Content uploaded by Jakub Orzechowski Westholm
Author content
All content in this area was uploaded by Jakub Orzechowski Westholm on Mar 14, 2019
Content may be subject to copyright.
18. T. Alerstam, D. Christie, A. Ulfstrand, Bird Migration
(Cambridge Univ. Press, 1993).
19. F. Liechti, W. Witvliet, R. Weber, E. Bächler, Nat. Commun. 4,
2554 (2013).
20. N. C. Rattenborg, Naturwissens chaften 93,413–425
(2006).
21. J. A. Lesku et al., Science 337, 1654–1658 (2012).
22. H. Weimerskirch, M. Louzao, S. de Grissac, K. Delord, Science
335, 211–214 (2012).
23. U. C. Mohanty, M. Mohapatra, O. P. Singh, B. K. Bandyopadhyay,
L. S. Rathore, Monitoring and Prediction of Tropical Cyclones in
the Indian Ocean and Climate Change (Springer, Dordrecht,
Netherlands, 2014).
24. H. J. J. Jonker, T. Heus, P. P. Sullivan, Geophys. Res. Lett. 35,
L07810 (2008).
ACKNOW LEDGM ENTS
The tracking data presented in the paper are available
from the Dryad Digital Repository. We thank the Forces Armées
de la Zone Sud de l’Océan Indien for transport and logistical
support on Europa Island and the TAAF Administration
for allowing us to work on Europa Island. We thank the
fieldworkers involved in the study on Europa, in particular
J. B. Pons and R. Weimerskirch; R. Spivey for help with
preparing the electrocardiogram and acceleration tags
and for the data processing of the heart rate recording;
and A. Corbeau for help with data analyses. The study is a
contribution to the Program EARLYLIFE funded by a
European Research Council Advanced Grant under the
European Community’s Seven Framework Program FP7/
2007–2013 (grant agreement ERC-2012-ADG_20120314 to
H.W.). We thank Y. Ropert-Coudert, Y. Cherel, and two
anonymous reviewers for helpful comments on earlier versions
of the manuscript.
SUPPLEMENTARY MATERIALS
www.sciencemag.org/content/353/6294/74/suppl/DC1
Materials and Methods
Supplementary Text
Figs. S1 to S10
Table S1
References (25–34)
11 February 2016; accepted 20 May 2016
10.1126/science.aaf4374
TRANSCRIPTION
Visualization and analysis of gene
expression in tissue sections by
spatial transcriptomics
Patrik L. Ståhl,
1,2
*Fredrik Salmén,
2
*Sanja Vickovic,
2
†Anna Lundmark,
2,3
†
José Fernández Navarro,
1,2
Jens Magnusson,
1
Stefania Giacomello,
2
Michaela Asp,
2
Jakub O. Westholm,
4
Mikael Huss,
4
Annelie Mollbrink,
2
Sten Linnarsson,
5
Simone Codeluppi,
5,6
Åke Borg,
7
Fredrik Pontén,
8
Paul Igor Costea,
2
Pelin Sahlén,
2
Jan Mulder,
9
Olaf Bergmann,
1
Joakim Lundeberg,
2
‡Jonas Frisén
1
Analysis of the pattern of proteins or messenger RNAs (mRNAs) in histological tissue sections
is a cornerstone in biomedical research and diagnostics.This typically involves the visualization
of a few proteins or expressed genes at a time.We have devised a strategy, which we call “spatial
transcriptomics,”that allows visualization and quantitative analysis of the transcriptome with
spatial resolution in individual tissue sections. By positioning histological sections on arrayed
reverse transcription primers with unique positional barcodes, we demonstrate high-quality
RNA-sequencing data with maintained two-dimensional positional information from the mouse
brain and human breast cancer. Spatial transcriptomics provides quantitative gene expression data
and visualization of the distribution of mRNAs within tissue sections and enables novel types of
bioinformatics analyses, valuable in research and diagnostics.
