF and
cohesin (RAD21, cohesin subunit) binding sites
(
77
)
(Fig. 6D), but
chromosomes in singl
e cells appeared to stochastically form domains from a subset of these sites (Fig.
6E, bottom). These structures can be observed in single chromosome pairwise spatial distance maps of
genomic loci as well as direct visualization of the chromosomal regions
(Fig. 6E, bottom). In addition,
even regions that did not show clear domain structures in the bulk averaged data contained domain
-
like
structures at the single cell level. For example, chromosome 7 (44.6
-
46.1 Mb) and chromosome 11
(97.4
-
98.9 Mb) regions co
ntained several domains in single cells even when no clear domains were
visible in the bulk level at the examined spatial scale (fig. S6B to E). Furthermore, there were
chromosomal regions that exhibited more deterministic boundaries, such as chromosome 6
(49.4
-
50.9
Mb) region (fig. S7A and S7B). However, single chromosome subpopulations showed heterogeneity in
.
CC-BY-NC-ND 4.0 International license
available under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (which
this version posted April 27, 2021.
;
https://doi.org/10.1101/2021.04.26.441547
doi:
bioRxiv preprint
the spatial proximity of inter
-
domain organization, as seen by the off
-
diagonal elements (fig. S7B, top
and middle) and the 3D reconstruction of the
chromosome structures (fig. S7B, bottom). Similarly, the
histone gene cluster in chromosome 13 (21.5
-
23.8 Mb), known to form a higher
-
order chromosome
structure in a population of cells
(
10
)
, showed heterogeneous higher
-
order chromosome organization in
individual cells (fig. S7C and S7D).
These results demonstrate that even in cells expressi
ng CTCF and cohesin, there are highly variable
domain boundaries and inter
-
and intra
-
domain interactions that can be obscured with ensemble
-
averaged measurements. Recent super
-
resolution imaging studies
(
33
,
34
)
have shown that CTCF and
cohesin have differential effects on chromosomal domain formation in single cells, where CTCF
preferentially promotes intra
-
domain interactions whereas cohesin promotes stochastic intermingling of
inter
-
domains. In addit
ion, cohesin appears to alter the instantaneous transcription activities of
boundary
-
proximal genes
(
33
)
. It is possible that the variabilities of single
-
cell domains we observed (Fig.
6 and fig. S6 and S7) are mediated by stochastic combinatorial
binding of the architectural proteins
(
78
)
such as CTCF and cohesin at individual chrom
osomes and may lead to instantaneous transcriptional
differences at those domain boundaries in single cells.
Active and inactive X chromosome organization in single cells.
Lastly, we examined the differences of chromatin states and chromosome conformation
s between the
active X chromosome (Xa) and the inactive X chromosome (Xi) from the female mouse brain cortex. X
-
chromosome inactivation in female mammalian cells has been extensively studied as a model system for
chromosome organization
(
59
)
. The imaging
-
based genomics data can straightforwardly distinguish the
Xa and Xi based on their mutual
ly exclusive associations with Xist RNA, a long noncoding RNA that is
specifically expressed from and associates with the Xi
(
59
)
(fig. S8A and S8B). As expected, the Xi showed
enrichment of repressive mH2A1 and H3K27me3 marks
(
79
,
80
)
, high nucleolar association
(
80
)
,
depletion of active chromatin mark
(
81
)
H3K4me2, and condensed LINE1 DNA elements
(
82
)
(fig. S8A
and S8B).
To gain insight into the structural differences between the Xa and Xi, we calculated the median spatial
d
istance and spatial proximity maps for the whole X chromosome at 1
-
Mb resolution as well as for a
targeted region (75.3
-
77.0 Mb) at 25
-
kb resolution in a cell
-
type
-
specific fashion (fig. S8C to F). We
observed that the Xa and Xi have distinct median distan
ces between pairs of loci at different genomics
length scales and in different cell types (fig. S8G and S8H). In particular, we observed that the Xi is
organized into two mega
-
domains separated at the macrosatellite DXZ4 locus at the whole chromosome
scale
, consistent with the literature
(
83
)
. Interestingly, although the Xi is more compact th
an the Xa
globally at the larger scale of tens of megabases (fig. S8C and S8G)
(
22
,
84
)
, we found the Xa appears to
be more structured and compact at the Mb or below length scales based on the populatio
n
-
averaged
spatial distance quantification (fig. S8C, D, G and H).
