FOCUS | technologY featuRe

FOCUS | technology feature

Method of the Year: spatially resolved
transcriptomics
Nature Methods has crowned spatially resolved transcriptomics Method of the Year 2020.

Vivien Marx

I
f a researcher is making a smoothie,
it might be snack time. Or it could be
the moment to prepare a sample for
bulk RNA sequencing, in which tissue
is homogenized and analyzed to yield
averaged gene expression from the mRNAs
in a tissue’s cells — its transcriptome.
When single-cell RNA sequencing
(scRNA-seq) was developed, it brought a
more fine-grained assessment of each cell’s
transcriptome. In scRNA-seq, researchers
dissociate cells from tissue to, for example,
discern cell types on the basis of gene
expression. Working with single cells is
more like digging into a fruit salad than a
smoothie, says Hongkui Zeng, who directs
the Allen Institute for Brain Science. Now,
with spatially resolved transcriptomic
methods, scientists can get transcriptomic
data and know the positional context
of those cells in a tissue1–3. “Fruit tart is
spatial transcriptomics,” says Bosiljka
Tasic, an Allen Institute researcher who
was interviewed jointly with Zeng. “You Starry skies invite space exploration. In transcriptomics, spatial resolution opens up new worlds too.
know exactly where each piece of fruit is Credit: bjdlzx/Getty Images
and what is the relationship of each piece
of fruit to the other,” she says. Given how
much one can learn from single cells and
spatially resolved cells, the ‘smoothie’ says Jay Shendure from the University says Alexander van Oudenaarden, a physicist
approach is fast becoming passé, she says. of Washington. He is most excited about turned biologist who directs the Hubrecht
It’s not yet routine for spatial analysis to ways to apply these methods to obtain Institute in Utrecht. He and his lab
deliver transcriptome-wide information of spatially resolved atlases of development have combined single-cell and spatial
all single cells, but the field is fast moving that combine cell states, histories and transcriptomics to compare gene expression
in that single-cell direction. Biology, says fates. “It’s a very active field,” says Zeng, in gastruloids and mouse embryos.
Sten Linnarsson, a Karolinska Institute with techniques that open up new kinds The research focus on single cells has
biologist, “is all about context.” Spatial of experimental opportunities. Harvard nineteenth-century roots, such as the
single-cell transcriptomics is the next wave University’s Xiaowei Zhuang, a Howard observation by Rudolf Virchow that disease
after single-cell analysis and, he says, will be Hughes Medical Institute investigator, has begins in cells, not whole organisms. A
particularly useful to labs studying human always lived in the world of imaging. She decade ago, as van Oudenaarden and others
disease, which often starts with single cells and her lab developed a super-resolution peered into microscopes to study cells as
and spreads spatially. “And if I can dream, microscopy method, stochastic optical they develop, grow and migrate in growing
in a few years we’ll have spatiotemporal reconstruction microscopy (STORM). organisms, “we were always a little jealous
single-cell ’omics in living tissues,” which When single-cell genomics emerged, she of the sequencing world.” He and his team
brings single-cell developments full circle. began thinking, dreaming at night, and could make time-lapse movies and capture
working on a way to bring together imaging complex processes, but only label gene
Welcome, family and genomics to “get the best sides of both expression with a handful of fluorescent
Labs keen on quantifying mRNAs in their worlds.” Methods, especially those that proteins. Genomics labs could measure
spatial context have embraced the family began to emerge from around 2012 onward thousands of genes. To do this with tissues,
of methods that deliver spatially resolved for capturing spatial information alongside he says, they “put them in a blender.” Now
transcriptomic data in neuroscience, cancer single-cell sequence, have been melding it’s become possible to assess many single
research or developmental biology, among the previously separate worlds of imaging, cells and many genes and keep track of the
other fields. “I think the future is bright,” sequencing and molecular characterization, cells’ tissue context, their spatial attributes.