Tissue transcriptomes are typically studied
by RNA-sequencing (RNA-seq) (1) of ho-
mogenized biopsies, which results in an
averaged transcriptome and loss of spatial
information. The positional context of gene
expression is of key importance to understand-
ing tissue functionality and pathological changes.
Several strategies have recently been developed
with this aim (2–5), but they have limitations in
the number of transcripts that can be analyzed,
rely on rich preexisting data sets, and/or are costly
and labor-intensive, and none of them are opera-
tional in the standard research and diagnostic
setting of regular histological tissue sections.
Weaskedwhetheritwouldbepossibletoin-
troduce positional molecular barcodes in the
complementary DNA (cDNA) synthesis reac-
tion within the context of an intacttissue section
before RNA-seq. We first assessed whether it was
feasible to generate cDNA from messenger RNA
(mRNA) in tissue sections on a surface. We im-
mobilized reverse-transcription oligo(dT) primers
on glass slides and placed on the slides sections
of adult mouse olfactory bulb, a brain region
with clear histological landmarks and ample gene-
expression reference data. The tissue was fixed,
stained, and imaged (Fig. 1A) (6).
After permeabilization, we added reverse-
transcriptionreagentsontopofthetissue.Weused
fluorescently labeled nucleotides to visualize the
synthesized cDNA (Fig. 1A and fig. S1). The tissue
was then enzymatically removed, which left cDNA
coupledtothearrayedoligonucleotidesontheslide
(6). The fluorescent cDNA showed a pattern in detail
corresponding to the tissue structure revealed by the
general histology (Fig. 1, B and C), and the cDNA was
strictly localized directly under individual cells (Fig. 1,
DtoG′). By comparing the hematoxylin-and-eosin
and fluorescent signals, we could measure the aver-
agedistanceofdiffusionoutsidetheborderofa
cell to 1.7 ± 2 mm (mean ± SD) (fig. S1, E to H).
The realization that it is possible to capture
mRNA in tissue sections with minimal diffusion
and maintained positional representation moti-
vated us to array oligonucleotides with positional
barcodes (Fig. 2A), and we denoted this strategy
“spatial transcriptomics.”We deposited ~200 million
oligonucleotides in each of 1007 features, with a
diameter of 100 mm and a center-to-center distance
of 200 mm, over an area of 6.2 mm by 6.6 mm (fig. S2).
After capturing and reverse-transcribing mRNA,
we generated sequencing libraries based on
amplification by in vitro transcription (fig. S3, A
and B) (7,8). Comparison with data from RNA
extracted and fragmented in solution revealed
that ~95% of the genes found with one of the
methods was also found with the other (fig. S3C).
The correlation between the surface and in-solution
libraries was r= 0.94, with even representation
of genes having high or low expression (fig. S3D).
Replicates of surface-based experiments of adja-
cent tissue sections showed a correlation of r=
0.97 (fig. S3E). Thus, cDNA synthesis from tissue
with arrayed oligonucleotides on a surface is ef-
ficient and does not introduce bias compared
with in-solution protocols (fig. S3F and table S1).
We sorted the RNA-seq data to its correspond-
ing array features by using the spatial barcodes
and aligned the tissue image with the features of
the array, which enabled visualization and analy-
ses. Examples of gene-expression patterns revealed
by spatial transcriptomics and validation by in situ
hybridization are shown in Fig. 2B and fig. S4, A to
C. Transcripts expressed at very low levels, such as
olfactory receptor mRNAs (9), were also detected
with spatial transcriptomics (fig. S4D).
The number of genes (10) (Fig. 2C) and unique
transcripts (fig. S5A) per individual feature varied
between cell layers with different cell density (Fig.
2D and table S2). For the vast majority of genes,
the coefficient of variation decreased as the aver-
age expression increased (fig. S5B). The number of
78 1JULY2016•VOL 353 ISSUE 6294 sciencemag.org SCIENCE
1
Department of Cell and Molecular Biology, Karolinska
Institute, SE-171 77 Stockholm, Sweden.
2
Science for Life
Laboratory, Division of Gene Technology, KTH Royal Institute
of Technology, SE-106 91 Stockholm, Sweden.