To further resolve the observed bulk structural differences between the Xa and Xi, we examined the Xa
and Xi conformation systematically at the single
-
cell level by applying the PCA
-
based
approach used for
the autosomal regions (Fig. 6A to E and fig. S6 and S7). Interestingly, both the Xa and Xi have
heterogeneous domain structures in individual cells (Fig. 6F and 6G and fig. S9). Even the region of the Xi
.
CC-BY-NC-ND 4.0 International license
available under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (which
this version posted April 27, 2021.
;
https://doi.org/10.1101/2021.04.26.441547
doi:
bioRxiv preprint
that appears unstructured in the b
ulk data from both DNA seqFISH+ (Fig. 6F and fig. S8D) and allele
-
specific Hi
-
C studies
(
83
,
85
)
appeared to adopt discrete domains in subsets of cells (Fig. 6G). We found
that similar domain subcluster
s can be used in both the Xa and Xi, but with different relative frequencies
for specific chromosome conformation (Fig. 6H), which leads to different average conformations for the
Xa and Xi (Fig. 6F and fig. S8D). Taken together, although the Xa and Xi sho
w very different chromatin
states and ensemble
-
averaged chromosome conformation, they can share similar underlying single
-
cell
domain structures with different relative conformational preferences at the finer scale, which has been
obscured with ensemble
-
av
eraged bulk measurements.
Discussion
Our work demonstrates cell type
-
dependent and
-
independent features in nuclear organization across
thousands of single cells in the mouse cerebral cortex, derived by integrated spatial genomics tools to
measure RNA,
DNA and chromatin marks. In particular, we examined nuclear morphologies, global
chromatin states, DNA
-
nuclear body associations, radial nuclear organization, as well as chromosome
and domain structures in transcriptionally defined cell types. Existing mic
roscopy and sequencing
-
based
datasets showed high concordance with our results at each modality, supporting the robustness and
accuracy of our approach and allowing the exploration of multimodal nuclear features in tissue sections.
At the global level, our
observations indicate that the chromosome organization in a cell reflects cell
-
type specific nuclear body arrangements. We observed that each chromosome contains a unique
pattern of fixed points associated with each nuclear body and chromatin mark such th
at the DNA loci
are reliably located on the exterior of the nuclear bodies in single cells at the scale of 1 Mb. Some of the
fixed points are cell type
-
dependent, whereas others, such as nuclear speckle
-
associated loci
(
68
)
, are
largely cell type
-
independent. Because nuclear body morphologies are different in different cell types,
these fixed
points lead to distinct organizations of the nucleus in each cell type. For example, because
most Pvalb inhibitory neurons have a large central heterochromatic globule, chromosomes 7 and 17
with fixed points to heterochromatin are organized around this ce
ntral hub. These chromosomes are
often found in the nuclear interior and interact with many other chromosomes. We had observed both
nuclear bodies and fixed points in mouse ES cells
(
32
)
, and now show that this principle operates in
tissues to drive cell type
-
specific genome organization. Similar observations of radial chromosome and
nuclear
body reorganization during brain and retinal development
(
18
,
19
,
61
,
74
)
suggest the same
principle c
ould also be applied to the developmental processes.
At the 25
-
kb scale organization of chromosomes, the single
-
cell resolution of the DNA seqFISH+ data
explains that nested structures and regions with ambiguous boundaries that are often observed in bulk
H
i
-
C contact maps
(
86
)
are due to different domains appearing in individual cells, possib
ly because
subsets of CTCF and cohesin sites are stochastically used to insulate individual chromosomes. In line
with this observation, single
-
cell domain structures were previously observed even in cohesin depleted
cells where domain structures were aboli
shed at the population
-
averaged level, suggesting the
population level domains are formed due to preferential domain boundary positions
(
23
)
. Our
systematic single
-
cell analysis further showed the prevalence of clear single
-
cell domains even in regions
which lack ensemble domain bou
ndaries such as the Xi region
(
83
,
85
)
, demonstrating the importance of
.