Nature Methods | VOL 18 | January 2021 | 9–14 | www.nature.com/naturemethods 9

technologY featuRe | FOCUS
technology feature | FOCUS

That, says Ståhl, gives you something to

Penk Doc2g Kctd12 image right where the synthesis of cDNA
mRNA

Tissue section from the captured mRNA takes place. “You

get a very nice fluorescent footprint of where
everything went,“ he says. “That was, for
AAAAAAAAA
TTTTTTTTT

us, a gigantic stepping stone into getting the

rest to work,” he says. This stepping stone
ID582 ID583 ID584 ID585 ID586 ID587 ID588 ID589 ID590
convinced many the method had potential
— even the “non-believers,” says Salmén,
laughing. Methods development means
plenty of failure, he says; in this case, four
Researchers in Sweden developed an approach in which fixed, stained tissue is imaged, permeabilized to five years of failure. The method was
and the mRNAs attach to an array of barcoded oligos. The RNAs are reverse-transcribed; the cDNAs commercialized through a spinout company,
are sequenced and yield spatially resolved transcriptomic information. Credit: Adapted with permission Spatial Transcriptomics, which was acquired
from ref. 4, AAAS by 10x Genomics in 2018 and led to the
product Visium.

“There are all kinds of smart ways to keep Next comes sequencing of these cDNAs. Other spatial travels
the spatial information,” he says. “It’s a very The positional barcodes provide spatially Imaging is inherently spatial and natively
exciting technology.” resolved transcriptomic information. The three-dimensional (3D), says California
As Joakim Lundeberg from KTH scientists worked hard on an array that Institute of Technology researcher Long
Royal Institute of Technology’s Science captures single cells. “We really wanted to Cai, whose career has also been shaped by
for Life Laboratory and colleagues go down to that level,” says Salmén, who was spatial methods. To get 3D information with
note3, spatial techniques can be divided interviewed jointly with Ståhl and is now a a microscope “you just do z sections,” he
into those that involve gene expression postdoctoral fellow in van Oudenaarden’s says. He recalls when his Caltech colleague
analysis on microdissected tissues and lab. The method’s resolution ended up being Barbara Wold mentioned to him how
those that involve in situ hybridization, 100 µm, which is tens of cells. Among their informative RNA sequencing could be. That
in situ sequencing, in situ capturing, and many technical challenges was that the was a decade ago, when second-generation
computational reconstruction of spatial mRNAs could diffuse in many directions, sequencers were nascent. “If you think about
data. Single-molecule fluorescence in situ which risked inaccurate spatial data or it, sequencing is actually single-molecule
hybridization (smFISH) is “the beginning mixed expression patterns, says Salmén. imaging in a box,” he says. A sequencer, “a
of the hybridization-based approaches” They developed ways to avoid that. “It was single-molecule microscope,” can generate
with spatial techniques, says Zeng. In this a long struggle,” says Ståhl but they and the 200-base-pair reads, which arise from
method, multiple oligonucleotides carry team managed, and Ståhl and Salmén share 200 rounds of chemical reactions and
fluorescent labels and bind to an RNA first authorship on the paper4. The paper and imaging. Genomics methods yield cellular
molecule. smFISH yields a quantitative the process of getting there — the melding of information, but to read out single-cell
mRNA readout with “a near 100% detection concepts and technology from microarrays, genomic information with imaging takes
sensitivity,” note researchers from the Allen imaging, sequencing, and bioinformatics more labels than smFISH. Cai applied
Institute, Science for Life Laboratory and analysis based on barcodes — has shaped super-resolution microscopy to cram
Karolinska Institute2. their science and their careers, says Ståhl. discernible fluorescent dots into an imaged
Patrik Ståhl from KTH explains that it Salmén’s key idea, says Ståhl, was setting cell. The dots can be tagged RNA, DNA
was around 2009 when Karolinska Institute up an initial reaction with fluorescent or protein. His lab’s methods seqFISH5
researcher Jonas Frisén set out to see how nucleotides to render visible where the cells’ and seqFISH+6 use sequential rounds of
his lab could reap more information from mRNAs meet the array’s surface probes. hybridization to barcode mRNAs with a set
the typical tissue slide, prepared with stains
that date back to the nineteenth century.
Frisén reached out to KTH colleague
Lundeberg, where Ståhl was a PhD student.
Fredrik Salmén was a master’s student at 2 µm
647 nm
KTH; Ståhl later joined the Frisén lab as 20 µm 100 nm
a postdoctoral fellow and Salmén became Sh3bp4
Cuedc2
a PhD student in Lundeberg’s lab. A
collaborative project in which Ståhl and Kcnd3
Pde8b
Fam26e
Salmén worked closely together resulted Afg3l2 Iqsec2
in a spatial analysis approach whereby Pcsk4 Cysltr1