3
Department
of Dental Medicine, Division of Periodontology, Karolinska
Institute, SE-141 04 Huddinge, Sweden.
4
Science for Life
Laboratory, Department of Biochemistry and Biophysics,
Stockholm University, Box 1031, SE-171 21 Solna, Sweden.
5
Division of Molecular Neuroscience, Department of Medical
Biochemistry and Biophysics, Karolinska Institute, SE-17177
Stockholm, Sweden.
6
Department of Physiology and
Pharmacology, Karolinska Institute, SE-17177 Stockholm,
Sweden.
7
Division of Oncology and Pathology, Department of
Clinical Sciences Lund, Lund University, SE-223 81 Lund,
Sweden.
8
Department of Immunology, Genetics and
Pathology, Uppsala University, SE-751 85 Uppsala, Sweden.
9
Science for Life Laboratory, Department of Neuroscience,
Karolinska Institute, SE-171 77 Stockholm, Sweden.
*These authors contributed equally to this work. †These authors
contributed equally to this work. ‡Corresponding author. Email:
joakim.lundeberg@scilifelab.se
RESEARCH |REPORTS
on July 1, 2016http://science.sciencemag.org/Downloaded from
genesandtranscriptscapturedwasatleasttwice
as high as when using laser capture microdis-
section (11), and spatial transcriptomics detected
almost twice as many genes as examination by in
situ hybridization in the Allen Brain Atlas (fig.
S5, C and D). Furthermore, we compared spatial
transcriptomics with the near-100% sensitivity of
single-molecule fluorescent in situ hybridization
in adjacent tissue sections. The sensitivity of spa-
tial transcriptomics was 6.9 ± 1.5% of single-
molecule fluorescent in situ hybridization (fig.
S6). By comparison, single-cell RNA sequencing
has been reported to have about 5 to 40% sen-
sitivity (12).
To further assess the potential lateral diffusion
of transcripts, we investigated the distribution of
the expression of 10 different genes with highly
enriched expression in the mitral cell layer (MCL),
SCIENCE sciencemag.org 1JULY2016•VOL 353 ISSUE 6294 79
Fig. 1. Spatially localized cDNA synthesis. (A) The tissue is sectioned, placed onto oligo(dT) primers, stained, and imaged. cDNA synthesis with Cy3-labeled
nucleotides reveals fluorescent cDNA after tissue removal. (B) Hematoxylin-and-eosin staining of olfactory bulbs and (C) fluorescent cDNA after tissue removal.
Scale bar, 500 mm. (Dand E) Magnification of boxes in (B) and (C). Cell layers: GL; OPL, outer plexiform layer; MCL; and GCL. Arrowheads and boxes indicate
individual cells and corresponding cDNA with overlapping positions. Scale bar, 40 mm. (Fto G′) Cells in (D) and (E) magnified showing cytoplasm and
corresponding cDNA.
RESEARCH |REPORTS
on July 1, 2016http://science.sciencemag.org/Downloaded from
80 1JULY2016•VOL 353 ISSUE 6294 sciencemag.org SCIENCE
Fig. 2. Spatially resolved gene expression. (A) Each array feature contains
uniqueDNA-barcodedprobescontainingacleavagesite,aT7amplificationand
sequencing handle, a spatial barcode, a unique molecular identifier (UMI), and
an oligo(dT) VN-capture region, where V is anything but T and where N is any
nucleotide. cDNA (red) is generated from captured mRNA by reverse tran-
scription. (B) Visualization of the expression of three genes by spatial tran-
scriptomics (top) and in situ hybridization (bottom). Penk and Kctd12 in situ
images are from the Allen Institute. Cutoff normalized counts, Penk,8;Doc2g,
13; and Kctd12,19.(C) Distribution of unique genes per feature under the
tissue. (D) Number of genes detected for different layers and entire tissue over
sequencing depth. (E) Lateral diffusion of transcripts from genes enriched in
MCL. The genes are expressed in MCL features but are not separable from the
background in features adjacent to the MCL. (F) Spatial expression and in situ
hybridization of four genes in (E).The leftmost feature overlaps the MCL, and
the three rightmost features are situated in the GCL.The colored bar depicts
the distances from feature center in (E).
