NUCLEAR ARCHITECTURE
Single-cell nuclear architecture across cell types in
the mouse brain
Yodai Takei^1 , Shiwei Zheng^2 , Jina Yun^1 , Sheel Shah^1 , Nico Pierson^1 , Jonathan White^1 ,
Simone Schindler^1 , Carsten H. Tischbirek^1 , Guo-Cheng Yuan^2 , Long Cai^1 *
Diverse cell types in tissues have distinct gene expression programs, chromatin states, and nuclear
architectures. To correlate such multimodal information across thousands of single cells in mouse brain
tissue sections, we use integrated spatial genomics, imaging thousands of genomic loci along with
RNAs and epigenetic markers simultaneously in individual cells. We reveal that cell typeÐspecific
association and scaffolding of DNA loci around nuclear bodies organize the nuclear architecture and
correlate with differential expression levels in different cell types. At the submegabase level, active and
inactive X chromosomes access similar domain structures in single cells despite distinct epigenetic
and expression states. This work represents a major step forward in linking single-cell three-dimensional
nuclear architecture, gene expression, and epigenetic modifications in a native tissue context.
A
key feature of modern biology is enabl-
ing measurement of many parameters
in the same cells to arrive at accurate
low-dimensional representations of bi-
ological systems ( 1 , 2 ). In the case of
three-dimensional (3D) organization of the
genome, which is critical for many cellular
processes, from regulating gene expression to
establishing cellular identity ( 3 ), it is crucial to
integrate chromosome structure measurements
( 4 , 5 ) with others that capture transcriptional
states ( 6 , 7 ), chromatin states ( 8 ), and nuclear
bodies ( 9 , 10 ). However, multimodal measure-
ments of chromosome structure, transcriptional
states, and epigenetic markers in the same cell
remain challenging with sequencing approaches
( 2 ). Imaging-based approaches, on the other
hand, allow direct integration of multimodal
measurements, including chromosome struc-
tures ( 6 , 7 , 11 – 17 ). We recently established an
integrated spatial genomics approach ( 17 ),
imaging thousands of genomic loci along
with transcriptional states and subnuclear
localization of histone modifications and nu-
clear bodies in single mouse embryonic stem
(ES) cells. This approach revealed determi-
nistic scaffolding of chromosomes across
multiple nuclear bodies and protein glob-
ules in individual cells ( 17 ). Nevertheless, it is
unknown to what extent principles for single-
cell nuclear organization extend to diverse cell
types in complex mammalian tissues.
Integrated spatial genomics in the brain
To investigate single-cell nuclear architecture
across cell types, we imaged sections of the
adult mouse cerebral cortex with 3660 DNA
loci, 76 cellular RNAs, and 8 chromatin marks
and nuclear bodies over 125 rounds of hybrid-
izations and imaging while maintaining tissue
integrity (see methods in the supplementary
materials) ( 18 ) (Fig. 1, A and B; figs. S1 to S4;
and tables S1 to S4). Specifically, we used DNA
sequential fluorescence in situ hybridization
(DNA seqFISH+) to image 2460 loci at ~1-Mb
resolution across the genome and an additional
1200 loci for at least a 1.5-Mb region in each
chromosome at 25-kb resolution ( 17 ) (Fig. 1,
C to E; fig. S1, B and C; and table S1). We de-
tected 2813.0 ± 1334.0 (median ± SD) spots per
cell in total with low false positives in 2762 cells
from three biological replicates (Fig. 1, F and
G, and fig. S2, A and H), corresponding to an
estimated detection efficiency of at least 38.4 ±
18.2% (median ± SD) (methods). The DNA
seqFISH+ data agreed well with sequencing-
based bulk Hi-C data ( 19 ) and among the
three biological replicates (Fig. 1H and fig.
S3), demonstrating the robustness of DNA
seqFISH+ for mapping 3D chromosome struc-
tures in tissue samples.