fixed, stained tissue is imaged and then Ppp2r5d

Shank1
permeabilized. The released mRNAs move Cortex
and attach to an array beneath the tissue
with barcoded oligonucleotides, which
fixes them in the position they had in the
tissue. After reverse transcription the tissue
is enzymatically removed. What remains Spatial techniques help with atlas-building by localizing expressed genes. Here, seqFISH+ was used to
are spatially barcoded complementary measure 10,000 genes in mouse cortex. Credit: Cai lab, Caltech, I. Strazhnik; adapted with permission
DNA molecules attached to the oligo array. from ref. 6, Springer Nature.

10 Nature Methods | VOL 18 | January 2021 | 9–14 | www.nature.com/naturemethods

 FOCUS | technologY
|
featuRe
FOCUS technology feature

of FISH probes that carry a fluorophore. FISH. Molecular changes can define tumor
The transcripts are fixed in a tissue’s cells. types, but tumors can vary greatly within
With each round of hybridization, probes those classes. “Cell types aren’t so binary,”
are removed for the next hybridization she says. She wonders if flow cytometry
round with a different fluorophore. The has shaped cell typing to date. Looking at
sequence of fluorescence signals delivers samples with many markers for cells now
spatially resolved transcriptomic data. might yield a continuum of cell identities.
Instead of converting imaging information For this work, team science grows more
into sequencing readouts, “you can do it important to iterate on findings and develop
directly, in situ,” he says. He has founded a data analysis and visualization approaches to
start-up called Spatial Genomics to make ferret out computational artifacts. The Fertig
seqFISH commercially available. lab is part the National Cancer Institute’s
In his view, researchers will want Xiaowei Zhuang developed the super-resolution Informatics Technology for Cancer Research
single-cell resolution, which scRNA-seq microscopy method STORM. Single-cell genomics program to build the frameworks for
provides but tissue imaging, as of yet, does got her thinking about melding genomics and interpreting the high-dimensional ’omics
not. Resolution matters because in a region imaging. Credit: Harvard University data that will increasingly include spatial
of interest tissues can have a mix of cell types data. With more markers per cell, the high
but “a completely wrong picture” can emerge dimensionality of the data have increased
without adequate transcriptomic resolution. Drop-seq’s barcoding scheme. They started the challenges of data interpretation.
And if a method has low efficiency, some out with 50-micrometer beads and moved “I think that’s kind of the current challenge
signals can be missed, such as the mRNAs to 10-micrometer beads. Sam Rodriques in the field for spatial,” she says. For
for transcription factors expressed at ten in the Chen lab and Robert Stickels in the integrating imaging and single-cell data
copies or less. Cai sees as a weak point of Macosko lab figured out how to arrange a for spatial transcriptional analysis, she and
many current spatial methods that they bead monolayer as a 2D array and developed her team apply transfer learning to overlay
lead to something that “in the end you have a protocol to transfer RNAs to the beads, gene correlation signatures and perform
to put in a sequencer.” High-throughput which was “easier said than done,” says dimension reduction.
genomics has had a head start, but “for Macosko. Slide-seq’s output approaches Macosko foresees a phase of
almost everything sequencing can do now, that of single-cell analysis, which made it computational development around
there will be a spatial assay that will do the possible to use single-cell computational finding spatially variable genes or for
same kind of thing or better.” tools “as if the beads were single cells,” he finding trajectories that leverage the spatial
Evan Macosko, a physician-scientist says. They have run single-cell analysis information. Despite much effort, it remains