RESEARCH |REPORTS
on July 1, 2016http://science.sciencemag.org/Downloaded from
andweaskedwhethertheycouldbedetectedin
the adjacent granular cell layer (GCL). All these
genes were confirmed to be highly expressed in
the MCL by spatial transcriptomics, but they were
undetectable or detected at very low levels within
the GCL, even with the border of the feature 0 to
5mm and the center of the feature 50 to 55 mm
from the MCL (Fig. 2, E and F, and fig. S7A).
Furthermore, we compared the distribution of
transcripts between areas obtained with laser cap-
ture microdissection (6) where there is no diffusion
of transcripts and with spatial transcriptomics
features, and we did not find evidence for a
difference between these methods in terms of
mRNA diffusion (fig. S7, B and C).
A common goal of gene expression analysis of
tissues is to define the transcriptome of specific
areas. Analysis between homologous regions re-
vealed very similar expression profiles (Fig. 3, A
and B, and fig. S8), with no differentially expressed
genes. In contrast, comparison of different domains
revealed different gene expression profiles (Fig.
3, A and C, and fig. S8). This included genes with
previously known restricted expression, such as
Doc2g in the glomerular layer (GL) and Penk in
the GCL (13), as well as novel layer-specific gene
expression profiles (Fig. 3C).
It is valuable to explore the gene expression
pattern of populations of cells or tissue domains
that can be defined by a combination of markers.
Spatial transcriptomics offers an alternative ap-
proach that circumvents multiplex labeling and
cell isolation. Any combination of presence or
absence of expression for a set of genes can be
used to define a marker profile of interest for
further analysis. Features were selected on the
basis of the presence and/or absence of the three
interneuron-marker genes Camk4,Th,andVip.
The distribution of features, where one of the
genes is expressed alone, is shown in Fig. 3D.
Comparing gene expression revealed specific
transcriptomes defined by these interneuron-
marker profiles (Fig. 3, E and F, and fig. S8).
To further explore gene expression profiles in
spatially defined domains within the olfactory bulb,
we used principal component analysis (fig . S9) or
the t-distributed stochastic neighbor embedding
(t-SNE) (14,15) machine-learning algorithm for
dimensionality reduction, followed by hierarchi-
cal clustering (Fig. 4A). When placing back the
clustered features on the tissue images, it was ap-
parent that each cluster of features largely corre-
sponded to well-defined morphological layers (Fig.
4B).Theclusterswerethencomparedwitheach
other, which allowed the identification and visu-
alization of cluster-specific marker genes (fig. S10,
A and B). This proved to be an efficient, unbiased
way to identify genes with expression enriched in
the cell layers of interest. Furthermore, we in-
vestigated the gene expression pattern in 10 sec-
tions from a total of five animals, as well as the
feature-to-feature correlation at the same location
in two adjacent sections (fig. S10, C to E).
Analysis of the histology and a set of markers
are routine in cancer diagnostics, although anal-
ysis of the expression of panels of genes has
started to enter the clinic. We asked whether
adding a spatial dimension to gene expression
analysis may add information in cancer diag-
nostics and applied spatial transcriptomics to
breast cancer biopsies. In Fig. 4, C and D (see also
fig. S11, A and B), an area with invasive ductal
SCIENCE sciencemag.org 1JULY2016•VOL 353 ISSUE 6294 81
Fig. 3. Visualization and bioinformatics analyses of tissue domains de-
fined by morphology or gene expression profile. (A) Ten selected features
in areas a(GCL), b(GCL), or g(GL) are indicated. (B) Scatterplot of gene
expression in areas aand bshows similar expression of layer-specific genes.