CC-BY-NC-ND 4.0 International license
available under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (which
this version posted April 27, 2021.
;
https://doi.org/10.1101/2021.04.26.441547
doi:
bioRxiv preprint
studying chromosome structures in individual cells to better interpret the organizational principles of 3D
genome architecture.
T
he robust demonstration of integrated spatial genomics in tissues indicates the same approach can be
applied to a diverse range of biological systems to further explore the diversity and invariant nuclear
architectures in single cells. At the global scale
of nuclear organization, it is still unclear how the distinct
nuclear body and chromosome arrangements as well as their associations arise in the first place in
different cell types. In addition, given the prevalence of the diverse single
-
cell domain struc
tures that
can differ from ensemble
-
averaged TADs, it would be critical to study both single
-
cell domain structures
and instantaneous transcriptional activity in each domain from the same single cells and understand
their relationships at a fine spatiotemp
oral resolution. We anticipate addressing these questions in
future studies by genome
-
scale chromosome imaging together with transcriptome
-
scale profiling
(
53
,
87
)
and protein imaging as well as by “tra
ck first and identify later” live
-
cell approaches
(
88
,
89
)
with
multiplexed chromosome labeling
(
90
,
91
)
.
References and Notes
1.
M. J. Rowley, V. G. Corces, Organizational principles of 3D genome architecture.
Nat. Rev. Genet.
19
, 789
–
800 (2018).
2.
B. van Steensel, E. E. M. Furlong, The role of transcription in shaping the spatial organization of the
genome.
Nat. Rev. Mol. Cell Biol.
20
, 327
–
337 (
2019).
3.
T. Misteli, The Self
-
Organizing Genome: Principles of Genome Architecture and Function.
Cell
.
183
,
28
–
45 (2020).
4.
J. Dekker, A. S. Belmont, M. Guttman, V. O. Leshyk, J. T. Lis, S. Lomvardas, L. A. Mirny, C. C. O’Shea,
P. J. Park, B. Ren, J. C. R. Politz, J. Shendure, S. Zhong, 4D Nucleome Network, The 4D nucleome
project.
Nature
.
549
, 219
–
226 (2017).
5.
R. Kempfer, A. Pombo, Methods for mapping 3D chromosome architecture.
Nat. Rev. Genet.
21
,
207
–
226 (2020).
6.
J. R. Dixon, S. Selvaraj, F. Yue, A. Kim, Y. Li, Y. Shen, M. Hu, J. S. Liu, B. Ren, Topological domains in
mammalian gen
omes identified by analysis of chromatin interactions.
Nature
.
485
, 376
–
380 (2012).
7.
E. P. Nora, B. R. Lajoie, E. G. Schulz, L. Giorgetti, I. Okamoto, N. Servant, T. Piolot, N. L. van Berkum,
J. Meisig, J. Sedat, J. Gribnau, E. Barill
ot, N. Blüthgen, J. Dekker, E. Heard, Spatial partitioning of the
regulatory landscape of the X
-
inactivation centre.
Nature
.
485
, 381
–
385 (2012).
8.
T. Sexton, E. Yaffe, E. Kenigsberg, F. Bantignies, B. Leblanc, M. Hoichman, H. Parrinello, A. Tanay, G.
Cavalli, Three
-
dimensional folding and functional organization principles of the Drosophila genome.
Cell
.
148
, 458
–
472 (2012).
9.
R.
A. Beagrie, A. Scialdone, M. Schueler, D. C. A. Kraemer, M. Chotalia, S. Q. Xie, M. Barbieri, I. de
Santiago, L.
-
M. Lavitas, M. R. Branco, J. Fraser, J. Dostie, L. Game, N. Dillon, P. A. W. Edwards, M.
.
CC-BY-NC-ND 4.0 International license
available under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (which
this version posted April 27, 2021.
;
https://doi.org/10.1101/2021.04.26.441547
doi:
bioRxiv preprint