at the Broad Institute, has been keen on tools such as velocyto and Monocle. Among a big challenge, says Shendure, to manage
using genomics to make new kinds of other projects, they have used the method data and integrate across platforms. Each
measurements in cells and tissues. He to study development of the ocular lens experiment leads to vast amounts of data,
co-developed Drop-seq while a postdoctoral and embryonic cortex. Lower-resolution and researchers face decisions about what
fellow in the lab of Steve McCarroll at spatial techniques have too many cells per to save and what to discard and do not want
Harvard Medical School. With Drop-seq, pixel, which makes deconvolution more to deviate from the principles of open data
it takes a day to prepare thousands of difficult, he says. Since Slide-seq uses beads, sharing in genomics.
sequencing libraries from many individual “we can continue to make the beads smaller
cells. In a microfluidic device cells travel and resolution better.” The team has been Data juggling
through narrow channels and end up working on the method’s efficiency. Slide-seq At the Allen Institute, large-scale projects
encapsulated in a droplet with beads, v2 has mRNA capture efficiency that such as cell typing in the brain are as
each outfitted with a unique 12-base-pair approaches that of scRNA-seq technology, commonplace as juggling large datasets.
oligonucleotide barcode. Cells are lysed in he says. Efficiency matters, because for rare It has even surprised them, says the
the droplet and mRNAs are bar-coded and transcripts enough information is needed Allen’s Tasic, how quickly they can hit
reverse-transcribed into cDNAs that are then to confidently assign them to a particular the boundaries of standard analysis
sequenced. Together with Fei Chen, then a location. Discussions on commercializing when looking at single-cell and spatial
Broad Fellow, who co-invented expansion Slide-seq are in early stages, he says. transcriptomic data with tens of thousands
microscopy and who now has a lab at the of properties per cell. “The data matrices
Broad, Macosko developed Slide-seq7. Compute that space become unmanageable,” she says. They
The method draws on “DNA barcoding “The fact that it’s not perfectly resolving reach out to work with others from outside
strategy and some of the other tricks,” says every cell, I’m ok with that,” says Elana biology who handle large-scale data
Macosko. As smFISH evolved, it grew into Fertig, computational cancer biologist matrices. The computational road ahead is
a workhorse in his lab. The multiplexed at Johns Hopkins University, although daunting and exciting, she says. Her dream
in situ hybridization strategies are useful, in an ideal world, she would want in graduate school of measuring all genes
but they take time and are technically single-cell resolution. She works on tumor in all single cells is starting to come true.
demanding. “It’s not something that a heterogeneity, which spatial techniques can Beyond data size and dimensionality, says
traditional molecular biology or cell biology help to resolve. One of her projects assesses Zeng, she and her team work on integrating
lab can immediately get set up,” he says. T cell infiltration into tumors. The lab has and correcting for batch effects. Even data
He sought a different way to get cellular or a multi-omics focus, integrating data from captured with the same modality can differ
near-cellular resolution on a genome-wide different modalities, and the researchers use slightly from one batch to the next and
scale that leverages the widespread data from more than one spatial analysis have to be normalized. To integrate across
infrastructure of Illumina-based sequencing method. She finds 10x Genomics’ Visium modalities, such as imaging, sequencing and
in genomics labs. He and the team applied technology more accessible than, say, electrophysiology, she and her team build