Examples of genes are indicated with purple and brown dots. Housekeeping
genes are orange. (C) Scatterplot of gene expression in areas aand gshows a
difference in gene expression. Examples from the 170 differentially expressed
genes are labeled. (D) The spatial expression of three interneuron-marker–
gene profiles.Ten features with the different expression profiles were randomly
selected for differential expression analysis. (E) Comparing the 10 Camk4
+
/Vip
–
/
Th
–
features with the 10 Vip
+
/Camk4
–
/Th
–
features. Examples, out of the 196
differentially expressed genes, are labeled. (F) Comparing the 10 Camk4
+
/Vip
–
/
Th
–
features with the 10 Th
+
/Camk4
–
/Vip
–
features. Examples from the 328
differentially expressed genes are labeled.
RESEARCH |REPORTS
on July 1, 2016http://science.sciencemag.org/Downloaded from
cancer, as well as six separate areas of ductal can-
cer in situ, were identified on the basis of mor-
phological criteria. Spatial transcriptomics analysis
of the invasive component revealed high expres-
sion of extracellular matrix–associated genes (Fig.
4E). Analysis of the ductal cancer in situ areas
revealed a surprisingly high degree of heteroge-
neity in gene expression between these regions,
probably reflecting different subclones, with vary-
ing expression of several genes implicated in can-
cer progression (Fig. 4E and fig. S11C). For example,
expression of KRT17 and GAS6, implicated in
epithelial-to-mesenchymal transition (16,17), was
high only in areas 1 and 5 (Fig. 4, C to E, and fig
S11). Thus, spatial transcriptomics revealed un-
expected heterogeneity within a biopsy, which
would not be possible to detect with regular tran-
scriptome analysis and which may give more de-
tailed prognostic information.
Spatial transcriptomics calls for only a few ex-
tra steps compared with RNA-seq analysis of
homogenized tissue, with the benefit of providing
spatial information enabling additional levels of
analysis. In contrast to standard methods, different
domains of the tissue are processed in the same
reaction in spatial transcriptomics, which removes
technical variation between samples. A unique fea-
ture of spatial transcriptomics is that any gene ex-
pression profile can be selected to specify a
molecularly defined domain for further analysis.
Finally, in contrast to when different regions of a
tissue are dissected for analysis, the information for
the whole section is maintained; hence, the analysis
is not limited to the initially selected regions. An
individual spatial transcriptomics experiment thus
serves as a permanent resource to investigate gene
expression patterns for future research questions.
REFERENCES AND NOTES
1. Z. Wang, M. Gerstein, M. Snyder, Nat. Rev. Genet. 10,57–63 (2009).
2. N. Crosetto, M. Bienko, A. van Oudenaarden, Nat. Rev. Genet.
16,57–66 (2015).
3. R. Satija, J. A. Farrell, D. Gennert, A. F. Schier, A. Regev,
Nat. Biotechnol. 33, 495–502 (2015).
4. P. A. Combs, M. B. Eisen, PLOS ONE 8, e71820 (2013).
5. K. Achim et al., Nat. Biotechnol. 33, 503–509 (2015).
6. Materials and methods are available as supplementary
materials on Science Online.
7. T. Hashimshony, F. Wagner, N. Sher, I. Yanai, Cell Reports 2,
666–673 (2012).
8. R. N. Van Gelder et al., Proc. Natl. Acad. Sci. U.S.A. 87,
1663–1667 (1990).
9. R. Vassar et al., Cell 79, 981–991 (1994).
10. K. D. Pruitt et al., Nucleic Acids Res. 42 (D1), D756–D763 (2014).
11. S. Zechel, P. Zajac, P. Lönnerberg, C. F. Ibáñez, S. Linnarsson,
Genome Biol. 15, 486 (2014).
12. D. Grün, A. van Oudenaarden, Cell 163, 799–810 (2015).
13. E. S. Lein et al., Nature 445, 168–176 (2007).
14. A. Mahfouz et al., Methods 73,79–89 (2015).
15. L. J. P. van der Maaten, G. E. Hinton, J. Mach. Learn. Res. 9,
2579–2605 (2008).
16. M. Kittaneh, A. J. Montero, S. Glück, Biomarkers Cancer 5,
61–70 (2013).
17. C. Gjerdrum et al., Proc. Natl. Acad. Sci. U.S.A. 107, 1124–1129
(2010).