We clustered the RNA seqFISH data and
obtained nine major cell type clusters within
the cerebral cortex that matched well with cell
types identified from single-cell RNA sequenc-
ing ( 20 ), including excitatory neurons; four
subclasses of inhibitory neurons expressing
parvalbumin (Pvalb), vasoactive intestinal
polypeptide (Vip), neuron-derived neuro-
trophic factor (Ndnf), or somatostatin (Sst);
three types of glial cells (astrocytes, microglia
and oligodendrocyte precursor cells, and oligo-
dendrocytes); and endothelial cells (Fig. 1I and
fig. S1, C to F).
In addition to the genome and RNA imag-
ing, we used sequential immunofluorescence
(IF) with oligonucleotide-conjugated primary
antibodies (methods) ( 17 ) to detect six his-
tone modifications or variants (H3K4me2,
H3K27me2, H3K27me3, H3K9me3, H4K20me3,
and mH2A1), nuclear speckles [splicing factor3A subunit 2 (SF3a66)], and the methyl-CpG-
binding protein MeCP2 (Fig. 2A, figs. S4 and
S5A, and table S3). Additionally, we mea-
sured three noncoding RNAs that mark the
inactive X chromosome (Xi, Xist), the nucleolus
[internal transcribed spacer region 1 (ITS1)],
nuclear speckles [metastasis-associated lung
adenocarcinoma transcript 1 (Malat1)], as
well as five repetitive regions [long inter-
spersed element 1 (LINE1), short interspersed
nuclear element B1 (SINEB1), major satel-
lite (MajSat), minor satellite (MinSat), and
Telomere] ( 21 , 22 ) (figs. S4 and S5A). The lo-
calization patterns of individual antibodies,
RNA, and DNA markers in our data were
consistent with literature ( 21 , 22 ) (fig. S4),
enabling the multimodal investigation of
single-cell nuclear architecture across cell
types with unprecedented detail.Distinct nuclear features across cell types in
the brain
The nine major cell types displayed clear dif-
ferences in the global levels of both repressive
marks (MeCP2, H3K27me3, and mH2A1) and
an active mark (H3K4me2) (Fig. 2A and figs.
S5 and S6). Clustering of single cells using the
multiplexed IF data was able to distinguish
the same nine cell types identified by RNA
seqFISH (Fig. 2B), supporting a strong corre-
lation between global chromatin states and
transcriptional states. Notably, even the 4′,6-
diamidino-2-phenylindole (DAPI) features
alone were sufficient to separate the major
cell types in the cortical areas of the mouse
brain (Fig. 2, A, C, and D, and fig. S5C).
The spatial versus genomic distance scal-
ing was distinct in different cell types in the
1-Mb-resolution data (Fig. 2, E and F, and
figs. S7 and S8), which cannot be explained
just by nuclear size differences (fig. S5C). In
contrast, in the 25-kb-resolution data at the
targeted megabase regions, the scaling rela-
tionship differed depending on the chromo-
somal regions and cell types (Fig. 2, E and F,
and figs. S7 and S8). To obtain a more inte-
grative picture of nuclear organization, we
need to examine the association between DNA
and nuclear bodies beyond characterizing in-
dividual components.Cell typeÐspecific chromatin profiles can
relate to differential transcriptional states
To characterize the spatial association be-
tween DNA loci, chromatin marks, and nu-
clear bodies, we calculated the fraction of
instances in which each DNA locus is detected
within 300 nm from the exterior of each mark
( 17 , 23 )(Fig.3,AtoD,andfigs.S9toS12).
With these imaging-based“chromatin profiles”
using super-resolved DNA seqFISH+ spots
and diffraction-limited IF images, we have
previously showed a high correlation with
sequencing-based bulk measurements such586 29 OCTOBER 2021•VOL 374 ISSUE 6567 science.orgSCIENCE
(^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 Sinai, New York, NY, USA.
*Corresponding author. Email: [email protected]
RESEARCH | RESEARCH ARTICLES