Nature Methods | VOL 18 | January 2021 | 9–14 | www.nature.com/naturemethods 11

technologY featuRe | FOCUS
technology feature | FOCUS

clustering approaches and draw on different it’s being done for the first time, much of
types of machine-learning methods. what scientists do is difficult and expensive, Astrocytes
Much opportunity exists, says van says Zeng, but technology development
Mitral cells
Oudenaarden, to computationally combine helps to change that. Commercialization
single-cell RNA-seq and spatial information has been crucial in propelling single-cell
to map single cells back onto space and to sequencing, says Tasic, and the same can
build and use reference maps and tools. He happen in spatial technologies. “I’m very
is intrigued by computational approaches much looking forward to a commercial
to reconstructing spatial maps from say, solution for spatial transcriptomics that will
in situ hybridization reference databases work at the single-cell level,” she says. Th+

or approaches that use no maps. One such Spatial techniques, especially in situ
approach, novoSpaRc, does ‘gene expression techniques, “first took off probably the
cartography’ without maps. It’s based on most in neuroscience,” says Michael
assumptions about how gene expression Schnall-Levin, 10x Genomics senior vice Astrocytes
varies across a tissue section. Having ever president of research and development.
more data about tissues is a nice problem Other fields eager for spatial analyses are
to face, says van Oudenaarden. This wealth developmental biology, where people study Spatial techniques can reveal a tissue’s complex
makes it “an even bigger puzzle” to get to gradients and spatial patterns of molecules cell type mixtures — here, the mouse brain’s
the important “bones,” the core aspects of in cells, and cancer research, especially for olfactory bulb. Credit: Cai lab, Caltech, I.
biology. To him, the merging of single-cell the assessment of tumor heterogeneity and Strazhnik Adapted with permission from ref. 6,
’omics and spatial methods is “really a way T cell infiltration. Since it was founded, Springer Nature.
to generate hypotheses,” he says. One needs says Schnall-Levin, who has been with the
to confirm and validate findings with data company since its start, the teams began
collected separately. with microfluidic technology, long-read the approaches from ReadCoor and Cartana
sequencing approaches and a focus on as complementary to the technology
Commercial takes single-cell and spatial analysis. What began underpinning Visium. The company
Spatial technologies have a rich history, as a potential comarketing agreement keeps relations with academic labs after
says Shendure, such as in the labs of Mats with Spatial Transcriptomics turned into acquisitions, since they have expertise that is
Nilsson, George Church and others, who an acquisition in 2018. Visium is not yet important for guidance and it’s part of being
have been working on projects such as single-cell resolution, but the company is “good citizens” of the research community.
genotyping or sequencing directly on tissue working on ways to get there and “we’re In 2019, nanoString began selling its
sections for well over a decade. Spatial pretty far along,” he says. He and his team instrument, GeoMx, for spatially resolved
analysis blossomed into a field in the past also focus on in situ–based high-resolution tissue analysis. The device had been in
year or two, he says. That’s in part due to analysis, says Schnall-Levin. The company testing with early adopters. Some labs
the way the single-cell field has matured acquired ReadCoor, a spinout of the worked with prototypes and others sent
and technologies have been turned into Church lab at Harvard Medical School, their samples to the company’s Seattle
commercially viable instruments with and Cartana, a spinout from the lab of headquarters and used the instrument
broad usage. “This last point is still one Stockholm University and Science for Life remotely. “We learned so much from that,“
of the challenges, though,” says Shendure. Laboratory researcher Mats Nilsson. Cartana says Joe Beechem, chief science officer
“Many of the most exciting methods are uses padlock probes that hybridize to the and senior vice president of research and
still somewhat bespoke and really only mRNA of genes of interest, followed by development. “They’re sitting in their living
operational in one or a few labs,” he says of rolling circle amplification to amplify the room with a glass of wine, controlling it,
spatial methods to date. The question, he RNA on-site to prepare it for sequencing. laughing like little kids.” Early on, GeoMx
says, is how more of these methods can be ReadCoor uses barcoding and probes for could capture around 100 mRNAs; now that
more readily adopted by other groups. When fluorescence in situ RNA sequencing and number is over 22,000, which shows how
protein detection, too. Cartana brings to fast this field is moving, says Brad Gray,
10x Genomics an intellectual property nanoString president and chief executive
portfolio, chemistry and scientists with officer. nanoString launched as a spinout
know-how, which is a seed one can build from the Institute for Systems Biology
around in an accelerated fashion, says in Seattle with technology for optical
Schnall-Levin. ReadCoor brought scientists, molecular barcodes and is now, says Gray,
a “foundational technology,” intellectual all about spatial analysis. Early adopters
property and technology development, since of the GeoMx platform were mainly in
the team that built the company’s instrument immuno-oncology, where labs sought to
had worked though some “non-trivial” see, for example, whether T cells were
engineering constraints, he says. Between trapped in the periphery of a tumor instead
the two companies 10x received 110 of bringing their destructive power deep
patents, leading to a company total of 935 inside it, says Beechem. One day, right
patents, says Shernaz Daver, who advises before giving a talk, he scribbled on a
10x Genomics. These technologies enable napkin how the company’s oligo-barcodes
Spatially resolved transcriptomics reveals markers pinpointing molecules down to subcellular could spatially resolve tissue and deliver a
of somitogenesis expressed in stem-cell-based locations, says Schnall-Levin. For now, they readout on a second-generation sequencer.
models of development as in mouse. Credit: A. measure sets of a few hundred genes, but The company had previously developed a
van Oudenaarden, Hubrecht Inst. “obviously, that has room to grow.” He views way to attach barcodes to antibodies with