ACKNO WLE DGME NTS
We thank K. Meletis and M. Nilsson for discussions. This study
was supported by Knut och Alice Wallenberg Foundation, the
Swedish Foundation for Strategic Research, the Swedish Research
Council, the Swedish Cancer Society, the Karolinska Institute,
Tobias Stiftelsen, Torsten Söderbergs Stiftelse, Ragnar Söderbergs
Stiftelse, StratRegen, Åke Wiberg Foundation, and the Jeansson
Foundations. P.L.S. was supported by a postdoctoral fellowship
from the Swedish Research Council. We thank the Swedish
National Genomics Infrastructure hosted at SciLifeLab, as well as
the Swedish National Infrastructure for Computing–Uppsala
Multidisciplinary Center for Advanced Computational Science and
Bioinformatics Long-Term Support for providing sequencing and
computational assistance and infrastructure. The sequencing data
are deposited at the National Center for Biotechnology Information,
NIH, with BioProject ID PRJNA316587. Gene counts and scripts can
be downloaded from www.spatialtranscriptomicsresearch.org.
P.L.S., F.S., J.L., and J.F. are authors on patents applied for by
Spatial Transcriptomics AB covering the technology.
SUPPLEMENTARY MATERIALS
www.sciencemag.org/content/353/6294/78/suppl/DC1
Materials and Methods
Figs. S1 to 11
Tables S1 and S2
References (18–25)
12 January 2016; accepted 31 May 2016
10.1126/science.aaf2403
82 1JULY2016•VOL 353 ISSUE 6294 sciencemag.org SCIENCE
Fig. 4. Comparative analyses of tissuedomains. (A) t-SNE analysis and hierarchical clustering of 551 features from two replicates creates five distinct clusters.
(B) The features placed back onto the two tissue images. (Cand D) Histological section of a breast cancer biopsy (C) containing invasive ductal cancer (INV) and
six separate areas of ductal cancer in situ (1 to 6), with analyzed spatial transcrip tomics features in (D). INVareas without, or with minimal, stromal infiltration were
selected. (E) Gene expression heat map over the different areas in four adjacent sections (D) and (fig. S11).
RESEARCH |REPORTS
on July 1, 2016http://science.sciencemag.org/Downloaded from
(6294), 78-82. [doi: 10.1126/science.aaf2403]353Science
Joakim Lundeberg and Jonas Frisén (June 30, 2016)
Pontén, Paul Igor Costea, Pelin Sahlén, Jan Mulder, Olaf Bergmann,
Mollbrink, Sten Linnarsson, Simone Codeluppi, Åke Borg, Fredrik
Michaela Asp, Jakub O. Westholm, Mikael Huss, Annelie
José Fernández Navarro, Jens Magnusson, Stefania Giacomello,
Patrik L. Ståhl, Fredrik Salmén, Sanja Vickovic, Anna Lundmark,
by spatial transcriptomics
Visualization and analysis of gene expression in tissue sections
Editor's Summary
, this issue p. 78Science
can do so for multiple genes.
performing reverse transcription followed by sequencing and computational reconstruction, and they
annealing fixed brain or cancer tissue samples directly to bar-coded reverse transcriptase primers,
have developed a way of measuring the spatial distribution of transcripts byet al.sequencing. Ståhl
methods typically lose positional information and many require arduous single-cell isolation and
RNA-seq and similar methods can record gene expression within and among cells. Current
Spatial structure of RNA expression
This copy is for your personal, non-commercial use only.
Article Tools
http://science.sciencemag.org/content/353/6294/78
article tools:
Visit the online version of this article to access the personalization and
Permissions http://www.sciencemag.org/about/permissions.dtl
Obtain information about reproducing this article:
is a registered trademark of AAAS. ScienceAdvancement of Science; all rights reserved. The title
Avenue NW, Washington, DC 20005. Copyright 2016 by the American Association for the
in December, by the American Association for the Advancement of Science, 1200 New York
(print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last weekScience
on July 1, 2016http://science.sciencemag.org/Downloaded from