12 Nature Methods | VOL 18 | January 2021 | 9–14 | www.nature.com/naturemethods

 FOCUS | technologY
|
featuRe
FOCUS technology feature

photocleavable linkers. His idea was to

spread the linkers and barcodes on a tissue
slide. The oligos hybridize to mRNAs.
A microscope takes snapshots of the tissue
and then emits a burst of light, which
pops off the photocleavable linker. The
mRNAs and barcodes are then sequenced.
As Gray explains, their customers have
long requested ways to study whole
transcriptomes and gain higher resolution
down to compartments in cells so they can
ask: “Is it on the surface, is it in the nucleus,
is it attached to certain organelles?” The
market, in his view, is going to split into two
instrument classes. There will be profilers
such as GeoMx and Visium that look at
multicellular regions of interest in high
throughput. Then there will be instruments
with cellular and subcellular resolution,
which may or may not cover the whole With MERFISH, the Zhuang lab captured the expression of 10,050 genes in individual cells in human
transcriptome but will cover thousands of cancer cells. RNA molecules from different genes are shown in different colors. Credit: X. Zhuang
genes at a time. nanoString has just launched laboratory, Harvard U./HHMI
its GeoMx Digital Spatial Profiler, which
has single-cell resolution. Gray knows
nanoString’s competitors include Fluidigm’s circularized cDNA made from captured project funded by the US National Institutes
Hyperion Imaging System, an imaging transcripts and include primers and adapters of Health in which many labs work together
mass cytometer, and Ionpath’s MIBIscope. ready for BGI’s array-based sequencing. to produce reference human, mouse
The new nanoString technology also uses Each bar-coded readout has as its spatial and non-human primate brain atlases.
photocleavable tags for gene-expression address a specific pitch on the array. It involves imaging, electrophysiology,
profiling of tissue, including formalin-fixed The tracking lines on the chips give the and molecular genetic analyses such as
paraffin-embedded tissue. It is being coordinates of what is measured in each cell, transcriptomics and epigenomics. Work
applied, for example, to study spatial aspects says Xu. The spatial method draws on BGI’s on the Allen Mouse Brain Atlas began
in autopsy samples from people who died acquisition of technology from Complete with situ expression profiling of all mouse
of COVID-19. With technology of this type, Genomics, Inc. and is slated for release genes to show anatomical and spatial gene
says Beechem, researchers will have all the sometime in 2021. BGI has made chips for expression patterns. “It’s a reference database
power of all the molecular tools they had in spatial analysis of mouse brain with around that has been widely used,” says Zeng. She
the past; “they can just do it in space now.” 2,000 genes and 4,000 gene transcripts per directs BICCN’s whole-brain atlas ventures
At BGI Research, a high-resolution nanoball bin. The team is thinking beyond for the mouse brain. The team is building
method for spatially resolved gene mRNA to other ’omics, says Xu. Ultimately a “periodic system” of cell types in the
expression analysis is under development. the instrument can help researchers build mouse and human brain, says Tasic. She
The idea is to bridge the gap between human and model organism reference tissue leads a BICCN project on cell-type-specific
single-cell sequencing and spatial analysis maps; they might use this spatial technique tools for brain-wide labeling and ways to
of where cells sit in tissue, which together to assess how genetic variants affect tissue study circuits in cell-type-specific ways.
will shape the study of function as well as function, such as in cancer, or use it for The research community will soon have
structure, says Xun Xu, CEO of BGI Group, cross-species comparative approaches, such cell-type-specific reference atlases onto
who directs BGI Research. He watches other as tissue-based studies of evolution and which they can map their own findings.
company activities in this area and believes adaptation of organ structure. Among other papers, the BICCN teams
the BGI approach will have the highest Companies involved in spatial areas are recently published a multimodal cell census
resolution. The approach captures a few active software developers. Informatics has and atlas of the mammalian primary motor
hundred data points per cell, says Xu. It’s been a core aspect of the company, says cortex8. BICCN phase two is starting, says
packaged on a silicon chip as a nanoarray Schnall-Levin, whose team is committed Zeng. Among the new projects are ones
built with photolithographic techniques to open-source software. Researchers can devoted to cell-type targeting tools to
adapted from the semiconductor industry. add their algorithms and tools to the 10x apply spatial and molecular information,
BGI has used this approach to profile Genomics pipeline. As Xu explains, the perturb cells and characterize their roles in a
DNA–protein interactions, and Chris Ao team has worked on interactive visualization circuit and behavior. “That’s how you build
Chen, a scientist at BGI Research, says he tools “to give people a real view of the causality,” says Zeng.
and his colleagues are now working on ways tissues.” nanoString’s Gray says, “Joe’s largest For the recent atlas of the mouse primary
to embed cells or tissues on these chips department within R&D is now software.” motor cortex, the teams note that single
to yield high-resolution spatially resolved The company has more programmers than cell genomics has swept across biology,
data. The square compartments of the molecular biologists and engineers. including neuroscience, and catalyzed a
silicon chip’s array each hold round areas, transformation from labs being able to
each containing barcoded DNA nanoballs Atlas at hand classify and describe phenotypes to having
resolvable with conventional wide-field Spatial analysis is crucial for the Brain a mechanistic and explanatory molecular
imaging. DNA nanoballs are bundles of Initiative Cell Census Network (BICCN), a genetic framework for the cellular basis of

Nature Methods | VOL 18 | January 2021 | 9–14 | www.nature.com/naturemethods 13

technologY featuRe | FOCUS
technology feature | FOCUS

brain organization, which includes complete Spatial information clearly matters,

transcriptomes. For mapping the spatial says Shendure, “and we’re still in the early
organization of more than 300,000 cells days as far as where one can imagine the
from the mouse’s primary motor cortex technologies eventually going.” It’s exciting
by transcriptomic cell type, they chose that spatially resolved transcriptomics is the
Zhuang’s MERFISH9, multiplex error-robust Method of the Year, says Macosko, and “I
fluorescence in situ hybridization. They think it’s really just the beginning.” To dig
combined MERFISH with other methods into a question of interest, people will likely
such as Patch-seq, in which cells are use several of the many tools and techniques
measured with electrophysiology and in the spatial analysis field. Standards,
then sequenced. They analyzed and Hongkui Zeng (left), director of the Allen Institute metrics and benchmarks will help the spatial
compared mouse, marmoset and human for Brain Science, and Bosiljka Tasic, also at the field, just as they are helping the single-cell
transcriptomic cell types. Then they selected Allen Institute. Credit: Allen Institute for Brain genomics field, he says.
a panel of over 250 genes to image and then Science What started with single molecule FISH
cluster by single-cell expression profiles, and single-cell genomic analysis has become
finding “excellent correspondence” with a group of ways to bring the large-scale
clusters identified by single-nucleus and measurement, not a sequential measurement hypothesis-generating technologies of
single-cell RNA-seq. This let them discern that accumulates errors. As she and her genomics into cell biology and tissue
the spatial distribution of cell types and team developed MERFISH, they worked biology — one that will ground thinking
gain a view of cortical layers that refined on imaging, built-in error correction and a and technology development, says Macosko.
traditionally defined layers. They also noted way to deliver results on thousands of RNA Researchers will be able to do fine-grained
spatial distributions along medial–lateral species transcriptome-wide. “That took studies of gene expression in a spatial
and anterior–posterior axes. On occasion, us a while,” she says. The method achieves context and make other measurements,
disagreements erupt about the brain’s 80% detection efficiency. In a separate too, such as of enzymatic processes and
cartography, but Tasic believes advancing BICCN-based project, Zhuang and her interactions between cells, among genes
technologies will slowly help to settle these team used MERFISH to analyze more than and between proteins. They will be able to
disagreements, which are sometimes based one million cells in the preoptic region of barcode and anchor aspects of interest in
on labs using different methods. With a the mouse hypothalamus. In the brains of space, and this stands to change the way cell
common coordinate system in place, the female mice that had not previously given biology as well as pathology and histology
disagreements can shift, for example, to birth to pups, exposure to pups activated a are practiced, he says.
the necessary discussion about function. neuron that is also activated in female and Zeng sees many future opportunities.
“Very often one gene is not sufficient to male mouse parents but not active in males “The techniques are not, in many cases,
identify a cell type,” says Zeng. “Black and that have not fathered a litter. She and her they are not perfect yet,” she says. “We
white frequently does not exist,” says Tasic team also apply MERFISH to characterize want more and better,” she says, to scale
in terms of cell types. “You have multiple subcellular spatial organization to reveal up techniques and be able to measure
genes that are expressed at multiple different RNAs in different cellular compartments thousands of genes routinely and efficiently.
levels and together they make a cell what and the way chromatin is organized10. Spatial methods will become standard, says
it is.” The ability to look at thousands of Ultimately this work shows “how 3D Tasic. One day people will send out tissue
genes expressed in individual cells “has genome organization and transcription and get back spatially resolved single-cell
already changed the field dramatically,” are related to each other,” she says. Along genetic information, she says. Commonplace
says Zeng. Spatial transcriptomics adds with Harvard Medical School researcher reactions to papers in this area might
richness: “maybe not thousands of genes David Walt and Jeffrey Moffitt, a former become: “Oh, you didn’t measure all the
yet,” depending on the method, but it’s the postdoctoral fellow of hers who now has genes? Why not?” ❐
same idea: researchers get a view of many of his own Harvard Medical School lab,
a cell’s genes in a spatially localized way. she has founded a company, Vizgen, to Vivien Marx ✉
Zhuang is happy MERFISH is being used commercialize MERFISH. Nature Methods.
for brain atlas building. “I’m just personally ✉e-mail: [email protected]
very fascinated by the brain,” she says. In Future in space
MERFISH, RNAs are barcoded during As a microscopist, Zhuang is happy to Published online: 6 January 2021
hybridization rounds. Non-fluorescent see the fields of imaging and single-cell https://doi.org/10.1038/s41592-020-01033-y
targeting probes bind the mRNA and genomic analysis grow closer. KTH’s Ståhl
fluorescent readout probes hybridize to the looks forward to high-resolution spatially References
1. Crossetto, N., Bienko, M. & van Oudenaarden, A. Nat. Genet. 16,
targeting probes. The readout sequences resolved transcriptomics. “Maybe, the single 57–66 (2015).
yield a barcode distinct to each mRNA. cell, like ten micrometers, may be the sweet 2. Lein, E., Borm, L. E. & Linnarsson, S. Science 358,
The sequential rounds of hybridization, spot,” he says, that delivers enough data per 64–69 (2017).
3. Bergenstråhle, J. & Lundeberg, J. Bioessays 42, 1900221 (2020).
the barcoding and error correction make pixel. Zeng is happy about the useful spatial 4. Ståhl, P. L. et al. Science 353, 78–82 (2016).
it possible to capture many genes. The analysis techniques that have emerged. Many 5. Lubeck, E., Coskun, A. F., Zhiyentayev, T., Ahmad, M. & Cai, L.
challenge with hybridization rounds, of the existing methods lead up to single-cell Nat. Methods 11, 360–361 (2014).
6. Eng, C.-H. L. et al. Nature 568, 235–239 (2019).
however, is the accumulation of errors, RNA sequencing, which still gives the 7. Rodriques, S. G. et al. Science 363, 1463–1467 (2019).
says Zhuang. Each round has a small error highest sensitivity readout for quantifying 8. BRAIN Initiative Cell Census Network (BICCN) et al. Preprint at
rate individually, but the errors accumulate gene expression. One day, though, she hopes bioRxiv https://doi.org/10.1101/2020.10.19.343129 (2020).
9. Chen, K. H., Boettiger, A. N., Moffitt, J. R., Wang, S. & Zhuang, X.
and become significant, which risks “we can just get completely rid of single-cell Science 348, aaa6090 (2015).
gene misidentification. smFISH has high sequencing” for transcriptome-wide 10. Su, J.-H. et al. Cell https://doi.org/10.1016/j.cell.2020.07.032
accuracy, says Zhuang. But it’s a one-time single-cell spatial analysis. (2020).

14 Nature Methods | VOL 18 | January 2021 | 9–14 | www.nature.com/naturemethods